Yan, Dongyao

Abstract Purpose: Digital proteomic profiling was performed to identify spatial context in relationship to patient response and survival after anti–PD-1 therapy in metastatic colorectal cancer. Experimental Design: Primary colorectal cancers with deficient mismatch repair from patients treated with anti–PD-1 antibodies were analyzed (N = 30) using digital spatial profiling (GeoMx nCounter). At the invasive margin, 71 proteins were profiled in 10 regions of interest/slide that were segmented into 3 compartments labeled with pan-cytokeratin (epithelia), CD45 (stromal cells), and SYTO13 (nuclei). In an independent cohort (n = 13), digital spatial profiling data and single-cell transcriptomic data were analyzed. Differential protein abundance, after Benjamini–Hochberg correction, was examined by clinical response and progression-free survival (PFS) using multivariable Cox regression. Results: Protein abundance varied significantly between epithelial and stromal compartments. Nonresponders to anti–PD-1 showed higher fibronectin and smooth muscle actin abundance in the epithelial compartment that was associated with significantly shorter PFS (adjusted HR: 6.49 and 4.52, respectively; P < 0.05). In CD45+ stroma, increased expression of proteins related to T cells (CD3 and CD4), NK cells (CD56), antigen presentation (CD40), immune activation (CD27, ICOS), and apoptosis (GZMA) were found in responders (vs nonresponders) to anti–PD-1; each marker was significantly associated with longer patient PFS (0.02 < adjusted HR < 0.17; P < 0.05). In a separate cohort, consistent results by compartment were found for fibronectin and CD56. Gene expression data revealed that fibronectin and smooth muscle actin were primarily derived from cancer-associated fibroblasts. Conclusions: Spatially resolved protein profiles within microenvironments of deficient mismatch repair colorectal cancers can influence patient response and survival after anti–PD-1, highlighting their potential clinical significance.

Abstract Deep learning may detect biologically important signals embedded in tumor morphologic features that confer distinct prognoses. Tumor morphologic features were quantified to enhance patient risk stratification within DNA mismatch repair (MMR) groups using deep learning. Using a quantitative segmentation algorithm (QuantCRC) that identifies 15 distinct morphologic features, we analyzed 402 resected stage III colon carcinomas [191 deficient (d)-MMR; 189 proficient (p)-MMR] from participants in a phase III trial of FOLFOX-based adjuvant chemotherapy. Results were validated in an independent cohort (176 d-MMR; 1,094 p-MMR). Association of morphologic features with clinicopathologic variables, MMR, KRAS, BRAFV600E, and time-to-recurrence (TTR) was determined. Multivariable Cox proportional hazards models were developed to predict TTR. Tumor morphologic features differed significantly by MMR status. Cancers with p-MMR had more immature desmoplastic stroma. Tumors with d-MMR had increased inflammatory stroma, epithelial tumor-infiltrating lymphocytes (TIL), high-grade histology, mucin, and signet ring cells. Stromal subtype did not differ by BRAFV600E or KRAS status. In p-MMR tumors, multivariable analysis identified tumor-stroma ratio (TSR) as the strongest feature associated with TTR [HRadj 2.02; 95% confidence interval (CI), 1.14–3.57; P = 0.018; 3-year recurrence: 40.2% vs. 20.4%; Q1 vs. Q2–4]. Among d-MMR tumors, extent of inflammatory stroma (continuous HRadj 0.98; 95% CI, 0.96–0.99; P = 0.028; 3-year recurrence: 13.3% vs. 33.4%, Q4 vs. Q1) and N stage were the most robust prognostically. Association of TSR with TTR was independently validated. In conclusion, QuantCRC can quantify morphologic differences within MMR groups in routine tumor sections to determine their relative contributions to patient prognosis, and may elucidate relevant pathophysiologic mechanisms driving prognosis. Significance: A deep learning algorithm can quantify tumor morphologic features that may reflect underlying mechanisms driving prognosis within MMR groups. TSR was the most robust morphologic feature associated with TTR in p-MMR colon cancers. Extent of inflammatory stroma and N stage were the strongest prognostic features in d-MMR tumors. TIL density was not independently prognostic in either MMR group.

Abstract Purpose: Targeting the programmed cell death protein 1 (PD-1)/programmed cell death ligand 1 (PD-L1) interaction has led to durable responses in fewer than half of patients with mismatch repair–deficient (MMR-d) advanced colorectal cancers. Immune contexture, including spatial distribution of immune cells in the tumor microenvironment (TME), may predict immunotherapy outcome. Experimental Design: Immune contexture and spatial distribution, including cell-to-cell distance measurements, were analyzed by multiplex immunofluorescence (mIF) in primary colorectal cancers with d-MMR (N = 33) from patients treated with anti–PD-1 antibodies. By digital image analysis, density, ratio, intensity, and spatial distribution of PD-L1, PD-1, CD8, CD3, CD68, LAG3, TGFβR2, MHC-I, CD14, B2M, and pan-cytokeratin were computed. Feature selection was performed by regularized Cox regression with LASSO, and a proportional hazards model was fitted to predict progression-free survival (PFS). Results: For predicting survival among patients with MMR-d advanced colorectal cancer receiving PD-1 blockade, cell-to-cell distance measurements, but not cell densities or ratios, achieved statistical significance univariately. By multivariable feature selection, only mean number of PD-1+ cells within 10 μm of a PD-L1+ cell was significantly predictive of PFS. Dichotomization of this variable revealed that those with high versus low values had significantly prolonged PFS [median not reached (>83 months) vs. 8.5 months (95% confidence interval (95% CI), 4.7–NR)] with a median PFS of 28.4 months for all patients [adjusted HR (HRadj) = 0.14; 95% CI, 0.04–0.56; P = 0.005]. Expression of PD-1 was observed on CD8+ T cells; PD-L1 on CD3+ and CD8+ T lymphocytes, macrophages (CD68+), and tumor cells. Conclusions: In d-MMR colorectal cancers, PD-1+ to PD-L1+ receptor to ligand proximity is a potential predictive biomarker for the effectiveness of PD-1 blockade.

3542 Background: Solid tumors can be characterized by distinct immune phenotypes based on the level of T cell infiltration in the tumor microenvironment (TME) which may provide prognostic information and also inform strategies to restore the anti-tumor immune response. Immune phenotypes were determined using an artificial intelligence (AI) algorithm in digitized whole-slide images (WSI) of tumors with deficient (d) mismatch repair (MMR) vs proficient (p) MMR. Methods: Stage III colon carcinomas (N = 401; 393 met QC) from participants in a phase III trial of FOLFOX-based adjuvant chemotherapy were analyzed including all available tumors with dMMR (n = 196) and a randomly selected cohort of pMMR tumors (n = 195). Using an AI algorithm (Lunit SCOPE) previously trained on solid tumors, digitized tumors were categorized into three immune phenotypes based on epithelial and stromal TIL data-driven cutpoints in dMMR and pMMR (inflamed: TIL high in epithelium (> upper 25%); excluded: TIL high in stroma (>lower 25%), low in epithelium; and desert: TIL low in epithelium and stroma). Phenotypes were then examined in relationship to disease-free survival (DFS) using Kaplan-Meier methodology. Results: Of the 3 immune phenotypes in dMMR tumors, 25% were inflamed, 50% were immune-excluded, and 25% were immune-desert. Based on univariate results, multivariable modeling was performed incorporating immune phenotype in addition to age, histological grade, N stage, T stage, performance status, treatment arm, BRAF and KRAS status and identified immune phenotypes and N stage to be significantly associated with DFS. Among dMMR tumors, immune desert phenotype was associated with the poorest DFS (Desert: HR adj 1.95, 95%CI (1.01, 3.80); Excluded: HR adj 0.89, 95%CI (0.47, 1.69); Inflamed: ref; p:0.01). When compared to immune desert tumors, immune-excluded tumors had significantly better DFS (HR adj 0.49, 95%CI (0.27, 0.88), p:0.01). At a median follow-up of 60 months, 3-year DFS of patients with dMMR was 71.2% with Immune inflamed, 78.5% with Excluded, and 54.5% with Desert phenotypes. In contrast to dMMR, univariate analyses of data-driven cutpoints in pMMR tumors was not prognostic. Using a different data-driven cutpoint (15%), the revised immune phenotypes remained non-significant for DFS within pMMR tumors. Conclusions: Distinct AI-derived immune phenotypes in the TME were identified that were significantly prognostic in patients with dMMR, but not pMMR colon cancers. A data-driven immune-desert phenotype was identified in dMMR tumors that was associated with significantly poorer survival. Further investigation of the potential predictive utility of these phenotypes for immunotherapy are planned. Support: https://acknowledgments.alliancefound.org . ClinicalTrials.gov Identifier: NCT00079274 .

3527 Background: Advances in deep learning may improve the ability to evaluate and quantify pathological features in solid tumors to improve prediction of patient disease-free survival (DFS). We used an artificial intelligence (AI) algorithm to quantify morphological features in the tumor microenvironment of stage III colon cancers. Methods: We analyzed digitized stage III colon carcinomas (N = 402; 382 met QC) from participants in a phase III trial of FOLFOX-based adjuvant chemotherapy that included all available tumors with dMMR and a randomly selected cohort of pMMR tumors (median follow-up 60 months) [NCCTG N0147; Alliance for Clinical Trials in Oncology]. Fifteen morphological features were extracted using the QuantCRC algorithm that segments images into carcinoma (low-grade, high-grade, signet ring cell), stroma (immature, mature, and inflammatory), mucin, tumor/stroma ratio, tumor budding/poorly differentiated clusters (TB/PDC), necrosis, smooth muscle, fat, and tumor-infiltrating lymphocytes (TILs) and tumor area (mucin, epithelium and TB/PDC). Analysis of AI-derived features with clinical variables, molecular alterations ( BRAF/KRAS), and DFS were examined using Kaplan-Meier methodology and multivariable Cox regression. Results: Among the 15 AI-derived morphological features, the following differed significantly between dMMR (n = 191) and pMMR (n = 189) whereby dMMR tumors had lower mature stroma, but higher inflammatory stroma and tumor, higher tumor grade and increased mucin, TB/PDCs, signet ring cells, and TILs (all p< 0.05). Among dMMR tumors, multivariable analysis revealed that tumor area [mucin, epithelium and TB/PDC] (HR adj 0.45, 95%CI (0.24, 0.84), p:0.01, [Q2/3 vs Q1]) and N stage were the only significant and independent prognostic variables. Among pMMR tumors, multivariable analysis identified that tumor budding/poorly differentiated clusters (TB/PDCs) (HR adj 0.20, 95%CI (0.06, 0.61), p:0.01, [Q1 vs Q4]) was the strongest prognostic variable and the only morphological feature that was significantly associated with DFS along with age, N stage and T stage. Conclusions: Using AI, we can extract and quantify distinct morphological features in tumor sections that differ between dMMR and pMMR and multivariately, can significantly and robustly enhance prognostication within each MMR group. Among pMMR tumors, tumor budding/poorly differentiated clusters was the strongest predictor of DFS. Support: NIH U10CA180821, U10CA180882, U24CA196171, R01 CA210509 (to FAS). Study NCCTG N0147 received funds from Sanofi https://acknowledgments.alliancefound.org . ClinicalTrials.gov Identifier: NCT00079274 Key words Artificial Intelligence; Tumor Microenvironment; Colonic Neoplasms; Disease-free survival.

Abstract Background: Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of patients with deficient DNA mismatch repair (d-MMR) metastatic colorectal cancer (CRC), yet less than half of patients receive clinical benefit. We hypothesized that immune contexture including spatial distribution of cells expressing immune molecules in the tumor microenvironment may predict immunotherapy outcome. Methods: Primary CRC tissues from consecutive patients with d-MMR metastatic CRC (N=33) were treated with anti-PD-1 antibodies at Mayo Clinic Comprehensive Cancer Center (2015-2018). Tumors were stained for PD-L1, PD-1, CD8, CD3, CD68, LAG3, TGFβR2, MHC-I, CD14, B2M, DAPI and pan-cytokeratin in 5 compartments (overall tumor, tumor epithelia, tumor stroma, peritumor inside vs outside) by multiplex immunofluorescence with digital image analysis. Features computed within each image analysis region included positive cell density, fraction of positive cell types, intensity of positive cells, and spatial distribution between distinct cell types (distances measured using image analysis software). Prior to model fitting, feature selection was performed using regularized Cox regression with LASSO. Regularization parameter was chosen based on 5-fold cross validation. A Cox proportional hazards model was fitted to predict patient progression-free survival (PFS). Results: Among patients with d-MMR CRCs, 16/33 (48.4%) were female, 10 received first-line ICI therapy, and 23 had received &gt; 1 prior chemotherapy regimen. Median age was 61.6 years (IQR of 49.4, 73.7). Eight patients (24.2%) harbored BRAFV600E, 9 (27.2%) had mutant KRAS, and median PFS was 22.2 months (95% CI: 11.8, NR) with 20 events. PD-L1 was expressed in tumor cells, CD68+ macrophages, and CD3+ T lymphocytes. PD-1 expression on CD8+ T lymphocytes was also observed. By univariate analysis, only cell-cell distance readouts achieved statistical significance. Multivariable feature selection identified the mean number of PD-1+ cells within 10 microns of a PD-L1+ cell in the overall tumor as the strongest predictor of anti-PD-1 efficacy, and was significantly associated with PFS (HR=0.87, 95% CI: 0.79-0.95, p&lt; 0.001). In contrast, the ratio of PD-1+/PD-L1+ cells was not predictive of treatment efficacy (p= 0.47), thereby underscoring the importance of spatial distribution for PFS prediction. Conclusion: The mean number of PD-1+ cells in proximity to PD-L1+ cells in tumors was a predictive biomarker of anti-PD-1 efficacy. Confirmatory studies are warranted. Citation Format: Bahar Saberzadeh-Ardestani, Rondell P. Graham, Qian Shi, Eze Ahanonu, Sara McMahon, Crystal Williams, Antony Hubbard, Wenjun Zhang, Andrea Muranyi, Dongyao Yan, Kandavel Shanmugam, Frank A. Sinicrope. Prediction of anti-PD-1 efficacy based on immune marker densities and their spatial distribution in colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6648.