ERRα mediates metabolic adaptations driving lapatinib resistance in breast cancer

Despite the initial benefits of treating HER2-amplified breast cancer patients with the tyrosine kinase inhibitor lapatinib, resistance inevitably develops. Here we report that lapatinib induces the degradation of the nuclear receptor ERRα, a master regulator of cellular metabolism, and that the expression of ERRα is restored in lapatinib-resistant breast cancer cells through reactivation of mTOR signalling. Re-expression of ERRα in resistant cells triggers metabolic adaptations favouring mitochondrial energy metabolism through increased glutamine metabolism, as well as ROS detoxification required for cell survival under therapeutic stress conditions. An ERRα inverse agonist counteracts these metabolic adaptations and overcomes lapatinib resistance in a HER2-induced mammary tumour mouse model. This work reveals a molecular mechanism by which ERRα-induced metabolic reprogramming promotes survival of lapatinib-resistant cancer cells and demonstrates the potential of ERRα inhibition as an effective adjuvant therapy in poor outcome HER2-positive breast cancer.

T he overexpression and aberrant activation of receptor tyrosine kinase (RTK) signalling pathways constitute a prominent driver of human breast cancer progression 1,2 . Lapatinib (Tykerb) is a dual epidermal growth factor receptor (EGFR)/human EGFR-2 (HER2) tyrosine kinase inhibitor (TKI) approved for patients with HER2-amplified breast tumours presenting with metastatic lesions 3 . Despite initial benefits of lapatinib treatment in breast cancer patients, resistance inevitably develops 4 . While several studies indicate that the EGFR/HER2 and downstream AKT/PI3K and ERK/MAPK signalling pathways often remain inhibited by RTK inhibitors in resistant cells, alternative redundant signalling routes are instead engaged and converge on re-activation of common downstream effectors [5][6][7] . Proposed bypassing routes include the activation of kinases downstream of b1 integrin 8,9 , stimulation of the mechanistic target of rapamycin (mTOR) complex 1 (mTORC1) [10][11][12][13] and enhanced autocrine mitogenic signalling [14][15][16][17][18] . While these observations suggest that reactivation of RTK or of alternative signalling routes may restore the proliferative potential of lapatinib-treated breast cancer cells, the downstream effects responsible for conferring resistance to lapatinib remain unknown.
The metabolic status of cancer cells impacts on their response to drugs. Indeed, targeting glycolysis sensitizes HER2-positive cells to HER2 inhibition by Herceptin treatment 19 and the redox status can predict the response to HER2-inhibiting drugs in breast cancer cells 20 . Furthermore, lapatinib-resistant breast cancer cells display upregulation of genes controlling the glucose deprivation response network, suggesting a potential influence of the metabolic state of the cell in the response to lapatinib 21 . However, the specific signalling pathways and transcriptional regulators responsible for the metabolic adaptations contributing to lapatinib resistance in breast cancer cells are undefined.
Oestrogen-related receptor a (ERRa, NR3B1), an orphan member of the superfamily of nuclear receptors 22 , is a master regulator of cellular energy metabolism in both normal and cancer cells [23][24][25] . ERRa expression positively correlates with HER2 status and with poor prognosis in breast tumours 26,27 , and we recently showed that it contributes to ERBB2-dependent mammary tumorigenesis in mice 28 . Mechanistically, EGF has been shown to induce the recruitment of ERRa to the promoter of the TFF1 gene 29 , while signals from HER2 impact on ERRa transcriptional activity 30,31 . These observations suggest that ERRa could participate as a downstream effector of mitogenic signals to mediate the metabolic adaptation of HER2-positive breast cancer cells and subsequently in their response to lapatinib.
Here we explore the hypothesis that ERRa acts as an effector of mitogenic signalling responsible for metabolic adaptations of breast cancer cells and further provide evidence of its implication in the therapeutic response and resistance to the RTK inhibitor lapatinib.

Results
Growth factor-dependent activity of ERRa in breast cancer cells. We previously showed that ablation of ERRa significantly delays ERBB2-induced tumour development in mice and lowers the levels of the ERBB2 amplicon transcripts 28 . To further investigate the interplay between RTK signalling and ERRa activity in breast cancer, we quantified the level of ERRapositive nuclear staining in breast tumour samples from various subtypes using a specific antibody for ERRa ( Supplementary  Fig. 1a) and observed that the HER2-positive/oestrogen receptor-negative tumours express the highest level of ERRa-positive nuclei with a median expression of 90% (Fig. 1a, one-way analysis of variance P value ¼ 0.0487).
We then considered whether RTK signalling impacts on the genetic programmes regulated by ERRa in breast cancer cells. Serum-starved ERBB2-positive SKBr3 breast cancer cells were exposed to EGF or heregulin (HRG) to activate HER2 through EGFR/HER2 and HER2/HER3 signalling routes, respectively. The ERRa cistromes were generated using chromatin-immunoprecipitation (ChIP) followed by massive parallel sequencing (ChIPseq). As shown in Fig. 1b,c, growth factor treatment induces a significant potentiation and reprogramming of ERRa binding without affecting the level of ERRa expression. A large number of genomic regions are significantly bound by ERRa only on growth factor treatment in SKBr3 cells (Fig. 1b,c and Supplementary  Fig. 1b,d), including the region within the TFF1 promoter originally described 29 as an EGF-induced ERRa binding site in MCF-7 breast cancer cells ( Supplementary Fig. 1b). Moreover, while ERRa binds to several common regions regardless of cell context, the addition of growth factors to the media significantly amplifies the signal intensity of ERRa recruitment to both these common sites and growth factor-specific sites ( Fig. 1c and Supplementary Fig. 1c). Similar reprogramming of ERRa binding by ChIP-seq was observed in another HER2amplified breast cancer cell line, BT-474, on growth factor treatment ( Supplementary Fig. 1e). De novo binding site analyses identified the ERRa response element (ERRE: Esrrb, Nr5a2 and Erra) as the most enriched motif in segments bound by ERRa both in serum-starved and growth factor-treated SKBr3 and BT-474 cells. However, treatment of both cell lines with EGF or HRG led to the specific co-enrichment of the binding sites for AP-1 (Jun/Fos) along with the ERRE in ERRa-bound segments ( Supplementary Fig. 1f,g). These observations reveal a growth factor-dependent cooperation between ERRa and AP-1 in ERBB2-positive breast cancer cells.
We then assessed the biological significance of growth factordependent potentiation of ERRa binding in breast cancer cells through analysis of enriched functional pathways. Target genes harbouring ERRa-binding sites within 20 kilobases (kb) of their transcriptional start site, commonly bound in growth factortreated cells and in untreated conditions, are involved in biological pathways related to previously reported functions of ERRa (ref. 32), including the control of mitochondrial metabolic functions and less-studied functions like stress response (Fig. 1d). The promoters bound by ERRa exclusively upon growth factor treatment are preferentially associated with genes regulating tumorigenic processes, including RTK and integrin signalling, proliferation and metastatic processes (Fig. 1d). To assess the functionality of this growth factor-modulated binding of ERRa, we further intersected our ChIP-seq data with gene expression data generated on depletion of ERRa in SKBr3 cells with and without growth factor treatment (Fig. 1e). While the depletion of ERRa in SKBr3 cells in all the conditions tested leads to altered regulation of target genes controlling mitochondrial metabolism (OXPHOS, tricarboxylic acid (TCA) cycle and glycolysis/ gluconeogenesis) (Supplementary Fig. 1h and Supplementary  Table 1), the depletion of ERRa in EGF-or HRG-treated cells also triggered the alteration of ERRa target genes controlling processes related to ribosome, fatty acid, glutathione, cysteine and methionine metabolism, as well as amino acid degradation (Fig. 1e and Supplementary Table 2). Moreover, the ERRabinding intensity in genes regulating these metabolic functions or of genes in the glutathione/detoxification pathway (list adapted from ref. 33) is increased on growth factor treatment ( Supplementary Fig. 1i). These results show that growth factorinduced RTK signalling in breast cancer cells potentiates the activity of ERRa to modulate the expression of target genes implicated in various aspects of cellular metabolism, including glutathione-mediated detoxification.
We further assessed whether pharmacological inhibition of EGF and HRG signalling could prevent this growth factordependent potentiation of ERRa recruitment to chromatin. SKBr3 cells were treated with the dual EGFR/HER2 inhibitor lapatinib to block growth factor signalling through EGFR/HER2/ HER3. Standard ChIP experiments revealed that the binding of ERRa to growth factor-reprogrammed sites is abolished upon exposure to lapatinib (Fig. 1f, four upper panels). Unexpectedly, lapatinib treatment also triggered a significant decrease in ERRa recruitment to regions that are constitutively bound in absence of growth factors as depicted by the ESRRA promoter (Fig. 1f,    ERRa recruitment to chromatin is not limited to growth factorreprogrammed sites but could affect all sites bound by ERRa. We therefore monitored the expression of ERRa in SKBr3 cells and observed that 24-h lapatinib treatment induces a sharp decrease in ERRa protein in SKBr3 and BT-474 cells, as well as in the mouse cell line NIC-5231 derived from ErbB2-driven mammary tumorigenesis 34 (Fig. 1g). Quantitative PCR analyses revealed a similar decrease in ESRRA mRNA levels (Fig. 1h), likely a consequence of the decrease in ERRa protein levels since ERRa autoregulates its own expression 35 . Treatment of SKBr3 cells with the potent proteasome inhibitor MG-132, which blocks ubiquitin-conjugated protein degradation, prevented the lapatinibdependent degradation of ERRa (Fig. 1i), indicating that the effect of lapatinib on ERRa protein levels occurs at least in part through proteosomal degradation. Taken together, these results reveal that growth factor-induced signalling potentiates ERRa transcriptional activity in regulating cellular metabolism and oxidative stress response and that inhibition of this signalling with RTK inhibitors leads to the degradation of ERRa.

Re-expression of ERRa in lapatinib-resistant cells by mTOR.
On the basis of these findings, we next investigated whether ERRa could play a role in the resistance to lapatinib treatment in breast cancer cells. SKBr3, BT-474 and NIC-5231 cells were made resistant to lapatinib on long-term exposure to gradually increasing doses of lapatinib as previously described 36 . We confirmed that lapatinib-resistant SKBR3 (LRSKBr3) cells survive in higher doses of lapatinib than parental cells (pSKBr3) (Fig. 2a). In contrast to parental cell lines where lapatinib treatment decreases ERRa protein levels, the expression of ERRa is unaffected by lapatinib treatment in LRSKBr3 cells, as well as in the lapatinib-resistant BT-474 (LR-BT-474) and NIC-5231 (LR-NIC-5231) cells (Fig. 2b).
To gain insight into the mechanisms leading to ERRa reexpression in lapatinib-resistant cells, we monitored the status of RTK and downstream effectors involved in EGFR/HER2/HER3 signalling. We first confirmed that acquired lapatinib resistance in LRSKBr3 cells does not lead to reactivation of EGFR/HER2/HER3 or of downstream signalling pathways in LRSKBr3 cells as monitored by the phosphorylation status of the receptors and their downstream effectors ( Supplementary Fig. 2a). We next assessed the status of the energy sensors AMPK and mTOR acting downstream of RTK signalling. While the levels of total AMPK are stable, phosphorylated AMPK, which is triggered in response to stressful stimuli, including oxidative stress, increases on lapatinib treatment both in parental and resistant cells, supporting the efficacy of the treatment in both cell lines ( Supplementary Fig. 2b). However, while lapatinib treatment in pSKBr3 cells induces a decrease in total mTOR protein levels and in the levels of the activated forms of its downstream phosphorylated effectors P-S6 and P-4E-BP1 (Fig. 2c), mTOR signalling is insensitive to the inhibitory action of lapatinib in resistant cells (Fig. 2c). This suggests that reactivation of mTOR activity becomes decoupled from EGFR/HER2/HER3 and AMPK signalling in lapatinib-resistant cells.
We next tested the potential involvement of the mTOR pathway in the re-expression of ERRa in lapatinib-resistant cells. As shown in Fig. 2d, co-treatment of parental SKBr3 cells with the mTOR inhibitor rapamycin exacerbates the lapatinib-mediated decrease in ERRa levels. While high levels of ERRa and P-S6 are observed in resistant cells treated with lapatinib, co-treatment with rapamycin induces a significant decrease in the levels of these proteins in resistant cells (Fig. 2d), indicating that mTOR re-activation in lapatinib-resistant cells is linked to ERRa reexpression. We observed a similar effect with the mTORC1specific inhibitor INK1341 that also induces a decrease in ERRa levels in lapatinib-resistant SKBr3 cells ( Supplementary Fig. 2c). Similarly, SKBr3 cells resistant to treatment with the HER2specific antibody trastuzumab also display increased levels of ERRa protein and activation of mTOR signalling through an increase in P-S6 levels when compared with parental cells. This suggests that reactivation of ERRa expression via mTOR signalling plays a key role in the development of HER2 targeted drug resistance in distinct therapeutic contexts ( Supplementary  Fig. 2d).
We next assessed whether the inhibitory effect of lapatinib on ERRa expression also occurs in vivo using both a mouse model of human HER2-driven breast cancer (MMTV-NIC) 37 and HER2positive patient-derived xenografts (PDXs) propagated in NOG mice. Following acute treatment of the mice with lapatinib or vehicle for 48 h before harvesting, we observed a considerable decrease in the expression of ERRa in the lapatinib-treated NIC and PDX tumours (Fig. 2e,f). As observed in the cell line models, this decrease in ERRa levels was concomitant with the inhibition of mTOR signalling on lapatinib treatment in vivo as assessed by the reduction in the phosphorylation of the mTOR effector P-S6 (Fig. 2e,f).
We further investigated whether the expression of ERRa was also reactivated in vivo in tumours that have acquired the capacity to grow in the presence of lapatinib. Mice bearing ERBB2dependent MMTV-NIC mammary tumours were treated with lapatinib or vehicle using previously published protocols 38,39 for a period of 6 weeks. Despite an initial response and tumour volume stabilization in the lapatinib-treated mice, the tumours eventually relapsed and acquired the ability to grow in the presence of the drug (Fig. 2g). As observed in cell culture models, monitoring the expression of ERRa by immunohistochemical staining and immunoblotting revealed a significant decrease of ERRa expression on acute lapatinib treatment compared with the control treatment arm, and an increase at end point in ERBB2dependent tumours that have relapsed (Fig. 2h,i). Further, decreased ERRa expression on acute lapatinib treatment in vivo in the NIC tumours is concomitant with decreased mTOR signalling, while lapatinib-resistant tumours exhibit restoration of mTOR signalling along with the re-expression of high levels of ERRa (Fig. 2i). These results suggest that development of lapatinib resistance in ERBB2-driven mammary tumours in vivo recapitulates the mechanisms observed in the cell line models.

ERRa dictates a metabolic signature in lapatinib-resistant cells.
Re-expression of ERRa in resistant cells prompted us to investigate the ERRa transcriptional programme in LRSKBr3 cells. Using a standard ChIP approach, we observed constitutive recruitment of ERRa to growth factor-induced binding sites in serum-starved lapatinib-resistant cells, even in absence of growth factor stimulation (veh bars in Supplementary Fig. 3a). Growth factor stimulation did not further enhance the recruitment of ERRa to these growth factor-induced sites in resistant cells. We therefore performed ERRa ChIP-seq to compare the ERRabinding profile of untreated pSKBr3 cells to that of LRSKBr3 cells maintained in lapatinib. ERRa is recruited to more sites in the lapatinib-treated resistant cells than in the untreated parental cells ( Fig. 3a and Supplementary Fig. 3b). The binding intensity of ERRa at both common and specific sites is significantly increased in LRSKBr3 cells compared with pSKBr3 cells (Fig. 3b-d).
Interrogation of de novo binding sites revealed enrichment of the ERRE motif both in pSKBr3 and in LRSKBr3 cells ( Supplementary Fig. 3c). As observed with the growth factor reprogramming of ERRa-binding profiles in parental SKBr3 cells ( Fig. 1), co-occurrence of the AP-1 motif is observed at ERRabinding sites that are specific to the resistant cells ( Supplementary  Fig. 3c).
Similar to results obtained in growth factor-treated breast cancer cells, Ingenuity Pathway Analyses of enriched biological processes reveal that the genes whose promoters are bound by ERRa both in pSKBr3 and LRSKBr3 cells are enriched for functions related to energy metabolism (Fig. 3e). The genes with ERRa promoter recruitment exclusively in LRSKBr3 cells regulate proliferation, invasion, metastasis, growth factor signalling and oxidative stress response. Comparison of the gene ontology also reveals an overlap between the functions of the genes whose promoters are bound by ERRa specifically in LRSKBr3 cells with those bound uniquely on growth factor treatment in parental cells ( Supplementary Fig. 3d). This prompted us to compare the binding profile of ERRa in LRSKBr3 cells to the growth factordependent binding profile of ERRa in SKBr3 cells. We observe that sites bound by ERRa in LRSKBr3 cells but not in pSKBr3 cells are also typically bound by ERRa specifically upon growth factor treatment in SKBr3 cells (Fig. 3f,g and Supplementary  Fig. 3e,f). We also observe that ERRa-binding intensity in lapatinib-resistant cells is closer to that observed upon growth factor treatment than in parental cells (Fig. 3h,i). This indicates   that re-expression of ERRa in resistant cells contributes to re-establishment of the metabolic signature prevailing in growth factor-treated cells. We next asked whether re-expression of ERRa in lapatinibresistant breast cancer cells also contributes to their drug-resistant phenotype. We performed gene expression profiling of LRSKBr3 and compared the data with gene expression profiles of pSKBr3 cells. The acquisition of lapatinib resistance in SKBr3 cells is characterized by the upregulation of genes involved in the control of cell cycle, DNA repair, metastasis, RTK signalling, hypoxia, oxidative stress response (Supplementary Table 3) and glutathione, cysteine and glutamine metabolism (Fig. 3j, left panels). Depletion of ERRa using specific siRNAs in LRSKBr3 cells induces the downregulation of genes controlling cancer signalling along with various metabolic functions, including amino-acid metabolism, acetyl-coA biosynthesis, oxidative stress response, detoxification and regulation of glutamate, cysteine and methionine metabolism (Fig. 3j, right panels) (Supplementary Table 4). Intersecting these gene expression profiles with ERRa-bound promoters in LRSKBr3 cells resulted in a subset of direct ERRa target genes that are oppositely modulated on depletion of ERRa and acquisition of lapatinib resistance in SKBr3 cells ( Supplementary Fig. 3g,h and Supplementary Tables 5, 6 and 7). Pathway analyses reveal that these ERRa direct target genes control key metabolic functions, including oxidative stress response, reactive oxygen species (ROS) detoxification, xenobiotic response, RTK signalling and cellular metabolism, including cysteine and glutamate metabolism ( Supplementary Fig. 3i). This observation prompted us to assess the expression of ERRa target genes involved in glutathionemediated ROS detoxification in parental and resistant cells. We used pharmacological inhibition of ERRa activity using the specific inhibitor Compound 29 (C29) 40 that leads to decreased ERRa expression in both cell lines (Fig. 3k). The expression of the detoxification enzymes SOD2, SOD3, GSR and GPX1 is increased on lapatinib treatment in parental cells and in lapatinib-resistant cells maintained in the drug compared with parental cells (Fig. 3k). Inhibition of ERRa activity by C29 treatment leads to a significant decrease in lapatinib-induced expression of these detoxification target genes, both in lapatinib-sensitive and -resistant SKBr3 cells (Fig. 3k).
ERRa drives the glutamine flux in lapatinib-resistant cells. We investigated the metabolic states prevailing in the resistant SKBr3 cells. We first assayed the flux of nutrients through glycolysis or glutamine oxidation using 13 C6-glucose or 13 C5-glutamine tracer analysis, respectively, on pharmacological inhibition of ERRa activity using C29. As shown in Supplementary Fig. 4, lapatinib treatment induces a decrease in glucose flux in pSKBr3 cells as assayed through 13 C6-glucose incorporation into pyruvate and lactate m þ 3 metabolite pool levels. While lapatinib treatment has no effect on glucose flux in LRSK cells, 13 C6-glucose tracer analysis shows that glucose flux is decreased in resistant cells compared with control-untreated parental cells ( Supplementary  Fig. 4), indicative of an incomplete re-establishment of the glucose flux in resistant cells. Pharmacological inhibition of ERRa activity by C29 treatment did not affect glucose flux in either parental or resistant cells, indicating that the lapatinib-induced decrease in glucose flux in parental cells is independent of ERRa.
Assessment of glutamine metabolism using 13 C5-glutamine tracer analysis allows the study of both forward and reverse glutamine fluxes through the TCA cycle (Fig. 4) 41 . Lapatinib treatment significantly decreased glutamine flux in parental cells while it had only a moderate effect in lapatinib-resistant cells as determined for all the TCA cycle metabolites tested, both in the forward and reverse directions (Fig. 4). Importantly, we show that pharmacological inhibition of ERRa activity using C29 induces a significant decrease in glutamine flux, in both the parental and lapatinib-resistant SKBr3 cells, upon either control or lapatinib treatment. This result indicates that re-expression of ERRa in the lapatinib-resistant cells is essential for re-instatement of glutamine metabolism.

Suppression of ERRa re-sensitizes resistant cells to lapatinib.
Given the involvement of ERRa in the genomic reprogramming of ERRa binding to promoters of genes involved in tumorigenic processes in lapatinib-resistant cells, we tested whether inhibiting ERRa activity could affect the proliferative and migratory potential of pSKBr3 and LRSKBr3 cells. Pharmacological inactivation of ERRa using C29 leads to a decrease in proliferation and migration of sensitive cells (Fig. 5a,b, left panels). While LRSKBr3 cells kept proliferating and migrating on exposure to lapatinib, pharmacological inhibition of ERRa activity using C29 led to a significant decrease in proliferation and migration of the resistant cells (Fig. 5a,b, right panels).
Given the effect of ERRa inhibition on glutamine metabolism, we investigated the effect of glutamine deprivation from the media in both parental and resistant SKBr3 cells. Glutamine deprivation significantly decreased the proliferation of sensitive SKBr3 cells (Fig. 5c). Treatment with C29 further potentiated the effect of glutamine deprivation on proliferation in the sensitive cells. In lapatinib-resistant cells, glutamine deprivation lead to a drastic and significant decrease in proliferation, suggesting that inhibition of glutamine utilization can re-sensitize the cells to lapatinib treatment.
Since the re-expression of ERRa in the resistant cells contributes to the re-instatement of glutamine metabolism, we assessed whether inhibition of ERRa contributes to re-sensitization of the cells to lower doses of lapatinib. We analysed lapatinib sensitivity in SKBr3 cells upon treatment with C29. While inhibition of ERRa activity with C29 had no significant effect on the lapatinib sensitivity of parental SKBr3 cells, inhibition of ERRa in the resistant cells contributed to the re-sensitization of the LRSKBr3 cells to lower doses of lapatinib treatment (Fig. 5d).
ERRa-driven metabolic adaptations restore detoxification capacity. Since the acquisition of lapatinib resistance in SKBr3 cells is characterized in part by ERRa-dependent re-establishment of glutamine metabolism and by an ERRa-target gene signature controlling oxidative stress response and ROS detoxification, we assessed whether re-expression of ERRa in lapatinib-resistant cells could also confer detoxification capacities enabling survival under lapatinib-induced oxidative stress conditions. We first demonstrated that a fraction of the pool of glutamine is used for generation of glutathione in the pSKBr3 (Supplementary Fig. 5). We next monitored the detoxification capacity of the cells by assessing the reduced versus oxidized glutathione (GSH/GSSG) ratio. We observed that inhibition of ERRa activity using C29 treatment leads to decreased GSH and increased GSSG levels, with an overall decrease in the GSH/GSSG ratio, indicative of reduced detoxification capacity and increased oxidative damage (Fig. 6a). Furthermore, we observed a reduction in the steadystate levels of several metabolites involved in the glutamineglutathione biosynthesis pathway on C29-mediated inhibition of ERRa activity in lapatinib-resistant SKBr3 cells (Fig. 6b,c).
We further assessed whether the sustained expression of ERRa in the lapatinib-LRSKBr3 cells contributes to the detoxification potential of resistant cells using antioxidants treatment to counteract the effect of C29. Treatment of the cells with the antioxidants N-acetylcysteine and 6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid (Trolox) rescued the survival of the lapatinib-resistant cells in the context of ERRa inhibition by C29   ( Fig. 6d). We next assessed whether ERRa directly contributes to controlling the levels of ROS produced in the lapatinib-resistant cells. Staining and in vivo imaging quantification of LRSKBr3 cells using the ROS-detection probe DCFDA revealed that therapeutic inhibition of ERRa using C29 leads to a significant increase in the levels of DCFDA-positive cells, suggesting that the sustained expression of ERRa in the lapatinib-resistant cells contributes to ROS detoxification (Fig. 6e).

Inhibition of ERRa impairs the growth of resistant tumours.
To determine the relevance of these observations to the growth of ErbB2-driven mammary tumorigenesis in vivo, we transplanted parental (lapatinib-sensitive) ErbB2-dependent NIC cells into the mammary fat pads of immunocompetent, coisogenic MMTV-Cre transgenic mice. These mice were then randomized to receive either vehicle, C29 alone, lapatinib alone or a combination of lapatinib and C29 treatment ( Fig. 7a and Supplementary Fig. 6a). The growth of lapatinib-sensitive NIC tumours was monitored over a period of 36 days. As seen in the original transgenic mouse model (Fig. 2g), lapatinib treatment decreased tumour growth. While the inhibition of ERRa using C29 alone had little effect on tumour growth, addition of C29 to lapatinib treatment significantly decreased tumour growth compared with either treatment alone (Fig. 7a). Mammary tumours that underwent a complete response, as determined by complete regression of the tumour to a non-palpable state, inevitably recurred in all mice treated with lapatinib alone (Fig. 7b). However, the combination of C29 and lapatinib not only increased the proportion of complete responses but also prevented recurrence in 30% of treated mice over a follow-up period of 20 weeks. To determine whether ERRa inhibition could also affect the growth of tumours that had already acquired lapatinib resistance in vivo, we transplanted an MMTV-NIC ErbB2-driven mammary tumour that had relapsed following lapatinib treatment (Fig. 2g-i) into the mammary fat pads of immunocompetent, coisogenic MMTV-Cre transgenic mice. These mice were then randomized to receive either lapatinib alone or in combination with C29. The resistant NIC tumours grew robustly despite treatment with a high dose of lapatinib, reaching the maximum tumour volume within a 24-day period ( Fig. 7c and Supplementary Fig. 6b). However, concomitant treatment with C29 markedly attenuated tumour growth, and the lapatinib/C29-treated tumours were significantly smaller at the end of the experiment (Fig. 7c, inset). Together, these results indicate an essential role for ERRa in establishing a favourable metabolic context that supports survival on lapatinib insult and favours proliferation of lapatinib-resistant cells, both in vitro and in vivo.

Discussion
Mitogenic signalling is a prominent driver of breast tumorigenic processes, and resistance to TKIs is a common issue in the treatment of HER2-amplified breast tumours in the clinical setting. In this work, the global metabolic regulator ERRa is first identified as a transcriptional mediator translating downstream mitogenic signals into a metabolic signature that is impaired on treatment with the RTK inhibitor lapatinib. The present study further uncovers a molecular mechanism sustaining resistance to lapatinib that involves the constitutive re-expression of ERRa through mTOR re-activation in resistant cells. Reactivation of ERRa transcriptional programmes further elicits a suitable metabolic context enabling cell survival despite constant exposure to lapatinib-induced inhibition of HER family RTK signalling. Importantly, this study demonstrates that pharmacological inhibition of ERRa prevents this metabolic adaptation and restores lapatinib sensitivity in resistant cells in culture and mammary tumours in vivo (Fig. 7d). The data presented herein reveal that growth factor signalling modulates the transcriptional activity of ERRa at a genome-wide level in breast cancer cells. Since poor-prognosis breast tumours display enhanced mitogenic signalling and increased activity of ERRa, our data imply that ERRa mediates a specific transcriptional programme in response to growth factor/RTK signalling in poor-prognosis breast tumours. In addition, ERRa directly regulates the expression of the ERBB2 gene in breast cancer cells in the absence of the oestrogen receptor 28 . Here we show that ERRa in turn regulates tumorigenic processes on stimulation of EGFR/HER2/HER3 signalling, supporting the existence of an ERRa/RTK feed-forward loop further enhancing the mitogenic effect of HER2 in HER2-amplified breast tumours. The mechanisms underlying the development of TKI resistance in breast cancer cells are diverse, yet they all involve the engagement of survival signals redundant to those transduced by the targeted kinases 6,42 . Our observation that the transcriptional profiles of ERRa in lapatinib-resistant cells mirror those observed in growth factor-treated sensitive cells reveals that the resistant cells provide a context favouring a mitogenic-like state affecting ERRa activity despite the constant blockade of EGFR/HER2/ HER3 signalling by lapatinib. The observation that re-expression of ERRa in lapatinib-resistant cells depends on mTOR reactivation also denotes an important role of ERRa in breast cancer biology. Re-activation of mTOR signalling has been described as a mechanism potentially contributing to the development of resistance to various TKIs in cancer cells but the downstream mechanisms by which it does so have been largely unknown. The results presented here describe the downstream effects of constitutive mTOR reactivation and support a model by which the consequential re-expression of ERRa, occurring independently of EGFR/HER2/HER3 and downstream effectors signalling, can mediate at least a subset of the mitogenic effects and mTOR functions in resistant cells, despite the sustained inhibition of RTK signalling. ERRa is a master regulator of energy and cellular metabolism. We demonstrate that lapatinib affects glucose and glutamine fluxes in breast cancer cells. Our observations indicate that re-expression of functional ERRa is required for optimal restoration of glutamine metabolism in resistant cells and that pharmacological inhibition of ERRa activity prevents this metabolic adaptation and desensitizes the cells to lapatinib treatment. Moreover, our data indicate that the constitutive re-expression of ERRa in resistant cells can restore the detoxification capacities and can prevent an increase in the levels of ROS, sustaining cell survival despite the oxidative stress conditions triggered on therapeutic insult (Fig. 7). Several of the genes regulating the oxidative stress response, which have previously been reported as upregulated in lapatinib-resistant cells 21 , are indeed direct targets of ERRa. In addition, the ERRa coactivator PGC-1a was shown to confer improved mitochondrial metabolism and detoxification capacities enabling survival under oxidative stress conditions in aggressive melanoma tumours 43 , and to determine a metabolic transcriptional programme promoting invasion and metastasis in breast cancer 44 . Our functional genomic and metabolomic data coupled with expression profiling suggest that these PGC-1a effects are likely mediated through ERRa in breast cancer cells since the re-expression of ERRa in resistant cells provides the optimal metabolic context required to survive the lapatinib-induced oxidative stress insult.
Importantly, our study suggests that pharmacological inhibition of ERRa activity represents a viable mechanism to counteract lapatinib resistance in breast cancer and to impact on metabolic adaptations occurring in resistant tumours. Our work also demonstrates the value of ERRa inhibition in the context of lapatinib treatment since co-treatment prevented tumour recurrence in 30% of treated mice. This study also supports the value of recent clinical trials assessing the therapeutic benefits of concomitant inhibition of mTOR and of RTK signalling in breast tumours 45 Figure 7 | Inhibition of ERRa impairs the growth of lapatinib-resistant tumours. (a) Lapatinib-naive murine ErbB2-driven mammary tumours were transplanted into the mammary fat pads of MMTV-Cre mice. Growth curves indicate the response of the tumours to vehicle (n ¼ 8), C29 monotherapy (n ¼ 9), lapatinib monotherapy (n ¼ 12) and lapatinib/C29 combination therapy (n ¼ 10). *Po0.05, one-way analysis of variance (ANOVA) with Tukey's post-test. (b) Quantification of the therapeutic response and recurrence in response to sustained treatment with lapatinib and C29 individually and in combination. Partial or no response was defined as no effect of treatment on tumour growth or a reduced rate of growth with incomplete tumour regression. Complete response was defined as total regression of the tumour mass to a non-palpable state. Recurrence was defined as the regrowth of any palpable tumour mass within the follow-up period of 20 weeks. (c) A lapatinib-resistant murine ErbB2-driven mammary tumour was transplanted into the mammary fat pads of MMTV-Cre mice. Growth curves indicate the response of the tumours to lapatinib monotherapy (n ¼ 8) and lapatinib/C29 combination therapy. **Po0.01, one-way ANOVA with Tukey's post-test. Box plot shows the tumour size for both treatment groups at the end of the experiment. P value was generated by comparing the mean tumour size using an unpaired Student's t-test. (d) A working model of the role of the mTOR-ERRa axis in dictating the metabolic response of ERBB2-driven breast cancer cells to lapatinib and the potential for ERRa antagonists to counteract lapatinib resistance in those cells.
effectors through inhibition of ERRa activity might represent an even more efficient way to impinge on metabolic adaptations triggered by mTOR reactivation. Taken together, our study demonstrates that modulation of ERRa promotes metabolic vulnerabilities that can be exploited therapeutically and supports inhibition of ERRa as an adjuvant therapy in advanced HER2positive breast cancer.
In vivo studies. MMTV-NIC and MMTV-Cre transgenic mice (FVB/N background) have been described previously 37,48 . Lapatinib (150 mg kg À 1 ) and C29 (10 mg kg À 1 ) were administered daily by oral gavage as suspensions in 0.5% hydroxypropylmethylcellulose/0.1% Tween 80 (Sigma). Mice were randomly assigned to treatment groups, and tumour size was assessed in a blinded fashion using calliper measurements. Lapatinib administration in MMTV-NIC mice: female MMTV-NIC mice on a pure FVB/N genetic background were monitored for spontaneous mammary tumour formation. These mice developed mammary tumours with an average latency of 124.6±15.3 days, in line with previously published observations 37 . Once tumours were measurable, mice were randomly assigned to receive lapatinib (150 mg kg À 1 ) or vehicle (0.5% hydroxypropylmethylcellulose, 0.1% Tween 80 -Sigma) daily by oral gavage. Group sizes were n ¼ 5 for vehicle and n ¼ 6 for lapatinib. Tumours were measured weekly using calipers, and tumour volume was calculated using the formula: 4.18879 Â (L/2) Â (W/2) 2 . Drug administration and tumour measurement were performed by distinct individuals and those measuring the tumours and analysing the data were blinded to the treatment received by each mouse. Mice were killed when their total tumour burden reached 5 cm 3 , as per guidelines approved by the McGill University Animal Care Committee. Once all vehicle-treated mice had reached the tumour burden end point (6 weeks following the initiation of the experiment), the experiment was terminated. Lapatinib/C29 administration in an immune-competent allograft model: a mammary tumour in an MMTV-NIC mouse that had relapsed following 6 weeks of treatment with lapatinib was surgically excised and transplanted into the mammary fat pads of female MMTV-Cre mice on the same genetic background (FVB/N). This mouse strain was chosen for its immune toleration to Cre recombinase, which we have found to be immunogenic in non-transgenic FVB/N mice receiving MMTV-NIC tumour transplants. Once tumours had reached a size of 5 Â 5 mm (65 mm 3 ), the mice were randomly assigned for treatment with lapatinib (150 mg kg À 1 ) or lapatinib (150 mg kg À 1 ) plus C29 (10 mg kg À 1 ) in 0.5% hydroxypropylmethylcellulose/0.1% Tween 80 by daily oral gavage as described above. Group sizes were n ¼ 8 for lapatinib monotherapy and n ¼ 9 for lapatinib þ C29 combination therapy. Tumour measurements and volume calculations were performed in a blinded fashion as described above. The experiment was terminated after 24 days of treatment when the majority of mice receiving lapatinib monotherapy had reached the maximum tumour size for a single mass (1.5 mm 3 ). Average tumour volumes for the two experimental groups at the end of the experiment were compared using an unpaired Student's t-test. All animal studies were approved by the McGill University Downtown Campus Animal Facility Care Committee (FACC), which is a branch of the McGill University Animal Care Committee (UACC).
Patient-derived xenografts. All human participants provided informed consent for this study, and tissue was collected at the McGill University Health Centre in accordance with the protocols approved by the research ethics board (SUR-99-780). All animal studies were approved by the UACC (2014-7514). PDXs were developed from 1-8-mm 3 fragments of freshly resected breast tumour, submerged in PBS:Matrigel (1:1) (Corning), and transplanted into the fourth mammary fat pad of 5-7 week SCID-beige mice (Charles River) housed under pathogen-free conditions. Tumours were measured twice per week using callipers and collected when the largest tumour dimension reached 10 mm. PDXs were passaged using fragments from collected tumours as described above.
ChIP assays and ChIP-sequencing. For ChIP-sequencing analyses, chromatin was prepared from SKBr3 or BT-474 cells cultured for 24 h in serum-depleted media and further treated with 100 mM EGF or HRG for 90 min prior harvesting. The ChIP primers are listed in Supplementary Table 8. ChIP-seq analyses were performed as previously described 49 . The sequences generated from ChIP-seq were treated as previously described 49 . Briefly, sequences were aligned to the human genome database (Hg19) using BWA v0.5.9 (ref. 50). Peaks were called using MACS v1.4.1 software (mfold ¼ 10, 30; bandwidth ¼ 300; P value cutoff ¼ 1E-5) 51 . Peak annotation, tag directory, bed file generation, merging and de novo motif discovery were performed with the Homer package v3.1 (http://biowhat.ucsd.edu/ homer/ngs/index.html). Tag heatmaps were obtained with Java TreeView (http://jtreeview.sourceforge.net/). Standard ChIP was performed as described previously 46 . Quantification of ChIP enrichment by real-time quantitative PCR was carried out using the LightCycler480 instrument (Roche). ChIPs are normalized against background enrichment on anti-IgG antibody ChIP control and enrichment on a negative control unbound region. Representative graph of three independent experiments performed in triplicates are shown. Statistical significance of standard ChIP is obtained with unpaired Student's t-tests. For ChIPsequencing, a library of enriched DNA segment following the ChIP experiment was prepared according to the ChIP-sequencing library protocol by Illumina and as previously described. Following quality control with an Agilent Bioanalyser 2100 the libraries were sequenced using an Illumina HiSeq 2000 (McGill University and Genome Quebec Innovation Centre).
Expression profiling and gene set enrichment. SKBr3 cells were maintained as described and transfected with siRNAs 60 h before collecting. Cells were treated with 100 mM EGF, HRG or veh for 3 h or with 2 mM lapatinib for 24 h before RNA extraction with the RNeasy Mini Kit (Qiagen) and reverse transcribed using Superscript II (Invitrogen) and analysed by quantitative RT-PCR with SYBR-green (Roche). For expression profiling, RNA was hybridized to Illumina HT12-v4 Expression BeadChIP array according to the manufacturer's protocol. Expression data were normalized using the lumi normalization function in FlexArray (http://genomequebec.mcgill.ca/FlexArray). Differential gene expression was computed using the SAM function in FlexArray, and statistical test for fold changes were performed using Student's t-tests. Normalized data and differentially expressed genes were used as input for determining gene set enrichment analysis (GSEA) using the GSEA software 52,53 and Ingenuity Pathways Analysis software using canonical pathways (Ingenuity(R) Systems, (Ingenuity(R) Systems, http:// www.ingenuity.com/ (IPA)).
Ingenuity pathway analyses for ChIP-seq. ERRa ChIP-seq target genes found within ± 10 kb of binding site were analysed for significant biological pathways using Ingenuity Pathways Analysis software. Fisher's exact test was used to calculate P values representing the probability of association between genes in the data set and the canonical pathways.
Cell proliferation and migration assays. For cell proliferation assays, cells were maintained as described above, seeded in 24-well plates and treated with 5 mM C29 or veh for 24, 48, 72 or 96 h. Cell viability was assessed using crystal violet as previously described 28 . For migration assays, cells were transferred to CIM-16 migration plates (Roche). Cells were allowed to migrate towards FBS and were monitored in real time using Xcelligence RTCA DP instrument (Roche). Statistical significance was calculated using Student's t-test and analysis of variance. For glutamine deprivation, cells were seeded in 96-well plates in either complete media or in glutamine-depleted media ( þ 10% dialysed serum). Next day, cells were treated with 5 mM C29 or vehicle and transferred to IncuCyte ZOOM Live cell analysis system (Essen BioScience) and proliferation was monitored using IncuCyte ZOOM phase-contrast quantification software (2014A). For antioxidant treatments, resistant cells were plated in 96-well plates in presence of lapatinib. The next day, cells were treated with either 500 mM N-acetylcysteine or 500 mM 6-hydroxy-2,5,7,8-tetramethylchroman-2-carboxylic acid (Trolox), in presence or absence of 5 mM C29. The cells were transferred to the IncuCyte ZOOM Live cell analysis system (Essen BioScience) and quantified as described above.
Live cell imaging and ROS quantification. Lapatinib-resistant SkBr3 cells were seeded in 24-well plates at a density of 2 Â 10 4 per well and treated with lapatinib alone (2 mM) or lapatinib þ C29 (5 mM) for 48 h. Cells were then stained with 5 mM H2DCFDA (Thermo Fisher, D399) for 30 min under normal growth conditions and imaged in phenol-red free DMEM at 37°C and 5% CO 2 using a Zeiss Axiovert 200M equipped with a Zeiss LD A-Plan Â 20 objective. Total and DCFDA-positive cells in a minimum of 10 independent fields (at least 1,000 total cells per condition) were quantified using ImageJ. The experiment was performed three times and statistical significance was determined using Student's t-test.
GC/MS, LC/MS and mass isotopomer distribution analysis. Gas chromatography/ mass spectrometry (GC/MS), liquid chromatography/MS (LC/MS) and Mass Isoprotomer Distribution analysis are described in Supplementary Methods. Briefly for GC/MS, pSKBr3 and LRSKBr3 cells were seeded in triplicate in 35-mm plates, grown to 50% confluence and treated with control (DMSO 0.1%) and/or Lapatinib and/or C29 for 24 h before changing media to 1 ml DMEM high glucose supplemented with 10% dialysed FBS (Wisent) and either 10 mM U-13 C 6 -glucose (uniformly labelled, Cambridge Isotope Laboratories, CLM-1396, 99%) or 2 mM U-13 C 5 -glutamine (Cambridge Isotope Laboratories, CLM-1822, 97-99%). After 30 min pulses, cells were collected and metabolites were derivatized using N-tert-Butyldimethylsilyl-N-methyltrifluoroacetamide (MTBSTA, 394882, Sigma) and subjected to GC/MS. For LC/MS analysis of glutamine flux into glutathione, parental and lapatinib-resistant SkBr3 cells were grown in DMEM containing 2 mM 13 C 5 -labelled glutamine (CML-1822; Cambridge Isotope Laboratories, Inc.) or unlabelled glutamine for 6 h. Metabolites were then extracted for LC/MS and analysed as described in Supplementary Methods (Supplementary Tables 9 and 10). Matrix corrections for tracer analysis were carried out as described 54 .
Data availability. All microarray and sequencing raw and processed data have been deposited in the GEO repository and are available under the accession numbers GSE81546 and GSE81651, respectively.