Discovery of Biomarkers Predictive of GSI Response in Triple-Negative Breast Cancer and Adenoid Cystic Carcinoma

The Notch signaling pathway is an evolutionarily conserved regulator of cell fate, differentiation, and growth. In mammals, Notch signaling is mediated by four Notch receptors (NOTCH1–4) and at least four functional ligands [Delta-like-1 (DLL1), DLL3, DLL4, JAG1, and JAG2]. Canonical Notch signaling is initiated by ligand binding to the Notch ectodomain. This triggers a series of proteolytic cleavage events, culminating in the release of the Notch intracellular domain (NICD) by gamma-secretase (GS). Upon GS cleavage, NICD translocates to the nucleus, where it forms a Notch transcription complex with the DNA-binding factor CSL (also known as RBPJ) and coactivators of the Mastermind-like (MAML) family (for review, see ref. 1).

Deregulated Notch signaling is oncogenic in specific cell types; for example, it is strongly associated with T-cell acute lymphoblastic leukemia (T-ALL), in which somatic activating mutations in NOTCH1 are present in >50% of cases (2). Most NOTCH1 mutations in human T-ALL fall into two classes: (i) in-frame mutations or indels in exons 25 to 28 that disrupt an extracellular juxtamembrane negative regulatory region (NRR), leading to ligand-independent receptor proteolysis and release of the NOTCH1 intracellular domain (N1-ICD); and (ii) stop codons or frameshift mutations in exon 34 that result in deletion of a C-terminal PEST degron domain, stabilizing N1-ICD. Less commonly in human T-ALL, NOTCH1 is the target of (7;9) translocations in which the 3′ end of NOTCH1 is fused to promoter/enhancer elements of TCRB (3). The rearranged NOTCH1 alleles in tumors with the t(7;9) drive expression of truncated mRNAs that initiate translation from a conserved methionine lying within the NOTCH1 transmembrane domain (4).

Oncogenic Notch signaling is also implicated in breast cancer. Recently, RNA-sequencing (RNA-seq) was used to identify abnormal NOTCH mRNAs in human breast carcinoma cell lines and primary tumors (5). The aberrant transcripts resulted from cytogenetically silent deletions involving either NOTCH1 or NOTCH2; although several of the rearrangements produce fusion genes, none encode chimeric proteins. Instead, the rearranged NOTCH1 genes drive expression of truncated mRNAs that initiate translation from the same internal methionine implicated in human T-ALLs with the t(7;9), whereas the rearranged NOTCH2 genes drive expression of truncated mRNAs that initiate translation from a methionine residue located within the intracellular domain of NOTCH2, internal to the GS cleavage site. Because of this distinction, NOTCH1-rearranged breast cancers are predicted to be GS inhibitor (GSI) sensitive, whereas NOTCH2-rearranged breast cancers are predicted to be GSI resistant.

In addition to its oncogenic roles, genetic evidence suggests that Notch is a tumor suppressor in human squamous cell carcinoma (SCC). Whole-exome deep sequencing identified likely loss-of-function mutations in at least one Notch signaling component in roughly 15% to 20% of head and neck SCC (6, 7); similarly, at least one putative loss-of-function mutation involving either NOTCH1 or NOTCH2 was identified in 19 of 26 primary cutaneous SCC or derived cell lines (8). In addition, one trial of a GSI in patients with Alzheimer disease reportedly led to an increase in skin cancers (9).

Despite concerns about the complications of long-term GSI treatment, preclinical studies in animals and clinical trials in patients with cancer suggest that intermittent treatment with GSIs is well tolerated, and GSIs continue to hold promise as targeted therapy for malignancies in which Notch is an oncogenic driver (9). However, clinical responses to GSIs have been modest, possibly because GSI trials to date have not used biomarkers that predict responsiveness as a criterion for enrollment. To address the need for biomarkers, we first screened large collections of cell lines, primary tumors, and metastases for NOTCH gene mutations, reasoning that tumors with gain-of-function mutations are most likely to be sensitive to GSI. Our studies show that triple-negative breast cancers (TNBC) are uniquely enriched among tumors screened for activating NOTCH1 and NOTCH2 deletions. Using xenograft models, we demonstrate that the GSI sensitivity of NOTCH1-rearranged breast cancer cell lines correlates with N1-ICD levels, and that NOTCH2-rearranged tumors are indeed GSI resistant. We also identify other human solid tumors with novel activating point mutations in the NOTCH1 NRR, including a subset of adenoid cystic carcinoma (ACC) cell lines and primary tumors. Like NOTCH1-rearranged breast cancers, xenografted NOTCH1-mutated ACCs are GSI sensitive, whereas Notch wild-type (WT) xenografts are not. Finally, we identify HES4, a known Notch target gene, as a gene whose expression is correlated with poor clinical outcome in TNBC. Our findings suggest that assessment of NOTCH gene mutational status, activated Notch protein levels, and expression levels of particular target genes, such as HES4 in TNBC, will be useful in selecting patients for GSI trials.

To detect NOTCH gene rearrangements, we used targeted exome sequencing (TES) data from human cancer cell lines and primary tumors to identify imbalances in exon coverage, which can be used to infer the presence of intragenic rearrangements, particularly deletions. Analysis of TES data from 608 cancer cell lines identified exon imbalances in NOTCH1 or NOTCH2 in five cell lines and in one cell line, respectively (Fig. 1A and Supplementary Fig. S1A). Exon imbalances were found only in TNBC lines and stemmed from the deletion of exons encoding the Notch extracellular EGF repeats and NRR. For example, in the MB-157 and MDA-MB157 cell lines, the read coverage of 5′ NOTCH1 exons was markedly lower than that of 3′ NOTCH1 exons (Fig. 1A and 1B). A similar imbalance was observed in the read coverage for 5′ and 3′ NOTCH2 exons in the HCC1187 cell line (Supplementary Fig. S1A). Our data are consistent with previously reported NOTCH gene breakpoints in the HCC1187 and HCC2218 cell lines, which were originally suspected based on RNA sequencing data (5), and indicate that exon read imbalances in TES data can be used to identify tumors with NOTCH1 and NOTCH2 rearrangements. This capacity was confirmed by the detection of novel NOTCH1 rearrangements in the MB-157 and MDA-MB-157 cell lines, both of which were derived from the same patient with TNBC. Importantly, a fusion transcript consisting of noncoding RNA derived from SEC16A, a gene flanking NOTCH1 on chromosome 9q34.3, and exons 27 to 34 of NOTCH1 was identified by sequencing of RNA prepared from the MB-157 cell line (data not shown), consistent with the presence of an interstitial deletion that created a SEC16A–NOTCH1 fusion gene. Similarly, sequencing of RNA prepared from the HCC1599 cell line confirmed the presence of an aberrant transcript consisting of exon 2 joined out of frame to exon 27 of NOTCH1, consistent with the presence of an intragenic deletion involving NOTCH1, as described elsewhere (5).

Next, we analyzed whole-exome sequencing data from 66 triple-negative primary breast tumors. We identified NOTCH1 or NOTCH2 5′ deletions in six tumors, as indicated by significantly decreased read coverage (P < 10⁻⁵) in 5′ exons as compared with 3′ exons (Fig. 1C and 1D; Supplementary Fig. S1B). An excess of transcripts containing 3′ NOTCH exons was confirmed by RT-qPCR (data not shown). Each of the inferred NOTCH1 deletions is predicted to create mutated alleles driving the expression of truncated transcripts encoding membrane-tethered NOTCH1 polypeptides that depend on GS for activity (Supplementary Fig. S1C; refs. 4 and 5). In contrast, as with previously reported NOTCH2 deletions (5), the rearranged NOTCH2 alleles are predicted to drive the expression of truncated transcripts encoding constitutively active, GS-independent NOTCH2 polypeptides (Supplementary Fig. S1C).

TNBC is a molecularly heterogeneous tumor that can be classified on the basis of patterns of gene expression into basal-like, immune, mesenchymal stem cell–like, and luminal androgen receptor subtypes (10). Consistent with previous reports, we observed the highest level of expression of NOTCH1 mRNA in the basal-like tumors (P < 0.001; data not shown). Next, we examined whether the presence of NOTCH1 or NOTCH2 rearrangements correlates with a particular subtype of TNBC. Of the 66 tumors analyzed, NOTCH gene rearrangements were observed in the basal and luminal androgen receptor subtypes (see Supplementary Table S1), but not in the immune or mesenchymal stem cell–like subtypes.

To determine the relationship between NOTCH gene mutational status and sensitivity to GSI, we evaluated the antiproliferative activity of MRK-003, a potent and selective GSI (11), in a panel of breast cancer cell lines. T-ALL and mantle cell lymphoma cell lines with mutations that disrupt the NOTCH1 NRR (2, 4) and TALL1, an additional GSI-sensitive T-ALL line with an activating NRR mutation in NOTCH3 (J.C. Aster, unpublished data), were included as positive controls. The IC₅₀ for MRK-003 in the TNBC cell lines HCC1599 and MB157 with NOTCH1 rearrangements was <1 μmol/L, similar to the sensitivity of lymphoid cell lines with NOTCH1 or NOTCH3 point mutations or gene rearrangements that disrupt the NRR coding region (Fig. 2A and Supplementary Fig. S2A). In contrast, the HER2⁺ breast cancer cell line HCC2218, which also harbors a SEC16A–NOTCH1 fusion gene (5), demonstrated intermediate sensitivity (Fig. 2A and Supplementary Fig. S2A). As a group, breast cancer cell lines with NOTCH gene rearrangements that disrupt the NRR coding region were significantly more sensitive to MRK-003 than cell lines with WT NOTCH alleles, with two exceptions, the HCC1187 and MDA-MB-157 cell lines. HCC1187, a breast cancer cell line harboring a SEC22–NOTCH2 fusion gene (5), was predicted to be GSI resistant because the translation start site in aberrant NOTCH2 transcripts is internal to the site of GS cleavage (Supplementary Fig. S1C). In contrast, the resistance of the cell line MDA-MB-157 (Supplementary Fig. S2A) was unexpected, as this line has a deletion that disrupts the NOTCH1 NRR coding region. We also noted that the mantle cell lymphoma cell lines (Mino and MT3), which harbor NOTCH1 mutations leading to deletion of the C-terminal PEST domain, were also resistant to MRK-003. Thus, GSI sensitivity is confined to cell lines with mutations that disrupt the NRR, but this association is not absolute.

To assess the relationship between NOTCH1 mutational status, Notch signal strength, and GSI sensitivity, we evaluated N1-ICD protein levels. We observed high levels of N1-ICD in most cell lines with NOTCH1 gene rearrangements or mutations involving the NRR (Fig. 2B). Of interest, the exception to this rule is the GSI-resistant MDA-MB157 cell line, which had low levels of N1-ICD despite the presence of a NOTCH1 rearrangement (Fig. 2B). As expected, N1-ICD levels were low or absent in GSI-resistant NOTCH1 WT cell lines, as well as in GSI-resistant mantle cell lymphoma lines (MT3 and Mino), which harbor only PEST domain mutations (Fig. 2B). This is consistent with studies showing that although PEST domain mutations stabilize NICD, they have no effect on NICD levels in the absence of NRR mutations or exposure to ligands (2). As a consequence, N1-ICD protein level was more highly correlated with GSI sensitivity than NOTCH gene mutational status (Supplementary Fig. S2B and S2C).

We also compared the protein levels of all four Notch receptors in the breast cell line panel (n = 8; Supplementary Fig. S2D) using antibodies that recognize full-length Notch receptors and their furin-cleaved transmembrane subunits. We observed variable levels of all four receptors, expression of which did not correlate with sensitivity to MRK-003. For example, high NOTCH1 polypeptide levels were observed in both GSI-sensitive HCC1599 cells and GSI-resistant HCC70 and SKBR3 cells. Similarly, NOTCH2–4 polypeptide levels did not correlate with sensitivity to MRK-003 in this breast cancer cell line panel. These results suggest that N1-ICD level is more highly associated with sensitivity to MRK-003 treatment than expression of Notch receptors per se, with the important caveat that reagents to identify other forms of activated Notch receptors (N2-4-ICD) are not yet available.

Although NICD would appear to be a promising biomarker, because sensitive and specific antibodies only exist for detection of N1-ICD, additional markers of Notch activation in diverse cellular lineages are needed. To identify Notch target genes in Notch-addicted cancer cell lines in an unbiased fashion, we performed RNA-seq on CUTLL1 T-ALL cells, REC-1 mantle cell lymphoma cells, and HCC1599 cells (Gene Bank accession number GSE59810), all of which have mutations that disrupt the NOTCH1 NRR. RNA-seq was performed in the Notch-off, GSI-treated state, and 4 hours following washout of GSI, a strategy that rapidly generates N1-ICD and upregulates NOTCH1 target genes in a temporally controlled fashion. Notably, although >80 genes were upregulated upon GSI washout in each cell line (FDR <0.05), only 6 target genes were common to all three cell lines—HEY1, MYC, NOTCH3, TFRC, NRARP, and HES4 (Fig. 3A). Chromatin immunoprecipitation sequencing (ChIP-Seq) analysis of CUTLL1 cells (12) showed that NOTCH1 and RBPJ bind to a site in the HES4 proximal promoter on chromosome 1 (Fig. 3B). To confirm that this site is highly dynamic, a feature that characterizes functional RBPJ–NOTCH1 binding sites (13), we performed local ChIP for RBPJ and NOTCH1 in REC-1 cells. This showed that GSI depleted NOTCH1 from the HES4 promoter, and that GSI washout resulted in rapid reloading of NOTCH1 onto this site (Fig. 3C). Finally, we confirmed that GSI markedly inhibited HES4 expression in HCC1599 cells, and that GSI washout rapidly restored expression (Fig. 3D).

To further study the relationship between activating NOTCH1 mutations and gene expression, we carried out Affymetrix-based microarray analysis in the panel of 608 cell lines (Gene Bank accession number GSE59242) screened previously for NOTCH gene mutations using TES. We observed that high HES4 expression was significantly correlated (FDR <0.05) with the presence of activating mutations involving NOTCH genes in several cancer cell lines (Supplementary Fig. S3 and Supplementary Table S2). We next compared expression levels of various Notch target genes among primary TNBCs, and noted that HES4 was the gene whose expression was most strongly correlated with NOTCH gene rearrangements in TNBC (Supplementary Table S3). High HES4 expression was also associated with poor patient outcome in TNBC (log-rank P < 0.0001), as patients with metastatic disease in the HES4-high group had a median survival of 0.89 years [n = 21; 95% confidence interval (CI), 0.31], whereas patients with metastatic disease in the HES4-low group had a median survival of 2.97 years (n = 139; 95% CI, 2.18; Supplementary Fig. S4). Similarly, in patients without measurable metastatic disease, high HES4 expression in primary tumors also predicted a poor prognosis (Supplementary Fig. S4). These results suggest that HES4 expression levels (and by extension, activated Notch levels) may be a useful prognostic biomarker in patients with TNBC.

To investigate how GSI affects cell growth in NOTCH gene–rearranged breast cancer models, we evaluated the effect of GSI treatment on molecular pathways regulated by Notch. MRK-003 treatment resulted in a dose-dependent decrease in the levels of activated Notch as measured by N1-ICD in the NOTCH1-rearranged TNBC cell lines MB-157 and HCC1599 (Fig. 4A). NOTCH1 inhibition was accompanied by decreases in MYC protein and RB phosphorylation and increases in the level of the cyclin-dependent kinase (CDK) inhibitor p21 (Fig. 4A). In addition, compensatory activation of the MAPK–PI3K pathway was observed, as indicated by GSI dose–dependent increases in phosphorylation of ERK and PRAS40, an AKT substrate (Fig. 4A). These results suggest the existence of a compensatory feedback loop between Notch and the PI3K–MAPK signaling pathways in Notch-dependent breast cancer cell lines. In contrast, MRK-003 treatment did not significantly affect any of these signaling events in the Notch WT cell line HCC1143.

To investigate possible synergistic antitumor effects of Notch and PI3K–MAPK pathway coinhibition, we performed cell proliferation assays in the presence of MRK-003 and a selective ERK inhibitor (SCH772984; ref. 14). MRK-003 treatment produced dose-dependent growth inhibition of both MB-157 and HCC1599, but ERK inhibition failed to potentiate the effect of MRK-003 (see Supplementary Fig. S5). Although these in vitro studies were negative, given the evidence of cross-talk between Notch and MAPK–PI3K signaling in breast cancer cells shown here and in other tumor cell types such as T-ALL cells (15), further evaluation of this combination in breast cancer may be warranted.

Next, we evaluated the effects of GSI treatment of NOTCH gene–rearranged breast cancer cell lines on cell-cycle progression and apoptosis. MRK-003 treatment resulted in a G₁ cell-cycle arrest and the appearance of a sub-G₁ population, suggestive of apoptotic cell death, in MB-157 cells (Fig. 4B). Similar effects were also observed in the HCC1599 cells (data not shown). Furthermore, MRK-003 treatment resulted in a dose-dependent increase in cleaved PARP in HCC1599 cells (Fig. 4C), whereas no significant change in cleaved PARP was observed in the NOTCH2-rearranged, GSI-resistant cell line HCC1187 (5). In the NOTCH1-rearranged MB-157 cell line, MRK-003 produced only a modest increase in apoptosis (data not shown), suggesting that GSI-mediated growth inhibition may stem from a different mechanism in this line. Consistent with this idea, we noted that prolonged MRK-003 treatment of MB-157 cells resulted in the induction of cellular senescence, as measured by β-galactosidase activity (Fig. 4D). Furthermore, robust GSI-mediated upregulation of p21 and downregulation of MYC, events implicated in induction of cellular senescence (16), were also observed in MB-157 cells (Fig. 4A).

The Notch pathway has also been suggested to be important in the maintenance of stem-like cells with tumor-initiating activity (17) and to promote the growth of breast epithelial cells as mammospheres in culture, a phenotype linked to stem-like properties (5, 18). We evaluated the effect of MRK-003 treatment on stem-like cells using the markers CD44/CD24 and aldehyde dehydrogenase (ALDH). When cultured in stem cell media, MB-157 cells contain a subpopulation (∼8%) of CD44⁺/CD24^low stem-like cells that is depleted by MRK-003 treatment (Supplementary Fig. S6A). Similarly, MRK-003 treatment of MB-157 cells also suppressed the ALDH⁺ cell population (Supplementary Fig. S6B), another marker of cells with stem-like properties. Unlike the effects of MRK-003 on growth, which were restricted to breast cancer lines with NOTCH1 rearrangements, MRK-003 treatment also suppressed the ALDH⁺ cell population in NCIH226 and MFM223 cells (Supplementary Fig. S6C), which have WT NOTCH alleles. Thus, NOTCH gene rearrangements do not correlate with the effect of MRK-003 on breast cancer stem–like cells, possibly because the dose of Notch that is required to maintain stem-like cells is substantially lower than that required to drive cell growth. Taken together, Notch signaling in NOTCH1-rearranged TNBC cell lines appears to support cell growth through multiple mechanisms, including promotion of cell-cycle progression, decreased apoptosis, and cellular senescence, and increasing the fraction of stem-like cells.

We next evaluated the effect of MRK-003 therapy on the growth of breast cancer xenograft models with NOTCH1 or NOTCH2 rearrangements (Table 1). MDA-MB-231, a TNBC cell line with WT NOTCH alleles, was used as a control model. To identify new primary tumor–derived xenograft models with NOTCH1 rearrangements, we used RT-PCR to screen for previously described NOTCH1 rearrangements (5). Of the 21 primary human tumor–derived TNBC xenograft models screened, two models, HBCx8 and HBCx14, yielded RT-PCR products consistent with the presence of NOTCH1 rearrangements. In HBCx8 cells, we observed a SEC16A–NOTCH1 fusion transcript identical to the previously described rearrangements (data not shown). In HBCx14 cells, we were unable to detect RT-PCR products containing both exons 26 and 28 of NOTCH1 (data not shown); although this was consistent with the presence of a NOTCH1 rearrangement, we were unable to identify fusion junctions corresponding to previously described NOTCH1 rearrangements, suggesting that this line harbors a novel NOTCH1 rearrangement.

To determine whether NOTCH1 rearrangements correlate with NOTCH1 activation in vivo, we evaluated N1-ICD levels using a sensitive immunohistochemical staining method that is specific for N1-ICD (19). High levels of N1-ICD staining were detected in the HCC1599 and MB-157 models (Supplementary Fig. S7A), a result that correlated well with the results of Western blotting for N1-ICD (Supplementary Fig. S7B). However, HBCx14 had lower levels of N1-ICD staining, and even less staining for N1-ICD levels was seen in the HBCx8 model, despite the presence of a SEC16A–NOTCH1 fusion gene. As expected, models with WT NOTCH1 alleles were negative for N1-ICD staining. Next, we evaluated the sensitivity of these xenograft models to GSI treatment, given either alone or in combination with paclitaxel in vivo. We noted that sensitivity to MRK-003, alone or in combination with paclitaxel, was associated with N1-ICD levels; these results are summarized in Table 1. For example, MRK-003 treatment produced dose-dependent growth cessation or regression of HCC1599 and MB-157 xenografts (Fig. 5A and Table 1), but had only modest antitumor activity in the MDA-MB-231 xenograft model with WT NOTCH genes (Supplementary Fig. S7C and Table 1). Although previous studies have demonstrated significant antitumor activity with MRK-003 in the MDA-MB-231 mode, we would like to note that the dosing scheme used in these studies was not identical to that of our study (20). In this study, we used a clinically tolerated dosing schedule (once a week dosing). Similarly, the HCC1187 xenograft model with a SEC-16–NOTCH2 fusion gene encoding NOTCH2 polypeptides that do not require GS cleavage for activation was resistant to GSI therapy (Fig. 5B). Monotherapy with MRK-003 had modest antitumor activity in the primary human tumor–derived xenografts HBCx8 and HBCx14 (Fig. 5 and Table 1), which have low N1-ICD levels (Supplementary Fig. S7A and S7B). These results indicate that N1-ICD levels can be substantially different in tumors with similar NOTCH1 rearrangements, and suggest that N1-ICD levels are more highly correlated with GSI responsiveness than NOTCH mutational status per se.

To further evaluate the association of the antitumor activity of MRK-003 with Notch pathway inhibition, we studied the effect of MRK-003 on Notch target gene expression in the HCC1599 xenograft model. Dose-dependent inhibition of multiple direct or indirect Notch target genes was observed in MRK-003–treated tumors (Fig. 5C), which also showed depletion of N1-ICD, as judged by immunohistochemistry (Fig. 5D). These data confirm that inhibition of the Notch pathway is correlated with the antitumor activity of MRK-003.

Multiple Notch targets may contribute to the antitumor activity of MRK-003 in vivo, but one that may be of special importance is MYC, which, as previously noted (Fig. 3), is one of the few common target genes across multiple types of Notch-“addicted” cancer cells in vitro. Treatment of HCC1599 xenografts with MRK-003 downregulated MYC protein levels, as judged by Western blotting and immunohistochemistry (Supplementary Fig. S7D), an alteration that was accompanied by decreases in cyclin E, which is involved in G₁–S phase progression. Regulation of MYC by Notch has been documented in T-ALL (21–24) and in murine models of breast cancer (25), and our results suggest that targeting of the Notch–Myc signaling axis also underlies the responsiveness of human breast cancer models to MRK-003.

We next evaluated the effect of combination therapy with GSI and paclitaxel, an agent that is frequently used to treat TNBC. Modest antitumor activity was observed with single-agent paclitaxel in the NOTCH1-rearranged breast tumor models. Although GSI monotherapy resulted in significant growth inhibition of HCC1599 or MB-157 cells, combination therapy with MRK-003 and paclitaxel significantly potentiated antitumor activity in both models and resulted in tumor regression in the MB-157 xenograft model (Fig. 5E). In the primary human tumor–derived xenograft models with NOTCH1 rearrangements, HBCx8 and HBCx14, monotherapy with MRK-003 or paclitaxel was ineffective in blocking tumor growth, but combination treatment showed significant antitumor activity (Fig. 5F and Table 1), whereas no potentiation was observed in a model with WT NOTCH alleles. These results suggest that GSI–paclitaxel combination therapy may be effective in treatment of NOTCH1-rearranged tumors.

TES of more than 4,000 tumors identified several novel NOTCH gene mutations. As described previously, Notch mutations can be activating or inactivating, depending on the location and nature of the substitution (26). To assess possible gain-of-function mutations, we focused on NOTCH1 mutations affecting the NRR region or the PEST domain (Supplementary Table S4). NOTCH gene mutations in the NRR region and PEST domain were restricted to a small fraction of tumors (<5%), and most of these mutations mapped to the core of the NRR (Fig. 6A), a region frequently involved by gain-of-function NOTCH1 mutations in human T-ALL (2). Tumor types in which recurrent Notch NRR and PEST domain mutations were identified are summarized in Supplementary Table S5. Of note, NRR or PEST domain mutations were identified in five of 105 TNBCs (Supplementary Table S5). NRR mutations were also observed in one of three ACCs, in line with recent sequencing studies that identified possible Notch gain-of-function mutations in a subset of ACCs (27). To test whether these newly identified NOTCH1 mutations affect function, they were scored for their ability to activate a Notch-responsive luciferase reporter gene (Fig. 6B). All mutations within the NOTCH1 NRR caused significant increases in MRK-003–sensitive luciferase activity, with amino acid substitutions at positions 1680, 1570, 1575, and 1683 having the greatest effect. In contrast, in accordance with previous studies, mutations in the PEST domain did not significantly increase luciferase reporter activity in the absence of ligand-mediated stimulation (data not shown).

To investigate the possible role of Notch in ACC, we sequenced NOTCH1 in six primary tumor–derived xenograft models (see Supplementary Table S6). This led to the identification of NOTCH1 mutations in two ACC models, both of which were tested for GSI sensitivity. MRK-003 treatment resulted in significant growth inhibition in the ACCx9 model (Fig. 6C) that harbors an I1680N NOTCH1 NRR substitution, a mutation known to cause ligand-independent NOTCH1 activation (28). In contrast, MRK-003 treatment did not have any antitumor effect on the POS-912 model (Fig. 6D), which harbors a NOTCH1 R365H point substitution in the EGF repeat region that is not expected to produce NOTCH1 gain-of-function. In line with these expectations, high levels of nuclear N1-ICD were observed in tumor cells in the ACCx9 model, whereas in the POS-912 model, only a minor subset of tumor cells had detectable N1-ICD levels (Fig. 6E). We also observed high levels of MYC positivity (>90% of tumor cells) in the ACCx9 model, whereas the POS-912 model showed much lower levels of MYC expression (see Supplementary Table S6), suggesting that MYC may be a target of Notch in NOTCH-mutated ACC. It should be noted, however, that high MYC expression was also observed in some ACCs with WT NOTCH1 genes, such as the ACCx6 model, suggesting that MYC may sometimes be dysregulated in ACC through mechanisms unrelated to NOTCH1 mutations or NOTCH1 activation. In other analyses, we did not detect an association between NOTCH mutational status and HES4 expression in ACC models (data not shown), emphasizing the need to validate biomarkers of Notch activation in each cellular context of interest, and suggesting that the most reliable biomarker of Notch activation is direct assessment of NICD levels. Further evaluation of N1-ICD in primary ACC is needed to determine the prevalence of Notch pathway activation in this neoplasm, which appears to be a candidate for treatment with anti-Notch therapies such as GSIs.

Aberrant Notch signaling has been implicated in numerous human diseases, including different types of cancers. Through unbiased sequencing of diverse cancer cell lines and primary tumors, we identified different types of activating Notch mutations in specific cancer subtypes. One striking finding is that deletions that remove the coding sequences of NOTCH1 and NOTCH2 ectodomains appear to be highly specific among human tumors for TNBC. It is of interest to note that structurally similar Notch1 deletions are common in murine T-ALL, where they are caused by DNA breakage at cryptic RAG recombinase sites (4). It is possible that underlying abnormalities of DNA repair make TNBC cells susceptible to activating NOTCH gene deletions caused by random DNA breakage followed by nonhomologous end joining (29). Recent studies by Shah and colleagues (30) observed a high degree of clonal and mutational diversity in TNBC, suggestive of genomic instability. In contrast, in human tumors with relatively small numbers of genetic changes, such as ACC and human T-ALL, Notch gain-of-function mutations tend to consist mainly of point substitutions and small indels.

Several recent tumor genome sequencing studies have identified activating mutations in the Notch signaling pathway in a minority of ACCs (27, 31, 33). Ross and colleagues (33) reported genomic alterations in NOTCH1 in 11% (3 of 28) of ACCs, whereas Ho and colleagues (31) reported alterations in NOTCH signaling pathway genes in 13% of samples. Stephens and colleagues (27) reported activating mutations in NOTCH2 and loss-of-function mutations in SPEN, a gene encoding a transcriptional repressor that forms a complex with RBPJ and downregulates Notch target genes. We did not observe mutations in SPEN in our primary tumor dataset, and it remains to be determined whether SPEN-mutated tumors and NOTCH-mutated tumors are comparable in terms of activation of downstream genes.

It might be anticipated that activating mutations in NOTCH genes would be robust predictors of tumor response to Notch pathway inhibitors such as GSIs, because recurrent mutation of oncogenes in particular types of tumors reliably identifies genes and pathways that are subject to selection during the initiation and progression of cancers. However, although GSIs have been proven safe (34), they are yet to be proven effective, and early experience with human T-ALL suggests that NOTCH gene mutational status, per se, is not highly correlated with response. Thus, additional biomarkers that are better predictors of response, preferably across a broad spectrum of cancers that are driven by Notch gain-of-function mutations, are highly desirable.

On the basis of the response of breast cancer and ACC xenografts to MRK-003, it appears that immunohistochemical staining for activated NOTCH1 (N1-ICD) may be one such biomarker. Xenografts with high levels of N1-ICD immunoreactivity showed excellent responses to GSI, alone and in combination with paclitaxel, whereas xenografts with low N1-ICD reactivity or GSI-refractory rearrangements in NOTCH2 show no response to GSI. It is notable in this regard that NOTCH1 mutations in human T-ALL were discovered through a cell line screen for GSI sensitivity in which the most sensitive cell lines were found to be those with dual NOTCH1 NRR and PEST domain mutations in cis, an alignment that produces high levels of N1-ICD (2). Our data suggest that it should be possible to select patients for clinical trials of GSI based on N1-ICD immunoreactivity in archival formalin-fixed paraffin-embedded tissue sections, which would enable rapid screening and identification of patients who are most likely to respond to GSI treatment. Of note, the neoepitopes created by GS cleavage of NOTCH2, NOTCH3, and NOTCH4 are distinct from that created by GS cleavage of NOTCH1, and, in principle, it should be possible to extend this approach to other members of the Notch receptor family, pending development of additional, NICD isoform–specific antibody reagents.

Another biomarker emerging at the interface of expression profiling and genomic analysis is HES4, which is one of a small number of genes that are common to the Notch-driven gene signatures in breast cancer, T-ALL, and B-cell lymphoma cell lines. HES4 expression is also well correlated with NOTCH mutational status in diverse cell lines, and identifies a group of patients with poor prognosis in TNBC. These data suggest that HES4 is a potentially valuable biomarker in certain tumor types such as breast cancer, with the important caveat that the results shown here need to be validated in independent clinical cohorts.

Detailed methods for cell growth assays, cell-cycle analysis, stem cell marker analysis, luciferase reporter assays, whole-exome sequencing, exome imbalance analysis, and RNA-seq analysis are provided in Supplementary Methods.

The T-ALL cell line CUTLL1 (kind gift from Adolfo Ferrando, Columbia University, New York, New York) was cultured in RPMI-1640 containing 10% FBS and 1% penicillin–streptomycin–glutamine. Other cell lines were purchased from cell line banks [ATCC, Japanese Collection of Research Bioresources Cell Bank (RIKEN), or Deutsche Sammlung von Mikroorganismen und Zellkulturen (DKMZ)] and were grown under culture conditions recommended by the vendors. The authenticity of the cell lines was verified by short tandem repeat (STR) profiling analysis or similar methodologies by the banks. In addition, the mutation and gene expression levels from the TES data were compared with the published mutation (Catalogue of Somatic Mutations in Cancer; Sanger database) and gene expression data.

Cells were lysed in 50 mmol/L Tris, pH 8.0, containing 150 mmol/L NaCl and 1% NP40 supplemented with Protease Inhibitor Cocktail (Thermo Scientific). Protein amounts were determined using the Bio-Rad DCTM Protein Assay Kit II according to the manufacturer's protocol. Samples were mixed with Laemmli sample buffer (Bio-Rad) containing 5% β-mercaptoethanol, separated by 6% or 4% to 15% SDS–PAGE (Bio-Rad), and transferred onto a polyvinylidene difluoride (PVDF) membrane using an iBlot dry transfer apparatus (Invitrogen). The membrane was blocked with 5% non fat dry milk (Bio-Rad) or 3% BSA (Sigma) in TBST (20 mmol/L Tris–HCl, 0.5 mol/L NaCl, and 0.1% Tween 20) and incubated with a primary antibody overnight at 4°C. Following washes with TBST, the membrane was incubated with horseradish peroxidase–conjugated secondary antibody (The Jackson Laboratory) and detected with ECL developing solution (Thermo Scientific). A list of primary antibodies used is provided in Supplementary Table S7.

Standard 5-μm paraffin-embedded tissue sections from xenografts were stained using an anti–N1-ICD rabbit monoclonal antibody (Cell Signaling Technology; clone D3B8, catalog #4147; final concentration, 17 μg/mL), as previously described (19).

CUTLL1, REC-1, and HCC1599 cells were cultured for 3 days with GSI (compound E; 1 μmol/L) to establish a Notch-off state. Notch was then reactivated by GSI washout as described (22) and harvested for analysis 4 hours later.

RNA was extracted from cultured cells or tumor xenografts using the RNeasy Mini Kit (Qiagen). cDNA was synthesized using SuperScript VILO MasterMix (Invitrogen). Quantitative PCR (qPCR) was performed on an ABI 7900 using TaqMan Gene Expression MasterMix (Invitrogen) and their inventoried TaqMan probes/primers (Supplementary Table S8); the resulting qPCR data were analyzed using the ΔΔC_t relative quantification protocol.

NOTCH1 and RBPJ ChIP were performed as described (8). REC-1 cells were crosslinked with 1% formaldehyde for 10 minutes at 37°C and sheared by sonication. Rabbit IgG (Jackson ImmunoResearch 011-000-003), NOTCH1 (35), and RBPJ (Cell Signaling Technology; #5313) antibodies were added to the sonicate and incubated overnight at 4°C. DNA–protein complexes were captured with protein A–conjugated agarose beads, washed, and eluted. After reversal of cross-links, DNA was purified using the QIAquick PCR Purification Kit (Qiagen). Input control was prepared in parallel without immunoprecipitation. Real-time PCR was performed in triplicate using primers specific for the HES4 promoter and genomic negative control. The primer sequences are as follows: HES4: forward, 5′-GGTGTGTGAACCCGGCTCCG-3′; reverse, 5′-CCGAGGCGTGACTGACAGCG-3′. Genomic negative control primers: forward, 5′-AATGCTGGGCTTCCAAGGA-3′; reverse, 5′-GACCTTGGTGACTGTTGAGGAAAC-3′.

From 1 to 8 × 10⁶ HCC1599, HCC1187, MB157, or MDA-MD-231 TNBC cells were inoculated subcutaneously into the left flank of 4- to 6-week-old immunodeficient (nu/nu or NOD/SCID) female mice (Charles River Laboratories). Patient-derived subcutaneous xenograft efficacy studies included the TNBC models HBCx8 and HBCx14 (Xentech) and the ACC models CTG-0007 (ACCx9) and CTG-0009 (POS-912; Champions Oncology Inc.). Upon reaching an average tumor size of 150 to 250 mm³, mice were randomized across control or treatment groups (n = 10–12 mice per group). Tumor size was measured with calipers, and body weight was recorded twice per week during the dosing phase. MRK-003 in 0.5% methylcellulose was given orally at the indicated dose/schedules, whereas paclitaxel in 0.9% NaCl was administered intraperitoneally at a 15 mg/kg once per week. Mice were euthanized at the indicated time points and portions of the tumors were snap-frozen in liquid nitrogen for biochemical analysis or fixed in 10% neutral buffered formalin for immunohistochemical analysis.

Detailed analyses of human tumor cell lines and breast tumors are provided in Supplementary Methods. A summary of the NOTCH gene coverage data from human tumors is provided (see Supplementary Table S9; cell line sequencing data can be accessed at Gene Bank accession number SRP044150). NOTCH gene signature analysis is provided in Supplementary Table S10.

C. Ware is Senior Scientist at Merck & Co., Inc. C. Moskaluk is a consultant/advisory board member for Novartis. X.S. Liu is a guest professor at Tongji University. J.C. Aster is a consultant/advisory board member for Cell Signaling Technology and CytomX, Inc. S. Sathyanarayanan was Principal Scientist at Merck & Co., Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Lejnine, L. Zawel, S. Fawell, T. Zhang, W.S. Pear, J.C Aster, S. Sathyanarayanan

Development of methodology: S. Lejnine, A. Truong, C. Zang, M. Kluk, B.B. Haines, B.E. Kremer, X.S. Liu, J.C Aster, S. Sathyanarayanan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Stoeck, A. Truong, L. Pan, H. Wang, J. Yuan, C. Ware, P.W. Garrett-Engele, M. Kluk, J. Laskey, C. Moskaluk, B.E. Kremer, B. Knoechel, W.S. Pear

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Stoeck, S. Lejnine, A. Truong, H. Wang, C. Zang, J. Yuan, J. MacLean, M. Kluk, J. Laskey, B.B. Haines, S. Fawell, G. Gilliland, T. Zhang, B.E. Kremer, B.E. Bernstein, X.S. Liu, J.C Aster, S. Sathyanarayanan

Writing, review, and/or revision of the manuscript: A. Stoeck, S. Lejnine, C. Zang, J. MacLean, M. Kluk, B.B. Haines, S. Fawell, G. Gilliland, T. Zhang, W.S. Pear, X.S. Liu, J.C Aster, S. Sathyanarayanan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Zang, J. Laskey, B. Knoechel

Study supervision: S. Fawell, G. Gilliland, B.E. Bernstein, X.S. Liu, J.C Aster, S. Sathyanarayanan

Other (performed immunohistochemistry): C. Ware, M. Kluk

The majority of this work was funded by Merck & Co., Inc. J.C. Aster and W.S. Pear were supported by NIH P01 grant (5P01CA119070-08) and W.S. Pear was supported in part by an Abramson Cancer Center Support Grant from the NCI. B.E. Bernstein acknowledges support from the Adenoid Cystic Carcinoma Research Foundation. B.E. Kremer is supported by NIH K32 Grant CA009615.