Whole-Genome Sequencing of Common Salivary Gland Carcinomas: Subtype-Restricted and Shared Genetic Alterations

Abstract

Purpose:

Salivary gland carcinomas (SGCs) are pathologically classified into several widely diverse subtypes, of which adenoid cystic (ACC), mucoepidermoid (MEC), and salivary duct (SDC) carcinomas are the most commonly encountered. A comparative genetic analysis of these subtypes provides detailed information on the genetic alterations that are associated with their tumorigenesis and may lead to the identification of biomarkers to guide tumor-specific clinical trials.

Experimental Design:

Whole-genome sequencing of 58 common SGCs (20 ACCs, 20 SDCs, and 18 MECs) was performed to catalogue structural variations, copy number, rearrangements, and driver mutations. Data were bioinformatically analyzed and correlated with clinicopathologic parameters, and selected targets were validated.

Results:

Novel and recurrent type-specific and shared genetic alterations were identified within and among 3 subtypes. Mutually exclusive canonical fusion and non-fusion genomic alterations were identified in both ACC and MEC. In ACCs, loss of chromosome 12q was dominant in MYB or MYBL1 fusion-positive tumors and mutations of NOTCH pathway were more common in these fusion-negatives. In MECs, CRTC1-MAML2 fusion-positive tumors showed frequent BAP1 mutation, and tumors lacking this fusion were enriched with LRFN1 mutation. SDCs displayed considerable genetic instability, lacked recurrent chromosomal rearrangements, and demonstrated non-overlapping TP53 mutation and ERBB2 amplification in a subset of tumors. Limited genetic alterations, including focal amplifications of 8q21-q23, were shared by all subtypes and were associated with poor survival.

Conclusions:

This study delineates type-specific and shared genetic alterations that are associated with early phenotypic commitment and the biologic progression of common SGCs. These alterations, upon validation, could serve as biomarkers in tumor-specific clinical trials.

INTRODUCTION

Salivary gland carcinomas (SGCs) comprise several morphologically and clinically heterogeneous subtypes, of which adenoid cystic carcinoma (ACC), mucoepidermoid carcinoma (MEC), and salivary duct carcinoma (SDC) are the most commonly encountered, representing 70% of diagnosed salivary gland carcinomas (1,2). Each of these malignancies displays variable cellular, phenotypic, genetic, and clinical characteristics. The primary therapy of salivary neoplasms is complete surgical excision, with or without post-operative radiotherapy (2,3). However, patients with advanced non-resectable primary, recurrent, and metastatic disease have limited therapeutic options available and are treated empirically (3). Several conventional and targeted therapeutic trials have been conducted on patients with advanced SGCs, but they were unsuccessful because of the inclusion of patients with different tumor subtypes and the lack of biomarkers to guide patient stratification (4–8).

Comparative molecular profiling allows for detailed genomic analysis, which may lead to the characterization of early type-specific and common progression associated biomarkers of potential clinical utility. We previously postulated that salivary carcinoma subtypes evolved from the neoplastic transformation of non-committed progenitors in different segments of the salivary gland duct-acinar structure (9,10). In that model, ACC evolves from the terminal duct and is composed of dual inner ductal epithelial and outer myoepithelial cells, while both MEC and SDC are derived from ductal epithelial cells. This tenet has empirically been sustained by the distinctive morphologic, biologic, genetic, and clinical differences among these subtypes (11–19). Experimental validation of this concept, however, has been hindered by the lack of defined pre-malignant precursors and faithful animal models for each tumor type. Although studies of individual subtypes have characterized recurrent genetic rearrangements and novel genetic events in ACCs, MECs, and SDCs (11–19), the temporal evolution and tumor context specificity of these findings are uncertain.

We performed whole genomic sequencing of 3 major SGCs (ACC, MEC, and SDC) to identify common and distinctive genetic alterations that were associated with their tumorigenesis and molecular targets and correlated with clinicopathologic factors.

MATERIALS AND METHODS

Primary SGCs

In a search of the Salivary Gland Tumor Biorepository database at The University of Texas MD Anderson Cancer Center (Houston, Texas) for all major SGCs, we identified 60 tumors (20 each of ACC, MEC, and SDC) with fresh-frozen tumor and matching non-tumorous salivary gland tissues (the majority of these tumors were used in our previous subtype studies (14,20–24)). All specimens were harvested from consented patients who presented at MD Anderson through an IRB-approved protocol. Clinical, pathologic, and follow-up data were extracted from pathology reports and patients’ medical records.

Whole Genome Sequencing and Data Processing

Fifty-eight salivary gland tumors (20 ACC, 20 SDC, and 18 MEC [2 cases of MEC failed quality control criteria]) and matched normal salivary gland tissue (Supplementary Table S1) were qualified for the analysis. Whole genome sequencing of extracted genomic DNAs was conducted using the Complete Genomics (CG) Analysis Platform (CGA™ Platform), which includes biochemical analysis, instrumentation, and software (25). In brief, the technology is based on self-assembling DNA nanoballs, their binding into nanoarrays, and sequencing by ligation. The sequencing average depth was approximately 40x, with a median gross mapping yield of 311 Gb. Because of the unique nature of the platform, a tailored sequencing data analysis was performed using Complete Genomics Analysis (CGA) tools (version 1.8.0) (26) integrated into CG Pipeline 2.2. The resulting sequencing reads were aligned to the reference genome (NCBI Build 37), and variants were called and scored using a local de novo assembly approach (25). The following databases were used for the initial annotation of the produced variants by CGA tools: NCBI build 37.2, dbSNP build 137, COSMIC v61, and miRBase version 19.

Single nucleotide variations (SNPs) and small InDels were called using CGA Tools on the basis of the local reassembly process; individual differences were evaluated between the assembled genome and the reference using the confidence score to call the variations. Copy number variations (CNVs) were called on the basis of depth of coverage. The coverage values were averaged, corrected for GC bias, and then normalized. A hidden Markov model was used to determine the copy number. Structural variation (SV) calls were based on an analysis of the local assemblies, namely when DNA nanoballs had an unexpected length or abnormal orientation.

Somatic Variation Calling

The called variations were considered somatic if they were found in the tumor but not in its matched normal sample. The somatic events were identified by the CGA tool “calldiff” using a somatic score and a rank. The score indicated the likelihood that the variant was present in the tumor only, and the somatic rank estimated the rank of the mutation among all true variants for the specific classification of the variant.

Copy Number Alterations

Focal copy number alterations for each subtype were determined using GISTIC 2.0 (27). To infer biologic processes impacted by CNVs in our cohort, we selected genes that were deleted or amplified in at least 5 tumors to ensure statistical reliability using Fisher’s exact test. This ensured the specificity of a given variation in a subtype with a probability of less than 99.99%. Amplified and deleted genes were further analyzed to identify significant overlap with known gene sets from Molecular Signatures Database v6.2, including hallmark, positional, and curated gene sets from the database (28,29). HUGO symbols of genes in the amplified/deleted loci were used as the input for the analysis. The significance of the overlap was evaluated by the p-value, calculated from the hypergeometric distribution, considering the number of genes in 1) the query set, 2) the known gene set, 3) the intersection of the query set with a known gene set, and 4) the total number of HUGO symbols. The FDR q-value was calculated using multiple hypothesis testing according to Benjamini and Hochberg. The threshold for the FDR q-value was set to 0.05. To identify potential driver genes, we compared the selected amplified and deleted genes with known tumor suppressor genes (TSGs) and oncogenes (OGs) downloaded from COSMIC Cancer Gene Census (COSMIC v91).

Germline Variations

Germline SNVs and small InDels were called if found in both tumor and matching normal samples. To determine the potential predisposition genes for SGC, we filtered germline variations to select for frameshift, insertion, deletion, nonsense, and missense non-synonymous variations. To identify germline variations that could be associated with SGCs and to minimize artifacts, we filtered variations that were frequent in our cohort (allele frequency > 0.10) but very rare in a healthy population (allele frequency < 0.01). To search for the reported rare variations, we updated annotations of the germline variants using ANNOVAR (version 2018Apr16) (30) and selected those variants that were annotated by the Exome Sequencing Project or by the Exome Aggregation Consortium database (v3) (31). The ratio of allele frequency in the study cohort to the population frequency was used to select top candidates for SGC predisposition.

Mutational Signatures

Mutational signatures were predicted by the R libraries “Somatic Signatures” and “deconstruct Sigs” (32) using somatic exonic SNVs generated for each sample as input. To use the libraries, annotation of the mutations was converted into VRanges objects using the R library “VariantAnnotation” (33). Known mutational signatures (v2) were downloaded from COSMIC (34).

Fusion Gene Transcript

Fusion genes were identified using the `junctions2events` CGA tools to align the fusion sequences of left and right junctions to the reference genome (Human NCBI Build 37). In-frame fusions with genomic annotations were identified using Ingenuity Variant Analysis software (QIAGEN). Canonical fusion genes (MYB-NFIB, MYBL1-NFIB, and CRTC1-MAML2), detected by whole genomic sequencing, were matched with those identified in previous studies, where fusion gene products by RT-PCR or 3’RACE were validated by Sanger sequencing (14,20–22).

Immunohistochemical Analysis of ERBB2

Immuno-stained tumor slides of ERBB2 expression were retrieved and reassessed independently of original scoring (24,35). An additional IHC analysis of previously untested tumors was performed as previously described (24,35). In brief, 4-μm-thick unstained sections were prepared from all SDC tissue blocks and stained using Autostainer Link 48 (Dako), according to the manufacturer’s instructions. Sections were incubated with primary anti-ERBB2 (HER2) mouse monoclonal (e2–4001, Thermo Scientific, 1:300) antibody, and then a secondary antibody was applied.

ERBB2 protein expression was scored in a binary fashion into high and low expression. High expression was scored as strong continuous membranous staining, with and without cytoplasmic expression. Faint cytoplasmic or membranous and heterogeneous lack of staining was scored as negative.

Statistical Analysis

Differences in categorical clinicopathologic characteristics among the 3 tumor types were evaluated by Pearson’s chi-squared test (‘chisq.test’ in R). The Kruskal-Wallis test, followed by Dunn’s multiple comparison test, was used to evaluate differences in quantitative clinicopathologic characteristics (age and tumor size).

The prognostic value of frequent somatic variations, the total number of variations, and clinical factors on the patients’ overall survival duration was estimated by the Cox proportional hazards regression model (36) using ‘coxph’ function in R library “survival”. Optimal cutpoints for continuous variables were determined on the basis of the maximally selected rank statistics from the ‘maxstat’ R package using function ‘surv_cutpoint’. Further categorization of the variable according to cutpoints was performed using function ‘surv_categorize’. The obtained results were visualized using Kaplan-Meier plots. A p-value < 0.05 was considered statistically significant.

RESULTS

Clinicopathologic Characteristics

Supplementary Table S2 presents the clinicopathologic characteristics of the cohort (total 58 tumors: 20 ACCs, 20 SDCs, and 18 MECs). The median age of patients was 55.5 years (range, 13 to 89 years; ACC, 50.0 years; MEC, 62.5 years; and SDC, 61.0 years). Patients with ACC were significantly younger than were those with MEC and SDC (p-values = 0.02 and 0.005, respectively, Supplementary Fig. S1). The tumor size ranged from 1.2 to 9.5 cm, with a mean of 3.5 cm (ACC, 4.0 cm; MEC, 3.0 cm; and SDC, 3.0 cm), and patients with ACC had larger tumors than did those with MEC and SDC (p-values = 0.03 and 0.02, respectively, Supplementary Fig. S1). Patients comprised 34 males (ACC, 13; MEC, 7; and SDC, 14) and 24 females (ACC, 7; MEC, 11; and SDC, 6). Morphologically, tubular and cribriform patterns were found in 15 ACCs and a solid component was present in 5 ACCs. MECs were comprised of low/intermediate grade (11 cases) and high grade (7 cases). Patients with SDCs were mainly stages III and IV, which was statistically different from ACCs and MECs (p-value = 0.041, Pearson’s chi-squared test [Supplementary Table S2]).

Somatic Alterations in SGCs

Fig. 1A-C displays somatic alterations among all salivary carcinoma subtypes. Tumor and matching salivary tissues were sequenced with a median coverage of 40x. On average, ~97% of the whole genome (tumor or adjacent normal sample) was called with median of 3,509,904 SNVs, 388,962 insertions, 374,383 deletions, and 407 CNVs (Supplementary Table S3-S4). SDCs displayed a higher incidence of SVs and amplified segments than did ACCs and MECs. Total SV events were correlated with the number of somatic CNVs (Spearman rank correlation = 0.687, p-value < 0.001) and exonic SNVs (Spearman rank correlation = 0.54, p-value < 0.001). Pairwise positive associations were found between amplified and deleted segments and the number of exonic mutations and SVs (Supplementary Table S5). Number of genes with somatic Mobile Element Insertions (MEI) (Supplementary Fig. S2 and Supplementary Table S6) was significantly higher in SDCs than in ACCs and MECs (Fisher test, p-values = 0.004 and = 0.009, respectively). The RHOD (ras homolog family member D) gene was significantly associated with MEI events (10% of all tumors).

Figure 1. Incidence of somatic variations in SGC.

(A) A higher number of structural variations (SVs) was found in SDC than in ACC and MEC. (B) Copy number variations (CNVs) show a higher incidence of amplified segments in SDC than in ACC or MEC. (C) SDC shows a relatively high incidence of variable single nucleotide variations (SNVs) in comparison to that in ACC and MEC. Lower grit plot highlights previously identified fusion genes and ERBB2 expression results. ACC = adenoid cystic carcinoma, MEC = mucoepidermoid carcinoma, SDC = salivary duct carcinoma.

Copy Number Variations (CNVs)

Hierarchical clustering and a Gene Set Enrichment analysis yielded tumor type–specific and a small number of shared genomic gain and loss within and among subtypes (Fig. 2, Tables 1 and 2, and Supplementary Table S7). The somatic copy number profile showed gains (> +1) on chromosomes 1q, 5p, 7p, 7q, 8q, 17q, 19q, 20q, and 22q regions and losses (≤ −1) at chromosomes 1p, 9p, and 12q regions among all subtypes (Fig. 2). A total of 1727 amplified and deleted genes was identified in different tumors (Supplementary Table S7). SDCs had the highest number of amplified genes, with 247 genes per tumor, followed by MECs with 140 genes and ACCs with 85 genes. ACCs showed the highest incidence of gene deletion (97 per tumor), followed by SDCs with 44 genes and MECs with 28 genes.

Figure 2. Hierarchical clustering of commonly amplified and deleted chromosomal regions of cancer genes.

Horizontal rows represent clustered genes, sorted by chromosome/loci on the left, and perpendicular columns represent tumor samples of each tumor subtype (top colored bar). Annotated genes are placed on the left side of the figure. SDCs and MECs displayed frequent gene amplification and a low incidence of genetic deletions compared to ACCs. Subtype-specific amplifications of the 17q12 region in SDC are noted. Deletion of chromosome 12q12-q13 region was detected in 5 fusion-positive ACCs. Of note, 2 MECs and 1 SDCs were omitted from the heatmap because of the lack of CNVs of the selected genes. ACC = adenoid cystic carcinoma, MEC = mucoepidermoid carcinoma, SDC = salivary duct carcinoma.

TABLE 1.

Shared Genetic Alterations among Common Salivary Carcinomas

Factor	Tumor type			Total
	ACC	MEC	SDC
Fusion gene
HFM1-RYR2	1	2	3	6
FSIP1-BAZ2A	2	2	1	5
SNVs/InDels
CIC	2	3	3	8
MUC16	2	4	2	8
KCNQ1OT1	3	2	4	9
NEAT1	2	2	3	7
CNVs
Loss
9p21	2	5	2	9
Gain
8q11-q13	1	3	4	8
8q21-q24	2	3	7	12
19q13	1	2	3	6
20q13	2	1	3	6

ACC: adenoid cystic carcinoma, MEC: mucoepidermoid carcinoma, SDC: salivary duct carcinoma, SNV: single nucleotide variation, CNV: copy number variation.

TABLE 2.

Tumor-Specific Genetic Alterations in Adenoid Cystic, Mucoepidermoid, and Salivary Duct Carcinomas

Tumor	Genetic alterations
	Fusion	SNVs/inDels	CNVs
ACC	MYB-NFIB MYBL1-NFIB ENOX1-TYRO3	NOTCH1 SPEN	12q12–13 (del)
MEC	CTRC1-MAML2 CACNA1B-NBPF10 EWSR1-ATF1	LRFN1	NA
SDC	NA	TP53 TPTE2P1 MALAT1	17q12 (amp)

SNV: single nucleotide variation, CNV: copy number variation, ACC: adenoid cystic carcinoma, MEC: mucoepidermoid carcinoma, NA: no alterations, SDC: salivary duct carcinoma.

To identify potential driver events, we compared amplified/deleted genes in at least 5 tumors with known driver genes in the Cancer Gene Census of COSMIC (V91). We identified 27 recurrently altered genes, including 3 TSG deletion and 13 OG amplifications (Supplementary Table S8). Shared amplification of the chromosome 8q21-q24 region harboring the OGs of HEY1, UBR5, CDH17, and MYC was found in 12 tumors (7 SDCs, 3 MECs, and 2 ACCs [Table 1]). The chromosome 8q13 region, which encodes OGs of PREX2 and NCOA2, was amplified in 8 tumors (4 SDCs, 3 MECs, and 1 ACCs). In addition, amplification of chromosome 19q13 (3 SDCs, 2 MECs, and 1 ACC) and 20q13 (3 SDCs, 2 ACCs, and 1 MEC) was shown in all subtypes. Shared deletions at the 9p21 region housing the CDKN2A (TSG) were found in 8 tumors ([14%] 5 MECs, 2 ACCs, and 1 SDC), including homozygous deletion of this region in only 1 MEC (MEC15).

Amplification of the chromosome 17q12 region that harbors the ERBB2 (15) and NEUROD2 (neurogenic differentiation factor 2) (37) genes were found in 5 SDCs with known overexpression of ERBB2 protein. Deletion of the 12q12-q13 region (38) housing the NAB2 (TSG), human epidermal growth factor receptor 3 (ERBB3), and several keratin-encoding genes (types I and II) was found in 5 ACCs with canonical fusions only.

Genetic Rearrangements

Figure 3 represents the incidence of fusion gene that was tumor-type specific or shared in all subtypes. Known gene fusions CRTC1-MAML2 and MYB/MYBL1-NFIB (Table 2), which had been previously identified in all MECs and ACCs by targeted molecular techniques, were confirmed (14,20–22,39). In total, 127 gene fusions were detected in at least 1 tumor; of these, 10 were recurrent in more than 2 tumors of a given subtype (Supplementary Fig. S3 and Supplementary Table S9). No specific recurrent gene fusion was identified in any other SDC. Two fusion genes, including HFM1-RYR2 and the FSIP1-BAZ2A, were detected in all types: HFM1-RYR2 genes in 3 MECs, 2 SDCs, and 1 ACC and ZFP37-UBR4 in 2 ACCs, 2 MECs, and 1 SDC. Two of these fusions were previously identified in other solid cancers: FSIP1-BAZ2A in breast cancer (40) and GPR128-TFG in cutaneous T-cell lymphoma (41). Two novel tumor-specific fusion genes were detected in canonical fusion-negative ACCs and MECs: CACNA1B-NBPF10 in 3 MECs and ENOX1-TYRO3 in 2 ACCs.

Figure 3. Integrative findings of genes with various somatic alterations.

The figure illustrates specific and common somatic alterations in ACC, MEC, and SDC. The alterations include gene fusions, single nucleotide variations (SNVs), small insertions and deletions (InDels), non-coding RNA (ncRNA), and gene amplifications (>5) are represented. Genes in SNVs/InDels/CNVs were included if alterations occurred in at least 3 tumors and the Cancer Gene Census genes (red text, COSMIC database). In addition to reported fusions, ENOX1-TYRO3 (ACC) CACNA1B-NBPF10 (MEC) were detected as a type-specific fusion. Although SDC had no recurrent fusion genes, this event occurred randomly. The most frequently mutated genes shared by subtypes were CIC and MUC16, and both had hotspot mutations (*, see Supplementary Fig. S4). Long non-coding RNAs (lncRNAs; NEAT1 and KCNQ1OT1) were also frequently mutated and shared in all subtypes. The most frequently altered subtype-specific genes were TP53 and ERBB2 in SDC, NOTCH1 and SPEN in ACC, and LRFN1 and BAP1 in MEC. Of note, all cases with ERBB2 amplification were validated as strong positive by IHC, and its locus (17q11-q12) was highly amplified (>5) along with some genes. Homozygous deletion of 9p21-p22 was identified in 1 case (MEC15, see Supplementary Table S7). ACC; adenoid cystic carcinoma, MEC; mucoepidermoid carcinoma, SDC; salivary duct carcinoma.

Single Nucleotide Variations (SNVs) and Insertions/Deletions (InDels)

Figure 3 demonstrates common single nucleotide substitutions and small InDels, copy number changes, and genetic rearrangements in all 3 SGC subtypes. Somatic SNVs and InDels (Supplementary Table S10) were higher in SDCs than in ACCs and MECs. The 6 most common subtype-specific mutated genes in at least in 3 tumors are presented in Table 2. These included the NOTCH1 and SPEN genes in ACC; the LRFN1 gene in MEC; and the TP53, TPTE2P1, and MALAT1 genes in SDC. Concurrent mutations of the ncRNAs and the TP53 gene were formed in 6 (30%) of the 20 SDCs (Fig. 3 and Supplementary Table S10). In addition, for the first time, all ERBB2-mutated SDCs showed concomitant gene amplification and high protein expression, while 1 MEC with ERBB2 mutation lacked gene amplification and expression. Three non-synonymous point mutations in the ERBB2 gene with a variant allele frequency ranging from 60% to 84% were identified, while 1 MEC had an ERBB2 mutation with an allele frequency of only 35%.

Genetic mutations shared by all subtypes were limited and included the transcription factor CIC (2 ACCs, 3 MECs, and 3 SDCs), MUC16 (2 ACCs, 4 MECs, and 2 SDCs), and the long non-coding RNAs, NEAT1 and KCNQ1OT1 (5 ACCs, 3 MECs, and 5 SDCs genes). Also, novel mutational hotspots in CIC (p.Q894H) and MUC16 (p.G12695V) genes were detected (Supplementary Fig. S4).

Mutational Signatures

Figure 4A displays the similarity of somatic base changes among different tumor types. A comparison of somatic base changes with those reported in other cancers (34) delineated 8 known mutational signatures (Fig. 4B). Five were associated with DNA repair pathways, including signature 3 (homologous recombination); signatures 6, 15, and 20 (mismatch repair); and signature 24 (nucleotide excision repair). Noticeably, these 5 mutational signatures representing DNA repair pathways showed a significant association with somatic single nucleotide alterations in all 3 tumor subtypes, with a notable increase of signature 24 in MECs (Supplementary Fig. S5).

Figure 4. Mutation signature and prognosis factors in salivary gland tumors.

(A) Frequencies of somatic base changes in SGC (ACC = adenoid cystic carcinoma, MEC = mucoepidermoid carcinoma, SDC = salivary duct carcinoma) demonstrate comparable profiles among all 3 tumor subtypes. (B) Heatmap of relative contributions of known mutational signatures in SGC. Blue color in rows of the heatmap shows the contribution of each signature in the mutational processes in each SGC sample, from 0 (not colored) to 0.6 (dark blue). All 3 tumor subtypes involve DNA repair processes (HR = homologous recombination, MMR = mismatch repair, and NER = nucleotide excision repair). (C) The Cox proportional hazards regression model (univariate analysis) revealed significant associations between a high incidence of structural variations and poor survival (left panel, p-value = 0.022, log-rank test). In addition, amplification of chromosome 8q21-q24 was significantly associated with poor survival (right panel, p-value = 0.022, log-rank test).

Germline Alterations

Common germline variations among our patients (frequency > 0.10), which are reported to be very rare in healthy populations (allele frequency < 0.01), were identified in 11 genes (Supplementary Fig. S6 and Supplementary Table S11) including 3 mucin-coding genes (MUC2, MUC12, and FCGBP) (42,43), TMEM52 (44), and HLA-B (45).

Correlative Analysis of Clinical Factors and Genetic Alterations

A univariate survival analysis showed that age (p-value < 0.0001, Supplementary Fig. S7A), total SVs (p-value = 0.022, Fig. 4C), 3 major types of SVs (deletion, tandem duplication, and interchromosomal alterations; p-values = 0.011, 0.012, and 0.035, respectively, Supplementary Fig. S8), amplification of chromosome 8q21-q24 region (p-value = 0.022, Fig. 4C), and CCDC58 gene mutation (p-value = 0.0079, Supplementary Fig. S9) were significantly associated with poor outcome. Interestingly, an analysis of the abundance of the molecular factors in different subtypes showed their enrichment in patients with SDCs (Supplementary Table S12). A survival analysis among patients with the 3 subtypes revealed a trend towards a better outcome for ACC patients than for SDC patients (p-value = 0.12, log-rank test, Fig. S7B) and intermediate outcome for patients with MEC.

DISCUSSION

We report type-specific and shared genetic alterations that are associated with the evolution and progression of common SGCs. The analysis also demonstrated considerable genetic differences between purely epithelial (MECs and SDCs) and dual epithelial and myoepithelial (ACC) cell–derived tumors. Epithelial-derived subtypes displayed frequent chromosomal gains and gene amplifications, while chromosomal deletions dominated tumors with dual epithelial/myoepithelial composition (ACC). These findings are concordant with previous genomic studies of individual tumor type (11–13,17,18) and lend empirical support to our segmental ductal derivation postulate (9). This is further highlighted by the low incidence of genetic alteration and protracted clinical behavior of ACC that could be attributed to the suppressive nature of myoepithelial cells (46–49).

Importantly, we report intra-subtype genetic heterogeneity within canonical gene fusion-positive and -negative ACCs and MECs. Deletions at the 12q12–13 region housing the ERBB3 gene (50) and the keratin type I and II genes (38) were detected in only canonical fusion-positive ACC, while alterations of the NOTCH pathway were limited to canonical fusion-negative ACC, which indicates solid and high-stage tumors (12,13,19,51–53). Analogously, mutations of LRFN1, CIC, and MUC16 were limited to CRTC1 fusion-negative MECs, while mutations of CCDC58, BAP1, KCNQ1OT1, and NEAT1 were only identified in CRTC1 fusion-positive MECs. These results underscore the involvement of mutually exclusive genomic pathways within canonical gene fusion-positive and -negative ACCs and MECs, implicating stochastic genetic pathways of equal differentiation consequence in the tumorigenesis of both entities. Interestingly, we also identified 2 novel fusion genes: ENOX1-TYRO3 in 2 canonical fusion-negative ACCs and CACNA1B-NBPF10 in 3 canonical fusion-negative MECs. The consistency and specificity of these fusion genes, however, are currently unknown and should await further validation. Nonetheless, the propensity for translocations in both ACC and MEC suggest the possibility of the non-random spatial proximity of certain chromosomes in early progenitor cells of these entities (54).

In contrast to both ACC and MEC, SDCs had broad genetic abnormalities, lacked recurrent chromosomal rearrangement, and had distinctive differential alterations of TP53 and ERBB2. These findings are concordant with those of previous studies of this entity (17). Interestingly, no alterations of the AR gene were found in both male and female tumors in spite of the high expression of AR protein, implicating the activation of this gene thorough epigenetic or post-translational modification in SDCs (23). We also report overexpression of the ERBB2 protein with and without amplification of this gene in SDC (23,55), indicating that both genomic and non-genomic mechanisms underlie ERBB2 overexpression in SDCs. This finding lends further support to the use of IHC in the evaluation of ERBB2 as a reliable method in patients with SDC. Our results, however, are at variance with those of mammary ductal carcinoma, most likely because of tumor or organ context differences (56). In addition, for the first time, we report mutually exclusive TP53 with ncRNA mutations that could lend to future biological sub-classification and therapeutic stratifications of this entity, similar to those of mammary ductal carcinomas (57). Our TP53 gene findings are concordant with those of previous studies of SDCs and underscore a significant role for this gene in a subset of this tumor (15–17).

Our study, for the first time, identified limited shared somatic alterations in high-stage tumors of all types, suggesting that they develop with progression irrespective of tumor type context. Shared hotspot mutations of the CIC and MUC16 genes, altered non-coding RNAs KCNQ1OT1 and NEAT1 genes, and CNVs at the chromosomes 8q21-q24 and 9p21 region housing the CDKN2A gene (58) were found, irrespective of tumor type. Of these alterations, amplification of chromosome 8q21-q24 housing the MYC gene was significantly correlated with poor outcome. We also observed an increased incidence of mobile element insertion at the RHOD gene that encodes for the Rho GTPase enzyme that is linked to focal adhesion and cancer cell invasion (59). The biologic and clinical implication of this are currently unknown. We also highlighted a potential association between mutational signature of DNA repair–related pathways and the development of salivary cancers (60). The dominant mutational signatures of DNA repair pathways in all three subtypes may suggest potential deficiencies of these pathways and induction of epithelial cells damage that are not well-protected by salivary mucins in the patients (61). Indeed, we found increased frequency of rare germline variations in 3 genes, MUC2 (62,63), MUC12, and FCGBP (42,43) in comparison to in the healthy population. These genes are linked to mucosal and salivary gland tissue homeostasis and defense against toxic and bacterial infection (42), suggesting that they play a role in the predisposition of certain individuals to SGC development. However, further studies of a large cohort and control subjects are needed to confirm these findings.

In conclusion, the results characterized type-restricted and shared genetic alterations that were associated with the development and progression of common salivary gland malignancies; we also identified genetic makers of potential clinical utility, upon validation.

Translational Relevance.

Salivary gland carcinomas (SGCs) are uncommon malignancies that are comprised of widely variable histopathologic subtypes, of which adenoid cystic, mucoepidermoid, and salivary duct carcinomas are the most often clinically encountered. Each of these SGC subtypes manifests distinctly different phenotypic, clinical, and molecular characteristics that reflect remarkable inter-tumor heterogeneity. Although the primary treatment for salivary tumors is surgery, patients with unresectable primary, recurrent, and metastatic disease have limited therapeutic options. We conducted whole genomic sequencing to characterize the molecular genetic events that are associated with the evolution, diversity, and progression of these uncommon salivary malignancies. Our whole genomic sequencing identified type-specific and shared genetic alterations to be associated with early phenotypic commitment and progression and delineated intra-subtype heterogeneity and potential molecular targets of biologic and therapeutic utilities to be validated.