PI3K Inhibition Activates SGK1 via a Feedback Loop to Promote Chromatin-Based Regulation of ER-Dependent Gene Expression

SUMMARY

The PI3K pathway integrates extracellular stimuli to phosphorylate effectors such as AKT and serum-and-glucocorticoid-regulated kinase (SGK1). We have previously reported that the PI3K pathway regulates estrogen receptor (ER)-dependent transcription in breast cancer through the phosphorylation of the lysine methyltransferase KMT2D by AKT. Here, we show that PI3Kα inhibition, via a negative-feedback loop, activates SGK1 to promote chromatin-based regulation of ER-dependent transcription. PI3K/AKT inhibitors activate ER, which promotes SGK1 transcription through direct binding to its promoter. Elevated SGK1, in turn, phosphorylates KMT2D, suppressing its function, leading to a loss of methylation of lysine 4 on histone H3 (H3K4) and a repressive chromatin state at ER loci to attenuate ER activity. Thus, SGK1 regulates the chromatin landscape and ER-dependent transcription via the direct phosphorylation of KMT2D. These findings reveal an ER-SGK1-KMT2D signaling circuit aimed to attenuate ER response through a role for SGK1 to program chromatin and ER transcriptional output.

In Brief

Toska, Castel, et al. show that the PI3K pathway propagates its effects to control chromatin and estrogen receptor (ER) function through SGK1, a PI3K effector. PI3K inhibitors, via a negative-feedback loop, activate SGK1, which phosphorylates the histone lysine methyltransferase KMT2D to attenuate its activity and regulate ER response.

INTRODUCTION

The serum and glucocorticoid-regulated kinase (SGK) family consists of three different isoforms—SGK1, SGK2, and SGK3—that are members of the AGC (PKA-, PKG-, and PKC-related) family of serine-threonine kinases of the human kinome (Pearce et al., 2010). SGK1, the most studied member of the SGK family, was initially identified in a screen for glucocorticoid-responsive transcripts as an immediate early gene transcriptionally induced by glucocorticoids (Webster et al., 1993). Additional studies showed that the expression of SGKs is rapidly induced by treatment with steroid hormones (Wang et al., 2011, 2014). As both SGK1 mRNA and protein exhibit a short half-life, the induction of SGK1 expression is transient, indicating that its activity is mostly regulated by a localized increased expression mechanism (Di Cristofano, 2017; Webster et al., 1993).

Several years after SGK1 identification, it was discovered that SGK1 functions downstream of the phosphoinositide 3-kinase (PI3K) pathway (Kobayashi and Cohen, 1999; Park et al., 1999). Activation of PI3K by mutations in PIK3CA, the gene coding for PI3Kα, or by amplification of upstream tyrosine kinase receptors leads to increased phosphatidylinositol-(3,4,5)-triphosphate (PIP3) synthesis, which, in turn, initiates a downstream signaling cascade involving the phosphorylation and activation of AKT by PDK1 at T308 and mammalian target of rapamycin mTORC2 at S473 (Alessi et al., 1997; Manning and Toker, 2017; Sarbassov et al., 2005). Similarly to AKT, activation of SGKs is mediated through a phosphorylation by mTORC2 at the hydrophobic motif (S422 in SGK1) and a subsequent PDK1-dependent phosphorylation at the activation loop (T256) (García-Martínez and Alessi, 2008; Pearce et al., 2010). Moreover, upon activation, like AKT, endogenous SGK1 rapidly translocates into the nucleus, where it can interact with nuclear substrates (Buse et al., 1999). However, in contrast to AKT, whose physiological targets are well studied, the cellular functions of SGKs and their physiological targets remain largely uncharacterized.

Recent studies from our group and others have demonstrated that SGKs seem to play a critical role in cancer cell survival, especially under scenarios where AKT activity has been challenged (Bago et al., 2016; Castel et al., 2016). This is most likely explained by the ability of SGK1 to replace AKT in phosphorylating its downstream targets. In this regard, we have shown that, upon PI3Kα inhibition and full suppression of AKT, SGK1 sustains mTORC1 activity and cell growth through the direct phosphorylation and inhibition of TSC2 and FOXO, bona fide AKT targets, thus conferring resistance to PI3Kα inhibitors (Castel et al., 2016). The reason for this observation likely resides in the fact that, within its catalytic domain, SGK1 is 54% homologous to AKT, and both kinases share the same phosphorylation consensus motif (RXRXXS/T) (Alessi et al., 2009).

Activation of the PI3K signaling regulates growth, survival, and metabolism of cells; as a result, pharmacological inhibition of this oncogenic pathway is of high interest (Cantley, 2002; Engelman, 2009; Thorpe et al., 2015). PI3Kα inhibitors have shown improved clinical outcome in estrogen receptor (ER)-positive and PIK3CA mutant breast cancers that have elevated PI3K signaling (Juric et al., 2017, 2018), but not all tumors are sensitive to these inhibitors. We have observed a highly uniform adaptive tumor response to PI3K inhibitors that is characterized by an increase in ER-dependent transcription, which mediates therapeutic resistance (Bosch et al., 2015). To this end, a recent large phase-III study has shown that the addition of PI3Kα inhibitor BYL719 to anti-endocrine therapy greatly improved progression-free survival in ER+/PIK3CA mutant metastatic breast cancer (F. André et al., 2018, ESMO 2018 Congress, conference).

Searching for the underlying mechanism(s), we have found that the oncogenic PI3K pathway tightly suppresses ER-mediated transcription via direct phosphorylation of the histone methyltransferase KMT2D by AKT1, and conversely, PI3Kα inhibitors activate KMT2D, influencing the chromatin landscape of breast cancer and unleashing ER-dependent transcription (Toska et al., 2017). KMT2D (also known as MLL2 or MLL4) is a member of the COMPASS (Complex Of Proteins Associated with Set1) family and has been identified as a major histone methyltransferase that activates transcription via mono- and di-methylation of histone 3 at lysine 4 (H3K4), which are active chromatin marks found mainly at enhancers (Froimchuk et al., 2017; Hu et al., 2013). Hence, KMT2D is essential for enhancer activation and transcriptional output.

As a mediator of PI3K signaling in the nucleus, and given that SGK1 shares high similarity with AKT at the catalytic domain and that both kinases share the same phosphorylation consensus motif (RXRXXS/T), we reasoned that SGK1 is likely to be an important regulator of PI3K-dependent biological processes, including chromatin-based transcriptional regulation via KMT2D phosphorylation. Likewise, the observation that SGK1 is an estrogen-inducible kinase led us to hypothesize that the upregulation of ER activity upon PI3K inhibition would induce SGK1 expression that, in turn, phosphorylates KMT2D to inhibit ER transcriptional output as a regulatory feedback loop. Therefore, in this work, we set out to elucidate whether ER activation by PI3K inhibition can generate a negative-feedback loop through SGK1 to phosphorylate KMT2D and promote chromatin-based regulation of gene expression.

RESULTS

ER Activates SGK upon PI3Kα Pathway Inhibition

SGK expression is tightly regulated by different hormone receptors, including estrogen and androgen receptors in breast and prostate cancer, respectively (Wang et al., 2011, 2014). Furthermore, the activity of these hormone receptors is also regulated by the PI3K/AKT pathway, which is commonly hyperactivated in these cancers. Inhibition of PI3K paradoxically activates ER-mediated transcription in breast cancer cells as a survival feedforward mechanism (Bosch et al., 2015).

In order to investigate whether SGK expression is upregulated upon PI3Kα inhibition via the ER, we measured the mRNA expression of SGK1, SGK2, and SGK3 isoforms in the ER+/PIK3CA mutant T47D and MCF7 breast cancer cells upon treatment with estradiol, a PI3Kα inhibitor (BYL719), or both treatments at different time points. BYL719 has been predominantly used throughout our study, as it is a highly specific and alpha-selective PI3K inhibitor (Fritsch et al., 2014). Administration of estradiol resulted in a modest increase in SGK1, SGK2, and SGK3 mRNA. On the other hand, PI3Kα inhibition by BYL719 resulted in a far greater enhanced transcription of SGK1, SGK2, and SGK3 mRNA, and this effect was further augmented if estradiol was also added (Figure 1A). An elevated expression of SGKs was also observed upon treatment of MCF7 and T47D cells with several PI3K/AKT inhibitors (Figures S1A and S1B). Next, we treated T47D and MCF7 cells with the ER degrader fulvestrant and observed decreased mRNA expression of SGKs transcripts, further suggesting that their expression is transcriptionally regulated by ER (Figures S1C and S1D). Moreover, by performing chromatin immunoprecipitation experiments followed by qPCR analysis (ChIP-qPCR), we observed that PI3Kα inhibition increased the occupancy of ER and phosphorylated (S5) Pol II, a marker of transcriptional activation, at the promoters of SGK1, SGK2, and SGK3 (Figure 1B). We then used genome-wide assays for transposase-accessible chromatin using sequencing (ATAC-seq) in cells treated with PI3Kα inhibitor (Toska et al., 2017) to study the chromatin accessibility at SGK promoters. ATAC-seq captures open chromatin by utilizing the transposase-mediated insertion of sequencing primers into accessible chromatin to profile the epigenome of cells (Buenrostro et al., 2013). Using this assay, we found that treatment with a PI3Kα inhibitor led to an enhanced chromatin accessibility at the promoters of SGKs (Figure 1C). Consistent with these findings, immunoblotting for SGK1 and SGK3, which have well-suited antibodies available, revealed an enhanced expression of SGK1 and SGK3 proteins at 24 or 32 h of PI3Kα inhibition in T47D or MCF7 cells (Figures 1D and S1E). Altogether, these results suggest that SGKs are directly regulated by ER and that enhanced ER activation by PI3K pathway inhibition results in a marked increase of SGK expression.

Figure 1. ER Activates SGK upon PI3Kα Pathway Inhibition.

(A) mRNA levels of SGK1, SGK2, and SGK3 measured by RT-qPCR in hormone-depleted MCF7 or T47D cells for 3 days followed by treatment with DMSO, estrogen E2 (100 nm), BYL719 (1 μM), or E2 + BYL719 for 2, 4, 8, 24, 48, and 72 h. Data are representative of at least three biological repeats.

(B) ChIP-qPCR for ER and Pol II (pS5) binding at SGK1, SGK2, and SGK3 promoters in T47D and MCF7 cells. Data represent two biological repeats.

(C) ATAC-seq (Toska et al., 2017) open chromatin at SGK1, SGK2, and SGK3 promoters in T47D cells upon treatment with BYL719 (1 μM) for 24 h. The peaks are presented as reads per million (RPMs). ATAC-seq data represent two biological repeats.

(D) Immunoblotting with the indicated antibodies in T47D cells treated with DMSO or BYL719 for 24, 32, or 48 h. Data are representative of at least three biological repeats.

The p values were calculated using Student’s t test. Error bars denote ± SD.

See also Figure S1.

SGK1 Interacts with and Phosphorylates KMT2D In Vivo and In Vitro

We have previously shown that AKT regulates the function of the epigenetic regulator KMT2D by a direct phosphorylation event at the AGC kinase consensus site of KMT2D, S1331 (Toska et al., 2017). SGKs share high similarity with AKT at the catalytic domain; as a consequence, many substrates that contain the AGC kinase consensus motif RXRXX(S/T) are mutual between the two kinases (Alessi et al., 2009). Therefore, we reasoned that SGKs could also phosphorylate and regulate KMT2D function. Using an in vitro kinase assay with recombinant active SGK1 and full-length hemagglutinin (HA)-tagged KMT2D as a substrate immunoprecipitated from 293T cells, we confirmed the ability of SGK1 to directly phosphorylate S1331, as detected by using a degenerated phospho-specific AGC motif antibody and phospho-KMT2D (S1331) antibodies (Figure 2A). Mutation of the phosphorylatable serine into the non-phosphorylatable alanine (S1331A) completely abolished the ability of SGK1 to phosphorylate KMT2D in vitro. SGK1 could also phosphorylate a short fragment of KMT2D (amino acids 1222–1819) where the phosphorylatable S1331 resides but not when the fragment contained the S1331A mutation (Figure S2A).

Figure 2. SGK1 Interacts with and Phosphorylates KMT2D In Vivo and In Vitro.

(A) In vitro kinase assay using HA-KMT2D wild-type (WT) or S1331A as a substrate immunoprecipitated from 293T cells and recombinant His-SGK1. Data are representative of at least three biological repeats. IP, immunoprecipitation. EV, empty vector.

(B) In vitro kinase assay using HA-KMT2D WT or S1331A as a substrate and recombinant His-SGK2. Data are representative of at least three biological repeats.

(C) In vitro kinase assay of HA-KMT2D wild-type or S1331A substrate and recombinant His-SGK3. Data are representative of at least three biological repeats.

(D) Co-immunoprecipitation assay in 293T cells transfected with HA-KMT2D and FLAG-SGK1 and probed with HA and FLAG antibodies. WCL, whole cell lysates. Data are representative of at least three biological repeats.

(E) Endogenous co-immunoprecipitation between KMT2D and SGK1 in JIMT1 cells. Data are representative of two biological repeats.

(F) MCF7 or T47D cells were treated with control siRNAs or siRNAs targeting SGK1 and were subjected to immunoblotting for phosphorylated KMT2D (pKMT2D) at S1331. Data are representative of two biological repeats.

(G) Immunoblot of phosphorylated KMT2D (S1331) in T47D cells treated with DMSO or SGK1 inhibitor (SGK1-inh) (10 μM). Data are representative of at least two biological repeats.

(H) Michaelis-Menten reactions to determine ATP K_m using recombinant glutathione S-transferase (GST)-tagged full-length AKT1 or SGK1 kinases with KMT2D (GRGRGRARLKSTA) peptide. Data represent two biological repeats.

See also Figure S2.

The SGK family consists of three separate but highly homologous isoforms (SGK1, SGK2, and SGK3) encoded by three different genes (Di Cristofano, 2017). To this end, we also performed similar in vitro kinase assays, using recombinant active SGK2 or SGK3, and confirmed that both isoforms phosphorylate the wild-type KMT2D at S1331 in vitro but not the phospho-dead mutant of KMT2D (S1331A) (Figures 2B and 2C). SGKs may have both overlapping and specific substrates, cellular functions, and subcellular localization, most of which remain to be elucidated. However, since the expression of SGK isoforms is upregulated upon PI3K inhibition and they all phosphorylate KMT2D in vitro, we posit that, in the context of KMT2D, their roles might be comparable. For this reason, and based on our previous findings regarding SGK1 upregulation as a mechanism of resistance to PI3Kα inhibitors (Castel et al., 2016), we focused on the SGK1 isoform for the remainder of our experiments. Next, we performed co-immunoprecipitation assays of recombinant HA-tagged KMT2D and FLAG-tagged SGK1, which revealed that KMT2D physically interacts with SGK1 in cells (Figure 2D). We further confirmed the interaction between endogenous KMT2D and SGK1 by co-immunoprecipitation (Figure 2E). To validate whether KMT2D is a physiological target of SGK1 in cells, we next transfected MCF7 or T47D breast cancer cells with small interfering RNA (siRNA) targeting SGK1 and probed for total and phospho-KMT2D (S1331). Knockdown of SGK1 attenuated KMT2D phosphorylation, as demonstrated by a reduction in phospho-KMT2D (S1331) levels compared to cells treated with control siRNAs (Figure 2F). Furthermore, we treated cells with a recently characterized SGK1 inhibitor (SGK1-inh) from our laboratory and checked the levels of phospho-KMT2D (S1331). SGK1-inh exhibits a half maximal inhibitory concentration (IC50) of 4.8 nM in recombinant SGK1 kinase assay and a concentration of 10 mM to fully inhibit endogenous SGK1. This is not due to a low activity against SGK, which is at the low nano-molar range, but because the drug exhibits poor plasma membrane permeability (Castel et al., 2016). Treatment with SGK1-inh also led to a reduction in the downstream effector of SGK1, phospho-NDRG1 (T346), and phospho-KMT2D (S1331) (Figure 2G). Of note, we have previously demonstrated that treatment with PI3K and AKT inhibitors reduced the phosphorylation of KMT2D at S1331 (Toska et al., 2017). Consistently, treatment of T47D cells with PI3K inhibitors (BKM120 and BYL719) or AKT inhibitor (MK2206) reduced the phosphorylated levels of KMT2D, but treatment with the mTOR inhibitor RAD001, which would inhibit S6K kinase, did not have such an effect (Figure S2B). Thus, we conclude that the AGC kinases AKT and SGK, but not S6K kinase, phosphorylate KMT2D at S1331.

We next performed enzyme-based Michaelis-Menten assays using recombinant active full-length SGK1 and AKT1 recombinant proteins in the presence of a KMT2D substrate peptide to characterize the kinetic properties of S1331 phosphorylation. Since both enzymes are able to phosphorylate similar substrates, we decided to assess their catalytic properties with the goal of providing an explanation for the biological significance of this redundancy. First, we confirmed that both SGK1 and AKT1 phosphorylate S1331 in vitro, perfectly following a Michaelis-Menten model (Figure 2H). We subsequently set out to quantify and compare the catalytic efficiency of both enzymes against KMT2D peptide at S1331 by determining the kinetic parameters: V_max, K_m, and K_cat. As reflected by the K_cat values, which measure the turnover rate of the enzymes, and K_cat/K_m values, which measure catalytic efficiency, we discovered that AKT1 was ~10 times more efficient than SGK1 at phosphorylating S1331 in vitro (Figure 2H). Intrigued by these results, we decided to perform enzymatic reactions to compare the catalytic efficiency of these enzymes in other shared substrates, such as the canonical AKT1 and SGK1 substrate, FOXO1 T24 (Biggs et al., 1999; Di Pietro et al., 2010), and two known AGC phosphorylation sites that were recently characterized by our laboratory to be phosphorylated by SGK1: TSC2 T1462 and TSC2 S1798 (Castel et al., 2016). The kinase assays confirmed that AKT1 and SGK1 kinases were both able to successfully phosphorylate these sites completely following a Michaelis-Menten model, and similarly to KMT2D at S1331, AKT1 was more efficient at phosphorylating FOXO1 T24, TSC2 T1462, and TSC2 at S1798 than SGK1 in vitro (Figure S2C). Of note, TSC2 S1798 has not been reported before as an AKT1 target, and our data demonstrate that this site is also a shared AKT1/SGK1 site. Although these results suggest that AKT1 is a more efficient kinase at phosphorylating these shared substrates, this would need to be validated in vivo where the local concentration of the kinase could modify the affinity toward the substrate.

SGK1 Induces a Repressive Chromatin State at Canonical ER Target Loci

Since SGK1 phosphorylates KMT2D, we hypothesized that SGK1 may impact the epigenome of breast cancer cells. To explore the role that SGK1 plays on the global chromatin landscape, we used genome-wide ATAC-seq in breast cancer cells transfected with SGK1 or control vector (Figure 3A). Since SGK1 is rapidly and transiently regulated both at the transcriptional and post-transcriptional levels, we have used for our experiments an SGK1 mutant that lacks the PEST domain present within the first 60 amino acids, increasing its protein stability. This construct also contained the phosphomimetic mutation S422D, which ensures the full activity of the enzyme. Consistent with this, an increase of phosphorylated NDRG1 (T346) and phosphorylated KMT2D (S1331) levels is observed upon SGK1 (S422D) expression (Figure S3A). ATAC-seq assays revealed that overexpression of SGK1 results in substantial changes in chromatin accessibility. This is observed by the marked number of accessible sites that are lost or gained genome-wide upon SGK1 expression in comparison to the control vector, as indicated in red in Figure 3A and represented in the heatmaps in Figure 3B. Motif analyses showed an enrichment of motifs for nuclear receptor, predictive of ER, and FOXA1, a key regulator of ER function, in the lost accessible sites upon SGK1 expression (Figure 3C). Indeed, when we overlay ER and FOXA1 ChIP sequencing (ChIP-seq) performed in our laboratory in T47D cells (Toska et al., 2017) or from ENCODE (Encyclopedia of DNA Elements), we observed a striking occupancy of these transcription factors (TFs) in the sites that will lose chromatin accessibility upon SGK1 expression. These results suggest that ER and FOXA1 occupancy will likely be reduced in the lost accessible sites, since these sites will be closed upon SGK1 expression (Figure 3D). We then examined how SGK1 affects the chromatin accessibility at the loci of ER canonical genes such as WISP2, LIF, XPB1, and IGFBP4, among others. Accessibility at these regulatory genes is markedly reduced upon SGK1 expression, suggesting that SGK1 alters ER-mediated transcription (Figures 3E and S3B).

Figure 3. SGK1 Induces a Repressive Chromatin State at Canonical ER Target Loci.

(A) MA plot of ATAC-seq results in T47D cells transfected with empty vector (EV) or constitutively active SGK1 (S422D). The x axis represents mean log₂ of normalized counts (normC counts), and the y axis represents log₂ fold change (log2FC). Red dots represent the significant differential accessible sites upon SGK1 expression.

(B) Heatmap of gained or lost accessible sites in EV control cells and SGK1 (S422D) cells, shown in a horizontal window of ± 5 kb from the peak center.

(C) Enriched motifs in lost accessible sites, nuclear hormone receptor (p = 1 × 10⁻¹⁰), and FOXA1 (p = 1 × 10⁻²²).

(D) Heatmap of ER ChIP-seq (Toska et al., 2017), ER ENCODE (ENCFF532GWP), FOXA1 ChIP-seq (Toska et al., 2017), and FOXA1 ENCODE (ENCFF845PAS) at the sites that lose accessibility upon SGK1 (S422D) expression.

(E) Examples of ATAC-seq of chromatin accessibility regions aligned with the ER ChIP-seq (Toska et al., 2017) of binding regions at canonical ER loci. The peaks are presented as reads per million (RPMs). ATAC-seq data are representative of a biological repeat.

See also Figure S3.

These results show that SGK1 influences the chromatin state at ER regulatory regions and define the regulation of chromatin accessibility as a previously unknown function for SGK1 in breast cancer.

SGK1 Attenuates KMT2D and H3K4me1/2 Occupancy at ER Loci

We have previously shown that phosphorylation of KMT2D attenuates its function and that inhibition of the PI3K/AKT pathway affects the occupancy of KMT2D at ER loci (Toska et al., 2017). We have also indicated that the phospho-dead mutant of KMT2D (S1331A) has a differential ability to bind DNA as it is recruited to the chromatin at a higher level, phenocopying the results observed with BYL719 (Toska et al., 2017). Thus, given these findings and since SGK1 phosphorylates KMT2D at S1331, we next sought to determine how SGK1 influences KMT2D binding at the chromatin. In this regard, we overex-pressed constitutively active SGK1 (S422D) or control empty vector (EV) in T47D cells and performed ChIP-qPCR experiments. We observed that expression of SGK1 resulted in a marked decrease of binding of KMT2D to the regulatory regions of ER target genes (Figure 4A). Next, we studied the effects of pharmacological inhibition of SGK1 in KMT2D binding in breast cancer cells. We treated T47D and MCF7 cells with a SGK1 inhibitor (SGK1-inh) and performed ChIP-qPCR analysis in T47D cells. Treatment of T47D cells, as well as MCF7 cells with SGK1 inhibitor led to a decrease in NDRG1 phosphorylation (Figures S3C and S3D). Moreover, an increase in KMT2D binding at ER loci was observed upon SGK1 inhibition in both cell lines, consistent with our aforementioned findings (Figures S3C and S3D). Likewise, similar results were also obtained when SGK1 was stably knocked down using a doxycycline-inducible construct targeting SGK1 (Figure S3E).

Figure 4. SGK1 Attenuates KMT2D and H3K4me1/2 Occupancy at ER Loci.

(A) KMT2D ChIP-qPCR in T47D cells overexpressing empty vector (EV) or SGK1 (S422D) at canonical ER target genes. The p values were calculated using Student’s t test. Error bars denote ± SD. Data represent three biological repeats.

(B) Heatmaps of H3K4me1 and H3K4me2 in T47D cells expressing EV or SGK1 (S422D) shown in a horizontal window of ± 5 kb from the peak center.

(C) Average H3K4me1 or H3K4me2 read density plots at bound peaks in T47D cells expressing EV or SGK1 (S422D) shown in a horizontal window of ± 5 kb from the peak center.

(D) Box plot representing mean signal across H3K4me1 or H3K4me2 peaks upon SGK1 (S422D) expression in T47D cells. The p values were calculated using the Mann-Whitney test.

(E) Heatmaps of H3K4me1 or H3K4me2 enrichment at ER sites (ENCODE and ENCFF532GWP) in T47D cells expressing control or SGK1. Also shown is the FOXA1 binding obtained from ENCODE (ENCFF845PAS) at ER sites ± 5 kb from the peak center.

(F) Box plot representing the mean signal of H3K4me1 or H3K4me2 peaks after SGK1 expression across ER sites (ENCODE and ENCFF532GWP). The p values were measured using the Mann-Whitney test.

(G) Box plot showing mean signal of H3K4me1 and H3K4me2 across peaks that lose chromatin accessibility upon SGK1 overexpression in T47D cells. The p values were calculated using the Mann-Whitney test. ChIP-seq data are representative of a biological repeat.

See also Figure S4.

KMT2D is a H3K4 mono- and di-methylation transferase responsible for catalyzing these modifications associated with active transcription (Shilatifard, 2012). To this end, we also explored whether SGK1 affects the H3K4me1 and H3K4me2 occupancy globally. We performed ChIP-seq analyses of H3K4me1 and H3K4me2 in T47D cells expressing control EV or SGK1 (S422D). Analyses of ChIP-seq data for H3K4me1 and H3K4me2 abundance revealed a global reduction of these marks genome-wide in T47D cells upon SGK1 expression (Figures 4B and 4C). A more quantitative analysis, as measured by the mean signal per peak for each condition, demonstrated that the changes in global occupancy of H3K4me1 and H3K4me2 upon SGK1 expression were statistically significant; p = 2.2 · 10⁻¹⁶ for both H3K4me1 and H3K4me2 (Figure 4D). Importantly, similar significant downregulation in global H3K4me1 and H3K4me2 abundance was also observed in MCF7 breast cancer cells upon SGK1 expression (Figures S4A, S4B, and S4C) (H3K4me1: p = 2.2 · 10⁻¹⁶; and H3K4me2: p = 2.6 · 10⁻⁷). Of note, overexpression of SGK1 (S422D) also led to an increase of phosphorylated KMT2D at S1331 and phosphorylated NDRG1 (T346) as controls (Figure S4D). These observations suggest that the effects of SGK1 on H3K4 mono- and di-methylation global occupancy are consistent across ER+ breast cancer.

To observe the effects of SGK1 on H3K4me1 and H3K4me2 binding at ER loci, we overlaid H3K4me1 and H3K4me2 peaks at ER sites obtained from ENCODE in T47D cells, given that ER ChIP-seq is available in these cells. Once again, in the ER binding sites, we also observed a significant downregulation of the binding of H3K4me1 (p = 8.392 · 10⁻¹⁴⁾) and H3K4me2 (p = 1.1 · 10⁻⁹) upon SGK1 expression (Figures 4E and 4F)— with a more striking effect on H3K4me1 that is most likely due to KMT2D being a major methyltransferase at enhancers (Hu et al., 2013)—and that ER mainly binds at enhancer regions (Carroll et al., 2005). ChIP-qPCR experiments in MCF7 cells also showed that SGK1 overexpression abrogated the occupancy of H3K4me1/2 at canonical ER target genes (Figure S4E). Of note, we observed an effect of SGK1 on H3K4me1, a marker mainly associated with enhancer elements, at the GREB1 promoter. Studies have shown that KMT2D and its paralog KMT2C may also provoke monomethylation of promoter regions (Cheng et al., 2014). However, as expected, our global ChIP-seq analyses shows that H3K4me1 is mainly enriched in the flanking enhancer regions and not at transcription start sites (TSSs), in contrast to H3K4me2, a marker found at enhancer and promoter regions, that is also enriched at TSSs (Figure S4F). Moreover, we also observed a significant reduction of H3K4me1 (p = 6.167 · 10⁻⁶) and H3K4me2 (p = 0.014) in the genomic sites that have lost chromatin accessibility in the setting of SGK1 expression (Figure 4G). Together, these results are consistent with the notion that the phosphorylation event of KMT2D by SGK1 suppresses KMT2D function, leading to a reduction of global H3K4me1/2 levels and at ER loci. Since these marks are essential for gene activation, these results also suggest that elevated SGK1 would alter ER target gene expression.

SGK1 Regulates ER-Dependent Transcription and the Recruitment of the ER-FOXA1-PBX1 Regulatory Network

ER is the master transcription factor in ER+ breast cancer, and its function is tightly controlled by cooperating transcription factors, including FOXA1 and PBX1, whose binding at cis-regulatory elements is associated with the active histone modifications, H3K4me1/2 (Hurtado et al., 2011; Jozwik and Carroll, 2012; Magnani et al., 2011). We have recently shown that the occupancy of ER and its cooperating transcription factors are regulated by KMT2D (Toska et al., 2017). Based on these findings, and since SGK1 disturbs KMT2D occupancy and the binding of H3K4me1 and H3K4me2 at ER target genes, we sought to investigate how SGK1 affects the occupancy of the ER-FOXA1-PBX1 regulatory network at shared loci. We performed ChIP-qPCR experiments in T47D cells where a constitutively active SGK1 or an EV was overexpressed. In the presence of SGK1, we observed a decrease in the occupancy of ER, FOXA1, and PBX1 (Figure 5A) at common cis-regulatory elements. In a similar way, we performed ChIP-qPCR experiments to evaluate ER-FOXA1-PBX1 occupancy in MCF7 cells where SGK1 was overexpressed, and similar results were obtained (Figure S5A). ChIP-qPCR experiments in T47D or MCF7 cells treated with the SGK1 inhibitor also revealed a significant increase of binding of this regulatory network at shared target genes loci upon SGK1 inhibition (Figures 5B and S5B). Consistent with this, SGK1 knockdown also led to an enhanced binding of the ER-FOXA1-PBX1 regulatory network at shared ER target genes (Figure S5C). Taken together, these results reveal that SGK1, through the posttranslational modification of KMT2D, affects the occupancy of the intricate network that contains ER and the coregulators critical for ER function, FOXA1 and PBX1.

Figure 5. SGK1 Regulates ER-Dependent Transcription and the Recruitment of the ER-FOXA1-PBX1 Regulatory Network.

(A) ChIP-qPCR for ER, FOXA1, PBX1, and immunoglobulin G (IgG) control in promoter or enhancer regions of shared target genes in T47D cells overexpressing empty vector (EV) or constitutively active SGK1 (S422D).

(B) ChIP-qPCR for ER, FOXA1, PBX1, and IgG in shared target gene loci in T47D cells treated with DMSO or the SGK1 inhibitor (SGK1-inh) (10 μM) for 24 h.

(C) Expression of candidate target genes in EV or SGK1 (S422D) overexpressed in T47D cells as measured by RT-qPCR in hormone-depleted cells for 3 days and treatment with DMSO or E2 (100 nM) for 24 h.

(D) Expression of candidate target genes in control EV, SGK1-sg3, or SGK1-sg4 MCF7 cell transduced with dCAS9-VP64–2A-GFP-HSF1-P65 vector. Also shown is the western blot testing of SGK1 in these cells.

(E) RT-qPCR in hormone-depleted T47D cells subjected to control or SGK1 siRNAs and treatment with DMSO or with E2 (100 nM), BYL719 (1 μM), or both for 24 h. The p values were calculated using Student’s t test. Error bars represent ± SD. All data represent at least two biological repeats.

See also Figure S5.

Next, we studied the transcriptional activity of ER upon SGK1 expression. RT-qPCR analyses demonstrated that overexpression of constitutively active SGK1 in MCF7 and T47D cells resulted in a reduction of the expression of ER canonical genes that are also shared with FOXA1 and PBX1, such as cFOS, IGFBP4, and GREB1, among others (Figures 5C and S5D). These significant changes in RNA expression upon SGK1 over-expression correlate with the repressive chromatin openness caused by SGK1, highlighting the importance of accessibility in the regulation of ER-mediated transcription. To observe whether these transcriptional changes can also be reflected at the protein level, we treated T47D cells with estradiol and/or BYL719 for 24 h and immunoblotted for a subset of canonical ER target genes. As expected, we observed that the protein up-regulation of these target genes upon E2 and/or BYL719 treatment is partially prevented upon transfection of SGK1 S422D (Figure S5E).

To study the transcriptional changes in a more physiologically relevant way, we generated MCF7 cells that show efficient transcriptional activation of the endogenous SGK1 promoter using the CRISPR-Cas9 Synergistic Activation Mediator (SAM) technology. As described by Feng Zhang and colleagues, the combination of (1) single guide RNA (sgRNA); (2) inactive CAS9 to prevent CAS9 cutting; and (3) the presence of the transcription coactivators VP64, HSF1, and P65, which are modeled after natural transcriptional activation processes, compromise a transcription activation system for achieving robust, sgRNA-mediated gene upregulation (Konermann et al., 2015). To this end, MCF7 cells were transduced with catalytically inactive CAS9 along with the transcription coactivators VP64, HSF1, and P65 and sgRNA targeting the promoter of the SGK1 gene. The positively transduced cells were transduced with individual sgRNA targeting the endogenous SGK1 promoter. RT-qPCR analyses revealed that overexpression of endogenous SGK1 (SGK1-sg3 clone) attenuated the expression of ER target genes in comparison with control cells or the cells transfected with a sgRNA (sgRNA-4) that did not show endogenous activation of SGK1 (Figure 5D).

Moreover, we have also studied the transcriptional effects of SGK1 in the presence of estrogen (E2), the PI3K inhibitor BYL719, or both. T47D cells were subjected to SGK1 siRNAs, followed by induction with E2, BYL719, or both. Subsequent analysis of mRNA levels by RT-qPCR revealed a further signifi-cant E2- and BYL719-mediated induction of classical ER target genes, including PGR, GREB1, and MYC in SGK1 silenced cells compared to control cells (Figure 5E). Altogether, these results determine that SGK1 plays a critical role in regulating ER transcriptional output via the posttranslational modification of KMT2D.

DISCUSSION

In this work, we show that SGK1 regulates the chromatin state and ER transcriptional output via the direct phosphorylation of KMT2D, which is the major mammalian histone H3K4 mono- and di-methyltransferase that elicits transcription, and that it is required for enhancer activation. We have discovered that the ER activity triggered by PI3Kα inhibition upregulates SGK1 transcription and protein levels. Elevated SGK1, in turn, directly phosphor-ylates KMT2D, attenuating its function, which leads to a loss of H3K4me1/2 occupancy and a decrease in the recruitment of ER transcriptional regulatory network to shared loci. The net effect of this process is a downregulation of ER transcriptional output (Figure 6).

Figure 6. Proposed Model: The Feedback Repression of ER-Dependent Transcription.

PI3Kα-mediated induction of ER upregulates SGK1 transcription, leading to elevated SGK1 protein levels. Elevated SGK1, in turn, directly phosphorylates KMT2D, attenuating its function, which leads to a global loss of H3K4me1/2 occupancy and at ER sites and a decrease in the binding of ER-FOXA1-PBX1 regulatory network to shared loci, resulting in the downregulation of ER target gene expression.

We have previously shown that KMT2D function and activity is regulated by the downstream effector of PI3K pathway, AKT1 (Toska et al., 2017). The decrease in phosphorylated KMT2D (S1331) upon SGK1 silencing indicates that, in vivo, KMT2D is a substrate for this kinase, contributing to both KMT2D and ER function. We propose that the induced expression of SGK1 by ER serves as a negative-feedback loop to attenuate KMT2D function and, subsequently, ER activity. This is achieved by the ER-SGK1-KMT2D regulatory loop to restore homeostasis of the ER transcriptional output (Figure 6). However, future studies will be necessary to determine the biological consequences of the activation of this loop.

We further speculate that the upregulation of SGKs could be an important mediator of resistance not only to PI3K inhibitors, as we have previously reported (Castel et al., 2016), but also to endocrine therapies, as high levels of active SGK1 would inhibit ER activity, making cells less dependent on the ER response. Based on the positive phase-III clinical trial results (F. Andre et al., 2018, ESMO 2018 Congress, conference), BYL719 may soon become the standard of care for ER+/PIK3CA mutant patients, and comprehensive, correlative studies of SGK1 expression to the dual anti-PI3K and endocrine therapy response may soon be available.

We further hypothesize that the single activity of AKT or SGK enzymes may be sufficient to propagate the effects of PI3K activation on chromatin-based regulation of gene expression via KMT2D phosphorylation. However, under pharmacological stress, such as upon PI3K inhibition where AKT1 is fully inhibited but SGK1 is not, cells can upregulate SGK1 to replace AKT1, making cells less dependent on the ER response. On the other hand, we suggest that the ER-dependent increase of SGK1 transcription observed by PI3K inhibition is transient, in line with previous reports showing that SGK1 expression is rapidly and transiently regulated at the transcriptional and post-transcriptional levels and that both SGK1 mRNA and protein have a short half-life (Di Cristofano, 2017; Webster et al., 1993). For this reason, we propose that the effect of SGK1 on the attenuation of ER-dependent transcription is acute and may only be active for a few hours after estrogen signaling or PI3K inhibition, while the overall ER function is sustained. In support of this, similar regulatory feedbacks are also observed in other signaling pathways, such as the mitogen-activated protein kinase (MAPK) pathway, where transcriptional targets of the pathway, such as DUSP and SPRY, are also negative regulators of the upstream pathway (Jeffrey et al., 2007; Lake et al., 2016).

In kinetic studies, we compared the catalytic efficiency of the two PI3K-regulated kinases, AKT1 and SGK1, that redundantly phosphorylate KMT2D. These assays demonstrate that the catalytic efficiency of AKT1 is significantly higher than that of SGK1 in phosphorylating KMT2D at S1331 and bona fide AKT1/SGK1 targets, such as FOXO1 T24, TSC2 T1462, and TSC2 S1798; prior to in our data, TSC2 S1798 was not reported before as an AKT1 site. This may be due to the nature of activation of these kinases where AKT1—unlike SGK1, which is not recruited to the plasma membrane—is recruited in close proximity to PDK1 at the plasma membrane and is activated within seconds upon growth factor stimuli. Additionally, it could be due to the lack of stability of the SGK1 expression, given its short mRNA half-life. However, because SGK1 is highly regulated at the mRNA and protein levels, and is rapidly upregulated in response to ER, we hypothesize that, despite inferior catalytic efficiency, the local abundance of SGK1 could compensate for the lower efficiency. These results, however, would need to be validated in cellular settings where the strength or duration of AKT1 or SGK1 activation, the expression level of these kinases, possible spatial effects between the kinase and the substrate, and the diverse extracellular stimuli might influence the substrate phosphorylation.

The regulation of KMT2D protein itself by posttranslational modifications had not been elucidated, likely due to the large size of KMT2D (~593 kDa), rendering biochemical characterization of posttranslational modifications challenging. However, our findings uncover a multi-faceted regulatory mechanism of KMT2D protein by distinct PI3K effectors, AKT1 and SGK1. We have yet to uncover any other posttranslational modifications of KMT2D and whether they may cross-talk with the AKT1/SGK1-mediated phosphorylation. In addition, we have previously shown that this phosphorylation event negatively regulates the activity of KMT2D, whereas inhibition of the PI3K pathway enhances the recruitment of KMT2D at ER loci, and the overexpression of catalytically active SGK1 attenuates KMT2D occupancy at ER loci. Future experiments will also need to address the mechanism(s) of how phosphorylation close to the PHD domain of KMT2D alters the function of KMT2D. This will likely involve differential recruitment of possibly uncharacterized cofactors. Of note, a recent study has shown that mutations in the PHD domain of a close paralog of KMT2D, the methyltransferase KMT2C, disrupt the interaction with a previously unknown cofactor, BAP1, reducing the recruitment of KMT2C to gene enhancers (Wang et al., 2018). Similar studies in combination with cryo-electron microscopy studies of full-length KMT2D will provide important structural and functional insights of the regulatory mechanism(s) of this posttranslational modification. It would also be important to explore the role of KMT2D phosphorylation by AKT1 or SGK1 kinases in settings independent of ER-dependent transcription.

In summary, our findings show that the regulation of KMT2D and its role in transcription are attributable to SGK1, revealing an ER-SGK1-KMT2D negative-feedback regulatory loop and directly connecting this kinase with chromatin-based gene expression. Other previously unknown functions of SGK1 in modulating biological mechanisms that are independent or in parallel to AKT now remain to be uncovered.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and request for resources and reagents should be directed and will be fulfilled by the Lead Contact, Eneda Toska (toskae@mskcc.org).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

The T47D (cat# HTB-133), HEK293T (cat# CRL-3216), and MCF7 (cat# HTB-22) cells lines used for this study were purchased from ATCC and used at low passages. JIMT1 were purchased from AddexBio (cat# C0006005) and used at low passages. T47D and JIMT1 cells were maintained in RPMI 1640 with 10% FBS, 1% L-glutamine, and 1% penicillin-streptomycin. MCF7 cells were maintained in DF-12/DMEM Dulbecco medium with 10% FBS, 1% L-glutamine, and 1% penicillin-streptomycin. HEK293T cells were maintained in DMEM Dulbecco medium supplemented with 10% FBS, 1% L-glutamine and 1% penicillin-streptomycin. All cell lines were maintained under normal oxygen conditions (5% CO₂, 37°C). When mentioned, MCF7 or T47D cells were hormonally deprived for 3 days in phenol-red-free DF12/DMEM or RPM1 media respectively that was supplemented with 5% charcoal/dextran-treated FCS. Following hormonal deprivation, the cells were treated with DMSO, E2 (100nm), BYL719 (1 μM), or both for 24 hours.

METHOD DETAILS

Reagents, Transfections, and Lentiviral Generation

Reagents such as BYL719, GDC0941, MK2206, BKM120, GDC0032 were purchased from Selleckchem. SGK1-inh was a gift from M. Nazare and N. Halland, dissolved in DMSO. GDC0077 was a kind gift under an MTA with Genentech, Inc.

Plasmids pLenti7.3-V5-SGK1 (Δ60, S422D) and control vector pLenti7.3luc were cloned in house as described by (Fellmann et al., 2013). Lentiviral doxycycline-inducible mirE-embedded shRNAs were cloned into the LT3GEPIR vector. This all-in-one vector contains the puromycin resistance and the reverse transactivator (rtTA3) under the control of PGK promoter. SGK1 shRNA was chosen experimentally based on five different hairpins validated at Castel et al., 2016.

SGK1#282shRNA: 5^′TGCTGTTGACAGTGAGCGCAGAAGTGTTCTATGCAGTCAATAGTGAAGCCACAGATGTATTGACTGCATA GAACACTTCTTTGCCTACTGCCTCGGA-3^′.

All Constructs were Validated by Sanger Sequencing

To generate lentiviruses, 293T cells were transfected with pCMV-VSVG, pCMV-dR8.2 and the indicated plasmids by lipofectamine 3000 according to the manufacture’s protocol. 72 hr post transfection the viruses were collected, filtered using a 0.45 μm filter and supplemented with 8 μg/μL of polybrene. T47D or MCF7 cells were infected twice and selected with puromycin (2 μg/μL). For the pLenti7.3 vectors, the cells were GFP sorted by Fluorescence Activated Cell Sorting (FACS).

To perform transient transfections, 3 SGK1 siRNAs and control siRNAs was obtained from Ambion. Cells were transfected using a pool of three SGK1 siRNAs and lipofectamine 3000. Transfected cells were harvested 48 hr post transfection.

Immunoblot and Co-immunoprecipitation

Cells were lysed in RIPA buffer supplemented with protease and phosphatase inhibitors and separated using SDS-PAGE gradient gels (4%–12%). The KMT2D probed gels were separated using SDS-PAGE gradient 3%–8%. After transfer to a PVDF membrane for 2 hours at 70 V, the membranes were probed using Vinculin, Actin, SGK1, SGK3, pNDRG1 (T346) and phospho-RXRXX(S/T); all were from Cell Signaling Technology (CST). The rabbit KMT2D was generated by Eurogentec (third bleed) and affinity purified against Ac-GRARLKSTASSI C-NH2 while the rabbit pKMT2D (S1331) was affinity purified against Ac-GRARLKS(PO₃H₂)TASSIC-NH2 peptide.

Cell lysates for co-immunoprecipitation assays were made using NP-40 buffer and incubated overnight with EZview Anti-HA agarose beads. The immunocomplexes were then washed three times with NP40 and immunoblotted with the indicated antibodies. For endogenous co-immunoprecipitation of SGK1 and KMT2D, we employed the KMT2D antibody A300BL1185 (Bethyl) in 5 mg of JIMT1 lysate and SGK1 was recognized using the CST antibody with a secondary conformational specific antibody from Clean Blot (Thermo).

In Vitro Kinase Assay

293T cells were transfected using equimolar amounts (1 μg/ml) of wild-type full-length pCMV-HA-KMT2D or S1331A mutant, or V5-KMT2D (1222–1819) or V5-KMT2D S1331A (1222–1819) using polyethylenimine method. Prior to collection cells were treated with the PI3K inhibitor, BKM120 (800nM). KMT2D was immunoprecipitated using EZview Anti-HA or Anti-V5 agarose beads. The immunoprecipitated substrate was incubated in a reaction with recombinant His-SGK1, His-SGK2, His-SGK3 (MRC-PPU reagents) and ATP (Signalchem) in kinase buffer (25 mM MOPS, pH 7.2, 12.5 mM β-glycerolphosphate, 25mM MgCl_2, 5 mM EGTA, 2mM EDTA, and 0.25 mM DTT) at 30°C for 30 minutes.

Michaelis-Menten Reactions

Recombinant human full-length AKT1, N-terminal GST-tagged, expressed in Sf9 insect cells (SignalChem, A16–10G) or recombinant human full-length SGK1, N-terminal GST-tagged, expressed in Sf9 insect cells (SignalChem, S06–10G) kinases were used for the kinetics reactions. Each of the kinases were incubated with the following peptide substrates: KMT2D (GRGRGRARLKSTA) where S1331 resides, TSC2 (GLRPRGYTI) where T1462 resides, TSC2 (VGQRKRLISSVE) where S1798 resides, or FOXO1 (PLPRPRCCTWPL) peptide where T24 resides. Note that the peptide for FOXO1 also has the S22C mutation to make sure the kinases are only able to phosphorylate the T24 residue. The peptides were custom made by GenScript. The reactions were performed by Reaction Biology as follows: The substrates were incubated in the assay buffer (20 mM HEPES ph 7.5, 10 mM MgCl₂, 1 mM EGTA, 0.02% Brij35, 0.02 mg/ml BSA, 0.1 mM Na₃VO₄, 2mM DTT, 1% DMSO) followed by the addition of the enzyme in the substrate mixer. Hot/cold 33P-ATP was added to initiate the reactions. 6 ATP concentrations were used: 1, 3, 10, 30, 100, 300 μM. The activity was measured with the filter-binding method. The concentrations of the enzymes were shown in the tables in the figures section together with the Km, Vmax, Kcat, and Kcat/km values.

CRISPR-Cas9 Synergistic Activation Mediator (SAM) System Generation

Endogenous SGK1 was transcriptional activated using the CRISPR-Cas9 Synergistic Activation Mediator (SAM) system. First the guide target sequences were cloned into the lenti-sgRNA(MS2)_zeo backbone (addgene #61427) according to the published protocol (Konermann et al., 2015). Briefly, the following paired oligos were synthesized by IDT and were annealed using T4 PNK (NEB) in a thermal cycler with the following conditions: 37°C for 30 minutes; 95°C for 5 minutes; Ramp to 25°C at 5°C/min.

Oligos

• SAM-SGK1-F1: CACCGCGCAGGCCCCGCCCCCGCGG

• SAM SGK1-R1: AAACCCGCGGGGGCGGGGCCTGCGC

• SAM SGK1-F2: CACCGGAGCCCCGGGCGGGGGCGCG

• SAM SGK1-R2: AAACCGCGCCCCCGCCCGGGGCTCC

• SAM SGK1-F3: CACCGGCCCGGGGACGGCCTGGCGC

• SAM SGK1-R3: AAACGCGCCAGGCCGTCCCCGGGCC

• SAM SGK1-F4: CACCGACCGCGAGGCGGCCGGGGCG

• SAM SGK1-R4: AAACCGCCCCGGCCGCCTCGCGGTC

• SAM-SGK1-F5: CACCGGAGGGGCGAGGCGAAGGGCG

• SAM-SGK1-R5 AAACCGCCCTTCGCCTCGCCCCTCC

The lenti-sgRNA (zeo) backbone (addgene #61427) was digested by BsmBI enzyme and ligated with the above annealed products by T7 ligase (NEB) in a thermal cycler with the following conditions: 37°C for 5 min; 20°C for 5 min ;repeat for 15 cycles total.

2 μL of products were transformed in Stbl3 competent cells (Invitrogen) and all the clones were confirmed by sanger sequencing. dCas9-VP64-GFP (addgene 61422), MS2-P65-HSF1 (addgene 61426) and the above cloned lenti-sgRNA vectors were transduced into MCF7 cells. GFP-positive cells were sorted by flow cytometry following by puromycin and zeocin selection. The consequences of the transcriptional activation of SGK1 was determined by western blot.

RNA-Extraction, cDNA-Synthesis, and RT-qPCR

RNA was isolated from MCF7 or T47D cells using the QIAGEN RNeasy kit and cDNA synthesis was made using the Bio-Rad cDNA synthesis kit following the manufacturer’s instructions. cDNA fragments were amplified with the primers listed below and SYBR green mix (Applied Biosystems) using the ViiA™ Real Time PCR system (Applied Biosystems). To obtain relative RNA expression, the data were analyzed by the change-in-threshold (2^−ΔΔCT) method of the specific gene of interest over the housekeeping genes, Actin or GAPDH. The RT-qPCR primers used are:

• GREB1: 5^′-GTGGTAGCCGAGTGGACAAT-3^′; 5^′-ATTTGTTTCCAGCCCTCCTT-3^′

• PGR: 5^′-GGCATGGTCCTTGGAGGT-3^′; 5^′-CCACTGGCTGTGGGAGAG-3^′

• cFOS: 5^′-TGATGACCTGGGCTTCCCAG-3^′; 5^′-CAAAGGGCTCGGTCTTCAGC-3^′

• EGR3: 5^′-GGAGCAAATGAAATGTTGGTG-3^′; 5^′-AGGAAAACCTATGGGGAATG-3^′

• MYC: 5^′-GCTGCTTAGACGCTGGATTT-3^′; 5^′-TAACGTTGAGGGGCATCG-3^′

• SGK1: 5^′-GACAGGACTGTGGACTGGTG-3^′; 5^′-TTTCAGCTGTGTTTCGGCTA-3^′

• SGK2: 5^′-ACCCTTCAACCCAAATGTGA-3^′; 5^′-GGGTGAACTCTGGGTCAAAA-3^′

• SGK3: 5^′-CAGCTGGGCTGACCTTGTA-3^′; 5^′-TGTCAAAGTTTCTGATATCATCTGG-3^′

• LIF: 5^′-TGCCAATGCCCTCTTTATTC-3^′; 5^′-GTCCAGGTTGTTGGGGAAC-3^′

• XBP1: 5^′-GCGCCTCACGCACCTG-3^′; 5^′-GCTGCTACTCTGTTTTTCAGTTTCC-3^′

• WISP2: 5^′-CAGGGGTCGCAGTCCACAAAA-3^′; 5^′-AGGCAGTGAGTTAGAGGAAAGG-3^′

• IGFBP4: 5^′-AACTTCCACCCCAAGCAGT-3^′; 5^′-GGTCCACACACCAGCACTT-3^′

• ACTIN: 5^′-CGTCTTCCCCTCCATCGT-3^′; 5^′-GAAGGTGTGGTGCCAGATTT-3^′

• GAPDH: 5^′-ACAGTCAGCCGCATCTTCTT-3^′; 5^′-ACGACCAAATCCGTTGACTC-3^′

Chromatin Immunoprecipitation (ChIP)

ChIP experiments were performed as previously described. Briefly, cells were crosslinked with 1% paraformaldehyde for 15 minutes followed by quenching with glycine for 5 minutes. After lysis with SDS buffer for 10 minutes, the cells were sonicated, and the sheared chromatin was incubated with Protein G Dynabeads or the following antibodies: FOXA1 (ab5089) Abcam, ER (sc-543) Santa Cruz, PBX1 (H00005087-MO1), Abnova, and KMT2D (HPA035977), Sigma. The immunocomplexes were washed twice with low salt wash buffer, high salt wash buffer, LiCl wash buffer, and 1x TE buffer. After elution, the samples were crosslinked at 65°C, and PCR-quick purified. The ChIP-qPCR primers are as follows:

• GREB1: 5^′-GAAGGGCAGAGCTGATAACG-3^′; 5^′-GACCCAGTTGCCACACTTTT-3^′

• PGR: 5^′-AGGGAGGAGAAAGTGGGTGT-3^′; 5^′-GGAGAACTCCCCGAGTTAGG-3^′

• SGK1: 5^′- GGGAGGGAGAGGTCAGGAAT-3^′; 5^′-TCGCTTGTTACCTCCTCACG-3^′

• SGK2: 5^′-GCCTTCCCTGTTCTCCAG-3^′; 5^′-TACTGATTACGGGTGACAAGTCA-3^′

• SGK3: 5^′−5^′-CAGAACTTCCCCTTCTTTAACTGC-3^′; 5^′-GCTGACCTACCAATTCAATGTTTC-3^′

• cFOS: 5^′-AATGCAAACAGGACCAAAGG-3^′; 5^′-TGGCAGTGTCAGGACAGAAG-3^′

• EGR3: 5^′-ACCTCCAAGAGGGAGAGGAG-3 3^′; 5^′-CTGTCCAGCCGGAGTTAGAG-3^′

• MYC: 5^′-GTCAGCCAATCTTCGCACTT-3^′; 5^′-TGCCAGAGGAAGCTACTGGT-3^′

• WISP2: 5^′-AGGGCTTCTCTCCCTGGAAC-3^′; 5^′-GCCCAGAGAAGGCAAGACAC-3^′

• ACTIN: 5^′-TGTTCCAGGCTCTGTTCCTC-3^′; 5^′-AGAAAAGAACGCAGGCAGAA-3^′

ChIP-Seq and ChIP-Seq Analysis

ChIP-seq of H3K4me1 and H3K4me2 was performed as previously described (Chen et al., 2015). Briefly, cell samples were cross-linked with 1% formaldehyde for 10 min, and quenched by glycine to a 125nM final concentration. The fixed cells were lysed in SDS buffer and the chromatin was sheared by Covaris sonication. The sheared chromatin was incubated with the indicated antibodies and protein G-Dynabeads. The samples underwent decrosslinking, RNase and proteinase K treatment. DNA fragments were eluted using AMP Pure beads, library was prepared and samples were subjected to high-throughput sequencing using HiSeq 2000 platform (Illumina). Deeptools (Ramírez et al., 2014), along with size factor scaled BigWig tracks, was used to generate heatmaps of peak profile. H3K4me1 and H3K4me2 FASTQ reads were aligned to human reference genome (hg19) using bowtie2 (Langmead and Salzberg, 2012; Quinlan and Hall, 2010) after trimming adaptors and low quality sequences with trim-galore tool (www.bioinformatics.babraham.ac.uk/projects/trim_galore). Duplicated reads are marked with Picard tools. (http://picard.sourceforge.net). Alignment BAMs are sorted and indexed via samtools (Li et al., 2009). Peaks are called with MACS2 against matched input. Genome coverage is calculated using bedtools (Quinlan and Hall, 2010) and normalized to 10 million reads. Deeptools (Ramírez et al., 2014), along with size factor scaled BigWig tracks, was used to generate heatmaps of peak profiles.

Transposase-Accessible Chromatin using Sequencing (ATAC-seq) and Analysis

ATAC-seq assays were performed as described by (Buenrostro et al., 2013) with the exception that 0.2% IGEBAL lysis buffer was used. Starting from fastq files containing ATAC-seq paired-end reads, sequencing adaptors were removed using Trimmomatic (Bolger et al., 2014). Trimmed reads were mapped to the hg19 human genome using Bowtie2 allowing at most 1 seed mismatch and keeping only uniquely aligned reads. Duplicates were removed using Picard (http://picard.sourceforge.net). For peak-calling the read start sites were adjusted (reads aligning to the ± strand were offset by +4bp/−5bp, respectively) to represent the center of the transposon binding-event, as described in (Buenrostro et al., 2013). Peak calling was performed using MACS2 with a permissive p value threshold (-p 1e-2). Only peaks with an IDR of 0.05 or less were kept for downstream analyses. To link site accessibility to regulation of gene expression, we associated each peak to its nearest gene in the human genome using ChIPpeakAnno package (Zhu et al., 2010). Differentially accessible peaks from this atlas were identified with edgeR (Robinson et al., 2010) by counting all read ends overlapping peaks in each condition. edgeR was also used to compute the robost normalization factor. edgeR was run with default settings, a fold-change threshold of 2, and FDR < 0.01. Deeptools (Ramírez et al., 2014), along with size factor scaled BigWig tracks, was used to generate heatmaps of peak profile.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analyses and number of replicates are demonstrated at the appropriate figure legends and the method details section. Briefly, p values were calculated using Student’s t test. Error bars are ± STD. Regarding the p values of the boxplots, they represent mean signal across ChIP-seq or ATAC-seq peaks and were measured by the Mann-Whitney test.

Data and Software Availability The ATAC-seq and ChIP-seq data are deposited at GEO database (GEO: GSE119522)

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
H3	Cell Signaling	Cat# 9715; RRID: AB_331563
Actin	Cell Signaling	Cat# 4970L; RRID: AB_2223172
SGK1	Cell Signaling	Cat# 12103; RRID: AB_2687476
SGK3	Cell Signaling	Cat# 8156; RRID: AB_10949507
pNDRGI (T346)	Cell Signaling	Cat# 5482; RRID: AB_10693451
Phospho-RXRXX(S/T)	Cell Signaling	Cat# 10001; RRID: AB_10950819
Phospho-KMT2D(S1331)	(Toska et al., 2017)	N/A
Vinculin	Cell Signaling	Cat# 1390; RRID: AB_2728768
Phospho-AKT (S473)	Cell Signaling	Cat# 4060; RRID: AB_2315049
H3K4me1	Abcam	Cat# ab8895; Lot#: Gr3236714–1; RRID: AB_306847
H3K4me2	Abcam	Cat# ab32356; Lot#: Gr253788–29; RRID: AB_732924
V5	Cell Signaling	Cat# 13202; RRID: AB_2687461
HA	Roche	Clone 3F10 Cat# 11867423001; RRID: AB_390918
FOXA1	Abcam	Cat# ab5089; RRID: AB_304744
PBX1	Abnova	Cat# H00005087-MO1; RRID: AB_606725
KMT2D (ChIP)	Sigma	Cat# HPA035977; RRID: AB_10670673
KMT2D (IP)	Bethyl	Cat# A300BL1185
ER	Santa Cruz	Cat# sc-543; RRID: AB_631471
Bacterial and Virus Strains
One Shot® Stbl3 Chemically Competent E. coli	Thomas Scientific	C7373–03
Chemicals, Peptides, and Recombinant Proteins
SGK1-inh	(Castel et al., 2016)	Gift
BYL719	Selleckchem	S2814
MK-2206	Selleckchem	S1078
GDC-0941	Selleckchem	S1065
GDC-0077	Genentech, inc	MTA
GDC-0032	Selleckchem	S7103
BKM120	Selleckchem	S2247
Recombinant human full-length AKT1	SignalChem	A16–10G
Recombinant human full-length SGK1	SignalChem	S06–10G
KMT2D GRGRGRARLKSTA peptide	Genscript	N/A
TSC2 GLRPRGYTI peptide	Genscript	N/A
TSC2 VGQRKRLISSVE peptide	Genscript	N/A
FOXO1 PLPRPRCCTWPL peptide	Genscript	N/A
Recombinant His-SGK1	MRC Protein Phosphorylation and Ubiquitination Unit (PPU)	N/A
Recombinant His-SGK2	MRC Protein Phosphorylation and Ubiquitination Unit (PPU)	N/A
Recombinant His-SGK3	MRC Protein Phosphorylation and Ubiquitination Unit (PPU)	N/A
ATP	SignalChem	A50-09-200
Critical Commercial Assays
QIAGEN RNeasy kit	QIAGEN	74104
Iscript cDNA synthesis	BIO-RAD	1708891
Lipofectamine 3000	Thermo Fisher	L3000–015
Michaelis-Menten Assays	Reaction Biology	N/A
Deposited Data
Raw sequencing data for ATAC-seq and ChIP-seq data	GEO database	GSE119522
Experimental Models: Cell Lines
Human T47D breast cancer cells	ATCC	HTB-133
Human MCF7 breast cancer cells	ATCC	HTB-22
Human HEK293T cells	ATCC	CRL-3216
Human JIMT1 breast cancer cells	AddexBio	C0006005
Oligonucleotides
Primers for RT-qPCR (see method below)	This paper	N/A
Primers for ChlP-qPCR (see method below)	This paper	N/A
siRNA targeting SGK1	Ambion Select	N/A
shRNA sequence targeting SGK1 (see method below)	(Castel et al.,2016)	N/A
Primers for SAM assays (see method below)	This paper	N/A
Recombinant DNA
pLenti7.3-V5-SGK1 (Δ60, S422D)	This paper and (Castel et al., 2016)	N/A
pLenti7.3luc	This paper and (Castel et al., 2016)	N/A
LT3GEPIR	MSKCC RNAi Core	N/A
pCMV-VSVG	(Stewart et al., 2003)	Addgene 8454
pCMV-dR8.2	(Stewart et al., 2003)	Addgene 8455
lenti-sgRNA(MS2)_zeo	(Konermann et al., 2015)	Addgene 61427
lenti-sgRNA (zeo)	(Konermann et al., 2015)	Addgene 61427
dCas9-VP64-GFP	(Konermann et al., 2015)	Addgene 61422
MS2-P65-HSF1	(Konermann et al., 2015)	Addgene 61426
pCMV-HA-KMT2D	(Zhang et al., 2015)	N/A
Software and Algorithms
Trimmomatic	(Bolger et al., 2014)	http://www.usadellab.org/cms/index.php?page=trimmomatic
Bowtie2	(Langmead and Salzberg, 2012)	http://bowtie-bio.sourceforge.net/bowtie2/index.shtml
ChlPpeakAnno package	(Zhu et al., 2010)	https://www.bioconductor.org/packages/devel/bioc/html/ChIPpeakAnno.html
edgeR	(Robinson et al., 2010)	https://bioconductor.org/packages/release/bioc/html/edgeR.html
deepTools	(Ramírez et al., 2014)	http://deeptools.ie-freiburg.mpg.de
Trim Galore	N/A	http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/
SAMtools	N/A	http://samtools.sourceforge.net/
Bedtools	(Quinlan and Hall, 2010)	http://code.google.com/p/bedtools

Highlights.

• Estrogen receptor (ER) activates SGK1 upon PI3Kα pathway inhibition

• SGK1 directly phosphorylates KMT2D, resulting in the attenuation of its function

• SGK1 induces a repressive chromatin state and a loss of H3K4me1/2 binding at ER loci

• SGK1 regulates ER-dependent gene expression