A Surviving Beta Cell Subpopulation Enriched in Patients with T1D- Max Spurrell MD PhD Candidate
On Demand video from the June 16 2026 talk
🎬 ON DEMAND VIDEO
A Surviving Beta Cell Subpopulation Enriched in Patients with T1D Max Spurrell MD PhD Candidate · June 16, 2026
📺 Now available on demand ▶ Watch the Full Talk📄 Read the Preprint
💬 Key Quote
“If we can identify the important features helping these cells survive, we could potentially leverage that information to engineer immune-resistant beta cells. That’s a really hot topic in the field right now — and here we have a naturally surviving population to learn from.”
— Max Spurrell
🔬 Foundational Insights as They Apply to T1D
The canonical story of T1D is a story of destruction — autoimmune T cells systematically eliminate insulin-producing beta cells until essentially none remain, and the patient becomes dependent on exogenous insulin for life. But that story is not entirely right, and the exceptions are biologically significant. A landmark 2010 study of the Joslin Medalist cohort — patients who had lived with T1D for at least five decades — found that roughly two-thirds still had detectable C-peptide, a marker of endogenous insulin production. A follow-up histological study staining for beta cells in cadaveric tissue from Medalists found residual insulin-positive cells in all 68 donors examined. Something protects these cells. The question is what.
Maxwell Spurrell, an MD/PhD candidate in the Department of Immunobiology at Yale, co-mentored by Drs. Kevan Herold and John Tsang, set out to answer that question using the most powerful single-cell resource the T1D field currently has: the Human Pancreas Analysis Program (HPAP) dataset. Rather than applying conventional gene-expression-based clustering — which, as Spurrell showed early in his talk, is severely confounded by technical noise arising from differences in 10x Genomics kit chemistries across the HPAP collection — he took a methodologically distinct approach. He applied SCENIC, a gene regulatory network (GRN) inference framework, to convert the standard cell-by-gene expression matrix into a cell-by-transcription-factor-activity matrix. This rank-based, diffusion-model approach is inherently more robust to the ambient RNA contamination and batch variation that plague dissociation-based single-cell data — and crucially, it asks a different biological question: not which genes are expressed, but which regulatory circuits are active.
The payoff was a discrete beta cell cluster — designated Cluster 3 (C3) — that emerged cleanly from GRN-based dimensionality reduction and was strikingly enriched among beta cells from T1D donors. In non-diabetic and autoantibody-positive donors, C3 beta cells comprised roughly 3–4% of the beta cell pool. In patients with established T1D, the median frequency rose to approximately 15% — representing, by definition, the surviving beta cell population in human disease. Spurrell recovered thousands of beta cells from 10 of 12 T1D donors in the dataset, enough to characterize the C3 state with confidence.
The transcriptional and regulatory identity of C3 was internally consistent and externally validated. The top transcription factors defining C3 activity — including IRF1, BCL6, and several others with established roles in beta cell survival and immunomodulation — had their transcripts also enriched at the mRNA level, providing an internal cross-check. Immunomodulatory genes were prominently upregulated: SOCS1 and SOCS3 (negative regulators of cytokine signaling), HLA-E (an inhibitor of NK and T cell killing), and CD47 (the “don’t eat me” signal being exploited in immune-evasion engineering strategies). Simultaneously, C3 cells showed evidence of dedifferentiation — downregulation of key beta cell lineage transcription factors including MAFA, MAFB, and NKX6.1, and reduced expression of insulin processing and secretion genes, some of which are also T1D autoantigens. The picture that emerged was a beta cell trading functional identity for immune invisibility: less antigen displayed, less secretory machinery advertised, more immune checkpoint signaling activated.
To understand what drives C3 formation, Spurrell leveraged a published dataset from the Millman lab in which primary human islets from non-diabetic donors were exposed to a panel of stress stimuli — ER stressors and inflammatory cytokines, alone and in combination — followed by single-cell RNA sequencing. The cytokine mixture of interferon, TNF, and IL-1 most faithfully reproduced the C3 gene expression signature: upregulating the C3-enriched genes and downregulating the C3-depleted genes. Importantly, each cytokine individually could also shift cells in the C3 direction, with the combination being most potent. This implicates the chronic inflammatory cytokine environment of the insulitic islet — not some intrinsic pre-existing property of a rare beta cell variant — as the likely environmental driver of the C3 state.
A final experiment tested whether the C3 program was beta cell-specific or a broader tissue response. Spurrell repeated the entire GRN inference and clustering pipeline on alpha cells from the same HPAP donors. An alpha cell Cluster 3 emerged with a nearly identical transcription factor activity profile and a nearly identical set of differentially expressed genes — upregulating the same immunomodulatory factors, downregulating the same lineage genes. And across donors, the frequency of C3 beta cells and C3 alpha cells correlated strongly: when one was high, the other was high; when one was near zero, so was the other. The C3 program is not a beta cell fate — it is a shared endocrine cell response to cytokine exposure, acting on whatever islet cell types happen to be in the inflamed microenvironment.
🎯 Core Premise
A discrete subpopulation of beta cells — identifiable by gene regulatory network inference rather than conventional expression clustering — is selectively enriched among the surviving beta cells of patients with established T1D. This Cluster 3 (C3) population is defined by elevated activity of IRF1 and related transcription factors, upregulation of immunomodulatory genes including SOCS1/3, HLA-E, and CD47, and partial dedifferentiation marked by reduced autoantigen and secretory gene expression. Inflammatory cytokines can drive normal beta cells into a C3-like state in vitro, implicating the insulitic cytokine environment as the inducer of this phenotype in vivo. The same program is active in a correlated fraction of alpha cells from T1D donors, suggesting a cell-extrinsic, tissue-wide response rather than a beta cell-intrinsic fate. C3 may represent a natural immune evasion program — one that could be studied, validated, and ultimately leveraged to engineer more immune-resistant beta cells for T1D therapy.
🌟 Why This Talk Matters to T1D Scientists and Clinicians
Nature’s own immune-resistant beta cell — now characterized at the transcription factor level.
For scientists: The application of GRN inference (SCENIC) to identify beta cell subpopulations rather than clustering on raw gene expression is a methodological contribution that should propagate through the field. Conventional clustering finds groups defined by what genes are expressed; GRN-based approaches find groups defined by which regulatory circuits are active. For a transcriptionally stressed and partially dedifferentiated population like surviving T1D beta cells, the regulatory layer is more stable and more informative than any individual gene. The identification of IRF1 as the dominant node in the C3 survival program opens a direct mechanistic hypothesis with multiple molecular handles — and the observation that a prior study from the Avezri lab found IRF1-positive beta cells focally concentrated in T1D islets but absent from non-diabetic tissue provides independent histological corroboration without any experimental overlap. The alpha cell finding adds a further layer: the C3 program is not a clonal beta cell variant but a cytokine-induced state, which means it is in principle inducible, modifiable, and potentially separable from the dedifferentiation that accompanies it.
For clinicians: Residual beta cell function — measured by C-peptide — is not just a research curiosity. It is a clinically meaningful phenotype associated with reduced hypoglycemia risk, more stable glucose control, and better long-term outcomes in T1D. Teplizumab, developed by Spurrell’s mentor Kevan Herold and now the first FDA-approved drug to delay T1D onset, works in part by preserving this residual function. Understanding what allows some beta cells to survive immune attack is therefore directly relevant to understanding why teplizumab works in some patients and not others, and to designing combination strategies that could augment or extend its effect. The C3 program — and IRF1 in particular — gives the field a molecularly specific target for that inquiry, linking the immunotherapy clinical program to a defined cellular mechanism for the first time.
The broader picture: The field of immune-resistant SC-islet engineering is currently built almost entirely on empirically derived combinations — beta-2 microglobulin knockout, CIITA knockout, CD47 overexpression, PD-L1 overexpression — chosen based on general immunological principles rather than on what surviving human beta cells in actual T1D actually do. Spurrell’s paper offers something the field has not had before: a biologically grounded, human-tissue-validated molecular blueprint of immune resistance drawn from nature’s own experiment. The C3 program — with its convergence of SOCS signaling, HLA-E upregulation, CD47 expression, autoantigen downregulation, and IRF1-driven transcriptional coordination — is not a theoretical construct. It is what the surviving cells in a human who has had T1D for a decade actually look like. That is the right place to start engineering from.
3️⃣ Big Takeaways
Surviving human T1D beta cells are not a random remnant — they are a transcriptionally defined state, and GRN inference is what makes them visible. Conventional expression-based clustering of the HPAP beta cell data is severely confounded by batch effects driven by differences in 10x kit chemistry across the multi-year HPAP collection. Standard integration methods like Harmony tend to overcorrect, merging real biological signal along with technical noise. Spurrell’s GRN inference approach sidesteps this problem: because SCENIC uses a rank-based diffusion model to score transcription factor activity, it is inherently more resistant to the ambient transcript contamination and batch structure that distort raw expression matrices. The result was a cluster — C3 — that is cleanly distributed across patients rather than tracking with technical variables, and that is enriched approximately fivefold in T1D donors relative to non-diabetic controls. Klaus Kaestner, who directs the HPAP initiative and was present in the discussion, explicitly praised the analytical approach and acknowledged the batch noise as a known limitation his group has long grappled with. The methodological contribution here is not just for beta cell biology — it is a demonstration that GRN inference can extract biologically coherent signal from noisy, multi-batch human tissue datasets where conventional approaches fail.
IRF1 sits at the center of the C3 survival program — and its focal enrichment in T1D islets has already been independently confirmed in human tissue. IRF1 — interferon regulatory factor 1 — was the top transcription factor defining C3 activity in the GRN analysis, with transcript-level enrichment confirming what the regulatory score suggested. Multiple published papers from the Avezri lab have demonstrated pro-survival and immunomodulatory roles for IRF1 specifically in beta cells in immune contexts. And a 2018 histological study, cited by Spurrell during the live discussion, immunostained human pancreatic sections for IRF1 and found that IRF1-positive beta cells were present focally in T1D donor islets but absent from non-diabetic tissue — an independent line of evidence entirely consistent with the HPAP computational finding. IRF1 activates SOCS1 and SOCS3 to dampen cytokine signal transduction, suppresses inflammatory chemokine production that would otherwise recruit additional immune effectors, and may regulate autoantigen expression. The convergence of computational, transcriptomic, and histological evidence around IRF1 as the key node makes it the most tractable immediate target for mechanistic follow-up.
The C3 program is a cytokine-induced tissue response, not a clonal beta cell variant — which means it is in principle inducible in any beta cell exposed to the right environment. The finding that the same transcriptional program appears in a correlated fraction of alpha cells from T1D donors — cells that are not being targeted by autoimmunity — was one of the most conceptually clarifying results of the talk. If C3 were an intrinsic survival fate restricted to a pre-existing rare beta cell subtype, it should not appear in alpha cells. Its presence there, with strongly correlated frequency across donors, points to a cell-extrinsic signal — cytokines produced in the inflamed islet microenvironment — acting on all endocrine cell types present. The in vitro cytokine stimulation experiment from the Millman lab dataset confirmed that interferon, TNF, and IL-1 can drive naive non-diabetic donor beta cells into a C3-like transcriptional state. Bart Roep from Leiden, during the live discussion, made the important conceptual point that Spurrell had moved between describing a “beta cell state” and a “different beta cell type” — and Spurrell agreed that the data support a continuous state interpretation rather than a discrete clonal population. This distinction matters enormously for the translational agenda: a state can be induced, a clonal type cannot.
❓ Key Questions from the Discussion
Are the surviving C3 beta cells found in typical islets, or as single cells or small beta cell clusters? Spurrell acknowledged this as an important limitation of the dissociation-based 10x Genomics approach: when tissue is processed for droplet-based single-cell sequencing, spatial coordinates are lost entirely. He could not determine from the HPAP data whether C3 beta cells are scattered uniformly across islets, concentrated in particular islets, or distributed as isolated single cells. He flagged spatial localization as a key future direction — and noted that a 2018 histological study from the Avezri lab, using immunofluorescence co-staining for insulin and IRF-1, found that IRF1-positive beta cells were focal, enriched in certain islets but not broadly distributed — providing an early hint that the C3 program may reflect local microenvironmental variation rather than a tissue-wide response.
It’s interesting that you find P21 in surviving beta cells — it has been suggested that elimination of senescent beta cells delays diabetes in NOD mice. Do you think these are senescent cells? Spurrell was careful not to claim senescence. P21 was the only classically senescence-associated gene he observed, and he noted that P21 is also a well-established interferon-stimulated gene — any cell type treated with interferon will upregulate P21, independent of senescence. Establishing senescence properly requires functional assays such as beta-galactosidase staining, which the cross-sectional HPAP data cannot provide. He concluded that “maybe” is the most appropriate characterization. Kevan Herold added an important nuance: the role of P21 in the NOD senescence story is about immune sensitization leading to killing, whereas in this context P21 is appearing in the context of beta cell survival — and there are published data showing that P21 can reduce ER stress and improve beta cell function in T2D, making the biological interpretation quite different. Herold suggested these are distinct questions that should not be conflated.
Have the alpha cell clusters been scored to see whether they are more related to the top or bottom beta cells from the original NOD mouse study? Spurrell noted that published RNA-seq data on sorted top and bottom beta cells from NOD mice does show some shared features — including dedifferentiation and interferon-stimulated gene expression — with the C3 population, but the one-to-one gene mappings are not direct. Species differences, technical differences between platforms, and the fact that the NOD bottom cell population was defined by flow cytometry scatter properties rather than transcriptional criteria all make direct comparison difficult. He described the similarities as directionally consistent and conceptually encouraging without claiming equivalence.
You’ve moved between calling these different “beta cell states” and different “beta cells” — I think you should stick with state. And separately: these T1D donors are on insulin, so their beta cells don’t need to function. Couldn’t the dedifferentiation simply reflect that they don’t need to secrete? On the state vs. type question, Spurrell fully agreed. He noted that the C3 transcription factor activity scores all load onto the same principal component, suggesting a continuous axis of variation rather than a discrete cluster — and that discrete clustering is an analytical convenience rather than a biological claim. On the exogenous insulin confound, Spurrell acknowledged it as real but noted that many HPAP T1D donors were individuals who presented in severe diabetic ketoacidosis and passed away acutely — they were not on stable insulin regimens prior to death. With only 12 T1D donors in the dataset, he argued that distinguishing between those with and without prior insulin exposure would produce subgroups too small for any meaningful analysis, though he agreed it was an important potential confounder to track in future larger cohorts.
The cytokine in vitro data is a nice correlation — but how comfortable are you that an in vitro cytokine response actually mimics what’s happening in the human pancreas in T1D? Spurrell was deliberately measured. He described the in vitro signal as “C3-like” rather than identical, and highlighted several reasons for caution: beta cells in vitro are not performing their physiological job of glucose-stimulated insulin secretion; the HPAP samples are themselves cultured for several days prior to sequencing, during which tissue-resident immune cells egress from the preparation; and a classic in vitro interferon response produces a broader and more uniform set of interferon-stimulated genes than the partial, focal ISG signature seen in C3. He concluded that the positive signal — the genes that do overlap — is convincing enough to implicate cytokines as an important driver, but that which specific cytokines, at what timing, and through what exact mechanism remain open questions that the cross-sectional data cannot resolve.
If you had unlimited time and money, what would your next experiments be? Spurrell outlined two priorities. First, he would want to confirm the C3 finding in a larger, less noisy cohort — ideally one where the GRN inference approach is unnecessary because the data quality is sufficient for conventional analysis, and where disease duration, insulin history, and other clinical metadata are complete enough to test correlations. Second, and more ambitiously, he would want to overexpress C3-defining transcription factors — particularly IRF1 — in beta cells and directly test whether they confer a survival advantage against immune challenge. He acknowledged the central tension here: C3 cells are dedifferentiated, and introducing their program into SC-derived beta cells would risk compromising the functional maturity the field has worked so hard to achieve. But he argued that the immune system is nuanced enough that the survival benefit of the C3 program might be partially uncoupled from its dedifferentiation signature — and that using the biology to guide engineering, rather than building immune resistance empirically, is the more principled long-term approach.
🔗 3 TSS Talks That Connect With This One
Kevan Herold, MD with Matthias Von Herrath, MD — Yale & UCSD · Ask the Expert Spurrell’s mentor Dr. Herold in conversation with Dr. Von Herrath, covering teplizumab, the clinical evidence for residual beta cell preservation, and the broader landscape of immunotherapy in T1D. This is the essential clinical context for why the question Spurrell is asking matters therapeutically: teplizumab works by slowing beta cell loss, and understanding what makes some beta cells naturally resistant to immune killing is the mechanistic complement to that clinical program. Herold’s presence in the live discussion — adding nuance on P21, senescence, and the NOD mouse comparisons — makes this pairing particularly direct.
Ruth Elgamal, PhD Candidate — UCSD · Ask the Expert · Integrated Pancreatic Islet Reference Map Elgamal is a co-author on the expanded HPAP scRNA-seq resource paper covering 258,000+ cells from 67 donors — the dataset Spurrell directly reanalyzes to identify C3. Her Ask the Expert talk is the best available public introduction to the HPAP resource: its structure, its donors, the technical challenges of cross-batch integration, and the analytical decisions that shaped it. Klaus Kaestner’s live comments during Spurrell’s talk — about the unavoidable noise introduced by changing kit chemistries across a multi-year collection — are directly addressed in Elgamal’s work, making this the mandatory companion piece for anyone who wants to work with the same data.
Kyle Gaulton, PhD — UCSD · Ask the Expert · T1D Risk, Genetics, and Single-Cell Epigenetics Gaulton leads one of the primary analytical efforts within HPAP and has done foundational work linking T1D genetic risk loci to regulatory elements active in specific islet cell types using single-cell epigenomics. Spurrell’s identification of IKZF4 — a known T1D risk gene — among the transcription factors enriched in C3 beta cells is a direct point of intersection: if a T1D risk variant alters IKZF4 activity, it could in principle affect the probability of a beta cell entering the C3 survival state. Gaulton’s framework for reading genetic risk through the lens of cell-type-specific regulatory programs is the conceptual bridge between Spurrell’s transcriptional findings and the human genetics of T1D susceptibility.
🔔 Subscribe to The Sugar Science newsletter to get the latest T1D research drops straight to your inbox.