Prediction of clinical outcomes of advanced cutaneous squamous cell carcinoma to PD-1 inhibition directly from histopathology slides using inferred transcriptomics

Introduction: Metastatic or locally advanced cutaneous squamous cell carcinoma (cSCC) that is not amenable to local therapy is treated with programmed death-1 (PD-1) inhibitors. Although response rates are relatively high, there are no validated predictive biomarkers to guide treatment. As a result, a subset of patients – particularly frail and elderly patients which can be treated with local palliative therapy – are exposed to immune-related adverse events without clinical benefit. Here, we present a retrospective evaluation of ENLIGHT-DP, a novel digital pathology biomarker which predicts response to PD-1 inhibition in advanced cSCC directly from histopathology slides using inferred transcriptomics.

Methods: We scanned high-resolution hematoxylin and eosin (H&E) slides from pretreatment tumor samples of 38 patients with advanced cSCC treated with cemiplimab and retrospectively generated an individualized prediction score using the ENLIGHT-DP pipeline in a two-step process: (i) inference of mRNA expression profiles directly from H&E slides using the DeepPT deep-learning algorithm, and (ii) integration of these inferred transcriptomes into ENLIGHT, a transcriptomics-based precision oncology platform that predicts therapeutic response. We unblinded clinical outcomes and assessed the predictive performance of ENLIGHT-DP.

Results: The cohort consisted primarily of frail, elderly patients (median age 81 years), with 18 patients having an ECOG performance status ≥2. Using a binary threshold for classification, ENLIGHT-DP significantly predicted response to cemiplimab, demonstrating a positive predictive value of 84.2% and an odds ratio (OR) of 4.8 (95% CI: 1.1-22.1), along with significant stratification for progression-free survival with HR = 0.22 (95% CI: 0.05-0.95, p = 0.023) and outperforming performance status, age and site of cancer. Comparative analyses of inferred immune-related transcriptomic signatures revealed significant differences between cSCC and head and neck squamous cell carcinoma, underscoring distinct tumor immunobiology.

Conclusion: This exploratory study introduces ENLIGHT-DP as a digital pathology biomarker shown to significantly predict clinical outcomes in patients with cSCC treated with PD-1 inhibitors.