Tumor characteristics of dissociated response to immune checkpoint inhibition in advanced melanoma

Patients with stage IV melanoma treated in the Netherlands Cancer Institute between 2010 till May 2019 with dissociated response patterns (defined as both regressive and progressive tumor lesions present on CT evaluation scans) to ICI treatment were screened for availability of tumor material. ICI treatment consisted of either pembrolizumab (anti-PD-1 mAb) 2 mg/kg or in a fixed dose of 150–200 mg intravenously every three weeks, nivolumab (anti-PD-1 mAb) 3 mg/kg or in a fixed dose of 240 mg intravenously every three weeks, ipilimumab (anti-CTLA-4mAb) 3 mg/kg every three weeks for a maximum of four cycles, or ipilimumab 1 mg/kg plus nivolumab 1 mg/kg for four cycles followed by maintenance nivolumab 240 mg every three weeks.

Patients with dissociated response, who did not opt out for use of remaining tumor material for research purpose or had given informed consent for performing additional biopsies for research purposes, were screened for availability of tumor samples. When both a regressive baseline tumor sample (taken within 3 months before start treatment) and a tumor sample of a progressive lesion with sufficient tumor cells (at least 30% tumor cells of HE stained frozen section) were present, patients were included.

Clinical characteristics and RECIST response were collected from patient records. This study was conducted in accordance with the Declaration of Helsinki after approval by the local institutional review board.

The formalin-fixed, paraffin-embedded (FFPE) samples were stained for both PD-L1 and CD8. Immunohistochemistry of the FFPE tumor samples was performed on a BenchMark Ultra autostainer (Ventana Medical Systems). Briefly, paraffin sections were cut at 3 µm, heated at 75 °C for 28 min and deparaffinized in the instrument with EZ prep solution (Ventana Medical Systems). Heat-induced antigen retrieval was carried out using Cell Conditioning 1 (CC1, Ventana Medical Systems) for 32 min at 95 °C (CD8) or 48 min at 95 °C (PD-L1). CD8 was detected using clone C8/144B (1/100 dilution, 32 min at 37 °C, Agilent/DAKO) and PD-L1 was detected using clone 22C3 (1/40 dilution, 1 h at RT, Agilent/DAKO). Bound antibody was detected using the OptiView DAB Detection Kit (Ventana Medical Systems). Slides were counterstained with Hematoxylin and Bluing Reagent (Ventana Medical Systems). A Pannoramic® 1000 scanner from 3DHISTECH was used to scan the slides at a 40 × magnification. The stained FFPE slides were scored by a blinded pathologist using Slidescore (www.​slidescore.​com). Of each biopsy, five representative areas of 0.2mm² were selected to assess the number of CD8⁺ cells/mm².

RNA and DNA were simultaneously isolated from FFPE sections (10 µm) with the AllPrep DNA/RNA FFPE kit (Qiagen, 80,234), according to manufacturers’ protocol, using the QIAcube. Non-tumor DNA to determine mutation load, was isolated from PBMCs using the AllPrep DNA/RNA FF kit (Qiagen, 80,224), and when no PBMCs were available, from whole blood samples, using the MagNa Pure Compact Nucleic Acid Isolation Kit.

Transcriptome and whole-exome sequencing were performed by CeGaT GmbH (Tübingen, Germany). Transcriptome libraries were generated using the KAPA RNA HyperPrep with RiboErase (HMR) & SMART-Seq stranded total RNA (Takara). Exome libraries were generated using the Twist Human Core Exome Plus (Twist Bioscience). These libraries were sequenced with 2 × 100 base pair reads on a NovaSeq 600 system according to manufacturer’s protocols, with a sequence quality value of > 93% for transcriptome and > 90% for exome libraries.

Data were analyzed in the CeGaT analysis pipeline. Briefly, demultiplexing of the sequencing reads was performed with Illumina bcl2fastq (2.20) and adapters trimmed with Skewer (v0.2.2) [15]. The quality of FASTQ files was analyzed with FastQC (v0.11.5-cegat) [16].

FASTQ files with DNA sequencing data were aligned to the human reference genome (GRCh38) using Burrows-Wheeler Aligner [17], duplicate reads were marked by Picard MarkDuplicates. Using GATK BaseRecalibrator base quality scores were recalibrated and single nucleotide variants were called using GATK MuTect2 [18].

To evaluate correct normal-tumor pairing and any underlying contamination in the tumor samples, GATK calculate contamination and BAMix were used. The tumor mutational burden was calculated by summarizing the total number of non-synonymous, somatic mutations per sample with minimal variant allele frequency of 0.05 (5%). FATHMM prediction was used to predict functional consequences of non-coding and coding sequence variation [19].

RNA sequencing data were mapped with STAR (v2.7.3a) [20] to human reference genome using default settings. The read counts were computed with HTseq-count (v0.12.4) [21] and were analyzed with DESeq2 (v1.38.2) [22]. For the downstream analyses, data were analyzed using R (v4.2.2). Centering of normalized gene expression was performed by subtracting the row means and scaling by dividing the columns by the standard deviation. The Danaher immune cell [23], interferon-gamma (IFNγ) [24], micro-environment cell population (MCP counter) [25] gene expression signatures were analyzed and expressed in normalized z-scores.

Descriptive statistics were performed on patient and tumor characteristics using IBM SPSS Statistics, version 27. Overviews of RNA gene expressions levels were visualized with the tidyverse and reshape2 library in R.

Ten patients were identified with available tumor samples at at least two time-points, with sufficient tumor material to send for sequencing. Of these patients, one patient had metachronous regressive and progressive tumor lesions of the same site (patient 1), two patients had metachronous regressive and novel tumor lesions at another site (patients 2 and 3), and three patients had metachronous regressive and progressive tumor lesions at different sites (patients 4–6) (Fig. 1). In addition, four patients had acquired resistant lesions without a comparative tumor lesion, as due to removal of the other tumor lesion for diagnostic purposes there was no known response of that specific lesion (patients 7–10).

These ten patients were in majority male (80%), had cutaneous melanoma (90%) and were treatment-naïve (80%) (Table 1). Most patients had either BRAFV600E/K (40%) or NRAS (30%) mutation-positive melanoma, and half of the patients received pembrolizumab. All seven patients of whom performance status was recorded at baseline, had a WHO performance status of 0. Details of treatment per patient are depicted in Suppl. Table 1.

In five out of six patients with paired tumor lesions, RNA sequencing data were available, while DNA sequencing data were available only in three patients, both due to insufficient material.

In patient 1 (metachronous progressive and regressive lesion), a higher number of CD8⁺ T cells was counted in the progressive lesion (+528 CD8⁺ cells/mm²). This was in contrast to both the Danaher and MCP counter immune cell subsets, which demonstrated lower T cell gene expression in the progressive lesion, ranging from −2.83 to −0.97 (Suppl. Fig. 1). In line with the progression, a lower IFNγ score (the average of the genes of the IFNγ signature, −1.937) was detected in the progressive lesion, while the PD-L1 expression on tumor cells was higher in the progressive lesion (10–50% vs. 1–10%) (Table 2).

Unlike for patient 1, both patients 2 and 3 had higher IFNγ scores (+ 0.488 and + 1.933) in their novel progressive versus the regressive lesion. Again, the CD8 T cell infiltration was similar or higher in the progressive lesion (1742 and 1629/mm², and 225 and 521/mm² in the regressive and progressive lesion, respectively), and this time in line with the Danaher and MCP counter immune cell subsets count (Suppl. Figs. 2 and 3). PD-L1 expression was similar between the regressive versus progressive lesion in both patients, but was high in one while low in the other patient (Table 2).

For the latter group of patients with progressive lesions at a different site (patients 4–6), data are incomplete. In general, again a higher CD8 infiltration and higher IFNγ signature levels in the progressive lesion was observed (Table 2), while the Danaher and MCP counter immune cell subsets were inconclusive (Suppl. Figures 4 and 5).

Due to lack of sufficient material, DNA sequencing analyses were only available for patient 2, 4 and 6. Patient 2 with a novel tumor lesion had only 2% difference in composition of mutational profile between lesions (Fig. 2), while TMB level (the total number of non-synonymous somatic mutations) was lower (-35%) in the novel lesion (Table 2). Patients 4 and 6, both with progressive lesions at a different site, had slight differences (14 and 4%, respectively) (Fig. 1), while TMB levels were evidently higher in the novel lesions (+ 540% and + 419%, respectively) (Table 2).

In addition, focused DNA mutation analyses were performed in both lesions (Suppl. Table 1). Only in patient 2 a pathogenic mutation (according to FATHMM prediction) was found in the regressive tumor lesion in the ATM gene (c.8494C > T) (Table 3). Although the progressive lesion of patient 4 had many mutations in the genes used for the focused DNA analyses, including different mutations in the HLA-A and HLA-C genes, the majority of these mutations was considered neutral according to FATHMM prediction (Table 3). No mutations were found in both lesions of patient 6 through focused DNA mutation analyses.

Additional analyses were available for four patients with progressive tumor lesions after initial response (the responding tumor lesions of these patients were not available for analysis). High CD8 infiltration (range 762–1161 CD8⁺ cells/mm²) and relatively high IFNγ signature levels (range −0.177 to 1.108) were observed compared to both the regressive and progressive lesions of the previously described patients (Table 2).

DNA sequencing analyses were available of three patients. TMB levels were variable between these patients (range 327–1221) (Table 2). Two non-sense (stop gained) mutations in the B2M (c.240G > A) and PTEN (c.821G > A, pathogenic according to FATHMM prediction) genes of patients 8 were observed (Table 3). Patient 7 had one pathologic missense mutations in POLE (c.4514C > T), while patient 10 had only missense mutations. In the lesions of patient 9, no mutations were found through the focused analyses.