Coordinate regulation of ELF5 and EHF at the chr11p13 CF modifier region

Abstract E74‐like factor 5 (ELF5) and ETS‐homologous factor (EHF) are epithelial selective ETS family transcription factors (TFs) encoded by genes at chr11p13, a region associated with cystic fibrosis (CF) lung disease severity. EHF controls many key processes in lung epithelial function so its regulatory mechanisms are important. Using CRISPR/Cas9 technology, we removed three key cis‐regulatory elements (CREs) from the chr11p13 region and also activated multiple open chromatin sites with CRISPRa in airway epithelial cells. Deletion of the CREs caused subtle changes in chromatin architecture and site‐specific increases in EHF and ELF5. CRISPRa had most effect on ELF5 transcription. ELF5 levels are low in airway cells but higher in LNCaP (prostate) and T47D (breast) cancer cells. ATAC‐seq in these lines revealed novel peaks of open chromatin at the 5’ end of chr11p13 associated with an expressed ELF5 gene. Furthermore, 4C‐seq assays identified direct interactions between the active ELF5 promoter and sites within the EHF locus, suggesting coordinate regulation between these TFs. ChIP‐seq for ELF5 in T47D cells revealed ELF5 occupancy within EHF introns 1 and 6, and siRNA‐mediated depletion of ELF5 enhanced EHF expression. These results define a new role for ELF5 in lung epithelial biology.

epithelial differentiation in the mouse. 16 It has not been well studied in the human lung, although it is a candidate gene for conferring susceptibility to asthma. 17 Furthermore, ELF5 was predicted to regulate EHF expression, since ChIP-seq with an antibody specific for ELF5 showed its occupancy at the EHF promoter and first intron in the breast cancer cell line T47D. 18 Although to date, the highest p-value SNPs identified in the replicated GWAS do not correlate with the transcription of genes at 11p13, this may reflect a paucity of data on gene expression at the single-cell level. However, revealing the regulatory interactions of the EHF and ELF5 TFs is critical to understanding their role in lung epithelial biology. Here we investigate the cis-regulatory elements (CREs) at chr11p13 that may control the expression of these two epithelial selective ETS transcription factors. Using CRISPR/Cas9 technology, three key CREs were deleted from the region in airway epithelial cells, and subsequent changes in gene expression and chromatin architecture were examined. We also took an unbiased screening approach to identify additional CREs at chr11p13 by activating multiple DNase I hypersensitive sites (DHS) across the region using CRISPRa, 19 followed by measuring nearby gene expression. In parallel, we pursued the predicted role of ELF5 in regulating EHF by investigating open chromatin and 3D architecture in cell types with different abundance of ELF5.
These genomic context studies revealed a critical role for ELF5 in coordinating the complex regulatory environment at chr11p13.
Our results are relevant to the involvement of both ELF5 and EHF in lung disease, since both genes are expressed in human bronchial epithelium.

| Luciferase-based reporter assays
Luciferase promoter:enhancer constructs were generated as described previously. 25 Constructs were transfected with Lipofectamine 2000 (Life Technologies) into A549 c9 and T47D cells. A modified pRL Renilla luciferase vector (Promega) was used as a transfection control. Cells were assayed for Renilla and firefly luciferase activity after 48 hours using the Dual-Luciferase Reporter Assay System (Promega). 26

| CRISPR guide design, CRISPR/Cas9 transfection and screening
Two pairs of gRNAs flanking EHF intron 6, 11.2521 and 11.2516 were identified using Benchling 27 (Table S1). Both gRNAs flanking each region were sequentially cloned into a single pBlueScript (pBS) with a modified multiple cloning site. A549 c9 cells were seeded onto 6-well plates and transfected after 24 hours with 0.1 pmol of pMJ920 (wild-type Cas9 plasmid tagged with GFP) (Addgene, plasmid #42234) and 0.2 pmol of pBS containing the 5' and 3' gRNAs using Lipofectamine 2000 (Life Technologies); 48 hours after transfection, cells were prepared for fluorescence-activated cell sorting of GFP-positive cells and single cells were plated onto 96-well plate.
Clones were expanded and screened for homozygous deletion of each site as described previously using primers shown in Table S2. 28

| Reverse transcription quantitative polymerase chain reaction (RT-qPCR)
RNA was extracted from cells using TRIzol (In Vitrogen) according to manufacturer's instructions. Reverse transcription used TaqMan Reverse Transcription Reagents (Life Technologies) by the standard protocol. qPCR assays were performed using SYBR Green reagents (Life Technologies) and primers listed in Table S3.

| Circular chromosome conformation capture and deep sequencing (4C-seq)
4C-seq was performed as previously described 28 in A549 c9 and its derivative CRE deletion clones, Calu3, LNCaP and T47D cells. The primers used to generate libraries for each viewpoint are listed in Table S5. Two independent 4C libraries were generated for each viewpoint for each cell type. The sequencing data were processed using the 4Cseqpipe pipeline. 31 All 4C-seq domainograms were generated using the default parameters of the pipeline.

| CRISPRa/VPR-mediated activation experiments
Single guide RNAs (sgRNAs) targeting ELF5 and EHF promoters were designed within 1 kb 5' to the transcriptional start sites (TSS). Two to four sgRNAs were designed to target DHS cores at chr11p13 using Benchling 27 (Table S6) and were chosen based on the highest on-target and lowest off-target scores. sgRNAs were cloned into the pSPgRNA plasmid (Addgene #47108) and nucleofected with the SP-dCas9-VPR plasmid (Addgene #63798) into either 16HBE14o-or A549 cells with Lonza 4D-Nucleofector kits (P3 Primary Cell V4XP-3032 for 16HBE14o-and SE Cell Line V4XC-1032 for A549). For each experiment, pSPgRNA alone was nucleofected with SP-dCas9-VPR and used as a negative control. Using protocols described above, RNA was extracted 48 hours post-nucleofection and reverse transcription quantitative polymerase chain reaction (RT-qPCR) assays were performed. Data were normalized individually to the empty vector control.

| Omni Assay for transposase accessible chromatin and deep sequencing
Omni-ATAC-seq was performed on 50,000 Calu3, LNCaP and T47D cells as described previously 32 with minor modifications. ATAC-seq libraries were purified with Agencourt AMPure XP magnetic beads (Beckman Coulter) with a sample to bead ratio of 1:1.2 and eluted in Buffer EB (Qiagen). Data were processed by the ENCODE-DCC/ atac-seq-pipeline.

| Statistical analysis
Error bars denote standard deviation (SD) or standard error of the mean (SEM) as noted in figure legends, which also define the statistical analysis used. This was either two-way ANOVA plus multiple comparisons tests or Student's unpaired t tests. All statistical analysis was performed in Prism software (GraphPad).

| Deletion of a CRE in EHF intron 6 enhances EHF expression while impairing recruitment of a strong enhancer to the gene promoter
The EHF locus coincides with an extended track of the H3K27ac active histone mark, indicative of a stretch enhancer in airway epithelial cells. 25,33 EHF intron 6 lies at the 3' end of this feature and coincides with a region of high occupancy by multiple transcription factors (ENCODE data [34][35][36]. To determine the role of this element in the genomic context within airway epithelial cells, CRISPR/Cas9 was used to generate A549 c9 clones lacking an ~3.1 kb fragment coinciding with the extent of TF occupancy. Three homozygous deletion clones were evaluated and compared both to non-targeted clones from the same experiment and wild-type (WT) A549 c9 cells. Models of the 11p13 region (left) and the impact of removal of EHF intron 6 (right) are shown in Figure 1A. First, expression of the ELF5, EHF and APIP genes was assayed by RT-qPCR ( Figure 1B). Removal of the EHF intron 6 element did not change ELF5 or APIP transcript levels; however, it was associated with an increase in EHF (~8-fold, P < .0001) expression.
These results suggest that either transcriptional repressors of EHF are recruited to the intron 6 element or that loss of this site alters chromatin architecture at the locus, thus facilitating occupancy of activating factors at the EHF promoter or its enhancers. To investigate the latter hypothesis, we first assayed CTCF occupancy at specific sites across the chr11p13 region by ChIP-qPCR. CCCTCbinding factor (CTCF) is an architectural protein involved in higher order chromatin organization. It is recruited to invariant sites in the genome (topologically associated domain (TAD) boundaries) in addition to variant structural elements that may be cell-type specific. [37][38][39] The EHF intron 6 element binds CTCF with high affinity in several airway cell types, 25 so its removal was predicted to impact CTCF recruitment at adjacent sites. Consistent with this suggestion, deletion of the EHF intron 6 element was associated with a substantial (~3fold) decrease in CTCF occupancy at the adjacent site DHS11.2516 ( Figure 1C), though this did not reach statistical significance.
To examine the impact of loss of the EHF intron 6 element on 3D chromatin structure more broadly, we performed 4C-seq on the deletion clones and compared them to WT A549 c9 cells ( Figure 1D). A minimum of 2 deletion clones were analysed for each viewpoint, and each clone was evaluated in replica experiments.

| Deletion of the 11.2521 CRE enhances ELF5 expression and increases interactions of the ELF5 promoter with novel sites
The 11.2521 CRE was shown previously to be a strong enhancer of both the ELF5 and EHF promoters (though not APIP), by luciferase assays in 16HBE14o-bronchial epithelial cells. 25 These data were confirmed here in A549 c9 cells ( Figure S1A Figure 2A. First, ELF5, EHF and APIP expressions were measured using RT-qPCR ( Figure 2B). Removal of this CRE did not alter EHF or APIP expression; however, it enhanced transcript abundance of ELF5 (~11-fold, P < .0001).
Since the 11.2521 CRE is an enhancer of the ELF5 and EHF promoters in transient assays, 25  shown earlier by 4C-seq to interact with many other sites of CTCF occupancy across the region, 25 so we predicted that its removal could alter occupancy at some other sites. Consistent with this prediction, removal of the 11.2521 CRE was associated with moderately elevated CTCF occupancy at HB11.1485 (~2-fold) ( Figure 2C), though this alteration did not reach statistical significance.
Next, to investigate the impact of removal of the 11.2521 CRE on chromatin architecture across chr11p13, we performed 4C-seq on the deletion clones and compared them to WT A549 c9 cells ( Figure 2D). Again, two deletion clones were analysed for each viewpoint and each clone was assayed in replica experiments. One experiment is shown in each 4C-seq panel and is consistent with its replicates. Using a viewpoint at the ELF5 promoter, loss of the 11.2521 CRE increased its interactions with DHS11.2512/11.2513 and an uncharacterized element immediately 3' to EHF (black arrows). However, the deletion had no impact on interactions with the HB11.1485 viewpoint, while associations with the EHF promoter were modestly reduced (black dotted line). These data suggest that the 11.2521 CRE may inhibit ELF5 promoter interactions with previously uncharacterized enhancer elements, and that its functions depend upon specific chromatin architecture at chr11p13.

| Deletion of the 11.2516 CRE increases ELF5 but not EHF expression and alters chromatin architecture
The 11.2516 CRE was also shown previously to be a strong enhancer of both the ELF5 and EHF promoters (though not APIP), by luciferase assays in 16HBE14o-cells. 25 These data were confirmed here in A549 c9 cells ( Figure S1A-B). To examine the F I G U R E 1 Impact of deletion of cis-element EHF intron 6 in A549 c9 cells. (A) EHF intron 6 maintains a structural role at 11p13 and helps to facilitate looping of the 11.2516 enhancer (green arrow) to the ELF5 and EHF promoters (yellow arrows). Upon deletion of EHF intron 6 (red X), EHF expression is enhanced. Additionally, CTCF occupancy at 11.2516 is reduced (smaller grey circle) and its interactions with the ELF5 and EHF promoters are weakened (yellow dashed arrows). (B) RT-qPCR for ELF5, EHF and APIP expression in non-targeted WT (NT) and deletion clones. Data are normalized to β2M and relative to clonal WT. Error bars are standard deviation (SD), n ≥ 3. ****P < .0001, ***P < .001, **P < .01, *P < .05, ns = not significant by a two-way ANOVA test with multiple comparisons. (C) ChIP-qPCR for CTCF enrichment at interacting sites across the 11p13 region in WT and deletion clones. ChIP results are shown as percent recovery over input and are normalized to a positive control (CFTR + 48.9 kb ( 28 Figure 3A. First, ELF5, EHF and APIP expressions were measured using RT-qPCR ( Figure 3B).
Removal of this CRE did not alter abundance of the EHF or APIP transcripts; however, it was associated with an increase in ELF5 expression (~7-fold, P = .0004).
As for the 11.2521 CRE, it seemed probable that the in vitro enhancer activity of the 11.2516 CRE might be dominated by features of the chromatin architecture in the genomic context. To test this prediction, we measured CTCF occupancy at known sites across chr11p13 using ChIP-qPCR. The 11.2516 CRE was shown previously to bind CTCF in multiple airway cell types. 25 Hence, it was not unexpected to observe that removal of this element reproducibly enhanced CTCF occupancy at adjacent sites HB11.1485 (~2.4-fold increase) and EHF intron 6 (~1.4-fold increase), though these alterations did not reach statistical significance ( Figure 3C).
This enhancement in occupancy at nearby sites is a common feature of removal of key sites of CTCF occupancy from a genomic region. 28,41 Next, to examine the impact of removal of the 11.2516 CRE on chromatin architecture, we performed 4C-seq on the deletion clones and compared them to WT A549 c9 cells ( Figure 3D). Again, two deletion clones were analysed for each viewpoint and each clone was evaluated in replica experiments. One experiment is shown in each 4C-seq panel and is consistent with its replicates.
Removal of the 11.2516 CRE enhanced interactions between the ELF5 promoter viewpoint and the 11.2512/11.2513 and EHF intron 6 elements (black arrows). These data are consistent with our previous results showing direct interaction of 11.2516 with these sites. 25 In contrast, reduced associations were evident between a viewpoint at HB11.1485 and multiple sites across the region, including the ELF5 promoter, EHF promoter and EHF intron 6 (black arrows). Loss of the 11.2516 CRE also coincided with a modest reduction of interactions between the EHF promoter and the rest of the region, with the exception of an unidentified element located between HB11.1485 and 11.2516 (black arrow), which showed an enhanced association. These data suggest that 11.2516 has an important role in the maintenance of chromatin architecture at the chr11p13 region.

| Expression of ELF5 is associated with altered chromatin architecture
None of the airway cell lines used in this work and our earlier analysis of the chr11p13 region 11,25 express detectable levels of ELF5 protein, and ELF5 transcript abundance is very low ( Figure S2) .2521 interacts with both the ELF5 and EHF promoters (blue arrows) and enhances their expression. Upon deletion of 11.2521 (red X), ELF5 expression is enhanced. Additionally, interactions between the ELF5 promoter and DHS 11.2512/11.2513 (pink arrow) and an unknown element 3' to the EHF locus (purple arrow) are evident. CTCF occupancy at 5' sub-TAD boundary HB11.1485 is slightly increased (two grey circles). (B) RT-qPCR for ELF5, EHF and APIP expression in non-targeted WT (NT) and deletion clones. Data are normalized to β2M and relative to clonal WT. Error bars are SD, n ≥ 3. ****P < .0001, ***P < .001, **P < .01, *P < .05, ns = not significant by a two-way ANOVA test with multiple comparisons. (C) ChIP-qPCR for CTCF enrichment at interacting sites across the 11p13 region in WT and deletion clones. ChIP results are shown as percent recovery over input and are normalized to a positive control (CFTR + 48.9 kb). Error bars are SD, n ≥ 3. (D) 4C-seq data as in Figure 1    has not yet been studied.
Since expression of ELF5 is associated with novel peaks of open chromatin at chr11p13, it seemed probable that altered chromatin architecture might also be seen in ELF5 expressing cells.
To investigate this hypothesis, we performed 4C-seq on Calu3, LNCaP and T47D cells using viewpoints at the ELF5 and EHF promoters ( Figure 4C). The Using the EHF promoter viewpoint, the major specific difference seen in Calu3, LNCaP and T47D cells was a great reduction in interactions with the HB11.1485 5' sub-TAD boundary when the ELF5 gene is highly expressed (T47D) (black arrows). A minor reduction of interactions with the EHF intron 6 CRE was also observed in LNCaP and

T47D cells (black arrows). In addition, in Calu3 cells only, interactions
were evident with the airway enhancers (11.2516 and 11.2521), EHF intron 6 and the 11.2525/11.2526 sites closest to the highest p-value SNP (rs10742326) from the GWAS 4 (black arrows). In summary, these data suggest that alterations in chromatin architecture at chr11p13 and the loss of a 5' sub-TAD boundary enable expression of ELF5. to target sites. A minimum of two guide RNAs were designed to target the ELF5 and EHF promoters and multiple airway DHS across chr11p13 in 16HBE14o-and A549 cells ( Figure 5 and Figure S3).

| Activation of CREs across chr11p13 enhances ELF5 expression, but has little impact on EHF or APIP
Targeting the ELF5 promoter increased its expression by ~2600-fold in 16HBE14o-and ~40-fold in A549 cells. Targeting the EHF pro- and EHF, or that a more complex regulatory mechanism exists for the genes in this region. The latter explanation would be consistent with our 4C-seq observations suggesting close interactions between the ELF5 and EHF genes and their CREs.

| ELF5 occupies regions of open chromatin at EHF and represses its expression
We next tested the hypothesis that the close physical interactions between the active ELF5 and EHF loci and their CREs might underlie the complex regulatory environment at chr11p13. We first performed siRNA-mediated depletion of ELF5 in T47D cells, which express ELF5 and EHF, and assayed expression of both genes by RT-qPCR. Primers located in exons 5 and 6 for ELF5 and exons 3 and 4 for EHF (Table S3) enabled detection of all major isoforms of both F I G U R E 3 Impact of deletion of cis-element 11.2516 in A549 c9 cells. (A) 11.2516 interacts with both the ELF5 and EHF promoters and enhances their expression (yellow arrows). 11.2516 also interacts with the 11.2521 enhancer element (yellow arrow) and helps to recruit this DHS to the gene promoters. Furthermore, 11.2516 interacts strongly with 5' sub-TAD boundary HB11.1485 and helps to loop this element to the ELF5 promoter, EHF promoter and EHF intron 6 (red arrows). Upon deletion of 11.2516 (red X), ELF5 expression is enhanced. Additionally, interactions between HB11.1485 and the ELF5 promoter, EHF promoter and EHF intron 6 are reduced (red dashed arrows), which allows the ELF5 promoter to gain interactions with CREs, particularly DHS 11.2512/11.2513 (pink arrow) and EHF intron 6 (purple arrow). Furthermore, CTCF occupancy increases at adjacent sites HB11.1485 and EHF intron 6 (two grey circles). (B) RT-qPCR for ELF5, EHF and APIP expression in non-targeted WT (NT) and deletion clones. Data are normalized to β2M and relative to clonal WT. Error bars are SD, n ≥ 3. ****P < .0001, ***P < .001, **P < .01, *P < .05, ns = not significant by a two-way ANOVA test with multiple comparisons. (C) ChIP-qPCR for CTCF enrichment at interacting sites across the 11p13 region in WT and deletion clones. ChIP results are shown as per cent recovery over input and are normalized to a positive control (CFTR + 48.9 kb). Error bars are SD, n ≥ 4. (D) 4C-seq data as in Figure 1 in WT and one 11.2516 deletion clone. Viewpoints at the ELF5 promoter, HB11.1485 and EHF promoter (red dotted lines). Open red arrowheads indicate deletion site. Informative interactions are shown as black arrows   Figure 6A). Correspondingly, ELF5 depletion also significantly increased EHF protein expression, suggesting that EHF transcription is repressed by ELF5 ( Figure 6B).
A previous study 18 suggested a direct repression of EHF by ELF5, though as ELF5 was overexpressed, there were concerns of potential off-target effects of the TF. To determine whether the repression of EHF by ELF5 observed in our experiments was dependent on direct occupancy of CREs for the EHF locus by ELF5, we performed ChIP-seq using an antibody specific for endogenous ELF5 in T47D cells. The Irreproducible Discovery Rate (IDR) analysis of two replica experiments is shown in Figure 6C and reveals multiple sites of ELF5 occupancy across the EHF locus, most notably in introns 1 and 6 and close to its 3' end (black arrows). Each of the peaks of ELF5 binding also coincides with a peak of open chromatin in T47D cells, mapped by ATAC-seq ( Figure 6C upper track). The sites of ELF5 occupancy at the EHF locus were confirmed by ChIP-qPCR analysis ( Figure 6D). In addition, ELF5 occupancy coinciding with sites of open chromatin is seen at the APIP/PDHX and ELF5 gene promoters (grey arrows), suggesting that ELF5 may have a major impact across 11p13, both through controlling other genes and by autoregulation. These results are highly relevant to primary bronchial and tracheal epithelial cells, which also express both ELF5 and EHF, albeit with much lower ELF5 abundance than in T47D cells.

| D ISCUSS I ON
Intense interest in the chr11p13 region arose through its association with lung disease severity in CF; however, the genetic mechanisms underlying the functions of this region are not well understood. Our data reveal a complex regulatory network at chr11p13 that coordinates the expression of the ETS transcription factors ELF5 and EHF.
We previously identified several putative airway epithelial cell-selective enhancer elements of ELF5 and EHF within the region and defined important aspects of the chromatin architecture relevant to the regulation of genes at chr11p13. 25 Here, we performed functional genomics experiments to elucidate the endogenous role of two predicted enhancers and a CRE that lies within a stretch enhancer at the EHF locus.
Deletion of the CRE in EHF intron 6 and the two airway epithelial cell-selective enhancers at 11.2516 and 11.2521 using CRISPR/Cas9 protocols had unexpected effects on gene expression, and were associated with modest changes in CTCF occupancy and local higher F I G U R E 5 Changes in 11p13 gene expression with VPR-mediated activation of promoters and cis-elements in 16HBE14o-cells. VPR-mediated activation of ELF5 promoter, EHF promoter and DHS at 11p13. EHF intron 6 encompasses Ca11.1058 and Ca11.1327. Error bars are SEM, n = 3. ****P < .0001, ***P < .001, **P < .01, *P < .05, ns = not significant by an unpaired Student's t test. † P < .0001 -statistically significant, but not biologically significant due to low expression levels. Inset panel shows    shows eRNA transcription at the three of CREs studied here, notably from the antisense strand at EHF intron 6 ( Figure S4). 47 Figure 2D) suggesting that this site is at least in part, responsible for the large increase in ELF5 transcript levels in these deletion clones.
It seems probable the 11.2513 and novel CREs either harbour enhancers of the ELF5 promoter and/or are sites of occupancy for this transcription factor.
In addition to the novel CREs observed in T47D and LNCaP cells, 4C-seq data showed a loss of interactions between the EHF promoter and the HB11.1485 CRE in comparison to the airway cell lines ( Figure 4C). In this context, our earlier work 25  Returning to the role of chr11p13 as a modifier of lung disease severity, we do not address here the mechanisms underlying the association of high p-value SNP close to the chr11.2525 site. However, our observations on cell-specific recruitment of CREs across the modifier region and most importantly on the direct regulation of EHF by ELF5 are significant advances in understanding the region. EHF is intimately involved in regulating the wound response, inflammation and tissue remodelling, all of which are integral components of CF lung pathology. 11,12 For example, EHF enhances expression of S100 Calcium Binding Proteins A8 and A9 (S100A8 and S100A9), which are essential for cell motility [52][53][54] and thus the wound response. Proper wound response is crucial for injury repair, and this process is delayed in the CF lung. 12 Also, EHF enhances the transcription of genes encoding the pro-inflammatory cytokines interleukin 8 (IL-8) and C-X-C motif chemokine ligand 6 (CXCL6). 11,12 These secreted cytokines are necessary for the recruitment of neutrophils to a site of infection or injury. 55,56 Increasing the neutrophil burden in the CF lung can lead to chronic inflammation and eventual tissue remodelling, 57 both of which significantly contribute to patient mortality. Furthermore, EHF promotes expression of Sam Pointed-Domain Containing ETS Transcription Factor (SPDEF), 12 which is involved in interleukin 13 (IL-13)-mediated goblet cell differentiation. 58 Enhanced goblet cell differentiation increases expression of the mucin 5AC (MUC5AC) gene, 59,60 which encodes one of the major secreted mucins in the lung (reviewed in 61 ) and is a hallmark of CF lung disease. Since our data show that ELF5 occupies open chromatin sites at the EHF locus and represses its expression, it is likely that this ETS factor also plays an important role in lung biology. Modulation of ELF5 expression would be predicted to impact many of the EHF-mediated processes discussed above. Although EHF remains the primary candidate for influencing lung phenotype differences observed in CF patients, ELF5 likely also plays an important role through its regulation of EHF expression. These observations are not only relevant to CF but also to other airway diseases associated with inflammation, defective wound response and lung tissue remodelling, including asthma, chronic obstructive pulmonary disease (COPD) and bronchiectasis. 62,63

CO N FLI C T S O F I NTE R E S T
The authors confirm that there are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Genome-wide data are deposited at GEO: GSE131084.