Clinical application of genomic profiling to find druggable targets for adolescent and young adult (AYA) cancer patients with metastasis

This study was approved by the Institutional Review Board (IRB) of Seoul National University Hospital (1206-086-414). We obtained written informed consent from the patients who participated to this study. All participants in this study gave us written informed consent for publication of their details. Written informed consent for publication of their clinical details and/or clinical images was obtained from the patients. A copy of the consent form is available for review by the Editor of this journal.

Samples from seven different tumors, prostate cancer, olfactory neuroblastoma, head and neck squamous cell carcinoma (HNSCC), urachal carcinoma, germ cell tumor, lung cancer, and liposarcoma, were prospectively obtained in three different forms (fresh-frozen tissue, formalin-fixed paraffin-embedded (FFPE), and pleurisy). The samples were analyzed using three different genomics platforms (whole-exome sequencing (WES), whole transcriptome sequencing (WTS), and OncoScan™) as the tumor sources permitted. We first intended to analyze samples from ten patients, but three samples were excluded because the amount of provided tumor sample was insufficient (AYA03) or sufficient DNA/RNA for a genome-scale analysis was not obtained (AYA05, and 08). For sample AYA04 (HNSCC), the HPV infection status was identified by IHC staining (data not shown).

A minimum of 3 μg of genomic DNA was randomly fragmented by Covaris, and the sizes of the library fragments were mainly distributed between 250 and 300 bp. adapters were then ligated to both ends of the fragments. Extracted DNA was amplified by ligation-mediated PCR (LM-PCR) and then purified and hybridized to the SureSelect XT Human All Exon v4 + UTR 71 Mb (Agilent Technologies, Santa Clara, CA, USA) for enrichment according to the manufacturer’s recommended protocol. After loading each captured library on the Hiseq2000 platform (Illumina, San Diego, CA, USA), we performed high-throughput sequencing for each captured library. Raw image files were processed by Illumina CASAVA v1.8.2 for base-calling with default parameters, and the sequences from each individual were generated as 101-bp pair-end reads.

WES data were processed using a series of steps. We aligned the sequenced files (Fastq file) to the reference genome (human reference genome g1k v37) using the Burrows-Wheeler Aligner (BWA v0.7.5a) [11] and then sorted the output and removed PCR duplicates using PICARD v1.95 [12]. Using the typical GATK workflow (The Genome Analysis Toolkit v2.6-5), we processed the data for local indel realignment and base quality recalibration [13]. For variant calling, we used MuTect v1.1.6 for single nucleotide variants (SNVs) and Somatic Indel Detector (from GATK v2.2-8) for indels [14]. Whereas we called the SNVs with the default setting value, we altered the tumor indel fraction from 0.3 to 0.05 (T_INDEL_F <0.05) for indel calling after considering false-negatives. The called variants interpreted as somatic mutations were tagged with “KEEP” or “SOMATIC” with MuTect and Somatic Indel Detector, respectively, and used for further study. To avoid false-positive indel variants, we filtered out variants with tumor alterative reads less than 6. All somatic variants were annotated by ANNOVAR [15]. The variants that passed through the steps were called ‘processed WES data’ (Additional file 1: Figure S1).

We used the 330-k OncoScan™ FFPE platform (Affymetrix, Santa Clara, CA, USA) to identify candidate CNVs (amplification/deletion and loss-of-heterozygosity (LOH)). AYA02 was excluded because the amount of DNA was insufficient for OncoScan™. A minimum of 80 ng of DNA from each sample was used for the OncoScan™ platform. The Nexus Express (Affymetrix) software was used to analyze the data and find CNVs. We filtered out CNVs with a CN ≤ 2.5, which were considered insignificant amplification, and analyzed chromosome-level CNVs and focal level CNVs.

To analyze CNVs in AYA02, we used VarScan2 as an alternative method to OncoScan™. After processing data up to the ‘Realignment/Recalibration’ step in WES processing (Additional file 1: Figure S1), we processed data based on the manufacturer’s recommendations. We generated mpileup data from recalibrated BAM files of both tumor and normal using SAMtools and used the ‘Copy caller’ module of VarScan2 to generate the relative copy number change (C), which was determined as follows: C = log ₂ ((D _T /D _N )*(I _N /I _T )), where ‘D’ stands for the average depth, ‘I’ for the number of uniquely mapped bases, ‘N’ for normal, and ‘T’ for tumor [16, 17]. After filtering out mapping quality values <15, we adjusted the relative copy number using the re-centering option in ‘Copy caller’ and segmented copy number regions based on the circular binary segmentation algorithm. After merging, the results were represented by IGV (Integrative Genomics Viewer) [18]. Because the VarScan2 results covered only exon regions, not all genome regions, we analyzed only chromosome-level CNVs that were similar to those obtained with OncoScan™ and did not analyze focal-level CNVs. To create graphs of relative copy number changes, we used data from re-centered relative copy number changes displayed on a log₂ scale.

The mutation frequency was analyzed by counting the number of variants annotated by ANNOVAR from WES data as nonsynonymous SNVs, synonymous SNVs, nonsense mutations, stop-loss mutations, splicing mutations, frameshift insertions/deletions (indels), in-frame indels, and noncoding RNA in exonic regions. These mutations had previously been described in published data from 12 major cancer studies [19]. To analyze the mutation spectrum, we used SNVs processed with MuTect in all sequenced regions not limited to coding regions.

After assigning the levels to the variants by pattern-based heuristic annotation, we investigated the biologic pathways of variants using DAVID or the literature [20]. To analyze the druggability of the variants, we concentrated mainly on level-1 (strong) variants using DGIdb [21].

We analyzed WTS data from four samples (AYA01, 02, 09, and 10) to identify cancer driving fusions using three different fusion tools, FusionMap, deFuse and ChimeraScan [22‐24]. From the results, we selected candidate cancer-driving fusions using a fusion gene list archived in COSMIC (download date: 2015-03-03).

Samples analyzed in this study were collected from AYA patients (median age = 32) who had metastatic tumors. Seven different tumor samples were obtained in three different forms (fresh-frozen tissues, FFPE and pleurisy) and analyzed using three different genomics platforms (WES, WTS and OncoScan™) (Table 1).

We processed WES data using well-known bioinformatics tools described in Supporting Fig. 1 following previously published studies [8, 10]. From the WES data of six tumors and matched normal blood, we generated total of 95 gigabases (Gb, range 12.5–20.0/sample) and 106 Gb (range 12.7–27.0/sample) of mapped sequences, respectively. The mean target coverage was 135X (range, 101X-190X) and 150X (range, 115X-205X) for tumor samples and normal blood, respectively. The coverage of mean target bases exceeded 30X for 89.8 and 92.5 % of the tumor and normal blood samples, respectively (Additional file 2: Table S1).

We analyzed the mutation frequency and mutation spectrum of six samples from processed WES data (Fig. 1). Except for AYA07, which contained a hypermutation, the coding regions (exon and splicing regions) of a total of 276 somatic mutations (median, 40; range, 26–133) were detected (Additional file 3: Table S2). Compared with the mutation frequency of 12 major cancer types obtained from published data, the somatic mutation frequency of most samples was low (median, 0.56/Mb; range, 0.37–1.87), except for AYA07 (Fig. 1a) [19]. This finding is consistent with data showing that somatic mutations are accumulated with age [1]. Although the mutation frequency of AYA07 was high (9.37/Mb), the overall mutation frequency of AYA cancers is low: a recent large-scale study of testicular germ cell tumors (TGCTs), the same tumor type as that of AYA07, demonstrated a low mutation frequency (mean 0.5/Mb) [25]. Additionally, three AYA cancers with higher mutation frequencies (AYA02, 04, and 07) harbored mutations in TP53 or in several DNA repair genes compared with other samples (mean, 3.98/Mb vs. 0.44/Mb, respectively), as shown in large studies [19].

Except for AYA07, the mutation spectrum of AYA cancers showed prominent C > T transitions, as has been demonstrated in many solid cancers (Fig. 1b) [26]. Interestingly, AYA07 showed a high proportion of C > A transversions (79.7 %). This result may be related to an over-representation of C > A transversions in the TCGT study [25] or multiple cycles of chemotherapy for the AYA07 patient, because increased C > A transversions were observed in all samples obtained from eight relapsed AML patients after chemotherapy [27].

After processing the WES data, we annotated variants using ANNOVAR as described in Additional file 1: Figure S1. All annotated variants are described in Additional file 4: Table S3. We then selected driving genetic alterations using our pattern-based annotation, because it is limited to analyzing rare types of cancers using the same statistical methods to select driving genetic alterations used in large-scale studies (Additional file 3: Figure S2) [4]. By applying our approach to TCGA AML data, we could detect all candidate genes that were previously defined using statistical methods (Additional file 3: Figure S3 and Additional file 5: Table S4) [28].

Except for the hypermutations in AYA07, we focused on level-1 variants to identify driving genetic alterations of AYA cancers that are specific to each patient and may be druggable (Fig. 2 and Additional file 6: Table S5). CNVs were analyzed to identify candidate driving CNVs (OncoScan™) and chromosome-level CNVs (OncoScan™ or VarScan2) (Fig. 3 and Additional file 1: Figure S4).

To investigate differences in the genomic profiles between AYA cancers and cancers found in other age groups, we compared candidate driving genes between AYAs and all age groups for the same cancer based on published data from analyses similar to ours (Table 2). The mutation pattern of AYA01 (prostate cancer) differed from that shown in a large-scale study analysis. Whereas AYA01 harbored alterations in the Ras pathway (NF1 and RASA2), prostate cancer from other age groups showed recurrent mutations in SPOP, TP53, and PTEN [39]. However, several commonly altered genes in HNSCC, such as TP53, FAT1, and NOTCH1, were identified in both AYA04 and in large-scale studies, whereas other mutations differed [34].

In this study, we described the genomic profiles of seven different rare types of AYA cancers using three different genomics platforms (WES, WTS and OncoScan™). After processing genomics data, we identified potential druggable targets for each cancer and selected existing anti-cancer drugs to treat individual patients (Fig. 2 and Additional file 6: Table S5).

We identified candidate driving genetic alterations specific to each patient using logical manual curation (pattern-based heuristic annotation) as alternative to statistical method (Additional file 9: Supplementary materials and methods and Additional file 10: Table S8). It is needed to alternative method to select candidate genes in rare type of cancers, like AYA cancers, since low number of samples is limited to the selection of candidate driving genes using statistical method as shown in large-scale studies [4].

Because the features of AYA cancers are distinct from those of other age groups, including the incidences and clinical outcomes, studying the genomic profiles of AYA cancers is important to identify the unique features of AYA cancers [5, 6]. When comparing candidate genes between AYAs and all age groups for the same cancer type, we identified different candidate genes in prostate cancer (AYA01) and a germ cell tumor (AYA07), although several common candidate genes (TP53, FAT1,and NOTCH1) were found in HNSCC (AYA04) in both AYAs and all age groups (Table 2). These results showed that AYA-specific genetic alterations may be different from those in other age groups; thus, further study is needed to define the significance of the differences in the genetic alterations between AYAs and other age groups.

In this study, we analyzed individual cancer genomes and suggested potential drugs for each patient based on his or her genetic alterations. Characterizing the genomes of patients and genomics-driven knowledge enabled personalized medicine and advanced cancer genomics for clinical implications [40]. Moreover, to establish the clinical validity of genetic tests, especially for NGS data, the FDA discussed ‘post-marketing pursuit’ to define the clinical implications of variants generated from NGS, which have remained unknown [41]. Therefore, we expect many prospective genomic studies, such as our study, to link the patient to therapy as well as diagnosis, prognosis, and monitoring [42].