An immunomodulatory signature of responsiveness to immune checkpoint blockade therapy

Dear Editor, We identified an immunomodulatory signature that was significantly associated with clinical improvement of immune checkpoint blockade (ICB) therapy response and better prognosis in urothelial carcinoma1 and melanoma.2 Our finding is helpful for identifying cancer patients who may benefit from ICB therapy. ICB therapy is an effective treatment regimen for cancer patients. Tumor mutation burden (TMB),3 expression of PD-1, PD-L1,4 and CTLA-4, are known factors associatedwith ICB therapy response. However, our understanding of biomarkers for response to ICB therapy remains incomplete. This necessitates further investigation on identification of reliable biomarkers for ICB therapy response. A flowchart depicting procedures of the study was shown in Figure 1A, which consisted of feature representation learning of gene expression from single cells and association of expression signatures with ICB therapy response in two clinical trials. We collected gene expression data from 71,494 single cells encompassing tumor cells, stroma cells, dendritic cells, CD4+, and CD8+ T cells from seven cancer types (Table S1). A deep learning model5,6 was developed by iteratively training on gene expression data of these 71,494 single cells. Subsequently, it was applied to extract 258 expression signatures for single cells and samples from two clinical trials. We ranked the extracted expression signatures of single cells with a gene set involved in immunomodulation7 (see Supporting InformationMethods, Table S2) and examined association of the top five ranking signatures with clinical improvement of ICB therapy and prognosis. The t-SNE embedded scatter plot shown in Figure 1B depicted single cell clusters unveiled from expression signatures learned by deep learning model. The same cell types (such asCD4+ orCD8+T cell) fromdifferent data set were clustered together, whereas different cell types were separated distinctively (Figures S1A andB).Notably, a clus-


Dear Editor,
We identified an immunomodulatory signature that was significantly associated with clinical improvement of immune checkpoint blockade (ICB) therapy response and better prognosis in urothelial carcinoma 1 and melanoma. 2 Our finding is helpful for identifying cancer patients who may benefit from ICB therapy.
ICB therapy is an effective treatment regimen for cancer patients. Tumor mutation burden (TMB), 3 expression of PD-1, PD-L1, 4 and CTLA-4, are known factors associated with ICB therapy response. However, our understanding of biomarkers for response to ICB therapy remains incomplete. This necessitates further investigation on identification of reliable biomarkers for ICB therapy response.
A flowchart depicting procedures of the study was shown in Figure 1A, which consisted of feature representation learning of gene expression from single cells and association of expression signatures with ICB therapy response in two clinical trials. We collected gene expression data from 71,494 single cells encompassing tumor cells, stroma cells, dendritic cells, CD4+, and CD8+ T cells from seven cancer types (Table S1). A deep learning model 5,6 was developed by iteratively training on gene expression data of these 71,494 single cells. Subsequently, it was applied to extract 258 expression signatures for single cells and samples from two clinical trials. We ranked the extracted expression signatures of single cells with a gene set involved in immunomodulation 7 (see Supporting Information Methods, Table S2) and examined association of the top five ranking signatures with clinical improvement of ICB therapy and prognosis.
The t-SNE embedded scatter plot shown in Figure 1B Figure S1C).
We categorized ICB therapy response into clinical and no clinical improvement (see Supporting Information Methods). The baseline characteristics are shown in Table  S3. The urothelial carcinoma clinical trial 1 consisted of 298 urothelial cancer patients all treated with PD-L1 inhibitor, 68 of which were categorized as clinical improvement and the rest 230 were categorized as no clinical improvement group. The melanoma clinical trial 2 consisted of 119 melanoma cancer patients, 47 of which achieved clinical improvement and the rest did not. In melanoma clinical trial, 72 patients received PD-1 inhibitor, whereas 47 patients received CTLA4 in conjunction with PD-1 inhibitor. The other available molecular markers related to ICB therapy response or prognosis such as TMB, the degree of cytotoxic lymphocytes infiltration (CTL), 8 expression levels of PD-1, PD-L1, and CTLA4, and clinical features are summarized in Table S2. TMB data were available for 78.5% (234/298) and 100% (119/119) of patients in these two clinical trials.
Among these top five ranking signature, one of the expression signature scores was significantly different between clinical improvement group and no clinical improvement group for the urothelial carcinoma  Figure 2D). We further examined the significance of identified immunomodulatory signature with respect to anti-PD1 and anti-PD-L1 inhibitors. In urothelial carcinoma clinical trial, all patients were treated with anti-PD-L1 inhibitors. In melanoma clinical trial, the identified immunomodulatory signature remained significant in patients treated with anti-PD-1 treatment (OR = 1.32, 95% CI = 1.05-1.67, P = 0.02; Figure S2A), while it also trended toward better association in patients treated with anti-PD-1 in conjunction with anti-CTLA4 ( Figure S2B).
Gene set enrichment analysis revealed that immunerelated signaling circuits 9 (Table S4) were overrepresented in patients with activated immunomodulatory signature, while signatures featuring cancer cell proliferation and invasiveness such as epithelial-to-mesenchymal transition and angiogenesis were underrepresented ( Figure S3). In addition, tumor immune microenvironment signatures obtained from CIBERSORT algorithm 10 showed that there were significant differences among infiltration of CD8+ T cells, activated CD4+ memory T cells, M1 macrophages, and dendritic cells in patients stratified by this signature ( Figure S4).
In summary, we reported that an immunomodulatory signature dissected from large-scale single cell expression data was significantly associated with clinical F I G U R E 3 Association between the identified immunomodulatory signature and prognosis in two ICB therapy clinical trials. Kaplan-Meier survival curves (A and B) and forest plot representation of multivariate Cox regression model (C and D) depicting the association between the identified immunomodulatory signature and prognosis in two ICB therapy clinical trials. The confounding factors include sex, TCGA molecular subtype, or melanoma stage and TMB. The age information was not available. Adjusted P-value < .05 was considered to be significant while .05 ≤ adjusted P-value ≤ .1 was considered to be marginally significant improvement of ICB therapy and better prognosis in the two independent ICB therapy clinical trials. However, further investigation in prospective randomized clinical trial is warranted.

A C K N O W L E D G M E N T S
We want to thank Prof. Wei Zhang from Wake Forest Cancer Center for his constructive comments and editing the manuscript and Liangtao Zheng from Peking University for his comments on data interpretation as well as researchers for their generosity to made their data publicly available.

E T H I C S A P P R O VA L A N D C O N S E N T T O PA R T I C I PAT E
This study was approved by the institutional review board (IRB) of Tianjin Cancer Hospital. Informed consent was exempted by the IRB given that data were obtained from public, open access database.

D ATA AVA I L A B I L I T Y S TAT E M E N T
Data are available in open access database. All data relevant to the study are included in the article or uploaded as supplementary information. All data generated or analyzed during this study are included in this manuscript.