571P - Artificial intelligence-powered analysis of tumor lymphocytes infiltration: A translational analysis of AtezoTRIBE trial

Background

Adding atezolizumab (atezo) to FOLFOXIRI/bev prolongs PFS of mCRC patients (pts) in AtezoTRIBE trial. We explored the predictive role of immune phenotypes (IPs) and Inflamed Score (IS), as assessed by means of Lunit SCOPE IO, an artificial intelligence (AI)-powered whole-slide images (WSI) analyzer of tumor infiltrating lymphocytes (TILs).

Methods

In AtezoTRIBE study mCRC pts were randomized 1:2 to upfront FOLFOXIRI/bev -/+ atezo. AI-powered spatial and quantitative analyses of TILs were conducted on pre-treatment hematoxylin & eosin-stained WSI from 173 (79%) enrolled pts. Three IPs (inflamed, excluded, desert) were defined according to TILs density in tumor epithelium and stroma, and IS was calculated as the proportion of grids classified as inflamed within a WSI. IPs and IS were correlated with immune-related biomarkers (MMR, TMB, Immunoscore IC, Immunoscore, TPS PD-L1, TILs by optical microscope) and clinical outcome.

Results

AI-powered analyses were successfully performed in 154 (84%) cases. 67 (44%) and 87 (56%) cases were classified as excluded and desert, respectively, while no inflamed cases (IS ≥ 33.3%) were detected. Pts with excluded and desert tumors had similar baseline features, except for a higher percentage of pts with Immunoscore IC-high (p= .006), Immunoscore-high (p= .008), and TILs-high (p= .006) tumors in the excluded vs desert group. A similar PFS benefit from adding atezo was reported in the excluded (HR 0.70, 95% CI 0.40-1.23) and desert (HR 0.73, 95% CI 0.45-1.17) groups (p_int=0.817). Tumors with high (≥3^rd quartile, 1.01%) IS (n=39) were more frequently dMMR (p= .031) and TMB-high (p= .020) than those with low (<3^rd quartile) IS (n=115). The PFS benefit from adding atezo was homogeneous between tumors with high IS (HR 0.80, 95% CI 0.38-1.67) and low IS (HR 0.69, 95% CI 0.44-1.07) (p_int=0.791). Similar results were found in the pMMR cohort.

Conclusions

IPs and IS by Lunit SCOPE IO provide a characterization of tumor microenvironment, being associated with Immunoscore, Immunoscore IC and TILs. Further development of AI-powered TILs analyses taking into account their densities may help identifying biomarkers of immunogenicity in mCRC.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

GONO Foundation.

Disclosure

S. Lonardi: Financial Interests, Personal, Advisory Board: Amgen, Merck Serono, Lilly, Servier, AstraZeneca, MSD, Incyte, Daiichi Sankyo, Bristol Myers Squibb; Financial Interests, Personal, Invited Speaker: Pierre Fabre, GSK, Roche, Astellas; Financial Interests, Institutional, Coordinating PI: Amgen, Merck Serono, Bayer, Roche, Lilly, AstraZeneca, Bristol Myers Squibb. M. Fassan: Financial Interests, Institutional, Funding: QED, Macrophage Pharma, Astellas, Diaceutics; Financial Interests, Personal, Speaker, Consultant, Advisor: Roche, Astellas, AstraZeneca, Incyte, Bristol Myers Squibb, Merck Serono, Pierre Fabre, GSK, Novartis, Amgen; Financial Interests, Personal, Advisory Board: Amgen, Astellas, Roche, Merck Serono, GSK, Novartis, Janssen. C. Cremolini: Financial Interests, Personal, Advisory Board: Roche, MSD, Amgen, Pierre Fabre, Nordic Pharma; Financial Interests, Personal, Invited Speaker: Bayer, Servier, Merck Serono; Financial Interests, Institutional, Coordinating PI: Roche, Bayer, Servier, Merck; Financial Interests, Institutional, Local PI: Seagen, Hutchinson. All other authors have declared no conflicts of interest.