Analysis of 3D pathology samples using weakly supervised AI.

Human tissue, which is inherently three-dimensional (3D), is traditionally examined through standard-of-care histopathology as limited two-dimensional (2D) cross-sections that can insufficiently represent the tissue due to sampling bias. To holistically characterize histomorphology, 3D imaging modalities have been developed, but clinical translation is hampered by complex manual evaluation and lack of computational platforms to distill clinical insights from large, high-resolution datasets. We present TriPath, a deep-learning platform for processing tissue volumes and efficiently predicting clinical outcomes based on 3D morphological features. Recurrence risk-stratification models were trained on prostate cancer specimens imaged with open-top light-sheet microscopy or microcomputed tomography. By comprehensively capturing 3D morphologies, 3D volume-based prognostication achieves superior performance to traditional 2D slice-based approaches, including clinical/histopathological baselines from six certified genitourinary pathologists. Incorporating greater tissue volume improves prognostic performance and mitigates risk prediction variability from sampling bias, further emphasizing the value of capturing larger extents of heterogeneous morphology.

Systemic bevacizumab as salvage therapy for persistent severe bleeding and anemia in heyde syndrome following aortic valve replacement.

Heyde syndrome is characterized by the co-occurrence of aortic stenosis and bleeding gastrointestinal angiodysplasias, often with acquired von Willebrand syndrome. Current management for bleeding includes hematologic support with red cell transfusion and intravenous iron and correction of aortic stenosis with valve replacement. However, persistent Heyde syndrome after valve replacement occurs in a significant minority of cases, and there is no accepted therapy for these patients. Given that the pathophysiology of angiodysplasia formation in Heyde syndrome involves dysregulated angiogenesis, targeting angiogenesis may be an effective therapeutic option. We describe two cases of persistent Heyde syndrome with severe bleeding and anemia in patients following aortic valve replacement who were treated with systemic bevacizumab, a monoclonal antibody directed against vascular endothelial growth factor. In both cases, treatment was successful, with resolution of bleeding, liberation from hematologic support, and normalization of hemoglobin. In addition to responding to therapy, neither patient had treatment-related adverse events (and both had recurrent anemia upon treatment discontinuation, further evidence of the therapeutic impact of bevacizumab). Additional investigation into the use of systemic antiangiogenic therapy for treatment of Heyde syndrome is warranted.

Systemic bevacizumab to facilitate anticoagulation in antiphospholipid syndrome and bleeding gastrointestinal angiodysplasia.

Bleeding gastrointestinal angiodysplasia may occur in patients with vasculitis and can be challenging to treat. We describe the novel use of bevacizumab therapy to treat bleeding gastrointestinal angiodysplasia and severe anemia in a patient with eosinophilic granulomatosis with angiitis complicated by antiphospholipid antibody syndrome requiring indefinite warfarin therapy. Studies confirmed multiple bleeding jejunal angiodysplasias unamenable to endoscopic intervention, and the patient required ongoing support with iron infusions and blood transfusions to maintain a minimally acceptable hemoglobin. Given the severe anemia, need for continued, indefinite antiplatelet and anticoagulation therapy, and failure of standard treatment approaches, the patient was initiated on systemic bevacizumab therapy, on the basis of prior documented success of bevacizumab to manage gastrointestinal telangiectasias in patients with hereditary hemorrhagic telangiectasia. Bevacizumab was highly effective, with rapid resolution of bleeding, normalization of hemoglobin, liberation from hematologic support and no adverse events, including no thromboembolic events. Vascular endothelial growth factor (VEGF-A) rose paradoxically after initiation of bevacizumab and normalized after its discontinuation. Given these findings, use of systemic bevacizumab to manage bleeding angiodysplasia in patients with acquired vascular disorders merits further study.

Thrombopoietin level predicts response to treatment with romiplostim in chemotherapy-induced thrombocytopenia.

Chemotherapy-induced thrombocytopenia (CIT) is a common complication of cancer treatment. Evidence has emerged supporting use of romiplostim to treat CIT but predicting clinical response to romiplostim is not possible. To determine utility of endogenous thrombopoietin (TPO) as a biomarker of romiplostim response, we performed an observational cohort study of patients with CIT and known baseline TPO levels receiving romiplostim. For weekly on-romiplostim platelet (Plt) count assessment, clinical response was defined as Plt ≥ 75 × 109 /L and ≥ 30 × 109 /L above pretreatment baseline. Overall, moderate, and superior classes of treatment response were defined based on fraction of Plt assessments meeting clinical response criteria (> 0, ≥ 0.6, and ≥ 0.8, respectively). Sixty-three patients with CIT were included; median age was 62 years, 41.3% were female, and median (IQR) romiplostim treatment duration was 14 (4-38) weeks. Median (IQR) TPO was lower in patients achieving moderate response to romiplostim vs those who did not, 234 (135-1085) pg/mL vs 665 (244-1970) pg/mL (p = .034) and lower still in patients achieving superior response vs those who did not, 212 (91-690) pg/mL versus 559 (173-1851) pg/mL (p = .023). Negative correlations were found between TPO level and baseline Plt and TPO level and response fraction. A positive correlation was found between TPO level and lowest effective romiplostim dose. In receiver operating characteristic (ROC) analysis, optimally discriminant TPO level thresholds (as defined by Youden's Index) were ≤ 457 pg/mL for moderate response and ≤ 260 pg/mL for superior response. In conclusion, TPO levels predict response to romiplostim in CIT, with lower levels predicting improved probability and depth of response.

Initial experience with scalp sparing radiation with concurrent temozolomide and tumor treatment fields (SPARE) for patients with newly diagnosed glioblastoma.

INTRODUCTION: Standard of care for glioblastoma includes concurrent chemoradiation and maintenance temozolomide with tumor treatment fields (TTFields). Preclinical studies suggest TTFields and radiation treatment have synergistic effects. We report our initial experience evaluating toxicity and tolerability of scalp-sparing radiation with concurrent TTFields. METHODS: This is a single arm pilot study (clinicaltrials.gov Identifier: NCT03477110). Adult patients (age ≥ 18 years) with KPS ≥ 60 with newly diagnosed glioblastoma were eligible. All patients received concurrent scalp-sparing radiation (60 Gy in 30 fractions), standard concurrent temozolomide (75 mg/m2 daily), and TTFields. Maintenance therapy included standard temozolomide and continuation of TTFields. Radiation treatment was delivered through TTFields arrays. The primary endpoint was safety and toxicity for concurrent TTFields with chemoradiation in newly diagnosed glioblastoma. RESULTS: We report the first ten patients on the trial. Eight were male, and two were female, with median age 61 years (range 49 to 73 years). Median KPS was 90 (range 70-90). Median follow-up was 7.9 months (2.8 to 17.9 months). Nine (90%) patients with unmethylated MGMT promotor, and one with methylated. Median time from surgery to radiation was 33 days (28 to 49 days). All patients completed concurrent chemoradiation plus TTFields without radiation or TTFields treatment interruption or discontinuation. Scalp dose constraints were achieved for all patients, with mean dose having a median value of 7.7 Gy (range 4.9 to 13.2 Gy), D20cc median 22.6 Gy (17.7 to 36.8 Gy), and D30cc median 19.8 Gy (14.8 to 33.4 Gy). Average daily use during concurrent phase had median value of 83.5% and 77% for maintenance. There was no related ≥ Grade 3 toxicity. Skin toxicity (erythema, dermatitis, pruritus) was noted in 80% of patients, however, these were limited to Grade 1 or 2 events which resolved spontaneously or responded to topical medications. Eight patients (80%) had progression, with median PFS of 6.9 months (range 2.8 to 9.6 months). CONCLUSIONS: Concurrent TTFields with scalp-sparing chemoradiation is a safe and feasible treatment option with limited toxicity. Future randomized prospective trial is warranted to define therapeutic advantages of concurrent TTFields with chemoradiation. TRIAL REGISTRATION: Clinicaltrials.gov Identifier NCT03477110.

Characterization of the rate, predictors, and thrombotic complications of thrombocytosis in iron deficiency anemia.

The association of thrombocytosis with iron deficiency anemia (IDA) is well-recognized, but data describing the rate, predictors, and risk of thrombotic complications associated with IDA-related thrombocytosis are limited. We queried an institutional patient data repository containing comprehensive chart data for over 6 million patients to identify IDA patients with and without thrombocytosis and thrombotic events over a 40-year time period (1979 to 2019). Demographics, hematological parameters, thrombosis history, and other medical history were collected. Fidelity of query data was assessed via detailed manual chart review of 700 patients, including confirmation of ferritin and hematologic parameters in addition to temporal and clinical association of thrombocytosis. Our queries identified 36 327 cases of IDA of which 15 022 had thrombocytosis. Following assessment for data integrity, we observed a thrombocytosis rate of 32.6% in patients with IDA. The rate of thrombosis was calculated to be 7.8% in patients with IDA and 15.8% in patients with IDA and thrombocytosis. Platelet mass index at time of peak thrombocytosis was significantly higher than at baseline and was strongly negatively correlated with hemoglobin at peak thrombocytosis. A multivariable model demonstrated a significant predictive relationship between decreasing hemoglobin and increasing platelet count at peak thrombocytosis. In conclusion, we observed reactive thrombocytosis in one-third of IDA patients, and a 2-fold thrombosis risk in patients with IDA and thrombocytosis compared with patients with IDA alone. Given the global burden of untreated and undertreated IDA, adequate IDA treatment may reduce thrombotic complications and associated morbidity and mortality.

A prospective cohort study of condensed low-dose total skin electron beam therapy for mycosis fungoides: Reduction of disease burden and improvement in quality of life.

BACKGROUND: Low-dose total skin electron beam therapy (TSEBT) for mycosis fungoides is popular because of reduced toxicity with effective palliation. We condensed TSEBT, reducing visits by half and overall treatment length by one third. OBJECTIVE: To determine the efficacy and safety of a novel condensed low-dose TSEBT for mycosis fungoides. METHODS: We conducted a cohort study (2014-2018) with a median follow-up of 22.8 months. We delivered 12 Gy per 6 fractions with the modified Stanford technique, 3 fractions per week, with boosts to shadowed sites at risk between treatments, completing in 2 weeks. Primary outcomes included clinical response, duration of and time to response, and toxicity. Secondary outcomes included patient-reported quality of life (pain, pruritus, and Dermatology Life Quality Index) and physician-scored disease burden (body surface area involvement and Modified Skin Weighted Assessment Tool). RESULTS: Of 25 patients, stage IB was most common at the time of TSEBT (36%). The overall response rate was 88%. Most common was a near complete response (36%), and complete response was achieved in 6 (24%) patients. The median duration of response was 17.5 months (3.5-44.2), and the median time to response was 2 months (range, 0.9-4.1). No patients had toxicity of grade 3 or greater. QOL and disease burden showed significant benefit after TSEBT (P < .001). LIMITATIONS: Cohort study with limited sample size. CONCLUSIONS: Condensed, low-dose TSEBT has favorable outcomes and toxicity with logistical convenience.

Perceptions of Oncologists About Sharing Clinic Notes with Patients.

In a large survey (n = 809) conducted to understand how oncologists differ from nononcologists regarding routinely sharing visit notes with patients, oncologists were less likely to agree patient safety would improve (p = .03) or that patients would be offended after reading notes (p = .01); however, they agreed with nononcologists that sharing notes would lead to less candid documentation (69% vs. 73%; p = .39). Oncologists share a high level of worry about the impact of sharing notes on documentation practices, a concern that will need to be addressed as the practice of sharing visit notes expands to cancer care.

Perinatal nicotine exposure alters Akt/GSK-3ß/mTOR/autophagy signaling, leading to development of hypoxic-ischemic-sensitive phenotype in rat neonatal brain.

Maternal cigarette smoking is a major perinatal insult that contributes to an increased risk of cardiovascular and neurodevelopmental diseases in offspring. Our previous studies revealed that perinatal nicotine exposure reprograms a sensitive phenotype in neonatal hypoxic-ischemic encephalopathy (HIE), yet the underlying molecular mechanisms remain largely elusive. The present study tested the hypothesis that perinatal nicotine exposure impacts autophagy signaling in the developing brain, resulting in enhanced susceptibility to neonatal HIE. Nicotine was administered to pregnant rats via subcutaneous osmotic minipumps. Neonatal HIE was conducted in 9-day-old male rat pups. Protein kinase B/glycogen synthase kinase-3ß/mammalian target of rapamycin (Akt/GSK-3ß/mTOR) signaling and key autophagy markers were determined by Western blotting analysis. Rapamycin and MK2206 were administered via intracerebroventricular injection. Nicotine exposure significantly inhibited autophagy activities in neonatal brain tissues, characterized by an increased ratio of phosphoylated (p-) to total mTOR protein expression but reduced levels of autophagy-related 5, Beclin 1, and LC3ßI/II. Treatment with mTOR inhibitor rapamycin effectively blocked nicotine-mediated autophagy deficiency and, more importantly, reversed the nicotine-induced increase in HI brain infarction. In addition, nicotine exposure significantly upregulated p-Akt and p-GSK-3ß. Treatment with the Akt selective inhibitor MK2206 reversed the enhanced p-Akt and p-GSK-3ß, restored basal autophagic flux, and abolished nicotine-mediated HI brain injury. These findings suggest that perinatal nicotine-mediated alteration of Akt/GSK-3ß/mTOR signaling plays a key role in downregulation of autophagic flux, which contributes to the development of hypoxia/ischemia-sensitive phenotype in the neonatal brain.

Pharmacological Modulation of Noradrenergic Arousal Circuitry Disrupts Functional Connectivity of the Locus Ceruleus in Humans.

State-dependent activity of locus ceruleus (LC) neurons has long suggested a role for noradrenergic modulation of arousal. However, in vivo insights into noradrenergic arousal circuitry have been constrained by the fundamental inaccessibility of the human brain for invasive studies. Functional magnetic resonance imaging (fMRI) studies performed during site-specific pharmacological manipulations of arousal levels may be used to study brain arousal circuitry. Dexmedetomidine is an anesthetic that alters the level of arousal by selectively targeting α2 adrenergic receptors on LC neurons, resulting in reduced firing rate and norepinephrine release. Thus, we hypothesized that dexmedetomidine-induced altered arousal would manifest with reduced functional connectivity between the LC and key brain regions involved in the regulation of arousal. To test this hypothesis, we acquired resting-state fMRI data in right-handed healthy volunteers 18-36 years of age (n = 15, 6 males) at baseline, during dexmedetomidine-induced altered arousal, and recovery states. As previously reported, seed-based resting-state fMRI analyses revealed that the LC was functionally connected to a broad network of regions including the reticular formation, basal ganglia, thalamus, posterior cingulate cortex (PCC), precuneus, and cerebellum. Functional connectivity of the LC to only a subset of these regions (PCC, thalamus, and caudate nucleus) covaried with the level of arousal. Functional connectivity of the PCC to the ventral tegmental area/pontine reticular formation and thalamus, in addition to the LC, also covaried with the level of arousal. We propose a framework in which the LC, PCC, thalamus, and basal ganglia comprise a functional arousal circuitry.SIGNIFICANCE STATEMENT Electrophysiological studies of locus ceruleus (LC) neurons have long suggested a role for noradrenergic mechanisms in mediating arousal. However, the fundamental inaccessibility of the human brain for invasive studies has limited a precise understanding of putative brain regions that integrate with the LC to regulate arousal. Our results suggest that the PCC, thalamus, and basal ganglia are key components of a LC-noradrenergic arousal circuit.

Bevacizumab and re-irradiation for recurrent high grade gliomas: does sequence matter?

PURPOSE/OBJECTIVES: We report the outcomes of the largest cohort to date of patients receiving both bevacizumab (BEV) and fractionated stereotactic radiotherapy (FSRT) for progressive or recurrent high grade glioma (HGG). Furthermore, the sequence of these two treatment regimens was analyzed to determine an optimal treatment paradigm for recurrent HGG. MATERIALS/METHODS: After Institutional Review Board approval, patients with pathologically confirmed WHO grade III anaplastic astrocytoma (AA) or IV glioblastoma multiforme (GBM) glioma who subsequently underwent re-irradiation at recurrence with FSRT were retrospectively reviewed. Patients from this group who had received BEV were also identified. Survival from initial diagnosis, as well as from recurrence and re-irradiation, were analyzed as study endpoints. Date of recurrence was defined as the date of radiographic evidence of progressive/recurrent disease. Kaplan-Meier curves were generated utilizing a log-rank test with a p-value ≤ 0.05 considered significant to compare treatment sequences in terms of survival outcomes. RESULTS: A total of 118 patients with recurrent/progressive HGG (GBM = 87, AA = 31) had received both BEV and FSRT. Patient characteristics were as follows: median KPS at recurrence was 80 (range 50-100); median age at recurrence was 57 years; median time to radiographic recurrence/progression was 10.8 months (mo) and 33.1% of patients had surgery for recurrence. The median time from the start of BEV to FSRT was 6.4 months and from FSRT to the start of BEV was 5.1 months. For the entire cohort, median overall survival (OS) was 26.7 months and median survival time (MST) from recurrence was 13.8 months (24.4 months and 11.9 months for GBM only). In patients that received BEV prior to FSRT (n = 50), median OS and MST from recurrence were 25.2 and 13.3 months respectively. In patients receiving FSRT first (n = 56), median OS and MST from recurrence were 28.8 months and 13.9 months, respectively. Sequencing of BEV and FSRT at recurrence was not significantly associated with OS (p = 0.08) or median survival from recurrence (p = 0.75). CONCLUSIONS: The combination of FSRT and BEV for recurrent/progressive HGG provides promising results in terms of overall survival and survival from recurrence. Combining these treatment modalities appears to improve upon the historic outcomes of either treatment alone. The outcomes data from this study support the ongoing RTOG trial exploring the combination of BEV and FSRT for recurrent HGG.

Genetic anticipation in BRCA1/BRCA2 families after controlling for ascertainment bias and cohort effect.

BACKGROUND: Genetic anticipation, the earlier onset of disease in successive generations, has been reported in hereditary breast and ovarian cancer syndrome (HBOC), but little is known about its underlying mechanisms. Ascertainment bias has been suggested as a reason in previous studies. Likewise, cohort effect, which may be caused by environmental factors, can be misinterpreted as genetic anticipation. METHODS: The authors reviewed the pedigrees of 176 kindreds, segregating those with deleterious mutations in breast cancer genes 1 and 2 (BRCA1/BRCA2) who had at least 2 consecutive generations of the same cancer (breast or ovarian). By using mutation probabilities as analytical weights in weighted random-effect models, generational differences in the age at onset of breast/ovarian cancer were calculated. The analyses were further controlled for ascertainment bias by excluding probands and adjusting for birth-cohort effect in the anticipation models. RESULTS: The mean age at the onset of breast cancer for the probands' generation was 41.9 years, which was 6.8 years and 9.8 years earlier than the parents' and grandparents' generations, respectively. The anticipation effect for breast cancer remained significant after excluding the probands. There was a birth-cohort effect: patients who were born in 1930s and 1940s had breast cancer 5.0 years and 7.6 years earlier than patients who were born before 1920. The difference in breast cancer age of onset across generations was no longer significant after adjusting for birth-cohort effect. CONCLUSIONS: The observed anticipation effect was driven mainly by a decrease in age of onset across birth cohorts, underscoring the need for risk-reducing interventions that target changing environmental/lifestyle factors in BRCA1/BRCA2 carriers. Cancer 2016;122:1913-20. © 2016 American Cancer Society.

Association Between Time From Surgery to Radiation Therapy and Multimodality Treatment Outcomes in HPV+ Head and Neck Cancer: A Multi-Institutional Cohort Experience.

Purpose: Oropharyngeal squamous cell cancers (OPSCCs) are traditionally managed with surgery and, if indicated, adjuvant radiation therapy (RT) with or without chemotherapy. NCCN recommends keeping the time from surgery to the start of RT (TSRT) within 6 weeks to avoid possibly compromising patient outcomes. HPV+ OPSCCs behave more favorably than HPV- OPSCCs. We hypothesized that TSRT beyond 6 weeks may not portend poorer outcomes for the former. Methods: We identified nonmetastatic, high-risk HPV+ OPSCCs treated with multimodal therapy at 2 institutions. Prolonged TSRT was defined as >6 weeks and was evaluated for association with recurrence-free survival (RFS). Radiation treatment time (RTT; time from the first to the last day of RT), total treatment package time (TTPT; time from surgery to the end of adjuvant treatments), de-escalated RT (dose ≤56 Gy), concurrent chemotherapy, smoking history, and treatment institution were evaluated as possible confounders. Results: In total, 96 patients were included. The median follow-up time was 62 months (4-123 months); 69 patients underwent transoral robotic surgeries, and 27 received open surgeries. The median postoperative RT dose was 60 Gy (50-70.8 Gy). The median TSRT, RTT, and TTPT were 38 days (11-208), 43 days (26-56 days), and 81 days (40-255 days), respectively. Ten patients failed treatment at a median of 8 months (4-64 months). Two locoregional and 4 distant failures occurred in the group without prolonged TSRT, whereas 2 locoregional and 2 distant failures were recorded in the prolonged TSRT group. Prolonged TTPT, de-escalated RT, chemotherapy, smoking history, and treatment institution were not associated with treatment failure. RTT was dropped from our analyses as no events appeared in the prolonged RTT group, and no reliable hazard ratio could be computed. Conclusions: TSRT > 6 weeks was not significantly associated with inferior outcomes in the postoperative management of HPV+ OPSCCs. Longer TSRT may facilitate better recovery from surgical toxicity, as needed, without compromising oncologic outcomes. The TSRT goal for these cancers should be investigated in future studies.

Vole genomics links determinate and indeterminate growth of teeth.

Continuously growing teeth are an important innovation in mammalian evolution, yet genetic regulation of continuous growth by stem cells remains incompletely understood. Dental stem cells responsible for tooth crown growth are lost at the onset of tooth root formation. Genetic signaling that initiates this loss is difficult to study with the ever-growing incisor and rooted molars of mice, the most common mammalian dental model species, because signals for root formation overlap with signals that pattern tooth size and shape (i.e., cusp patterns). Different species of voles (Cricetidae, Rodentia, Glires) have evolved rooted and unrooted molars that have similar size and shape, providing alternative models for studying roots. We assembled a de novo genome of Myodes glareolus, a vole with high-crowned, rooted molars, and performed genomic and transcriptomic analyses in a broad phylogenetic context of Glires (rodents and lagomorphs) to assess differential selection and evolution in tooth forming genes. We identified 15 dental genes with changing synteny relationships and six dental genes undergoing positive selection across Glires, two of which were undergoing positive selection in species with unrooted molars, Dspp and Aqp1. Decreased expression of both genes in prairie voles with unrooted molars compared to bank voles supports the presence of positive selection and may underlie differences in root formation. Bulk transcriptomics analyses of embryonic molar development in bank voles also demonstrated conserved patterns of dental gene expression compared to mice, with species-specific variation likely related to developmental timing and morphological differences between mouse and vole molars. Our results support ongoing evolution of dental genes across Glires, revealing the complex evolutionary background of convergent evolution for ever-growing molars.

Demographic bias in misdiagnosis by computational pathology models.

Despite increasing numbers of regulatory approvals, deep learning-based computational pathology systems often overlook the impact of demographic factors on performance, potentially leading to biases. This concern is all the more important as computational pathology has leveraged large public datasets that underrepresent certain demographic groups. Using publicly available data from The Cancer Genome Atlas and the EBRAINS brain tumor atlas, as well as internal patient data, we show that whole-slide image classification models display marked performance disparities across different demographic groups when used to subtype breast and lung carcinomas and to predict IDH1 mutations in gliomas. For example, when using common modeling approaches, we observed performance gaps (in area under the receiver operating characteristic curve) between white and Black patients of 3.0% for breast cancer subtyping, 10.9% for lung cancer subtyping and 16.0% for IDH1 mutation prediction in gliomas. We found that richer feature representations obtained from self-supervised vision foundation models reduce performance variations between groups. These representations provide improvements upon weaker models even when those weaker models are combined with state-of-the-art bias mitigation strategies and modeling choices. Nevertheless, self-supervised vision foundation models do not fully eliminate these discrepancies, highlighting the continuing need for bias mitigation efforts in computational pathology. Finally, we demonstrate that our results extend to other demographic factors beyond patient race. Given these findings, we encourage regulatory and policy agencies to integrate demographic-stratified evaluation into their assessment guidelines.

AI-driven Discovery of Morphomolecular Signatures in Toxicology.

Early identification of drug toxicity is essential yet challenging in drug development. At the preclinical stage, toxicity is assessed with histopathological examination of tissue sections from animal models to detect morphological lesions. To complement this analysis, toxicogenomics is increasingly employed to understand the mechanism of action of the compound and ultimately identify lesion-specific safety biomarkers for which in vitro assays can be designed. However, existing works that aim to identify morphological correlates of expression changes rely on qualitative or semi-quantitative morphological characterization and remain limited in scale or morphological diversity. Artificial intelligence (AI) offers a promising approach for quantitatively modeling this relationship at an unprecedented scale. Here, we introduce GEESE, an AI model designed to impute morphomolecular signatures in toxicology data. Our model was trained to predict 1,536 gene targets on a cohort of 8,231 hematoxylin and eosin-stained liver sections from Rattus norvegicus across 127 preclinical toxicity studies. The model, evaluated on 2,002 tissue sections from 29 held-out studies, can yield pseudo-spatially resolved gene expression maps, which we correlate with six key drug-induced liver injuries (DILI). From the resulting 25 million lesion-expression pairs, we established quantitative relations between up and downregulated genes and lesions. Validation of these signatures against toxicogenomic databases, pathway enrichment analyses, and human hepatocyte cell lines asserted their relevance. Overall, our study introduces new methods for characterizing toxicity at an unprecedented scale and granularity, paving the way for AI-driven discovery of toxicity biomarkers.

Developmental role of macrophages modeled in human pluripotent stem cell-derived intestinal tissue.

Macrophages populate the embryo early in gestation, but their role in development is not well defined. In particular, specification and function of macrophages in intestinal development remain little explored. To study this event in the human developmental context, we derived and combined human intestinal organoid and macrophages from pluripotent stem cells. Macrophages migrate into the organoid, proliferate, and occupy the emerging microanatomical niches of epithelial crypts and ganglia. They also acquire a transcriptomic profile similar to that of fetal intestinal macrophages and display tissue macrophage behaviors, such as recruitment to tissue injury. Using this model, we show that macrophages reduce glycolysis in mesenchymal cells and limit tissue growth without affecting tissue architecture, in contrast to the pro-growth effect of enteric neurons. In short, we engineered an intestinal tissue model populated with macrophages, and we suggest that resident macrophages contribute to the regulation of metabolism and growth of the developing intestine.

Towards a general-purpose foundation model for computational pathology.

Quantitative evaluation of tissue images is crucial for computational pathology (CPath) tasks, requiring the objective characterization of histopathological entities from whole-slide images (WSIs). The high resolution of WSIs and the variability of morphological features present significant challenges, complicating the large-scale annotation of data for high-performance applications. To address this challenge, current efforts have proposed the use of pretrained image encoders through transfer learning from natural image datasets or self-supervised learning on publicly available histopathology datasets, but have not been extensively developed and evaluated across diverse tissue types at scale. We introduce UNI, a general-purpose self-supervised model for pathology, pretrained using more than 100 million images from over 100,000 diagnostic H&E-stained WSIs (>77 TB of data) across 20 major tissue types. The model was evaluated on 34 representative CPath tasks of varying diagnostic difficulty. In addition to outperforming previous state-of-the-art models, we demonstrate new modeling capabilities in CPath such as resolution-agnostic tissue classification, slide classification using few-shot class prototypes, and disease subtyping generalization in classifying up to 108 cancer types in the OncoTree classification system. UNI advances unsupervised representation learning at scale in CPath in terms of both pretraining data and downstream evaluation, enabling data-efficient artificial intelligence models that can generalize and transfer to a wide range of diagnostically challenging tasks and clinical workflows in anatomic pathology.