Specific IHC markers supporting the breast origin of an unknown carcinoma are important and helpful for diagnosis, especially ER-negative or triple-negative tumors. Currently, GATA3, GCDFP-15, and mammaglobin are commonly used panels to support breast origin, of which GATA3 is the most widely used. GATA-binding protein 3 (GATA3) is considered to be the most prevalent transcription factor involved in the proliferation and differentiation of ductal epithelium of the breast [32,33,34], which is linked to ER signaling [35, 36]. A number of studies have demonstrated that GATA3 is a superior marker for ER+ breast carcinoma than GCDFP-15 or mammaglobin, with a sensitivity consistently over 90% [7, 16, 35]. However, the sensitivity of GATA3 is significantly lower in ER-negative or TNBC subtypes, ranging from 15 to 60% in various studies [7, 16, 31, 35, 37, 38]. Our data also show that GATA3 exhibited extremely high sensitivity in 93.4% of ER/PR+ breast carcinomas, while up to 23.0% HER2 + BC and 60.4% TNBC were negative for GATA3. In addition, GATA3 is not a breast-specific marker that can label other common sources of tumors [39], including urothelial carcinomas, squamous cell carcinomas, lung adenocarcinoma, pancreatic adenocarcinomas, endocrine tumors, soft tissue sarcomas, and others. Currently, no single IHC marker is entirely breast-specific; GATA3 should be applied as part of an IHC panel, and more specific biomarkers are still required in the diagnostic setting.

Using mRNA sequencing and proteomic data from TCGA and CPTAC of 24 different solid tumors, we identified six potential genes that are specifically upregulated in breast carcinoma: NAT1, MGP, SCGB2A2, LMX1B, TFAP2B, and TRPS1. Similar to the approach used by Ai et al. [29], the genes highly expressed in breast carcinoma compared to all other tumor types and equally highly expressed in all four PAM50 subtypes were identified as candidate markers. Moreover, our approach includes two more steps. Since gene expression may change during the process of metastasis, we also removed genes showing decreased expression in metastasis using RNA-Seq data from primary breast lesions and their paired metastases. In addition, we aimed to look for potential biomarkers that can be used in IHC-based assays, which are cellular protein labeling techniques; thus, we further selected genes with a high correlation between protein and RNA expression using protein expression data from CPTAC. SCGB2A2 (mammaglobin) and the most recently reported breast-specific marker TRPS1 are both listed in our final candidate markers, which suggests the robustness of our approach. This bioinformatic analysis process (Fig. 2) not only identifies MGP and TRPS1 as novel candidate IHC markers to support breast origin but also provides a new approach for the future selection of specific biomarkers in other tumor types.

Trichorhinophalangeal syndrome type 1 (TRPS1) is named for a very rare hereditary disease with mainly autosomal dominant inheritance features characterized by craniofacial and skeletal abnormalities with damage and mutation affecting chromosome 8q [40]. TRPS1 is reported to be a transcriptional repressor that binds specifically to GATA sequences and represses the expression of GATA-regulated genes which function in vertebrate development, especially in the process of chondrocyte proliferation and differentiation [41, 42]. Some studies have suggested that TRPS1 may also act as a critical modulator in mammary epithelial cell growth, differentiation, and breast cancer development via epithelial–mesenchymal transformation, DNA replication, and mitosis [43, 44]. A recent study by Ai et al. reported that TRPS1 could serve as a sensitive and specific marker for breast carcinomas [31]. TRPS1 exhibited high sensitivity in ER/PR+ (98%), HER2+ (87%), and TNBC (86%) subtypes on TMAs. On the other hand, TRPS1 showed no or little expression in other tumor types. Parkinson et al. [37] and Yoon et al. [38] further verified the utility of TRPS1, showing higher sensitivity in the HER2+ and TNBC subgroups. Although the commercial TRPS1 antibody used in our study (EPR16171 from Abcam) is different from Ai’s and Parkinson’s study (TRPS: PA5-845874 from Invitrogen/Thermo Fisher), similarly high TRPS1 expression (91.2–93.1%) is also found in all types of breast carcinoma with the largest sample size thus far. Our findings confirm that both clones of TRPS1 are sensitive markers supporting breast origin. In addition, previously reported abnormal membranous expression of TRPS1 was not observed in our cohort.

Previous studies reported that MGP is mainly secreted by chondrocytes [45, 46] and vascular smooth muscle cells [47], and it is considered a marker of vitamin K status in bone and vasculature, substantiating the role of MGP in extracellular matrix calcification regulation [48, 49]. MGP was recently found to be overexpressed in various types of cancer [50,51,52] and was reported to promote tumor progression by regulating angiogenesis [53]. In breast carcinoma, Yoshimura et al. and Gong et al. demonstrated that high MGP mRNA expression was associated with poor prognosis [52, 54]. However, whether MGP can serve as a breast-specific marker is unknown.

In our cohort of 1201 breast carcinomas, every case matched benign breast ducts in a separate TMA or whole-slide section. MGP displays cytoplasmic labeling in nearly all ductal epithelial cells with various strengths but not in myoepithelial cells. Perivascular smooth muscle can be used as an internal positive control for MGP (Fig. 1). MGP was verified as a reliable marker with extremely high sensitivity in all subtypes of breast carcinoma (87.3–91.2%), which is comparable to TRPS1 and much higher than GATA3 in HER2+ and TNBC subtypes.

We noticed that most MGP-positive cases demonstrated moderate and multifocal cytoplasmic staining patterns. There were generally more cases showing extensive and strong positivity for GATA3 and TRPS1 (greater than 49%) than for MGP (less than 40%, adjusted p < 0.001), except for the TNBC group (Table 1). Among the positive staining cases, 26.6% (286/1075, Table 1) showed mild positive of MGP, significantly higher than that of TRPS1 (14.1%, 157/1109, adjusted p < 0.001) and GATA3 (10.4%, 96/922, adjusted p < 0.001). More cases were categorized as mildly positive for MGP may be due to its cytoplasmic staining pattern, which is not preferable or easy to interpret subjectively like nuclear staining markers such as TRPS1 and GATA3. Mild cytoplasmic positivity tends to be more easily recognized as nonspecific or unstable as compared with mild nucleus positivity, which does affect the value of MGP as a single marker to determine breast origin in the clinical practice. Other commercial MGP antibodies could be also tested and verified in further studies. Even so, our data suggest that MGP has much better and more stable sensitivity than conventional nuclear (GATA3, SOX10 [38, 55, 56]) or cytoplasmic biomarkers (GCDFP15, mammaglobin) used to determine breast origin. The moderate–high positivity rate of MGP was significantly higher than that of GATA3 in TNBC-NST (65.3% vs. 34.8%, adjusted p < 0.001) and TNBC-MBC subtypes (70.7% vs. 17.4%, adjusted p < 0.001), suggesting the high sensitivity of MGP specially in the most troubling TNBCs. In addition, we observed that 239 GATA3-negative cases and 75 TRPS1-negative cases were positive for MGP, and 69 GATA3-mild positive cases and 96 TRPS1-mild positive cases showed moderate–high positive for MGP. Thus, using MGP, GATA3, and TRPS1 as a novel IHC panel significantly increased the sensitivity from 76.8–92.3% of the single marker (MGP, GATA3, or TRPS1) to 93.0–99.3% (≥ 1 positive or ≥ 2 markers positive for the GATA3, MGP, and TRPS1 panel).

Although our IHC data of MGP were collected from primary breast carcinomas, our bioinformatics analysis revealed that MGP mRNA was not significantly changed between paired primary tumors and their metastases (Additional file 3: Table S2), which suggests that similar MGP expression could be found in metastatic breast carcinomas. Further verification of MGP is required in metastatic lesions as well as special types of invasive breast carcinoma, such as salivary gland-type tumors and neuroendocrine carcinoma.

In the present study, we included 144 TNBC-NSTs and 140 TNBC-MBCs. GATA3 was expressed in only 39.6% of TNBC-NSTs and 47.1% of TNBC-MBCs and was mostly weakly positive, which is consistent with previous studies [31, 37, 38]. TRPS1 and MGP maintained high sensitivity in both TNBC-NSTs (84.7% and 86.1%) and TNBC-MBCs (97.9% and 88.6%). Focusing on MBCs, the sensitivity of TRPS1 (137/140, 97.9%) in our cohort was slightly higher than those reported by Ai et al. [31], Parkinson et al. [37], and Yoon et al. [38], which were 86.5% (45/52, adjusted p < 0.05), 91.0% (61/67, raw p = 0.061), and 95.0% (134/141, raw p = 0.33), respectively. TRPS1 exhibited a larger portion of strong positivity among the positive cases in TNBC-MBCs (112/137 [81.8%]) compared with the ER/PR+ (313/526 [59.5%], adjusted p < 0.001) or HER2+ (177/324 [54.6%], adjusted p < 0.001) group (Table 1). When MBCs were stratified by subtype, we observed that GATA3 showed relatively higher sensitivity in SqCCs (62.1%) than in other subtypes (22.5–37.5%). The majority of cases (85%, 34/40) in MBC with mesenchymal differentiation group showed chondroid/osseous differentiation, and both TRPS1 and MGP had the highest positivity in MBCs with chondroid/osseous differentiation (100% and 88.2%, respectively), followed by SqCCs and SpCCs. Interestingly, fibromatosis-like metaplastic carcinomas (FMCs) in our cohort were all positive for MGP (8/8). A total of 37.5% (3/8) and 87.5% (7/8) of FMCs were positive for TRPS1 and GATA3, respectively, which is inconsistent with Parkinson et al. [37] showing no single case with positive staining of TRPS1 or GATA3 in FMCs (0/3). In addition, we observed that 64 GATA3-negative MBCs and 2 TRPS1-negative MBCs were positive for MGP. The combined use of MGP, GATA3, and TRPS1 increased the sensitivity from 47.1–97.9% of the single marker (MGP, GATA3, or TRPS1) to 90.7–100.0% (≥ 1 positive or ≥ 2 markers positive for the GATA3, MGP, and TRPS1 panel) in MBCs. All these data suggest that both MGP and TRPS1 maintain excellent sensitivity in different subtypes of metaplastic breast carcinomas. However, TRPS1 may play a role in chondro-osseous differentiation. Wang et al. [57] found that TRPS1 was highly expressed in chondro-osseous sarcomas from both breast and extramammary sites, including heterologous components within malignant phyllodes tumors. Coincidentally, MGP is highly abundant in cartilage and acts as a critical regulator of calcification and turnover of bone and cartilage. The previous study showed that tumors exhibiting cartilaginous/osseous differentiation such as osteosarcoma and chondrosarcoma had high MGP expression [46, 58], and they can also metastasize to bone and lung-like breast cancer [59,60,61,62]. It would be hard to differentiate metaplastic breast carcinoma with cartilaginous/osseous differentiation from these tumors simply based on MGP positivity. Thus, pathologists should be cautious when faced with positive expression of MGP or TRPS1 in chondroid/osteoid components, especially with limited biopsy tissue. Further verification of MGP is also required in sarcomas and malignant phyllodes tumors.

MGP was first isolated from bovine bone matrix in the 1980s [43]; since then, its expression has been demonstrated in normal endothelial cells, fibroblasts, chondrocytes, and vascular smooth muscle cells. Our data show that MGP had negative expression in normal organs, including the ovary, biliary duct, lung, colorectum, stomach, bladder, and thyroid. Interestingly, we observed that MGP was constantly expressed in normal hepatocytes, but the positive expression was detected in only 31.1% of hepatocellular carcinomas. In addition, previous studies also demonstrated that MGP was abundantly expressed in normal kidneys, specifically in the epithelium of Bowman's capsule and proximal tubules, where the activated protein contributes to maintaining renal microvascular traits [63, 64]. Consistently, we found that MGP was predominantly expressed in normal renal tubules, but the positivity rate significantly dropped to 5.0% in renal cell carcinomas. It is difficult for MGP itself to differentiate breast carcinoma from hepatocellular or renal cell carcinoma. The joint application of TRPS1 and GATA3 may be helpful since TRPS1 [31, 37] and GATA3 [7, 65] have been proven to be rarely positive in these tumors.

Since high-grade ovarian serous carcinoma and breast carcinoma share similar morphologies and immunophenotypes, such as a micropapillary architecture and ER positivity, the diagnosis can be challenging. Our results showed that only 2.4% of ovarian serous carcinomas had focal MGP expression, which is lower than the reported positivity of GATA3 (~6%, [7, 66]) and TRPS1 (14%, [31]). Thus, this GATA3-MGP-TRPS1 panel may need inclusion with Pax-8 and WT-1 to differentiate breast carcinoma from serous carcinoma.

Poorly differentiated lung adenocarcinomas have been frequently reported as TTF-1-negative and occasionally labeled for ER [67], while individual cases of breast carcinoma may show TTF-1 staining [68]. Thus, differentiating breast carcinoma and lung adenocarcinoma is common and sometimes difficult in clinical practice. In the current study, MGP was rarely expressed in lung adenocarcinomas (0.7%), which is lower than the previously reported positivity of GATA3 (∽8%, [7, 69]) and TRPS1 (2–3%, [31, 37]), indicating that MGP is a good marker to differentiate breast cancer from lung adenocarcinoma.

We also found that MGP was positive in only 2 of 218 cases (0.9%) of urothelial carcinoma, which is known to be frequently labeled with GATA3 (70–90% [7, 70],). According to the documented literature and our data, the positivity of MGP, GATA3, and TRPS1 is extremely rare in other tumor types, such as cholangiocarcinoma and colorectal, gastric, and thyroid carcinomas. Ai et al. [31] found low TRPS1 expression in one melanoma, while none of the melanomas enrolled in our cohort was MGP-positive. Further investigation of MGP expression in other tumor types is needed, especially those for which relatively high TRPS1 or GATA3 expression has been reported, such as salivary duct carcinomas and pancreatic adenocarcinoma.

A limitation of this study is that the TMA samples we used may not be able to adequately represent the intra-tumor expression heterogeneity of the IHC markers [62, 63]. A multicenter prospective study using standard whole-tissue sections should be undertaken to fully validate the value of MGP in determining breast origin. Our study used a relatively higher number of total breast carcinomas and metaplastic breast carcinomas than the recently published studies to identify new breast cancer marker [31, 37, 38]. Nevertheless, more cases are included in our ongoing study to further evaluate the sensitivity and specificity of MGP in metastatic breast carcinomas, special type of invasive breast carcinomas, neuroendocrine neoplasms, salivary gland-type tumors (either primary in breast or salivary gland), as well as tumors exhibiting cartilaginous/osseous differentiation.