SEARCH · Search PubMed

Results for “spatial genomics”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 559 records · Page 31Linked to original sources

PubMed1999

Structural constraints on RNA virus evolution.

The recently discovered hepatitis G virus (HGV) or GB virus C (GBV-C) is widely distributed in human populations, and homologues such as HGV/GBV-CCPZ and GBV-A are found in a variety of different primate species. Both epidemiological and phylogenetic analyses support the hypothesis that GB viruses coevolved with their primate hosts, although their degree of sequence similarity appears incompatible with the high rate of sequence change of HGV/GBV-C over short observation periods. Comparison of complete coding sequences (8,500 bases) of different genotypes of HGV/GBV-C showed an excess of invariant synonymous sites (at 23% of all codons) compared with the frequency expected by chance (10%). To investigate the hypothesis that RNA secondary-structure formation through internal base pairing limited sequence variability at these sites, an algorithm was developed to detect covariant sites among HGV/GBV-C sequences of different genotypes. At least 35 covariant sites that were spatially associated with potential stem-loop structures were detected, whose positions correlated with positions in the genome that showed reductions in synonymous variability. Although the functional roles of the predicted secondary structures remain unclear, the restriction of sequence change imposed by secondary-structure formation provides a mechanism for differences in net rate of accumulation of nucleotide substitutions at different sites. However, the resulting disparity between short- and long-term rates of sequence change of HGV/GBV-C violates the assumptions of the "molecular clock." This places a major restriction on the use of nucleotide or amino acid sequence comparisons to calculate times of divergence of other viruses evolving under the same structural constraints as GB viruses.

Base Sequence↗

PubMed2026

HIF1A+CSF3R+ neutrophils-dominated hypoxic niche induced metabolic reprogramming for neoadjuvant therapy resistance in NSCLC.

BACKGROUND: Non-small cell lung cancer (NSCLC) is one of the frequently occurring cancers characterized by molecular heterogeneity and multiple immune cell infiltration patterns, which are associated with treatment sensitivity and resistance. However, the specific microenvironmental cells and their mechanisms that lead to treatment resistance in patients need to be explored in greater depth. METHODS: On the basis of patients receiving neoadjuvant therapy in our center, a multicenter, multicohort NSCLC spatial transcriptome, single-cell transcriptome, T-cell receptor repertoire sequencing, bulk RNA transcriptome, phosphorylated proteome, genome mutation, and clinical data were included for a comprehensive assessment of the therapeutic and prognostic impact of HIF1A+ CSF3R+ neutrophils in NSCLC. In vitro experiments validated the functional phenotype of HIF1A+ CSF3R+ neutrophils and co-localization interactions with other cellular subpopulations. Gradient boosting machine (GBM) constructed region of interest (ROI) models for evaluation. Computer-aided drug design (CADD) was used to predict targeted small molecule drugs, and in vivo mouse models were constructed to assess the effectiveness of the combination treatment regimen. RESULTS: Centered on HIF1A+ CSF3R+ neutrophils, recruited exhausted T cells and stromal cells form a hypoxic niche within the tumor region, which was enriched in non-response patients. ROI composed of these specific cellular subpopulations, associated with senescence and glycolysis, accurately predicting NSCLC progression, prognosis, and microenvironment composition. CADD analysis identified that platycodin-D2 specifically targeted CSF3R, reducing HIF1A expression and inhibiting neutrophil activity. Combining navitoclax, platycodin-D2 with anti-programmed cell death protein 1 (PD-1) significantly suppressed tumor proliferation and improved the immunosuppressive microenvironment. CONCLUSION: Our study emphasized the role of HIF1A+ CSF3R+ neutrophils in immunotherapeutic resistance of NSCLC, constructed a microenvironmental immune dysregulation network in a hypoxic ecological niche with HIF1A+ CSF3R+ neutrophils as the center. Platycodin-D2 specifically targeted HIF1A+ CSF3R+ neutrophils, enhancing the efficacy of anti-PD-1 therapy in NSCLC.

Humans↗

PubMed1989

Three Strongylocentrotus purpuratus actin genes show correct cell-specific expression in hybrid embryos of S. purpuratus and Lytechinus pictus.

The cell-specific expression of three actin genes from the sea urchin species Strongylocentrotus purpuratus was examined in hybrid embryos of S. purpuratus and another species, Lytechinus pictus, by in situ hybridization. The mRNAs from each of these genes displayed distinct spatial patterns of expression in late-stage hybrid embryos (constructed in either direction), being detected only in the cell lineages where they are normally found in S. purpuratus embryos (i.e. CyIIIa, only in the aboral ectoderm lineage; CyI, in the gut, oral ectoderm and some mesenchyme cells of plutei, and preferentially in the archenteron of gastrulae; M, only in two small clusters of cells near the esophagus in plutei). These results, together with our previous observation that expression of each of these genes is activated at the same stage in these hybrid embryos as in normal S. purpuratus embryos, demonstrate that the trans-acting factors which are necessary to regulate both the temporal and spatial expression of these genes are present in the hybrid embryos. Previous experiments have shown that the expression of a chimeric gene containing the CyIIIa promoter fused to a bacterial chloramphenicol actetyltransferase (CAT) gene is not confined to the correct cell lineage (aboral ectoderm) when injected into Lytechinus embryos. The conclusion from these sets of data is that the factor(s) that regulate the spatial expression of at least one of the actin genes must derive from transcription of the zygotic genome.

Actins↗

PubMed2002

REDOX regulation of early embryo development.

Preimplantation embryonic development is associated with a change in preference in energy metabolism pathways. Although oxidative phosphorylation is obligatory in most species throughout preimplantation development, an increasing role for energy derived from glycolysis is associated with compaction and blastulation. Such a shift in metabolic pathway preference is desirable as the embryo faces an increasingly hypoxic environment in utero. We hypothesize that this shift in metabolic preference is associated with a change in the reduction-oxidation (REDOX) state within the embryo, affecting not only the energy production required for development, but also the activity of REDOX-sensitive transcription factors, which may alter gene expression patterns. Shifts in intracellular REDOX state may also contribute to spatial differences in cell activity, especially after compaction, and perhaps even major embryonic events such as fertilization, genome activation and cellular differentiation.

Animals↗

PubMed1998

Minimal sizes of deletions detected by comparative genomic hybridization.

Comparative genomic hybridization (CGH) has been used widely for the molecular cytogenetic analysis of tumors. Until now, the spatial resolution of this technique for diagnosing deletions of chromosomal sequences has not been assessed in detail. In the present study, we performed CGH analyses on five DNA samples derived from B-cell leukemias with 11q deletions, the sizes of which ranged from 3 Mbp to 14-18 Mbp. CGH experiments were evaluated by two established commercial analysis systems. Deletions down to a size of 10-12 Mbp were diagnosed based on a diagnostic threshold value of 0.8, if the vast majority of cells carried the deletion. For cases with smaller deletions, the ratio profiles were shifted toward underrepresentation at the respective chromosomal bands; however, the diagnostic threshold value was not reached. In all five cases, there was complete agreement between the two image analysis systems.

Chromosome Deletion↗

PubMed2026

Lactylation-related immune-metabolic dysregulation defines prognostic and therapeutic stratification in lung adenocarcinoma.

BACKGROUND: Lactylation links lactate metabolism with inflammatory signaling and immune regulation in tumors. However, its cellular distribution and translational value in lung adenocarcinoma (LUAD) remain unclear. METHODS: Single-cell RNA-sequencing datasets GSE189357 and GSE171145 were integrated to characterize lactylation-related activity, intercellular communication, and malignant epithelial cell states in LUAD. Single-cell-derived lactylation-related differentially expressed genes were mapped to TCGA-LUAD and multiple GEO cohorts. Univariate Cox regression and machine learning algorithms were used to construct a lactylation-related prognostic signature (LRPS). The associations of LRPS with prognosis, immunotherapy response, drug sensitivity, genomic alterations, immune infiltration, and inflammation- and metabolism-related pathways were evaluated. KRT7 was further validated using virtual knockout analysis, spatial transcriptomics, and in vitro and in vivo experiments. RESULTS: lactylation-related transcriptional activity showed heterogeneous distribution across LUAD cell populations and was associated with altered cell-cell communication. In malignant epithelial cells, LRTS-high and LRTS-low states exhibited distinct metabolic, inflammatory, and tumor-related pathway activities. LRPS showed stable prognostic performance in TCGA-LUAD and multiple GEO cohorts and remained an independent prognostic factor. Low LRPS was associated with greater potential benefit from immunotherapy, whereas different LRPS groups displayed distinct drug sensitivity, genomic alteration, and immune microenvironment patterns. KRT7 was highly expressed in LUAD and associated with poor prognosis. KRT7 knockdown suppressed LUAD cell proliferation, migration, invasion, colony formation, and tumor growth in vivo. CONCLUSIONS: This study identifies lactylation-related immune-metabolic dysregulation as a clinically relevant feature of LUAD and develops a single-cell-guided LRPS for prognosis and therapeutic stratification. KRT7 emerged as an LRPS-related functional candidate with experimentally supported roles in malignant LUAD phenotypes.

Immunotherapy↗

PubMed2006

Folding and organization of a contiguous chromosome region according to the gene distribution pattern in primary genomic sequence.

Specific mammalian genes functionally and dynamically associate together within the nucleus. Yet, how an array of many genes along the chromosome sequence can be spatially organized and folded together is unknown. We investigated the 3D structure of a well-annotated, highly conserved 4.3-Mb region on mouse chromosome 14 that contains four clusters of genes separated by gene "deserts." In nuclei, this region forms multiple, nonrandom "higher order" structures. These structures are based on the gene distribution pattern in primary sequence and are marked by preferential associations among multiple gene clusters. Associating gene clusters represent expressed chromatin, but their aggregation is not simply dependent on ongoing transcription. In chromosomes with aggregated gene clusters, gene deserts preferentially align with the nuclear periphery, providing evidence for chromosomal region architecture by specific associations with functional nuclear domains. Together, these data suggest dynamic, probabilistic 3D folding states for a contiguous megabase-scale chromosomal region, supporting the diverse activities of multiple genes and their conserved primary sequence organization.

Animals↗

PubMed1999

Regionalization of Sonic hedgehog transcription along the anteroposterior axis of the mouse central nervous system is regulated by Hnf3-dependent and -independent mechanisms.

The axial midline mesoderm and the ventral midline of the neural tube, the floor plate, share the property of being a source of the secreted protein, Sonic hedgehog (Shh), which has the capacity to induce a variety of ventral cell types along the length of the mouse CNS. To gain insight into the mechanisms by which Shh transcription is initiated in these tissues, we set out to identify the cis-acting sequences regulating Shh gene expression. As an approach, we have tested genomic clones encompassing 35 kb of the Shh locus for their ability to direct a lacZ reporter gene to the temporally and spatially restricted confines of the Shh expression domains in transgenic mice. Three enhancers were identified that directed lacZ expression to distinct regions along the anteroposterior axis including the ventral midline of the spinal cord, hindbrain, rostral midbrain and caudal diencephalon, suggesting that multiple transcriptional regulators are required to initiate Shh gene expression within the CNS. In addition, regulatory sequences were also identified that directed reporter expression to the notochord, albeit, under limited circumstances. Sequence analysis of the genomic clones responsible for enhancer activity from a variety of organisms, including mouse, chicken and human, have identified highly conserved binding sites for the hepatocyte nuclear factor 3 (Hnf3) family of transcriptional regulators in some, but not all, of the enhancers. Moreover, the generation of mutations in the Hnf3-binding sites showed their requirement in certain, but not all, aspects of Shh reporter expression. Taken together, our results support the existence of Hnf3-dependent and -independent mechanisms in the direct activation of Shh transcription within the CNS and axial mesoderm.

Animals↗

PubMed2026

Molecular epidemiology and phylogeographic architecture of oncogenic intracellular bacteria in cervical cancer patients across Northern China.

BACKGROUND: Oncogenic intracellular bacteria, including Chlamydia trachomatis, Mycoplasma genitalium, and Fusobacterium nucleatum, have emerged as significant contributors to cervical carcinogenesis. Despite growing interest in microbial oncology, the molecular epidemiological landscape and phylogeographic distribution of these pathogens in Northern China remain poorly characterized. This study aimed to determine the prevalence, co-infection patterns, genotypic diversity, and spatial phylogeographic clustering of oncogenic intracellular bacteria among cervical cancer patients across five provinces of Northern China. METHODS: A cross-sectional, multi-center study was conducted between March 2022 and November 2024 across Shaanxi, Heilongjiang, Beijing, Shandong, and Inner Mongolia. Cervical swab specimens were collected from 1247 confirmed cervical cancer patients. Pathogen detection was performed using multiplex real-time polymerase chain reaction, 16S rRNA gene amplicon sequencing, and whole-genome sequencing. Phylogeographic analyses employed maximum likelihood and Bayesian evolutionary inference frameworks. Statistical analyses included multivariate logistic regression and geographic information system-based spatial clustering. RESULTS: The overall prevalence of at least one oncogenic intracellular bacterium was 68.3% (n = 852). Chlamydia trachomatis was the most prevalent pathogen detected in 41.2% of participants. Co-infection with two or more bacteria was identified in 29.7% of cases and was independently associated with advanced-stage cervical cancer (adjusted odds ratio = 2.87; 95% confidence interval: 1.94 to 4.23; p < 0.001). Phylogeographic analysis revealed three distinct molecular clades with evidence of bidirectional gene flow between Shaanxi and Heilongjiang. Whole-genome sequencing identified 14 novel virulence gene variants not previously characterized in Chinese clinical isolates. CONCLUSIONS: Oncogenic intracellular bacteria are highly prevalent and genotypically diverse among cervical cancer patients in Northern China. The identified phylogeographic clustering and novel virulence variants have direct implications for regional screening programs, targeted antimicrobial strategies, and the development of region-specific molecular diagnostic panels.

Cervical cancer↗

PubMed1996

[The problem of systemic mutation].

The problems of the rearrangement of genome architecture during speciation are considered. The history of the problem of the role of systemic mutations in evolution is analyzed. The author's own phenomenological data on the spatial rearrangement of the chromosomal apparatus of germline tissue (that is regarded as systemic mutations) during the phylogeny of a number of dipterous insects are systematized. A concept on the essence of systemic mutations and their crucial role in saltatory transformation of genomes in the course of species evolution is advanced.

Animals↗

PubMed1996

Prerequisites for tissue specific and position independent expression of a gene locus in transgenic mice.

The elucidation of general parameters influencing the transcriptional activation of gene loci at distinct stages of development is an essential prerequisite for a reproducibly successful gene transfer in both gene therapy protocols and biotechnology. Up to now research has focused mostly on the identification and characterization of individual cis-regulatory elements by transient transfection and in vitro assays. However, the most relevant assay system to test gene constructs designed for gene therapy protocols is the transgenic animal. In such an experimental system exogenous genes are usually integrated randomly in the chromatin. For gene constructs not fulfilling the requirements for correct gene locus activation this can lead to genomic position effects on gene expression. The consequences are highly variable expression levels and a disturbance of temporal and spatial expression patterns. Hence it is important to examine how cis-elements function in a chromatin context, and how they cooperate during the developmentally controlled activation of an entire gene locus. One among a few gene loci which are sufficiently characterized to enable such investigations is the chicken lysozyme locus. This review summarizes recent results aimed at identifying the necessary prerequisites for a reproducibly correct expression of the lysozyme locus in transgenic mice and the implications of our findings for gene transfer. The complete lysozyme locus is expressed independent of the chromosomal position and at a high level in macrophages of transgenic mice. Correct transgene regulation requires the cooperation of all cis-regulatory elements. Chromatin of the lysozymes locus in both the active and the inactive state is highly structured. Each cis-regulatory element on the chicken lysozyme locus is organized in its own unique chromatin environment, with nucleosomes specifically placed on specific sequences. The transcriptional activation of the lysozyme locus is accompanied by extensive rearrangements of its chromatin structure, which are disturbed when the transgenes are subjects to genomic position effects. Based on these results, we propose that a complete locus is resistant to genomic position effects, and that a distinct chromatin architecture of a gene locus is required for its correct activation.

Animals↗

PubMed2005

An analysis of genetic diversity across the maize genome using microsatellites.

How domestication bottlenecks and artificial selection shaped the amount and distribution of genetic variation in the genomes of modern crops is poorly understood. We analyzed diversity at 462 simple sequence repeats (SSRs) or microsatellites spread throughout the maize genome and compared the diversity observed at these SSRs in maize to that observed in its wild progenitor, teosinte. The results reveal a modest genome-wide deficit of diversity in maize relative to teosinte. The relative deficit of diversity is less for SSRs with dinucleotide repeat motifs than for SSRs with repeat motifs of more than two nucleotides, suggesting that the former with their higher mutation rate have partially recovered from the domestication bottleneck. We analyzed the relationship between SSR diversity and proximity to QTL for domestication traits and observed no relationship between these factors. However, we did observe a weak, although significant, spatial correlation for diversity statistics among SSRs within 2 cM of one another, suggesting that SSR diversity is weakly patterned across the genome. Twenty-four of 462 SSRs (5%) show some evidence of positive selection in maize under multiple tests. Overall, the pattern of genetic diversity at maize SSRs can be explained largely by a bottleneck effect with a smaller effect from selection.

DNA, Plant↗

PubMed2026

Agentic AI for Spatial Omics.

This highlight summarises recent advances in agentic artificial intelligence (AI) systems for spatial omics analysis. These systems are compared along two central tensions: autonomy versus accountability, and adaptability versus reproducibility. We argue that progress will depend not on maximising automation, but on defining where autonomy is appropriate.

Artificial Intelligence↗

PubMed2026

Evolutionary and resistance dynamics in oligometastatic and oligoprogressive cancer treated with stereotactic radiotherapy and systemic therapies: A systematic review and focused meta-analysis.

BACKGROUND: Oligometastatic and oligoprogressive disease treated with stereotactic ablative radiotherapy (SABR) represents a clinically heterogeneous entity. Increasing evidence suggests that anatomical definitions alone may not adequately capture underlying biological diversity. This systematic review aimed to synthesize translational evidence exploring evolutionary dynamics, resistance mechanisms, and biomarker-driven stratification in patients treated with SABR. METHODS: A systematic literature review was performed including prospective and retrospective studies evaluating translational biomarkers in oligometastatic or oligoprogressive settings treated with SABR. Studies assessing genomic, transcriptomic, circulating or immune-related biomarkers were included. Data were summarized qualitatively according to predefined translational domains: (i) evolutionary dynamics under systemic therapy pressure, (ii) baseline biological stratification, (iii) longitudinal circulating biomarkers, and (iv) systemic immune remodeling. Exploratory quantitative visual syntheses were performed using reported hazard ratios when conceptually comparable endpoints were available. RESULTS: 19 studies comprising 1527 patients were included. Across tumor types and treatment contexts, translational analyses consistently indicated that anatomically defined oligometastatic states encompass biologically distinct subgroups with different risks of systemic progression. Studies evaluating oligoprogression under ongoing systemic therapy suggested a distinction between spatially constrained resistance and systemic molecular escape, supported by circulating tumor DNA and tissue- or plasma-based molecular profiling (including genomic and transcriptomic analyses). Baseline biological features, including adverse genomic signatures and circulating biomarkers, were associated with inferior progression outcomes despite metastasis-directed therapy. Longitudinal biomarkers provided early signals of treatment response and systemic control. Immune remodeling after SABR showed context-dependent effects, both systemic immune activation and treatment-related immunosuppression reported across studies.

Humans↗

PubMed2005

Identification of a hypaxial somite enhancer element regulating Pax3 expression in migrating myoblasts and characterization of hypaxial muscle Cre transgenic mice.

Pax3 encodes a transcription factor that functions in the embryonic central nervous system, neural crest, and somitic mesoderm. Prior studies suggest that distinct regulatory elements regulate temporal and spatial expression of Pax3 in neural crest and mesoderm. Here, we describe a discrete enhancer element, conserved between mouse and human genomes, that directs Pax3 expression in the ventral-lateral lip of interlimb somites. These regions give rise to hypaxial musculature including limb, ventral body wall, diaphragm, and tongue muscles. Transgenic mice harboring the hypaxial muscle enhancer driving lacZ expression initiate beta-galactosidase expression at E10.0, significantly later than endogenous Pax3 expression in presomitic and segmented mesoderm. Initiation of transgene expression is not dependent on Pax3 itself, since expression is detectable in homozygous Splotch embryos. Transgenic mice expressing Cre recombinase in hypaxial myoblasts were generated and characterized. These results suggest that Pax3 is differentially regulated within the somite in both spatial and temporal domains. Hypaxial muscle Cre mice will allow for specific manipulation of gene expression in this subset of developing skeletal muscle.

Animals↗

PubMed2005

Isolation of a cotton reversibly glycosylated polypeptide (GhRGP1) promoter and its expression activity in transgenic tobacco.

Reversibly glycosylated polypeptides (RGPs) are thought to be involved in polysaccharide metabolism. A cDNA of the cotton (Gossypium hirsutum) RGP gene, designated GhRGP1, has previously been characterized, and is preferentially expressed in fiber cells. In order to investigate its temporal and spatial control, we isolated a 624bp fragment upstream of the GhRGP1 coding sequence using a polymerase chain reaction (PCR)-based genomic walking method, transcriptionally fused the 624bp promoter sequence to the beta-glucuronidase (GUS) gene, and analyzed the stable gene expression in tobacco (Nicotiana tabacum). In 4-week-old transgenic tobacco plants, the highest expression level was observed in roots, and the GUS activity was 1.13- and 6.65-fold higher than that in stems and leaves, respectively. In the reproductive growth stage, the GUS expression level was highest in the pistils and the GUS activity in the stigmas and styles were 17.6-fold higher than that in the ovaries. High GUS activity was also detected in the anthers. In addition, histochemical staining for GUS activity on transgenic tobacco plants further indicated a higher expression in the trichomes, seeds and vascular tissues of stems. Abiotic stress treatments on transgenic tobacco plants showed that wounding and dehydration induced GUS expression. These results demonstrated the spatial and temporal regulation of a cotton RGP promoter in a model plant, and provided an important insight into the factors that control the fiber development and stress responses of the gene.

Base Sequence↗

PubMed2001

The effect of MAR elements on variation in spatial and temporal regulation of transgene expression.

The level of transgene expression often differs among independent transformants. This is generally ascribed to different integration sites of the transgene into the plant genome in each independently obtained transformant (position effect). It has been shown that in tobacco transformants expressing, for example, a cauliflower mosaic virus (CaMV) 35S promoter-driven beta-glucuronidase (GUS) reporter gene, these position-induced quantitative differences among individual transformants were reduced by the introduction of matrix-associated regions (MAR elements) on the T-DNA. We have previously shown by imaging of in planta firefly luciferase (luc) reporter gene activity that quantitative differences in transgene activity can be the result of either a variation in (1) level, (2) spatial distribution and/or (3) temporal regulation of transgene expression between independent transformants. It is not known which of these three different aspects of transgene expression is affected when the transgene is flanked by MAR elements. Here we have used the firefly luciferase reporter system to analyse the influence of MAR elements on the activity of a CaMV 35S-luc transgene in a population of independently transformed tobacco plants. Imaging of in planta LUC activity in these tobacco plant populations showed that the presence of MAR elements does not result in less variation in the average level of transgene expression between individual transformants. This result is different from that obtained previously with a 35S-GUS reporter gene flanked by MAR elements and reflects the differences in the stability of the LUC and GUS reporter proteins. Also the variation in spatial patterns of in vivo LUC activity is not reduced between independent transformants when the transgene is flanked by MAR elements. However, MAR elements do seem to affect the variation in temporal regulation of transgene expression between individual transformants. The potential effects of MAR elements on the variability of transgene expression and the relation to the stability of the (trans)gene product are discussed.

Gene Expression Regulation, Developmental↗

PubMed2025

Deep Learning on Histologic Slides Accurately Predicts Consensus Molecular Subtypes and Spatial Heterogeneity in Colon Cancer.

Colon cancer (CC) is the third most prevalent cancer type. It is highly heterogeneous, particularly in terms of molecular profiles, which have both prognostic and predictive impacts on the treatment efficacy. However, CC treatment in adjuvant situations is currently guided solely by T and N staging. In this context, consensus molecular subtypes (CMSs) were introduced to stratify patients with CC based on molecular profiles. Recent studies have shown that CMS can be heterogeneous in CC, leading to a worse prognosis. This study focused on predicting CMS and its heterogeneity in CC using deep learning on digitized hematoxylin and eosin ± saffron-stained whole-slide images. Data and whole-slide images of 1996 patients from the PETACC-8, The Cancer Genome Atlas-COAD, and PRODIGE-13 cohorts were used. The model is trained to predict a 4-dimensional CMS vector, reflecting intratumor heterogeneity (ITH). It comprises a self-supervised model for embedding image patches into vectors and a weakly supervised model predicting CMS calls. Ground-truth CMS scores are obtained with the CMSclassifier package. Interpretability analyses are performed at the slide and patch levels. For homogeneous tumors, the model trained on PETACC-8 achieves 93.0% (±1.4%) macroaverage area under the curve in internal cross-validation and 94.4% macroaverage area under the curve in external validation over PRODIGE-13, whereas the The Cancer Genome Atlas-COAD model reaches 85.4% (±3.0%) in cross-validation and 92.4% over PRODIGE-13. The trained models also provide spatial distributions of CMS across tumor slides and associate specific histologic features with each CMS. Finally, the models are able to predict ITH. The results show that a deep learning model trained on routine histology slides is capable of providing an efficient and robust method for predicting CMS and characterizing a patient's ITH, paving the way for the routine consideration of CMS/ITH in clinical decision making in the adjuvant setting.

Humans↗