168P - Biosimulation coupled with personalized tumor microenvironment (TME) modeling predicts response to immunotherapy treatment in NSCLC patients

Background

Although immunotherapy (IO) has emerged as a standard treatment option in patients with non-small cell lung cancer (NSCLC), current biomarkers are imperfect and could be improved upon. Understanding variations in both tumor microenvironment (TME) and tumor genomic composition may help. To explore this, we have combined output from a mechanistic computational biology model that integrates a patient’s tumor-based genomic profile to predict patient-specific IO signaling pathway dysregulation and drug response, with predicted TME cell composition using deconvoluted bulk transcriptomic data.

Methods

We developed and validated an algorithm for deconvolution of cell proportion and cell specific gene expression from bulk transcriptome data. Reference matrix was generated from single-cell NSCLC data (n= 823,622 cells). The locked algorithm was applied to bulk transcript data from 59 NSCLC patients (PMID:37024582) treated with immunotherapy to generate TME cell proportions (TCP). Computational biosimulation (Cellworks) was performed using WES data and used to generate an immunotherapy efficacy score (ES), simulating composite cell growth in response to disease and IO. Cox proportional hazards models were generated to test the hypothesis that TCP and ES are predictive of progression free survival (PFS).

Results

Bulk patient transcriptomic data was deconvoluted into 19 predicted TME cell types. Patient-specific TCPs correlated well with clinical response. Notably, a high proportion (>5%) of an M1-like macrophage subtype (high CXCL9/10, STAT1; low SPP1, CD206) was associated exclusively with patients showing some clinical benefit (n=15). Both ES and TCP significantly predicted PFS in a Cox proportional hazards model (p < 0.001), and combined use of ES and TCP further improved model performance (C-Index= 0.792, LRT=49.49, p = 0.002) compared to ES (C-Index= 0.61, LRT=8.32, p = 0.004) or TCP (C-Index= 0.75, LRT=35.57, p = 0.008) alone.

Conclusions

Integrating genomic and transcriptomic data to capture TME signaling offers a powerful tool for predicting IO response in NSCLC. These results warrant further prospective evaluation in larger cohorts.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

Cellworks Group, Inc.

Funding

Cellworks Group, Inc.

Disclosure

H. Grover, A. Kumar, M. KS, P. Kumari, M. Patil, S. C, A. Chakraborty, J. Chauhan, R. Mandal, R. Sethia, S. Kulkarni, A. Tyagi, S. Kapoor, M.P. Castro, J. Wingrove: Financial Interests, Personal, Full or part-time Employment: Cellworks.