Immunotherapy has revolutionized survival outcomes for many patients diagnosed with cancer. However, biomarkers that can reliably distinguish treatment responders from nonresponders, predict potential life-threatening and life-changing drug-induced toxicities, or rationalize treatment choices are still lacking. In response to this unmet clinical need, we introduce Multiomic ANalysis of Immunotherapy Features Evidencing Success and Toxicity, a tumor type-agnostic platform to provide deep profiling of patients receiving immunotherapy that will enable integrative identification of biomarkers and discovery of novel targets using artificial intelligence and machine learning.

Flow cytometry is a well-established method to analyze cell populations using antibody-based fluorescent detection of protein biomarkers. In this study, we demonstrate the ability to generate intact single cells and perform flow cytometry analysis with two types of formalin fixed tissue: FFPE curls and Representative Samples (RS). RS are homogenized, well-mixed tissue samples from formalin fixed tumors dissected from leftover surgical material. We demonstrate biomarker expression results which correlate with IHC scores. This new method for biomarker quantification may be considered alongside other methods (e.g. digital pathology). Intact single cells were dissociated from RS and FFPE curls using a non-enzymatic, mechanical dissociation method. Cells were stained in suspension for Cytokeratin 8&18 (CK8&18), Ki67, Her2, and DNA content was assessed via DAPI staining and analyzed by flow cytometry. Samples were analyzed on a BD FACSMelody or BD LSR II flow cytometer, and analysis was performed using FCS Express 7 software. Immunohistochemistry (IHC) was performed on the Benchmark ULTRA to compare Ki67 expression (n=78) and Her2 expression (n=16) to the flow analysis. Ki67 expression by IHC was assessed using the international Ki67 working group scoring methods to generate a positive percentage. Her2 expression by IHC was assessed by a pathologist and assigned a score ranging 0 to 3+. Formalin fixed tissue such as RS and FFPE curls can be mechanically dissociated into single cells with intact surface biomarkers. These single cells can be stained in suspension, analyzed via flow cytometry and generate correlating data with both weighted IHC-based scores and clinical IHC scores. The flow analysis of Her2 positive percentage correlates with the IHC scores showing an increasing trend and significant difference between scores for both FFPE and RS. Ki67 expression varied by tumor region using IHC analysis, however by flow cytometry showed a strong correlation with a weighted average across multiregional quantification of Ki67 expression. We demonstrate that millions of intact single cells can be generated from RS and FFPE curls for breast tissue, using non-enzymatic mechanical dissociation methods. Staining in suspension and flow cytometry analysis can be performed in a day for rapid biomarker quantification. Flow cytometry has the potential to analyze FFPE samples using workflows and technologies that have existed in the hematopathology space for decades. Samantha M. Hill, Hannah L. Veloz, Brian Hanley, Tracy Davis, Lisa L. Gallegos, Harold Sasano, Samra Turajlic, Nelson R. Alexander. Optimizing fixed flow cytometry for breast cancer biomarker expression in representative samples and FFPE curls [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 670.

ccRCC is marked by niches of immune evasion in late-stage disease, that are associated with resistance to immune checkpoint inhibitors (CPI). Intratumoral microbes are emerging as key modulators of the tumor immune microenvironment and may therefore play a yet unrecognised role in ccRCC disease progression and therapy resistance, marking them as potential biomarkers, and novel therapeutic targets. We extracted bacterial reads from three cohorts of treatment-naïve primary tumors and one cohort of pre- and post-CPI treated tumors: Genomics England Renal (GEL) (636 patients, WGS), TRACERx Renal (154 patients, 629 samples, WGS); TCGA Renal (494 patients, RNA) & ADAPTeR (15 patients, 56 samples, RNA). Bespoke denoising, and decontamination were applied. Live presence of intratumoral bacteria was confirmed through culture and RNAscope. Genera associated with survival were identified using ElasticNet feature selection. We observed significant heterogeneity in bacterial abundance within and across tumors, driven by differences in Cutibacterium abundance. Cutibacterium makes up ∼90% of bacteria in each tumor sample and is enriched in tumors compared to adjacent normal tissue (p=7.1e-10). 9 of 11 colonies grown from two positive tumors were genotyped as Cutibacterium acnes, confirming its live presence and relative abundance in ccRCC tumors. Cutibacterium is higher in late-stage tumors (III&IV) than early-stage tumors (I&II)(p=7.8e-3). Its abundance also separates patients by progression free survival (PFS) and overall survival (OS) (p=9.9e-3, p=0.016) and is higher in CPI non-responders than responders (p=0.028). Association with survival is confirmed in the TCGA cohort independently of stage (PFS = 0.031, OS = 7.3e-4). While other genera associate with survival in either TRACERx or TCGA, only Cutibacterium is prognostic in both. Enrichment analyses of bulk RNA (TRACERx & TCGA), show upregulation of leukocyte taxis and innate immune response with Cutibacterium abundance. To further probe this interaction with the immune microenvironment we are performing single cell spatial transcriptomics, and in vitro co-cultures. Results of these ongoing experiments will be presented at the conference. Cutibacterium either creates or exploits a disease- and CPI-resistance-promoting environment in ccRCC tumors and can therefore act as a prognostic and predictive biomarker. The genus is known for causing chronic inflammation in sebaceous skin follicles, induces Nf-kB signalling, and M2-macrophage differentiation in vitro, and associates with myeloid response in ccRCC tumors. Together this supports a compelling hypothesis that Cutibacterium reduces survival in ccRCC by creating immunosuppressive niches, thereby fostering disease progression and CPI-resistance. Alice C. Martin, Anne-Laure Cattin, Irene Lobon, Zayd Tippu, Fiona Byrne, Charlotte Spencer, Clara Becker, Martha Zepeda Rivera, Krupa Thakkar, Hongui Cha, Angel Fernandez-Sanroman, Annika Fendler, Taja Barber, Leo Bickley, Daqi Deng, Scott Shepherd, Parise Lockwood, Maximiliano Gutierrez, Susan Bullman, Kevin Litchfield, Samra Turajlic. Intratumoral bacteria predict survival in clear cell renal cell carcinoma (ccRCC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2207.

572 Background: Anti-vascular endothelial growth factor (VEGF) tyrosine kinase inhibitors and checkpoint inhibitors (CPI) a standard-of-care treatment for clear cell renal cell carcinoma (ccRCC). We investigated the biology underpinning benefit of anti-VEGFR TKI in the phase II A-PREDICT trial (NCT01693822), evaluating pre- and post-treatment, fresh multiregion tumour biopsies in patients with metastatic ccRCC treated with first-line axitinib. Methods: We analysed 123 tumour samples from 52 patients, 28 with paired pre/post-treatment samples. Post-treatment samples included week-9, nephrectomy, and on-progression timepoints. ‘Responders’ had progression-free survival (PFS) ≥6 months (n=35), ‘non-responders’ with PFS <6 months (n=17). We applied a custom Nanostring panel for gene expression analysis and multiplex immunofluorescence (mIF) for orthogonal validation. Wilcoxin test was used to analyze paired observations. Results: At baseline, angiogenesis scores were similar between responders and non-responders (p=0.22). Post-treatment, the angiogenesis, vascular sprouting, and endothelial cell proliferation signature scores were significantly decreased (p=0.023, 0.0034, & 0.0082, respectively) in all patients, suggesting suppression of angiogenesis and neovascularisation irrespective of clinical outcomes. mIF in 3 patients (with PFS of 3, 5.6, & 100 months) confirms widespread intratumoral vessel depletion. Immune deconvolution analysis shows total levels of T cells and CD8 + T cells were similar pre- and post-treatment, suggesting axitinib did not enhance immune cell trafficking. Rather, axitinib promoted increased levels of exhausted CD8 + T cells post-treatment (p=0.01). M2 tumour-associated macrophages increased post-treatment in responders (p=0.033) but not in non-responders (p=0.44). A minority of patients had durable (>2 years) responses to axitinib (n=7/65, 6 with tissue for analysis). In these patients, we found higher levels of pre-treatment intratumoral cytotoxic immune cells (p=0.041) and NK cells (p=0.015) compared to patients with primary resistant disease. Conclusions: Axitinib suppressed angiogenesis and neovascularisation leading to intratumoral vessel depletion, and therapy response associates with features of an immunosuppressive TME. Baseline endogenous immune priming appears critical for durable response to anti-VEGF therapy. These data are relevant to understanding the clinical efficacy of combined anti-VEGF and CPI regimens. Clinical trial information: NCT01693822 .

A genomic and transcriptomic analysis of patients with clear-cell renal cell carcinoma reveals clinically relevant patterns of nongenetic evolution, including progressive immune dysfunction and cGAS–STING suppression.

Self-supervised foundation models for digital pathology encode small patches from H\&E whole slide images into latent representations used for downstream tasks. However, the invariance of these representations to patch rotation remains unexplored. This study investigates the rotational invariance of latent representations across twelve foundation models by quantifying the alignment between non-rotated and rotated patches using mutual $k$-nearest neighbours and cosine distance. Models that incorporated rotation augmentation during self-supervised training exhibited significantly greater invariance to rotations. We hypothesise that the absence of rotational inductive bias in the transformer architecture necessitates rotation augmentation during training to achieve learned invariance. Code: https://github.com/MatousE/rot-invariance-analysis.