Proteomic analysis of transitional cell carcinoma—like variant of tubo-ovarian high-grade serous carcinoma*^,**

Abstract

The current World Health Organization classification does not distinguish transitional cell carcinoma of the ovary (TCC) from conventional tubo-ovarian high-grade serous carcinoma (HGSC), despite evidence suggesting improved prognosis for patients with TCC; instead, it is considered a morphologic variant of HGSC. The immunohistochemical (IHC) markers applied to date do not distinguish between TCC and HGSC. Therefore, we sought to compare the proteomic profiles of TCC and conventional HGSC to identify proteins enriched in TCC. Prognostic biomarkers in HGSC have proven to be elusive, and our aim was to identify biomarkers of TCC as a way of reliably and reproducibly identifying patients with a favorable prognosis and better response to chemotherapy compared with those with conventional HGSC. Quantitative global proteome analysis was performed on archival material of 12 cases of TCC and 16 cases of HGSC using SP3 (single-pot, solid phase—enhanced, sample preparation)-Clinical Tissue Proteomics, a recently described protocol for full-proteome analysis from formalin-fixed paraffin-embedded tissues. We identified 430 proteins that were significantly enriched in TCC over HGSC. Unsupervised co-clustering perfectly distinguished TCC from HGSC based on protein expression. Pathway analysis showed that proteins associated with cell death, necrosis, and apoptosis were highly expressed in TCCs, whereas proteins associated with DNA homologous recombination, cell mitosis, proliferation and survival, and cell cycle progression pathways had reduced expression. From the proteomic analysis, three potential biomarkers for TCC were identified, claudin-4 (CLDN4), ubiquitin carboxyl-terminal esterase L1 (UCHL1), and minichromosome maintenance protein 7 (MCM7), and tested by IHC analysis on tissue microarrays. In agreement with the proteomic analysis, IHC expression of those proteins was stronger in TCC than in HGSC (p < 0.0001). Using global proteomic analysis, we are able to distinguish TCC from conventional HGSC. Follow-up studies will be necessary to confirm that these molecular and morphologic differences are clinically significant.

1. Introduction

Transitional cell carcinoma of the ovary (TCC), the transitional cell carcinoma—like variant of tubo-ovarian high-grade serous carcinoma (HGSC), was first described as a separate entity from Brenner tumors of the ovary by Austin and Norris [1], with the largest cohort described by Eichhorn and Young [2]. Studies from MD Anderson were the first to suggest that TCC has a better prognosis than HGSC [3–6], possibly owing to better response to chemotherapy and a less infiltrative invasion pattern; they were followed by work from a German group that supported their findings [7]. Other studies, however, failed to show a survival advantage for patients with TCC [8–10]. Tumors showing a combination of solid, pseudoendometrioid, and transitional cell carcinoma–like morphology are more likely to harbor a BRCA1/2 mutation (and BRCA1/2 mutations are associated with chemosensitivity [11]), providing further evidence that the TCC subset of HGSC has distinct characteristics [12]. Immunohistochemical (IHC) studies have established that TCC is not a urothelial tumor and differs significantly from Brenner tumors; instead, it is a variant of HGSC of tubo-ovarian origin based on the similar immunoprofile and mutations [13–15]. As a result, the 2014 World Health Organization (WHO) classification no longer regards TCC as a separate entity and identifies TCC as a morphological variant of HGSC [16]. The TCC variant is characterized by a resemblance to carcinomas originating from the urothelial epithelium, but without evidence of a benign or borderline Brenner tumor component [1,5]. They typically have a papillary architecture with multilayered epithelium but can also be nested or be admixed with conventional HGSCs [17,18]. Conventional HGSCs show a variety of morphologic patterns including glandular, papillary, and solid patterns. The glands are classically slit-like or irregular and are composed of highly pleomorphic malignant cells [16].

Research into TCC is limited by the subjective nature of the diagnosis, which is based purely on hematoxylin and eosin morphology. Some centers make this diagnosis with regularity, whereas it is never diagnosed in others. An important means to advance our understanding of TCC would be to identify one or more biomarkers that will accurately distinguish it from HGSC. Such a biomarker could serve to anchor study diagnoses, allowing for more reproducible diagnoses and more accurate assessment of the clinical significance, if any, of this HGSC subtype. This situation is reminiscent of the conundrum involving HGSC and high-grade endometrioid carcinoma of the ovary, which until the discovery of, (Wilms’ tumor suppressor gene 1) WT-1 could not be reliably and consistently distinguished [19,20].

Advances in sample handling and data acquisition strategies for proteomic analyses based on mass spectrometry (MS) allow quantitative assessment of thousands of proteins from modest amounts of formalin-fixed paraffin-embedded (FFPE) tissue [21]. Previously, our group successfully compared the protein abundance profiles of tubo-ovarian HGSC, endometrioid carcinoma, and clear-cell carcinoma in FFPE tissue with those in snap-frozen tissue using a modified version of the SP3 (single-pot, solid phase—enhanced, sample preparation) single-tube sample handling approach coupled to a quantitative proteomics workflow based on isobaric tagging (tandem mass tags) that was termed SP3-Clinical Tissue Proteomics (SP3-CTP) [22]. Using this approach, a comparable number of proteins were identified in the FFPE and snap-frozen tissue samples. The SP3-CTP workflow enabled deep proteome coverage from exceedingly small input quantities of tissue and demonstrated this technology could be effectively used as a biomarker discovery tool. Therefore, in this work, we used the SP3-CTP method to compare the full proteomes of a series of tubo-ovarian TCC and HGSC, to identify potential diagnostic biomarkers of this disease.

2. Materials and methods

2.1. Tissue acquisition

The slides and blocks of FFPE tumor tissues were obtained under the auspices of the University of British Columbia Research Ethics Boards (REB# H17-00569). A large TCC cohort (89 cases) of archival FFPE cases from Canada, Germany, and the United States was selected for analysis [23]. The cases chosen were included based on the modified morphologic criteria established by Eichhorn and Young [2]: (1) an ovarian epithelial neoplasm with destructive stromal invasion, (2) a histologic component closely resembling transitional cell carcinoma of the urinary bladder accounting for more than 50% of the tumor (macropapillae lined by thick bands of atypical cells, a solid nested pattern, or thick undulating bands of cells lining cystic spaces), (3) an absence of benign or borderline Brenner elements or endometrioid carcinoma, and (4) the absence of a synchronous or prior transitional cell carcinoma of the urinary tract. Cases showing 100% TCC features were classified as pure TCC; cases showing a minor (<50%) component of conventional HGSC were classified as mixed TCC. After review by a gynecologic pathologist (F.K.), one case was redefined as conventional HGSC, leaving 88 TCC cases: 59 pure and 29 mixed. Two representative 0.6-mm cores from the TCC blocks were used to construct a tissue microarray (TMA), including duplicate cores from both components (TCC-like and non-TCC–like) of 29 mixed TCC cases. Outcome and pathologic data were not available for the TCC cohorts. For comparison, a preexisting TMA at Vancouver General Hospital, Vancouver, BC, containing duplicate 0.6-mm cores from 237 cases of conventional HGSC was used [24]. Five centrally reviewed low-grade serous carcinomas (LGSCs) were also included for comparison in the proteomic experiment.

2.2. SP3-CTP sample preparation and MS analysis

From the perviously described TCC cohort, 12 TCC cases (10 pure TCC and 2 mixed TCC cases) were selected based on abundant tumor tissue and highest tumor content and cellularity. Sixteen conventional HGSC cases with no transitional morphologic features and 5 LGSC cases were included for comparison. The samples were separated into 4 batches of 8 or 9 samples per batch. Two 10-μm scrolled sections were cut from archival FFPE blocks, equivalent to 3 cm × 1 cm of tumor surface area, and selected based on having a tumor cellularity higher than 50%. Tissue sections were processed using the SP3-CTP protocol, as described previously (Supplemental Methods) [22].

2.3. Protein expression control analysis and volcano plot

Peptide-level proteomic data were median normalized using linear models for microarray data (LIMMA) and batch effect—corrected using ComBat (combatting batch effects when combining batches of microarray data) [25,26]. The protein expression control analysis (probe-level expression change averaging) R package was used to generate log2-transformed, protein-level expression differentials (signal log ratio) and, (false discovery rate) FDR-adjusted p values (p.fdr) for the 5010 proteins detected in all samples. For each protein, a score was generated (abs_SCORE) using the aforementioned values. Specifically, the absolute value of the expression differential for each protein was multiplied by its negative log10-transformed p.fdr value. This yielded a list of the proteins with the highest absolute expression differential in either direction and the lowest FDR-adjusted p-values. A subset of all proteins with abs_SCORE values higher than the population average of the proteins detected was then used in the downstream pathway analysis. These proteins are shown as blue (enriched in HGSC relative to TCC) and green (enriched in TCC relative to HGSC) in the volcano plot (Fig. 1).

Fig. 1.

Volcano plot showing proteins differentially expressed between TCC and HGSC. Volcano plot showing the proteins most significantly increased between cases with TCC features (toward the left) and conventional HGSC features (toward the right). The y-axis represents the statistical significance (p-value), whereas the x-axis shows the protein abundance (fold change). The proteins in color (green for TCC and blue for HGSC) highlight significance cutoff of proteins with abs_SCORE values higher than the population average. TCC, transitional cell carcinoma of the ovary; HGSC, high-grade serous carcinoma; PECA, protein expression control analysis.

2.4. Co-clustering and principal component analysis

Reproducible detection of specific peptides in MS-based proteomics is often incomplete or variable, especially for peptides with low abundance. As a result, those missing peptides would be excluded from a standard clustering analysis. To include peptides that were only detected in a subset of the 4 batches (eg, in samples from batches 1, 2, or 3, in which they were identified in every sample within that batch) into the clustering analysis, we clustered them in independent subsets and then statistically collapsed the results of all possible subset combinations into a final co-cluster defined in more detail in the following section. The peptides were grouped into the following subsets: common to all batches; found in only batch 1, 2, 3, or 4; and common to batches 1 + 2, 1 + 3, 1 + 4, 2 + 3, 2 + 4, 3 + 4, 1 + 2+3, 1 + 2+4, 1 + 3+4, or 2 + 3+4. The peptide subsets were independently normalized with LIMMA and corrected for batch effect using ComBat [25,26]. Hierarchical clustering on each subset was carried out using the hclust program (fastcluster, version 1.1.25) using a Ward method, in which the number of clusters for each subset was determined by the number of tumor types represented by the samples included in each subset [27]. When clustering on the subsets, a cutoff was used to determine which peptides would be used for clustering, based on the expression variance between the samples. This cutoff was adjusted iteratively, stopping when there was the best agreement between clustering and tumor type. Once clustering was completed, the groupings for each sample were converted into a distance using a weighted Jaccard index, with weights determined by the number of peptides in each subset. Overall clustering on the 15 peptide subsets was again carried out using the hclust package program using a Ward method [27]. To validate the cluster analysis, principal component analysis (PCA) was carried out only on peptides present in all batches, and the two most variable principle components were used to plot the samples. The first ten principle components were calculated using the pca package with the svd method from the pcaMethods package and were plotted using R version 3.4.3 [28].

2.5. Pathway analysis

The pathway analysis was conducted using the core analysis function of the Qiagen Ingenuity Pathway Analysis program, 1001 Marshall Street, Redwood City, CA 94063, United States. The Ingenuity Knowledge Base (Genes only) was used as the background reference set, and all direct and indirect gene relationships in humans and mice were included in the analysis. All proteins with abs_SCORE values higher than the population average of the proteins detected were used in the analysis (n = 964). The statistical parameters to define biologic pathways’ significance included a p-value lower than 0.05 and a z-score higher than 2 (increased activation state) or lower than −2 (decreased activation state).

2.6. Biomarker selection and immunohistochemistry

Proteins were selected based on three main factors: (1) protein enrichment, from the MS proteomic analysis which isolated the proteins that were significantly enriched in TCC compared with HGSC; (2) biologic interest, which was guided by a literature review and pathway analysis of significantly enriched proteins; (3) IHC antibody availability and performance; after selection of differentially expressed proteins, IHC analysis was optimized on FFPE material for those proteins in which commercially available antibodies could be identified. Taking these into account, we selected proteins for which expression was highest in TCC, lowest in HGSC, and with the lowest p-value measured. All 14 proteins were evaluated using commercially available antibodies, and only ubiquitin carboxyl-terminal hydrolase isozyme L1 (UCHL1), claudin-4, and minichromosome maintenance protein 7 (MCM7) IHC antibodies could be successfully optimized on FFPE material.

Immunohistochemistry was performed on 4-μm-thick FFPE sections of TMAs at the Department of Genetic Pathology Evaluation Centre at the University of British Columbia in Vancouver using the Ventana Discovery XT automated system (Ventana Medical Systems, Tucson, AZ). The sections were mounted on charged glass slides and baked at 60 °C for 15 min. Heat-induced antigen retrieval was performed in Cell Conditioning solution CC1 (Ventana). For UCHL1 (also known as protein gene product 9.5 [PGP 9.5]), slides were incubated with the clone 13C4/I3C4 (mouse monoclonal, DAKO) at 1:400 dilution in Diamond antibody diluent at room temperature for 30 min, 6600 Sierra College Blvd., Rocklin, California 95677, United States. The slides were washed and incubated with the Mach2 Mouse-HRP polymer, Biocare Medical, 60 Berry Dr, Pacheco, CA 94553 for 30 min at room temperature and detected using the IP DAB chromogen for 5 min. For claudin-4, slides were incubated with the clone 3E2C1 (mouse monoclonal; Invitrogen, (Fisher Scientific Company, 112 Colonnade Road, Ottawa, Ontario, K2E 7L6, Canada)) at 1:300 dilution in Da Vinci Green diluent at room temperature for 37 min. The slides were then washed and incubated with the Mach2 Mouse-HRP polymer for 32 min at room temperature and detected using the IP DAB chromogen for 5 min. For MCM7, slides were incubated with the clone D10A11 (rabbit monoclonal; Cell Signaling), 32 Tozer Road, Beverly, MA, 01915, United States at 1:100 dilution in Diamond antibody diluent at room temperature for 30 min. The slides were washed and incubated with the Mach2 Mouse-HRP polymer for 30 min at room temperature and detected using the IP DAB chromogen for 5 min. Nuclei were counterstained with a 1:10 dilution of CAT hematoxylin, Biocare Medical, 60 Berry Dr, Pacheco, CA 94553; then, slides were again washed, air-dried, and coverslipped manually.

2.7. IHC analysis interpretation

TMAs were scored by a pathologist (F.K.) and a senior pathology resident (B.T.-C.), and consensus was achieved on all cases. An immunoreactive score was assigned to each core: a product of staining intensity (scale of 0–3) and percentage of tumor cell staining (scale of 0–4). Staining intensity was scored as 0 (none or minimal staining in occasional tumor cells), 1 (weak), 2 (moderate), or 3 (strong). Percentage of tumor cell staining was scored as 0 (<1%), 1 (>0–25), 2 (>25–50), 3 (>50–75), or 4 (>75–100) based on the number of cells stained. Therefore, the final immunoreactive score ranged from 0 to 12.

3. Results

3.1. Histopathologic description of the TCC cohort

The TCC samples collected for this study predominantly composed of a macropapillary pattern, consisting of large plump stromal papillae covered by a stratified or microcystic epithelium, or thick undulating epithelial bands lining cystic spaces composed of a stratified malignant epithelium resembling urothelium and displaying high-grade cytologic features. A nested architectural pattern with pushing borders was also observed. All cases of mixed TCC had a minor (<50%) component of conventional HGSC, including slit-like spaces and papillary, glandular, and cribriform areas (Fig. 2). Mixed transitional tumors with a component of either endometrioid carcinoma or Brenner tumor were excluded.

Fig. 2.

Representative examples of TCC histologies. H&E staining results of TCC showing a macropapillary pattern (A—B) covered by stratified (C) or microcystic epithelium (D) or thick undulating epithelial bands lining cystic spaces (E). Solid pattern with undulating epithelial bands (F). Images were captured using the following magnification: ×40 for panels A, B, E, and F and ×400 for panels C and D. TCC, transitional cell carcinoma of the ovary; H&E, hematoxylin and eosin.

3.2. Whole-proteome analysis and clustering

Proteomic analysis of the 33 tumors (10 TCC, 2 mixed TCC, 16 HGSC, and 5 LGSC cases) submitted for proteomic analysis identified 83,676 peptides associated with 5010 different proteins. Between TCC and HGSC, 964 proteins were significantly differentially expressed (p < 0.05); 430 were significantly enriched in TCC over HGSC (including UCHL1, claudin-4, and MCM7), and 534 were significantly enriched in HGSC compared with TCC (Fig. 1, Supplemental Table 1a). However, even the most statistically significant proteins did not show the ability to distinguish both tumor types with 100% sensitivity and specificity (Supplemental Fig. 1). Between LGSC and HGSC, 658 proteins were significantly differentially expressed in either direction (Supplemental Table 1b and c). Unsupervised co-clustering and PCA of peptide expression (Fig. 3A and B) accurately distinguishes all three groups: TCC, conventional HGSC, and LGSC. Based on the clustering algorithm data, TCC and HGSC were nonoverlapping with each other and with LGSC.

Fig. 3.

Unsupervised proteomic co-clustering of TCC, HGSC, and LGSC. A, Unsupervised co-clustering showing the result of multiple independent cluster analyses. To take into account expression of peptides present at low concentration and thus not identified in some samples, clustering was performed for peptides that were detected in every sample in the batches included in the clustering analysis but were not detected in one or more samples from the batch(es) not included in the clustering analysis. The dendrogram arm lengths are proportional to the degree of relatedness of the tumors, as determined by clustering. Just under the dendrogram, the individual tumors are depicted, with LGSC in red, HGSC in blue, and TCC in green. Note that there is perfect distinction of the three groups based on the co-clustering analysis. The next row indicates the batch number each sample was processed in (from 1 to 4). The remaining rows represent 15 different clustering analyses, each including a different subset of peptides exclusive to a given batch or combination of batches (eg, the first row represents a cluster analysis of all peptides only present in batch 1, whereas the last row represents a cluster analysis of all peptides only present in every batch). The batches included in each row and the number of peptides (n = X) on which the clustering was based are shown at the right of each row. The cases that clustered together in each clustering analysis are shown in the same color, ie, blue, orange, or green. Those 15 cluster analyses were statistically weighted based on the number of peptides and collapsed into the dendrogram above. B, PCA analyses of peptide quantities was performed on 12 TCC, 16 conventional HGSC, and 5 LGSC endometrial cancers. For each case, histotype and batch number are described. The algorithm accounts for all proteins that were expressed in at least one batch. TCC, transitional cell carcinoma of the ovary; HGSC, high-grade serous carcinoma; LGSC, low-grade serous carcinoma; PCA, principal component analysis. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

3.3. Pathway analysis

Pathway analysis of proteins enriched in the TCC cases shows that proteins associated with cell death, necrosis, and apoptosis were more highly expressed, whereas proteins associated with DNA homologous recombination, cell mitosis, proliferation and survival, cell cycle progression, and carbohydrate mechanisms were less expressed. In the group of HGSC cases, proteins associated with cell movement, invasion, viral infection, endocytosis, carbohydrate metabolism, cell survival, and proliferation were more highly expressed, whereas proteins associated with cell death and anoikis were less expressed (Supplemental Table 2).

3.4. Biomarker identification and validation

Based on protein high expression values in TCC compared with HGSC and biological interest, 15 proteins were shortlisted including UCHL1, claudin-4, claudin-3, MCM2, MCM7, eukaryotic translation initiation factor 3 subunit C, melanoma-associated antigen 4, periostin, mothers against decapentaplegic homolog 2, secretogranin-1, arachidonate 8S-lipoxygenase, B-cell CLL/lymphoma 7 protein family member C, prostatic acid phosphatase, desmin, and dynein assembly factor 1. Of these, UCHL1, claudin-4, and MCM7 had commercially available IHC antibodies, which were ordered and successfully optimized on FFPE samples (Fig. 4). The signal log ratio of the relative protein abundance for UCHL1, claudin-4, and MCM7 (TCC/HGSC) was −1.02 (p = 0.00003), −0.61 (p = 0.04956), and −0.54 (p < 0.00001), respectively. Only UCHL1 and claudin-4 were significantly higher in TCC than in HGSC.

Fig. 4.

UCHL1 and claudin-4 IHC analysis in TCC and HGSC. H&E staining results of a case of TCC (A) and conventional HGSC (B) followed by IHC staining in both cases for UCHL1 (C—D) and claudin-4 (E—F). All images were captured using ×200 magnification. H&E, hematoxylin and eosin; TCC, transitional cell carcinoma of the ovary; HGSC, high-grade serous carcinoma; UCHL1, ubiquitin carboxyl-terminal esterase L1; IHC, immunohistochemical.

Initial IHC validation was performed on a TMA with 16 cases of conventional HGSC and 88 cases of TCC (59 pure TCC, 29 mixed TCC cases) using antibodies for UCHL1, claudin-4, and MCM7 (Table). Additional 221 HGSC cases were added on a second course of validation for UCHL1 and claudin-4. The cores for two cases within the HGSC TMA were missing on the claudin-4 IHC section and could not be scored (total n = 235). For cases of mixed TCC, immunoscores were assigned to both the TCC-like (n = 29) and HGSC-like (n = 28) components. On review of the pattern of staining (eg, focal or diffuse), no association with either conventional HGSC or TCC was observed. Preliminary analysis of MCM7 IHC scores showed that it could not distinguish between TCC and the conventional HGSC morphology. High UCHL1 expression was significantly correlated with the pure TCC when compared with HGSC (p < 0.00001) but showed no significant correlation when comparing the mixed TCC-like and HGSC-like components (p = 0.29). Given that there were no significant differences between the components of the mixed tumors, we averaged the UCHL1 scores of TCC-like and HGSC-like components of mixed TCC cases (n = 28). Mixed TCC was significantly associated with high expression of UCHL1 compared with HGSC (p = 0.035), whereas the difference between mixed TCC and pure TCC was not significantly different (p = 0.17). For claudin-4, IHC expression was significantly higher in pure TCC than in HGSC (p < 0.00001). Again, the comparison between both components of mixed TCC showed no statistical difference (p = 0.13), so these were averaged to give a single result for each mixed tumor. The average expression of claudin-4 in mixed TCC was similar to that in pure TCC (p = 0.93) and was significantly higher than in HGSC (p < 0.00001). Any positivity (immunoscores 1–12) for both biomarkers had reasonable good sensitivity for pure TCC over conventional HGSC (UCHL1 = 81%, claudin-4 = 97%) but poor specificity (UCHL1 = 37%, claudin-4 = 22%). Strong positivity (immunoscores 8–12) was more specific for pure TCC than for conventional HGSC (UCHL1 = 86%, claudin-4 = 86%) but was not sensitive (UCHL1 = 44%, claudin-4 = 57%). When trying to predict the TCC (including pure TCC and mixed TCC-like) over HGSC morphology (including conventional HGSC and mixed HGSC-like), results were similar: any positivity for UCHL1 and claudin-4 was reasonably sensitive (UCHL1 = 83%, claudin-4 = 93%) but not specific (UCHL1 = 37%, claudin-4 = 20%), and using strong positivity as the threshold would show better specificity (UCHL1 = 86%, claudin-4 = 80%) at the cost of sensitivity (UCHL1 = 39%, claudin-4 = 55%) (Supplemental Fig. 2). Combining both IHC did not improve their ability to distinguish both tumor types. In pure TCC vs conventional HGSC, their combined positivity (immunoscores 1–12) showed low sensitivity (79.3%) and low specificity (52.9%), whereas strong positivity threshold had great specificity (97.7%) but low sensitivity (24.1%). When comparing TCC-like morphology with HGSC morphology, sensitivity was 77.3% and specificity was 52.2% for any combined positivity, and sensitivity was 19.3% and specificity was 96% for strong positivity.

Table.

IHC immunoscore results for UCHL1 and claudin-4 in pure TCC cases, TCC-like, and HGSC-like components of mixed TCC and conventional HGSC.

Proteins	Immunoscores	A	B	C	D	E
		Pure TCC (%), n = 59	Mixed (TCC-like) (%), n = 29	Mixed (HGSC-like) (%), n = 28	Mixed (TCC-HGSC average), (%), n = 28a	HGSC (%)
UCHL1b	0 (0)	11 (19)	4 (14)	11 (39)	7.5 (27)	88 (37)
	1 (1,2,3)	9 (15)	9 (31)	6 (21)	7.5 (27)	55 (23)
	2 (4,6)	13 (22)	8 (28)	5 (18)	6 (21)	62 (26)
	3 (8,9,12)	26 (44)	8 (28)	6 (21)	7 (25)	32 (14)
Claudin-4c	0 (0)	2 (3)	4 (14)	2 (7)	3 (11)	51 (22)
	1 (1,2,3)	9 (15)	4 (14)	1 (4)	2.5 (9)	86 (37)
	2 (4,6)	14 (24)	7 (24)	6 (21)	6 (21)	65 (27)
	3 (8,9,12)	34 (58)	14 (48)	19 (68)	16.5 (59)	33 (14)

Abbreviations: IHC, immunohistochemical; UCHL1, ubiquitin carboxyl-terminal esterase L1; TCC, transitional cell carcinoma of the ovary; HGSC, high-grade serous carcinoma.

^aRepresents the average of the scores between the TCC-like and HGSC-like components of the mixed TCC cases. Only cases for which both components could be scored were included in the average.

^bA vs E (p < 0.00001); B vs C (p = 0.29); A vs D (p = 0.17); D vs E (p = 0.035).

^cA vs E (p < 0.00001); B vs C (p = 0.13); A vs D (p = 0.93); D vs E (p < 0.00001).

In comparison, the expression of UCHL1 and claudin-4 in LGSC cases was comparable with what was seen in HGSC (p = 0.11 and 0.49, respectively) but weaker than TCC (p = 0.0023 and 0.00064, respectively).

4. Discussion

The clinical relevance of the TCC variant of HGSC is controversial; the subclassification evidence is contradictory and contaminated by outdated and irreproducible diagnostic criteria that make it infeasible to compare study cohorts. Despite the genomic similarities supporting consideration of TCC as a variant of HGSC in the 2014 WHO classification system, we sought to compare global protein expression profiles in these closely related tumors with the goal of better understanding TCC and to identify protein biomarkers that could be used diagnostically.

4.1. TCC and HGSC show different protein expression profiles

We were able to accurately distinguish TCC from HGSC based on protein expression, using unsupervised analysis, which supports the hypothesis that they have different tumor biology, as suggested by their different histological features. There were 964 proteins that were found to be differentially expressed between TCC and HGSC. Importantly, this suggests the possibility for development of diagnostic biomarkers for TCC that could be used in research and, potentially, clinical practice. In comparison, LGSC showed 658 differentially expressed proteins when compared with HGSC, ie, the number of differentially expressed proteins is similar in magnitude between TCC vs HGSC and LGSC vs HGSC, with the latter comparison being between closely related but distinct tumor types with important differences in tumor biology and response to treatment. That the clear distinction between TCC and HGSC was possible and was based on a large number of differentially expressed proteins is the main result of this study.

4.2. Pathways associated with aggressive behavior are downregulated in TCC compared with HGSC

Studies demonstrating better prognosis for TCC have hypothesized that the different types of observed tumor growth, a pushing margin for TCC as opposed to an infiltrating border for HGSC, could explain the survival difference [7]. Our pathway analysis could be related, in theory, to a less aggressive behavior for TCC. We identified a downregulation of homologous recombination in TCC, which would support previous evidence of the positive association between TCC and BRCA1/2 mutations [12]. Homologous recombination deficiency in HGSC is associated with a better prognosis, related to increased sensitivity to current therapies [29–32].

4.3. UCHL1 and claudin-4 expression are higher in TCC than in HGSC

Proteomic analysis identified 964 proteins significantly more highly expressed in TCC than in HGSC. IHC validation of proteomic analysis confirmed higher expression of two proteins in TCC than in HGSC: UCHL1 and claudin-4. Those were selected based on our proteomic results, biologic relevance such as previous association with malignancies, and antibody availability. As part of the biomarker validation process, we also evaluated commercially available antibodies for 12 other proteins high in TCC compared with HGSC; however, they could not be used successfully on FFPE material, and validation for those proteins was not possible. Thus, availability of suitable antibodies remains a significant barrier to translating proteomic results into clinical practice, if antibody-based detection of proteins is to be used. Nevertheless, our data suggest that there are many additional candidate TCC biomarkers, providing suitable antibodies, to be obtained.

UCHL1, a ubiquitin-protein hydrolase, also known as PGP 9.5, has been associated with a worse prognosis in breast cancer, with an increased rate of metastasis potentially mediated through UCHL1 deubiquitinating the transcription factor Hypoxia-inducible factor 1 (HIF-1) alpha to promote epithelial-to-mesenchymal transition [33–35]. In gastric carcinomas, it is regarded as an oncogene [36,37]. However, others have suggested it to be a tumor suppressor because UCHL1 methylation is seen in tubo-ovarian carcinomas [38] as well as in nasopharyngeal carcinomas, colorectal carcinomas, and giant-cell tumors in which it was suggested to be responsible for malignant transformation [39–42]. Moreover, UCHL1 knockdown in ovarian cancer cell lines had increased cell growth, decreased apoptosis, and increased cisplatin resistance [43]. UCHL1 was also shown to be expressed in tumors of neuroendocrine origin such as gastroenteropancreatic neuroendocrine tumor, Merkel cell carcinoma, granular cell tumor, and melanoma and was used to assist pathologists in certain diagnostic scenarios [44–46].

Claudin-4 is a tight junction protein associated with worse prognosis in oral squamous cell carcinomas and breast cancer [47,48]. Evidence also suggests that it acts as a tumor suppressor by restraining pro-oncogenic activation of EPHA2 (ephrin type-A receptor 2) and limits metastatic potential by sustaining the expression of E-cadherin and limiting beta-catenin signaling [49,50]. Low expression correlates with metastasis, recurrence, and poor prognosis in esophageal cancer [51]. Knockdown of claudin-4 in tubo-ovarian cancer cell lines increases chemoresistance [52]. In the endometrium, the expression of claudin-4 in epithelial malignancies was associated with worse prognosis and is considered a potential therapeutic target [53–55]. IHC validation, consistent with the proteomic analysis, showed a higher frequency of high immunoscores in TCC than in HGSC. Unlike conventional HGSC, the HGSC-like component of our series of mixed TCC showed a high percentage of strongly staining cases (66%). This potentially suggests a biological difference between TCC and conventional HGSC.

MCM7 is part of a highly conserved group of DNA-binding proteins in the prereplication complex, important for DNA synthesis during the cell cycle [56]. It has been associated with tumorigenesis in ovarian, gastroesophageal, colorectal, prostatic, and lymphoid malignancy [57–63]. In our analysis, despite MCM7 emerging as a strong differential protein in the global proteomic analysis (p < 0.00001), the IHC validation failed to show any potential in distinguishing between conventional HGSC and TCC. This discrepancy could be a result of poor IHC specificity for the MCM7 protein or an overestimation of the expression difference between both tumor types by the proteomic analysis.

4.4. MS-based proteomic analysis can identify tumor-type—specific biomarkers from FFPE samples

Our proteomic results were matched by the IHC staining pattern for the two proteins we validated, ie, higher expression in TCC than in HGSC. Unfortunately, although statistically significant, the marker was neither sensitive nor specific enough to be used in isolation to distinguish between TCC and HGSC in practice. Further investigation of other highly expressed proteins in TCC in the proteomic data is underway to identify other biomarkers to distinguish these entities. It is possible that a single robust or small panel of TCC biomarkers might not differentiate TCC and HGSC owing to the very close relationship between TCC and HGSC.

The mixed tumors are between TCC and HGSC in terms of IHC expression of UCHL1; the differences between mixed TCC and HGSC are statistically significant, whereas the differences between pure TCC and mixed TCC are not. This may simply reflect an increase in statistical power owing to the very high number of HGSC cases (n = 237) compared with the TCC cohorts. The results of analysis of UCHL1 in pure TCC, mixed TCC, and pure HGSC suggest that TCC changes may occur as a continuum that is part of the morphological spectrum of HGSC. Although these morphological changes may correlate with outcomes, or BRCA mutation, there is not a clear-cut diagnostic metric between TCC and non-TCC that allows TCC to be established as a robust diagnosis. In other words, the biological phenomenon, so clearly documented in the proteomic data, may exist on a continuum and never be established as a binary diagnostic variable. One of our limitations was availability of effective antibodies for IHC, and it is possible that a non—antibody-mediated multiplex detection system to characterize expression levels of multiple proteins, such as MS, would allow for clear distinction of TCC and HGSC in clinical practice.

In conclusion, this is the first study to report a comprehensive proteomic analysis for TCC and that shows a clear distinction between TCC and HGSC based on global protein expression profiles. We also show a proof of concept that proteomic analysis of FFPE samples can be translated into IHC biomarkers and was able to demonstrate that UCHL1 and claudin-4 are more highly expressed in TCC than in conventional HGSC. The pathway analysis data hint at biologic distinctions between TCC and HGSC, which may explain the differences in survival observed in some studies. Nonetheless, an important negative result of this study is that there are no single proteins for which the expression level can serve as a sensitive and specific biomarker of TCC vs HGSC; none of the 5010 proteins provided clear distinction between these tumors, indicating the closely related nature of TCC and HGSC and suggesting that they exist on a continuum.