Profiling colon cancer architecture with spatial transcriptomics identifies clinically relevant stromal ecotypes

Profiling colon cancer architecture with spatial transcriptomics identifies clinically relevant stromal ecotypes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Profiling colon cancer architecture with spatial transcriptomics identifies clinically relevant stromal ecotypes Sophie Mouillet-Richard, Antoine Cazelles, Camilla Pilati, Delphine Le Corre, and 17 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8583137/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Despite advances in characterizing intra-tumor heterogeneity in colon cancer (CC), its spatial organization remains to be fully delineated. We generated a large spatial transcriptomic atlas of 48 stage III CC and identified recurrent, biologically relevant spatial ecosystems. Recurrence-associated analysis revealed stromal-specific upregulation of a set of genes including the regenerative/revival-stem cell (REC/RSC) marker ANXA1, together with enrichment of a YAP-TEAD program. These signals converged into a refined three-gene YAP-Rev signature (AHNAK, ANXA1, CAPN2) capturing the fetal-like state associated with relapse. Deeper dissection of the stromal compartment revealed additionally layers of heterogeneity. Further Visium HD profiling allowed spotting ANXA1-expressing tumor cells surrounded by collagen-producing cancer-associated fibroblasts (matCAFs). Finally, we translated these findings to bulk transcriptomics and demonstrated a synergistic interaction between matCAF enriched stroma derived-signature and ANXA1-REC/RSC score with strong prognostic value across three independent cohorts comprising >3,500 stage II and stage III CC patients. Biological sciences/Cancer/Gastrointestinal cancer/Colorectal cancer/Colon cancer Biological sciences/Cancer/Cancer microenvironment Biological sciences/Cancer/Tumour biomarkers Health sciences/Diseases/Cancer/Gastrointestinal cancer/Colorectal cancer/Colon cancer Figures Figure 1 Figure 1 Figure 2 Figure 2 Figure 3 Figure 4 Figure 5 Figure 5 Figure 6 Introduction Colon cancer (CC) remains the fourth diagnosed cancer worldwide, with over 1.1 million new cases in 2020 1 . By 2040, colorectal cancer is expected to affect over 3 million individuals, leading to over 1.5 million deaths due to increasing incidence in emerging economies 2 . As other cancer types, CC is a heterogenous disease, which may notably vary according to the primary tumor location (left or right-sided), disease stage at diagnosis, carcinogenic pathway (serrated or tubular), as well as molecular features. Beyond the distinction between microsatellite stable (MSS) or unstable (MSI) tumors and the specificities related to several key driver mutations (e.g. KRAS or BRAF ), the molecular taxonomy of CC is usually depicted through the lens of the consensus molecular classification into 4 consensus molecular subtypes CMS1 to CMS4, following the landmark study by Guinney et al. 3 . This molecular classification, based on bulk gene expression, has prognostic implications and was shown to be accompanied by specific immune and stromal contextures 4 . Adding some complexity to the picture, we demonstrated through deconvolution approaches that a large proportion of CC tumors actually belong to multiple CMS, entailing a high degree of intra-tumor heterogeneity 5 . Along the same line, Langerud et al. recently documented a high level of CMS heterogeneity based on multiregional sampling 6 . Despite these overall advances, our understanding of the spatial organization of tumor cells and their surrounding tumor microenvironment (TME) in colon cancer is still far from complete. In recent years, the advent of spatial transcriptomics technologies has revolutionized studies focusing on tissue heterogeneity, notably in the field of cancer 7,8 . Until recently, these were restricted to fresh frozen tissue, and the extension of existing tools to formalin-fixed paraffin embedded (FFPE) samples in 2021 9 has substantially expanded the possibility to investigate retrospective collections, and hence to evaluate tumor architecture in relation with patient outcome 8 . In CC, spatial transcriptome analyses have allowed to infer cell communication events in the tumor-stroma interface of CMS2 carcinomas 10 , shed light on the heterogeneity of liver metastases 11–13 , or reveal interactions between cancer-associated fibroblasts (CAFs) and macrophages subtypes 14 . Nevertheless, in most instances, the number of samples studied was limited, possibly due to the cost of these technologies, thus precluding the statistical power to bring to light some particular traits related to individual patient features. Another point that is worth considering when attempting to discover spatial molecular features with broad clinical implications is that samples should belong to a homogeneous category of disease (e.g. in term of disease stage) while at the same time representing the inter-patient tumor heterogeneity (e.g. in terms of age, sex, clinical and molecular profiles). Overall, achieving this goal necessitates to analyze a significant number of samples within a homogeneous collection. On another hand, while single-cell and spatial transcriptomics data are powerful tools to investigate tumor heterogeneity, they are not compatible with routine clinical practice for individual patient care. For this reason, it remains often difficult to translate spatial transcriptomics-related discoveries into molecular signatures applicable in daily care 15 . With these considerations in mind, our goal was to generate and analyze a spatial transcriptomics dataset with the following specifications: (1) for sample selection, combine homogeneity of disease stage and clinical diversity in terms of disease outcome and (2) for data analysis, mine for spatially-relevant features associated with disease outcome to be further validated in bulk transcriptomic datasets. Here we present the largest spatial transcriptomic dataset to date in non-metastatic colon cancer, generated from 48 stage III tumors collected in the randomized PETACC8 clinical trial 16 . Using FFPE-compatible Visium technology (10x Genomics), we chart the spatial architecture of tumor and tumor-associated stromal compartments and identify recurrent spatial ecosystems, including tumor and stromal, with specific molecular hallmarks and signatures. Comparative analysis based on recurrence status uncovered a stromal-specific increase in ANXA1 expression together with an associated regenerative cells (REC) / revival stem cell (RSC) program in tumors that later relapse. Unsupervised dissection of the stromal compartment unveiled additional layers of heterogeneity. Re-profiling a subset of 4 samples with Visium HD allowed to spot ANXA1 -expressing tumor cells embedded in stroma. Translating these findings to bulk transcriptomic data, we show that spatially-derived signatures carry prognostic values in two large independent cohorts of stage III colon cancer (PETACC8, IDEA-France) and a newly generated cohort including 529 stage II patients (PRODIGE13), altogether comprising over 3,500 patients. Finally, multivariate analyses demonstrated a synergistic interaction between matCAF-enriched stroma derived-signature and ANXA1 -REC/RSC score, and identified their combined high expression as a biologically grounded, clinically meaningful marker of tumor aggressiveness. Results Generation of a spatial transcriptomic dataset of stage III colon cancer With the aim of understanding the architecture of localized colon cancer, we generated spatial transcriptomic profiles of a set of 48 primary resected tumors at baseline (Fig. 1a). Samples originated from the PETACC8 study, an open-label, randomized phase 3 trial of oxaliplatin, fluorouracil, and leucovorin with or without cetuximab in patients with resected stage III colon cancer 16 . Among those, 14 patients subsequently experienced disease recurrence (Relapse) while the remaining 34 patients did not (No relapse) (see Supplementary Table S1 for patient characteristics). 10X Genomics Visium spatial transcriptomics was applied to 5 µm-thick FFPE sections enriched in tumor tissue (see Methods). In order to identify recurrent histo-molecular spatial patterns, or spatial archetypes, across our 48 colon cancer samples, we designed a four-step strategy (Fig. 1b; see also Materials and Methods). First, we applied high-resolution Leiden clustering independently to each Visium sample, aiming to delineate transcriptionally coherent spot clusters (Fig. 1b). In the second step, we summarized each cluster into a “meta-spot” by averaging gene expression values across all constituent spots, yielding a median of 61 meta-spots per sample. Importantly, this approach ensured that each meta-spot represented a single cluster regardless of its size, allowing small but biologically meaningful regions to be preserved and preventing them from being overshadowed by larger clusters. The median spot number per meta-spot across the 48 samples was 57 (range 46-133) (Supplementary Fig. S1a and 1b). In the third step, we aggregated all meta-spots from the 48 samples into a single dataset, resulting in 3,231 meta-spots overall (Fig. 1b). To integrate this multi-sample meta-spot dataset, we used the Harmony algorithm and built a unified Seurat object. In the final step, we performed unsupervised clustering on this object and sought to select an appropriate clustering resolution to balance cluster granularity and stability (Supplementary Fig. S1c and 1d). This led to the identification of 11 meta-clusters. These meta-clusters grouped meta-spots from different patients (Supplementary Fig. S1e), highlighting successful integration. Furthermore, when meta-cluster assignments were projected back onto each sample, we observed spatially organized patterns composed of multiple meta-clusters (Fig. 1b and Supplementary Fig. S2a), analytically confirmed by determining Moran’s I spatial autocorrelation indexes (Supplementary Fig. S2b). Meta-clustering identifies spatial ecotypes with specific CMS and TME combinations The above results support the existence of spatial ecotypes: shared, spatially structured transcriptional units recurrently found across tumors, both in tumor cores and in adjacent non-malignant tissue 15 . We then sought to characterize these meta-clusters by assessing molecular (Fig 2a and Supplementary Fig. S4) and morpho-histological features (Fig 2b and Supplementary Fig. S3 for representative sample #74_D1). Molecular analyses were performed by computing signatures for CMS, CMS-related features as well as immune-stromal hallmarks. Projections of ecotypes on Visium capture areas were reviewed by a pathologist (JFE). Histological examination revealed the presence of glandular tumor cells in four clusters (clusters 0, 1, 7 and 8) (Fig 2b). The heatmap built on the 3,231 meta-spots pre-classified through unsupervised clustering showed molecular similarity between 3 of those meta-clusters, with the highest CMS2 and CMS3 scores and epithelial signatures, including that derived from the single cell atlas of colon cancer 17 (Fig 2a and c and Supplementary Fig. S4). One of them (meta-cluster 0, red) had the highest mean Copy Number Variation (CNV) fraction as determined with our recently developed FastCNV algorithm (https://github.com/must-bioinfo/fastCNV) (Supplementary Fig. S4). It also displayed the highest score for Wnt and cell cycle signature and was thus annotated as “tumor-cycling” (Fig 2b). Meta-cluster 1 (blue) had the second highest scores (Supplementary Fig. S4) and was termed “tumor-intermediate”. Meta-cluster 7 (pink) was transcriptionally close to the “tumor-cycling” and “tumor-intermediate” meta-cluster according to unsupervised clustering (Fig 2a), albeit with a lower cell cycle score and Wnt signature, suggesting a lower content of tumor cells and a higher abundance of non-tumor cells (Supplementary Fig. S4), and was thus denoted “tumor-infiltrated”. The last one (meta-cluster 8, grey) had meta-cluster 3 (purple) as its closest meta-cluster according to Fig 2a, which corresponded to the muscularis propria. Meta-spots from meta-cluster 8 overlapped with myofibroblasts and it was therefore termed “tumor-myofibroblasts” (Fig 2c). Two other meta-clusters were enriched in CMS1/CMS4 as well as various mesenchymal and stromal (in particular fibroblast) signatures (Fig 2a and c and Supplementary Fig. S4). Histological examination confirmed their stromal identity (Fig 2b). Analysis of meta-spots neighborhood indicated that one of them, designated “stroma-A” (meta-cluster 2, green), was closer to tumor ecotypes, in line with a higher epithelial score (Fig 2c and Supplementary Fig. S4), while the other, annotated as “stroma-B” (meta-cluster 4, orange), was further away from tumor ecotypes (Fig 2d). Stroma-A accounted for 77.7% of all stromal contacts in the immediate tumor neighborhood, compared with only 22.3% for stroma-B (3.48-fold difference, p < 0.0001). Finally, we identified meta-clusters corresponding to necrotic regions (characterized by the highest neutrophil infiltration score, meta-cluster 9, black), lymphoid aggregates enriched in B and T cells (meta-cluster 10, bright blue), connective tissue including nerves or arteries (meta-cluster 5, yellow), and normal colon epithelium (meta-cluster 6, brown) (Fig 2b and Supplementary Fig. S4). While tumor and stromal ecotypes were consistently present across most samples, their relative abundances varied markedly between individuals (Supplementary Fig. S5). Together, these findings indicate that we have constructed a spatial atlas of stage III colon cancer from a clinically homogeneous cohort, yet displaying substantial heterogeneity in spatial organization and microenvironmental composition. Stromal ecotypes from tumors of patients with relapse display an RSC-like program enriched in ANXA1 We next leveraged the diversity of our cohort to explore specific spatial features associated with disease recurrence. For each sample, we calculated the proportion of spots assigned to each meta-cluster and examined the distribution of ecotypes relative to tumor or stromal content (Supplementary Fig. S6). Focusing on tumor and stromal ecotypes only, we found that the distributions of ecotypes proportions were highly variable among patients, with a tendency towards lower global proportions of tumor spots and higher proportions of stromal spots in patients with relapse versus patients without relapse, especially when restricting analyses to pMMR patients (Supplementary Fig. S7). These observations align well with the well-established notion that stromal-rich tumors are associated with a poorer prognosis than stromal-poor tumors in colon cancer 18,19 . The above results prompted us to further scrutinize the stromal ecotypes and to perform differential gene analysis between stromal meta-spots from patients with relapse (n=247) and patients without relapse (n=527). We identified a set of 16 genes upregulated in the global stromal compartment of patients that later relapsed (Fig .3a). The levels of these genes were not different when considering all meta-spots combined (Fig. 3b). However, we found that among them, 15 genes were systematically decreased in at least one out of the 4 tumor ecotypes (Supplementary Fig S8A) and that they were all increased in both the stroma-A and stroma-B ecotypes of tumors that later relapsed (Fig. 3b). Interestingly, 4 out of 16 of these genes ( AHNAK , ANXA1 , CAPN2 and LRP10 ) feature among those specifying the recently described Regenerative Cells (REC) and their related inflammatory RECs (iRECs), two colon cancer cell states associated with metastasis 20 (Fig. 3c). This suggests the presence of such cancer cells in the stroma of patients who will experience disease recurrence. In accordance, REC and iREC gene signatures computed with ssGSEA were enriched in the stromal compartment of patients with relapse as compared to those without relapse (Fig. 3d and Supplementary Fig S8B). To identify potential transcriptional regulators of these four genes, we performed Enrichr analysis on this gene set against the ENCODE-TF database, which highlighted a significant enrichment in TEAD4, a transcription factor that relays YAP activity and targets AHNAK , ANXA1 and CAPN2 (Fig. 3e). These results are consistent with the presence of YAP–TEAD–activated cancer cells within stromal regions of patients with relapse (Fig. 3f). In line with this hypothesis, we found a differential enrichment in YAP-associated signatures 21,22 in the stromal meta-spots of patients with versus without relapse (Fig. 3g and Supplementary Fig S8b). The YAP pathway is a major determinant of the Revival Stem Cell (RSC) state 23 , which globally corresponds to RECs 20 . Accordingly, we also found significant enrichments in the Revival Stem Cell (RSC) signature 24 in the stromal compartment of patients with versus without relapse (Fig. 3h and Supplementary Fig S8b). Finally, in view of the demonstration that mouse CC organoids are reprogrammed into an RSC-like (also termed fetal-like) state upon stromal collagen-induced YAP activation 25 , we probed potential differential associations between ANXA1 and collagen-expressing genes according to relapse. Interestingly, stronger correlations were observed between levels of ANXA1 and those of different collagen type VI-expressing genes in the two stromal compartments of patients with as compared to without relapse (Supplementary Fig. S9). Altogether, by using an unbiased approach, we uncovered enriched ANXA1 expression and a YAP-associated REC/RSC state specifically within the stromal compartment of tumors that later relapsed. Taking into account pre-clinical studies and single-cell analyses 20,23,25,26 , our spatially resolved analysis supports the notion that, in patients who eventually relapse, the stromal compartment harbors cancer cells in an ANXA1 ⁺ REC/RSC state promoted by a YAP-dependent transcriptional program, itself potentially sustained by local collagen enrichment. Stromal meta-clusters split into distinct subclusters with specific molecular features Given the consistent enrichment of ANXA1 + REC/RSC-like cells in the stromal compartment of tumors that later relapse, we next sought to refine stromal states by sub-setting the 774 meta-spots assigned to stromal compartments and performing unsupervised re-clustering. This revealed five transcriptionally distinct stromal subclusters (Fig. 4a). The stroma-A ecotype broke down into 2 stromal subclusters. The first one, denoted “stroma-A1”, was ubiquitous across samples and exhibited the highest universal fibroblast signature (normal fibroblasts, NFs) and, to a lesser extent, various CAF signatures (Fig. 4b and Supplementary Fig. S10 and S11). The latter, which include the CXCL14 + CAF and the GREM1 + CAF signatures from 17 and the matCAF signature from 27 , were more enriched in the stroma-A2 subcluster, present in 44 of the 48 samples (Fig. 4b and Supplementary Fig. S10 and S11). Stroma-A2-specifying genes included the prototypical CAF markers LRRC15 and POSTN (Supplementary Table S2), both described to sustain tumor growth 28,29 . Among the stroma-B meta-cluster, three subclusters were identified. The stroma-B1 subcluster, found in 43 samples, was enriched in various immune cells, including monocytes, T, NK, B, and cytotoxic lymphocytes cells as well as myeloid-dendritic cells, consistent with an immune-active niche (Fig. 4b). The less frequent stroma-B2 subcluster (27 of the 48 samples, Supplementary Fig. S11), was characterized by high signatures of myofibroblasts, and expression of the contractile CAF marker ACTG2 30 (Supplementary Table S2). Lastly, the stroma-B3 subcluster, which gathered only 24 meta-spots from 20 samples, was defined by high expression of various cytokines and chemokines, including IL11 (Supplementary Table S2). It was characterized by signatures of Inflammation-associated fibroblasts (IAFs) and an enrichment in neutrophils and endothelial cells. This subcluster is also marked by an elevated signature of MMP3 + CAF described by Pelka et al to be highly active in dMMR patients 17 . In agreement, it was significantly more represented in dMMR vs pMMR patients and in patients without relapse vs patients with relapse (Fig. 4c). Then, we examined the relative expression of the set of genes found to be enriched in the stromal compartment of patients with relapse, in each stromal subcluster, according to disease recurrence (Supplementary Fig. S12). Analyses were restricted to pMMR patients in view of the disbalanced proportions of the different stromal subclusters in dMMR patients (see above). Besides, analyses specific to the stroma-B3 subcluster were considered irrelevant due to the limited number of meta-spots. In line with the stromal-A1 subcluster exhibiting a generic fibroblast phenotype, there was barely any change in this ecotype between patients with or without relapse (Supplementary Fig. S12). In contrast, nearly all genes were more abundantly expressed in the stroma-A2 and stroma-B2 compartments of patients with relapse vs patients without relapse (Supplementary Fig. S12). Taken together, these findings unveil distinct stromal archetypes with different fibroblast and immune contextures. Visium HD reveals the spatial arrangement of stromal subclusters and maps ANXA1 -expressing REC/RSC-like cells in surrounding stroma Having identified stromal archetypes across our spatial transcriptomic atlas, we sought to refine their respective distribution at a microscale. To this purpose, we re-profiled 4 samples from our initial Visium dataset with Visium HD (Fig. 5a), allowing a single cell-state resolution 31 . We additionally leveraged the publicly available the P2CRC sample from 10X Genomics 31 . EPCAM and CEACAM6 (as in 31 ) were selected as markers for tumor cells (Supplementary Fig. S13). Each stromal subcluster was mapped via the projection of the combined list of its specific marker genes (log-normalized average expression) (Supplementary Fig. S13). The stroma-A1 signature yielded a pervasive, diffuse and faint signal, in contrast to the stroma-A2 signature that intensively stained specific regions outside glandular tumor zones. The stroma-B1 signature produced a faint signal lining tumor cells. The stroma-B2 signature was enriched in regions annotated as smooth muscle and gave a weaker non region-specific signal. Finally, the stroma-B3 signature yielded intense signals at defined zones, notably tumor borders, which may correspond to ulcerative regions. We then sought to determine which cell type(s) within the stroma express the 3 YAP target genes AHNAK , ANXA1 and CAPN2 . Upon inspection of the stroma-A2 enriched region of 74D1_HD sample at high magnification, we observed clusters of cells that were positive for the 3-gene signature, denoted “YAP-Rev”, while they expressed low levels of EPCAM as compared with glandular tumor cells (Fig. 5b). These clusters were annotated by a pathologist (JFE) as poorly differentiated tumor cells (arrows in Fig. 5b). Co-expression analysis distinctly localized such clusters of cancer cells embedded in matCAF-enriched stroma A2 (Fig. 5b). As a whole, these observations reveal a specific, spatially organized distribution of stromal sub-clusters at high resolution and provide in situ evidence for the presence of “YAP-Rev” cancer cells within a matCAF-enriched microenvironment. Spatial transcriptomics-derived signatures are applicable to bulk transcriptomic data and display prognostic value While single-cell and spatial transcriptomics analyses are not designed for large clinical cohorts, they may nevertheless bring to light single or composite biomarkers of interest 15,32 . We therefore leveraged our large bulk RNAseq datasets of stage III colon cancer, totaling nearly 3,000 patients 33 , to test whether spatially derived stromal and tumor-intrinsic programs retain prognostic value at the population level. Two independent phase 3 cohorts were analyzed: the full PETACC8 cohort (n=1,733 patients, stage III), from which the spatial transcriptomics samples originated, and 1,248 stage III patients from IDEA-France cohort. Prognostic performance (time to recurrence) of the spatially-derived signatures, computed as continuous z-scores in uni- and multi-variate COX models, is summarized in Fig. 6a-b. Across both cohorts, tumor-intrinsic signatures were consistently associated with favorable prognosis, whereas stromal signatures exhibited adverse effects. Among stromal sub-clusters, the stroma-A1 signature correlated with good outcome, while stroma-A2 reproducibly associated with early recurrence. For stroma-B1 to B3 signatures, results differed between cohorts, likely due to distinct sampling strategies (whole tumor slides for PETACC8, intratumoral punches for IDEA-France). Finally, the “YAP-Rev” signature composed of AHNAK , ANXA1 and CAPN2 was associated with poor prognosis in both datasets. We next evaluated these signatures in stage II colon cancer, where stratification tools are scarce 34 . To this purpose, we generated a bulk transcriptomic dataset of n=529 stage II patients from the PRODIGE13 study 35 (see Supplementary Table S3 for patient characteristics). As shown in Fig. 6c, results from PETACC8 and IDEA-France cohorts were globally recapitulated in the PRODIGE13 cohort. To determine whether ANXA1 + REC/RSC cells provide clinical information beyond the stromal-A2 program, we first evaluated each signature separately in multivariable Cox models adjusted for grade, WHO performance status, baseline recurrence risk, mismatch repair status, and bowel complications at diagnosis (occlusion or perforation). In these covariate-adjusted models, higher expression of Stroma-A2 was significantly associated with shorter recurrence-free survival (HR = 1.22, 95% CI 1.10–1.36, p = 1.7×10⁻⁴), and YAP-Rev showed a similarly adverse effect (HR = 1.24, 95% CI 1.12–1.38, p = 2.6×10⁻⁵) (Fig. 6a). When both signatures were included simultaneously, each remained independently prognostic (stroma-A2: HR = 1.17, 95% CI 1.06–1.30, p = 0.0025; YAP-Rev: HR = 1.21, 95% CI 1.09–1.34, p = 0.0004). Notably, the combined model revealed a statistically significant interaction between the two programs (HR = 1.15, 95% CI 1.04–1.27, p = 0.0047) when both signatures were included as continuous variables in the multivariable Cox model, demonstrating that the effect of each transcriptional program on recurrence risk depends on the activity of the other. Interaction curves further highlighted this statistical interaction, compatible with a biological synergy between the two programs: across increasing stroma-A2 values, patients with high YAP-Rev exhibited a markedly steeper risk gradient, indicating that stromal activation amplifies the detrimental effect of YAP-Rev score. Reciprocally, the risk associated with YAP-Rev was substantially greater in tumors with high stroma-A2 activity. Together, these findings identify a biologically aggressive subgroup characterized by simultaneous stromal activation and YAP-REC/RSC cells (Fig. 6d). To externally validate this interaction structure, we computed unscaled, cohort-independent single-sample ssGSVA scores for stroma-A2 and YAP-Rev and applied PETACC8-derived median thresholds to define four groups (Low/Low, High/Low, Low/High, High/High) across cohorts. In PETACC8, the high stroma-A2 /high YAP-Rev group showed the poorest prognosis (HR 2.04, 95% CI 1.59–2.60, p < 0.001) (Fig. 6e). This phenotype was reproduced in both validation cohorts. In PRODIGE13, the High/High group displayed a markedly elevated recurrence risk (HR = 2.13, 95% CI 1.17–3.85, p = 0.013), whereas the single-high groups showed no significant effect. IDEA-France showed the same risk hierarchy (High/High: HR = 1.43, 95% CI 1.10–1.87, p = 0.008) (Fig. 6f), while intermediate groups exhibited weaker or absent associations. The slightly attenuated effect in IDEA-France aligns with its sampling approach: intratumoral punch biopsies underrepresent the stromal compartment. Overall, across three independent cohorts and using PETACC8-derived thresholds, the high stroma-A2 /high YAP-Rev phenotype consistently identified the subgroup with the worst recurrence-free survival, whereas isolated elevation of only one program provided limited prognostic information. These findings robustly support a synergistic interaction between stromal activation and YAP-Rev activity, and position their combined high expression as a clinically meaningful marker of tumor aggressiveness. Discussion In this study, we have built a spatial transcriptomics atlas of 48 stage III colon cancers, enabling us to mine the diversity of cellular ecosystems in relation to patient characteristics, uncover molecular features associated with disease relapse and derive spatially-informed transcriptomic signatures with prognostic relevance. Our main objective was to generate a comprehensive dataset that balances homogeneity (stage III) with biological and clinical diversity (age, sex, MMR status, disease recurrence), while providing a sufficiently large sample size to identify both shared features and statistically robust group-specific traits. A second objective was to leverage spatial transcriptomics data to generate ecosystem-level insights that could be translated into bulk-compatible clinical biomarkers 15 . We identified a structured set of recurrent ecosystems, including tumor- and stroma-enriched, defined by distinct molecular hallmarks and whose spatial organization is consistent with histology. We further described a second level of stroma heterogeneity, allowing to define 5 stromal subclusters characterized by distinct fibroblast subtype signatures, consistent with the notion that fibroblasts dominate the tumor microenvironment 31 . Re-profiling a subset of samples with Visium HD allowed to map these different contingents and revealed in particular a wide-spread distribution of the stroma-A2 ecosystem, sometimes encircling tumor zones, and a more confined pattern for stroma-B3, lining the lumen and most likely corresponding to ulcerative zones. In agreement with this ecotype being enriched in various signatures of IAFs, including one derived from the single cell atlas of colon 17 , the B3 ecotype was more represented in dMMR tumors and in patients who did not experience relapse. Searching for features associated with recurrence allowed identification of a set of 16 genes significantly upregulated in the stromal compartment of patients who later experienced relapse. Importantly, these DEGs exhibited no different expression according to disease recurrence when meta-spots were considered as a global entity, while they displayed a concomitant reduction in at least one tumor ecotypes and increase in the two stromal ecotypes, clearly emphasizing spatial compartment specific differences. One prototypical representative of these DEGs is ANXA1 , a well-defined marker of regenerative cell / revival stem / fetal-like marker 23,26 . ANXA1 was also identified as a marker associated with tumor budding in colorectal cancer 36 . REC/RSC signatures were specifically increased in stromal ecotypes of relapsing patients, but not in pooled meta-spots, supporting the notion that the stromal niche constitutes a master regulator of colon cancer cell fate transitions 37 . Furthermore, in agreement with YAP being a central driver of the REC/RSC/fetal-like phenotype 23,26 , our data point to two other DEGs beyond ANXA1 , namely AHNAK and CAPN2 , as YAP target genes, and to stromal YAP activation as an accompanying feature of disease recurrence. At the second level of stromal ecosystem diversity, the recurrence-associated increase in ANXA1 , AHNAK and CAPN expression and RSC signature are recovered in the stroma-A2 subcluster, marked by the expression of prototypical immunosuppressive (e.g. LRCC15 ) or pro-tumoral (e.g. POSTN ) CAF markers. The stroma-A2 gene signature also features COL10A1 and CTHCR1 , which together with POSTN specify the matCAF state defined in a pan-cancer single cell study of CAF 27 . Taken together, these data suggest that collagen-producing CAFs may contribute to a niche favoring the emergence and persistence of ANXA1 ⁺ REC/RSC cells, potentially through YAP activation driven by POSTN -encoded periostin 29 and/or collagen-dependent mechano-transduction 25 . Consistent with this model, we observed stronger correlations between ANXA1 expression and that of collagen genes ( COL6A1 , COL6A2 , and/or COL6A3 ) in the stroma of patients who experienced relapse compared with those who did not. Our Visium HD data, revealing clusters of undifferentiated tumor cells expressing AHNAK , ANXA1 , and CAPN2 (our YAP-Rev signature) embedded within an A2-enriched stromal environment, provides in situ micro-scale validation of this model. Building on this framework, our data support a model in which matCAF-driven tumor compression triggers local budding. Budding cells then re-activate a YAP-dependent fetal-like program that sustains survival and plasticity within the stromal niche. Based on pre-clinical and clinical studies, these observations raise the possibility that such cells may contribute to resistance to chemotherapy and dissemination 20,36,38 . One remaining is question is how to relate our findings to other cell states associated with disease relapse, in particular the EMP1+ HRCs (High Relapse Cells) 39 . It is noteworthy that this cell population is also marked by high AHNAK expression (see 39 , which may represent a broad marker of recurrence-associated cells. Second, the HRC population appears to be elevated in KRAS G12D mutant versus KRAS non-mutant CC 39 . Third, while fibroblasts were shown to polarize human CC organoids to a RSC state via TGFb and YAP, this transition did not occur in KRAS G12D mutant organoids 40 . Hence, while metastatic dissemination of colon cancer cells relies on cell plasticity 41 , multiple metastasis-initiating cell states may be adopted according to the genomic landscape of the tumor 39,40,42 . We next assessed the clinical value of these ecotype-derived signatures across independent bulk cohorts. Three key features of the studied cohorts were essential to achieving this goal: a large patient population, detailed clinical annotations, and long-term follow-up. Beyond the two stage III colon cancer cohorts that we recently exploited to derive transcriptomic-based prognostic models of recurrence 33 , we here produced a third dataset with stage II patients from the PRODIGE13 trial. Indeed, stage II tumors comprise the majority of colon cancers 43 , and, although most patients are cured after surgery alone, about 15% of patients experience disease recurrence. One key result of our study is the demonstration that the combination of the stroma-A2 with the YAP-Rev signature robustly identifies patients at high risk of relapse not only in stage III but also in stage II patients. Beyond their individual association with relapse, the stroma-A2 and YAP-Rev programs displayed a clinically meaningful synergy. In multivariable models, both remained independently prognostic, yet their significant interaction showed that each program’s impact depends on the other, pointing to a biologically aggressive subtype driven by mutual reinforcement between a collagen-rich CAF niche and YAP-dependent epithelial plasticity. This cooperation was consistently reproduced across cohorts: only tumors simultaneously high for both programs showed a strong increase in recurrence risk, whereas isolated elevation of either had limited effect. In PRODIGE13, this High/High subgroup also included patients who relapsed despite receiving chemotherapy, suggesting that fetal-like ANXA1 ⁺states may withstand cytotoxic stress within matCAF-protected niches. Overall, these findings indicate that stromal activation and epithelial revival-like programs must co-occur within the same spatial ecosystem to drive aggressive disease. Altogether, our findings reinforce the idea that spatial transcriptomics is not only a descriptive tool but a platform to generate actionable biomarkers and translate into clinical applications 15 with the demonstration that ecosystem-specific signatures have prognostic value. Current standard-of-care recommendations for localized colon cancer are largely imperfect, with both overtreated patients at low risk of recurrence and undertreated patients at high risk of recurrence. By translating spatially resolved programs into bulk-compatible signatures, we pave the way for future clinical implementation. Such approaches may improve patient stratification beyond current standards and help identify patients at high risk of relapse who might benefit from specific therapeutic strategies, including CAF-targeting agents 44 . Importantly, these signatures were not trained on clinical outcome, yet they robustly stratify patients across cohorts and treatment contexts. This underscores a fundamental principle: biologically grounded transcriptional programs inherently carry prognostic information. As such, they represent not only biomarkers for patient stratification but also a window into the mechanisms underpinning treatment failure and disease recurrence. Our work provides a framework to functionally dissect the tumor microenvironment and nominates molecular targets, particularly within the stromal compartment, which may inform future therapeutic strategies. The spatial dissection of tumors thus emerges as a powerful tool, not only to classify disease more accurately, but to uncover actionable biology with direct clinical relevance. Methods Study Population and Sample Selection for Spatial Transcriptomics Forty-eight stage III colon adenocarcinomas were selected from patients enrolled in the PETACC8 clinical trial (DOI: 10.1016/S1470-2045(14)70227-X) for spatial transcriptomics analysis. Tumor samples were preserved as FFPE tissue blocks and selected to reflect the clinical and molecular diversity of the cohort, including 14 patients who experienced disease recurrence and 34 who did not, as previously described 33 . Patients enrolled in the PETACC8 trial had histologically confirmed stage III colon adenocarcinoma (pTxN+M0) and received adjuvant chemotherapy: either FOLFOX (oxaliplatin, fluorouracil, leucovorin) or FOLFOX plus cetuximab. Clinical, biological, histological, and molecular data, including survival outcomes, were collected prospectively and made available by the Fédération Francophone de Cancérologie Digestive (FFCD). All patients provided written informed consent for specific translational research and the study protocol was approved by appropriate institutional review boards. Tissue Selection and Processing and Visium SD Spatial Transcriptomics Tissue blocks were selected based on centralized pathological review to ensure the presence of viable invasive carcinoma, adequate RNA quality, and sufficient tissue area compatible with Visium processing. Sections were chosen to be representative of the biological and clinical heterogeneity of the cohort and, when feasible, encompassed tumors with different mismatch repair status and histological features. Most selected sections contained both epithelial and stromal compartments, including tumor–stroma interfaces, rather than being chosen to maximize tumor purity or target a specific anatomical region. Spatial transcriptomics was performed on 5 µm-thick FFPE sections using the 10X Genomics Visium Spatial Gene Expression assay for FFPE tissue (https://www.10xgenomics.com/support/spatial-gene-expression-ffpe/) (RRID RRID:SCR_023571), following the manufacturer’s instructions. Sequencing data were processed using Space Ranger (RRID:SCR_025848) (10X Genomics, version 1.3.0, aligned to the GRCh38 reference transcriptome. Libraries were prepared according to the standard Visium protocol (10X Genomics; Spatial gene expression assay protocol CG000407). Imaging and Histological Annotation H&E-stained images were annotated by a senior pathologist (J.-F. Emile) using NDP.view 2 software. Annotated features included tumor and non-tumor regions, mucinous areas, tertiary lymphoid structures (TLS), and tumor invasion fronts. Annotations were imported in LoupeBrowser v8. The corresponding spatial barcodes were stored in the metadata as qualitative variables. Representative H&E staining sections of the various ecotypes were visualized using the QuPath software. Copy Number Variation Inference Copy number variation was inferred using the fastCNV method (https://github.com/must-bioinfo/fastCNV) applied to each Visium dataset. The CNV burden was summarized as the mean absolute deviation across the genome for each spot or cluster. Meta-spot Generation and Meta-cluster Integration For cross-sample spatial analysis, a second workflow was implemented using Seurat (v5.3.0) (RRID:SCR_016341). Data were normalized using NormalizeData() and scaled with ScaleData(). Within each sample, high-resolution clustering was first performed using FindClusters() at a resolution of 10. For each resulting cluster, gene expression values were averaged on the scaled expression matrix (scale.data) across all constituent spots to generate a “meta-spot”, resulting in 3,231 meta-spots across the 48 tumors. A new Seurat object was created using the meta-spot expression matrix, and the averaged matrix was directly assigned to the scale.data slot. Dimensionality reduction was performed using principal component analysis (PCA) on all genes (50 PCs), followed by batch correction using the Harmony algorithm (v1.2.0) (RRID:SCR_022206), where batch was defined by the sample identity encoded in meta-spot names. The Harmony-reduced matrix was used for UMAP embedding (dims = 1:15), neighborhood graph construction, and clustering (resolution = 0.5). Eleven meta-clusters were identified and annotated based on CMS, TME, histology, and CNV profiles. Two meta-clusters (MC2 and MC4) were identified as stromal and used for downstream stromal subclustering. Stromal Subcluster Identification Meta-spots from MC2 and MC4 were extracted (n=774) and reclustered using Seurat at resolution 0.5. Five distinct stromal subclusters were identified and annotated based on their transcriptional profiles and known CAF or immune markers. Differential Gene Expression Differentially expressed genes in meta-spots of the stromal compartment of patients with versus without relapse were identified using a Wilcoxon rank-sum test using Seurat’s FindMarkers function with the following filters: logfc.threshold >0.6, min.pct = 0.5, adjusted p.value <0.01. Analysis of ENCODE-TF binding motifs was performed with the enrichR package https://github.com/wjawaid/enrichR (RRID:SCR_001575). Signature Generation and Projection For each meta-cluster and stromal subcluster, transcriptional signatures were defined by selecting the top 20 differentially expressed genes (DEGs), identified using FindAllMarkers() with thresholds of adjusted p-value 1. The complete gene lists used to define each signature are provided in Supplementary Table S2. Signature scores were computed using single-sample gene set enrichment analysis (ssGSEA via GSVA package (v1.50.0) RRID:SCR_021058) applied to the scaled expression matrix (scale.data) of each sample. The enrichment was calculated using normalized scores (normalize = TRUE), allowing comparison across gene sets within a given sample. These scores were used for both intra-sample characterization of ecotypes and for projection onto bulk transcriptomic datasets (PETACC8, IDEA-France, PRODIGE13). Spatial neighbor analysis of Visium metaclusters Spatial interactions between metaclusters were quantified from 10x Genomics Visium sections by identifying the six closest tissue neighbors of each spot (first spatial crown). Tissue‐restricted barcodes were extracted using Seurat and mapped to physical coordinates. For every spot, the metacluster identities of its six immediate neighbors were determined using euclidean proximity computed with the knearneigh function ( spdep R package v1.3-5). For each stromal spot, neighboring spots were visualized in an Alluvial plot. Visium HD Spatial Transcriptomics H&E staining and imaging were performed according to the 10X Visium HD FFPE Tissue Preparation Handbook (CG000684). Sample processing and spatial transcriptomics were performed using the 10X Visium HD Spatial Gene Expression Reagents Kits User Guide (CG000685). The Visium HD tissue section was processed using SpaceRanger (10x Genomics, version 3.1.1). Transcript alignment was performed against the GRCh38-2020-A reference transcriptome (refdata-gex-GRCh38-2020-A). Transcript detection was carried out using the Visium Human Transcriptome Probe Set v2.0 (GRCh38-2020-A) provided by 10x Genomics. Fiducial alignment and spatial registration quality control were performed using Loupe Browser version 8. All analyses of the Visium HD slide were performed using Loupe Browser version 8. Gene expression plots were generated using normalized log-transformed expression values, scaled between 0 and 7. For signatures, plots were generated using means of individual normalized log-transformed expression values, scaled between 0 and 7. 3’RNA Sequencing of Tumor Samples The PRODIGE13 study (FFCD PRODIGE-13; ClinicalTrials.gov identifier NCT00995202) is a randomized phase III trial that evaluated the benefit of intensive radiological and CEA monitoring versus standard surveillance in patients with stage II and III colorectal cancer 35 . A subset of 529 stage II tumors from this cohort underwent 3′ RNA-sequencing as part of a translational research program. Briefly, tumor RNA was extracted from macrodissected FFPE tissue sections using the Maxwell RSC RNA FFPE kit (Promega). The PolyA-RNA sequencing (RNAseq) library preparation protocols were performed using 400 ng of template RNA and the QuantSeq 3’mRNA-Seq Kit FWD for Illumina (Lexogen, Vienna, Austria) according to the manufacturer’s instructions. Libraries were sequenced on NovaSeq6000 (Illumina, San Diego, CA). Raw sequencing data was processed as in 33 to generate the dataset. Clinical annotation and survival data were available for all sequenced samples. Survival Analyses Survival analyses were performed on bulk transcriptomic data from PETACC8 (stage III), IDEA-France (stage III) and PRODIGE13 (stage II). Signature scores were treated as continuous variables, quantified by ssGSEA ( GSVA package (v1.50.0)) and standardized per signature (mean = 0, SD = 1) to allow comparability of hazard ratios. Univariate and multivariable Cox proportional hazards models were fitted using the survival package in R (v.3.8-3) (RRID:SCR_021137); multivariable models were adjusted for grade, WHO performance status, baseline high-risk features (pT4 and/or pN2 when available), mismatch repair status and bowel complications at diagnosis (occlusion or perforation). Interaction was assessed using a continuous Stroma-A2 × YAP-Rev term and visualized by estimating predicted hazard ratios with the interacting program fixed at representative low or high values (25th or 75th percentile, Q1/Q3) and centering predictions on a reference patient with median program expression and standard clinical covariates. For categorical stratification, patients were assigned to Low/Low, High/Low, Low/High and High/High groups using PETACC8 median thresholds computed on raw ssGSEA values and applied unchanged to all cohorts. Kaplan–Meier curves and categorical hazard ratios were compared using two-sided log-rank tests and unadjusted Cox models (survminer, ggplot2, broom). Significance was set at p < 0.05. Statistical Analysis Spearman correlations were computed for variable pairs and integrated using the median coefficient. Group comparisons were performed using non-parametric tests (Mann–Whitney U test, Kruskal–Wallis) or parametric equivalents where applicable. All statistical analyses were conducted using R (v4.1.1) on RStudio Server (RRID:SCR_000432) hosted by the IFB-core cluster. Data and Code Availability Raw spatial transcriptomics data has been deposited at the European Genome-Phenome Archive under Study ID EGAD50000002091. Images can be downloaded from https://doi.org/10.5281/zenodo.17711406 and the complete code used to generate the main figures is available on GitHub at https://github.com/crcordeliers/CCVisiumAtlas. Bulk transcriptomic datasets (PETACC8, IDEA-France) are available as described in 33 . Bulk transcriptomic data for PRODIGE13 cohort are available within the framework of FFCD data-sharing policies upon request to the corresponding authors. Declarations Competing interests The authors declare no conflict of interest in the context of the present study Acknowledgments : This work was supported by Institut National du Cancer (INCa, IMPROCOCA project, 2023-119), the SIRIC CARPEM (INCa-DGOS-Inserm-ITMO Cancer_18006), the Labex Onco-Immunology (investissement d’avenir) and the InidEx Immuno-Onco (Initiatives d’excellence, Université Paris Cité). The group is supported by the Ligue Nationale Contre le Cancer (Equipe Labellisée). A. Cazelles was supported by a fellowship from Association pour la Recherche Contre le Cancer and M. Sroussi was funded by Fondation pour la Recherche Médicale (grant FDM202006011237). I. Hernández-Verdin was founded by BETPSY project, overseen by the French National Research Agency, as part of the second “Investissements d’Avenir” program (grant number ANR-18-RHUS-0012). We thank the CAIBI platform at the Centre de Recherche des Cordeliers for help in data deposition and Pauline Hamon for fruitful discussion. Author Contributions AC : Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Writing – original draft CP : Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Writing – original draft, Writing – review & editing DLC : Investigation, Data curation GC : Formal analysis, Visualization, Data curation MS : Formal analysis CM : Investigation, Resources CG : Resources, Clinical data curation GM : Resources, Clinical data curation MK: Resources, Clinical data curation IH : Data curation AB : Investigation WHF : Supervision LP: Data curation AdR : Methodology, Supervision CL : Resources, Clinical data curation TA : Resources, Clinical data curation JFE : Validation, Data curation TZH : Validation, Writing – review & editing JT : Resources, Clinical data curation, Writing – review & editing PLP : Conceptualization, Methodology, Formal analysis, Data curation, Supervision, Writing – review & editing, Funding acquisition SMR : Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Supervision, Writing – original draft, Writing – review & editing, Project administration, Funding acquisition References Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA. Cancer J. Clin. 71, 209–249 (2021). Eng, C. et al. Colorectal cancer. Lancet 404, 294–310 (2024). Guinney, J. et al. The consensus molecular subtypes of colorectal cancer. Nat. Med. 21, 1350–1356 (2015). Becht, E. et al. Immune and Stromal Classification of Colorectal Cancer Is Associated with Molecular Subtypes and Relevant for Precision Immunotherapy. Clin. Cancer Res. 22, 4057–66 (2016). Marisa, L. et al. Intratumor CMS Heterogeneity Impacts Patient Prognosis in Localized Colon Cancer. Clin. Cancer Res. 27, 4768–4780 (2021). Langerud, J. et al. Multiregional transcriptomics identifies congruent consensus subtypes with prognostic value beyond tumor heterogeneity of colorectal cancer. Nat. Commun. 15, 4342 (2024). Elhanani, O., Ben-Uri, R. & Keren, L. Spatial profiling technologies illuminate the tumor microenvironment. Cancer Cell 41, 404–420 (2023). Chen, J., Larsson, L., Swarbrick, A. & Lundeberg, J. Spatial landscapes of cancers: insights and opportunities. Nat. Rev. Clin. Oncol. 21, 660–674 (2024). Gracia Villacampa, E. et al. Genome-wide spatial expression profiling in formalin-fixed tissues. Cell Genomics 1, 100065 (2021). Valdeolivas, A. et al. Profiling the heterogeneity of colorectal cancer consensus molecular subtypes using spatial transcriptomics. NPJ Precis. Oncol. 8, 10 (2024). Wang, F. et al. Single-cell and spatial transcriptome analysis reveals the cellular heterogeneity of liver metastatic colorectal cancer. Sci. Adv. 9, eadf5464 (2023). Wood, C. S. et al. Spatially Resolved Transcriptomics Deconvolutes Prognostic Histological Subgroups in Patients with Colorectal Cancer and Synchronous Liver Metastases. Cancer Res. 83, 1329–1344 (2023). Wu, Y. et al. Spatiotemporal Immune Landscape of Colorectal Cancer Liver Metastasis at Single-Cell Level. Cancer Discov. 12, 134–153 (2022). Qi, J. et al. Single-cell and spatial analysis reveal interaction of FAP+ fibroblasts and SPP1+ macrophages in colorectal cancer. Nat. Commun. 13, 1742 (2022). Gulati, G. S., D’Silva, J. P., Liu, Y., Wang, L. & Newman, A. M. Profiling cell identity and tissue architecture with single-cell and spatial transcriptomics. Nat. Rev. Mol. Cell Biol. 26, 11–31 (2025). Taieb, J. et al. Oxaliplatin, fluorouracil, and leucovorin with or without cetuximab in patients with resected stage III colon cancer (PETACC-8): an open-label, randomised phase 3 trial. Lancet Oncol. 15, 862–873 (2014). Pelka, K. et al. Spatially organized multicellular immune hubs in human colorectal cancer. Cell 184, 4734-4752.e20 (2021). Corry, S. M. et al. Activation of innate-adaptive immune machinery by poly(I:C) exposes a therapeutic vulnerability to prevent relapse in stroma-rich colon cancer. Gut 71, 2502–2517 (2022). Mouillet-Richard, S. et al. Clinical Challenges of Consensus Molecular Subtype CMS4 Colon Cancer in the Era of Precision Medicine. Clin. Cancer Res. 30, 2351–2358 (2024). Ogden, S. et al. Phenotypic heterogeneity and plasticity in colorectal cancer metastasis. Cell Genomics 5, 100881 (2025). Gregorieff, A., Liu, Y., Inanlou, M. R., Khomchuk, Y. & Wrana, J. L. Yap-dependent reprogramming of Lgr5(+) stem cells drives intestinal regeneration and cancer. Nature 526, 715–718 (2015). Serra, D. et al. Self-organization and symmetry breaking in intestinal organoid development. Nature 569, 66–72 (2019). Fey, S. K., Vaquero-Siguero, N. & Jackstadt, R. Dark force rising: Reawakening and targeting of fetal-like stem cells in colorectal cancer. Cell Rep. 43, 114270 (2024). Ayyaz, A. et al. Single-cell transcriptomes of the regenerating intestine reveal a revival stem cell. Nature 569, 121–125 (2019). van der Net, M. C. et al. Mechanosensitive calcium channels and integrins coordinate the reprogramming of colorectal cancer cells into a fetal-like state. Cell Rep. 44, 116308 (2025). Tape, C. J. Plastic persisters: revival stem cells in colorectal cancer. Trends Cancer 10, 185–195 (2024). Chen, B. et al. The molecular classification of cancer-associated fibroblasts on a pan-cancer single-cell transcriptional atlas. Clin. Transl. Med. 13, e1516 (2023). Krishnamurty, A. T. et al. LRRC15+ myofibroblasts dictate the stromal setpoint to suppress tumour immunity. Nature 611, 148–154 (2022). Ma, H. et al. Periostin Promotes Colorectal Tumorigenesis through Integrin-FAK-Src Pathway-Mediated YAP/TAZ Activation. Cell Rep. 30, 793-806.e6 (2020). Giguelay, A. et al. The landscape of cancer-associated fibroblasts in colorectal cancer liver metastases. Theranostics 12, 7624–7639 (2022). Oliveira, M. F. de et al. High-definition spatial transcriptomic profiling of immune cell populations in colorectal cancer. Nat. Genet. 57, 1512–1523 (2025). Lázár, E. & Lundeberg, J. Spatial architecture of development and disease. Nat. Rev. Genet. https://doi.org/10.1038/s41576-025-00892-5 (2025) doi:10.1038/s41576-025-00892-5. Gallois, C. et al. Prognostic Models From Transcriptomic Signatures of the Tumor Microenvironment and Cell Cycle in Stage III Colon Cancer From PETACC-8 and IDEA-France Trials. J. Clin. Oncol. JCO2302262 (2025) doi:10.1200/JCO.23.02262. Taieb, J., Karoui, M. & Basile, D. How I treat stage II colon cancer patients. ESMO Open 6, 100184 (2021). Lepage, C. et al. Effect of 5 years of CT-scan and CEA follow-up on survival endpoints in patients with colorectal cancer. Ann. Oncol. S0923-7534(25)04701–5 (2025) doi:10.1016/j.annonc.2025.09.004. De Smedt, L. et al. Expression profiling of budding cells in colorectal cancer reveals an EMT-like phenotype and molecular subtype switching. Br. J. Cancer 116, 58–65 (2017). England, F. J., Lin, M., Sigal, M. & Leedham, S. J. Defining the mucosal ecosystem: epithelial-mesenchymal interdependence in gastrointestinal health and disease. Nat. Rev. Gastroenterol. Hepatol. 22, 741–754 (2025). Ramos Zapatero, M. et al. Trellis tree-based analysis reveals stromal regulation of patient-derived organoid drug responses. Cell 186, 5606-5619.e24 (2023). Cañellas-Socias, A. et al. Metastatic recurrence in colorectal cancer arises from residual EMP1+ cells. Nature 611, 603–613 (2022). Qin, X. et al. An oncogenic phenoscape of colonic stem cell polarization. Cell 186, 5554-5568.e18 (2023). Moorman, A. et al. Progressive plasticity during colorectal cancer metastasis. Nature 637, 947–954 (2025). Cammareri, P. et al. Loss of colonic fidelity enables multilineage plasticity and metastasis. Nature 644, 547–556 (2025). IARC Working Group on the Evaluation of Cancer-Preventive Interventions. Colorectal Cancer Screening . (International Agency for Research on Cancer, Lyon (FR), 2019). Timperi, E. et al. At the Interface of Tumor-Associated Macrophages and Fibroblasts: Immune-Suppressive Networks and Emerging Exploitable Targets. Clin. Cancer Res. 30, 5242–5251 (2024). Buechler, M. B. et al. Cross-tissue organization of the fibroblast lineage. Nature 593, 575–579 (2021). Smillie, C. S. et al. Intra- and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis. Cell 178, 714-730.e22 (2019). Additional Declarations There is NO Competing Interest. Supplementary Files TableS2.xlsx SupplFig1jan2026.pdf SupplFig8jan2026.pdf SupplFig2jan2026.pdf SupplFig1jan2026.pdf SupplFig7jan2026.pdf Supplementary Figure S7 SupplFig3jan2026.pdf SupplFig9jan2026.pdf SupplFig4jan2026.pdf SupplFig6jan2026.pdf Supplementary Figure S6 SupplFig5jan2026.pdf SupplFig12jan2026.pdf SupplFig12jan2026.pdf SupplFig13jan2026.pdf Supplementary Figure S13 SupplFig6jan2026.pdf TableS1.pdf SupplFig4jan2026.pdf Supplementary Figure S4 SupplFig7jan2026.pdf Legendstosupplementaryfiguresdec2025.docx SupplFig8jan2026.pdf SupplFig4jan2026.pdf SupplFig9jan2026.pdf SupplFig5jan2026.pdf SupplFig5jan2026.pdf SupplFig10jan2026.pdf SupplFig6jan2026.pdf SupplFig11jan2026.pdf SupplFig7jan2026.pdf SupplFig12jan2026.pdf SupplFig13jan2026.pdf SupplFig12jan2026.pdf Supplementary Figure S12 SupplFig13jan2026.pdf SupplFig9jan2026.pdf Supplementary Figure S9 TableS2.xlsx Figure5Visiumjan2026.png SupplFig1jan2026.pdf Supplementary Figure S1 TableS1.pdf TableS3.pdf Figure2Visiumjan2026.png TableS3.pdf Figure1Visiumjan2026.png Figure1Visiumjan2026.png Legendstosupplementaryfiguresdec20251.docx Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8583137","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":597666337,"identity":"2b851524-9fba-4d8e-aee2-9425d1d65f96","order_by":0,"name":"Sophie Mouillet-Richard","email":"data:image/png;base64,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","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":true,"prefix":"","firstName":"Sophie","middleName":"","lastName":"Mouillet-Richard","suffix":""},{"id":597666338,"identity":"c12f3a32-167a-4cfa-bb1a-e52ca8ff7b84","order_by":1,"name":"Antoine Cazelles","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Antoine","middleName":"","lastName":"Cazelles","suffix":""},{"id":597666339,"identity":"eb549f5a-381d-4651-9efe-dc498cdd24f3","order_by":2,"name":"Camilla Pilati","email":"","orcid":"https://orcid.org/0000-0002-1781-5180","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Camilla","middleName":"","lastName":"Pilati","suffix":""},{"id":597666340,"identity":"e697f7b8-ca33-4c1a-804b-87ea17880c99","order_by":3,"name":"Delphine Le Corre","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Delphine","middleName":"Le","lastName":"Corre","suffix":""},{"id":597666341,"identity":"3d3c7b39-4bba-43a8-842f-36692f83780f","order_by":4,"name":"Gadea Cabrejas","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Gadea","middleName":"","lastName":"Cabrejas","suffix":""},{"id":597666342,"identity":"1420aa08-d962-4f8b-9805-2f3ec0c6c3e7","order_by":5,"name":"Marine Sroussi","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Marine","middleName":"","lastName":"Sroussi","suffix":""},{"id":597666343,"identity":"97a93b79-4435-455d-80a4-b7f7cd8696cf","order_by":6,"name":"Claire Mulot","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Claire","middleName":"","lastName":"Mulot","suffix":""},{"id":597666344,"identity":"95d732a8-24e8-42c0-8b68-af44f9c19907","order_by":7,"name":"Claire Gallois","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Claire","middleName":"","lastName":"Gallois","suffix":""},{"id":597666345,"identity":"0e957dc6-5775-4073-8741-11086ee446d7","order_by":8,"name":"Gilles Manceau","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Gilles","middleName":"","lastName":"Manceau","suffix":""},{"id":597666346,"identity":"d3d98d68-2f09-43a3-a8f2-39d94073e4e5","order_by":9,"name":"Mehdi Karoui","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Mehdi","middleName":"","lastName":"Karoui","suffix":""},{"id":597666347,"identity":"6e77db65-ba9f-45e7-b52d-301114e0db13","order_by":10,"name":"Isaias Hernandez","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Isaias","middleName":"","lastName":"Hernandez","suffix":""},{"id":597666348,"identity":"ccd1257f-42aa-46c6-a2dc-3d989c8b1582","order_by":11,"name":"Antoine Bougouin","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Antoine","middleName":"","lastName":"Bougouin","suffix":""},{"id":597666349,"identity":"f08ad483-abdc-4e47-87a8-29abdda01e6b","order_by":12,"name":"Wolf-Hervé Fridman","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Wolf-Hervé","middleName":"","lastName":"Fridman","suffix":""},{"id":597666350,"identity":"f9c466fc-c661-46fc-bb6a-22933de9f479","order_by":13,"name":"Lucie Poupel","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Lucie","middleName":"","lastName":"Poupel","suffix":""},{"id":597666351,"identity":"1bdcd5d1-2324-4a6c-a2e7-60b495406815","order_by":14,"name":"Aurelien de Reynies","email":"","orcid":"","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Aurelien","middleName":"","lastName":"de Reynies","suffix":""},{"id":597666352,"identity":"81e0001b-c8b5-4f72-b654-c3a9564a7b7e","order_by":15,"name":"Come Lepage","email":"","orcid":"","institution":"Universite de Bourgogne","correspondingAuthor":false,"prefix":"","firstName":"Come","middleName":"","lastName":"Lepage","suffix":""},{"id":597666353,"identity":"8ce3e309-d31d-4626-8ab5-91792111f5fd","order_by":16,"name":"Thierry André","email":"","orcid":"https://orcid.org/0000-0002-5103-7095","institution":"Sorbonne university and hôpital Saint Antoine APHP","correspondingAuthor":false,"prefix":"","firstName":"Thierry","middleName":"","lastName":"André","suffix":""},{"id":597666354,"identity":"aad2c497-137e-4765-94e9-eee27c4a5e6a","order_by":17,"name":"Jean-François Emile","email":"","orcid":"https://orcid.org/0000-0002-6073-4466","institution":"Versailles SQY university \u0026 AP-HP","correspondingAuthor":false,"prefix":"","firstName":"Jean-François","middleName":"","lastName":"Emile","suffix":""},{"id":597666355,"identity":"914d588d-7b26-457f-a8e2-7cc4ed07cf24","order_by":18,"name":"Théo Hirsch","email":"","orcid":"https://orcid.org/0000-0003-4428-2997","institution":"Centre de Recherche des Cordeliers, Inserm U1138, Université Paris Cité, Sorbonne Université","correspondingAuthor":false,"prefix":"","firstName":"Théo","middleName":"","lastName":"Hirsch","suffix":""},{"id":597666356,"identity":"067cec2a-2095-4a42-bf61-d00240a4b597","order_by":19,"name":"Julien Taieb","email":"","orcid":"","institution":"Centre de recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Julien","middleName":"","lastName":"Taieb","suffix":""},{"id":597666357,"identity":"d684fb86-5b6b-45c6-9ea4-c335e289790c","order_by":20,"name":"Pierre Laurent-Puig","email":"","orcid":"https://orcid.org/0000-0001-8475-5459","institution":"Centre de Recherche des Cordeliers","correspondingAuthor":false,"prefix":"","firstName":"Pierre","middleName":"","lastName":"Laurent-Puig","suffix":""}],"badges":[],"createdAt":"2026-01-12 15:03:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8583137/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8583137/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104399461,"identity":"08c45372-522f-4031-8cf3-6402998dda93","added_by":"auto","created_at":"2026-03-11 12:06:13","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1802764,"visible":true,"origin":"","legend":"\u003cp\u003eDesign of the spatial transcriptomics study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Simplified flow chart of experimental design. 48 samples were selected from the PETACC8 stage III colon cancer trial for spatial transcriptomic analysis using the 10X Visium technology. Out of the 48 samples, 11 have a MSI phenotype, 14 corresponded to patients who subsequently relapsed. \u003cstrong\u003eb\u003c/strong\u003e, Illustration of data integration across the 48-sample dataset. UMAP and projection of Leiden clusters across the Visium capture area for sample #74_D1. Scheme explaining the reduction of spots from Seurat clusters into meta-spots in a given sample. Scheme explaining the aggregation of 3,231 merged meta-spots across the 48 samples. UMAP of Seurat meta-clusters obtained on the aggregated dataset. UMAP showing the projection of meta-cluster identity for sample #74_D1.\u003c/p\u003e","description":"","filename":"Figure1Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/2ec226facc9fa452c460471d.png"},{"id":104027732,"identity":"2fdb1687-d560-4a8d-8e5f-679a0156b7ff","added_by":"auto","created_at":"2026-03-05 21:40:02","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1802764,"visible":true,"origin":"","legend":"\u003cp\u003eDesign of the spatial transcriptomics study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Simplified flow chart of experimental design. 48 samples were selected from the PETACC8 stage III colon cancer trial for spatial transcriptomic analysis using the 10X Visium technology. Out of the 48 samples, 11 have a MSI phenotype, 14 corresponded to patients who subsequently relapsed. \u003cstrong\u003eb\u003c/strong\u003e, Illustration of data integration across the 48-sample dataset. UMAP and projection of Leiden clusters across the Visium capture area for sample #74_D1. Scheme explaining the reduction of spots from Seurat clusters into meta-spots in a given sample. Scheme explaining the aggregation of 3,231 merged meta-spots across the 48 samples. UMAP of Seurat meta-clusters obtained on the aggregated dataset. UMAP showing the projection of meta-cluster identity for sample #74_D1.\u003c/p\u003e","description":"","filename":"Figure1Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/5d5cd88acba709eea648ac71.png"},{"id":104399549,"identity":"2215f722-1655-4613-9dd7-12ebfaa093bc","added_by":"auto","created_at":"2026-03-11 12:06:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2653833,"visible":true,"origin":"","legend":"\u003cp\u003eIdentification of 11 spatial ecotypes with specific morpho-molecular patterns\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea,\u003c/strong\u003e Heatmap showing signatures of CMS1 to CMS4, immune and stromal components, as well as CMS-associated signatures from the Guinney et al. study and immune-related signatures from the Becht et al. study for 3,231 meta-spots grouped into meta-clusters. \u003cstrong\u003eb,\u003c/strong\u003e Representative H\u0026amp;E areas of FFPE slides corresponding to the various tumor, stromal and other ecotypes.\u003cstrong\u003e c,\u003c/strong\u003e Violin plots summarizing the levels of scores corresponding to various signatures from \u003csup\u003e17\u003c/sup\u003e across the 10 ecotypes.\u003cstrong\u003e d,\u003c/strong\u003e Alluvial plot summarizing the ecotype neighborhood (right) of stroma A and stroma B meta-spots (left).\u003c/p\u003e","description":"","filename":"Figure2Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/c1212437933fce25089a00e3.png"},{"id":104027731,"identity":"95934740-7983-4052-9267-c2ea9f6662b0","added_by":"auto","created_at":"2026-03-05 21:40:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2653833,"visible":true,"origin":"","legend":"\u003cp\u003eIdentification of 11 spatial ecotypes with specific morpho-molecular patterns\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea,\u003c/strong\u003e Heatmap showing signatures of CMS1 to CMS4, immune and stromal components, as well as CMS-associated signatures from the Guinney et al. study and immune-related signatures from the Becht et al. study for 3,231 meta-spots grouped into meta-clusters. \u003cstrong\u003eb,\u003c/strong\u003e Representative H\u0026amp;E areas of FFPE slides corresponding to the various tumor, stromal and other ecotypes.\u003cstrong\u003e c,\u003c/strong\u003e Violin plots summarizing the levels of scores corresponding to various signatures from 17 across the 10 ecotypes.\u003cstrong\u003e d,\u003c/strong\u003e Alluvial plot summarizing the ecotype neighborhood (right) of stroma A and stroma B meta-spots (left).\u003c/p\u003e","description":"","filename":"Figure2Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/68ff97c9903c1c80482c8f90.png"},{"id":103607938,"identity":"3f70f13e-6f66-45e6-b0d7-081bf9e3f468","added_by":"auto","created_at":"2026-02-27 15:14:48","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1025611,"visible":true,"origin":"","legend":"\u003cp\u003eMolecular hallmarks of disease recurrence specific to stromal ecotypes\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea,\u003c/strong\u003e Volcano plot summarizing the differentially expressed genes (DEGs) in the stromal compartment of patients having experienced relapse or not. Genes with significantly (adjusted \u003cem\u003ep\u003c/em\u003e-value \u0026lt;0.01) upregulated expression in the stroma of patients without or with relapse are labelled in blue and red, respectively. \u003cstrong\u003eb,\u003c/strong\u003e Boxplots showing the scaled expression of the DEGs obtained in panel A, in all meta-spots combined (top, no significant change), stroma-A meta-spots (middle, all p\u0026lt;0.05 except \u003cem\u003eCTXN1\u003c/em\u003e not significant) or stroma-B meta-spots (bottom, all p\u0026lt;0.05), according to disease recurrence. \u003cstrong\u003ec,\u003c/strong\u003e Venn diagram illustrating the overlap between the list of genes upregulated in the stromal compartment of patients who subsequently relapsed and the genes signatures of REC and iREC \u003csup\u003e20\u003c/sup\u003e. \u003cstrong\u003ed,\u003c/strong\u003e Boxplots showing the enrichment score of REC (left panel) or iREC (right panel) signatures in all meta-spots combined or stromal (pooled stroma-A and stroma-B) meta-spots, according to disease recurrence. \u003cstrong\u003ee,\u003c/strong\u003e Table summarizing the analysis of TF ChIPseq binding on \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, \u003cem\u003eCAPN2\u003c/em\u003e and \u003cem\u003eLRP10\u003c/em\u003e with EnrichR. \u003cstrong\u003ef,\u003c/strong\u003e Model explaining the activation of \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, and \u003cem\u003eCAPN2 \u003c/em\u003etranscription downstream from YAP in cancer cells with a regenerative phenotype in a local environment enriched in CAFs (created with Biorender). \u003cstrong\u003eg-h,\u003c/strong\u003e Boxplots showing the enrichment score of YAP-associated signatures from \u003csup\u003e21\u003c/sup\u003e (\u003cstrong\u003eg\u003c/strong\u003e, left and middle) or \u003csup\u003e22\u003c/sup\u003e (\u003cstrong\u003eg\u003c/strong\u003e, right) or those of the RSC signature \u003csup\u003e24\u003c/sup\u003e (\u003cstrong\u003eh\u003c/strong\u003e) in all meta-spots combined or stromal meta-spots, according to disease recurrence.\u003c/p\u003e","description":"","filename":"Figure3Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/70eadcc60b7792e83efb8de2.png"},{"id":103607933,"identity":"80438638-acf0-47c1-bf99-65d2f7f645aa","added_by":"auto","created_at":"2026-02-27 15:14:48","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1332699,"visible":true,"origin":"","legend":"\u003cp\u003eStromal ecotypes break into five stromal sub-clusters.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea,\u003c/strong\u003e Scheme summarizing the analytical workflow to assess stromal heterogeneity and projection of stromal ecotypes and stromal sub-clusters in sample #74_B1. \u003cstrong\u003eb,\u003c/strong\u003e Bubble plots of average scores in each stromal sub-cluster for signatures derived from various studies \u003csup\u003e17,27,45,46\u003c/sup\u003e or obtained with MCPcounter. Data were re-scaled to enhance contrast between sub-clusters. \u003cstrong\u003ec,\u003c/strong\u003e Boxplots comparing the proportions of spots corresponding to the stroma-B3 subcluster in dMMR versus pMMR patients (left) and in patients with versus without relapse( right) in the whole dataset (n=48).\u003c/p\u003e","description":"","filename":"Figure4Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/b012747007f3f5b8b688963a.png"},{"id":104779162,"identity":"36e65780-6f97-4d91-8693-ca0ab819c0e7","added_by":"auto","created_at":"2026-03-17 07:36:00","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":3198505,"visible":true,"origin":"","legend":"\u003cp\u003eLocalization of \u003cem\u003eANXA1\u003c/em\u003e+ REC/RSC with Visium HD\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Simplified flow chart of experimental design and selection of 4 samples for reprofiling using the 10X Visium HD technology. \u003cstrong\u003eb\u003c/strong\u003e, Expression of the stroma-A2 signature score of the entire capture area or in magnified zone (top first and second panels). Expression of \u003cem\u003eEPCAM\u003c/em\u003e or the YAP-Rev (\u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, \u003cem\u003eCAPN2\u003c/em\u003e) signature score in magnified zone (top third and fourth panels). Histology (bottom left) and co-expression analysis of the stroma-A2 (yellow) and the YAP-Rev (blue) signatures (bottom right) in further magnified zone. Arrows indicate clusters of undifferentiated tumor cells. Scale bars: 2 mm (full capture zone), 100 µm (magnified zone), 200 µm (further magnified zone).\u003c/p\u003e","description":"","filename":"Figure5Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/061c74c6a2e67947dfaba222.png"},{"id":104027730,"identity":"b57dfe97-7a29-4426-af63-1cfb7698faf5","added_by":"auto","created_at":"2026-03-05 21:40:01","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":3198505,"visible":true,"origin":"","legend":"\u003cp\u003eLocalization of \u003cem\u003eANXA1\u003c/em\u003e+ REC/RSC with Visium HD\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e, Simplified flow chart of experimental design and selection of 4 samples for reprofiling using the 10X Visium HD technology. \u003cstrong\u003eb\u003c/strong\u003e, Expression of the stroma-A2 signature score of the entire capture area or in magnified zone (top first and second panels). Expression of \u003cem\u003eEPCAM\u003c/em\u003e or the YAP-Rev (\u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, \u003cem\u003eCAPN2\u003c/em\u003e) signature score in magnified zone (top third and fourth panels). Histology (bottom left) and co-expression analysis of the stroma-A2 (yellow) and the YAP-Rev (blue) signatures (bottom right) in further magnified zone. Arrows indicate clusters of undifferentiated tumor cells. Scale bars: 2 mm (full capture zone), 100 µm (magnified zone), 200 µm (further magnified zone).\u003c/p\u003e","description":"","filename":"Figure5Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/6ba4d5c719d5f111c9893439.png"},{"id":103607937,"identity":"49df53ce-7f41-4ad5-b984-a112ba566849","added_by":"auto","created_at":"2026-02-27 15:14:48","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":782223,"visible":true,"origin":"","legend":"\u003cp\u003ePrognostic value of spatially-derived signatures across three bulk RNA-seq colorectal cancer cohorts.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ea-c\u003c/strong\u003e, Forest plots summarizing the hazard ratio (HR) associated with a 1-SD increase in continuous ssGSEA scores for the four tumor and two stromal ecotypes, the five stromal subclusters, and the YAP-Rev program in PETACC8 (a\u003cstrong\u003eA\u003c/strong\u003e, n = 1,733, stage III), IDEA-France (\u003cstrong\u003eb\u003c/strong\u003e, n = 1,248, stage III) and PRODIGE13 (\u003cstrong\u003ec\u003c/strong\u003e, n = 529, stage II). Circles and squares indicate univariate and multivariable Cox models, respectively. Analyses were adjusted for grade, WHO performance status, baseline high-risk features (pT4 and/or pN2 when available), mismatch repair status, and bowel complications at diagnosis (occlusion or perforation). \u003cstrong\u003ed,\u003c/strong\u003e Interaction effects between stroma-A2 and YAP-Rev in PETACC8. Curves display the predicted hazard ratio associated with increasing levels of stroma-A2 (left) or YAP-Rev (right). For each curve, the interacting program was fixed at representative low or high expression levels, defined as the 25th (Q1) or 75th percentile (Q3) values of the cohort. Hazard ratios are centered on a reference patient exhibiting median expression of both programs and standard clinical covariates. Analyses were adjusted for grade, WHO performance status, baseline recurrence risk (pT4 and/or pN2 disease), mismatch repair status, and bowel complications at diagnosis when available. Shaded envelopes indicate 95% confidence intervals. \u003cstrong\u003ee–g,\u003c/strong\u003e Kaplan–Meier curves for time-to-recurrence (TTR) according to combined Stroma-A2 and YAP-Rev levels in PETACC8 (\u003cstrong\u003ee\u003c/strong\u003e), IDEA-France (\u003cstrong\u003ef\u003c/strong\u003e) and PRODIGE13 (\u003cstrong\u003eg\u003c/strong\u003e). Patients were stratified into four subgroups (Low/Low, High/Low, Low/High, High/High) using PETACC8-derived median thresholds computed from non-standardized ssGSEA scores, which were then applied unchanged to the other cohorts to ensure consistent biological cut-points. Statistical significance was assessed using two-sided log-rank tests and unadjusted Cox models.\u003c/p\u003e","description":"","filename":"Figure6Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/ac931ed14a374ba575f1f06f.png"},{"id":109203696,"identity":"408f8cef-f585-490e-923d-846a81ee9662","added_by":"auto","created_at":"2026-05-13 14:43:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":19570551,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/d167f74c-3111-4d1d-8029-f701c72b2371.pdf"},{"id":104403083,"identity":"af5b5846-486d-433a-b805-ed83e715617c","added_by":"auto","created_at":"2026-03-11 12:17:26","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":171578,"visible":true,"origin":"","legend":"","description":"","filename":"TableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/963cdb7ce6f3aee91bbe4b88.xlsx"},{"id":104081811,"identity":"b66520db-4d76-4543-87a4-c23923fdaa55","added_by":"auto","created_at":"2026-03-06 14:29:58","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1046218,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig1jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/9ac687f71860b1ddd2f5a17d.pdf"},{"id":104402706,"identity":"5687181e-8f5c-4dc9-a67f-ccc1d1f551bd","added_by":"auto","created_at":"2026-03-11 12:16:10","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":488565,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig8jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/3a55ccd96ea619a840490448.pdf"},{"id":104081936,"identity":"e67e28a9-3fb2-438e-bd7d-46855bca030c","added_by":"auto","created_at":"2026-03-06 14:34:47","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":49943733,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig2jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/32d166bdbf21a88e4e989193.pdf"},{"id":104402868,"identity":"bceec721-73ca-466b-85d0-9c4385f6f745","added_by":"auto","created_at":"2026-03-11 12:16:44","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1046218,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig1jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/da124218607a19d1d0742968.pdf"},{"id":104399340,"identity":"3fa896ea-3b9b-4e7d-93fc-f006e6ef2070","added_by":"auto","created_at":"2026-03-11 12:05:36","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":177769,"visible":true,"origin":"","legend":"Supplementary Figure S7","description":"","filename":"SupplFig7jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/68bf59368f571adeff32d6cc.pdf"},{"id":104082013,"identity":"7ddf4a95-fd0c-4e59-b65a-cad7184f16cf","added_by":"auto","created_at":"2026-03-06 14:37:55","extension":"pdf","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":9911945,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig3jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/7778cd3063ce3405b623e11c.pdf"},{"id":104779409,"identity":"fb732161-6c6f-48c2-a6ae-a132e503a1a1","added_by":"auto","created_at":"2026-03-17 07:39:56","extension":"pdf","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":165331,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig9jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/0b63da3d618a6037cf07364b.pdf"},{"id":104082662,"identity":"6e3a71d7-7cb4-4907-913d-78d48b398255","added_by":"auto","created_at":"2026-03-06 14:41:36","extension":"pdf","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":3200835,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig4jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/1f61a02773bf34756f7e5a90.pdf"},{"id":104399447,"identity":"b4d14eef-613e-4e8e-8633-d199e33fee13","added_by":"auto","created_at":"2026-03-11 12:06:10","extension":"pdf","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":1132806,"visible":true,"origin":"","legend":"Supplementary Figure S6","description":"","filename":"SupplFig6jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/0c776757b61480c05820a88c.pdf"},{"id":104082483,"identity":"7588ae7e-900f-40de-b94a-14abb595e050","added_by":"auto","created_at":"2026-03-06 14:40:50","extension":"pdf","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":796657,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig5jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/22b9127f271b51ed2f59027d.pdf"},{"id":105727671,"identity":"48c820be-3919-4cd7-a5d0-b19990f5ab9f","added_by":"auto","created_at":"2026-03-30 10:59:38","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":401597,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig12jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/444bebf6070a90f003c14b93.pdf"},{"id":104402667,"identity":"fff2d3b0-0b21-43a8-838b-3316c530b46f","added_by":"auto","created_at":"2026-03-11 12:16:03","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":401597,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig12jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/ad8e3e016b634c09b7060b63.pdf"},{"id":104399372,"identity":"01fcb375-5f3f-4a38-9b76-4d5cf9c84821","added_by":"auto","created_at":"2026-03-11 12:05:46","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":2171173,"visible":true,"origin":"","legend":"Supplementary Figure S13","description":"","filename":"SupplFig13jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/63fbfe9a0f8499596614554f.pdf"},{"id":104082804,"identity":"b8f16f9a-db2e-4e66-b583-236e263248ef","added_by":"auto","created_at":"2026-03-06 14:42:14","extension":"pdf","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":1132806,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig6jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/8f1cbf1eedf6ec1d4c461b00.pdf"},{"id":104403014,"identity":"9407c804-a7a6-45c0-8d9c-c4d1bf22f090","added_by":"auto","created_at":"2026-03-11 12:17:11","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":41544,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/632e7c8306134e652e793d58.pdf"},{"id":104398304,"identity":"43259af6-9c6e-4e8e-aa67-af2873700aaf","added_by":"auto","created_at":"2026-03-11 12:01:33","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":3200835,"visible":true,"origin":"","legend":"Supplementary Figure S4","description":"","filename":"SupplFig4jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/b13ce02aa274c68def87c805.pdf"},{"id":104082829,"identity":"50b2ebf6-442f-4cf8-8127-8d28028d9c5c","added_by":"auto","created_at":"2026-03-06 14:43:10","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":177769,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig7jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/d3ec12d84fbed13e98f148bf.pdf"},{"id":104402165,"identity":"7cb0f5e7-94e0-4207-b509-8e230f26c9d3","added_by":"auto","created_at":"2026-03-11 12:14:32","extension":"docx","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":28675,"visible":true,"origin":"","legend":"","description":"","filename":"Legendstosupplementaryfiguresdec2025.docx","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/7986c583e5b17d7cd84cf429.docx"},{"id":104085117,"identity":"aabf2b49-2691-40a8-af50-3d67be498945","added_by":"auto","created_at":"2026-03-06 15:11:47","extension":"pdf","order_by":8,"title":"","display":"","copyAsset":false,"role":"supplement","size":488565,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig8jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/669d3b13468e4f4af2b5d625.pdf"},{"id":104402092,"identity":"91f512c3-7485-4f20-87d4-cda39f93707e","added_by":"auto","created_at":"2026-03-11 12:14:16","extension":"pdf","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":3200835,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig4jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/4d3adcfe07db50c98f273e94.pdf"},{"id":104085192,"identity":"610f754b-c0bb-4173-918a-fd3b7bda3b00","added_by":"auto","created_at":"2026-03-06 15:12:11","extension":"pdf","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":165331,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig9jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/67ea71bd6c63c1cbaa64735a.pdf"},{"id":104779362,"identity":"906b22af-6767-49c2-917b-66151178301a","added_by":"auto","created_at":"2026-03-17 07:39:15","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":796657,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig5jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/4641fdba12e7981d9326656f.pdf"},{"id":104402883,"identity":"d4c673e9-ddf1-4efa-baca-556eb7b718de","added_by":"auto","created_at":"2026-03-11 12:16:47","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":796657,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig5jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/228b9f63439028efcf02e606.pdf"},{"id":104085288,"identity":"334b0159-ab2f-41e9-96bb-ce25c5f4baa9","added_by":"auto","created_at":"2026-03-06 15:13:43","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":49674543,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig10jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/6456eba902addd9dd2cafcf8.pdf"},{"id":104402213,"identity":"f641bef7-f50d-45e9-b7ec-57562d76a71b","added_by":"auto","created_at":"2026-03-11 12:14:40","extension":"pdf","order_by":11,"title":"","display":"","copyAsset":false,"role":"supplement","size":1132806,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig6jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/bf4e483c90f3608d17d0de9d.pdf"},{"id":104086644,"identity":"114aa3a2-c59c-4a3d-85c0-aacdff23c0fc","added_by":"auto","created_at":"2026-03-06 15:30:33","extension":"pdf","order_by":11,"title":"","display":"","copyAsset":false,"role":"supplement","size":113602,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig11jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/c81aaddb74585e25a4948638.pdf"},{"id":104402767,"identity":"01fe7af4-a82f-4a53-9603-8f1e2c38166e","added_by":"auto","created_at":"2026-03-11 12:16:21","extension":"pdf","order_by":12,"title":"","display":"","copyAsset":false,"role":"supplement","size":177769,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig7jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/1064b3bcc96b656df3349f29.pdf"},{"id":104088261,"identity":"8f94d6dc-c78b-46dd-90f5-92b86e730252","added_by":"auto","created_at":"2026-03-06 15:47:39","extension":"pdf","order_by":12,"title":"","display":"","copyAsset":false,"role":"supplement","size":401597,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig12jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/409ac28fb7f3ed85d0cc3213.pdf"},{"id":104403350,"identity":"2137b4f0-0733-40fd-bc11-1322b405dc89","added_by":"auto","created_at":"2026-03-11 12:18:07","extension":"pdf","order_by":13,"title":"","display":"","copyAsset":false,"role":"supplement","size":2171173,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig13jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/0afffde4f279ccf947bf2d6d.pdf"},{"id":104399344,"identity":"32ebb332-836b-4ddd-ba04-3c6033dbd608","added_by":"auto","created_at":"2026-03-11 12:05:37","extension":"pdf","order_by":13,"title":"","display":"","copyAsset":false,"role":"supplement","size":401597,"visible":true,"origin":"","legend":"Supplementary Figure S12","description":"","filename":"SupplFig12jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/912e6580303a57073689c928.pdf"},{"id":104088371,"identity":"fd0ae50c-bf3c-4437-b708-90d9162f5c98","added_by":"auto","created_at":"2026-03-06 15:48:34","extension":"pdf","order_by":13,"title":"","display":"","copyAsset":false,"role":"supplement","size":2171173,"visible":true,"origin":"","legend":"","description":"","filename":"SupplFig13jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/d488a4db31b9b2b88e829610.pdf"},{"id":104399194,"identity":"0957a45e-bd5f-4745-991c-854e0d92c851","added_by":"auto","created_at":"2026-03-11 12:05:03","extension":"pdf","order_by":14,"title":"","display":"","copyAsset":false,"role":"supplement","size":165331,"visible":true,"origin":"","legend":"Supplementary Figure S9","description":"","filename":"SupplFig9jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/a3b3549dd410fa2feafada94.pdf"},{"id":104089007,"identity":"eb8c83ae-aa8b-4714-b7bc-402561450560","added_by":"auto","created_at":"2026-03-06 15:55:21","extension":"xlsx","order_by":14,"title":"","display":"","copyAsset":false,"role":"supplement","size":171578,"visible":true,"origin":"","legend":"","description":"","filename":"TableS2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/76c4971f82ee9baba43bb9d1.xlsx"},{"id":104402456,"identity":"9d7e2485-49c7-456a-8367-a1eb4f47b1bd","added_by":"auto","created_at":"2026-03-11 12:15:25","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"supplement","size":3198505,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/af1d3222551a5364b0db3f47.png"},{"id":104399511,"identity":"b42d1816-1dfb-46cc-abb1-f01295eafcf2","added_by":"auto","created_at":"2026-03-11 12:06:26","extension":"pdf","order_by":15,"title":"","display":"","copyAsset":false,"role":"supplement","size":1046218,"visible":true,"origin":"","legend":"Supplementary Figure S1","description":"","filename":"SupplFig1jan2026.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/fe70440f97e9496267069ddb.pdf"},{"id":104089119,"identity":"86d27589-de85-4cd7-8c2a-14e54ab46334","added_by":"auto","created_at":"2026-03-06 15:55:50","extension":"pdf","order_by":15,"title":"","display":"","copyAsset":false,"role":"supplement","size":41544,"visible":true,"origin":"","legend":"","description":"","filename":"TableS1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/5da2de5983bdf1376f8b3d21.pdf"},{"id":104403150,"identity":"edab9039-b833-48dc-a146-544293b7be0e","added_by":"auto","created_at":"2026-03-11 12:17:37","extension":"pdf","order_by":16,"title":"","display":"","copyAsset":false,"role":"supplement","size":42701,"visible":true,"origin":"","legend":"","description":"","filename":"TableS3.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/396ac313f51523e8ceae6042.pdf"},{"id":104403060,"identity":"18677e0a-a0a8-4fa0-9af5-73db4a566ca2","added_by":"auto","created_at":"2026-03-11 12:17:20","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"supplement","size":2653833,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/f40ab857fb43c81686e0c024.png"},{"id":104089513,"identity":"186d4f02-65c0-4b14-bcc3-bfd347c73aaa","added_by":"auto","created_at":"2026-03-06 16:02:16","extension":"pdf","order_by":16,"title":"","display":"","copyAsset":false,"role":"supplement","size":42701,"visible":true,"origin":"","legend":"","description":"","filename":"TableS3.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/6f94716abd12a67f74221ee2.pdf"},{"id":104403300,"identity":"ec96cf5f-adda-45db-8771-27073cfdfe5c","added_by":"auto","created_at":"2026-03-11 12:17:58","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"supplement","size":1802764,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/ed658eedf34be5583da154b1.png"},{"id":104402862,"identity":"89b953c2-a227-4594-964a-f34a4a814b62","added_by":"auto","created_at":"2026-03-11 12:16:43","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"supplement","size":1802764,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1Visiumjan2026.png","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/8b628acb3fbff6747122dfca.png"},{"id":104090215,"identity":"f390d793-97fd-4ace-93f1-685b9ad6a697","added_by":"auto","created_at":"2026-03-06 16:10:26","extension":"docx","order_by":17,"title":"","display":"","copyAsset":false,"role":"supplement","size":28675,"visible":true,"origin":"","legend":"","description":"","filename":"Legendstosupplementaryfiguresdec20251.docx","url":"https://assets-eu.researchsquare.com/files/rs-8583137/v1/b28f3e147da3f6c848eaad49.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Profiling colon cancer architecture with spatial transcriptomics identifies clinically relevant stromal ecotypes","fulltext":[{"header":"Introduction","content":"\u003cp\u003eColon cancer (CC) remains the fourth diagnosed cancer worldwide, with over 1.1 million new cases in 2020\u0026nbsp;\u003csup\u003e1\u003c/sup\u003e. By 2040, colorectal cancer is expected to affect over 3 million individuals, leading to over 1.5 million deaths due to increasing incidence in emerging economies\u0026nbsp;\u003csup\u003e2\u003c/sup\u003e. As other cancer types, CC is a heterogenous disease, which may notably vary according to the primary tumor location (left or right-sided), disease stage at diagnosis, carcinogenic pathway (serrated or tubular), as well as molecular features. Beyond the distinction between microsatellite stable (MSS) or unstable (MSI) tumors and the specificities related to several key driver mutations (e.g. \u003cem\u003eKRAS\u003c/em\u003e or \u003cem\u003eBRAF\u003c/em\u003e), the molecular taxonomy of CC is usually depicted through the lens of the consensus molecular classification into 4 consensus molecular subtypes CMS1 to CMS4, following the landmark study by Guinney et al.\u0026nbsp;\u003csup\u003e3\u003c/sup\u003e. This molecular classification, based on bulk gene expression, has prognostic implications and was shown to be accompanied by specific immune and stromal contextures\u0026nbsp;\u003csup\u003e4\u003c/sup\u003e. Adding some complexity to the picture, we demonstrated through deconvolution approaches that a large proportion of CC tumors actually belong to multiple CMS, entailing a high degree of intra-tumor heterogeneity\u0026nbsp;\u003csup\u003e5\u003c/sup\u003e. Along the same line, Langerud et al. recently documented a high level of CMS heterogeneity based on multiregional sampling\u0026nbsp;\u003csup\u003e6\u003c/sup\u003e. Despite these overall advances, our understanding of the spatial organization of tumor cells and their surrounding tumor microenvironment (TME) in colon cancer is still far from complete.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn recent years, the advent of spatial transcriptomics technologies has revolutionized studies focusing on tissue heterogeneity, notably in the field of cancer\u0026nbsp;\u003csup\u003e7,8\u003c/sup\u003e. Until recently, these were restricted to fresh frozen tissue, and the extension of existing tools to formalin-fixed paraffin embedded (FFPE) samples in 2021\u0026nbsp;\u003csup\u003e9\u003c/sup\u003e has substantially expanded the possibility to investigate retrospective collections, and hence to evaluate tumor architecture in relation with patient outcome\u0026nbsp;\u003csup\u003e8\u003c/sup\u003e. In CC, spatial transcriptome analyses have allowed to infer cell communication events in the tumor-stroma interface of CMS2 carcinomas\u0026nbsp;\u003csup\u003e10\u003c/sup\u003e, shed light on the heterogeneity of liver metastases\u0026nbsp;\u003csup\u003e11\u0026ndash;13\u003c/sup\u003e, or reveal interactions between cancer-associated fibroblasts (CAFs) and macrophages subtypes\u0026nbsp;\u003csup\u003e14\u003c/sup\u003e. Nevertheless, in most instances, the number of samples studied was limited, possibly due to the cost of these technologies, thus precluding the statistical power to bring to light some particular traits related to individual patient features. Another point that is worth considering when attempting to discover spatial molecular features with broad clinical implications is that samples should belong to a homogeneous category of disease (e.g. in term of disease stage) while at the same time representing the inter-patient tumor heterogeneity (e.g. in terms of age, sex, clinical and molecular profiles). Overall, achieving this goal necessitates to analyze a significant number of samples within a homogeneous collection. On another hand, while single-cell and spatial transcriptomics data are powerful tools to investigate tumor heterogeneity, they are not compatible with routine clinical practice for individual patient care. For this reason, it remains often difficult to translate spatial transcriptomics-related discoveries into molecular signatures applicable in daily care\u0026nbsp;\u003csup\u003e15\u003c/sup\u003e. With these considerations in mind, our goal was to generate and analyze a spatial transcriptomics dataset with the following specifications: (1) for sample selection, combine homogeneity of disease stage and clinical diversity in terms of disease outcome and (2) for data analysis, mine for spatially-relevant features associated with disease outcome to be further validated in bulk transcriptomic datasets.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eHere we present the largest spatial transcriptomic dataset to date in non-metastatic colon cancer, generated from 48 stage III tumors collected in the randomized PETACC8 clinical trial\u0026nbsp;\u003csup\u003e16\u003c/sup\u003e.\u0026nbsp;Using FFPE-compatible Visium technology (10x Genomics), we chart the spatial architecture of tumor and tumor-associated stromal compartments and identify recurrent spatial ecosystems, including tumor and stromal, with specific molecular hallmarks and signatures. Comparative analysis based on recurrence status uncovered a stromal-specific increase in \u003cem\u003eANXA1\u003c/em\u003e expression together with an associated regenerative cells (REC) / revival stem cell (RSC) program in tumors that later relapse. Unsupervised dissection of the stromal compartment unveiled additional layers of heterogeneity. Re-profiling a subset of 4 samples with Visium HD allowed to spot \u003cem\u003eANXA1\u003c/em\u003e-expressing tumor cells embedded in stroma. Translating these findings to bulk transcriptomic data, we show that spatially-derived signatures carry prognostic values in two large independent cohorts of stage III colon cancer (PETACC8, IDEA-France) and a newly generated cohort including 529 stage II patients (PRODIGE13), altogether comprising over 3,500 patients. Finally, multivariate analyses demonstrated a synergistic interaction between matCAF-enriched stroma derived-signature and \u003cem\u003eANXA1\u003c/em\u003e-REC/RSC score, and identified their combined high expression as a biologically grounded, clinically meaningful marker of tumor aggressiveness.\u003c/p\u003e\n"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eGeneration of a spatial transcriptomic dataset of stage III colon cancer\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWith the aim of understanding the architecture of localized colon cancer, we generated spatial transcriptomic profiles of a set of 48 primary resected tumors at baseline (Fig. 1a). Samples originated from the PETACC8 study, an open-label, randomized phase 3 trial of oxaliplatin, fluorouracil, and leucovorin with or without cetuximab in patients with resected stage III colon cancer \u003csup\u003e16\u003c/sup\u003e. Among those, 14 patients subsequently experienced disease recurrence (Relapse) while the remaining 34 patients did not (No relapse) (see Supplementary Table S1 for patient characteristics). 10X Genomics Visium spatial transcriptomics was applied to 5 \u0026micro;m-thick FFPE sections enriched in tumor tissue (see Methods). In order to identify recurrent histo-molecular spatial patterns, or spatial archetypes, across our 48 colon cancer samples, we designed a four-step strategy (Fig. 1b; see also Materials and Methods). First, we applied high-resolution Leiden clustering independently to each Visium sample, aiming to delineate transcriptionally coherent spot clusters (Fig. 1b). In the second step, we summarized each cluster into a \u0026ldquo;meta-spot\u0026rdquo; by averaging gene expression values across all constituent spots, yielding a median of 61 meta-spots per sample. Importantly, this approach ensured that each meta-spot represented a single cluster regardless of its size, allowing small but biologically meaningful regions to be preserved and preventing them from being overshadowed by larger clusters. The median spot number per meta-spot across the 48 samples was 57 (range 46-133) (Supplementary Fig. S1a and 1b). In the third step, we aggregated all meta-spots from the 48 samples into a single dataset, resulting in 3,231 meta-spots overall (Fig. 1b). To integrate this multi-sample meta-spot dataset, we used the Harmony algorithm and built a unified Seurat object. In the final step, we performed unsupervised clustering on this object and sought to select an appropriate clustering resolution to balance cluster granularity and stability (Supplementary Fig. S1c and 1d). This led to the identification of 11 meta-clusters. These meta-clusters grouped meta-spots from different patients (Supplementary Fig. S1e), highlighting successful integration. Furthermore, when meta-cluster assignments were projected back onto each sample, we observed spatially organized patterns composed of multiple meta-clusters (Fig. 1b and Supplementary Fig. S2a), analytically confirmed by determining Moran\u0026rsquo;s I spatial autocorrelation indexes (Supplementary Fig. S2b). \u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eMeta-clustering identifies spatial ecotypes with specific CMS and TME combinations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe above results support the existence of spatial ecotypes: shared, spatially structured transcriptional units recurrently found across tumors, both in tumor cores and in adjacent non-malignant tissue \u003csup\u003e15\u003c/sup\u003e. We then sought to characterize these meta-clusters by assessing molecular (Fig 2a and Supplementary Fig. S4) and morpho-histological features (Fig 2b and Supplementary Fig. S3 for representative sample #74_D1). Molecular analyses were performed by computing signatures for CMS, CMS-related features as well as immune-stromal hallmarks. Projections of ecotypes on Visium capture areas were reviewed by a pathologist (JFE). Histological examination revealed the presence of glandular tumor cells in four clusters (clusters 0, 1, 7 and 8) (Fig 2b). The heatmap built on the 3,231 meta-spots pre-classified through unsupervised clustering showed molecular similarity between 3 of those meta-clusters, with the highest CMS2 and CMS3 scores and epithelial signatures, including that derived from the single cell atlas of colon cancer \u003csup\u003e17\u003c/sup\u003e (Fig 2a and c and Supplementary Fig. S4). One of them (meta-cluster 0, red) had the highest mean Copy Number Variation (CNV) fraction as determined with our recently developed FastCNV algorithm (https://github.com/must-bioinfo/fastCNV) (Supplementary Fig. S4). It also displayed the highest score for Wnt and cell cycle signature and was thus annotated as \u0026ldquo;tumor-cycling\u0026rdquo; (Fig 2b). Meta-cluster 1 (blue) had the second highest scores (Supplementary Fig. S4) and was termed \u0026ldquo;tumor-intermediate\u0026rdquo;. Meta-cluster 7 (pink) was transcriptionally close to the \u0026ldquo;tumor-cycling\u0026rdquo; and \u0026ldquo;tumor-intermediate\u0026rdquo; meta-cluster according to unsupervised clustering (Fig 2a), albeit with a lower cell cycle score and Wnt signature, suggesting a lower content of tumor cells and a higher abundance of non-tumor cells (Supplementary Fig. S4), and was thus denoted \u0026ldquo;tumor-infiltrated\u0026rdquo;. The last one (meta-cluster 8, grey) had meta-cluster 3 (purple) as its closest meta-cluster according to Fig 2a, which corresponded to the muscularis propria. Meta-spots from meta-cluster 8 overlapped with myofibroblasts and it was therefore termed \u0026ldquo;tumor-myofibroblasts\u0026rdquo; (Fig 2c). Two other meta-clusters were enriched in CMS1/CMS4 as well as various mesenchymal and stromal (in particular fibroblast) signatures (Fig 2a and c and Supplementary Fig. S4). Histological examination confirmed their stromal identity (Fig 2b). Analysis of meta-spots neighborhood indicated that one of them, designated \u0026ldquo;stroma-A\u0026rdquo; (meta-cluster 2, green), was closer to tumor ecotypes, in line with a higher epithelial score (Fig 2c and Supplementary Fig. S4), while the other, annotated as \u0026ldquo;stroma-B\u0026rdquo; (meta-cluster 4, orange), was further away from tumor ecotypes (Fig 2d). Stroma-A accounted for 77.7% of all stromal contacts in the immediate tumor neighborhood, compared with only 22.3% for stroma-B (3.48-fold difference, p \u0026lt; 0.0001). Finally, we identified meta-clusters corresponding to necrotic regions (characterized by the highest neutrophil infiltration score, meta-cluster 9, black), lymphoid aggregates enriched in B and T cells (meta-cluster 10, bright blue), connective tissue including nerves or arteries (meta-cluster 5, yellow), and normal colon epithelium (meta-cluster 6, brown) (Fig 2b and Supplementary Fig. S4). While tumor and stromal ecotypes were consistently present across most samples, their relative abundances varied markedly between individuals (Supplementary Fig. S5). Together, these findings indicate that we have constructed a spatial atlas of stage III colon cancer from a clinically homogeneous cohort, yet displaying substantial heterogeneity in spatial organization and microenvironmental composition.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStromal ecotypes from tumors of patients with relapse display an RSC-like program enriched in \u003cem\u003eANXA1\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe next leveraged the diversity of our cohort to explore specific spatial features associated with disease recurrence. For each sample, we calculated the proportion of spots assigned to each meta-cluster and examined the distribution of ecotypes relative to tumor or stromal content (Supplementary Fig. S6). Focusing on tumor and stromal ecotypes only, we found that the distributions of ecotypes proportions were highly variable among patients, with a tendency towards lower global proportions of tumor spots and higher proportions of stromal spots in patients with relapse versus patients without relapse, especially when restricting analyses to pMMR patients (Supplementary Fig. S7). These observations align well with the well-established notion that stromal-rich tumors are associated with a poorer prognosis than stromal-poor tumors in colon cancer \u003csup\u003e18,19\u003c/sup\u003e. \u003c/p\u003e\n\u003cp\u003eThe above results prompted us to further scrutinize the stromal ecotypes and to perform differential gene analysis between stromal meta-spots from patients with relapse (n=247) and patients without relapse (n=527). We identified a set of 16 genes upregulated in the global stromal compartment of patients that later relapsed (Fig .3a). The levels of these genes were not different when considering all meta-spots combined (Fig. 3b). However, we found that among them, 15 genes were systematically decreased in at least one out of the 4 tumor ecotypes (Supplementary Fig S8A) and that they were all increased in both the stroma-A and stroma-B ecotypes of tumors that later relapsed (Fig. 3b). Interestingly, 4 out of 16 of these genes (\u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, \u003cem\u003eCAPN2\u003c/em\u003e and \u003cem\u003eLRP10\u003c/em\u003e) feature among those specifying the recently described Regenerative Cells (REC) and their related inflammatory RECs (iRECs), two colon cancer cell states associated with metastasis \u003csup\u003e20\u003c/sup\u003e (Fig. 3c). This suggests the presence of such cancer cells in the stroma of patients who will experience disease recurrence. In accordance, REC and iREC gene signatures computed with ssGSEA were enriched in the stromal compartment of patients with relapse as compared to those without relapse (Fig. 3d and Supplementary Fig S8B). To identify potential transcriptional regulators of these four genes, we performed Enrichr analysis on this gene set against the ENCODE-TF database, which highlighted a significant enrichment in TEAD4, a transcription factor that relays YAP activity and targets \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e and \u003cem\u003eCAPN2\u003c/em\u003e (Fig. 3e). These results are consistent with the presence of YAP\u0026ndash;TEAD\u0026ndash;activated cancer cells within stromal regions of patients with relapse (Fig. 3f). In line with this hypothesis, we found a differential enrichment in YAP-associated signatures \u003csup\u003e21,22\u003c/sup\u003e in the stromal meta-spots of patients with versus without relapse (Fig. 3g and Supplementary Fig S8b). The YAP pathway is a major determinant of the Revival Stem Cell (RSC) state \u003csup\u003e23\u003c/sup\u003e, which globally corresponds to RECs \u003csup\u003e20\u003c/sup\u003e. Accordingly, we also found significant enrichments in the Revival Stem Cell (RSC) signature \u003csup\u003e24\u003c/sup\u003e in the stromal compartment of patients with versus without relapse (Fig. 3h and Supplementary Fig S8b). Finally, in view of the demonstration that mouse CC organoids are reprogrammed into an RSC-like (also termed fetal-like) state upon stromal collagen-induced YAP activation \u003csup\u003e25\u003c/sup\u003e, we probed potential differential associations between \u003cem\u003eANXA1\u003c/em\u003e and collagen-expressing genes according to relapse. Interestingly, stronger correlations were observed between levels of \u003cem\u003eANXA1\u003c/em\u003e and those of different collagen type VI-expressing genes in the two stromal compartments of patients with as compared to without relapse (Supplementary Fig. S9). \u003c/p\u003e\n\u003cp\u003eAltogether, by using an unbiased approach, we uncovered enriched \u003cem\u003eANXA1\u003c/em\u003e expression and a YAP-associated REC/RSC state specifically within the stromal compartment of tumors that later relapsed. Taking into account pre-clinical studies and single-cell analyses \u003csup\u003e20,23,25,26\u003c/sup\u003e, our spatially resolved analysis supports the notion that, in patients who eventually relapse, the stromal compartment harbors cancer cells in an \u003cem\u003eANXA1\u003c/em\u003e⁺ REC/RSC state promoted by a YAP-dependent transcriptional program, itself potentially sustained by local collagen enrichment.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStromal meta-clusters split into distinct subclusters with specific molecular features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGiven the consistent enrichment of \u003cem\u003eANXA1\u003c/em\u003e+ REC/RSC-like cells in the stromal compartment of tumors that later relapse, we next sought to refine stromal states by sub-setting the 774 meta-spots assigned to stromal compartments and performing unsupervised re-clustering. This revealed five transcriptionally distinct stromal subclusters (Fig. 4a). The stroma-A ecotype broke down into 2 stromal subclusters. The first one, denoted \u0026ldquo;stroma-A1\u0026rdquo;, was ubiquitous across samples and exhibited the highest universal fibroblast signature (normal fibroblasts, NFs) and, to a lesser extent, various CAF signatures (Fig. 4b and Supplementary Fig. S10 and S11). The latter, which include the CXCL14\u003csup\u003e+\u003c/sup\u003eCAF and the GREM1\u003csup\u003e+\u003c/sup\u003eCAF signatures from \u003csup\u003e17\u003c/sup\u003e and the matCAF signature from \u003csup\u003e27\u003c/sup\u003e, were more enriched in the stroma-A2 subcluster, present in 44 of the 48 samples (Fig. 4b and Supplementary Fig. S10 and S11). Stroma-A2-specifying genes included the prototypical CAF markers \u003cem\u003eLRRC15\u003c/em\u003e and \u003cem\u003ePOSTN \u003c/em\u003e(Supplementary Table S2), both described to sustain tumor growth \u003csup\u003e28,29\u003c/sup\u003e. Among the stroma-B meta-cluster, three subclusters were identified. The stroma-B1 subcluster, found in 43 samples, was enriched in various immune cells, including monocytes, T, NK, B, and cytotoxic lymphocytes cells as well as myeloid-dendritic cells, consistent with an immune-active niche (Fig. 4b). The less frequent stroma-B2 subcluster (27 of the 48 samples, Supplementary Fig. S11), was characterized by high signatures of myofibroblasts, and expression of the contractile CAF marker \u003cem\u003eACTG2\u003c/em\u003e \u003csup\u003e30\u003c/sup\u003e (Supplementary Table S2). Lastly, the stroma-B3 subcluster, which gathered only 24 meta-spots from 20 samples, was defined by high expression of various cytokines and chemokines, including \u003cem\u003eIL11\u003c/em\u003e (Supplementary Table S2). It was characterized by signatures of Inflammation-associated fibroblasts (IAFs) and an enrichment in neutrophils and endothelial cells. This subcluster is also marked by an elevated signature of \u003cem\u003eMMP3\u003c/em\u003e+ CAF described by Pelka et al to be highly active in dMMR patients \u003csup\u003e17\u003c/sup\u003e. In agreement, it was significantly more represented in dMMR vs pMMR patients and in patients without relapse vs patients with relapse (Fig. 4c). Then, we examined the relative expression of the set of genes found to be enriched in the stromal compartment of patients with relapse, in each stromal subcluster, according to disease recurrence (Supplementary Fig. S12). Analyses were restricted to pMMR patients in view of the disbalanced proportions of the different stromal subclusters in dMMR patients (see above). Besides, analyses specific to the stroma-B3 subcluster were considered irrelevant due to the limited number of meta-spots. In line with the stromal-A1 subcluster exhibiting a generic fibroblast phenotype, there was barely any change in this ecotype between patients with or without relapse (Supplementary Fig. S12). In contrast, nearly all genes were more abundantly expressed in the stroma-A2 and stroma-B2 compartments of patients with relapse vs patients without relapse (Supplementary Fig. S12). Taken together, these findings unveil distinct stromal archetypes with different fibroblast and immune contextures.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eVisium HD reveals the spatial arrangement of stromal subclusters and maps \u003cem\u003eANXA1\u003c/em\u003e-expressing REC/RSC-like cells in surrounding stroma\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHaving identified stromal archetypes across our spatial transcriptomic atlas, we sought to refine their respective distribution at a microscale. To this purpose, we re-profiled 4 samples from our initial Visium dataset with Visium HD (Fig. 5a), allowing a single cell-state resolution \u003csup\u003e31\u003c/sup\u003e. We additionally leveraged the publicly available the P2CRC sample from 10X Genomics \u003csup\u003e31\u003c/sup\u003e. \u003cem\u003eEPCAM\u003c/em\u003e and \u003cem\u003eCEACAM6\u003c/em\u003e (as in \u003csup\u003e31\u003c/sup\u003e) were selected as markers for tumor cells (Supplementary Fig. S13). Each stromal subcluster was mapped via the projection of the combined list of its specific marker genes (log-normalized average expression) (Supplementary Fig. S13). The stroma-A1 signature yielded a pervasive, diffuse and faint signal, in contrast to the stroma-A2 signature that intensively stained specific regions outside glandular tumor zones. The stroma-B1 signature produced a faint signal lining tumor cells. The stroma-B2 signature was enriched in regions annotated as smooth muscle and gave a weaker non region-specific signal. Finally, the stroma-B3 signature yielded intense signals at defined zones, notably tumor borders, which may correspond to ulcerative regions. \u003c/p\u003e\n\u003cp\u003eWe then sought to determine which cell type(s) within the stroma express the 3 YAP target genes \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e and \u003cem\u003eCAPN2\u003c/em\u003e. Upon inspection of the stroma-A2 enriched region of 74D1_HD sample at high magnification, we observed clusters of cells that were positive for the 3-gene signature, denoted \u0026ldquo;YAP-Rev\u0026rdquo;, while they expressed low levels of \u003cem\u003eEPCAM\u003c/em\u003e as compared with glandular tumor cells (Fig. 5b). These clusters were annotated by a pathologist (JFE) as poorly differentiated tumor cells (arrows in Fig. 5b). Co-expression analysis distinctly localized such clusters of cancer cells embedded in matCAF-enriched stroma A2 (Fig. 5b). As a whole, these observations reveal a specific, spatially organized distribution of stromal sub-clusters at high resolution and provide in situ evidence for the presence of \u0026ldquo;YAP-Rev\u0026rdquo;\u003cem\u003e \u003c/em\u003ecancer cells within a matCAF-enriched microenvironment. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSpatial transcriptomics-derived signatures are applicable to bulk transcriptomic data and display prognostic value\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile single-cell and spatial transcriptomics analyses are not designed for large clinical cohorts, they may nevertheless bring to light single or composite biomarkers of interest \u003csup\u003e15,32\u003c/sup\u003e. We therefore leveraged our large bulk RNAseq datasets of stage III colon cancer, totaling nearly 3,000 patients \u003csup\u003e33\u003c/sup\u003e, to test whether spatially derived stromal and tumor-intrinsic programs retain prognostic value at the population level. Two independent phase 3 cohorts were analyzed: the full PETACC8 cohort (n=1,733 patients, stage III), from which the spatial transcriptomics samples originated, and 1,248 stage III patients from IDEA-France cohort. Prognostic performance (time to recurrence) of the spatially-derived signatures, computed as continuous z-scores in uni- and multi-variate COX models, is summarized in Fig. 6a-b. Across both cohorts, tumor-intrinsic signatures were consistently associated with favorable prognosis, whereas stromal signatures exhibited adverse effects. Among stromal sub-clusters, the stroma-A1 signature correlated with good outcome, while stroma-A2 reproducibly associated with early recurrence. For stroma-B1 to B3 signatures, results differed between cohorts, likely due to distinct sampling strategies (whole tumor slides for PETACC8, intratumoral punches for IDEA-France). Finally, the \u0026ldquo;YAP-Rev\u0026rdquo; signature composed of \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e and \u003cem\u003eCAPN2\u003c/em\u003e was associated with poor prognosis in both datasets.\u003c/p\u003e\n\u003cp\u003eWe next evaluated these signatures in stage II colon cancer, where stratification tools are scarce \u003csup\u003e34\u003c/sup\u003e. To this purpose, we generated a bulk transcriptomic dataset of n=529 stage II patients from the PRODIGE13 study \u003csup\u003e35\u003c/sup\u003e (see Supplementary Table S3 for patient characteristics). As shown in Fig. 6c, results from PETACC8 and IDEA-France cohorts were globally recapitulated in the PRODIGE13 cohort. \u003c/p\u003e\n\u003cp\u003eTo determine whether \u003cem\u003eANXA1\u003csup\u003e+\u003c/sup\u003e\u003c/em\u003e REC/RSC cells provide clinical information beyond the stromal-A2 program, we first evaluated each signature separately in multivariable Cox models adjusted for grade, WHO performance status, baseline recurrence risk, mismatch repair status, and bowel complications at diagnosis (occlusion or perforation). In these covariate-adjusted models, higher expression of Stroma-A2 was significantly associated with shorter recurrence-free survival (HR = 1.22, 95% CI 1.10\u0026ndash;1.36, p = 1.7\u0026times;10⁻⁴), and YAP-Rev showed a similarly adverse effect (HR = 1.24, 95% CI 1.12\u0026ndash;1.38, p = 2.6\u0026times;10⁻⁵) (Fig. 6a). When both signatures were included simultaneously, each remained independently prognostic (stroma-A2: HR = 1.17, 95% CI 1.06\u0026ndash;1.30, p = 0.0025; YAP-Rev: HR = 1.21, 95% CI 1.09\u0026ndash;1.34, p = 0.0004). Notably, the combined model revealed a statistically significant interaction between the two programs (HR = 1.15, 95% CI 1.04\u0026ndash;1.27, p = 0.0047) when both signatures were included as continuous variables in the multivariable Cox model, demonstrating that the effect of each transcriptional program on recurrence risk depends on the activity of the other.\u003c/p\u003e\n\u003cp\u003eInteraction curves further highlighted this statistical interaction, compatible with a biological synergy between the two programs: across increasing stroma-A2 values, patients with high YAP-Rev exhibited a markedly steeper risk gradient, indicating that stromal activation amplifies the detrimental effect of YAP-Rev score. Reciprocally, the risk associated with YAP-Rev was substantially greater in tumors with high stroma-A2 activity. Together, these findings identify a biologically aggressive subgroup characterized by simultaneous stromal activation and YAP-REC/RSC cells (Fig. 6d).\u003c/p\u003e\n\u003cp\u003eTo externally validate this interaction structure, we computed unscaled, cohort-independent single-sample ssGSVA scores for stroma-A2 and YAP-Rev and applied PETACC8-derived median thresholds to define four groups (Low/Low, High/Low, Low/High, High/High) across cohorts. In PETACC8, the high stroma-A2 /high YAP-Rev group showed the poorest prognosis (HR 2.04, 95% CI 1.59\u0026ndash;2.60, p \u0026lt; 0.001) (Fig. 6e). This phenotype was reproduced in both validation cohorts. In PRODIGE13, the High/High group displayed a markedly elevated recurrence risk (HR = 2.13, 95% CI 1.17\u0026ndash;3.85, p = 0.013), whereas the single-high groups showed no significant effect. IDEA-France showed the same risk hierarchy (High/High: HR = 1.43, 95% CI 1.10\u0026ndash;1.87, p = 0.008) (Fig. 6f), while intermediate groups exhibited weaker or absent associations. The slightly attenuated effect in IDEA-France aligns with its sampling approach: intratumoral punch biopsies underrepresent the stromal compartment.\u003c/p\u003e\n\u003cp\u003eOverall, across three independent cohorts and using PETACC8-derived thresholds, the high stroma-A2 /high YAP-Rev phenotype consistently identified the subgroup with the worst recurrence-free survival, whereas isolated elevation of only one program provided limited prognostic information. These findings robustly support a synergistic interaction between stromal activation and YAP-Rev activity, and position their combined high expression as a clinically meaningful marker of tumor aggressiveness.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study, we have built a spatial transcriptomics atlas of 48 stage III colon cancers, enabling us to mine the diversity of cellular ecosystems in relation to patient characteristics, uncover molecular features associated with disease relapse and derive spatially-informed transcriptomic signatures with prognostic relevance. Our main objective was to generate a comprehensive dataset that balances homogeneity (stage III) with biological and clinical diversity (age, sex, MMR status, disease recurrence), while providing a sufficiently large sample size to identify both shared features and statistically robust group-specific traits. A second objective was to leverage spatial transcriptomics data to generate ecosystem-level insights that could be translated into bulk-compatible clinical biomarkers\u0026nbsp;\u003csup\u003e15\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe identified a structured set of recurrent ecosystems, including tumor- and stroma-enriched, defined by distinct molecular hallmarks and whose spatial organization is consistent with histology. We further described a second level of stroma heterogeneity, allowing to define 5 stromal subclusters characterized by distinct fibroblast subtype signatures, consistent with the notion that fibroblasts dominate the tumor microenvironment\u0026nbsp;\u003csup\u003e31\u003c/sup\u003e. Re-profiling a subset of samples with Visium HD allowed to map these different contingents and revealed in particular a wide-spread distribution of the stroma-A2 ecosystem, sometimes encircling tumor zones, and a more confined pattern for stroma-B3, lining the lumen and most likely corresponding to ulcerative zones. In agreement with this ecotype being enriched in various signatures of IAFs, including one derived from the single cell atlas of colon\u0026nbsp;\u003csup\u003e17\u003c/sup\u003e, the B3 ecotype was more represented in dMMR tumors and in patients who did not experience relapse.\u003c/p\u003e\n\u003cp\u003eSearching for features associated with recurrence allowed identification of a set of 16 genes significantly upregulated in the stromal compartment of patients who later experienced relapse. Importantly, these DEGs exhibited no different expression according to disease recurrence when meta-spots were considered as a global entity, while they displayed a concomitant reduction in at least one tumor ecotypes and increase in the two stromal ecotypes, clearly emphasizing spatial compartment specific differences. One prototypical representative of these DEGs is \u003cem\u003eANXA1\u003c/em\u003e, a well-defined marker of regenerative cell / revival stem / fetal-like marker\u0026nbsp;\u003csup\u003e23,26\u003c/sup\u003e. \u003cem\u003eANXA1\u003c/em\u003e was also identified as a marker associated with tumor budding in colorectal cancer\u0026nbsp;\u003csup\u003e36\u003c/sup\u003e.\u0026nbsp;REC/RSC signatures were specifically increased in stromal ecotypes of relapsing patients, but not in pooled meta-spots, supporting the notion that the stromal niche constitutes a master regulator of colon cancer cell fate transitions \u003csup\u003e37\u003c/sup\u003e. Furthermore, in agreement with YAP being a central driver of the REC/RSC/fetal-like phenotype\u0026nbsp;\u003csup\u003e23,26\u003c/sup\u003e, our data point to two other DEGs beyond \u003cem\u003eANXA1\u003c/em\u003e, namely \u003cem\u003eAHNAK\u003c/em\u003e and \u003cem\u003eCAPN2\u003c/em\u003e, as YAP target genes, and to stromal YAP activation as an accompanying feature of disease recurrence. At the second level of stromal ecosystem diversity, the recurrence-associated increase in \u003cem\u003eANXA1\u003c/em\u003e, \u003cem\u003eAHNAK\u003c/em\u003e and \u003cem\u003eCAPN\u003c/em\u003e expression and RSC signature are recovered in the stroma-A2 subcluster, marked by the expression of prototypical immunosuppressive (e.g. \u003cem\u003eLRCC15\u003c/em\u003e) or pro-tumoral (e.g. \u003cem\u003ePOSTN\u003c/em\u003e) CAF markers. The stroma-A2 gene signature also features \u003cem\u003eCOL10A1\u003c/em\u003e and \u003cem\u003eCTHCR1\u003c/em\u003e, which together with \u003cem\u003ePOSTN\u003c/em\u003e specify the matCAF state defined in a pan-cancer single cell study of CAF\u0026nbsp;\u003csup\u003e27\u003c/sup\u003e. Taken together, these data suggest that collagen-producing CAFs may contribute to a niche favoring the emergence and persistence of \u003cem\u003eANXA1\u003c/em\u003e⁺ REC/RSC cells, potentially through YAP activation driven by \u003cem\u003ePOSTN\u003c/em\u003e-encoded periostin\u0026nbsp;\u003csup\u003e29\u003c/sup\u003e and/or collagen-dependent mechano-transduction\u0026nbsp;\u003csup\u003e25\u003c/sup\u003e. Consistent with this model, we observed stronger correlations between \u003cem\u003eANXA1\u003c/em\u003e expression and that of collagen genes (\u003cem\u003eCOL6A1\u003c/em\u003e, \u003cem\u003eCOL6A2\u003c/em\u003e, and/or \u003cem\u003eCOL6A3\u003c/em\u003e) in the stroma of patients who experienced relapse compared with those who did not. Our Visium HD data, revealing clusters of undifferentiated tumor cells expressing \u003cem\u003eAHNAK\u003c/em\u003e, \u003cem\u003eANXA1\u003c/em\u003e, and \u003cem\u003eCAPN2\u003c/em\u003e (our YAP-Rev signature) embedded within an A2-enriched stromal environment, provides \u003cem\u003ein situ\u003c/em\u003e micro-scale validation of this model. Building on this framework, our data support a model in which matCAF-driven tumor compression triggers local budding. Budding cells then re-activate a YAP-dependent fetal-like program that sustains survival and plasticity within the stromal niche. Based on pre-clinical and clinical studies, these observations raise the possibility that such cells may contribute to resistance to chemotherapy and dissemination\u0026nbsp;\u003csup\u003e20,36,38\u003c/sup\u003e. One remaining is question is how to relate our findings to other cell states associated with disease relapse, in particular the EMP1+ HRCs (High Relapse Cells)\u0026nbsp;\u003csup\u003e39\u003c/sup\u003e. It is noteworthy that this cell population is also marked by high \u003cem\u003eAHNAK\u003c/em\u003e expression (see\u0026nbsp;\u003csup\u003e39\u003c/sup\u003e, which may represent a broad marker of recurrence-associated cells. Second, the HRC population appears to be elevated in KRAS\u003csup\u003eG12D\u003c/sup\u003e mutant versus KRAS non-mutant CC\u0026nbsp;\u003csup\u003e39\u003c/sup\u003e. Third, while fibroblasts were shown to polarize human CC organoids to a RSC state via TGFb\u0026nbsp;and YAP, this transition did not occur in KRAS\u003csup\u003eG12D\u003c/sup\u003e mutant organoids\u0026nbsp;\u003csup\u003e40\u003c/sup\u003e. Hence, while metastatic dissemination of colon cancer cells relies on cell plasticity\u0026nbsp;\u003csup\u003e41\u003c/sup\u003e, multiple metastasis-initiating cell states may be adopted according to the genomic landscape of the tumor\u0026nbsp;\u003csup\u003e39,40,42\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe next assessed the clinical value of these ecotype-derived signatures across independent bulk cohorts. Three key features of the studied cohorts were essential to achieving this goal: a large patient population, detailed clinical annotations, and long-term follow-up. Beyond the two stage III colon cancer cohorts that we recently exploited to derive transcriptomic-based prognostic models of recurrence\u0026nbsp;\u003csup\u003e33\u003c/sup\u003e, we here produced a third dataset with stage II patients from the PRODIGE13 trial. Indeed, stage II tumors comprise the majority of colon cancers\u0026nbsp;\u003csup\u003e43\u003c/sup\u003e, and, although most patients are cured after surgery alone, about 15% of patients experience disease recurrence. One key result of our study is the demonstration that the combination of the stroma-A2 with the YAP-Rev signature robustly identifies patients at high risk of relapse not only in stage III but also in stage II patients.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBeyond their individual association with relapse, the stroma-A2 and YAP-Rev programs displayed a clinically meaningful synergy. In multivariable models, both remained independently prognostic, yet their significant interaction showed that each program\u0026rsquo;s impact depends on the other, pointing to a biologically aggressive subtype driven by mutual reinforcement between a collagen-rich CAF niche and YAP-dependent epithelial plasticity. This cooperation was consistently reproduced across cohorts: only tumors simultaneously high for both programs showed a strong increase in recurrence risk, whereas isolated elevation of either had limited effect. In PRODIGE13, this High/High subgroup also included patients who relapsed despite receiving chemotherapy, suggesting that fetal-like \u003cem\u003eANXA1\u003c/em\u003e⁺states may withstand cytotoxic stress within matCAF-protected niches. Overall, these findings indicate that stromal activation and epithelial revival-like programs must co-occur within the same spatial ecosystem to drive aggressive disease.\u003c/p\u003e\n\u003cp\u003eAltogether, our findings reinforce the idea that spatial transcriptomics is not only a descriptive tool but a platform to generate actionable biomarkers and translate into clinical applications\u0026nbsp;\u003csup\u003e15\u003c/sup\u003e with the demonstration that ecosystem-specific signatures have prognostic value. Current standard-of-care recommendations for localized colon cancer are largely imperfect, with both overtreated patients at low risk of recurrence and undertreated patients at high risk of recurrence. By translating spatially resolved programs into bulk-compatible signatures, we pave the way for future clinical implementation. Such approaches may improve patient stratification beyond current standards and help identify patients at high risk of relapse who might benefit from specific therapeutic strategies, including CAF-targeting agents\u0026nbsp;\u003csup\u003e44\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eImportantly, these signatures were not trained on clinical outcome, yet they robustly stratify patients across cohorts and treatment contexts. This underscores a fundamental principle: biologically grounded transcriptional programs inherently carry prognostic information. As such, they represent not only biomarkers for patient stratification but also a window into the mechanisms underpinning treatment failure and disease recurrence.\u003c/p\u003e\n\u003cp\u003eOur work provides a framework to functionally dissect the tumor microenvironment and nominates molecular targets, particularly within the stromal compartment, which may inform future therapeutic strategies. The spatial dissection of tumors thus emerges as a powerful tool, not only to classify disease more accurately, but to uncover actionable biology with direct clinical relevance.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eStudy Population and Sample Selection for Spatial Transcriptomics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eForty-eight stage III colon adenocarcinomas were selected from patients enrolled in the PETACC8 clinical trial (DOI: 10.1016/S1470-2045(14)70227-X) for spatial transcriptomics analysis. Tumor samples were preserved as FFPE tissue blocks and selected to reflect the clinical and molecular diversity of the cohort, including 14 patients who experienced disease recurrence and 34 who did not, as previously described \u003csup\u003e33\u003c/sup\u003e. Patients enrolled in the PETACC8 trial had histologically confirmed stage III colon adenocarcinoma (pTxN+M0) and received adjuvant chemotherapy: either FOLFOX (oxaliplatin, fluorouracil, leucovorin) or FOLFOX plus cetuximab. Clinical, biological, histological, and molecular data, including survival outcomes, were collected prospectively and made available by the F\u0026eacute;d\u0026eacute;ration Francophone de Canc\u0026eacute;rologie Digestive (FFCD). All patients provided written informed consent for specific translational research and the study protocol was approved by appropriate institutional review boards.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eTissue Selection and Processing and Visium SD Spatial Transcriptomics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTissue blocks were selected based on centralized pathological review to ensure the presence of viable invasive carcinoma, adequate RNA quality, and sufficient tissue area compatible with Visium processing. Sections were chosen to be representative of the biological and clinical heterogeneity of the cohort and, when feasible, encompassed tumors with different mismatch repair status and histological features. Most selected sections contained both epithelial and stromal compartments, including tumor\u0026ndash;stroma interfaces, rather than being chosen to maximize tumor purity or target a specific anatomical region. Spatial transcriptomics was performed on 5 \u0026micro;m-thick FFPE sections using the 10X Genomics Visium Spatial Gene Expression assay for FFPE tissue (https://www.10xgenomics.com/support/spatial-gene-expression-ffpe/) (RRID RRID:SCR_023571), following the manufacturer\u0026rsquo;s instructions. Sequencing data were processed using \u003cem\u003eSpace Ranger\u003c/em\u003e (RRID:SCR_025848) (10X Genomics, version 1.3.0, aligned to the \u003cem\u003eGRCh38\u003c/em\u003e reference transcriptome. Libraries were prepared according to the standard Visium protocol (10X Genomics; Spatial gene expression assay protocol CG000407).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eImaging and Histological Annotation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH\u0026amp;E-stained images were annotated by a senior pathologist (J.-F. Emile) using \u003cem\u003eNDP.view 2\u003c/em\u003e software. Annotated features included tumor and non-tumor regions, mucinous areas, tertiary lymphoid structures (TLS), and tumor invasion fronts. Annotations were imported in LoupeBrowser v8. The corresponding spatial barcodes were stored in the metadata as qualitative variables. Representative H\u0026amp;E staining sections of the various ecotypes were visualized using the QuPath software.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eCopy Number Variation Inference\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eCopy number variation was inferred using the fastCNV method (https://github.com/must-bioinfo/fastCNV) applied to each Visium dataset. The CNV burden was summarized as the mean absolute deviation across the genome for each spot or cluster.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eMeta-spot Generation and Meta-cluster Integration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor cross-sample spatial analysis, a second workflow was implemented using \u003cem\u003eSeurat\u003c/em\u003e (v5.3.0) (RRID:SCR_016341). Data were normalized using NormalizeData() and scaled with ScaleData(). Within each sample, high-resolution clustering was first performed using FindClusters() at a resolution of 10. For each resulting cluster, gene expression values were averaged on the scaled expression matrix (scale.data) across all constituent spots to generate a \u0026ldquo;meta-spot\u0026rdquo;, resulting in 3,231 meta-spots across the 48 tumors. A new Seurat object was created using the meta-spot expression matrix, and the averaged matrix was directly assigned to the scale.data slot. Dimensionality reduction was performed using principal component analysis (PCA) on all genes (50 PCs), followed by batch correction using the Harmony algorithm (v1.2.0) (RRID:SCR_022206), where batch was defined by the sample identity encoded in meta-spot names. The Harmony-reduced matrix was used for UMAP embedding (dims = 1:15), neighborhood graph construction, and clustering (resolution = 0.5). Eleven meta-clusters were identified and annotated based on CMS, TME, histology, and CNV profiles. Two meta-clusters (MC2 and MC4) were identified as stromal and used for downstream stromal subclustering.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eStromal Subcluster Identification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMeta-spots from MC2 and MC4 were extracted (n=774) and reclustered using \u003cem\u003eSeurat\u003c/em\u003e at resolution 0.5. Five distinct stromal subclusters were identified and annotated based on their transcriptional profiles and known CAF or immune markers.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eDifferential Gene Expression \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDifferentially expressed genes in meta-spots of the stromal compartment of patients with versus without relapse were identified using a Wilcoxon rank-sum test using Seurat\u0026rsquo;s FindMarkers function with the following filters: logfc.threshold \u0026gt;0.6, min.pct = 0.5, adjusted p.value \u0026lt;0.01. Analysis of ENCODE-TF binding motifs was performed with the enrichR package https://github.com/wjawaid/enrichR (RRID:SCR_001575).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eSignature Generation and Projection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor each meta-cluster and stromal subcluster, transcriptional signatures were defined by selecting the top 20 differentially expressed genes (DEGs), identified using FindAllMarkers() with thresholds of adjusted p-value \u0026lt; 0.05 and log2 fold-change \u0026gt; 1. The complete gene lists used to define each signature are provided in Supplementary Table S2. Signature scores were computed using single-sample gene set enrichment analysis (ssGSEA via \u003cem\u003eGSVA\u003c/em\u003e package (v1.50.0) RRID:SCR_021058) applied to the scaled expression matrix (scale.data) of each sample. The enrichment was calculated using normalized scores (normalize = TRUE), allowing comparison across gene sets within a given sample. These scores were used for both intra-sample characterization of ecotypes and for projection onto bulk transcriptomic datasets (PETACC8, IDEA-France, PRODIGE13).\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eSpatial neighbor analysis of Visium metaclusters\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSpatial interactions between metaclusters were quantified from 10x Genomics Visium sections by identifying the six closest tissue neighbors of each spot (first spatial crown). Tissue‐restricted barcodes were extracted using Seurat and mapped to physical coordinates. For every spot, the metacluster identities of its six immediate neighbors were determined using euclidean proximity computed with the knearneigh function (\u003cem\u003espdep\u003c/em\u003e R package v1.3-5). For each stromal spot, neighboring spots were visualized in an Alluvial plot.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eVisium HD Spatial Transcriptomics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH\u0026amp;E staining and imaging were performed according to the 10X Visium HD FFPE Tissue Preparation Handbook (CG000684). Sample processing and spatial transcriptomics were performed using the 10X Visium HD Spatial Gene Expression Reagents Kits User Guide (CG000685). The Visium HD tissue section was processed using SpaceRanger (10x Genomics, version 3.1.1). Transcript alignment was performed against the GRCh38-2020-A reference transcriptome (refdata-gex-GRCh38-2020-A). Transcript detection was carried out using the Visium Human Transcriptome Probe Set v2.0 (GRCh38-2020-A) provided by 10x Genomics. Fiducial alignment and spatial registration quality control were performed using Loupe Browser version 8. All analyses of the Visium HD slide were performed using Loupe Browser version 8. Gene expression plots were generated using normalized log-transformed expression values, scaled between 0 and 7. For signatures, plots were generated using means of individual normalized log-transformed expression values, scaled between 0 and 7.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003e3\u0026rsquo;RNA Sequencing of Tumor Samples\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe PRODIGE13 study (FFCD PRODIGE-13; ClinicalTrials.gov identifier NCT00995202) is a randomized phase III trial that evaluated the benefit of intensive radiological and CEA monitoring versus standard surveillance in patients with stage II and III colorectal cancer \u003csup\u003e35\u003c/sup\u003e. A subset of 529 stage II tumors from this cohort underwent 3\u0026prime; RNA-sequencing as part of a translational research program. Briefly, tumor RNA was extracted from macrodissected FFPE tissue sections using the Maxwell RSC RNA FFPE kit (Promega). The PolyA-RNA sequencing (RNAseq) library preparation protocols were performed using 400 ng of template RNA and the QuantSeq 3\u0026rsquo;mRNA-Seq Kit FWD for Illumina (Lexogen, Vienna, Austria) according to the manufacturer\u0026rsquo;s instructions. Libraries were sequenced on NovaSeq6000 (Illumina, San Diego, CA). Raw sequencing data was processed as in \u003csup\u003e33\u003c/sup\u003e to generate the dataset. Clinical annotation and survival data were available for all sequenced samples. \u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eSurvival Analyses \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSurvival analyses were performed on bulk transcriptomic data from PETACC8 (stage III), IDEA-France (stage III) and PRODIGE13 (stage II). Signature scores were treated as continuous variables, quantified by ssGSEA (\u003cem\u003eGSVA\u003c/em\u003e package (v1.50.0)) and standardized per signature (mean = 0, SD = 1) to allow comparability of hazard ratios. Univariate and multivariable Cox proportional hazards models were fitted using the survival package in R (v.3.8-3) (RRID:SCR_021137); multivariable models were adjusted for grade, WHO performance status, baseline high-risk features (pT4 and/or pN2 when available), mismatch repair status and bowel complications at diagnosis (occlusion or perforation). Interaction was assessed using a continuous Stroma-A2 \u0026times; YAP-Rev term and visualized by estimating predicted hazard ratios with the interacting program fixed at representative low or high values (25th or 75th percentile, Q1/Q3) and centering predictions on a reference patient with median program expression and standard clinical covariates. For categorical stratification, patients were assigned to Low/Low, High/Low, Low/High and High/High groups using PETACC8 median thresholds computed on raw ssGSEA values and applied unchanged to all cohorts. Kaplan\u0026ndash;Meier curves and categorical hazard ratios were compared using two-sided log-rank tests and unadjusted Cox models (survminer, ggplot2, broom). Significance was set at p \u0026lt; 0.05.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSpearman correlations were computed for variable pairs and integrated using the median coefficient. Group comparisons were performed using non-parametric tests (Mann\u0026ndash;Whitney U test, Kruskal\u0026ndash;Wallis) or parametric equivalents where applicable. All statistical analyses were conducted using R (v4.1.1) on \u003cem\u003eRStudio Server\u003c/em\u003e (RRID:SCR_000432) hosted by the IFB-core cluster.\u003c/p\u003e\n\n\u003cp\u003e\u003cstrong\u003eData and Code Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRaw spatial transcriptomics data has been deposited at the European Genome-Phenome Archive under Study ID EGAD50000002091. Images can be downloaded from https://doi.org/10.5281/zenodo.17711406 and the complete code used to generate the main figures is available on GitHub at https://github.com/crcordeliers/CCVisiumAtlas. Bulk transcriptomic datasets (PETACC8, IDEA-France) are available as described in \u003csup\u003e33\u003c/sup\u003e. Bulk transcriptomic data for PRODIGE13 cohort are available within the framework of FFCD data-sharing policies upon request to the corresponding authors. \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflict of interest in the context of the present study\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003cstrong\u003eAcknowledgments\u0026nbsp;:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by Institut National du Cancer (INCa, IMPROCOCA project, 2023-119), the SIRIC CARPEM (INCa-DGOS-Inserm-ITMO Cancer_18006), the Labex Onco-Immunology (investissement d\u0026rsquo;avenir) and the InidEx Immuno-Onco (Initiatives d\u0026rsquo;excellence, Universit\u0026eacute; Paris Cit\u0026eacute;). The group is supported by the Ligue Nationale Contre le Cancer (Equipe Labellis\u0026eacute;e). A. Cazelles was supported by a fellowship from Association pour la Recherche Contre le Cancer and M. Sroussi was funded by Fondation pour la Recherche M\u0026eacute;dicale (grant FDM202006011237). I. Hern\u0026aacute;ndez-Verdin was founded by BETPSY project, overseen by the French National Research Agency, as part of the second \u0026ldquo;Investissements d\u0026rsquo;Avenir\u0026rdquo; program (grant number ANR-18-RHUS-0012). We thank the CAIBI platform at the Centre de Recherche des Cordeliers for help in data deposition and Pauline Hamon for fruitful discussion.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAC\u003c/strong\u003e: Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Writing \u0026ndash; original draft\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCP\u003c/strong\u003e: Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDLC\u003c/strong\u003e: Investigation, Data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGC\u003c/strong\u003e: Formal analysis, Visualization, Data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMS\u003c/strong\u003e: Formal analysis\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCM\u003c/strong\u003e: Investigation, Resources\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCG\u003c/strong\u003e: Resources, Clinical data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGM\u003c/strong\u003e: Resources, Clinical data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMK:\u003c/strong\u003e Resources, Clinical data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIH\u003c/strong\u003e: Data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAB\u003c/strong\u003e: Investigation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWHF\u003c/strong\u003e: Supervision\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLP:\u003c/strong\u003e Data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdR\u003c/strong\u003e: Methodology, Supervision\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCL\u003c/strong\u003e: Resources, Clinical data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTA\u003c/strong\u003e: Resources, Clinical data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eJFE\u003c/strong\u003e: Validation, Data curation\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTZH\u003c/strong\u003e: Validation, Writing \u0026ndash; review \u0026amp; editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eJT\u003c/strong\u003e: Resources, Clinical data curation, Writing \u0026ndash; review \u0026amp; editing\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePLP\u003c/strong\u003e: Conceptualization, Methodology, Formal analysis, Data curation, Supervision, Writing \u0026ndash; review \u0026amp; editing, Funding acquisition\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSMR\u003c/strong\u003e: Conceptualization, Methodology, Formal analysis, Data curation, Visualization, Supervision, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing, Project administration, Funding acquisition\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSung, H. \u003cem\u003eet al.\u003c/em\u003e Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. \u003cem\u003eCA. Cancer J. Clin.\u003c/em\u003e 71, 209\u0026ndash;249 (2021).\u003c/li\u003e\n\u003cli\u003eEng, C. \u003cem\u003eet al.\u003c/em\u003e Colorectal cancer. \u003cem\u003eLancet\u003c/em\u003e 404, 294\u0026ndash;310 (2024).\u003c/li\u003e\n\u003cli\u003eGuinney, J. \u003cem\u003eet al.\u003c/em\u003e The consensus molecular subtypes of colorectal cancer. \u003cem\u003eNat. Med.\u003c/em\u003e 21, 1350\u0026ndash;1356 (2015).\u003c/li\u003e\n\u003cli\u003eBecht, E. \u003cem\u003eet al.\u003c/em\u003e Immune and Stromal Classification of Colorectal Cancer Is Associated with Molecular Subtypes and Relevant for Precision Immunotherapy. \u003cem\u003eClin. Cancer Res.\u003c/em\u003e 22, 4057\u0026ndash;66 (2016).\u003c/li\u003e\n\u003cli\u003eMarisa, L. \u003cem\u003eet al.\u003c/em\u003e Intratumor CMS Heterogeneity Impacts Patient Prognosis in Localized Colon Cancer. \u003cem\u003eClin. Cancer Res.\u003c/em\u003e 27, 4768\u0026ndash;4780 (2021).\u003c/li\u003e\n\u003cli\u003eLangerud, J. \u003cem\u003eet al.\u003c/em\u003e Multiregional transcriptomics identifies congruent consensus subtypes with prognostic value beyond tumor heterogeneity of colorectal cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e 15, 4342 (2024).\u003c/li\u003e\n\u003cli\u003eElhanani, O., Ben-Uri, R. \u0026amp; Keren, L. Spatial profiling technologies illuminate the tumor microenvironment. \u003cem\u003eCancer Cell\u003c/em\u003e 41, 404\u0026ndash;420 (2023).\u003c/li\u003e\n\u003cli\u003eChen, J., Larsson, L., Swarbrick, A. \u0026amp; Lundeberg, J. Spatial landscapes of cancers: insights and opportunities. \u003cem\u003eNat. Rev. Clin. Oncol.\u003c/em\u003e 21, 660\u0026ndash;674 (2024).\u003c/li\u003e\n\u003cli\u003eGracia Villacampa, E. \u003cem\u003eet al.\u003c/em\u003e Genome-wide spatial expression profiling in formalin-fixed tissues. \u003cem\u003eCell Genomics\u003c/em\u003e 1, 100065 (2021).\u003c/li\u003e\n\u003cli\u003eValdeolivas, A. \u003cem\u003eet al.\u003c/em\u003e Profiling the heterogeneity of colorectal cancer consensus molecular subtypes using spatial transcriptomics. \u003cem\u003eNPJ Precis. Oncol.\u003c/em\u003e 8, 10 (2024).\u003c/li\u003e\n\u003cli\u003eWang, F. \u003cem\u003eet al.\u003c/em\u003e Single-cell and spatial transcriptome analysis reveals the cellular heterogeneity of liver metastatic colorectal cancer. \u003cem\u003eSci. Adv.\u003c/em\u003e 9, eadf5464 (2023).\u003c/li\u003e\n\u003cli\u003eWood, C. S. \u003cem\u003eet al.\u003c/em\u003e Spatially Resolved Transcriptomics Deconvolutes Prognostic Histological Subgroups in Patients with Colorectal Cancer and Synchronous Liver Metastases. \u003cem\u003eCancer Res.\u003c/em\u003e 83, 1329\u0026ndash;1344 (2023).\u003c/li\u003e\n\u003cli\u003eWu, Y. \u003cem\u003eet al.\u003c/em\u003e Spatiotemporal Immune Landscape of Colorectal Cancer Liver Metastasis at Single-Cell Level. \u003cem\u003eCancer Discov.\u003c/em\u003e 12, 134\u0026ndash;153 (2022).\u003c/li\u003e\n\u003cli\u003eQi, J. \u003cem\u003eet al.\u003c/em\u003e Single-cell and spatial analysis reveal interaction of FAP+ fibroblasts and SPP1+ macrophages in colorectal cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e 13, 1742 (2022).\u003c/li\u003e\n\u003cli\u003eGulati, G. S., D\u0026rsquo;Silva, J. P., Liu, Y., Wang, L. \u0026amp; Newman, A. M. Profiling cell identity and tissue architecture with single-cell and spatial transcriptomics. \u003cem\u003eNat. Rev. Mol. Cell Biol.\u003c/em\u003e 26, 11\u0026ndash;31 (2025).\u003c/li\u003e\n\u003cli\u003eTaieb, J. \u003cem\u003eet al.\u003c/em\u003e Oxaliplatin, fluorouracil, and leucovorin with or without cetuximab in patients with resected stage III colon cancer (PETACC-8): an open-label, randomised phase 3 trial. \u003cem\u003eLancet Oncol.\u003c/em\u003e 15, 862\u0026ndash;873 (2014).\u003c/li\u003e\n\u003cli\u003ePelka, K. \u003cem\u003eet al.\u003c/em\u003e Spatially organized multicellular immune hubs in human colorectal cancer. \u003cem\u003eCell\u003c/em\u003e 184, 4734-4752.e20 (2021).\u003c/li\u003e\n\u003cli\u003eCorry, S. M. \u003cem\u003eet al.\u003c/em\u003e Activation of innate-adaptive immune machinery by poly(I:C) exposes a therapeutic vulnerability to prevent relapse in stroma-rich colon cancer. \u003cem\u003eGut\u003c/em\u003e 71, 2502\u0026ndash;2517 (2022).\u003c/li\u003e\n\u003cli\u003eMouillet-Richard, S. \u003cem\u003eet al.\u003c/em\u003e Clinical Challenges of Consensus Molecular Subtype CMS4 Colon Cancer in the Era of Precision Medicine. \u003cem\u003eClin. Cancer Res.\u003c/em\u003e 30, 2351\u0026ndash;2358 (2024).\u003c/li\u003e\n\u003cli\u003eOgden, S. \u003cem\u003eet al.\u003c/em\u003e Phenotypic heterogeneity and plasticity in colorectal cancer metastasis. \u003cem\u003eCell Genomics\u003c/em\u003e 5, 100881 (2025).\u003c/li\u003e\n\u003cli\u003eGregorieff, A., Liu, Y., Inanlou, M. R., Khomchuk, Y. \u0026amp; Wrana, J. L. Yap-dependent reprogramming of Lgr5(+) stem cells drives intestinal regeneration and cancer. \u003cem\u003eNature\u003c/em\u003e 526, 715\u0026ndash;718 (2015).\u003c/li\u003e\n\u003cli\u003eSerra, D. \u003cem\u003eet al.\u003c/em\u003e Self-organization and symmetry breaking in intestinal organoid development. \u003cem\u003eNature\u003c/em\u003e 569, 66\u0026ndash;72 (2019).\u003c/li\u003e\n\u003cli\u003eFey, S. K., Vaquero-Siguero, N. \u0026amp; Jackstadt, R. Dark force rising: Reawakening and targeting of fetal-like stem cells in colorectal cancer. \u003cem\u003eCell Rep.\u003c/em\u003e 43, 114270 (2024).\u003c/li\u003e\n\u003cli\u003eAyyaz, A. \u003cem\u003eet al.\u003c/em\u003e Single-cell transcriptomes of the regenerating intestine reveal a revival stem cell. \u003cem\u003eNature\u003c/em\u003e 569, 121\u0026ndash;125 (2019).\u003c/li\u003e\n\u003cli\u003evan der Net, M. C. \u003cem\u003eet al.\u003c/em\u003e Mechanosensitive calcium channels and integrins coordinate the reprogramming of colorectal cancer cells into a fetal-like state. \u003cem\u003eCell Rep.\u003c/em\u003e 44, 116308 (2025).\u003c/li\u003e\n\u003cli\u003eTape, C. J. Plastic persisters: revival stem cells in colorectal cancer. \u003cem\u003eTrends Cancer\u003c/em\u003e 10, 185\u0026ndash;195 (2024).\u003c/li\u003e\n\u003cli\u003eChen, B. \u003cem\u003eet al.\u003c/em\u003e The molecular classification of cancer-associated fibroblasts on a pan-cancer single-cell transcriptional atlas. \u003cem\u003eClin. Transl. Med.\u003c/em\u003e 13, e1516 (2023).\u003c/li\u003e\n\u003cli\u003eKrishnamurty, A. T. \u003cem\u003eet al.\u003c/em\u003e LRRC15+ myofibroblasts dictate the stromal setpoint to suppress tumour immunity. \u003cem\u003eNature\u003c/em\u003e 611, 148\u0026ndash;154 (2022).\u003c/li\u003e\n\u003cli\u003eMa, H. \u003cem\u003eet al.\u003c/em\u003e Periostin Promotes Colorectal Tumorigenesis through Integrin-FAK-Src Pathway-Mediated YAP/TAZ Activation. \u003cem\u003eCell Rep.\u003c/em\u003e 30, 793-806.e6 (2020).\u003c/li\u003e\n\u003cli\u003eGiguelay, A. \u003cem\u003eet al.\u003c/em\u003e The landscape of cancer-associated fibroblasts in colorectal cancer liver metastases. \u003cem\u003eTheranostics\u003c/em\u003e 12, 7624\u0026ndash;7639 (2022).\u003c/li\u003e\n\u003cli\u003eOliveira, M. F. de \u003cem\u003eet al.\u003c/em\u003e High-definition spatial transcriptomic profiling of immune cell populations in colorectal cancer. \u003cem\u003eNat. Genet.\u003c/em\u003e 57, 1512\u0026ndash;1523 (2025).\u003c/li\u003e\n\u003cli\u003eL\u0026aacute;z\u0026aacute;r, E. \u0026amp; Lundeberg, J. Spatial architecture of development and disease. \u003cem\u003eNat. Rev. Genet.\u003c/em\u003e https://doi.org/10.1038/s41576-025-00892-5 (2025) doi:10.1038/s41576-025-00892-5.\u003c/li\u003e\n\u003cli\u003eGallois, C. \u003cem\u003eet al.\u003c/em\u003e Prognostic Models From Transcriptomic Signatures of the Tumor Microenvironment and Cell Cycle in Stage III Colon Cancer From PETACC-8 and IDEA-France Trials. \u003cem\u003eJ. Clin. Oncol.\u003c/em\u003e JCO2302262 (2025) doi:10.1200/JCO.23.02262.\u003c/li\u003e\n\u003cli\u003eTaieb, J., Karoui, M. \u0026amp; Basile, D. How I treat stage II colon cancer patients. \u003cem\u003eESMO Open\u003c/em\u003e 6, 100184 (2021).\u003c/li\u003e\n\u003cli\u003eLepage, C. \u003cem\u003eet al.\u003c/em\u003e Effect of 5 years of CT-scan and CEA follow-up on survival endpoints in patients with colorectal cancer. \u003cem\u003eAnn. Oncol.\u003c/em\u003e S0923-7534(25)04701\u0026ndash;5 (2025) doi:10.1016/j.annonc.2025.09.004.\u003c/li\u003e\n\u003cli\u003eDe Smedt, L. \u003cem\u003eet al.\u003c/em\u003e Expression profiling of budding cells in colorectal cancer reveals an EMT-like phenotype and molecular subtype switching. \u003cem\u003eBr. J. Cancer\u003c/em\u003e 116, 58\u0026ndash;65 (2017).\u003c/li\u003e\n\u003cli\u003eEngland, F. J., Lin, M., Sigal, M. \u0026amp; Leedham, S. J. Defining the mucosal ecosystem: epithelial-mesenchymal interdependence in gastrointestinal health and disease. \u003cem\u003eNat. Rev. Gastroenterol. Hepatol.\u003c/em\u003e 22, 741\u0026ndash;754 (2025).\u003c/li\u003e\n\u003cli\u003eRamos Zapatero, M. \u003cem\u003eet al.\u003c/em\u003e Trellis tree-based analysis reveals stromal regulation of patient-derived organoid drug responses. \u003cem\u003eCell\u003c/em\u003e 186, 5606-5619.e24 (2023).\u003c/li\u003e\n\u003cli\u003eCa\u0026ntilde;ellas-Socias, A. \u003cem\u003eet al.\u003c/em\u003e Metastatic recurrence in colorectal cancer arises from residual EMP1+ cells. \u003cem\u003eNature\u003c/em\u003e 611, 603\u0026ndash;613 (2022).\u003c/li\u003e\n\u003cli\u003eQin, X. \u003cem\u003eet al.\u003c/em\u003e An oncogenic phenoscape of colonic stem cell polarization. \u003cem\u003eCell\u003c/em\u003e 186, 5554-5568.e18 (2023).\u003c/li\u003e\n\u003cli\u003eMoorman, A. \u003cem\u003eet al.\u003c/em\u003e Progressive plasticity during colorectal cancer metastasis. \u003cem\u003eNature\u003c/em\u003e 637, 947\u0026ndash;954 (2025).\u003c/li\u003e\n\u003cli\u003eCammareri, P. \u003cem\u003eet al.\u003c/em\u003e Loss of colonic fidelity enables multilineage plasticity and metastasis. \u003cem\u003eNature\u003c/em\u003e 644, 547\u0026ndash;556 (2025).\u003c/li\u003e\n\u003cli\u003eIARC Working Group on the Evaluation of Cancer-Preventive Interventions. \u003cem\u003eColorectal Cancer Screening\u003c/em\u003e. (International Agency for Research on Cancer, Lyon (FR), 2019).\u003c/li\u003e\n\u003cli\u003eTimperi, E. \u003cem\u003eet al.\u003c/em\u003e At the Interface of Tumor-Associated Macrophages and Fibroblasts: Immune-Suppressive Networks and Emerging Exploitable Targets. \u003cem\u003eClin. Cancer Res.\u003c/em\u003e 30, 5242\u0026ndash;5251 (2024).\u003c/li\u003e\n\u003cli\u003eBuechler, M. B. \u003cem\u003eet al.\u003c/em\u003e Cross-tissue organization of the fibroblast lineage. \u003cem\u003eNature\u003c/em\u003e 593, 575\u0026ndash;579 (2021).\u003c/li\u003e\n\u003cli\u003eSmillie, C. S. \u003cem\u003eet al.\u003c/em\u003e Intra- and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis. \u003cem\u003eCell\u003c/em\u003e 178, 714-730.e22 (2019).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8583137/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8583137/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Despite advances in characterizing intra-tumor heterogeneity in colon cancer (CC), its spatial organization remains to be fully delineated. We generated a large spatial transcriptomic atlas of 48 stage III CC and identified recurrent, biologically relevant spatial ecosystems. Recurrence-associated analysis revealed stromal-specific upregulation of a set of genes including the regenerative/revival-stem cell (REC/RSC) marker ANXA1, together with enrichment of a YAP-TEAD program. These signals converged into a refined three-gene YAP-Rev signature (AHNAK, ANXA1, CAPN2) capturing the fetal-like state associated with relapse. Deeper dissection of the stromal compartment revealed additionally layers of heterogeneity. Further Visium HD profiling allowed spotting ANXA1-expressing tumor cells surrounded by collagen-producing cancer-associated fibroblasts (matCAFs). Finally, we translated these findings to bulk transcriptomics and demonstrated a synergistic interaction between matCAF enriched stroma derived-signature and ANXA1-REC/RSC score with strong prognostic value across three independent cohorts comprising \u003e3,500 stage II and stage III CC patients.","manuscriptTitle":"Profiling colon cancer architecture with spatial transcriptomics identifies clinically relevant stromal ecotypes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-27 15:14:43","doi":"10.21203/rs.3.rs-8583137/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-genetics","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"ng","sideBox":"Learn more about [Nature Genetics](http://www.nature.com/ng/)","snPcode":"","submissionUrl":"","title":"Nature Genetics","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Research","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"b5bafccc-c0fe-4ed7-b11f-7259c5882ed4","owner":[],"postedDate":"February 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":63603090,"name":"Biological sciences/Cancer/Gastrointestinal cancer/Colorectal cancer/Colon cancer"},{"id":63603091,"name":"Biological sciences/Cancer/Cancer microenvironment"},{"id":63603092,"name":"Biological sciences/Cancer/Tumour biomarkers"},{"id":63603093,"name":"Health sciences/Diseases/Cancer/Gastrointestinal cancer/Colorectal cancer/Colon cancer"}],"tags":[],"updatedAt":"2026-03-05T22:07:03+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-27 15:14:43","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8583137","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8583137","identity":"rs-8583137","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}