KLF4 recruits SWI/SNF to increase chromatin accessibility and reprogram the endothelial enhancer landscape under laminar shear stress

Physiologic laminar shear stress (LSS) induces an endothelial gene expression profile that is vasculo-protective. In this report, we delineate how LSS mediates changes in the epigenetic landscape to promote this beneficial response. We show that under LSS, KLF4 interacts with the SWI/SNF nucleosome remodeling complex to increase accessibility at enhancer sites that promote the expression of homeostatic endothelial genes. By combining molecular and computational approaches we discover enhancers that loop to promoters of KLF4- and LSS-responsive genes that stabilize endothelial cells and suppress inflammation, such as BMPR2, SMAD5, and DUSP5. By linking enhancers to genes that they regulate under physiologic LSS, our work establishes a foundation for interpreting how non-coding DNA variants in these regions might disrupt protective gene expression to influence vascular disease.

Regions of the arterial circulation exposed to uniform physiologic laminar shear stress (LSS) are associated with expression of vasculoprotective genes such as endothelial nitric oxide synthase (NOS3), whereas regions of low and disturbed shear stress, near arterial branches and bifurcations, are more prone to develop disease [1][2][3][4][5] . However, the nature and severity of vascular lesions depends on whether there are mitigating genetic and environmental factors that not only worsen the propensity to disease in vulnerable regions of the circulation, but also perturb the normal response to LSS.
Major effort has been directed at finding mechanosensing complexes in endothelial cells (EC) that respond to LSS to initiate a pattern of protective gene regulation that is altered in disease [6][7][8] . Krüppel-like factors (KLF) 2 and 4 have been identified as key transcription factors induced by LSS that control endothelial homeostatic gene regulation [9][10][11][12][13] . Beyond transcriptional activation at gene promoters, these factors might function as chromatin organizers to regulate genes by altering the chromatin landscape at distal enhancers. Knowing how the enhancer landscape changes in response to LSS provides a blue-print that will better inform how variants in non-coding DNA can influence vascular diseases, including pulmonary arterial hypertension (PAH), which is characterized by progressive occlusion of distal pulmonary arteries 14,15 . Pathogenic gene variants in BMPR2 are the most common genetic risk factor for PAH, but have been associated with only 15% of patients with the non-familial form of the disease 16,17 , suggesting that variants in non-coding DNA regions might contribute to a reduction in BMPR2 levels or in downstream effectors of BMPR2 mediated gene regulation in patients with PAH.
Whereas variants in coding and proximal regulatory DNA regions can be readily related to dysfunction of a particular disease-causing gene, ascribing variants in distal non-coding regions to genes related to vascular disease has been more elusive. New bioinformatic approaches are overcoming this hurdle. For example, a variant in a distal enhancer that targets endothelin-1 was recently linked to five vascular diseases 18 . With more effort directed at whole genome sequencing in cohorts of patients such as those with PAH, establishing the functional significance of variants in non-coding DNA has become increasingly important. Moreover, variants identified in enhancers that are active under physiologic LSS might have the greatest adverse impact on gene regulation.
We, therefore, sought to map the enhancer landscape of pulmonary artery EC (PAEC) exposed to LSS by first relating changes in chromatin accessibility, determined by ATAC-Seq, to gene expression, assessed by RNA-Seq. We find an overall increase in chromatin accessibility, coinciding with the induction of protective genes and suppression of disease-related genes. Motif analyses of differentially accessible regions (DAR) reveal enrichment for KLF binding sites in regions with increased accessibility under LSS, which we confirm by KLF4 ChIP-Seq. We demonstrate, using KLF gain-and-loss-of-function experiments, that KLF is required to mediate chromatin accessibility changes at proximal and distal sites, adding to the known role for KLF in transcriptional activation of flow responsive genes.
Consistent with this finding, we identify components of the SWItch/Sucrose Non-Fermentable (SWI/SNF) nucleosome remodeling complex that are recruited by KLF4 and required for KLF-mediated gene regulation. H3K27ac HiC Chromatin Immunoprecipitation (HiChIP) 19,20 shows increased enhancer looping to LSS-responsive genes, often spanning multiple genes. Application of the Activity-by-Contact (ABC) algorithm 21 , which identifies regulatory elements of specific genes based upon chromatin accessibility and H3K27ac levels, predicts more than 70% of genes differentially expressed under LSS to be regulated by KLF4. These data provide a blueprint of the endothelial enhancer landscape under physiologic LSS, which informs future studies on how variants in non-coding DNA might impair protective gene expression and promote vascular disease.

LSS alters chromatin accessibility, increasing the expression of vasculo-protective genes
Genome-wide changes in chromatin accessibility and gene expression were studied by ATAC-Seq and RNA-Seq in PAEC exposed to a physiologic level of 15 dyn/cm 2 of LSS 5,22 for 24 h (Fig. 1a). Exposure to LSS increased accessibility of 4699 genomic regions while 2191 regions showed a loss of accessibility compared to PAEC cultured under static conditions (ST) (Fig. 1b). Whereas a subset of these DAR contains gene promoters, most DAR are located in introns or at intergenic sites (Fig. 1c). When assigning DAR to their nearest genes, we found that the majority of DAR are located within 100 Kb of a transcription start site (Fig. 1d). We then performed motif enrichment analyses on all DAR. In regions with increased accessibility under LSS, we found enrichment for motifs that are recognized by members of the KLF family (77% of DAR, P = 1E−227), while activator protein-1 (AP1) motifs were most enriched in regions where accessibility was lost (58% of DAR, P = 1E −205). Motifs for E26 transformation-specific (ETS) family members were present in about 80% of DAR under both LSS and static culture conditions, which is in line with previous reports indicating ETS as indispensable for endothelial gene regulation 23,24 (Fig. 1e and Supplementary Fig. 1).
Immunofluorescence was next applied to assess nuclear expression of these transcription factor families under LSS and ST conditions, which confirmed the increased expression of KLF4 with LSS, while the AP1 family member activating transcription factor-2 (ATF2), which is increased in endothelial cells in atherosclerotic lesions 25 , was increased under static conditions. ETS1 was constitutively expressed under both conditions (Fig. 1f).
To relate chromatin accessibility changes to altered gene expression, we performed RNA-Seq of PAEC cultured under identical conditions. Induction of the LSS-responsive transcription factors KLF2 and KLF4 was confirmed by RNA-Seq and RT-qPCR ( Supplementary  Fig. 2a, b). We observed 720 and 663 genes significantly up-or downregulated, respectively, with LSS compared to ST conditions (Fig. 1g). Exposure to LSS resulted in induction of a quiescent vasculoprotective gene expression profile, as evident in enrichment for genes related to blood vessel development, improved barrier function and enhanced cell-matrix interaction (Fig. 1h, upper panel). In contrast, the gene expression profile of PAEC cultured under ST conditions is typical of a more activated and proliferative phenotype, including genes that promote cell migration, cell cycle molecules, and genes related to oxidative stress (Fig. 1h, lower panel). We found that the gene expression profile of PAEC cultured under static conditions is very similar to that of PAEC exposed to +/−3 dyn/cm 2 of oscillatory shear stress (OSS) ( Supplementary Fig. 2c-e) which agrees with previous observations by Qiao et al. 26 .
DEG without proximal accessibility changes were enriched for pathways related to suppression of cell proliferation, e.g., a decrease in nucleosome assembly genes (CENPA, CENPM, CENPN, and CENPP), multiple histone genes, and suppression of genes related to the resolution of sister chromatids (PLK1 and AURKB), as well as several minichromosome maintenance (MCM) genes that are involved in the initiation of eukaryotic genome replication (MCM2, MCM4, MCM5, and MCM7) ( Supplementary Fig. 3a, c).

KLF regulates chromatin accessibility of vasculoprotective genes under LSS
To determine whether KLF regulates changes in chromatin accessibility under LSS, we performed ATAC-Seq under conditions of KLF gain-or loss-of-function. For gain-of-function studies, we transduced PAEC with an adenoviral vector encoding a constitutively active mutant of mitogen-activated protein kinase kinase 5 (caMEK5), which leads to phosphoactivation of signal-regulated kinase 5 (ERK5) and the induction of KLF2 and KLF4 27,28 . . PAEC transduced with caMEK5 showed a 14-and 22-fold increase in expression of KLF2 and KLF4, respectively, compared to controls transduced with a vector encoding green fluorescent protein (GFP) (Supplementary Fig. 4a) which is similar to what we observed in wildtype cells exposed to LSS ( Supplementary Fig. 2a, b). Almost half of the DAR previously identified under LSS versus ST were also differentially accessible in caMEK5-transduced PAEC compared to GFP controls (n = 3211/6890), and accessibility changes were highly correlated (r = 0.89, P < 0.0001) (Fig. 2a, left panel). When plotting the correlation of all LSS vs ST DAR with accessibility changes in caMEK5-transduced PAEC at those sites, we found a correlation of r = 0.78 (P < 0.0001) ( Supplementary  Fig. 4b), indicating that activation of ERK5 and induction of KLF2/4 as previously described with LSS 28 is largely responsible for the changes in chromatin accessibility. Similarly, the gene expression changes in caMEK5-transduced cells assessed by RNA-Seq strongly correlated with those observed with LSS (n = 1164, r = 0.82, P < 0.0001). (Fig. 2a Fig. 4c). We note that this still corresponds to 4.9-and 7.8-fold higher gene expression compared to ST conditions. KLF2/4 protein expression was reduced by 60% (Supplementary Fig. 4d). A subset of regions did not show the increase in accessibility that we previously observed with LSS (Fig. 2b, left panel) which coincided with a subset of genes not being induced (Fig. 2b, right panel). When plotting all LSS vs ST DAR with accessibility changes in KLF RNAi-treated PAEC at those sites, we found a correlation of r = −0.73 (P < 0.0001) ( Supplementary Fig. 4e).
We observed a modest increase in accessibility at the promoter regions of a known endothelial flow responsive gene and KLF4 target, NOS3 11 ( Supplementary Fig. 5), as these sites were also highly accessible under static conditions. We also identified prominent DAR associated with flow-responsive genes. An example is DUSP5, a dual specificity b Volcano plot showing the differentially accessible regions (DAR) determined by ATAC-Seq in PAEC that were exposed to 15 dyn/cm 2 of LSS for 24 h vs static controls. Each dot represents a DAR. Red dots are DAR with increased accessibility, blue dots are DAR with decreased accessibility under LSS. Highlighted genes are described in subsequent sections. n = 3 experimental replicates. P values were determined by a two-tailed Wald test with Benjamini-Hochberg adjustment. c Pie chart showing the annotation of DAR peaks, analyzed using Homer. d Bar graphs showing the distance of DAR to its nearest gene, analyzed using Homer. e Motif enrichment of DAR with increased or decreased accessibility with LSS, analyzed using Homer. P values were determined by binomial test. f Representative immunofluorescent images (left) with quantitation on the right, of PAEC that were exposed to 15 dyn/cm 2 of LSS for 24 h vs static controls (ST), showing the nuclear expression of KLF4 (green); ETS1 (gray), and ATF2 (red). Nuclei were stained with DAPI, (blue). n = 4 experimental replicates. Bars indicate mean ± s.e.m. P values were determined by Student's two-tailed t-test. ns: not significant. Source data are provided as a Source Data file. Scale bar, 20 μm. g Volcano plot showing the differentially expressed genes (DEG) determined by RNA-Seq in PAEC exposed to 15 dyn/cm 2 of LSS for 24 h vs Static controls. Each dot represents a gene. Red dots are DEG with increased expression under LSS, blue dots are DEG with decreased expression under LSS. Highlighted genes are described in subsequent sections. n = 2 experimental replicates. P values were determined by a two-tailed Wald test with Benjamini-Hochberg adjustment. h Gene enrichment analysis using Metascape of genes that are induced by LSS (red bars) and genes that are repressed by LSS (blue bars). P values were determined by a one-tailed hypergeometric test with Benjamin-Hochberg adjustment. i Scatterplot showing a positive correlation of accessibility changes with the expression of its nearest differentially expressed gene. Highlighted genes are described in subsequent sections. r = 0.41 with P < 0.0001 determined by two-tailed Pearson R test. phosphatase that inhibits ERK1/2 signaling 29 . ATAC-Seq identified a potential regulatory DAR located 39 Kb upstream of the DUSP5 transcription start site (TSS). ATAC-qPCR confirmed the increased accessibility at that site with LSS, which coincided with higher DUSP5 mRNA transcript levels (Fig. 2c). When investigating a broader region surrounding the DUSP5 locus, many sites where similarly accessible under LSS versus ST conditions ( Supplementary Fig. 6). caMEK5 transduced cells had increased accessibility at the regulatory region located 39 kb upstream of the TSS, which was confirmed by qPCR ( Fig. 2d and Supplementary Fig. 6). Upon silencing of KLF2 and KLF4 by RNAi prior to exposure to LSS, neither an increase in this DAR, nor the induction of DUSP5 gene expression were observed ( Fig. 2e and Supplementary  Fig. 6). DUSP5 loss-of-function experiments revealed that DUSP5 is responsible for repressing multiple chemoattractant, antiviral and interferon genes under LSS ( Supplementary Fig. 7). In addition to DUSP5, we investigated other DEG that had DAR associated with it, such as those illustrated in Supplementary Fig. 3. For COL3A1 and DLL1, we found a single DAR proximal to the gene TSS. For ITGB4, ITGA6 and HYAL1, we found multiple DAR near the TSS (Supplementary Fig. 8). As we observed at the DUSP5 locus, across a broader region, many sites were similarly accessible under LSS and ST conditions, reinforcing that the DAR represent flow-responsive regulatory sites. Overall, we found lower intensity of ATAC-Seq peaks with KLF siRNA, compared to untreated PAEC exposed to LSS or those transduced with caMEK5, likely for technical reasons.

KLF4 interacts with the SWI/SNF nucleosome remodeling complex to increase chromatin accessibility
Having established that loss of KLF reduces chromatin accessibility, we next sought to identify a chromatin remodeler with which it might interact to exert these changes. We transduced PAEC with an adenoviral vector encoding a Flag-tagged mutant of KLF4 or with a GFP vector as control, and performed affinity purification followed by mass spectrometry (AP-MS) using anti-KLF and anti-Flag antibodies. We found, with both antibodies, that KLF4 co-purified with ten components of the SWI/SNF nucleosome remodeling complex, including the ATP-dependent helicase Brahma-Related Gene 1 (BRG1), also known as SMARCA4, and the invariable core subunit SMARCC2 (Fig. 3a). In addition to the SWI/SNF complex, we found KLF4 co-purified with several other chromatin remodelers that enhance chromatin c Representative image of KLF4-BRG1 PLA in healthy rat lung tissue. The original PLA protocol was modified with longer incubation times to allow the reagents to fully penetrate the 350 μM thick sections. Note the interaction of KLF4 with BRG1 at straight sites of the vasculature that are exposed to LSS (upper left insert), while sites at bifurcations which are exposed to disturbed shear stress (DSS) do not show the KLF4-BRG1 interaction (bottom right insert). Bar graphs show the average number of foci per nucleus quantified in regions of DSS or LSS at 6 random bifurcations. KLF4-BRG1 interaction (magenta); vWF (gray, pseudo-color); αSMA (green). Nuclei were stained with DAPI, (blue). Scale bar, 20 and 10 μm in the higher magnifications. Data shown as the mean +/− s.e.m. P value was determined by Student's two-tailed t-test. Source data are provided as a Source Data file. accessibility and facilitate transcriptional activation, including the lysine demethylase KDM2A and Bromodomain Containing 4 (BRD4), as well as remodelers that decrease accessibility resulting in transcriptional repression, such as histone deacetylase HDAC2 and DNA methyltransferases DNMT1 and DNMT3A (Fig. 3a).
To confirm the interaction between KLF4 and the SWI/SNF complex in cells and tissue, we performed in situ proximity ligation assays (PLA) for KLF4 and either BRG1 or SMARCC2, which requires the epitopes of both proteins to be within 40 nm of each other to be detected. Indeed, in PAEC exposed to LSS, we observed the interaction between KLF4 and both SWI/SNF components (Fig. 3b). To determine if this interaction also occurs in the intact vessel wall, we modified the PLA protocol to allow us to study the interaction between KLF4 and the SWI/SNF complex in the rat pulmonary vasculature. Since SMARCC2 is a core subunit of the SWI/SNF chromatin remodeling complex that colocalizes with BRG1, we focused these studies on the interaction between KLF4 and BRG1. Using deep tissue imaging we studied regions with bifurcations to allow us to compare straight parts of the vessel exposed to LSS, with branching points that are exposed to disturbed shear stress ( Supplementary Fig. 9). The KLF4-BRG1 interaction was prominent at sites of LSS and much less apparent at regions with disturbed flow (Fig. 3c).

KLF4 and BRG1 co-occupied regions show an enhancer chromatin signature
To study the genome wide chromatin accessibility changes coregulated by KLF4 and BRG1, we intersected KLF4 and BRG1 peaks from the caMEK5 ChIP-Seq data with the ATAC-Seq data of PAEC exposed to LSS. Of the 8,404 regions that were enriched for both KLF4 and BRG1, 2,955 were found at DAR that open with LSS, representing 62.9% of all LSS DAR (Fig. 4a). In contrast, KLF4-BRG1 co-occupancy was only found at 4.2% of accessible regions that were not differentially accessible with LSS ( Supplementary Fig. 10a) and was only present in a small subset of DAR (2.1%) that lose accessibility under LSS (Supplementary Fig. 10b). An increase in KLF4 binding at KLF4-BRG1 cooccupied DAR was associated with increased accessibility (Fig. 4b).
Increased BRG1 peaks were found at 77.6% of DAR with increased accessibility under LSS (Fig. 4a), as well as at 53.7% of DAR that have increased accessibility under ST conditions ( Supplementary Fig. 10b), but only in 13.5% of regions that were not differentially accessible ( Supplementary Fig. 10a), demonstrating that BRG1 binding coincides with increased chromatin accessibility even under ST conditions. In addition, when plotting changes in BRG1 binding at LSS vs ST DAR, we found that increased BRG1 binding correlates with an increase in accessibility, and a decrease in BRG1 with loss of accessibility (r = 0.89, p < 0.0001) (Supplementary Fig. 10c).
An example of a KLF4-BRG1 co-occupied site is the DAR 39 Kb upstream of DUSP5 that we had previously identified by ATAC-Seq ( Fig. 4c and Supplementary Fig. 11a). We also studied KLF4-BRG1 binding at DAR located at the COL3A1, ITGB4, ITGA6, DLL1, and HYAL1 loci that were illustrated before ( Supplementary Fig. 8). In all cases, we found enrichment for KLF4, generally coinciding with increased BRG1 occupancy (Supplementary Fig. 11b-f). In one of the DAR upstream of ITGB4, BRG1 occupancy and chromatin accessibility were also high under ST conditions, suggesting that this site is regulated by other factors under ST conditions. ( Supplementary Fig. 11c).
To confirm that BRG1 is required for the chromatin accessibility changes, we transfected PAEC with siRNA targeting BRG1 prior to exposure to LSS ( Supplementary Fig. 10d). Indeed, loss of BRG1 decreased chromatin accessibility of the DAR upstream of DUSP5 as assessed by ATAC-qPCR, coinciding with reduced DUSP5 mRNA transcript levels (Fig. 4d).
To further characterize KLF4-BRG1 co-occupied regions, we performed ChIP-Seq for the H3K4me1 enhancer mark, as well as for H3K27ac which is typically present at active enhancers and promoter regions. The majority of KLF4-BRG1 co-occupied regions were enhancers marked by H3K4me1 (91%, Fig. 4e) two-thirds of which were also marked by H3K27ac (66%, Fig. 4e).

KLF4 regulates gene expression by binding at enhancer loops
Thus far, we have related chromatin accessibility changes to gene expression based on the nearest gene. However, we found a direct relationship in only a subset of DEG, possibly because several enhancers target a particular gene, or because enhancer loops can skip promoters of proximal genes to target a more distal gene. We, therefore, applied two complementary strategies to map enhancerpromoter regulation: H3K27ac HiC Chromatin Immunoprecipitation (HiChIP), which measures the frequencies of 3D contacts between enhancers and promoters 19,20 , and the Activity-By-Contact (ABC) algorithm, that predicts enhancer-promoter interactions based on ATAC-Seq and H3K27ac ChIP-Seq data 21 .
Significant chromatin loops identified by HiChIP were determined using FitHiChIP 30 . At a one Kb resolution, we identified 4698 enhancerpromoter (EP) loops, of which 1,299 were differential between LSS and ST conditions. We then integrated these data with KLF4 ChIP-Seq to study KLF4 binding at enhancer anchor and found differential KLF4 binding at 856 enhancer anchors that regulate 325 DEG (Fig. 5a). To determine whether KLF is required to form these EP loops, we performed H3K27ac HiChIP in PAEC treated with siRNA targeting KLF2/4 prior to exposure to LSS. More than a third of the enhancer anchors were lost with KLF siRNA. The enhancer anchors that were lost targeted more than half of the DEG that are regulated by EP loops in wildtype PAEC exposed to LSS (Fig. 5b).
The ABC algorithm predicts which distal elements regulate which genes, considering elements located at both shorter (<10 Kb) and longer (>10 Kb) ranges 21 . The algorithm scores elements based on chromatin accessibility determined by ATAC-Seq and the strength of the H3K27ac enhancer signal at those sites and has been validated by perturbing thousands of putative enhancers using CRISPR interference (CRISPRi) 21 . The algorithm predicted 996/1,383 DEG under LSS to be regulated by enhancer elements that were occupied by KLF4 (Fig. 5c). ABC confirmed most of the enhancer loops that were identified by HiChIP, for example, the DAR that was observed near DUSP5 ( Fig. 5d and Supplementary Figs. 11a, 12), as well as one targeting SMAD5, a transcriptional effector of BMPR2 signaling ( Fig. 5e and Supplementary Fig. 13). Both loops were lost when PAEC were treated with KLF RNAi prior to exposure to LSS (Fig. 5d, e).
KLF4 also occupied enhancer anchors of loops that were lost with LSS, associated with a reduction in H3K27ac levels at those sites and a decrease in target gene expression. These include two enhancer anchors of EP loops that we identified to target endothelin-1 (EDN1), a potent vasoconstrictor that is elevated in patients with PAH, and a major therapeutic target 31,32 . In caMEK5-transduced cells, there was increased binding of KLF4 at these sites with a concomitant decrease in H3K27ac levels ( Fig. 5f and Supplementary Fig 14). When we performed H3K27ac HiChIP in PAEC treated with KLF siRNA prior to exposure to LSS, both loops remained present (Fig. 5f). Based on our AP-MS data (Fig. 3a), we speculate that under this condition, KLF4 might interact with HDAC2 to inactivate the enhancer, resulting in decreased EDN1 mRNA levels (Fig. 5f). In smooth muscle cells, the interaction between KLF4 and HDAC2 promotes transcriptional silencing of SM22α 33 .
We then went on to functionally validate these enhancers using CRISPRi. We transduced PAEC with a lentiviral vector encoding a doxycycline-inducible mutant of a catalytically dead Cas9 fused to the bipartite repressor domain MeCP2-KRAB 34 . We designed synthetic guide RNAs (sgRNA) to target the predicted KLF binding motifs in the candidate enhancer regions, and analyzed target gene expression by RT-qPCR following 24 h of exposure to LSS. Disruption of the DUSP5 and SMAD5 enhancers by CRISPRi resulted in decreased expression of their target genes (Fig. 5d, e), illustrating that they are bonafide enhancers. CRISPRi-mediated disruption of those same sites under ST conditions had no effect on target gene expression, indicating that these regulatory sites are specifically activated under LSS (Supplementary Fig. 15). CRISPRi-mediated inactivation of the EDN1 enhancer sites resulted in an increase in EDN1 gene expression under LSS, confirming that these enhancer elements are repressed under LSS ( Fig. 5f and Supplementary Fig. 15).
The ABC algorithm predicted many shorter-range EP interactions that were not discovered by HiChIP, including two enhancers for the canonical Notch target gene HES2, that has previously been described to be regulated by shear stress 35 (Supplementary Fig. 16). One of these enhancers was identified based on increased chromatin accessibility, despite low H3K27ac expression levels at that site. HiChIP also identified long-range EP loops, often spanning multiple genes, which were not predicted by the ABC algorithm, likely because the distance of an enhancer element to its target gene is a component of the algorithm, favoring shorter range interactions as enhancers are more likely to affect proximal genes. An example is an enhancer loop that spans several genes to target the promoter region of BMPR2. Mutations in the coding and proximal regulatory regions of this gene have been identified in patients with PAH 16,17 , but there might also be mutations in this distal enhancer that have not previously been linked to BMPR2. We identified a KLF4 bound EP loop with increased H3K27ac levels at the enhancer anchor located 1.3 Mb downstream of the BMPR2 transcription start site, associated with a 2-fold increase in BMPR2 mRNA levels under LSS ( Supplementary  Fig. 17). When attempting to silence the enhancer using CRISPRi, we did not observe a significant decrease in BMPR2 mRNA expression, possibly because the proximal enhancers have more pronounced affects or they act in concert with each other, or because the gRNAs did not effectively target the enhancer region.
Non-coding variants at KLF4 bound enhancers are associated with vascular disease More than 90% of single nucleotide polymorphisms (SNPs) identified in whole genome-wide association studies occur in non-coding regions 36 . Since our data indicate that KLF4 regulates endothelial gene expression through enhancer activation, we explored whether genetic variants occur in KLF4 bound distal enhancers, as this may explain how SNPs in non-coding regions might predispose to vascular disease. We intersected KLF4 bound enhancer elements identified by ABC with variants associated with blood pressure traits from a recent genetic association study that included over 1 million people 37 . We found 22 SNPs associated with hypertension in enhancer elements predicted to target 35 genes (Supplementary Fig. 18a). This included rs933795, which is also associated with coronary artery disease 38,39 . This SNP is predicted to target an enhancer for junctional cadherin 5 associated protein (JACD), which was found to regulate the response of endothelial cells to shear stress 40 .
Another example is rs3918226, a variant located in the promoter of NOS3, which is associated with an increased risk for systemic hypertension 41 . This variant is thought to negatively affect transcription of the NOS3 gene, resulting in decreased NO bioavailability 40 , however healthy subjects with this variant do not have altered plasma nitrite levels, suggesting a NOS3 independent effect 42 . Interestingly, it is located at a KLF4 bound enhancer that under LSS, targets AGAP3, a GTPase-activating protein that interacts with myosin phosphatase which regulates endothelial permeability 43 (Supplementary Fig. 18b). The polymorphism may result in increased permeability, facilitating pro-inflammatory activation that could lead to vascular stiffening. While it is probable that this variant impairs NOS3 expression by interfering with NOS3 promotor activation, these data indicate that dysregulation of AGAP3 may also contribute to the increased risk for systemic hypertension.

Discussion
KLF2 and KLF4 are well-recognized LSS-induced transcription factors that control endothelial homeostatic gene regulation [9][10][11][12][13] . Our studies indicate that KLF4, beyond its previously established role as a transcriptional activator of vasculoprotective gene expression, can act as a chromatin organizer by recruiting the SWI/SNF chromatin remodeling complex to alter chromatin accessibility and regulate the endothelial enhancer landscape (Fig. 6).
Shear forces exerted by blood flow are sensed by endothelial cells and converted to protective or vulnerable gene expression profiles through mechanotransduction. Identification of mechanosensing complexes and their downstream signaling pathways has been the subject of extensive investigation and has greatly improved our understanding of how spatiotemporal changes in hemodynamics affect endothelial function and vascular homeostasis [6][7][8] . More recently, the epigenetic regulation of shear stress-mediated gene expression has gained increasing attention. Illi et al. were the first to report the importance of H3 and H4 histone modifications in the regulation of LSS-induced gene expression 44 . Many studies followed that investigated the role of histone acetyltransferases and deacetylases, as well as the histone methyltransferase EZH2, in mediating shear stress-induced gene expression changes both in vitro and in vivo [45][46][47][48][49] . Other studies have addressed the role of DNA methylation and the contribution of DNA methyltransferases (DNMT) in the regulation of gene expression by shear stress [50][51][52] , including the identification of DNMT3A-dependent KLF4 promoter hypermethylation by disturbed flow 52 .
He and colleagues recently demonstrated that KLF4 regulates Inositol 1,4,5-trisphosphate receptor type 3 (IPTR3) expression by increasing chromatin accessibility and H3K27 acetylation at the ITPR3 promoter region 53 . In addition, many microRNAs are shear stressregulated 54 , perhaps most notably miR-92a, that targets KLF2 and KLF4 under conditions of disturbed flow 55,56 . More recently, long non-coding RNAs (lncRNAs) have been studied as epigenetic regulators of shear stress-induced gene expression, leading to the discovery of splicedtranscript endothelial-enriched lncRNA (STEEL) 57 and a lncRNA that enhances NOS3 expression (LEENE) 58 .
Yet despite these advances, no studies have comprehensively mapped the endothelial regulatory landscape under physiologic LSS. While transcription factor binding to gene promoters is sufficient to drive basal levels of transcription 59 , activation of enhancers is required for full transcriptional activity 60 and to ensure phenotypic robustness 61 . In cultured cells, persistent changes in the epigenome may be particularly revealing of mechanisms that converge to cause disease 62 .
To investigate LSS-induced changes to the chromatin landscape, we performed ATAC-Seq on PAEC that were exposed to physiologic LSS of 15 dyn/cm 2 for 24 h. Only a small subset of DAR were located in gene promoter regions, the majority being putative regulatory elements in introns or intergenic sites. For KLF gain-of-function studies, and to have sufficient cell numbers for ChIP-Seq analyses, we used PAEC that were transduced with a vector encoding a constitutively active mutant of MEK5, resulting in activation of ERK5 and induction of KLF2 and KLF4 via MEF2 26,27 . KLF2 and KLF4 expression levels were induced to physiologic levels, i.e., similar to those observed in wildtype cells exposed to LSS. Although we cannot rule out KLF-independent effects resulting from ERK5 activation, such as through induction of NF-E2-related Factor 2 (Nrf2) 63 or direct effects of MEF2 activation, motif enrichment analyses of wildtype cells exposed to LSS revealed that MEF2 motifs were not in the top 20 most enriched motifs, with 17% of DAR having predicted MEF2 binding sites. We did not find enrichment for Nrf2 binding sites in DAR under LSS, corroborating the role of KLF in mediating the changes in accessibility. We focused our analyses on KLF4-occupied DAR that were common to wildtype PAEC exposed to LSS and those transduced with caMEK5. While recent technological advances, e.g., CUT&RUN 64 , allow for lower cell inputs to study protein-chromatin interactions based on histone modifications, these strategies are generally less comprehensive when studying transcription factor binding.
ChIP-Seq analyses of H3K4me1 and H3K27ac confirmed that most of these DAR have a chromatin signature typical of enhancers. Motif enrichment analysis of DAR with increased accessibility under LSS pinpointed the KLF transcription factor family as potential regulators of the accessibility changes. KLF4, in particular, has gained a lot of interest, being one of the four original reprogramming factors used by Takahashi and Yamanaka to create induced pluripotent stem cells (iPSC) 65 . This finding initiated further investigations into its role as a pioneer factor. Upon initial engagement of closed chromatin, pioneer factors recruit chromatin remodelers to allow stabilization of accessible chromatin, enabling other transcription factors to bind alongside them and recruit transcriptional machinery 66 . In vitro, purified KLF4 binds nucleosomes, and in vivo KLF4 preferentially targets closed the relation between differential KLF4 binding at enhancer anchors identified by HiChIP, with target gene expression, in PAEC exposed to 15  In the induction of pluripotency, KLF4 predominantly binds distal enhancers of genes that promote reprogramming, unless it partners with c-myc, in which case it tends to bind promoter regions 68 . This was confirmed in our data, in which 70.3% of KLF4 binding occurs outside of gene promoters that are within 3 Kb of transcription start sites. It would be interesting to identify which transcriptional co-factors guide KLF4 to either distal enhancers or promoter regions. Given the enrichment for the ETS motifs in DAR, and the well-established roles for ETS factors in regulation of endothelial gene expression 23,24 , we speculate that distinct ETS family members may play important roles in guiding these processes. At sites that were accessible under both LSS and ST conditions, a change in chromatin accessibility may not be a prerequisite for KLF to increase enhancer activity. Conversely, while the majority of DAR depend on KLF to increase accessibility, additional factors may be required for enhancer activation.
Besides DNA methylation and histone modifications, ATPdependent chromatin remodeling processes such as the mammalian SWI/SNF complex, are important regulators of chromatin accessibility by disrupting histone-DNA contacts 69 . The SWI/SNF remodeling complex can assist transcriptional activators to access closed chromatin [70][71][72] or facilitate subsequent transcriptional activation 73 . For example, in mouse embryonic stem cell differentiation, the pioneer factor Forkhead Box A2 (FOXA2) binds regions of closed chromatin and recruits SWI/SNF to regulate nucleosome depletion and to promote endodermal differentiation 74 . More recently, BRG1, one of two mutually exclusive ATPases that functions as the catalytic subunit of the SWI-SNF complex 75 , was related to the reprogramming of iPSCs by increasing the accessibility of pluripotency enhancers 76 . Alver and colleagues described SWI/SNF as required for maintenance of lineagespecific enhancers 77 .
While BRG1 co-localizes with pluripotency factors OCT4, SOX2, and NANOG in embryonic stem cells 78 , to our knowledge there have been no prior reports of KLF4 and SWI/SNF co-localization or interaction, other than a high throughput lentiviral proteomics study that identified an interaction between KLF4 and Brahma (BRM) which is closely related to BRG1, but mutually exclusive 79 . Because BRG1 is not regulated by shear stress, we speculate that under conditions of disturbed shear stress BRG1 might interact with AP1 family members, as we found enrichment for these motifs in DAR that are losing accessibility with LSS.
Since physical distance is an important indicator of the probability that a regulatory element targets a certain gene, we initially related DAR to their nearest gene. While accessibility changes generally correlated with gene expression changes, this could only explain a subset of gene regulation. The seminal work by Dekker and colleagues, who first described chromosome confirmation capture (3C) 80 , initiated further development of 3C-derived methods, including landmark studies applying Hi-C 81,82 . This greatly improved our understanding of the 3D genome and helped explain how regulatory elements that are megabases away from their target promoter region in the linear genome are in fact in close physical proximity. To map the endothelial enhancer landscape and study the contribution of KLF4 and BRG1 in regulating enhancer activity and target gene expression, we applied two distinct but complementary strategies to study enhancerpromoter (EP) looping, i.e., H3K27ac HiChIP 19,20 and the Activity-by-Contact model (ABC) 21 . The experimental H3K27ac HiChIP approach detects the frequency of 3D contacts and identified many long-range EP loops that often span genomic regions containing multiple genes. The computational ABC algorithm, that was extensively validated by thousands of CRISPRi perturbations 21 , confirmed most of the EP loops that were identified by HiChIP and discovered many additional EP loops. ABC predicted more than 70% of differentially expressed genes to be regulated by KLF4-bound enhancers, reinforcing the pivotal role of KLF4 as the mediator of the vasculoprotective effect of LSS. The importance of KLF4 in regulating the endothelial enhancer landscape is further supported by a recent study that uncovered a critical role for KLF4 in organizing the pluripotency-associated enhancer network 83 . Here we show, using KLF loss-of-function by siRNA, that a large subset of EP loops is indeed KLF dependent. We applied CRISPRi to functionally validate candidate enhancers that we found are regulated by LSS, and are particularly relevant to pulmonary vascular homeostasis and PAH, i.e., BMPR2, SMAD5, and EDN1. We observed KLF4 binding at enhancers of other genes associated with PAH, such as ACVRL1, CAV1, ENG, SMAD9, and SOX17, but their expression levels did not significantly change under LSS. Additional genetic and/or environmental factors can contribute to the altered regulation of these genes, and more extensive whole genome sequencing studies may reveal mutations in binding sites for other transcription factors that might act in concert with KLF.
In most cases, KLF4 binding at enhancers resulted in increased target gene expression. However, in some, such as the EP loops targeting endothelin-1, KLF4 binding resulted in reduced H3K27ac and target gene expression, possibly by recruiting histone deacetylases such as HDAC2 that we identified to interact with KLF4 by AP-MS, and that others have shown to interact with KLF4 in smooth muscle cells 32 . These opposing changes in gene regulation reflect the diversity of remodelers that we identified as KLF4 interactors and warrant further study to understand how KLF4 binding results in different outcomes at different sites.
Our findings can be leveraged to better relate genetic variants in non-coding regions identified by genome-wide association studies (GWAS) to protective or pathogenic gene expression. An example that we described is a variant in the NOS3 promotor that may target AGAP3. Most single nucleotide polymorphisms (SNPs) identified by GWAS occur in non-coding regions 35 , and individually often contribute small or indirect effects to complex traits. Knowing the endothelial enhancer landscape under physiologic LSS will facilitate linking variants identified by GWAS to specific genes, and to uncover other genes that have not been related to cardiovascular disease before, especially in rare conditions such as PAH. Future studies applying CRISPRi or base editing approaches to target SNPs in candidate enhancers, or broader approaches using pooled CRISPR screens combined with single cell RNA-Seq, such as Perturb-Seq 84 , are needed to validate the impact of these non-coding variants on the genes that they regulate, and to determine how this may contribute to the prevention or progression of disease.

Ethical approval
Procedures were compliant with all ethical regulations regarding animal research. Care and housing of the rats was in accordance with the guidelines from the Stanford University Administrative Panel on Laboratory Animal Care and approved under APLAC protocol 31608. Human cells provided by the PHBI Initiative were obtained under the PHBI network IRB protocol, with informed consent and IRB approvals at the transplant procurement sites. The cell lines used were coded with no identifying information and therefore considered non-human subject research for the purposes of this study. Cells purchased from PromoCell were derived from tissues of donors who have signed an informed consent form. In both cases, the informed consent outlines in detail the purpose of the donation and the procedure for processing the tissue.

Cell culture studies
Primary human pulmonary artery endothelial cells (PAEC) were either commercially obtained (PromoCell) or harvested from unused donor control lungs obtained though the Pulmonary Hypertension Breakthrough Initiative (PHBI) funded by NIH (R24 HL123767) and the Cardiovascular Medical Research and Education Fund (CMREF; UL 1RR024986). PAEC were grown in commercial EC media containing 5% FBS (Sciencell) in a 5% CO 2 air atmosphere and used at passages 3-7. Cells were routinely tested for mycoplasma contamination. For shear stress experiments, PAEC were seeded in flow chamber slides (µ-Slide I 0.4 mm ibiTreat; Ibidi) and grown to confluence before exposure to 15 dyn/cm 2 of unidirectional uniform laminar shear stress (LSS) for 24 h. LSS was generated using the Ibidi Perfusion System (Ibidi). Static controls were performed simultaneously with the shear stress experiments, cultured on standard tissue culture treated plates or dishes using the same EC media (Sciencell), to avoid nutrient deprivation that can occur in flow chamber slides that are not perfused.

Plasmids
PAEC were transduced with adenoviral constructs encoding a constitutive active mutant of dual specificity mitogen-activated protein kinase 5 (caMEK5) (#000101 A, Applied Biological Materials Inc); Flagtagged KLF4 (#VH829440, Vigene Biosciences) or GFP control (AVP004, GenTarget Inc) for 12 h after which cells were allowed to recover for 90 h before being harvested for subsequent experiments. For CRISPRi studies, PAEC were transduced with a lentiviral vector encoding a doxycylin-inducible mutant of a catalytically dead Cas9 fused to the bipartite repressor domain MeCP2-KRAB (Addgene Plasmid #122205) for 12 h and allowed to recover for 90 h followed by selection with blasticidin (10 µg/ml).

ATAC-seq sample preparation and data analysis
ATAC-seq was performed as described in Buenrostro et al. 85 . Briefly, endothelial cells were trypsinized to create a single-cell suspension. After counting, nuclei were isolated from 100,000 cells and sequencing adapters were transposed for 30 min at 37°C using 5 μl of TDE1 (Nextera Tn5 transposase, Illumina). After PCR and gel purification, libraries were subjected to 2 × 151 paired end sequencing on the Illumina HiSeq 4000 to obtain an average of approximately 50 million uniquely mapped reads per sample (Stanford Center for Genomics and Personalized Medicine, supported by NIH grant S10OD020141). The resulting data were processed using the Kundaje Lab ATAC-seq processing pipeline (https://github.com/kundajelab/atac_dnase_ pipelines). Briefly, this pipeline takes FASTQ files as input, and outputs peak calls (accessible regions, AR). Alignments to AR were counted using DiffBind v2.4.8 (https://rdrr.io/bioc/DiffBind/) to produce a count matrix. DAR were detected using DESeq2 (https:// bioconductor.org/packages/release/bioc/html/DESeq2.html) with a Pvalue cutoff of <0.1. The HOMER (http://homer.salk.edu/homer/) function findMotifsGenome was used with default parameters to search for motif enrichment in the full accessible regions.

RNA-seq sample preparation and data analysis
RNA was extracted using the RNeasy Mini Kit (#74136, Qiagen). Libraries were prepared using TruSeq Stranded Total RNA Library Prep Kit with Ribo-Zero Gold (Illumina) for experiments with PAEC exposed to LSS vs ST, caMEK5 vs GFP, and KLF2/4 RNAi studies, and using QuantSeq 3′ mRNA-Seq Library Prep Kit REV for Illumina (#016.96, Lexogen) for the DUSP5 RNAi study. Sequencing on the Illumina HiSeq 4000 yielded an average of approximately 8 million uniquely mapped reads for total RNA-Seq, and 7 million uniquely mapped reads per sample for mRNA-Seq (Supplementary Table 1; Stanford Center for Genomics and Personalized Medicine, supported by NIH grant S10OD020141). The resulting data were aligned to the human genome (GRCh37.p13) by STAR v.2.5.4b (https://github.com/alexdobin/STAR). The aligned transcripts were quantitated based on features in the GENCODE annotation database (GRCh37, version 19) by RSEM v. 1.3.1 (http://deweylab.biostat.wisc.edu/rsem/). Differentially expressed genes were detected using DESeq2 v. 1.20.0 (https://bioconductor.org/ packages/release/bioc/html/DESeq2.html) with a P-value cutoff of <0.1. Functional enrichment for the differentially expressed genes was performed using Metascape.

ChIP-seq sample preparation and data analysis
For KLF4, H3K4me1, and H3K27ac ChIP-Seq, cells were trypsinized and cross-linked with 1% formaldehyde (EMD Millipore) for 10 min at RT. To quench the formaldehyde, 2 M glycine (ThermoFisher Scientific) was added and incubated for 5 min at room temperature. For BRG1 ChIP-Seq, cells were first cross-linked using 2 mM of DSG (Pierce) for 45 min at RT, washed with PBS, and then cross-linked with 1% formaldehyde as described above. Cells were washed with ice cold PBS twice, snapfrozen and stored at −80°C. For ChIP-DNA preparation, cells were thawed by adding PBS and incubated at 4°C with rotation. Cells were treated with hypotonic buffer (20 mM HEPES pH 7.9, 10 mM KCl, 1 mM EDTA pH 8.0,10% glycerol) for 10 min on ice in the presence of protease inhibitors (G6521, Promega), then were homogenized using a glass homogenizer. Nuclear pellets were resuspended in RIPA buffer (Millipore) and incubated for 30 min on ice. Chromatin corresponding to 20 million cells for transcription factors, or 5 million cells for histone modifications was sheared with SFX250 Sonifier (Branson) and immunoprecipitated with antibodies targeting H3K27ac (#8173, Cell signaling Technology), H3K4me1 (#5326, Cell Signaling Technology), KLF4 (sc20691, Santa Cruz Biotechnology) and BRG1 (A303-877A, Bethyl Laboratories); 5 μg for each condition, at 4°C overnight on a nutator. For the input sample, 100 μl of sheared nuclear lysate was removed and stored overnight at 4°C. The next day, protein A/G agarose beads (Millipore) were added to the chromatin-antibody complex and incubated for one hour at 4°C on a nutator, after which the beads were eluted with SDS buffer (Santa Cruz Biotechnology) and incubated at 65°C for 10 min. Supernatant containing ChIP-DNA was reverse crosslinked by incubating overnight at 65°C. On the third day, ChIP-DNA was treated with RNase A (Qiagen) and proteinase K (Ther-moFisher Scientific) and then purified. The ChIP-DNA samples were end repaired using End-It DNA End Repair Kit (Lucigen) and A-tailed using Klenow Fragment and dATP (New England Biolabs). Illumina TruSeq adapters (Illumina) were ligated using LigaFast (#M8221, Promega) and size-selected by gel extraction before PCR amplification. The purified libraries were subjected to 2 × 151 paired end sequencing on the Illumina HiSeq 4000 to obtain an average of approximately 37 million uniquely mapped reads for each sample (Supplementary Table 1; Stanford Center for Genomics and Personalized Medicine, supported by NIH grant S10OD020141). The resulting data were processed using the Kundaje Lab ChIP-seq processing pipeline (https:// github.com/kundajelab/chipseq_pipeline). Briefly, this pipeline takes FASTQ files as input and outputs peak calls (bound regions; BR). Alignments to BR were counted using DiffBind 2.4.8 (https://rdrr.io/ bioc/DiffBind/) to produce a count matrix. Differentially bound regions (DBR) were detected using DESeq2 (https://bioconductor.org/ packages/release/bioc/html/DESeq2.html) with a P-value cutoff of <0.1. Data visualization was performed using the Integrative Genomics Viewer (http://www.broadinstitute.org/igv).
Immunofluorescence PAEC were cultured on flow slides (µ-Slide I 0.4 mm ibiTreat; Ibidi), washed with PBS, and fixed with ice-cold methanol at −20°C for 30 min. Methanol was aspirated and the slides were rehydrated with PBS at room temperature for 10 min. After washing with PBS, slides were blocked with 5% normal donkey serum and 2% BSA (Sigma Aldrich) in PBS at room temperature for 1 h. Incubation with primary antibodies targeting KLF4 (1:100, sc20691, Santa Cruz Biotechnology), ATF2 (1:100 sc-242, Santa Cruz Biotechnology) and ETS1 (1:100, sc55581, Santa Cruz Biotechnology) were carried out in the blocking buffer at 4°C overnight. Secondary antibody incubations were performed using 1:600 dilutions of A-21206 (for KLF4); A-21203 (for ATF2) and A-31571 (for ETS1), all from ThermoFisher Scientific, in the blocking buffer at room temperature for 1 h. Slides were mounted with DAPI Fluoromount-G (DAPI, 4,6-diamidino-2-phenylindole) (SouthernBiotech). Stained slides were imaged using Leica Application Suite X software on a Leica Sp8 (Leica). Quantification of the nuclear fluorescence intensities was performed using ImageJ.

Reverse Transcription (RT-) and ATAC-qPCR
For RT-qPCR, total RNA was extracted and purified using the Quick-RNA MiniPrep Kit (Zymo Research). The quantity and quality of RNA was determined using a spectrophotometer. RNA was reverse transcribed using the High-Capacity RNA to cDNA Kit (Applied Biosystems) according to the manufacturer's instructions. For ATAC-qPCR, cells were processed as described above for ATAC-Seq. For qPCR, nuclei were isolated from 50,000 cells and sequencing adapters were transposed for 30 min at 37°C using 2.5 μl of TDE1 (Nextera Tn5 transposase, Illumina). The reaction was terminated and purified using MinElute reaction cleanup (Qiagen) and used as template. In both cases, qPCR was performed using 1 μl of 5 μM Powerup SYBR green PCR Master Mix (Applied Biosystems), 2 μl of dH 2 O and 2 μL of cDNA sample in a 10 μl reaction. Each measurement was carried out in a duplicate using a CFX384 Real-Time System (Bio-Rad). The PCR conditions were: 95°C for 2 min, followed by 40 cycles of 95°C for 15 s, and 60°C for 60 s. For ATAC-qPCR, primers were designed to flank the center of the DAR as determined by ATAC-Seq. Primer sequences used are listed in Supplementary Table 2. Gene expression levels were normalized to β-actin, and accessibility changes were normalized to GAPDH.

Affinity purification followed by mass spectrometry (AP-MS)
Nuclear fractionation and AP were performed using the Nuclear Complex Co-IP kit (ActiveMotif) according to manufacturer's instructions. Briefly, cells were trypsinized and washed with cold PBS. The cell pellets were resuspended in hypotonic buffer and incubated on ice for 15 min. Detergent was added and the suspension was centrifuged at for 30 s at 14,000 × g in a pre-cooled centrifuge. The nuclear pellet was resuspended in complete digestion buffer with 0.75 μl of enzymatic shearing cocktail and incubated at 37°C for 10 min. 0.5 M EDTA (3 μl) was added to stop the reaction. The suspension was placed on ice for 5 min and then centrifuged for 10 min at 14,000 × g. For AP, the supernatant was pre-cleared by adding 30 μl of Dynabeads Protein G (10004D, ThermoFisher Scientific) and incubated on a rotator at 4°C for 1 h. The beads were removed and the supernatant was used for subsequent AP. Antibodies (2 μg) targeting either KLF (sc166238, Santa Cruz Biotechnology) or FLAG (F7425, Millipore Sigma) were added per 1 mg of pre-cleared supernatant and incubated on a rotator at 4°C overnight. The next day, samples were incubated with 30 μl of Dynabeads Protein G (10004D, ThermoFisher Scientific) on a rotator at 4°C for 1 h. After three successive washes with ice-cold washing buffer, the proteins were eluted from the beads using 100 μl of IgG Elution Buffer (Thermo-Fisher Scientific) at a gentle vortex at room temperature (RT) for 7 min. The eluate was then immediately neutralized with 1:10 1 M Tris-HCl, pH 8.5. For MS analysis, samples were reduced with 5 mM DTT in 120 μl of 50 mM ammonium bicarbonate. Following reduction, proteins were alkylated using 10 mM acrylamide for 30 min at room temperature to cap cysteines. Digestion was performed using Trypsin/LysC (Promega) overnight at 37°C. Following digestion and acid quenching, samples were passed over HILIC resin (Resyn Biosciences), dried in a speed vac and then reconstituted in 10 μl reconstitution buffer (2% acetonitrile with 0.1% Formic acid); 3 μl of the reconstituted peptides were injected on the instrument.
All mass spectrometry experiments were performed using an Orbitrap Fusion Tribrid mass spectrometer (ThermoFisher Scientific) with an attach Acquity M-Class UPLC (Waters Corporation) liquid chromatograph. A pulled-and-packed fused silica C18 reverse phase column containing Dr. Maisch 1.8 μm C18 beads and a length of~25 cm was used over a 80 min gradient. A flow rate of 300 nL/min was used with the mobile phase A consisting of aqueous 0.2% formic acid and mobile phase B consisting of 0.2% formic acid in acetonitrile. Peptides were directly injected onto the analytical column. The mass spectrometer was operated in a data dependent fashion, with MS1 survey spectra collected in the Orbitrap and MS2 fragmentation using CID for in the ion trap.
For data analysis, the raw data files were processed using Byonic v2.14.27 (Protein Metrics) to identify peptides and infer proteins against the human UniProt database containing isoforms concatenated with synthesized sequences. Proteolysis was assumed to be tryptic in nature and allowed for ragged n-terminal digestion and up to two missed cleavage sites. Precursor mass accuracies were held within 12 ppm, with MS/MS fragments held to a 0.4 Da mass accuracy. Proteins were held to a false discovery rate of 1%, using standard approaches.

Proximity ligation assays (PLA)
PLA in cultured PAEC were performed using Duolink PLA protein detection technology (Millipore Sigma) according to manufacturer's instructions. Briefly, slides with PAEC were crosslinked with 4% PFA for 10 min, washed and incubated with Duolink Blocking solution for 1 h at 37°C. Slides were incubated with rabbit antibodies targeting KLF4 (1:100, sc20691, Santa Cruz Biotechnology) and mouse antibodies targeting BRG1 (1:75, sc17796, Santa Cruz Biotechnology) or SMARCC2 (1:75, sc17838, Santa Cruz Biotechnology) overnight at 4°C. Rabbit and mouse IgG were used as controls. The next day, slides were washed, and then incubated with the Duolink PLUS and MINUS probes for 1 h at 37°C. After washing the slides, the probes were ligated for 30 min at 37°C. Slides were washed and rolling circle amplification was performed for 100 min at 37°C, after which slides underwent final washes and were mounted using Duolink in situ mounting medium with DAPI (Millipore Sigma). Stained slides were imaged using Leica Application Suite X software on a Leica Sp8 (Leica). Quantification of the average number of foci per nucleus was performed using ImageJ.
For PLA in rat lung tissue, the protocol was modified using longer incubation times to allow reagents to fully penetrate the tissues. Healthy untreated 8-week-old Sprague Dawley rats (n = 2) were sacrificed, and lungs were flushed and perfusion fixed with 4% PFA for 30 min on ice, washed with PBS, and dehydrated in methanol. Following rehydration, tissue was cut in 350 μm thick sections. Sections were bleached using 3% H 2 O 2 for 1 h at RT, washed with 0.5% triton-PBS and blocked using 5% DS 5% BSA in 0.5% triton-PBS for 3 h at RT. Sections were incubated with primary antibodies targeting KLF4 (1:50, sc20691, Santa Cruz Biotechnology) and BRG1 (1:30, sc17796, Santa Cruz Biotechnology) in 3% DS 1% BSA in 0.1% triton-PBS overnight at 4°C. The following day, sections were washed, blocked using 3% DS 1% BSA in 0.1% triton-PBS (dilution buffer) for 4 h at RT, after which they were incubated with the Duolink PLUS and MINUS probes overnight at 4°C. On the next day, sections were washed, and the probes were ligated for 2 h at 37°C. After washing, rolling circle amplification was performed for 3 h at 37°C, after which slides underwent final washes. Sections were post-fixed using 4% PFA for 25 min at RT, washed, and blocked with anti-Rabbit IgG in dilution buffer at RT. Sections were then incubated with FITC-conjugated anti-αSMA antibodies (1:400, F377, Sigma-Aldrich) and antibodies targeting vWF (1:500, ab6994, Abcam) overnight at 4°C. After washing, sections were incubated with anti-Rabbit Alexa Fluor 647-conjugated secondary antibodies (1:200, A32795, Thermo Fisher Scientific) for 3 h at RT, washed in triton-PBS with DAPI for 2 h at RT, and post-fixed with 4% PFA for 30 min at RT. Finally, sections were washed, dehydrated in methanol, and cleared using Benzyl Alcohol/ Benzyl Benzoate (BABB). Sections were imaged using Leica Application Suite X software on a Leica Sp8 (Leica). Threedimensional reconstructions were made with Imaris version 9.3.0 (Bitplane).

H3K27ac HiC chromatin immunoprecipitation (HiChIP)
H3K27ac HiChIP was performed as previously described 19 . PAEC were crosslinked in 1% formaldehyde for 10 min at room temperature and then quenched by 125 mM Glycine for 5 min at RT. Nuclei were isolated from 1 million crosslinked cells by 30 min of lysis at 4°C. Nuclei were permeabilized in 0.5% SDS for 10 min at 62°C and quenched using Triton X-100 for 15 min at 37°C. MboI restriction enzyme (R0147, New England Biolabs) was added to digest chromatin for 2 h at 37°C and then heat-inactivated for 20 min at 62°C. Klenow was then used to fill in restriction fragment overhangs and mark the DNA ends with biotin (M0210, New England Biolabs). Proximity ligation contact (PLC) pellets were then created by incubation with DNA ligase for 4 h at room temperature followed by centrifugation. PLC pellets were then sonicated and immunoprecipitated using 5 μg of H3K27ac antibodies (#8173, Cell signaling Technology) as previously described. The eluted fragments labeled by biotin were then captured by streptavidin bead pull-down. DNA was then adaptor-labeled using Tn5 transposase (Illumina) and subjected to PCR amplification. Samples were then sequenced by 2 × 101 paired-end sequencing on an Illumina NovaSeq 6000 to an average yield of 200 million reads per sample (Supplementary Table 1). The resulting data were filtered for duplicate reads, aligned to the hg19 genome, and filtered for valid interactions using the HiC-Pro pipeline v.2.11.1 (https://github.com/nservant/HiC-Pro) using the default settings. FitHiChIP (https://github.com/ay-lab/ FitHiChIP) was used to determine statistically significant interactions using default settings, with the exception of allowing interactions with a minimum size of 1Kb. The diffLoop package (https://github.com/ aryeelab/diffloop) was used to test for differential interactions between conditions, and to infer gene enhancer to promoter relationships.

Activity-by-contact model (ABC)
The ABC v0.2 pipeline was cloned from the GitHub repository (https:// github.com/broadinstitute/ABC-Enhancer-Gene-Prediction/). First, ATAC peaks were called by MACS2 v.2.1.2 (https://github.com/taoliu/ MACS) from each ATAC BAM file with a P-value cutoff of 0.1. Candidate enhancer regions were then defined by the ABC script makeCandida-teRegions.py, which: (1) Resized each peak to be 250 base pairs centered on the peak summit. (2) Counted ATAC-seq reads in each peak and retained the top 150,000 peaks with the most read counts. (3) Removed any regions that are blacklisted due to known propensity for errors (hg19-blacklist.v2.bed from https://github.com/Boyle-Lab/ Blacklist) and (4), merged any overlapping regions. Enhancer activity was then quantified by the ABC script run.neighborhoods.py, which counted ATAC-seq and H3K27ac ChIP-seq reads in the candidate enhancer regions that were generated in the previous step, gene bodies, and promoter regions. Lastly, the ABC score was calculated using the ABC script predict.py; which combined information from the enhancer and promoter activities, calculated in the previous step, with contact frequency data from average Hi-C profiles of 10 cell lines (ftp:// ftp.broadinstitute.org/outgoing/lincRNA/average_hic/average_hic.v2. 191020.tar.gz. The default threshold of 0.02 was applied which corresponds to approximately 70% recall and 60% precision 21 .

CRISPR-interference
PAEC were transduced with a lentiviral vector encoding a doxycylininducible mutant of a catalytically dead Cas9 fused to the bipartite repressor domain MeCP2-KRAB, as described above. Cells were treated with 500 ng/μl of doxycycline 48 h prior to nucleofection of synthetic guide RNAs (sgRNA) (Synthego) using the Basic Nucleofector Kit for Primary Mammalian Endothelial Cells (Lonza). sgRNA were designed to target the KLF binding sites in the candidate enhancers using the GPP sgRNA Design tool from the Broad Institute (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrnadesign). The sequences used are listed in Supplementary Table 2. Cells were allowed to recover for 24 h and then exposed to 24 h of LSS, after which target gene expression was assessed by RT-qPCR.

Statistical analysis
Values from multiple experiments are shown as arithmetical mean ± SEM. Statistical significance was determined using unpaired two-tailed Student's t test. Correlations were calculated by Pearson R test. A Pvalue of <0.05 was considered significant. The number of samples in each group, the statistical test used, and the statistical significance is indicated in the figures. Data were analyzed using Prism version 8.4 (Graphpad).

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request. ATAC-Seq, RNA-Seq, ChIP-seq and HiChIP data are deposited in the Gene Expression Omnibus (GEO)