Abstract
Triple-negative breast cancers (TNBC) are genetically characterized by aberrations in TP53 and a low rate of activating point mutations in common oncogenes, rendering it challenging in applying targeted therapies. We performed whole-exome sequencing (WES) and RNA sequencing (RNA-seq) to identify somatic genetic alterations in mouse models of TNBCs driven by loss of Trp53 alone or in combination with Brca1. Amplifications or translocations that resulted in elevated oncoprotein expression or oncoprotein-containing fusions, respectively, as well as frameshift mutations of tumor suppressors were identified in approximately 50% of the tumors evaluated. Although the spectrum of sporadic genetic alterations was diverse, the majority had in common the ability to activate the MAPK/PI3K pathways. Importantly, we demonstrated that approved or experimental drugs efficiently induce tumor regression specifically in tumors harboring somatic aberrations of the drug target. Our study suggests that the combination of WES and RNA-seq on human TNBC will lead to the identification of actionable therapeutic targets for precision medicine–guided TNBC treatment.
Significance: Using combined WES and RNA-seq analyses, we identified sporadic oncogenic events in TNBC mouse models that share the capacity to activate the MAPK and/or PI3K pathways. Our data support a treatment tailored to the genetics of individual tumors that parallels the approaches being investigated in the ongoing NCI-MATCH, My Pathway Trial, and ESMART clinical trials. Cancer Discov; 8(3); 354–69. ©2017 AACR.
See related commentary by Natrajan et al., p. 272.
See related article by Matissek et al., p. 336.
This article is highlighted in the In This Issue feature, p. 253
Introduction
Targeted therapies for breast cancer treatment using estrogen receptor (ER) antagonists, mAbs, or small molecules directed against HER2 for tumors expressing these cognate targets have resulted in great improvement in patient survival in ER+, progesterone receptor–positive (PR+), and HER2-amplified patient populations (1–3). In contrast, there are no similar specific targeted therapies available for triple-negative breast cancer (TNBC), which is defined by the absence of ER, PR, and HER2. This disease occurs more commonly in younger women and in women of African and Hispanic descent, and patients are at higher risk of local or distant recurrence and worse prognosis when compared with other breast cancer subtypes (4). Despite significant efforts, little progress has been made, and chemotherapy remains the standard of care for patients with TNBC.
Patients with TNBC exhibit highly variable responses to chemotherapy, which is likely linked to the significant molecular differences observed among these tumors (4). Recent profiling of the tumor transcriptomes has highlighted the heterogeneity in TNBC (5, 6), and transcriptional profiling allowed the classification of TNBC into discrete subgroups, each with distinct expression profiles and, importantly, clinical implications (5, 7). Large-scale whole-exome sequencing (WES) and whole-genome sequencing projects further demonstrated the molecular basis for TNBC tumor heterogeneity, revealing a diverse range and number of mutations, chromosomal number variations, and translocations (7–9). Although over 80% of patients with TNBC harbor point mutations or deletions in the TP53 locus and a smaller subset have point mutations in genes controlling the PI3K pathway (PTEN, PIK3CA, and INPP4B), gain-of-function point mutations that are common in other cancers, such as those found in BRAF, KRAS, and EGFR, are rare events in TNBCs (8–11). Several of these oncogenic drivers are potentially targetable using small molecules that are either FDA approved or currently undergoing clinical evaluation. However, the low frequency of each oncogenic mutation in patients with TNBC precludes the broad usage of any single targeted agent in this patient population. Rather, these data provide a rationale for pursuing a precision medicine approach to identify potential drivers, and to tailor treatment regimens based upon the somatic alteration spectrum specific to each patient's tumor.
Several issues confound applying precision medicine approaches to TNBCs. Most trials do not require genomic profiling on biopsies at the time of enrollment, rendering correlations between clinical responses and genetic aberrations difficult to establish. In recent years, patient-derived xenografts (PDX) have been used to test personalized medicine approaches in treating patient-derived breast cancers (12). However, these models use immunocompromised mice, and their clinical applicability remains unclear.
Here, we took a complementary approach to establish genetically engineered mouse models (GEMM) harboring breast-specific Trp53 deletion with or without concomitant Brca1 deletion. We carried out WES and RNA sequencing (RNA-seq) on the tumors that these TNBC models developed and confirmed that they faithfully recapitulated features of human TNBC disease. We were able to identify putative driver aberrations, including chromosomal amplifications and deletions, chromosomal translocations, mutations, and proto-oncogene overexpression specific to individual tumors. Interestingly, although different tumors harbor different genetic aberrations, many of these share the ability of enhancing MAPK and PI3K pathway activation. In addition, when a particular tumor was treated with first-line therapeutic drugs targeting predicted oncogenic drivers specific to that tumor, sustained tumor remission was achieved. These data further support the tumor-promoting functions of these genetic events and support the paradigm of genomics-guided treatment in TNBC. Our results suggest that the combination of WES and RNA-seq on human TNBC can reveal genetic aberrations that would be missed by conventional approaches that evaluate mutational events in panels of known oncogenes. Importantly, our results suggest that this approach can significantly extend the use of approved or experimental drugs for patients with TNBC who progress on standard-of-care treatments.
Results
Establishment of Brca1-deficient and Brca1-WT TNBC Mouse Models
The majority of TNBC tumors possess TP53 mutations, and approximately 15% of patients carry germline BRCA1 mutations (4). We generated two cohorts of mice to model both BRCA1-germline mutation carriers and sporadic TNBCs: K14cre; Trp53flox/flox; Brca1flox/flox, defined as Brca1-deficient cohort; and K14cre; Trp53flox/flox; Brca1wt/wt, defined as Brca1 wild-type (WT) cohort (Supplementary Fig. S1A; ref. 13). The majority of tumors developed from these mouse models were ER−, PR−, and HER2−, with histopathologic characteristics reminiscent of human TNBCs (Fig. 1A; Supplementary Fig. S1B). To further investigate whether these mouse models resembled human TNBC, we performed RNA-seq on RNA isolated from tumors and from age-matched normal mammary glands to analyze their transcriptional profiles. Both AIMS and PAM50 classifier algorithms (14, 15), as well as unsupervised hierarchical clustering, were used to identify the intrinsic subtypes of these mouse tumors. AIMS assigned almost all mouse tumors with high probability into basal-like breast cancer tumor subtypes, whereas all three normal mammary glands were assigned to the “normal-like” subtype (Fig. 1B). Similarly, PAM50 assigned most of the tumors to the basal-like breast cancer subtype, which is the subtype most commonly associated with TNBC (Supplementary Fig. S1C; ref. 16). In addition, unsupervised hierarchical clustering of the mouse tumors with RNA-seq expression profiles from breast cancer samples from The Cancer Genome Atlas (TCGA) on the 138 shared genes between mouse and human revealed that the mouse tumors segregated with human TNBCs, and not with other breast cancer subtypes (Supplementary Fig. S1D; Supplementary Table S1). Therefore, both immunohistologic analysis and transcriptional profiling indicated that these mouse tumors, developed on a Trp53-deficient background with or without Brca1, provide reasonable models for human TNBC.
Establishment of TNBC mouse models that recapitulate many aspects of human TNBC, including heterogeneous genomic alterations. A, Hematoxylin and eosin (H&E) and IHC staining of ER, PR, and HER2 for tumors developed in GEMMs. B, Transcriptional classification using AIMS reveals endogenous tumors developed in the mouse models are most similar to basal-like breast cancer. C, Endogenous mouse tumors show a range of mutations.D, Mutational signature for each Brca1 genotype. The six boxes show the relative frequency of each of the six substitution types, and within each box are the 16 substitution contexts depending on the 5′ and 3′ base pair context. E, Global CNA profile shows heterogeneous amplifications and deletions among different tumors. F, Recurrent amplification at the Met and Yap1 loci in the mouse tumors and their corresponding mRNA levels. Amplified samples are in red. G, Individual tumors with Egfr or Fgfr2 amplification express higher mRNA levels (expressed as fragments per kilobase million or PRKM) of Egfr or Fgfr2, respectively, whereas a tumor with Pten deletion shows low Pten transcript level. Arrows indicating amplified or deleted samples are in red.
Establishment of TNBC mouse models that recapitulate many aspects of human TNBC, including heterogeneous genomic alterations. A, Hematoxylin and eosin (H&E) and IHC staining of ER, PR, and HER2 for tumors developed in GEMMs. B, Transcriptional classification using AIMS reveals endogenous tumors developed in the mouse models are most similar to basal-like breast cancer. C, Endogenous mouse tumors show a range of mutations.D, Mutational signature for each Brca1 genotype. The six boxes show the relative frequency of each of the six substitution types, and within each box are the 16 substitution contexts depending on the 5′ and 3′ base pair context. E, Global CNA profile shows heterogeneous amplifications and deletions among different tumors. F, Recurrent amplification at the Met and Yap1 loci in the mouse tumors and their corresponding mRNA levels. Amplified samples are in red. G, Individual tumors with Egfr or Fgfr2 amplification express higher mRNA levels (expressed as fragments per kilobase million or PRKM) of Egfr or Fgfr2, respectively, whereas a tumor with Pten deletion shows low Pten transcript level. Arrows indicating amplified or deleted samples are in red.
RNA-seq and WES Revealed Heterogeneous Aberrations in Murine TNBC
To identify spontaneous genetic aberrations that potentially are oncogenic drivers of these tumors, we performed WES on tumor DNA and matched germline DNA. We observed low somatic mutation rates; an average of 30 somatic mutations (and 17 nonsynonymous mutations) were found in the coding exons per tumor, ranging from 0 to 104 mutations across our set (Fig. 1C; Supplementary Table S2). Among all nonsynonymous mutations, we identified only KrasQ61H and HrasQ61K as potential oncogenic driver mutations in two of the primary tumors. Both mutations have been observed in human tumors and have been shown to drive transformation in cell lines (http://cancer.sanger.ac.uk/cosmic/mutation/overview?id=554 and refs. 17, 18). Although other mutations were identified, they were not at hotspots in human oncogenes, and their oncogenic potential remains to be determined (Supplementary Fig. S1E). We compared the mouse tumor mutational signatures to the 21 mutation signatures identified in human cancers (19) and found that Brca1-deficient tumors were most similar to signature 3, which is associated with BRCA1/2 mutations in human breast cancer (Fig. 1D; ref. 8). We also found that the mutation signature for Brca1-WT tumors was most similar to signature 17, which has an unknown etiology (Fig. 1D; Supplementary Fig. S1F).
Next, we analyzed somatic copy-number alterations (CNA) to investigate chromosomal gains and losses in these murine tumors. Overall, the CNA profiles exhibited marked variability among the tumors (Fig. 1E; Supplementary Table S3). Consistent with human breast cancer (20), the Brca1-deficient subgroup exhibited more amplifications and deletions than the Brca1-WT subgroup (Supplementary Fig. S1G), supporting that loss of Brca1 contributes to genomic instability. The most recurrent events were focal amplifications on chromosome 6 centered on the Met locus and amplifications on chromosome 9 centered on the Yap1 locus (Fig. 1F; Supplementary Fig. S1H). Amplification of the Met gene was observed in 10 of the 32 Brca1-WT primary tumors but in none of the Brca1-deletion tumors. Amplifications in the Yap1 gene were found in 8 of 32 Brca1-WT and 1 of 30 Brca1-deletion tumors. We also observed recurrent chromosomal gain at the Myc locus on chromosome 15, and deletion of the Rb locus within a broad region on chromosomal 14 (Supplementary Fig. S1H). In addition to these recurrent amplification events, we identified sporadic high focal amplification on chromosome 7 that includes the Fgfr2 locus and on chromosome 11 that includes the Egfr locus, and a biallelic deletion on chromosome 19 that includes the Pten locus in one tumor (Fig. 1G; Supplementary Fig. S1I).
We integrated WES data with RNA-seq data to investigate whether CNA alterations led to changes in the expression levels of the affected genes. Among the recurrent amplifications, 7 of the 10 Met-amplified tumors indeed showed increased Met transcript levels, and 6 of the 9 Yap1 amplifications resulted in elevated mRNA expression (Fig. 1F). In addition, we found that tumor #1 with the Fgfr2 amplification expressed a significantly higher level of Fgfr2 mRNA compared with other tumors that were CNA-neutral at this locus. The tumor with a focal amplification of Egfr was among the ones expressing the highest Egfr mRNA levels. The tumor harboring a biallelic Pten deletion expressed the lowest Pten mRNA levels among all tumors (Fig. 1G). It has been shown that overexpression of MET (21, 22), EGFR (23–25), YAP1 (26–28), or FGFR2 (29, 30), as well as deletion of PTEN (31, 32), contribute to TNBC tumorigenesis, and our data suggest that through combined RNA-seq and WES data analyses, potential oncogenic drivers of TNBC tumors can be identified.
Chromosomal Rearrangements in Primary and Transplanted Tumors
Analysis of RNA-seq data also revealed the presence of chromosomal translocation events, which varied from tumor to tumor (Supplementary Fig. S2A), and a subset of these rearrangements resulted in mRNA fusions (Fig. 2A and B). Several of the fusions involved genes encoding proto-oncogenic protein kinases (Supplementary Fig. S2B; Supplementary Table S4). Three different Fgfr2 fusions were observed: Fgfr2–Dnm3 (Dynamin 3) in tumor #1, Fgfr2–Tns1 (Tensin 1) in tumor #6, and Fgfr2–Zmynd8 where the distal portion of the Fgfr2 gene was replaced by the translocation partners, potentially generating fusion proteins (Fig. 2C) with intact kinase domains. Proteins generated from these Fgfr2-fusion transcripts share the same domain architecture as FGFR2 fusions identified in human breast cancer (33, 34). Notably, in tumor #1 where Fgfr2–Dnm3 translocation was observed, the chromosomal region encoding Fgfr2 was highly amplified, whereas in tumor #6 with an Fgfr2–Tns1 fusion, the chromosomal region covering Fgfr2 exhibited some gain in copy number (Supplementary Fig. S1I). Translocations involving the Raf family kinases were also detected, including a Dlg1–Braf fusion in tumor #3, a Dhx9–Raf1 fusion in tumor #5, and a Rpl32–Raf1 fusion in tumor #25. In all cases, the fusion proteins had an intact RAF kinase domain, whereas the RAF N-terminal regulatory domain was replaced by the fusion partners, similar to BRAF fusions identified in metastatic breast cancer (35) and other BRAF and RAF1 fusions found in different human tumor types (Fig. 2C and Supplementary Fig. S2C; refs. 36–38).
Next-generation sequencing of the endogenous TNBC tumors from GEMMs identifies multiple chromosomal rearrangement events. A, An overall view of fusion events in mouse breast tumors. B, Representative Circos plots show different numbers and patterns in chromosomal translocations among different tumors. C, Domain structure of FGFR2, BRAF, and RAF1 fusion kinases. D and E, RT-PCR (D) and Sanger sequencing (E) confirming the presence of Fgfr2–Dnm3, Fgfr2–Tns1, and Dhx9–Raf1 fusion transcripts in each of the spontaneous tumors.
Next-generation sequencing of the endogenous TNBC tumors from GEMMs identifies multiple chromosomal rearrangement events. A, An overall view of fusion events in mouse breast tumors. B, Representative Circos plots show different numbers and patterns in chromosomal translocations among different tumors. C, Domain structure of FGFR2, BRAF, and RAF1 fusion kinases. D and E, RT-PCR (D) and Sanger sequencing (E) confirming the presence of Fgfr2–Dnm3, Fgfr2–Tns1, and Dhx9–Raf1 fusion transcripts in each of the spontaneous tumors.
We validated the presence of fusion transcripts in tumors harboring Fgfr2 and Raf1 fusions by RT-PCR and traditional Sanger sequencing using primer pairs that either match the input RNA (match primers) or not (mismatch primers). PCR products of expected sizes were observed for matching RNA/primers, but not for mismatch RNA/primers (Fig. 2D). When RT-PCR products were gel extracted and sequenced by traditional Sanger sequencing, the junctions were confirmed (Fig. 2E; Supplementary Fig. S2D–S2F).
Thus, through combined WES and RNA-seq analyses, we detected spontaneous chromosomal amplifications of Met, Yap1, Egfr, and Fgfr2, a biallelic deletion of Pten, oncogenic mutations in Kras and Hras, as well as sporadic fusions of Fgfr2–Dnm3, Fgfr2–Tns1, Fgfr2–Zmydn8, Dhx9–Raf1, and Rpl32–Raf1 and Dlg1–Braf in distinct tumors.
Aberrant Genetic Events Possess Oncogenic Activity by Enhancing MAPK and/or PI3K Pathway Activations
We next investigated whether the fused transcripts resulted in functional fusion proteins. We detected the FGFR2–DNM3 fusion protein in tumor #1 at the predicted size of 99 kDa using antibodies recognizing the N-terminus of FGFR2 and the C-terminus of DNM3. Importantly, the presence of FGFR2–DNM3 coincided with elevated phosphorylation of the FGFR2 substrate FRS2 (Fig. 3A and Supplementary Fig. S3A for sample key). To confirm the presence of FGFR2–TNS1 fusion proteins, we immunoprecipitated FGFR2 and subjected proteins at the predicted molecular mass (265 kDa) for mass spectrometry and identified peptides from both FGFR2 and TNS1 (Fig. 3B and Supplementary Fig. S3B for peptide sequences matching FGFR2 or TNS1). We also confirmed the presence of the DHX9–RAF1 fusion protein in tumor #5 at the expected molecular weight (51 kDa; Fig. 3C), and this tumor exhibited elevated activation of the MAPK signaling pathway as measured by ERK phosphorylation, compared with tumors that lacked RAF fusions (Fig. 3C; Supplementary Fig. S3C).
Western blotting confirming the presence of fusion proteins in tumor lysates and showing increased signaling pathway activation in respective tumors. A, Tumor #1 series express FGFR2–DNM3 fusion proteins at expected molecular weight (left). At the same molecular weight, its fusion partner DNM3 is also detected. FGFR2–DNM3 expression coincides with enhanced FRS2 phosphorylation. B, Detection of FGFR2–TNS1 in tumor #6 by mass spectrometry (MS) upon FGFR2 immunoprecipitation. Short gray blocks show the locations of peptides identified by MS. C, Tumor #3 series express DHX9–RAF1 fusion proteins at the expected molecular weight with increased pERK signals. D, Tumor #2 expresses a high level of c-MET at the protein levels, which coincides with increased MET phosphorylation. E, Tumor #17 that harbors biallelic PTEN deletion expresses undetectable PTEN at the protein levels, and high AKT and PRAS40 phosphorylation.
Western blotting confirming the presence of fusion proteins in tumor lysates and showing increased signaling pathway activation in respective tumors. A, Tumor #1 series express FGFR2–DNM3 fusion proteins at expected molecular weight (left). At the same molecular weight, its fusion partner DNM3 is also detected. FGFR2–DNM3 expression coincides with enhanced FRS2 phosphorylation. B, Detection of FGFR2–TNS1 in tumor #6 by mass spectrometry (MS) upon FGFR2 immunoprecipitation. Short gray blocks show the locations of peptides identified by MS. C, Tumor #3 series express DHX9–RAF1 fusion proteins at the expected molecular weight with increased pERK signals. D, Tumor #2 expresses a high level of c-MET at the protein levels, which coincides with increased MET phosphorylation. E, Tumor #17 that harbors biallelic PTEN deletion expresses undetectable PTEN at the protein levels, and high AKT and PRAS40 phosphorylation.
The tumor with high Met mRNA expression had elevated tyrosine-phosphorylated MET compared with tumors that lacked MET overexpression (Fig. 3D; Supplementary Fig. S3D). The tumor with biallelic deletion of Pten (#17) exhibited low levels of PTEN protein and elevated PI3K pathway activation detected by high levels of phospho-AKT and high levels of phosphorylation of the AKT substrate PRAS40 (Fig. 3E; Supplementary Fig. S3E).
FGFR2, BRAF, and RAF1 Fusion Kinases Exhibit Enhanced MAPK and/or PI3K Pathway Activation and Oncogenic Activity
To evaluate the oncogenic potential of the novel fusion proteins, we cloned HA-tagged Fgfr2–Dnm3, Dhx9–Raf1, and Rpl32–Raf1 into pBabe-puro and generated stable NIH3T3 cells and carried out signaling analyses and soft-agar colony formation assays. FGFR2 fusions, similar to the ones we observed in mice albeit with different translocation partners, have been reported in human breast and thyroid cancers (33, 34), and we included one of these fusions, FGFR2–CCDC6 (Supplementary Fig. S4A) in our assays.
Expression of the fusion proteins was confirmed by blotting for the HA-tag, FGFR2, and RAF1 (Supplementary Fig. S4B and S4C). We found that expression of FGFR2–DNM3 and FGFR2–CCDC6 resulted in FGFR2 activation detected by phospho-Tyr653/654 of FGFR2, and substrate phosphorylation detected by phospho-FRS2 at Tyr436. As a consequence, downstream PI3K activation was enhanced as measured by phospho-AKT (Fig. 4A; Supplementary Fig. S4B and S4C). Moreover, both FGFR2–DNM3 and FGFR2–CCDC6 induced anchorage-independent growth of NIH3T3 cells in soft agar (Fig. 4B). To understand the mechanisms for enhanced pathway activation, we transiently expressed epitope-tagged FGFR2 fusions in HEK293T cells and assayed for ligand-independent oligomerization/dimerization by coimmunoprecipitation in serum-free condition, and we found both FGFR2–CCDC6 and FGFR2–DNM3 undergo homo-oligomerization in the absence of FGF but fail to oligomerize with endogenous FGFR2s (Fig. 4C). These results are consistent with a model in which the C-terminal fusion partner (CCDC6 and DNM3) mediates the homo-multimerization.
Kinase fusions function as oncogenic drivers via enhanced activation of the MAPK and/or PI3K pathways. A, Expression of FGFR2–CCDC6 or FGFR2–DNM3 results in increased FGFR2, FRS2, and downstream AKT phosphorylation in NIH3T3 cells. B, FGFR2–CCDC6 and FGFR2–DNM3 confer soft-agar colony formation capacity in NIH3T3 cells. C, Both FGFR2–DNM3 and FGFR2–CCDC6 dimerize/oligomerize in a ligand-independent manner with fusion kinases but not with endogenous FGFR2. D, Expressing of DHX9–RAF1 or RPL32–RAF1 fusion kinases significantly enhances the MAPK pathway in NIH3T3 cells. E, DHX9–RAF1 and RPL32–RAF1 fusion kinases confer soft-agar colony formation ability. F, Expression of FGFR2–CCDC6, FGFR2–DNM3, DHX9–RAF1, or RPL32–RAF1 increases HMEC proliferation. G, MET overexpression in HMECs induces spontaneous MET phosphorylation. H, Potential oncogenic driver events identified through combined WES and RNA-seq analyses.
Kinase fusions function as oncogenic drivers via enhanced activation of the MAPK and/or PI3K pathways. A, Expression of FGFR2–CCDC6 or FGFR2–DNM3 results in increased FGFR2, FRS2, and downstream AKT phosphorylation in NIH3T3 cells. B, FGFR2–CCDC6 and FGFR2–DNM3 confer soft-agar colony formation capacity in NIH3T3 cells. C, Both FGFR2–DNM3 and FGFR2–CCDC6 dimerize/oligomerize in a ligand-independent manner with fusion kinases but not with endogenous FGFR2. D, Expressing of DHX9–RAF1 or RPL32–RAF1 fusion kinases significantly enhances the MAPK pathway in NIH3T3 cells. E, DHX9–RAF1 and RPL32–RAF1 fusion kinases confer soft-agar colony formation ability. F, Expression of FGFR2–CCDC6, FGFR2–DNM3, DHX9–RAF1, or RPL32–RAF1 increases HMEC proliferation. G, MET overexpression in HMECs induces spontaneous MET phosphorylation. H, Potential oncogenic driver events identified through combined WES and RNA-seq analyses.
In contrast, expression of DHX9–RAF1 and RPL32–RAF1 stimulated MAPK pathway activation with minimal impact on PI3K pathway activation in NIH3T3 cells (Fig. 4D), and this also resulted in the significantly increased capacity to form colonies in soft agar (Fig. 4E).
We also analyzed the effects of these fusion proteins in immortalized human mammary epithelial cells (HMEC). Consistent with the results obtained in NIH3T3 cells, FGFR2–DNM3 expression increased both MAPK and PI3K signaling in epithelial cells (Supplementary Fig. S4D). Surprisingly, DHX9–RAF1 expression in epithelial cells also enhanced activation of the PI3K pathway (Supplementary Fig. S4D). Previous studies have shown autocrine production of various growth factors contribute to RAF1-mediated PI3K activation in mammary epithelial cells (39–42), which may explain the signaling differences observed between NIH3T3 cells and HMECs. In addition, expression of the fusion proteins resulted in increased proliferation of HMECs (Fig. 4F). c-MET overexpression resulted in spontaneous c-MET phosphorylation in the absence of growth factors and enhanced phospho-AKT and phospho-ERK levels (Fig. 4G), as well as increased cell proliferation (Supplementary Fig. S4E).
Altogether, our results identify a likely driver event in approximately 50% of the 72 tumors evaluated (Fig. 4H). Our data show that the majority of the genetic alterations we identified have in common the ability to enhance activation of MAPK and/or PI3K signaling.
Tumors Harboring FGFR2 Fusion Proteins Are Responsive to FGFR2 Inhibition In Vivo, Resulting in Complete Tumor Regression
Next, we asked whether the murine TNBC tumors that spontaneously acquired the Fgfr2 or Raf1 translocations would respond to drugs that target these protein kinases. To this end, we transplanted mouse TNBC tumors with identifiable potential genetic drivers in cohorts of nude mice (43), and each mouse received a designated treatment regimen (Supplementary Fig. S5A). Targeted agents were selected based on (i) high specificity and high competence indicated by low IC50; (ii) good in vivo bioavailability with low in vivo dosing; and (iii) having been approved for cancer therapy or currently in phase II or III clinical trials. On the basis of these criteria, NVP-BGJ398 was chosen as an FGFR inhibitor (Supplementary Fig. S5B), trametinib (GSK1120212) as a MEK inhibitor (Supplementary Fig. S6A), and crizotinib as a MET inhibitor (Supplementary Fig. S6D) for in vivo treatments.
For tumor #1, which is Brca1-deficient with a spontaneous Fgfr2–Dnm3 translocation, we evaluated the efficacy of FGFR inhibitor BGJ398 alone, or in combination with olaparib (a PARP inhibitor approved for BRCA1-mutant cancers; Fig. 5A), anticipating that tumors may become resistant on monotherapy. We also included crizotinib in this trial, reasoning that MET is not the driver in this particular tumor, and it should therefore not respond to this treatment. We determined that the dose of BGJ398 needed to suppress FRS2 phosphorylation was 30 mg/kg (Fig. 5B), which is estimated to be equivalent to a human dose of 2.4 mg/kg (44), and the dose of olaparib had previously been reported (45). We did not observe general toxicity over prolonged treatment (Fig. 5C). At this dose, BGJ398 alone was sufficient to induce complete tumor regression (Fig. 5D). This outcome was a significant improvement compared with a pan-PI3K inhibitor BKM120 treatment, which resulted only in a slower rate of tumor progression (Supplementary Fig. S5C). Importantly, crizotinib had no effect on the growth of tumor #1 (Fig. 5D), although it was capable of delaying tumor progression when MET levels are elevated (Fig. 6F), supporting the notion that the choice of treatment should be based on the genetics of the tumor. At early time points (up to 33 days), BGJ398 and the combination of BGJ398 and olaparib were equally effective in inducing tumor remission. However, three of six tumors on BGJ398 monotherapy developed resistance, and on average at day 43 (days 36, 47, and 57) tumors relapsed to the initial volume (Fig. 5E). In contrast, tumors treated with BGJ398 and olaparib combination therapy did not relapse when the experiment ended, and all tumors reached complete response (CR; Fig. 5E). Using a second PARP inhibitor, BioMarine BMN673 (46), in combination with BGJ398, no tumor relapse was observed for up to 80 days (Supplementary Fig. S5D). Our data suggest that spontaneous Fgfr2–Dnm3 fusions can drive tumorigenesis in the breast.
Targeting FGFR2 fusion–containing tumors with the FGFR inhibitor BGJ398 results in CR. A, Treatment designs for tumor #1 that is Brca1-deficient and harbors Fgfr2–Dnm3 translocation. B, In vivo dosing of BGJ398 at 30 mg/kg achieves target inhibition measured by pFRS2 and pERK levels. C, BGJ398 alone, or in combination with olaparib, does not cause general toxicity in nude mice. D, BGJ398 alone or in combination with olaparib results in complete tumor remission. E, Spider plot showing tumor responses to each treatment over long duration. F, BGJ398 alone results in slow tumor regression, whereas BKM120 and BGJ398 cause fast tumor remission.
Targeting FGFR2 fusion–containing tumors with the FGFR inhibitor BGJ398 results in CR. A, Treatment designs for tumor #1 that is Brca1-deficient and harbors Fgfr2–Dnm3 translocation. B, In vivo dosing of BGJ398 at 30 mg/kg achieves target inhibition measured by pFRS2 and pERK levels. C, BGJ398 alone, or in combination with olaparib, does not cause general toxicity in nude mice. D, BGJ398 alone or in combination with olaparib results in complete tumor remission. E, Spider plot showing tumor responses to each treatment over long duration. F, BGJ398 alone results in slow tumor regression, whereas BKM120 and BGJ398 cause fast tumor remission.
Targeting DHX9–RAF1 and c-MET with MEK and MET inhibitor, respectively, results in tumor regression or delayed progression. A, The MEK inhibitor trametinib (GSK1120212) achieves target inhibition at 3 mg/kg. B, Treatment designs for BRCA1-deficient tumor #5 harboring Dhx9–Raf1 translocation. C, Trametinib alone delays tumor progression, whereas trametinib plus olaparib results in complete tumor remission. D, At 50 mg/kg, crizotinib inhibits MET phosphorylation and downstream signaling pathway activation. E, Targeting MET-overexpressing endogenous tumor with MET inhibitor delays tumor progression, and combination therapy results in stable disease. F, Treatment designs for tumor #2 that is Brca1-WT and overexpresses c-MET.
Targeting DHX9–RAF1 and c-MET with MEK and MET inhibitor, respectively, results in tumor regression or delayed progression. A, The MEK inhibitor trametinib (GSK1120212) achieves target inhibition at 3 mg/kg. B, Treatment designs for BRCA1-deficient tumor #5 harboring Dhx9–Raf1 translocation. C, Trametinib alone delays tumor progression, whereas trametinib plus olaparib results in complete tumor remission. D, At 50 mg/kg, crizotinib inhibits MET phosphorylation and downstream signaling pathway activation. E, Targeting MET-overexpressing endogenous tumor with MET inhibitor delays tumor progression, and combination therapy results in stable disease. F, Treatment designs for tumor #2 that is Brca1-WT and overexpresses c-MET.
For tumor #6, which is BRCA1-WT with a spontaneous Fgfr2–Tns1 translocation, we compared BGJ398 alone with the PI3K inhibitor BKM120 (Supplementary Fig. S5E), because BRCA1-WT murine tumors are less sensitive to PARP inhibition relative to PI3K inhibition (Supplementary Fig. S5E). We treated tumor-bearing mice with BGJ398 alone, BKM120 alone, or both drugs in combination (Supplementary Fig. S5F). We observed that BKM120 treatment resulted in stable disease, although with general toxicity (Supplementary Fig. S5G). However, BGJ398 alone induced tumor regression, and the combination of both drugs caused fast and complete tumor remission until no tumor tissue could be detected (Fig. 5F). The combination did not cause further toxicity beyond what was elicited by BKM120 alone (Supplementary Fig. S5G). Our data show that spontaneous tumors with Fgfr2 aberrations are responsive to FGFR inhibitor treatments, suggesting complete tumor regression can be achieved when genetic alterations of each tumor are carefully considered.
Tumors with Raf1 Fusion or MET Overexpression Responded to Specific Inhibitors Targeting These Pathways
As MEK inhibitors have been shown to effectively treat human tumors driven by mutated RAF family members, we treated tumor #5 that expresses the DHX9–RAF1 fusion protein with the MEK inhibitor trametinib (GSK1120212). A 3 mg/kg/day dose of trametinib was effective in blocking ERK phosphorylation (Fig. 6A). Mice implanted with tumor #5 were given trametinib either as a single agent, or in combination with olaparib because tumor #5 is a Brca1-deficient tumor (Fig. 6B). Either single-agent or combination therapy did not cause general toxicity at the doses used (Supplementary Fig. S6B). Consistent with our hypothesis, trametinib alone significantly delayed tumor growth, with an initial phase of stable disease, which transitioned to progressive disease after 14 days (Supplementary Fig. S6C). Olaparib alone initially did not halt tumor growth, but slowly resulted in tumor regression. When mice were treated with a combination of both the MEK and PARP inhibitors, tumor #5 completely regressed without relapse upon extended treatment for up to 50 days till the experiment ended (Fig. 6C). It is also important to note that when treated with the FGFR inhibitor BGJ398, tumor #5 (Dhx9–Raf1 fusion) failed to respond (Fig. 6C; Supplementary Fig. S6C), further supporting the importance of identifying and suppressing specific oncogenic activities in specific patient subpopulations.
RNA-seq revealed that a group of tumors derived from the same primary tumor (tumor #2; Fig. 1F) exhibited approximately a 30-fold increase in Met mRNA levels compared with other tumors evaluated, despite no amplification in the Met locus. Because MET overexpression enhances spontaneous MET activation in cells, we evaluated the effect of crizotinib, an inhibitor of both MET and ALK tyrosine kinases, on the growth of these tumors. At 50 mg/kg, crizotinib completely blocked c-MET phosphorylation in tumors, as well as downstream AKT, ERK, and S6 phosphorylation (Fig. 6D) without general toxicity (Supplementary Fig. S6D). Although overall survival for mice receiving control treatment is only 8 days, crizotinib treatment delayed tumor progression and tripled survival to 24 days (Fig. 6E). Combining crizotinib with BKM120 (Fig. 6F) achieved stable disease in this particularly aggressive tumor (Fig. 6E). The effect of crizotinib on the tumor was probably a consequence of inhibiting MET because, as discussed above, crizotinib had little effect on tumor #1 (Fig. 5D), which does not overexpress MET.
The results presented in Figs. 5 and 6 indicate that either complete tumor regression or stable disease can be achieved, using monotherapy or combination therapy with inhibitors that target the pathways that are affected by the genetic aberrations specific to individual tumors. These in vivo treatment results not only confirm the functional importance of these genetic events in tumor development, but also provide evidence that individualized treatment design based on genomic information can lead to significantly improved outcomes.
Human TNBCs Harbor a Broad Range of Targetable Genetic Alterations
Our data show that despite the use of highly inbred mice and common initiating events (breast-specific deletion of Trp53 alone or deletion of both Trp53 and Brca1), tumors took diverse evolutionary pathways to become TNBC. Yet, the majority of the tumors converged on genetic alterations enhancing MAPK and PI3K signaling pathways. Therefore, we analyzed TCGA human breast cancer data to investigate genetic alterations known to enhance MAPK and/or PI3K pathway activation.
We first interrogated 82 tumors in the TCGA database that were annotated as TNBC. Specifically, we searched for mutations and amplifications in KRAS, BRAF, and RAF1 genes; mutation, amplification, and overexpression in genes encoding for receptor tyrosine kinases (RTK); as well as genes involved in the PI3K pathway. We found that approximately 90% of all TNBCs have at least one such genetic aberration (Fig. 7A; Supplementary Fig. S7A). To determine whether this high prevalence is specific to TNBC, or is generally true for all breast cancer, we investigated the TCGA breast cancer database (816 cases) and the Molecular Taxonomy of Breast Cancer International Consortium database (METABRIC; 2,509 cases) and found that 72% and 56% of breast tumor samples, respectively, showed at least one of these changes, with the majority of the tumors displaying two or more alterations (Fig. 7A; Supplementary Fig. S7B). The result from this search suggests that although genetic changes leading to potential MAPK/PI3K activation are common, such alterations may be particularly important for TNBC development.
Multiple, different genetic aberrations lead to common elevated MAPK and/or PI3K pathway activation in patients with human breast cancer. A, Frequent genetic alterations that are known to activate the MAPK and/or PI3K pathways in patients with TNBC (cBioPortal TCGA data). B, Frequency of amplification at the FGFR1, FGFR2, and FGF19 loci. C, Genomic amplification of FGFR1 and FGFR2 results in increased mRNA expression levels. D, Location and frequency of mutations found in FGFR1 and FGFR2 in breast cancer. E, A model showing multiple genetic aberrations in TNBC leading to the activation of MAPK and/or PI3K pathways.
Multiple, different genetic aberrations lead to common elevated MAPK and/or PI3K pathway activation in patients with human breast cancer. A, Frequent genetic alterations that are known to activate the MAPK and/or PI3K pathways in patients with TNBC (cBioPortal TCGA data). B, Frequency of amplification at the FGFR1, FGFR2, and FGF19 loci. C, Genomic amplification of FGFR1 and FGFR2 results in increased mRNA expression levels. D, Location and frequency of mutations found in FGFR1 and FGFR2 in breast cancer. E, A model showing multiple genetic aberrations in TNBC leading to the activation of MAPK and/or PI3K pathways.
Because mouse tumors with spontaneous FGFR signaling pathway activation are sensitive to the FGFR inhibitor BGJ398, we were interested in further investigating genetic alterations of the FGFR family of RTKs and their ligands. In addition to translocations, mutations and focal amplifications can also activate the FGFR family (33, 34, 47, 48). Therefore, we also analyzed the Catalogue of Somatic Mutations in Cancer (COSMIC) databases as well as the TCGA and METABRIC databases to identify mutations for FGFR and FGF family members. We found that chromosomal amplifications of the FGFR1, FGFR2, and FGF3/4/19 loci (Fig. 7B; Supplementary Fig. S7C) resulted in increased mRNA levels of FGFR1, FGFR2, and FGF4/19 (Fig. 7C; Supplementary Fig. S7D). Overall, 34% and 26% of all patients with breast cancer showed FGFR1, FGFR2, and/or FGF3 mutation/amplification/overexpression in TCGA and METABRIC, respectively (Supplementary Fig. S7E). Mutations in FGFR1 are rare, but 1.2% of all patients with breast cancer have FGFR2 mutations, including known activating mutations such as S252W, N549K, and K659E (Fig. 7D).
These data indicate that in both GEMMs and human patients, the genetic alterations in TNBC are quite diverse but converge on a relatively limited number of ways to activate the MAPK and/or PI3K pathways (Fig. 7E).
Discussion
In this work, we generated two GEMMs that are Trp53- deficient, with or without Brca1 deficiency, and carried out WES and RNA-seq on these spontaneously developed tumors. We found that these TNBC mouse models recapitulated many aspects of human TNBC, including the overall transcriptional profile, mutational burden and signature, as well as heterogeneous CNAs. We identified spontaneous genetic aberrations that were likely drivers of the evolution of these tumors. These genetic aberrations included both recurrent chromosomal amplifications of the Met and Yap1 loci and sporadic amplification at the Fgfr2 and Egfr loci, a biallelic deletion of Pten in individual tumors. In addition, spontaneous Fgfr2–Dnm3, Fgfr2–Tns1, Fgfr2–Zymnd8, Dhx9–Raf, Rpl32–Raf1, and Dlg1–Braf translocations that generate overexpressed and/or constitutively active protein kinases were also discovered as oncogenic drivers. Point mutations of known oncogenes were rare; however, we identified tumors with activating mutations in Kras and Hras. Our results also showed that the majority of these genetic aberrations have in common the ability to enhance MAPK and PI3K pathway signaling. Next, we asked to what extent these pathways are activated by genetic alterations known to occur in human breast cancer. We searched the TCGA breast cancer database for KRAS, BRAF, and RAF1 mutations and amplifications, RTK gene mutations, amplifications, and overexpressions, YAP1 amplification/gain/overexpression, PIK3CA mutations and amplifications, as well as PTEN and INPP4B deletions, all of which can potentially activate the MAPK and/or PI3K pathways. More than 60% of breast tumors bear at least one of these aberrations, whereas over 90% of the TNBC subgroup of breast cancers exhibit these genetic lesions, suggesting the importance of MAPK and PI3K signaling pathway activation in TNBC tumorigenesis.
Through combined WES and RNA-seq analyses, we identified potential oncogenic drivers in approximately 50% of tumors against which first-line drugs are also available (Supplementary Table S5). It is worth noting that almost all of these genetic driver fusions would not be detected by conventional targeted sequencing approaches that only investigate mutational events in panels of known oncogenes. The mutational burden of murine tumors is relatively low, averaging 30 somatic mutations per tumor with 17 nonsynonymous mutations. These numbers are comparatively lower than those of human TNBC, which averages 127 somatic mutations per tumor with 107 being nonsynonymous (TCGA data analysis, and ref. 9), reflecting cross-species/age differences. Nevertheless, we do observe similar mutational patterns between Brca1-deficient murine tumors and BRCA1-mutant human breast tumors (Fig. 1).
We found recurrent Met and Yap amplification in 20% and 22% of primary mouse tumors investigated, as well as individual tumors showing increased mRNA levels without chromosomal amplification. Overall, approximately 25% of the primary tumors show amplification/gain/overexpression of Met or Yap (Fig. 1; Supplementary Fig. S1). Our data are consistent with human TCGA data, where 27% or 29% of patients with TNBC show amplification/gain/overexpression of MET or YAP, supporting the importance of these genetic events in TNBC development (21, 22, 28). Current data also point to the role of YAP1 in activation of the PI3K/MAPK pathway via increased PIK3CB, HRAS, or GAB2 expression, or increased IRS phosphorylation in diverse types of tumors (49–52). We also observed recurrent Myc amplification and Rb1 deletion in the mouse tumors, although both tend to occur in broader chromosomal regions. These data are consistent with a report by Holstege and colleagues in mouse tumors (20), and also cBioPortal analyses of human TNBC (http://www.cbioportal.org/index.do?session_id=59edd065498e5df2e29710eb&show_samples=false&). In addition to these recurrent CNAs, sporadic amplifications and deletions that affect cancer-related genes were identified, including high amplification of Fgfr2 and Egfr loci and deep deletion of the Pten locus, as well as shallow amplification of the Cdk6, Aurka, and Jun loci and deletion of the Nf1 loci (Supplementary Table S3). Our data suggest that although normal mouse and human chromosomes bear major differences in their architectures, mouse tumors developed over 7 to 12 months exhibited similar changes in their cognate regions to human tumor genomes.
In addition to chromosomal amplifications and deletions at known oncogenic loci, chromosomal translocations were found to be the oncogenic drivers in a subset of the murine tumors. Interestingly, although mutational burden is low in human breast cancer, ranking 20th among 30 different cancer types (19), the chromosomal rearrangement frequency ranks among the highest (Supplementary Fig. S7F and data generated on the basis of the “TCGA Fusion Gene Data Portal,” http://54.84.12.177/PanCanFusV2/), although the frequency of recurrent translocations is low. Traditionally viewed as important oncogenic drivers for hematologic cancers (53), chromosomal rearrangement events have been identified and increasingly recognized as potent drivers in human solid tumors (34, 35, 37, 54–56). Through next-generation sequencing, a number of gene fusions with low frequency were identified in patients with breast cancer, including FGFR2, FGFR3, BRAF, NTRK3, and MET (33, 34, 37, 55–57). We also identified chromosomal translocations involving Fgfr2, Braf, and Raf1 that resulted in constitutively active fusion kinases in the mouse tumors. We noticed that the fusion partners in the mouse tumors are different from those found in the human counterparts. However, it has been shown that these kinases tend to fuse with different partners in human tumors, while maintaining similar domain architecture. Typically, fusion events involving FGFR2 and FGFR3 occur at the C-terminus, replacing the C-terminus of FGFRs with different fusion partners that provide the ability of the fused protein to undergo spontaneous homo-multimerization (Supplementary Fig. S7G; refs. 34, 47, 56). Our data confirmed that the fusion partner found in mouse tumors had similar functions (Fig. 4C). Similarly, it is commonly observed that a diverse set of genes can fuse with RAF family members, generating fusion kinases where the N-terminal regulatory domain of RAF is replaced by the fusion partner, resulting in constitutive activation of the RAF family kinases (Supplementary Fig. S7H; refs. 35–38, 55).
Further evidence for the oncogenic function of a genetic alteration is provided when such alterations confer sensitivity to targeted therapy that results in an objective response only in patients harboring the genetic alteration. This, in fact, represents the litmus test for the precision medicine approach in cancer treatment. We carried out individualized treatment studies based on the specific driver(s) present in a particular tumor. This allowed us to confirm that these predicted drivers indeed are important for tumor growth in vivo. We show that two tumors with FGFR2 fusions responded to the FGFR inhibitor BGJ398, a tumor with a RAF1 fusion responded to the MEK inhibitor trametinib, and a tumor with overexpressed MET responded to the MET inhibitor crizotinib, verifying that these spontaneous genetic aberrations were driving tumor growth in each of the individual tumors. Importantly, the tumor with an FGFR2–DNM3 fusion did not respond to MET inhibition, nor was the FGFR2 inhibitor BGJ398 effective on the MET-overexpressing tumor, highlighting the importance of matching therapeutic drugs with oncogenic drivers.
Our study indicates that although the evolutionary courses that lead to breast tumor formation in the context of Trp53 deletion, with or without Brca1 deletion, are diverse, they have in common the ability to activate the MAPK and/or PI3K pathway. However, targeting both PI3K and MAPK pathways resulted in high toxicity and is not a viable option for patients. Our study suggests approved drugs or experimental drugs could be effective therapies when targeting upstream driver events, which can be identified by combined WES and RNA-seq efforts.
One such upstream driver event is FGFR activation. The importance of the FGFR pathway in tumorigenesis has increasingly been recognized (58, 59), and FGFR fusions have been found in bladder, thyroid, prostate, and lung cancers, glioblastoma, and cholangiocarcinomas (Supplementary Fig. S7H; refs. 34, 47, 56). We demonstrated that mouse FGFR2–DNM3 and human FGFR2–CCDC6 fusion kinases share similar oncogenic potential and activation mechanism. Our result that the FGFR inhibitor BGJ398 (60) is effective in treating murine tumors harboring FGFR2 fusions is also consistent with recent data obtained in a cholangiocarcinomas PDX model harboring FGFR2–CCDC6 fusion treated with BGJ398 (61). Together, these results support ongoing clinical trials targeting the FGFR in various cancers (https://clinicaltrials.gov), some of which have shown therapeutic benefits. A phase II trial evaluating the efficacy of the TKI258 FGFR inhibitor in patients with or without FGFR2 mutations showed increased overall survival when mutations were detected (20.2 months with mutations vs. 9.3 months without; Clinical Trial # NCT01379534). A recently published basket trial targeting patients with potential FGFR pathway activations also found BGJ398 provided disease control in 37% of patients (49/132 patients) who had failed all available standard therapies, and for patients with breast cancer with FGFR1/2 amplification or mutation, BGJ398 alone resulted in stable disease in 31% (10/31) patients, although the disease eventually progressed on this treatment (62). These data support the potential of FGFR inhibitors in cancer treatment, but also highlight their limitation as monotherapy in long-term disease management.
Similar to available human trials, we found that although tumors harboring FGFR2 fusions responded well initially to BGJ398 monotherapy, this response is not sustained and tumors relapsed despite continuous treatment (62). Given our success with the FGFR inhibitor BGJ398 as part of combination therapy with PARP inhibitor in treating BRCA1-deficient tumors, or with BKM120 in treating BRCA1-WT tumors, we suggest that including BGJ398 as part of combination treatment regimens warrants further investigation in patients with FGFR family genetic aberrations and either germline or sporadic defects in DNA repair pathways.
Fusion kinases involving the RAF family members (BRAF and RAF1) have been found with high frequency in pilocytic astrocytomas (63, 64) and at low frequency in melanoma (65, 66) and pancreatic (36), prostate (38), thyroid (67), and metastatic breast cancers (35, 68). We reasoned that although RAF1 fusions have not been found in human TNBC, DHX9–RAF1 identified in the murine models shares the same architecture as fusion RAF1s discovered in other tumor types. Therefore, it is important to consider and evaluate fusion kinases involving the RAF family members as therapeutic targets in vivo and identify effective treatments for these fusion-containing tumors in breast cancer and beyond. We report here that a TNBC tumor with RAF1 fusion kinase was responsive to trametinib as a single agent, and note that as this tumor was BRCA1-deficient, the efficacy of trametinib was improved when combined with olaparib. Our data suggest the potential of treating RAF1 fusion–containing tumors with MEK inhibitors.
Given that Met overexpression synergizes with Trp53 loss to induce basal-like breast cancer (typically TNBCs; refs. 21, 22), and Met amplification and overexpression correlate with TNBC, it is encouraging that the murine tumor with MET overexpression responded to the MET inhibitor crizotinib either as monotherapy or in combination with the PI3K inhibitor BKM120. This result is consistent with reports that targeting MET in MET-amplified TNBC mouse tumors results in an initial CR (69). Combining MET and PARP inhibitors has been shown to suppress olaparib-resistant TNBC tumor cell growth in vivo (70).
Another potential therapeutic target in TNBC that emerged from this study is YAP1, a transcription factor that is normally suppressed by the Hippo tumor suppressor pathway. YAP1 collaborates with other transcription factors, including TEAD and AP1, to promote tumorigenesis and metastasis (27, 71, 72). In fact, verteporfin, which inhibits YAP1 and TEAD interaction and thus transcriptional activity, has been used in vitro and in vivo (51, 73). It will be interesting to investigate the efficacy of this approach in our TNBC mouse model.
In summary, we have generated mouse models that recapitulate human TNBC. Through combined WES and RNA-seq, we identified amplifications and fusions of oncogenic drivers that would have been missed by evaluating mutations in panels of known oncogenes. We investigated the efficacy of this precision medicine approach by treating each tumor as an individual patient using first-line drugs targeting specific oncogenic drivers. Our results indicate the importance of performing RNA-seq and WES on TNBC to identify the genetic aberrations driving the disease. These data also underscore the importance of combination therapy to elicit prolonged treatment efficacy. This work provides proof-of-principle evidence for the ongoing basket trials NCI-MATCH (NCT02465060), My Pathway Trial (NCT02091141), and the European ESMART (NCT02813135), which are designed to match the specific abnormalities with therapeutic drugs.
Methods
Endogenous Tumor Generation
All animal studies were reviewed and approved by the Institutional Animal Care and Use Committee at the Beth Israel Deaconess Medical Center (Boston, MA). K14cre; Brca1flox/flox; Trp53flox/flox mice were obtained from Dr. Jos Jonkers's laboratory (Netherlands Cancer Institute, Amsterdam, the Netherlands).
Orthotopic Tumor Implantation
Tumor pieces were cut to 2 mm in diameter and inserted into the fourth mammary fat pad of 8-week-old recipient mice via a 0.5-cm incision in the skin, and the skin was closed with VetBond as described previously (45).
Tumor Treatment and Tumor Measurement
Once tumors reached 8 mm in diameter as measured by electrical caliper (Thermo Fisher Scientific), mice were treated with indicated drugs obtained from MedChemExpress, LLC. For oral gavage, 100 μL of drug suspension was administrated daily for 6 consecutive days, followed by one drug holiday. Tumor sizes were measured twice a week (length and width), and tumor volume was calculated as (3.14 × length × width × width/6).
RNA and Library Preparation
Total RNA was prepared following the protocol for the Promega ReliaPrep RNA Tissue Miniprep System (Z6111), and RNA integrity and concentration were measured using the Agilent 2100 Bioanalyzer (Agilent Technologies). cDNA libraries were prepared from 15 to 35 ng RNA starting material (RIN values >6.0), using the TruSeq RNA Sample Preparation Kit (Illumina) according to the manufacturer's instructions, and quality was checked on an Agilent 2100 Bioanalyzer (Agilent Technologies). Sequencing was carried out on the HiSeq 2500 (Illumina) using paired-end clustering and 51 × 2 cycles sequencing.
Genomic DNA and Library Preparation
Genomic DNA from tumor or liver samples was prepared following the protocol for the Promega ReliaPrep Tissue DNA Miniprep System (A2051). SureSelect or NimbleGen Mouse exome capture kits were used to generate DNA library according to the manufacturer's instructions. Sequencing was carried out using HiSeq4000 (Illumina) using paired-end clustering and 51 × 2 cycles sequencing. The NCBI BioProject accession number for both RNA-seq and WES sequences reported in this article is PRJNA398328.
Sequence Analysis
The quality of the raw FASTQ files was checked with FastQC. RNA-seq reads were mapped to GRCm38 using STAR and expression levels (FPKM) quantified using Cufflinks with default parameters. WES reads were mapped to GRCm38 using BWA, then deduplicated, realigned around indels, and base recalibrated. We classified the mouse tumors into the breast cancer–intrinsic subtypes using AIMs and PAM50 classifiers after conversion of mouse gene symbols to human Entrez gene IDs and human gene symbols, respectively. PAM50 centroids were recomputed using RNA-seq expression profiles from TCGA breast cancer data and their associated, published PAM50 classification as gold standard. Gene fusions were called using FusionCatcher and filtered for false positives.
Mutation and Copy-Number Analysis
Somatic mutations were identified upon removing any mutations found in any tail, liver, or normal mammary control samples, in mouse dbSNP, or with insufficient coverage in the control samples. Mutations were annotated with SnpEff. Copy-number variants were called using CNVkit after removing low-quality reads. Sample-specific thresholds were computed to call amplifications and deletions.
Cell Culture
Primary HMECs were isolated and immortalized on passage 3 as described (74) upon informed consent (DFHCC-IRB legacy 04-405). Cells were maintained in MEGM medium (Lonza) and used within 10 passages. NIH3T3 cells were obtained from and authenticated by ATCC in 2014 and cultured in DMEM with 10% FCS and were used within 10 passages. Retrovirus and lentivirus preparation and infection were carried out as described previously (75) using Lipofectamine 2000 (Thermo Fisher Scientific).
Protein Lysis, Western Blots, Immunoprecipitation, Silver Staining, and Mass Spectrometry
Tumors and cells were lysed in RIPA lysis buffer with protease inhibitors and phosphatase inhibitors (Sigma), and protein quantifications were performed using BCA Protein Assay (Pierce). An equal amount of total proteins were used for Western blots or immunoprecipitation, gels were stained using Pierce Silver Staining Kit (#24612), and mass spectrometry was carried out as described previously (76). See Supplementary Materials and Methods for detailed antibody information.
Cell Proliferation and Soft-Agar Formation Assay
For cell proliferation assay, 1,500 cells were plated in 96-well plates and measured with CellTiter-Glo (Promega, G7572). For soft-agar colony assay, 10,000 cells were resuspended in 0.4% agar (SeaPlaque low melting agar, Lonza) in cell growth medium and plated on 0.8% agar in 6-well plates. Medium was changed every week, and after 3 weeks, cells were fixed and stained in 0.005% Crystal Violet in 20% methanol in PBS.
Disclosure of Potential Conflicts of Interest
L.C. Cantley is a founder and member of the BOD of Agios Pharmaceuticals and is a founder of and receives research support from Petra Pharmaceuticals. These companies are developing novel therapies for cancer. No potential conflicts of interest were disclosed by the other authors.
One of the Editors-in-Chief is an author on this article. In keeping with the AACR's editorial policy, the peer review of this submission was managed by a senior member of Cancer Discovery's editorial team; a member of the AACR Publications Committee rendered the final decision concerning acceptability.
Authors' Contributions
Conception and design: H. Liu, C.J. Murphy, G.M. Wulf, L.C. Cantley
Development of methodology: H. Liu, C.J. Murphy, F.A. Karreth, G.M. Wulf
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Liu, F.A. Karreth, K.B. Emdal, K. Yang, G.M. Wulf
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Liu, C.J. Murphy, K.B. Emdal, F.M. White, O. Elemento, L.C. Cantley
Writing, review, and/or revision of the manuscript: H. Liu, C.J. Murphy, F.A. Karreth, O. Elemento, A. Toker, G.M. Wulf, L.C. Cantley
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Liu, C.J. Murphy, K. Yang
Study supervision: H. Liu, F.M. White, O. Elemento, A. Toker, G.M. Wulf, L.C. Cantley
Acknowledgments
We thank Drs. Jos Jonkers, Takehiko Sasaki, and Michael Begley for providing critical reagents, members of the Cantley, Wulf, and Toker laboratories for suggestions, and Dr. Xiang and her team at the Genomics and Epigenomics Core Facility of Weill Cornell Medical College for next-generation sequencing. This work was supported by NIH grants R35 CA197588 (to L.C. Cantley), R01 GM041890 (to L.C. Cantley), PSOC U54 CA210184 (to L.C. Cantley), U24 CA210989 (to O. Elemento), R01 CA194547 (to O. Elemento), P50CA211024 (to O. Elemento), 5P50 CA168504-03 (SPORE; to G.M. Wulf); the Breast Cancer Research Foundation awards BCRF-16-021 (to L.C. Cantley) and BCRF-17-174 (to G.M. Wulf); the Jon and Mindy Gray Foundation (to L.C. Cantley); a Susan Komen postdoctoral fellowship (to H. Liu); the Ludwig Center at Harvard (to G.M. Wulf and A. Toker); The Mary Kay Ash Foundation (to G.M. Wulf); The Breast Cancer Alliance (to G.M. Wulf and L.C. Cantley); DFCI Men's Initiative to Cure Women's Cancers (to G.M. Wulf); and a Novo Nordisk STAR postdoctoral fellowship (to K.B. Emdal). Research supported by a Stand Up To Cancer Colorectal Cancer Dream Team Translational Research Grant (grant number: SU2C-AACR-DT22-17). Research grants are administered by the American Association for Cancer Research, the scientific partner of SU2C (to L.C. Cantley).