PI3K pathway regulates ER-dependent transcription in breast
cancer through the epigenetic regulator KMT2D
Eneda Toska
1
, Hatice U. Osmanbeyoglu
2,*
, Pau Castel
1,3,*
, Carmen Chan
1
, Ronald C.
Hendrickson
4
, Moshe Elkabets
1,5
, Maura N. Dickler
6
, Maurizio Scaltriti
1,7
, Christina S.
Leslie
2
, Scott A. Armstrong
8,9
, and José Baselga
1,6
1
Human Oncology and Pathogenesis Program, Memorial Sloan Kettering Cancer Center, 1275
York Avenue, Box 20, New York, NY 10065, USA
2
Computational Biology Program, Memorial Sloan Kettering Cancer Center, 1275 York Avenue,
Box 460, New York, NY 10065, USA
3
Helen Diller Family Comprehensive Cancer Center, University of California–San Francisco, 1450
3rd Street, San Francisco, CA 94158, USA
4
Microchemistry and Proteomics Core Laboratory, Memorial Sloan Kettering Cancer Center, New
York, NY 10065, USA
5
The Shraga Segal Department of Microbiology, Immunology and Genetics, Faculty of Health
Sciences, Ben-Gurion University of the Negev, Beer-Sheva 84105, Israel
6
Department of Medicine, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA
7
Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA
8
Cancer Biology and Genetics Program, Memorial Sloan Kettering Cancer Center, New York, NY
10065, USA
9
Department of Pediatric Oncology, Dana-Farber Cancer Institute, 450 Brookline Avenue, Boston,
MA 02215, USA
Abstract
Activating mutations in
PIK3CA
, the gene encoding phosphoinositide-(3)-kinase α (PI3Kα), are
frequently found in estrogen receptor (ER)–positive breast cancer. PI3Kα inhibitors, now in late-
stage clinical development, elicit a robust compensatory increase in ER-dependent transcription
that limits therapeutic efficacy. We investigated the chromatin-based mechanisms leading to the
activation of ER upon PI3Kα inhibition. We found that PI3Kα inhibition mediates an open
chromatin state at the ER target loci in breast cancer models and clinical samples. KMT2D, a
histone H3 lysine 4 methyltransferase, is required for FOXA1, PBX1, and ER recruitment and
activation. AKT binds and phosphorylates KMT2D, attenuating methyltransferase activity and ER
function, whereas PI3Kα inhibition enhances KMT2D activity. These findings uncover a
Correspondence to: Scott A. Armstrong; José Baselga.
*
These authors contributed equally to this work
Supplementary Materials: www.sciencemag.org/content/355/6331/1324/suppl/DC1 Materials and Methods Figs. S1 to S8 Tables S1 to
S3 References (34–44)
HHS Public Access
Author manuscript
Science
. Author manuscript; available in PMC 2018 March 24.
Published in final edited form as:
Science
. 2017 March 24; 355(6331): 1324–1330. doi:10.1126/science.aah6893.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
mechanism that controls the activation of ER by the posttranslational modification of epigenetic
regulators, providing a rationale for epigenetic therapy in ER-positive breast cancer.
Activating mutations in the
PIK3CA
gene occur at high frequency in estrogen receptor
(ER)–positive breast cancer, indicating that the PI3K signaling pathway plays an important
role in tumorigenesis (1, 2). Consistent with this, PI3Kα inhibitors have shown antitumor
activity in patients with
PIK3CA
-mutant, ER-positive breast cancer (3, 4). However, a
number of mechanisms of resistance have emerged that could potentially limit their efficacy
(5, 6). We have observed an adaptive tumor response to PI3Kα inhibitors that is
characterized by an increase in ER-dependent transcription, which mediates therapeutic
resistance and can be reversed by the addition of anti-ER therapies (7). These findings have
led to pilot clinical studies of combination therapies consisting of PI3Kα inhibitors and anti-
ER agents that have produced high response rates and prolonged clinical benefit, even in
tumors refractory to ER therapies (8–10). These findings have triggered the launch of two
phase III registration clinical studies testing PI3Kα inhibitors in combination with the anti-
ER agent fulvestrant (11, 12). The mechanisms by which tumor cells exposed to PI3Kα
inhibitors activate ER signaling are not well understood. In luminal breast cancers, ER is a
master regulator of transcription, activating expression of genes that favor tumor growth and
survival (13). Binding of ER to estrogen-responsive elements (EREs) leads to recruitment of
additional cooperating transcription factors (TFs) that are essential for the function of this
intricate network (13–15).
To investigate the cis-regulatory elements and cofactors involved in ER-dependent
transcription induced by treatment with a PI3Kα inhibitor (namely, BYL719) (16), we
performed ER chromatin immunoprecipitation followed by high-throughput sequencing
(ChIP-seq) in the breast cancer cell line T47D. PI3Kα inhibition induced differential ER
binding, as demonstrated by gained or lost ER-binding events (Fig. 1A). As expected, the
ERE motif was highly enriched in the gained ER-binding sites upon treatment with
BYL719. The forkhead box A1 (FOXA1) motif and a homeobox class motif were also
enriched, suggesting the presence of these TFs at the ER-binding sites. FOXA1 is required
for ER-chromatin interactions (14). PBX1 (pre–B cell leukemia homeobox 1), a FOXA1
collaborator and a member of the homeobox family that is required for optimal ER activity
(15), was most likely associated with the homeobox class motif found in our analysis (Fig.
1A and fig. S1A).
FOXA1 ChIP-seq analysis in T47D cells revealed a differential binding profile of FOXA1
upon PI3Kα inhibition. A nuclear receptor and a homeobox class motif, suggestive of ER
and PBX1, respectively, were among the most enriched motifs within the enhanced FOXA1-
binding regions (Fig. 1B and fig. S1B). An increased co-occupancy of ER and FOXA1 was
also observed upon BYL719 treatment at specific target loci (Fig. 1C and fig. S1C). Hence,
we hypothesized that (i) PI3Kα inhibition enhances the binding of the FOXA1-PBX1-ER
regulatory network to chromatin and (ii) FOXA1 or PBX1 may be involved in ER activation
in this setting. To assess the role of FOXA1 and PBX1 in the activation of ER at the
chromatin level, we performed ChIP-quantitative polymerase chain reaction (qPCR) in
T47D and MCF7 breast cancer cells to examine the binding of these TFs to the enhancers or
Toska et al.
Page 2
Science
. Author manuscript; available in PMC 2018 March 24.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
promoters of canonical ER target genes. The binding of ER, FOXA1, and PBX1 was
substantially enhanced upon PI3Kα inhibition (fig. S1, D and E). FOXA1 and PBX1 were
required for ER function in the context of PI3Kα therapeutic inhibition, as evidenced by the
loss of ER binding and activity upon FOXA1 or PBX1 knockdown (Fig. 1, D and E, and
figs. S2 and S3). We also conducted gene set enrichment analysis using our previously
published transcriptomic data set (7) in T47D cells, MCF7 cells, and MCF7 xenografts to
study the enrichment of nuclear receptors, FOXA1, and PBX1 upon PI3Kα inhibition. As
expected, we observed strong enrichment of ER, FOXA1, and PBX1 in response to PI3Kα
blockage (tables S1 to S3). PI3Kα inhibition also resulted in enrichment for several other
nuclear receptors, such as PR (progesterone receptor), GR (glucocorticoid receptor), AR
(androgen receptor), VDR (vitamin D receptor), and RAR (retinoic acid receptor), but at
lower levels compared with those for ER (tables S1 to S3). These results suggest that ER is
the nuclear receptor affected the most by PI3Kα inhibition in breast cancer.
We next explored the role of FOXA1 and PBX1 in enhancing ER function in response to
PI3Kα inhibitors. FOXA1 or PBX1 silencing by either constitutive or inducible short
hairpin RNAs (shRNAs) was sufficient to decrease the viability of cells treated with
BYL719 (fig. S4, A to F). To investigate the in vivo activity of BYL719, we engineered
MCF7 human breast cancer cells to express inducible shRNA against FOXA1 and then
injected the cells into mice. Although FOXA1 silencing alone had no discernible effects on
the growth of tumor xenografts, it markedly increased the antitumor activity of BYL719
(Fig. 1, F and G). Similar results were observed when PBX1 was knocked down in MCF7
xenografts (fig. S4, G and H). In these tumors, we also confirmed the reduction of ER
occupancy when either FOXA1 or PBX1 was silenced (Fig. 1H and fig. S4I). Thus, FOXA1
or PBX1 silencing sensitizes cells to PI3Kα inhibition through suppression of ER activity.
To study the impact of PI3Kα inhibition on the global chromatin structure of the breast
cancer epigenome, we used ATAC-seq (assay for transposase-accessible chromatin with
sequencing), which enables the study of chromatin openness and the interplay between TFs
and accessible chromatin regions. Genome-wide ATAC-seq in T47D cells revealed that
PI3Kα inhibition induces substantial changes in chromatin state, as demonstrated by an
increase in the gained and lost accessible chromatin sites (Fig. 2A). Analysis of the open
chromatin patterns (Fig. 2B), a feature of active transcription, revealed responsive elements
enriched for ER, FOXA1, and PBX1 (Fig. 2C), consistent with our previous findings.
ATAC-seq signals peaked at regions where we had detected ER- and FOXA1-binding events
by ChIP-seq—namely, ER canonical target genes (fig. S5, A and B).
Taking advantage of the feasibility of performing ATAC-seq with a limited number of cells
(17), we next explored the epigenetic landscape of breast tumor samples from patients who
had participated in a clinical trial with the PI3Kα inhibitor BYL719 (4). We studied the
chromatin accessibility of two paired tumor biopsies that had been obtained before BYL719
therapy and while the patients were on therapy. We found that drug treatment induced an
increase in chromatin accessibility in the regions previously identified to be responsive to
PI3Kα inhibition (Fig. 2D and fig. S5C). Notably, the ATAC-seq signals detected in the
patient samples peaked at the same genomic regions that had shown enhanced binding of ER
and FOXA1 and enhanced chromatin accessibility in our cell line model. Moreover, similar
Toska et al.
Page 3
Science
. Author manuscript; available in PMC 2018 March 24.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
to our in vitro data, responsive elements for ER, FOXA1, and PBX1 were present in the
open chromatin regions after in silico motif analysis, highlighting the specificity of our assay
in the patient samples (Fig. 2E). We also performed RNA sequencing on 15 pre-and on-
treatment tumor biopsies and found a substantial enrichment of ER-associated signatures
induced by BYL719 treatment (fig. S5, D and E). Together, these data indicate that PI3Kα
inhibition results in remodeling of the chromatin landscape to enhance ER-dependent
transcription in breast cancer cell lines and patient tumor samples.
We next aimed to identify the epigenetic regulators that orchestrate the chromatin
remodeling induced by PI3Kα inhibition. It has been previously shown that active histone
modifications—mono- and dimethylated histone H3 lysine 4 (H3K4me1/2)—are associated
with FOXA1 and PBX1 binding at their cis-regulatory elements (15, 18, 19). Additionally,
KMT2D (also known as MLL2 or MLL4), a member of the COMPASS (complex of
proteins associated with Set1)–like family, is a major H3K4me1/2 methyltransferase that
facilitates gene expression (20–22) and has the ability to interact with ER through LXXLL
motifs (L, leucine; X, any amino acid) (23). To explore whether KMT2D participates in the
activation of the ER regulatory network after PI3Kα inhibition, we knocked down KMT2D
and examined the recruitment of ER, FOXA1, and PBX1 and the expression of ER target
genes. KMT2D knockdown in T47D and MCF7 cells resulted in a marked loss of binding of
these TFs at ER-FOXA1-PBX1 shared target genes and a decrease in ER target gene
expression upon BYL719 treatment (Fig. 3, A and B, and fig. S6, A and B). KMT2D
suppression also led to a considerable decrease in H3K4me1/2 occupancy at representative
ER canonical genes, consistent with the hypothesis that KMT2D catalyzes these histone
modifications in breast cancer cells (fig. S6C). Although overexpression of wild-type
KMT2D in cells depleted of endogenous KMT2D could restore the binding of these TFs,
overexpression of KMT2D mutants deficient in methyltransferase activity (24) failed to do
so (fig. S6D). Similarly, overexpression of the H3K4me1/2 demethylase KDM1 (25), which
antagonizes KMT2D enzymatic action, abrogated the occupancy of ER, FOXA1, and PBX1
and the expression of ER-dependent genes upon PI3Kα inhibition (fig. S6, E to G).
Collectively, these findings suggest that KMT2D is required for the activation of ER-
dependent transcription upon PI3Kα inhibition and that it does this by catalyzing
H3K4me1/2 modifications that support FOXA1, PBX1, and ER recruitment to specific
chromatin sites.
Given our observation that KMT2D silencing suppresses ER activation, we hypothesized
that loss of function of KMT2D might augment the therapeutic activity of BYL719 in ER-
positive cells. Indeed, we found that KMT2D knockdown decreased cell viability in ER-
positive T47D and MCF7 cell lines and had a greater effect when combined with BYL719
(fig. S7, A to C). In mice bearing MCF7 xenografts with doxycycline-inducible shRNA
against KMT2D, silencing KMT2D resulted in a marked antitumor effect that was enhanced
in the presence of BYL719 (Fig. 3, C and D). By performing ChIP-qPCR experiments in the
in vivo tumor samples, we confirmed that binding of ER, FOXA1, and PBX1 was lost when
KMT2D was knocked down (Fig. 3E). These findings suggest that targeting KMT2D in ER-
positive breast cancers could be a potential therapeutic strategy to enhance the sensitivity of
PI3Kα inhibitors.
Toska et al.
Page 4
Science
. Author manuscript; available in PMC 2018 March 24.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Our finding that KMT2D is a critical determinant of ER function upon PI3Kα targeting
prompted us to study how PI3K inhibition affects the binding dynamics of KMT2D at the
chromatin level. We found that KMT2D recruitment at gene regulatory regions peaked early
in BYL719 treatment at ER target genes but not at control genes—namely, ER-FOXA1-
PBX1 targets whose expression did not change upon BYL719 treatment (Fig. 3F and fig. S7,
D and E). The abundance of H3K4me1/2 occupying these regions was also elevated at the
same time points, as were total H3K4 methyltransferase activity and H3K4me1/2 protein
levels (Fig. 3, F to H, and fig. S7D). In addition, KMT2D ChIP-seq assay in T47D cells
revealed that KMT2D binding peaked at several regions that showed differential binding of
ER and FOXA1 upon treatment with the PI3Kα inhibitor (fig. S7F). The tight temporal
correlation between PI3Kα inhibition and changes in KMT2D recruitment, H3K4
methyltransferase activity, and abundance of H3K4me1/2 suggests a direct regulation of
KMT2D activity by the PI3K pathway.
Oncogenic PI3K signaling regulates several downstream effectors through kinases that can
phosphorylate these effectors. Among these kinases, the serine/threonine (S/T) kinase AKT
phosphorylates a large number of substrates by recognizing the consensus motif
RXRXX(S/T) (R, arginine) (26). We observed that the KMT2D protein sequence contains
two RXRXX(S/T) consensus sites that are evolutionarily conserved (Fig. 4A). Although the
consensus site at S1762 could not be detected by mass spectrometry because of the
surrounding lysine-rich region, which renders it sensitive to trypsin digestion, we found that
S1331 was phosphorylated (fig. S8, A and B). We therefore focused on the S1331 site for
the remaining experiments. Coimmunoprecipitation assay using ectopically expressed HA
(hemagglutinin epitope)–tagged KMT2D and V5-tagged AKT1 in 293T cells revealed that
KMT2D physically interacts with AKT1. We further confirmed the interaction between
endogenous KMT2D and AKT by coimmunoprecipitation in T47D cells (Fig. 4, B and C).
We next identified the region of KMT2D that interacts with AKT1. Given the large size of
KMT2D (∼593 kDa), we performed coimmunoprecipitation assays using five non-
overlapping fragments of KMT2D (Fig. 4D) and found that AKT1 binds the 1222–1819
amino acid region of KMT2D, where the S1331 phosphorylation site resides (Fig. 4E). In
vitro kinase assays confirmed the ability of AKT to directly phosphorylate S1331, as
detected by the phospho-RXRXX(S/T) and phospho-KMT2D (S1331) antibodies (Fig. 4F
and fig. S8C). Endogenous KMT2D was found to be more phosphorylated at S1331 in
isogenic MCF10A cells that harbor the activating H1047R mutation of
PIK
3
CA,
commonly
found in breast cancer, as compared with those harboring wild-type
PIK3CA
(fig. S8D).
Treatment of ER-positive cell lines with either PI3K or AKT inhibitors reduced S1331
phosphorylation (fig. S8, E and F). Next, we sought to determine how phosphorylation of
KMT2D by AKT affects KMT2D activity. Mutation of KMT2D S1331 to the
nonphosphorylatable amino acid alanine (S1331A) increased H3K4 methyltransferase
activity and the levels of H3K4me1/2. Conversely, mutation of this site to the
phosphomimetic amino acid aspartic acid (S1331D) decreased the levels of KMT2D activity
(Fig. 4, G and H), suggesting that this phosphorylation event suppresses KMT2D
methyltransferase activity, leading to a decrease in ER function. Consistent with this
hypothesis, we observed an increase in ER, FOXA1, PBX1, and H3K4me1/2 binding at ER
loci in S1331A KMT2D–expressing cells (Fig. 4I and fig. S8G).
Toska et al.
Page 5
Science
. Author manuscript; available in PMC 2018 March 24.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript