NRG1 is a critical regulator of differentiation in TP63-driven squamous cell carcinoma

Squamous cell carcinomas (SCCs) account for the majority of cancer mortalities. Although TP63 is an established lineage-survival oncogene in SCCs, therapeutic strategies have not been developed to target TP63 or it’s downstream effectors. In this study we demonstrate that TP63 directly regulates NRG1 expression in human SCC cell lines and that NRG1 is a critical component of the TP63 transcriptional program. Notably, we show that squamous tumors are dependent NRG1 signaling in vivo, in both genetically engineered mouse models and human xenograft models, and demonstrate that inhibition of NRG1 induces keratinization and terminal squamous differentiation of tumor cells, blocking proliferation and inhibiting tumor growth. Together, our findings identify a lineage-specific function of NRG1 in SCCs of diverse anatomic origin.


Introduction
Within the past decade, lineage addiction has emerged as a common paradigm to explain how certain tumors depend on co-opted survival and self-renewal programs that drive the normal development of the tissues from which they arise (Garraway and Sellers, 2006). During normal development and tissue homeostasis, 'master regulator' transcription factors control large sets of genes regulating cellular identity, differentiation and survival (Chan and Kyba, 2013). Amplifications of master regulators act as oncogenic drivers in cancers arising in the tissues whose development they normally control. Examples include MITF, which directs melanocyte development and is amplified in some melanomas and NKX2.1, which directs development of the distal lung epithelium and is amplified in some lung adenocarcinomas (Kendall et al., 2007). Some cancers remain fully dependent on transcription factors expressed by precursor cells of the lineage from which they develop, even in the absence of genetic alterations in these genes. Examples include AR in prostate cancer and ESR1 in luminal breast cancers (Garraway and Sellers, 2006).
Despite varied anatomic origins, squamous cell cancers (SCCs) share many common properties, including genetic and epigenetic alterations (Dotto and Rustgi, 2016). The TP63 transcription factor exemplifies an important lineage dependency in SCCs (Ramsey et al., 2013). Amplification of TP63 is prevalent in SCCs, and TP63 expression is used to distinguish SCCs from other cancer subtypes in multiple tissues (Dotto and Rustgi, 2016). SCCs arise in numerous organ systems that contain stratified or pseudo-stratified epithelia, including the lung, head and neck, esophagus, skin, bladder and cervix. Interestingly, like other lineage survival oncogenes, TP63 is a key regulator of the progenitor cells in the basal cell compartment during normal development and homeostasis of most stratified or pseudostratified epithelia (Mills et al., 1999;Yang et al., 1999). Despite its established role as a driver of lineage dependency, TP63 is a transcription factor, and as such is challenging to target therapeutically.
Here we show that NRG1 expression is directly regulated by TP63 in SCCs of various organs, and that co-expression of NRG1 and its receptor ERBB3 is prevalent in SCCs. Moreover, we find that many of the SCC models that co-express NRG1 and ERBB3 depend on NRG1 autocrine signaling in vivo, in contrast to non-squamous cancers that exhibit NRG1 autocrine signaling but are not dependent on it.
Results and discussion TP63 regulates NRG1 expression TP63 is highly expressed in basal cells of various epithelia and is required for the progenitor cell function. In addition, TP63 acts as a key survival factor and driver of SCCs (Rocco et al., 2006;Thurfjell et al., 2005). Interestingly, studies of normal mammary basal cells established that TP63 can directly activate NRG1 transcription (Forster et al., 2014). Therefore, we evaluated whether this transcriptional wiring exists in SCCs. We found that NRG1 and TP63 expression significantly correlated in both esophageal and lung squamous cell carcinomas (LUSC) as determined from The Cancer Genome Atlas (TCGA) transcriptome data ( Figure 1A). TP63 has two isoform classes that either contain (TA-TP63) or lack (DeltaN-TP63) an N-terminal transactivation domain. Despite lacking this domain, deltaN-TP63, the major isoform expressed in SCCs, functions as both a positive and negative transcriptional regulator of different target gene subsets (Hibi et al., 2000;Moll and Slade, 2004). We evaluated whether TP63 regulates NRG1 expression in SCCs using siRNAs to knockdown all TP63 isoforms (siTP63 # 14) or only the TA-TP63 isoforms (siTA-TP63 # 13) and assessing expression of both isoforms of the NRG1 EGF-like domain, NRG1a and NRG1b, by qPCR in the OE-21 and KYSE-140 SCC cell lines. Knockdown of just the TA isoforms modestly reduced NRG1 expression, whereas silencing of all isoforms robustly decreased NRG1 expression ( Figure 1B). We further expanded this finding in an additional cell line, KYSE-180 ( Figure 1C). Knockdown of deltaN-TP63 significantly reduced NRG1 transcripts, confirming that deltaN-TP63 regulates NRG1 expression in SCCs. Immunoblotting confirmed knockdown of deltaN-TP63 at the protein level. In addition, we determined whether NRG1 is a direct transcriptional target of deltaN-TP63 by ChIP-PCR using antibodies for TP63alpha and deltaN-TP63 and PCR primers that amplify the NRG1 promoter. Binding of both TP63 isoforms was significantly enriched at the region À30 kB from the transcriptional start site (TSS), which encompasses the TP63 binding motif, compared to the control locus at À21 kB ( Figure 1D). Together, these data suggest that deltaN-TP63 directly regulates NRG1 transcription in SCC.

Efficacy of anti-NRG1 in in vivo models of SCC
Emerging evidence suggest that high ERBB3 or high NRG1 expression is associated with poor clinical outcome in SCCs (Qian et al., 2015). We reasoned that in order for NRG1 to promote SCC growth, tumors would have to co-express NRG1 and its receptor ERBB3. Interestingly, NRG1/ERBB3 co-expression appeared prevalent in lung and head and neck SCCs, but was notably rare in lung adenocarcinoma across the various cancer datasets available in TCGA ( Figure 2A). To ascertain whether SCCs co-expressing NRG1 and ERBB3 are responsive to inhibition of NRG1 signaling, we screened a panel of cell lines from SCC indications including lung, esophageal and skin for growth sensitivity to an NRG1 blocking antibody (Hegde et al., 2013) in vitro. Anti-NRG1 treatment modestly inhibited the growth of cell lines expressing both NRG1 and ERBB3 ( Figure 2B). Because the p63 transcriptional program is critical for both maintenance and cell fate determination of epithelial progenitor cells, we reasoned that perturbation of this program would be more impactful in vivo. We assessed the effect of NRG1 inhibition on xenograft tumors derived from the FaDu head and neck, HCC95 lung and KYSE-180 esophageal squamous cell carcinoma lines. Importantly, anti-NRG1 treatment markedly inhibited tumor growth in each of these models to an extent that far exceeded  that observed in vitro. ( Figure 3A). In addition, increased keratinization and changes in tumor cell morphology were observed in tumors from anti-NRG1 treated mice (Figure 4-figure supplement 1). To further evaluate the NRG1-dependency in ERBB3/NRG1 co-expressing squamous cell cancers, we tested the efficacy of anti-NRG1 in lung SCC PDX models. Again, anti-NRG1 significantly inhibited the tumor growth of three models that co-express NRG1 and ERBB3, resulting in tumor stasis ( Figure 3B and Although NRG1 is not widely recognized as an important factor in cutaneous SCC, downregulation of endogenous NRG1 expression occurs during differentiation of cultured keratinocytes, and activation of NRG1 signaling inhibits keratinocyte differentiation and promotes neo-epidermal outgrowth in human skin explant cultures (De Potter et al., 2001). Furthermore, mice with Krt5-Credriven knockout of Erbb3 in basal progenitor cells are resistant to carcinogen-induced skin tumorigenesis and exhibit defective wound healing (Dahlhoff et al., 2015), while Krt5-driven Average growth is presented as is mean with standard error of mean relative to anti-Ragweed from four independent experiments with more than three replicates in every experiment.  overexpression of deltaN-TP63 enhances carcinogen-induced skin tumorigenesis (Devos et al., 2017). Therefore, we tested the effect of inhibiting NRG1 in the Lgr5 CreERT2 ; Pten flox/flox ; Kras LSL-G12D/+ genetically engineered model of cutaneous SCC. In this highly aggressive tumor model, anti-NRG1 dramatically increased the progression free survival (PFS) compared to control treatment, nearly doubling time to progression from 14 to 27 days (p<0.002) ( Figure 3C-D). Progression was defined as enlargement and redness of the lips and snout, as this was the only clinical observation in the animals and macroscopic skin tumors were observed only upon necropsy. In control animals, the dermis was expanded by confluent nests of squamous cells forming tumors that > 80% of the dermis in skin lesions sampled from 5 of 7 animals. In contrast, in 5 of 6 anti-NRG1 treated animals, lesions were limited to central dilated keratinization without dermal expansion, consistent with epidermal inclusion cyst, a benign squamous proliferation ( Figure 3D). Our analysis of RNAseq data across tumor indications also revealed prevalent NRG1/ERBB3 coexpression in ovarian cancer. We tested the effect of anti-NRG1 on a panel of ovarian cancer cell lines and indeed proliferation of cell lines expressing both NRG1 and ERBB3 was inhibited by anti-NRG1 in vitro (Figure 3-figure supplement 2A). However, unlike the SCC models, ovarian PDX models expressing NRG1 and ERBB3 were not sensitive to anti-NRG1 treatment in vivo (Figure 3figure supplement 2B-C) suggesting that dependence on NRG1 signaling may be specific to the TP63 driven cancer types. Together these data support a role for NRG1 in mediating p63 lineage dependency in SCCs.

Anti-NRG1 induces squamous differentiation
To explore the mechanism mediating the robust tumor growth inhibition observed in anti-NRG1 treated SCC models, we repeated the in vivo studies for three SCC xenograft models of different anatomic origins and collected tumors after treatment. Histological analysis revealed that in all models anti-NRG1-treated tumors exhibited a more well-differentiated appearance, with increased eosinophilic cytoplasm consistent with enhanced keratinization, indicating that anti-NRG1 treatment was driving differentiation in these tumors (Figure 4-figure supplement 1 and Figure 4A). Moreover, anti-NRG1-treated tumors showed a dramatic increase in the expression of KRT10, a differentiationspecific keratin normally restricted to the post-mitotic layers of stratified-keratinizing and cornifying epithelia ( Figure 4A). Immunoblot analyses of treated tumors showed that anti-NRG1 inhibited ERBB3 activation and decreased the levels of multiple proliferation markers, consistent with a mechanism in which differentiation is induced at the expense of proliferation ( Figure 4B and Figure 4figure supplement 2). Markers of apoptosis were not affected ( Figure 4B).
To broadly assess changes in expression of differentiation markers, we analyzed RNAseq data, focusing on a gene expression signature of human airway basal cells (Hackett et al., 2011). Anti-NRG1 treatment caused significant changes in the levels of nearly all the genes in the panel ( Figure 4C). Notably, many of the genes upregulated by anti-NRG1 treatment are associated with differentiation of stratified epithelia, such as KLK7, BNC1 and ADAMTS1, demonstrating a profound effect on the differentiation state of the tumor cells that becomes more pronounced with ongoing treatment. Of note, NRG1 is itself a marker of airway basal cells, and anti-NRG1 treatment resulted in downregulation of the NRG1 transcript. Upon differentiation, airway basal cells cluster with keratinocytes in unsupervised analyses (Hackett et al., 2011). Therefore, we evaluated markers of keratinocyte differentiation and found that anti-NRG1 treatment resulted in strong upregulation of several well-established keratinocyte differentiation markers ( Figure 4D and NRG1 signaling supports progenitor cell function and regeneration in a diverse set of normal tissues (Bartus et al., 2016;Bersell et al., 2009). Our data suggest that NRG1 may serve a similar function downstream of TP63 in tumors derived from squamous epithelia. We show that treating squamous tumors with an NRG1 blocking antibody reduced proliferation, concomitant with induced cellular differentiation and increased keratin production. The marked difference in response between in vitro and in vivo models is consistent with a pro-differentiation effect of anti-NRG1 treatment, as cancer cells are known to undergo dedifferentiation and lose tissue-specific expression patterns and   LEPREL1  SERPINE1  PLAU  IL1RL1  PHLDA1  STC2  TRIB3  SLC16A1  PSAT1  ALDH1L2  KLK7  LIPG  BNC1  ITGA6  CAY1  COL17A1  IL13RA2  NRG1  SERPINE2  THBS1  ADAMTS1  SLC7A5  FSCN1  EREG  TUBB6  CALD1  EHD2  CAMK2N1  IL1RN  KRT6A  SPRR1A  SCEL  SPRR3  SPRR1B  KRT16  DSG3  KRT6B  GJB6  AREG  GJB2  CRCT1 UPP1 RGS20 DKK1 differentiation capacity when cultured long term on plastic. In contrast, although NRG1 and ERBB3 co-expression is prevalent in ovarian tumors, ovarian cancer models did not respond to NRG1 inhibition in vivo. Our model predicts such a result, given that fewer than 10% of epithelial ovarian tumors express p63 and it is not a lineage oncogene in this cancer type.

KRT10-IHC
EGFR signaling is known to be an important driver in SCCs and anti-EGFR therapies have been approved for the treatment of head and neck and lung SCCs (Sacco and Worden, 2016;Thakur and Wozniak, 2017). In contrast, NRG1-ERBB3 signaling has been implicated as a resistance mechanism to anti-EGFR therapies (Wheeler et al., 2008). However, dual inhibition of ERBB3 and EGFR did not show significant clinical benefit compared to EGFR inhibition alone. Our findings on the role of NRG1 in SCCs raise the possibility that NRG1 may still provide resistance to EGFR therapeutics through ERBB4 receptor activation when ERBB3 is inhibited, and provide rationale for exploring the clinical benefit of NRG1 inhibition in combination with EGFR inhibition in SCCs. Continued on next page

Cell culture
Cancer cell lines were sourced, authenticated, tested for mycoplasma and maintained by the Genentech cell bank (gCELL) as described (Yu et al., 2015). Cell lines were cultured in either RPMI-1640 or Dulbecco's modified Eagle's medium (DMEM) growth medium supplemented with 10% of fetal bovine serum (FBS), 2 mmoL glutamine and penicillin/streptomycin. For growth assays, tumor cells were cultured in media containing 2% FBS in 96-well plates, and treated for 96 hr with 20 ug/ml of anti-NRG1 YW538.24.71 (previously described in Hegde et al., 2013) or anti-Ragweed (control) antibodies in triplicate, and assayed using cell titer blue (Promega). For effect of antibodies on differentiation in vitro, FaDu, HCC95 and KYSE-180 tumor cells were cultured with 20 ug/ml of antibodies for three days. RNA was analyzed and qPCR was performed using ABI TaqMan primer/probes as explained below. Detailed characterization of this anti-NRG1 (YW538.24.71) was previously described (Hegde et al., 2013). Statistical significance was determined by t-test from at least three independent experiments. Expression of NRG1, ERBB3 and ERBB4 was assessed by RNAseq. sonicated using a Bioruptor sonicator (Diagenode) to produce chromatin fragments averaging 200-500 bp. Sheared chromatin was incubated overnight at 4˚C with protein A-coated magnetic beads plus either anti-TP63 alpha (CST 13109), anti-deltaNTP63 (BioLegend 619001) or isotype control antibody. Beads were washed, DNA was eluted and crosslinks were reversed during an incubation overnight at 65˚C. Samples were treated with iPure beads and DNA was purified. Quantitative PCR was performed using sybr green PCR master mix (Applied Biosystem) to assess enrichment at the NRG1 promoter. The results were validated by three independent experiments. The ChIP ratio was calculated as enrichment over noise, normalized to the input. Statistical significance was determined by one-way ANOVA. Primers used for CHIP PCR were À21 kB NRG1 promoter (5'-TTCAAAAGG-GAGTGCCAACTTTTCC-3', 5'-GGTGCCTCACCTTTCTTCTTCCTGTCC-3') and À30 KB NRG1 promoter (5'-GCCCCAAATTCTTTTGCCCCTTAT-3', 5'-TTGGTTGGCTTGCTGAAGCTGGTGT-3') from NRG1 transcript start site as described earlier (Forster et al., 2014).

Immunohistochemistry
Tumors collected after one or three dose of antibodies were fixed in 10% neutral-buffered formalin overnight then transferred to 70% ethanol, processed and embedded into paraffin. Tumor sections were subjected to H and E and IHC using rabbit polyclonal anti-Cytokeratin10 (KRT10) antibody (Covance Biologicals, PRB-159P), incubated at a concentration of 1.0 ug/mL for 60 min at room temperature and binding was visualized using ABC-Peroxidase Elite followed by DAB chromagen and counter stained with Mayer's hematoxylin. KRT10 expression was reviewed manually by a translational pathologist (JMG) and scored for the percentage of cells demonstrating moderate to strong immunoreactivity, excluding areas of necrosis (0-100%).

RNA-seq
RNA-seq libraries were prepared using TruSeq RNA Sample Preparation kit (Illumina, CA) and sequenced on Illumina HiSeq 2500 sequencers, yielding an average of 34 million single-end reads (50 bp) per sample. Reads were aligned to the human genome version NCBI GRCh37 using GSNAP. Expression counts per gene were obtained by counting the number of reads aligned concordantly within a pair and uniquely to each gene locus as defined by NCBI, Ensembl gene annotations, and RefSeq mRNA sequences. Differential gene expression analysis was performed using edgeR. Gene enrichment analysis was performed on the edgeR differential expression results using the Gsea Preranked tool available through the Broad's GSEA application. DESeq was used to compute the variance stabilized expression values for plotting the expression heat maps.

Statistical analysis
Graphical and statistical data were generated with Microsoft Excel or GraphPad Prism (GraphPad Software, La Jolla, CA, USA). Statistical significance of differences between the results was assessed using a standard 2-tailed t-test or one-way ANOVA using Prism. p<0.05 was considered statistically significant.