Profiling of 149 Salivary Duct Carcinomas, Carcinoma Ex Pleomorphic Adenomas, and Adenocarcinomas, Not Otherwise Specified Reveals Actionable Genomic Alterations

Malignant salivary gland carcinomas (SGC) are rare cancers affecting less than 2,500 adults in the United States per year (1). Moreover, SGCs are heterogeneous tumors, with 24 distinct malignant histotypes recognized in the most recent World Health Organization (WHO) classification of salivary gland tumors (2). Although there are several unique salivary gland cancer histologies, such as mucoepidermoid carcinoma, adenoid cystic carcinoma (ACC), acinic cell carcinomas (AciCC), salivary duct carcinoma (SDC), and carcinoma ex pleomorphic adenoma (ca ex PA), a proportion of cases without specific histologic or immunophenotypic features are best assigned to the category adenocarcinoma, not otherwise specified (NOS; refs. 2, 3). Conventional and molecular cytogenetic analysis has identified recurring translocations in a variety of SGCs, including ACC (MYB-NFIB), mucoepidermoid carcinoma (MECT1-MAML2), mammary analogue–secretory carcinoma (MASC; ETV6-NTRK3), and hyalinizing clear cell carcinoma (HCCC; EWSR1-ATF1; refs. 4–7). However, little is known about the molecular features of SDC, adenocarcinomas, NOS, and ca ex PA. In general, conventional chemotherapy and radiation for SGCs has limited efficacy (8). Similarly, to date, targeted therapies in nonmolecularly preselected SGCs have been disappointing (8). HER2 (ERBB2) and the androgen receptor (AR) have been recognized as important drivers of SDC and adenocarcinomas, NOS (9, 10). Multiplex mutation or limited next-generation sequencing (NGS) analysis has identified TP53, activating PIK3CA, and RAS mutations as leading genetic alterations in SDC (11, 12). However, these studies have been limited by small mutation profiles (<50 genomic alterations), insensitive assays for low purity tumor samples, and/or small sample size. The genetic underpinnings of adenocarcinoma, NOS are largely unknown. Both adenocarcinoma, NOS and SDC are associated with a worse prognosis than other SGCs, heightening the need for novel, precise therapies (13).

In the following study, we used comprehensive genomic profiling (CGP) to survey a large group of clinically advanced salivary adenocarcinoma, NOS, SDC, ca ex PA, and unclassifiable salivary carcinoma, NOS to search for novel therapy targets and demonstrate the impact of biomarker-selected targeted therapy in selected cases.

Local site permissions were utilized for this study, and a full description of the methods can be found in the Supplementary Methods. SGC samples submitted for commercial CGP as salivary adenocarcinoma or SDC underwent pathologic assessment using hematoxylin and eosin (H&E) and review of prior IHC profiling. Images from each submitted tumor specimen were individually reviewed. Cases of other WHO SGC categories, including ACC, mucoepidermoid carcinoma, AciCC, epithelial-myoepithelial carcinoma, myoepithelial carcinoma, MASC, basal cell adenocarcinoma, HCCC, low-grade polymorphous adenocarcinoma, and squamous cell carcinomas were excluded. The resulting samples were then classified into the WHO categories SDC, adenocarcinoma, NOS, and ca ex PA, SDC. Carcinoma samples with a clinical history as an SGC that did not fall into a clear WHO grouping were classified as carcinoma, NOS. DNA was extracted from formalin-fixed and paraffin-embedded samples. Captured libraries were sequenced to a median exon coverage depth of 600× for up to 315 genes, and resultant sequences were analyzed for base substitutions, short insertions/deletions, copy number alterations (focal amplifications and homozygous deletions), and gene fusions/rearrangements, as described previously (14).

Clinically relevant genomic alterations were defined as those identifying anticancer drugs on the market or in registered clinical trials. Statistical analysis was performed with the Fisher exact test, with the level of significance set at P ≤ 0.05 using JMP (SAS).

Of the 149 patient samples included in this study 52 (34.9%) were salivary adenocarcinoma, NOS, 41 (27.5%) were SDC, 24 (16.1%) were ca ex PA, and 32 (21.5%) were carcinoma, NOS (Table 1). The male-to-female ratio differed by tumor type, with relatively more men diagnosed with SDC (87.8%) and adenocarcinoma, NOS (78.8%) (P < 0.001). The tumor source from which the sample was obtained did not differ significantly between groups with most samples taken from the parotid gland (32.2%), followed by a head and neck nonspecified site (12.1%), lung (12.8%), lymph node (10.1%), and salivary gland (8.7%; P = 0.442). The majority of tissue samples came from local/regional (94/149, 63%) versus metastatic sites (55/149, 36.9%).

A total of 590 GAs were found in 157 unique genes (Figs. 1 and 2; Table 2). One hundred and ninety-eight (33.6%) were base substitutions or short indels, 163 (27.6%) were amplifications, 44 (7.4%) were homozygous deletions, 23 (3.9%) were rearrangements or fusions, and 168 (28.5%) were gene truncations. The mean number ± SD of GAs per sample was 3.9 (2.8), with a difference across histotypes: adenocarcinoma, NOS (3.8 ± 3.1); SDC (3.6 ± 1.9); ca ex PA (3 ± 2), carcinoma, NOS (5.2 ± 4.1; P = 0.033). The majority of patients (142/149, 95.3%) had at least one GA identified. Clinically relevant GAs were identified in 117 (78.5%) patient samples.

There were many similarities in GA profiles among histotypes. All had GAs in the PI3K/AKT/mTOR signaling pathway, cyclin-dependent kinases (CKD), and RAS (Figs. 1 and 2; Table 3 and Supplementary Table S1). In the PI3K/AKT/mTOR pathway, GAs occurred most commonly in SDC [22 (53.6%)], followed by adenocarcinoma, NOS [19 (36.5%)], carcinoma, NOS [10 (31.2%)], and ca ex PA [4 (16.7%); P = 0.019]. There was a difference in CDK GAs among adenocarcinoma, NOS [18 (34.6%)], SDC [5 (12.2%)], ca ex PA [4 (16.7%)] and carcinoma, NOS [10 (31.2%); P = 0.043]. RAS family GAs appeared more frequently in SDC [11 (26.8%)] and adenocarcinoma, NOS [9 (17.3%)] than other histologies (<10%) but did not reach statistical significance (P = 0.054). Simultaneous GAs in the PI3K/mTOR pathway and RAS were more common in SDC [11 (26.8%)] than other histologies (P = 0.006). In fact each RAS mutation in an SDC was accompanied by PI3K/mTOR pathway activation. There was no difference in concurrent GAs in PI3K pathway and CDKs between histologies (P = 0.473).

Other notable GAs were identified. The most common GA in all tumor types involved TP53: adenocarcinoma, NOS [29 (55.7%)]; SDC [22 (53.7%)]; ca ex PA [11 (45.8%)]; carcinoma, NOS [19 (59.4%); P = 0.783]. ERBB2 GAs, including both amplifications and activating mutations, were also seen across histologies: adenocarcinoma, NOS [7 (13.5%)]; SDC [11 (26.8%)]; ca ex PA [7 (29.2%)]; carcinoma, NOS [6 (18.8%); P = 0.249]. NOTCH1 GAs were not seen in SDC but were observed in adenocarcinoma, NOS [4 (7.7%)], ca ex PA [2 (8.3%)], and carcinoma, NOS [7 (21.9%); P = 0.006. ARID1A GAs were observed in adenocarcinoma, NOS [4 (7.7%)], SDC [1 (2.4%)], ca ex PA [3 (12.5%)], and carcinoma, NOS [5 (15.6%); P = 0.182]. NF1 GAs were seen more common in SDC (7 [17.1%]) and carcinoma, NOS [6 (18.8)] than ca ex PA [0 (0%)] or adenocarcinoma, NOS [5 (9.6%); P = 0.036]. BRAF mutations were found in both adenocarcinoma, NOS [3 (5.8%)] and SDC [2 (4.9%)], but not other histologies. ETV6–NTRK3 translocations were seen only in 2 (3.8%) adenocarcinomas, NOS. Neither of these tumors was clearly suggestive of MASC tumors by light microscopy (Supplementary Fig. S1). One CCDC6–RET fusion was observed in an adenocarcinoma, NOS (1.9%), and two NCOA4–RET fusions were found in SDC (4.9%).

Despite their rarity and the retrospective nature of this study, examples of successful use of precisions therapies were identified. A 62-year-old man with a stage IVA (T4a N2b M0) HER2-positive, AR-positive parotid SDC who refused local therapy had a brisk partial response following 2 cycles of carboplatin, docetaxel, and trastuzumab, an anti-HER2 antibody (Fig. 3A). Two patients with NCOA4–RET translocations derived benefit from RET-targeted therapy. A 68-year-old man with a stage IVC (T4a M2c M1) AR-positive parotid SDC who had progressed following concurrent chemoradiation, combination chemotherapy, dual androgen deprivation, cetuximab, and everolimus had a chest wall lesion biopsied that harbored an NCOA–RET fusion. He was given cabozantinib, a tyrosine kinase inhibitor targeting RET and other kinases and had a dramatic improvement in his chest wall lesions and mediastinal adenopathy after approximately 10 weeks (Fig. 3). A 79-year-old man with a stage IVA (T2 N2b Mx) AR-positive parotid SDC had a neck lesion biopsied after progression following surgery, radiation, and dual androgen deprivation therapy. A NCOA4–RET fusion was identified and patient had improvement in his palpable neck relapse when treated with cabozantinib.

SGCs are uncommon, diverse malignancies with limited benefit from nontargeted systemic therapies. In this study, we identified that many patients with salivary gland adenocarcinomas, NOS, salivary duct carcinomas, ca ex PA, and carcinoma, NOS have genetic alterations that may allow for precision therapy selection. To our knowledge, this is the most comprehensive study examining the genetic profiles of these tumors.

This study identified unique GAs that may be exploited therapeutically in SGC. Consistent with the great heterogeneity in SGCs, the SGCs in this series had more actionable and nonactionable GAs than ACC, another SGC histotype (15). RET fusions were observed in 1 adenocarcinoma, NOS (CCDC6-RET) and two SDCs (NCOA4-RET). NCOA4-RET has been reported in 20% to 30% of papillary thyroid carcinoma cases and 1% of non–small cell lung carcinoma (NSCLC; refs. 16–20). RET fusions have not been reported in SGCs to our knowledge. NSCLC patients with KIF5B–RET fusion have benefited from the RET inhibitor cabozantinib, as have NSCLC patients with CCDC6–RET fusions treated with vandetanib, another RET tyrosine kinase inhibitor (21, 22). In an ongoing phase II study of cabozantinib, 5 responses have been achieved by NSCLC patients with RET fusions in their tumors (21). Similar to these other solid tumors, we found evidence of anticancer activity targeting RET in RET-activated tumors, as two patients with RET fusions in our retrospective series had tumor shrinkage with cabozantinib.

The PI3K/AKT/mTOR was frequently altered in both histotypes, particularly SDCs. PIK3CA encodes the catalytically active subunit of PI3K that is involved in cell growth, proliferation, differentiation, and survival (23). There are numerous PI3K inhibitors under investigation, and prior studies have shown that activation of the PI3K/AKT/mTOR pathway across different tumor types may predict responses to PI3K/AKT/mTOR pathway inhibitors (24–27). In SDC, a small series of patients with PI3K/AKT/mTOR–activating GAs demonstrated clinical improvement when they were treated with temsirolimus, an mTOR inhibitor (28). Prior studies demonstrated PIK3CA mutations occurred in 19% of SDC (11, 12), and they were reported in 7% of SGCs in COSMIC (29). In our datasets, we identified PIK3CA mutations in 19.2% of adenocarcinoma, NOS and 36.6% in SDC, a higher number than previously reported. Interestingly, although PIK3CA GAs were also identified in carcinoma, NOS, they were not identified in ductal ca ex PA. PTEN GAs were reported in 2% of salivary gland carcinomas in COSMIC (29), whereas we found them in higher numbers in adenocarcinoma, NOS (7.7%), SDC (12.2%), and ductal ca ex PA (12.5%). This study expands upon prior studies by identifying frequent GAs not just in PIK3CA, but AKT, PTEN, and RICTOR that may make patients candidates for PI3K, AKT, or mTOR inhibitors.

Frequent simultaneous GAs in the PI3K/AKT/mTOR pathway and the RAS family of genes were observed. In a series of multiple solid tumor histologies, paired mutations in KRAS were commonly observed in PIK3CA-mutated tumors (30). A similar pattern with PIK3CA and HRAS has been noted in SDC (12). Targeting both pathways may have a therapeutic advantage. For instance, colon cancer cells harboring both KRAS and PIK3CA mutations were resistant to PI3K/mTOR inhibition alone, although solitary PI3K inhibition could be overcome with combined MEK and PI3K blockade (31, 32). In NRAS-mutated melanoma, combined MEK and PI3K inhibition is more effective than blockade of either pathway alone (33). Thus, dual pathway inhibition may be worth evaluating in PIK3CA and HRAS-mutated tumors.

A novel finding in this study was the frequent alteration of a variety of CDKs in all histologies. CDKs are a diverse set of critical regulatory proteins responsible for cell-cycle transitions (34, 35). Recently, CDK GAs were reported in a wide variety of solid tumors at varying frequencies (0%–81%), where they were associated with poorer survival (36). In this study, CDKs were altered in more than 30% of adenocarcinoma, NOS and carcinoma, NOS. Because of diversity and complexity of CDK signaling, it has been difficult to develop CDK-targeted therapies (37). However, palbociclib, a CDK4/6 inhibitor, was recently shown to improve progression-free survival when combined with endocrine therapy in estrogen receptor–positive, HER2-negative breast cancer (38). SGCs harboring CDK GAs may be good candidates for trials targeting CDKs or their partners.

Other common and rare GAs were identified that may have therapeutic and/or diagnostic relevance. Consistent with prior studies, amplification of ERBB2 was commonly seen in SDC but less so in adenocarcinomas, NOS (12, 39, 40). In addition, we also identified a high number of ERBB2 GAs in ductal ca ex PA. New to this study was the identification of activating ERBB2 mutations in adenocarcinoma, NOS, SDC, and ductal ca ex PA. HER2-targeted therapies, such as trastuzumab, have previously demonstrated efficacy in ERBB2-amplified salivary gland cancers and contributed a partial response in our series (41, 42). Select GAs in genes with precision drugs currently commercially available included ROS1, MET, and BRAF, although BRAF GAs were seen less frequently than previously reported (11). BRAF inhibition was recently shown to be of benefit in one patient with a BRAF-mutated SDC (43). Interestingly, this study identified characteristic GAs in unexpected histologies. For instance, ETV6–NTRK3 translocations, a translocation characteristic of MASCs, were observed in 2 (3.8%) of tumors submitted as adenocarcinoma, NOS (5). The finding of this translocation in the adenocarcinoma, NOS may speak to the lack of consensus for the IHC profile for MASC tumors and suggest that CGP profiling may be of diagnostic benefit in histologically difficult cases (44). Finally, the carcinoma, NOS group in this study did not clearly cluster with any of the three other histotypes, yet they harbored frequent actionable GAs. These findings suggest CGP may help clarify diagnoses and identify GAs for therapeutic exploitation in difficult to diagnose cases.

This study has limitations. The greatest limitation is the lack of clinical correlations between identified GAs and disease characteristics or patients outcomes. As this was a retrospective evaluation of samples submitted for clinical care, data about tumor grade, cancer stage, response to therapies, and patient survival are not available. Another limitation is that, although the CGP platform used in this study covers a wide range of genes, there may be clinically relevant GAs that were not assessed. For instance, although the assay routinely identifies the MYB–NFIB translocation commonly seen in ACC, it does not assess for the MECT1–MAML2 translocation commonly identified in mucoepidermoid carcinoma (4, 6). Designing a CGP panel for salivary gland cancers, including MECT-MAML2, may be more clinically applicable to a broad range of tumors. Similarly, although the pathology reports and prior immunohistochemical staining patterns were reviewed, we could not perform additional/supplementary IHC characterization because specimens were submitted strictly for CGP analysis. For SDC, AR staining aids in the diagnosis and can be exploited therapeutically (9, 39, 45). Another limitation is that the tumors tested in this study may not fully represent the genetic spectrum of these diagnoses, as commercial CGP tends to be performed or more aggressive, advanced/metastatic tumors, and many salivary gland cancers are somewhat indolent (46). Moreover, CGP was performed on both locoregional disease and metastatic sites. There are no data in SGCs to demonstrate whether the GA profile of locoregional disease is representative of metastatic sites, or vice versa, although there appears to be high concordance in other tumors (47, 48). Finally, although this study demonstrates frequent actionable GAs, it does not address intratumoral heterogeneity and the relative importance of each GA in patient outcomes. Recent studies in renal cell and lung cancers have demonstrated that different clones have varying driver mutations whose importance can fluctuate over time (49, 50). Determining the inciting GA and relative importance of subsequent GAs in SGC is worth exploring in future studies. Despite these limitations, this study greatly contributes to the knowledge of the genetic drivers of these rare tumors.

In summary, deep genomic profiling with a comprehensive genomic profiling assay of salivary adenocarcinomas, NOS, salivary duct carcinomas, ca ex PA, and salivary carcinoma, NOS identified a high percentage of potentially actionable GAs that can influence therapy selection and direct patients to enter clinical trials using precision therapies. The GAs identified in this study are an important step to opening pathways for new therapeutic approaches in these notoriously difficult to treat cancers.

K. Wang, J.A. Elvin, D. Khaira, A. Johnson, S.M. Ali, M. Murray, J. Chmieckecki, R. Yelensky, D. Lipson, V.A. Miller, P.J. Stephens, and J.S. Ross have ownership interests (including patents) in Foundation Medicine. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Wang, J.S. Russell, J.A. Elvin, D. Khaira, S.J. Wong, V.A. Miller, J.S. Ross, D.W. Bowles

Development of methodology: K. Wang, R. Yelensky, D. Lipson, J.S. Ross, D.W. Bowles

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.S. Russell, J.A. Elvin, D. Khaira, T.A. Jennings, S.M. Ali, J.S. Ross, D.W. Bowles

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Wang, J.D. McDermott, D. Khaira, A. Johnson, S.M. Ali, J. Chmielecki, R. Yelensky, P.J. Stephens, J.S. Ross, D.W. Bowles

Writing, review, and/or revision of the manuscript: K. Wang, J.S. Russell, J.D. McDermott, J.A. Elvin, D. Khaira, S.M. Ali, M. Murray, C. Marshall, D. Washburn, S.J. Wong, V.A. Miller, P.J. Stephens, H.S. Serracino, J.S. Ross, D.W. Bowles

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.D. McDermott, M. Murray, J.S. Ross, D.W. Bowles

Study supervision: K. Wang, J.S. Ross, D.W. Bowles

Other (case report contribution to the content of the manuscript): D. Oldham

Other (minor contribution by detail writing and editing content by physician): D. Washburn

We would like to thank Dr. Marino Leon from Moffitt Cancer Center for the H&E and androgen receptor photomicrographs of the RET fusion patient described in this article.

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.