Transcriptomic alterations underlying metaplasia into specific metaplastic components in metaplastic breast carcinoma

The study protocol was approved by the Institutional Review Board of National Taiwan University Hospital (Approval No. 201711051RINC). From the Department of Pathology of the hospital, we retrieved formalin-fixed, paraffin-embedded (FFPE) surgical specimens collected between 1998 and 2019 from 27 patients with biphasic MpBC who had dissectible tumor components. The 27 cases comprised metaplastic carcinoma with heterologous mesenchymal differentiation (n = 12), spindle cell carcinoma (n = 9), squamous cell carcinoma (n = 3), and mixed metaplastic carcinoma (n = 3). Ten 10-µm hematoxylin-counterstained slides of each dissectible NST component and paired metaplastic component were prepared for needle-assisted microdissection, in which a 27-gauge needle was used under 40 × magnification. A total of 59 dissected tumor components were collected for RNA extraction: these comprised NST components (n = 27) and paired metaplastic components, namely SPS (n = 12), RHA (n = 6), matrix-producing (chondroid, n = 9; osteoid, n = 1), and SQC (n = 4). All the six RHA components showed no convincing staining for myogenic markers (desmin and myogenin). The chondroid and osteoid matrix-producing components are hereafter referred to as MAT and OGS, respectively. The tumor size and the status of lymph node metastasis were recorded for all specimens. Lymph node metastasis was observed in 10 cases of MpBC with both carcinomatous and sarcomatous components. Seven of these cases involved only carcinomatous components in the metastatic lymph nodes. The other three cases involved both carcinomatous and sarcomatous components, with the carcinomatous components being predominant.

ER (SP1, Ventana, Tucson, AZ, USA), PR (1E2, Ventana), and HER2 (4B5, Ventana) staining was performed using the Ventana iVIEW DAB detection kit with an autoimmunostainer (Ventana BenchMark). Specimens demonstrating HER2 (2 +) were further tested for HER2 through fluorescence in situ hybridization (FISH; PathVysion, Abbott, Abbott Park, IL, USA). Immunohistochemistry was performed to verify the presence of differentially expressed genes in the metaplastic components, the intrinsic gene sets of NST components, and the differentially expressed genes associated with lymph node metastasis. Primary antibodies against EPAS1 (SC13596; Santa Cruz Biotech, Dallas, TX, USA), SLC2A1 (SC377228), IL1RA (SC374084), CAV1 (EP353; Bio SB, Santa Barbara, CA, USA), FBN1 (HPA021057; Sigma-Aldrich, St. Louis, MO, USA), HAPLN1 (HPA019482), COL9A3 (HPA040125), PYCARD (HPA054496), PDGFRA (HPA004947), NCAM1 (MRQ-42, Ventana), and SOX10 (SP267, Cell Marque, Rocklin, CA, USA) were used.

RNA isolation was conducted using the Roche High Pure FFPE RNA Isolation Kit (Roche Molecular Systems, Pleasanton, CA, USA). To ensure sample purity (optical density 260/280 nm; ratio 1.7–2.5), the RNA concentration was estimated using the NanoDrop ND-1000 spectrophotometer and the Qubit 3.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). The GEP was analyzed using the NanoString Breast Cancer 360 (BC360) Panel on a NanoString nCounter FLEX platform (NanoString Technologies, Seattle, WA, USA). The BC360 Panel contains 770 genes across 23 breast cancer-related pathways and processes as well as 30 signatures for measuring tumor and immune activities [14, 15]. Intrinsic molecular subtypes of PAM50 were used to classify breast cancer into four subtypes (luminal A, luminal B, HER2-enriched, and basal-like) [16, 17]. Risk of recurrence (ROR) scores are derived from the expression profile of 46 PAM50 genes with a weighted sum of the proliferation score, the four subtype correlations and tumor size information to calculate a score between 0 and 100 [18, 19]. The whole tumor size was used in this calculation, so the comparison of ROR score between NST and its paired metaplastic component mainly indirectly compared the expression of proliferation-associated genes of these two components. BC360-defined signatures were scored using nSolver software (NanoString Technologies).

Gene set enrichment analysis (GSEA) was performed using the enrichment analysis function in the clusterProfiler R package. The gene sets used in the GSEA were obtained from the c2.cp, c5.bp, and hallmark collections in the Molecular Signatures Database (MSigDB; version 7.0). We used a preranked GSEA to analyze gene lists ranked by the − log₁₀(p) * sign(log₂ fold-change), where p was derived from paired t tests for paired samples or from t tests for unpaired samples.

Raw data from this study were deposited in the Gene Expression Omnibus (GEO) with Accession Number GSE212245.

Processing, analyses, and plotting were conducted using R3.5.2 software (http://​www.​r-project.​org/​). The paired t test, t test, and analysis of variance (ANOVA) were applied to paired samples, unpaired samples, and multigroup samples, respectively. A hierarchical clustering analysis was performed using the pheatmap R package with the clustering distance set to the “euclidean” default and with the clustering method set to “ward.D2.”

In total, 59 dissected NST components and paired metaplastic components were collected and subjected to gene expression profiling by using the NanoString BC360 Panel (Fig. 1). The clinicopathological characteristics and molecular intrinsic subtypes of PAM50 are presented with ROR scores in Table 1, and the information of patients’ treatment and outcome is shown in Additional file 4: Table S1. All 31 metaplastic components were classified as the basal-like subtype and were determined to be ER− , PR− , and HER2− . Four and two NST components were classified as the HER2-enriched and luminal A subtypes, respectively. The components were consistent in immunohistochemistry, except case BT83, which was classified as the HER2-enriched subtype and HER2 2+ in immunohistochemistry testing, but HER2 testing by FISH was negative. The remaining 21 NST components were classified as the basal-like subtype, with 19 components being ER− , PR− , and HER2− and the other 2 components demonstrating 5% and 20% ER in immunohistochemical staining. Notably, compared with that of the paired NST components, the ROR score was higher or equal in 83.3% (10/12) of SPS components and in 100% (6/6) of RHA components. However, it was lower in 77.8% (7/9) of MAT components and was lower or equal in 75% (3/4) and 25% (1/4) of SQC components, respectively (Fig. 2).

We performed hierarchical clustering and principal component analysis (PCA) to illustrate the relationship among the 59 components on the basis of the overall GEPs (Fig. 3A-E and Additional file 5: Table S2). Although hierarchical clustering revealed a modest distinction among the metaplastic component subtypes, intracase clustering of GEPs was noted in 14 of the 27 MpBC cases (Fig. 3A), suggesting that the overall intercase tumor heterogeneity was high. PCA indicated relatively clustered GEPs in each metaplastic subtype, but the NST specimens displayed greater heterogeneity than the specimens corresponding to the metaplastic subtypes (Fig. 3B). The high intercase heterogeneity among the NST components was further demonstrated by the relatively wide range of Euclidean distances between those components in the GEPs (Fig. 3C). The distribution of the specimens in the PCA further revealed a modest overlap between the RHA and SPS components, differentiating them from the MAT and SQC components (Fig. 3B). However, when metaplastic subtype-specific differentiation was considered, GEP differences between NST components and paired SPS or RHA components were greater than those between NST components and paired MAT or SQC components (Fig. 3D). This finding was supported by the fact that a larger Euclidean distance was observed in the GEPs between NST components and paired SPS or RHA components than between paired MAT or SQC components in the four MpBC cases with multiple metaplastic components (Fig. 3E). These results suggest that the GEPs of metaplastic components are subtype dependent.

To investigate the functional difference between NST components and metaplastic components, the genes differentially expressed between the 59 NST components and paired metaplastic components were identified. Of these genes, 55 (31 upregulated, 24 downregulated) were identified in SPS components, 22 (14 upregulated, 8 downregulated) were identified in MAT components, 31 (12 upregulated, 19 downregulated) were identified in RHA components, and 7 (4 upregulated and 3 downregulated) were identified in SQC components (paired t test, p < 0.01). These differentially expressed genes were clustered into four groups (G1–G4) according to their expression patterns across all samples (Fig. 4A and Additional file 6: Table S3). To verify the genes that were differentially upregulated between the NST components and the paired SPS, MAT, RHA, or SQC components, we selected a representative gene from each of the four metaplastic components for immunohistochemical analysis. FBN1, SLC2A1, EPAS1, and IL1RN were differentially expressed in the SPS, MAT, RHA, and SQC components, respectively, and were subjected to immunohistochemical verification (Fig. 4B). The biological processes involving the differentially expressed genes are displayed in Fig. 4C. Most genes upregulated in the SPS components belonged to G3. These genes were enriched in functions such as stem cell, cell adhesion, epithelial–mesenchymal transition (EMT), extracellular matrix organization, and growth factor responses (Fig. 4C). Most SPS-specific downregulated genes, which belonged to G1 and G2, were associated with nucleosome organization (e.g., HMGA1, HIST3H2BB, MIS18A, and ARID1A) as well as with cell cycle (e.g., PRKAA2, WEE1, MDM2, CDC7, and XRCC2) and cell development (e.g., EFNA3, CD24, and PIK3R3). The upregulated and downregulated genes corresponding to the RHA components demonstrated an overall similarity to those of the SPS components, which belonged to G3 and G1, respectively. Furthermore, despite some overlap in enriched functions, such as cell adhesion, cell development, stem cell upregulation (e.g., RHA-specific gene EPAS1 [20]), and EMT (e.g., RHA-specific gene BDNF [21]), the specific differentially expressed genes differed between the RHA and SPS components (Fig. 4A and C). By contrast, certain RHA-specific upregulated genes were associated with vascular endothelial growth factor (VEGF) signaling. Moreover, some RHA-specific downregulated genes were linked to cell adhesion and hypoxia. In addition, the SPS-specific downregulated genes that were associated with nucleosome organization and cell cycle were not downregulated in the RHA components. Mainly belonging to G4, the upregulated genes corresponding to the MAT components were associated with functions such as hypoxia (namely VEGFA, BNIP3, ADM, and SLC2A1) and apoptosis (namely BBC3, BNIP3, INHBB, FGFR3, and COL2A1) (Fig. 4A, C). The downregulated genes corresponding to the MAT components, primarily belonging to G2, were associated with cell-cycle control (e.g., BAX, SPRY1, PSMB7, and PLCB1). Mainly belonging to G4, the upregulated genes in the SQC components, including NOD2, IL20RB, BCL2A1, and IL1RN, were linked to apoptosis, immune responses, and cell adhesion. Overall, despite some overlap between SPS and RHA components, the functions of the differentially expressed genes in each metaplastic component revealed distinct intrinsic molecular characteristics.

To identify the gene expression signatures underlying metaplastic processes in MpBC, the differentially expressed signatures defined by the BC360 Panel were identified (paired t test, p < 0.05) and visualized as a heat map (Fig. 5A and Additional file 7: Table S4). Consistent with the findings presented in Fig. 4A, C, we observed the expression of some overlapping signatures in the SPS and RHA components. The expression of the claudin-low, stroma, and macrophage signatures and various genes (including TGF-β and inhibitory immune genes PD-L2 and B7.H3) was higher in the SPS components than in the NST components (Fig. 5A, B). By contrast, differentiation signatures and genes including ESR1, ERBB2, and T cell immunoreceptor with Ig and immunoreceptor tyrosine-based inhibitory motif domains (TIGIT) were downregulated in the SPS components. We further performed a GSEA on external gene sets related to EMT, which in turn is related to claudin-low signatures and TGF-β signaling. The SPS components exhibited high activity of EMT and TGF-β signaling (Fig. 5B). Compared with the NST components, the RHA components had a more distinct claudin-low signature and greater macrophage abundance but a less distinct differentiation signature. This was further demonstrated at the gene expression level of macrophage-related genes (CD84, CD163, and CD68) and validated through GSEA on gene sets linked to the EMT and cell differentiation (Fig. 5C). TGF-β was upregulated in the SQC components; this finding was validated through the GSEA of TGF-β-responsive genes (Fig. 5A, D). As presented in Fig. 4A, because many MAT-specific genes were linked to hypoxia, we examined the expression of hypoxia-responsive genes collected from MSigDB. These genes were relatively highly expressed in the MAT components compared with in the NST components (Fig. 5E). Compared with those in the NST components, the MHC2 signature and TIGIT in the MAT components were downregulated (Fig. 5A). These results were consistently obtained for the four cases with multiple metaplastic components, with claudin-low, macrophage, and TGF-β signatures scores being higher and differentiation signature scores being lower in the SPS and RHA components than in the paired NST components. Two specimens contained MAT components, and both had higher hypoxia signature scores than the paired NST components. The only case with SQC component had a higher TGF-β score than did the paired NST component (Additional file 8: Table S5).

We next investigated whether metaplastic types were determined by the intrinsic gene expression of their paired NST components. We restricted our analysis to the 22 NST components with only one type (n = 19) or predominantly one type (> 95%; n = 3) of a paired metaplastic component. As displayed in Fig. 6A, 44 differentially expressed genes were identified among the metaplastic types (ANOVA, p < 0.05). Organized according to these 44 genes, the 22 NST components were separated into three clusters dominated by two sets of genes, namely subgroups S and subgroup M (Additional file 9: Table S6). The clusters were highly correlated with types of associated paired metaplastic components as follows: MAT (M-high/S-low), SQC (M-low/S-high), and SPS/RHA. We further clustered the 31 metaplastic components on the basis of the same 44 genes and identified three clusters with reference to the genes in subgroups S and M. The analysis achieved an accuracy of 74.2% (23/31), although one RHA component and 5 SPS components were misclassified into the MAT cluster and two MAT components were misclassified into the SPS/RHA cluster. The differentially expressed genes among the 31 metaplastic components were employed in separating 31 metaplastic components modestly correlated with metaplastic type. However, under these gene sets, the 22 NST components could not be clustered with their corresponding paired metaplastic types (Additional file 1: Fig. S1). The 27 MpBC samples were subjected to immunohistochemical analysis for M-subgroup (SOX10, NCAM1, HAPLN1, and COL9A3) and S-subgroup (PYCARD) proteins (Fig. 6B–G and Additional file 10: Table S7). Of the 27 MpBC cases, 22 had only one type or predominantly one type of paired metaplastic component. In nearly all MpBC cases with the MAT component, the coexpression of all four M-subgroup genes was observed in the MAT and NST components. In MpBC cases with the SQC component, the coexpression of the S-subgroup gene was observed in the SQC and NST components. However, in MpBC cases with SPS or RHA metaplasia, both M-subgroup and S-subgroup genes were generally not or less frequently expressed in the NST and metaplastic SPS and RHA components. A similar trend was observed in MpBC cases with multiple metaplastic components. The immunohistochemistry results were consistent with those of intrinsic gene clustering. To validate the significance of the intrinsic gene sets, 28 frozen samples from GSE57544 [12] were scored using gene set variation analysis (GSVA) against the M-subgroup and S-subgroup genes. Although no paired NST components were available for analysis, a high MAT and SQC signature score was observed in the chondroid and squamous components, respectively (Additional file 2: Fig. S2), thereby supporting our finding. Together, the observations presented suggest that the intrinsic gene expression of NST determines the metaplastic type.

Because the carcinomatous component was the predominant one present in metastatic axillary lymph nodes in the MpBC cases, we investigated whether any specific GEP in the NST was linked to nodal metastasis. The comparisons of GSEA of hallmark gene sets of NST components between cases with and without nodal metastasis revealed that cell cycle-related and cell proliferation-related pathways were negatively enriched in tumors with nodal metastasis (Fig. 7A). However, genes linked to the EMT and stem cells tended to be upregulated in tumors with nodal metastasis (Fig. 7A, B and Additional file 11: Table S8). Among these genes, CAV1, which is functionally associated with stem cell upregulation, and PDGFRA, which is functionally associated with EMT and stem cell upregulation, were selected for immunohistochemical verification in the 27 MpBC tissues. PDGFRA and CAV1 expression were both significantly more frequent in the NST components of MpBC cases with lymph node metastasis than in the NST components of those without metastasis (Fig. 7C, D), supporting the GSEA results. Because most NST components of MpBC are negative for ER, PR, and HER2, we validated the biological significance of the metastasis-associated genes by using data on triple-negative breast cancers from the Cancer Genome Atlas (TCGA). The results indicated a worse progression-free and overall survival in cases with high expression of the metastasis-associated genes (Additional file 3: Fig. S3). This finding supports the biological significance of the metastasis-associated genes.