Convergent and Divergent Cellular Responses by ErbB4 Isoforms in Mammary Epithelial Cells Free

This article is featured in Highlights of This Issue, p. 1067

The four receptor kinases in the epidermal growth factor (EGF) family, EGFR, ERBB2, ERBB3, and ERBB4 regulate developmental processes in the nervous system, cardiovascular system, and in epithelia. EGFR, ERBB2, and ERBB3 are common drivers in human carcinoma and glioblastoma and are targets for cancer therapeutics approved by the US FDA. But, ERBB4 has a more ambiguous influence on cancer. ERBB4 is overexpressed in medulloblastoma, and candidate ERBB4–activating mutations have been identified in lung cancer, melanoma, and other cancers (1–4). Nonetheless, conflicting reports have been published on ERBB4 as a prognostic marker, with both positive and negative clinical outcome correlations (5–7).

Inconsistent associations of ERBB4 with cancer may be explained by the diversity of ERBB4-regulated signaling processes enabled by mRNA splice variants. JM-a and JM-b isoforms differ in the extracellular juxtamembrane domain (8). JM-b isoforms are conventional receptor tyrosine kinases (RTK): The ligands, including neuregulin 1 (NRG1), induce receptor phosphorylation and activate subsequent signal transduction. In contrast, JM-a isoforms have a metalloproteinase cleavage site that is clipped by tumor necrosis factor-α-converting enzyme (TACE) in response to NRG1 binding. This releases the extracellular domain, leaving the membrane-anchored m80 form. ERBB4 m80 can then undergo intramembrane cleavage by γ-secretase to release the soluble s80 form comprising the intracellular domain (ICD). s80 relocalizes to mitochondria and the nucleus (9, 10), in which it binds transcriptional coregulators and transcription factors.

A second alternatively spliced region in the ICD includes (CYT-1) or excludes (CYT-2) an exon that encodes a binding site for the p85 adaptor subunit of phosphatidyl inositol (3′) kinase, and an overlapping WW domain PPXY-binding site. Divergence of signaling processes incited by the four ERBB4 isoforms may explain the discordance in the ERBB4 cancer literature: Most studies fail to consider these isoforms separately, and the isoform(s) expressed and subcellular localization of ERBB4 have an impact on prognosis (11, 12).

We previously identified binding of both ERBB4 ICD isoforms (CYT-1 and CYT-2) with the transcriptional corepressor KAP1, and identified 16 other candidate interactors, including ubiquitin ligases ITCH and WWP2 (13). The ERBB4 ICD has been reported by others to associate with transcription factors ERα and Stat5, with transcriptional coregulators, including YAP, WWOX, ETO2, and a TAB2/N-CoR complex, and with ubiquitin ligases Itch and Mdm2 (14–20). To better understand the diverse biologic outcomes associated with activity of the full-length (FL) and truncated ERBB4 isoforms, we have explored the phenotypic, transcriptional and signaling consequences of introduction and activation of ERBB4 isoforms, and identified candidate gene target interactions by chromatin immunoprecipitation-sequencing (ChIP-seq).

MCF10A cells were maintained in DMEM/F12 supplemented with 5% horse serum, 20 ng/mL EGF, 0.5 mg/mL hydrocortisone, 100 ng/mL cholera toxin, 10 μg/mL insulin, 100 U/mL penicillin, and 100 μg/mL streptomycin. MCF10A cells stably expressing FL JM-a CYT-1–ERBB4 isoform (CYT-1 MCF10A) or JM-a CYT-2–ERBB4 isoform (CYT-2 MCF10A) or vector only (V-MCF10A) were generated by lentiviral infection and selection with 10 μg/mL puromycin and maintained in 1 μg/mL puromycin. MCF10A cells stably expressing either of the ICD ERBB4 isoforms: CYT-1 or CYT-2 were produced by lentiviral infection, selection in with 10 μg/mL blastocidin and maintenance in 7 μg/mL blastocidin. T47D and MDA-MB-231 cells were cultured in RPMI-1640 with glutamate (Gibco) containing 100 U/mL penicillin, 100 μg/mL streptomycin, and 10% fetal bovine serum (BioWest). FuGENE 6 (Roche) or Lipofectamine 2000 reagent (Invitrogen Corporation) were used for transfections. T47D cells were transduced with pLKO ERBB4 3′ untranslated region (UTR)–directed shRNA (Sigma; TRCN0000314628) or scrambled control and selected in 1 μg/mL puromycin. These ERBB4 knockdown (KD) T47D stable cell lines were subsequently infected with pInducer20 ERBB4 JM-a CYT-1, CYT-2, or vector control and selected in 400 μg/mL G418. T47D ERBB4 KD, pInducer20 CYT-1 or CYT-2 stable cell lines were maintained in 1 μg/mL puromycin, 200 μg/mL G418, and ERBB4 KD and doxycycline (DOX)-inducible ERBB4 isoform reexpression was confirmed by Western blot analysis.

Lentiviral expression plasmids for JM-a FL CYT-1 ERBB4 (EX-A0212-Lv105), CYT-2 ERBB4 (EX-Z4265-Lv105), and negative control vector (EX-EGFP-Lv105), including the CMV promoter followed by the ERBB4 coding sequences, puromycin selection cassette, 3′ long terminal repeat (LTR), poly adenylation sites, ampicillin cassette, pUC Ori, 5′LTR and packaging elements, were obtained from GeneCopoeia. ERBB4 plasmids were packaged as lentivirus by cotransfecting 293T cells with pLP/VSV-G, pLP1(Gag/Pol), pLP2(rev), and pcTat (tat) using Lipofectamine 2000 (Invitrogen Corporation). MCF10A cells were infected with a multiplicity of infection of approximately 5 in the presence of 8 μg/mL polybrene. Expression of ERBB4 in these MCF10A cells was tested 24 and 72 hours after infection. Polyclonal stable cell lines were selected with puromycin. The above FL CYT-1 and CYT-2 ERBB4 constructs were also packaged into pInducer20 DOX-inducible expression plasmids that were used to infect ERBB4 KD T47D stables to reexpress specific JM-a ERBB4 CYT-1 or CYT-2 isoform. The pInducer20 DOX-inducible expression plasmid used for cloning was generously provided by Dr. Stephen Elledge, Department of Genetics, Harvard Medical School (21).

ICD expression cDNAs encoding CYT-1 (amino acids 676–1308) and CYT-2 (amino acids 676–1294) isoforms (GeneCopoeia) were cloned into the lentiviral TA cloning vector Lenti6.3-V5 in frame with the 3′ V5 epitope tag (Life Sciences Technologies). Stable cell lines for the ICDs and the vector control (vector) were selected with blasticidin.

For NRG1 stimulation, cells were plated at 1 × 10⁶ cells per 100-mm plate. The following day, cells were incubated in serum-free OptiMEM medium for 48 hours, followed by incubation with 100 ng/mL NRG1. Sample buffer lysates normalized for protein concentration were analyzed by electrophoresis in 4% to 12% NuPAGE SDS–polyacrylamide midigels (Life Technologies Corporation). For immunoblotting, polyvinylidene difluoride membranes were blocked with 2% BSA in 10 mmol/L Tris-HCl, 50 mmol/L NaCl, 0.1% Tween 20, pH 7.4 (TBST), and incubated with anti–phospho-ERBB4 (Tyr 1056; ref. 22), phospho-ERBB4 Tyr 1284 (Cell Signaling Technology; #4757), ERBB4 (sc-283), GAPDH (Santa Cruz Biotechnology), phospho-MAPK (Thr202/Tyr204), or phospho-AKT (Ser473;Cell Signaling Technology) diluted 1:5,000 to 1:20,000 in TBST/2% BSA for 2 hours. Membranes were washed five times with TBST, incubated with horseradish peroxide–conjugated secondary antibodies in TBST/2% BSA for 1 hour, rinsed with TBST, and detected by chemiluminescence (SuperSignal West Pico Chemiluminescent Substrate; Pierce).

In Fig. 2A–C, cells were plated at 1,000 cells per well in 96-well plates. The next day, four wells per group were fed either serum-free OptiMEM medium or 5% horse serum containing medium, ± 100 ng/mL NRG1, and incubated for 5 days with refeeding day 2. Proliferation was assayed daily using the ATP-based CellTiter-Glo Luminescent Cell Viability Assay (Promega). The difference among groups after 5 days was determined by ANOVA followed by the Newman–Keuls multiple comparison test, with P < 0.05 considered to be statistically significant.

In Fig. 2E and F, 1,000 cells were seeded in 96-well plates in triplicate in complete medium. For scoring, cells were washed three times with PBS and then harvested with 0.25% Trypsin (Invitrogen), stained with Trypan blue, and viable cells were enumerated using a cell counter (The Countess; Invitrogen).

BD BioCoat Matrigel Invasion Chambers (BD Biosciences) with 8-μm pore PET Matrigel membranes were used. Inserts were hydrated in 500 μL of OptiMEM for 2 hours and transferred to with 5% horse serum–containing medium as chemoattractant. Cells were suspended in OptiMEM with 0.1% horse serum and plated at 50,000 cells per insert in triplicate for both Matrigel and control inserts (lacking Matrigel) in 24-well plates, followed by incubation for 24 hours. Noninvading cells were removed by scrubbing the upper membrane surface. Invading cells on the lower surface of membrane were stained with Diff-Quick (Invitrogen Corporation) and counted in three microscopic fields per membrane. The percentage of invasion was calculated as the mean number of cells invading through Matrigel insert membrane/mean number of cells migrating through control insert ×100. The invasion index is the percentage of invasion of test cells relative to control vector-infected MCF10A cells. Differences among groups were determined by ANOVA, followed by the Newman–Keuls multiple comparison test, with P < 0.05 considered to be statistically significant.

FL ERBB4 cell lines and controls were plated at 1 × 10⁶ cells per 100-mm plate and incubated in serum-free OptiMem for 48 hours. The next day, cells were incubated in fresh OptiMem with or without 100 ng/mL NRG1 for 2 hours. RNA was extracted with the RNeasy Plus Mini Kit (Qiagen). For ICD ERBB4 isoforms, RNA was extracted from ICD ERBB4 cell lines maintained in complete medium. RNA samples were analyzed by the Yale Center for Genome Analysis using the Illumina HumanHT-12 v4 Expression BeadChIP (Illumina Inc.), with more than 47,000 probes derived from NCBI RefSeq Release 38, and also legacy UniGene content. Both FL and ICD ERBB4 experiments were performed with two biologic replicates run in parallel, with each sample analyzed in technical duplicate. The microarray data are available at Gene Expression Omnibus GEO website through accession numbers: GSE57346 (FL ERBB4 experiment) and GSE57339 (for ICD ERBB4 experiment). The threshold for significant changes in gene expression was set at P < 0.05 and fold change (FC) > 1.5. T47D pLKO ERBB4 KD, pInducer20 ERBB4–expressing stables were similarly serum-starved and treated with NRG1 (100 ng/mL) for 2 hours in the presence of 100 ng/mL DOX (24 hours). RNA was isolated using the RNeasy Plus Mini Kit (Qiagen).

Data from gene-expression microarrays were analyzed through the use of Ingenuity Pathway Analysis (IPA) (Ingenuity Systems; www.ingenuity.com). Detailed procedures for each analysis are included in the legends of Fig. 3 and Table 3.

Total RNA was isolated using the RNeasy Plus Mini Kit (Qiagen) and reverse transcribed with the iScript cDNA Synthesis Kit from Bio-Rad using 1 μg of RNA per reaction. Universal TaqMan Master Mix (Applied Biosystems) was used to conduct quantitative real-time PCR (qRT-PCR) analysis. Primers included ADAMTSL4 (Hs00417524_m1), ALDH1A3 (Hs00167476_m1), BTG2 (Hs0098887_m1), CDCA5 (Hs00293564_m1), CDC20 (Hs00426680_m1), CENPF (Hs01118845_m1), CTGF (Hs01026927_g1), CYR61 (Hs00998500_g1), DHCR7 (Hs01023087_m1), DKK1 (Hs00183740_m1), ERBB4 (Hs00171783_m1), FZD2 (Hs00361432_s1), FZD5 (Hs00361869_g1), GAPDH (Hs02758991_m1), HMGCR (Hs00168352_m1), HMGCS1 (Hs00940429_m1), KRT14 (Hs00265033_m1), LDLR (Hs01092524_m1), MMP3 (Hs00968305_m1), MMP9 (Hs00234579_m1), MSBR3 (Hs00827017_m1), MXD4 (Hs01557630_m1), PHLDA1 (Hs00378285_g1), PKMYT1 (Hs00993620_m1), SERPINE2 (Hs00385730_m1), SOCS2 (Hs00919620_m1), SPARC (Hs00234160_m1), TP63 (Hs00978343_m1), TPX2 (Hs00234160_m1), TOP2A (Hs01032137_m1), WNT5A (Hs00998537_m1; Applied Biosystems). Relative mRNA expression was determined with the ΔC_t method, with GAPDH as the reference gene.

Cells were cross-linked with dimethyl 3,3′-dithiobispropionimidate (DTBP; Pierce); chromatin was extracted and sonicated to an average size of 300 to 500 bp; and individual ChIP assays were performed using antibodies to V5 protein tag and protein G-coupled magnetic beads (23). Sequencing libraries were produced by the Yale Center for Genome Analysis, using 15 to 18 cycles of amplification, gel purified from 2% agarose gels, quantified, and sequenced on an Illumina Gene Analyser II. Sequence tags (24 bp) were mapped to the human genome (hg19/NCBI Build 37) from the UCSC Genome Browser (http://genome.ucsc.edu/) using ELAND (24). Sequence tags were extended to 200 bp and converted to signal map files representing the integer count of mapped tags overlapping at each genomic position. The signal maps were scored using PeakSeq to identify factor binding sites (25). Statistical significance was calculated using a binomial test followed by the Benjamini–Hochberg correction for multiple hypothesis testing to yield a q value for each candidate region. High-confidence–bound regions were selected with a q value cutoff of 0.01, corresponding to an overall FDR of 1%. A q value cutoff of 0.05 was also used to identify a set of lower confidence regions.

Standard ChIP (26) was performed on 5 × 10⁹ CYT-1 and CYT-2 ICD cells cross-linked with DTBP. Nuclear extracts were divided into three aliquots and precipitated with anti-IgG, anti-histone, (Cell Signaling Technology), or anti-V5 tag (Invitrogen) antibodies using protein G magnetic beads. The beads were washed extensively and the antibody complexes were released from the beads. The DNA cross-links were reversed and DNA was purified using a Qiagen PCR kit. The DNA from each antibody reaction was used in quantitative PCR. The primers used were:

• APOE_F-1, Sequence:GCT ATC TTC CCA TCC GGA AC

• APOE_R-1, Sequence:CAT CTC TGC TGC TGC AGT CT

• SPARC_F-1, Sequence:CAG AGC TCC ACA GAA TGC AG

• SPARC_R-1, Sequence:CAC CCG TCT CTT CTT CTC GA

• STMN1_F-1 Sequence:TCC CAA AGT GCT GGG ATT AG

• STMN1_R-1 Sequence:GCA GGG TGC TGT CTT TGT CT

• ADAP1_F-1 Sequence:AAC ACT ACT GCC CGA TGG TC

• ADAP1_R-1 Sequence:CAG GTG CCA TCT CTT GAG G

• MXD4_F1 Sequence:TTT ACA GCC CAG GAA ACA GG

• MXD4_R1 Sequence:GGC AGG TTC TAG GTC AGT GG

Three biologically independent experiments were done for each ChIP. Binding for each target sequence was calculated as the percentage of input binding to that sequence.

ERBB4 has unusually broad signaling potential for a RTK owing to its atypical nuclear functions, and the diversification of ICD isoforms by the CYT-1/CYT-2 splice choice. To compare the functionality of FL versus ICD, and CYT-1 versus CYT-2 isoforms, we produced stable cell lines overexpressing different ERBB4 isoforms. MCF10A cells were used first because they express little or no endogenous ERBB4, and because they are a normal-like, nontransformed human mammary cell line.

We first engineered MCF10A cells to express FL CYT-1 and CYT-2 ERBB4 with the cleavable JM-a domain. DNA-mediated gene transfer yielded robust expression of ERBB4 that was stable in T47D cells, which possess endogenous ERBB4. However, ERBB4 expression was lost within 3 days of DNA-mediated gene transfer in MCF10A cells, consistent with a counter selection against high ERBB4 in MCF10A background (Fig. 1A). Nonetheless, expression of FL ERBB4 was stable in MCF10A cells 3 days after infection with lentivirus (which integrates efficiently) and after puromycin selection. Engineered ERBB4 mRNA expression was higher than endogenous expression in T47D cells (Fig. 1B). Although mRNA levels for FL CYT-1 and CYT-2 were comparable (Fig. 1B), CYT-2 ERBB4 protein was slightly higher at steady state, and this difference was augmented by stimulation with NRG1 (Fig. 1C). This is consistent with the reports that CYT-2 protein is more stable, because it lacks the ubiquitin ligase–binding site present in CYT-1 (27, 28).

NRG1-induced phosphorylation of ERBB4 was greater and more sustained in cells expressing CYT-2 than CYT-1. Vector cells responded with increased MAPK phosphorylation (Thr202/Tyr204) and AKT phosphorylation (Ser473), presumably through activation of endogenous ERBB3. Expression of CYT-1 or CYT-2 enhanced the NRG1 response (Fig. 1C) with greater phosphorylation of MAPK in CYT-2. There was little or no difference between control and ERBB4-expressing MCF10A cell lines in relative protein levels of cyclin D1, E-cadherin, vimentin, phospho-YAP(Tyr357), or phospho-YAP(Ser127) over a 24-hour period (data not shown).

Proliferation rates were similar in FL CYT-1, CYT-2, and control cell lines grown in 5% horse serum, with or without addition of NRG1 (Fig. 2A). However, in serum-free medium, FL CYT-2 cells grew significantly faster than vector control or FL CYT-1 cells (Fig. 2B), both in the absence and presence of NRG1 (P < 0.001). With NRG1, FL CYT-1 cells grew significantly more slowly than vector cells (P < 0.01) after 5 days (Fig. 2B), so CYT-1 ERBB4 actively reduces growth of these cells. In Boyden chamber Matrigel invasion assays, FL CYT-2 MCF10A cells scored significantly higher (24.30 ± 1.67%; P < 0.05) than both control vector (13.1 ± 1.70%) and FL CYT-1 MCF10A cells (15.33 ± 2.50%; Fig. 2C). The invasion index of FL CYT-2 MCF10A cells (1.84) was nearly twice that of vector control cells (1.0), but similar for FL CYT-1 MCF10A and control cells (1.1). Overall, expression of FL CYT-1 reduces proliferation in the absence of serum and the presence of NRG1, whereas FL CYT-2 promotes proliferation and invasion, concordant with somewhat higher relative MAPK signaling.

The NRG1-activated outputs of the FL JM-a isoforms are a composite of conventional RTK signaling and the activities of the m80 and s80 forms released by cleavage. To evaluate the signaling activities of soluble ICD isoforms, we expressed V5-tagged constructs beginning just beyond the basic residues marking the cytoplasmic face of the transmembrane domain (Fig. 2D). These are structurally similar to the forms produced by γ-secretase cleavage of ERBB4, but the actual amino terminus of ERBB4 s80 has not been determined by peptide sequencing. We have reported that these ICDs are Tyr-phosphorylated, concordant with earlier findings (29). Similar to cells expressing FL ERBB4, ICD CYT-2 MCF-10A lines proliferate more rapidly than the vector or ICD CYT-1 cell lines, whereas ICD CYT-1 cell lines were growth suppressed compared with the ICD CYT-2 lines or control lines in medium containing 5% horse serum (Fig. 2E). The ICD CYT-2 cells invaded through a Matrigel membrane more efficiently (70% ± 1.8) than the vector or ICD CYT-1 lines, which had low invasive potential.

We used transcriptional profiling to identify the genes commonly and differentially affected by expression of the two FL ERBB4 isoforms. Nonsupervised hierarchical clustering of the top genes up- or downregulated by ERBB4, based on all features with a SD/mean > 0.05 in biologic replicates, grouped vector, CYT-1/CYT-2 without NRG1 and CYT-1/CYT-2 with NRG1 (data not shown). Under serum-free conditions, stimulation with NRG1 (100 ng/mL) for 2 hours did not significantly change expression of genes in vector-MCF10A cells but significantly altered gene expression in FL CYT-1 and FL CYT-2 MCF10A cell lines (Supplementary Fig. S2A, left; Supplementary Table S1). Without NRG1, expression of 61 genes in CYT-1 MCF10A and 143 genes in CYT-2 MCF10A cells was significantly altered relative to vector control cells (adjusted P value < 0.05; FC > 1.5). Of these, 52 genes were altered in both CYT-1 and CYT-2 cells (Supplementary Fig. S2A, center). With NRG1 stimulation, 37 genes in CYT-1 and 57 genes in CYT-2 MCF10A cells were significantly different, of which 24 were commonly affected in both, in comparison with NRG1-stimulated vector-MCF10A cells (Supplementary Fig. S2A, right). Overall, genes encoding cytokines IL8, CCL20, CXCL1, matrix metalloproteinase MMP3, protease inhibitor SERPINE2, signaling adaptor VAV3, CTGF, and phosphorylation-dependent ubiquitin ligase adaptor FBX032, were the top genes exclusively altered in CYT-2 MCF10A, whereas poly A–binding protein PABPC1 and secreted serine protease kallikrein-10 (KLK10) were exclusively altered in CYT-1 MCF10A cells (Supplementary Table S1). qRT-PCR assays validated 10 of 11 candidate transcriptional changes in this group, including SERPINE2 and MMP3 (Supplementary Fig. S1A). Some of the genes shared between FL ERBB4 CYT-1 and CYT-2 are listed in Table 1.

There were many transcriptional differences among MCF10A ICD CYT-1 or CYT-2 and vector control cells (Supplementary Fig. S2B). Of note, 918 genes were altered in common between the ICD CYT-1 and ICD CYT-2 cells, with 482 ICD CYT-1–specific genes and 168 ICD CYT-2–specific genes (Supplementary Fig. S2B). Some of the shared top upregulated (13/20) and downregulated (16/20) genes are listed (Table 2). A subset of genes was validated by qRT-PCR (Supplementary Fig. S1B).

FL and ICD ERBB4 differ in ligand dependence (ICD are constitutively Tyr phosphorylated; refs. 13, 29), subcellular localization, and stability. Overall, NRG1-stimulated cells expressing FL ERBB4 (CYT-1 and CYT-2) altered fewer pathways than the ICD, and with less significance. Endogenous ERBB3-regulated genes were factored out of our analysis by comparison against NRG1-treated controls. In comparison with the respective vector controls, genes encoding transforming growth factor alpha (TGFα), angiopoietin-like 4 (ANGPTL4), plasminogen activator inhibitor SERPINE1, and signaling protein PHLDA1 were among the top commonly affected genes in both FL and ICD, CYT-1 and CYT-2 ERBB4–expressing MCF10A cells (Supplementary Tables S1 and S2). HSP70-encoding HSPA1A was the only gene that was altered exclusively in CYT-1 isoform in both FL and ICD forms, consistent with a stress response. Similarly, genes, including chemokine receptor CXCL1, growth-activating MYC and FOSB, and neural receptor/adhesion protein CNTNAP2, were altered uniquely in cells expressing FL and ICD CYT-2.

We used Ingenuity Systems IPA analysis to evaluate pathways altered by ICD ERBB4 expression. We chose to focus on ERBB4 ICD because the transcriptional changes elicited by the ICD were both stronger and more statistically significant than FL ERBB4, thereby facilitating a more robust analysis. IPA analysis tests for overrepresentation of genes in a particular annotated process to infer altered pathways from patterns of gene expression (Supplementary Fig. S3). Interestingly, both CYT-1– and CYT-2–upregulated genes in cholesterol biosynthesis (mevalonate pathway) and ketogenesis. ICD CYT-1 was uniquely linked to increased colorectal cancer metastasis signaling and pancreatic carcinoma signaling and epithelial-to-mesenchymal transition (EMT)–associated genes. ERBB4 CYT-2 ICD uniquely affected cyclins/cell-cycle regulation and DNA damage induced 14-3-3 signaling through increases in cyclins and CDK1. ERBB4 ICD CYT-2 and ERBB4 FL CYT-2 both regulated pathways involved in cell cycle and cell-cycle regulation (Supplementary Fig. S3).

IPA analysis was used to test for overlap of ICD ERBB4–induced genes with genes grouped according to functional annotation. ERBB4 ICD CYT-1 was connected with increases in processes, including regulation of microtubules, cytoskeleton, metastasis, neoplasia, cell death, and tumorigenesis, and decreases in skin abnormality, and cell cycle, and ploidy processes (Table 3, left; Supplementary Table S4). ERBB4 ICD CYT-2 was associated with increases in proliferation, angiogenesis, blood vessel development, cardiovascular development, M phase, cytostasis, and malignant tumor growth, and decreases in ploidy. The genes comprising each pathway are listed in Supplementary Tables S2 and S4.

Ingenuity Upstream Regulator Analysis (Table 3, right, and Supplementary Table S3) predicted several transcription factors to be activated in both ERBB4 CYT-1 ICD and CYT-2 ICD HIF1α, SREBF1, MTPN, E2F1, FOXO1, SREBF2, and NFκBI. Network analysis of gene expression involving predicted transcriptional regulators suggested activation of YAP/TAZ (Hippo pathway), HIF1α, and TGFβ in both ERBB4 CYT-1 ICD and CYT-2 ICD (Fig. 3). ERBB4 CYT-1 ICD was associated with factors, including components of the NFκB pathway, EMT, and TGFβ. ERBB4 CYT-2 ICD was predicted to repress negative regulators of the cell cycle—CDKN2A and RB1 (Fig. 3).

ERBB4 is unusual among RTKs in the ability of the soluble ICD to associate with transcription factors and coregulators and directly participate in gene regulation. Only a small number of nuclear ERBB4 regulatory targets have been identified, so we performed ChIP-seq experiments with ERBB4 ICD CYT-1 to identify the global set of DNA sequences to which ERBB4 binds. Because these binding interactions are likely to be indirect, both protein/protein and protein/nucleic acid cross-linking agents were used before shearing and ChIP. Two biologic repeats were conducted for the ICD CYT-1. Although there were many fewer reads in the repeat experiment, 236 immunoprecipitated DNA segments were assigned to the same gene between the two sets, and 94 of these were within the same annotated feature (data not shown).

Candidate binding sites based on ChIP-Seq were found in intergenic regions, introns, and promoters. The intergenic regions were most numerous, but they are of uncertain significance because we have not yet attempted to validate them by ChIP. As we were most interested in binding targets with functional impact on gene transcription, initial validation experiments were confined to segments of DNA within 20 kb 5′ of the transcriptional start site (TSS) of genes that overlapped with genes altered in RNA profiling (Supplementary Table S5). Of the 10 genes that were tested, SPARC, SERPINE 1, and MXD4 were upregulated ICD CYT-1–specific genes, and STMN1 was an ICD CYT-2–specific gene (Supplementary Fig. S1). ADAP1/CENTA1, SOCS2, and HS3ST1 were downregulated by both ICD isoforms whereas APOE, CDKN2AIPNL, and TK1 were upregulated by both. In ChIP validation experiments in cells expressing CYT-1 and CYT-2, no ERBB4 enrichment was seen over the histone and IgG controls for SERPINE1, ANGPTL4, CDKN2AIPNL, HS3ST1, and TK1 (data not shown). Specific ChIP of ERBB4 was detected for each of the other sites analyzed (Fig. 4). ERBB4 ChIP enriched for ADAP1/CENTA1 and APOE equivalently from cells expressing CYT-1 ICD and CYT-2 ICD, consistent with the upregulation of APOE and downregulation of ADAP1/CENTA1 in CYT-1 and CYT-2 ICD lines. SPARC and STMN1 are preferentially upregulated in CYT-1 and CYT-2 cells, respectively, but were also evenly enriched by ChIP from cells expressing either of the isoforms. Finally, MXD4 was preferentially enriched by ERBB4 ChIP from CYT-1 ICD cells, mirroring CYT-1 preferential transcriptional upregulation.

ERBB4 is not known to be a sequence-specific DNA-binding protein, but instead modulates transcription through association with DNA-binding proteins and transcriptional coregulators. A search for predicted transcription factor–binding sites and transcription factor ChIP–defined sites [Human Genome Build (GRCh37/hg19)] within our validated ChIP-Seq targets identified several sites of interest. The ADAP1 ERBB4–binding site includes a CEBPB site found in four of four cell lines tested by CEBPB ChIP-Seq (30). Within the SPARC ERBB4–binding site are BRACH, ATF6, ATF and XBP1 predicted transcription factor–binding sites (31). Within the STMN1 ERBB4 site is a TCF4 ChIP-Seq site that was validated in one of the two cell lines tested. APOE ERBB4 sites are both HNF4 and ETS1 sites that were found by transcription factor ChIP-Seq in K562 cells, and the MAD ERBB4–binding site includes ChIP-Seq–defined sites for MYC and CTCF in K562 cells. Overall, these results identify five new gene candidates for direct regulation through ERBB4 ICD, and a number of other candidates remain to be tested (Supplementary Table S5). Of these, both ADAP1 and STMN1 have been implicated in intracellular signaling and microtubule regulation, APOE in lipoprotein metabolism, and SPARC, in growth and calcium regulation in the extracellular matrix. MXD4 encodes a MAD that antagonizes MYC. MXD4 may contribute to growth suppression in CYT-1 ICD cells, in which it was preferentially expressed in validation experiments (Supplementary Fig. S1B).

We initially chose to analyze different ERBB4 isoforms expressed in MCF10 cells, as they do not express significant endogenous ERBB4. These cells have a basal phenotype. We next evaluated ERBB4-induced transcription in a more biologically relevant context, T47D cells, which express endogenous ERBB4 and which have a luminal phenotype. We knocked down endogenous ERBB4 and reintroduced vectors encoding specific ERBB4 isoforms (Fig. 5A). Knock-down (KD) by 3′UTR-specific shRNA resulted in >50% reduction in ERBB4 protein levels in T47D ERBB4 KD stable cell lines, but NRG1-induced ERBB4 Tyr phosphorylation (B4 sh3 and pI20 V) was maintained and possibly slightly increased in a possible compensatory circuit. Introduction of vectors expressing FL ERBB4 CYT-1 (B4 sh3, pI20 C1) or CYT-2 (B4 sh3, pI20 C2) resulted in greater expression of ERBB4 protein (Fig. 5A), higher basal ERBB4 phosphorylation, and greater NRG1-induced ERBB4 phosphorylation.

We determined whether ERBB4-regulated genes identified in MCF10A background were similarly regulated in T47D cell lines. The genes most responsive to NRG1 or ERBB4 expression in T47D cells included CTGF, CYR61, DKK1, LDLR, SPARC, HMGCR, HMGCS1, TP63, and KRT14, all of which were upregulated except for SPARC, which was reduced with ERBB4 expression (Fig. 5B). Most followed the same trend seen in MCF10A cells with the exception of SPARC, which was up in MCF10A cells and down in T47D cells, and TP63 and SOCS2, which were both reduced in MCF10A cells and up in T47D cells. In addition, PHLDA1, TOP2A, SOCS2, and SERPINE2 increased with NRG1/ERBB4 in the T47D cells (Supplementary Fig. S1C). Of the top genes altered in MCF10A ERBB4 ICD microarray analysis, CTGF, CYR61, PHLDA1, SERPINE2, and TOP2A all validated, but isoform-specific effects were not as strong as with the ICD. The ERBB4-altered genes overlapping between MCF10A and T47D cells mainly grouped into the mevalonate/cholesterol pathway (HMGCR, HMGCS1, LDLR, and DHCR7) and the YAP/Hippo pathway (CTGF, CYR61, and SPARC) in addition to luminal/basal markers (KRT14 and TP63), DKK1 (Wnt-negative regulator), and PHLDA1, an important negative regulator and effector of Aurora A kinase in breast cancer (Fig. 5B).

Because the cholesterol pathway genes HMGCR and LDLR were among the novel and prominent genes altered in both MCF10A and T47D cells by ERBB4 and its ligand, NRG1, we investigated their expression in pregnant and lactating mouse mammary glands using existing gene-expression data (32). Erbb4 and its ligands are highly expressed and necessary for normal mammary development (33, 34). Intriguingly, Hmgcr and Ldlr expression is elevated during late pregnancy and early lactation along with Erbb4 and several of its ligands (Fig. 5C).

We report here the first direct comparison of the four JM-a FL and artificially truncated CYT-1 and CYT-2 ERBB4 isoforms in an isogenic background. In MCF10A cells, which lack endogenous ERBB4, CYT-1 ERBB4 suppresses growth whereas CYT-2 ERBB4 increases cell proliferation and promotes invasion. Transcriptional profiling revealed genes that are commonly altered by multiple ERBB4 isoforms and also genes that are uniquely affected by each ERBB4 isoform. ERBB4 was knocked down in T47D cells and ERBB4 isoforms reexpressed to confirm functionally relevant genes in a luminal-like cell background in which ERBB4 is normally expressed. Novel ERBB4 ICD DNA–binding regions and candidate ERBB4 target genes were identified by ChIP-Seq.

Our findings of growth inhibitory potential of CYT-1 and growth promoting potential of CYT-2 are consistent with studies in HC11 mouse mammary epithelial cells expressing s80 (35), and with comparisons of MCF7 cells expressing FL JM-a CYT-1 and CYT-2 (36) and MCF10A cells expressing CYT-1 s80 (37). Moreover, NR6 mouse fibroblasts expressing JM-a CYT-2 ErbB4 showed enhanced growth compared with JM-b CYT-2, consistent with a potent role for the intracellular cleaved forms in growth promotion by ERBB4 (38). Although ERBB4 is involved in mammary differentiation (33, 34, 39), we did not find that NRG1-activated FL ERBB4 promotes differentiation or EMT. This may be because MCF10A cells express little or no prolactin receptor or STAT5, both of which regulate mammary differentiation cooperatively with ERBB4 (33, 40). Similarities in biologic activities of FL and ICD isoforms suggest that the dominant signaling output of ERBB4 JM-a is mediated by the s80 cleavage products, and implies that ERBB4 JM-a will have very different signaling qualities from JM-b, which is not cleaved.

Major categories of genes regulated by ERBB4 encode proteases/protease inhibitors (MMP3, SERPINE2, and KLK-10), YAP/Hippo pathway targets (CTGF, CYR61, and SPARC), mevalonate/cholesterol pathway genes (HMGCR, HMGCS1, LDLR, and DHCR7), and cytokines (IL8, CCL20, and CXCL1). In FL ERBB4 cell lines, a cleaved ERBB4 band was detected even in the absence of NRG1, indicating some basal ERBB4 activity. In comparison of nonstimulated ERBB4 MCF10A versus vector-MCF10A, genes encoding Rho-GEF VAV3, and protease inhibitor SERPINE2 were upregulated by CYT-2. VAV3, along with VAV2, control a lung metastasis-specific transcriptional program in breast cancer (41), and VAV2 is regulated in hippocampal neurons by NRG1-ERBB4 signaling (42). SERPINE2 is upregulated by oncogenic activation of the MAPK pathway and has been proposed as a therapeutic target in colorectal cancer. Top NRG1-dependent genes unique to CYT-2 cells include MMP3, SERPINE2, IL8, CCL20 (upregulated), and FBX032 (downregulated). The invasive properties of MMP3, upregulation of SERPINE2 by MAPK, and proangiogenic disposition of VAV-3 agree with the higher MAPK activity, pro-proliferative, and invasive behavior of CYT-2 cells. Activation of ERBB4 by cytokines is well documented (43, 44). We report a converse upregulation of cytokines IL8 and CCL20 by CYT-2. FBXO32 is a novel TGFβ/SMAD4 target gene and a tumor suppressor. Thus, modulation of proteases, cytokines and TGFβ pathways by CYT-2 ERBB4 may contribute to the highly proliferative and invasive phenotype of these cells.

We found that the free ERBB4 ICD is a much more potent transcriptional regulator than FL ERBB4, which may be associated with its constitutive Tyr phosphorylation and chronic signaling activity. There was overlap of predicted pathways in ERBB4 CYT-2 ICD and FL CYT-2 but no overlap for unique CYT-1 ERBB4 ICD and FL. CYT-1 contains a binding site p85 PI3K, and the PI(3′) kinase requires membrane localization for signaling, so CYT-1 may have unique signaling roles at the membrane that are bypassed following release of the s80 ICD. The overlapping ERBB4 CYT-2 pathways were cell-cycle related, consistent with the fact that ERBB4 CYT-2 increases proliferation in MCF10A cells.

The predicted upstream transcription factors activated or repressed by ERBB4 ICD (Fig. 3) include YAP and HIF1α, both known to bind ERBB4 (45, 46). TWIST1 and SNAI1 were predicted as active only in ERBB4 CYT-1 ICD, raising the possibility that CYT-1 has a role in cancer progression through EMT, consistent with the pathway analysis (Table 3). In contrast, many of the predicted CYT-2 ICD pathways were connected with cell cycle and proliferation (Table 3), and ERBB4 CYT-2 ICD gene expression predicted repression of CDKN2A and RB1 (Fig. 3), critical negative regulators of cell-cycle progression. This prediction is reinforced by the increase in proliferation and M phase–associated genes (Table 3) as well as the finding that ERBB4 ICD CYT-2 increases proliferation of MCF10A cells (Fig. 2). These findings identify candidate mediators for the differential growth regulation by CYT-1 versus CYT-2, which is not explained by major differences in MAPK or PI3K/AKT pathway signaling (Fig. 1C).

We were surprised to find that ERBB4 upregulated several enzymes in the mevalonate/cholesterol pathway, as this is a novel function of ERBB4. The upregulation of HMG-CoA reductase (HMGCR), HMG-CoA synthase (HMGCS1), 7-dehydrocholesterol reductase (DHCR7), and low-density lipoprotein receptor (LDLR) could result from either direct or indirect transcriptional regulation by ERBB4. Interestingly, sterol regulatory element–binding proteins 1 and 2 (SREBF1 and SREBF2), major transcriptional regulators of enzymes critical for sterol biosynthesis, including LDLR, HMGCS1, and HMGCR, were two of the top predicted activated transcription factors in the IPA analysis. ChIP-seq identified ERBB4 binding very close to the TSS (62–776 bp) of SREBF1 and SREBF2, although expression of these genes was not significantly altered in the MCF10A ERBB4 microarrays. DHCR7 was also identified in the CYT-1 ERBB4 ChIP-seq experiment at approximately 70 kb from the putative TSS. In addition, APOE, another component of cholesterol metabolism, was transcriptionally upregulated in MCF10A ERBB4 ICD cells and was validated from the ERBB4-ChIP. HMGCR and HMGCS1 are regulated by PPARα (47), whose activity is modulated by NCOR1, known to form a nuclear complex with ERBB4 (17, 48). Collectively, these data lead us to hypothesize that ERBB4 interacts with SREBP1/2 to directly regulate expression of mevalonate/cholesterol genes and we are currently exploring this mechanism.

In addition to expressing ERBB4 isoforms in MCF10A cells that otherwise lack ErbB4 expression, we also reduced endogenous ERBB4 in luminal T47D cells and then reexpressed specific FL ERBB4 CYT-1 or CYT-2 isoforms. Expression analysis revealed that, similar to MCF10A cells, genes regulating cholesterol and Hippo pathway genes are also significantly altered by ERBB4 expression in T47D cells. Notable Hippo pathway genes include CTGF, CYR61, and SPARC, all of which are YAP/TEAD–regulated genes (49). We are currently investigating the mechanism and biologic consequences of the ERBB4-YAP biochemical interaction reported previously (46). In T47D cells, ERBB4 also upregulated HMGCR, HMGCS1, and LDLR genes, which are chief regulators of the cholesterol pathway. As ERBB4 plays a critical role in mammary gland development during pregnancy and lactation (33, 34), and the temporal pattern of Hmgcr and Ldlr expression is similar to that of Erbb4 and its ligands in mouse mammary glands, there is a potential functional relationship between ERBB4 and the mevalonate pathway. ERBB4 regulates milk proteins through activation of mammary differentiation factor STAT5, so we hypothesize based on these findings that ERBB4 coordinately regulates cholesterol synthesis as another nutritional component of milk. Indeed, cholesterol is synthesized locally in mammary glands in addition to the liver (50). These findings in normal basal-like MCF10A and cancerous luminal T47D cells are notable in light of the recent finding that cholesterol promotes breast cancer growth and metastasis (51), and we further speculate that ERBB4 regulation of cholesterol metabolism, normally occurring in mammary development, could be hijacked during tumorigenesis or cancer progression. It was recently shown that YAP/TAZ activity can be controlled by the SREBP/mevalonate pathway (52), raising the possibility that ERBB4 could be a RTK facilitating mevalonate/Hippo signaling.

A major goal of this work was to identify common and differential cellular responses associated with ERBB4 signaling in mammary background, and to find candidate pathways and protein mediators of these responses. This work underscores the diverging phenotypes of CYT-1 and CYT-2 isoforms despite related responses linked to activation of Hippo and HIF1α pathway genes. Differential ERBB4 isoform-dependent changes implicate cytokines, growth factors/mitotic cell-cycle regulators, and extracellular matrix mediators. The linkage of ERBB4 with cholesterol metabolism, intracellular cytoskeletal regulators, and novel candidate target genes detected by ChIP, as well as TGFβ and NFκB pathways significantly extends the universe of potential processes connected with nuclear signaling by ERBB4. Our findings implicate ERBB4 as a coordinate regulator of growth through the Hippo pathway. The possibility that ERBB4 regulates cholesterol metabolism may have important implications for milk production and, more generally, anabolic processes in normal epithelia and cancer.

No potential conflicts of interest were disclosed.

Conception and design: V.B. Wali, M. Gilmore-Hebert, D.F. Stern

Development of methodology: V.B. Wali, M. Gilmore-Hebert

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V.B. Wali, J.W. Haskins, M. Gilmore-Hebert

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V.B. Wali, J.W. Haskins, M. Gilmore-Hebert, J.T. Platt, Z. Liu

Writing, review, and/or revision of the manuscript: V.B. Wali, J.W. Haskins, M. Gilmore-Hebert, D.F. Stern

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V.B. Wali

Study supervision: V.B. Wali, D.F. Stern

This work was supported by USPHS grant R01 CA80065 from the National Cancer Institute, and NIH training grant T32GM07223 (to J.W. Haskins).

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.