Genomic Correlates of Outcome in Tumor-Infiltrating Lymphocyte Therapy for Metastatic Melanoma

Abstract

Purpose: Adoptive cell therapy (ACT) of tumor-infiltrating lymphocytes (TIL) historically yields a 40%-50% response rate in metastatic melanoma. However, the determinants of outcome are largely unknown.

Experimental design: We investigated tumor-based genomic correlates of overall survival (OS), progression-free survival (PFS), and response to therapy by interrogating tumor samples initially collected to generate TIL infusion products.

Results: Whole-exome sequencing (WES) data from 64 samples indicated a positive correlation between neoantigen load and OS, but not PFS or response to therapy. RNA sequencing analysis of 34 samples showed that expression of PDE1C, RTKN2, and NGFR was enriched in responders who had improved PFS and OS. In contrast, the expression of ELFN1 was enriched in patients with unfavorable response, poor PFS and OS, whereas enhanced methylation of ELFN1 was observed in patients with favorable outcomes. Expression of ELFN1, NGFR, and PDE1C was mainly found in cancer-associated fibroblasts and endothelial cells in tumor tissues across different cancer types in publicly available single-cell RNA sequencing datasets, suggesting a role for elements of the tumor microenvironment in defining the outcome of TIL therapy.

Conclusions: Our findings suggest that transcriptional features of melanomas correlate with outcomes after TIL therapy and may provide candidates to guide patient selection.

Figure 1.. Resected tumor workflow on the metastatic melanoma ACT of TIL study.

Schematic depicting the distribution of the metastatic melanoma tumor tissue after surgical resection at MDACC (n=64) and MCC (n=9). Numbers in circles indicate initial sample prioritization and distribution for 1) generation of the clinical TIL product for patient treatment and 2) formalin-fixed, paraffin embedded (FFPE) samples. The 73 patients with generated clinical TIL product, thus had 73 corresponding FFPE samples. DNA was extracted from all 73 FFPE samples, 64 samples passed quality control (QC) for mutation profiling via WES and 30 samples were also used for DNA methylation profiling (7 samples passed DNA methylation QC but did not pass WES QC). Extracted RNA from 34 patients passed QC for transcriptomic profiling for RNA-Seq, with remaining RNA or subsequently extracted RNA utilized for ddPCR from 16 samples.

Figure 2.. Mutation analysis of pre-treatment tumors from TIL treated patients identifies enriched recurrently mutated genes in this cohort.

A, Mutation load in the TCGA cutaneous melanoma cohort and our combined TIL cohort from MDACC and MCC, consisting of the pre-treatment tumor of the 64 patients treated with TIL. Red line identifies median number of mutations. B, The upper panel shows the mutation signature of SNVs in the TIL patient cohort (n=64), while the lower panel depicts the signature of UV-induced mutations as described by COSMIC Signature 7. The two signatures have a cosine similarity of 0.99. C The MutSig2 plot of the recurrent mutations (left side) found within the sample set (q< 0.1 is significantly enriched) and the frequency (in percentage) of the mutations (right side). The type of mutation, response to therapy, and cancer center of origin for each sample are denoted below the main plot.

Figure 3.. Non-silent mutation burden and predicted HLA class I neoantigen load associate with overall survival.

The three graphs represent pre-treatment tumor samples from 64 TIL treated patients. A, Graph showing non-silent mutation load and OS of patients color-coded by their response to TIL therapy. The p-value was generated using the Cox proportional hazards model. B, Predicted neoantigen load versus OS color-coded by patient response status. The dotted lines in both plots indicates ongoing survival. Statistical differences were determined by Cox proportional hazards model.

Figure 4.. Identification of gene expression profile associated with outcome to TIL therapy.

Volcano plots of genes enriched by response A, PFS B, and OS C. Genes associated with response to therapy, longer OS, and longer PFS are listed on the right of each plot, with the inverse indicated on the left. The most significantly enriched genes are labeled on the plots. Plots were generated using log (PFS or OS +1) as a continuous variable in DESeq2, where PFS and OS are in months. D, Euler diagrams represent the overlapping genes enriched in response, OS, or PFS. The upper diagram represents genes associated with improved OS, improved PFS, and response, with NGFR, PDE1C, and RTKN2 enriched in these criteria. The lower diagram identifies ELFN1 enriched in patients with poor OS, poor PFS, and lack of response to therapy. Graphs E and F depict 5 samples from responder patients and 11 samples from non-responder patients. E, Scatterplot depicts correlation between ELFN1 expression by RNA-Seq (transcripts per million; tpm) and the mean ddPCR ratio of concentration of target gene to reference gene (ELFN1/HPRT1) per sample, plotted with the line of best fit. Statistical differences were calculated by Spearman correlation and the r² value is represented on the graph. Samples are color coded by response to TIL therapy. F, Graph depicts the mean ddPCR ratio of concentration of target gene to reference gene (ELFN1/HPRT1) per sample stratified by response to TIL therapy. Statistical differences were determined by Mann-Whitney test. Error bars represents mean ± standard deviation.

Figure 5.. ELFN1 DNA methylation status in pre-treatment tissue samples correlates with response.

A, Global mean methylation of ELFN1 in 30 tumors harvested for TIL propagation, comparing responders (blue, n = 20, CR+PR) to non-responders (red, n = 10, PD) patients. Statistical differences were determined using a Wilcoxon test. Error bars represents mean ± standard deviation. B, Unsupervised hierarchical clustering of the 30 tumors shown in (a) by methylation site within ELFN1. Statistical differences were determined using Fisher’s exact test. Location of the methylation site (M sites) is presented within the gene (5’ to 3’, top of the heat map) with the CpG islands in maroon. Each patient’s clinical response and PFS is annotated on the right. The location of ELFN1 on chromosome 7 is also shown above.