Conditional overexpression of histone reader Tripartite motif containing protein 24 (TRIM24) in mouse mammary epithelia (Trim24COE) drives spontaneous development of mammary carcinosarcoma tumors, lacking ER, PR and HER2. Human carcinosarcomas or metaplastic breast cancers (MpBC) are a rare, chemorefractory subclass of triple-negative breast cancers (TNBC). Comparison of Trim24COE metaplastic carcinosarcoma morphology, TRIM24 protein levels and a derived Trim24COE gene signature reveals strong correlation with human MpBC tumors and MpBC patient-derived xenograft (PDX) models. Global and single-cell tumor profiling reveal Met as a direct oncogenic target of TRIM24, leading to aberrant PI3K/mTOR activation. Here, we find that pharmacological inhibition of these pathways in primary Trim24COE tumor cells and TRIM24-PROTAC treatment of MpBC TNBC PDX tumorspheres decreased cellular viability, suggesting potential in therapeutically targeting TRIM24 and its regulated pathways in TRIM24-expressing TNBC.
Advances in next-generation sequencing (NGS) and mapping of genome-wide epigenetic marks show that epigenetic regulators are frequently mutated and/or aberrantly expressed in numerous diseases, including cancers, often in correlation with poor survival or therapeutic resistance1. Among these epigenetic regulators are proteins with structural domains that “read” epigenetic modifications and selectively interact with specific histone or non-histone post-translational modifications (PTMs). These protein PTM readers are critical to interpretation of regulatory information encoded by protein PTMs. The central role of reader domain binding to histone PTMs offers therapeutic opportunities to exploit small molecule ligands that disrupt reader-PTM interactions. As a seminal example, small molecule ligand JQ1 binds the BRD4 bromodomain and disrupts BRD4 interactions with acetylated histone lysines, which leads to inhibition of oncogenic myc-regulated gene networks and potential therapeutic application2,3.
Our unbiased proteomic screen of embryonic stem cells for endogenous protein interactions with tumor suppressor p53 uncovered Tripartite Motif Protein 24 (TRIM24). TRIM24 is a multi-functional histone reader that ubiquitinates p53, via an N-terminal RING domain, and binds a specific signature of histone PTMs (H3K4me0/H3K23ac) via a combinatorial Plant Homeodomain (PHD) and bromodomain4,5,6. We found that TRIM24 is a co-regulator of estrogen receptor (ER)-regulated genes in breast cancer-derived MCF7 cells5. In immortalized human mammary epithelial cells (iHMECs), TRIM24 promotes loss of p53 and metabolic reprogramming that transform iHMECs, which engraft and form high-grade epithelial tumors7. High levels of TRIM24, determined by tumor immunohistochemistry (IHC), correlate with poor survival of breast cancer patients5. Interestingly, recent reports associate aberrantly high TRIM24 with poor patient prognosis, poor differentiation, advanced stage, chemo-resistance, and recurrence in a wide range of cancers, especially solid tumors8,9,10.
Given the functional studies in cellulo and correlative links between TRIM24 and human cancers, we hypothesized that TRIM24 is a potent oncogene that reprograms transcriptional networks and promotes development of aggressive breast cancers. We tested our hypothesis in a physiologically relevant setting by piggyBac transposon-mediated integration of a single copy of Trim24 under conditional expression control in the mouse genome. Here, we show that <3-fold overexpression of Trim24 in mammary epithelia is sufficient to induce mammary tumors, 67% of which are characterized as carcinosarcomas.
In breast cancer patients, carcinosarcomas are included within the diagnosis of metaplastic breast cancers (MpBC), a rare subgroup of breast cancers that are commonly triple negative (TNBC) and diagnosed by histological evidence of at least two cell types within an individual tumor: an epithelial component (carcinoma) and a non-epithelial component (sarcoma, matrix producing or squamous morphology)11. MpBCs are chemo-refractory and have a worse prognosis compared to non-metaplastic breast cancers, though little is known about the origin or drivers of these tumors11. Markers of epithelial–mesenchymal transition (EMT) are increased in expression in MpBC, compared to other subtypes of breast cancer12. MpBCs are usually high grade with larger tumor size, though often lymph node negative, at the time of presentation13. MpBCs also have frequent aberrations in TP53 and PI3K/AKT/mammalian target of rapamycin (mTOR) pathways, as well as prominent expression of vascular endothelial growth factor (VEGF) and programmed cell death-ligand 1 (PD-L1)14,15. Molecular characterization by next-generation sequencing (NGS) of MpBC patient tumors and patient-derived xenografts (PDXs) of MpBC revealed some potential targets; however, there is limited ability to define a molecular landscape with a large patient cohort in this rare TNBC subtype and few animal models exist to test novel targeted regimens14,16,17,18.
Here, we present evidence that TRIM24 is an oncogenic histone reader with the ability to drive an aggressive breast cancer phenotype in a mouse model with pathologic and molecular features that parallel MpBC human patient data. Our data suggest that TRIM24 may serve as a prognostic biomarker, as well as a potential predictor of disease response to select targeted therapies. Global gene expression analyses of TRIM24-driven carcinosarcomas were used to generate a TRIM24 metaplastic signature that revealed highly up-regulated glycolysis, EMT and PI3K pathways, the latter confirmed by global and single-cell analysis of tumor protein levels. TRIM24 aberrantly activates c-MET-PI3K-mTOR pathways, nominating pharmacological approaches that proved effective in inhibition of metaplastic carcinosarcoma primary cell viability. We assessed degradation of TRIM24, as a potential therapeutic approach, using MpBC PDX models that express high levels of TRIM24. A recently developed proteolysis-targeting chimera (PROTAC) fusion of small molecule IACS-9571, a bromodomain inhibitor of TRIM24, and the Von-Hippel Lindaeu (VHL) ligand19,20 inhibited viability of TRIM24-expressing MpBC and non-MpBC TNBC PDX tumorspheres. Thus, a mouse model of human MpBC, driven by histone reader TRIM24, was used to nominate PI3K/mTOR inhibitors and a TRIM24-specific PROTAC for further assessment as potential therapeutic approaches for a rare subclass of chemorefractory breast cancer patients.
TRIM24 over-expression drives mammary gland tumorigenesis
To conditionally express Flag-tagged, murine Trim24 in mammary epithelia, MMTV-driven Cre mice were bred with mice bearing a floxed ß-Geo-Stop-mTrim24-flag cassette transgene. Trim24LSL;MMTV-CreTg/0, hereafter referred to as Trim24COE, were viable, fertile, and over-produced TRIM24 in mammary epithelia (Supplementary Fig. S1a–d). The Flag-tagged mTrim24 transposon integrated as a single copy, into the q-arm of chromosome 1 without disrupting any known genes or regulatory elements, as determined by whole-genome sequencing and PCR (Supplementary Fig. S1e and Supplementary Data 3). Trim24 transcript expression in the whole mammary gland was increased ~2.5–3-fold in Trim24COE mice, as compared to controls (Supplementary Fig. S1c). TRIM24 and exogenous Flag-TRIM24 protein levels in the mammary epithelium were confirmed by western blot analysis (Supplementary Fig. S1d). Assessment of an aging cohort of virgin Trim24COE females (n = 40) revealed atypical hyperplasia of mammary ductal and lobular epithelia, ductal thickening, as well as increases in both side branching and terminal endbuds within 2 months (Fig. 1a, b). Trim24COE mice developed palpable mammary tumors within 5–14 months of age (penetrance of 46%). Necropsy revealed no other primary tumor sites and two mice with metastases: one to the lung and the other to the liver (Supplementary Data 4). Trim24LSL, lacking MMTV-Cre-expression, lacked any palpable tumor formation within 600 days and MMTV-Cre-only mice (n = 27) had occasional hyperplasia and spontaneous mammary tumor development (Fig. 1c). Trim24COE tumors overexpressed TRIM24 compared to controls, as determined by IHC (Fig. 1d). Trim24COE tumors exhibited overexpression of FLAG and TRIM24 proteins and lacked detectable expression of p53 (Fig. 1e), compared to Cre-only control mammary glands. These results show that Trim24COE mice develop mammary tumors at high penetrance, and these tumors maintain expression of Flag-tagged TRIM24.
A majority of TRIM24 over-expressing tumors are carcinosarcomas
Pathologic examination of mammary tumors formed in Trim24COE mice revealed three tumor types, including adenoma (8%), carcinoma (25%), and carcinosarcoma (67%) (Fig. 2a and Supplementary Data 4). Carcinosarcoma or metaplastic tumors are composed of two different tissue types: carcinoma (cancer of epithelial cells) and sarcoma (cancer of connective tissues or mesenchymal cells). A majority of human carcinosarcoma or metaplastic breast cancers (MpBC) are classified broadly as triple-negative breast cancer (TNBC) or claudin low and are highly associated with two major hallmarks of EMT, a loss of E-cadherin and gain of vimentin12. Assessment of Trim24COE carcinosarcoma tumors by IHC revealed gain of vimentin, loss of E-cadherin, estrogen receptor (ER) and progesterone receptor (PR), and no over-expression of receptor tyrosine kinase erbB-2 (ERBB2) (Fig. 2b and Supplementary Fig. 2). Thus, like a majority of human MpBC, Trim24COE carcinosarcomas are triple-negative.
Human MpBC is diagnosed by pathologists via light microscopy to determine the presence of carcinoid and sarcomatoid regions in sectioned tumor samples21. Pathologist evaluation of H&E stained, formalin-fixed, and paraffin-embedded (FFPE) sections of Trim24COE mouse carcinosarcomas (Fig. 2c) revealed considerable, morphological similarities between human metaplastic breast and Trim24COE tumors with both sarcomatoid and carcinoid regions. Expression of TRIM24 was additionally assessed by IHC of a tumor array with 46 MpBC patient biopsies and quantified according to previously published convention (Supplementary Data 5)22. This revealed four different expression patterns: (i) low staining of TRIM24 in the nucleus and cytoplasm, (ii) high nuclear TRIM24 expression and little cytoplasmic staining, (iii) high TRIM24 expression in the cytoplasm and little or no TRIM24 in the nucleus and, (iv) high TRIM24 expression in both nucleus and cytoplasm (Fig. 2d). Survival curves were generated with patients divided into classes i–iv by scored TRIM24 levels. MpBC patients were morphologically classified as subtypes: spindle cell (32%), matrix producing (43%), mixed spindle cell/matrix producing (15%), chondrosarcomatoid (2%), osteochondroid (2%), and squamous (2%). High nuclear TRIM24 levels were found in 27% spindle and 45% matrix producing subtyped MpBC patients, similar to the cohort distribution of subtypes (Supplementary Data 5). Interestingly, high nuclear TRIM24, and not associated cytoplasmic TRIM24, levels correlate with worse 5-year overall survival (p = 0.04) (Fig. 2e). Thus, nuclear functions of TRIM24 in chromatin association and coregulation of transcription may have greater oncogenic significance than more recently reported cytoplasmic functions of TRIM24 in antiviral immunity23.
TRIM24 up-regulates EMT and glycolysis in mammary tumors
TRIM24 is a histone “reader” protein, which impacts transcription of a number of genes involved in metabolism and nuclear receptor-regulated functions, as we previously showed by Nanostring analyses of genes differentially regulated in response to ectopic expression in cellulo5,7. To determine the regulatory impact of TRIM24-driven mammary tumorigenesis in a physiologically relevant mouse model, we performed deep sequencing of RNA (RNA-seq) isolated from Trim24COE tumors (n = 6, 3 metaplastic carcinosarcoma and 3 carcinomas), control un-diseased mammary glands from age-matched MMTV-Cre animals (n = 4) and mammary tumors spontaneously generated in either FVB or FVB/129 mice in a non-TRIM24 overexpressing background (n = 3). We plotted the overall distribution of differentially expressed genes (DEGs), using a volcano plot to illustrate up- and down-regulated genes, significantly altered by TRIM24 over-expression (Fig. 3a). Expression of 2221 of these genes was up-regulated, whereas 2826 were down-regulated. RNA-sequencing data were analyzed using DESeq2 to determine DEGs with TRIM24 over-expression24. Using a false discovery rate (FDR) of 5%, we found 2006 genes, 529 up-regulated and 1477 down-regulated, had at least 5-fold change (Supplementary Data 6). An unbiased clustering analysis showed a clear distinction between gene expression profiles of MMTV-Cre-expressing, normal control mammary glands, non-TRIM24-overexpressing mammary tumors and TRIM24-overexpressing mammary tumors. Metaplastic carcinosarcomas clustered separately from carcinomas; considerable heterogeneity occurs most notably among carcinomas (Fig. 3b).
To validate the expression of genes within the top 1% (based on FDR) of DEGs in our RNA-Seq analysis, we used qRT-PCR analysis of RNA isolated from Trim24COE tumors (carcinoma, n = 3: tumor number 2735, 3956, and 89; metaplastic carcinosarcoma, n = 3: tumor number 567, 64, and 897) and MMTV-cre controls (Supplementary Fig. 3a). As predicted by RNA-seq, qRT-PCR shows that Trim24, gelactin-1 (Lgals1), macrophage migration inhibitory factor (Mif), cyclin-dependent kinase Inhibitor 2 A (Cdkn2a), vimentin (Vim), and fibronectin 1 (Fn1) were highly expressed in TRIM24-driven tumors compared to Cre-only mammary glands (Supplementary Fig. 3b). In comparison, p53 (Trp53), aldehyde dehydrogenase 1 family member A1 (Aldh1a1), V-set domain-containing T-cell activation inhibitor 1 (Vtcn1), synuclein gamma (Sncg), leukocyte receptor tyrosine kinase (Ltk), E-cadehrin (Cdh1), and synaptotagmin 9 (Syt9) were down-regulated in TRIM24-driven tumors compared to Cre-controls (Supplementary Fig. 3a, b). Consistent with high EMT association with MpBC, we saw that differential expression of EMT markers, for example, up-regulated Vim and Lgals1 and down-regulated Cdh1, was greater in metaplastic carcinosarcomas compared to carcinomas, consistent with protein levels of E-cadherin and vimentin, identified via IHC in TRIM24-driven tumors (Fig. 2b). Together, these results show that TRIM24-driven tumors express a distinct set of genes that can distinguish between metaplastic and carcinoma tumors.
To determine the functional significance of TRIM24-impacted gene regulation in mammary metaplastic carcinosarcoma tumors, we identified hallmark pathways of differentially expressed genes using the ssGSEA algorithm25. We assessed differential pathway activity using a Student’s t-test, correcting for multiple hypotheses and accepting pathways with a 5% FDR cutoff. Although the molecular profiles of transcription showed some heterogeneity between tumor samples of the same pathological classification (Fig. 3b), we identified several significant hallmark pathways of Trim24COE metaplastic carcinosarcomas (Fig. 3c). The top five, significantly up-regulated pathways in TRIM24-driven tumors were the glycolysis pathway, EMT, angiogenesis, allograft rejection, and mTORC1 signaling (Fig. 3c). The only significant pathway down-regulated by TRIM24 expression is a group of genes (KRas signaling DN) likewise down-regulated in response to oncogenic KRAS expression26. The predicted scores of statistically significant pathways (>5% FDR) were clustered, revealing that metaplastic carcinosarcoma tumors (M, n = 3) had similar scores that were significantly different from normal tissue27. In contrast, the scores for carcinomas (C, n = 3) were heterogeneous and in between the scores of metaplastic carcinosarcoma tumors (Fig. 3d)28. Additionally, EMT scores, calculated by an ssGSEA algorithm, showed significant differences between TRIM24-driven metaplastic carcinosarcoma tumors and TRIM24-driven carcinoma tumors. Interestingly, the EMT scores of TRIM24-driven and non-TRIM24-driven carcinomas were not significantly different (Supplementary Fig. 3c).
To investigate the association between TRIM24 activity and metaplastic tumors, we analyzed previously published tumor expression data sets. Having shown previously5,6 that TRIM24 is regulated by post-transcriptional mechanisms, we used a Trim24COE expression signature of downstream marker genes to assess TRIM24 activity levels, rather than evaluating TRIM24 gene expression alone. Weigelt et al.29 determined gene expression and copy number variation of 26 human MpBC patients and reported specific features that correlate with histological subtypes. We compared the gene expression analysis of Weiget et al. by means of the calculated TRIM24 metaplastic score and found that this cohort of human MpBC tumors has a significantly higher TRIM24 metaplastic score than TRIM24-driven carcinomas, as did TRIM24-driven metaplastic carcinosarcomas (Supplementary Fig. 3d), predicting that TRIM24 is activated in human metaplastic tumors. Previously a mouse model with MMTV-driven expression of Hepatocyte Growth Factor Receptor Met tyrosine kinase was reported to develop diverse EMT-high and claudin-low, metaplastic carcinosarcoma tumors27,30. Comparison of our calculated TRIM24 metaplastic score to microarray expression analysis of MMTV-Met-driven and MMTV-Met;Trp53fl/+ tumors revealed that the TRIM24 signature is significantly higher in the two MET-driven metaplastic models, but not the non-metaplastic p53/MET model (Supplementary Fig. 3e). Next, the association between TRIM24-driven metaplastic carcinosarcoma tumors and a previously reported human MpBC gene expression signature was calculated using GSEA12. This assessment revealed significant alignment between Trim24COE carcinosarcoma DEGs and human metaplastic gene signature data (NES = 1.615 and FDR 0.002) (Supplementary Fig. 3f). Together these expression analyses support our morphological and molecular characterization of mammary-specific Trim24COE tumors as a model of MpBC.
PI3K pathways are up-regulated in TRIM24-driven mammary tumors
We used Reverse Phase Protein Array (RPPA) to quantify specific proteins in TRIM24-driven metaplastic carcinosarcomas compared to MMTV-Cre mammary glands. Two biological replicates of Trim24COE metaplastic carcinosarcomas with two technical replicates were compared to MMTV-Cre mammary glands. There were 247 protein probes used to determine protein levels and all relative protein levels were normalized using protein loading controls31. Normalized protein loading and transformed linear values were used to determine fold change and later log 2 transformed fold change values (Supplementary Data 7). The normalized values were also used to generate a heatmap representation of differentially produced (57 up-regulated and 47 down-regulated) proteins in Trim24COE tumors compared to control mammary glands (Fig. 4a). We converted protein probes to gene identifiers (IDs) to perform GSEA and determine hallmark pathways of differentially expressed proteins25. Based on RPPA, we identified PI3K-mTOR signaling, apoptosis, UV response and Notch signaling as the top up-regulated pathways (Fig. 4b). mTOR complex pathway and G2/M check point genes were both present in up-regulated hallmark pathways determined by RNA-Seq and RPPA (Figs. 3c, d and 4b).
To assess mTOR signaling in Trim24COE metaplastic carcinosarcomas at a single-cell level, we performed cytometry by time-of-flight (CyTOF) using allografted tumors. These allografted tumors were generated by injecting mouse primary cell lines, derived from Trim24COE metaplastic carcinosarcoma tumors, into mammary fat pads of nude mice32. CyTOF analysis revealed that Trim24COE metaplastic carcinosarcomas exhibited high levels of TRIM24, phospho-PI3K (pPI3K), phospho-AKT (pAKT), and mTOR within the same spatial orientation (Fig. 4c). These results were quantified using an overall median of protein levels in Trim24COE metaplastic carcinosarcomas compared to MMTV-Cre mammary glands along with pan cytokeratin to differentiate sarcomatoid carcinoma from sarcoma (Fig. 4d). Thus, CyTOF supports the conclusion that overexpression of TRIM24 up-regulates key members of the PI3K-AKT-mTOR pathway, leading to aberrant activation frequently associated with tumorigenesis, including MpBC in human patients14.
TRIM24 activates c-Met and PI3K pathway in carcinosarcomas
Activation of the PI3K-AKT-mTOR pathway in TRIM24-driven carcinosarcoma tumors nominates potential therapeutic targets14. In particular, both RNA-seq and qRT-PCR revealed that Met, an upstream regulator of the PI3K-AKT-mTOR pathway, is overexpressed (RNA-Seq log2 fold = 3.9) in TRIM24-driven tumors, compared to Cre-only mammary glands (Fig. 5a). To determine a mechanism for TRIM24-mediated activation of Met expression (c-Met), previously reported to drive formation of mammary metaplastic carcinosarcoma tumors when overexpressed27,30, we performed Assay for Transposase Accessible Chromatin with high-throughput sequencing (ATAC-Seq) of murine primary cell lines; an MMTV-Cre-only cell line (823), derived from a spontaneous mammary outgrowth, and a TRIM24COE metaplastic carcinosarcoma-derived primary line (897). The TRIM24-driven metaplastic carcinosarcoma cell line (897) showed higher chromatin accessibility at specific sites along the Met gene region, compared to the MMTV-Cre-control cell line (823) (Fig. 5b). We used chromatin immunoprecipitation (ChIP) PCR analysis of primary tumor cell lines and confirmed that TRIM24 was significantly enriched at regions of increased chromatin accessibility near −2.0 kb and the proximal promoter (−50 bp) of Met in the primary metaplastic carcinosarcoma line (897) but not the MMTV-Cre control (823) (Fig. 5c). Further, using a tet-inducible system, we showed that knock-down of Trim24 leads to a parallel decrease in mRNA expression of cMet (Supplementary Fig. 4a). Enrichment of TRIM24, as a co-regulator of transcription that recognizes a dual signature of acetylated and unmethylated histone H3 (H3K23ac:H3K4me0) through its combinatorial PHD/bromodomain5, offers a mechanism for increased chromatin accessibility, activated transcription and high levels of Met gene expression revealed by RNA-seq (Fig. 5a).
We leveraged our finding that TRIM24 directly activates Met expression and TRIM24COE tumor-derived primary cells to test the potential of therapeutically targeting MET. Crizotinib is a dual inhibitor of MET and anaplastic lymphoma kinase (ALK) that has shown promise in multiple cancer types, including breast cancer33,34,35. Crizotinib is an FDA-approved drug for metastatic non-small cell lung cancer (NSLC) and anaplastic large cell lymphoma (ALCL), which has been used in more than 70 clinical trials in United States (clinicaltrials.gov) for various cancers36. Crizotinib treatment of TRIM24COE primary carcinosarcoma cell lines (567, 64, and 897), compared to a murine TNBC-like cell line 4T1, led to a significant, TRIM24COE metaplastic carcinosarcoma cell-specific reduction in cell viability in a dose-dependent manner (Fig. 5d). To further assess MET as a drug target, we used a selective inhibitor of MET kinase, PHA-665752, previously shown to reduce tumor burden in a mouse model of gastric cancers where MET is amplified37. PHA-665752 treatment and targeting of MET significantly reduced cell viability in a dose-dependent and metaplastic carcinosarcoma-specific manner (Supplementary Fig. 4b).
Aberrant activity of MET potentially disrupts cellular homeostasis by activation of one or more downstream pathways, including PI3K38. PI3K pathway components exhibited high protein levels in RPPA and CyTOF analysis of TRIM24COE metaplastic carcinosarcoma tumors (Fig. 4), likely activated by post-transcriptional regulation (Supplementary Fig. 4c). To focus downstream of MET and determine whether the PI3K pathway plays an essential role in TRIM24COE metaplastic carcinosarcomas, we used Dactolisib, a dual inhibitor of PI3K isoforms and mTOR39,40. We found that TRIM24COE metaplastic carcinosarcoma cell lines showed a marked reduction in cell viability with low doses of Dactolisib, which was not recapitulated in the 4T1 TNBC-like cell line (Supplementary Fig. 4d).
Together, we show that TRIM24-mediated, epigenetic activation of c-Met is disrupted by selective inhibitors, Crizotinib and PHA-665752, in a dose-dependent manner in primary metaplastic carcinosarcoma tumor cells. c-Met induces up-regulation of PI3K and mTOR pathways at a protein level, consistent with our RPPA and CyTOF analysis of TRIM24COE metaplastic carcinosarcomas. Using a dual inhibitor developed for human PI3K isoforms and mTOR, we showed that viability of murine TRIM24COE metaplastic carcinosarcoma cells was likewise inhibited, further supporting the parallels between human cancers and tumors arising in the TRIM24COE mouse model.
TRIM24 metaplastic carcinosarcomas are similar to human disease
To compare Trim24COE metaplastic carcinosarcoma mammary tumors directly to human metaplastic breast cancer (MpBC) samples, we analyzed patient data generated by an ongoing clinical trial: A Robust TNBC Evaluation fraMework to Improve Survival (ARTEMIS: MD Anderson protocol 2014-0185; NCT02276443) at the University of Texas MD Anderson Cancer Center. We used RNA-Seq analysis of MpBC patient tumor biopsies, acquired at the time of enrollment in the clinical trial, to generate a human TRIM24 metaplastic breast cancer signature using the SIGNATURE system, as previously described41, and developed a heatmap representation of the TRIM24 metaplastic signature expressed by murine metaplastic carcinosarcoma (M) versus carcinoma tumors (C) (Fig. 6a and Supplementary Data 8). Hallmark pathway analysis showed that the TRIM24 MpBC signature exhibit significant up-regulation of Glycolysis, EMT, E2F, and mTORC1 signaling pathways (Fig. 6b). This analysis shows a consistent overlap at gene expression level between human MpBC tumors and TRIM24-driven metaplastic carcinosarcoma.
Global expression profiles of human TNBC tumors, both MpBC and non-metaplastic cases of the ARTEMIS cohort, were assessed by our generated TRIM24 metaplastic signature and found significantly higher in MpBC TNBC but not non-MpBC TNBC (Fig. 6c). Further, TRIM24-signature-high-scoring MpBC tumors (red dots) also showed a higher percentage of vimentin IHC staining, indicative of EMT, compared to non-metaplastic TNBC tumors (Fig. 6d and Supplementary Data 9). Patients with TNBC, including MpBC, are subjected to chemotherapy of anthracyclines followed by taxanes, as standard-of-care primary therapy. Residual cancer burden (RCB) is calculated after neoadjuvant therapy to predict disease recurrence and patient survival, hence, a higher RCB index correlates with worse patient survival42. RCB-II and RCB-III tumors had higher TRIM24 scores, but did not achieve statistical significance due to low numbers of chemosensitive tumors (Fig. 6e and Supplementary Data 9), indicating little to no response to primary chemotherapy. In the subtypes of TNBCs, described by Lehman et al.43, we found that MpBC tumors with significant TRIM24 metaplastic scores were classified mainly as immunomodulatory (IM) and mesenchymal (M*) subtypes (Fig. 6f and Supplementary Data 9). Enrichment of Mesenchymal subtypes is highly consistent with the association of TRIM24-driven tumors and up-regulation of EMT43. Overall, these results suggest that TRIM24-driven mouse metaplastic carcinosarcomas exhibit significant similarities to human MpBC tumors by gene expression and hallmark pathway profiling, offering an animal model of human triple-negative MpBC suitable for development of potential therapeutics.
PROTAC targeting of TRIM24 decreases tumor cell viability
TNBC, both non-MpBC and MpBC, patient-derived xenograft (PDX) mouse models were established with patient biopsies obtained through the ARTEMIS trial. Non-MpBC and MpBC subtyped patient biopsies were further classified by Vanderbilt-Based subtyping (Supplementary Fig. 5a). PDXs were created by embedding human patient needle biopsies in mammary fat pads of nude mice, which were “humanized” with immortalized human mammary stromal fibroblasts44. Implanted and established PDX tumors were grown and passaged two times in nude mice prior to collection of PDX tumors from passage 3 for analyses44. Human mitochondrial protein IHC for nonglycosylated components of mitochondria distinguishes human tissue in the mouse mammary environment of PDX models (Supplementary Fig. 5b, d). PDX tumors were further analyzed by TRIM24 IHC to identify positive TRIM24 expression (Fig. 7a). Both MpBC and TNBC (non-MpBC) PDX tumors are positive for TRIM24 expression. MpBC PDXs have high levels of TRIM24 expression in both nuclear and cytoplasmic compartments, while non-MpBC TNBC PDXs have more variable TRIM24 expression (Fig. 7a and Supplementary Fig. 5c).
Previously, our collaborative efforts yielded a high-affinity, small molecule inhibitor of the TRIM24 bromodomain, IACS957120. IACS9571 proved to be highly effective in displacement of TRIM24 from chromatin, as measured biochemically in vitro20 and by loss of TRIM24 SUMOylation in cellulo45, but had little effect on proliferation of established breast and other cancer cell lines grown as 2-D cultures. However, IACS9571 bromodomain inhibitor was further developed as a VHL-ligand-based proteolysis targeting chimera (PROTAC), which proved to be an effective TRIM24 protein degrader (dTRIM24) in cellulo (19 and Supplementary Fig. 5e). In this previously reported study, dTRIM24 added at 5 μM concentration inhibited growth of leukemia-origin, cultured MOLM-13 cells but had little effect on the only breast cancer-origin cells tested, MCF7, a Luminal A breast cancer-derived cell line19. The bioavailability of dTRIM24 for in vivo assessment remains undocumented; thus, we assayed dTRIM24 (kind gift of L. Gechijian, N. Gray and J. Bradner) and its chemical enantiomeric negative control (eTRIM24) using TRIM24-driven primary carcinosarcoma cell lines and control cell line 823. The dTRIM24 drug successfully degraded TRIM24 compared to eTRIM24 and DMSO (Supplementary Fig. 5e). To ascertain potential relevance of TRIM24 targeting in human TNBC, we compared the ability of dTRIM24 and eTRIM24 to impact viability of TNBC PDX tumorspheres, both MpBC and non-MpBC (Fig. 7a and Supplementary Fig. 5a), using a high-throughput drug screening platform (Fig. 7b). We found that treatment with dTRIM24 led to significant loss of cell viability for all four MpBC PDX tumorspheres and two of four non-MpBC TNBC PDX tumorspheres, versus negative control eTRIM24. Overall, these results suggest that targeting TRIM24 or TRIM24-regulated pathways may have therapeutic potential in the future for patients with TNBC, especially the rare, aggressive and chemoresistant MpBC subtype.
Here, we present a transposon-mediated, conditional expression Trim24 mouse model and show that increased mammary-specific expression of TRIM24 (Trim24COE) drives mammary tumorigenesis. Unexpectedly, a majority (67%) of Trim24COE tumors are carcinosarcomas with coexisting malignant epithelial and spindle (mesenchymal) cellular components, a defining characteristic of human breast carcinosarcoma or MpBCs11. Metaplastic breast carcinomas (MpBCs) are a rare, aggressive, chemorefractory and understudied subtype of TNBC. With few effective treatment options for these patients, it is essential that the molecular mechanisms governing the development of these rare tumors be understood. In a cohort of 46 MpBC patients, we showed that nuclear TRIM24 is associated with poor overall survival; whereas, cytoplasmic TRIM24 is not. Cytoplasmic TRIM24 reportedly ubiquitinates TRAF3 and facilitates antiviral immunity4,23, a function that may not be relevant to MpBC tumor development and patient survival.
In the current study, we focused on nuclear roles of TRIM24, as a histone reader and transcriptional regulator in MpBC. Differential gene expression analysis identified glycolysis and EMT as the top up-regulated pathways in TRIM24-driven mammary tumors, robustly active even in end-stage tumors, analyzed here. Our previous in cellulo studies showed that immortalized human mammary epithelial cells were transformed and formed high-grade, xenograft tumors when TRIM24 was expressed by lentiviral transduction7. Nanostring analyses of genes differentially expressed in TRIM24-transformed cells showed similar patterns of up-regulated glycolysis and mTOR pathways and cell cycle check point alteration. These data, generated both in cellulo and in vivo, suggest that high levels of TRIM24, in addition to reprogramming epithelial cells to a mesenchymal phenotype, promote an environment in which transformed cells alter metabolic response to survive and proliferate during tumorigenesis.
We developed a TRIM24-driven metaplastic carcinosarcoma gene signature and compared it to human MpBC tumor global gene expression to derive a TRIM24 metaplastic score for each MpBC patient in the cohort reported in this study. Gene expression profiles of human MpBC tumors exhibit significant EMT expression signatures, consistent with the presence of epithelial/mesenchymal compartments within these highly aggressive tumors12, and loss of E-cadherin and gain of robust vimentin expression in Trim24COE carcinosarcomas, reported here. Glycolysis and EMT are linked in many cancer types46, and it has been shown that up-regulation of glycolysis modulates the actin cytoskeleton, which may promote EMT47. Molecular subtyping of TNBCs shows an enrichment of EMT-associated gene expression in mesenchymal and mesenchymal stem-like subtypes, 10–30% of which are classified as MpBC by light microscopy12. TNBC tumors with high TRIM24 metaplastic signature scores are mainly IM and M subtypes and exhibit high vimentin expression. These MpBC tumors also showed poor outcomes in response to systemic chemotherapy (anthracyclines followed by taxanes) in a neoadjuvant setting (NACT), which is reflected by their extensive residual disease (RCB-II and RCB-III) status (Fig. 6e). A recent in-depth proteomic landscape of MpBC patients further defined subtypes of MpBC and compared MpBC to the broader TNBC classification18. This cutting-edge report revealed that MpBC patient samples, as well as spindle MpBC-like tumors that develop in a mammary epithelial-specific Ccn6 knockout mouse48, have enrichment in EMT and E2F pathways at protein levels consistent with our transcriptomic analysis of both MpBC TRIM24 signature and TRIM24-driven metaplastic carcinosarcoma tumors. This subtype classification additionally underscores heterogeneity of MpBC tumor development and further need for multiple MpBC mouse models, as down-regulation of CCN6 and cellular metabolism and protein translation up-regulation are not significant in our epigenetic-driven model. Our analyses here are limited to TNBCs. While previous studies and public data sets showed that TRIM24 RNA is expressed across a broad range of cancers, it is not yet established whether TRIM24 plays a similar role in promoting malignancy through alterations of EMT, E2F, and metabolic pathways. The use of our TRIM24 signature may partially address this issue, although tissue- and batch-specific differences across gene expression data sets may confound the analyses.
Our molecular analyses of Trim24COE metaplastic carcinosarcoma tumors nominated c-MET receptor tyrosine kinase, an upstream activator of PI3K/AKT/mTOR pathways, as a potential therapeutic target. A mechanism for increased Met expression in Trim24COE metaplastic carcinosarcomas lies in specific enrichment of PHD/bromodomain protein TRIM24, a co-regulator of transcription, at accessible chromatin of the activated Met promoter, consistent with TRIM24-induced RNA expression and downstream activation of the PI3K and mTOR pathways. Interestingly, a previously reported mouse model with MMTV-driven Metmt developed a high percentage of claudin-low, mixed pathology mammary carcinomas that undergo EMT, and a low percentage of adenocarcinoma and adenomyoepithelial carcinoma30. Comparison of RNA expression changes in MMTV-driven Metmt tumors, as well as to publicly available human MpBC (GSE57544), revealed significant concurrence with our derived TRIM24 Metaplastic signature score (Supplementary Fig. 3d, e). Although we did not see activation of downstream PI3K genes by RNA expression analysis of TRIM24-driven tumors, CyTOF and RPPA analysis showed post-transcriptional activation of PI3K pathway components, consistent with c-MET kinase functions. Previously, Coussy et al.17 showed that a combination of PI3KCA and MEK inhibitors significantly reduces tumor growth in a limited number of MpBC PDX models. To inhibit the PI3K pathway, we used multiple approaches and found that cell viability of primary Trim24COE metaplastic carcinosarcoma cells was reduced in a dose-dependent manner by inhibiting either c-MET, with Critzotinib and PHA-665752, or PI3K/mTOR, with dual inhibitor Dactolisib. Crizotinib is an FDA-approved drug for metastatic NSLC and ALCL; whereas, PHA-665752 and Dactolisib are bioavailable and have been tested in clinical trials for various cancers although not reported as potential therapeutic treatments of MpBC.
Knowing that single-agent therapeutics often lead to resistance and relapse in human cancers, we pursued assessment of TRIM24 inhibition and degradation, using a previously reported PROTAC version of TRIM24 bromodomain inhibitor IACS-957119. PROTAC-TRIM24 [IACS-9571] (dTRIM24) was tested for inhibition of human MpBC PDX tumorsphere viability, as a surrogate for MpBC patient tumor samples. dTRIM24 was previously shown effective in inhibition of a cultured leukemia cell line (MOLM13) viability but, to our knowledge, has not been tested for inhibition of cell viability in solid tumor PDXs19. Recent advances in drug discovery have shown the potential of PROTACs for therapeutic success, such as ER PROTAC ARV-471 in clinical trials for ER-positive breast cancer49. We propose that combinatorial therapeutics featuring either a c-MET or dual PI3K/mTOR inhibitor in combination with epigenetic targeting, via a bioavailable TRIM24 degrader, may show promise for treatment of MpBC and other TRIM24-expressing TNBC patients with otherwise poor prognosis and survival.
Our studies establish TRIM24 as a potent oncoprotein in mammary epithelia, acting as a histone reader in chromatin association and disruption of metabolic homeostasis, as well as activation of EMT. Similar to human MpBC patients, our TRIM24-driven mouse model also shows activation of MET/PI3K pathways, which can be therapeutically targeted. Importantly, we observed a TRIM24 signature in a rare and aggressive form of human breast cancer, MpBC. Preclinical studies are needed to validate that inhibiting TRIM24 and its downstream targets, such as c-MET and PI3K pathway activities, has therapeutic potential for this specific subset of TNBC patients.
A piggyBac transposon vector was constructed bearing a human UBC promoter-driven loxP-lacZ-stop-loxP cassette 5′ of a FLAG-tagged mTrim24 cDNA (pBac-UbC-lox-stop-Trim24) (Supplementary Fig. S1a and Supplementary Table 1)50. Uncut pBac-UBC-lox-stop-Trim24 (2 ng/μl) was injected into the pronuclei of FVB/NCr (Charles River Laboratories) zygotes along with transposase RNA (15 ng/μl). The FVB/NCr strain is reportedly more permissive for mammary tumor development51. Transgenic Trim24 (Trim24LSL) founder mice were identified by PCR of tail genomic DNA (Fig S1b), using primers listed in Supplementary Data 1. Trim24LSL mice were bred with the Tg(MMTV-cre)4Mam (The Jackson laboratory, stock 003553) recombinase mouse line D (MMTV-CreTg/0)52 for Cre-mediated excision of the floxed stop cassette to drive expression of transgenic mTrim24 in mammary epithelial cells. The resulting mice were called Trim24COE and maintained in pathogen-fee conditions. We also maintained an aging virgin female colony by breeding MMTV-Cre mice with FVB to generate control mice of similar background as Trim24COE and Trim24LSL mice. Virgin female mice were aged until there was evidence of a palpable tumor or maximum of 2 years. At killing, mammary glands, liver, lung, and spleens were dissected and preserved for further analyses. Tumor penetrance was calculated as percentage of mice in tumor bearing genotype Trim24COE divided by total number of mice in Trim24COE cohort. Animal handling and experiments were performed as approved by the Institutional Animal Care and Use Committee at the MD Anderson Cancer Center.
To determine the genomic location of transgenes, we performed whole-genome sequencing using tail tissue-derived DNA of three different Trim24LSL progeny of the same founder. The library preparation and sequencing were done at the MD Anderson Next Generation Sequencing Core Facility. Briefly, the libraries were prepared using a KAPA Hyperprep kit, Roche (KK850) and samples were run for 76 bp paired-end to achieve 5–10x coverage per sample. The sequence reads were mapped by Burrows-Wheeler Aligner (BWA) (version 0.7.15)53 to the mouse genome (version GRCm38). In order to detect possible locations of Trim24 transgene, we included the Trim24LSL vector sequence into the mouse genome as an extra “chromosome”. The mapped bam files were generated by GATK pipeline 3.7 developed by the Broad Institute54. The discordantly mapped reads with Trim24 gene sequences were then retrieved by samtools55. The discordant reads of these samples are included in Supplementary Data 3.
Carmine alum staining
Inguinal mammary glands were dissected from littermate female mice and compared only between mice of the same age and harvested at the same time of day. Mammary glands were flattened onto a glass slide and stained as previously described52 with the following alterations: mammary glands were fixed in ice cold 4% paraformaldehyde in phosphate-buffered saline (PBS) overnight, washed in PBS 3 × 10 min then submerged in carmine alum staining solution (4 mM carmine, 10 mM aluminum potassium sulfate) until staining thoroughly permeated the tissue. Tissues were then dehydrated as described52, cleared in Histo-clear (National Diagnostics) overnight and stored in toluene. Cleared mammary glands were imaged using a stereomicroscope. To count the number of branches, one major branch was selected, and individual sub-branches were counted manually. The graph was generated using Prism (v9) and student t-test was performed to calculate significance between control and Trim24COE mammary branches.
All hematoxylin and eosin (H&E) staining was performed using standard procedures by the MD Anderson Department of Veterinary Medicine Histology Core. IHC for Flag (Sigma,sc-7945, 1:5000), ER (Santa Cruz, sc-8002,1:500), PR (Abcam, ab63605,1:4000), E-cadherin (Proteintech, 20874-1-AP, 1:100) Vimentin (Abcam, ab92547, 1:500), human mitochondria (Chemicon # MAB1273, 1:2000), and TRIM24 (Proteintech,14208-1-AP, 1:500) was performed by the MD Anderson Cancer Center Research Histology Pathology and Imaging Core using protocols as previously described7. Forty-six archived MpBC paraffin tissue blocks from the Department of Pathology at MD Anderson Cancer Center, with up to 22 years of clinical follow-up data, were prepared for TRIM24 IHC using the standardized departmental protocols. The tissues were scored as previously described22. Next, patients were stratified based on scoring and percentage of total cells expressing TIRM24 either in nucleus or cytoplasm or both. The group of patients were divided into four groups and overall survival was calculated. Statistical analyses were carried out using GraphPad Prism (version 8) software. The cumulative overall survival was calculated using the Kaplan–Meier method, and log-rank test was used to analyze differences in the survival times.
Protein and RNA analyses
To prepare whole cell lysates, cells were lysed as indicated in either radioimmunoprecipitation assay (RIPA) buffer (20 mM Tris pH 8.0, 150 mM NaCl, 1 mM EDTA, 1% NP-40, 1, Triton X-100, 0.5% deoxycholic acid or NETN buffer (150 mM NaCl, 1 mM EDTA, 50 mM Tris pH 7.5, 0.1% NP-40) or 1% SDS supplemented with protease inhibitor cocktail (EMD Millipore). Immunoblotting was performed using standard techniques for antibodies listed in Supplementary Data 1. Total RNA was isolated with TRIzol reagent (Invitrogen) and qRT-PCR was performed as described previously7 using primers listed in Supplementary Data 2. Data were calculated using 2−ΔΔCT method in Microsoft Excel (365) and graphs were prepared using Prism (v8).
Primary cell lines
Tumor fragments designated for cell isolation were harvested as previously described56. Briefly, tissue sections were minced and placed in 10 ml digestion media (Dulbecco’s Modified Eagle Medium (DMEM)-F12, 250 u/ml hyaluronidase, 3 mg/ml collagenase, 100 U/mL penicillin/streptomycin (PS), 100 μ/mL antibiotic-antimycotic) per gram of tissue. Tissues were digested while shaking at 1500 × g for 1–3 h at 37 °C, filtered through a 40 μm sterile filter and centrifuged at 1200 × g for 10 min at 4 °C. Cells were resuspended in DMEM/F12 and washed 4X sequentially by centrifugation at 1200 × g for 5 min. The resulting suspensions were plated onto 60 mm tissue culture dishes in complete media (DMEM 10% fetal bovine serum (FBS), and PS) at 37 °C. Cells were grown to confluence, during which time observable colonies were formed. These cell colonies were then subjected to differential trypsinization, removed and gently tapped to release colonies enriched for epithelial cells, leaving behind mesenchymal and fibroblast-like cells. The resulting population was termed epithelial-enriched and then plated onto a new dish and cultured in complete media. Three primary cell lines, 567, 64, and 897, were generated from TRIM24-driven metaplastic carcinosarcoma and cultured using complete media. Primary cell line 823 was generated from a spontaneous mammary tumor generated in an MMTV-Cre mouse and used as a control. Cells were passaged and frozen after Mycoplasma testing (Lonza MycoAlert Mycoplasma Detection kit, LT07218) as per manufacturer’s instruction. The passages of primary cells used in all experiments were between 8 and 15. For tet-inducible shRNA cell lines, we used a lentivirus system and Lipofectamine 3000 (ThermoFisher Scientific, L3000008) to transfect 293FT cells using manufacturer’s recommendations on virus production. Both shControl (Horizon, Catalog No. RHS4743) and shTrim24 (Horizon, Catalog No. RMM4431-200368247,) were purchased through the functional genomic core at University of Texas M.D. Anderson Cancer Center for cloning in a TRIPZ inducible lentivirus vector. Next, lentivirus was collected from 293FT cells to transduce the TRIM24-driven primary metaplastic carcinoma cell line 897. The shControl and shTRIM24 897 cells were grown under puromycin selection and sorted based on RFP after tetracycline induction (1 μg/ml). To collect RNA, 897 sh-lines were induced with tetracycline at 1 μg/ml for 24 h. Cells were tested for pathogens using IMPACT testing (IDEXX BioResearch) prior to inject into mouse for generating allograft tumors. Mouse mammary tumor-derived 4T1 cell line was purchased from ATCC (CRL-2539) and grown in RPMI-1640 media.
RNA library preparation and deep-sequencing
To perform deep sequencing of RNA (RNA-seq), total RNA was isolated from individual tumors, in parallel with freshly isolated, disease free mammary glands (controls), using Zymo Direct-zol RNA miniprep kit (R2050). RNA libraries were prepared using TrueSeq RNA Library Prep Kit v1 (Illumina) followed by sequencing at the MD Anderson Next Generation Sequencing Core Facility.
Pre-preprocessing and analysis of RNA-Seq data used pipelines developed by BETSY57. To summarize, we checked the quality of the FASTQ files using FastQC (v 0.11.8). Then, we trimmed adapters with Trimmomatic (0.39) using default settings58. Trimmed reads were checked again with FastQC, and then aligned to the GRCm38 mouse genome assembly with the GENCODE Release M14 transcriptome using the STAR aligner for MMTV-CreTg/0 mammary gland59. The reads were counted with HTSeq-Count60 using the union mode to resolve multiple features. Gene expression values were estimated as TPM using RSEM (version 1.3.1), using the STAR (2.7.2b) aligner option61. We evaluated the quality of the alignments using the CollectAlignmentSummaryMetrics and CollectRnaSeqMetrics tools from Picard (2.9)62. The data are submitted on Gene Expression Omnibus (GEO) with accession GSE179036 for murine samples.
To determine gene expression profiles of TRIM24-driven tumors, a differential expression analysis was performed to compare profiles of tumors with non-tumor mammary tissue using DESeq2 (3.13)24. We used script developed in house to calculate the average expression of tumors and control groups, the differences and fold change expression between groups. Genes were selected using a false discovery rate cutoff of 5% and at least a 5-fold change between groups. This resulted in a list of 2006 genes (Supplementary Data 6). Hallmark pathways63 were scored on gene expression profiles (read counts) using the ssGSEA algorithm (2.0). Differential pathways were identified using a Student’s t-test correcting for multiple hypotheses and accepting pathways with a 5% FDR cutoff.
The association between Trim24COE tumors and a previously established metaplastic gene expression signature were calculated using GSEA (4.1.0)25. The LIEN_BREAST_CARCINOMA_METAPLASTIC_VS_DUCTAL_UP gene set was downloaded from the MSigDB database.
Reverse Phase Protein Arrays analysis
Flash-frozen tumor samples and mammary glands were submitted to the Functional Proteomics RPPA Core Facility at MD Anderson Cancer Center. The samples were processed as described previously31. Samples were probed with 244 antibodies listed in Supplementary Data 7. The staining intensity was normalized to an internal loading control and customized software to produce Normalized Linear values and transformed Log2 values. The normalized raw values were used to generate a heatmap using R function. The protein names were transformed to gene names to calculate GSEA scores, subjected to Hallmark pathway analysis and plotted using Prism (v8)25,63.
Cytometry by time of flight
To validate RPPA analysis, we generated allograft tumors by injecting primary cell lines derived from TRIM24 overexpressing tumors into mammary fat pads of Foxn1nu mice, also known as athymic mice32. Tumors were dissected once they reached 2 cm3 in size. We also collected normal mammary glands of MMTV-CreTg/0 mice as controls. The samples were prepared and processed for CyTOF as previously described64. FlowJo_v10.3 was used to remove beads, debris and to obtain single cells. CytoBank (version 6.7) was used to generate viSNE plots.
c-Met and PI3K/mTOR inhibitor treatment
MET inhibitors Crizotinib and PHA-665752 were purchased from Cayman Chemical (Item numbers 12087 and 14703, respectively). A dual inhibitor of PI3K and mTOR, Dactolisib or NVP-BEZ235, was also purchased from Cayman Chemical (Item number 21185). Primary murine cells obtained from TRIM24-driven carcinosarcomas (567, 64, and 897) and a TNBC cell line (4T1) were plated at a confluency of 2000 cells per well in 96-well plates. Each drug of choice or vehicle control (Dimethyl sulfoxide or DMSO) were added the next day and incubated for 72 h. At the end of 72 h, post-treatment cell viability assays were performed.
Cell viability was measured using CellTiter-Glo® (CTG) Luminescent Cell Viability Assay (Promega G7570), as per manufacturer’s recommendations. On the last day of treatment, CTG reagent was added at a 1:1 ratio, as per manufacturer’s recommendation. Viability was measured using luminescence on a FLUOstar Omega platform as per manufacturer’s guidance.
TRIM24 metaplastic signature
We scored the TRIM24 metaplastic signature on a panel of 68 triple-negative human breast cancer tumors from the ARTEMIS trial (NCT02276443) that have been profiled by RNA-Seq. They were processed using a similar pipeline as above, except that they were aligned to the human reference genome hg19 and a transcriptome from RefSeq65. To compare the mouse-derived metaplastic signature with the gene expression profiles of human tumors, we converted the signature to human genes based on the homologs in the Homologene database. We then scored the signature on human tumors using the SIGNATURE system, as described previously41. We then compared the scores between the metaplastic tumors (defined by a pathologist) and the non-metaplastic ones using a two-tailed T-test via GraphPad Prism (Supplementary Data 8). We also compared the scores against the vimentin percentage as determined by IHC on tissue slides and calculated the p-value using a Pearson test as implemented in the R programming language. The data are submitted to GEO with accession number GSE165407.
TRIM24 degrader treatment
Human MpBC PDX tumors were harvested and cells were extracted in cell suspension media as previously described44. These cells were distributed on 384-well plates and subjected to a high-throughput robotic imaging and drug delivery platform (Texas A&M Combinatorial Drug Discovery Program, Houston, TX). The drugs used for treatment were eTRIM24, a negative control, and dTRIM24, a degrader of TRIM24, and diluted to 10 mM in DMSO19. DMSO was used as a vehicle control. Cell viability was measured using CellTiter-Glo. The measured cell viability values were normalized to DMSO controls and plotted as percent inhibition using GraphPad Prism. To determine the protein expression level under the treatment of DMSO, eTRIM24, and dTRIM24, cell lines 823, 567, 64, and 897 were treated with 4μM of eTRIM24 and dTRIM24 along with equivalent amount of DMSO in 6 well plate in triplicates. Treated cells were collected after 48 h for western blot analysis as described earlier.
ATAC-seq library generation
ATAC-seq in replicates was performed following the Omni-ATAC protocol as previously described66. Briefly, cells grown in tissue cultures were pretreated with 200 U/ml DNase (Worthington) for 30 min at 37 °C to remove free-floating DNA and to digest DNA from dead cells. The cell cultures were washed with PBS followed by harvesting and counting. 50,000 cells were lysed and nuclei isolated after centrifugation at 500 × g for 10 min in lysis buffer containing 0.1% NP-40 and 0.01% Digitonin. Pelleted nuclei were tagmented with Nextera Tn5 Transposase (TDE1, Illumina 15027865) in TD Tagment DNA buffer (Illumina 15027866) for 30 min at 37 °C, and the resulting library fragments were purified using a Qiagen MinElute kit. Libraries were amplified by 4–6 PCR cycles as described66 and purified using AMPure XP beads (Beckman Coulter A63881). ATAC-seq libraries were sequenced 2 × 75 bp on an Illumina HiSeq4000 to obtain at least 50 million high quality mapping reads per sample.
The libraries for ATAC-Seq were sequenced using 2 × 75 bases paired-end protocol on Illumina HiSeq 3000 instrument. 58–67 million pairs of reads were generated per sample. Each pair of reads represents a DNA fragment from the library. Adapter sequences were removed from 3′ends of reads by Trimmomatic (0.39) using default settings58. Then the reads were mapped to the mouse genome (mm10) using Bowtie (version 1.1.2)67 with parameters “-allow-contain -maxins 2000 -v 2 -m 1 -best -strata”. About 89–91% pairs of reads were mapped to the mouse genome, with 84–86% uniquely mapped. To avoid PCR bias for multiple pairs of reads that were mapped to the same genomic position, only one copy was retained for further analysis. After the removal of reads by chrM, 40–46 million pairs of reads were used in the final analysis. For each fragment (e.g. a pair of reads), the 5′ end was offset by +4 bp and the 3′ end was offset by −5 bp to adjust both ends to represent the center of a transposon binding event. To generate the signal landscape, binding events (i.e. ends of fragments) were smoothed by 73 bp (i.e. extended 36 bp upstream and 36 bp downstream), piled up and normalized to 10 M total binding events. The resulting values were averaged over every 10 bp window and displayed in Integrative Genomics Viewer (IGV) (2.8)68. The data are submitted on GEO (GSE149685).
ChIP assays were performed, as previously described5, using murine MMTV-Cre expressing tumor primary cell line 823 and TRIM24 overexpressing metaplastic carcinosarcoma primary cell line 897. Additionally, after sonication by a BioRuptor 300, chromatin was further treated with MNase (Worthington, LS004798), 30 min at room temperature. TRIM24 (Proteintech, 14208-1-AP; Bethyl, A300-815A) antibodies were used for immunoprecipitation, and the precipitate was collected by Dynabeads Protein A (Invitrogen, 10002D).
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The RNA-Seq data generated for TRIM24-driven mouse model (Fig. 3) are deposited in the GEO database under accession code GSE179036. The RNA-Seq data associated with human metaplastic breast cancer patients (Fig. 6) are deposited in the GEO database under accession code GSE165407. The ATAC-Seq data of TRIM24-driven metaplastic carcinosarcoma cell line and control cell line are available at GEO database under accession code GSE149685. WGS data associated with identification of Flag-tagged mTrim24 transposon in transgenic mice are available at figshare repository (Sample 1 R1 file at https://doi.org/10.6084/m9.figshare.14818542.v1(2021), Sample 1 R2 file at https://doi.org/10.6084/m9.figshare.14818560.v1(2021), Sample 2 R1 file at https://doi.org/10.6084/m9.figshare.14818536.v1(2021) and Sample 2 R2 files at https://doi.org/10.6084/m9.figshare.14818575.v1(2021)). MsigDB LIEN_BREAST_CARCINOMA_METAPLASTIC_VS_DUCTAL_UP was used to compare gene expression of TRIM24-driven metaplastic tumors. Source data are available as a Source data file. The remaining data and information associated with this manuscript are available within the article as a supplementary information. Source data are provided with this paper.
Jones, P. A., Issa, J. P. & Baylin, S. Targeting the cancer epigenome for therapy. Nat. Rev. Genet. 17, 630–641 (2016).
Filippakopoulos, P. & Knapp, S. Targeting bromodomains: epigenetic readers of lysine acetylation. Nat. Rev. Drug Discov. 13, 337–356 (2014).
Mertz, J. A. et al. Targeting MYC dependence in cancer by inhibiting BET bromodomains. Proc. Natl Acad. Sci. USA 108, 16669–16674 (2011).
Allton, K. et al. Trim24 targets endogenous p53 for degradation. Proc. Natl Acad. Sci. USA 106, 11612–11616 (2009).
Tsai, W. W. et al. TRIM24 links a non-canonical histone signature to breast cancer. Nature 468, 927–932 (2010).
Jain, A. K. et al. TRIM24 Is a p53-induced E3-ubiquitin ligase that undergoes ATM-mediated phosphorylation and autodegradation during DNA damage. Mol. Cell. Biol. 34, 2695–2709 (2014).
Pathiraja, T. N. et al. TRIM24 links glucose metabolism with transformation of human mammary epithelial cells. Oncogene 34, 2836–2845 (2015).
Li, H. et al. Overexpression of TRIM24 correlates with tumor progression in non-small cell lung cancer. PLoS ONE 7, e37657 (2012).
Zhang, L. H. et al. TRIM24 promotes glioma progression and enhances chemoresistance through activation of the PI3K/Akt signaling pathway. Oncogene 34, 600–610 (2015).
Groner, A. C. et al. TRIM24 is an oncogenic transcriptional activator in prostate cancer. Cancer Cell 29, 846–858 (2016).
Park, H. S. et al. Clinicopathologic features and outcomes of metaplastic breast carcinoma: comparison with invasive ductal carcinoma of the breast. Yonsei Med. J. 51, 864–869 (2010).
Lien, H. C. et al. Molecular signatures of metaplastic carcinoma of the breast by large-scale transcriptional profiling: identification of genes potentially related to epithelial-mesenchymal transition. Oncogene 26, 7859–7871 (2007).
Shah, D. R., Tseng, W. H. & Martinez, S. R. Treatment options for metaplastic breast cancer. ISRN Oncol. 2012, 706162–706162 (2012).
Basho, R. K. et al. Comparative effectiveness of an mtor-based systemic therapy regimen in advanced, metaplastic and nonmetaplastic triple-negative breast cancer. Oncologist 23, 1300–1309 (2018).
Tray, N., Taff, N. & Adams, S. Therapeutic landscape of metaplastic breast cancer. Cancer Treat. Rev. 79, 101888 (2019).
Abouharb, S. & Moulder, S. Metaplastic breast cancer: clinical overview and molecular aberrations for potential targeted therapy. Curr. Oncol. Rep. 17, 431 (2015).
Coussy, F. et al. Combination of PI3K and MEK inhibitors yields durable remission in PDX models of PIK3CA-mutated metaplastic breast cancers. J. Hematol. Oncol. 13, 13 (2020).
Djomehri, S. I. et al. Quantitative proteomic landscape of metaplastic breast carcinoma pathological subtypes and their relationship to triple-negative tumors. Nat. Commun. 11, 1723 (2020).
Gechijian, L. N. et al. Functional TRIM24 degrader via conjugation of ineffectual bromodomain and VHL ligands. Nat. Chem. Biol. 14, 405–412 (2018).
Palmer, W. S. et al. Structure-guided design of IACS-9571, a Selective High-Affinity Dual TRIM24-BRPF1 Bromodomain Inhibitor. J. Med. Chem. 59, 1440–1454 (2015).
He, X. et al. Prognosis in different subtypes of metaplastic breast cancer: a population-based analysis. Breast Cancer Res. Treat. 173, 329–341 (2019).
Ma, L. et al. Histone H3 lysine 23 acetylation is associated with oncogene TRIM24 expression and a poor prognosis in breast cancer. Tumour Biol. 37, 14803–14812 (2016).
Zhu, Q. et al. TRIM24 facilitates antiviral immunity through mediating K63-linked TRAF3 ubiquitination. J. Exp. Med. 217, e20192083 (2020).
Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014).
Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).
Serra, R. W. et al. A KRAS-directed transcriptional silencing pathway that mediates the CpG island methylator phenotype. eLife 3, e02313 (2014).
Ponzo, M. G. et al. Met induces mammary tumors with diverse histologies and is associated with poor outcome and human basal breast cancer. Proc. Natl Acad. Sci. USA 106, 12903–12908 (2009).
Eisen, M. B. et al. Cluster analysis and display of genome-wide expression patterns. Proc. Natl Acad. Sci. USA 95, 14863–14868 (1998).
Weigelt, B. et al. Metastatic breast carcinomas display genomic and transcriptomic heterogeneity. Mod. Pathol. 28, 340–351 (2015).
Knight, J. F. et al. Met synergizes with p53 loss to induce mammary tumors that possess features of claudin-low breast cancer. Proc. Natl Acad. Sci. USA 110, E1301–E1310 (2013).
Grote, T. et al. Validation of reverse phase protein array for practical screening of potential biomarkers in serum and plasma: accurate detection of CA19-9 levels in pancreatic cancer. Proteomics 8, 3051–3060 (2008).
Kocatürk, B. & Versteeg, H. H. Orthotopic injection of breast cancer cells into the mammary fat pad of mice to study tumor growth. J. Vis. Exp. e51967 (2015).
Hallberg, B. & Palmer, R. H. The role of the ALK receptor in cancer biology. Ann. Oncol. 27, iii4–iii15 (2016).
Heigener, D. F. & Reck, M. Crizotinib. Recent Results Cancer Res 211, 57–65 (2018).
Lee, J., Tran, P. & Klempner, S. J. Targeting the MET pathway in gastric and oesophageal cancers: refining the optimal approach. Clin. Oncol. 28, e35–e44 (2016).
Gerber, D. E. Crizotinib in Treating Patients with Stage IB-IIIA Non-small Cell Lung Cancer That Has Been Removed by Surgery and ALK Fusion Mutations (An ALCHEMIST Treatment Trial) (ECOG-ACRIN Cancer Research Group: National Cancer Institute, 2014).
Christensen, J. G. et al. A selective small molecule inhibitor of c-Met kinase inhibits c-Met-dependent phenotypes in vitro and exhibits cytoreductive antitumor activity in vivo. Cancer Res. 63, 7345–7355 (2003).
Organ, S. L. & Tsao, M.-S. An overview of the c-MET signaling pathway. Ther. Adv. Med. Oncol. 3, S7–S19 (2011).
Shi, F. et al. Correction: the dual PI3K/mTOR inhibitor dactolisib elicits anti-tumor activity in vitro and in vivo. Oncotarget 9, 17255 (2018).
Janku, F., Yap, T. A. & Meric-Bernstam, F. Targeting the PI3K pathway in cancer: are we making headway? Nat. Rev. Clin. Oncol. 15, 273–291 (2018).
Chang, J. T. et al. SIGNATURE: a workbench for gene expression signature analysis. BMC Bioinform. 12, 443 (2011).
Hamy, A.-S. et al. Prognostic value of the Residual Cancer Burden index according to breast cancer subtype: validation on a cohort of BC patients treated by neoadjuvant chemotherapy. PLoS ONE 15, e0234191–e0234191 (2020).
Lehmann, B. D., Pietenpol, J. A. & Tan, A. R. Triple-negative Breast Cancer: Molecular Subtypes And New Targets For Therapy. p. e31-e39 (American Society of Clinical Oncology Educational Book, 2015).
Echeverria, G. V. et al. High-resolution clonal mapping of multi-organ metastasis in triple negative breast cancer. Nat. Commun. 9, 5079–5079 (2018).
Appikonda, S. et al. Cross-talk between chromatin acetylation and SUMOylation of tripartite motif-containing protein 24 (TRIM24) impacts cell adhesion. J. Biol. Chem. 293, 7476–7485 (2018).
Morandi, A. et al. Targeting the metabolic reprogramming that controls epithelial-to-mesenchymal transition in aggressive tumors. Front. Oncol. 7, 40–40 (2017).
Park, J. S. et al. Mechanical regulation of glycolysis via cytoskeleton architecture. Nature 578, 621–626 (2020).
Martin, E. E. et al. MMTV-cre;Ccn6 knockout mice develop tumors recapitulating human metaplastic breast carcinomas. Oncogene 36, 2275–2285 (2017).
Burslem, G. M. & Crews, C. M. Proteolysis-targeting chimeras as therapeutics and tools for biological discovery. Cell 181, 102–114 (2020).
Chen, Y.-T. et al. PiggyBac transposon-mediated, reversible gene transfer in human embryonic stem cells. Stem Cells Dev. 19, 763–771 (2010).
Davie, S. A. et al. Effects of FVB/NJ and C57Bl/6J strain backgrounds on mammary tumor phenotype in inducible nitric oxide synthase deficient mice. Transgenic Res. 16, 193–201 (2007).
Wagner, K. U. et al. Spatial and temporal expression of the Cre gene under the control of the MMTV-LTR in different lines of transgenic mice. Transgenic Res. 10, 545–553 (2001).
Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 (2010).
Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).
Roarty, K. et al. Ror2 regulates branching, differentiation, and actin-cytoskeletal dynamics within the mammary epithelium. J. Cell Biol. 208, 351–366 (2015).
Chen, X. & Chang, J. T. Planning bioinformatics workflows using an expert system. Bioinformatics 33, 1210–1215 (2017).
Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30, 2114–2120 (2014).
Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013).
Anders, S., Pyl, P. T. & Huber, W. HTSeq-a Python framework to work with high-throughput sequencing data. Bioinformatics 31, 166–169 (2015).
Li, B. et al. RNA-Seq gene expression estimation with read mapping uncertainty. Bioinformatics 26, 493–500 (2010).
DePristo, M. A. et al. A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat. Genet. 43, 491–498 (2011).
Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 1, 417–425 (2015).
McCarthy, R. L., Duncan, A. D. & Barton, M. C. Sample preparation for mass cytometry analysis. J. Vis. Exp. e54394 (2017).
O’Leary, N. A. et al. Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation. Nucleic Acids Res. 44, D733–D745 (2016).
Corces, M. R. et al. An improved ATAC-seq protocol reduces background and enables interrogation of frozen tissues. Nat. Methods 14, 959–962 (2017).
Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359 (2012).
Thorvaldsdóttir, H., Robinson, J. T. & Mesirov, J. P. Integrative Genomics Viewer (IGV): high-performance genomics data visualization and exploration. Brief Bioinform. 14, 178–192 (2013).
We thank L. R. Patel and the Lozano lab for their help and valuable input, as well as L. Gechijian, N. Gray and J. Bradner for their kind gift of a TRIM24 PROTAC. This study was supported by National Institutes of Health grant RO1 CA214871 to MCB and GL, the generous philanthropic contributions to The University of Texas MD Anderson Cancer Center Moon Shots ProgramTM, and the MD Anderson Cancer Center National Cancer Institute Cancer Center Support Grant (CCSG) CA016672. We are grateful to the patients who provided tumor biopsies for PDX model establishment. PDX models and derivatives were obtained from the Cazalot Breast Cancer Model Resource, established through a gift from the Cazalot family and from funds from the MDACC Breast Cancer Moon Shot Program. Additional funding sources that supported this work include the Cancer Prevention and Research Institute of Texas (CPRIT) grants RP160710 (to H.P-W., J.T.C. and P.J.D.) and RP150578 (to P.J.D). Transgenic animals were created at the MD Anderson Cancer Center Genetically Engineered Mouse Facility with the support of CCSG (P30CA016672) and R50 (R50CA211121) grants. RNA-Sequencing was performed at MD Anderson Cancer Center Science Park Next-Generation Sequencing Facility supported by CPRIT Core Facility Support Grants (RP120348 and RP170002). IHC was performed with the support of CCSG-funded Science Park Research Histology, Pathology and Imaging Core (RHPI) at University of Texas MD Anderson Cancer Center Science Park, the Flow Cytometry and Cell Imaging Core shared resource and Histopathology core of the Department of Veterinary Medicine & Surgery are partially funded by NCI Cancer Center Support Grant P30CA16672. High-throughput drug screening was performed at the Texas A&M Combinatorial Drug Discovery Program (CPRIT CFSA RP150578).
S. Moulder has partial support for clinical trials from Pfizer, Genetech, Seattle Genetics, Lilly, Merck, EMDSerono, Bayer, AstraZeneca and Novartis, and no other conflict of interest. All other authors have affirmed no conflict of interest.
Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
Shah, V.V., Duncan, A.D., Jiang, S. et al. Mammary-specific expression of Trim24 establishes a mouse model of human metaplastic breast cancer. Nat Commun 12, 5389 (2021). https://doi.org/10.1038/s41467-021-25650-z