Introduction

New York’s Diamond District glitters less brightly than its name might imply. You can find all sorts of things there: hawkers muttering to themselves in decrepit doorways, crumpled envelopes skittering in the streets, discarded once deflated of cash, fresh street grime blackening their edges where someone once licked.

You can also find many a true artist.

These artists are typically men, short in stature and long in years, skin leathered and blackened with jeweler’s rouge like the markings of a curse. You can watch them bent over their workbenches, focus ferocious, transferring minuscule glimmers from one station to the next with pipette-like tools worn down beyond recognition. This is no factory floor; here, the phrase ‘at scale’ would elicit expressions equally agog and askance.

Which is fine, because there is no need for scale: people only need one engagement ring or wedding band, maybe two if they’re optimistic post-divorce.

This process — bent in supplication over a bench, precision over priceless objects, in its own time — calls to mind stem-cell manufacturing at today’s scale: artisanal. But in the case of stem cell-derived islets, you need much more than one or maybe two of the artist’s wares. With a therapeutic dose being around half a million islet equivalents (IEQs), you need billions of cells for each person.

Induced pluripotent stem cells (iPSCs) are mature adult cells that are genetically reprogrammed to behave like embryonic stem cells, meaning they can differentiate into virtually any type of cell. Cells on demand! And in the words of Sana CEO Steve Harr, almost every single disease is caused by one of two things: a missing cell, or a damaged cell. The implication being, if you could generate any cell in the body on demand, you could cure all diseases. The challenging part is creating safe, high-quality versions of those cells at the scale of billions per person.

Twenty years after the discovery of induced pluripotent stem cells, the field is facing a coming-of-age moment. The science has advanced to the point where many promising cell therapies, including islet replacement for type 1 diabetes, are closer than ever.

So what’s holding these therapies back from the clinic? Now, instead of scientific conundrums, manufacturing and scaling challenges define cell therapy’s next frontier.

The State of Stem Cell-Derived Islet Manufacturing Today

Creating iPSC islets involves immense variability and seemingly endless mission-critical decisions, much like type 1 diabetes itself. These decisions include differentiation protocols, equipment ranging from manual plasticware to the latest 3D bioreactors, what circadian rhythm staging during islet development, and, perhaps most crucially, choice of master cell line or starting material. Not to mention quite a few more unknown unknowns.

Any and all of these variables determine the quality of the final islet product. For example, when Horikawa and colleagues tested fifteen clinical-grade cell lines that they guided through two differentiation protocols, they found staggering variation: the highest-performing line was twenty times more effective at producing insulin than the lowest-performing. When the end product is a living organism, even slight variations in the manufacturing process have an avalanche of downstream effects.

Islets are extremely biologically complex. By all accounts, the beta cell is a card-carrying international cell of mystery, complete with a dark rumored past, a trail of broken hearts left in its wake, and probably a tiny black fedora styled after old noir films. And whole islets take the intrigue to another level. In the words of Raphael Scharfmann and Arnaud Zaldumbide, “We may not have all the tools to study the complex crosstalk between alpha and beta cells, or even further between the beta cells and other endocrine cells” (emphasis mine).

Like all stem cell manufacturing, iPSC islet manufacturing carries a high risk of off-target cells. Protocols inevitably generate unintended cell types or polyhormonal cells (cells that awkwardly secrete both insulin and glucagon simultaneously). And if even just a few cells among billions remain undifferentiated or become off-target proliferative subversives, they risk forming tumors.

The flip side of this complexity is that the field is making new and potentially transformative discoveries about iPSC islet biology every day. But exploring these new discoveries is not an excuse to avoid tackling unanswered questions of manufacturing and scaling.

Part of islet manufacturing’s challenge is mindset. To scale to the millions of people who would benefit from this therapy, stem cell scientists must learn to think about the product of their work differently. As the field matures, they can no longer afford to think of the end product as the answer to a research question; they must instead think of it as the curative, life-changing therapy that it is.

This perspective is part of a tidal shift sweeping the biomedical field toward “patient centricity,” or epistemic justice — thinking always of people with T1D as people, with moment-to-moment lived experiences, rather than in abstraction as data points or subjects or … patients. The goal of islet manufacturing is not to answer a research question, or even to manufacture the best islets; it is to transform real people’s lives.

Strategies to Scale Stem Cell Islet Manufacturing

The Process Systems Engineering Approach

Thankfully, stem cell manufacturing is not the first field to encounter astronomically high-stakes scientific challenges. Consider your local power plant, the source of the energy that allows you to be online reading this, whether chemical or nuclear. Nuclear engineering’s stakes are so high that they consumed the collective psyche of Western culture for nearly half a century. And for good reason. Manufacturing nuclear power — and other highly volatile products essential to the infrastructure of everyday life — nearly puts stem cell manufacturing to shame with its levels of complexity, volatility, obscenely expensive equipment, and specialized labor that are positively mind-blowing.

How could this possibly be managed? The answer lies in a process systems engineering approach. Was that the most boring sentence you’ve ever read? Definitely in the running. But that is what we need right now.

Process systems engineering (PSE) is a formal toolkit that integrates supervisory process control, dynamic scheduling, real-time optimization, and predictive modeling of individual unit operations for managing processes that are dynamic, time-constrained, and inextricably interdependent (Nieri et al., 2026). This approach “reframes the manufacturing challenge from one of artisanal repetition to one of system-level design and control, offering a structured route to reduce turnaround time, contain batch-to-batch variability, and increase throughput without compromising product quality” (Nieri et al., 2026).

For manufacturing iPSC islets and other cell therapies, a process systems engineering approach has three parts: standardization toward interoperability, real-time monitoring and feedback loops, and machine-assisted strategies.

Standardization Toward Interoperability

In 2017, Stephen Sullivan was serving as Program Manager for the Global Alliance for iPSC Therapies. He sent a questionnaire to its several hundred members asking two questions that seemed pretty routine: How do you define quality in your cells? How do you measure it? He was gobsmacked when the answers came back, in his words, all over the place.

So he started mailing cells to members: blinded samples with instructions to test them as the lab normally would and send back the results. The results of this exercise exposed that the field was considerably less aligned than anyone believed (and least of all on flow cytometry, that critical measure of cell identity).

There must be a unified definition of what a clinical-grade iPSC islet product should look like (sterility, genomic stability, marker expression, dimensions, etc.) and how to measure it and reproduce it. This should include platforms, reagent specifications, assay methods, and, especially, the data formats results get reported in, so that the findings can build on each other.

Most cell manufacturing today is open. A lab tech breaks a seal, moves material across the open air to the next station, and repeats that sequence dozens of times across a month. Each transfer introduces variability, room for contamination and error: the force with which any given object was handled, how long the vessel sat open, whose hands handled it and the state of that person’s microbiome that day, and so on.

Ultimately, the goal of standardization is to remove as many variables as possible, then define the remaining variables and put them into conversation with one another — interoperability. Closed system processing is the paradigm for achieving this.

This closed-system principle explains why the most promising iPSC islet scale-up approach keeps the entire differentiation inside a single suspension vessel. Within this closed system, the consistent conditions turned a fivefold increase in vessel size into a twelvefold increase in output.

Real-Time Monitoring and Feedback Loops

Scaling exacerbates unmeasured microenvironmental risks to islets. At-scale volumes also mean higher stakes and a greater imperative to prevent losses. While a run lost in a research lab costs a flask, a run lost in a 0.5-liter vessel costs 183,000 islet equivalents, nearly half a therapeutic dose.

Real-time monitoring can prevent losses like these with early detection. Today’s imaging technology can detect cluster size and morphology. Raman and near-infrared spectroscopy can track what the cells are consuming without removing any of them, which matters more the larger the batch gets. Systems that monitor the cell product in real time enable alerts and intervention at critical points. This supports scalable operations by allowing one person to supervise many loops rather than being manually responsible for each one.

Real-time monitoring also provides invaluable training data for algorithms that can improve monitoring itself. This creates a feedback loop in which continuous learning drives process improvements for fewer failures, higher yields, and probably other factors that the naked human eye and mind alone cannot conceptualize but that will prove critical to production.

Machine-Assisted Strategies

Machine-assisted strategies for scaling cell therapy manufacturing are twofold: automation and algorithmic support.

Cell therapies rely heavily on skilled operators with many years of training, talent which is few and far between. Automation can chip away at this labor gap while also reducing waste from potential human error. A machine dispensing reagents into a sealed vessel delivers the specified volume at the specified rate every time. Much better than a shaky, exhausted hand during a late night in the lab. Highly dexterous automation can also perform painstaking tasks at a scale that is simply impossible for humans. For example, automated single-cell seeding deposits one cell per well and images the well for two weeks to prove the colony descended from that one cell — a task Janet Rothberg of CCRM describes as unverifiable and nearly impossible to scale when done by hand with a pipette.

Continuous real-time monitoring produces more data and more opportunities for algorithmic decision support. More insight helps on a batch-to-batch basis, but its deeper value is in building objectively better processes over time. While the field still lacks the high-fidelity datasets to train these systems fully, it may not for much longer. CCRM built its DeepTech Bio Lab on this premise, pairing with IonQ on quantum computing toward building the best digital twin of a cell. Along these lines, GenBio AI recently released AIDO Cell, a model that predicts how a cell’s molecular machinery behaves under the weight of the world’s conditions.

And what begins as a manufacturing tool could evolve into an end-to-end learning system that integrates clinical insights as feedback into R&D and manufacturing. Connecting insights from actual clinical use will translate real-world experiences into better therapies.

Conclusion

Scaling cell therapy manufacturing has a long road ahead. But the infrastructure is beginning to catch up to the science. In February, Japan granted the world’s first conditional approvals for iPSC-derived therapies. In the US, pivotal trials are running in several indications. As for the Diamond District, it may lose its shine to reactor-generated gems. But as diamonds are becoming increasingly lab-grown, cell therapies are finally making their way out of the lab and into the world. The Canadian cyberpunk sci-fi writer William Gibson said it best: “The future is already here — it’s just not evenly distributed yet.”

***

References

🧬 PREGAME: Helena Hu — Boyden & Langer Labs, MIT

Thousandfold Expansion Microscopy: Toward a Complete Molecular Map of Life

TheSugarScience T1D Th1nk Tank

📅 Registration

Date: Tuesday, September 8, 2026 · 9:00 AM Pacific · 12:00 PM Eastern Format: Free virtual seminar for the global T1D research community.

👉 Register here on Zoom

👤 About the Speaker

Helena Hu is a PhD student in Biological Engineering at MIT, jointly advised by Professor Ed Boyden — inventor of optogenetics and expansion microscopy — and Professor Bob Langer, one of the most cited engineers in history and a pioneer of controlled drug delivery and tissue engineering. Her research sits at the outer boundary of what is physically observable in biology: she is extending expansion microscopy to the scale at which individual amino acid residues of proteins become visible using ordinary light microscopes.

The ambition of her work is stated with precision on the Boyden lab website: “Molecules exert their function by physically contacting other molecules, and these interactions occur at distances below one nanometer. This is the resolution at which signaling occurs and therapeutics intervene.” Her goal — mapping all biomolecules and their interactions across cells and tissues to build predictive, bottom-up models of living systems — is not a modest one. Today’s paper is the first major step toward it.

01 · The Paper

“Thousandfold Expansion Microscopy” Hu H, Krah D, Ntolkeras A, Chanda S, Heimbrodt A, Mondal M, Altendorf J, Jing B, Berger B, Shaib AH, Rizzoli SO & Boyden ES · bioRxiv, June 2026 DOI: 10.64898/2026.05.31.729018

To understand why this paper matters — and why it matters specifically for T1D — it helps to understand what has been invisible until now.

The pancreatic islet is one of the most studied tissues in medicine. Decades of microscopy, electron microscopy, and increasingly powerful super-resolution techniques have revealed its architecture at the level of cells and organelles. But the functional events that determine whether a beta cell secrets insulin correctly, responds to stress appropriately, or presents a neoantigen to an autoreactive T cell — all of these occur at the scale of molecular interactions: protein-protein contacts, receptor-ligand binding, post-translational modifications, conformational changes. These events happen at distances of 1–10 nanometers. Until this paper, no light microscope could see them.

The innovation: The Boyden lab hypothesized that individual protein residues could be imaged by anchoring their side chains to a swellable polymer, cleaving backbone amide bonds, and expanding residues away from each other to a degree that enables them to be visualized separately.</cite> The result is 1000ExM — a four-network interpenetrating hydrogel architecture. The chemistry is sequential: a first expansion network anchors biomolecules; additional interpenetrating networks amplify the expansion multiplicatively. This enables successive expansion from approximately 18-fold to greater than 1000-fold — one billion-fold in volume.</cite>

To put that in physical terms: a gel the size of a few centimeters, once fully expanded, reaches the volume of an Olympic swimming pool. Structures that were separated by one nanometer in the original tissue are now separated by one millimeter — well within the resolution of a standard confocal or widefield microscope.

The validation: Protein and peptide structures are maintained across these expansion factors, as verified by analyses of proteins with known structures (nanobodies and GFP) and a well-studied peptide (mCLING).</cite> This is a critical technical achievement: expanding a sample a thousand-fold while preserving molecular relationships and structural information is far harder than simply expanding the gel. Deformation, loss of epitopes, and spatial distortion are all failure modes that had to be characterized and controlled. The use of nanobodies and GFP — proteins with known crystallographic structures at sub-Ångstrom resolution — as internal calibration standards provides the most rigorous available test of structural fidelity. Computational analysis indicates that 1000ExM resolves adjacent amino acid residues, thereby achieving sub-nanometer precision on conventional light microscopes.</cite>

This means that for the first time, a researcher can take a piece of islet tissue, embed it in a 1000ExM gel, expand it a thousand-fold, and image individual protein residues — not just protein blobs, not just organelles, but the actual molecular architecture of the proteins that govern beta cell function — using an instrument that already exists in most university core facilities.

The T1D application: The islet is an ideal early target for 1000ExM precisely because its molecular pathology is so well characterized at the cellular level but so poorly understood at the nanoscale. The insulin secretory granule membrane fusion machinery (SNAREs, synaptotagmins, Munc18), the MHC class I peptide loading complex that presents islet antigens to autoreactive CD8⁺ T cells, the gap junction channels (connexin 36) that coordinate calcium waves across the islet, the neoantigen structures formed by post-translational modifications of beta cell proteins — all of these are molecular-scale phenomena that 1000ExM could resolve for the first time in native tissue.

Tags: expansion microscopy · 1000ExM · super-resolution · nanoscale imaging · protein structure · amino acid resolution · hydrogel · islet biology · beta cell · T1D · molecular mapping · Boyden lab · MIT · bioimaging

02 · Why This Matters

For scientists: The history of cell biology is largely a history of new microscopes creating new fields. The electron microscope revealed organelle ultrastructure. Confocal microscopy enabled three-dimensional imaging of intact tissues. Super-resolution techniques (STORM, PALM, STED) broke the diffraction barrier and made 20–50 nm resolution routine. Each transition unlocked a new layer of biological understanding. 1000ExM is the next transition — from the scale of organelles and protein complexes to the scale of individual amino acid residues, in fixed tissue, using equipment that labs already own. For islet biologists, this means the molecular architecture of the secretory machinery, the immune synapse, and the stress response can now be studied with structural precision that was previously available only in purified protein crystallography — but now in the biological context of the islet itself.

For clinicians: The most therapeutically actionable insights from 1000ExM for T1D are likely to come from two directions. First, neoantigen structural biology in situ: the post-translational modifications that generate T1D autoantigens (hybrid insulin peptides, citrullination, transamidation) occur at specific molecular sites on beta cell proteins, in specific cellular compartments, under specific stress conditions. Understanding the structural context of these modifications — which proteins are adjacent, which chaperones are present, what the local molecular environment looks like — could inform both the development of antigen-specific tolerance therapies and the design of biomarkers that detect neoantigen formation. Second, the molecular basis of beta cell dysfunction: the difference between a functionally competent beta cell and a stressed, dysfunctional one is ultimately a difference in molecular organization — in the arrangement of SNAREs at the granule membrane, in the conformation of ion channels, in the clustering of receptors. 1000ExM can map these differences at the amino acid level.

The broader picture: Helena Hu’s stated goal is to build predictive, bottom-up models of living systems by mapping all biomolecules and their interactions across cells and tissues. This is not hyperbole — it is a research program. The Human Cell Atlas, the Allen Brain Cell Atlas, and comparable projects have mapped cell types and gene expression at single-cell resolution. 1000ExM could be the tool that extends those maps to the molecular interaction layer: not just which genes are expressed, but where every protein is, what it is touching, and what conformation it is in. For T1D specifically, a molecular atlas of the human islet — in health, in prediabetes, in early autoimmunity, and in established T1D — built with 1000ExM and anchored in nPOD tissue, would be the most comprehensive picture of beta cell pathology ever assembled.

03 · Four Questions We Will Ask the Speaker

Q1. The validation of structural fidelity uses nanobodies, GFP, and mCLING — proteins with known crystallographic structures. But islet proteins like proinsulin, SNARE complexes, and connexin 36 hemichannels have very different biophysical properties. What expansion artifacts are most likely when you apply 1000ExM to densely packed secretory granule membranes or tight junction complexes — and how do you characterize and correct for deformation at those interfaces?

Q2. Achieving sub-nanometer precision requires not just physical expansion but also labeling that doesn’t introduce spatial error. Conventional antibody labels are 10–15 nm in size — comparable to or larger than the structures you’re trying to resolve. How does 1000ExM approach the labeling problem: are you using nanobodies, click chemistry, or direct chemical anchoring to the polymer — and what is the effective spatial uncertainty introduced by the labeling step?

Q3. The paper demonstrates 1000ExM in model proteins and peptides. What does the workflow look like for applying this to human pancreatic tissue from nPOD donors — specifically, what are the practical constraints around fixation, sectioning, and expansion of post-mortem human tissue that differs from fresh cell culture preparations?

Q4. The ultimate application you describe is a complete molecular map of cellular interactions — mapping all biomolecules and their contacts. For the islet, where would you start? Is there a specific molecular interaction — an autoantigen modification, a secretory complex, a gap junction nanodomain — that you believe 1000ExM could resolve in the next two years and that would most directly change our understanding of T1D pathogenesis?

04 · Four Key Associated Papers

1. Boyden ES et al. (Chen F, Tillberg PW & Boyden ES) (2015) Expansion microscopy Science, 347(6221):543–548 · PubMed → The original expansion microscopy paper — the foundational method that 1000ExM extends a thousandfold. Reading this first establishes the conceptual framework: physical magnification of fixed specimens by embedding in a swellable hydrogel, enabling standard microscopes to achieve super-resolution.

2. Chang J-B, Chen F, Yoon Y-G, … Boyden ES (2017) Iterative expansion microscopy Nature Methods, 14:593–599 · PubMed → The iterative ExM paper achieving approximately 20-fold expansion — the prior art that 1000ExM improves upon by ~50-fold and one billion-fold in volume. Essential for understanding the engineering advances in the four-network hydrogel architecture.

3. Delong T, Wiles TA, Baker RL, … Haskins K (2016) Pathogenic CD4 T cells in type 1 diabetes recognize epitopes formed by peptide fusion Science, 351(6274):711–714 · PubMed → The hybrid insulin peptide paper — establishing that T1D autoantigens include non-canonical molecular structures formed by post-translational peptide fusion events. These are precisely the nanoscale molecular modifications that 1000ExM could for the first time visualize in their native cellular context.

4. Stanton AE, Kang J, Blanchard JW, … Langer R, Boyden E & Tsai LH (2026) Expansion revealing of pathology resolves nanostructures associated with inflammatory phenotypes in COVID-19 decedent human brain tissue bioRxiv · Read preprint → The most recent expansion microscopy application from the same Boyden/Langer ecosystem — applying expansion microscopy to post-mortem human disease tissue (COVID-19 brain) to resolve inflammatory nanostructures. The closest existing proof of concept for applying this class of technology to human disease pathology, directly analogous to what 1000ExM could do in nPOD islet tissue.

05 · Four Videos to Watch First

▶ State of the Science: Bioimaging in Human T1D — Challenges, Progress, and the Roadmap to Clarity The essential TSS primer on the pancreatic imaging landscape in T1D — covers the biological and technical challenges that motivate the kind of nanoscale structural resolution 1000ExM now offers. Watch this first to understand what the field has been missing.

▶ Ask the Expert: Matthew Merrins, PhD & Richard Benninger, PhD — 3D Light-Sheet Imaging of the Islet The current state of the art in high-resolution live islet imaging — essential context for appreciating the spatial and temporal gap between existing light microscopy of islets and the sub-nanometer structural resolution that 1000ExM achieves in fixed tissue.

▶ Ask the Expert: Jacob Hecksher-Sørensen — Gubra, Live Demo of Diabetes Imaging A live demonstration of whole-pancreas 3D imaging for diabetes research — another dimension of the imaging frontier that 1000ExM’s molecular-resolution approach would complement: one zooms out to the whole organ, the other zooms in to individual amino acids.

▶ Ask the Expert: Joana Almaça, PhD — University of Miami Almaça works on the islet microenvironment and pericyte-beta cell interactions using live imaging — her focus on the nanoscale cellular contact events that regulate islet function is the closest biological parallel in the TSS library to the molecular interaction mapping that 1000ExM enables.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

REGISTER for this THOUGHT PROVOKING PANEL

Advancing T1D Research Toward Curative Therapies- Lessons Learned and New Directions

David M. Harlan, Matthias von Herrath, Bart O. Roep

Anyone attending a research conference devoted to understanding type 1 diabetes (T1D) pathogenesis so as to rationally develop curative therapy will invariably hear someone point out that that lofty goal has for decades been promised “in 5 more years.” Indeed, even in retrospect, there was reason for that optimism based on then recent research breakthroughs, only for that optimism to be fractured when new and unanticipated problems were encountered. In the mid-1970s, Robert M. Pirsig published Zen and the Art of Motorcycle Maintenance, a philosophical novel that quickly became a must read for young people of that era. In Zen, Persig coined a phenomenon he called the “gumption trap”, i.e. an incorrect assumption that prevents the victim from observing something that would otherwise be quite obvious. In that context, in the mid-1980s our mutual friend and great scientist and mentor George Eisenbarth 1^st proposed that T1D was a human anti- β-cell specific autoimmune disorder. Dr. Eisenbarth had ample reason for that hypothesis including the ability to predict disease susceptibility based upon anti- β-cell specific autoantibodies, genetic studies which found many immune genes (HLA in particular) profoundly influenced T1D risk, the presence of T cells in and around pancreatic islets (insulitis) for those who died shortly after T1D onset, and the fact that immunosuppressive drugs could delay the inexorable decline in endogenous insulin producing capacity that occurs in those afflicted with T1D. And yet, over the past, nearly 100 immuno-therapeutic clinical trials have been completed at a cost of several billion dollars, with only one (anti-CD3, or Teplizumab) making it to the marketplace. A remarkable achievement, to be sure, but that therapy is very expensive and merely delays the onset of hyperglycemia for 2 to 3 years and for only a subset of those treated. While not a popular thing to ask among scientists who have devoted careers in pursuit of the T1D cure based upon developing an immunotherapy, we ask the rhetorical question, “Have we been lulled into the “gumption trap” that has blinded us to leads we might otherwise have pursued?

Such thinking is no longer so apocryphal. Over the past several years, many research groups have pursued research themes that fall quite outside “the gumption trap” that T1D is an anti- β-cell specific autoimmune disorder, and we will highlight some of those themes. We will also point out some other topics not (yet) as widely discussed.

Starting with increasingly well appreciated and researched observations, T1D is NOT an anti-β-cell specific disease. Evidence going back decades has existed demonstrating that pancreatic exocrine dysfunction exists in individuals with T1D. Recent studies have also shown that the entire pancreas is smaller in individuals at genetic risk for developing T1D and long before any evidence of anti- β-cell autoimmunity can be identified. Similarly, recent clinical trial data suggests that subtle hyperglycemia (perhaps reflecting β-cell dysfunction or inadequate mass) precedes the earliest serologic evidence of autoimmunity. In other words, might a diseased β-cell or islet environment be initiating an immune response?

Many groups are now pursuing other studies that suggest the β-cell is not a passive victim of the immune system and that some β-cell pathology may have triggered an immune response. Indeed, a histologic hallmark of developing T1D (i.e. observed in individuals found to be anti- β-cell autoantibody positive and yet with normal glycemia at their death) is general islet cell HLA class I overexpression. It is certainly formally possible that some event has caused general pancreatic islet inflammation, to which the β-cell is simply the most susceptible.

All immunologists agree that the anti-β-cell immune response differs markedly from better-understood immune responses. For example, although the entire β-cell mass weighs only a few grams, the immune process that is associated with β-cell dysfunction and death is remarkably slow, appearing to unfold over years, if not decades. Even among identical twins discordant for T1D (only one afflicted for a long period yet ultimately fated to develop the disease), or among individuals with multiple anti-β-cell autoantibodies, hyperglycemia may not develop for decades. It is also quite true that T1D appears to be quite heterogeneous in other ways. This raises a central question: how do we identify the ideal subjects for therapy, and when should immunotherapy be given if the outcome it seeks to prevent may still be decades away? Together, these often overlooked but well-known observations underscore that the optimal timing of immunotherapy remains unresolved. In general, although there are many exceptions, a therapy is more likely to have a favorable risk–benefit profile when it is specific and targets a well-understood disease mechanism. We argue that such understanding is still lacking in T1D. Indeed, experimental immunologists all agree that it is very difficult to break tolerance to a particular cell type or tissue. The factors causing that break in self-tolerance for those destined to develop T1D remains obscure. Trials to date

Last, and to our knowledge not often discussed, a seminal feature of the immune system is its memory. It is the feature underlying vaccine efficacy such that one or a few doses can provide long-term protection from that pathogen. And yet, the vast majority of immunotherapeutic trials employ agents with little or no antigen specificity meaning that even if effective in delaying the loss of β-cell function and mass, one might anticipate other immune responses to be similarly weakened. That prediction has been validated for the most aggressive past immunotherapeutic approaches (e.g. the autologous bone marrow transplant studies of the early 2000s). Fortunately, other immunotherapeutic approaches have to date not been marred by infectious complications. And yet, such immunosuppressive agent complications can be rare and observed late, for example the rare but well validated sepsis risk splenectomized patients face from encapsulated bacteria We suggest therefore that all patients enrolled in future immunotherapeutic trials designed to delay or treat T1D be enrolled in a lifelong monitoring program to look for an increased risk for infection or cancer. This is especially important since modern insulin-based therapies for T1D, while expensive, imperfect, and troublesome, are known to promote an excellent prognosis.

One possible alternative approach would be research efforts to increase β-cell mass or resilience to stressors. It is known, for instance, that if the immune system tries unsuccessfully to destroy a target, that immune response can fatigue creating a détente of peaceful coexistence. In recent years, many groups have begun working on strategies to promote β-cell proliferation, or to create heartier β-cells by altering their metabolic or cellular neighborhood, gene modification, or by developing drugs that target β-cell susceptibilities. Any future curative or preventative therapy will also need to be cost effective if it is to reach the millions afflicted with T1D worldwide.

📅 Registration

Date: Tuesday, September 1, 2026 · 12:00 PM Eastern Time Format: Free virtual seminar for the global diabetes research community — clinicians and scientists are welcome.

👉 Register here on Zoom

Figure adapted from DOI: 10.3389/fgene.2026.1833788.

👤 About the Speaker

Michael Robben, PhD, is an Assistant Professor at the University of Illinois, Urbana-Champaign (Department of Animal Science, Division of Nutritional Science, and Institute for Genomic Biology), researching environmental factors that change the risk for developing Type 1 Diabetes. His group published one of the first single-cell RNA and TCR-seq studies of NOD mouse pancreatic T cells (Islam et al., 2025). As a computational immunologist, he develops machine learning models that combine sequencing data with clinical disease risk factors, and works to improve in-silico approaches to disease monitoring — necessary, he argues, as sequencing becomes more accessible and datasets grow larger and more multi-omic.

📄 The Paper

Yang L, Lindahl A, Robben M (2026). Immunogenetics and TCR expression patterns together predict future T1D onset from multi-omic datasets. Frontiers in Genetics 17:1833788. DOI: 10.3389/fgene.2026.1833788

What the paper actually found:

• The problem: T1D is currently flagged via autoantibodies (AAB) — but single-AAB detection carries a 50–60% false-positive rate, and up to 20% of true cases are missed entirely. T cells drive the disease mechanistically but have proven much harder than B cells to turn into a clinical biomarker, since TCR epitopes are numerous and restricted by each patient’s own HLA genetics.

• The dataset: A meta-analysis of 993 TCR profiling files across 7 studies (321 patients, ~118 million unique clones, via the iReceptor database), plus 1,430 bulk whole-blood RNA-seq files from 4 studies (466 patients), plus a single-cell RNA-seq dataset of 220,338 T cells from 35 donors (T1D, single-AAB+, and healthy).

• Individual CDR3 clones don’t generalize. Any two patients shared less than 1% of clones on average, and only 6 CDR3 sequences were significantly enriched in case samples out of tens of millions.

• TCR gene usage is the more tractable signal. Aggregating by V/D/J gene segment (rather than exact clone sequence) was the most disease-correlative feature overall, though its predictive value depended heavily on the patient’s HLA allelic risk background and on timing of sample collection.

• Genetics + TCR beats either alone. A GLM combining HLA identity and TCR gene abundance outperformed TCR-only models. Training on later (”terminal”) timepoint samples to predict earlier ones pushed accuracy to 92%, with both false-positive and false-negative rates under 10%.

• Recurring genes: TRBV11-3, TRBV11-2, TRBV10-1, and TRBV7-3 were consistently important across statistical models, multiple ML approaches (Random Forest, SVM, gradient boosting, etc.), and polygenic models built around classic high-risk DR3/DR4 HLA alleles.

• Same signal recoverable from cheaper bulk RNA-seq, not just dedicated TCR sequencing — meaning a standard RNA-seq blood test could in principle capture both HLA genotype (99.95% per-patient accuracy via T1K) and TCR gene signal in one assay.

• Functional relevance, not just correlation: T cells carrying certain TCR genes (e.g., TRBV5-1) were disproportionately represented in Tregs and memory subsets, and disease-relevant TCR gene clusters showed differential expression of activation, cytokine, and cytotoxicity genes.

• Mouse validation: dPCR on NOD mouse PBMCs confirmed candidate TCR genes (including mouse homologs of TRBV11-2 and TRAV12-1) show the predicted disease-timing dynamics ahead of clinical progression.

• Bottom line: no single universal TCR biomarker works across all genetic backgrounds — but genotype-specific (”polygenic”) modeling enables meaningful marker selection at the individual level, potentially deliverable via one affordable RNA-seq test.

  ---

📌 Why This Matters

For scientists: A rare large-scale unification of 7 TCR-seq studies, 4 RNA-seq studies, and a single-cell dataset under one modeling framework — it directly addresses why T-cell repertoire biomarkers have historically underperformed relative to autoantibodies.

For clinicians / clinical researchers: HLA genotyping plus TCR gene expression — both obtainable from one relatively cheap RNA-seq run — could form a new prognostic blood test substantially outperforming current AAB screening, particularly relevant given that roughly 85% of new T1D cases have no family history and go undetected until symptomatic.

For the broader field: A case study in how immunogenetics-aware machine learning changes what “predictive marker” means — usefulness is conditional on the patient’s own HLA background, with implications for other polygenic autoimmune diseases.

❓ Questions for the Speaker

1. Individual CDR3 clones don’t generalize (<1% overlap between patients), while TCR gene-level usage is far more robust. What’s the biological explanation for why gene segment usage carries more reproducible signal than exact clonal sequence?

2. TRBV11-2 and TRBV11-3 recur across very different modeling approaches. Beyond their association with known T1D-relevant epitope-specific clones (Table 1), is there a mechanistic hypothesis for why these particular genes are so central?

3. Training on terminal samples to predict non-terminal ones reached 92% accuracy overall, but type I error rose substantially in the RNA-seq-derived version. What’s the main obstacle to closing that gap before clinical translation?

4. The HLA-only model hit near-perfect accuracy due to high multicollinearity — flagged in the paper as an overfitting risk rather than a real result. How do you disentangle genuine TCR-driven signal from genetic confounding going forward?

5. Polygenic models only had enough sample size for a subset of allele combinations. What’s the minimum viable cohort size/diversity needed to make this clinically actionable across a genetically diverse population, rather than DR3/DR4-enriched cohorts?

   ---

📚 Key Associated Papers

1. Nakayama M, Michels AW (2021). Using the T cell receptor as a biomarker in type 1 diabetes. Front. Immunol. 12:777788. https://doi.org/10.3389/fimmu.2021.777788 — the direct conceptual precursor, used to cross-reference TCR genes against known T1D-specific clones.

2. Islam MZ, Zimmerman S, Lindahl A, et al. (2025). Single-cell RNA-seq reveals disease-specific CD8+ T cell clonal expansion... in diabetic NOD mice. PLOS ONE 20:e0317987. https://doi.org/10.1371/journal.pone.0317987 — Robben’s own prior NOD mouse work.

3. Ishigaki K, Lagattuta KA, Luo Y, et al. (2022). HLA autoimmune risk alleles restrict the hypervariable region of T cell receptors. Nat. Genet. 54:393–402. https://doi.org/10.1038/s41588-022-01032-z

4. Rawat P, Shapiro MR, Peters LD, et al. (2026). Identification of a type 1 diabetes–associated T cell receptor repertoire signature from the human peripheral blood. Sci. Adv. 12:eadx7448. https://doi.org/10.1126/sciadv.adx7448 — a contemporaneous, partly conflicting approach.

5. Suomi T, Starskaia I, Kalim UU, et al. (2023). Gene expression signature predicts rate of type 1 diabetes progression. eBioMedicine 92:104625. https://doi.org/10.1016/j.ebiom.2023.104625

6. McGrail C, Sears TJ, Griffin EN, et al. (2026). Genetic association and machine learning improve the prediction of type 1 diabetes risk. Nat. Genet. 58:1062–1072. https://doi.org/10.1038/s41588-026-02578-y

   ---

Videos to Watch First

▶ Ask the Expert: Abhijeet Patil, PhD, University of Pennsylvania — Modeling Type 1 Diabetes Progression using Machine-Learning and Single-Cell Transcriptomic Measurements in Human Islets
https://www.youtube.com/@thesugarscience8602/search?query=Abhijeet%20Patil%20Machine%20Learning%20Type%201%20Diabetes
The closest prior TSS talk to Robben’s own methodology — applying ML to genomic/transcriptomic data to model T1D progression.

▶ T1D Th1nk Tank: Brian Corrie, PhD — iReceptor Technical Director
https://www.youtube.com/@thesugarscience8602/search?query=Brian%20Corrie%20iReceptor
Corrie leads iReceptor, the exact immune repertoire database Robben’s team pulled 993 TCR profiling files from for this study.

▶ Debate — Autoantibodies in T1D: Biomarkers or Mediators of Destruction?
https://www.youtube.com/@thesugarscience8602/search?query=Debate%20Autoantibodies%20T1D%20Biomarkers%20Mediators%20Destruction
Robben’s paper opens by citing the real limitations of AAB-based screening as the motivating gap for a TCR-based alternative.

▶ Panel: What Would the Earliest Detection of Type 1 Diabetes Look Like?
https://www.youtube.com/@thesugarscience8602/search?query=Panel%20Earliest%20Detection%20Type%201%20Diabetes
A direct framing companion to the field-level question Robben’s paper offers one concrete computational answer to.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

CLICK here to view full version

Week in Review includes: notable T1D papers, upcoming Think Tanks and panels from the(sugar)science, on-demand talks, upcoming events for the T1D field.

CLICK here to view full version

Week in Review includes: notable T1D papers, upcoming Think Tanks and panels from the(sugar)science, on-demand talks, upcoming events for the T1D field.

CLICK here to view full version

Week in Review includes: notable T1D papers, upcoming Think Tanks and panels from the(sugar)science, on-demand talks, upcoming events for the T1D field.

📅 Registration

Date: August 18, 2026 · 12:00 PM Eastern Format: Free virtual seminar for the global T1D research community — clinicians and scientists are welcome.

👉 Register here on Zoom

Figure adapted from DOI: 10.3390/cells15110992 ·

👤 About the Speaker

Dr. Tzu-Wen Hong is a Research Fellow in the Department of Diabetes and Obesity, School of Cardiovascular and Metabolic Medicine & Sciences at King’s College London, funded by the Steve Morgan Foundation and King’s Health Partners Diabetes, Endocrinology and Obesity (KHP DEO) programme. Her work sits at a productive intersection of cell biology, regenerative medicine, and translational diabetes research — specifically focused on how supporting cells can be leveraged, either directly or through their secreted products, to improve the functional survival of both cadaveric human islets and stem cell-derived beta cells.

Her scientific journey is unusually international and methodologically diverse. She completed her MSci and PhD at the University of Tokyo, where her doctoral research centered on the physiology of bile acid receptor signaling — studying the roles of FXR and TGR5 in energy metabolism during and after fasting, and characterizing the regulatory mechanisms and physiological functions of TGR5 activation. After her PhD, she undertook postdoctoral training at Imperial College London and the MRC Laboratory of Medical Sciences, where she investigated how oestrogen and estrogen receptor signaling operate within bone, using transgenic mouse models targeting multiple mesenchymal cell subtypes. This grounding in mesenchymal cell biology — how stromal cells communicate, regulate, and support neighboring tissue — now informs her current work at King’s, where she applies that lens to the pancreatic islet microenvironment.

At King’s, Hong works in the laboratory of Professor Peter Jones, whose group has spent more than a decade systematically characterizing how mesenchymal stromal cells (MSCs) improve the functional survival of isolated islets — demonstrating beneficial effects of MSC co-culture in vitro, of MSC co-transplantation in vivo, and of defined MSC-secreted peptide cocktails on human islet glucose-stimulated insulin secretion and cytokine-induced apoptosis. Hong’s current work takes this program a step forward by asking whether the full beneficial effect of MSCs can be recapitulated by their secreted extracellular vesicles alone — replacing a complex living cell with a purified, potentially GMP-grade, cell-free product. Her research interests span metabolic regulation, tissue repair and remodelling, and the mechanisms of tissue regeneration — all threads that converge in the question she is pursuing: what does the islet microenvironment need to survive, and can we deliver it without the cell?

01 · The Paper

Hong T-W, Caxaria S, Daniels Gatward LF, Hussain S, Zhao M, King AJF, Rackham CL & Jones PM (2026) Mesenchymal Stromal Cells Improve Islet β-Cell Functional Survival: Analysis of Extracellular Vesicle-Trafficked Proteins and miRNAs · Cells, 15(11):992 DOI: 10.3390/cells15110992 · https://www.mdpi.com/2073-4409/15/11/992

The problem that motivates this work is one of attrition. Clinical islet transplantation offers the most physiologically complete beta cell replacement available for type 1 diabetes, but the procedure is badly hampered by rapid, extensive loss of beta cell functional mass — up to 60% of transplanted beta cells perish within the first 72 hours after implantation. They face a hostile environment: ischemia, hypoxia, inflammatory cytokines from the instant blood-mediated inflammatory reaction, and the absence of the supporting stromal cell niche they occupied in the native pancreas. Even before transplantation, isolated islets lose around 20% of their beta cell mass every 24 hours during the culture period required for quality testing and recipient preparation.

One approach to mitigating this loss is co-incubating or co-transplanting islets with mesenchymal stromal cells, which have well-documented anti-inflammatory, pro-survival, and trophic properties. Multiple research groups, including the Jones laboratory at King’s College London, have demonstrated that MSC co-culture maintains islet functional viability in vitro, reduces cytokine-induced apoptosis, and improves the outcomes of islet transplantation in experimental rodent models. But MSCs are living, complex, heterogeneous cells that are difficult to characterize, scale, and manufacture to GMP standards. The practical question is: what exactly are MSCs doing to help islets survive — and can that protective effect be captured in something simpler and more scalable?

Hong and colleagues address this by testing whether MSC-derived extracellular vesicles (MSC-EVs) — membrane-enclosed nanoparticles released by MSCs that carry a cargo of proteins, miRNAs, and other bioactive molecules — are alone sufficient to reproduce the beneficial effects of whole MSCs on islet functional survival. Extracellular vesicles have emerged as a major mechanism of intercellular communication, enabling cells to transfer functional molecular cargo to neighboring cells without direct contact. If the active molecules responsible for MSC’s islet-protective effects are trafficked via EVs, the EVs themselves could serve as defined, cell-free therapeutic agents — easier to characterize, manufacture, and potentially administer than living MSC co-cultures.

The study uses proteomics (LC-MS) and small RNA sequencing to comprehensively characterize the molecular cargo of MSC-EVs, and functional assays to assess their effects on islet survival and insulin secretion.

The key findings:

MSC-EVs alone are sufficient to recapitulate key beneficial effects of whole MSCs on islet functional survival. Without the parent MSC cells present, MSC-EVs treated islets showed improved glucose-stimulated insulin secretion and protection from cytokine-induced apoptosis — demonstrating that the EV cargo, rather than ongoing paracrine signaling from living MSCs, carries a meaningful portion of the islet-protective activity.

Proteomic analysis of MSC-EV cargo identified more than 100 proteins, including β-cell GPCR agonists. Among the proteins identified from the Uniprot Mouse Database were G protein-coupled receptor (GPCR) agonists — specifically including annexin A1 (ANXA1), stromal cell-derived factor-1 (SDF-1/CXCL12), and complement component C3a — that the Jones lab had previously shown, when applied as a defined recombinant cocktail, to enhance beta cell functional survival. This convergence between the unbiased proteomics result and prior hypothesis-driven work provides mutual validation: the protective molecules the lab had identified through rational screening are genuinely present in the EVs that reproduce the MSC effect.

MSC-EVs contain approximately 300 distinct miRNAs; five are highly enriched and significantly upregulated in MSC-EV-treated islets. Small RNA sequencing revealed a rich miRNA cargo, with five miRNAs notably enriched and measurably elevated in islets following MSC-EV treatment. The most prominent was miR-21a-5p — a miRNA with a complex and context-dependent role in beta cell biology, previously shown by other groups to both promote and inhibit beta cell apoptosis depending on origin and cellular context. In the MSC-EV setting, the enrichment of miR-21a-5p in treated islets suggests active transfer of miRNA cargo that may reprogram islet gene expression toward a more survival-competent state.

MSC-EV treatment altered a distinct set of islet mRNA targets involved in islet metabolism and function. Transcriptomic analysis of MSC-EV-treated islets identified a specific signature of upregulated and downregulated mRNAs associated with beta cell metabolic and functional pathways — consistent with EV cargo actively reprogramming islet gene expression rather than simply modulating cytokine signaling at the surface.

The findings open a path toward GMP-grade defined cell-free pretreatment protocols. Rather than using heterogeneous whole MSC populations (which vary by donor, passage, and culture conditions), the characterization of the active cargo — specific GPCR-agonist proteins and miRNAs — creates the possibility of replacing the MSC co-culture step with a defined cocktail of purified, recombinant, or synthetic molecules that could be manufactured to pharmaceutical grade and applied consistently across islet preparations before transplantation.

Tags: MSC · mesenchymal stromal cells · extracellular vesicles · exosomes · islet transplantation · beta cell survival · miR-21a-5p · miRNA · proteomics · GPCR · ANXA1 · SDF-1 · CXCL12 · C3a · cell-free therapy · GMP · T1D · beta cell apoptosis · islet functional mass · King’s College London

02 · Why This Matters

For scientists: The framing shift in this paper is methodologically important: from asking whether MSCs help islets (established) to asking what specifically in MSC-EVs accounts for that help (open question). The combination of unbiased cargo profiling with functional validation in a disease-relevant model is a rigorous approach that yields both discovery and mechanistic hypothesis. The identification of GPCR agonists in the EV proteome that are independently known to be beta-cell-protective — a convergence of unbiased and hypothesis-driven evidence — is particularly compelling.

The miR-21a-5p finding opens an interesting complexity. In some contexts, miR-21 has been shown to promote beta cell apoptosis via BCL2 degradation; in others, MSC-derived exosomal miR-21-5p has been shown to protect transplanted islets by suppressing PDCD4. The context dependence of this miRNA in the islet setting — whether the MSC-EV source results in a protective or damaging effect — is exactly the kind of question that now needs to be resolved mechanistically, and this paper positions the field to ask it rigorously.

More broadly, the EV approach addresses a key bottleneck in MSC-based therapies generally: the difficulty of manufacturing living cells at consistent quality and scale. If the therapeutic activity resides in the EV cargo rather than in the live cell, then the path to clinical translation runs through GMP EV production, cargo standardization, and defined molecular pretreatment protocols — a substantially more tractable regulatory and manufacturing challenge than whole MSC co-transplantation.

For clinicians: The clinical implication is direct: better islet transplant outcomes with a more practical, scalable tool. The loss of beta cell mass in the immediate post-transplantation period is the central technical limitation of clinical islet transplantation — it is the reason most patients require islets from multiple donors to achieve insulin independence, a constraint that severely limits the number of patients who can benefit from a given donor pancreas. Any treatment that reliably reduces that early post-transplant cell death translates directly into better outcomes per donor — more patients achieving insulin independence from a single donor, and potentially lower total islet mass requirements.

A defined, GMP-grade cell-free pretreatment — derived from EV cargo characterization — would be particularly attractive because it could be added to existing isolation and culture protocols without introducing additional regulatory complexity from living cell co-transplantation. Stem cell-derived beta cells face the same attrition problem as cadaveric islet transplants, and a pretreatment that improves their functional survival during the engraftment period would be equally valuable for that growing clinical program.

The broader picture: Hong’s work connects two converging fields: the islet transplantation improvement field (how do we protect more of the beta cells we transplant?) and the extracellular vesicle therapeutics field (can EVs be developed as defined cell-free medicines?). Both have been advancing rapidly and largely independently. This paper is a productive point of intersection — demonstrating that MSC-EV cargo carries real islet-protective activity, characterizing that cargo at molecular resolution, and identifying specific targets that could anchor a next-generation defined pretreatment protocol. That protocol would be relevant not just to cadaveric islet transplantation but to the rapidly advancing clinical program of SC-derived beta cell replacement.

03 · Four Questions We Will Ask the Speaker

Q1. You showed that MSC-EVs alone can recapitulate key protective effects of whole MSC co-culture — improved GSIS and reduced cytokine-induced apoptosis. How complete is that recapitulation? Are there aspects of MSC protection on islets — such as support for revascularization, modulation of allograft immune responses, or effects on endothelial cell function within the islet — that EVs alone do not capture, and how important are those missing effects for transplantation outcomes?

Q2. miR-21a-5p was the most prominent miRNA enriched in MSC-EV-treated islets, and its role in beta cell biology is complex — other groups have shown it can both promote and inhibit apoptosis depending on context. Do you have evidence that the miR-21a-5p transferred by MSC-EVs in your system is protective rather than damaging, and what downstream targets would you prioritize investigating to resolve its functional role in the islet setting?

Q3. The proteomics identified GPCR agonists including ANXA1, SDF-1/CXCL12 and C3a as EV cargo — molecules the Jones lab had previously validated individually as beta cell-protective. Does the EV-delivered version of these molecules appear more or less potent than exogenously applied recombinant versions, and do you think the vesicular delivery format provides a pharmacological advantage — for example, through enhanced uptake or intracellular trafficking — beyond what the free proteins achieve?

Q4. You mention that these findings may enable simplification of the islet pretreatment strategy by focusing on defined, GMP-grade biologically active molecules. What would a realistic next step toward clinical translation look like — are you moving toward testing a defined synthetic cargo cocktail, or toward GMP EV production as a pharmaceutical product? And how would you approach the scalability challenge of EV manufacturing for a clinical pretreatment protocol?

04 · Four Key Associated Papers

1. Hong T-W, Caxaria S, Daniels Gatward LF, Hussain S, Zhao M, King AJF, Rackham CL & Jones PM (2023) Mesenchymal stromal cell secretory molecules improve the functional survival of human islets · Diabetic Medicine, 40(12):e15227 · READ HERE · The direct predecessor to today’s paper, also co-authored by Dr. Hong, establishing that a defined cocktail of three MSC-secreted peptides — annexin A1 (ANXA1), SDF-1/CXCL12, and complement C3a — improves glucose-stimulated insulin secretion and protects human islets from cytokine-induced apoptosis in vitro, and supports islet function in vivo in a human-to-mouse transplantation model. This paper provides both the biological rationale and the specific molecular hypotheses that today’s EV proteomics study confirms: the GPCR agonists found in the MSC-EV cargo are the same molecules the lab had identified through rational secretome screening. Essential reading for understanding why specific proteins in the EV cargo matter.

2. Hubber EL, Rackham CL & Jones PM (2021) Protecting islet functional viability using mesenchymal stromal cells · Stem Cells Translational Medicine, 10(5):659–668 · READ HERE · A comprehensive review from the Jones laboratory covering the full evidence base for MSC-mediated protection of islet functional viability — including co-culture effects, co-transplantation outcomes, and the role of the MSC secretome. This paper explains why the field moved from whole-MSC approaches toward secretome and EV-based strategies: the need for scalability and mechanistic clarity. Reading this review establishes the conceptual journey that led directly to Hong’s EV characterization work.

3. Chen J, Chen J, Cheng Y, Fu Y, Zhao H, Tang M, Zhao H, Lin N, Shi X, Lei Y, Wang S, Huang L, Wu W, Tan J (2020) Mesenchymal stem cell-derived exosomes protect beta cells against hypoxia-induced apoptosis via miR-21 by alleviating ER stress and inhibiting p38 MAPK phosphorylation · Stem Cell Research & Therapy, 11(1):97 · READ HERE · A key paper from another group identifying miR-21 as a protective cargo in MSC-derived exosomes for beta cells under hypoxic stress — one of the primary prior studies establishing the mechanistic rationale for why miR-21a-5p, the most prominently enriched miRNA in Hong’s EV-treated islets, might be functionally relevant to islet survival. Understanding the ER stress and p38 MAPK pathway suppression mediated by miR-21 in this context provides the mechanistic framework for interpreting the miRNA data in today’s paper.

4. Arzouni AA, Vargas-Seymour A, Dhadda PK, Rackham CL, Huang GC, Choudhary P, King AJF & Jones PM (2019) Characterization of the effects of mesenchymal stromal cells on mouse and human islet function · Stem Cells Translational Medicine, 8(9):935–944 · READ HERE · The Jones laboratory’s landmark characterization paper establishing in detail how MSC co-culture affects mouse and human islet glucose-stimulated insulin secretion and cytokine-induced apoptosis across different MSC tissue sources and co-culture configurations. This paper is the experimental foundation on which Hong’s EV work rests: it establishes what “recapitulating MSC effects” means functionally, what conditions and endpoints were used, and which aspects of beta cell biology MSCs most robustly improve. The EV paper’s demonstration of equivalent functional effects becomes meaningful only against this established benchmark.

05 · Four Videos to Watch First

▶ Ask the Expert: Braulio Marfil-Garza, PhD, and James Shapiro, MD — Islet Transplantation in T1D · TheSugarScience Ask the Expert with Dr. Marfil-Garza and Dr. Shapiro, one of the world’s most prominent islet transplant surgeons, discussing islet transplantation outcomes over two decades. This talk is the essential clinical context for today’s basic science talk: it establishes what islet transplantation currently achieves, how far outcomes have come since the Edmonton Protocol, and where the remaining barriers lie — including the early post-transplant beta cell attrition that Hong’s MSC-EV pretreatment strategy is designed to address. Watch this first to understand the clinical problem Hong is working to solve.

▶ Ask the Expert: Raniero Chimienti, PhD — Can Stem Cell-Derived Islets Evade the Immune System Safely? · TheSugarScience Ask the Expert with Dr. Chimienti on immune-evasive SC-derived islets. Hong’s research extends beyond cadaveric islet transplantation to also improving functional survival of SC-derived beta cells — the next generation of transplantable beta cell products. Chimienti’s work addresses the immune evasion challenge for SC-islets; Hong’s work addresses the early engraftment and beta cell survival challenge. These are the two central obstacles to clinical SC-islet therapy, and understanding both in parallel frames the full problem space that beta cell transplantation research is currently navigating.

▶ Ask the Expert: Gladys Teitelman, PhD — The Human Beta Cell: Victim or Instigator of Type 1 Diabetes Development? · TheSugarScience Ask the Expert with Dr. Teitelman on the intrinsic biology of the human beta cell and how it participates in its own destruction in T1D. This talk provides a mechanistically grounded picture of beta cell vulnerability — the stress pathways, the apoptotic machinery, the ER stress and inflammatory cascades that make beta cells susceptible to the hostile transplantation environment. Understanding why beta cells die after transplantation (and in T1D) is the biological prerequisite for appreciating what MSC-EVs are doing to protect them. The GPCR agonists and miRNAs in the EV cargo are engaging precisely the survival and death pathways Teitelman describes.

▶ Ask the Expert: Dario Gerace, PhD — Engineering Immune-Evasive Stem Cell-Derived Islet Cells (Harvard) · TheSugarScience Ask the Expert with Dr. Gerace from the Melton lab at Harvard on engineering SC-islets to evade immune rejection. Gerace’s work and Hong’s work address complementary problems in cell therapy: immune protection versus functional survival enhancement. Together they represent the two-part engineering challenge — keep the beta cells safe from immune attack and keep them alive and functional after implantation. The paracrine support that MSC-EVs provide during the critical early post-transplant period is distinct from but complementary to immune evasion strategies, and both types of intervention are likely to be needed for a durable clinical outcome.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

Thymic Selection of Multiple Islet Antigen-Reactive HLA Class II-Restricted Human T Cell Clones Camillo Bechi Genzano MD • July 21 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“Not all antigen-reactive T cell clones are created equal in the thymus.” — Camillo Bechi Genzano MD

🔬 Foundational Insights as They Apply to T1D

T1D happens because autoreactive T cells that should have been screened out during development instead reach the pancreas and destroy beta cells. The checkpoint that is supposed to prevent this is thymic negative selection: as hematopoietic stem cells differentiate step-by-step into T cells — moving from CD4⁻CD8⁻ double-negative thymocytes, to CD4⁺CD8⁺ double-positive thymocytes, to CD4 or CD8 single-positive cells — any developing T cell whose receptor binds a self-peptide/HLA complex too strongly is supposed to be deleted. Some thymocytes instead get diverted into the regulatory T cell (Treg) lineage rather than deleted outright — a fate previously shown in NOD mice to depend on which insulin epitope the cell recognizes (insulin B:21-32-reactive clones convert to Treg far more readily than B:9-23-reactive clones, and are correspondingly far less common among the T cells that end up infiltrating NOD islets).

Prior work from the Creusot and Sykes labs (Madley et al., 2020) established the platform for testing this directly in human cells: immunodeficient NSG mice are thymectomized, implanted with a piece of human thymus under the kidney capsule, and reconstituted with hematopoietic stem cells (HSCs) from the same human fetal donor — creating a functioning, HLA-defined human thymic environment. That 2020 paper showed that a single TCR clone (Clone 5, HLA-DQ8-restricted, reactive to insulin B:9-23, isolated from a T1D patient) is efficiently deleted when its restricting HLA molecule is present on the hematopoietic compartment, but escapes when it is not — the first direct demonstration of physiological negative selection of a human islet antigen-reactive T cell.

Bechi Genzano’s talk scaled this from one clone to six, transducing HSCs with six different antigen-reactive TCRs — each carrying a unique fluorescent or surface reporter (GFP, NGFR, mStrawberry) — before injecting them into thymus-grafted, HLA-DQ8/DR4-defined humanized mice. The six clones, all previously identified in the blood or islets of T1D patients: Clone 5 and Clone 2011 (both HLA-DQ8-restricted, insulin B:9-23-reactive), A19 and A55 (HLA-DQ8-restricted, reactive to insulin C-peptide fragments C:10-18 and C:9-17), A310 (HLA-DQ8-restricted, reactive to a hybrid insulin peptide formed from a fragment of insulin C-peptide fused to a fragment of islet amyloid polypeptide, IAPP), and R164 (HLA-DR4-restricted, reactive to GAD65 555-567).

The results split cleanly into two groups. Clone 5 and Clone 2011 — the insulin B:9-23-reactive clones — were found at extremely low frequency in the spleen relative to the thymus, consistent with efficient negative selection. R164, the GAD65-reactive clone, was also found at very low frequency in the periphery, but for a different structural reason: population-marker analysis showed R164 thymocytes were disproportionately stuck at the CD4⁺CD8⁺ double-positive stage, failing to progress to single-positive maturity at all — a distinct developmental block rather than deletion after progressing further. By contrast, A19, A55, and A310 — the insulin C-peptide and hybrid-insulin-peptide-reactive clones — escaped to the spleen at much higher frequency, closely mirroring the developmental trajectory of non-transgenic thymocytes. Single-cell RNA and TCR sequencing of sorted thymocytes confirmed this split and added a third, less-expected outcome: alongside the canonical CD8-single-positive, CD4-single-positive, and Treg (FOXP3⁺) branches emerging from the double-positive stage, the data revealed a distinct Th17-like cluster (marked by high CCR6 expression) — and Clone 5 landed predominantly in this Th17 cluster, suggesting its “deletion” is better described as an agonist-selection outcome that diverts the clone toward a Th17-like fate rather than eliminating it outright.

The mechanistic explanation the team proposes centers on peptide-HLA binding affinity. Insulin C-peptide (the 9-18 region in particular) binds HLA-DQ8 only extremely weakly — weakly enough to support positive selection (getting a thymocyte through initial development) but not strongly enough to trigger negative selection. GAD65 555-567 and insulin B:9-23, by contrast, bind their respective HLA molecules (DR4 and DQ8) far more strongly, driving either deletion or the developmental arrest seen with R164. AlphaFold modeling of the TCR–peptide–HLA interaction added a further layer: Clone 5’s predicted binding energy for its peptide-HLA complex is substantially lower than R164’s, explaining why the weaker-binding Clone 5 gets diverted toward Th17 (agonist selection) rather than staying arrested at the double-positive stage the way the more strongly-binding R164 does.

In the periphery, most escaped clones displayed a naive phenotype — notable because the humanized mouse platform allows generation of genuinely naive antigen-reactive T cell clones, sidestepping the forced activation that standard T cell engineering approaches require. Clone 5 was the exception, showing a more effector-like phenotype and a higher percentage of peripheral Tregs, though whether this reflects conversion from the thymic Th17-like state or separate peripheral Treg expansion is not yet resolved. Functional validation with an independent cohort of A55-clone mice confirmed that these escaped, naive antigen-reactive T cells could be specifically activated (measured by CD69 upregulation) when cultured with autologous HSC-derived dendritic cells presenting the cognate peptide. Bechi Genzano closed by connecting this work to the lab’s recent (published last month) modeling of immune responses to stem-cell-derived islet grafts: while that prior work could not yet reconstitute a full autoimmune response against an autologous stem-cell-derived islet graft, combining it with today’s platform — same-donor iPSC-derived beta cells, human HSCs, a human thymus, and physiologically thymus-derived (rather than artificially activated) antigen-reactive T cell clones — puts the field within reach of a genuinely complete humanized model of human T1D.

🎯 Core Premise

Using a humanized mouse platform combining a grafted human thymus with HLA-matched, TCR-transduced hematopoietic stem cells, Bechi Genzano shows that human islet antigen-reactive CD4⁺ T cell clones are not selected uniformly in the thymus: outcome depends on the strength of the TCR–peptide–HLA interaction. Clones with strong peptide-HLA binding (insulin B:9-23/HLA-DQ8, GAD65 555-567/HLA-DR4) are efficiently deleted or developmentally arrested; clones with very weak peptide-HLA binding (insulin C-peptide fragments, a hybrid insulin-IAPP peptide, all HLA-DQ8-restricted) escape thymic negative selection at high frequency and enter the periphery as naive, functionally responsive T cells; and at least one clone with intermediate/low TCR-peptide-HLA binding energy is diverted toward a non-canonical Th17-like fate rather than being cleanly deleted. This establishes, for the first time in a human cellular system, a mechanistic and predictive link between peptide-HLA binding affinity and which islet-reactive T cell clones the thymus fails to remove — the clones most likely, by this logic, to go on to attack beta cells.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

For scientists: The prior single-clone demonstration (Clone 5, Madley et al. 2020) established proof of principle that human thymic negative selection of an islet antigen-reactive T cell could be modeled directly; Bechi Genzano’s six-clone extension is the first systematic test of whether that behavior generalizes across the broader landscape of islet autoreactivity — and the answer is clearly no, not uniformly. Some clones are deleted, some are arrested, one appears diverted to a Th17-like fate, and others escape efficiently, with the difference tracking peptide-HLA binding affinity rather than the antigen’s identity alone. This is a mechanistically specific answer to a question — why do HLA-DQ8 and other T1D-risk alleles allow disease-driving autoreactivity through — that the field has previously only been able to address indirectly, through mouse models or population genetics.

For clinicians: The direct translational implication is for antigen-specific immunotherapy, which has had only modest clinical success to date — plausibly in part because therapies have targeted antigens or epitopes without knowing which autoreactive clones actually escape a given patient’s thymic tolerance. Bechi Genzano’s platform offers a principled way to identify escaped, disease-relevant clones (like the insulin C-peptide- and hybrid-peptide-reactive clones shown here) as priority targets for tolerogenic intervention, while providing a testbed — now funded by a Breakthrough T1D postdoctoral fellowship — for directly testing whether antigen-specific therapies can prevent escaped clones from expanding in the periphery, or instead enhance thymic deletion itself.

3️⃣ Big Takeaways

1. Whether an islet-reactive T cell clone is deleted or escapes the thymus tracks the strength of its peptide-HLA binding, not simply which antigen it targets. Insulin B:9-23 (via HLA-DQ8) and GAD65 555-567 (via HLA-DR4) bind their restricting HLA molecules strongly and are associated with deletion (Clone 5, Clone 2011) or developmental arrest at the double-positive stage (R164). Insulin C-peptide fragments and a hybrid insulin-IAPP peptide bind HLA-DQ8 only weakly — enough for positive selection but not negative selection — and the corresponding clones (A19, A55, A310) escape to the periphery at high frequency, closely mirroring normal (non-autoreactive) thymocyte development.

2. Thymic selection outcomes for autoreactive clones are not limited to simple deletion versus escape — a Th17-like diversion is a third, less-appreciated outcome. Single-cell RNA/TCR sequencing revealed a distinct CCR6-high Th17-like thymocyte cluster alongside the canonical CD8-SP, CD4-SP, and Treg branches, and Clone 5 — despite showing very low peripheral frequency consistent with “deletion” by flow cytometry — landed predominantly in this cluster, indicating its true fate is agonist-driven diversion rather than straightforward elimination. AlphaFold-modeled TCR–peptide–HLA binding energy distinguished this outcome from R164’s double-positive-stage arrest, tying peripheral fate directly back to binding strength.

3. Escaped clones enter the periphery as genuinely naive, antigen-responsive T cells — a capability this humanized mouse platform provides that standard T cell engineering cannot. Because thymocytes develop physiologically from HSCs rather than being artificially activated during TCR transduction, most escaped clones (A19, A55, A310) retain a naive phenotype in the spleen; functional testing confirmed that these naive, escaped A55-clone T cells could be specifically activated by autologous dendritic cells presenting their cognate peptide. Combined with the lab’s parallel work modeling autoimmune responses to stem-cell-derived islet grafts, this brings a fully integrated, single-donor humanized model of human T1D — beta cells, immune system, thymus, and physiologically-selected autoreactive T cells all from one source — within reach.

   ---

❓ Key Questions from the Discussion

Do selection outcomes really differ meaningfully across clones, and what predicts the outcome? A questioner asked whether some islet antigen-reactive T cells are more efficiently deleted than others, and what features of the clone, antigen, or HLA restriction element predict which outcome occurs. Bechi Genzano confirmed that the insulin C-peptide-reactive clones (particularly the region around insulin C:9-18) escape at much higher frequency, and that peptide-HLA binding affinity appears to be a genuinely critical, and somewhat unexpected, determinant of thymic selection outcome. He also raised an important caveat: the peptide these clones were tested against in the thymus is the natural, unmodified peptide, but what may actually be recognized in the inflamed islet could be a modified version of that same peptide — meaning the thymic escape advantage measured here may not fully capture what happens once these T cells reach the pancreas.

What will the Breakthrough T1D-funded antigen-specific immunotherapy experiments actually look like? Asked whether the goal is to prevent expansion of escaped clones in the periphery or to augment thymic deletion itself, Bechi Genzano said the lab is pursuing both directions, including a project using antigen-transduced hematopoietic stem/progenitor cells designed to diffuse broadly through the body — with the goal of getting the relevant antigen into the thymus itself, where it could be presented more effectively than it currently is (addressing the low-binding-affinity problem directly). Technically, this could be tested by re-transducing HSCs with candidate antigens after the humanized immune system is already established, and then checking whether the previously-escaping clones are now deleted.

Does the specific HLA class II allele — not just DQ8 versus DR4, but subtypes like DRB1*04:01 versus 04:04 — affect selection outcomes? Todd Brusko (University of Florida) asked this directly. Bechi Genzano noted that five of the six clones are HLA-DQ8-restricted, while R164 (originally discovered in an HLA-DRB104:01 context) was also tested against *04:04 and *04:05 — the *04:05 allele was shown to present the antigen to the R164 clone similarly (data to appear as supplementary material in the group’s forthcoming paper), and across all DR4 subtypes tested, R164 was consistently deleted.

If the mice were allowed to age longer, would more of the GAD65-reactive R164 clone appear in the periphery, given known age-related changes in thymic function? Dr. Sally Kent (UMass Chan), noting that R164 was originally derived from the blood of a T1D patient with fairly advanced disease, asked whether an aging thymus might let more of that clone through over time. Bechi Genzano explained this specific question can’t be tested in the current model: T cells first emerge from the grafted thymus around 10-12 weeks post-reconstitution, and his experiments were taken down at 18-20 weeks, because mice that are allowed to age past roughly 24-25 weeks begin developing graft-versus-host-disease-like autoimmune pathology driven by the human immune cells. Dr. Kent noted this connects to a broader, unresolved mystery in the field — that some people without diabetes also carry GAD-reactive T cells and autoantibodies, sometimes against the very same epitope, yet never develop disease.

Was it a coincidence that the TCR constructs using the NGFR reporter showed the best transduction efficiency? Dr. Teresa DiLorenzo asked this technical question. Bechi Genzano explained that NGFR, unlike the fluorescent reporters GFP and mStrawberry, is more prone to detectable “pseudotransduction” at the early time point (2-3 days post-transduction) when transduction efficiency is checked — likely accounting for the difference, despite using the same multiplicity of infection and vector copy number across all constructs; achieving exactly matched efficiencies across different reporter constructs in practice is essentially impossible.

🔗 3 TSS Talks That Connect With This One

1. Ask the Expert: Roberto Mallone, MD, PhD — INSERM Dr. Mallone — a co-author with Bechi Genzano on a review of epitope-based precision immunotherapy for T1D — is one of the field’s leading authorities on the antigen-specific T cell landscape in human disease. His discussion of which islet autoantigens and T cell epitopes have been identified, and what HLA restriction means in practice, is the essential vocabulary for understanding exactly which epitopes today’s six clones represent and why that specificity matters therapeutically.

2. Ask the Expert: Billur Akkaya, MD DPhil — Ohio State University Dr. Akkaya’s work on regulatory T cell specificity and function speaks directly to one of today’s talk’s most important findings: that thymic selection outcomes for autoreactive clones aren’t limited to simple deletion, but can include diversion toward Treg or, as shown here, Th17-like fates. Understanding what drives Treg induction versus other agonist-selection outcomes is necessary context for interpreting Clone 5’s unexpected Th17 fate.

3. TheSugarScience State of the Science 2022 #1: Is T1D ONE Disease or Endotypal? This panel’s discussion of T1D heterogeneity — different patients with different HLA backgrounds, dominant autoantigens, and immunological drivers — is exactly the premise that makes Bechi Genzano’s multi-clone approach necessary in the first place. A therapy designed around Clone 5’s insulin B:9-23/HLA-DQ8 reactivity may do nothing for a patient whose dominant risk allele is HLA-DR4 with escaped clones reactive to a completely different antigen — which is precisely the landscape today’s talk begins to map.

   ---

   🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

Cyborg Organoids Implanted with Flexible Electronics Reveal Principles of Human Islet Cell Electrical Maturation Juan R. Alvarez-Dominguez PhD · July 16, 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“One of the curious observations in our study is that... between month one and month two of maturation, broadly speaking, [we see] the induction of HLA and the induction of beta-2M. So we do know that there’s a cost to becoming a more metabolically active cell, and that cost seems to be coupled with induction of signals that will make you more discoverable by the immune system. Perhaps one of the applications could be to uncouple one from the other.”

— Juan R. Alvarez-Dominguez

🔬 Foundational Insights as They Apply to T1D

Alvarez-Dominguez opened by framing the core gap in the field’s toolbox: most methods for studying islet cell function operate on the timescale of minutes to hours, but processes like functional maturation — how beta cells acquire the precise, glucose-responsive secretory behavior of an adult islet — unfold over weeks and months. Understanding how islet “circuits” develop (how alpha, beta, and delta cell hormone secretion becomes coordinated) matters directly for T1D, because that coordination is what breaks down during disease onset and progression. The technical bottleneck has been that existing tools — calcium imaging and patch-clamp electrophysiology chief among them — either capture only a single plane of view, or require disrupting the cell membrane in a way that makes long-term, repeated measurement impossible. Conventional extracellular microelectrode arrays (MEAs) solve the membrane-disruption problem but still only contact cells at a single surface plane of the tissue.

The solution presented is the “cyborg organoid” platform, developed in collaboration with Jia Liu’s bioelectronics lab at Harvard. Stretchable mesh nanoelectronics — soft enough to bend and curl on their own — are introduced at the 2D progenitor-cell stage, before the tissue folds into a 3D structure. As the cells undergo their natural organogenesis (extruding alpha cells to the outside and beta cells to the inside, recapitulating in vivo islet architecture), they envelop the mesh electronics and integrate them throughout the tissue volume. The resulting “cyborg pancreatic organoid” allows continuous, single-cell-resolution extracellular voltage recording — with millisecond resolution — over weeks to months, without disrupting the tissue. Alvarez-Dominguez’s lab produces the underlying stem cell-derived pancreatic organoids at scale using a published, bioreactor-based suspension culture protocol they developed to unify small- and large-volume differentiation approaches. He noted that this general class of stem cell-derived islet product has already reached clinical trials, where treated patients with severe hypoglycemia have shown blood glucose normalization roughly a year after treatment — but that the in vitro-differentiated cells lag behind natural transplanted islets, whose insulin/C-peptide activity normalizes within about a week, motivating the search for what specifically remains immature.

Validation of the platform showed that embedding the electrode mesh does not measurably change cell-type composition (endocrine vs. exocrine lineage proportions were consistent with or without the device) and that electrode impedance remains stable over at least 10 weeks in physiological medium. Individual cells recorded by a given electrode are distinguished computationally by clustering their waveform features (a spike-sorting approach adapted from neuroscience), and cell identity (alpha vs. beta) is cross-validated by clearing and sequencing the same fixed tissue afterward, confirming that electrically classified “alpha” and “beta” cells match transcriptionally defined glucagon- and insulin-expressing cells, respectively.

Using this platform, the team traced extracellular spike dynamics over months of maturation in parallel with hormone secretion (GSIS) measurements. Two competing models were formally distinguishable with this data: a “cell-level” model where individual immature cells gradually become more mature over time, versus a “population-level” model where only two discrete electrical states exist and maturation reflects a growing proportion of cells occupying the more mature state. The data supported the population-level model: beta cells were classified into an immature-fitting state (active under both low and high glucose) and a mature-fitting state (silent under low glucose, strongly active under high glucose), with the mature-state population growing in number over the culture time course; alpha cells showed the reciprocal pattern (immature cells active under both conditions; mature cells active only under low glucose). This finding — that a higher glucose threshold for action-potential firing itself distinguishes mature from immature cells — was notable because prior thinking attributed glucose-threshold sensitivity to downstream calcium-handling machinery (e.g., synaptotagmin isoform switching) rather than to the upstream membrane-potential firing threshold itself.

A second line of investigation addressed circadian entrainment. Building on Alvarez-Dominguez’s earlier finding that daily feeding/fasting cycles drive islet maturation, the team applied a defined entrainment protocol (four days of high-glucose pulses every 12 hours, followed by constant medium to observe self-sustained, cell-autonomous oscillations) while continuously recording electrical activity. This revealed entrained oscillations not just in population-average insulin/glucagon secretion, but in single-cell waveform parameters themselves — meaning circadian coordination operates at the level of individual cells’ firing properties, not merely as an averaged population effect. The talk also covered bidirectional use of the electrode mesh: applying a defined electrical stimulation protocol (2-millisecond pulses every 2 seconds at 500 mV) for three hours produced improved post-stimulation glucose responsiveness specifically in the physiologically relevant direction for each cell type (improved low-glucose responsiveness in alpha cells, improved high-glucose responsiveness in beta cells), raising the possibility of closed-loop electrical control of islet function.

Looking ahead, Alvarez-Dominguez described ongoing work to combine single-cell electrical recordings with single-cell gene expression (via clearing and sequencing the same recorded tissue) to map which transcriptional programs underlie which electrical behaviors, in vivo spatial transcriptomic mapping of transplanted grafts over time (referenced as an existing bioRxiv preprint from his group, though no specific title or DOI for that separate preprint was stated in the talk), and AI-driven (rather than fixed) stimulation protocols — illustrated with an example from cardiac organoids — as a strategy to normalize or accelerate organoid maturation without needing to hand-specify a stimulation policy in advance.

🎯 Core Premise

By embedding stretchable mesh nanoelectronics directly into stem cell-derived pancreatic organoids as they undergo organogenesis, this platform enables continuous, single-cell-resolution electrical recording of the same living islet-like tissue over months — something no prior electrophysiological method could achieve. Applying it reveals that functional maturation of SC-alpha and SC-beta cells reflects a population-level shift between two discrete electrical firing states (defined by glucose threshold) rather than a gradual, uniform change in every cell, and that circadian entrainment synchronizes both hormone secretion and individual cell firing waveforms across the islet. The same electrode mesh can also deliver stimulation that improves cell-type-appropriate glucose responsiveness, pointing toward electrical read/write control as a tool for both studying and potentially accelerating or correcting islet maturation.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

A chronic, single-cell “electrode implant” for the living, developing human islet.

For scientists: Islet electrophysiology has always faced a trade-off between resolution (single cell, via patch clamp) and durability/scale (whole-islet, via surface MEAs) — never both, and never over the months-long timescale that functional maturation actually requires. The cyborg islet platform breaks that trade-off directly, and the biological payoff is a genuine reframing of what maturation is: not each cell becoming more capable individually, but a population redistributing between two discrete, glucose-threshold-defined firing states. The demonstration that circadian entrainment produces single-cell-level waveform synchrony — not just averaged secretory rhythms — extends this further, suggesting maturation and circadian coordination are properties of the network as much as of any individual cell.

For islet organoid and neural/gut organoid scientists specifically: the platform offers a new, direct, real-time functional readout — proportion of high- vs. low-basal-firing cells — that could sit alongside existing SC-islet quality-control metrics. The same core question (how does an electrically excitable cell acquire its mature firing identity, and can implanted electronics both measure and drive that process) and the same joint transcriptomic/electrophysiological approach apply directly to any organoid system built from excitable cells, as underscored by the platform’s prior application to cardiac organoids.

For clinicians: SC-derived islets are already in clinical trials, and Alvarez-Dominguez’s own data show that transplanted in vitro-differentiated organoids currently take substantially longer to normalize blood glucose than transplanted primary islets — a gap in functional maturity that this platform is built to characterize and, potentially, close. A specific and clinically relevant finding from the discussion: as SC-islet cells mature and become more metabolically/electrically active, they also show increased induction of HLA and beta-2 microglobulin — markers that make a cell more visible to the immune system. Alvarez-Dominguez raised the possibility of trying to uncouple functional maturation from this immune-visibility signature as a specific, addressable engineering goal for transplant durability, though he was clear this is a hypothesis to be tested rather than a finding already established.

The broader picture: The platform was explicitly framed as a bidirectional tool — not just recording, but stimulating. Ongoing directions discussed included implanting the device in vivo (analogous to chronic neural electrodes in living animals) to monitor or actively control graft survival and function after transplantation, and moving toward AI-driven (rather than fixed) stimulation protocols that could be used to normalize organoid function in vitro or accelerate/correct maturation in vivo after transplant-related metabolic or immune stress.

3️⃣ Big Takeaways

Functional maturation of SC-islet cells looks, electrically, like a population shifting between two discrete states — not a smooth, uniform change in every cell. By recording the same cells continuously over a seven-week culture time course, the team could distinguish a “cell-level” maturation model (every cell gradually improving) from a “population-level” model (a fixed repertoire of immature and mature states, with cells moving into the mature state over time). The data supported the latter for both SC-beta cells (defined by glucose threshold for action-potential firing, silent at low glucose in the mature state) and SC-alpha cells (the reciprocal pattern), with the number of mature-state cells increasing over time in both populations. Notably, the glucose-threshold difference was detectable upstream of calcium signaling, at the level of membrane-potential firing itself — challenging the prior assumption that threshold sensitivity was set only by downstream calcium-handling machinery.

Circadian entrainment synchronizes islet function down to the level of individual cell firing waveforms, not just averaged hormone output. Using a defined four-day high-glucose entrainment protocol followed by constant medium, the team detected self-sustained oscillations not only in population-averaged insulin and glucagon secretion, but in the shape of individual cells’ voltage waveforms across an 11-parameter feature space — direct single-cell evidence that circadian coordination operates as a network-level phenomenon linking individual electrical identities together, a mechanism Alvarez-Dominguez described as still not fully understood (in particular, whether synchrony precedes or follows induction of gap-junction/exocytic connectivity remains an open, “chicken-and-egg” question by his own account).

The same implanted electrodes that record activity can also deliver targeted stimulation that improves cell-type-appropriate glucose responsiveness — opening a path toward closed-loop, bidirectional control of islet function. A three-hour stimulation protocol produced improved post-stimulation responsiveness specifically to low glucose in alpha cells and specifically to high glucose in beta cells, with no such difference before stimulation. Combined with in-progress work on in vivo implantation and AI-driven (rather than fixed) stimulation policies, this raises the practical possibility of using electrical stimulation not just to study maturation, but to actively accelerate it or correct it after transplant-related stress — alongside an open, unresolved question the group flagged about whether increased metabolic/electrical maturity necessarily comes coupled with increased immune visibility (via HLA/beta-2M induction), and whether the two could be engineered apart.

❓ Key Questions from the Discussion

Does electrical synchrony during circadian entrainment come before or after induction of the cell-cell communication and exocytic networks it correlates with — is coordinated firing driving connectivity, or is connectivity enabling the synchrony? Could synchrony be induced directly (without circadian entrainment) by stimulating exocytic or gap-junction networks directly? (Monica, moderator) Alvarez-Dominguez called this “a wonderful question” and said the causal direction is genuinely unknown — a “chicken and egg” problem. He proposed the process is likely autoregulated/bidirectional: some baseline level of connexin expression may already exist, and inducing activity could feed back to increase connexin transcription further, but connectivity may need to already exist to some degree for that feedback to work efficiently. He distinguished electrical synchronization itself (which he said is reasonably well understood and requires gap-junction connections) from the self-sustained 24-hour periodicity of that synchronization (which is less well understood and can, in principle, be entrained by multiple external cues — temperature, secreted factors, metabolic factors). He suggested pharmacological or genetic manipulation of connexins as a concrete way to test this going forward, noting they have already used a toxin to broadly silence electrical activity in past work.

Cyborg islets and transplanted islets lack vagal nerve innervation — does that matter for islet electrical coordination, or is vagal input more of a fine-tuning mechanism? (Monica, moderator) Alvarez-Dominguez said there are many plausible applications of vagal input beyond core secretory function — he mentioned having seen conference presentations on vagal involvement in beta cell proliferation and in stress-induced hypoglycemia — and that meal anticipation tied to daily behavioral cycling likely does involve nerve input. He noted that islets transplanted under the eye (as opposed to the kidney) do have some access to innervation, which he considered an interesting, underexplored angle. He said the field currently has no clear analog to deep-brain-stimulation parameters for the pancreas — i.e., no established sense of what type, frequency, or voltage of neural-equivalent input a stressed or decaying islet would need to resume function — and suggested that because the possibility space is so large and unexplored, machine-learning-driven trial-and-error stimulation could be a practical way to search it, testing empirically whether a given stimulation pattern moves cells toward more mature or worse function.

How might this platform be used to study the electrical signature of islets as they’re affected on the path to type 1 diabetes pathogenesis — for instance, under viral insult (e.g., a Coxsackievirus B–like challenge)? Does the “electrical symphony” get disrupted by that kind of stress? (Monica, moderator) Alvarez-Dominguez said this can’t be answered without directly doing the experiment. He shared a related observation from their existing data: single-cell-resolution recording lets them order cells by relative maturity with more precision than bulk measurements would allow, and between roughly month one and month two of maturation they observed induction of HLA and beta-2 microglobulin — indicating that becoming more metabolically/electrically active is coupled to becoming more visible to the immune system. He suggested that uncoupling functional maturity from this immune-visibility signature could be one practical application of the platform, but said this needs to be tested empirically rather than assumed.

Connecting to senescence research (referencing work associated with Peter Thompson and Anil Bhushan) showing that some stressed beta cells appear to go transcriptionally “silent” or dark under attack — what electrical signatures might occur in such senescent cells or senescent-like phenotypes? (Monica, moderator) Alvarez-Dominguez agreed this was worth exploring and reflected that the field’s default assumption — that the goal is simply to make non-functioning cells “work again” — might be naive, since there may be a protective reason cells go silent under stress or immune insult. He noted that senescence signatures are seen more frequently in at-risk individuals, and suggested that using electrical activity monitoring to detect — or potentially influence — this kind of vulnerability signal is a “virtually unexplored” but promising direction.

🔗 3 TSS Talks That Connect With This One

1. Ask the Expert: Raniero Chimienti, PhD — Can Stem Cell-Derived Islets Evade the Immune System Safely?
   Dr. Chimienti’s discussion of immune-evasive SC-derived islets frames the clinical program that Alvarez-Dominguez’s work is ultimately in service of: SC-islets are already in trials, and making them work well requires solving both the immune-evasion problem and the functional-maturation problem side by side. That pairing is directly relevant to a specific finding from Alvarez-Dominguez’s talk — that increasing electrical/metabolic maturity in SC-islet cells comes coupled with induction of HLA and beta-2 microglobulin, making more mature cells more visible to the immune system. Watching the two talks together gives the fuller picture of what a transplant-ready SC-islet actually needs to achieve on both fronts.

2. Ask the Expert: Kyle Gaulton, PhD — UC San Diego — T1D Risk, Genetics, and Single-Cell Epigenetics
   Dr. Gaulton’s work on the regulatory genomics of human islet cells — the chromatin accessibility and gene expression programs that define each islet cell type — supplies the molecular reference frame needed to interpret what the cyborg islet platform is finding electrically. The two discrete firing states (high- and low-basal glucose threshold) that Alvarez-Dominguez identifies in SC-α and SC-β cells are only biologically meaningful once mapped onto the underlying gene expression programs those states correspond to; Gaulton’s talk builds the vocabulary for making that connection.

3. Ask the Expert: Ruth Elgamal, PhD Candidate — UC San Diego — Integrated Pancreatic Islet Reference Map
   Elgamal’s talk on the HPAP integrated single-cell reference atlas of human islet cell types — their transcriptional states, marker genes, and proportions across donors — is the biological benchmark against which SC-islet cell populations and their maturation states get compared. Understanding what that reference atlas looks like prepares you to appreciate what it actually means when cyborg islet recordings reveal two discrete electrical states in SC-α and SC-β cells, and how those electrical states relate to the transcriptional heterogeneity already documented in adult human islets.

🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

Moving Small Molecule DYRK1A Inhibitors for Human Beta Cell Regeneration From High Throughput Screen to People With Diabetes Andrew F. Stewart MD · July, 14 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“Until 2014, the world thought that it was impossible to make human beta cells replicate. They do in infancy, but by the time you get to be a late adolescent or young adult, they’ve stopped.”

— Andrew F. Stewart

🔬 Foundational Insights as They Apply to T1D

The premise Stewart’s laboratory set out to test decades ago was blunt: everybody with type 1 or type 2 diabetes has too few functioning pancreatic beta cells, and if you can’t replace them, maybe you can regrow the ones that are left. The dogma standing in the way was that adult human beta cells are essentially post-mitotic — unlike the beta cells of an infant, they had, by every measure the field could apply, stopped dividing for good. Stewart’s group ran a high-throughput screen of more than 100,000 compounds against human beta cells and, in 2015, found harmine — a naturally occurring beta-carboline alkaloid and inhibitor of the kinase DYRK1A — as a hit capable of making adult human beta cells replicate in vitro (Wang et al., Nature Medicine 2015). Adding a GLP-1 receptor agonist on top produced even more proliferation, and a 2020 paper (Ackeifi et al., Science Translational Medicine) showed the combination is synergistic and beta-cell selective, driving replication rates of 5–6% per day — far above what either agent achieves alone.

The harder question was whether this worked in a living pancreas, not just a dish. Working with Adolfo Garcia-Ocaña, Sarah Stanley, and Peng Wang, Stewart’s group transplanted human islets into mice and infused them with harmine, or harmine plus exenatide, for up to three months. Harmine alone produced roughly a three-fold increase in human beta cell mass; adding the GLP-1RA pushed that to roughly seven-fold — a result consistent with the ~700% mass expansion later quantified with 3D iDISCO+ imaging in the 2024 Science Translational Medicine paper (Rosselot et al.). Notably, beta cell mass took the full three months to build, but blood glucose in the diabetic xenograft mice normalized within days — evidence that harmine’s early benefit is not just about making more cells, but about making the existing ones work better, faster.

That second effect turned out to be real and mechanistically distinct. Stewart’s team worried that forcing beta cells to divide would make them worse — dedifferentiated, dysfunctional. Instead, a panel of beta cell identity and function genes (PDX1, MAFA, MAFB, NKX6.1, GLUT2, prohormone convertase, GLP1R) went up, not down, both at the RNA level and the protein level. Single-cell RNA sequencing work from Ezra Karakose at Mount Sinai localized most of this induction to beta cells themselves, though two other islet populations — an “unknown endocrine cell” cluster and a population of cycling alpha cells — also responded, raising the possibility (not yet proven) that harmine is coaxing other islet cell types to become beta cells, a question tied to a 2024 Cell Reports Medicine paper from the group on cycling alpha cells as potential beta cell progenitors.

The most striking piece of new mechanism presented was the discovery of a second molecular target. Not all DYRK1A inhibitors are equal: harmine, and close analogs 2-2c and 5-IT, induce both proliferation and the identity/function gene program, while other well-validated DYRK1A inhibitors — INDY, leucettine, CC-401, and several Novartis/GNF compounds — drive proliferation without touching identity or function. Genetic silencing of DYRK1A and its homolog DYRK1B reproduced the proliferation effect but had no effect on PDX1, MAFA, or NKX6.1 induction — and harmine still induced those genes normally even when DYRK1A/1B were silenced. In other words, DYRK1A is necessary for proliferation but not for the differentiation benefit. A biotinylated harmine analog (2-2c modified for bead-based pulldown) used in proteomic screening of human islet extracts identified protein kinase A regulatory subunit 1A (PRKAR1A) as a second target: silencing PRKAR1A alone induced PDX1 and MAFA, and harmine activates PKA signaling (via PKA substrate phosphorylation) independent of DYRK1A. A thermal shift assay showed cyclic AMP directly binds and stabilizes PKA-R1A (raising its melting point from 43°C to 70°C) — but harmine itself did not produce that shift, suggesting harmine’s interaction with the PKA regulatory complex, whatever it is, is indirect rather than a simple ligand-binding event. Identifying that true “Target 2” mechanism is the group’s current central focus.

On the safety side, a 2021 FDA-approved Phase 1 trial of pharmaceutical-grade oral harmine in healthy volunteers (Ables et al., Journal of Psychopharmacology, 2024) established a maximum tolerated dose below 2.7 mg/kg with no serious adverse events; the dose-limiting effects were gastrointestinal and neurological, consistent with harmine’s known activity as a monoamine oxidase inhibitor. Because harmine is also a component of the psychoactive brew ayahuasca, Stewart addressed this directly: harmine itself is not the psychoactive ingredient in that context — dimethyltryptamine (DMT) is, and harmine’s role there is only to block DMT’s degradation. In the Phase 1 trial, pure harmine on its own produced no psychoactive effects at doses studied, and volunteers on a required tyramine-free diet showed no hypertensive events. Comprehensive tissue screening in the mouse studies — liver, heart, lung, spleen, kidney, brain, and exocrine pancreas — found no evidence of off-target proliferation, consistent with the roughly 250-fold gap between the micromolar concentrations used in vitro and the ~20 nanomolar circulating levels achieved therapeutically in vivo. Next-generation DYRK1A inhibitors engineered by Robert DeVita’s chemistry group, including 2-2c, are designed to minimize blood-brain-barrier penetration relative to harmine itself.

🎯 Core Premise

Small-molecule DYRK1A inhibitors — discovered via an unbiased, >100,000-compound high-throughput screen and led by harmine — are the first validated pharmacologic class capable of making adult human beta cells re-enter the cell cycle. Combined with a GLP-1 receptor agonist, they drive synergistic, beta-cell-selective proliferation (5–6%/day) and expand human beta cell mass in vivo by roughly 700% over three months in a mouse xenograft model, while simultaneously improving — not degrading — beta cell identity and insulin-secretory function. That functional benefit is mediated by a second target, protein kinase A (via PRKAR1A), distinct from the DYRK1A-mediated proliferative effect — meaning only a subset of DYRK1A inhibitors (harmine, 2-2c, 5-IT) are “dual-target” and clinically attractive. A completed Phase 1 human safety trial of pure harmine, extensive off-target tissue screening, and next-generation CNS-sparing analogs now under development (with PlexBio leading clinical development) mark the program’s transition from bench discovery toward first-in-human regenerative trials in both T1D and T2D — with T1D requiring eventual combination with immunomodulatory therapy to protect newly regenerated cells from ongoing autoimmunity.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

For scientists: This is arguably the most clinically advanced pharmacologic beta cell regeneration program in existence, and the discovery arc — an unbiased screen yielding a single validated hit, a decade of mechanistic dissection, and now a completed Phase 1 human trial — is a template for how academic drug discovery is supposed to work. The dual-target finding is the scientifically richest result presented: proliferation (DYRK1A) and functional maturation (PKA/PRKAR1A) are mechanistically separable, which means the field can now, in principle, screen and design molecules that hit both targets deliberately rather than relying on a lucky natural product. The unresolved question — what harmine’s biotinylated analog is really pulling down in the PKA regulatory complex, given that harmine itself does not directly stabilize PRKAR1A in a thermal shift assay — is a precisely defined biochemical puzzle that the group is actively pursuing.

For clinicians: The translational logic is unusually direct for a regenerative therapy. Most people with T1D, especially soon after diagnosis, retain some residual beta cells; many with longstanding disease still have a measurable functional reserve. An oral small molecule that could expand what remains — without requiring transplantation, immunosuppression, or a stem-cell-derived cell product — would be far cheaper and more scalable than existing beta cell replacement strategies. The safety picture so far is reassuring: the Phase 1 trial found no serious adverse events, comprehensive tissue screening found no evidence of off-target cell proliferation, and the psychoactive and hypertensive risks associated with harmine’s MAO-inhibitor pharmacology have not materialized at the doses studied. The central unmet need for T1D specifically is immunological: newly regenerated beta cells will face the same autoimmune attack that destroyed the originals, which is why Stewart’s collaborators are actively testing combinations with anti-CD3 immunotherapy in preclinical models, with the expectation that regeneration will need to be paired with — not substituted for — immune protection.

3️⃣ Big Takeaways

1. Harmine and related DYRK1A inhibitors are the first drug class proven to make adult human beta cells divide — and the effect scales with a GLP-1 receptor agonist already on the market. A screen of over 100,000 compounds found harmine as a hit against a dogma the field had accepted for decades: that adult human beta cells are terminally post-mitotic. Alone, harmine drives modest human beta cell replication (~2%/day) and roughly a three-fold expansion of beta cell mass in vivo over three months. Paired with a GLP-1RA (exenatide/exendin-4), replication rises to 5–6%/day and in vivo mass expansion reaches roughly seven-fold (~700%), with blood glucose normalizing within days — faster than the mass expansion itself, implying an early functional benefit that precedes the proliferative one.

2. New beta cells made this way are not degraded or dedifferentiated — they are, if anything, more functionally mature, and that benefit runs through a second, independent molecular target. Rather than losing identity markers as feared, treated beta cells upregulate PDX1, MAFA, MAFB, NKX6.1, GLUT2, and GLP1R, and this translates to improved glucose-stimulated insulin secretion and reversal of diabetes in xenograft mice. This differentiation/function effect is driven by protein kinase A (via the regulatory subunit PRKAR1A), independently of DYRK1A inhibition — genetic silencing of DYRK1A/1B eliminates proliferation but not the identity gene induction, and vice versa. Because of this, not all DYRK1A inhibitors behave alike: harmine, 2-2c, and 5-IT hit both targets and improve function; other well-established DYRK1A inhibitors (INDY, leucettine, CC-401, several Novartis/GNF compounds) only drive proliferation, making the “dual-target” subset the more clinically attractive candidates.

3. The program has already cleared Phase 1 human safety testing and is moving toward next-generation, CNS-sparing compounds and combination immunotherapy for T1D. A completed Phase 1 trial of pure oral harmine found a tolerable dose range below 2.7 mg/kg with no serious adverse events, and extensive tissue screening across liver, heart, lung, spleen, kidney, brain, and exocrine pancreas found no evidence of off-target proliferation — reassuring given that circulating therapeutic concentrations (~20 nM) are roughly 250-fold lower than the concentrations typically used in vitro. Next-generation DYRK1A inhibitors engineered to minimize blood-brain-barrier penetration are in development, and PlexBio is now leading clinical development for both T1D and T2D indications. For T1D specifically, preclinical work combining harmine with anti-CD3 immunotherapy in NOD mice (in preparation for publication) suggests regeneration will likely need to be paired with immune protection for durable benefit, rather than deployed alone.

   ---

❓ Key Questions from the Discussion

Whatever happened to the cycling alpha cell story? Host Monica asked about the group’s 2024 Cell Reports Medicine paper identifying cycling alpha cells — the population most responsive to DYRK1A inhibition besides beta cells themselves — as potential beta cell progenitors. Stewart declined to preempt the answer, noting that Ezra Karakose, who generated the single-cell RNA-seq data behind that finding, would present that story directly in an upcoming talk in the same series (July 28).

Is the newly expanded beta cell mass resistant to the inflammatory and autoimmune stress seen in T1D — and could regeneration be combined with immune protection? Anna asked directly whether expanded beta cell mass has been tested against inflammatory stress relevant to T1D, and whether regeneration could be paired with immunological protection. Stewart described an in-preparation paper from Adolfo Garcia-Ocaña combining harmine with anti-CD3 antibody in NOD mice, noting harmine has a modest immunomodulatory benefit on its own (building on prior work from Greg Korbutt in the T2D setting) but is “very nicely supplemented” by anti-CD3. His overall prediction: DYRK1A inhibitors alone may help in some autoimmune contexts, but will most likely need to be combined with an immunomodulatory agent for durable T1D benefit.

Does DYRK1A play a role in cells other than beta cells, and could off-target proliferation be a safety concern? Klaus Pfeffer asked about DYRK1A’s ubiquitous expression outside the pancreas. Stewart confirmed the group screened liver, heart, lung, spleen, kidney, brain, and the exocrine pancreas in a blinded histopathology review and found no evidence of proliferation or neoplastic change. He attributed this partly to dose: effective in vivo circulating concentrations (~20 nM) are far below the micromolar concentrations typically used in vitro, where little proliferation is seen at 20 nM either. He also noted the source of new beta cells beyond direct proliferation remains an open question — possibly transdifferentiation of alpha cells or other non-beta endocrine populations — that lineage-tracing work now underway (again, Ezra Karakose’s) is designed to resolve.

Is the thermal-shift “shoulder” seen with harmine biologically meaningful, or noise? Matt Merrins (Yale) asked about a minor shift observed in the harmine thermal-stability assay. Stewart agreed with outside experts that it is likely not meaningful at the protein concentrations used, but called it an interesting detail worth further discussion.

Could increases in “beta cell” gene expression actually reflect more beta cells being present, rather than more expression per cell? Klaus Pfeffer also raised this technical confound, common to bulk islet gene expression work. Stewart responded that single-cell RNA-seq performed at short time points (4 days to 1 week) — before meaningful proliferation has occurred — shows no change in the number of beta cells, supporting the interpretation that the observed gene induction reflects a genuine per-cell upregulation rather than a shift in cell-population composition.

Given harmine’s psychoactive and monoamine-oxidase-inhibitor pharmacology, how are those risks being managed, and could non-proliferative DYRK1A inhibitors instead be used to improve beta cell health early in disease rather than to drive division? David Harlan asked both about psychotropic risk and about a strategy — raised in a prior presentation the two had both attended — of using DYRK1A inhibitors that don’t drive proliferation to make existing beta cells healthier rather than to expand their number. Stewart explained that harmine’s psychoactivity in ayahuasca depends on co-administration of DMT (harmine itself blocks DMT’s breakdown but is not the psychoactive agent), that the completed Phase 1 trial found no psychoactive effects of pure harmine at studied doses, and that a tyramine-free diet precaution produced no hypertensive events. He noted next-generation compounds like 2-2c are engineered to reduce blood-brain-barrier penetration relative to harmine. On the immunology question, he deferred definitive comment on non-proliferative-but-function-enhancing strategies until human data are available, but noted Garcia-Ocaña’s data support benefit from combining regeneration approaches with immunomodulation, particularly anti-CD3.

🔗 3 TSS Talks That Connect With This One

1. Ask the Expert: Braulio Marfil-Garza, PhD, and James Shapiro, MD — Islet Transplantation in T1D Islet transplantation is the current clinical benchmark for beta cell replacement in T1D — donor-scarce, immunosuppression-dependent, and limited in long-term graft durability. Understanding what transplantation achieves and where it falls short is the clinical context for why a scalable, oral, pharmacologic approach to regenerating a patient’s own residual beta cells would be such a different — and potentially more accessible — proposition than the approach Stewart’s program is pursuing.

2. Ask the Expert: Kevan Herold, MD, and Matthias Von Herrath, MD — Teplizumab in T1D Teplizumab works by preserving the residual beta cells a person still has at diagnosis — the same cells DYRK1A inhibitors are designed to expand. This conversation sharpens the case, raised directly in today’s Q&A, that immune protection and beta cell regeneration are likely complementary rather than substitutable strategies: teplizumab delays disease but does not reverse it, which is exactly the gap a regenerative therapy paired with immunomodulation would need to fill.

3. Ask the Expert: Raniero Chimienti, PhD — Can Stem Cell-Derived Islets Evade the Immune System Safely? Chimienti’s work on engineering immune-evasive stem cell-derived islets represents a parallel strategy for the same underlying goal — restoring functional beta cell mass in T1D — but through cell replacement rather than pharmacologic regeneration of a patient’s own cells. Watching the two talks together frames the current landscape of curative-intent T1D strategies: regenerate what’s left (Stewart) versus replace it with engineered cells (Chimienti), each with different risk, cost, and scalability profiles.

   ---

🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

Structural Clues to HLA-B Mediated Type 1 Diabetes Risk Ruby Sharma PhD • July 7 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“Even a single difference in the amino acid change can entirely change the disease spectrum.” — Ruby Sharma, PhD

🔬 Foundational Insights as They Apply to T1D

Most of what the T1D field knows about HLA risk concerns class II molecules — HLA-DR and HLA-DQ, which present peptides to CD4+ T cells and account for the largest share of genetic susceptibility. But CD8+ cytotoxic T cells are the most abundant immune cell population infiltrating inflamed human islets, and it is HLA class I molecules — not class II — that present peptides to them. Among class I alleles, HLA-B39:06 is the single most strongly predisposing allele identified in T1D, associated with earlier disease onset and disproportionately common in Latino populations, where T1D incidence is rising. What makes this genetically strange is that B39:06 differs from two close relatives — the also-predisposing (but less so) B39:01, and the protective B38:01 — by only a handful of amino acids, despite the three sharing greater than 97% identity. Sharma’s talk was about what those few amino acids actually do, physically, inside the peptide-binding groove.

The basic architecture of any MHC class I molecule is an alpha-1 and alpha-2 helix pair forming a peptide-binding groove, subdivided into pockets labeled A through F. Sharma’s work concentrates on the F pocket — the pocket at the groove’s C-terminal end that anchors the last residue of a bound peptide — because this is exactly where B39:06 and B39:01 differ: B39:06 carries a bulky tryptophan at position 95, while B39:01 (and B38:01) carry a much smaller leucine at that position. To compare the two structurally, Sharma’s team first needed a peptide that both alleles would bind. They identified NRVMLPKA — derived from the human protein NLRP2 and eluted directly from C1R cells in the DiLorenzo lab — as the strongest known binder to B39:06 (IC50 = 14 nM) that also binds B*39:01 (IC50 = 35 nM), making it an ideal comparative tool.

Producing crystallizable protein required real troubleshooting: a single-chain, disulfide-trapped HLA-B*39:06 construct (via a Tyr84Cys mutation linking peptide to the MHC heavy chain) was expressed in FreeStyle 293 cells and purified by nickel-affinity and gel filtration chromatography, but early preparations resisted crystallization until the team recognized and removed a glycosylation site at position 86 using PNGase F. The deglycosylated protein yielded two distinct crystal forms (space groups C222₁ and P2₁2₁2₁), diffracted at a synchrotron beamline, and solved to 1.7 Å resolution using the CCP4i software suite and visualized in PyMOL.

The resulting structure clarified something the field had assumed differently for other MHC class I molecules: rather than the canonical position-9 anchor seen in most previously solved class I structures, NRVMLPKA is an 8-mer, and its C-terminal alanine at position 8 — not a position-9 residue — serves as the dominant anchor, forming strong hydrogen bonds with Asp80, Lys146, and Tyr147. Arginine at position 2 is the primary N-terminal anchor (hydrogen-bonding into the B pocket via Gln45, Asp63, and Ser24), with valine at position 3 as a secondary anchor. Methionine at position 4 and lysine at position 7 remain solvent-exposed — meaning these are the residues actually available for T cell receptor contact. Notably, this 8-mer preference lines up with the only known B39:06-restricted diabetogenic T cell epitope, MRLLA, itself an 8-mer — reinforcing that B39:06 has a genuine, biologically relevant preference for 8-mer peptides rather than the 9-mers typically favored by other class I alleles.

The comparative payoff came from the F pocket itself: B39:06’s bulky tryptophan-95 sterically restricts the pocket, meaning only small C-terminal residues (like alanine) fit — explaining why B39:06 favors 8-mer, alanine-anchored peptides. B39:01’s smaller leucine-95 opens up more room, favoring larger C-terminal residues (leucine, in a structural model built by superimposing NRVMLPKA onto a previously solved B39:01 structure). Because a full B38:01 crystal proved difficult to obtain, Sharma instead built a structural model: B38:01 also carries leucine at position 95, but its F pocket is comparatively neutral and larger, disfavoring basic residues while accommodating bulkier ones — meaning that despite sharing leucine-95 with B39:01, B38:01 preferentially binds a different peptide repertoire. Consistent with this, all three allotypes showed differential binding to peptides derived from beta cell autoantigens (including insulin and glucose-6-phosphatase 2, per the published paper), tying the structural differences directly to differences in which autoantigen peptides each allele can present to autoreactive CD8+ T cells.

🎯 Core Premise

X-ray crystal structures of HLA-B39:06 (1.7 Å resolution) bound to an 8-mer NLRP2-derived peptide, compared against HLA-B39:01 and a structural model of the protective HLA-B38:01, reveal that a single bulky tryptophan at F-pocket position 95 in B39:06 restricts the pocket to small C-terminal residues (favoring alanine-anchored 8-mers), while the smaller leucine-95 shared by B39:01 and B38:01 permits larger residues and a distinct peptide repertoire — with B38:01's additional groove differences further disfavoring basic residues. Because these three alleles — despite >97% sequence identity — differ in which peptides they can present, and because they show differential binding to beta cell autoantigen-derived peptides, the tiny structural differences between them translate directly into differences in which autoreactive CD8+ T cell responses each allele can support: the molecular basis for why B39:06 is the most aggressive class I T1D risk allele known, while B*38:01 is protective.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

For scientists: Structural biology transformed understanding of class II–mediated T1D risk years ago, when crystal structures of HLA-DQ8 and HLA-DQ2 revealed how an unusual empty position 57 in DQ8’s groove creates its risk-conferring peptide-binding preferences. Sharma’s work opens the equivalent chapter for class I. With atomic-resolution structures of B39:06 and B39:01 now in hand — plus a structural model of protective B*38:01 — the field has a physical scaffold for mapping autoreactive CD8+ T cell responses, designing peptide-binding inhibitors, and explaining, mechanistically, why a two-decade-old genetic association has taken this long to become structurally legible. The F pocket finding is especially consequential: because C-terminal anchor residues are typically the single most important determinant of whether a peptide binds an MHC molecule stably enough to reach the cell surface, a small geometric difference here has outsized downstream consequences for the entire presented peptide repertoire — and for which autoreactive T cells get activated in the first place.

For clinicians: The translational logic, while early, is direct. If specific beta cell peptides bind B39:06 preferentially and drive the CD8+ T cell responses that destroy beta cells, those peptides become candidate biomarkers for tracking autoreactive T cell activity in B39:06-positive individuals — potentially before clinical onset. Structure-guided small molecules or peptide mimetics could, in principle, competitively block the F pocket and prevent autoantigen presentation without broadly immunosuppressing the patient — an approach Sharma’s group is actively pursuing via a compound-library screen. And because B*39:06 is disproportionately common in Latino populations, a group historically underrepresented in T1D research, this structural program has a direct path toward addressing a population-specific gap in risk stratification and, eventually, in immunosuppression-dosing decisions for interventions like islet transplantation.

3️⃣ Big Takeaways

1. Two or three amino acid differences are enough to convert the most aggressive known HLA class I T1D risk allele into a protective one — and the F pocket is where that difference lives. HLA-B39:06, B39:01, and B38:01 share more than 97% amino acid identity, yet range from strongly predisposing (B39:06) to protective (B38:01). The crystal structures show this comes down largely to residue 95 in the F pocket: a bulky tryptophan in B39:06 sterically restricts the pocket to small C-terminal residues, while a smaller leucine in the other two allotypes opens the pocket to larger residues — directly shaping which peptides each allele can stably present.

2. HLA-B39:06 has a genuine structural preference for 8-mer, alanine-anchored peptides — overturning the canonical assumption that position 9 is always the dominant anchor in class I peptide binding. Using an 8-mer peptide (NRVMLPKA) that binds both B39:06 and B39:01, the structure shows alanine at position 8 (not a position-9 residue) forms the strongest anchoring hydrogen bonds at the C-terminus, while methionine-4 and lysine-7 remain solvent-exposed for TCR contact. This matches the only known B39:06-restricted diabetogenic T cell epitope, MRLLA — itself an 8-mer — giving the structural finding direct biological grounding.

3. The structural differences between risk and protective alleles translate into differential presentation of actual beta cell autoantigen peptides — including insulin and glucose-6-phosphatase 2 — meaning the molecular story isn’t just about general immunogenicity but about specific capacity to display the peptides autoreactive T cells are trained to recognize. This sets up a concrete translational pipeline already underway in Sharma’s lab: a ~5,000-compound library screen targeting the F pocket for structure-guided blockers, a collaboration with Dr. Sally Kent (UMass Chan) to identify patient-derived B39:06-restricted T cell epitopes and eventually solve full TCR–peptide–MHC ternary complexes, and population-specific peptide-binding studies motivated by GWAS data at Einstein showing B39:06 is predominant in the Latino population.

   ---

❓ Key Questions from the Discussion

Could this structural work eventually become a clinical assay for stratifying autoimmune aggressiveness — for example, to guide immunosuppression dosing around future islet transplants? An audience member framed this in terms of minimizing immunosuppressant burden by predicting a given patient’s rate of autoimmunity ahead of a transplant. Sharma noted that Einstein has a large ongoing GWAS effort in T1D and has found that B*39:06 is predominant in the Latino population, which is highly predisposed; the lab has begun working out which peptides would be most useful for a population-specific assay, with the goal of building toward that kind of clinical tool as more data accumulates.

Is the peptide NRVMLPKA, or any of its mimics, a known T1D epitope? Ky Gerder (Medical College of Wisconsin) asked this directly. Sharma explained the peptide was eluted from both B39:06 and B39:01 as their strongest binder, with an IC50 comparable to known T cell epitopes, but that confirming true mimicry will require completing ongoing work with the actual diabetogenic epitope MRLLA — at which point a direct comparison will be possible.

If the goal is structure-guided therapeutics, is blocking the F pocket the best molecular handle, or are there other more selectively targetable sites in the groove? A questioner pushed on whether F-pocket blockade risks broader immune effects versus more selective alternatives. Sharma said her group is currently screening a library of roughly 5,000 compounds and does not yet know which will show the best inhibition, though she expects the F pocket to be the most promising target given that it is the site of the sharpest structural divergence between B*39:06 and the other allotypes; a definitive answer awaits further screening data.

Does the difference between predisposing and protective HLA-B alleles come down to different epitope preferences? Peter Lindqvist (Benaroya Research Institute) asked this. Sharma agreed this is the core logic: because the mature alleles differ in the amino acids lining their pockets, they cannot be expected to present the same peptide repertoire, and epitope preference will vary allele by allele.

Do you plan to test point mutations — for example, at the key tryptophan-95 residue — to see their effect on binding affinity? Ky Gerder also asked this. Sharma said this hadn’t been done yet but agreed it was an interesting next step, speculating that altering tryptophan-95 specifically could substantially change B*39:06’s binding behavior, since it’s the residue most responsible for the allele’s distinctive F pocket geometry.

Are you pursuing full TCR–peptide–MHC ternary complex structures, and what would that reveal about where the true risk-conferring features live? A questioner noted that understanding TCR recognition ultimately requires the ternary complex, not just the peptide-MHC structure. Sharma confirmed an active collaboration with Dr. Sally Kent at UMass Chan Medical School, who works on patient-derived B*39:06-restricted T cell epitopes; preliminary studies are just beginning, with fuller results expected as the crystallography (which she noted can take anywhere from a single attempt to six months of condition optimization) matures.

Has this been tested in non-diabetic individuals, to see whether B39:06 carries risk outside the autoimmune T1D context? Dan Heller asked this. Sharma confirmed HLA-B39:06 is present in the general non-T1D population but is not highly predisposing outside the T1D context — prompting a brief exchange about how identical-twin studies might illuminate this further, with Sharma noting such personalized studies tend to be more resource-intensive than standard population studies.

🔗 3 TSS Talks That Connect With This One

1. The Search for Diabetes-Specific T-cell Repertoires — Michel Edwar Mickael, PhD (TheSugarScience T1D Th1nk Tank, January 2026) Mickael’s talk covers how T1D-associated TCR sequences are identified from the broader repertoire — the receptor side of the exact molecular handshake Sharma’s structures illuminate from the MHC side. Watching this alongside today’s talk completes the picture: which TCRs recognize which peptide-MHC complexes, and why.

2. Bart Roep, PhD — The Quest to Cure T1D: Antigen-Specific Immunotherapy and the Reverse Vaccine (TheSugarScience Podcast, Episode 53) Roep’s “reverse vaccine” concept — using specific beta cell peptides to restore tolerance rather than broadly immunosuppress — is the therapeutic paradigm Sharma’s structural work feeds most directly. Designing a reverse vaccine for B*39:06-positive individuals requires knowing exactly which peptides that allele presents to autoreactive T cells — the atomic-level knowledge Sharma’s crystal structures are building.

3. WAVE T1D: Multicenter RCT Combination Therapy in New-Onset Type 1 Diabetes (TheSugarScience Clinical Trial Th1nk Tank, February 2026) This talk covers the current front line of combination immunotherapy in new-onset T1D — therapies that broadly target the adaptive immune system. It provides the clinical contrast for Sharma’s structural program: today’s treatments aren’t HLA-restricted, but the next generation of precision immunotherapy may need to specifically target B*39:06-restricted autoreactive responses, which is exactly the structural groundwork this talk describes.

   ---

   🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

When Beta Cell Specification Fails: A Knockout Village Reveals Lineage Rewiring in Human Islet Development Dingyu Liu PhD Candidate • June 30 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“It's really the village that gives us the scale and also single-cell resolution to reveal this other side of monogenic diabetes. It's not just about the loss of beta cells — it's also about what the cells become.” — Dingyu Liu, PhD Candidate

🔬 Foundational Insights as They Apply to T1D

The standard story of monogenic diabetes is a story of arrest: a mutation in a transcription factor essential to beta cell development — PDX1, PAX6, RFX6, and roughly a dozen others — stalls the developmental program, beta cells fail to form in adequate numbers, and diabetes results. That account has been built gene by gene, over decades, largely from individual mouse knockouts studied one at a time — PDX1 loss producing near-total pancreatic agenesis, RFX6 loss producing endocrine cells that fail to turn on insulin or glucagon. What that gene-by-gene approach has never been able to answer is what those failed cells actually become, and whether there is a shared logic across different monogenic diabetes genes rather than 15 separate, disconnected stories.

Liu’s PhD work, from the Huangfu lab at Memorial Sloan Kettering, answers that question using a strategy she calls the “knockout village.” Rather than differentiating each CRISPR knockout line through the roughly 20-day stem-cell-to-islet protocol one at a time — an approach that is slow and vulnerable to batch and clonal variation that can obscure real phenotypes — the lab generated 76 CRISPR-engineered human embryonic stem cell lines (36 genotypes) targeting 30 diabetes- and islet-agenesis-associated genes, gave each line a unique heritable genetic barcode, and then pooled 79 of those clones into a single shared differentiation dish. The cells were profiled by single-cell RNA sequencing at five time points spanning pluripotency through early SC-islet stage, and demultiplexed by barcode afterward to recover the phenotype of each individual knockout — all cultured, differentiated, and sequenced under identical conditions.

The village recapitulated the expected phenotype: most knockout genotypes showed a reduced fraction of beta cells relative to wild type, with RFX6 among the most severely affected — consistent with decades of prior mouse and human data. But tracing the data back to earlier time points revealed something the gene-by-gene literature could not show directly: failing cells were not simply dying or stalling. They were being actively redirected toward alternative lineages, and which alternative lineage depended on which gene was lost — FOXA2 knockout cells were biased toward liver fate, GSC (goosecoid) knockout cells toward stromal fate. These redirections held up in individual (non-pooled) differentiations, confirming the effects are cell-autonomous rather than artifacts of pooled culture.

The most striking convergence, and the center of the talk, involved three genotypes — RFX6, PDX1, and PAX6 knockouts — which all showed the same paired signature: severe loss of beta cells alongside a corresponding gain of enterochromaffin (EC)-like cells, an endocrine cell type not normally considered part of the islet. This was reproducible in individual differentiations and, critically, was not just an in vitro artifact: reanalysis of a 20-year-old human RFX6-mutant patient sample (which had not originally been checked for EC markers) and of Rfx6-knockout mouse embryonic pancreas (E15.5) both showed the same pattern — CHGA-positive endocrine cells that had lost insulin, glucagon, and somatostatin expression, with the mouse data additionally showing a significant increase in the EC markers TPH1 and SLC18A1. That cross-species, cross-system convergence gave the team confidence that the EC-like fate is a genuine pathological cell state, not a cell-culture accident — and, intriguingly, the EC-specific gene program the team identified showed even stronger neuronal-feature enrichment than other endocrine gene programs, echoed by cross-referencing against a human fetal single-cell atlas spanning 15 organs (gestational weeks 11–13).

Using non-negative matrix factorization to define 36 gene expression programs across the dataset, and the SCENIC+ computational method to infer transcription factor regulators from single-cell multiome data, Liu’s team found that beta cell and EC gene programs are governed by largely distinct transcription factor sets — and identified ISL1 as a candidate downstream repressor of the EC fate, based on its expression being down in EC-high knockouts and negatively correlated with EC:beta ratio across genotypes, together with prior human genetic reports linking ISL1 mutations to early-onset diabetes. Overexpressing ISL1 in wild-type cells, ISL1-knockout cells, and — critically — in PDX1-knockout and PAX6-knockout cells was sufficient to repress the EC gene program across all of them, showing ISL1 acts downstream of at least two other monogenic diabetes genes to guard against the EC fate. ISL1 overexpression also reactivated the beta cell program in wild-type and ISL1-knockout cells, but notably not in the PDX1 or PAX6 knockouts — indicating those two genes have additional beta-cell-promoting roles beyond simply working through ISL1.

🎯 Core Premise

Using a pooled, barcoded "knockout village" of 79 CRISPR-engineered human stem cell lines spanning 30 monogenic-diabetes-associated genes, profiled by single-cell RNA sequencing across five time points of islet differentiation, Liu and colleagues show that loss of beta cell specification regulators does not simply block beta cell formation — it actively redirects developing cells toward alternative, gene-dependent fates. Three genotypes in particular (RFX6, PDX1, PAX6) converge on the same non-canonical outcome: a gain of enterochromaffin (EC)-like cells with serotonergic and neuronal features, a pattern independently confirmed in human patient tissue and mouse knockout embryos. ISL1 was computationally predicted and experimentally validated as a downstream repressor of this EC fate, acting under both PDX1 and PAX6. The result reframes monogenic diabetes pathology as a story about cell fate redirection, not merely beta cell insufficiency — and the knockout village itself is a reusable, scalable platform for studying many developmental regulators simultaneously with single-cell resolution.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

For scientists: The knockout village is the methodological advance the field is likely to adopt widely. hPSC differentiation studies have long been limited by batch effects and clonal variability that make it hard to compare phenotypes across knockout lines run in separate experiments. By barcoding every line and pooling them into a single shared differentiation, Liu’s approach recovers high-confidence, directly comparable phenotypes for 30 genes across five developmental time points in one experiment — a platform any lab studying hPSC developmental genetics could, in principle, adapt. The finding that developmental transcription factors function as much by suppressing alternative fates as by promoting the target fate is a general principle now demonstrated at scale in human cells, and the convergence of three independent monogenic diabetes genotypes on the same EC-like exit fate — validated computationally, then experimentally with ISL1 — is a concrete proof that this atlas-scale approach can generate specific, testable mechanistic hypotheses about understudied regulators.

For clinicians: Monogenic diabetes is frequently misdiagnosed as type 1 or type 2 disease, in part because the underlying developmental mechanisms have been understood gene by gene and incompletely. This work suggests that for at least three of the most severe monogenic forms (RFX6, PDX1, PAX6-associated disease), the pathology may not be simple beta cell insufficiency but active misdirection of progenitors into a non-canonical, serotonergic, neuron-like cell state whose functional consequences in the human islet are still unknown. The stage-specific failure map the village produces — showing that different genes fail at different points in the roughly 20-day differentiation program — is also directly relevant to the growing number of SC-derived islet cell therapy programs now in clinical trials for T1D, since those same differentiation protocols can, off-target, generate EC-like cells instead of the intended beta cells. Identifying ISL1 as a factor that can redirect cells away from the EC fate and back toward beta cell identity — including downstream of PDX1 and PAX6 loss — is a concrete, actionable lead for improving those protocols.

3️⃣ Big Takeaways

1. Losing a beta cell transcription factor doesn’t just erase beta cells — it reroutes the cells somewhere else, and where they end up depends on which gene was lost. Profiling 79 pooled, barcoded knockout lines across 30 genes and five time points, the team found that most knockouts reduce beta cell yield, as expected, but also actively increase alternative-fate populations — liver-like cells in FOXA2 knockouts, stromal-like cells in GSC knockouts — and these effects held up when tested in individual, non-pooled differentiations, ruling out pooled-culture artifacts as the explanation.

2. Three of the most clinically severe monogenic diabetes genes — RFX6, PDX1, and PAX6 — converge on the same non-canonical outcome: a gain of enterochromaffin (EC)-like cells. This EC-like population expresses serotonin pathway genes (TPH1, SLC18A1) and shows unusually strong neuronal gene-program features even relative to other endocrine cell types, and the same pattern was independently confirmed in a re-examined human RFX6-mutant patient sample and in Rfx6-knockout mouse embryos (E15.5) — evidence that this is a genuine pathological cell state relevant in vivo, not an artifact specific to stem-cell culture.

3. ISL1 was computationally predicted and then experimentally validated as a downstream guardian of beta cell identity against the EC fate, acting under both PDX1 and PAX6. Using gene program analysis (NMF) and transcription-factor inference (SCENIC+) across the whole knockout village dataset, the team identified ISL1 as a candidate repressor of the EC program; overexpressing ISL1 suppressed the EC gene program in wild-type, ISL1-knockout, PDX1-knockout, and PAX6-knockout cells alike, though it only restored the beta cell program in the first two — indicating PDX1 and PAX6 have beta-cell-promoting functions beyond what ISL1 alone can rescue.

   ---

❓ Key Questions from the Discussion

How late into differentiation does ISL1 overexpression rescue the beta cell fate, and does the effect persist? Katarina Zorc asked how far into maturation the EC-to-beta-cell recovery with ISL1 overexpression had been tested. Liu explained the results shown were from early-stage differentiation, but preliminary data extending culture out to roughly 10 days of additional maturation suggest the EC-suppressing effect of ISL1 overexpression is maintained across multiple time points, with longer maturation time courses still ongoing.

Do ISL1-overexpressing cells show improved hormone specificity (reduced polyhormonality) and actual insulin-secretory function? A questioner asked whether ISL1 overexpression improves polyhormonality or was tested by glucose-stimulated insulin secretion (GSIS). Liu said GSIS testing hasn’t been done yet but is planned, and noted that assessing true polyhormonality is complicated in this in vitro system because nearly all endocrine cells in the differentiation carry some baseline insulin expression alongside glucagon or somatostatin — making single-cell sequencing after ISL1 overexpression a better path to clarifying exactly which cells become what.

Could the large proportion of wild-type cells in the pooled village be rescuing knockout cells’ phenotypes, masking a stronger true effect? Diego Matías asked whether co-culturing knockouts with a majority of wild-type cells (roughly 70% wild type to 30% knockout in the village) could create a favorable local environment that partially rescues the knockout phenotype, potentially underestimating the true effect size. Liu agreed this is plausible — that ratio was chosen specifically to mitigate knockout-knockout interactions confounding results — but noted the defects observed were nonetheless substantial, and that most reported phenotypes were independently confirmed in individual (unpooled) differentiations.

Is there direct in vivo evidence of increased EC-like cells in human monogenic diabetes patients, beyond the reanalyzed RFX6 sample? Ed Sanchez asked whether there’s evidence for a higher fraction of EC-like cells in monogenic diabetes patient pancreata versus non-diabetic controls. Liu acknowledged that direct human evidence is currently limited to the single reanalyzed RFX6-mutant patient sample, since monogenic diabetes is rare and pancreatic biopsy is not standard clinical practice; supporting in vivo evidence otherwise comes from the Rfx6-knockout mouse model, with gene expression data from the other knockout mouse models not yet available but of active interest to the group.

Do beta cells arising in the non-canonical (EC) state show altered cell-cell communication within the islet? Nina Ostrer (referred to as “Nina Itis” in the transcript) asked whether non-canonical-state cells show reduced or altered signaling with other islet cell types. Liu noted the pooled-culture design makes it difficult to control or attribute exactly which cell types are signaling to which within the village, so this question hasn’t yet been directly addressed.

How reproducible are cell-fate outcomes across independent clones of the same knockout genotype? Juan Alvarez asked about village-to-village reproducibility of cell type allocation across different clonal lines of the same knockout. Liu reported that for most genotypes, two independent clonal lines were tested and showed generally reproducible cell type composition across differentiation stages.

Does NKX6-1 mark EC cells as well as beta cells, and could that inform strategies to generate C-peptide-positive, NKX6-1-negative cells? Christian Schutz asked about a slide suggesting NKX6-1 positively regulates EC cells as well as beta cells, and whether it might therefore be beneficial to generate C-peptide-positive but NKX6-1-negative cells. Liu called it a great and still-unresolved question: NKX6-1 is expressed in EC cells, and its expression increases in ISL1 knockouts — suggesting NKX6-1 marks both cell states rather than being beta-cell-specific — but the group does not yet have genetic evidence clarifying how NKX6-1 mechanistically shapes EC versus beta cell development or function.

🔗 3 TSS Talks That Connect With This One

1. Ask the Expert: Dario Gerace, PhD — Harvard University Dr. Gerace’s work engineering immune-evasive stem cell-derived islet cells (Melton lab) situates today’s developmental biology inside the broader SC-islet cell therapy program that is now in clinical trials for T1D. Because the differentiation protocols used to make clinical SC-islets can, off-target, generate non-canonical populations like the EC-like cells Liu’s team describes, understanding the clinical stakes of getting beta cell specification right — rather than just efficient — makes Liu’s mechanistic findings directly actionable.

2. Ask the Expert: Lorenzo Pasquali, PhD — Pompeu Fabra University Dr. Pasquali’s work on beta cell noncoding regulatory function provides the gene-regulatory-network framework — how transcription factors bind regulatory elements to control identity programs — that underlies everything in Liu’s talk. Watching this alongside today’s talk clarifies what it actually means, mechanistically, for a transcription factor like PDX1 or ISL1 to be “guarding” one cell fate against another.

3. Ask the Expert: Ruth Elgamal, PhD Candidate — UCSD Elgamal’s discussion of the integrated pancreatic islet reference map from the Human Pancreas Analysis Program (HPAP) supplies the single-cell classification vocabulary and marker-gene frameworks — SC-β, SC-α, SC-δ, and non-canonical populations like SC-EC — that are central to interpreting Liu’s single-cell RNA-seq results and the novel EC-like cell state she describes.

   ---

   🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

Generation of Hypoimmunogenic Gastric Insulin-Secreting Organoids Anna Ada Dattoli PhD • June 23 2026

📺 Now available on demand ▶ Watch the Full Talk

💬 Key Quote

“My work is on cell therapy, and a successful cell therapy is one that lasts.” — Anna Ada Dattoli PhD

🔬 Foundational Insights as They Apply to T1D

The GINS (gastric insulin-secreting) platform, developed in the Zhou lab at Weill Cornell, starts from an unusual premise: instead of the pancreas or a pluripotent stem cell line, human gastric stem cells — accessible by routine endoscopic biopsy — are the source material. Sequential induction of three transcription factors (NGN3, PDX1, and MAFA) converts these gastric stem cells into functional, glucose-responsive insulin-secreting organoids in about ten days, and prior work from the lab showed these organoids can restore normal blood glucose in diabetic mice — with hyperglycemia returning if the graft is removed, confirming the effect is graft-dependent.

Dattoli’s project, since joining Weill Cornell, asks the next logical question: are GINS organoids naturally less visible to the immune system than the stem-cell-derived beta cells the field usually works with? To test this cleanly, she compared GINS organoids and iPSC-derived islets generated from the same donor — gastric stem cells from a stomach biopsy, iPSCs from splenocytes, both differentiated in parallel, making the comparison isogenic. Insulin content and glucose-stimulated insulin secretion were equivalent between the two cell types, but single-cell RNA sequencing (focused on the purest, insulin-high populations of each) showed that two clinically important T1D autoantigens — GAD65 and ZnT8 — were specifically reduced in GINS organoids compared to iPSC-islets, a finding Dattoli confirmed at the protein level by flow cytometry (with ZnT8 showing a strong trend that didn’t fully reach significance).

To test functional immune evasion directly, Dattoli built an “avatar T cell” challenge system: naive CD8+ T cells isolated from donor PBMCs are transduced with a TCR specific for a preproinsulin-derived peptide (linked via a T2A sequence to a GFP reporter for sorting), creating a population of T cells engineered to specifically attack insulin-producing cells. Organoids were first treated with a carefully calibrated low dose of interferon-gamma (6 ng/mL) — enough to raise MHC class I expression (mimicking what happens to beta cells during T1D pathogenesis) without substantially inducing PD-L1, which can itself be stress-induced and confound the results. In a chromium-release cytotoxicity assay across multiple effector:target ratios, GINS organoids were consistently less sensitive to avatar T cell killing than iPSC-islets, with roughly a 60% reduction in cytotoxicity at a 5:1 effector:target ratio — evidence of an intrinsic immune-evasion property in GINS cells even before any genetic engineering.

Dattoli then asked whether this natural resistance could be reinforced. She engineered GINS organoids to overexpress PD-L1 — a molecule well established in tumor immunology (and previously used to protect stem-cell-derived beta cells in autoimmune contexts) that binds PD-1 on T cells and inactivates them. PD-L1 overexpression did not impair insulin content or glucose responsiveness. In a repeat immune challenge — this time loading organoids with synthetic preproinsulin peptide to generate a maximally strong T cell attack — PD-L1-positive GINS showed an additional ~40% reduction in cytotoxicity at the 5:1 ratio. A neutralizing anti-PD-1 antibody experiment confirmed the mechanism was specific: blocking PD-1 had no effect on control (PD-L1-negative) organoids, but restored cell death in PD-L1-positive GINS back up to control levels — ruling out a nonspecific genetic-engineering artifact.

To extend beyond a simple co-culture assay, Dattoli used a vascularized organ-on-a-chip system: endothelial cells combined with 3D organoids self-assemble into perfusable vessel-like structures within 48 hours. Validating the chip first with iPSC-islets, she showed that non-specific control T cells spread throughout the device without effect, while preproinsulin-reactive avatar T cells specifically attacked the islets, reducing the survival fraction by roughly 1.4-fold. Repeating this with GINS organoids and autologous avatar T cells (derived from the same donor’s splenocytes), PD-L1-positive GINS showed approximately a 20% increase in survival compared to PD-L1-negative GINS under active T cell attack.

Looking ahead under her Breakthrough T1D Innovation Award, Dattoli presented preliminary evidence that GINS organoids carry a broader natural protective signature: higher expression, relative to iPSC-islets from the same donor, of classic immune-evasion genes, of stress-resistance genes (drawn from Dr. Todd Brusko’s diabetic pancreas gene database), and of beta cell survival genes recently published by Dr. Shuibing Chen’s group. She also found that T cells co-cultured with GINS release less TNF-alpha than T cells co-cultured with iPSC-islets, and that T cells exposed to GINS adopt an inhibitory phenotype (low CD25, high PD-1) — suggesting GINS organoids may naturally suppress the T cells around them, independent of the engineered PD-L1 effect. She connected this to a related, very recent finding from Dr. Todd Brusko’s group showing that a gene called PBR can protect stem-cell-derived beta cells when its expression is experimentally induced — whereas GINS organoids appear to express PBR naturally, without requiring induction, hinting that GINS may retain a kind of protective molecular “memory” from their gastric tissue origin.

🎯 Core Premise

Gastric insulin-secreting (GINS) organoids — generated from a patient's own endoscopically accessible gastric stem cells via NGN3/PDX1/MAFA induction — show reduced expression of the T1D autoantigens GAD65 and ZnT8 relative to isogenic iPSC-derived islets, and are intrinsically less sensitive to antigen-specific T cell killing in an engineered "avatar T cell" challenge assay (~60% reduction in cytotoxicity at a 5:1 effector:target ratio). Engineering GINS organoids to overexpress PD-L1 adds a further, PD-1-dependent layer of protection (~40% additional reduction in cytotoxicity, and ~20% improved survival under active T cell attack in a vascularized organ-on-chip model) without requiring HLA or MHC class I knockout and without impairing insulin secretory function. Together with a broader natural signature of immune-evasion, stress-resistance, and beta-cell-survival gene expression, this positions PD-L1-engineered GINS organoids — now the subject of a US provisional patent — as a candidate off-the-shelf, immunosuppression-free cell replacement therapy for T1D.

🌟 Why This Talk Matters to T1D Scientists and Clinicians

For scientists: The field’s dominant hypoimmunogenic engineering strategy — knocking out B2M and CIITA (eliminating MHC class I and II) while overexpressing CD47, established by Deuse and colleagues in 2019 — reliably evades allogeneic rejection but creates a well-known tradeoff: removing MHC class I also removes the signal that keeps natural killer (NK) cells in check, creating vulnerability to NK-mediated killing that later work (such as selective HLA deletion approaches retaining HLA-E/F/G) has had to specifically engineer around. Dattoli’s finding that GINS organoids achieve substantial immune protection — both intrinsically and via PD-L1 overexpression — without requiring any HLA or MHC knockout at all is a genuinely distinctive result, since it may sidestep the NK-cell tradeoff altogether. The isogenic, same-donor comparison design (GINS versus iPSC-islets from identical genetic backgrounds) is also methodologically important: it isolates a cell-of-origin effect (gastric versus pluripotent-stem-cell-derived) from genetic background, strengthening the claim that GINS cells carry some innate protective biology tied to their tissue origin.

For clinicians: T1D cell therapy has generally treated the alloimmune and autoimmune rejection problems as separate challenges — autologous cell sourcing addresses the former, immunosuppression or genetic engineering addresses the latter. If GINS organoids combine a practical, endoscopically accessible cell source with genuinely reduced expression of autoantigens like GAD65 and ZnT8, plus PD-L1-mediated protection against ongoing autoimmune attack, that combination could in principle address both problems at once without chronic systemic immunosuppression — a milestone the field has been moving toward since Sana Biotechnology’s January 2025 report of the first human patient producing insulin from transplanted hypoimmune islet cells without immunosuppression. Dattoli was explicit in discussion that her near-term therapeutic vision is an off-the-shelf, allogeneic product (enabled by PD-L1 and GINS’s natural properties) rather than a biopsy-dependent autologous approach, which would meaningfully broaden accessibility if it holds up in further preclinical validation — chiefly, the humanized mouse model she described as her next and most important experiment.

3️⃣ Big Takeaways

1. GINS organoids naturally express less of two clinically important T1D autoantigens — GAD65 and ZnT8 — compared to iPSC-derived islets from the identical donor. This reduction was confirmed at both the RNA (single-cell sequencing) and protein (flow cytometry) level, using a same-donor, isogenic comparison specifically designed to isolate the effect of cell-of-origin (gastric stem cell versus iPSC) from genetic background.

2. GINS organoids are intrinsically more resistant to antigen-specific T cell killing than iPSC-islets — and engineered PD-L1 overexpression adds a second, independently verified layer of protection, without requiring MHC knockout. In a chromium-release cytotoxicity assay using preproinsulin-specific “avatar” T cells, GINS alone showed roughly 60% less cytotoxicity than iPSC-islets at a 5:1 effector:target ratio; PD-L1 overexpression added a further ~40% reduction, confirmed as PD-1-dependent via a neutralizing antibody control, and translated into a ~20% survival benefit in a vascularized organ-on-chip model under active autologous T cell attack.

3. GINS organoids appear to carry a broader, naturally occurring protective gene signature — including immune-evasion, stress-resistance, and beta-cell-survival genes, plus a T cell-suppressive effect on neighboring T cells (lower TNF-alpha release, an inhibitory CD25-low/PD-1-high phenotype) — that may trace back to their gastric tissue origin. This includes natural expression of PBR, a gene shown elsewhere (Brusko lab) to require induction for protective effect in stem-cell-derived beta cells, but which GINS organoids appear to express constitutively.

   ---

❓ Key Questions from the Discussion

Did PD-L1 expression itself change after interferon-gamma treatment, since IFN-gamma is known to increase both MHC class I and PD-L1 — and was there a cytotoxicity control run without IFN-gamma? Christian Schutz asked this directly. Dattoli confirmed she checked PD-L1 expression by flow cytometry, and that the interferon-gamma dose (6 ng/mL) was deliberately chosen as the minimum needed to raise MHC class I without substantially raising PD-L1 (she observed roughly a two-fold PD-L1 increase in GINS and essentially none in iPSC-islets at that dose). She confirmed that no cytotoxicity assay was run without interferon-gamma treatment — all experiments included it, to model the interferon-gamma-rich environment of inflamed islets in T1D.

Is the immune-evasion and stress-resistance gene signature data a true same-donor comparison, and how many donor pairs have been examined? Dr. Rémi Creusot asked this. Dattoli confirmed the comparison is fully isogenic (same donor for both GINS and iPSC-islets) — “the most perfect comparison I could” make — but that complete data currently comes from one donor pair (an earlier second donor’s GINS organoids failed to reach adequate insulin expression and were set aside). She is now working with an additional new donor: NGN3/PDX1/MAFA/PD-L1-expressing GINS cells have already been generated and are undergoing flow cytometry and functional testing, though iPSCs from that donor’s splenocytes have not yet been generated.

Knocking out MHC class I removes CD8 T cell recognition but creates NK cell vulnerability — a tradeoff that has pushed the field toward HLA-E or CD47 co-expression strategies. What is the current immune-shielding architecture for GINS, and how will protection against both T cell and NK cell attack be validated? Dattoli said her planned next step — the namesake experiment for her current grant — is building a humanized mouse model incorporating whole blood (including NK cells, not just T cells) to directly compare GINS and iPSC-derived beta cells from the same donor. She emphasized that GINS already achieve strong protection — both naturally and via PD-L1 — without requiring any HLA or MHC knockout, a result she described as genuinely surprising and exciting. Her longer-term vision includes not only developing GINS as a therapeutic product in its own right, but using it to learn transferable protective mechanisms that could be applied to other beta cell platforms, including stem-cell-derived beta cells and human islet grafts.

Given the provisional patent and Innovation Award, what are the critical remaining preclinical validation steps before first-in-human testing could be considered? Dattoli said she recently participated in a bio-innovation challenge exploring founding a company around this work, and identified the humanized mouse model — described above — as the single most important remaining experiment, expecting it to demonstrate that GINS organoids are less immunogenic than stem-cell-derived beta cells, with hopes to pursue this work in collaboration going forward.

Would a patient on the path toward T1D need a gastric biopsy to access this therapy, or is an allogeneic, off-the-shelf approach envisioned? Monica asked this directly, framing it as a thought experiment about a patient progressing toward T1D. Dattoli clarified her therapy is envisioned as off-the-shelf and allogeneic — enabled specifically by PD-L1 overexpression and GINS’s natural immune-evasive properties — rather than requiring an autologous biopsy-based approach for each patient.

Would use of this therapy be limited to later-stage (Stage 3, clinical) T1D, or could it have earlier applications? Following up, Monica asked whether Dattoli saw a broader use case. Dattoli indicated she is developing additional immunomodulatory targets identified in GINS organoids that could have multiple applications, potentially extending beyond T1D to other autoimmune diseases, rather than restricting the platform’s use case to late-stage disease alone.

🔗 3 TSS Talks That Connect With This One

1. Ask the Expert: Joe Qiao Zhou, PhD, and Steve Xiaofeng Huang, PhD — Restoring Glucose Homeostasis with Stomach-Derived Human Insulin-Secreting Organoids This is the direct scientific foundation for today’s talk, featuring Dattoli’s own mentors discussing the original GINS platform published in Nature Cell Biology — the biology, the advantages over iPSC-derived approaches, and the immune-evasion challenge that Dattoli’s work is now built to address. Essential background for understanding what GINS organoids are before hearing how they’re being engineered for immune protection.

2. Ask the Expert: Braulio Marfil-Garza, PhD, with James Shapiro, MD — Islet Transplantation in T1D This conversation on pancreatic islet transplantation outcomes and the legacy of the Edmonton Protocol frames the clinical stakes directly: what islet transplantation can achieve when it works, and why the systemic immunosuppression it requires limits accessibility so severely. That limitation is exactly the problem hypoimmunogenic GINS organoids are designed to solve.

3. Curing T1D with Hypoimmune Cells — Short Explainer A concise, accessible primer on what hypoimmunogenic engineering means and how it relates to the broader immune barrier problem in beta cell transplantation — useful grounding for understanding the B2M/CIITA knockout and PD-L1/HLA-E strategies discussed in today’s talk, without requiring prior familiarity with the more technical literature.

   ---

   🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.

The Future of Immune Protection for Islet Replacement Therapy

I. in the I.S.L.E.T. framework series

Introduction

Say what you will about the IRA, but you can’t deny that their correspondence had a certain literary quality to it. So much so that, in 2024, one of their threats ended up floating around Facebook as an inspirational quote:

While originally directed at Margaret Thatcher after a failed assassination attempt, this threat (and erstwhile Facebook inspo) perfectly describes the relationship between the immune system and transplanted islets in type 1 diabetes (T1D).

Immune Rejection Mechanisms in Islet Replacement for T1D

The immune system has many ways to kill transplanted cells, means and methods and pathways of unmatched persistence. In islet replacement for T1D, there are several overlapping, mutually reinforcing immune rejection mechanisms in play.

To understand ways of protecting against this attack, it’s useful to understand how it happens: through allorejection — the immune system’s response to cells that aren’t its own — and then the recurrent autoimmune response characteristic of the disease.

First, there is allorejection, which itself comes in multiple waves: innate and adaptive. In innate allo-rejection, natural killer (NK) cells recognize “non-self” cells, punching holes in the new cells’ walls and pouring in self-destruct enzymes, while monocytes release inflammatory chemicals (cytokines) toxic to beta cells. Then there is the adaptive wave of allorejection. This is where T cells read the foreign cells’ surface protein ID (HLA) to kill on sight, and B cells produce antibodies, sticky Y-shaped proteins that work like a flare, calling in complement killers and the immune cells that grab the tags and pull the cell apart. The antibody arm is the hardest to stop; it works even beneath the drugs built to restrain T cells.

And then comes recurrent autoimmunity. The immune system has already killed these cells once and has proven itself more than willing to do it again (and again) in people with T1D. The autoreactive T cells that killed the beta cells the first time still have an affinity for killing beta cells. While the T-cell arm, like allorejection, works through HLA presentation (which will matter later), the antibody arm targets autoantigens (insulin, GAD65, IA-2, ZnT8, and TSPAN7) that pop up like little flares on the surface of beta cells, attracting immune killing like moths to a light.

Strategies for Immune Protection in Islet Replacement for T1D

This section sorts immune protection strategies onto a spectrum. At one end is modifying the person receiving the cells. At the other end is modifying the cells themselves.

At the far left edge of the spectrum sits immunosuppression: it asks the most of the person through the downstream effects of suppressing their immune system, in addition to a lifelong medication regimen.

In the middle, there is modifying the local environment the cells live in: capsules, engineered materials, co-delivered signals. At the far right, you ask the least of the person receiving the cells, and instead change the cells themselves, editing them for immune protection and, at the far end, outright evasion.

Strategies That Modify the Person’s Immune Function

Broad immunosuppression

The tried-and-true method for protecting transplanted islets is to weaken the immune system via immunosuppressive drugs: broad immunosuppression. While the current standard of care, broad immunosuppression has that old-timey feel to it: heavy-handed, labor-intensive, and kind of dangerous. It means taking a combination of medications on a strict schedule throughout the day, drugs that weaken the whole immune system and not only the parts implicated in killing islets.

Central to this combination of medications are calcineurin inhibitors, which work by blocking the enzymes that T cells need to activate. Because T cells play such a central role in killing, this method works well enough that experienced islet transplant centers report insulin independence rates above 50% at five years. But this comes at a steep price: the immunosuppressive medications that comprise the standard of care are nephrotoxic, can cause encephalitis, increase risks for cancers and deadly opportunistic infections. And, in a particularly cruel twist, calcineurin inhibitors are toxic to beta cells themselves.

Targeted blockades

For a T cell to activate, it requires two distinct immune signals: antigen recognition and binding to the cell that presents the antigen. This is often thought of as a “handshake.” Newer drugs work as targeted blockades to interrupt the specific handshake a T cell needs before it can target cells, an approach called a costimulation blockade.

Examples include belatacept, which acts as a decoy that grabs the binding sites first, so that the T cell finds nothing to hold. Tegobrubart (Eledon, not yet clinically approved) interrupts the handshake that authorizes the T-cell attack. Costimulation blockades have tremendous potential: tegobrubart paired with deceased donor islets achieved 100% insulin independence in all participants in its investigator-led trial. No recipients have reported any side effects as of the time of this writing, with one even working in the front lines of healthcare.

While these drugs spare people and their islets from a certain level of toxicity, they do not spare the person a lifetime of medication and all of the potential side effects that come with it. (See: Tzield’s boxed warning from the FDA for life-threatening cases of viral reactivation, or efluzimab’s withdrawal from the market after being linked to cases of a rare fatal brain infection; more on this shortly). And even the most precise immunosuppressant is a resignation to a weaker immune system in a world where you probably really need a strong one.

Tolerance induction

A more ambitious idea than suppressing the immune system for life is to persuade it to accept the transplanted cells, an approach known as tolerance. There are several proposed mechanisms for achieving this.

The first is central tolerance, also known as mixed chimerism. It starts in the thymus, where developing T cells that react to “self vs. other” are deleted before reaching the bloodstream. This central tolerance can be recreated for a graft by transplanting a sliver of the donor’s bone marrow so the recipient’s thymus also learns to delete anti-donor T cells. But that demands harsh conditioning involving irradiation and chemotherapy. And it has not proven resilient; even more, in T1D specifically, a clinical trial investigating this approach to central tolerance failed to achieve it.

Another strategy is Treg therapy, or peripheral tolerance. It relies on regulatory T cells (Tregs), the cells responsible for dampening immune responses after an infection is beaten and reducing inflammation overall. This strategy includes Treg infusion, Treg expansion, tolerogenic dendritic cells, and slow drug withdrawal.

Tolerance’s Proof of Concept

As ambitious as tolerance may sound, it has achieved proof of concept in a human being, albeit accidentally.

This happened when Wisel, Stock, and colleagues tested calcineurin-sparing regimens built on belatacept or efalizumab in a cohort of ten people with T1D who received islet replacement therapy. Then, while the study was humming along, tragedy struck. Efalizumab was unexpectedly withdrawn from the market after reported side effects of a fatal brain virus. This meant the people taking efalizumab had to switch to another regimen. One person formerly on efalizumab, referred to in the study as EFA-4, could not tolerate the alternative immunosuppressants, developing life-threatening infections and side effects. Eventually nothing was left that she could safely take, and her physicians stopped all of it, expecting she would lose the islet transplant and insulin independence with it.

But she didn’t. At the time of the report, she had been off immunosuppression for nine years, with an islet transplant placed roughly fourteen years earlier, and remained insulin independent.

Notably, nothing about this was engineered. The study had no Treg product, no tolerogenic cells, meaning that whatever tolerance she reached, her body reached on its own. It happened completely by accident, which is both the most hopeful and the maddening part.

Strategies That Modify the Local Immune Environment

Encapsulation

The idea of encapsulation is to create a physical barrier between the immune system and the transplanted cells that isolates them from the immune system. The barrier would ideally have pores sized so that glucose, insulin, and oxygen could pass freely, while immune cells and the large proteins of immunity could not.

There are three scales of encapsulation, in the order of impacting the person to impacting the cells. In macroencapsulation, many islets are placed in one retrievable device, which is then surgically placed within a person. In microencapsulation, a handful of islets are placed in a capsule. In nanoencapsulation, a conformal coating is applied a few molecules thick around a single islet.

Encapsulation approaches face many challenges, both for the encapsulated cells and the person receiving them. Since its inception, it has been plagued by fibrosis (scarring) resulting from a foreign body response to capsules or a device. Barriers create problems with diffusion, carrying over one of the core problems with exogenous insulin. Additionally, encapsulated cells tend to suffocate: they struggle with hypoxia, or a lack of oxygen, from poor vascularization. Islets are quite metabolically demanding. They need some of the most oxygen and blood supply in the body; they cannot have either compromised by a barrier and survive. No encapsulation strategy has yet achieved durable, immunosuppression-free islet function that results in insulin independence in a human being.

Biomaterials and co-delivery

In response to hypoxia and fibrosis challenges, researchers have been developing biomaterial strategies to engineer a more protective local environment for transplanted islets. One example is building the chemokine CXCL12 into alginate microcapsules, which protects islets without immunosuppression by recruiting regulatory cells and reducing inflammation. Another is FasL on microgels, which induced acceptance of islets in nonhuman primates. Another promising strategy is co-delivering mesenchymal stem cells with islets, which has been shown to calm inflammation and help them vascularize. There is also, of course, combining these approaches.

Strategies That Modify the Cells

Gene Editing: Ghost Cells

On the far end of the spectrum, we have a strategy that changes only the cells: gene editing approaches to immune protection and evasion.

The idea of gene-edited cells for immune protection comes from how the body already runs immune-privileged sites — the testis, the eye, the maternal-fetal interface — where foreign tissue survives indefinitely: the body has designed the cells in these sites to help them evade recognition and stay protected.

Remove immune recognition

HLA is the foundation of immune recognition and evaluation. Gene editing approaches that modify cells to not express HLA class I or II have shown significant promise. They achieve this through two gene knockouts that determine HLA presentation. The first knockout is β2M, pronounced “beta-2 microglobulin,” which is required for HLA class I to be shown on the cell surface. The second knockout is removing CIITA, or Class II HLA Transactivator, which is, well, the activator for HLA class II.

Some programs target HLA-A/B/C (class I) directly rather than targeting their mechanisms of presentation. Alternatively, some target TAP, which prevents peptides from being loaded onto HLA for expression. Because every T-cell arm of the attack against transplanted islets depends on HLA presentation, preventing HLA expression blinds them all at once.

Add immune inhibitors

A cell showing no HLA, unfortunately, is its own alarm. Natural killer cells patrol for exactly that absence — missing-self recognition — on the logic that a cell hiding its identity usually has something to hide. One approach to address this is to add a countersignal that inhibits this response cascade. CD47 is the canonical one, a “don’t-eat-me” signal that many healthy, naturally occurring cells already display, engaging the SIRPα checkpoint to send macrophages and NK cells along their merry way. Other approaches include adding inhibitors via HLA-E or G, which also ward off NK cells, and PD-L1 and FasL, which act on T cells by exhausting or killing them if they get too close.

Taken together, these gene edits are referred to as immune evasion, a kind of invisibility cloak that allows these previously vulnerable cells to avoid being targeted by the immune system – or even seen by it at all. Islets that receive these edits are called hypoimmune islets, operating beneath the immune system’s awareness – ghost cells.

Gene editing islets for immune evasion has shown some pretty exciting proof of concept in the UPF 421 Carlsson case. Like the tolerance case above, this is also N=1. But in this study, it was done on purpose. This study involved deceased donor islets that were edited to remove immune recognition (the β2M and CIITA knockouts discussed above) and to add immune inhibitor CD47, injected into a man’s forearm muscle.

Investigators injected three types of cells: unedited ones, partially edited ones that received only the HLA knockouts, and ones that received both the HLA knockouts and overexpression-of-CD47 edits to become hypoimmune. Over fourteen months, the fully edited hypoimmune cells survived untouched. In the same forearm, at the same time, the unedited and partially edited cells in the product were rejected and killed right on schedule, with a strong T-cell attack peaking at day 7, and antibodies class-switching by week three. The fully edited cells sat right next to that carnage and were ignored, secreting glucose-responsive C-peptide the whole time.

Unfortunately, the hypoimmune strategy comes with serious risks. Gene editing can be dangerous. Each knockout, knock-in, and round of expansion represents another chance to miscopy three billion letters of DNA. The more edits, the better the odds that a slightly off cell gains a growth advantage or becomes malignant. A nickname making the rounds for heavily edited cells is Franken-cells, or Frankenstein cells, implying that the more edits you add, the more you lose control of your potentially dangerous creation.

The need for a kill switch

These risks are why a kill switch belongs in the design of any hypoimmune islet product. The hypoimmune platform in the UPF 421 study built one in: because the cells overexpress CD47, a systemic anti-CD47 antibody clears them selectively and spares the person’s own CD47-expressing cells. Other programs bolt on a kill switch, such as HSV-thymidine kinase (vulnerability to ganciclovir), or RapaCasp9 (cell death on demand). Both work but neither is complete, since reported kill switch systems clear about 95 percent of cells, and in a dividing population the surviving 5 percent is significant.

Tolerance vs. Evasion

Only two strategies have ever produced drug-free, device-free islet survival in a human being. Tolerance did it once, by accident, in the Wisel et al study where efalizumab was withdrawn. Evasion did it once, on purpose, in a study designed to do so. Everything else on the spectrum still needs drugs, a foreign body such as capsules or devices, or both.

Tolerance is an equilibrium, and equilibria tip. The immune system is always evolving. Out of the blue, a viral infection, or even a change in gut bacteria, might provoke a broad T-cell response and expand the immune memory pool. Infection simultaneously blunts Treg suppression and raises HLA, brightening the target exactly when the immune system is most agitated.

This means that tolerance would have to hold not against the current immune system but against every future version of it.

Evasion has one piece of evidence tolerance cannot match: edited cells survive inside a hostile environment of active rejection. In primates, and then in the UP421 recipient, they lived while unedited cells died a few centimeters away in the same body at the same time, over multiple versions of who-knows-what kind of immune system evolution.

Conclusion and the Path Forward: Why I Believe in Ghosts

Over the course of researching and writing this article, I’ve encountered countless comparison tables that score immune protection approaches on efficacy, safety, durability, cost, and scalability. I would like to add the attributes that dominate life with T1D: attention and labor. Physical, intellectual, emotional. Islet replacement is worth the risks not only because it reduces complications and physical suffering, but because it could give people back energy and agency.

If there were one word I would use to describe this state of no-days-off, it would be precarity. Evasion is the approach that leaves the least precarity intact. Not because it can’t fail (there is no system which cannot), but because it does not depend on a fragile détente with the immune system, officially the most haunted system in the human body.

So, the ghosts I believe in: immune evasion along with biomaterial microenvironmental engineering.

This ghost approach is a combined one: hypoimmmune cells produced by evasion edits, and biomaterial “ghosts” too. In local microenvironmental engineering, the most sophisticated materials disappear. The forefront of biomaterial engineering is moving toward resorption, an ECM scaffold that helps with vascularization and oxygenation through the vulnerable early window of islet engraftment, before disappearing without a trace.

Kim and colleagues put it best: “Neither genetic immune modulation nor extrinsic microenvironmental control alone is sufficient to ensure durable graft function.” While gene edits would protect the islets from attack, biomaterial engineering would make sure they were not just surviving but thriving. Taken together, this combinatorial ghost approach could create an environment where a haunted immune system, replacement islets, and the person they inhabit may rest in peace.

And besides, sometimes there is good in what you cannot see. If the immune system is reading this, somewhere, I would remind it: the more you stare into the abyss, the more the abyss stares back into you.

***

References

Date: Tuesday, August 25, 2026 · 9:00 AM Pacific · 12:00 PM Eastern
Format: Free virtual seminar for the global diabetes research community — clinicians and scientists are welcome.

👉 Register here on Zoom

Figure adapted from https://doi.org/10.64898/2026.05.25.727705

👤 About the Speaker

Cristina Aguayo-Mazzucato is an Assistant Professor at the Joslin Diabetes Center and Harvard Medical School, where she holds the Margaret A. Congleton Chair. Her laboratory investigates how pancreatic β-cell senescence contributes to islet health and disease, using animal and human disease models to identify therapeutic targets capable of reversing dysfunction and early cellular death during the progression of Type 2 Diabetes. As part of the NIH SenNet Consortium (KAPPSen Tissue Mapping Center), her group has generated one of the most comprehensive single-cell, multi-omic maps of senescence in the human endocrine pancreas to date.

01 · The Paper

“Distinct Senescent β-Cell Senotypes Differentially Drive Islet Aging and Dysfunction” Iwasaki K, Pan H, Dreyfuss J, et al. (Aguayo-Mazzucato C, corresponding author) · bioRxiv, posted May 28, 2026
DOI: https://doi.org/10.64898/2026.05.25.727705

For nearly a decade, findings on β-cell senescence in diabetes have appeared contradictory — some studies showing senescence enhances insulin secretion, others showing it destroys β-cell identity and function. This paper resolves that contradiction directly: it identifies two molecularly and functionally distinct senescent β-cell “senotypes,” defined by differential expression of CDKN1A (p21) and CDKN2A (p16), and shows they have opposite consequences for islet health.

Using single-cell–resolved spatial proteomics and transcriptomics (Xenium, Visium, CODEX) on intact pancreas from 26 donors (ages 20–80), plus scRNA-seq and functional glucose-stimulated insulin secretion (GSIS) assays on dispersed islets from 14 donors (ages 34–69), the team defined:

• CDKN1A⁺/CDKN2A⁻ (maladaptive senotype): loss of β-cell identity genes (INS, MAFA, PDX1, GK, ABCC8), abolished insulin secretion in response to glucose and IBMX, and a proinflammatory, immune-regulatory SASP associated with increased islet immune infiltration.

• CDKN1A⁻/CDKN2A⁺ (adaptive senotype): preserved transcriptional identity, intact and even improved GSIS, and a lower-inflammation SASP enriched for cell signaling and differentiation pathways (including PAPPA, a known longevity regulator).

• Double-positive (CDKN1A⁺/CDKN2A⁺): mixed profile — retained identity but with the same immune-regulatory SASP as the maladaptive senotype.

Both senotypes increase with chronological age across all endocrine cell types, and both β- and γ-cells show the highest senescent-cell burden overall (45% and 42%, respectively), with CDKN1A⁺/CDKN2A⁻ as the dominant senotype in both. Nuclear HMGB1 tracked specifically with the adaptive (p16) senotype, while SA-β-Gal activity tracked with the maladaptive one — giving the field its first senotype-specific biomarker panel.

Tags: β-cell senescence · senotypes · CDKN1A · CDKN2A · p21 · p16 · islet aging · SenNet · SASP · senolytics · Type 2 Diabetes · GSIS · spatial transcriptomics · HMGB1 · Joslin Diabetes Center

02 · Why This Matters

For scientists: This paper directly resolves a long-standing contradiction in the β-cell senescence literature — why some studies (e.g., p16-overexpression models) showed senescence improving insulin secretion while others showed senescence driving dysfunction. The answer, per this data, is that these studies were likely observing different senotypes. That reframes how senescence data across the field should be interpreted going forward: a senescence signature alone is no longer sufficient — senotype identity now needs to be specified. The paper also provides the field’s first senotype-specific biomarker panel (nuclear HMGB1 for adaptive/CDKN2A⁺; SA-β-Gal for maladaptive/CDKN1A⁺), which could be adopted directly by other labs.

For clinicians: The finding that a maladaptive senotype (CDKN1A⁺/CDKN2A⁻) drives β-cell dysfunction and immune infiltration, while an adaptive senotype (CDKN2A⁺) preserves function, argues against blanket senolytic clearance strategies in diabetes. A senolytic that indiscriminately clears p16-high and p21-high cells could remove a protective population along with a harmful one. This paper provides a mechanistic rationale for senotype-selective senotherapeutics rather than pan-senescent clearance — directly relevant as dasatinib+quercetin and related senolytics move toward diabetes-focused clinical testing.

The broader picture: The paper explicitly flags an unresolved tension with prior work: p21 has been shown to be protective in mouse models of Type 1 Diabetes (Lee et al., 2023), while this study characterizes CDKN1A⁺ as the maladaptive senotype in the T2D/aging context. The authors attribute this to differences in the type of senescence-inducing stressor — a limitation they acknowledge directly and one worth asking about live. Separately, the paper’s data is cross-sectional and cannot establish whether CDKN1A⁺ and CDKN2A⁺ populations represent a temporal progression (one senotype transitioning into the other) or parallel, independent states — a distinction with major implications for when a senotype-targeted therapy would need to be given.

03 · Four Questions We Will Ask the Speaker

Q1. Your paper identifies p21 as central to a maladaptive senotype here, but prior work (including Lee et al., 2023) found p21-driven senescence protective in T1D models. You attribute this to differences in the inducing stressor — can you say more about what kind of stressor would push a β-cell toward the adaptive (CDKN2A⁺) versus maladaptive (CDKN1A⁺) path in the first place?

Q2. This is a cross-sectional study, so it can’t directly test whether CDKN1A⁺ and CDKN2A⁺ senotypes represent sequential stages of the same senescence program versus two independent, parallel fates. Do you have a working hypothesis — and what longitudinal or lineage-tracing experiment would be needed to resolve it?

Q3. You found that CDKN1A⁺/CDKN2A⁻ β-cells are spatially closer to extra-islet endothelial cells and farther from ductal cells, and you propose a bidirectional SASP crosstalk between endothelial cells and β-cells. What would be the most direct experiment to test whether senescent endothelial cells are actively inducing the maladaptive β-cell senotype, versus the two simply co-occurring with age?

Q4. Given that nuclear HMGB1 and SA-β-Gal now distinguish the two senotypes, do you see either biomarker as ready for use in screening human islet preparations — for research or even transplant-quality assessment — ahead of a full single-cell profiling pipeline?

04 · Four Key Associated Papers

1. Helman A, Klochendler A, Azazmeh N, et al. (2016) p16(Ink4a)-induced senescence of pancreatic beta cells enhances insulin secretion · Nature Medicine, 22(4):412–420 · DOI: 10.1038/nm.4054
   The original finding that p16/CDKN2A-driven senescence can enhance β-cell function — the earlier, seemingly contradictory result that this week’s paper’s adaptive senotype now mechanistically explains and extends.

2. Aguayo-Mazzucato C, Andle J, Lee TB Jr, et al. (2019) Acceleration of β Cell Aging Determines Diabetes and Senolysis Improves Disease Outcomes · Cell Metabolism, 30(1):129–142.e4 · DOI: 10.1016/j.cmet.2019.05.006
   The speaker’s own foundational paper establishing that senolytic clearance of senescent β-cells improves diabetes outcomes — the therapeutic premise this week’s senotype-resolved data now complicates and refines.

3. Lee H, et al. (2023) Stress-induced beta cell early senescence confers protection against type 1 diabetes · Cell Metabolism, 35(12):2200–2215.e9
   Directly cited in this week’s paper as an apparent contradiction — p21-driven senescence shown as protective in T1D — which the speaker attributes to differences in the senescence-inducing stressor. Essential reading for Q1 above.

4. Thompson PJ, Shah A, Apostolopolou H, Bhushan A (2019) Targeted Elimination of Senescent Beta Cells Prevents Type 1 Diabetes · Cell Metabolism, 29(5):1045–1060.e10 · DOI: 10.1016/j.cmet.2019.01.021
   Shows that senescent β-cell clearance delays T1D onset in mouse models — background for why senotype-selective (rather than pan-senescent) clearance, as this week’s paper argues for, may matter clinically.

   ---

05 · Four Videos to Watch First

▶ Old Dogs, New Tricks: T Cells Drive Beta Cell Senescence in Type 1 Diabetes — Jasmine Pipella https://www.youtube.com/@thesugarscience8602/search?query=Old%20Dogs%20New%20Tricks%20Pipella
The most directly on-topic prior talk in the TSS library — Pipella’s work on T cell–driven β-cell senescence in T1D is the adaptive-immunity mirror image of this week’s talk, which characterizes senescence senotypes in T2D/aging without reference to T cells at all. Watching this first raises the natural question: do the same CDKN1A⁺/CDKN2A⁺ senotypes exist in a T1D islet under immune attack, and if so, which one does immune infiltration select for?

▶ 2025 Most Watched Highlights: Peter Thompson, PhD, & Jasmine Pipella, PhD Candidate https://www.youtube.com/@thesugarscience8602/search?query=Peter%20Thompson%20Jasmine%20Pipella%20Highlights
Peter Thompson’s own paper (Thompson et al., Cell Metabolism 2019, “Targeted Elimination of Senescent Beta Cells Prevents Type 1 Diabetes”) is cited directly in this week’s paper’s reference list as evidence that senescent β-cell clearance delays T1D onset. This highlight reel is a fast, direct way to hear Thompson’s framing in his own words before the talk revisits — and complicates — his clearance-based premise with senotype specificity.

▶ Ask the Expert: Rohit Kulkarni, MD, PhD, Joslin Diabetes/Harvard Med https://www.youtube.com/@thesugarscience8602/search?query=Rohit%20Kulkarni%20Joslin%20Ask%20the%20Expert
Kulkarni is a senior β-cell biologist at the same institution as this week’s speaker — Joslin Diabetes Center/Harvard Medical School. Useful institutional and scientific context for the broader β-cell biology program Aguayo-Mazzucato’s senescence work sits within, from a colleague working on adjacent questions of β-cell dysfunction and identity.

▶ Site Visit Vanderbilt SOM: Danielle Dean, PhD — What makes an alpha cell an alpha cell? https://www.youtube.com/@thesugarscience8602/search?query=Danielle%20Dean%20alpha%20cell%20Vanderbilt
This week’s paper reports that α-cells show far less senescence-associated identity loss than β-cells (ARX is preserved better than MAFA is lost, proportionally, in the senotype data) — an asymmetry that echoes across multiple T1D and T2D contexts. Dean’s work on what defines α-cell identity in the first place is useful background for why α-cells may be structurally more resistant to the identity erosion this paper documents in β-cells.

Graphical Abstract courtesy: Amina Bedrat PhD

📅 Registration

Date: Thursday, August 13, 2026 · 9:00 AM Pacific · 12:00 PM Eastern Format: Free virtual seminar for the global T1D research community — all are welcome.

👉 Register here on Zoom

👤 About the Speaker

Amina Bedrat, PhD is a postdoctoral research fellow in the laboratory of Martin J. Hessner, PhD at the Medical College of Wisconsin, Milwaukee. She earned her PhD from the University of Bordeaux, France, and completed her first postdoctoral fellowship at Harvard University, where she worked at the intersection of bioinformatics and molecular biology — studying alternative DNA secondary structures and their role as epigenetic features in human disease, as well as cancer epigenetics and autoimmune disease heterogeneity through the lens of ribosomal DNA copy number variation. At MCW, she has applied her bioinformatics and statistical expertise to the clinical heterogeneity of T1D, leading to the identification of two immunologically distinct endotypes with differential disease progression rates and dramatically different responses to immunotherapy. She presents today a paper that is both methodologically rigorous and clinically consequential — one that could reshape how T1D clinical trials are designed and analyzed.

01 · The Paper

“Age-Independent Immune Subtypes in Type 1 Diabetes Exhibit Distinct Post-Onset Progression Rates and Immunotherapeutic Responses” Bedrat A, Truchan NA, Pant T, Jia S, Roethle MF, Chen Y-G, Lin C-W, … & Hessner MJ · Diabetologia, 2026 DOI: 10.1007/s00125-026-06794-3

The T1D immunotherapy trial record is littered with drugs that “failed” — not because they don’t work, but because they may have been given to the wrong patients. Every major immunotherapy trial has enrolled a heterogeneous T1D population and looked for an average treatment effect. If that population contains two biologically distinct subgroups — one that responds well to drug A and poorly to drug B, and another with the reverse pattern — the average effect for either drug may be zero or near-zero, and both would be called failures. This paper, from the Hessner lab at the Medical College of Wisconsin, provides the most rigorous evidence yet that this is exactly what has been happening.

The methodology is built on plasma-induced transcription — a functional bioassay in which participants’ plasma is co-cultured with a standardized peripheral blood mononuclear cell (PBMC) reporter population, followed by microarray transcriptomic profiling of the resulting gene expression response. Rather than measuring a fixed panel of proteins or cells, this approach captures the integrated biological activity of everything in plasma — cytokines, chemokines, peptides, lipids, metabolites, microbial antigens — in a single functional readout. The Hessner lab has used this bioassay across more than a decade of T1D research.

Here, Bedrat and colleagues applied plasma-induced transcription to pre-intervention samples from 560 participants enrolled across six TrialNet new-onset T1D immunotherapy trials (rituximab/anti-CD20, abatacept/CTLA4-Ig, MMF-DZB, GAD vaccine, canakinumab/anti-IL-1β, and ATG-GCSF). Unsupervised hierarchical clustering of 2,854 highly variable transcripts identified two reproducible subgroups — designated Subgroup 1 and Subgroup 2 — that were stable over time (p=1.4×10⁻¹⁴), confirming they reflect durable immune states rather than transient disease fluctuations.

Subgroup 1 was enriched for participants with neutral or low-risk HLA haplotypes, with the youngest participants showing the most rapid C-peptide decline. Its transcriptional signature was consistent with a Th1-skewed, innate immune-activated state — featuring elevated IFN-γ, TNF-α, CCL2, and IL-8 signaling pathways, increased circulating CD4⁺CXCR3⁺CCR6⁻ Th1 T cells, and a higher Th1/Treg ratio. Pathway analysis predicted activation by LPS and proinflammatory cytokines, and subgroup 1 participants showed modestly elevated plasma acetate — raising the hypothesis that gut dysbiosis and impaired mucosal immunity may contribute to this inflammatory phenotype.

Subgroup 2 was enriched for participants with insulin autoantibody titres and showed a Th17/regulatory miRNA profile — with elevated plasma miR-155-5p (4.13-fold, p=0.003) and miR-409-3p (2.11-fold, p=0.01), miRNAs that suppress negative regulators of inflammation and promote STAT3-driven Th17 differentiation.

The clinical implications of the drug-response data are the paper’s most striking contribution:

Anti-CD20 (rituximab): Subgroup 1 showed a pattern consistent with a substantially greater benefit — with higher C-peptide AUC in the treatment arm at 6, 12, and 24 months (meta p=0.002), and an 11.2-month delay in C-peptide decline compared to placebo. Subgroup 2 showed no significant effect (meta p=0.27). This aligns mechanistically: Subgroup 1’s high miR-155-5p levels in B cells may promote antigen presentation and immune activation — the very process anti-CD20 targets.

CTLA4-Ig (abatacept): The pattern reversed. Subgroup 2 showed significantly higher C-peptide AUC in the treatment arm (meta p=3.0×10⁻⁴), with an 8.70-month delay in C-peptide decline and greater reduction of CD4⁺CD45RO⁺CD62L⁺ central memory T cells (p=8.1×10⁻⁵) and comparatively less Treg contraction (p=0.02). Subgroup 1 showed no significant response. The mechanistic interpretation: Subgroup 1’s high inflammatory cytokine milieu (IFN-γ, IL-6, TNF-α) may provide alternative co-stimulatory signals that bypass CD28 dependency, directly limiting abatacept’s mechanism of action.

The subgroup classifications were validated in an independent local cohort using a K Top Scoring Pairs classifier trained on TrialNet samples, achieving 83.1% accuracy (AUC=0.96, p=3.02×10⁻⁹).

Tags: T1D endotypes · immune subtypes · plasma-induced transcription · TrialNet · abatacept · rituximab · Th1 · Th17 · miR-155 · precision immunotherapy · C-peptide · heterogeneity · HLA · biomarker · WGCNA

02 · Why This Matters

For scientists: This paper provides the most direct evidence yet that T1D immunotherapy trials have been systematically averaging across biologically distinct patient populations — and that doing so obscures real drug effects in specific subgroups while generating apparent null results across the whole cohort. The plasma-induced transcription bioassay is a methodologically elegant solution to a hard problem: it captures the integrated immune state without requiring prior knowledge of which markers to measure. The 2,854-transcript signature is stable, reproducible across independent cohorts and trial datasets, and predicts differential drug response — the trifecta required of a clinical biomarker. The finding that subgroup 2’s elevated miR-155-5p and miR-409-3p overlap with a recently published miRNA-defined T1D subgroup from a completely independent team further strengthens the biological reality of these endotypes.

For clinicians: The therapeutic implications are direct and actionable. A patient with new-onset T1D characterized by a Th1-high, innate-activated immune profile (Subgroup 1) would be expected to respond well to rituximab but not to abatacept. The same patient’s twin with a Th17/miRNA regulatory profile (Subgroup 2) would show the opposite pattern. If confirmed prospectively, this means the choice between available immunotherapies — or future therapies targeting these pathways — could be guided by a blood-based transcriptional assay performed at the time of diagnosis. For the teplizumab era specifically, the question this paper immediately raises is: which subgroup does anti-CD3 therapy preferentially benefit, and does the subgroup assignment predict who will gain the full three-year delay in onset versus a more modest response?

The broader picture: The endotype concept in T1D has been discussed for over a decade, primarily in the context of age-at-onset (T1DE1 vs T1DE2) and pancreatic histopathology from nPOD donors. This paper achieves something qualitatively different: an age-independent, blood-accessible, functionally validated endotype classification that predicts real-world therapeutic outcomes across six independent clinical trials. This is not a research tool — it is a prototype precision medicine classifier for T1D immunotherapy.

03 · Four Questions We Will Ask the Speaker

Q1. The plasma-induced transcription bioassay captures the integrated activity of everything in plasma. The subgroup classification is stable over the post-onset period, suggesting it reflects a durable immune state. But does the subgroup assignment precede disease onset — and do you have data from prediabetic TrialNet participants or TEDDY samples showing the same two subgroups emerge before clinical T1D diagnosis?

Q2. Subgroup 1’s apparent response to anti-CD20 and Subgroup 2’s to CTLA4-Ig are striking — but neither is fully statistically significant at the C-peptide rate-of-decline level, and sample sizes within each trial-by-subgroup cell are modest. What prospective study design would provide definitive confirmation — and is there an ongoing or planned trial that will prospectively randomize patients by subgroup assignment?

Q3. The mechanistic hypothesis for Subgroup 1’s resistance to abatacept — that high inflammatory cytokines provide alternative CD28-independent co-stimulatory signals — is elegant. Is there direct evidence from the TrialNet flow cytometry data that CTLA4-Ig fails to reduce T cell activation markers in Subgroup 1 treated patients, even while it successfully reduces central memory T cells in Subgroup 2?

Q4. Subgroup 1’s elevated plasma acetate raises the gut microbiome hypothesis — that dysbiosis and impaired gut barrier function may be driving or amplifying the innate immune activation that defines this subgroup. Is there microbiome data from any of the TrialNet cohorts that could test this, and does the miR-155/Th17 signature in Subgroup 2 suggest a different but equally specific environmental trigger?

04 · Four Key Associated Papers

1. Cabrera SM, Engle S, Kaldunski M, … Hessner MJ (2018) Innate immune activity as a predictor of persistent insulin secretion and association with responsiveness to CTLA4-Ig treatment in recent-onset type 1 diabetes Diabetologia, 61(11):2356–2370 · Read paper → The foundational Hessner Lab paper establishing that plasma-induced transcriptional innate immune activity predicts C-peptide preservation and CTLA4-Ig response — the direct scientific predecessor to today’s two-subgroup discovery.

2. Redondo MJ & Morgan NG (2023) Heterogeneity and endotypes in type 1 diabetes mellitus Nature Reviews Endocrinology, 19(9):542–554 · PubMed → The definitive review of T1D endotyping from tissue-based and clinical perspectives — the conceptual framework that today’s blood-accessible, functionally validated subgroup classification now extends into a clinical tool.

3. Sebastiani G, Grieco GE, Bruttini M et al (2024) A set of circulating microRNAs belonging to the 14q32 chromosome locus identifies two subgroups of individuals with recent-onset type 1 diabetes Cell Reports Medicine, 5(6):101591 · Read paper → The independent miRNA-based T1D subgroup paper — identifying two recent-onset T1D groups with elevated miR-409-3p that overlap with today’s Subgroup 2, providing critical external validation of the miRNA endotype signature.

4. Pant T, Lin CW, Bedrat A, … Hessner MJ (2024) Monocytes in type 1 diabetes families exhibit high cytolytic activity and subset abundances that correlate with clinical progression Science Advances, 10(20):eadn2136 · Read paper → The companion Hessner/Bedrat paper characterizing monocyte biology in T1D families using the same plasma-induced transcription platform — establishing that innate immune dysregulation tracks with T1D progression and contextualizing the Subgroup 1 innate activation signature.

05 · Four Videos to Watch First

▶ Ask the Expert: Martin Hessner, PhD & Pinar Sargin, MD — University of Wisconsin & Wellington Regional Today’s speaker’s PI in a prior TSS appearance — Hessner on the plasma-induced transcription bioassay, the T1D immune heterogeneity program at MCW, and the foundational science underlying today’s two-subgroup discovery. Watch this first.

▶ Ask the Expert: Peter Linsley, PhD — Benaroya Research Institute Linsley works on computational immune biomarker discovery in T1D — the closest methodological neighbor in the TSS library to Bedrat’s plasma transcriptomics approach, and essential context for understanding how blood-based immune signatures are validated across cohorts.

▶ Ask the Expert: Laura Jacobsen, MD — University of Florida Jacobsen’s work on ATG responder identification using immune phenotyping is the most direct clinical parallel to today’s paper — demonstrating that within a single T1D immunotherapy, biological subgroups predict who responds and why. A perfect conceptual companion.

▶ Ask the Expert: Kevan Herold, MD & Matthias Von Herrath, MD — Yale and UCSD Herold and Von Herrath lead T1D immunotherapy trials including teplizumab — the clinical translation destination for the precision patient-stratification approach Bedrat’s subgroup classifier would directly enable.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

Date: Tuesday, August 11, 2026 · 9:00 AM Pacific · 12:00 PM Eastern Format: Free virtual seminar for the global T1D research community — all are welcome.

👉 Register here on Zoom

Citation: https://www.benningerlab.com/research.html Figure 9A

About the Speakers

Jason Bini, PhD is an Assistant Professor of Radiology and Biomedical Imaging at Yale University, where he works within the Yale Biomedical Imaging Institute and the PET Core. He earned his PhD in Biomedical Engineering at The City College of New York (CUNY), with doctoral work focused on quantitative PET of atherosclerotic plaque metabolic activity using simultaneous PET/MR systems, and completed his postdoctoral training at Yale. His research applies PET and MR imaging to investigate mechanisms of disease and receptor/enzyme pharmacology in metabolic diseases including diabetes and obesity. Current projects include using PET to image dopamine signaling in the pancreas in both T1D and T2D, whole-body distribution of the cortisol-activating enzyme 11β-hydroxysteroid dehydrogenase type 1, PET imaging of whole-body inflammation, and method development for whole-body kinetic modeling of novel PET radioligands.

Richard Benninger, PhD is an Associate Professor of Bioengineering and a researcher at the Barbara Davis Center for Diabetes, University of Colorado Anschutz Medical Campus. He received his PhD in Physics from Imperial College London and completed postdoctoral training at Vanderbilt University in Molecular Physiology and Biophysics. His research has focused on applying imaging technologies and biophysical approaches to understand how emergent multicellular dynamics govern insulin and glucagon secretion from the endocrine pancreas. More recently, his lab has pioneered contrast-enhanced ultrasound (CEUS) with nanobubble contrast agents as a non-invasive diagnostic tool for detecting insulitis and monitoring beta cell mass decline in presymptomatic T1D — a platform that also doubles as a therapeutic delivery vehicle for tolerance-inducing peptides. Both speakers are returning TSS guests.

01 · The Problem: You Cannot Cure What You Cannot See

One of the most consequential gaps in T1D medicine is deceptively simple: we cannot directly observe what is happening in the pancreas of a living person with T1D or at risk for it. We can measure autoantibodies in blood and genetic risk scores from saliva — but neither tells us how much beta cell mass remains, whether insulitis is progressing or regressing, or whether a therapy is actually protecting the pancreas right now. The tools that could answer these questions — biopsy, histology, nPOD — require the pancreas to be removed or the donor to be deceased. The result is that T1D prevention trials are designed and evaluated largely blind to what is happening inside the organ they are trying to protect.

Today’s two speakers are approaching this problem from different angles — one with the high-sensitivity molecular specificity of PET, the other with the real-time accessibility and theranostic potential of ultrasound — and together they map the frontier of non-invasive pancreatic imaging in T1D.

Jason Bini’s angle — PET imaging of pancreatic dopamine and metabolism: Bini’s lab at Yale uses PET imaging to investigate mechanisms of disease and receptor/enzyme pharmacology in metabolic diseases, with current projects including using PET to image dopamine in the pancreas in both T1D and T2D. The dopamine connection to the pancreas is less well known than it should be: beta cells synthesize dopamine from L-DOPA, use it to regulate insulin secretion through autocrine/paracrine signaling, and the dopaminergic system in the pancreas is measurably disrupted in T1D and T2D. By developing and validating PET radioligands that report on pancreatic dopamine signaling, Bini’s group is building a molecular window into beta cell physiology that can be monitored longitudinally — tracking disease progression and therapeutic response from outside the body.

Richard Benninger’s angle — ultrasound nanobubbles for insulitis detection: During T1D, the islet microvasculature increases permeability, allowing nanoparticles to access the microenvironment. Contrast-enhanced ultrasound uses shell-stabilized gas bubbles to provide acoustic backscatter in vasculature, and sub-micron sized nanobubble contrast agents can be used to measure increased islet microvasculature permeability and indicate asymptomatic T1D. In Benninger’s Nature Communications 2020 paper, pre-clinical T1D models showed accumulation of nanobubbles specifically within pancreatic islets correlating with insulitis, detected early in disease progression and decreasing with successful therapeutic intervention. The platform has since been extended to phase-change nanodroplets capable of detecting insulitis in mouse models and human islets, and — in the most recent 2026 Theranostics paper — to peptide-loaded nanobubbles that simultaneously image insulitis via CEUS and deliver insulin-reactive tolerogenic peptides to the islet microenvironment to expand islet-resident regulatory T cells and delay diabetes onset.

Tags: PET imaging · contrast-enhanced ultrasound · nanobubbles · insulitis · beta cell mass · dopamine · pancreas imaging · T1D diagnostics · biomarker · theranostics · non-invasive · preclinical · GLP-1 receptor · VMAT2 · islet microvasculature

02 · Why This Matters

For scientists: The imaging tools Bini and Benninger are developing would transform how T1D research is conducted at every level. Right now, measuring the effect of a therapeutic intervention on the beta cell mass or insulitis requires waiting for functional readouts (C-peptide, HbA1c) or endpoint histology. A validated in vivo imaging readout — whether PET-based molecular signaling or CEUS-based vascular permeability — would allow researchers to directly observe whether a therapy is working inside the pancreas in real time, dramatically accelerating the iterative cycle of preclinical and clinical development. Benninger’s nanobubble platform adds a further dimension: the same contrast agent that images insulitis can deliver therapeutic payloads directly to the inflamed islet microenvironment, creating a feedback-guided therapeutic system.

For clinicians: The clinical deployment of teplizumab for Stage 2 T1D has created a new problem: how do you decide when to treat, how long to treat, and whether treatment is working, when you cannot see the organ being treated? A validated non-invasive pancreatic imaging biomarker would change this calculus entirely. Benninger’s CEUS nanobubble signal tracks with insulitis severity and normalizes with successful therapeutic intervention in preclinical models — if translated to humans, it could serve as a real-time treatment response monitor. Bini’s dopamine PET data could provide a complementary readout of beta cell metabolic health, independent of immune activity. Together they would give clinicians a two-dimensional window — immune activity and beta cell function — that currently does not exist.

The broader picture: The pancreatic imaging field for T1D has historically lagged far behind brain and cardiac imaging in tool development, partly because the pancreas is difficult to image (deep, surrounded by bowel, small islet volume relative to total organ size) and partly because T1D has not historically attracted the same imaging investment as oncology or neurology. That is changing rapidly — driven by the urgent need for biomarkers in prevention trials, the maturation of GLP-1 receptor PET tracers for beta cell mass measurement, and now the nanobubble theranostic platform. Today’s talk represents the state of that frontier.

03 · Four Questions We Will Ask the Speakers

Q1 — For Bini: Pancreatic dopamine imaging is an unusual entry point into T1D diagnostics. Walk us through the mechanistic rationale — what does reduced dopaminergic signaling in the pancreas tell us about beta cell function or immune status, and how does the PET signal change between healthy individuals, prediabetic individuals, and those with established T1D?

Q2 — For Benninger: Your nanobubble CEUS signal tracks with insulitis severity in NOD mice and normalizes with successful therapeutic intervention. What is the translational path to human imaging — what regulatory, safety, and sensitivity challenges need to be cleared before CEUS nanobubbles can be used to monitor insulitis in a Stage 1 or Stage 2 T1D clinical trial participant?

Q3 — For both: The current gold standard for beta cell mass measurement in living humans is GLP-1 receptor PET (68Ga-exendin) — but it has shown only modest sensitivity for detecting differences between T1D and healthy controls at the whole-pancreas level. How do your approaches compare in sensitivity and specificity, and is there a case for combining PET-based beta cell mass measurement with CEUS-based insulitis detection as a multimodal T1D staging panel?

Q4 — For Benninger: The 2026 Theranostics paper takes your nanobubble platform from diagnostic to theranostic — imaging insulitis while simultaneously delivering tolerogenic peptides. What is the evidence that the peptide-nanobubble treatment works through the imaging mechanism (islet accumulation via microvasculature permeability) rather than systemic peptide exposure — and how do you design the clinical trial for a combined imaging-therapeutic agent?

04 · Four Key Associated Papers

1. Ciccaglione M, Pham VT, Ramirez DG & Benninger RKP (2021) Detecting insulitis in type 1 diabetes with ultrasound phase-change contrast agents PNAS, 118(41):e2022523118 · Read paper → The landmark PNAS paper demonstrating that ultrasound phase-change nanodroplets detect insulitis in mouse models of T1D through islet microvasculature permeability — the most advanced iteration of the CEUS insulitis imaging platform before the 2026 theranostic extension.

2. Ramirez DG, Ciccaglione M, Pham VT, … Benninger RKP (2020) Contrast-enhanced ultrasound with sub-micron sized contrast agents detects insulitis in mouse models of type 1 diabetes Nature Communications, 11:2238 · Read paper → The original Nature Communications nanobubble paper establishing CEUS insulitis detection — showing nanobubble accumulation in pancreatic islets correlating with insulitis, detected early in disease and reversing with successful therapeutic intervention.

3. Normandin MD, Petersen KF, Bhatt DL, … Bini J (2022) In vivo imaging of endogenous pancreatic beta-cell mass in healthy and type 1 diabetic subjects using [18F]FP-(+)-DTBZ PET Journal of Nuclear Medicine · PubMed → The foundational VMAT2 PET paper establishing [18F]FP-(+)-DTBZ as a radiotracer for in vivo beta cell mass imaging — the methodological precedent for Bini’s dopamine-related PET imaging approach in the pancreas.

4. Andralojc K, Srinivas M, Brom M, … Boerman OC & Gotthardt M (2012) Obstacles on the way to the clinical visualisation of beta cells: looking for the Aeneas of the beta cell imaging world Diabetologia, 55(5):1247–1257 · PubMed → The definitive review of the challenges in beta cell imaging — covering PET tracers, MRI approaches, and the biological obstacles (low islet volume, deep location, off-target binding) that Bini’s and Benninger’s platforms are designed to overcome.

05 · Four Videos to Watch First

▶ State of the Science: Bioimaging in Human T1D — Challenges, Progress, and the Roadmap to Clarity The essential TSS primer on the pancreatic imaging field — covers the full landscape of PET, MRI, ultrasound, and optical approaches to T1D diagnostics. Watch this first to understand the context both speakers are operating in.

▶ Ask the Expert: Jason Bini, PhD — Yale Today’s co-speaker in his prior TSS appearance — Bini on PET imaging of dopamine, cortisol, and metabolic signaling in the diabetic pancreas. Essential background before today’s talk.

▶ Ask the Expert: Richard Benninger, PhD — University of Colorado Boulder Today’s co-speaker in his prior TSS appearance — Benninger on islet imaging, calcium dynamics, and the biophysical approaches that underpin his ultrasound nanobubble platform.

▶ Ask the Expert: Matthew Merrins, PhD & Richard Benninger, PhD — 3D Light-Sheet Imaging of the Islet Benninger’s second TSS appearance, focused on 3D light-sheet imaging of islet calcium dynamics and multicellular coordination — the functional imaging foundation that complements the structural/diagnostic imaging work in today’s talk.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

Beta Cells in Type 1 Diabetes: Murder or Suicide?

Revisiting the Critical Question in T1D Pathogenesis

TheSugarScience T1D Th1nk Tank · 60-Minute Debate including 15-Minute Q&A

📅 Registration

Format: Free virtual seminar for the global T1D research community — all are welcome.

👉 Register here on Zoom

👤 The Debaters

Team Murder — The Immune System as Executioner

Remi Creusot, PhD — Associate Professor, Columbia Center for Translational Immunology & Naomi Berrie Diabetes Center, Columbia University Irving Medical Center. His case: HLA-DR3/DR4-DQ8 alleles confer the single largest genetic risk for T1D outside the MHC by shaping peptide-binding repertoire and enabling autoreactive T cells to escape thymic deletion. The INS-VNTR class I allele reduces thymic insulin expression — a causal immune tolerance defect acting upstream of any beta cell contact. Soluble antigen arrays delivering proinsulin peptides protect NOD mice from T1D by restoring tolerance: proof that the therapeutic leverage point is entirely in the immune compartment.

Andrea Schietinger, PhD — Investigator, Immunology Program, Memorial Sloan Kettering Cancer Center / Weill Cornell Medicine. Her case: a stem-like autoreactive CD8⁺ T cell population (Nature, 2022) initiates and sustains beta cell destruction in NOD mice by continuously replenishing effector autoreactive T cells from a self-renewing progenitor — a property entirely intrinsic to the immune compartment. Critically, unlike tumor-reactive T cells that exhaust, autoreactive T cells in T1D do NOT become dysfunctional. Their epigenetic programs are fundamentally different from exhaustion — making them permanently dangerous without immune intervention.

Team Suicide — The Beta Cell as Complicit in Its Own Demise

Shuibing Chen, PhD — Kilts Family Professor of Surgery; Director, Center for Genomic Health, Hartman Institute, Weill Cornell Medicine. Her case: hiPSC-derived beta cells carrying GWAS-risk knock-ins for GLIS3, CLEC16A, and INS-VNTR class I demonstrate cell-autonomous functional failure, impaired stress response, and increased apoptosis — independent of any immune effector. In vascularized macrophage-islet organoids (Cell Stem Cell, 2024), SARS-CoV-2 drives beta cell pyroptosis via macrophage activation — but the beta cell’s own ACE2/FURIN/NRP1 expression determines viral tropism. The beta cell sets up its own destruction. ML analysis of the HPAP database reveals beta cell stress signatures upstream of immune infiltration.

Shahin Rafii, MD — Arthur B. Belfer Professor; Director, Hartman Institute for Therapeutic Organ Regeneration & Ansary Stem Cell Institute, Weill Cornell Medicine. His case: islet-specific endothelial cells (ISECs) provide critical angiocrine survival and regeneration signals to beta cells. When the ISEC niche is disrupted, beta cell mass fails — not from immune attack, but from loss of vascular support. R-VEC co-transplantation reverses diabetes without immune manipulation, demonstrating that restoring the vascular niche is sufficient to rescue beta cell function.

Moderator: Anna Ava Dattoli PhD, Weill Cornell Medical. She is a senior T1D clinician-scientist with no prior published position on the debate question.

01 · The Debate

The question traces directly to a 1986 lecture by Gian Franco Bottazzo — the immunologist who first described insulitis in human T1D pancreata — titled “Death of a Beta Cell: Homicide or Suicide?” It was formally revisited in 2011 by Atkinson, Bluestone, Eisenbarth, Hebrok, Herold, Accili, Von Herrath and colleagues in Diabetes — and after 15 more years of evidence, remains unresolved. This debate asks four New York City scientists to argue their assigned position forcefully, with data. The goal is not a verdict. It is a sharper understanding of the causal hierarchy that must be resolved to deliver durable cures.

The central question: Is beta cell destruction in T1D primarily a crime committed upon an innocent bystander — murder by immune dysregulation — or does the beta cell carry intrinsic genetic, metabolic, and structural vulnerabilities that make it complicit in its own demise — suicide by fragility?

The murder case in brief: Without HLA susceptibility alleles enabling escape of autoreactive T cells from thymic negative selection, and without a self-sustaining stem-like autoreactive T cell pool that does not exhaust, there is no disease. Teplizumab delays T1D by suppressing autoreactive T cells — the strongest proof of causal primacy available. The immune system is the executioner.

The suicide case in brief: The identical twin discordance in T1D — same immune genetics, different outcomes — demands a non-immune explanation. Beta cells carry GWAS-encoded vulnerabilities (GLIS3, CLEC16A, INS-VNTR) that produce cell-autonomous failure. They express viral entry receptors (ACE2, FURIN, NRP1) that make them uniquely susceptible to direct viral destruction. HPAP multi-omic data shows beta cell stress signatures that appear upstream of immune infiltration. Teplizumab delays but does not cure T1D — because the underlying beta cell fragility is not addressed.

Where the debate gets complicated: The COVID-19 data may be the most destabilizing evidence yet for both sides. SARS-CoV-2 appears to trigger both direct beta cell pyroptosis and autoimmunity simultaneously — through a beta cell-intrinsic viral entry program that activates a macrophage-mediated immune cascade. This is neither pure murder nor pure suicide. It is complicity. And it may be the model for how T1D begins in every case.

Tags: beta cell · T1D pathogenesis · immune tolerance · autoreactive T cells · beta cell stress · GLIS3 · CLEC16A · HLA · stem-like T cells · T cell exhaustion · islet endothelial cells · angiocrine niche · hiPSC islets · COVID-19 · nPOD · HPAP · murder · suicide · debate

02 · Why This Matters

For scientists: The causal hierarchy question is not academic — it determines where you aim your therapeutic intervention. If the immune system initiates and the beta cell merely receives, then immunotherapy is the primary target and beta cell protection is supportive. If beta cell fragility generates the immunogenic signals that prime the immune response, then no immunotherapy will produce a durable cure without also addressing the underlying beta cell vulnerability. The HPAP and nPOD databases are currently being mined by ML approaches that are — right now — generating the temporal data to answer this question empirically. Tonight’s debate is the field thinking out loud about what that answer might be before the data arrives.

For clinicians: Teplizumab delays T1D onset by approximately three years. It does not prevent T1D permanently. If the immune system is the sole cause, why doesn’t immunotherapy cure the disease? The suicide team’s answer — that the underlying beta cell fragility eventually overwhelms immune protection — has direct implications for how we counsel Stage 2 T1D patients receiving teplizumab. Are we buying time while doing nothing about the condition that made the time-buying necessary? And for clinicians treating COVID-associated new-onset T1D, the Chen organoid data raises a pressing mechanistic question: is that a beta cell disease, an immune disease, or both simultaneously?

The broader picture: More than 90% of T1D genetic risk loci are linked to immune function. But the ones linked to beta cell biology — GLIS3, CLEC16A, INS-VNTR — are among the most mechanistically studied and the most actionable for direct beta cell engineering. The endotype hypothesis — that different patients have T1D for different primary reasons — may be the resolution the field has been circling: some patients are murdered, some are suicidal, and the therapeutic implications differ. Tonight’s debate is a stress test of that hypothesis.

03 · Six Questions the Debate Will Turn On

Opening (pre-vote): Before the first panelist speaks, the audience votes: Murder or Suicide? The post-debate shift in that vote is the night’s most honest data point.

Q1 — The teplizumab challenge (to Murder team): Teplizumab delays T1D by ~3 years but does not prevent it permanently. If immune dysregulation is the sole cause, why doesn’t immunotherapy cure the disease? Does the beta cell’s intrinsic vulnerability eventually overwhelm immune protection?

Q2 — The twin challenge (to Suicide team): If intrinsic beta cell fragility is primary, why don’t all individuals with high-risk HLA alleles and viral exposure develop T1D? What is the rate-limiting step — and isn’t it the immune response?

Q3 — The COVID challenge (to all): SARS-CoV-2 appears to trigger both direct beta cell pyroptosis and autoimmunity simultaneously. Does COVID-driven T1D break the murder-or-suicide dichotomy — and if so, what does it replace it with?

Q4 — The alpha cell challenge (to Suicide team): If beta cell-intrinsic stress is the primary driver, why are alpha cells and delta cells in the same islet not destroyed to the same extent? Does the selectivity of immune targeting argue more strongly for murder than suicide?

Q5 — The HPAP empirical question (to all): ML clustering of HPAP data is mapping early beta cell stress signatures relative to immune infiltration. What single finding from those databases — a temporal relationship, a cell type, a gene signature — would definitively settle this debate for you?

Q6 — The restoration challenge (cross-examination): To Chen/Rafii: If you restored beta cell mass and vascular niche function in a T1D patient without immune intervention, would autoimmunity recur? To Creusot/Schietinger: If thymic selection were perfect and all autoreactive T cells were deleted, would beta cells with high-risk genetic backgrounds still fail?

04 · Four Key Associated Papers

1. Atkinson MA, Bluestone JA, Eisenbarth GS, Hebrok M, Herold KC, Accili D, … Von Herrath M (2011) How does type 1 diabetes develop?: the notion of homicide or β-cell suicide revisited Diabetes, 60(5):1370–1379 · Read paper → The direct intellectual ancestor of tonight’s debate — the 2011 revisitation of Bottazzo’s 1986 question by the field’s leading authorities. Required reading before the panel.

2. Gearty SV, Dündar F, Zumbo P, … Schietinger A (2022) An autoimmune stem-like CD8 T cell population drives type 1 diabetes Nature, 602(7895):156–161 · PubMed → Schietinger’s landmark paper identifying the self-renewing stem-like autoreactive CD8⁺ T cell population that sustains beta cell destruction in NOD mice — the murder team’s most powerful mechanistic argument.

3. Roep BO, Thomaidou S, van Tienhoven R & Zaldumbide A (2021) Type 1 diabetes mellitus as a disease of the β-cell (do not blame the immune system?) Nature Reviews Endocrinology, 17:150–161 · PubMed → The suicide team’s most provocative position paper — arguing that T1D is fundamentally a disease of the beta cell, with the immune system responding to stress signals the beta cell generates. The rebuttal target for the murder team.

4. Robertson CC, Elgamal RM, Henry-Kanarek BA, Arvan P, Chen S, … Soleimanpour SA (2024) Untangling the genetics of beta cell dysfunction and death in type 1 diabetes Molecular Metabolism, 86:101973 · PMC → The definitive human genetic evidence review — mapping GWAS loci including GLIS3, CLEC16A, and INS-VNTR to beta cell regulatory elements and establishing the genomic case for intrinsic beta cell vulnerability.

05 · Four Videos to Watch First

▶ T1D Th1nk Tank: Shuibing Chen, PhD & Stephen Parker, PhD — Weill Cornell and University of Michigan Tonight’s Team Suicide lead in a prior TSS appearance — Chen on single-cell omics, hiPSC-derived islets, and T1D-associated regulatory networks within beta cells. Essential context for her GLIS3/CLEC16A/ACE2 arguments tonight.

▶ Ask the Expert: Andrea Schietinger, PhD — Memorial Sloan Kettering Cancer Center Tonight’s Team Murder co-lead in her TSS appearance — Schietinger on T cell exhaustion, epigenetic programming, and why autoreactive T cells in T1D do not exhaust the way tumor-reactive T cells do.

▶ Ask the Expert: Remi Creusot, PhD — Columbia University Tonight’s Team Murder lead in his TSS appearance — Creusot on antigen presentation, thymic tolerance failure, and precision immunotherapy approaches in T1D.

▶ T1D Th1nk Tank: Mollie Huber, PhD — University of Florida Huber’s live pancreas slice data showing beta cell dysfunction equally impaired in T cell-infiltrated and non-infiltrated islets is the hottest recent evidence for the suicide case — and the paper the murder team will need to explain away tonight.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org

Paper: https://www.sciencedirect.com/science/article/pii/S2212877825000900?via%3Dihub

📅 Registration

Date: Tuesday, August 4, 2026 · 12:00 PM Eastern Format: Free virtual seminar for the global T1D research community — clinicians and scientists are welcome.

👉 Register here on Zoom

👤 About the Speaker

Jonathan Belin is a PhD student in the laboratory of Prof. Yuval Dor at the Department of Developmental Biology and Cancer Research, Institute for Medical Research Israel-Canada, Faculty of Medicine at the Hebrew University of Jerusalem. He grew up in Jerusalem, where he decided to study biomedical sciences at the Hebrew University of Jerusalem, graduating with highest honors before entering the university’s Abisch-Frenkel excellence program for direct-track PhD studies.

His research sits at an unusual and productive intersection: he asks not just how the innate immune system responds to beta cell stress, but whether the adaptive immune system — the T and B cells classically considered the executioners of T1D — is actually required for disease to unfold in the first place. Working with mouse models in which the RNA-editing enzyme ADAR is selectively deleted in beta cells, he investigates how a beta cell’s failure to manage its own endogenous double-stranded RNA triggers islet inflammation and whether that process can proceed to frank diabetes without the involvement of adaptive immune cells at all.

His mentor, Prof. Yuval Dor, is one of the most creative and prolific beta cell biologists working today. The Dor lab at Hebrew University has made seminal contributions to understanding beta cell mass dynamics, the origins of new beta cells, and most recently, the mechanism by which disrupted RNA editing in beta cells triggers an innate antiviral response that closely mimics early-stage T1D — without any virus required. That foundational work, published in Cell Metabolismin January 2024, established the experimental system Jonathan Belin is now using to ask the next question: what does adaptive immunity actually contribute to this disease, and when?

01 · Background: The Model and the Question

The canonical story of T1D is a T cell story. Autoreactive CD4⁺ and CD8⁺ T cells — primed against beta cell antigens, infiltrating the islets, and eventually destroying insulin-producing cells — are the accepted protagonists of disease. This understanding underlies every clinical immunotherapy program in T1D, from teplizumab (anti-CD3) to a growing roster of antigen-specific and regulatory T cell approaches. The adaptive immune system, in this framework, is not merely involved — it is the disease.

But there is a question the field has not fully resolved: when does the adaptive immune system become essential, and what happens before it arrives?

The Dor lab’s ADAR-deficiency model has opened a window onto precisely this question. When ADAR — the enzyme responsible for adenosine-to-inosine editing of double-stranded RNA structures generated from inverted retroelements in the beta cell genome — is selectively deleted in beta cells, the unedited endogenous dsRNA accumulates and is sensed by MDA5/IFIH1, the same cytosolic dsRNA sensor that recognizes viral RNA. The result is a massive type I interferon response, islet inflammation with extensive immune infiltration, beta cell failure, and diabetes — all without any viral infection. Critically, the model shares striking hallmarks with early-stage human T1D: an interferon transcriptional signature in islets, HLA class I upregulation, beta cell dedifferentiation and loss, and selective sparing of alpha cells.

What this model has not yet resolved is the causal role of adaptive immunity in disease. The inflammatory infiltrate in βAdarKO mice contains both innate and adaptive immune cells. But are the T and B cells required to initiate or execute disease in this setting, or is the innate response — the interferon storm, the recruitment of innate immune effectors — sufficient to destroy beta cells on its own? This is the question Jonathan Belin’s talk addresses, and it goes to the heart of how we understand T1D initiation.

The key findings expected from his work:

Belin’s research uses genetic tools to deplete or disable adaptive immune cells in the βAdarKO background, asking whether mice lacking T cells, B cells, or both still develop insulitis and diabetes when ADAR is knocked out in beta cells. The anticipated findings — based on the framing of his talk title and his published contributions to the Dor lab — suggest that innate mechanisms may be more autonomous in early beta cell damage than the field has assumed. If adaptive immunity is dispensable for the initial phases of disease in this model, it would mean the innate inflammatory cascade triggered by endogenous dsRNA can cause substantial beta cell damage without any help from the T cell–mediated autoimmunity classically considered central to T1D. This would challenge the prevailing model and suggest that innate-directed interventions — not just immunotherapies targeting T cells — deserve serious attention as prevention strategies in early or pre-clinical T1D.

The alpha cell sparing effect adds a further dimension. Prior work from the lab (Peleg, Zamashanski, Belin et al., Molecular Metabolism, 2025) demonstrated that ADAR deletion in alpha cells does not trigger the same inflammatory cascade — alpha cells are far less sensitive to RNA editing deficiency and to cytokine-mediated killing than beta cells. That differential vulnerability provides a molecularly grounded explanation for one of T1D’s most puzzling features: why the immune system appears to selectively destroy beta cells while alpha cells in the same islet are spared. Whether adaptive immune cells sharpen that selectivity — or whether it emerges entirely from beta cell-intrinsic biology — is a question Belin’s work is positioned to answer.

02 · Why This Matters

For scientists: The innate versus adaptive immunity question in T1D is not an academic exercise. It has direct implications for how we design experiments, interpret mouse models, and think about the temporal sequence of events in human disease. If the innate interferon response triggered by endogenous dsRNA can damage beta cells substantially before adaptive immune cells are required, then studies examining T1D pathogenesis exclusively through the lens of autoreactive T cells may be systematically missing an earlier causal window. The βAdarKO model offers a temporally clean system to dissect this — ADAR deletion can be induced acutely, the innate response follows within days, and the timing of adaptive immune involvement can be precisely probed. Understanding when adaptive immunity becomes required (rather than assuming it is always required) would reshape how we stage T1D pathogenesis and where we place the therapeutic window for prevention.

The differential immunogenicity of alpha versus beta cells in the ADAR-deficiency setting also has mechanistic implications beyond T1D. Why do beta cells generate more endogenous dsRNA under RNA editing deficiency? Is it the volume of retroelement expression, the particular pattern of B1/B2 repeats in beta cell transcriptomes, the heightened metabolic activity, or some combination of these? Understanding the molecular basis of this susceptibility difference could reveal beta cell-specific vulnerabilities that extend to other forms of cellular stress — including those that might precede autoimmunity in human T1D.

For clinicians: The translational stakes here are high. If innate immunity is sufficient to initiate meaningful beta cell damage in this model, it suggests that the conventional immunotherapy window — targeting autoreactive T cells near or after clinical diagnosis — may be missing the earliest events. Teplizumab, which has demonstrated the first-ever ability to delay T1D onset by modulating adaptive immune cells in Stage 2 disease, works at a point when adaptive immunity is already established. But could earlier interventions — targeting the innate interferon response, JAK/STAT signaling, or endogenous dsRNA sensing — protect beta cells before adaptive autoimmunity is even engaged? This talk won’t answer that question definitively, but it will provide the foundational experimental evidence needed to take it seriously.

The broader picture: The virologic hypothesis for T1D has been pursued for decades, supported by the interferon signature in early islets, the associations between enterovirus seropositivity and T1D risk, and the protective GWAS signals around IFIH1 and IFTH1. The Dor lab’s work offers a compelling alternative framing: that the interferon signature in early T1D does not require a virus — it requires only a beta cell under sufficient stress to allow its endogenous dsRNA to escape editing and trigger the same innate alarm. Jonathan Belin’s contribution is to ask what happens next: does the immune system’s adaptive arm amplify a pre-existing innate fire, or is it responsible for starting it?

03 · Four Questions We Will Ask the Speaker

Q1. In the βAdarKO model, the inflammatory infiltrate contains both innate and adaptive immune cells. When you deplete or disable adaptive immunity, what is the residual disease phenotype — is there still insulitis, beta cell loss, and metabolic dysfunction, and how does it compare in severity and kinetics to immunocompetent controls?

Q2. The Dor lab has shown that glycolysis and calcium signaling amplify the interferon response in ADAR-deficient beta cells, suggesting a vicious cycle where increased beta cell workload drives more inflammation. Does disabling adaptive immunity interrupt this cycle, or does the innate arm sustain it independently?

Q3. Alpha cells sharing the same islet microenvironment as ADAR-deficient beta cells do not develop the same inflammatory phenotype, even when ADAR is deleted in them as well. How does your work inform whether this differential vulnerability is truly cell-intrinsic — and does the presence or absence of adaptive immune cells change the fate of alpha cells in the inflamed islet?

Q4. If innate mechanisms are sufficient for early beta cell damage in this model, what does that imply for the therapeutic window in human T1D? Is there an innate-targeted intervention that you think is most plausible, and what would need to be true about the human disease for it to be testable clinically?

04 · Four Key Associated Papers

1. Knebel UE, Peleg S, Dai C, Cohen-Fultheim R, Jonsson S, Poznyak K, Israeli M, Zamashanski L, Glaser B, Levanon EY, Powers AC, Klochendler A & Dor Y (2024) Disrupted RNA editing in beta cells mimics early-stage type 1 diabetes · Cell Metabolism, 36(1):48–61 · READ HERE

The foundational paper establishing the βAdarKO model. ADAR deletion in beta cells triggers a massive MDA5-dependent interferon response, islet infiltration, and beta cell failure with features strikingly similar to early-stage human T1D — all without viral infection. Jonathan Belin’s talk builds directly on this experimental system, asking what happens to the disease phenotype when adaptive immune cells are removed from the equation.

2. Peleg S, Zamashanski L, Belin J, Novoselsky R, Cohen-Fultheim R, Knebel UE, Glaser B, Itzkovitz S, Kaestner KH, Powers AC, Levanon EY, Klochendler A & Dor Y (2025) RNA editing deficiency models differential immunogenicity of pancreatic α- and β-cells · Molecular Metabolism, 98:102183 · READ HERE

Jonathan Belin is a co-author on this paper, which directly addresses a central mystery of T1D: why beta cells are destroyed while neighboring alpha cells survive. Mosaic ADAR deletion in beta cells triggers severe inflammation and loss of wild-type bystander beta cells, while ADAR deletion in alpha cells produces only a mild interferon elevation with no inflammation or metabolic phenotype. The paper establishes molecularly why beta cells are uniquely vulnerable — a finding essential for interpreting the adaptive immunity question Belin addresses in this talk.

3. Szymczak F, Cohen-Fultheim R, Thomaidou S, Coomans de Brachène A, Castela A, Colli M, Marchetti P, Levanon E, Eizirik D & Zaldumbide A (2022) ADAR1-dependent editing regulates human β cell transcriptome diversity during inflammation · Frontiers in Endocrinology, 13:1058345 · READ HERE

A complementary human-islet study from the Eizirik and Levanon groups, showing that ADAR1-mediated RNA editing in human beta cells is modulated by inflammatory cytokines and that editing deficiency alters the beta cell transcriptome in ways relevant to T1D pathogenesis. This paper situates the Dor lab’s mouse findings in the context of human beta cell biology and directly links the ADAR-editing axis to the cytokine environment of insulitis.

4. Eizirik DL, Szymczak F & Mallone R (2023) Why does the immune system destroy pancreatic β-cells but not α-cells in type 1 diabetes? · Nature Reviews Endocrinology, 19(7):425–434 · READ HERE

A major conceptual review directly relevant to the α/β differential vulnerability question Belin’s work addresses. Eizirik and colleagues synthesize the evidence that beta cells are intrinsically more immunogenic — expressing more MHC class I, more autoantigens, more ER stress markers, and a more pronounced interferon response — than alpha cells. This paper provides the human disease context and the mechanistic framework within which the differential ADAR sensitivity Belin’s lab has documented should be interpreted.

05 · Four Videos to Watch First

▶ The Search for Diabetes-Specific T-cell Repertoires — Michel Edwar Mickael, PhD TheSugarScience T1D Th1nk Tank, January 2026 WATCH HERE This recent TSS Th1nk Tank talk on the search for diabetes-specific T cell receptor repertoires is the most direct contextual primer for Jonathan Belin’s question about adaptive immunity. If you want to understand what the adaptive immune arm is doing in T1D — and why identifying its specific role is both scientifically and therapeutically important — start here. Belin’s work asks whether that arm is required for disease initiation; Mickael’s work illuminates what T cell-mediated disease looks like when it is.

▶ What Would the Earliest Detection of Type 1 Diabetes Look Like? — Expert Panel TheSugarScience T1D Th1nk Tank Expert Panel WATCH HERE A TSS expert panel discussion on the earliest detection of T1D, featuring leading researchers discussing what biological events precede autoantibody appearance and clinical onset. Essential viewing for contextualizing Belin’s work: if innate immune events driven by endogenous dsRNA can damage beta cells before adaptive autoimmunity is engaged, the earliest detection window may need to be pushed even earlier than current autoantibody screening captures.

▶ WAVE T1D: Multicenter RCT Combo Therapy in New-Onset Type 1 Diabetes — Clinical Trial Th1nk TankTheSugarScience Clinical Trial Th1nk Tank, February 2026 WATCH HERE A deep look at combination immunotherapy for new-onset T1D, targeting the adaptive immune system at the clinical stage of disease. Watching this alongside Belin’s talk illuminates the implicit question his science raises: are we intervening at the right moment in the disease cascade, or is the adaptive immune contribution arriving after innate-driven damage has already set the stage? The clinical stakes of the innate vs. adaptive question become tangible here.

▶ Heard on the Street: Dr. Yuval Dror at Hebrew University — nPOD 2024 Scientific Meeting TheSugarScience Podcast / nPOD 2024 FIND IN TSS LIBRARY A brief “Heard on the Street” recording from the nPOD 2024 scientific meeting featuring a researcher from the Hebrew University sharing current work — a direct window into the Dor lab’s engagement with the international T1D community. This brief clip situates Jonathan Belin’s work within the live scientific conversation happening at the intersection of islet biology and T1D pathogenesis, and illustrates the kind of rapid-cycle translational science the Dor lab is contributing.

TheSugarScience · Expediting a cure for T1D by curating the scientific conversation · thesugarscience.org