Monoallelic expression in melanoma

Fifteen melanoma cell lines and tumor infiltrating lymphocytes were derived from metastatic melanoma lesions from patients treated at the Surgery Branch, National Cancer Institute (NCI), Bethesda, MD, USA kindly donated by Dr. Steven A. Rosenberg. Identity of melanoma cell lines and parental tissues was carried out by HLA phenotyping as previously described [20].

All patients signed an informed consent approved by the Institutional Review Board of the National Cancer Institute.

Genomic DNA was isolated from tumour infiltrating lymphocyte samples using the QIAmp DNA Mini Kit (Qiagen, Germantown, MD) according to the manufacturers guidelines. DNA quality and quantity was assessed using Nanodrop (ThermoScientific, Pittsburgh, PA). Samples were genotyped using Illumina Human 1 M Beadchip. Samples have been processed according to the Illumina procedure for the Infinium II assay. Data was extracted by the Illumina BeadArray reader. Samples and markers with call rate below 95% were excluded from analysis and a GenCall cutoff score of 0.15 was used for all Infinium II products. The SNP data was filtered by GenTrain score with a hard cut off of < 0.5 being excluded from further analysis. Heterozygous SNP calls for each sample were exported and subsequently compared to the genotype derived from the RNA-seq data.

Total RNA was isolated from the cell lines using miRNeasy minikit (Qiagen) according to the manufacturer’s protocol. RNA quality and quantity was estimated using Nanodrop (Thermo Scientific) and Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Enrichment in mRNA molecules was obtained using oligo (dT) magnetic beads (Ambion^® Poly (A)Purist™ MAG Kit). Subsequently, mRNA was fragmented into shorter fragments (approximately 200 bp), cDNA was synthesized by random hexamer primers (Illumina TruSeq Stranded mRNA Library Prep Kit). The double-stranded cDNA was purified by QiaQuick PCR extraction kit (Qiagen) and went through an end repair process with the addition of a single ‘A’ base, and then ligation of the adapters. These products were then purified by agarose gel electrophoresis and enriched with PCR to create the final cDNA library. The library products were sequenced and 300 bp sequences were generated via the GAIIx Illumina sequencing platform. Raw reads were imported on a commercial data analysis platform CLC Genomics Workbench (CLC bio, MA, USA). Quality control checks on raw sequence data from each sample were performed using the QC analysis application tool. Adapter trimming was done to remove ligated adapter from 3′ end of the sequenced reads with only one mismatch allowed. After reads have been processed to meet a quality standard, they were aligned to the Human reference genome UCSC-Hg19, using the ultra high-throughput short aligner provided by CLC bio software. Next, after a Transcript Discovery analysis we obtained a transcript annotation file with an estimation of the relative abundances of each transcript by counting the number of reads that mapped to the genomic location of that transcript. Transcription level assessment has been obtained by the number of fragments per kilobase of transcript per million fragments mapped (RPKM).

Genotyping data of the 15 cell lines has been filtered to include only SNPs with heterozygous calls. Genotypes for the corresponding positions in the RNA-seq data were calculated, the genotypes were assigned using GATK. In order to increase the number of informative SNPs (with heterozygous calls at the DNA level) we have performed imputation analysis using IMPUTE v2 [21]. The following data analysis steps were automated using custom Perl scripts. We have ascertained the B-allele frequency in the RNA sequencing data where the SNP was heterozygous in the microarray analysis and mapped to the RNA sequencing data with > 5 reads. SNPs were mapped to UCSC gene CDS. SNPs outside CDS were excluded as potentially not representative of the gene. The number of SNPs monoallelically expressed was calculated, across each sample and each gene using user defined threshold for number of samples and the number of SNPs per gene. Functional network analysis was performed using the Ingenuity Pathway Analysis (IPA) tools 3.0 which transforms large data sets into a group of relevant networks containing direct and indirect relationships between genes based on known interactions in the literature (http://​www.​ingenuity.​com, Ingenuity System Inc., Redwood City, CA, USA).

MAE frequency was calculated as the sum of the homozygous SNPs at the RNAseq data divided by the sum of all informative SNPs (with heterozygous calls at the DNA level and mapped at the RNA-seq data). To increase the number of informative SNPs we performed imputation analysis (Table 1). MAE frequency ranged between 0.17 and 0.77 across the 15 pairs of melanoma cell lines—TILs (non-imputed average 0.58 ± 0.23, imputed average 0.60 ± 0.21), suggesting that MAE occurs in metastatic melanoma with an overall high frequency as compared to the MAE frequency reported in normal conditions. In order to investigate whether DAE does also occur in melanoma, B-allele frequency of the SNPs with heterozygous calls at the DNA level was calculated for the RNA-seq data and plotted against the occurrence count (Fig. 1a). B-allele frequency showed a normal distribution when ranging between 0.1 and 0.9. However, the occurrence count peaked when B-allele frequency was 0 or 1, suggesting that monoallelic expression in melanoma seems to be complete rather than skewed.

The MAE frequency for imputed data was overall similar to the one calculated for the non-imputed data, suggesting that the MAE frequency was not biased by the imputation analysis. For all the further analyses we used the imputed data as this was considered to be more informative.

With the hypothesis that activation of oncogenic pathways might modify the overall transcriptional program and therefore, the MAE pattern, we assessed whether MAE rate was different between NRAS and BRAF mutated cell lines. The MAE rate in NRAS mutated cell lines was slightly higher than in the BRAF mutated cell lines (NRAS average 0.58, standard deviation 0.18, BRAF average 0.27, standard deviation 0.18, Fig. 1b), however the difference was not significant (t test p = 0.7). The only BRAF and NRAS wild type cell line available in this study had a MAE rate of 0.26.

To assess whether MAE was functionally enriched in given chromosomes, we have calculated the MAE frequency by cell lines and according chromosomes across all the 15 pairs of samples (Table 2, Additional file 1: Figure S1).

Interestingly, when we looked at the MAE frequency by chromosome across cell lines, we found chromosome 9 to be frequently MAE expressed (11 out 15 cell lines, Fisher’s test of chromosome 9 vs all, two-tailed p = 0.0001).

As some of the samples analyzed showed high differences between homozygous and heterozygous calls, we sought to determine whether loci under MAE were located within regions of copy number changes (see Additional file 2: Figure S2). We found that not in all instances MAE was explainable by CN, suggesting that additional mechanisms may influence MAE occurrence in melanoma. As an example in Mel2744, MAE was completely explained by CN in chromosomes 1, 9, 17 and 18 (Additional file 1: Figure S1).

With the assumption that MAE plays a role in melanoma development, we sought to determine whether the genes under MAE were related to tumor pathways. We have first filtered genes with at least two informative SNPs showing MAE, and then selected the genes that were shared by at least 8 cell lines, this allowed us to identify 515 genes. When imported on IPA, these genes pointed ‘Cancer’ as the top disease with a p-value range of 2.36E−03–9.06E−24 and 431 molecules represented in this category, suggesting that the genes under MAE in melanoma are indeed related to cancer.