Abstract Pancreatic Ductal Adenocarcinoma (PDAC) remains one of the most lethal cancers, with only 20% of patients surviving 5 years post pancreatectomy. Despite rationale for targeting immune pathways, existing immunotherapies have provided minimal benefit to date. The immune checkpoint molecule, B7-H3, has emerged as a promising therapeutic target across solid tumours. However, its cellular origins, spatial context and clinical relevance in PDAC is poorly defined. This study aims to address this unmet need. An integrated spatial multi-omics strategy was applied across two treatment naïve PDAC cohorts. These FFPE tissue microarrays were divided into a test cohort (n=27), and a validation cohort (n=74). GeoMx regional proteomics (Bruker) was applied on section of the test cohort, while GeoMx regional whole transcriptome (Bruker), CosMx single cell 64-plex protein (Bruker) and CosMx 6K RNA (Bruker) assays were applied to serial sections of the validation cohort. Additional CosMx single cell whole transcriptome (WTx) data was generated from an Intraductal Papillary Mucinous Neoplasm (IPMN) whole section cohort (n=6). GeoMx samples were fluorescently stained with Syto13, PanCk, CD45 and αSMA, and masks were selected for PanCk+ve epithelium and tumour microenvironment (TME). CosMx samples were fluorescently stained for DAPI, PanCk, CD298/B2M, CD45 and CD68. Extensive clinicopathological data was available for all samples. Across regional protein and transcriptome profiling, B7-H3 expression was consistently elevated in TME and fibroblast enriched compartments compared to epithelium. Reduced epithelial expression significantly correlated with improved disease specific survival. B7-H3 ranked transcriptomic regional signatures revealed suppression of T-cell activation and antigen presentation pathways, alongside enrichment of fibro-inflammatory, myeloid and stromal signalling pathways. Single-cell spatial proteomics and RNA validated and resolved these signatures to discrete cellular niches. B7-H3 high tumours were characterised by dense clustering of fibroblasts, antigen-presenting macrophages and exhausted T-cell phenotypes, whereas B7-H3 low tumours demonstrated epithelial dominant clustering and increased spatial separation between B7-H3 enriched cell types and cytotoxic lymphocytes. WTx data further demonstrated graded B7-H3 expression variation from normal to dysplasia to IPMN cancer, highlighting spatial-temporal remodelling throughout oncogenesis. Across all spatial platforms, B7-H3 consistently mapped within fibro-inflammatory and immunosuppressive niches, features strongly associated with aggressive disease and poor prognosis. Collectively, these findings establish B7-H3 as a robust spatial biomarker in PDAC and provide compelling biological rationale for therapeutic targeting of B7-H3 in a malignancy historically resistant to immune-based treatments. Citation Format: Assya Legrini, Tengyu Zhang, Mari-Claire McGuigan, Colin Wood, Luke McNickle, Claire Kennedy-Dietrich, Ghazal Latifi, Yoana Doncheva, Josefina Vasquez, Hannah Morgan, Pamela McCall, Michail Doukas, Joanne Edwards, Nigel Jamieson. Spatial mapping of B7-H3 mediated cellular plasticity in treatment naive pancreatic cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 3951.

Abstract Background: Locally advanced rectal cancer (LARC) shows highly variable responses to neoadjuvant radiotherapy with a minority of patients achieving complete tumor regression. Existing clinical tools cannot reliably predict responders using pre-treatment samples. Understanding the dynamic molecular and immune changes occurring during and after radiotherapy is initiated is therefore critical to guide personalized treatment strategies and improve outcomes. Aims/Objectives: This study employed a serial rectal tumor sampling protocol during neoadjuvant treatment. We aim to characterize longitudinal transcriptomic changes in LARC tumors during radiotherapy and to identify early molecular and microenvironmental features associated with treatment response. Methods/Results: Tumor biopsies were collected from 38 patients at baseline, week 2, week 6, and week 12 during short-course or long-course radiotherapy. Bulk RNA-seq was used to profile gene expression across time. PCA and differential expression analyses showed that most transcriptional reprogramming occurred between baseline and week 2, with minimal changes at later time points. Responders exhibited early downregulation of proliferation-related genes alongside sustained activation of immune pathways. Divergence from non-responders was most evident by week 2, highlighting a critical window for treatment response. Consensus Molecular Subtype (CMS) classification showed enrichment of CMS2 in non-responders and CMS4 in responders. Tissue composition further influenced transcriptional patterns, with stromal-rich samples in responders displaying stronger immune signatures. Based on Cell deconvolution tools, responders showed pronounced week-2 activation of myeloid and lymphoid populations, while non-responders exhibited delayed immune activation. Conclusions: This study shows that radiotherapy response in LARC is determined early, with week 2 emerging as a key point where responders and non-responders clearly diverge. Responders rapidly suppress proliferation and activate strong immune programs, while non-responders show delayed or minimal immune engagement. These early molecular and microenvironmental shifts highlight an important role for early immune activation in directing successful RT responses. These features, present early during therapy suggest that on-treatment predictive biomarker development should be pursued to support development of adaptive, personalized treatment strategies. Citation Format: Fiza Ishaqwala, Ashley McCulloch, Lily Hillson, Liang Tang, Leonor Schubert Santana, Chia Yew Kong, Ross McMahon, Lynsey Devlin, Walaiphorn Woraharn, Noori Maka, Timothy Mitchell, Sean O'Cathail, Jean Quinn, Simon WF Milling, Philip Dunne, Colin W. Steele, Joanne Edwards, Campbell S. Roxburgh. Dynamic molecular and immune characterization for personalized treatment in locally advanced rectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 7368.

Among colorectal cancers (CRC), DNA mismatch repair-deficient (dMMR) tumors exhibit high mutational burden, increased immunogenicity, and better responsiveness to immune checkpoint blockade, when compared to MMR-proficient (pMMR) tumors. As a result, conventional T cells have been extensively studied in this disease, while the function of gδ T cells is understudied. Here, we report on a novel dMMR CRC mouse model whose cancer progression relies on IL-17A-producing gδ T cells. Deletion of Mlh1 in mouse intestinal epithelial cells resulted in tumors with a long latency period and high mutational burden. Mlh1 -deficient tumors were characterized by inflammatory transcriptional signatures. IL-17A-producing gδ T cells were the most abundant T cell in these tumors, and deletion or inhibition of these cells extended survival of tumor-bearing mice, indicative of their pro-tumorigenic role. Interfering with gδ T cells or IL-17A increased CD8 T cell and reduced neutrophil infiltration into tumors. Mlh1 -deficient tumors were unresponsive to anti-PD-1 immunotherapy. Resistance to this immunotherapy was facilitated by IL-17A-producing gδ T cells. However, tumors were sensitive to COX-2 inhibition, as celecoxib extended survival and reduced tumor-associated gδ T cells. Analysis of human CRC indicated that gδ T cells are more frequent in dMMR than pMMR and exhibit activation of IL-17 pathway. Tumor-associated macrophages exhibited the highest expression of PTGS2 , mirroring observations in mice. This study demonstrates that IL-17A-producing gδ T cells under regulation of COX-2 play an important role in dMMR CRC progression, revealing potential new targets for immunotherapy-resistant dMMR tumors.

Supplementary Table 4. Summary data reclassified using a positional based system based on proximity to functional domains, which may hold value in exploring atypical BRAF variants not yet functionally assessed (Classes A, B, and C). A single instance of co-occurring class 2 and 3 BRAF is omitted from the total counts yet presented as MSS with no additional clinical data available.

Supplementary Figure 2. Scaled Schoenfeld residuals for corresponding fixed-effect models, inclusive of study as a categorical variable, relating to mixed-effects survival models in Table 2. A. Model I compares BRAF classes across all patients irrespective of additional Ras mutation; B. Model II compares Class 1 without Ras vs Class 3 with Ras mutations; C. Model III examines additional Ras mutation status within Class 3. P-values represent significance of violation of proportional hazards assumptions.

Suplementary Table 3a-b. Summarised BRAF mutation characteristics per cohort following the 1-2-3 classification system (Table 3a) and the A-B-C classification system (Table 3b). Table cells with values less than 5 have been censored to facilitate Genomics England reporting restrictions. An uncensored version of Supplementary Table 3 is available within the Genomics England 100,000 Genomes Project Research Environment upon request.

Introduction Patients undergoing polypectomy at colonoscopy remain at risk of metachronous neoplasia despite surveillance guided by histopathological features. Mutational profiling of adenomas, including canonical driver mutations in APC, KRAS, and TP53, may offer additional predictive value. This study aimed to determine whether mutational status in index adenomas was associated with metachronous lesion risk. Methods The INCISE cohort included patients aged 50 to 74 years who underwent polypectomy within the Scottish Bowel Screening Programme and subsequent surveillance colonoscopy within 6 years. Targeted next-generation sequencing was performed on formalin-fixed paraffin-embedded polyps. Driver mutation frequency, tumour mutational burden (TMB), and variant allele frequency (VAF) were analysed and correlated with histopathological features and metachronous outcomes using appropriate statistical models. Results A total of 895 adenomas from 723 patients were analysed. In conventional adenomas, as the number of high-risk histopathological features (size >=10mm, villous architecture, and high-grade dysplasia) increased there was a stepwise increase in the proportion of samples with a mutation in KRAS from 13% to 51% (padj<0.001) and TP53 from 8% to 35% (padj<0.001). However, neither mutation frequency (p=0.901), nor median tumour mutation burden (TMB) (2.27 vs 2.15 mut/Mb, p=0.242), in index adenomas was associated with the development of metachronous lesions. Conclusions While classical driver mutations reflect histopathological progression within adenomas, they do not predict metachronous lesion risk post-polypectomy. Targeted mutation profiling alone is insufficient for surveillance risk stratification, highlighting the need for integrated molecular approaches in this setting.

BACKGROUND:Response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer (LARC) is heterogeneous. The Neoadjuvant Rectal (NAR) score quantifies treatment response via tumour downstaging and nodal status and correlates with survival outcomes. This work set out to explore whether low NAR and high NAR tumours represent biologically distinct groups pre-treatment. METHODS:Transcriptome microarrays quantified RNA expression within pre-treatment biopsies from LARC patients in GSE87211 and S:CORT consortium Grampian datasets. A validation cohort from the Beatson West of Scotland Cancer Centre was analysed using multiplex immunofluorescence and H&E-derived tumour-stromal ratios. RESULTS:Seventeen MSigDB Hallmarks were enriched in high NAR tumours; 'Epithelial-Mesenchymal Transition' was notably enriched (padj < 0.001). Immune-related signatures were enriched in low NAR tumours (p < 0.25). Both cohorts showed higher Microenvironment Cell Population counter 'Cytotoxic Lymphocytes' in low NAR (p < 0.05) and more stromal cells ('Endothelial Cells', 'Fibroblasts'; p < 0.05) in high NAR tumours. CD8 densities confirmed higher immune infiltration in low NAR tumours. DISCUSSION:Pre-treatment biopsies show consistent microenvironmental differences between high and low NAR tumours. While immune and stromal features are associated with response, the observed differences are insufficient to serve as standalone predictive biomarkers.

Abstract The increasing prevalence of incidentally diagnosed pre-malignant pancreatic Intraductal Papillary Mucinous Neoplasms (IPMNs), the cost of surveillance and the low rate of progression to cancer, underscores the importance of identifying IPMNs likely to progress to cancer. Currently available imaging and endoscopic tools cannot assess the complexity of IPMNs in their evolution to cancer. Understanding the spatial and molecular heterogeneity of pancreatic pre-malignant lesions is a critical strategy to enhance biological understanding, improve early detection and inform therapeutic strategies. Spatial transcriptomics offers the ability to robustly profile the spatial molecular landscape of IPMNs; however, cell type heterogeneity poses challenges. A same slide multiomic approach, combining high-plex proteomic and transcriptomic profiling, overcomes the limitations of transcriptionally defined cell type heterogeneity while preserving the exploration of cellular functionality through transcriptomic pathway and gene module analysis. A 64 plex protein panel followed by a whole transcriptome RNA panel using CosMx SMI (Bruker) was applied to a Tissue Microarray (TMA) of 40 x 1.5mm cores from patients with pancreatic cancer originating in IPMNs from various histological subtypes. Experimental time was 9 days to first visualisation of the data. From the RNA analysis 340,069 cells were identified with a mean of 1,545 transcripts and 1,157 unique genes per cell. In the protein 416,766 cells were identified, expression analysis demonstrated a mean fluorescence intensity of 16,287. After alignment of the decoded RNA transcripts to the co-ordinates of the protein data there were 412,680 cells across 335 fields of view, enabling multiomic integration within the same tissue regions. The analysis pipeline incorporated three distinct approaches for cell typing: RNA: 33 clusters RNA & Protein: 28 clusters Protein : 44 clusters Cell type annotation of the RNA was performed using a hybrid computational-manual approach. The top 20 most highly expressed genes and top 20 most differentially expressed genes per cluster were identified and provided to a large language model (Claude, Anthropic), generating cell types at three levels of granularity. All annotations were reviewed to ensure consistency with established pancreatic cell type markers. Proteins were clustered and cell-typed using differential expression analysis. Protein-based cell typing showed enhanced alignment with multiomic cell typing compared to RNA alone, demonstrating the value of multiomic data in validating RNA-based cell type annotations with protein data. This multiomic approach provides unprecedented resolution of cellular heterogeneity in IPMNs highlighting the complexity the pre-malignant state and establishes a framework for identifying early markers of malignant progression. Citation Format: Mari-Claire McGuigan, Luke McNickle, Assya Legrini, Ghazal Latifi, Claire Kennedy-Dietrich, Hannah Morgan, Olivia McCabe, Fraser Duthie, Tengyu Zhang, Michail Doukas, Andrea Gonzelz Cisar, Yoana Doncheva, Josefina Martinez Vasquez, Yi Cui, Sanghamithra korukonda, Ashley Heck, Kim Young, Joanne Edwards, Joseph Beechem, Nigel Jamieson. First single slide spatially resolved multiomic integration of pancreatic cancer: High-plex proteomic and whole transcriptome analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2687.

Supplementary Table 5. 100,000 Genomes Project: Clonality estimates for atypical variants harbouring additional Ras pathway mutation with sufficient data available.

Foundation models for computational pathology are expected to facilitate the development of high-performing, generalisable deep learning systems. However, in addition to biologically relevant features, current foundation models also capture pre-analytic and scanner-specific variation that bias the predictions made by downstream task-specific models trained on these features. Here we show that introducing novel robustness losses during downstream model training reduces sensitivity to technical variability. A purpose-designed comprehensive experimentation setup with 27,042 whole-slide images from 6,155 patients is used to train thousands of models from the features of eight well-known foundation models for computational pathology. In addition to a substantial improvement in robustness, our approach improves classification accuracy by focusing on biologically relevant features. It mitigates robustness limitations of foundation models for computational pathology without retraining the foundation models themselves, enabling development of models that are more suitable in real-world clinical use.

Supplementary Table 8. Differential gene expression results (post-DESeq2 analysis) for BRAF, KRAS, NRAS, and the EGFR-ligands AREG and EREG across two normal-colorectum RNAseq cohorts. Sorted by adjusted p.values (Benjamini-Hochberg).

INTRODUCTION:Colorectal Cancer (CRC) is a common cause of cancer death and prognostic factors are used to determine management. Patients with advanced disease often become cachectic, losing skeletal muscle mass and density, as well as subcutaneous fat. CT can be used to assess body composition by measuring skeletal muscle area, density and subcutaneous and visceral fat. We hypothesise that evidence of sarcopenia or myosteatosis at diagnosis is associated with an increased risk of cancer recurrence. METHODS:Patients discussed at the CRC multidisciplinary team meeting between 2015 and 2021 at a single institution were collated prospectively. We retrospectively gathered patient information, tumour characteristics and initial CT staging scans. Patients with stage II-III colon cancer were included. Patients with rectal cancer and those who underwent neoadjuvant therapy were excluded. We analysed CT images at the level of the L3 transverse processes using pre-specified criteria with a machine learning model (Mosamatic) to determine body composition. Pre-defined definitions of sarcopenia and myosteatosis were taken from the literature. The primary outcome assessed was recurrence-free survival (RFS). RESULTS:615 patients with stage II-III colon cancer underwent surgery and 120 developed a recurrence. Skeletal muscle radiation attenuation was lower in patients with recurrence (34.3 HU vs 32.2 HU, p = 0.012) but there was no significant difference in Skeletal Muscle Index (44.2 cm2/m2 vs 42.2 cm2/m2, p = 0.124). 71.7% of patients who developed recurrence had myosteatosis in comparison to 54.9% in the no recurrence group (p = 0.001). At 1-year the RFS was 92.0% (95% CI, 88.7-95.4) in the normal group and 88.3% (95% CI, 84.9-91.7) for those with myosteatosis; but at 5-years the RFS was 86.2% (95% CI, 82.0-90.6) in the normal group and 74.6% (95% CI, 70.0-79.4) in the myosteatosis group (p < 0.001). Myosteatosis was an independent predictor of recurrence on multivariate analysis (HR = 1.62, p = 0.026). There was no difference in RFS looking at sarcopenia alone. At 1-year RFS was 92.0% (95% CI, 88.7-95.3) in the normal group and 88.2% (95% CI, 84.8-91.7) in the sarcopenic group; at 5-years RFS was 81.9% (95% CI, 77.3-86.8) in the normal group and 77.5% (95% CI, 73.1-82.2) in the sarcopenic group (p = 0.17). CONCLUSION:Myosteatosis was associated with increased risk of recurrence in our cohort. This could be a useful prognostic factor in colon cancer and should be validated in prospective studies.

The PIWI-interacting RNA (piRNA) biosynthesis pathway is best studied for its role in suppressing Drosophila germline transposable elements. Piwi, the founding member of the pathway, is involved in adult intestinal stem cell (ISC) homeostasis. Whether a broader role of the PIWI pathway exists in the intestine remains unknown. Here, we characterize a role of the PIWI family protein Aubergine (Aub) in ISCs. While dispensable for basal ISC self-renewal, upregulation of Aub by damage-induced reactive oxygen species drives regenerative ISC proliferation through increased protein synthesis, including translation of ISC factors Myc and Sox21a. Unexpectedly, such roles of Aub in ISCs appear uncoupled from its piRNA regulatory function. Additionally, Aub and mammalian PIWIL1 mediate tumorigenic intestinal growth in Drosophila and human organoids, respectively. Our results reveal regulated protein translation as a fundamental aspect of regenerative ISC function and discover a central role of Aub in such process.

Supplementary Table 6. Logistic regression model of atypical BRAF mutation versus V600E status with covariates MSI status, age, sex, and tumour location.

Supplementary Figure 4. Kaplan-Meier survival curves of unadjusted overall survival (OS, months) for CRCs harboring a Class 1, 2, or 3 BRAF mutation. A: Overall survival by BRAF class irrespective of additional Ras status. B: Class 1, 2, and 3 in the absence of additional Ras. C: Class 1 and 3 absent additional Ras, and class 3 with additional Ras. Survival as shown is not corrected for covariates, whereas mixed-effects models including several covariates are reported in the main text and Table 2. Class 2 samples with survival data containing additional Ras status were insufficient for this analysis, as discussed in the main text.

PURPOSE:Two dominant neoadjuvant radiotherapy (RT) regimens are in common use for locally advanced rectal cancer (LARC): long-course RT (LCRT; 25 × 1.8 Gy) with concomitant chemotherapy and short-course RT (SCRT; 5 × 5 Gy), typically followed by systemic chemotherapy. This study uses serial sampling to investigate the evolution of systemic and local immune responses to different RT regimens in LARC. EXPERIMENTAL DESIGN:We conducted a serial sampling study involving patients receiving RT for LARC, in which longitudinal blood and tumor biospecimens were collected at baseline and at 2, 6, and 12 weeks after treatment initiation. Leukocyte concentrations via full blood count and a multiplex cytokine ELISA of plasma samples measured systemic changes. Multiplex immunofluorescence (CD8 and FOXP3) and RNA sequencing of tumor biopsies were used to assess local changes. RESULTS:Circulating lymphocyte concentrations tended to increase in SCRT patients and decrease in LCRT patients between week 2 and week 6 and were significantly lower in LCRT patients compared with SCRT patients at week 6 (P < 0.0001) and week 12 (P = 0.019). Additionally, we report higher densities of CD8+ and FOXP3+ tumoral lymphocytes from SCRT patients compared with LCRT patients at week 2 (CD8 P = 0.053; FOXP3 P = 0.023) and week 6 (CD8 P = 0.035; FOXP3 P = 0.0016). CONCLUSIONS:SCRT is less lymphodepleting and induces more frequent increases in intratumoral T-cell infiltration compared with LCRT. These results are relevant to the field of radiation-immune-oncology combination studies in LARC. Furthermore, these findings may underpin early trial results, in which higher rates of response to RT-immune checkpoint inhibitor combinations have been reported with SCRT-based regimens.

Background Current British Society of Gastroenterology (BSG) guidelines misclassify metachronous lesion risk after polypectomy in approximately 40% of patients. Building on evidence that immune exclusion drives progression of adenomas to colorectal cancer, this study examined immune profiles in screen-detected adenomas as a predictive biomarker for metachronous lesion risk. Methods Patients undergoing polypectomy within the Scottish Bowel Screening Programme, with surveillance colonoscopy between 6 months and 6 years were included. Chromogenic immunohistochemistry (IHC; n=2642), 6-plex multiplex immunofluorescence (mIF; n=334), and spatially resolved 6000-plex single cell transcriptomics (n=7) were applied to adenoma microarrays. Cell density and location were measured using QuPath. Hierarchical then K-means clustering was used to define immune cell density-based clusters, which were compared to future lesion events using Kaplan-Meier curves and the log rank test. Results After adjustment for age, sex, site, size and dysplasia, adenoma CD3+ T cell density was significantly associated with future colorectal neoplasia (HR 1.43, 95% CI 1.19-1.71, p<0.001). Using mIF three immune cell density clusters were identified; 1) high T cell density, low macrophage density, 2) low T cell density, low macrophage density, and 3) high T cell, macrophage and αSMA density, with significant differences in future lesion risk (Cluster 1: 22%, Cluster 2: 41%, Cluster 3: 36%, p=0.032). Bulk RNAseq and spatial transcriptomic analysis revealed significant variation in T cell and macrophage co-location and gene expression profiles between clusters. Conclusion Adenoma immune contexture emerges as a determinant of future metachronous lesion risk, offering a novel biomarker to refine surveillance and reduce disease burden. ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement This study was funded by; Innovate UK: Awards 10054829 and105858 Guts UK: ECR2023_03 Medical Research Scotland: PhD Studentship - PhD-50246-2020 Chief Scientist Office: Early Postdoctoral Fellowship - EPD/25/14 Cancer Research UK Scotland Institute: CTRQQR-2021\100006 ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: Ethical approval for this work was obtained for data processing from NHS Greater Glasgow and Clyde and Glasgow SafeHaven(GSH/20/CO/002), and tissue analysis from the West of Scotland Research Ethics Committee (22/WS/0020). I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Anonymised clinical and pathological data are available on the secure Glasgow Safe Haven TRE platform. Access can be arranged by application to the authors and via the Glasgow SafeHaven TRE following ethical approval and the completion of mandatory information governance modules. Mutational and transcriptomic are available publicly at ArrayExpress(E-MTAB-15349, E-MTAB-15346) and IHC data access can be discussed on reasonable application to the corresponding author.

Supplementary Table 1. Details of each cohort included in the study