Citations

Abstract Scientific foundation models learn high-dimensional representations from diverse data modalities, yet what they encode and how to extract that knowledge remain open questions. Here we show that probing the internal representations of Evo 2, a 7-billion-parameter genomic foundation model, enables accurate and interpretable genetic variant effect prediction. We introduce a covariance-based probe that captures second-order structure from Evo 2 sequence embeddings to predict variant pathogenicity across variant types and functional consequences, matching or exceeding specialized predictors within their domains. To ground these predictions in known biological mechanisms, we train a complementary panel of probes on existing annotations to detect which genomic properties are disrupted by a variant. This categorized evidence is then integrated with each variant’s genomic context through a language model to generate variant-specific mechanistic hypotheses. Our pathogenicity predictions correlate with experimental measures of variant function, clinical penetrance, and biobank disease associations while the mechanistic hypotheses are consistent with expert reviews, known mechanism classes, and downstream molecular readouts. We release pathogenicity scores, disruption profiles, and contextualized interpretations for 4.2 million variants from the ClinVar database as an open resource through the Evo Variant Effect Explorer (EVEE). More broadly, this structured probing approach offers a general framework for interrogating foundation models across scientific disciplines and grounding their outputs in existing domain concepts.

ABSTRACT Background and Aims Primary sclerosing cholangitis (PSC) is a progressive cholestatic liver disease with a high risk of malignancy and limited therapeutic options. Both the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) recommend structured baseline evaluation, fibrosis staging and longitudinal surveillance. However, real‐world adherence to these practices in North America remains unclear. Methods We conducted a retrospective cohort study of 1300 patients with PSC enrolled in the Consortium for Autoimmune Liver Diseases (CALiD) registry between 2018 and 2024 across 19 centres in North America. Guideline adherence was assessed for diagnostic measures at baseline (colonoscopy to establish IBD status and noninvasive fibrosis staging) and longitudinal surveillance (colonoscopy in PSC‐IBD, abdominal imaging for hepatobiliary cancer surveillance and CA 19‐9 testing). Prescribing patterns for ursodeoxycholic acid (UDCA) were also characterised. Multivariable logistic regression was used to evaluate demographic, clinical and socioeconomic predictors of adherence. Results At diagnosis, 78.8% of eligible patients with available data underwent colonoscopy to establish IBD status, and 45% underwent noninvasive fibrosis staging. During follow‐up, only 30%–35% of patients with PSC‐IBD underwent colonoscopy annually, while abdominal imaging for hepatobiliary cancer surveillance declined from 74% at Year 1 to 51% by Year 4. CA 19‐9 testing was documented in 47% over 5‐year period. UDCA was prescribed in 51% at baseline, with median dose of 12.5 mg/kg/day. Socioeconomic factors, including lower education and household income, were associated with reduced adherence, while cirrhosis strongly predicted completion of surveillance imaging. Conclusions Adherence to guideline‐based care in PSC remains suboptimal in North America Suboptimal adherence was observed for strongly recommended practices with longstanding guideline support (colonoscopy to establish IBD status; annual CRC surveillance in PSC‐IBD) as well as practices where guideline consensus has evolved more recently or where societies differ (baseline fibrosis assessment; hepatobiliary cancer imaging). Efforts to standardise care and address socioeconomic barriers are urgently needed to improve outcomes.

Current spatial T cell receptor (TCR) profiling approaches lack the resolution needed to link clonal identity, transcriptional state, and spatial positioning of individual T cells in the tumor microenvironment. Here, we introduce a spatial TCR profiling strategy that resolves individual T cell clones together with their transcriptional states at single-cell resolution and applied the method to human head and neck squamous cell carcinoma. Presumed tumor-specific T cells were broadly dispersed throughout the tumor microenvironment, and cells of the same clone occupied distinct transcriptional states in different locations: Immune-rich regions contained more plastic or progenitor cells, whereas tumor-dense regions were enriched for exhausted states. Patients exhibited notably different spatial architectures of antitumor T cell responses, revealing variation that was not captured by high-resolution, spatially agnostic methods such as spectral flow cytometry and single-cell RNA sequencing. These results provide a blueprint for dissecting antigen-specific T cell states in human tumors and reveal how T cell states are spatially coordinated with local cues across the tumor microenvironment.

Abstract The Bridging the Gaps (BTG) in Leukemia, Lymphoma and Multiple Myeloma Consensus Conference 2025 brought together a multidisciplinary group of oncology experts to address the complexities of lymphoma management, focusing on mantle cell lymphoma (MCL), chronic lymphocytic leukemia (CLL), and diffuse large B‐cell lymphoma (DLBCL). This article presents consensus recommendations developed through a modified Delphi process, which emphasize the need for tailored therapeutic strategies in light of recent advancements in treatment options. Key recommendations include the screening for high‐risk features in MCL, use of the BOVen regimen (zanubrutinib, obinutuzumab, and venetoclax) for TP53 ‐aberrant cases, and integration of chimeric antigen receptor T‐cell therapy for patients with mantle cell lymphoma that is refractory to covalent Bruton tyrosine kinase inhibitors. For CLL, recommendations include consideration of time‐limited therapies for younger patients and a “watch and wait” strategy for asymptomatic patients despite the improved activity and safety of current treatment regimens. For DLBCL, this article highlights the challenges in treatment sequencing and the role of circulating tumor DNA and minimal residual disease testing in monitoring disease progression. Overall, the conference describes the importance of ongoing research to refine management strategies and improve patient outcomes in lymphoma care, addressing the gaps in clinical practice where high‐level evidence is lacking.

Background Psychiatric comorbidities are increasingly recognised in demyelinating diseases, yet their frequency and timing in myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD) are not well defined. We aimed to characterise the spectrum and temporal pattern of psychiatric illness in MOGAD and compare to multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) with aquaporin-4 antibody (AQP4-IgG). Methods We conducted a cross-sectional study of patients with MOGAD (n=388), age-matched and sex-matched MS (n=257) and AQP4-IgG-positive NMOSD (n=58) evaluated at Mayo Clinic (1991–2025). Psychiatric diagnoses, timing relative to the first attack and patient-reported outcomes were extracted from electronic health records (EHRs). Complementary analyses were conducted using the NeuroBlu real-world EHR database. Results Among 388 patients with MOGAD, 54.8% had ≥1 psychiatric disorder, most commonly anxiety (47.7%), depression (36.6%) and attention-deficit/hyperactivity disorder (ADHD, 9%). After the first attack, 33.3% of patients with MOGAD developed a new psychiatric disorder, often within the first year. MOGAD showed a lower frequency of depression versus MS (36.6% vs 47.5%, p=0.02) but similar rates of anxiety and ADHD. Temporal patterns differed: psychiatric disorders often appeared years before onset in MS but clustered shortly after onset in MOGAD. NeuroBlu analyses supported high rates of psychiatric disorders across the three conditions. Conclusion Psychiatric comorbidities are common in MOGAD, particularly depression and anxiety, typically emerging close after onset, whereas in MS they often preceded onset by years. These findings underscore the need for routine psychiatric screening in MOGAD and highlight unresolved questions regarding whether, and to what extent, MOG-IgG-mediated inflammatory demyelination contributes to psychiatric vulnerability.

This article describes the typical and atypical chest imaging appearances of Pneumocystis jirovecii pneumonia.

The NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Non-Small Cell Lung Cancer (NSCLC) provide recommendations for the treatment of patients with NSCLC, including diagnosis, primary disease management, surveillance, and subsequent treatment. The panel has updated the list of recommended targeted therapies based on recent FDA approvals and clinical data. This selection from the NCCN Guidelines for NSCLC focuses on treatment recommendations for advanced or metastatic NSCLC with actionable biomarkers.

Objective Immunohistochemically (IHC) measured transactive response DNA‐binding protein 43 (TDP‐43) inclusions are observed in Alzheimer's disease (AD) and are associated with medial temporal lobe atrophy. Accumulation of cryptic exons occurs in AD in response to TDP‐43 pathology. We aimed to assess relationships between IHC and biochemically measured insoluble TDP‐43 and cryptic exons and assess associations with hippocampal and amygdala volume loss and atrophy rates on magnetic resonance imaging (MRI). Methods Eighty‐one neuropathologically diagnosed AD cases were analyzed. For biochemistry, insoluble TDP‐43 was quantified using a Meso‐scale discovery (MSD) immunoassay. IHC‐TDP burden was quantified with digital histopathology. Cryptic RNAs were assessed via quantitative real‐time polymerase chain reaction (qRT‐PCR). Thirty‐eight cases had serial brain MRI. Hippocampal and amygdala volumes were calculated using FreeSurfer. Regression models were used to investigate associations among IHC‐TDP‐43 status/burden, MSD‐TDP status/levels, cryptic RNAs, and hippocampal and amygdala volumes and atrophy rates. Results IHC‐TDP(+) cases exhibited elevated levels of MSD‐TDP and cryptic RNAs ( KCNQ2 , STMN2 , and UNC13A ) and increased MSD‐TDP levels were associated with increased cryptic RNA levels, in the hippocampus and amygdala. IHC‐TDP(+) cases had smaller hippocampal and amygdala volumes compared to IHC‐TDP(−) cases. MSD‐TDP(+) cases had smaller hippocampal volumes and faster amygdala rates of atrophy compared with MSD‐TDP(−) cases. Higher KCNQ2 and UNC13A levels were associated with smaller amygdala volumes. Interpretation MSD‐TDP level is a reliable surrogate for IHC‐based TDP‐43 status. Both TDP‐43 and cryptic RNA levels are associated with reduced medial temporal volumes, suggesting cryptic exons may be playing a role in brain volume loss in AD. ANN NEUROL 2026;100:193–205

Background. Donor-derived cell-free DNA (dd-cfDNA) has emerged as a promising noninvasive biomarker for detecting allograft rejection in solid organ transplantation. However, its diagnostic utility in pancreas transplantation, including in multiorgan recipients, remains underexplored. Methods. We conducted a retrospective analysis of 42 pancreas transplant recipients (38 simultaneous pancreas-kidney, 4 pancreas transplant alone) who underwent for-cause organ biopsy and dd-cfDNA testing between April 2020 and December 2024. dd-cfDNA was measured using the Prospera assay, with a threshold of ≥1% considered high risk and <1% low-risk. Biopsy results were correlated with dd-cfDNA to assess test performance. Results. A total of 50 dd-cfDNA tests and 45 organ biopsies were analyzed. Among simultaneous pancreas-kidney recipients, dd-cfDNA demonstrated a sensitivity of 86.7%, specificity of 52.2%, positive predictive value (PPV) of 54.2%, and negative predictive value (NPV) of 85.7% for rejection. In pancreas transplant alone recipients, all high-risk dd-cfDNA results corresponded to biopsy-proven rejection, yielding 100% sensitivity, specificity, PPV, and NPV. When combining both cohorts, dd-cfDNA testing achieved an overall sensitivity of 88.2%, specificity of 57.7%, PPV of 57.7%, and NPV of 88.2%. Additionally, dd-cfDNA kinetics suggested its potential utility as a surveillance tool posttreatment, potentially reducing the need for closure biopsies. Conclusions. dd-cfDNA is a valuable noninvasive adjunct for monitoring of rejection in pancreas transplant recipients. Further prospective studies are needed to validate these findings and define the optimal role of dd-cfDNA in posttransplant care.