Abstract
Adenoid cystic carcinomas (ACC) of the salivary glands are challenging to understand, treat, and cure. To better understand the genetic alterations underlying the pathogenesis of these tumors, we performed comprehensive genome analyses of 25 fresh-frozen tumors, including whole-genome sequencing and expression and pathway analyses. In addition to the well-described MYB–NFIB fusion that was found in 11 tumors (44%), we observed five different rearrangements involving the NFIB transcription factor gene in seven tumors (28%). Taken together, NFIB translocations occurred in 15 of 25 samples (60%, 95% CI, 41%–77%). In addition, mRNA expression analysis of 17 tumors revealed overexpression of NFIB in ACC tumors compared with normal tissues (P = 0.002). There was no difference in NFIB mRNA expression in tumors with NFIB fusions compared with those without. We also report somatic mutations of genes involved in the axonal guidance and Rho family signaling pathways. Finally, we confirm previously described alterations in genes related to chromatin regulation and Notch signaling. Our findings suggest a separate role for NFIB in ACC oncogenesis and highlight important signaling pathways for future functional characterization and potential therapeutic targeting. Cancer Prev Res; 9(4); 265–74. ©2016 AACR.
Introduction
Adenoid cystic carcinomas (ACC) of the salivary glands are often indolent but can aggressively invade local structures and metastasize to distant sites many years after initial treatment (1). ACC frequently spreads to nearby nerves, a process termed perineural invasion (PNI), which portends poor prognosis and higher likelihood of local recurrence (2, 3). There are at present no reliable chemotherapeutic options for long-term disease control.
Recently, molecular and whole-exome sequencing studies have begun to shed light on the genetic underpinnings of this relatively rare disease (4–7). A t(6;9)(q22–23;p23–24) chromosomal translocation resulting in fusion of the MYB and NFIB genes has been described in 29% to 86% of cases (4, 5, 8–11). Whether the MYB oncogene or the transcription factor encoded by NFIB is responsible for the selective advantage afforded by these translocations is unknown. Frequent alterations of genes involved in chromatin regulation, Notch signaling, and several other pathways has also been reported (5–7). Whole-genome sequencing, which is capable of revealing chromosomal rearrangements not detectable by traditional cytogenetic techniques, has been applied to a limited number of tumors (5, 12).
We performed whole-genome sequencing of 25 fresh-frozen surgically resected ACC tumors, and mRNA expression analysis of a subset of these tumors, to better understand the genetic alterations that drive ACC pathogenesis. We found novel rearrangements involving NFIB, and showed that NFIB was translocated in the majority (15 of 25, 60%) of tumors. NFIB mRNA was also overexpressed in ACCs compared with normal tissues. We also noted frequent disruption of axonal guidance and Rho family signaling pathways. Our findings suggest a critical role for NFIB in ACC oncogenesis and identify important signaling pathways for functional characterization and potential therapeutic targeting in the future.
Materials and Methods
Sample preparation
Fresh-frozen surgically resected tumor and matched blood/normal tissue were obtained from patients under an Institutional Review Board protocol at the Johns Hopkins Hospital and from the Salivary Gland Tumor Biorepository. Informed consent was obtained from each patient. Tumor tissue was analyzed by frozen section histology to estimate neoplastic cellularity. In order to enrich the samples for neoplastic cells, normal tissue was removed from the samples using macro-dissection based on the frozen section histology. An estimated average of at least 60% neoplastic cellularity was obtained. DNA and RNA were purified using AllPrep (Qiagen, cat# 80204).
DNA sequencing
Whole-genome sequencing was performed in tumors from 25 patients. Tumor samples were carefully selected or dissected in order to achieve a neoplastic cellularity of >60%. DNA was purified from these tumors and from matched nonneoplastic tissue, and used to generate libraries suitable for massively parallel sequencing. Sample library construction, next-generation sequencing, and bioinformatic analyses of tumor and normal samples were performed at Personal Genome Diagnostics (Baltimore, MD). In brief, genomic DNA from tumor and normal samples was fragmented and used for Illumina TruSeq library construction (Illumina). Paired-end sequencing, resulting in 100 bases from each end of the fragments, was performed using a HiSeq 2000 Genome Analyzer (Illumina).
Somatic mutations were identified using VariantDx custom software for identifying mutations in matched tumor and normal samples. Prior to mutation calling, primary processing of sequence data for both tumor and normal samples was performed using Illumina CASAVA software (v1.8), including masking of adapter sequences. Sequence reads were aligned against the human reference genome (version hg19) using ELAND with additional realignment of select regions. Candidate somatic mutations, consisting of point mutations and small (<50 bp) insertions and deletions, were then identified using VariantDx across the exonic regions. VariantDx examines sequence alignments of tumor samples against a matched normal while applying filters to exclude alignment and sequencing artifacts. In brief, an alignment filter was applied to exclude quality-failed reads, unpaired reads, and poorly mapped reads in the tumor. A base quality filter was applied to limit inclusion of bases with reported Phred quality scores >30 for the tumor and >20 for the normal (13). A mutation in the tumor was identified as a candidate somatic mutation only when (i) distinct paired reads contained the mutation in the tumor; (ii) the number of distinct paired reads containing a particular mutation in the tumor was at least 2% of the total distinct read pairs for targeted analyses and 10% of read pairs for exome; (iii) the mismatched base was not present in >1% of the reads in the matched normal sample as well as not present in a custom database of common germline variants derived from dbSNP; and (iv) the position was covered in both the tumor and normal. Mutations arising from misplaced genome alignments, including paralogous sequences, were identified and excluded by searching the reference genome. Candidate somatic mutations were further filtered based on gene annotation to identify those occurring in protein coding regions. Functional consequences were predicted using snpEff and a custom database of CCDS, RefSeq, and Ensembl annotations using the latest transcript versions available on hg19 from University of California Santa Cruz (UCSC) (14). Predictions were ordered to prefer transcripts with canonical start and stop codons and CCDS or Refseq transcripts over Ensembl when available. Finally, mutations were filtered to exclude intronic and silent changes, while retaining mutations resulting in missense mutations, nonsense mutations, frameshifts, or splice-site alterations. A manual visual inspection step was used to further remove artifactual changes. We have previously optimized our sequencing and bioinformatics approaches so that specificity of mutations is extremely high. This has been extensively validated by not only Sanger Sequencing but also by NGS at high depth. A minimum of 95% of the mutations identified using these approaches are bona fide (15, 16).
Copy-number alterations were identified by comparing normalized average per-base coverage for a particular gene in a tumor sample to the normalized average per-base coverage in a matched normal sample for that patient (17). Focal amplifications (≥3-fold or six copies) and homozygous deletions were reported. Genomic rearrangements were identified through an analysis of discordantly mapping paired-end reads. The discordantly mapping paired-end reads were grouped into 1-kb bins when at least 5 distinct tag pairs (with distinct start sites) spanned the same two 1-kb bins, and further annotated based on the approximate breakpoint (18).
Randomization-based statistical test for proportion of truncating mutations
We performed a randomization-based statistical test of increased proportion of truncating mutations (S) out of total non-silent mutations (N) for genes involved in chromatin regulation, controlling for the effect of gene sequence and mutational context. For each gene i, our test statistic was
Truncating mutations were defined as any nonsense, conserved di-nucleotide splice-site mutations, or out-of-frame insertions/deletions (frameshift). Monte Carlo simulations were performed to approximate the null probability distribution of the test statistic |${T_i}$|. Each Monte Carlo sample was generated by moving every observed non-silent single-base substitution (SBS) mutation to uniformly sampled nucleotide positions within the same gene, with matched single-nucleotide mutational context (e.g., if a C→T was observed, the mutation had to be moved to a C reference position). However, the mutation consequence type of the C→T might change. For example, a SBS that generated a missense mutation might generate a nonsense mutation in its new position. Because frameshift mutations do not change consequence when moved to a different position, in the Monte Carlo sample, they were retained with probability equal to the observed proportion of frameshift mutations out of all mutations (maximum likelihood estimate), otherwise they were changed to a non-truncating mutation. After each iteration of this sampling procedure, the number of mutations in a gene is always the same, but the mutation consequence of each mutation may change. Thus, the test statistic |${T_i}$| for the gene will change values at each iteration, and repeated iterations yield a null distribution of test statistics to estimate the P value of the gene's observed test statistic. For the gene group analysis, our test statistic was
and it was computed both for the observed and simulated mutations. A one-tailed empirical P value was calculated as the fraction of Monte Carlo samples in which the observed value of the test statistic was equal to or higher than the simulated value. Increasing the number of iterations of Monte Carlo sampling increases the precision of the P value; 10,000,000 iterations were chosen to achieve adequate precision.
mRNA expression
Strand-specific sequencing libraries were prepared using the TruSeq Stranded Total RNA library kit (Illumina). Barcoded libraries were quality controlled using the Kappa PCR kit and pooled in equimolar ratios for subsequent cluster generation and sequencing on an Illumina HiSeq 2000 instrument to yield >50,000,000 paired-end 100 × 100 bp tags for each sample. Paired-end reads were then mapped to the human genome (build hg19), and expression levels of all known gene isoforms annotated in RefSeq, Ensembl, and UCSC gene annotations were derived using the RSEM package (19), which uses an expectation maximization algorithm to derive the abundance of each gene isoform after taking into account the read mapping uncertainty with a statistical model. TopHat-Fusion was used to identify fusion transcripts from the RNA sequencing data (20).
RSEM-expected counts for genes were upper quartile normalized perl scripts from the TCGA RNA-seq V2 normalization pipeline (22). Each gene was further normalized to a maximum value of 1 for visualization in a heatmap. Upper quartile normalized read counts were voom transformed for differential expression statistics (23). All genes with overall log fold change greater than 0.25 were compared with empirical Bayes moderated t-statistics with equal variance implemented in the R/Bioconductor package LIMMA version 3.26.0 (24). P values reported are Benjamini–Hotchberg corrected among the subset of seven genes in NFIB fusions and retained for differential expression analysis.
Pathway analyses
Pathways analysis was performed on the DNA sequencing analysis results to identify biological pathways that are likely to be altered by the somatic mutations alone, and then by both somatic mutations and chromosomal rearrangements observed in the 25 ACC samples analyzed, using Ingenuity Pathway Analysis (IPA) Build: 321501M (Qiagen). Two gene sets (mutations and mutations-plus-rearrangements) were separately uploaded to IPA and compared with their gene Knowledgebase to determine the known canonical pathways that contained the affected genes. The statistical significance of each pathway was reported as a P value, determined using the right-tailed Fisher exact test, representing the likelihood that the given pathway was chosen by chance. Figures are provided for selected pathways, wherein the affected gene symbols are color coded.
Results
Sequence mutations
Whole-genome sequencing was performed on 25 samples as described in the Materials and Methods section (Supplementary Table S1 and Methods). The average coverage of each base in tumors was 51.7-fold overall (65.2-fold for tumors and 38.1-fold for matched nonneoplastic tissues), and 98.2% of targeted bases were represented by at least 10 reads (Supplementary Table S1). We identified 396 somatic mutations (point mutations and small indels) in 372 genes among the 25 tumors. The range of mutations per tumor was 2 to 36, with a median of 14 (Table 1 and Supplementary Table S2). Somatic mutations were identified in more than one tumor in 21 different genes. NOTCH1 was the most frequently altered gene, with four mutations in three tumors including two substitutions (one nonsense and one nonsynonymous), one frameshift and one in-frame deletion. A nonsynonymous substitution was also found in NOTCH2. Other genes found to have multiple mutations are recorded in Table 2.
Summary of common genetic alterations in each tumor sample
Abbreviations: intrachr., intrachromosomal; interchr., interchromosomal; IQR, interquartile range.
aIncludes point mutations and small indels.
b“X” indicates tumors with the listed characteristic; “–” indicates lack of the listed characteristic.
cBold: mRNA transcript also detected. Italicized: not tested for mRNA.
Top mutated genes in adenoid cystic carcinomaa
Gene . | Number mutations (N = 396) . | Number tumors (N = 25) . | Gene size . | Mutations/Mb . |
---|---|---|---|---|
NOTCH1 | 4 | 3 | 7,668 | 21 |
MLL2 | 3 | 2 | 16,614 | 7 |
ABCA4 | 2 | 2 | 6,822 | 12 |
COL4A2‐AS2 | 2 | 2 | 1,062 | 75 |
EP300 | 2 | 2 | 7,245 | 11 |
KANK4 | 2 | 2 | 2,988 | 27 |
KDM6A | 2 | 2 | 4,227 | 19 |
MLL3 | 2 | 2 | 14,907 | 5 |
PLXNB1 | 2 | 2 | 6,408 | 12 |
PRKD1 | 2 | 2 | 2,763 | 29 |
RHOA | 2 | 2 | 582 | 137 |
RP1L1 | 2 | 2 | 7,203 | 11 |
SF3B1 | 2 | 2 | 3,915 | 20 |
SKP2 | 2 | 2 | 1,275 | 63 |
TTN | 2 | 2 | 100,272 | 1 |
ZNF292 | 2 | 2 | 8,172 | 10 |
APOA1 | 2 | 1 | 804 | 100 |
BRD1 | 2 | 1 | 3,570 | 22 |
HBS1L | 2 | 1 | 2,055 | 39 |
R3HDM2 | 2 | 1 | 3,033 | 26 |
SPHKAP | 2 | 1 | 5,103 | 16 |
Gene . | Number mutations (N = 396) . | Number tumors (N = 25) . | Gene size . | Mutations/Mb . |
---|---|---|---|---|
NOTCH1 | 4 | 3 | 7,668 | 21 |
MLL2 | 3 | 2 | 16,614 | 7 |
ABCA4 | 2 | 2 | 6,822 | 12 |
COL4A2‐AS2 | 2 | 2 | 1,062 | 75 |
EP300 | 2 | 2 | 7,245 | 11 |
KANK4 | 2 | 2 | 2,988 | 27 |
KDM6A | 2 | 2 | 4,227 | 19 |
MLL3 | 2 | 2 | 14,907 | 5 |
PLXNB1 | 2 | 2 | 6,408 | 12 |
PRKD1 | 2 | 2 | 2,763 | 29 |
RHOA | 2 | 2 | 582 | 137 |
RP1L1 | 2 | 2 | 7,203 | 11 |
SF3B1 | 2 | 2 | 3,915 | 20 |
SKP2 | 2 | 2 | 1,275 | 63 |
TTN | 2 | 2 | 100,272 | 1 |
ZNF292 | 2 | 2 | 8,172 | 10 |
APOA1 | 2 | 1 | 804 | 100 |
BRD1 | 2 | 1 | 3,570 | 22 |
HBS1L | 2 | 1 | 2,055 | 39 |
R3HDM2 | 2 | 1 | 3,033 | 26 |
SPHKAP | 2 | 1 | 5,103 | 16 |
Abbreviation: Mb, megabase.
aIncludes point mutations and small indels.
Consistent with previous findings, several genes with well-known roles in chromatin regulation were mutated in multiple tumors: MLL2, MLL3, EP300, SMARCA2, SMARCC1, and KDM6A (Table 1 and Supplementary Table S2). The proportion of truncating mutations (nonsense codons, splice-site alterations, or out-of-frame insertions and deletions) out of the total number of nonsilent mutations in these genes was high (6 of 11), significantly greater than expected by chance (P = 3.8 × 10−6, randomization-based test). Furthermore, MLL2 and EP300, when considered individually, had a significantly higher proportion of truncating mutations than expected by chance (P = 0.008 for MLL2 and P = 0.01 for EP300). This finding is consistent with the hypothesis that several of these genes played an important role in the cancers in which they occurred.
Copy-number variation
We identified 41 copy-number variations (CNV) in 12 of the 25 tumors sequenced, with a range of 0 to 9 CNVs per tumor (Table 1; Supplementary Table S3). The most common CNVs included amplification of genes on 7p14.1 (five tumors) and 14q11.2 (four tumors). Both cytobands contain T-cell receptor genes. Several deletions of 10q26.3, containing DUX4 and DUX4L (DUX4-like) homeobox genes, were also observed (in three tumors).
Rearrangements
We observed a total of 253 chromosomal rearrangements in the 25 tumors (median 7 per tumor, range 0 to 42 rearrangements per tumor), including 118 interchromosomal (median 3 per tumor) and 135 intrachromosomal (median 3 per tumor) rearrangements (Table 1 and Supplementary Table S4; Fig. 1 and Supplementary Fig. S1). Chromosomal breakpoints were concentrated in several regions of the genome, particularly chromosome 6q (144 of 506 breakpoints, 28%) and 9p (71 breakpoints, 14%; Fig. 1).
Chromosomal rearrangements identified in 25 adenoid cystic carcinomas.
ACCs have been shown to harbor t(6;9)(q23.3;p22.3) translocations resulting in MYB–NFIB fusion gene products (4, 9, 11). We identified this translocation in 11 of 25 tumors (44%, 95% CI, 27%–63%; Table 1). We also identified five gene fusions involving NFIB that did not involve MYB, four of which are previously undescribed. These included t(6;9)(q23.3;p22.3) fusions involving MAP3K5–NFIB and t(8;9)(q13.1;p22.3) fusions involving MYBL1–NFIB in two tumors each (Table 1; Fig. 2). NFIB also recombined with three other genes on chromosome 6q, RPS6KA2, MYO6, and RIMS1. In contrast, MYB recombined with only one gene other than NFIB (Supplementary Table S4). Overall, NFIB translocations occurred in 15 of 25 tumors (60%, 95% CI, 41%–77%). Three tumors with novel NFIB fusions also harbored MYB–NFIB fusions. The NFIB breakpoint locations for both MYB–NFIB and other NFIB fusions were heterogeneous (Supplementary Table S5).
Circos plots for adenoid cystic carcinomas containing interchromosomal rearrangements in which NFIB recombines with gene other than MYB.
Circos plots for adenoid cystic carcinomas containing interchromosomal rearrangements in which NFIB recombines with gene other than MYB.
The most frequent intrachromosomal rearrangements resulted in deletions (N = 67 of 135 intrachromosomal rearrangements, 50%), with inversions (N = 47, 35%) and duplications (N = 21, 16%) less common (Supplementary Table S4). The median size of intrachromosomally rearranged segments was 4,074 kb (range, 2–137,666 kb; IQR, 167–29,034 kb). NFIB was involved in intrachromosomal rearrangements in three tumors (two deletions and one inversion), including one tumor (HN 324) that did not also have an NFIB translocation.
RNA sequencing fusion analysis and differential mRNA expression
To compare chromosomal rearrangements with mRNA fusion transcript expression, mRNA expression was assessed in the 17 tumors for which freshly frozen tumor tissue was available (see Materials and Methods). Two of the six tumors evaluated with non-MYB NFIB rearrangements expressed corresponding mRNA transcripts (RIMS1–NFIB and MYBL1–NFIB; Table 1). Of the nine tumors with MYB–NFIB rearrangement and available RNA, all were found to have MYB–NFIB mRNA transcripts. One additional tumor that did not have a MYB–NFIB rearrangement detected on whole-genome sequencing was found to have MYB–NFIB transcript expression (tumor HN 335 PT).
Differential expression of the genes in involved in NFIB fusions was also assessed. NFIB was overexpressed in tumors relative to normal tissues (P = 0.002; Fig. 3). NFIB expression was not significantly different in tumors with NFIB fusion genes compared with those without NFIB fusion genes (P = 0.91).
Heatmap of mRNA expression for genes involved in NFIB fusions in ACC tumors compared with normal tissues. Black outline indicates fusion gene DNA transcripts detected on whole-genome sequencing.
Heatmap of mRNA expression for genes involved in NFIB fusions in ACC tumors compared with normal tissues. Black outline indicates fusion gene DNA transcripts detected on whole-genome sequencing.
MYB expression was also significantly higher in tumors than in normal samples (P < 0.001). MYB was more highly expressed in tumors with MYB–NFIB fusion genes when compared with those without (P < 0.001). The only two tumors that did not express MYB were those with MYBL1–NFIB fusion genes (HN 333 PT and HN 320 PT). These two tumors had significantly lower MYB expression compared with the other tumors (P < 0.001), but had higher MYBL1 expression than other tumors (P = 0.07), although this latter difference was not statistically significant.
Genes other than MYB that were involved in NFIB fusions, including MYO6, MAP3K5, RPS6KA2, RIMS1, and MYBL1, did not have significantly different levels of expression in tumors compared with normal tissues (P > 0.13 for all).
Pathway analyses
IPA software was used to identify biological pathways that might be altered by the somatic mutations observed in the 25 ACC samples. Interestingly, IPA revealed significant involvement of the Rho family GTPase signaling pathway, with 11 (44%, 95% CI, 27%–63%) tumors containing somatic mutations in one or more members of this pathway (P = 3.9 × 10−6), including RHOA (two tumors), CDH2, CDH13, ARHGEF3, ARHGEF5, ARHGEF18, ACTB, and others (Fig. 4). The axon guidance signaling pathway was also highly disrupted [14 tumors (56%, 95% CI, 37–73%), P = 8.3 × 10−5], with mutations in PLXNB1 (two tumors), PRKD1 (two tumors), SLIT1, PLXNA1, SRGAP2, SRGAP3, ADAM2, ADAMTS5, ADAMTS13, SEMA3G (one tumor each), and others (Supplementary Fig. S2).
Rho Family GTPase signaling pathway alterations in adenoid cystic carcinoma.
IPA was also performed for all genes affected by any chromosomal rearrangement and/or somatic mutation, and identified disruptions of pathways involved in a variety of complex cellular processes that have previously been implicated in ACC, corroborating previous reports and validating our findings. These include protein kinase A (PKA) signaling (21 tumors), FGF signaling (15 tumors), Wnt/ß-catenin signaling (13 tumors), PI3K/AKT signaling (13 tumors), and Notch signaling (9 tumors; refs. 5–7, 25).
Discussion
Our whole-genome sequencing data and analysis reveal previously unreported NFIB rearrangements and NFIB overexpression in ACC tumors, as well as genetic alterations of the Rho family signaling pathway. We also corroborate previous reports of alterations in axonal guidance signaling, chromatin regulation, NOTCH1/2 somatic mutations, and involvement of key signaling pathways.
Our finding that NFIB rearrangements occur independent of MYB and are present in the majority of tumors, and that NFIB is significantly overexpressed in ACCs in comparison with normal tissues regardless of NFIB fusion status, suggests that NFIB may have a role in ACC oncogenesis independent of MYB. MYB is a transcription factor involved in proliferation, survival, and differentiation, with well-known oncogenic capabilities (26). Although MYB is overexpressed in 80% of ACCs, and is often, but not always, associated with the MYB–NFIB fusion, MYB overexpression does not appear to be a necessary step in the development of ACC (6, 10, 11, 27). NFIB is a member of the nuclear factor I transcription family reported to be involved in rearrangements in other tumor types, although the significance of these rearrangements has not been well described (28, 29). NFIB recombined with several genes in our cohort other than MYB, including two cases each with MYBL1, a transcriptional activator in the MYB family that has been implicated in pediatric low-grade gliomas (30) and a recently described fusion partner with NFIB in a subset of ACCs (12, 31). Notably, the two tumors with MYBL1–NFIB fusions uniquely did not express MYB at all, consistent with reports of a mutually exclusive relationship between MYB and MYBL1 alterations in ACC (12, 31). The significance of the other NFIB fusion partners is unclear. Although we found recurrent fusions of NFIB with MAP3K5 (also known as apoptosis signal-regulating kinase I, ASK1), a serine/threonine protein kinase with critical regulatory functions in the apoptotic pathway and reported roles in tumorigenesis (32), the other three fusions occurred in only one tumor each; few were confirmed on mRNA expression analysis. None of these genes were identified in other recent reports of novel NFIB fusion partners in ACC (5, 12). Interestingly, three of the 25 tumors in our study, and one of the five reported by Ho and colleagues, contained NFIB fusions both with MYB and with a different gene.
The overexpression of the NFIB mRNA in ACC has not been previously reported to our knowledge. This did not appear related to the presence of NFIB fusions, implicating alternative mechanisms of NFIB dysregulation. NFIB is reported to behave as an oncogene in small-cell lung cancer (33) and is overexpressed in estrogen receptor–negative breast cancer (34). It is conceivable, therefore, that there is a role for NFIB overexpression in ACC oncogenesis. Taken together with the high prevalence of NFIB fusion genes, this constellation of findings suggests that NFIB may contribute more significantly to ACC biology than has been previously appreciated. Further studies are necessary to evaluate the functional importance and potential therapeutic implications of NFIB alterations in ACC (5, 12).
The Rho signaling pathway was significantly affected by somatic mutations in our cohort, and this has not been previously reported in ACC. The Rho GTPase family, including the Rho and Rac subfamilies, is involved in many aspects of carcinogenesis (35). In particular, Rho signaling is a major regulator of motility, cytoskeleton structure, and adhesion in cancer cells, facilitating the tumor cell plasticity that allows for adaptation to and invasion of diverse tumor microenvironments (36, 37). Rho protein expression and activity is frequently observed in human malignancies (38). Although mutations of Rho GTPase proteins are rare (35, 36, 39), a recurrent RHOA somatic mutation was recently identified in angioimmunoblastic T-cell lymphoma. Interestingly, we found two different mutations in RHOA in two of the 25 ACC tumors analyzed in this study, and pathway analysis revealed significant disruption of the Rho pathway overall. This suggests a potential role for Rho signaling in ACC, but additional functional and genomic work is required to solidify this conjecture.
The known predilection of ACC for perineural invasion, and the association of PNI with disease recurrence (2, 3), add importance to our finding that the axonal guidance signaling pathway is significantly altered in ACC. Somatic mutations of genes involved in axonal guidance signaling (including NTNG1, SEMA3G, and SEMA5A) were also previously noted by Ho and colleagues (5), and an identical SEMA3G nonsynonymous substitution (474S>P) was observed in our cohort. Interestingly, significant disruptions of axonal guidance signaling pathways have also been noted in pancreatic adenocarcinoma, another malignancy characterized by frequent PNI that portends poor prognosis (40, 41).
Our findings also confirmed several observations made in ACC exome sequencing studies. Somatic mutation of NOTCH1, NOTCH2, and other Notch pathway molecules has been previously reported in ACC (5–7), and indeed NOTCH1 was the most commonly disrupted gene in our cohort (12% of tumors). Notch signaling has pleiotropic context-dependent effects on cell differentiation, survival, and growth and is disrupted in many human malignancies (42, 43). NOTCH1 alterations can result in either tumor suppression or oncogenesis, depending on the tumor type (44–46). Additional characterization of Notch signaling alterations in ACC is indicated, particularly because Notch signaling is a potentially targetable pathway (42). Our findings also corroborate a high prevalence of somatic mutations in genes related to chromatin remodeling, including MLL2, MLL3, and EP300, among others. The finding that chromatin regulators are consistently altered in ACC supports an important role for epigenetic regulation of gene expression in ACC oncogenesis (5, 6).
We observed recurrent amplifications of two loci containing T-cell receptor genes; however, these findings should be interpreted with caution as they may be an artifact of T-cell receptor gene rearrangement (47).
Our increasing understanding of ACC's molecular characteristics holds future promise for both more sophisticated application of existing chemotherapeutic drugs and design of new ones. In particular, the MYB–NFIB fusion is an appealing target in that it is highly prevalent in and specific for ACC (48). Our data implicate that targeting NFIB alone may also be of high yield. Rho signaling has already been the subject of research as a target of precision anticancer therapy given its role in many human cancers (49), and axonal guidance signaling has also been proposed as a viable therapeutic target (50). These pathways may be worth investigation in the context of ACC (50).
This whole-genome sequencing study highlights new pathways that may contribute to the carcinogenesis of ACC and suggests novel therapeutic areas that may be relevant to the management of ACC. Importantly, we also confirm observations from several previous sequencing studies. Further research will be required to elucidate the functional significance of our findings, which ultimately may contribute to the development of targeted, effective therapeutics for long-term control of this rare but aggressive disease.
Disclosure of Potential Conflicts of Interest
L.D. Wood is a consultant at Personal Genome Diagnostics. N. Papadopoulos has ownership interest (including patents) in Personal Genome Diagnostic, Inc., and PapGene, Inc.; is a consultant/advisory board member for Sysmex-Inostics, Inc.; and has licensed inventions through Johns Hopkins University. K.W. Kinzler has ownership interest (including patents) in PapGene Inc. and Personal Genome Diagnostics and is a consultant/advisory board member for Sysmex-Inostics, PapGene Inc., and Personal Genome Diagnostics. B. Vogelstein has ownership interest (including patents) in PGDx and PapGene Inc., is a consultant/advisory board member for Personal Genome Diagnostic, PapGene, Inc., Sysmex Inostics, Inc., and Morphotek, Inc. and has licensed inventions through Johns Hopkins University. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: N. Papadopoulos, K.W. Kinzler, B. Vogelstein, P.K. Ha, N. Agrawal
Development of methodology: M. Sausen, P.K. Ha, N. Agrawal
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Sausen, J.A. Bishop, L.D. Wood, C. Fakhry, K.W. Kinzler, P.K. Ha, N. Agrawal
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.M. Rettig, C.C. Talbot Jr, M. Sausen, S. Jones, J.A. Bishop, C. Tokheim, N. Niknafs, R. Karchin, E.J. Fertig, S.J. Wheelan, L. Marchionni, M. Considine, C. Fakhry, K.W. Kinzler, P.K. Ha, N. Agrawal
Writing, review, and/or revision of the manuscript: E.M. Rettig, C.C. Talbot Jr, M. Sausen, S. Jones, J.A. Bishop, L.D. Wood, E.J. Fertig, C. Fakhry, N. Papadopoulos, K.W. Kinzler, B. Vogelstein, P.K. Ha, N. Agrawal
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.M. Rettig, P.K. Ha, N. Agrawal
Study supervision: K.W. Kinzler, P.K. Ha, N. Agrawal
Acknowledgments
The authors thank the patients for their courage and generosity and Dr. Srinivasan Yegnasubramanian and the Next Generation Sequencing Center at the Johns Hopkins Sidney Kimmel Comprehensive Cancer Center, supported by NIH grant P30 CA006973, for their support with the high-throughput RNA sequencing experiments.
Grant Support
This work was supported by the Adenoid Cystic Carcinoma Research Foundation (N. Agrawal and P. Ha), NIH/NIDCR grant DE023218 (N. Agrawal), NIH/NIDCR Specialized Program of Research Excellence grant DE019032 (N. Agrawal), the Virginia and D.K. Ludwig Fund for Cancer Research (N. Agrawal, N. Papadopoulos, K. W. Kinzler, and B. Vogelstein), The Sol Goldman Center for Pancreatic Cancer Research (N. Papadopoulos, K.W. Kinzler, and B. Vogelstein), NIH/NIDCR R01-DE023227 (P. Ha), and NIH/NIDCR grant T32 DC000027 (E. Rettig).
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.