Recurrent Mutations of Chromatin-Remodeling Genes and Kinase Receptors in Pheochromocytomas and Paragangliomas

Pheochromocytomas and paragangliomas (PPGL) are catecholamine-secreting tumors of neural crest origin that arise from the sympathetic lineage cells of the adrenal medulla and paraganglia, respectively. More than 40% of these tumors are caused by a dominant driver mutation in one of various susceptibility genes involving a broad range of pathways (1). Remarkably, in more than one-third of the patients, the mutation is found in the germline and is transmitted in an autosomal dominant manner (1, 2).

To explore new genetic events underlying familial and sporadic cases, we sequenced the exomes or transcriptomes of 43 PPGL samples from 41 individuals (Supplementary Table S1 and Fig. 1). Sequences were generated from 26 tumors, 13 germline samples, and 4 tumor–germline pairs. Three separate tumors were from the same patient. Germline samples were selected from individuals at high risk for hereditary PPGL, that is, family history of PPGLs, disease onset before 30 years of age, and/or multiple tumors (Supplementary Table S1). Seven patients (17%) had a history of metastatic PPGLs. Here we report novel variants in genes coding for tyrosine kinases and chromatin-remodeling proteins in these samples.

One-hundred seventy-seven PPGLs, obtained after written informed consent or in an anonymized fashion after deidentification, were collected through a tumor repository approved by the University of Texas Health Science Center at San Antonio (UTHSCSA; San Antonio, TX) Institutional Review Board. A summary of the main clinical features of this cohort is shown in Supplementary Table S1. Forty-three samples from 41 individuals were used for whole exome (n = 40) or transcriptome sequencing (n = 3). A separate cohort of 136 pheochromocytoma or paraganglioma tumor only samples was used for verification of some of the detected mutations (Supplementary Table S2). Clinical features of one patient with associated PPGL and GCT has been reported elsewhere (3). Additional details of the samples used in the study are available in Supplementary Data.

Whole-exome sequencing was performed in 40 and RNA sequencing was performed in 3 PPGL samples (Supplementary Table S1). Of the 43, 26 were tumors, 13 were germline samples from blood, and four were matched tumor–germline pairs. Three of the tumors were from the same patient. The four paired samples and two tumors from the same patient were previously reported (4). For the remaining samples, libraries were prepared using Agilent SureSelect XT2 Target Enrichment System for Illumina Multiplexed Sequencing, version C.2, and exomes enriched using Agilent Sure Select 44 Mbp or NimbleGen 44 Mb SeqCap EZ Exome Kit v2 (at the Beijing Genomics Institute) or Agilent Sure Select XT2 Human All Exon V5 capture (at University of Texas at Austin Genomic Core facilities) and sequenced on Illumina HiSeq2000. Sequences were aligned and annotated as detailed in Supplementary Methods. Distinct filtering thresholds were applied to germline and tumor samples and consensus variants were considered for further analysis. To prioritize significant germline or somatic missense variants, we used separate tools of the Cancer-Related Analysis of Variants Toolkit (CRAVAT; Supplementary Methods; refs. 5, 6). The somatic nature of selected variants was verified in the four paired cases and in an additional 11 samples for which matched germline or tumor DNA was available. RNAseq data were processed for sequence variant detection only.

DNA from fresh-frozen or formalin-fixed paraffin-embedded (FFPE) sections from the various tumor and nontumor tissues from two cases carrying a H3F3A c.103 G>T, p.G34W mutation, and three unrelated PPGLs, and four sporadic GCTs were used to quantify variant allele representation. A similar approach was used to detect the MET c.2416G>A; p.V806M variant in samples from three individuals with familial pheochromocytoma without a known driver mutation, including germline DNA from two affected siblings and FFPE from a nonpheochromocytoma tissue section from their affected father, along with unrelated samples. Primers spanning the two variants contained adaptors to allow for detection of the variant allele at ultra-deep sequencing using Illumina TruSeq Custom Amplicon with slight modification, as detailed in Supplementary Methods. The sequenced traces were annotated using VarScan2 (7) and the specific variants of interest were detected at an average read depth of 67700x (±47969SD). Samples with variant read counts below the threshold of quality detection (below 1%) were considered negative.

Exons spanning the MERTK kinase domain, MET semaphorin, transcription factor immunoglobin (TIG), juxtamembrane and kinase domains, H3F3A, and H3F3B exon 2 were sequenced in the relevant samples from the exome/transcriptome cohort and in the validation cohort of 136 PPGLs by Sanger sequencing (primers and PCR conditions available upon request). Sequencing was processed at Beckman Genomics and analyzed with Mutation Surveyor (Softgenetics), as previously reported (4). The Mutation Quantifier tool of Mutation Surveyor was used to measure frequency of the H3F3A c.103 G>T, p.G34W mutation, expressed as percentage of the mutant allele in the tumor samples, and calculated as described in Supplementary Methods.

Gene expression data generated using the Affymetrix U1332.0 platform were previously reported (GEO accessions GSE2841 and GSE19987; ref. 8). Normalized data from the three tumors with H3F3A G34W mutation and from 36 tumors without this mutation were used for gene set enrichment analysis (GSEA), as detailed in Supplementary Methods. DAVID Bioinformatics Resources 6.7 NIAID/NIH toolset was used for further annotation of the GSEA differentially expressed genes onto gene ontology terms, protein–protein interactions, and protein functional domains (9). Results were expressed as an enrichment score and associated P value.

Structural predictions of histone 3.3 variants were performed using the I-TASSER server (10). I-TASSER selected the nucleosome PDB structure 1KX5, determined by X-ray crystallography at 1.9Å resolution, as the highest significance template for predicting the final models. We used the full-length amino acid sequence for human histone H3F3A (ENST00000366813) and computationally modeled the wild-type (WT) and the G34W mutant. The C-scores (confidence score, range -5 to 2), TM-scores, (template modeling) and root mean square deviation values are shown in Supplementary Fig. S2A for all the structural models provided by I-TASSER. Additional details of model selection, three-dimensional structure visualization, and calculations of electrostatic surface potential and hydrophobicity are described in Supplementary Methods.

HEK293 and Kelly neuroblastoma cell lines were cultured at 37°C in 5% CO₂ in DMEM (CellGro) with 10% FBS (Thermo Fisher Scientific, Life Technologies). HEK293 cell line was preexistent in our group and was obtained earlier from ATCC. The Kelly cell line was obtained from Dr. Gail Tomlinson (Department of Pediatrics, UTHSCSA, San Antonio, TX). The Kelly cell line was not formally authenticated but its identity was verified by this cell's characteristic amplification of MYCN at RNA and protein levels by quantitative real-time PCR and Western blotting, respectively. Both cell lines were tested and cleared of Mycoplasma contamination before the start of the project. Transfection was performed as described (11). A construct containing wild-type MERTK was obtained from (Addgene, #23900), subcloned into pHM6, and the relevant mutations introduced using site-directed mutagenesis using primers containing the mutation (Sigma) and Phusion Taq polymerase (Thermo Fisher Scientific). Recombinant MERTK ligand Gas6 was purchased from R&D Systems and applied to culture media or HBSS (400 nmol/L) for 10 minutes before harvesting.

Whole-cell lysates from tumors (12) or cell lines were prepared as described previously (11) and detailed in Supplementary Methods. Proteins were detected with the following antibodies: H3K27me3, H3K36me3, Histone H3, MYCN, MERTK, phospho ERK (all from Cell Signaling Technology), or β-actin (Sigma-Aldrich, #A2228).

Immunohistochemical staining was performed using the standard 3-step staining ImmunoCruz TM LSAB Staining System (Santa Cruz Biotechnology) with histone H3 trimethyl-K36 antibody (Epigentek #A-4042) and H3 tri-methyl K27 (Cell Signaling Technology C36B11) both at 1:200 dilution. In negative controls the primary antibody was replaced by normal rabbit IgG. Slides were counterstained with hematoxylin. Nuclear staining was evaluated on the basis of intensity and percentage of positive cells. Five fields were analyzed from each slide.

The mean depth of coverage of the 40 exomes was 107× and the average read number of the RNAseq samples was 140 Mbp. The number of variants per tumor was small across the entire group: the median somatic mutation frequency was 0.8 per Mb in paired exomes (range 0.3–1.3), comparable with the rate recently described in other series (13–16). The overall mutation frequency was 9.7 mutations per Mb (range 1–26) in nonpaired samples, including both somatic and germline variants. Germline and somatic mutations in known PPGL genes were detected in 16 samples (41%, including two somatic NF1 mutations in the same tumor, Supplementary Table S2). Some variants had low allelic fraction (Table 1, Supplementary Table S2), suggesting that they may represent subclonal mutant populations within the tumor.

We next explored genes that had not been implicated in PPGL tumorigenesis, specifically searching for variants belonging to the same functional or structural class and found two gene categories that were recurrently mutated in our series (Fig. 1 and Table 1): 11 high-scoring variants (11/43, 25.6%) were found in chromatin-remodeling genes, including histone (H3F3A), histone chaperone ATRX, methyltransferases (KMT2B, EZH2, and SETD2), demethylases (JMJD1C, KDM2B), of 8 patients. The second category encompassed kinase receptor–encoding genes, in which we identified 16 mutations (16/43, 35%) in 15 distinct individuals. These 27 variants were found somatically and in the germline, had few overlaps between the two groups (only three tumors had variants in both gene groups), and predominantly targeted PPGLs without a recognizable germline mutation (Fig. 1, Supplementary Fig. S1A and S1B, and Table 1).

Within the first group, we detected a mutation of the histone 3.3 encoding gene, H3F3A, which was shared by the exome of three tumors from the same patient, suggesting that the mutation was inherited as a driver event in these tumors (Fig. 2A). The patient was a 31-year-old female who had bilateral pheochromocytomas and later bladder and periaortic paragangliomas, but without a family history of PPGLs. The H3F3A mutation (c.103 G>T, p.G34W, COSM1732355) was identical to that reported recently as the main oncogenic driver of sporadic giant cell tumor of bone (GCTs; refs. 17, 18). Upon reviewing the patient's records, we identified a previous history of recurrent tibial GCT. These bone tumors occur mainly in young adults, preferentially affect epiphyseal sites and have the potential to metastasize (19). The same glycine is also mutated in about 15% of pediatric gliomas, although distinct amino acid changes (either G34R or G34V) are seen in these tumors (20, 21), suggesting a tight association between histologic type and the mutated amino acid (18). To expand on this initial finding, we obtained samples from an unrelated 20-year-old male patient who was previously reported with an aggressive retroperitoneal paraganglioma with liver metastasis and recurrent, metastatic GCTs (3), and found the same H3F3A mutation (Table 2 and Supplementary Fig. S2). In both patients, the G34W mutation was detected in all tumors and their metastasis, but was either absent (first case, C1), or detected at a low fraction (7.8%, second case, C2) in adjacent, histologically tumor-free tissue by ultra-deep targeted sequencing (Table 2). This mutation pattern excluded a germline event and suggested instead that the mutation might have occurred postzygotically. This differs from H3F3A mutations in bone and brain cancers, which have been detected exclusively at the somatic level (18, 20, 21). No other mutations of H3F3A, or its homolog H3F3B, were found in an independent set of 136 PPGLs not associated with GCT (Supplementary Table S3).

We next performed computational modeling to infer the tridimensional structure of the G34W mutant and found that this variant is predicted to change the electrostatic potential (Supplementary Fig. S3A and S3B) and surface hydrophobicity of H3.3 (Supplementary Fig. S3C and S3D). These alterations are expected to promote a marked shift on the side chains of K27 and K36 (Fig. 2B and Supplementary Fig. S3E), two key posttranslationally regulated residues relevant for modulating DNA accessibility, thus impacting on gene transcription (17). To determine whether the G34W mutation modulated the levels of methylation of these two histone residues, we examined tumor sections from the two H3F3A-mutant patients by IHC. Both trimethylated K36 (Fig. 2C and Supplementary Fig. S4A) and K27 (Supplementary Fig. S4B) were detectable in the majority of paraganglioma nuclei, and in a significant proportion of GCT cells. These immunohistochemical findings were confirmed by Western blots of lysates from the PPGL from one of the patients (Fig. 2D). These observations suggest that the mutation does not cause overt changes in K36 or K27 methylation, although it is not possible to unequivocally determine in these experiments that the mutant allele binds to the methylation-specific antibodies with the same affinity as wild-type H3.3. Nonetheless, these results are similar to the findings in G34-mutant brain tumors (22–24). However, it should be noted that the amino acid changes in gliomas, G34V or G34R, are distinct from the specific variant detected in our samples, G34W, thus structural and functional differences may exist among these variants and lead to unique phenotypes.

We also examined the global gene transcription profile of the three G34-mutant PPGLs from the first patient (8). These mutants clustered with tumors carrying mutations in RET, NF1, or TMEM127 genes, known as cluster 2 PPGLs (ref. 8; Supplementary Fig. S5A). GSEA and gene ontology assessment revealed significant association with ontologies related to neurogenesis and neural differentiation in the three G34W-mutant PPGLs (Supplementary Fig. S5C and S5C;Supplementary Table S4), which mirrors the profile found in G34-mutant pediatric gliomas (25, 26). We next investigated whether MYCN, which is overexpressed in G34-mutant gliomas (25), could also be deregulated in PPGLs with H3F3A mutation. Using Western blot of primary tumors, we demonstrated that the MYCN protein is upregulated in mutant G34W-PPGLs when compared to tumors with intact H3F3A sequence (Fig. 2E). Future studies will be required to establish a causative connection between the G34W histone mutation and MYCN upregulation, but our current results suggest that MYCN may be overexpressed in paragangliomas with this mutation, and may contribute to the pathogenesis of PPGLs.

Besides the H3.3 mutation, variants were also identified in other genes involved in chromatin-mediated gene regulation. Among the methyltransferases, we identified a truncating mutation in SETD2 in an otherwise mutation-free, sporadic PPGL (Table 1 and Supplementary Table S1). SETD2 is a K36 methyltransferase frequently mutated in renal cancers and leukemias (27, 28), suggesting that this variant is pathogenically relevant. An in-frame, single amino acid germline deletion of the K27 methyltransferase EZH2 was found in another sporadic pheochromocytoma (Table 1 and Supplementary Table S1). Activating and inactivating EZH2 mutations are involved in, respectively, lymphomas (29) and myeloid malignancies (30). In addition, KMT2B, a member of the K4 methyltransferase 2 (KMT2) family formerly known as MLL, was mutated in germline DNA of a patient with multiple paraganglioma recurrences since the age of 14 years (Table 1). KMT2 mutations occur in various cancers (31) and both germline and somatic variants of the KMT2D isoform were recently described in pheochromocytomas (16) supporting the likely pathogenetic relevance of the KMT2B variant that we identified. Two demethylases, JMJD1C and the KDM2B, members of the Jumonji C (JmjC)-domain containing demethylase family were mutated, respectively, in a benign, sporadic and in a malignant PPGL carrying a mutation in the PPGL susceptibility gene, SDHB (Table 1). Finally, ATRX, a histone 3.3 chaperone, was mutated in one sporadic tumor (Table 1). Somatic ATRX mutations, including the same variant detected in one of our samples, were recently reported in PPGLs (13–15) and found to associate with malignant forms of these tumors (15). Together, these data suggest a broad role of chromatin-remodeling genes in PPGLs.

A second group of recurrently mutated genes comprised tyrosine kinase receptors (Table 1), a class of proteins often disrupted in cancers (32). We evaluated three of these variant kinase-encoding genes in detail. A germline mutation was detected in a residue within the tyrosine kinase domain of the MERTK gene (c.2273G>A, p.R758H, COSM1398803; Fig. 3A) in a 32-year-old patient diagnosed with pheochromocytoma, recurrent and metastatic paragangliomas, and medullary thyroid carcinoma (MTC). The association of pheochromocytoma and MTC is reminiscent of multiple endocrine neoplasia type 2 (MEN 2), but no mutations were detected in the MEN2 susceptibility gene, RET, in this patient. MERTK is one of the three components of the TAM kinase receptor family (TYRO3, AXL, MERTK), which is overexpressed in cancers (33). We sequenced the MERTK kinase domain in a separate cohort of 136 PPGLs and identified a germline mutation targeting the same residue, R758C, in a 36-year-old female with a sporadic pheochromocytoma (Fig. 3A). Ectopic expression of both 758R- and 758C-mutant MERTK in HEK293 cells led to sustained phosphorylation of the downstream signaling effector ERK after stimulation with the MERTK ligand GAS6, compared with cells expressing an empty vector (EV), WT construct, or a dominant negative MERTK mutant Y754F (Fig. 3B). Importantly, ERK phosphorylation was maintained in these mutants after nutrient starvation (Fig. 3C), suggesting that kinase domain changes may lead to constitutive activation of the receptor. Furthermore, somatic MERTK mutations were identified in two pheochromocytomas of the TCGA dataset (preliminary release at refs. 34–36). These results are consistent with MERTK mutations playing a role in PPGL pathogenesis.

We also identified a novel germline mutation of the MET kinase receptor in the exome of the index case of a three-generation family with pheochromocytomas but without a known driver mutation. MET is somatically mutated in multiple cancers and germline-activating mutations cause familial papillary renal carcinoma (37). The germline c.2416G>A; p.V806M mutation found in this patient resides within the transcription factor immunoglobin (TIG) juxtamembrane domain, and segregates with the disease in this family: it is present in the affected sibling and father, but absent in the unaffected mother, when tested by Sanger and targeted next-generation sequencing (Fig. 3D and Supplementary Fig. S5a and S5b). There is no report of renal cancer in this family but one of the affected individuals had bladder cancer diagnosed after 70 years of age. Additional MET variants were identified within the semaphorin domain, a hotspot region for cancer mutations (38), in the exome of three sporadic PPGLs (Table 1). We next sequenced 118 unrelated PPGLs for exons spanning these domains and identified 11 other variants present in COSMIC, TCGA, and/or predicted to be pathogenic, or present in reference databases with a minor allele frequency <0.05 (Fig. 3E and Supplementary Table S5). Three of the variants occurred in more than one tumor, whereas a variant of controversial pathogenicity was seen in five tumors (39). Overall, we identified 15 different samples carrying MET variants, both somatic and germline (Fig. 3E), suggesting that this gene may be more commonly mutated in PPGLs. In support of our findings, somatic MET mutations were recently reported in an independent series of PPGLs (14).

A somatic mutation in the main hotspot residue of the FGFR1 receptor (c.1638C>A; p.N546K COSM19176/302229), known to constitutively activate the receptor in glioblastomas (40) and other cancers (41), was detected in one patient with a sporadic pheochromocytoma without an identifiable PPGL susceptibility gene mutation (Table 1). No additional N546K mutations were found in our validation group of 136 samples, but this variant was detected in a sample of the PPGL TCGA dataset (35, 36), suggesting that FGFR1 mutation is a rare but recurrent event in PPGLs.

Mutations in PPGLs have been considered mutually exclusive (1). Only three samples in our series carried mutations in both chromatin-related and kinase genes (3/23 separate tumors from 20 individuals; Fig. 1 and Table 1). We also verified whether novel mutations (of either class) coexisted with those of well-known PPGL genes. These variants tended to occur predominantly in tumors with no mutation of an established PPGL gene (13/23, log OR, −1.28; P = 0.06; Table 1), suggesting that they might play a dominant role in tumor development.

Our study uncovered new candidate germline, postzygotic, and somatic driver mutations in PPGLs. Overall, mutations in genes related to chromatin remodeling or kinase receptors were found in 20 of the 41 patients, and when cases with known PPGL gene mutations are combined, driver mutations could be assigned to 68% of the samples in this series.

Our data point to mutations in chromatin-remodeling genes as a recurrent feature of PPGLs, implying that epigenetic modifications occur in these tumors. Indeed, epigenetic changes associated with enhanced DNA and histone methylation have been linked to loss-of-function mutants of the well-known PPGL susceptibility genes SDH and FH, engendered as a result of metabolic imbalances that decrease the activity of histone demethylases (42). Our data now suggest that the chromatin architecture of a sizeable proportion of PPGLs is disrupted by different mechanisms, involving direct mutations of chromatin regulators. At least in the case of histone 3.3 mutations, such misregulation occurs in a specific clinical and developmental context. Furthermore, the postzygotic H3.3 mutation pattern found in our cases, not previously reported in other H3.3-mutated tumors, may now instigate studies to revisit the cell of origin of sympathetic lineage and bone tumors, not hitherto thought to share common precursors (43), and may also have implications for measuring transmission risk of the mutation in affected patients. More broadly, identification of specific consequences of each individual and combinatorial variant on chromatin deposition throughout the PPGL genome may provide key insights into gene regulation and tumorigenesis.

Our findings also expand the role of kinase receptors in PPGLs. The RET kinase receptor, the oncogene mutated in MEN2 syndrome and the related syndrome familial MTC, is one of the earliest recognized PPGL susceptibility genes (44). The mutated MERTK R758, found in two patients of our study, is homologous to RET R912, previously detected in FMTC (45). R758 is adjacent to a conserved KWIAIES sequence within the kinase domain, known as the TAM domain (46), which is highly homologous to RET and encompasses a hotspot residue for mutation of this gene in human cancers (M918T, COSM965). Moreover, the phenotype of one of the affected patients in our cohort raises the intriguing possibility that MERTK may be related to rare RET-negative cases with MEN2-like clinical manifestations.

The relevance of mutations of the MET receptor found in our study is corroborated by recent findings of somatic mutations of this gene in PPGLs (14). We expand on these observations by also identifying MET as a susceptibility gene to familial pheochromocytoma. Somatic MET mutations occur in multiple cancer types, but germline MET mutations have only been previously reported in papillary renal cancers (37). Although the patients with MET mutations in our series had no history of renal tumors, there is increasing recognition of an overlap between genes that cause susceptibility to pheochromocytoma and renal cancers (47, 48). Taken together, these findings suggest that MET may also be involved in PPGL predisposition.

The identification of variants in either chromatin or kinase genes in patients without a known germline PPGL mutation, detection of additional mutations of these new candidate genes in our validation cohort and also in other, independent series (13–16) offer support for their role in PPGL pathogenesis. Definitive validation of these new candidates as susceptibility genes and their potential value for genetic risk assessment, prognosis, and surveillance awaits further analyses. Importantly, although these variants were not specifically enhanced in malignant tumors, the availability of clinical grade tyrosine kinase inhibitors, including those targeting MET and FGFR1, and agents directed at chromatin modifications may expand the number of experimental therapeutic options to malignant or inoperable PPGLs (32, 49).

Notably, the low mutation frequency and limited overlap between individual genes found in our and other PPGL series (13–16) suggests that the spectrum of genetic driver events in these rare tumors is not saturated, and that multiple cohorts will be required to complete the catalog of genetic variations implicated in PPGLs. The relevance of pursuing the full genetic repertoire of these tumors is especially compelling given their high rates of heritability, at present conservatively estimated at 40%, and the heightened transmission risk of affected alleles.

No potential conflicts of interest were disclosed.

Conception and design: R.A. Toledo, R.C.T. Aguiar, P.L.M. Dahia

Development of methodology: R.A. Toledo, Z.-M. Cheng, Q. Gao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Qin, Q. Gao, S. Iwata, M.L. Prasad, I.T. Ocal, S. Rao, N. Aronin, M. Barontini, J. Bruder, P.L.M. Dahia

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.A. Toledo, Y. Qin, S. Iwata, G.M. Silva, I.T. Ocal, R.L. Reddick, Y. Chen, R.C.T. Aguiar, P.L.M. Dahia

Writing, review, and/or revision of the manuscript: R.A. Toledo, Y. Qin, I.T. Ocal, N. Aronin, R.L. Reddick, R.C.T. Aguiar, P.L.M. Dahia

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.A. Toledo, I.T. Ocal

Study supervision: P.L.M. Dahia

The authors thank Dr. Jonathan Liu (SoftGenetics) for support and assistance with NextGEne analysis, Dr. Scott Hunicke-Smith for processing whole exome and transcriptome sequencing, Dr. Zhao Lai and Dawn Garcia for help with targeted next-generation sequencing, Jamie Myers for assistance with Sanger sequencing, Dr. Josefine Heim-Hall for access to Pathology tumor bank samples, Jamie Parra for support with tumor bank material and IHC, and Dr. Gail Tomlinson for providing a cell line (Kelly). The authors also thank many colleagues who have contributed to the Familial Pheochromocytoma Consortium in the past.

This study was supported by grants from the Max and Minnie Voelcker Fund, and Greehey Children's Cancer Research Institute (GCCRI; to P.L.M. Dahia). R.A. Toledo was a recipient of a research fellowship from the National Council for Scientific and Technological Development (CNPq). The CTRC Pathology tumor bank and GCCRI Next Generation Sequencing core are supported by UTHSCSA, NIH-NCI P30 CA54174.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.