Notch ligands signal through one of four receptors on neighboring cells to mediate cell–cell communication and control cell fate, proliferation, and survival. Although aberrant Notch activation has been implicated in numerous malignancies, including breast cancer, the importance of individual receptors in distinct breast cancer subtypes and the mechanisms of receptor activation remain unclear. Using a novel antibody to detect active NOTCH3, we report here that NOTCH3 signals constitutively in a panel of basal breast cancer cell lines and in more than one third of basal tumors. Selective inhibition of individual ligands revealed that this signal does not require canonical ligand induction. A NOTCH3 antagonist antibody inhibited growth of basal lines, whereas a NOTCH3 agonist antibody enhanced the transformed phenotype in vitro and in tumor xenografts. Transcriptomic analyses generated a Notch gene signature that included Notch pathway components, the oncogene c-Myc, and the mammary stem cell regulator Id4. This signature drove clustering of breast cancer cell lines and tumors into the common subtypes and correlated with the basal classification. Our results highlight an unexpected ligand-independent induction mechanism and suggest that constitutive NOTCH3 signaling can drive an oncogenic program in a subset of basal breast cancers. Cancer Res; 77(6); 1439–52. ©2017 AACR.
Breast cancer is the second most common cancer among women in the United States and is the second leading cause of cancer death. Three main subtypes have been distinguished on the basis of the expression of progesterone and estrogen receptors and the HER2 receptor tyrosine kinase: luminal (hormone receptor–positive; ∼50%–60%), HER2-enriched (∼10%), and basal (triple-negative; ∼10%–20%; ref. 1). The basal subtype lacks targeted therapies and suffers from the poorest prognosis. Illuminating the molecular mechanisms supporting maintenance of this subtype thus holds therapeutic promise.
Notch ligands and receptors comprise a conserved family of transmembrane proteins that conduct cell–cell communication to regulate cell fate and stem/progenitor cell proliferation and differentiation. Current understanding holds that Notch receptors remain quiescent until ligand binding triggers a conformational change that enables ADAM-protease cleavage (at site S2) within the juxtamembrane negative regulatory region (NRR; refs. 2, 3). The γ-secretase complex subsequently cleaves within the membrane (site S3), liberating the Notch intracellular domain (NICD*) to translocate to the nucleus and direct the Notch transcriptional program (4). Mammals express 4 Notch receptors (NOTCH1–4) and 4 canonical ligands, including members of the Jagged (Jag)-Serrate family, JAG1 and JAG2, and the Delta-like family, DLL1 and DLL4.
Widely expressed and functioning broadly in tissue homeostasis, the Notch pathway has been linked to disease, including cancer. The strongest case supporting Notch as an oncogenic driver in humans stems from T-cell acute lymphoblastic leukemia (T-ALL), where NOTCH1 signal–stimulating mutations, including missense changes that destabilize the NRR and enable ligand-independent signaling, are found in more than half of the patients (5). Hyperactive Notch signaling has been implicated in numerous other leukemias and solid tumors, although the rationale for Notch as a cancer driver in these other indications generally lacks such “smoking gun” data of frequent mutations (6). These links to cancer have compelled therapeutic targeting, beginning with small-molecule γ-secretase inhibitors (GSI). However, GSIs target all Notch signaling plus other γ-secretase–dependent signals and are not well-tolerated.
Genetic and expression studies have linked Notch signaling to breast cancer. A subset of breast cancers display JAG1 and NOTCH1 overexpression, correlating with a poor prognosis (7, 8), and genetic translocations leading to constitutive Notch1 and Notch2 signaling occur in 2.6% of basal breast tumors (9). In mice, NOTCH3 expression marks a self-renewing population of luminal progenitor cells in the normal gland (10), whereas constitutive NOTCH3 signaling generates tumors and reveals NOTCH3 oncogenic potential (11). In the human disease, basal breast cancers show NOTCH3 amplification and overexpression (12, 13), and Notch3 knockdown reduces proliferation of breast cancer cell lines (14). Jag1 or Jag2 knockdown reduced Notch activity in these lines, suggesting that these ligands were required to induce the NOTCH3 signal (14). While these studies provide a foundation, the mechanism and prevalence of NOTCH3 activation in breast cancer have not been rigorously characterized.
To address these questions, (i) we developed a new reagent antibody that selectively detects activated NOTCH3 and (ii) we exploited therapeutic antibodies that selectively agonize or antagonize individual Notch ligands and receptors, moving beyond GSIs to precisely manipulate Notch function and minimize toxicity. We discovered that NOTCH3 signals constitutively in breast cancer cell lines, independent of canonical ligand induction, as well as in human breast cancers, particularly those of the basal/triple-negative subtype. Selective NOTCH3 inhibition antagonizes breast cancer cell line and tumor growth, whereas NOTCH3 agonism promotes a malignant phenotype and increases proliferation. Transcriptomic analyses establish a Notch signature that includes increased expression of the c-Myc oncogene as well as the mammary stem cell regulator Id4. This signature is sufficient to cluster breast cancer cell lines and patient tumors into the major subtypes and positively correlates with basal samples, indicating that NOTCH3 activity may be clinically relevant in this subclass.
Materials and Methods
Antibody characterization, immunohistochemistry
Hybridoma supernatants (rabbits immunized with human NOTCH3 peptide 1662–1675) were screened by immunoblotting of MDA-MB-468 nuclear lysates, after using N-[N-(3,5-difluorophenylacetyl)-l-alanyl]-S-phenylglycine tert-butyl ester (DAPT) or 10 mmol/L EDTA to inhibit or stimulate NOTCH3, respectively. This concentration of EDTA chelates calcium ions needed for proper folding of the LNR modules of the NRR and disrupts receptor quiescence without inhibiting ADAM protease cleavage (15–17). Tissues from 4- to 8-week-old knockout mice (B6;129S1-NOTCH3tm1Grid/J, Jackson Labs) and age-matched controls (B6129SF1/J) were fixed in formalin and embedded in paraffin. After rehydration, slides were pretreated with Target Retrieval solution (Dako) for 20 minutes at 99°C, rinsed, treated with KPL (Kierkegaard and Perry Laboratories) and avidin/biotin blocking solution (Vector Labs), TNB buffer (Perkin Elmer), and primary antibody (2.5 or 10 μg/mL) for 1 hour at room temperature. Slides were rinsed and incubated with donkey anti-rabbit biotinylated secondary (Jackson Immunoresearch) followed by tyramide signal amplification (Perkin Elmer). Visualization used metal-enhanced DAB staining (Thermo Scientific), counterstaining, and dehydration. Isotype control antibody controlled for staining specificity. Tissue arrays were from Biomax.
Short tandem repeat (STR) profiles using Promega PowerPlex16 were compared with external profiles to authenticate cell lines (ATCC).
Notch expression, activation
Cell lines were cultured in: RPMI/10% FBS/1% GlutaMAX (Life Technologies; breast cancer), DMEM/20% FBS/1% GlutaMAX (C2C12 myoblasts), and DMEM with 10% FBS/1% GlutaMAX/1% nonessential amino acids (Life Technologies; U87 cells). Cells grown to about 80% to 90% confluence were treated with 5 μmol/L DAPT (EMD), DMSO (0.02% final concentration), or antibodies. Isotype controls were α-gD (hIgG1) or α-ragweed (mIgG2a). Antibody concentrations were kept constant between samples by adding control antibody as needed. For treatments lasting more than 24 hours, DAPT was re-added 3 hours before harvest. MG132 (EMD; 20 μmol/L) was added at this time, as needed. For Notch induction, cells were washed with PBS and incubated for 20 minutes at 37°C in 10 mmol/L EDTA. JAG1-induced signaling was triggered for 3 hours with rat recombinant JAG1-Fc (R&D) immobilized to anti-Fc–coated paramagnetic particles (Bangs Labs).
Extracts were prepared in RIPA buffer after 72-hour transfections with 50 nmol/L OTP-modified siRNA pools (Dharmacon) or nontargeting siRNA. For IHC, cells were treated with 5 μmol/L DAPT after 48-hour transfections, incubated for 24 hour, washed, and incubated 20 for minutes with 10 mmol/L EDTA ± 5 μmol/L DAPT.
Nuclear and cytoplasmic/membrane fractions were prepared after lysis (RIPA, Pierce; ref. 18), separated using 4%–12% NuPAGE gels/MOPS (Life Technologies), transferred to polyvinylidene difluoride (PVDF; TransBlot Turbo, BioRad), blocked with 5% milk, and probed for 12 to 16 hours with: 1 μg/mL anti-NICD3* (V1662), 0.2 μg/mL anti-NICD2* (V1697), 1:1,000 Cell Signaling antibodies (anti-NICD1* D3B8, anti-NOTCH3 2889, anti-JAG2 C83A8, anti-CREB 48H2), 1:500 anti-JAG1 (Santa Cruz H-114), or 1:5,000 anti-tubulin (Sigma DM1a). Secondary antibodies (1:15000) were from BioRad. Chemiluminescent detection used Super Signal West Pico or Femto reagents (Thermo).
Cells (n = 1,500) in 40 μL 0.29% low-melting temperature agarose (Life Technologies) with 1× RPMI/10% FBS were plated in 96-well plates prefilled with 40 μL 0.6% agarose in the same medium and then overlaid with 40 μL medium plus a 3× concentration of test articles: 15 μmol/L DAPT/0.06% DMSO or 60 μg/mL antibody. Cells were fed twice weekly over 12 to 19 days with 40 μL medium plus 1× test articles. Plates were scanned using Gelcount (Oxford Optronix Ltd.).
Cells (n = 1,000) in 40 μL drops (test articles as in soft agar) were prepared in GravityPLUS plates using 10 mL PBS/0.1% Triton X-100 in the humidifier pad (InSphero). Medium and test articles were changed twice weekly over 8 days before centrifuging to 96-well plates, lysing cells (30 minutes, 100 μL), and processing (CellTiter-Glo, Promega).
A total of 1.7 × 104 cells were plated in 1.5 mL per well in a 6-well plate in medium plus test articles, which were refreshed twice weekly over 11 to 18 days before counting cells.
Animal studies were conducted in accordance with the Guide for the Care and Use of Laboratory Animals, National Academy Press (2006) and approved by the Institutional Animal Care and Use Committee (IACUC, Genentech). C.B-17 SCID.bg mice (Charles River Labs) were inoculated with 5 million cells in 0.2 mL HBSS/Matrigel into the 2/3 mammary fat pad. When tumors grew to 80 to 150 mm3 [length and width measured with UltraCal-IV calipers 54-10-111, Fred V. Fowler Company; tumor volume = (length × width2) × 0.5], mice were randomly assigned to treatment groups (10/group) and injected intraperitoneally twice weekly with 30 mg/kg mouse IgG2a antibodies: α-ragweed, anti-N3.A4, or anti-N3.A13. In the HCC1143 study, anti-N3.A13 was dosed once a week for 2 doses. For the MDA-MB-468 gene expression studies, the anti-N3.A13 dose amount was lowered to 20 mg/kg after the initial dose and supplemented with anti-ragweed to keep total IgG at 30 mg/kg. Animal weights were monitored; none of the antibody regimens caused weight loss. Tumors were fixed, embedded, sectioned, and stained with hematoxylin or anti-V1662.
For the NOTCH3 antagonism experiment, 2.5 × 106 MDA-MB-468 cells/10-cm plate were treated after 40 hours with 10 μg/mL antibody (anti-gD or anti-N3.A4) + 0.05% DMSO ± 5 μmol/L DAPT and harvested 24 hours later. For NOTCH3 agonism, 6.25 × 106 cells/10-cm plates were treated with 0.05% DMSO (baseline) or 5 μmol/L DAPT/0.05% DMSO (Notch inhibited) for 16 hours and then cells were washed and incubated for 2 hours in medium alone. After washout, cells were treated with DMSO, 5 μmol/L DAPT/DMSO, 20 μg/mL control IgG, 20 μg/mL anti-N3.A13, JAG1-Fc–coated beads, or Fc-coated beads. DAPT or DMSO was refreshed in their corresponding plates 22 hours later. Cells were harvested after 24 hours. RNA was extracted (Qiagen RNeasy Plus kit) and yields analyzed (Agilent Bioanalyzer). Microarray methods used Affymetrix HGU133plus2 expression microarrays and GeneChip protocols, and data were processed using Affymetrix Microarray Analysis Suite software version 5 to yield signal data; results were deposited to GEO (GSE82298). Three arrays were performed per condition using biologic replicates. RNASeq reads were mapped to the hg19 human reference genome using “GSNAP” (19) and normalized for each gene by transcript length and total number of mapped reads to yield RPKM. Differential expression was estimated using DESeq2 (Bioconductor; ref. 20).
Detection of active NOTCH3
Antibodies recognizing the active form (S3-cleaved) of Notch receptors, rather than just receptor expression, have proven valuable for measuring NOTCH1 and NOTCH2 activity (21, 22), and recent work has described a similar antibody for studying NOTCH3 signaling in T-ALL (23). We characterized a new tool for assessing NOTCH3 activity, with a focus on NOTCH3 signaling in basal breast cancer, starting with the generation of a monoclonal antibody directed at the human NOTCH3 S3 site at amino acid V1662 (Supplementary Fig. S1). We tested whether the V1662 antibody recognized active NICD3 (NICD3*) using the basal breast cancer cell line MDA-MB-468 because this line could be manipulated to create a range of NICD3* levels. V1662 recognized a polyprotein of the molecular mass expected for NICD3* (Fig. 1A, Supplementary Fig. S1). The band intensity correlated with signal strength, increasing or decreasing after treating the cells with a NOTCH3 agonist (anti-N3.A13) or antagonist antibody (anti-N3.A4), respectively (Fig. 1A and B); similarly, treatment with EDTA to destabilize the NRR and induce signaling (15–17) or with a GSI (DAPT) to inhibit signaling affected band intensity in the predicted manner (Fig. 1A). Targeting NOTCH3 with siRNAs efficiently silenced NOTCH3 expression (Fig. 1C) and eliminated the V1662 signal, confirming antibody specificity. Furthermore, V1662 recognized a band of the expected size, and only after Notch activation, in the mouse cell line C2C12 (Fig. 1D, left), yet did not reveal any signal at the expected positions for either NICD1* or NICD2* (Fig. 1D, left), despite the fact that these NICDs were clearly detected by anti-NOTCH1 V1774 in 3T3-N1 cells and by anti-NOTCH2 V1697 in U87 cells, respectively (Fig. 1D, right). These data demonstrate that anti-NOTCH3 V1662 specifically detects human and mouse NICD3* but does not cross-react with NICD1* or NICD2*.
To enable assessment of activated NOTCH3 in clinical samples, we tested V1662 for IHC applications in formalin-fixed, paraffin-embedded (FFPE) material. EDTA activation yielded a strong V1662 signal that was significantly reduced following DAPT or siRNA inhibition (Fig. 1E, top). To rigorously test staining specificity, we compared wild-type (WT) or NOTCH3 knockout (KO) mice, focusing on NOTCH3 in brain vasculature and muscle satellite cells. Whereas WT tissues showed a clear nuclear signal of V1662 staining, KO tissues did not (Fig. 1E, bottom), supporting the assertion that V1662 specifically detects active NICD3*.
NOTCH3 activation is constitutive and ligand-independent
Our detection of constitutively active NOTCH3 signaling seemed surprising, as Notch signaling typically requires induction with a ligand expressed on neighboring cells. Mutational destabilization of the NRR, as found in certain cancers, provides one hypothesis to explain this result. However, DNA sequencing of cell lines with constitutively active NOTCH3 signaling did not identify any alterations in the Notch3 NRR (data not shown). Consistent with previous results (23), our sequencing did reveal that MDA-MB-468 cells express both a WT and a mutant Notch3 allele with a truncated PEST domain (data not shown), expected to prolong but not induce signaling. Moreover, such truncations are irrelevant with regards to the full-length, WT NOTCH3 protein on which we focus. We hypothesized that signaling might be induced by endogenously expressed ligand. Immunoblots revealed expression of JAG1 in MDA-MB-468 and JAG1 plus JAG2 in HCC1143 cells, a second basal line with active NICD3* (Fig. 2A and B). However, potent antibody inhibitors of these JAG ligands (24) failed to block the constitutive NOTCH3 signal, in contrast to the inhibition from a GSI or anti-N3.A4 (Fig. 2B and C) and as observed in additional basal lines (data not shown). Likewise, anti-DLL1 (25) and anti-DLL4 (26) blocking antibodies, used alone or in combination with high concentrations of all 4 anti-DLL/JAG inhibitors, did not affect NOTCH3 signaling (Fig. 2D). To test whether ligand-induced signaling could originate in an intracellular compartment inaccessible to antibodies, we used siRNA to silence Jag1, the most abundantly expressed canonical ligand in MDA-MB-468 cells (Fig. 2A). While knockdown of Notch3 significantly inhibited the NOTCH3 signal, evidenced by the reduction in NICD3* and JAG1 protein (NOTCH3 signaling induces Jag1 expression; see below), knockdown of Jag1 had little or no effect on the NOTCH3 signal (Fig. 2E). Inhibition using the NOTCH3 antagonist antibody further argues against this hypothesis of intracellular signal induction. Our results using multiple methods thus demonstrate (i) constitutive NOTCH3 signaling that occurs (ii) independent of canonical Notch ligands.
Constitutive NOTCH3 signaling is prevalent in basal cell lines and tumors
To determine the frequency at which constitutive NOTCH3 signaling occurs in breast cancer cell lines, we expanded our screen to include 17 basal plus 5 hormone receptor–positive lines. NICD3* levels roughly correlated with expression of total NOTCH3, with clearly detectable NOTCH3 signaling in 13 of 17 basal lines and weaker but detectable signaling in 3 of 5 hormone receptor-positive lines (Fig. 3A). Although the majority of lines expressed JAG ligands, JAG levels did not correlate with NICD3* levels (Fig. 3A), consistent with NOTCH3 occurring independent of canonical ligands. DLL1 and DLL4 were expressed weakly, if at all, across the entire cell line panel (data not shown).
We exploited our NICD3* IHC method to determine the prevalence of NOTCH3 signaling in human breast cancers. We observed a range of NICD3* signals, from strong nuclear staining in the majority of tumor cells to weak staining in a fraction of cells (Fig. 3B), in 73 of 371 samples (20%; Fig. 3C). Triple-negative breast cancer (TNBC) samples were enriched for NICD3* staining, observed in 52 of 153 cases (34%; Fig. 3C). Consistent with published results (27), we detected NICD1* in tumor cells in only 4% of total samples (Fig. 3C). Thus, NOTCH3 signaling is evident in a significant percentage of TNBCs and is found more frequently than is NOTCH1 signaling.
NOTCH3 signaling promotes breast cancer cell line growth
To assess whether NOTCH3 signaling fostered basal tumor cell growth, we exploited NOTCH3 function–modulating antibodies in assays that involve cell–cell contact, a hallmark of Notch signaling. These therapeutic antibodies selectively target the human NOTCH3 NRR, either antagonizing (anti-N3.A4) or agonizing (anti-N3.A13) signaling (28). We first examined anchorage-independent growth in soft agar and found that selective NOTCH3 blockade using anti-N3.A4 significantly inhibited colony growth, to an extent similar to that seen using the pan-Notch inhibitor DAPT (Fig. 4A and B). Selective inhibition of NOTCH1, NOTCH2, or JAG1 had little or no effect (Fig. 4B and Supplementary Fig. S2A). In contrast, NOTCH3 agonism using anti-N3.A13 modestly stimulated colony growth, an effect that was reversed when anti-N3.A4 was included (Fig. 4A and B and Supplementary Fig. S2A). We observed similar results with a variety of basal lines and growth assays (Fig. 4C and D, Supplementary Fig. S2B, data not shown). Colony formation on plates revealed the most striking effects, with NOTCH3 blockade dramatically inhibiting growth (Fig. 4E and F). Our results with multiple cell lines and assays, using selective and general Notch inhibitors, indicate that NOTCH3 signaling promotes growth or survival in a subset of basal lines and is sufficient to account for the observed growth effects.
The colony formation assay also provided a means to investigate cell morphology. Control colonies displayed a mix of 2 cell morphologies, with some cells appearing in a cobblestone pattern, typical of epithelial cells, and other cells appearing rounded, consistent with a transformed phenotype (Fig. 4G, middle). NOTCH3 antagonism shifted the colony phenotype to one almost exclusively comprised of a cobblestone epithelial layer (Fig. 4G, left), whereas NOTCH3 agonism shifted the phenotype to one primarily composed of rounded cells (Fig. 4G, right). Such NOTCH3-controlled shifts in phenotype evoke classic views of Notch regulating binary cell fate decisions and are consistent with NOTCH3 signaling promoting a transformed state.
NOTCH3 activation drives tumor growth in vivo
To examine whether NOTCH3 signaling affected tumor growth in vivo, we grew basal lines as xenografts in the mammary fat pads of immunocompromised mice, allowed tumors to establish, and then treated with our NOTCH3 antibodies. Tumors grown from HCC1143 cells, which display high levels of active NOTCH3 signaling (Fig. 3A), dramatically responded to NOTCH3 agonism, doubling in volume approximately twice as rapidly in the presence of anti-N3.A13 compared with control (Fig. 5A). NOTCH3 antagonism using anti-N3.A4 slowed tumor growth, although the effect was modest relative to the growth stimulation induced by agonism (Fig. 5A). Tumors from other basal lines responded similarly, although the magnitude of the responses varied, with MDA-MB-468 cells showing the most aggressive growth acceleration following NOTCH3 stimulation (Fig. 5B, Supplementary Fig. S3). Immunoblotting and IHC using V1662 showed high levels of active NOTCH3 signaling in the vast majority of tumor cells following anti-N3.A13 treatment, indicating that tumor growth rates correlated with levels of nuclear NICD3* (Fig. 5C and D). NOTCH3-agonized tumor cells appeared more aggressive than did their control counterparts, with increased numbers of active mitoses (Fig. 5D). Given that NOTCH3 antagonism significantly inhibited colony formation in vitro (Fig. 4), the modest growth inhibition observed in vivo suggested that differences in the growth settings impacted the responses to NOTCH3 inhibition. Indeed, MDA-MB-468 cells expressed lower levels of both total and active NOTCH3 in vivo compared with in vitro (Fig. 5C, note the near absence of NICD3* in the control anti-RW tumors and compare with Fig. 3), consistent with slow tumor growth and only a modest further slowing after NOTCH3 inhibition (Fig. 5B).
NOTCH3 drives an oncogenic expression signature sufficient to cluster breast cancer subtypes and correlating with basal phenotype
We pharmacologically toggled NOTCH3 signaling up or down from its baseline state in basal lines (i) to compare NOTCH3-specific versus pan-NOTCH manipulation and (ii) to define the downstream gene signature (Fig. 6A). Gene expression changes following 24-hour treatment with DAPT or anti-N3.A4 appeared very similar (Fig. 6B, Supplementary Table S1). To ascertain whether the expression changes in the baseline state differed from those driven by a full “on” signal after NOTCH3 agonism, we performed a wash-out experiment (29). Following DAPT incubation to inhibit Notch signaling, cells were replated without DAPT but with (i) anti-gD isotype control antibody, to allow the cells to re-establish baseline signaling (the de-repressed state) or (ii) anti-N3.A13, to agonize NOTCH3 signaling (the stimulated state). Expression changes from the same set of genes were observed under both conditions, although the magnitude of the changes was greater following NOTCH3 agonism, as expected (Fig. 6C, Supplementary Table S2). We also replated in the presence of (i) immobilized JAG1, to provide pan-Notch stimulation or (ii) anti-N3.A13, to provide NOTCH3-specific agonism and found that both inducers caused similar expression changes (Fig. 6D, Supplementary Table S3). These studies support the conclusion that NOTCH3 signaling is constitutively “on” and that this baseline signal can be both inhibited and stimulated. Moreover, these results indicate that NOTCH3 signaling is sufficient to account for the Notch effects in these cells, without invoking contributions from other Notch receptors.
To elucidate a Notch basal transcriptional program, we defined a Notch gene signature in MDA-MB-468 cells by comparing DAPT and anti-N3.A13 treatments. This comparison identified 83 genes for which expression varied significantly (adj. P < 0.01, Benjamini–Hochberg correction), with at least a 2-fold change (Fig. 6D, Supplementary Table S4). NOTCH3 positively regulated a number of oncogenes and cell-cycle genes as well as Notch transcriptional targets identified in other cell types (Fig. 6E, Supplementary Table S4). For example, 2 well-established Notch targets, Nrarp (30) and Olfm4 (31), increased expression following NOTCH3 agonism, underscoring the validity the signature. The downstream oncogenic program includes the oncogene c-Myc as well as stem cell markers/oncogenes Id4 and Kit. Consistent with its driving cell-cycle progression and proliferation, NOTCH3 activity also increased expression of CCND1 and Cdk6. The Notch ligand Jag1 itself appeared as downstream target, suggesting a positive feedback loop. Intriguingly, NOTCH3 signaling downregulated expression of Dkk1, a negative regulator of Wnt signaling, and gene set enrichment analysis revealed a positive correlation of Notch and Wnt signaling (data not shown), suggesting that NOTCH3 may act upstream to stimulate the Wnt pathway. Supporting the notion that Notch may favor a basal-over-luminal lineage decision in breast cancer, NOTCH3 signaling correlated with higher levels of keratins 5, 6B, and 14—markers of the breast myeoepithelial lineage and basal cancers—and with lower levels of FOXA1 and keratins 7, 8, and 18—markers of the luminal lineage (32, 33).
We next sought to determine whether this Notch gene signature, defined in the MDA-MB-468 cell line, could demarcate triple-negative samples and, more importantly, reveal Notch activity in patient tumor samples. We first performed nonsupervised hierarchical clustering on a RNA sequencing data set derived from a panel of 70 breast cancer cell lines. Our Notch basal signature was sufficient to cluster the lines into 2 main groups, one enriched for basal and the other for luminal and Her2 overexpression (Fig. 7A). Notably, a similar analysis using of a large set of patient tumors (34) showed that the Notch signature sufficed to tightly cluster the patient samples into the main subclasses of breast tumors (Fig. 7B).
We calculated the average correlation (Pearson coefficient) of signature gene expression within each cancer subtype to our signature (Fig. 6E). The basal tumors displayed the highest correlation (basal correlation = 0.26), indicating that the NOTCH3- controlled transcriptional network is active in this subtype of cancer and supporting the notion that a Notch signal may be clinically relevant in basal tumors. In contrast, the other groups showed a negative correlation (Her2, Luminal A, and Luminal B correlations = −0.17, −0.08, and −0.17, respectively), indicating that the state of the signature in these subtypes is inconsistent with NOTCH3 activation.
Notch receptor activation appears oncogenic in certain cellular contexts. Pan- Notch inhibition using GSIs supports this conclusion (13, 35), although use of GSIs fails to distinguish the particular Notch receptor driving growth. In human cancers, the clearest cases supporting Notch as a tumor driver derive from genomic studies revealing Notch1 mutations in the (i) NRR, which activate signaling in a ligand-independent manner, and (ii) PEST domain, which prolong signaling. Such mutations were reported over a decade ago in T-ALL (5), and this leukemia remains the best-studied example of oncogenic Notch signaling in humans. However, the subsequent wealth of tumor genomic data revealed that such mutations are rare in most tumor types, including breast (13). Arguments for Notch as a cancer driver largely turned to receptor and ligand overexpression, including in breast cancers (7, 9, 12), with Notch1 highlighted in the basal subtype. The discovery of translocations that activate NOTCH1 or NOTCH2 (9), as well as PEST domain mutations in multiple Notch receptors (13), although rare, lend genetic weight to the argument that Notch is important in basal cancers. Notch3 gene amplifications (12, 13) as well as mouse models and in vitro studies (7, 14) reinforced the possible relevance of NOTCH3 in this cancer subtype.
In the absence of the “smoking gun” of frequent activating mutations, we argue that it is imperative not to focus on receptor expression but rather on activity. Notch target gene expression (36, 37) can provide a surrogate for activity, but Notch target gene profiles vary between cell types and context. Thus, we have focused on the presence of γ-secretase–cleaved NICD*. Indeed, an antibody that specifically detects NICD1* has proven valuable for assessing NOTCH1 activity, and we recently exploited a similar antibody for NOTCH2 (21). However, such an antibody has only recently been described for NOTCH3 (23), and use of this antibody centered on Notch in T-ALL. The fact that NICD1* was detected in only 3 of 78 breast tumor samples examined (27) highlights the importance of examining activity of other Notch receptors, including NOTCH3, in breast cancer. Our characterization of anti-NICD3* V1662, including the use of Notch3 knockout tissues, shows that this antibody specifically detects NICD3* from human and mouse by both immunoblotting and IHC. We note that the ability to detect the NICD3* neo-epitope generated by γ-secretase cleavage demonstrates that NOTCH3 activity in these samples does not originate from the type of NOTCH3 fusions reported in glomus tumors (38)—an important consideration for the deployment of therapeutic antibodies because such fusions delete the amino terminus of NICD3 and, thus, would not be targetable by antibodies.
We first detected active NOTCH3 signaling in several basal cell lines. Surprisingly, we discovered that this signaling appears to be ligand-independent (although it remains possible that signaling is induced by a noncanonical ligand not targeted in our antibody panel). While Drosophila studies provide precedence for ligand-independent Notch signaling (39), the examples of ligand-independent signaling in mammals have largely centered on NRR destabilizing mutations, such as those found in T-ALL. However, studies of NRR3 dynamics point to a less stable autoinhibited state for NRR3 compared with other Notch receptor NRRs, and a recent structural analysis of NRR3 bolsters this assertion, further revealing elevated basal activity in reporter assays of NOTCH3 compared with NOTCH1 or NOTCH2 (40).
Our findings suggest that this less stable autoinhibited state is relevant in cancer. Specifically, we discovered that endogenously expressed NOTCH3 signals at a “de- repressed” baseline level in the absence of ligand induction. Given the elevated baseline activity of NOTCH3, one might view NOTCH3 as a WT equivalent of the mutationally activated NOTCH1 found in T-ALL, providing a moderate Notch signal in the absence of any receptor mutation. Indeed, we detected constitutively active NOTCH3 in a broad panel of breast cancer cell lines and in patient samples. Proliferation and viability studies, performed using a suite of assays in vitro and in vivo, suggested that this activity is physiologically meaningful, revealing that the level of NOTCH3 signaling, but not expression levels of NOTCH1, NOTCH2, or JAG1, correlated with growth.
We exploited our ability to toggle the level of NOTCH3 signaling with NOTCH3 agonist or antagonist antibodies to define a Notch gene signature in basal cancer and to compare NOTCH3-specific versus pan-Notch effects. NOTCH3 appeared sufficient to explain the downstream Notch transcriptional response, following either inhibition or agonism, without the need to invoke signaling from other Notch receptors. The gene signature provocatively suggests mechanisms that may function downstream of NOTCH3 signaling to drive an oncogenic phenotype. JAG1 and NOTCH3 are downstream targets, and NOTCH3 agonism correlated with NOTCH3 increases—which our data now indicate are sufficient for increased activity—highlighting a positive feedback loop. Our observation that c-Myc expression clearly depends on NOTCH3 activity is intriguing, given that c-Myc is an established NOTCH1 target and a key node in the oncogenic Notch network driving T-ALL (41). Two other oncogenes, Id4 and Kit, are also downstream targets of NOTCH3. Both genes are stem cell markers, prompting speculation that NOTCH3 may function in basal breast cancer to maintain an undifferentiated state. ID4 has also been described as a master regulator of the basal lineage (42), suggesting that NOTCH3, through control of a downstream transcriptional network that includes ID4, may regulate the same cell fate. Extending this rationale to the malignant state, it is tempting to speculate that NOTCH3 may help maintain a basal tumor subtype and oppose establishment of luminal or other subtypes. Our data provide preliminary support for this idea, given that NOTCH3 induces keratin markers of the basal subtype, downregulates markers of the luminal fate, and controls a cell morphologic switch consistent with a fate switch.
Our gene signature was sufficient to cluster not just cell lines, but also a large panel of patient tumors, into the well-established subtypes. Most importantly, the Notch “on” signature correlated with the basal phenotype, consistent with our IHC measurements indicating that NOTCH3 is active in more than one third of basal tumors. This finding suggests that therapeutic targeting of Notch3 could provide therapeutic benefit without the known toxicities associated with pan-Notch inhibition. As a whole, our studies highlight constitutive NOTCH3 signaling in a subset of TNBCs and underscore the importance of additional studies aimed at thoroughly ascertaining how strongly these tumors depend on Notch signaling for growth and survival.
Disclosure of Potential Conflicts of Interest
L. Choy reports receiving other commercial research support from Genentech. No potential conflicts of interest were disclosed by the other authors.
Conception and design: L. Choy, C.W. Siebel
Development of methodology: L. Choy, M. Solon, D.M. French, A. Shelton
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Choy, T. Hagenbeek, D. Finkle, A. Shelton, R. Venook
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Choy, T. Hagenbeek, A. Shelton, M.J. Brauer, C.W. Siebel
Writing, review, and/or revision of the manuscript: L. Choy, C.W. Siebel
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Hagenbeek, A. Shelton, R. Venook, C.W. Siebel
Study supervision: C.W. Siebel
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received June 2, 2016.
- Revision received December 1, 2016.
- Accepted December 19, 2016.
- ©2017 American Association for Cancer Research.