In this post I will summarise the FLBx landscape and then take a deeper-dive into Aqtual’s technology and the data they have shared publicly.

Part I: The Functional Liquid Biopsy Landscape

1.1 What Is Functional Liquid Biopsy?

Traditional ctDNA liquid biopsy tells us if cancer is present, how to treat it, and, is the tratment working. It can be used as a cancer genome surveillance tool detecting somatic mutations, copy number alterations, or, tumour-derived methylation patterns and we use it in assays designed for therapy selection, tumor response monitoring and MRD (and early cancer detection). What it cannot do is tell you what the tumour is actually doing, which genes are switched on, how the immune microenvironment is organised, or whether a resistance programme is being activated. Functional Liquid Biopsy (FLBx) is the term I coined for approaches that go beyond sequence or methylation to read the regulatory and transcriptional state of tumour and immune cells from a blood draw.

The biological basis is straightforward: when cells die, the DNA they release into circulation retains, however transiently, the epigenetic imprint of the parent cell. Active regulatory regions are packaged differently from silent ones. Transcription factors, RNA polymerase, and histone modifications all leave physical and chemical marks on the DNA and on the nucleosome complexes that wrap it. If you can capture and decode those signatures from plasma, you are reading, in effect, a non-invasive transcriptome of the cells shedding into the blood.

The field is early in development, but maturing fast. In my initial FLBx posts (one was aqtually my first Substack blog) I described the basic premise and the various technological approaches, which are now clearly differentiated:

• Antibody-based active Chromatin Immuno-Precipitation (cfChIP-seq) used by Precede Biosciences and Senseera.

• Antibody-free physical chromatin enrichment (cfDNAac) used by Aqtual.

• Methylation analysis and gene expression inference, used by LiquidCell Dx to identify TME state and predict immuno-therapy response by spatial ecotypes.

• WGS fragmentomics and gene expression inference, the Triton/Proteus approach from Fred Hutch (Robert Patton & Gavin Ha).

A word of caution that cuts across many/all FLBx approaches, and which I should have stated more plainly in my original post: these assays, particularly those relying on enriching a tumour-origin signal, perform best at elevated tumour fraction. They are most powerful in the metastatic, high-tumour-burden setting where tissue is old, unrepresentative, or unavailable, something that many clinical triallists are challenged by, and precisely the clinical scenario where a functional readout could change decisions. I hope they can be improved to perform well in the earlier disease setting.

1.2 The FLBx Landscape Summary Table

Part II: Aqtual Technology, How It Works, How It Differs, and Where the Edge Lies

2.1 The Core Biological Insight

Functional liquid biopsies are interrogating the tumour-state, and all rely on inferring gene expression (I’ll come back to Resero.bio in a later post), from the regulatory information available in plasma via the cfNucleosome or the cfDNA itself. Distilling the idea to it’s simplest form all of these companies look at the active regulatory regions, the open chromatin, where the promoters and enhancers of expressed genes sit. These regions are physically distinct from the inactive genome: the DNA is either unspooled from nucleosomes, or bound by large multi-protein complexes including transcription factors, co-activators, and RNA Polymerase II. Whereas silenced regions are wrapped tightly into compact nucleosomes, protected from the transcriptional machinery. Here endeth the lesson so-to-speak.

Aqtual’s founding insight was that the regions of the genome that are regulatory active shed the larger fragments of cfDNA. Their preferentially enrichment of these regions is akin to other Reduced Representation methods, i.e. it’s genome wide, but you only have to sequence a fraction of the genome to get the useful signal.

Where off-the-shelf cfDNA extraction kits are optimised to capture the short, canonical 167bp fragments, Aqtual’s proprietary cfDNA extraction enriches for the longer active chromatin (cfDNAac), which contains the most biologically meaningful regulatory information.

2.2 The cfDNAac Workflow

The Aqtual platform, as described in Lai et al. Extracting regulatory active chromatin footprint from cell-free DNA; Communications Biology, 2024, operates as follows:

Controlled pre-analytics: This is now pretty standardised across the industry as we know that poor collection and/or handling of blood for plasma liquid biopsy can severely limit downstream analysis. Aqtual used Streck BCT, but there are many proprietary tubes and many researchers stick to double-spun EDTA (follow the protocol and all will be well). However, the current cfDNA LBx protocols were developed to maximise enrichment of the canonical 167bp fragments, and NOT for FLBx. As such, additional optimisation of the collection conditions such as introducing mild cross-linking agents to maintain the integrity of the active chromatin protein-DNA complexes from blood draw to bench might be the focus of future research.

Chromatin capture (the proprietary wet-lab step): This is the heart of the platform and the key trade secret for Aqtual. Plasma cfDNA is processed through a modified magnetic nanoparticle-based extraction using optimised lysis and binding chemistry. Aqtual’s workflow adjusts binding and lysis conditions to selectively bind the larger, physically more “open” chromatin complexes, exploiting the electrostatic and steric properties of the protein-DNA complexes, while allowing the tight, nucleosomal fragments to be washed away.

The result is that whilst the off-the-shelf kit from Apostle Bio produces a size distribution with 77% of fragments under 200 bp, i.e. dominated by the ~167 bp mono-nucleosome cfDNA. Aqtual’s chromatin capture workflow produces a strikingly different distribution where only ~8% of fragments under 200 bp, such that the distribution is shifted to substantially longer fragments, with the mono-nucleosomal peak appearing broader and the fragment distribution enriched for protein-protected active chromatin complexes (see Lai’24 Fig. 1).

Sequencing and mapping: The enriched cfDNAac library is sequenced by standard next-generation sequencing. Because the pre-sequencing enrichment step has already removed most of the uninformative nucleosomal DNA, the 500M-ish reads (~30x WGS) depth that Aqtual use results deep genome-wide regulatory profiling. Every read that reaches the sequencer is pretty likely to map to a functionally relevant active regulatory region, likely achieving 30-50x coverage at regulatory loci.

Computational deconvolution: Approximately 52% of fragments in the Aqtual workflow are classified as cfDNAac through an “unsupervised clustering of the average [count] profile” across different fragment size bins. This step is critical: not every longer fragment is active chromatin, some nucleosomal DNA will also be captured at the longer end. Aqtual separates the cfDNAac signal from residual cfDNAnuc background using fragment size distributions and count profiles, generating two distinct fractions without requiring any prior knowledge of which genes should be active. The cfDNAnuc fraction serves as an internal baseline control, used computationally to normalise and subtract the inactive background signal. Cool. But way outside my area of expertise!

Mapping and feature engineering: Reads are mapped to the reference genome and read pileups identify active promoters, enhancers, transcription factor binding sites, and gene bodies. The final analytical output is a per-gene regulatory activity profile across multiple feature classes in a single assay. In the most recent Lopes et al. study (npj Precision Oncology, 2026), 1,570 features were selected across these classes (305 promoters, 613 exons, 95 enhancers, 557 TFBS) for leiomyosarcoma clinical benefit prediction.

2.3 What Aqtual Actually Measure and Why It’s Unusual

The validation data from Lai et al. (2024) are worth dwelling on, because they establish exactly what signal cfDNAac is capturing: Gene body cfDNAac vs GTEx whole blood expression. r = 0.95**, i.e. the cfDNAac signal from gene bodies is an extraordinarily accurate proxy for gene expression levels in the contributing cell types.

And for Promoters, Enhancers (H3K27ac) and RNAPolII, cfDNAac vs GTEx expression were r = 0.89, 0.73 & 0.67 resepctivlely, i.e. active promoters map faithfully to expressed genes, the platform captures enhancer activity comparable to H3K27ac ChIP, and Pol II occupancy at promoters is reflected in cfDNAac signal. This means a single cfDNAac assay effectively substitutes for three separate ChIP assays: H3K4me3 (active promoters), H3K27ac (active enhancers), and Pol II ChIP (transcriptional activity).

This antibody-free approach is a clear technical distinction in the FLBx field and one that I originally misunderstood. Least-of-all, but still important, it removes the challenges of antibody lot-to-lot variability.

2.4 How cfDNAac Differs from Computational Fragmentomics

The Patton et al. preprint represents the most direct scientific comparison to Aqtual’s approach because both are attempting to infer regulatory/transcriptional state from cfDNA. However, there are key differences in: Sample input; Patton use conventionally extracted cfDNA whilst Aqtual use their proprietary cfDNAac chemistry. Sequencing depth is not too dissimilar; 30–120x WGS for Patton versus ~30x for Aqtual (but deep at regulatory loci). But the inference models are different; either from nucleosome positioning patterns and fragment length distributions for Patton versus directly sequenced enriched active chromatin fragments for Aqtual. There are also differences in the data requirements; Patton et al can work with pretty standard WGS, i.e. no new assay required, whislt Aqtual requires their proprietary extraction and library preparation protocol.

The critical distinction is that cfDNAac directly enriches the signal of interest before sequencing. Proteus/Triton is performing pattern recognition on a highly diluted signal within the full genomic background. It is an impressive deep learning achievement, but one that may face signal-to-noise constraints at lower tumour fractions.

This also has a direct consequence for the ability to read immune and stromal cell biology. The immune and stromal cells in the blood and tumour microenvironment shed their own cfDNA at low tumour fractions, in fact, at most stages of cancer, the majority of cfDNA in plasma originates from haematopoietic cells and the tumour stroma, not the tumour cells themselves. Computational fragmentomics tools are primarily designed to deconvolute tumour signal from this background. Aqtual’s platform, by contrast, treats that immune/stromal signal not as noise but as a target: the cfDNAac from T-cells, B-cells, macrophages, and fibroblasts directly tells you what those cells are doing in the patient’s body at the time of blood draw. This is arguably the most important feature for immunotherapy monitoring.

Some caveats before looking at the clnical data: Aqtual is still a small company with limited published clinical oncology data. The LMS DAPPER study (n = 30) is compelling but small. Aqtual’s commercial focus to date has been autoimmune (PRIMA-102), and the oncology pivot is recent. Precede Biosciences has a significantly larger published oncology dataset and more established pharma partnerships. LiquidCell have a clearer focus, and a fantstic foundational paper. Patton et al may eb the future. The question is not whether Aqtual’s technology is superior but whether the various analytical niches of Functional Liquid Biopsy, e.g. ADC expression prediction, tumor subtype or gene expression state, TME, etc, can be captured by a single assay or whether the competition is on to find the company that will dominate this new space. Finaly, Aqtual does not have a cost advantage over simple deep WGS with the current methods, we wait to see what might be possible with a targeted Aqtual prep.

Part III: Clinical Evidence Base

Autoimmune Disease Is The Foundational Proof of Concept: Aqtual’s strongest clinical evidence base remains in Rheumatoid Arthritis (RA), which is treated with a hierarchy of biologic and targeted agents. Primary non-response rates are substantial. Treatment switching is costly and delays remission. There’s a lot of data available on their website: see https://aqtual.com/publications to take a deeper dive yourself.

PRIMA-102 is Aqtual’s largest published dataset. The study enrolled >1,300 RA patients from the CorEvitas RA Registry across multiple sites to assess their cfDNAac test for predicting response biologic and targeted synthetic DMARDs. The classifier achieved AUC 0.80–0.85 for predicting biologic class response and mirrored results from synovial tissue biopsies.

EULAR 2026 (POS1269, poster): Is their most recent data showing that baseline cfDNAac active chromatin profiles reflect “synovial pathotype-associated programmes” distinguishing responders and non-responders to TNFi, JAKi, and abatacept.

Oncology Data Are Early But Compelling: Aqtual’s oncology programme is newer, centred on the DAPPER trial collaboration with Princess Margaret Cancer Centre and the University Health Network (Toronto), and currently focussed on the leiomyosarcoma (LMS) cohort.

LMS is a rare, aggressive soft-tissue sarcoma. In the metastatic setting, 5-year survival is 7–10%. Checkpoint inhibitor (CPI) monotherapy has minimal activity in LMS, but combination approaches (CPI + PARP inhibitors, CPI + anti-VEGF) have shown promising signals in some patients. The challenge is identifying which patients benefit, LMS has limited somatic mutation burden, and conventional ctDNA mutation profiling is uninformative for predicting CPI response. Tissue-based transcriptomics has identified immune cell infiltration and tertiary lymphoid structures as favourable markers, but repeat biopsy is impractical.

ASCO 2025 (Abstract 11539)presents the first LMS cfDNAac data. It introduced the cfDNAac assay as a predictive biomarker platform for CPI benefit in LMS (n = 30, DAPPER trial). Established feasibility and a preliminary biomarker signal. This was the first public oncology data from Aqtual.

The follow-up Lopes et al., npj Precision Oncology (May 2026) is the primary publication (and there’s an ASCO 2026 poster). 30 LMS patients treated with durvalumab + olaparib (Arm A) or durvalumab + cediranib (Arm B) in the DAPPER trial (NCT03851614). Baseline plasma samples analysed by cfDNAac. Clinical benefit rate (CBR) defined as PR or SD > 6 months. CBR achieved in 26.7% overall (40% Arm A; 13.3% Arm B). Feature class analysis showed that the immune cell cfDNAac signal was among the most predictive features, not just the tumour-derived regulatory signal. They also present a concordance analysis with tissue RNA-seq in 17 baseline samples with matched tumour RNA-seq, which was high. The ASCO 2026 poster extended the analysis to include longitudinal samples collected at disease progression (n = 22 baseline + progression pairs) - cfDNAac provided a non-invasive, longitudinal view of tumor microenvironment.

Other onclolgy data include an ASCO 2025 Colorectal cancer poster, and an AACR 2024 poster that presented detection of NSCLC and Bladder cancer signatures in blood.

So what does the clinical evidence suggest? Aqtual’s platform can distinguish disease from health and one disease subtype from another with high accuracy (RA diagnostic data). It can predict therapeutic response at baseline with clinically actionable discrimination (PRIMA-102 for RA; DAPPER for LMS/IO). The cfDNAac signal accurately mirrors matched tissue gene expression and tissue pathotype. The assay can monitor disease evolution longitudinally, tracking TME state changes from sensitive to resistant (ASCO 2026 LMS longitudinal) and simultaneously captures immune cell biology (blood TFBS: 413 of 1,570 features in LMS) and tumour regulatory state - neither is discarded as noise

But we need to see much larger oncology cohorts, e.g. the LMS DAPPER dataset is small (n = 30), ideally through prospective validation in larger, multi-site cohorts and across additional tumour types. And we need to see performance at lower tumour fraction to understand the lower-limit-of-detection at realistic ctDNA concentrations in earlier-stage disease.

I’d also love to see head-to-head comparisons of cfDNAac versus cfChIP-seq versus Proteus/Triton on the same patient cohorts. Such comparisons would clarify information content differences and define which approach best serves which clinical question.

Phew. That was a big one. Let me know what you think in the comments or engage on LinkedIN with my other #FLBx posts.

The paper (Göring et al. 2022) presents data from a ²²³RaCl₂, an alpha emitting molecule. The alpha radiation, helium nuclei, are “heavy” and only travel a tiny distance (a few cell diameters), but it deposits a massive amount of energy along a perfectly straight path, seen in Fig1 above as the cluoud chamber like traces in the cell nucleus. This was visualised by staining with DAPI (blue) and γ-H2AX as a DNA damage marker (green).

They found that DNA damage induction is dose-dependent, and DNA repair rates are faster at lower absorbed doses. This provides crucial quantitative insight into how healthy cells resolve radiation damage during systemic therapies.

Currently, FDA-approved RDCs rely on beta-emitting isotopes (such as Lutetium-177). Beta particles are essentially high-energy electrons. Unlike alpha-particles, they are light, easily deflected, but can travel a few millimeters through tissue. Because their energy deposition is relatively sparse, they primarily cause single-strand DNA breaks (SSBs) or isolated double-strand breaks. So they won’t necessarily create the stunning images in the Göring paper.

The Original Particle Tracker as a Biological Solid-State Detector

In early particle physics, the cloud chamber was the ultimate tool for visualizing the invisible. The chamber contains a supersaturated alcohol vapor. When an ionizing particle, such as a high-energy alpha particle, shoots through this environment, it collides with vapor molecules, knocking off electrons to create a trail of ions. These newly formed ions act as nucleation centers, causing the surrounding vapor to instantly condense into tiny liquid droplets. The result is a transient, visible physical wake that maps the particle’s exact trajectory. It allows us to track radiation not by observing the particle itself, but by the localized physical disruption it leaves behind.

Instead of a phase change in a gas, the cells in the Göring study act sort of as biological solid-state track detectors, capturing damage across the chromatin rather than condensation in a vapor.

#coolscience

PS: I first learned about this when I read Issac Asimov’s New Guide To Science at about age 14. Nerdy then…and more proudly now.

If Leah’s story moves you, please consider donating via muchloved.com to Blood Cancer UK (and Brooke Action for Working Horses and Donkeys) to support better research, better treatments and better outcomes for people facing blood cancers.

Leah Clissold, as I knew her before she married Stuart Catchpole, was a fantastic colleague in the JIC, which we joined at almost the same time back in the late ‘90s. We both ended up in the JGL, the John Innes Genome Lab, the first iteration of what was temporarily TGAC, but ultimately became the Earlham Institute, where Leah worked until her diagnosis meant she could not.

We worked hard at JGL (not everyone we worked with put in quite as much effort - you know who you are!), Leah worked on genomic libraries, BACs, YACs and other types with Gawain Bennett, Jane and Sam; Dave Baker ran the Sanger sequencing services with Nigel; I worked on establishing microarrays (anyone remember UKAffy’06 and the open-top bus trip around Norwich), all back in the days before The Human Genome had been cracked or NGS had been invented.

And we had fun. From the simple game of Spoof to see who would make the tea (sorry Maija but thanks, you were so bad, or Lee cheated so often that we almost never had to make it ourselves) and Tweaky Weaky (sorry Lewin, for leaving you with fag ash on your forehead all day) to the many memorable trips to conferences.

The JGL was early in establishing what today would be called a genomics core facility, and our boss pushed us to drum up business outside our host institute. This led to many memorable trips to conferences to present our work and to “sell” our services to other academic groups.

Two of these stick in my mind:

The first was a trip to Berlin, just the two of us, where the mission was to entice in big academic projects, but the conference our manager sent us to ended up being a bad choice with very few prospects. We did not waste all our time, Leah dragged me half-way across Berlin to see a car. She was so excited to sit in it and all I can remember is that it was green. I think it was a famous version of a racing car, but it may have been from the first The Fast and Furious movie. And we went out drinking to a bar (Café Zapata - sadly now closed) where they had a Dragon on the counter that breathed fire during the evening, and where we had our first digital DJ experience with a dance floor packed for 99 Red Balloons. FYI: Leah was an Ibiza club classics kind of girl and Born Slippy and Children played at her funeral.

The second was a group “outing” to a conference in York. We drove a minibus from Norwich without the boss and stayed in student accommodation together like we were all mates from Uni. It was awesome. It was also hilarious when Leah collected her badge from the reception desk - they had mis-spelled her name and she was Leach Clissold, but the joke did not stick. Leah was more light than leech.

Rest in peace Leah - you’d have wanted more dancing at the party ;-)

ultra-rare cancers enter the classic epidemiological paradox, while any single ultra-rare cancer affects only a few hundred people worldwide, collectively, hundreds of thousands of patients are navigating an ultra-rare diagnosis at any given time.

What can we do about rare cancers?

Leah had a Hepatosplenic T-Cell Lymphoma (HSTCL) after many years of immunosuppression. Her treatment gave her many years of relatively healthy and certainly active and exuberant life, but ultimately it was likely the cause of her cancer. That is a balance that fortunately few of us have to live with.

HSTCL is extraordinarily rare. Population-based studies suggest the incidence of HSTCL is about 0.1 cases per million person-years, which means around 800 new cases diagnosed globally each year, or less than 2% of all peripheral T-cell lymphomas (PTCLs), which are already an uncommon subset of non-Hodgkin lymphomas (NHL).

The Top 10 Rarest Cancers:

I do not believe there is an official “Top 10 Rarest Cancers” because cancer rarity depends entirely on how finely you slice the diagnostic criteria (e.g., by broad histology, distinct World Health Organization entities, or specific molecular sub-clones) but there are certainly different categories of rarity:

• Common Cancers, e.g., breast, prostate, lung, have incidence rates of roughly 1,000 to 1,500 per million people per year but have massive research programs in academia and Pharma with new drugs or combinations and $Billions invested in R&D.

• Rare Cancers, e.g., Ewing sarcoma, gallbladder cancer, are cancers that fall under defined thresholds of incidence. In the US this is 0.015% (or fewer than 150 per million), and in the EU this is 0.006% (or fewer than 60 per million), though many sit around 10 to 30 per million….NB: this is the number of diagnoses in a million people in the general population, not out of a million cancer patients.

• Ultra-Rare Cancers, e.g., HSTCL, NUT carcinoma and Blastic plasmacytoid dendritic cell neoplasm (BPDCN), have an incidence of close to or under 0.0001%, i.e. they are around one-in-a-million.

Because HSTCL, NUT carcinoma & BPDCN are up to 100 times rarer than diseases that already meet the regulatory definitions of a “rare cancer,” they firmly belong in the most extreme tier of ultra-rare malignancies. And that means very few patients are affected by these cancers, in the United States ~350 new cases per year, and ~750 in Europe, but this “ultra-rare” tail likely accounts for roughly 1% to 2% of total global oncology incidence. Thus, ultra-rare cancers enter the classic epidemiological paradox, while any single ultra-rare cancer affects only a few hundred people worldwide, collectively, hundreds of thousands of patients are navigating an ultra-rare diagnosis at any given time.

And those ultra-rare cancers often have poor outcomes with data from Europe’s RARECARE project demonstrating that the 5-year relative survival for rare cancers as a group is significantly worse (47%) compared to common cancers (65%). This is partly driven by a lack of large-scale Phase III clinical trials, delayed or inaccurate primary diagnoses, a scarcity of targeted therapies, and the reality that many patients are not treated at specialized academic centers equipped to manage obscure disease presentations.

A quick scan (I am not the expert here) of the literature suggests that some/many of these ultra-rare cancers are driven by recurrent mutations and they have pretty “quiet” genomes (I’d like to know more about the epigenomes too).

Ultra-rare cancers as a vanguard for genomic medicine

IMHO it is precisely this recurrent nature, the same mutations appearing across patients rather than being unique to any one individual's tumour, that makes ultra-rare cancers the perfect vanguard for genomic medicine. Given that we now have the tools to find these recurrent mutations easily if we look for them, specifically the deep WGS that the latest NGS instruments enable at affordable cost: Ultima, Element and Illumina, or the rich panoply of epigenome, spatial, single-cell and other ‘omics tech - why are we not using it in cases like Leah’s?

These small populations are surely ideal for diving deep into broad genomic medicine without ballooning costs to understand what mix of technologies fits best to give patients options.

1. Roll-out for WGS+Epigenome+RNA-Seq to ALL ultra-rare cancers: given the small numbers I’d argue strongly for National programs to deliver WGS on the tumor tissue for all ultra-rare cancer patients with a roadmap to rare and then common cancers (more likely with cancer panels here due to volume).

   1. For the UK NHS this would be 1000-1500 ultra-rare adult solid tumour genomes per year, many of whom have limited standard-of-care options, making a strong case for a WGS-first approach. That number is dwarfed by the 20,000 or so CNS, sarcomas, pediatric tumours and all acute leukaemias, which are already eligible for WGS*. In theory this is almost achievable on a single NextSeq…although I make that statement to stimulate discussion!

   2. Sadly, according to the Tessa Jowell Brain Cancer Mission report (2023), “Closing the Gap”, fewer than 5% of eligible adult brain tumour patients were actually accessing WGS through NHS commissioned pathways in 2023. So we have a long way to go in the UK.

2. Cost-Effective Multi-Omics Prototyping: Deploying comprehensive genomic profiling platforms, experimental cell-free chromatin profiling, or deep spatial on fresh frozen tissue across a trial of 60 patients with an ultra-rare disease will not bankrupt a trial sponsor or make an academic grant impossibly large to fund. It would allow researchers to generate exhaustive, multi-omic baseline data that would be financially and logistically prohibitive if piloted directly on a massive, frontline lung or breast cancer population. But getting those patients together requires global registries.

3. High Biological Signal-to-Noise Ratio: Common cancers are often heavily influenced by environmental factors, resulting in chaotic genomes with immense mutational “noise.” In contrast, ultra-rare cancers are frequently driven by singular, highly penetrant molecular events (such as the STAT5B or SETD2 mutations in HSTCL). This pristine biological environment creates a much clearer “signal” e.g. for tracking clonal evolution.

4. The “Trickle-Down” Paradigm: What we learn in these ultra-rare cohorts about the best mix of technologies to identify relevant biological variation, or which AI tools are best placed to mine these multi-omic data and make sense of data coming from across the globe (remember these need to be global studies), and/or to inform new therapeutic options or combinations that may be relevant in building new “basket” trials for less rare cancers that carry similar mutational, epigenetic or permissive-IO states.

As sequencing costs continue to drop and the multi-omics technologies are optimized, those pipelines can then be rolled out to identify and track the rare molecular sub-clones present within much larger, heterogeneous cancer populations.

What do you think?

Is this proposal a sensible forward-looking strategy for precision oncology? Can we, or should we, use ultra-rare cancers as a vanguard to prototype advanced genomic and molecular technologies in health services? Are you already doing this?!

Please feel free to add your thoughts in the comments and/or get in touch directly.

Some useful links:

• *NHS test directoy incl. WGS: https://www.england.nhs.uk/publication/national-genomic-test-directories

• Tessa Jowell Brain Cancer Mission report (2023), "Closing the Gap": https://www.tessajowellbraincancermission.org/wp-content/uploads/2024/09/Closing-the-Gap-Report.pdf

• Management of patients with rare adult solid cancers: objectives and evaluation of European reference networks (ERN) EURACAN: https://www.thelancet.com/journals/lanepe/article/PIIS2666-7762(24)00027-9/fulltext

• The EURACAN consortium of 100 European expert centers designed to connect patients with rare adult solid cancers to highly specialized healthcare: https://www.euracan.eu

• Rare cancers are not so rare: the rare cancer burden in Europe: https://pubmed.ncbi.nlm.nih.gov/22033323

• The RARECAREnet database on the epidemiology of rare cancers in Europe:

  http://rarecarenet.istitutotumori.mi.it

• The EURORDIS Rare Cancer Advocates Network: https://www.eurordis.org/our-priorities/rare-cancers

• The NCI Rare Tumors Initiative, focused on understanding the molecular drivers of rare malignancies to pioneer precision therapies: https://ccr.cancer.gov/research/rare-diseases, FYI no HSTL on their list.

• NCI MyPART (My Pediatric and Adult Rare Tumor Network): https://www.cancer.gov/pediatric-adult-rare-tumor

• NORD (National Organization for Rare Disorders), which covers all rare diseases (not just oncology), but who are the most powerful US advocacy and funding organization for ultra-rare conditions: https://rarediseases.org

As I look toward the future of MRD, a few themes loom large: ultra-sensitive tumor-informed MRD detection should not be gated behind massive infrastructure, patients should not have to wait weeks for results, and sensitivity can always be improved. In the post below I highlight a few enabling technologies that I think will impact MRD solutions now or might do so in the not-too-distant future.

FYI: in this post I am intentionally not considering tumor-naïve solutions although they certainly have their place in the testing landscape.

Why are MRD tests all run on Illumina’s NovaSeq?

Because it is the most cost effective way to run a sequencing factory and reduce the costs of the biollions of reads needed to generate the tumor fingerprint to identify varinats for MRD tracking, the normal genome to remove CHIP variants, and then finally the deep plasma sequencing to detect MRD.

This means most of the MRD companies out there use a ton of NGS to get a result and that requires a NovaSeq…or an Ultima (I’m less sure if anyone is delivering MRD on Element - drop me a message if you are).

The challenge with needing a massive sequencer is you have to fill it and you probably won’t find it anywhere near your local hospital. So what would happen if you could reduce the number of reads you needed and move MRD from NovaSeq to NextSeq for instance?

Enspyre: MRD with >90% less sequencing?!

At AACR 2026, Rita Zhou in my team presented Abstract 1143: Enspyre: A novel enrichment technology enables ultra-sensitive ctDNA detection with 98% reduction in sequencing requirements. The major advances in ultra-sensitive MRD have come from a move to WGS-informed fingerprinting and increasing the numbers of variants being tracked (e.g. Personalis NeXT Personal (1800 variants), Myriad MyChoice (1000 variants) and Signatera, who moved from Exome+16 variants (E16) to Genome + 256 variants (G256, incl phased-variants). For the companies built on hyb-cap of large numbers of variants there’s a lot of sequencing data needed to get to the ultra-low PPM levels and that means these assay runs on Illumina’s NovaSeq line, usually found in larger centres.

In our AACR 2026 poster and a recent publication (Scientific Reports 2026), we demonstrated how the Biofidelity Enspyre technology can reduce sequencing depth by well over 90%. This is in line with their earlier paper and their AACR 2026 Analytical Validation poster, where Biofidelity achieved 5PPM LOD in cell lines with just 10M reads.

If you do not know the Enspyre technology it is worth taking a look at. It works similarly to standard hybridisation-capture but instead of using a simple change in pH or temperature to release the captured DNA fragments, BioFidelity use selective pyrophosphorolysis to enrich variant-containing molecules. This essentially PCR in reverse, and it digests the biotinlyated probe sequence in an exquisitely sequence specific manner such that any mismatch causes digestion to stop. As such only mutant alleles are releaseed back into solution for downstream NGS, massively reducing background wild-type sequence, and therefore reducing the sequencing needed to find mutant molecules.

In our experiments using MATRIX samples (the plasma-in-plasma reference materials we designed to assess new technologies) we were able to achieve ultra-sensitive ctDNA detection at 10 ppm with a 98% reduction in sequencing depth. This means it should be possible to achieve 100% sensitivity at 10PPM entirely on a benchtop NextSeq 550 and/or increasing the numbers of variants to many thousands to increase sensitivity whilst maintaining reasonable cost per test. This takes MRD out of the realm of massive NovaSeq runs and so could be a massive leap for democratization. It means smaller labs and community oncology centres can run ultra-sensitive, kitted MRD solutions in-house rather than shipping samples to centralized hubs.

TWIST MRD Express: Closing the TIA vs. TNA Turnaround Gap

Sequencing depth is only one of the challenges for ultra-senstiive tumor-informed MRD; the other is turnaround time (TAT). The clinical utility of personalized, tumor-informed MRD tracking is often constrained by the complexity of panel design and somewhat lengthy manufacturing lead times. This delay has historically given off-the-shelf, tumor-naïve panels a distinct logistical advantage.

However, the recent AACR poster from TWIST Bioscience on their MRD Express Panel showcases a scalable target enrichment workflow that fundamentally changes this math. Leveraging a silicon-based DNA synthesis platform and proprietary design algorithms, their automated pipeline can achieve a rapid turnaround time of as little as one business day from synthesis to shipment. And they can do this for panel sizes up to 5,000 probes.

To ensure robust performance across challenging genomic regions, Twist implemented an innovative Locked Nucleic Acid (LNA) boosting strategy that enhances coverage uniformity for probes with extreme GC content. When paired with a double-stranded staggered probe design, they can dramatically expand the available unique sequencing depth.

A 24-hour panel generation timeline shifts the paradigm for tumor-informed assays. It allows highly optimized, personalized tracking to be deployed almost as quickly as TNA assays, enabling early detection of molecular relapse within clinically actionable decision windows.

Pushing the Floor: Aarhus Vignettes

The innovation certainly doesn’t stop there. I recently had the pleasure of speaking at the International Symposium on ctDNA in Aarhus this past May, and I was struck by a couple of remarkable early-stage posters from Syndex Bio and Amplifyer Bio that tackle the signal-to-noise floor from entirely unique angles.

• Syndex Bio: As shown in their “mcPCR: PCR For Methylation” poster, David McBride and his team presented their technology that enables amplification of native DNA whilst preserving its methylation status. In a single-tube reaction they perform PCR and a “methyl-copy” reaction that synchronizes base extension and methylation preservation using novel enzymes. This overcomes a massive bottleneck by avoiding destructive bisulfite conversion while completely preserving epigenetic signals from tiny inputs of cfDNA.

• Amplifyer Bio: In the poster titled “Improving Liquid Biopsies by Boosting ctDNA Yield with Humanized Priming Agents”, Julia Amaral, Fujiko Duke, and their co-authors introduced an engineered DNA-binding antibody that acts as an in vivo “priming agent”. By shielding cfDNA from biological clearance, they demonstrated an incredible up to 100-fold increase in ctDNA recovery, significantly shifting the boundaries of what is stochastically catchable in a standard blood draw.

  Whilst they do not explain it quite as such I see this technology as analogous to a contrast agents for MRI, where the Gadolinium dramatically improves the detail of soft tissues and tumors, but this contrast agent for ctDNA boosts the signal we can detect using standard liquid biopsy techniques. See their Science paper from 2024 for more detail.

Both Syndex Bio’s non-destructive methylation amplification and Amplifyer Bio’s in vivo target preservation protocols could improve detection sensitivity of current tests by adding in a signal boost. The biggest impact could be for tumor-naïve methods, which are currently sitting at around 500PPM LOD. Would adding either of them push the limit of detection down to 100PPM? And if we stacked these technologies in a single workflow, would the sensitivity gains be simply additive, or truly multiplicative, maybe pushing down even further into the single-digit parts per million range?

These are exactly the questions we need to answer as we build out the next generation of liquid biopsy. But I’ll save that deep dive for another post.

What Is HSTCL?

Hepatosplenic T-cell lymphoma is an extremely rare and aggressive form of non-Hodgkin lymphoma. It most often arises from a specific subset of immune cells, gamma-delta T-cells, which normally patrol the body’s organs. In HSTCL, these cells undergo malignant transformation and accumulate predominantly in the liver, spleen, and bone marrow, causing marked enlargement of the liver and spleen, disruption to normal blood cell production, and often rapid clinical deterioration.

The disease typically strikes young adults, often men in their twenties and thirties, though it can appear at any age. Many cases are diagnosed in individuals with a history of immunosuppression, inflammatory bowel disease patients on long-term thiopurine or anti-TNF therapy, solid organ transplant recipients, and others. Diagnosis is frequently delayed because the early symptoms, fatigue, abdominal swelling, weight loss, recurrent fevers, overlap with many more common conditions.

For HSTCL, genomic sequencing is not currently used to guide treatment in the way it is for some more common cancers. In most cases, management is still driven primarily by the histopathological diagnosis, the patient’s clinical condition, and whether intensive chemotherapy and allogeneic stem cell transplantation are feasible. Sequencing may still be informative for research, biological understanding, and occasionally trial identification, but it rarely changes standard treatment directly.

It might have been interesting to sequence my friend, she could even have done it herself as she worked in a Genomics Institute, but it appears to be extremely unlikely that it would have changed anything in terms of her treatment, or her outcome.

A Disease Defined by What We Don’t Know

HSTCL is estimated to account for fewer than 1% of all non-Hodgkin lymphomas worldwide, with only a few hundred cases reported in the medical literature. That rarity is not just a statistical footnote, it is the root cause of almost every challenge patients and clinicians face.

Because so few cases exist, assembling the kind of large patient cohorts needed for robust genomic studies is enormously difficult. What genomic data we do have has been pieced together from small case series and retrospective analyses. Recurrent alterations have been identified, including mutations in chromatin-remodelling genes such as SETD2, INO80, and ARID1B, and abnormalities of chromosomes 7 and 8, but the landscape remains incompletely mapped. We do not yet have a clear picture of which mutations drive disease initiation, which drive progression, and which might represent vulnerabilities that drugs could exploit.

Without that genomic foundation, the rational design of targeted therapies, the approach that has transformed outcomes in cancer, has barely begun. There is no approved targeted agent for HSTCL. No companion diagnostic. No genomically stratified treatment algorithm.

Treatment: Borrowed Tools That Often Fall Short

The standard of care for HSTCL relies on chemotherapy regimens designed for other aggressive lymphomas, most commonly intensive combinations such as ICE or IVAC, often bridging to allogeneic stem cell transplantation in patients fit enough to receive it. Transplant remains the only intervention associated with durable remissions in a meaningful proportion of patients - unfortunately in my friend’s case it didn’t work and she ran out of options.

For many patients the initial response rates to induction chemotherapy are variable, and relapse is quick and common. Even with transplant, long-term survival rates remain low. Patients who are older, unfit, or unable to access a suitable donor have few options. For them, there is no second-line standard of care; treatment decisions are made case by case, drawing on anecdotal reports and small series rather than evidence from randomised trials.

Novel agents, including histone deacetylase inhibitors, the anti-CCR4 antibody mogamulizumab, and other agents active in T-cell lymphomas more broadly have been explored in small numbers of HSTCL patients, sometimes with encouraging signals. But without dedicated trials, these observations remain fragmentary.

There are just four trials on ClinicalTrials.gov:

• NCT01804166: A Research Study to Bank Samples for Future Evaluation to Identify Biomarkers That Predispose Crohn’s Disease and Ulcerative Colitis Patients to Develop Hepatosplenic T-Cell Lymphoma (HSTCL) Completed but no data published.

• NCT04021082: CELTIC-1: A Phase 2B Study of Cerdulatinib in Patients With Relapsed/Refractory Peripheral T-Cell Lymphoma (PTCL) Withdrawn.

• NCT02087878: A Blood and Tissue Sample Collection Study of Patients Who Have Inflammatory Bowel Disease, Who Have Been Treated With Adalimumab and Who Developed Hepatosplenic T-Cell Lymphoma Withdrawn.

• NCT05475925: A Study of DR-01 in Subjects With Large Granular Lymphocytic Leukemia or Cytotoxic Lymphomas. This Phase 1 trial, which is currently active and recruiting, demonstrated that the CD94-targeting antibody DR-01 is safe, tolerable, and active, achieving a 33.3% objective response rate in patients with relapsed or refractory cytotoxic lymphomas.

The Rare Disease Problem Affects Cancer Patients Too

HSTCL sits within a wider and deeply troubling pattern. Rare cancers, individually, each affect only a small number of people. Collectively, they account for roughly 20% of all cancer deaths. Yet because no single rare cancer generates sufficient commercial or scientific momentum on its own, they tend to attract a fraction of the research investment directed at common tumour types.

Pharmaceutical development is driven, in part, by market size. A drug developed specifically for HSTCL would serve a global patient population in the hundreds per year. The economics are brutal, and without regulatory incentives, academic consortia, and patient advocacy, these diseases remain in the shadows.

International collaboration offers one way forward. Rare cancer registries, centralised biobanking, and multi-centre genomic studies can aggregate data that no single institution could collect alone. Basket trials, which enrol patients across multiple rare tumour types sharing a common molecular feature (no use to my friend as WGS is not done by default - although JAK/STAT pathway mutations, which are common in HSTCL and can sometimes be targeted with JAK inhibitors like ruxolitinib in compassionate use cases), offer another route to generating evidence without requiring the large, histology-specific cohorts that rare diseases can never provide.

Why All Of This Matters

My friend did not die for lack of courage, or care, or effort on the part of those treating her. She died, in part, because the medical science simply was not there. The genomic map of her disease was never considered. The targeted therapy that might have exploited a specific vulnerability in her cancer could not have been found and may not yet have been developed. The clinical trial that might have offered her an alternative did not exist.

That is not inevitable. It is a consequence of how we have chosen, or failed to choose, to direct resources and attention. Rare cancers deserve better. The patients who face them, and the people who love them, deserve better.

If you want to support the broader effort to improve outcomes for rare and poorly understood cancers, consider looking into organisations such as the Lymphoma Research Foundation, the T-Cell Lymphoma Forum, or the Rare Cancers Europe initiative. Awareness is not a cure, but it is a beginning.

RIP my friend and colleague. I’ll always remeber our bunking off in Berlin!

While Swanton and Pandya used blood proteins to catch the “spark” of lung cancer before it forms, Arutha’s team show that that similar blood-based proteomic profiling can dynamically track the cancer once it is established and undergoing treatment.

I’m a ctDNA guy but I am open to other ideas and actively pursuing a “Beyond ctDNA” agenda in my day job. As such, plasma-proteomics excites me, but I am by no means an expert. Here is how I think the npj paper directly links and adds weight to the earlier work.

1. Validating the Power of Blood Proteomics

Swanton and Pandya demonstrated that a 14-protein signature in the blood could predict lung cancer risk years in advance. The npj paper validates that blood proteomics is equally powerful for monitoring the disease later in its lifecycle.

They used the Illumina SomaScan* v4.1, which comprises reagents to detect 7596 plasma proteins, and the Alamar Biosciences NULISA** targeted panel to detect 250 proteins, before and after surgery, and before and after immune checkpoint immunotherapy (ICI), to discover potential prognostic and diagnostic biomarkers. They identified a 21-protein plasma signature that changes in response to surgical resection and ICI therapy.

Both papers cement the idea that the blood proteome is a highly accurate, real-time window into lung cancer biology.

2. The Central Role of Inflammation

Swanton and Pandya’s breakthrough hinged on inflammation (specifically IL-1β) as the catalyst that wakes up dormant mutant cells due to PM2.5 exposure. The established NSCLC tumor cells do not produce the signature. It is coming from proteins secreted by the surrounding inflamed lung tissue, specifically myeloid immune cells and a highly plastic, transitional cell state they call “KAC” cells.

The signature remains present in the blood; it just represents the toxic, inflammatory “soil” that allowed the cancer “seed” to grow in the first place, rather than representing the cancer itself.

The npj paper intentionally incorporates the NULISA platform specifically to measure 250 inflammation-related proteins. This shows that the inflammatory cascade doesn’t just start the cancer, it continues to dictate the tumor microenvironment and heavily influences whether a patient will respond to immunotherapy. Inflammation is the common thread from the first precancerous cell to late-stage immune resistance.

3. Completing the Liquid Biopsy Timeline

Taken together, these two bodies of research map out a complete timeline for the future of lung cancer care:

• Early Interception (Pandya/Swanton): Using protein signatures to screen high-risk, never-smokers (like those exposed to high cooking indices) to determine risk status and enable interventional studies that might demonstrate cancer prevention. And the signature may act as an early cacner detection tool fitting into the mission of the Susan Wojcicki Foundation that I described in my earlier post.

• Treatment Response & Resistance Monitoring (Naei/Kilgallon/Kulasinghe): Using alternative and complementary protein biomarker tech they saw expression changes consistent with biological adaptations following surgery, signals associated with immunotherapy response, and that could be used to predict prognosis and survival in NSCLC. The work suggests that blood proteomic biomarkers can offer a real-time and non-invasive way to capture dynamic tumour and immune-related signals to better guide disease stratification and therapeutic outcomes for surgery and targeted therapies.

Ultimately, the npj paper reinforces Swanton’s earlier premise: the keys to understanding, catching, and treating lung cancer are actively circulating in the blood, waiting for us to read the not just ctDNA, methylome or cfRNA but also protein, and probably glycomic, metabolomic and other signatures. We just need to look with new tools.

Technology focus

A quick bit on the tech as this is what usualy draws me into the biology…

*Somascan as The Wide-Angle Lens: The SomaScan assay (v5 interrogates 11k proteins) uses SOMAmers (Slow Off-rate Modified Aptamer), short, chemically modified synthetic DNA sequences (aptamers) engineered to fold into complex 3D shapes that bind tightly to specific target proteins (epitopes). They act similarly to antibodies, binding tightly to specific target protein, but because they are made of DNA, they can be read by DNA sequencing machines.

After mixing with plasma or serum, streptavidin beads are used to caputre aptamer-bound proteins allowing the unboud, weakly bound, and off-target proteins to be washed away. UV light is used to break a chemical linker, releasing only the highly specific protein-SOMAmer complexes for downstream processing.

Previously a microarray readout was used and after SOMAmers were hybridized to a physical chip, the fluorescence intensity was used to estimate how much protein was present. But nowadays an NGS readout is used. In this the isolated DNA SOMAmers are treated exactly like a genomic sample and used in an Illumina library prep. The sequencing read counts of each SOMAmer are directly proportional to the starting protein conentrations. Illumina bought SOMAlogic for $350M last year.

**NULISA employs a highly sensitive dual-antibody sandwich approach with DNA barcoding and NGS, achieving attomolar sensitivity

**NULISA as a High-Powered Microscope: The Alamar Biosciences NULISAseq Inflammation panel interrogates 250 inflammatory cytokines, chemokines, and immune regulators. NULISA (Nucleic Acid-Linked Immunosorbent Assay), it uses a dual-antibody sandwich approach, where both antibodies are tagged with unique DNA barcodes. Only when both antibodies successfully bind the target protein do their barcodes connect to form a sequence that can be read via Next-Generation Sequencing (NGS). This dramatically reduces background noise (read the Feng 2023 Nat Comms paper)

It achieves attomolar (femtogram/mL) sensitivity, allowing it to detect targets at levels standard immunoassays simply cannot reach.

This slightly longer post aims to highlight two advances in lung cancer research, the launch of The Susan Wojcicki Foundation and the recent work from Charlie Swanton and Tej Pandya, and outlines my own personal cancer journey (my nan and two mums - that is not a typo).

The Silent Shift: Non-Smoking Lung Cancer and The Susan Wojcicki Foundation:

Susan Wojcicki sadly passed away from lung cancer (NSCLC) almost two years ago but in the two years since her diagnosis she’d kicked off a research portfolio many cancer institutes would be proud of. The Susan Wojcicki Foundation is building on this work, developing an accelerator for lung cancer early detection and prevention. Rather than competing with cancer research behemoths like the NCI ($7.35 billion) and CRUK ($510 million), it aims to fill critical gaps like the underfunding of lung cancer research relative to its mortality rate, leveraged by the family’s deep ties to the tech industry. It looks like they will leave the fundamental biology and therapeutics development to the big guns and fill a more specialized, agile niche, while collaborating with organizations like Stand Up To Cancer and the Lung Cancer Research Foundation, but bringing their tech focus and specialized resources, such as Google’s AI capabilities, to rapidly scale new diagnostic tools and redefine screening for non-traditional demographics like younger, never-smoking women.

The Foundation aims to aggressively target the systemic gaps Susan’s journey exposed. It is laser-focused on redefining risk by integrating genetic, environmental, and clinical data; they want to improve lung cancer early detection by liquid biopsy; and they are supporting the development of cancer vaccines to something that can be clinically deployed.

As such, the Foundation appears to be stepping into a crucial niche that complements, rather than competes with, the NCI and CRUK by rapidly accelerating highly targeted, tech-driven research in a demographic that the broader public health guidelines currently miss.

Susan Wojcicki had no known risk factors, was physically active, and had never smoked. Her death made her the latest high-profile individual lost to Lung Cancer In an individual who had Never Smoked (LCINS), a form of the disease that highlights how lung cancer is a disease that appears to be steadily moving away from the traditional demographic, smokers.

Today, approximately 10-25% of new lung cancer diagnoses worldwide occur in never-smokers

Samet JM, Avila-Tang E, Boffetta P, et al. Lung cancer in never smokers: clinical epidemiology and environmental risk factors. Clin Can Rese. 2009;15(18):5626–5645.

Why is this the case - pollution seems to be a big part of the answer: In regions like Southeast Asia, air pollution is a major driver of PM2.5 exposure. This may extend to indoor air pollution, exacerbated by poor ventilation and cooking fumes, an environmental factor often called the Cooking Index, which offers an explaination of the disproportionately high rates of EGFR-mutant NSCLC among non-smoking women of Asian descent [Ref 1]. However, the major epidemiological studies linking PM2.5 to lung cancer is all outdoor pollution.

The environmental catalyst for lung cancer has been the focus of Professor Charlie Swanton’s most recent work. His groundbreaking TRACERx research has fundamentally rewritten our understanding of how Lung cancers evolve [Ref 2 has some wonderful MRD data from Personalis], but this latest work demonstrated that the pollution (specifically PM2.5 particulates) that appears to drive non-smoking lung cancer does not mutate DNA directly like tobacco smoke [Ref 3]. Rather, when the lungs are repeatedly damaged by environmental pollutants, i.e. PM2.5’s, they release an inflammatory signal called IL-1β, which initiates tumorigenesis through pre-existing naturally occurring EGFR mutations and leads to the development of NSCLC [Ref 4].

Swanton and Pandya’s breakthrough bridges the gap between environmental exposure (like the cooking index) and the actual biological mechanism that causes non-smokers to develop lung cancer. They identified a signature of 14 proteins in the blood that can predict a person’s risk of developing lung cancer more than five years before a diagnosis - so surely these proteins come from undetected cancers?

Apparently not. Their work appears to show what is happening: as pollution causes the immune system to release IL-1β, it floods the lungs and drives normal lung cells into a stressed and inflammed “hot” state, and it is this inflammatory environment that allows EGFR mutant “normal” lung cells to flourish and develop into non-small cell lung cancer. These biological changes are be seen in the blood as the 14-protein signature, released by those stressed normal lung cells, that act as an alarm bell for this highly specific, pre-cancerous state of chronic lung inflammation.

And crucially it opens a new opportunity for lung cancer prevention using anti-inflammatory drugs (like those blocking IL-1β) to soothe or “cool” the lung environment and stop the dormant mutant cells from ever becoming a tumor.

My personal cancer journey - my nan and two mums.

Cancer research, like the breakthroughs described above, makes for good news headlines, and hopefully will lead to fewer cancers in the future, but the diagnoses and deaths that it causes hits home for real people every day.

For me, born in the early 1970s, everyone around me smoked: my dad, aunts and uncles, and my nan. She chain-smoked us over the Alps every year in her VW camper van, through Reims, Lucerne and then to Lake Como for our Summer holidays - bliss. But she died in 1991 from brain cancer, which I have long suspected could have been a NSCLC metastasis, just a few years after retiring.

My extra Mum died two years ago after three different cancers over about twenty years. She was my best mates Mum, and took me in when I left home at 15 after dropping out of high-school. She turned me around and put me firmly on the path to the LinkedIN page you may well have seen this post on.

And my Mum, who left when I was 2 and I never knew, died from lung cancer last year. Whilst I never had any relationship with her, it was still sad to hear that she’d died from a disease that is highly preventable (she smoked).

Other relatives, my mother-in-law, and friends, an ex-colleague was quote recently diagnosed with a hepatosplenic T-cell lymphoma (any advice gratefully received)…the list goes on.

Finally, I remember the cultural shock of Roy Castle’s diagnosis. For those of you in the states who do not know him, Roy Castle was a lifelong non-smoker who spent years playing trumpet in poorly ventilated, smoke-filled jazz clubs; he was also co-host of the Guiness World Records on TV when I was a kid. He was diagnosed in early 1992 and passed away in 1994; there’s now a Roy Castle Lung Cancer Foundation (the only UK charity solely dedicated lung cancer - I hope they are in touch with The Susan Wojcicki Foundation). His death shattered the illusion that non-smokers were immune to lung cancer.

My 30 year (so far) personal journey in cancer research

Most of my career has been dedicated to cancer genomics and diagnostics. Working across the The Big C, the John Innes Centre (non-cancer), Cancer Research UK, and now AstraZeneca, I’ve been luck enough to have front-row seats to the evolution of genome sciences, cancer genomics and oncology diagnostics.

My journey started with an early effort in biomarker testing for HER2+ breast cancer, establishing a differential PCR assay for the detection of c-erbB-2 (HER2) amplification in 1996 [Ref 5]. I moved into genome sciences via plant biology at the John Innes Centre where I set up one of the first Affymetrix labs in the UK (hence my recent MIAMIE-MRD post). But I focused on cancer genomics when I moved to CRUK’s Cambridge Institute to build the Genomics Core lab. There I worked on a huge variety of projects but two stand out for me in the context of this post: firstly the redefinition of the molecular subgroups of breast cancer and how we view clonal evolution with pioneers like Christina Curtis, Samuel Aparicio, and Carols Caldas [Ref 6]; and secondly on the development of liquid biopsy data that led to the spin-out of Inivata with Tim Forshew, Davina Gale and Nitzan Rosenfeld [Ref 7].

Today, my focus has shifted toward the vanguard of non-invasive diagnostics at AstraZeneca. My small team is building the tools to evaluate and standardize liquid biopsy tests, as part of the wider Biomarker, Science & Technologies function. For us, this includes pushing the boundaries of MRD monitoring and FLBx, and most recently, expanding the utility of cell-free DNA beyond just efficacy, exploring its potential to detect and monitor toxicity signals in real-time [Ref 8].

Closing: Looking back at the people I’ve lost to cancer, and considering the systemic gaps that initiatives like the Wojcicki Foundation are now rushing to fill, the stakes for our work have never been clearer. The era of simply reacting to cancer is ending. Today, we are moving beyond just finding the tumor to understanding its oncogenic state (is it a particular subtype, willit respond to therapy, is it becoming resistant, what is its immune environment), predicting its trajectory years in advance, and managing the toll of its treatment with cfDNA-Tox or deciding whether to treat or not (e.g. MRD de-escalation).

Every extracted data point, every new cancer genome, every optimized diagnostics assay, and every new drug is a step toward a reality where families don’t have to rely on luck or late-stage interventions. The wider cancer research community is making a much-needed mark on lung cancer - one that is earlier, smarter, and profoundly personal.

#1 LoPiccolo, J., Gusev, A., Christiani, D. C., & Jänne, P. A. (2024). Lung cancer in patients who have never smoked — an emerging disease. Nature Reviews Clinical Oncology, 21, 121–146. https://doi.org/10.1038/s41571-023-00844-0

#2 Black, J. R. M., Bartha, G., Abbott, C. W., et al. (2025). Ultrasensitive ctDNA detection for preoperative disease stratification in early-stage lung adenocarcinoma. Nature Medicine, 31, 70–76. https://doi.org/10.1038/s41591-024-03216-y

#3 Pleasance, E. D., Stephens, P. J., O’Meara, S., et al. (2009). A small-cell lung cancer genome with complex signatures of tobacco exposure. Nature, 463, 184–190. https://doi.org/10.1038/nature08629

#4 Pandya, T., et al. (2026). Plasma signals of lung tumor promotion for molecular cancer prevention. Cell. https://doi.org/10.1016/j.cell.2026.05.005

#5 Jennings, B. A., Hadfield, J. E., Worsley, S. D., Girling, A., & Willis, G. (1997). A differential PCR assay for the detection of c-erbB 2 amplification used in a prospective study of breast cancer. Molecular Pathology, 50, 254–256. https://doi.org/10.1136/mp.50.5.254

#6 Curtis, C., Shah, S. P., Chin, S.-F., et al. (2012). The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature, 486, 346–352. https://doi.org/10.1038/nature10983

#7 Forshew, T., Murtaza, M., Parkinson, C., et al. (2012). Noninvasive Identification and Monitoring of Cancer Mutations by Targeted Deep Sequencing of Plasma DNA. Science Translational Medicine, 4. https://doi.org/10.1126/scitranslmed.3003726

#8 Zhao, Y., Gallon, J., Kushner, M. H., et al. (2026). Beyond Circulating Tumor DNA for Ef cacy: Can We Use Cell-Free DNA to Detect and Monitor Toxicity Signals? JCO Precision Oncology, 10:e2600178. https://doi.org/10.1200/PO-26-00178

The comments turned into an absolute goldmine of niche scientific and professional humot that were simply too good to let disappear into the feed. As such, I’ve decided to collate them all into a highly official “Community Resource.”

Where are the jokes?

Here is the collation of the jokes from the thread, organized into themes. If you have any more to add, keep them coming here or send them my way in the Comments.

🧬 Genetics, Genomics & Molecular Biology

• Genetics: “I have a genetics joke, but I don’t know how at peas you would be hearing it.”

• Epigenetics: “I have a methylation joke, but it’s been silenced.” (Also: “I have a genetics joke, but it’s methylated.”)

• CRISPR: “I have a CRISPR joke, but it’s a bit off-target.” (Also: “...but it cuts straight to the point.”)

• GWAS: “I have a GWAS joke, but it lacks power.”

• Sequencing/Diagnostics: “I have an MRD joke but it may be too sensitive, if you are well informed.”

• Early Detection: “I have an early detection joke, but it’s buried in too much noise.” (With an honorable mention to “ONT jokes hiding in that noise”).

🧫 Microbiology & Immunology

• Microbiology: “I have a microbiology joke, but it may develop resistance.”

• Microbiology: “I have a microbiology joke, but it depends on the culture.”

• Immunology: “I have an autoimmune joke, but you may feel attacked.”

🧪 Chemistry & Biochemistry

• Chemistry: “I have a chemistry joke, but it gets mixed reactions…”

• Chemistry: “I have a chemistry joke but it’s probably just going to get no reaction.” (Follow-up: “Maybe it needs a catalyst.”)

• Biochemistry: “I have a biochemistry joke but it’s lower than detection limit.”

• Metabolism: “I have a metabolism joke, but it’s not too palatable.”

🩺 Medicine & Anatomy

• Surgery: “I have a surgical joke… well, it’s really more of a cutting remark.”

• Anatomy: “I have an anatomy joke, but I can’t put my finger on it.”

• Nursing: “I have a nursing joke, but it’s currently being triaged.”

• Cryobiology: “I have a cryopreservation joke but it won’t break the ice.”

💻 Computer Science, AI & Bioinformatics

• Bioinformatics: “I have a bioinformatics joke, but I need an environment to run it.”

• Bioinformatics: “I have a bioinformatics joke but the pipe(punch)-line is broken.”

• AI/LLMs: “I have a Claude joke, but I’m out of tokens.”

• AI: “I have an AI joke but it’s all over the place.”

• Programming: “I have a bash scripting joke, but it’s not executable!”

• Programming: “I have a code joke but it has no comments!!”

• Databases: “I have a database joke, but it is denormalized....and will not JOIN well.”

• Hardware/Data: “I have a joke about memory, but its out of cache.” (Also: “I have a data joke, but I am out of memory.”)

⚛️ Physics & Engineering

• Quantum Physics: “I have a quantum joke, but it’s both funny and not funny.” (Follow-up: “That’s a Schroedinger joke!”)

• Thermodynamics: “Entropy just isn’t what it used to be.”

• Optics: “I have a joke about optics but it lacks perspective.”

• Haptics: “I have a joke about haptics, but it’s out of touch.”

• Engineering: “I have an engineering joke, but it doesn’t hold up under stress.”

📊 Business, Finance & Management

• Human Resources: “I have an HR joke, but it’s confidential.”

• Organizational Development: “I have an Org Development joke...but it’s not very well structured.”

• Management: “I have a management joke, but it rarely goes as planned.”

• Accounting: “I heard an accounting joke, but I can’t repeat it because it was accrual one.”

• Finance: “I have a money joke, but it lacks funding.”

• Sales: “I have a sales joke, but its all features and no benefits.”

🧠 Humanities & Soft Skills

• Psychology: “I have a psych joke, but I’m not sure how it makes me feel.”

• Philosophy: “I have a philosophy joke, but why.”

• Communication: “I have a communication joke, but I don’t know how to say it.”

• Storytelling: “I have a storytelling joke, but you had to be there!”

For pharma, across many modalities, but perhaps most imapactfully for Antibody-Drug Conjugates (ADCs) and radiolconjugates (RCs) and Immunotherapy (IO), FLBx may be the holy grail. New methods allow us to track lineage plasticity, identify emerging resistance mechanisms, and understand TME heterogeneity in real time. But up until now, the major players in FLBx have relied on brilliant, sometimes complex, biochemical extractions to get this signal.

• cfChIP-Seq (Senseera Health, Precede Biosciences) uses highly specific antibodies to physically pull down circulating nucleosomes bearing active histone marks.

• cfDNAac (Aqtual) uses proprietary biochemical footprinting to physically separate open, active chromatin from tightly wound, silent nucleosomes.

• cfDNA methylation (LiquidCell) uses WGMS to deconvolve the tumor microenvironment. Their LiquidTME AI reads the epigenetic signatures in the plasma to map "spatial ecotypes".

These approaches are incredibly powerful, but they require specialized wet-lab workflows that are distinct from standard whole genomic sequencing.

In Deep learning-based non-invasive profiling of tumor transcriptomes from cell-free DNA for precision oncology, Gavin Ha’s team from Fred Hutch and WashU present an approach that means we may not need complex biochemistry at all?

The WGS+AI Disruptor: Triton/Proteus

The Hutch team introduced Triton/Proteus, an AI-driven framework where Triton uses standard (30-12x) plasma WGS to determine the fragmentome and inferred nucleosome positioning patterns of promoters and gene bodies, and then Proteus estimates tumor fraction and predicts RNA-Seq equivalent expression scores for Gene-Set Enrichment (e.g. PAM50), drug target monitoring (e.g. STEAP2, DLL3, NECTIN4) and molecular subtyping (e.g. neuroendocrine-state or SPAIN for SCLC).

No antibodies. No specialized pulldowns. Just smart dry-lab deep learning applied to standard wet-lab sequencing - it’s #moistlab compatible (anyone else a SwampThing fan?). By analyzing the physical “bumps and cuts” of cell-free DNA fragments, the AI infers which genes are actively being transcribed.

To build their predictive models, the researchers used matched tissue RNA-Seq as the “ground truth” to teach their AI framework how to read the WGS fragmentome and inferred nucleosome data. They did this by feeding both datasets into the deep learning model to correlate specific physical structures with actual transcriptomic activity. For instance, the model learned that a lack of fragments (indicating a nucleosome-depleted, “open” region) directly at a promoter, combined with specific fragment-length phasing across the gene body, strongly correlates with high RNA-Seq expression for that gene.

Once the model was fully trained on this paired data, the RNA-Seq “training wheels” were removed. The resulting algorithm can now take standard plasma WGS alone, analyze the fragmentome and nucleosome footprints, and accurately output an inferred RNA-Seq-equivalent expression score. However, the team is clear to point out that there is a way to go before this could be used for MRD and early cancer detection.

So if this method holds in larger validation studies it’ll be interesting to see how it compares to the approaches from other FLBx companies.

Fragmentome as a diagnostic modality:

When I heard fragmentomics and cancer my first thought was of DELFI Diagnostics (DNA evaluation of fragments for early interception). But there are big differences in what DELFI is doing compared to the Triton/Proteus approach.

DELFI utilizes shallow WGS (sWGS), typically around 1x to 2x depth. It looks at massive, megabase-scale bins to find global fragmentation shifts and copy number variations. It is a macro-level tool optimized for binary classification: Is cancer present or not? This is the underpinning of their FirstLook Lung test, a rule-in test for USPTF high-risk individuals to help decide if they needs a low-dose CT (LDCT) scan as part of national lung cancer screening programs. It has an NPV of >99% and a postivie results should help push screening-recalcitrant indivduals to attend the CT-scan.

Triton/Proteus, however, requires deep WGS, 30-120x depth because it tries to resolve more precise information about specific cancer genes like HER2 or TROP2. It’s intersting to think about whether a FirstLook Lung could refelx after a positive result, to re-sequencing at high-depth to deliver a Triton/Proteus FLBx analysis of the tumor biology? Obviously this is a big stretch, especially so in early cancer detection, but as USPTF screening is aimed at stage shifting in advanced disease from IV to III (as well as detecting early localized disease (Stage I and II)), then thos advanced cancers may have Tumor Fractions high enough to do the FLBx?

Generalist vs. Specialist: FLBx Niches or a future Monopoly?

This raises a question for the FLBx landscape: Can a deep WGS algorithm do everything reasonably well, or do the complex biochemical methods still do specific things very well?

The Generalist (WGS): Deep WGS is highly scalable. The commercial reality here is massive. While Fred Hutch hasn’t announced a NewCo (that I am aware of), the authors have filed a patent. WO2022217096A2 (many of the authors are named) appears to cover the core use of deep learning models to infer real-time, single-gene expression directly from fragmentomic patterns and nucleosome footprints in standard-depth cell-free DNA sequencing (maybe Nava could take a deeper dive over at his ASeq substack?).

The Specialist (cfChIP-Seq / cfDNAac): While WGS offers an elegant software solution, some sort of physical enrichment or molecular biology still holds a biological advantage for specific niches. Precede boosts the signal-to-noise ratioby pulling down only the active chromatin, cfChIP-Seq, and Aqtual gets to a simlar position by modifying the pre-analytical cfDNA extraction to isolate the larger, "open" chromatin fragments that still have transcriptional machinery attached to them (I need to do a deeper-dive on Aqtual). When hunting for ultra-rare lineage plasticity, subtle enhancer rewiring, or when dealing with exceptionally low tumor fractions, the specialized biochemical approaches may well show superior sensitivity.

Will the FLBx field split into specialized niches? Ultimately, the quest continues for the perfect balance of scalability and depth, but it remains to be seen if the market will sustain this ecosystem of specialized assays, or if a single approach will eventually become the “one FLBx to rule them all” (I’m not just a comic book fan) - maybe Resero Bio’s RARE-Seq.

But to assess TME you need access to tissue and for a cancer patient that usually means FFPE from the surgical resection or biopsies. But a new company on the #LBx scene, LiquidCell Dx, appears to translate spatial tumor biology into a simple blood draw!

What’s the data look like: Two posters caught my attention at American Association of Cancer Research from the company, which I recently posted on LinkedIN and I thought I’d highlight them both here.

Poster #1:

Abstract 94: Liquid biopsy profiling of the tumor microenvironment to determine response to immunotherapy regimens across solid tumors, details the blinded clinical validation of their new LiquidTME assay in advanced melanoma. Instead of just looking for tumor mutations, this approach uses deep learning on plasma cfDNA methylation to infer SpatialEcotypes (SEs; multicellular spatial states natively found within the tumor microenvironment).

In a blinded cohort of 34 metastatic melanoma patients treated with combination immunotherapies, these noninvasively detected SEs successfully stratified responders from non-responders. And it maintained its performance in patients where standard TMB profiling failed to stratify progression-free survival; in fact the comparison of LiquidTME to TMB showed that the former had a better HR.

The data were generated from 2ml plasma and just 15x EM-Seq (see my earlier post on the "systematic errors in enzymatic conversion" preprint).

This poster tantalises us with a citation for a paper in press at Nature to watch out for: Zhang*, Brown*, Usmani*, et al. Non-invasive profiling of the tumour microenvironment with spatial ecotypes. DOI: 10.1038/s41586-026-10452-4.

So, can a simple blood draw capture the complex spatial organization of a tumor microenvironment? And are we finally moving beyond TMB and PD-L1 for immunotherapy patient selection? Onto poster #2!

Poster #2:

Abstract 1040: Blinded clinical validation of LiquidTME, a cell-free DNA assay for predicting response to immunotherapy by noninvasively profiling the tumor microenvironment.

This second poster highlights the work LiquidCell Dx has been doing in developing Spatial Ecotypes. They discovered 9 SEs, including some associated with IO response and resistance, in an analysis of almost 1250 RNA-seq profiles from carcinomas and melanoma. In this poster they describe the transfer of Spatial Ecotype profiling from tissue to plasma cell-free DNA methylation and extension into NSCLC and Bladder.

By measuring SE’s directly from pretreatment plasma cfDNA, such as the immune-rich SE7 (associated with durable benefit) or the fibroblast-driven SE4 (associated with resistance), the team demonstrated striking associations with IO response across multiple solid tumors.

In the 25 patient NSCLC cohort (Stage II-IV) they showed that SE’s were better predictors of outcome (PFS or OS) than PD-L1 or TMB. And in a 10 patient Bladder cancer cohort the pretreatment plasma levels of SE’s associated with pathologic complete response with high AUC.

What next for #FLBx?

Both posters report small clinical cohorts but the initial results are promising. I look forward to larger datasets and some link to tumor fraction as I’d like to know when this assay stops working or whether the lead time for seeing an emerging SE might be better or worse than an emerging mutation to predict resistance!

If you’re interested in the burgeoning new field of Functional Liquid Biopsy #FLBx, or just curious about the potential to rationally design IO combinations or track “TME plasticity” longitudinally without invasive biopsies, then check out my other posts about #FLBx on LinkedIN or watch for more here on my Substack.

But a fascinating new preprint from Robert Patton, Peter Nelson, and Gavin Ha et al. at Fred Hutch proposes a new tool to predict individual gene expression levels from cfDNA in a manner that is directly analogous to tissue RNA-Seq.

In Deep learning-based non-invasive profiling of tumor transcriptomes from cell-free DNA for precision oncology, the authors describe a pair of tools, TRITON & PROTEUS to extract the tumor transcriptome from the standard Whole Genome Sequencing (WGS) data we already use every day.

Enter The AI-Driven Transcriptome

Current methods for inferring gene expression from ctDNA require specialized assays or ultra-deep, targeted sequencing, which preclude transcriptome-wide profiling at single-gene resolution. To solve this, the researchers built a two-part computational framework that relies entirely on standard depth (~30-120x) WGS of cell-free DNA:

1. Triton (The Signal Extractor): This tool comprehensively extracts fragmentomic and nucleosome positioning features at base-pair resolution. Crucially, it looks beyond just the promoter region and analyzes the full gene body to capture a complete picture of active chromatin remodeling.

2. Proteus (The AI Translator): Proteus is a multi-modal deep learning framework that takes Triton’s spatial signals and translates them into predicted single-gene expression levels. Uniquely, it outputs these predictions in standard units of counts per million, making it directly analogous to standard tissue RNA-Seq outputs.

The Proof is in the Plasma

The team didn’t just build an algorithm; they initially validated it on a mix of simulated data and original ctDNA samples from PDX and patient plasma and compared it to matched tumor RNA-Seq. They did this in castration-resistant prostate cancer (CRPC), small-cell lung cancer (SCLC), and bladder cancer (BLCA) and found that the ctDNA WGS predicted transcriptome (18,488 genes) strongly correlated with RNA-Seq (r = >0.9, see thier Fig. 3a), which is similar to replicates in tissue.

Proteus predictions were significantly correlated with matched tissue-derived gene expression measures across samples in each cohort. And they describe the application in a couple of scenarios that are quite exciting:

• SCLC subtyping: They observed significant correlations of gene expression in PDX of SCLC for NEUROD1, ASCL1, and POU2F3 which define SCLC transcriptional subtypes - an indication where LBx subtyping could impact therapy selection. PS: Siyu Liu presented our work in SCLC subtyping at AACR’25 along with researchers in CRUK Manchester we assessed TWIST methylome and T7-MBD-seq methods to show that liquid biopsies can offer a non-invasive method for subtyping SCLC, potentially improving personalized treatment feasibility.

• ADC Targets: Proteus successfully predicted the expression of key actionable targets right from the blood. This included STEAP1 in CRPC , DLL3 in neuroendocrine tumors , and NECTIN4—the validated target for the FDA-approved ADC Enfortumab vedotin—in bladder cancer.

• Radioligand Resistance: Similar to Precede’s ASCO data, Proteus proved it could track pathway-level resistance. In a cohort of CRPC patients evaluated prior to receiving Pluvicto (177Lu-PSMA-617), Proteus identified that a neuroendocrine signature (NE.10) and cell-cycle progression signatures were associated with increased risk for progression and therapeutic resistance.

• Breast cancer subtyping: I’d have seen this as an obvious one to test but there’s nothing on comparisons to Mammaprint or Oncotype Dx assay, which use 70-gene or 21-gene signatures urespectively. Whilst these are primarily used in early-stage breast cancer to guide adjuvant chemotherapy, I wonder if they could have been applied in high cfDNA samples as a test case for some of the few Dx that rely on Gx?

A Word of Caution (Again)

Just like the cfChIP-Seq assays I discussed previously, there are biological limits to what can be seen in the blood. Deep learning models are inherently restricted by the variability in observed data. The researchers themselves note that additional optimization and training will be required to apply Proteus in the settings of minimal residual disease (MRD) and early cancer detection. This technology shines, as do the other Functional Liquid Biopsies, in advanced, metastatic settings where tumor fraction is higher, and where tissue biopsies are often too dangerous, difficult, or outdated to rely on.

We are watching the rapid evolution of Functional Liquid Biopsy (apparently I even coined the term in my last post!). Lead on Mac Duff.

On last week’s “Defining the Next Frontier in MRD” webinar, organised by Decibio, I was asked about the need for standardisation in MRD; I referred everyone listening to the original MIAME paper from 2001 and said maybe we needed a MIMAE for MRD...so I built one.

Let’s start at the very beginning…back in 1999, the genomics world was facing a bit of a data crisis. Microarray technology was booming and everyone was having Chips with everything, but everyone was also reporting their data differently. It was sometimes impossible to reproduce experiments or confidently compare results across different studies.

Enter the MIAME standard (Minimum Information About a Microarray Experiment). Published in 2001, it established the critical elements needed to meaningfully interpret a microarray experiment. MIAME completely transformed the scientific landscape. It birthed a technical XML format for software communication (MAGE-ML) in 2002. And More importantly, it kicked off a broader “Minimum Information” movement across biology, including the MIQE guidelines (Minimum Information for Publication of Quantitative Real-Time PCR Experiments) in 2009, to ensure the reliability, reproducibility, and transparency of qPCR data. The 2012 saw the MINSEQE (Minimum Information About a High-throughput SEQuencing Experiment) published as the standard for reporting RNA-seq and other sequencing data. And most recently, the MUMIE (Minimal Urine Methods in Experiments) framework is a set of guidelines designed to standardize the reporting of key pre-analytical factors in urine collection, processing, and DNA extraction for cancer detection. Also see Wikipedia for 24 standards!!!

Now, we are facing the exact same problem in a new, critical field: MRD.

The Standardization Gap in MRD

Molecular Residual Disease (MRD) detection using circulating tumor DNA has emerged as a powerful biomarker in solid-tumor oncology. However, the rapid advancement of this technology has completely outpaced our reporting standards.

Even at major oncology conferences, researchers routinely present MRD data without including fundamental details like assay name, specifications, the limit of detection (LOD), or standard sample collection timepoints. This inconsistency creates massive roadblocks:

• Cross-study comparisons become nearly impossible without standardized reporting.

• Clinicians struggle to interpret MRD results without clear assay specifications.

• Regulatory review is complicated by variable reporting standards.

• Meta-analyses are prevented by a lack of structured, comparable data.

Proposing MIAME/MRD

Major organizations like the BLOODPAC Consortium and the FDA are actively establishing technical and regulatory baselines for the rapidly growing liquid biopsy space. Specifically, BLOODPAC recently published consensus protocols for analytical validation and a standardized lexicon, such as recommending “Molecular” over “Minimal” for solid tumors, while the FDA has issued draft guidance focused on standardizing methodologies and assay considerations for clinical trials.

But, just as the original MIAME framework fixed microarray reporting, I think we need a MIAME/MRD standard to ensure reproducibility and comparability in liquid biopsy research.

In my AI-generated concept draft, I proposed six core components that a true MIAME/MRD standard should require:

• Clinical context and study design: Documenting cancer type, stage, treatment history, study design, and clinical endpoints.

• Sample collection and processing: Detailing collection timepoints, blood and plasma volumes, tube types, and cfDNA extraction protocols.

• Assay specifications and methodology: Defining the platform, methodology, variant selection, and whether the approach is tumor-informed or tumor-agnostic.

• Analytical performance characteristics: Explicitly stating the limit of detection (LOD95), analytical sensitivity/specificity, and precision.

• Data processing and analysis: Outlining bioinformatic pipelines, variant calling, MRD detection logic, and quality control metrics.

• Results and quality metrics: Reporting per-sample MRD status, quantitative results, and confidence intervals.

A Call to Action

Widespread adoption of a reporting standard will accelerate clinical translation and enable meaningful meta-analyses across studies. But an AI-generated PDF isn’t going to get us there.

Reach out in the comments or send me a message if you want to collaborate on making MIAME/MRD a reality!

• Berchuck ‘25 (ASCO) – Radiopharmaceutical Resistance (mCRPC): This study showed the platform can quantify PSMA expression from blood with high accuracy. Crucially, it revealed that patients with elevated Wnt signaling and lower immune pathway activity at baseline had significantly poorer responses to 177Lu-PSMA-617 (Pluvicto).

  • Read the ASCO 2025 Presentation Summary

• Beagan ‘25 (ASCO) – ADC Resistance in Breast Cancer: Focusing on the TROP2-targeting ADC Sacituzumab Govitecan, Precede identified that specific tumor-intrinsic transcriptional programs—including EMT (Epithelial-to-Mesenchymal Transition) and proliferative pathways—are strong predictors of drug resistance in HR+/HER2- metastatic breast cancer.

  • Read the Breast Cancer Resistance Data

• Barrett ‘25 (ESMO) – SCLC Subtyping and Targets: In Small Cell Lung Cancer, Precede demonstrated the ability to non-invasively resolve expression for key targets like DLL3, SEZ6, and CEACAM5. The platform also successfully classified SCLC into its molecular subtypes (ASCL1, NEUROD1, POU2F3), which is critical for selecting next-gen precision therapies.

  • Read the ESMO 2025 SCLC Data

• Verjee ‘25 (AACR) – Genome-Wide Expression Scale: This technical milestone proved the platform can systematically predict the expression of over 2,500 tumor-specific genes from 1mL of plasma. This includes key ADC and immune targets such as NECTIN4, B7H4, HER3, and MUC1, along with master regulators like AR and FOXA1.

  • Read the AACR 2025 Scalability Data

• Morganti ‘24 (SABCS/ASCO) – HER2 & ER Pathway Status: Precede showcased its ability to accurately classify HER2 status across multiple solid tumors and introduced the PERDI (ER Transcriptional Dependency Index). This index goes beyond traditional staining to determine if a breast cancer is truly “dependent” on estrogen signaling, guiding the use of oral SERDs vs. ADCs.

  • Read the HER2/ER Status Publication

Are you considering using Functional Liquid Biopsy - tell me what you think it's good for in the comments.

The post How big would ABI377 plates need to be to deliver NovaSeq X Plus output? appeared first on Enseqlopedia.

The post cfDNA fragmentomics reviewed appeared first on Enseqlopedia.

The post Unlocking the Potential of Urine-Based Liquid Biopsies appeared first on Enseqlopedia.

This vision is rapidly moving from concept to reality, driven by advancements in both molecular diagnostics and, critically, revolutionary methods of blood collection. The promise lies in enabling cancer monitoring to be more accessible, less intrusive, and seamlessly integrated into a patient’s life.

PS: This post was inspired by Brad Evans (Merck) and Jordan Feeny (Natera)’s recent poster at ASCO’25: Feasibility of ctDNA detection with a low-volume blood collection device – using Tasso+. And it follows on from a blog post about longitudinal monitoring of ctDNA from blood spots, but I’ve dug a little deeper into what’s out there today even if there’s not a ton of ctDNA data yet!

The Power of Molecular Assays: A Quick Look

For many of my readers, the significance of molecular assays that use circulating tumor DNA (ctDNA) detection is well understood. These powerful tools offer non-invasive insights into a patient’s cancer, enabling molecular diagnostics, minimal residual disease (MRD) detection and treatment response monitoring for early recurrence detection. Currently, high-sensitivity ctDNA assays typically rely on larger volume venous blood draws, usually one or two 10-20 mL tubes of whole blood that are double-spun, to ensure sufficient material for detecting these often scarce biomarkers. The challenge, then, is to achieve this same level of precision with significantly smaller, patient-friendly samples.

Emerging At-Home Blood Collection Methods: A Revolution in Miniaturization

The true game-changer for at-home oncology testing lies in a new generation of blood collection devices designed for patient self-collection with minimal training. These innovations are reshaping how we think about obtaining diagnostic samples:

1. Capillary Blood & User-Friendly Microcollection:

While traditional microcollection tubes have been around for a while, newer systems are specifically engineered for ease of home use, collecting capillary blood (typically from a finger or forearm stick) with improved accuracy and patient comfort.

• BD MiniDraw™ Capillary Blood Collection System: Designed with simplicity in mind, this system aims to make capillary blood collection straightforward for patients, yielding approximately 1 mL of whole blood.

• RedDrop One: This device offers a unique approach to self-collection of capillary blood from the forearm, emphasizing ease and consistency for diagnostic purposes, with volumes typically ranging from 10 to 50 µL.

2. Volumetric Microsampling: Precision in Tiny Drops:

Perhaps the most exciting frontier for oncology monitoring, these devices collect extremely precise, small volumes of blood, often then dried onto a substrate for stable transport. This combination of minimal invasiveness and volumetric accuracy is key.

• TASSO Devices (TASSO+, TASSO-SST, TASSO-M20): Tasso’s range of devices allows for patient self-collection of whole blood, serum, or micro-samples from the upper arm. The TASSO+ can collect 200-600 µL of whole capillary blood, TASSO-SST collects 150-500 µL of serum, and the TASSO-M20 provides four precise ~17.5 µL micro-samples. Their “push-button” activation simplifies the process, making it highly amenable to remote use.

• Mitra® Devices (Neoteryx): Utilizing Volumetric Absorptive Microsampling (VAMS) technology, Mitra devices precisely collect a fixed volume of blood (typically 10-30 µL per tip) from a fingerstick. This volumetric precision is crucial for quantitative analyses, and the dried format aids stability during shipping.

• hemaPEN® (Trajan Scientific and Medical): This innovative device enables the collection of four exact 10 µL dried blood samples. Its multi-sample capability from a single fingerstick is valuable for repeat testing or multi-analyte assays.

• Capitainer® B 50: Capitainer devices are designed for highly accurate, fixed-volume dried blood spot collection (50 µL) from a fingerstick. They ensure that the exact volume of blood is absorbed, addressing a common variability challenge with traditional DBS cards.

• Timepoint LCM: An at-home device for for collection and stable recovery of ctDNA from upper arm capillary blood (500 ul)

Prominent Blood Collection Methods and Devices

The “Ifs” and “Buts”: Overcoming Challenges for Oncology

While the collection technology is advancing rapidly, significant hurdles remain, particularly when considering highly sensitive molecular oncology assays:

• The Volume Challenge: Current high-sensitivity ctDNA assays are often optimized for larger plasma volumes derived from 10-20 mL of whole blood. These micro-collection methods yield blood volumes in the microliter range. The critical question is whether molecular oncology assays can be re-optimized or new ones developed that maintain the same diagnostic sensitivity and specificity with significantly smaller input volumes. This is an intense area of ongoing research.

• The “Low Levels” Challenge: ctDNA is often present at extremely low concentrations, especially in the MRD setting or in early-stage cancers. The sensitivity of a molecular assay is directly linked to the amount of input material. Achieving the same robust limit of detection (LOD) and quantification (LOQ) with lower input volumes would be great but may be impossible. While recent data, from Personalis e.g. ESMO 2024 TRACERx NeXT Personal from Charlie Swanton, or Natera e.g. their ASCO 2025 presentations on the Signatera Genome assay, or Foresight Diagnostics e.g. their ASCO 2025 data on PhasED-Seq in DLBCL, demonstrates phenomenal sensitivity in detecting ctDNA across various cancer types, it’s vital to note that these studies typically leverage standard, higher-volume venous blood draws. Future work must validate if this same level of clinical sensitivity can be achieved using micro-samples from home-based collection devices, or if those highly-sensitive tests can still be useful if constrained by blood volume.

• Sample Quality and Stability: Ensuring the integrity and stability of delicate biomarkers like ctDNA during transit from a patient’s home to the lab, especially when collected in a dried format, is another key consideration. Appropriate preservatives and storage conditions are essential.

• Regulatory and Clinical Validation: Rigorous clinical validation studies demonstrating the equivalence or superiority of these at-home methods to traditional venipuncture are indispensable for widespread adoption and regulatory approval in oncology.

The Transformative Potential: Why It Matters

Despite these challenges, the potential impact of at-home blood collection in oncology is immense:

• Patient Empowerment: Granting patients greater control and comfort in their care journey.

• Increased Access: Breaking down geographical and mobility barriers, ensuring patients in rural areas or with limited mobility can still receive cutting-edge monitoring.

• More Frequent Monitoring: Easier access could enable more frequent testing, potentially allowing for earlier detection of disease changes or recurrence, leading to timelier intervention.

• Enabling Decentralized Clinical Trials: A cornerstone for modern cancer research, allowing broader patient participation regardless of location.

Conclusion: The Road Ahead

The convergence of innovative, low-volume blood collection technologies and increasingly sophisticated molecular assays promises a new era for oncology diagnostics. While the scientific community is actively working to overcome the technical challenges related to sample volume and ultra-low ctDNA levels, the advancements are undeniable – and I’d call out the Priming Agents work published last year (see my blog) and being spun out by Amplifyer.bio. As such, I believe the vision of oncology patients confidently managing parts of their care from home, leading to earlier insights and more personalized treatment paths, is not just a dream – it’s becoming a tangible goal, set to redefine the patient experience in cancer care.

FYI: This post The Future of At-Home Oncology Testing: Tiny Drops, Big Insights? appeared first on Enseqlopedia.

The post The word according to Fred appeared first on Enseqlopedia.

The post Roche emerges from the Lazarus pit appeared first on Enseqlopedia.

The post cfTAPS: a new method for cfDNA epigenomes appeared first on Enseqlopedia.