Automated prediction of response to neoadjuvant chemotherapy from digitized H&E slides of pre-treatment core needle biopsies in INFORM (TBCRC 031) patients with low stromal TILs.

Abstract

611 Background: We previously performed anautomated analysis of whole slide images (WSI) of H&E-stained pre-treatment core needle biopsies (CNBs) from patients (pts) enrolled in the INFORM phase II trial of neoadjuvant cisplatin vs doxorubicin-cyclophosphamide in HER2-negative germline BRCA carriers. That analysis demonstrated that a digital biomarker of complex immune response (CmbI) which combines immune heterogeneity index (IHI), proliferative, and cell cycle G1/S deregulation signatures, was significantly predictive of response (RCB 0,1) to neoadjuvant chemotherapy (NAC) in all pts, in sub-cohorts including TNBC, and in both therapy arms. Further, a lower IHI, indicating less heterogeneity of stromal tumor infiltrating lymphocytes (sTIL), was predictive of a better response to NAC (RCB 0,1), whereas a higher IHI, indicating greater heterogeneity of sTIL was associated with a worse response (RCB 2,3). The predictive performance of IHI alone was modest compared to CmbI but superior to sTIL. High sTIL is associated with favorable prognosis for NAC, especially in TNBC. However, the impact of heterogeneity of immune cell distribution on NAC response, particularly in those with low sTIL is unknown. The current study evaluated if IHI could augment sTIL assessment by identifying NAC responders in patients with tumors demonstrating low sTIL. Methods: CNBs scanned at 40x on a Hamamatsu Nanozoomer scanner were evaluated using the 4D QPOR platform to generate IHI as a continuous index. Among 88 QPOR analyzable pts, 85 had sTIL scores available from prior visual pathologic review. Tumors with low sTIL ( < 30%, a previously documented clinically significant cut-off),were further stratified into low vs high IHI using median IHI for the population as the cut-off. We then determined the relationship between IHI and likelihood of response to NAC (RCB 0,1) in the overall cohort, and in the TNBC and ER low ( < 10%) sub cohorts. The analysis was also performed using the median sTIL cut-off < 20% previously used in INFORM. Results: Low IHI was significantly predictive of NAC response (RCB 0,1) in low sTIL pts in the overall cohort (N = 64, OR = 4.75; 95% CI1.50, 16.21), p = 0.005, PPV = 70%) and in the ER low sub cohort (N = 46, OR = 4.04;95% CI 1.04-17.38, p = 0.04, PPV = 72%). IHI was modestly predictive in the low sTIL pts in TNBC sub cohort (N = 41, OR = 4.28 (0.99, 20.77), p = 0.03, PPV = 71%) but not predictive in high sTIL pts ( < 1/3 pts in each subgroup). For sTIL < 20%, IHI had stronger predictive ability in all pts (N = 64, OR = 7.63 (1.83, 40.09), p = 0.0016, PPV = 80%). Conclusions: Heterogeneity of immune cell distribution determined by computational analysis of WSI of pre-treatment CNB of patients with germline BRCA mutations and HER2-negative cancers in the INFORM trial improves response prediction to NAC in patients with low baseline sTIL as determined by visual analysis.