Digital Immunophenotyping of Lung Atypical Carcinoids and Large Cell Neuroendocrine Carcinomas Identifies Three Subtypes with Specific Tumor-Immune Microenvironment Features

Background: The WHO Classification defines Atypical Carcinoids (AC) and Large Cell Neuroendocrine Carcinomas (LCNEC) as intermediate and high grade lung neuroendocrine neoplasms, respectively. Treatment strategies based on this classification are still highly controversial and not fully effective. Several studies have suggested a significant role of tumor microenvironment (TME) including tumor-infiltrating lymphocytes (TiLs) in cancer treatment. However, the TME and TiLs composition of ACs and LCNECs remains largely unknown. Design: TME markers of T cell (CD8,CD3), immune suppression (PD-L1,PD-1), HLA (HLA-DR,HLA-I), fibroblast α-SMA and vascular CD31 were analyzed by immunohistochemistry (IHC) on a histological revaluated series of 56 ACs and 104 LCNECs and evaluated in tumor center and in peritumoral stroma by two expert pathologists. Tumor (tTiLs) and stromal (sTiLs) Tils from digital CD3 and CD8 IHC whole-slide images were quantified in entire slide (ES) and separately at the invasive margin (IM) and in tumor center (CT) using machine-learning based image analysis. Random forest was used to identify robust ACs and LCNECs Tils parameters and an unsupervised algorithm was used to identify clusters of patients. Relationships between clusters with clinicopathological and TME features were analyzed. Results: Overall, LCNECs showed more T cell, PD-L1, PD-1 and HLA-I in the stroma compared to ACs (p<0.001). ACs had more CD31 and α-SMA than LCNECs (p<0.001). No AC showed PD-L1 tumor expression. Digital pathology showed that LCNECs had more tTiLs and sTiLs in ES, IM and CT than ACs (p<0.001). Moderate concordance with manual count were obtained. Significant positive associations were found between tTiLs and sTiLs with immune checkpoint PD-1 and PD-L1. For both AC and LCNEC, IM showed significant higher densities of tTiLs and sTiLs compared to CT (p<0.001). TiLs robust parameters identified three clusters of patients from unsupervised analysis: cluster 1 enriched of ACs (39 AC and 12 LCNEC), low TiLs and better prognosis; cluster 2 heterogeneous in terms of histology (15 ACs and 50 LCNEC), with intermediate TiLs and moderate prognosis; cluster 3 mainly represented by LCNECs (2 AC and 42 LCNEC), with high TiLs and worst prognosis. TME markers specifically characterized each cluster. Tumor PD-L1 expression was strongly associated with cluster 3. Conclusions: ACs and LCNECs belong to three different and clinically relevant clusters driven by specific immune and tumor microenvironments features.