Melanoma adapts to RAF/MEK inhibitors through FOXD3-mediated upregulation of ERBB3

Identifying the FOXD3 transcriptome in melanoma.

To understand the transcriptional impact of FOXD3 in melanoma cells, we utilized a microarray approach. We collected RNA from 3 unrelated mutant BRAF melanoma cell lines (WM115, WM793, and A375) that were engineered to inducibly express FOXD3 or the control gene β-galactosidase (LacZ) after 5 days of transgene induction (Figure 1A). This time point was chosen based on maximal phenotypic changes previously observed (22). Comparison of gene signatures among the 3 cell lines produced approximately 2,600 common genes differentially regulated by FOXD3-expressing cells compared with the LacZ controls (Figure 1B and Supplemental Table 1; supplemental material available online with this article; doi: 10.1172/JCI65780DS1). Since a large number of altered genes may represent secondary targets of FOXD3, we sought to narrow the scope of FOXD3-regulated genes to direct transcriptional targets. We performed ChIP-seq on V5-tagged FOXD3 IP from WM115TR-FOXD3. The results showed specific, reproducible enrichment foci across the genome with a preference for promoter regions (Figure 1C) and bidirectional promoters (data not shown). Analysis of genes located proximal to FOXD3 enrichment sites and showing regulation by FOXD3 indicated a preference for genes involved in focal adhesions, ECM-receptor interactions, MAPK and mTOR signaling, and other processes involved in cancer (Table 1 and Supplemental Figure 1), suggesting that FOXD3 is able to act as a major orchestrator of transcription in melanoma.

ERBB3 is a direct transcriptional target of FOXD3.

Based on our previous data showing that FOXD3 promotes resistance to BRAF inhibition (23), we focused on genes that were druggable, given the translational nature of the study. We identified ERBB3 as a target upregulated by FOXD3 in the expression arrays and strongly enriched by FOXD3 in the ChIP-seq analysis (Figure 2A and Supplemental Table 1). ERBB3 expression is increased in response to targeted therapies such as lapatinib in breast cancer and gefitinib in lung cancer (24–27) and is also important for melanoma survival and proliferation (28, 29). ChIP-seq analysis showed that the first intron of ERBB3 was enriched by FOXD3. This region is well conserved between species and functions as an enhancer region for ERBB3 (30–32). Quantitative PCR (qPCR) showed dramatic enrichment of intron 1 over normal IgG only following FOXD3 expression (Figure 2B). Importantly, the V5 antibody did not enrich the promoter of an irrelevant gene, β-actin (ACTB), in a doxycycline-dependent (Dox-dependent) manner, verifying the specificity of FOXD3 enrichment. Enhanced expression on our microarrays coupled with binding of FOXD3 to the enhancer region suggests that FOXD3 directly upregulates the transcription of ERBB3. In support of this, IP of RNA polymerase II phosphoserine 2 (RNA pol II pSer2), a marker for transcriptional elongation (33, 34), significantly enriched ERBB3 intron 1 in cells expressing FOXD3 (Figure 2C). In addition we found that FOXD3 increased the expression of ERBB3 at both the mRNA (Figure 2D) and protein (Figure 2E) levels in WM115TR-FOXD3 cells. Similarly, induction of FOXD3 consistently enhanced the expression of ERBB3 in a panel of melanoma cells while consistently having no effect on the expression of other receptor tyrosine kinases (RTKs) known to convey resistance to targeted therapies (refs. 13–15, 25, and Figure 2F).

ERBB3 expression is enhanced by RAF/MEK inhibition in melanoma.

Previous studies showed that FOXD3 is upregulated in response to BRAF/MEK inhibition in mutant BRAF melanoma (22). We sought to determine whether inhibition of BRAF or MEK1/2 could recapitulate the effects on ERBB3 seen by the ectopic expression of FOXD3. Knockdown of BRAF by siRNA resulted in an increase in ERBB3 protein in WM115 cells (Figure 3A). Similarly, inhibition of BRAF or MEK with PLX4032 or AZD6244, respectively, induced both FOXD3 and ERBB3 in WM115 and 1205Lu cells (Figure 3B). This observation was reinforced by microarray data showing upregulation of ERBB3 in response to BRAF knockdown (Supplemental Figure 2A). Similarly, increased ERBB3 mRNA expression was also observed in 1205Lu cells treated with PLX4032 or AZD6244 (Figure 3C). In both WM115 and 1205Lu cells, the ERBB3 signal on microarrays was also reduced by FOXD3 targeting siRNA, both alone or in combination with BRAF siRNA or PLX4720 (Supplemental Figure 2, A and B). Another cell line, A375, showed enhanced surface expression of ERBB3 (Figure 3D) as well as a concomitant upregulation of ERBB3 mRNA in response to either PLX4032 or AZD6244 (data not shown). These data indicate that BRAF/MEK inhibition, like FOXD3 overexpression, positively regulates ERBB3 expression levels.

NRG1/ERBB3 signaling to AKT is enhanced by RAF/MEK inhibition in a FOXD3-dependent manner.

To assess the impact of FOXD3 expression on ligand-induced ERBB3 signaling, we treated WM115TR-FOXD3 cells with increasing concentrations of NRG1β, a potent ERBB3 ligand (35), in either the presence or absence of FOXD3 induction. Upregulation of ERBB3 by FOXD3 was associated with an enhanced sensitivity to NRG1β at all doses analyzed, as assessed by phosphorylation of ERBB3 (Figure 4A). Phosphorylated YXXM motifs in ERBB3 recruit PI3K, leading to activation of AKT (36). Consistent with enhanced ERBB3 signaling, FOXD3-expressing cells displayed enhanced NRG1β-dependent phosphorylation of AKT (Figure 4A).

To determine whether inhibition of BRAF could elicit a similar result in melanoma cells, WM115 cells were treated overnight with PLX4032 to induce endogenous FOXD3 and ERBB3, or with vehicle DMSO. PLX4032 treatment increased the sensitivity of ERBB3 to NRG1β and also enhanced AKT phosphorylation in WM115 (Figure 4B) and A375 cells (Supplemental Figure 3A). PLX4032 not only enhanced the intensity of response to NRG1β stimulation (Figure 4B), but also the duration of downstream AKT phosphorylation (Supplemental Figure 3B). A transient increase in ERK1/2 phosphorylation was observed in PLX4032-treated cells after stimulation with NRG1β, but this was largely dissipated within 1 hour (Supplemental Figure 3B). Similar to PLX4032, treatment of cells with AZD6244 enhanced both ERBB3 and AKT phosphorylation in response to NRG1β stimulation (Figure 4C). The enhancement of NRG1β/ERBB3 signaling was seen in multiple cell lines in response to either PLX4032 or AZD6244 pretreatment (Figure 4C, Supplemental Figure 3, C and D, and data not shown). Of note, phosphorylation of AKT was potently induced in melanoma cells regardless of PTEN status, as A375 cells are PTEN competent, while WM115 and 1205Lu cells are PTEN deficient. Importantly, phosphorylation of p70/p85 S6-kinase and S6 ribosomal protein were inhibited by treatment with PLX4032 or AZD6244, but restored by treatment with NRG1β (Figure 4C and Supplemental Figure 3C), indicating a restoration of translational activity by NRG1β/ERBB3 signaling. In addition to NRG1β, enhanced ERBB3 and AKT activation in PLX4032-treated cells was also observed following stimulation with NRG1α and neuroglycan (Supplemental Figure 4).

We next examined the temporal relationship among RAF inhibition, FOXD3 induction, and enhanced NRG1β/ERBB3 signaling. Induction of FOXD3 could be seen as early as 2 hours after treatment with PLX4032 and steadily increased up until 16 hours. Enhanced NRG1β/ERBB3 signaling could be observed after 4 hours of PLX4032 treatment, gradually increasing through 16 hours (Figure 4D). These data suggest that FOXD3 upregulation precedes enhancement of NRG1β/ERBB3 signaling. Importantly, depletion of FOXD3 by siRNA ablated ERBB3 protein expression, both basal and PLX4032 induced, and prevented responsiveness to NRG1β stimulation in both WM115 and 1205Lu cells (Figure 4E and Supplemental Figure 3E).

RAF inhibitors enhance ERBB3 phosphorylation in vivo.

We extended our analysis of RAF inhibitors on ERBB3 phosphorylation to the in vivo setting. First, we administered PLX4720 to nude mice with intradermal A375 xenografts for 5 days. PLX4720 is the nonclinical analog for vemurafenib. Analysis of the harvested tumors by immunohistochemistry (IHC) showed a statistically significant (P < 0.05) increase in the proportion of cells with high levels of membrane-associated staining for phosphorylated ERBB3 (phospho-ERBB3) in PLX4720-treated tumors compared with controls (Figure 5A). These findings indicate that increased ERBB3 sensitivity following RAF inhibition in melanoma cells occurs in vivo as well as in vitro.

Next, to analyze whether enhanced ERBB3 phosphorylation occurs in patients receiving vemurafenib, IHC was performed using biopsies taken before vemurafenib treatment, 15 days on-treatment, and following disease progression. In 2 patients analyzed, we observed low ERBB3 phosphorylation prior to treatment. A statistically significant increase in ERBB3 phosphorylation was observed in 1 of the 2 patients following treatment with vemurafenib and persisting through relapse (Figure 5B). An additional biopsy from a long-term on-treatment patient, who had not yet progressed, also showed upregulation of phospho-ERBB3 staining (Figure 5B). This suggests that ERBB3 phosphorylation can be enhanced in patients undergoing vemurafenib treatment.

We extended our analysis to a larger set for which pretreatment and progression samples were available. This set of 9 paired samples came from mutant BRAF melanoma patients who had received either RAF inhibitor or combined RAF/MEK inhibitor. The latter combination has been shown to provide increased progression-free survival in mutant BRAF melanoma patients compared with RAF inhibitor alone (37). Three out of the 9 progression samples showed a statistically significant increase in ERBB3 phosphorylation compared with the match pretreatment sample (Figure 5C). Statistical analysis across samples using an ordered logistic regression model with random intercept for each patient showed that progression samples have 2.16 times (95% CI, P < 0.001) higher odds of having greater scores compared with pretreatment and that on-treatment samples have 3.30 times (95% CI, P < 0.001) higher odds of having greater scores compared with pretreatment (Figure 5D). These findings suggest that upregulation of ERBB3 is maintained in some cases of chronic vemurafenib treatment.

ERBB3 activation promotes resistance to RAF/MEK inhibitors.

Increased expression and activation of RTKs has been associated with acquired resistance to PLX4032 in both patients and cultured melanoma cells (14, 15). To determine whether the rapid sensitization of cells to NRG1β stimulation could provide a form of adaptive resistance to PLX4032 and AZD6244, we plated A375 cells at low density in the presence of DMSO, PLX4032, or AZD6244 with or without NRG1β. DMSO-treated cells rapidly grew to confluency regardless of NRG1β stimulation (Figure 6A). As expected, treatment of A375 cells with either PLX4032 or AZD6244 potently blocked the growth of colonies, while addition of NRG1β to PLX4032- or AZD6244 treated cells promoted colony growth (Figure 6, A and B). Additionally, NRG1β enhanced the viability of WM115, WM266-4, and WM239A cells treated with PLX4032 or AZD6244 for 72 hours, but did not enhance the viability of DMSO-treated cells (Figure 6C). These data indicate that NRG1β is able to partially restore viability and colony growth in RAF/MEK inhibitor–treated cells.

To test the requirement for ERBB3 in responsiveness to NRG1β, 1205LuTR cells stably expressing control (LacZ-targeting) shRNA or ERBB3-targeting shRNA were created. Depletion of ERBB3 with 2 independent shRNAs effectively inhibited AKT phosphorylation in response to NRG1β stimulation in vitro (Figure 6D). To determine whether ERBB3 was important for resistance to RAF inhibitors in vivo, 1205LuTR xenografts harboring LacZ- or ERBB3-targeting shRNAs were established in nude mice, and the animals were subsequently fed vehicle or PLX4720-laden chow. 1205Lu cells were utilized, given that they displayed a high level of intrinsic resistance to PLX4720 in our previous studies (23). ERBB3-knockdown cells did not significantly alter the growth of xenografts in the vehicle group (Figure 6E). In contrast, ERBB3-knockdown cells showed a marked reduction in tumor growth in the PLX4720 treatment group (Figure 6E). These data indicate that ERBB3 signaling is important in the response to RAF inhibitors both in vitro and in vivo.

NRG1β/ERBB3 signaling requires ERBB2 in melanoma.

ERBB3 is deficient in intrinsic kinase activity and relies upon other ERBB family members to phosphorylate it in response to ligand binding (38). As such, we sought to identify the kinase responsible for ERBB3 phosphorylation. Concomitant with ERBB3 phosphorylation in cells, enhanced ERBB2 (also known as HER2) phosphorylation in response to NRG1β was observed (Figure 4, B and D, and Supplemental Figure 3, A–D). We also observed a statistically significant (P < 0.05) increase in cells expressing high levels of membrane-associated phospho-ERBB2 (Y1221/Y1222) in A375 xenografts fed PLX4720 chow for 5 days (Figure 7, A and B). To determine whether ERBB2 was responsible for phosphorylating ERBB3, WM115 cells were depleted of ERBB2 by RNA interference. Knockdown of ERBB2 abolished NRG1β/ERBB3 signaling (Figure 7C). Additionally, treatment of cells with increasing doses of lapatinib (Tykerb/Tyverb), a clinical ERBB2/EGFR inhibitor, effectively inhibited NRG1β-stimulated ERBB3 and AKT phosphorylation in a dose-dependent manner in both A375 and WM115 cells (Figure 7D and Supplemental Figure 5A). EGFR-specific inhibitors gefitinib (Iressa) and erlotinib (Tarceva) failed to inhibit NRG1β/ERBB3 signaling in WM115 cells (Supplemental Figure 5B), indicating EGFR is not the kinase responsible for ERBB3 phosphorylation. ERBB4, which is also a receptor for NRG1β, is mutated in a subset of melanomas (39) and can be inhibited by lapatinib (39, 40). However, ERBB4 was poorly detected in the cells used in this study and depletion of ERBB4 with siRNA did not inhibit NRG1β/ERBB3 signaling in WM115 cells (Supplemental Figure 5C), arguing against ERBB4 phosphorylation of ERBB3. These data indicate that ERBB2 is the coreceptor for ERBB3 when cells are challenged with BRAF/MEK inhibitors and is responsible for its phosphorylation.

Combining RAF/MEK inhibitors with lapatinib provides a therapeutic benefit in vitro and in vivo.

To determine whether lapatinib prevents NRG1β/ERBB3-mediated resistance to PLX4032, A375 cells were plated at low density in the presence of PLX4032 and treated with either NRG1β alone, lapatinib alone, or both in combination. After 10 days, PLX4032-treated cells formed sizeable colonies in the presence of NRG1β alone, but failed to do so in the presence of lapatinib (Figure 8, A and B). Of note, lapatinib alone did not prevent the growth of A375 cells (Supplemental Figure 6A). Lapatinib could also ablate cell viability promoted by NRG1β in the presence of PLX4032 or AZD6244 in WM115 and 1205Lu cells (Supplemental Figure 6, B and C). To test the combination of lapatinib with BRAF inhibitors in vivo, we treated nude mice carrying 1205Lu or A375 xenografts with or without lapatinib in combination with PLX4720 or placebo. 1205Lu tumors showed a modest but statistically significant (P < 0.05) inhibition of tumor growth when treated with lapatinib alone (Figure 8C). In contrast, A375 tumors rapidly progressed in both vehicle and lapatinib-treated animals and showed no statistical difference in tumor burden (Figure 8D). PLX4720-treated animals showed a long latency in tumor progression, with both cell lines followed by steady tumor growth after about 14–15 days (Figure 8, C and D). Nearly half of the 1205Lu and A375 xenografts treated with PLX4720 alone reached a sacrificial threshold by 28 and 26 days, respectively (Figure 8, E and F). Remarkably, the combination of PLX4720 with lapatinib almost completely abolished 1205Lu tumor growth, with no mice reaching the sacrificial threshold (Figure 8, C and E). Similarly, A375 tumors in PLX4720/lapatinib-treated animals showed a longer latency period followed by slower tumor growth than PLX4720 alone, with only 1 out of 16 animals reaching a tumor volume necessitating animal sacrifice (Figure 8, E and F). These results indicate that lapatinib enhances the efficacy of PLX4720 and impairs the regrowth of PLX4720-resistant tumors.

Cell culture.

Human melanoma cell lines WM793, WM115, 1205Lu, WM266-4, and WM239A were donated by Meenhard Herlyn (Wistar Institute, Philadelphia, Pennsylvania, USA). SK-MEL-28 and A375 cells were purchased from ATCC. Tetracycline repressor–expressing (TR-expressing) sublines WM793TR, WM115TR, A375TR, and SK-MEL-28TR cells expressing Dox-inducible FOXD3 or LacZ have been previously described (22). 1205LuTR cells expressing Dox-inducible FOXD3 were generated in the same manner. A375 and A375TR were cultured in DMEM with 10% FBS and nonessential amino acids. All other cells except A375 and A375TR were cultured in MCDB 153 medium containing 20% Leibovitz L-15 medium, 2% fetal bovine serum, 0.2% sodium bicarbonate, and 5 μg/ml insulin.

Inhibitors, growth factors, and function-blocking antibodies.

AZD6244 and lapatinib for in vitro use were purchased from Selleck Chemicals. Lapatinib for in vivo use was provided by the Thomas Jefferson University Hospital pharmacy. PLX4032, PLX4720, and PLX4720 rodent chow were provided by Gideon Bollag at Plexxikon. Recombinant human NRG1β was purchased from Cell Signaling Technology. Gefitinib and erlotinib were provided by Ulrich Rodeck (Thomas Jefferson University).

RNA interference.

1205Lu and WM115 cells were transfected for 5 hours with chemically synthesized siRNAs (Dharmacon) at a final concentration of 25 nM using Lipofectamine RNAiMAX (Invitrogen). For in vivo experiments, 1205LuTR cells stably expressing Dox-inducible shRNAs were generated by lentiviral transduction. Sequences for siRNA and shRNA and lentivirus information can be found in the Supplemental Methods.

Microarray analysis.

Total cellular RNA was extracted using the PerfectPure RNA Cultured Cell Kit (5 Prime). For FOXD3 overexpression experiments, RNA was collected after 5 days of either FOXD3 or LacZ induction. Microarrays were performed by MOgene LC using Agilent-014850 Whole Human Genome Microarrays, and analysis was performed by Kimmel Cancer Center Genomics facility. False discovery rates were estimated using the procedure introduced by Storey (52). Genes with an absolute fold change of at least 1.5 and false discovery rate of less than 25% were considered significant. Microarray data were deposited in the GEO database (GSE43962).

ChIP and ChIP-seq.

WM115TR/FOXD3-V5 cells were induced with Dox for 24 hours and then fixed with 1% formaldehyde for 10 minutes. ChIP was performed using the EZ-ChIP kit and protocol (Millipore). Precleared lysates were incubated overnight with protein G Dynabeads (Invitrogen); beads were washed and eluted overnight at 65°C in ChIP elution buffer (25 mM Tris, pH 7.5, 10 mM EDTA, 0.5% SDS, and 200 mM NaCl). Eluate was treated with RNase A and proteinase K followed by removal of beads and purification of DNA. Antibodies used were normal IgG (Cell Signaling Technology), V5 (Invitrogen), and anti-RNA pol II CTD repeat YSPTSPS (phospho S2) antibody (Abcam). Purified DNA was analyzed by qPCR using iQ SYBR Green Supermix (Bio-Rad), 0.8 μM oligonucleotide primers, and 5 μl ChIP product. The primers used are listed in Supplemental Methods. Primer specificity was confirmed by melt curve analysis and TAE gel electrophoresis. Reaction conditions were as follows: denaturation at 94°C for 30 seconds, annealing at 50°C for 30 seconds, and elongation at 72°C for 30 seconds, with 50 cycles in total. PCR was performed on an iCycler with MyiQ version 1.0 software (Bio-Rad). Relative DNA enrichment levels were calculated using the Comparative Ct method (ΔCt) (53). For ChIP-seq, cells were treated with Dox for 48 hours prior to ChIP. Next generation sequencing and analysis were performed on V5-IP and input DNA by the Kimmel Cancer Center Genomics facility.

ChIP-seq read-mapping, peak-finding, and annotation.

Alignment of ChIP-seq reads to the human hg19 genome was performed using Applied Biosystems Bioscope 1.3 software ChIP-seq analysis pipeline, with default settings. Model-based Analysis of ChIP-Seq (MACS) software version 1.4.1 (54, 55) was used to predict ChIP-binding peaks, comparing the IP samples against total chromatin input. Default peak-calling parameters were used, except the P value cutoff for peak detection was set to a more stringent value of 1 × 10^–12. The resulting set of predicted ChIP-binding peaks was analyzed for enrichment of genomic features, including introns, exons, promoter, and intergenic regions, using Cis-regulatory Element Annotation System software, version 1.0.2 (56). Promoter occupancy rates were estimated in regions 3 kb upstream and downstream of transcription start sites.

Western blotting.

Cells were lysed and analyzed by Western blotting, as previously described (16). A list of antibodies can be found in the Supplemental Methods. Chemiluminescence was visualized on a VersaDoc MultiImager and quantitated using Quantity-One software (Bio-Rad).

qRT-PCR.

Total cellular RNA was extracted using the PerfectPure RNA Cultured Cell Kit. cDNA was made using the iScript cDNA Synthesis Kit (Bio-Rad). qPCR and analysis, including statistics, was performed as with ChIP experiments. The primers used are listed in Supplemental Methods.

Flow cytometry.

Detached cells were incubated in PBS with 2% BSA and 50 μl PE-conjugated anti-ERBB3 antibody (R&D Systems) on ice for 45 minutes. Washed cells were analyzed by flow cytometry on a BD FACSCalibur flow cytometer (BD Biosciences). Data were analyzed by FlowJo software (Tree Star Inc.).

Cell viability assays.

Cells (2 × 10⁵) were plated in complete medium in the presence/absence of 10 ng/ml NRG1β and treated with either DMSO, PLX4032 (1 μM), AZD6244 (3.3 μM), lapatinib (1 μM), or combinations of lapatinib with either PLX4032 or AZD6244. Cells were cultured for 72 hours, at which time medium was replaced with complete medium containing 1× AlamarBlue (Invitrogen) with respective inhibitors/NRG1β added. Cells were allowed to reduce AlamarBlue for approximately 2 hours. Medium was collected in triplicate from each condition, and the absorbances of oxidized and reduced AlamarBlue were measured at wavelengths 600 nM and 570 nM, respectively, in a Multiskan Spectrum spectrophotometer (Thermo Fisher Scientific). The change in viability was calculated from the resulting absorbances using the manufacturer’s guidelines. All conditions were normalized to the DMSO control.

Colony formation assays.

A375 (1 × 10³) cells were plated per 10-cm dish in complete medium with inhibitors or NRG1β, which were replenished every 3 days. After 7 days, cells were stained with crystal violet in formalin, plates were imaged by scanner, and colonies were imaged on a Nikon Eclipse Ti inverted microscope (Nikon) with NIS-Elements AR 3.00 software (Nikon). The percentage plate coverage is indicated as determined from 5 independent areas using ImageJ software ( http://rsbweb.nih.gov/ij/).

In vivo growth and survival assays.

Melanoma cells (1 × 10⁶) were injected intradermally into female athymic mice (NU/J: Jackson) and allowed to grow for 10–14 days to reach appropriate volume (∼40 mm³ for A375, ∼200 mm³ for 1205Lu, and ∼100 mm³ for 1205LuTR). Mice were fed either AIN-76A chow or AIN-76A with 417 mg/kg PLX4720 chow. For lapatinib experiments, mice received either vehicle (0.5% methylcellulose, 0.1% Tween 80 in sterile PBS) or 100 mg/kg lapatinib suspended in vehicle by oral gavage daily (except on days 16, 17, 23, and 24 for A375 experiments). For shRNA experiments, mice were exposed to 2 mg/ml Dox in drinking water beginning 3 days prior to chow treatment. Measurements of tumor size were taken every 3–4 days using digital calipers, and tumor volume was determined by the following formula: volume = (length × width²) × 0.52. Time-to-event (survival) was determined by a 10-fold increase in baseline volume for the A375 experiment and a 3-fold increase in baseline volume for the 1205Lu experiment. The maximum allowable tumor size for 1205Lu and 1205LuTR cells was limited by the development of skin necrosis requiring euthanasia.

IHC.

Tissue samples from A375 intradermal xenografts were obtained from mice that were fed either control or PLX4720 chow for 5 days. Tissue was fixed in formalin and paraffin embedded. Sections were stained with anti–phospho-ERBB3 Y1289 (21D3) and phospho-ERBB2 Y1221/Y1222 (6B12) antibodies (Cell Signaling Technology) and scored in a blinded manner for staining intensity utilizing a digital Aperio ScanScope GL system and ImageScope software. Statistical analysis of staining quantitation was determined separately for each antibody using a proportional odds mixed model accounting for random effects to adjust for sample variation (R Statistical Software).

Patient samples.

Samples were formalin fixed and paraffin embedded immediately following isolation. IHC was performed using anti–phospho-ERBB3 Y1289 (21D3). Staining was scored in a blinded manner, as above.

Statistics.

For statistical analysis of qPCR and cell viability assays, 2-tailed t tests assuming unequal variances were performed using Excel (Microsoft). Statistical analysis for tumor growth data was conducted using a mixed-effects model and Tukey’s corrected pairwise comparisons of mean fold change in volume between treatment groups (SAS statistical software). Statistical analysis for time-to-event (survival) was conducted using log-rank comparison of Kaplan-Meier curves (SAS statistical software), and α for all experiments was 0.05. Additionally, analysis was performed across samples from all 9 patients that displayed staining for phospho-ERBB3 (3 sets of samples were uniformly negative and were not included). We utilized an ordered logistic regression model (i.e., proportional odds model) with random intercept for each patient. The ordered logistic regression model assumes that the odds of receiving a score greater than or equal to k is odds ratio (OR) times higher for progression than pretreatment, where the number OR is a constant for k = 1 or 2. We used the package ordinal of software R. For all analyses, P values of less than 0.05 were considered statistically significant.

Study approval.

All animal experiments were approved by the IACUC and performed in a facility at Thomas Jefferson University accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC). Patient samples were collected under a protocol approved by the IRB at the The University of Pennsylvania. All patients gave informed consent.