Targeted molecular profiling of salivary duct carcinoma with rhabdoid features highlights parallels to other apocrine and discohesive neoplasms: which phenotype should drive classification?

Abstract

Background

Salivary duct carcinoma with rhabdoid features (SDC-RF) is a recently-described salivary gland tumor that bears striking histologic similarity to lobular carcinoma of the breast. While this tumor has an apocrine phenotype that supports classification as a variant of SDC, it infrequently arises in association with conventional SDC. Furthermore, discohesive architecture can be seen in non-apocrine salivary carcinomas, raising the possibility that discohesive growth should define a separate entity. In this study, we aimed to perform comprehensive molecular profiling of SDC-RF to better understand its pathogenesis and classification.

Methods

We documented the clinicopathologic features of 9 cases of SDC-RF and performed immunostains including AR, GCDFP, and e-cadherin on all cases. We also performed targeted next generation sequencing (NGS) panels on 7 cases that had sufficient tissue available.

Results

The SDC-RF represented 8 men and 1 woman with a median age of 67 years (range 63–83 years) and included 6 parotid, 2 buccal, and 1 submandibular primary. All tumors were uniformly composed of discohesive cells with abundant eosinophilic cytoplasm; signet-ring cell features were seen in 2 cases. All tumors were also positive for AR (100%) and GCDFP (100%), and 7 tumors (78%) displayed lost or abnormal e-cadherin. NGS highlighted concomitant PIK3CA and HRAS mutations in 4 tumors, with additional cases harboring TP53, PTEN, and AKT1 mutations. Furthermore, CDH1 alterations were seen in 6 cases, including a novel CDH1::CORO7 fusion. Among 5 patients with follow-up available, 3 (60%) developed local recurrence and widespread distant metastasis and died of disease at a median 20 months (range 10–48 months).

Conclusions

Overall, our findings confirm frequent CDH1 mutations and e-cadherin inactivation in SDC-RF, similar to discohesive tumors from other sites. We also highlight an apocrine molecular profile similar to conventional SDC. However, occasional AKT1 mutation and signet-ring features suggest SDC-RF may also be related to mucinous adenocarcinoma. As more salivary tumors with discohesive growth are identified, it may become clearer whether SDC-RF should remain in the SDC family or be recognized as a separate entity.

Introduction

Salivary duct carcinoma (SDC) is a high-grade salivary gland carcinoma that was originally defined by histologic similarity to ductal carcinoma of the breast, including cribriform architecture with rigid bridge formation and prominent comedonecrosis [1–4]. More recently, however, an apocrine phenotype has become central to the diagnosis, with consistent abundant eosinophilic cytoplasm, prominent nucleoli, and positivity for androgen receptor (AR) [5–7]. These shifting criteria have facilitated broader recognition of the morphologic spectrum of SDC, including specific micropapillary, mucin-rich, sarcomatoid, and oncocytic variants [8–13]. Indeed, almost 40% of salivary tumors previously classified as adenocarcinoma not otherwise specified can be reclassified as SDC based on this expanded histologic profile and AR positivity [14]. Additionally, identification of characteristic molecular alterations in SDC has further clarified the pathogenesis of this entity. Most cases of SDC harbor recurrent TP53 mutations, HRAS hotspot mutations, PIK3CA hotspot mutations, PTEN loss, or ERBB2 amplification, with PLAG1 or HMGA2 fusions in tumors that arise ex pleomorphic adenoma [5, 15–23]. Similar mutations have also been noted in micropapillary and sarcomatoid variants of SDC, indicating both immunophenotypic and molecular unity across the SDC spectrum [16, 21, 24].

Recently, Kusafuka and colleagues described a novel variant of SDC with rhabdoid features (SDC-RF) [25]. These tumors display the apocrine morphology and consistent AR positivity now regarded as definitional for SDC. However, they also show a prominent discohesive growth pattern and frequent loss of membranous e-cadherin expression [25–27]. These unique histologic and immunohistochemical findings are strikingly similar to histiocytoid variant of lobular carcinoma of the breast. While the apocrine phenotype certainly is in keeping with SDC, SDC-RF are only rarely associated with conventional SDC morphology or other SDC variants and appear to have an even worse prognosis [27–29]. Moreover, other salivary tumors that lack apocrine differentiation have recently been reported to show a similar discohesive morphology, including the signet-ring subtype of mucinous adenocarcinoma and a case reported as mammary lobular carcinoma-like salivary gland carcinoma [30, 31]. It is not entirely clear whether SDC-RF is best classified as a variant of salivary duct carcinoma or a separate tumor type that happens to show frequent apocrine differentiation. Beyond documentation of recurrent CDH1 mutations [27], the genetic underpinnings of SDC-RF have not yet been investigated. This study aims to perform detailed clinicopathologic and molecular analysis of SDC-RF to help understand the classification and pathogenesis of this unique entity.

Materials and methods

Case selection

With institutional review board approval (Johns Hopkins Medicine IRB 00176183 and UT Southwestern IRB 112017-073), we identified 9 cases of SDC-RF from the authors’ surgical pathology and consultation files. Two of the cases had been previously reclassified from the adenocarcinoma not otherwise specified category [14]. For classification as SDC-RF and inclusion in this study, tumors had to demonstrate (1) apocrine cytomorphology, (2) positivity for AR, and (3) a dominant discohesive growth pattern. Immunohistochemical loss of e-cadherin was not required for classification as SDC-RF. We reviewed all available histologic sections for all cases to confirm the diagnosis, and salient morphologic, prognostic and staging parameters were documented. Clinical, demographic, and any available follow-up data were obtained from the electronic medical record.

Immunostains

We performed immunohistochemistry on all cases using mouse monoclonal antibodies for AR (clone SP107; CellMarque/Sigma- Aldrich, St. Louis, MO; prediluted), gross cystic disease fluid protein (GCDFP; clone D6; BioLegend, San Diego, CA; 1:200), S100 protein (clone 4C4.9; Ventana Medical Systems, Tucson, AZ; prediluted), p40 (clone BC28; BioCare Medical, Pacheco, CA; 1:100), and e-cadherin (clone EP700Y; Ventana Medical Systems; prediluted). In a subset of cases, we also performed immunohistochemistry for GATA3 (clone L50-823; BioCare Medical; 1:100) and estrogen receptor (ER; clone SP-1; Ventana Medical Systems; prediluted). Stains were performed on Ventana BenchMark Ultra autostainers (Ventana Medical Systems) using standardized automated protocols and appropriate positive and negative controls. The ultraView polymer detection kit (Ventana Medical Systems) was used to visualize signals.

Targeted next generation sequencing

We performed targeted next generation sequencing (NGS) on 7 cases that had sufficient tissue available. Five cases underwent NGS at the University of Texas Southwestern Medical Center as previously described [32]. DNA and RNA were isolated using Qiagen AllPrep kits (Qiagen, Germantown, MD), custom NimbleGen probes (Roche, Indianapolis, IN) were used to create an enriched library containing all exons from > 1,425 cancer-related gene, and sequencing was performed on a NextSeq 550 (Illumina, San Diego, CA) with a median 900x target exon coverage. Two cases underwent NGS at The Johns Hopkins Hospital, as detailed elsewhere [33, 34]. DNA was isolated using the Siemens Tissue Preparation System (Siemens Healthcare Diagnostics, Tarrytown, NY), the SureSelect XT Target Enrichment System (Agilent Technologies, Santa Clara, CA) was used to create libraries containing the full coding regions of 644 cancer-associated genes, and sequencing was performed on the HiSeq 2500 platform (Illumina) to an average 500 to 1000× read depth. For all cases, variants were reviewed using the Integrated Genomics Viewer (Broad Institute, Cambridge, MA) and annotated using the gnomAD and dbSNP databases.

Results

Clinical and demographic features

Clinical and demographic information is summarized in Table 1. The patients included 8 men and 1 woman with a median age of 67 years (range 63–83 years). There were 6 tumors centered in the parotid gland, 2 in the buccal mucosa, and 1 in the submandibular gland. Microscopic evidence of extraparenchymal extension was seen in 4 of the tumors that arose in major glands (57%). Tumors had a median size of 2.7 cm (range 1–6 cm). Lymph node sampling was performed in 7 cases, of which 5 (71%) had lymph node metastasis; 4 of these cases had extensive nodal disease with an average of 38 positive nodes each (range 6–66) and widespread extranodal extension.

Table 1.

Clinical and demographic information

Case	Age	Sex	Site	Size (cm)	Treatment	Follow-up	Progression	Status
1	65	M	Parotid	2.2	Surgery, chemo, XRT	48	Locoregional recurrence, widespread distant metastasis	DOD
2	63	M	Parotid	5	Surgery, chemo, XRT	20	Locoregional recurrence, widespread distant metastasis	DOD
3	78	M	Parotid	6	Surgery, chemo, XRT	5	None	NED
4	66	M	Buccal	3	Surgery, chemo, XRT	10	Locoregional recurrence, widespread distant metastasis	DOD
5	75	M	Parotid	2.1	NA	NA	NA	NA
6	83	M	Parotid	1.5	NA	NA	NA	NA
7	67	M	Parotid	1	NA	NA	NA	NA
8	67	M	Submandibular	2.7	Surgery, chemo, XRT	6	None	NED
9	75	F	Buccal	4.9	NA	NA	NA	NA

Chemo: chemotherapy; DOD: dead of disease; NA: not available; NED: no evidence of disease; XRT: external beam radiation

Histologic features

All 9 tumors were entirely composed of sheets (Fig. 1 A), cords (Fig. 1B), and singly dispersed eosinophilic cells (Fig. 1 C) that demonstrated highly infiltrative growth through salivary parenchyma and fibroadipose tissue. At least focal intraductal colonization was seen in 5 cases, including extensive intraductal spread in 1 case (Fig. 1D). No classic salivary duct architecture, including organized cribriform, solid, glandular, or micropapillary components, was seen in any case. Although 1 case had prominent stromal sclerosis, no definite histologic evidence of underlying pleomorphic adenoma was present. Perineural invasion (Fig. 2 A) and lymphovascular invasion (Fig. 2B) were both identified in 7 cases. The tumor cells in all cases had numerous mitotic figures including atypical mitotic forms (Fig. 2 C) and 5 cases showed zones of necrosis (Fig. 2D). All tumor cells had an apocrine appearance with abundant eosinophilic cytoplasm and irregular nuclei with vesicular chromatin and prominent nucleoli. However, these cells ranged from overtly rhabdoid, with hard, hyalinized cytoplasm and eccentric nuclei (Fig. 3 A) to somewhat histiocytoid, with more central nuclei and irregular, foamy cytoplasm (Fig. 3B). Two cases displayed an occasional signet-ring appearance with intracytoplasmic mucin vacuoles (Fig. 3 C) that were highlighted by mucicarmine stain (Fig. 3D).

Fig. 1.

SDC-RF were entirely composed of highly infiltrative sheets (A), cords (B), and singly dispersed cells (C) with intraductal colonization in a subset of cases (D)

Fig. 2.

Perineural invasion (A) and lymphovascular invasion (B) were frequently seen, and mitotic figures including atypical mitotic forms (C) and necrosis (Figure D) were common

Fig. 3.

Tumor cells had eosinophilic cytoplasm, vesicular chromatin, and prominent nucleoli with a rhabdoid (A) to histiocytoid (B) appearance. As subset of cases had signet-ring features (C) with intracytoplasmic mucin vacuoles on mucicarmine stain (D)

Immunohistochemical findings

The immunohistochemical features of SDC-RF are summarized in Table 2. All tumors were positive for AR (100%; Fig. 4 A) and GCDFP (100%; Fig. 4B) and negative for S100 protein (0%) and p40 (0%). Of a subset of cases tested, all tumors were also positive for GATA3 (100%) and negative for ER (0%). Six tumors (67%) showed total loss of e-cadherin labeling (Fig. 4 C) while one (11%) displayed diminished expression with abnormal cytoplasmic localization (Fig. 4D); two cases (22%) had intact e-cadherin. Areas of intraductal colonization displayed the same e-cadherin pattern as surrounding invasive carcinoma, with loss of e-cadherin labeling in most cases.

Table 2.

Immunohistochemical findings

Case	AR	GCDFP	GATA3	S100 protein	p40	ER	e-cadherin	Mucicarmine
1	+	+	+	-	-	ND	Lost	+
2	+	+	+	-	-	ND	Lost	-
3	+	+	ND	-	-	-	Lost	ND
4	+	+	+	-	-	-	Cytoplasmic	-
5	+	+	+	-	-	-	Intact	-
6	+	+	ND	-	-	-	Lost	+
7	+	+	+	-	-	-	Intact	-
8	+	+	+	-	-	-	Lost	-
9	+	+	+	-	-	-	Lost	-

AR: androgen receptor; ER: estrogen receptor; GCDFP: gross cystic disease fluid protein; ND: not done

Fig. 4.

All tumors were positive for AR (A) and GCDFP (B); the majority showed total loss of e-cadherin (C) or abnormal cytoplasmic localization (D)

Molecular findings

Molecular findings are summarized in Fig. 5. Of the 7 tumors that underwent targeted NGS, oncogenic driver mutations were identified in all cases. 4 cases (57%) had concomitant PIK3CA and HRAS hotspot mutations, 1 case (14%) showed PTEN loss and TP53 mutation, 1 case (14%) showed AKT1 E17K hotspot mutation and TP53 mutation, and 1 case (14%) had TP53 mutation. Other potentially significant alterations included single copy number loss of CDKN2A and CDKN2B in 3 cases (50%), monoallelic RB1 inactivation via single copy loss or mutation in 2 cases (29%), and SMAD4 mutation in 1 case (14%). Six cases (86%) showed evidence of CDH1 alterations, including biallelic inactivation in 2 cases and monoallelic mutations in 4 cases. These alterations included single copy number loss in 3 cases, frameshift mutations in 2 cases, nonsense mutation in 1 case, splice site mutation in 1 case and a novel CDH1::CORO7 fusion in 1 case. Because the two genes were out of frame, this fusion was predicted to lead to a loss of function. Five of these CDH1 mutant cases, including both with biallelic alterations, had lost or abnormal e-cadherin protein expression.

Fig. 5.

NGS highlighted frequent PIK3CA and HRAS hotspot mutations with additional TP53 mutations, AKT1 hotspot mutation, and PTEN loss. Most cases also had CDH1 alterations, including a novel CDH1::CORO7 fusion

Follow-up

Follow-up information is also summarized in Table 1. Detailed clinical follow-up was available for 5 patients, with median of 10 months (range 5–48 months). All patients were treated with surgical resection and adjuvant external beam radiation and chemotherapy. Three patients (60%) developed both local recurrence and widespread distant metastases involving the bones, lungs, liver, and dermis of the chest wall and axilla, which were treated with additional radiation and chemotherapy. All 3 of these patients died of disease at a median of 20 months (range 10–48 months).

Conclusions

The diagnostic boundaries of SDC have been refined in recent years, with increasing emphasis on an apocrine phenotype and AR positivity to confirm the classification. SDC-RF is a newly-described salivary gland tumor that shows a unique discohesive growth pattern but is currently classified as a variant of SDC due to consistent apocrine morphology. However, SDC-RF usually is not associated with conventional SDC elements, displays even more aggressive clinical behavior, and carries similarities to other non-apocrine salivary tumor types, raising the possibility that these tumors represent a separate entity. In this study, we performed comprehensive molecular analysis of SDC-RF to better understand its classification and pathogenesis.

First, our findings confirm the presence of recurrent CDH1 gene alterations in SDC-RF. CDH1 is a gene on chromosome 16q22.1 that encodes e-cadherin, a trans-membrane glycoprotein that serves as a tumor suppressor gene by regulating cellular adhesion and maintenance of cellular polarity [35–37]. Alterations in CDH1 are well-established in several tumor types that show discohesive morphology, not only including lobular breast carcinoma but also diffuse-type gastric carcinoma and plasmacytoid urothelial carcinoma [38–42]. Kusafuka et al. recently reported that CDH1 mutations are also present in 72% of SDC-RF [27]. In our series, 86% of cases also showed CDH1 alterations, with corresponding lost or abnormal e-cadherin expression in 78%. Interestingly, one of these tumors displayed a CDH1::CORO7 fusion that conferred loss of function- a mechanism of inactivation that has only been rarely reported as a pathway of CDH1 inactivation in lobular breast carcinoma, diffuse-type gastric carcinoma, and plasmacytoid urothelial carcinoma [43–45]. Notably, although biallelic CDH1 alterations were only seen in a subset of cases, inactivation of the second allele frequently occurs through other mechanisms such as promoter methylation [46, 47]. Likewise, neither CDH1 mutations nor e-cadherin loss were uniformly present in SDC-RF in this or previous series, analogous to lobular breast cancer where a subset of cases show intact CDH1 and e-cadherin despite discohesive growth [48, 49]. These findings place SDC-RF squarely within the spectrum of neoplasms defined by discohesive growth.

These results also demonstrate that SDC-RF have oncogenic driver mutations similar to those seen in conventional SDC and other salivary tumors with apocrine features. Molecular profiling of SDC has consistently identified alterations in the PI3K and MAPK pathways, most commonly including PIK3CA and HRAS hotspot mutations and PTEN loss; they also frequently harbor TP53 mutations and ERBB2 amplification as well as PLAG1 or HMGA2 fusions in the setting of origin ex pleomorphic adenoma [5, 15–23]. The vast majority of SDC-RF cases in this series also display alterations within this spectrum, including four cases with concomitant HRAS and PIK3CA mutations and one case with PTEN loss. AKT1 mutation, which has rarely been reported in SDC but also is part of the PI3K pathway [19, 20, 23], was also present in one SDC-RF. Although HER2/Neu immunohistochemistry was not performed, ERBB2 amplification was not observed in any cases at the molecular level. Likewise, the lack of PLAG1 or HMGA2 fusions in this series was in keeping with the absence of precursor pleomorphism adenoma. These findings suggest that SDC-RF does fall in a similar molecular spectrum with conventional SDC. However, while these mutations are characteristic of SDC they are not specific for this diagnosis. HRAS, PIK3CA, and PTEN mutations are also consistent features of other salivary tumors that show an apocrine phenotype, including low-grade apocrine intraductal carcinoma and even benign sclerosing polycystic adenoma [32, 50–53]. In many ways, this molecular spectrum may more directly correspond to apocrine differentiation in general than SDC as a specific entity.

Despite similarities to SDC, these morphologic and molecular findings also raise parallels to other rare salivary gland tumors that lack an apocrine phenotype but demonstrate discohesive architecture. Signet-ring carcinomas are now considered part of the emerging entity mucinous adenocarcinoma, a category of salivary carcinoma that shows heterogeneous histologic patterns but frequent AKT1 E17K mutations. We previously reported 1 case of mucinous adenocarcinoma with signet-ring features that was AR negative but otherwise displayed striking histologic similarity to SDC-RF, including a uniform population of discohesive eosinophilic cells with intracellular mucin vacuoles and loss of e-cadherin expression [31]. Interestingly, that tumor harbored a CDH1 mutation as well as AKT1 E17K mutation. The overlapping presence of signet-ring features and AKT1 mutation in cases in this series suggest that SDC-RF may have some relationship to mucinous adenocarcinoma as well as SDC. Lei et. al recently reported another unique AR-negative case as a mammary lobular carcinoma-like salivary gland carcinoma, which showed discohesive growth and abnormal cytoplasmic localization of e-cadherin [30]. This tumor had a mutation in CTNNA1- a gene that encodes for the alpha-E-catenin protein that interacts with e-cadherin during cellular adhesion and can also be mutated in lobular breast carcinomas and diffuse gastric cancer [54]. This unique case suggests that an even broader spectrum of discohesive salivary tumors may exist outside the apocrine spectrum. Both of these AR-negative carcinomas with discohesive architecture behaved in an extremely aggressive fashion with extensive lymph node involvement or metastatic spread, also in keeping with the highly aggressive profile of SDC-RF. While rare, these isolated AR-negative cases raise the possibility that salivary tumors with discohesive growth represent a unified spectrum regardless of apocrine differentiation.

In light of these parallels, it is worth reconsidering the classification of SDC-RF in the larger context of how we define the boundaries of SDC. While most tumors in the SDC spectrum have concordant clinical, histologic, immunohistochemical, and molecular features to support their diagnosis, classification becomes more challenging when all of these data points are not aligned. A small subset of tumors with conventional SDC morphology have been shown to have NTRK3, ALK, or RET fusions, suggesting that they may be related to secretory carcinoma or intercalated duct-type intraductal carcinoma [16, 55–57]. However, there is currently substantial consensus that these tumors should be classified as SDC regardless of their divergent genetic profiles, indicating that molecular heterogeneity is acceptable in SDC in the setting of classic morphologic features. So-called basal-like SDC are high grade salivary gland adenocarcinomas that lack AR expression but have morphology and CK 5/6 expression that is comparable to basal-like breast carcinomas [58, 59]. It is still controversial whether these tumors should be included within the SDC spectrum, but their classification as SDC is increasingly questionable as an apocrine immunophenotype becomes definitional for the diagnosis. As a tumor that shows AR positivity and molecular concordance with SDC but displays distinctive discohesive architecture and uniquely aggressive behavior, SDC-RF raises a new quandary. Ironically, despite emphasis on similarity to lobular carcinoma of the breast, the current classification of SDC-RF diverges from that of mammary carcinomas, where architectural cohesion is the primary distinguishing factor between ductal and lobular carcinoma and apocrine differentiation is considered a secondary feature. Even if an apocrine immunophenotype is necessary for classification as SDC, it is not yet clear whether this feature is sufficient in the setting of other divergent findings. It is possible that apocrine differentiation and discohesive architecture represent separate but frequently overlapping morphologic patterns that high grade salivary tumors can assume.

Overall, the findings in this series add to the emerging clinicopathologic profile of SDC-RF. We validate the presence of frequent CDH1 gene alterations in SDC-RF in keeping with its discohesive phenotype and demonstrate that its consistent apocrine morphology correlates to an apocrine molecular profile that is similar to more conventional SDC as well as other apocrine salivary tumors. However, we also highlight striking parallels between SDC-RF and non-apocrine salivary carcinomas such as the signet-ring variant of mucinous adenocarcinoma, and confirm its uniquely aggressive behavior and lack of a consistent association with conventional SDC components. As more cases with this distinctive growth pattern are accumulated, it will be essential to further consider whether SDC-RF should remain in the SDC family or be recognized as a separate entity.

Authors’ contributions

LMR and JAB designed the study, contributed tumor samples, performed data collection and interpretation, and prepared the manuscript. JG performed data collection and interpretation. All authors read and approved the final paper.

Funding

This study was funded in part by the Jane B. and Edwin P. Jenevein M.D Endowment for Pathology at UT Southwestern Medical Center. No external funding was obtained for this study.

Data Availability

All data generated or analyzed during this study are included in this published article.

Code Availability

Not applicable.

Declarations and compliance with ethical Standards

Conflicts of interest/Competing interests

All authors certify that they have no affiliations with or involvement in any organization or entity with any financial interest or non-financial interest in the subject matter or materials discussed in this manuscript.

Ethics approval

All procedures performed in this retrospective data analysis involving human participants were in accordance with the ethical standards of the institutional review board (Johns Hopkins Medicine IRB 00176183 and UT Southwestern IRB 112017-073).

Consent to participate/for publication

The IRB-approved study did not require informed consent.