Biomarkers of success of anti-PD-(L)1 immunotherapy for non-small cell lung cancer derived from RNA- and whole-exome sequencing: results of a prospective observational study on a cohort of 85 patients

Abstract

Background: Immune checkpoint inhibitors (ICIs) treatment have shown high efficacy for about 15 cancer types. However, this therapy is only effective in 20-30% of cancer patients. Thus, the precise biomarkers of ICI response are an urgent need.

Methods: We conducted a prospective observational study of the prognostic potential ofseveral existing and putative biomarkers of response to immunotherapy in acohort of 85 patients with lung cancer (LC) receiving PD-1 or PD-L1 targeted ICIs. Tumor biosamples were obtained prior to ICI treatment and profiled by whole exome and RNA sequencing. The entire 403 putative biomarkers were screened, including tumor mutation burden (TMB) and number of cancer neoantigens, 131 specific HLA alleles, homozygous state of 11 HLA alleles and their superfamilies; four gene mutation biomarkers, expression of 45 immune checkpoint genes and closely related genes, and three previously published diagnostic gene signatures; for the first time, activation levels of 188 molecular pathways containing immune checkpoint genes and activation levels of 19 pathways algorithmically generated using a human interactome model centered around immune checkpoint genes. Treatment outcomes and/or progression-free survival (PFS) times were available for 61 of 85 patients with LC, including 24 patients with adenocarcinoma and 27 patients with squamous cell LC, whose samples were further analyzed. For the rest 24 patients, both treatment outcomes and PFS data could not be collected. Of these, 54 patients were treated with PD1-specific and 7 patients with PD-L1-specific ICIs. We evaluated the potential of biomarkers based on PFS and RECIST treatment response data.

Results: In our sample, 45 biomarkers were statistically significantly associated with PFS and 44 with response to treatment, of which eight were shared. Five of these (CD3G and NCAM1 gene expression levels, and levels of activation of Adrenergic signaling in cardiomyocytes, Growth hormone signaling, and Endothelin molecular pathways) were used in our signature that showed an AUC of 0.73 and HR of 0.27 (p=0.00034) on the experimental dataset. This signature was also reliable (AUC 0.76, 0.87) for the independent publicly available LC datasets GSE207422, GSE126044 annotated with ICI response data and demonstrated same survival trends on independent dataset GSE135222 annotated with PFS data. In both experimental and one independent datasets annotated with samples' histotypes, the signature worked better for lung adenocarcinoma than for squamous cell LC.

Conclusion: The high reliability of our signature to predict response and PFS after ICI treatment was demonstrated using experimental and 3 independent datasets. Additionally, annotated molecular profiles obtained in this study were made publicly accessible.

Keywords: RNA sequencing; gene expression biomarker; immune checkpoint therapy; ipilimumab; nivolumab; non-small cell lung cancer; pembrolizumab; personalized medicine.

Figure 1

Characteristics of biosamples of the experimental NSCLC cohort. (A) Technical characteristics of biosamples under analysis. Sample IDs are given on the left. Color markers indicate availability of the RNAseq, WES profiles; tumor Histotypes established by pathologists; primary or metastatic origin of tumor samples; treatment with PD-1 or PD-L1 ICI therapeutics; treatment with CTLA4 ICI therapeutics; Availability of PFS data and availability of RECIST treatment response data. More detailed characteristics are given in Supplementary Table 1 . (B) Principal component analysis (PCA) plot of RNA sequencing profiles in the experimental NSCLC cohort in comparison with the healthy lung tissue controls. Dot color indicates tissue type.

Figure 2

Outline of the putative PD(L)1 ICI NSCLC response biomarkers investigated in this study. The putative biomarkers used were assessed using RNA sequencing (left) or WES data (right).

Figure 3

Correlation heat map representing statistically significant Pearson correlation coefficients obtained for the expression of immune checkpoint genes in the experimental NSCLC cohort. Color reflects correlation coefficients (see color scale) and dot size corresponds to the -log10(p-value) of the Pearson correlation test.

Figure 4

Intersection of PFS biomarkers and RECIST response biomarkers identified in this study for NSCLC patients in relation to responsiveness to PD(L)1 ICI therapeutics. (A) Intersection diagram of PFS and RECIST response biomarkers. (B) Statistics for the intersected biomarkers identified including PFS-based HR values with p-values, AUC values for differentiating treatment responders and non-responders (R vs NR), and Mann-Whitney test p-values for differentiating treatment responders and non-responders.

Figure 5

Progression-free survival analysis of tumor mutation burden (TMB) as the biomarker in the experimental NSCLC sampling. The Kaplan-Meier plots are given for the whole NSCLC dataset, and separately for the lung adenocarcinoma and squamous cell lung carcinoma sub-datasets. TMB-high status was defined as TMB greater than 10 per megabase.

Figure 6

Activation profile of the Growth hormone signaling pathway (gene expression via SRF, ELK1, STAT5B, CEBPD, STAT1, STAT3) in the experimental NSCLC groups of RECIST responders (A) and non-responders (B) to PD(L)1 ICI immunotherapy. Color reflects the logarithm of the case-to-normal ratio (CNR) of the pathway nodes, color scale is given (green – upregulated, red – downregulated, white – intact). Arrows show molecular interactions within a pathway: green stands for activation, red for inhibition. PAL values were calculated for the averaged biosamples in the treatment responder and non-responder groups.

Figure 7

Assessment of biomarker potential of NCAM1 gene expression level as the PD(L)1 ICI response biomarker in the experimental NSCLC sampling. (A–C) Progression-free survival analysis in the experimental NSCLC sampling. The Kaplan-Meier plots are given for the whole NSCLC (A) dataset, and separately for the lung adenocarcinoma (B) and squamous cell lung carcinoma (C) sub-datasets. (D–F), ROC AUC analysis of RECIST response status assessed for the whole NSCLC cohort (D) dataset, and separately for the lung adenocarcinoma (E) and squamous cell lung carcinoma (F) sub-datasets. “R” means treatment responder, “NR” – non-responder.

Figure 8

Activation profile of the Interleukin 10 signaling pathway in the experimental NSCLC groups of RECIST responders (A) and non-responders (B) to PD(L)1 ICI immunotherapy. Color reflects the logarithm of the case-to-normal ratio (CNR) of the pathway nodes, color scale is given (green – upregulated, red – downregulated, white – intact). Arrows show molecular interactions within a pathway: green stands for activation, red for inhibition. PAL values were calculated for the averaged biosamples in the treatment responder and non-responder groups.

Figure 9

Assessment of biomarker potential of the Oncobox gene signature as the PD(L)1 ICI response biomarker in the experimental NSCLC sampling. (A) heatmap outlining normalized values of signature components, signature risk score, response statuses, and PFS times. (B, D) F, ROC AUC analysis of RECIST response status assessed for the whole NSCLC cohort (B) and separately for the lung adenocarcinoma (D) and squamous cell lung carcinoma (F) sub-cohorts. (C, E, G) Progression-free survival analysis in the experimental NSCLC sampling. The Kaplan-Meier plots are given for the whole NSCLC (C) dataset, and separately for the lung adenocarcinoma (E) and squamous cell lung carcinoma (G) sub-datasets. “R” means treatment responder, “NR” – non-responder. (H) Box-and-whisker plot for signature values in a whole experimental dataset and in sub-datasets of lung adenocarcinoma and squamous cell carcinoma. p-values are presented for one-sided Mann-Whitney tests between responders (R) and non-responders (NR).

Figure 10

Characteristics of biosamples of the GSE207422 cohort. Sample IDs are given on the left. Color markers indicate availability of the RNAseq profiles; tumor Histotypes established by pathologists; treatment with PD-1 ICI immunotherapeutics; availability of treatment response data (R means treatment responder and NR – non-responder).

Figure 11

Assessment of biomarker potential of the Oncobox gene signature as the PD-1 ICI response biomarker in the literature GSE207422, GSE126044 and GSE135222 NSCLC cohort. (A) Heatmap outlining normalized values of signature components, signature risk score, and response statuses. (B–D) ROC AUC analysis of PD-1 immunotherapy response status assessed for the whole GSE207422 NSCLC cohort (B), and separately for the lung adenocarcinoma (C) and squamous cell lung carcinoma (D) sub-cohorts. “R” means treatment responder, “NR” – non-responder. (E) Box-and-whisker plot for signature values in a whole test dataset and in sub-datasets of lung adenocarcinoma and squamous cell carcinoma of GSE207422 dataset. p-values are presented for one-sided Mann-Whitney tests between responders (R) and non-responders (NR) in GSE207422 dataset. (F) ROC AUC analysis of PD-(L)1 immunotherapy response status assessed for the GSE126044 NSCLC cohort. (G) Progression-free survival analysis in the GSE135222 dataset. The Kaplan-Meier plots are given for the whole NSCLC.