BRD2 impedes iPSC reprogramming by regulating lipid biosynthesis and the matrisome via its ET tail

Abstract

Induction of human pluripotent stem cells (hiPSCs) from somatic cells encounters significant barriers that remain poorly understood. Lipids are fundamental cellular molecules with many essential roles. However, the roles of lipids in reprogramming are unknown. Here, we show that BRD2 is a barrier to hiPSC reprogramming by suppressing lipid genes and maintaining the somatic transcriptional program, particularly the somatic matrisome program. Strikingly, the acetylation epigenetic reader BRD2 unorthodoxly suppresses the expression of lipid-related genes in the reprogramming cells. The two rate-limiting enzymes of lipid biosynthesis, stearoyl-CoA desaturase (SCD), 3-hydroxy-3-methyl-glutaryl-coenzyme A reductase (HMGCR), and lipid supplements all enhance iPSC reprogramming. Interestingly, the extra terminal (ET) tail of BRD2 suppresses reprogramming and the lipid transcriptional program but positively regulates the matrisome program. In line with its transcriptional suppression of lipid genes, BRD2 binds to genes of lipid biosynthesis and negatively regulates their H3K27Ac status. These discoveries advance our molecular understanding of hiPSC reprogramming, as well as of the transcriptional regulation of matrisome and lipid programs.

BRD2, via its ET tail, impairs iPSC generation in early reprogramming by maintaining the somatic transcription, especially, the somatic matrisome program, and by suppressing lipid biosynthesis. Lipid deficiency is a barrier to reprogramming.

Introduction

Human somatic cells can be converted into embryonic stem cell- (ESC-)like cells by the Yamanaka factors, OCT4, SOX2, KLF4 and MYC (collectively OSKM)^1–5. Such factor-induced pluripotent stem cells (iPSCs) share molecular and functional characteristics with ESCs and are valuable resources for basic research, regenerative medicine, drug screening, disease modeling, and others^4,6,7. Reprogramming towards hiPSCs, however, is very inefficient, slow, inconsistent, and stochastic with significant barriers^5,8. Past investigations have revealed many reprogramming barriers such as DNA methylation, somatic transcription factors (TF), and others. Our molecular understanding of iPSC reprogramming, particularly its molecular barriers, remains limited. It is well-known that the culture of human pluripotent stem cells (hPSCs) uniquely requires support of feeder cells^7,9 or protein matrix precoated on the culture vessels¹⁰, indicating that hPSCs have a unique matrix biology¹¹, and thus matrisome reprogramming may be critical for hiPSC generation.

Bromodomain extra terminal (BET) proteins are characterized by two N-terminal tandem bromodomains (BD) and one extra terminal (ET) domain in their C-terminal tails. BET proteins bind to acetylated lysine in histones via their bromodomains and thus regulate transcription¹². Although BET proteins are known as positive regulators of transcription, their suppression of transcription has also been observed^13,14. Mammalian BET proteins include BRD2, BRD3, BRD4, and BRDt, with the first three being the ubiquitous members expressed in almost all cell types. BET proteins regulate a variety of cellular and physiological processes including transcription, cell cycle, tumorigenesis, inflammation, diabetes, stemness, and others^15,16. They are also implicated in cellular reprogramming^13,17–22, but the exact underlying molecular mechanisms remain elusive. Furthermore, the roles of each member are not defined. BRD4 enhanced mouse iPSC reprogramming in one report²⁰ but another group showed that BRD4 inhibition facilitated it²¹. We previously reported that mild pharmaceutical BET inhibition with the pan BET inhibitors enhanced hiPSC generation and dampened somatic transcription¹³. In line with our observation, Li et al. showed that chemical BET inhibition promoted mouse fibroblast reprogramming to functional neurons likely by suppressing somatic transcription¹⁹.

Lipids are one of the four fundamental cellular molecules along with proteins, nucleic acids, and carbohydrates. Humans may have >800 species of lipids²³. They are basic cellular structural molecules essential for all biological membranes. Lipids are also cellular energy sources, nutrients, hormones, and signaling molecules, as well as a chemical group for protein covalent modification (lipidation)^24–26. The complex lipid biosynthesis involves multiple pathways and many enzymes. Of those, stearoyl-CoA desaturase (SCD) catalyzes the generation of the first double bond as the rate-limiting step in the fundamental fatty acid (FA) biosynthesis²⁷. At the same time, HMG-CoA reductase (HMGCR) catalyzes the rate-limiting step of cholesterol biosynthesis²⁸. De novo lipogenesis is a characteristic feature of the naïve mammalian PSCs, and of the traditional hPSCs cultured in the chemically defined E8 media^29,30, implying a need for lipogenesis reprogramming as part of the pluripotency establishment. However, the roles of lipids and their biosynthesis in iPSC reprogramming are not clear.

Here, we report that BRD2 is a major hiPSC reprogramming barrier. We first showed that BRD2 may impair iPSC reprogramming by maintaining somatic transcription, particularly for the genes of matrisome and its regulation. Surprisingly, we discovered that BRD2, known as a positive transcriptional regulator, unconventionally suppresses genes in lipid biosynthesis and metabolism likely by direct binding to those genes and regulating their epigenetic status. Critically, we showed that the key enzymes of lipid biosynthesis promote iPSC reprogramming and that lipid supplements in the reprogramming media can overcome lipid deficiency during reprogramming.

Results

Human BRD2 is an early barrier of iPSC reprogramming

We previously reported that inhibition of BET proteins with the pan-BET chemical inhibitors at the widely used concentrations (e.g., ≥500 nM for JQ1) had little impact on or even impaired hiPSC reprogramming¹⁷, but their mild pharmacological inhibition improved it (e.g., 50 nM for JQ1)¹³. The present study confirmed our previous observations (Supplementary Fig. 1 and Supplementary Data 3). The mammalian BET family has three ubiquitously expressed members, BRD2, BRD3 and BRD4, all of which are embryonic lethal when knocked out (KO) individually (BRD3 embryonic lethality is based on our unpublished observation)^31,32, indicating their essential but distinct roles. The current BD-targeting BET inhibitors do not distinguish individual members³³. Given the above facts, we asked whether mild BET chemical inhibition targets a specific BET member to promote hiPSC reprogramming. We first established two efficient shRNAs against each human BET member (sheet “shRNA” of Supplementary Data 2, Fig. 1b and Supplementary Figs. 2 and 7b). Those knockdowns did not cause compensation from member species as estimated by western blot (Supplementary Fig. 2) and RNA-seq (Supplementary Fig. 7c–e). Reprogramming experiments (Fig. 1a) showed that BRD3 and BRD4 shRNAs compromised hiPSC generation under the OSKM condition (Fig. 1c, d and Supplementary Data 3). In contrast, both shRNAs against BRD2 significantly enhanced hiPSC reprogramming (Fig. 1c, d). Similar results were observed under the Yamanaka 3-factor (OCT4, SOX2, and KLF4, collectively OSK) condition (Fig. 1e). We verified these observations using additional pluripotency markers (TRA-1-60 and NANOG, Fig. 1e, f, h). In agreement with BRD2 shRNA knockdown, BRD2 overexpression (Supplementary Figs. 7a, 14 and 15) significantly decreased the reprogramming while overexpression of BRD3 and BRD4S had little impact on reprogramming (Fig. 1g).

Fig. 1. Human BRD2 is an early reprogramming barrier.

a Schematic for evaluation of the reprogramming efficiency in the clinic-relevant chemically defined E8 medium. b Western blots on human fibroblasts with the lentiviral shRNAs against human BRD2, BRD3 and BRD4. Two shRNA vectors (1 and 2) were used for each BET member. Beta actin was probed as a protein loading control. Proteins were extracted at 48 h of knockdown. c Average total alkaline phosphatase positive (ALP⁺) colony counts of three independent experiments on day 20 of OSKM reprogramming using the shRNAs shown in (b). n = 3; **p < 0.01; **p < 0.001. d Representative images of ALP staining of experiments shown in (c). e Average colony counts of TRA-1-60 positive colonies in three independent OSK reprogramming experiments. f Representative images of colonies expressing NANOG on day 20 of OSK reprogramming experiments. g Average total ALP⁺ colony counts at day 20 of three independent OSK reprogramming experiments with the overexpression of BET proteins. h Percentage of TRA-1-60⁺ cells at days 5, 8 and 11 during reprogramming using a doxycycline (dox) inducible shBRD2 vector as measured by flow cytometry. i Fold changes of ALP⁺ colonies under different times of BRD2 knockdown relative to no knockdown control using the dox inducible shBRD2 vector. Dox was added at different time windows during the first 15 days of OSK reprogramming. n = 3 (independent experiments); *p < 0.05; Statistical tests are one-way ANNOVA test followed by post hoc Dunnett’s test in (c, e, g, i). Error bars are standard deviations in (c, e, g, i).

To corroborate the above results from shRNA knockdown and overexpression, we further used CRISPR constructs for human BRD2 KO and established a human fibroblast line with doxycycline (dox) inducible Cas9 (Supplementary Fig. 3a). Two combinations of the three single guide RNA (sgRNAs) (See sheet “sgRNA” of Supplementary Data 2) resulted in efficient disruption of BRD2 in the reprogramming starting cells (Supplementary Figs. 3b–d, and 10). With this system, we observed larger and more colonies before day 15 (day 12 to 15) (Supplementary Fig. 3e), in agreement with the results of BRD2 shRNA knockdown and overexpression. After this point, however, the hiPSC colonies started to differentiate, and we could not establish stable hiPSC lines from our BRD2 KO experiments. This latter observation indicates that BRD2 might be critical for the pluripotency maintenance of the already established iPSCs although it impairs reprogramming during the process of pluripotency establishment, especially in the early reprogramming (see more below).

We previously showed that pharmacological BET inhibition promotes iPSC reprogramming in the early stages¹³. We then designed inducible BRD2 shRNA knockdown (Supplementary Fig. 4a, f and Supplementary Data 2). Upon dox induction, BRD2 was efficiently knocked down (Supplementary Figs. 4b–d and 7b). Induced knockdown of BRD2 increased numbers of TRA-1-60⁺ and TRA-1-81⁺ cells on day 8 and 11 of reprogramming (Fig. 1h and Supplementary Fig. 4e). Late knockdown (day 11 to 15) had little impact on reprogramming; BRD2 knockdown in the middle and middle to late stages increased the colony number but did not pass the significance tests (Fig. 1i). Early knockdown (day 3 to 7), however, significantly increased reprogramming (Fig. 1i). Similar to chemical BET inhibition¹³, continuous knockdown from day 3 to 11 or 15 demonstrated the highest reprogramming activities (Fig. 1i). In summary, BRD2 is an early reprogramming barrier, and this barrier lingers for at least the first 15 days.

We established hiPSC lines with BRD2 shRNA (shBRD2iPSCs), although we could not do so with BRD2 KO. The shBRD2iPSCs are pluripotent based on pluripotency marker expression (Supplementary Fig. 5a, b)³⁴, silence of the reprogramming factors (Supplementary Fig. 5c and Fig. 4a)³⁵, teratoma formation (Supplementary Fig. 5d), and a pluripotency transcriptome (Supplementary Fig. 5f). The shBRD2iPSCs maintain a normal karyotype (Supplementary Fig. 5e).

Fig. 4. Rate-limiting enzymes of lipogenesis and lipid supplements overcome BRD2 impairment of reprogramming.

a Lentiviral plasmid map for GFP co-expression with the candidate genes via P2A peptide. b Screening of the reprogramming activity of SREBF-regulated metabolic genes. n = 4 independent experiments; *p < 0.05; one-way ANNOVA test followed by post hoc Dunnett’s test. The right side of the bar plots are representative ALP staining images of the reprogramming cells in 24-well plates. Each candidate gene was overexpressed via lentiviral transduction with a MOI of 3. c The HMGCR inhibitor lovastatin abrogated iPSC reprogramming (day 20, ALP staining). Bar plots are iPSC colony numbers for different conditions as indicated; right panels are representative images of the ALP-stained reprogramming wells of 12-well plates. n = 4 independent experiments; *p < 0.05, one-way ANNOVA test followed by post hoc Dunnett’s test. Lova, lovastatin. d A chemically defined lipid (CDL) cocktail enhanced iPSC reprogramming (day 20, ALP staining). Bar plots are ALP⁺ colony numbers for conditions indicated; right panels are representative images of the ALP-stained reprogramming wells of 12-well plates. n = 4 independent experiments, *p < 0.05; one-way ANNOVA test followed by post hoc Dunnett’s test. Error bars are standard deviations in (b–d).

BRD2 is the more abundant BET species in the starting cells

The mammalian BRD2, BRD3, and BRD4 are ubiquitous proteins, and the BD-targeting BET chemical inhibitors are pan-BET. Given that mild BET chemical inhibition enhanced hiPSC reprogramming (Supplementary Fig. 1 and Supplementary Data 3)¹³, we wondered why BRD2 shRNA knockdown enhanced hiPSC reprogramming, whereas knockdown of BRD3 and BRD4 did not. We first reviewed our RNA-seq data. As anticipated, our RNA-seq data showed no expression of the testis-specific BRDt in fibroblasts, PSCs, and the reprogramming cells (Supplementary Figs. 6a and 7e), but high normalized read counts for the three ubiquitous members (Supplementary Figs. 6a and 7b–d). Interestingly, BRD2 demonstrated the highest expression in fibroblasts, while BRD3 displayed the lowest expression. BRD4 expression is half that of BRD2. The same trend was observed in hPSCs. Our quantitative RT-PCR confirmed that BRD2 displayed the highest expression, while BRD3 showed the lowest expression (Supplementary Fig. 6b). Further ELISA quantification using BET protein standards showed similar results to those of RT-qPCR and RNA-seq (Supplementary Fig. 6c).

BRD2 shRNA legitimately down-regulates genes of matrisome and its regulation

To reveal the underlying molecular underpinnings of reprogramming enhancement by BRD2 knockdown, we conducted RNA sequencing (RNA-seq) on the reprogramming cells with BRD2 shRNA knockdown versus luciferase shRNA controls. In agreement with our western blots (Fig. 1b and Supplementary Fig. 2), RNA-seq data indicated efficient BRD2 knockdown without significantly affecting expression of BRD3, BRD4 and BRDt (Supplementary Fig. 7b–e). As expected, the testis-specific BRDt was not expressed in all RNA-seq samples, and the three ubiquitous BET genes were expressed in all samples and all cell types (Supplementary Fig. 7), indicating the high quality of our RNA-seq. There were 1970 differentially expressed genes (DEG) upon BRD2 knockdown (Fig. 2a and Supplementary Fig. 8a), indicating a broad and significant transcriptional impact of BRD2 on the reprogramming cells. Of those DEGs, the majority (1163 genes, 59%) were significantly downregulated by BRD2 shRNA in agreement with its reported positive role in transcription. However, the fraction of up-regulated genes (805 genes, 40.9%) is noteworthy (Fig. 2a and Supplementary Fig. 8a). We previously also observed a large fraction of up-regulated genes for JQ1 inhibition of BET proteins in the reprogramming cells¹³.

Fig. 2. BRD2 knockdown legitimately reprograms genes of matrisome and its regulation as revealed by RNA-seq at 72 h of reprogramming.

a Number of DEG and the legitimately reprogrammed genes by BRD2 shRNA vs luciferase shRNA in the reprogramming cells. b Percentage of legitimate reprogramming by BRD2 shRNA vs luciferase shRNA in the reprogramming cells. c Top 10 enriched pathways for the legitimately down-regulated genes by BRD2 shRNA in early reprogramming cells. d cnetplot for the top 8 enriched pathways of the legitimately down-regulated genes. e Down-regulation of 57 matrisome and related genes by BRD2 shRNA towards their pluripotent levels. F, fibroblasts, n = 9 RNA-seq; S, BRD2 shRNA, n = 2; L, luciferase shRNA, n = 2; E, embryonic stem cells, n = 7. f Violin plots for the 410 DEG matrisome genes between human ESC and fibroblasts. The box represents the 25th (lower border), 50th (white dot) and 75th (upper border) percentiles with the violin lines representing the data density at each percentile point. g Waterfall plot on log2(fold changes) for the 410 DEG matrisome genes.

We then evaluated the reprogramming legitimacy of those DEGs^36,37. In short, if a change in expression levels of a gene is towards the pluripotent level, the transcriptional change is legitimate; otherwise (the gap becomes even greater), the reprogramming of this gene is illegitimate. Overall, 30% (592 out of 1970 genes) DEGs underwent legitimate reprogramming (Fig. 2a, b). More downregulated genes were legitimately reprogrammed towards the lower pluripotent levels (39.7%, 462 out of 1163). Only 16.2% (130 out of 805 genes) of the upregulated genes were legitimately reprogrammed upward to the higher pluripotent direction. Like the Yamanaka factors that cause significant aberrant transcriptional changes³⁶, not all transcriptional changes by BRD2 shRNA are legitimate in terms of reprogramming. In sum, BRD2 knockdown predominantly leads to legitimate down-regulation of the somatic genes, in agreement with our previous chemical BET inhibition in pluripotency reprogramming¹³ and the reported chemical neuron reprogramming of mouse fibroblasts¹⁹.

To understand the nature of the predominantly legitimate down-regulation of the somatic program, we ran a pathway enrichment analysis for the 462 legitimately down-regulated genes. The top 10 enriched pathways are overwhelmed by the terms of the extracellular matrix (ECM) theme (Fig. 2c), and those ECM terms shared many genes as revealed by category-network plot (cnetplot) of the top 8 terms (Fig. 2d). Of note, 17 out of the 30 enriched pathway terms are in the category of matrisome or are related to matrisome (Sheet 1 in Supplementary Data 1). The only independent term is “Butyrophilin (BTN) family interactions” (4 genes only) (Supplementary Fig. 8b). The remaining 12 terms still include shared genes with the matrisome gene network although those terms are not directly related to the matrisome terms (Supplementary Fig. 8b). Fifty-seven matrisome and matrisome-related genes are legitimately downregulated by BRD2 shRNA (Fig. 2e). In contrast, among the 93 enriched biological process (BP) terms and 13 pathway terms for the upregulated genes by BRD2 shRNA, we did not see any enriched ECM or ECM-related terms. Given that, we compared the transcription levels of the Naba matrisome gene set³⁸. Of the 1027 curated matrisome genes, 410 are differentially expressed between ESCs and the starting fibroblasts. Among the 410 DEGs, 283 (69%) are expressed at significantly lower levels in ESCs, but 127 (31%) displayed significantly higher expression in ESCs. We then compared the expression levels for those DEG matrisome genes as a whole. Overall, human ESCs displayed a much lower expression for the matrisome gene set (Fig. 2f). Waterfall plots³⁷ also indicated that ESCs have a much weaker matrisome program (Fig. 2g). In summary, BRD2 knockdown facilitates legitimate reprogramming of 592 genes and the majority is legitimate downregulation of the somatic genes, particularly the genes of matrisome and its regulation, towards the pluripotent levels.

BRD2 negatively regulates transcription of lipid genes

Next, we conducted gene ontology (GO) analyses of the upregulated gene list (805 genes) by BRD2 shRNA. Because too many enriched BP terms (93) were observed, we focused on pathway enrichment analyses of the upregulated genes with the Reactome knowledgebase. In total, far fewer pathway terms (13) are statistically enriched. Intersection and cnetplot analyses indicate that several terms share the exact sets of genes. Thus, we reduced the number of the enriched terms to 6 by removing the redundant terms. Cnetplot analyses revealed that the enriched pathways formed three clusters. One term (“tRNA aminoacylation”) was independent of the others (magenta in Fig. 3a). The term of “response of EIF2AK1 (HRI) to heme deficiency” shared one out of its 7 genes with the “cellular senescence” term. These two independent clusters each included fewer genes (8 each). The remaining four terms are intertwined. “ESR-mediated signaling” and the adipogenesis/steatosis groups shared 11 genes and are semantically related since ESR (estrogen receptor) is related to lipids/steroids. This merged steroid/lipid larger group includes 32 genes.

Fig. 3. BRD2 negatively regulates lipogenesis and lipid metabolism genes.

a Cnetplot summarizing all enriched pathways for genes upregulated by BRD2 shRNA knockdown vs luciferase shRNA in the early reprogramming cells (at 72 h). b Cnetplot for the three lipid-related enriched BP terms for the downregulated genes by BRD2 overexpression in the reprogramming cells relative to BRD2 shRNA knockdown. Black dots in a and b represent gene functional categories. Each colored dot represents one gene with its relative expression level. RNA-seq was conducted at 72 h of reprogramming. c Dot plots for the 30 enriched BP terms in the categories of lipid biosynthesis, metabolism and regulation for the genes compromised by BRD2 (n = 2, 72 h) overexpression in reprogramming cells in reference to BRD2 KO (n = 4). d Heatmap for the 145 lipid genes suppressed by BRD2 in the reprogramming cells. Selected genes are listed beneath the heatmap, and the full list of genes can be found in sheet 2 of Supplementary Data 1. BRD2KO48 and BRD2KO72, BRD2 KO treatment at 48 and 72 h post induced KO in the reprogramming cells, respectively. Heat maps were generated from two RNA-seq samples for BRD2 overexpression (a, b) and four RNA-seq samples for BRD2 KO (48a, 48b, 72a and 72b).

In addition to reprogramming enhancement by BRD2 shRNA, BRD2 overexpression almost abolished the iPSC reprogramming (Figs. 1g and 5a, b). We then focused on this dramatic impact of BRD2 on reprogramming. When the RNA-seq data of the reprogramming cells are compared between BRD2 overexpression and its shRNA knockdown, only two pathway terms were enriched for the downregulated genes by BRD2 overexpression relative to its knockdown, “transcriptional regulation by TP53” and “Cholesterol biosynthesis” (Supplementary Fig. 9). The P53 pathway is a reported barrier for reprogramming but our data indicates that BRD2 mitigates and its shRNA worsens the P53 barrier, contradicting BRD2 role as a reprogramming barrier. Our attention then turned to the “cholesterol biosynthesis” pathway given that BRD2 knockdown enhanced adipogenesis and ESR-mediated signaling as discussed above. Scrutiny of enriched BP terms for BRD2-downregulated genes revealed that three lipid-related BP terms are enriched: “cholesterol biosynthetic process”, “fatty acid metabolic process”, and “secondary alcohol biosynthetic process” (Fig. 3b). These data indicate that BRD2 impairs cholesterol biosynthesis and FA metabolism.

Fig. 5. The ET domain of BRD2 mediates the suppression of lipid genes in the reprogramming cells.

a Schematic of individual deletion of the three BRD2 characteristic domains and summary of their reprogramming activities. n = 3 independent experiments; one-way ANNOVA test followed by post hoc Dunnett’s test. Error bars are standard deviations. b Representative images of ALP staining of the reprogramming wells upon overexpression of different BRD2 domain-deletion variants as labeled. c Dot plot of the top 10 enriched pathway GO terms for the gene list upregulated by BRD2ΔET forced expression relative to BRD2 in the reprogramming cells. The lipid-related terms are in red text. d Cnetplot of the top 5 enriched pathway terms for gene list upregulated by BRD2ΔET forced expression relative to BRD2 forced expression in the reprogramming cells (at 72 h). SCD and HMGCR are highlighted in red. e Heat maps of the 119 lipid genes (combined BP and reactome pathway terms) upregulated by BRD2ΔET overexpression in reprogramming cells as compared to BRD2 WT at 72 h of reprogramming. Heat maps were generated from two RNA-seq samples for BRD2 overexpression (a and b) and two RNA-seq samples for BRD2ΔET (T1 and T2). f Graphic summary for BRD2-centered molecule roles in hiPSC reprogramming. Green arrows, enhancing activities; red T-bar, suppression activities.

To further confirm the role of BRD2 in lipid biology, we compared the RNA-seq data of the reprogramming cells between the two extreme conditions: induced BRD2 KO (Supplementary Figs. 3a–d and10) and its overexpression (Supplementary Figs. 7a and 14). There were 2354 DEGs with similar numbers of up- and down-regulated genes (1175; 1179, respectively). The top enriched pathway terms for the upregulated genes by BRD2 are predominantly of the category of cell cycle (Supplementary Fig. 11) in agreement with the reported BRD2 role in the regulation of cell cycles. On the other hand, the top GO and pathway terms for the BRD2-downregulated genes are lipid biosynthesis, metabolism, and regulation (Supplementary Fig. 12). We tallied all lipid BP terms for the genes downregulated by BRD2 overexpression vs KO and found 30 such terms (Fig. 3c and sheet 2 in Supplementary Data 1) with 145 lipid genes (Fig. 3d). In conclusion, BRD2 suppresses a large set of genes in lipid biosynthesis and metabolism in the reprogramming cells and abrogates hiPSC reprogramming when overexpressed.

SCD and HMGCR overcome BRD2 impairment of reprogramming

Given that BRD2 suppresses a large set of lipid genes in the reprogramming cells, we chose 9 such genes (red text in Fig. 3d) that were downregulated by BRD2 in the reprogramming cells with the hypothesis that some of them might enhance iPSC reprogramming when overexpressed. We previously reported that hiPSC reprogramming encounters reprogramming stress¹⁸. We thus include NAD(P)H quinone dehydrogenase 1 (NQO1) because it has antioxidant properties and is upregulated by BRD2 KO by 4.4-fold (Supplementary Fig. 13). To this end, we cloned these 10 human cDNAs into our optimized lentiviral vector with GFP co-expression via P2A peptide³⁵ (Fig. 4a) so that the transgene titers can be accurately and readily determined using GFP. Our reprogramming experiments showed that SCD and HMGCR enhanced iPSC reprogramming but not NQO1 (Fig. 4b). At the same time, the HMGCR chemical inhibitor (lovastatin) abrogated iPSC reprogramming (Fig. 4c). Enhancement of reprogramming by BRD2 shRNA was erased by lovastatin (Fig. 4c), indicating that enhancement of reprogramming by BRD2 knockdown relies on HMGCR. Given the critical and rate-limiting roles of SCD and HMGCR in lipid biosynthesis, we supplemented the reprogramming medium with lipids. Indeed, lipid supplements promoted iPSC reprogramming (Fig. 4d). Lipid and BRD2 shRNA demonstrated small but insignificant synergism (Fig. 4d), indicating mainly overlapping functions with some possible distinct roles in reprogramming. This is supported by the fact that BRD2 shRNA restored some reprogramming activity in the presence of HMGCR inhibitor (Fig. 4c).

BRD2 ET domain is the major intramolecular regulator of reprogramming, lipogenesis and matrisome

Previously, we reported that a truncated isoform of BRD3 (BRD3R) lacking its ET tail promotes the Yamanaka reprogramming¹⁷. We have also shown that deletion of the three characteristic domains releases reprogramming activity from BET proteins including BRD2¹⁸. We wanted to know whether any of these three domains is more critical in terms of reprogramming inhibition. For this purpose, we deleted each domain individually (Fig. 5a). Those BRD2 deletion constructs can be efficiently expressed at the expected sizes (Supplementary Figs. 14 and 15). Like wildtype BRD2, the reprogramming efficiency remained significantly lower than the GFP control when each bromodomain was deleted alone, but ET deletion alone significantly increased the reprogramming (Fig. 5a, b). In fact, all three BRD2 variants that retained their ET tail displayed significantly lower reprogramming as compared to GFP, blank (OSK only), and the BRD2ΔET (Fig. 5a, b), indicating that the ET tail suppressed iPSC reprogramming but some sequences flanking the two bromodomains promotes reprogramming (Fig. 5a, and our published work)¹⁸. In addition, colonies from the BRD2ΔET condition were larger as compared to that of the controls and had a more rounded morphology (Fig. 5b).

Given the ET-mediated inhibition of reprogramming observed above, we investigated the transcriptional impact of the ET tail in the reprogramming cells. Compared to BRD2, BRD2ΔET resulted in 2521 DEGs (Supplementary Fig. 17a). Similar amounts of genes are upregulated as downregulated (1257 vs 1264), indicating that ET tail plays both positive and negative roles in transcription. Reactome pathway analysis of the upregulated genes by BRD2ΔET relative to BRD2 indicated that 5 of the top 10 pathway terms are related to lipids (Fig. 5c); cnetplot analysis of the enriched reactome pathway terms showed that lipid terms are the most enriched network and all the top 5 pathway terms in cnetplot are related to lipids (Fig. 5d). GO enrichment analysis revealed that 9 of the top 10 enriched BP terms are related to lipids (Supplementary Fig. 16a, b). There were 11 lipid-related BP terms for the upregulated genes by BRD2ΔET after removing the redundant terms (sheet 3 in Supplementary Data 1). In total, 119 genes in lipid pathways are de-repressed by BRD2ΔET (Fig. 5e). In sum, BRD2 ET tail is an intramolecular inhibitor for transcription of lipid-related genes.

We then conducted pathway enrichment analysis using the Reactome database for the genes downregulated by BRD2ΔET relative to BRD2. Among the top 10 Reactome terms, 8 are related to ECM (Supplementary Fig. 17b, c). This result is perfectly aligned with observations from BRD2 shRNA and its KO reported above (Fig. 2 and Supplementary Fig. 8).

We then examined the reprogramming legitimacy of the DEG by BRD2ΔET. Out of the 1257 up-regulated DEGs, 494 (39.3%) are expressed significantly higher in ESC, representing a legitimate up-regulation; on the other hand, 726 out of the 1264 down-regulated DEGs (57.4%) underwent legitimate down-regulation (Supplementary Fig. 17d). We then ran pathway enrichment analyses for the legitimate reprogramming gene lists. Among all the 15 enriched reactome terms for the legitimate down-reprogramming gene list, 12 are in the category of ECM and two of the remaining three terms still have connection to the mastrisome network (Supplementary Fig. 18). The remaining three terms are associated with 9 independent genes only. For the legitimate up-reprogramming genes, the top reactome terms are related to cell cycle (Supplementary Fig. 17e) as we reported before^17,18. Three enriched Reactome terms of lipid biosynthesis are among the top 10 legitimate reprogramming terms for the up-regulated genes by BRD2ΔET (Supplementary Fig. 17e). In sum, BRD2 ET tail impairs iPSC reprogramming and suppresses the gene transcription of lipid biosynthesis in the reprogramming cells. BRD2 ET tail, on the other hand, positively regulates transcription of somatic genes, in particular, ECM-related genes. While unexpected, those opposite functions of ET are likely context dependent and mediated by different unidentified factors on local chromatin environments.

BRD2 binds to genes of lipogenesis and regulates their epigenetics

To gain epigenetic insights into our findings above, we conducted ChIP-seq with a BRD2 antibody. As expected, ChIP PCR with the chosen antibody can pull down the target sequences of the known BRD2-regulated genes (Supplementary Fig. 19a). Intersection analysis indicates that our ChIP-seq duplicates shared majority of the binding peaks (19,101, 64%) (Supplementary Fig. 19b). To increase credibility, we used the 19,101 common peaks only in the analyses below. We considered those peaks with binding scores above 75% quantile (Supplementary Fig. 19c, d) to be strong BRD2 binding. Seventy-nine percent of the strong binding peaks are within 1 kb surrounding TSS (Supplementary Fig. 19e). Of the BRD2 strong binding sites, KEGG terms of “fatty acid metabolism” and “terpenoid backbone biosynthesis” are enriched (Fig. 6a). These data agree with our RNA-seq observation that BRD2 regulates genes of lipid biosynthesis and metabolism. Sterol regulatory element-binding transcription factor 1 (SREBF1) is the master transcription factor (TF) to regulate genes of cholesterol and FA biosynthesis. Indeed, BRD2 and SREBF1 co-bind 9654 sites (Fig. 6b, c). We then focused on the gene set of “metabolic process” and found that BRD2 bound to those genes and the binding became significantly weaker upon BRD2 disruption indicating specificity of our antibody (Fig. 6d, upper panels). Unexpectedly but interestingly, upon BRD2 knockdown, the H3K27Ac marks were increased substantially on those genes (lower panels, Fig. 6d). We further analyzed binding on genes of “fatty acid metabolism” and “terpenoid biosynthesis”. Similarly, BRD2 binding decreased significantly upon its disruption, and the active transcription mark H3K27Ac increased significantly upon BRD2 disruption (Fig. 6e). This is in line with our RNA-seq data that showed negative regulation of lipid genes by BRD2. We then examined the promoters of HMGCR and SCD. BRD2 and SREBF1 co-bind both promoters (Fig. 6f). Upon BRD2 disruption, the H3K27Ac mark was increased on both SCD and HMGCR (Fig. 6f). These ChIP-seq data support our previous conclusion of BRD2 suppression of lipogenesis revealed by our RNA-seq data.

Fig. 6. BRD2 directly binds to promoters of metabolism genes including lipid metabolism and negatively regulates their H3K27Ac status.

a Top 20 KEGG terms of the category of “metabolism” enriched in the set of genes with strong BRD2 binding. b BRD2 and SREBF1 share 9654 binding regions. Upper, Venn diagram of SREBF1 and BRD2 binding sites (all binding regions are included). Lower, counts of binding sites around TSS. c Heatmaps of SREBF1 and BRD2 binding regions showing all SREBF1-specific regions close to TSS. d ChIP-seq read density heatmaps of BRD2 and H3K27Ac marks in naïve human fibroblasts at metabolic related genes associated with the GO terms of metabolic processes found in annotated BRD2 interaction regions close to TSS. e ChIP-seq normalized read density heatmaps of BRD2 and H3K27Ac marks in naïve human fibroblasts around TSS of genes associated with the enriched KEGG pathways “Fatty acid metabolism” and “Terpenoid biosynthesis” shown in a, that were annotated with strong BRD2 bindings. f ChIP-seq normalized read density tracks of BRD2 (green), H3K27Ac (blue) and SREBF1 (red) on HMGCR and SCD genes in naïve human fibroblasts. A side-by-side comparison between BRD2 shRNA (shBRD2), or sgRNA and scrambled control was included.

BRD2 binds to TF promoters that regulate matrisome genes

To understand how BRD2 positively regulates somatic matrisome genes, we analyzed BRD2 ChIP-seq data in the context of matrisome genes. We subset the top 5% binding peaks and obtained 2805 strong BRD2 binding peaks around TSS (±2 kb). Those peaks were annotated and analyzed for GO term enrichment, but we did not see any enriched terms related to matrisome. We did, however, found that out of the 2805 BRD2 promoter targets 34 were matrisome gene promoters, but just 3 (PLAU, MMRN2 and MDK) of them were significantly downregulated by BRD2 shRNA. We noticed the enriched terms in transcription, chromatin, and histone across all three GO categories (Fig. 7a1–a3), consistent with BRD2 roles in transcriptional regulation. We then conducted pathway enrichment with the reactome database, and subset for significant terms (q-value < 0.01) with keywords of transcription, chromatin, or histone. The resulting network reveals a clear BRD2 regulation of transcription (Fig. 7b). Of note, “transcriptional regulation by RUNX2” and “by RUNX3” are 2 of the 7 terms about transcriptional regulation. RUNX2 and RUNX3 are TFs that regulate matrisome³⁹. We then focused on TFs among the 2805 promoter targets bound by BRD2. To this end, we performed regulon analysis of the 2805 BRD2 peaks within 2 kb of TSS using the R package DoRothEA⁴⁰ for all human TFs obtained from the Scenic database⁴¹ (Fig. 7c). Out of the 357 putative BRD2-regulated TFs, 47 genes were regarded as high-confidence regulon TFs based on DoRothEA classifications. Among these 47 high-confidence BRD2-regulated TFs are those with roles in matrisome gene regulation such as RUNX2, SMAD1, HIF1A, ETS1, and TCF4. The regulons of the 47 TFs are composed of 1940 unique target genes, of which 173 were associated with the matrisome. Using these 173 matrisome-related regulon targets we constructed a regulatory network (Fig. 7d). We found that 25 target genes associated with the matrisome were downregulated by BRD2 shRNA, of which 8 genes (TNC, IGFBP3, LTBP4, NTN1, IGFBP4, COL13A1, COL16A1, and NTNG2) were of the core matrisome. The above data indicate that BRD2 may regulate matrisome genes by regulating matrisome TFs, although it directly binds to some matrisome genes.

Fig. 7. Regulon analysis of BRD2 binding sites reveal that BRD2 may regulate matrisome genes in the starting fibroblasts via regulating the matrisome transcription factors.

a Gene ontology analysis showing the top 10 terms in each of the biological process (a1), molecular function (a2) and cellular component (a3), associated with the top 5% strongest BRD2 binding interactions (2805 promoter targets). Terms associated with transcriptional regulation are shown in red. b Cnetplot showing significant pathway enrichment of terms related to the terms of transcription, chromatin and histone. All terms with a q-value < 0.05 were selected. c Regulon analysis of the top BRD2 interactions. TFs Transcription Factors. TSS Transcription Start Site. d Regulatory network of all matrisome related regulons. Annotated regulon targets (red) correspond to genes found to be downregulated by BRD2 shRNA during reprogramming. Targets associated with the matrisome were selected from the Naba matrisome database.

Discussion

We previously identified that the short isoform of human BRD3 lacking its ET-containing tail enhances hiPSC reprogramming¹⁷. We later found that the pan BET chemical inhibitors promote hiPSC reprogramming at low concentrations while the widely used high concentrations impair it¹³. The present study shows that BRD2 appears to be the target of the beneficial mild chemical inhibition with the pan BET inhibitors in promotion of reprogramming (Fig. 5f). Pan BET chemical inhibition enhances mouse fibroblast reprogramming to functional neurons¹⁹. Liu et al. reported that BRD4 enhanced mouse iPSC reprogramming²⁰, but it might do so only at the late stage since another group showed that early BRD4 inhibition slightly increases mouse fibroblast reprogramming towards iPSCs²¹. The latter report, however, did not examine shRNA against BRD2 and BRD3²¹. Using shRNA against each of the expressed BET members, we found that only BRD2 acts as a reprogramming barrier. It is important to recognize that all other BET studies have used mouse cells^19–21 while our investigation focused on human iPSC reprogramming. Species might be sources of some discrepancies.

We reported here that BRD2 shRNA and its dominant negative (DN) mutants legitimately downregulated somatic genes of matrisome and matrisome regulation towards the lower pluripotent levels. Human PSCs appear to have a unique matrisome¹¹. Human ESC culture needs the support of feeder cells⁹. In feeder-free systems, human PSCs still require coating of growth surface with matrix proteins¹⁰. Human PSC passage is also special in that the traditional trypsin detachment kills hPSCs³⁵, indicating a unique cell-matrix or cell-surface environment. In addition, hPSCs rely on cell-cell contacts for self-renewal and form tight colonies in culture³⁵. Individualized hPSCs die extensively. Of note, the Yamanaka factors downregulate somatic transcriptional program. Given this, BRD2 role in maintaining the somatic transcriptional program including genes of matrisome may be masked to some extent by the Yamanaka factors. Nevertheless, with our concept of reprogramming legitimacy we can still demonstrate that BRD2 maintains the somatic program and is thus an adverse factor of pluripotency reprogramming.

We showed here that lipid deficiency is a reprogramming barrier. This aligns with the fact that PSCs have a truncated and fast cell cycle as compared to the somatic ones, and establishment of this short pluripotent cell cycle is part of iPSC reprogramming⁴². The fast-proliferating iPSCs require more lipids as the key cellular building blocks of new cellular membranes. In support of our results, lipids regulate genome integrity, self-renewal, and pluripotency in PSCs⁴³. Our data showed that the desaturase SCD enhanced hiPSC generation. This is in line with the report that mouse ESCs have significantly higher monounsaturated FA (18:1, 20:1) than fibroblasts⁴⁴, and that 4–5 unsaturated cardiopilin, a major component of mitochondria inner membrane, is more abundant in mouse ESCs⁴⁴. In both human and mouse blastocysts, unsaturated lipids and membrane fluidity increased as compared to those in the 2-cell to 8-cell stage embryos⁴⁵. SCD regulates the membrane fluidity and cell polarity of the embryonic cells in the pre-implantation embryos⁴⁵. A recent report suggests that de novo lipogenesis is active between mouse 8-cell and pre-implantation embryos when lipid supply is not available from oocyte reserves or the pregnant mother²⁹. Those authors also showed that de novo lipogenesis is active in human PSCs cultured in the chemically defined E8 medium. Interestingly, our data suggests that BRD2 suppresses expression of lipid genes even though it is generally considered as a positive transcriptional regulator. Our BRD2 and H3K27Ac ChIP-seq data support our RNA-seq observations that BRD2 suppresses expression of lipid genes but our observed increase of H3K27Ac marks on those genes upon BRD2 loss challenges the current paradigm about the relationship of BET proteins and acetylation. We here report a negative correlation between BRD2 and H3K27AC, but it remains an open question about how BRD2 regulates the H3K27Ac status. In support of paradigm-changing findings here, a previous study indicates that mice with lower BRD2 develop more fat indicating BRD2 suppression of lipogenesis in animals⁴⁶. While this manuscript focuses on reprogramming it will be an exciting direction to study how BRD2 negatively affects acetylation status and lipid programs.

Here, we propose that BRD2 impedes hiPSC reprogramming largely by suppressing the lipid programs (Fig. 5f). This does not exclude other mechanisms (Fig. 5f). We do see that BRD2 shRNA and BRD2ΔET legitimately down-regulate the somatic genes, especially those of matrisome and its regulation. However, those positive impacts may not be seen phenotypically by the commonly used measurements of reprogramming efficiency³⁴. We previously proposed that BET mini proteins enhanced reprogramming by mitigating reprogramming stress¹⁸. We do see that many stress response GO terms are among the enriched lists for DEGs by BRD2 shRNA and BRD2ΔET. However, the mechanism of lipid deficiency to explain BRD2 as a reprogramming barrier is clearly supported by genetic, pharmaceutical, and lipid supplement data with SCD and HMGCR, two rate-limiting enzymes in lipid biosynthesis. One future direction is to investigate whether lipid deficiency has any relationship with reprogramming stress. It is well known that lipids are targets of various cellular stress. For example, lipids are one of the major targets of oxidative stress⁴⁷. ER stress alters lipid gene expression since ER is the major site of lipogenesis. ER stress pathway and lipogenesis also crosstalk⁴⁸. Lipids regulate PSC genome integrity⁴³. It is possible that reprogramming methods based on our current findings may improve iPSC quality in addition to efficiency. Future works may test whether iPSCs generated with lipid supplements and BRD2 manipulation harbor less mutations. Another open question relates to the lipids we used. Although the lipids are defined it is still a mixture of more than 10 species. Identification of possible individual active components in the lipid mixture will further shed insight into the precise molecular mechanisms regarding possible lipid regulation of reprogramming.

Methods

Production of lentivirus particles

Lentivirus vectors were generated using PEI-mediated transfection into Lenti-X^TM 293 T (Takara, Cat. 632180) cells. Briefly, 1 × 10⁷ Lenti-X^TM 293 T cells were seeded into one 150 mm dish per vector and cultured in 24–30 mL expansion medium: DMEM-F12 (Gibco, Cat. 12400-024) supplemented with 10% FBS (Gibco, Cat. 10437-028) to 80–90% confluence. At 80–90% confluence, medium was replaced with 24 mL fresh expansion medium at least 2 h before transfection. Transfection solution was prepared by mixing 3 mL of plasmid/DNA solution, containing envelope, packaging and transfer plasmids at a ratio of 1:3:4 (total amount of 60 µg of plasmid), with 3 mL of PEI solution, containing 60 µg/mL of polyethylenimine (PEI, Polysciences Inc., Cat. 24765-2). The DNA:PEI ratio is 1:3 by weight. Plasmid and PEI solutions were prepared in plain DMEM/F12 and mixed to get a final transfection solution of 6 mL. The transfection solution was incubated for 15 min at room temperature and added dropwise into the cell cultures of 80–90% confluence. The cultures were incubated for 16 hours at 37 °C, 5% CO₂. After the transfection, the medium was replaced with 20 mL fresh complete expansion medium, and the cells were incubated for 72 h. Medium with lentiviral particles was filtered using Stericup low-protein-retention 0.45-µm PVDF membrane filters (Millipore). The viruses in the filtered media were concentrated by PEG6000 (8.5% final concentration) and NaCl (0.4 M final concentration) for 3 h of incubation at 4 °C with periodical mixing. The virus-PEG complex was then centrifuged at 4500 × g for 45 min. The resulting lentiviral pellet was re-suspended into 150 µL sterile cooled PBS. This virus concentrate was aliquoted in small volume and then stored at −80 °C. Titration of lentiviral concentrate was performed by transducing HeLa cells and analyzing the co-expressed GFP signal with flow cytometry (Fortessa Flow Cytometer, BD) 72 h post transduction as described³⁵. Briefly, concentrated viruses were diluted in cold PBS 100x times. Various amounts of diluted viruses (1, 2, 5, and 10 μl) were added into wells of 24-well plates containing Hela cells along with polybrene (4 μg/ml). One well was used to count the cell numbers at the time of transduction. The next day, transduction media is replaced with fresh Hela cell growth media. Cells were analyzed with flowcytometry for GFP expression. The titers were calculated based on percentage of GFP and numbers of Hela cells at the time of transduction.

Human fibroblast reprogramming into hiPSCs³⁵

BJ fibroblasts (ATCC, CRL-2522) at passages 3 to 6, were seeded into 12-well plates at a density of 6.5 × 10⁴ cells/cm². Twenty-four hours after plating, OCT4, SOX2, and KLF4 (OSK) lentiviral vectors, or OCT4, SOX2, KLF4, and MYC (OSKM, Addgene Cat#, 130692, 130693, 130694, and 130695, respectively)¹⁷ lentiviral mix was added to the fibroblast culture in expansion medium supplemented with 5 μg/ml hexadimethrine bromide. As reported before, the virus Multiplicity Of Infection (MOI) for the Yamanaka factors are 8 for OCT4, 5 for SOX2 and KLF4, but one for MYC. When indicated, a shRNA lentiviral vector targeting BRD2, BRD3, BRD4, or Luciferase (Scramble control), or a lentiviral vector for the overexpression of BET proteins (BRD2, BRD3, BRD4, or BRD2 deletions) or GFP (mock control), was added into the medium. To be consistent, we used 3 MOI of transgene lentiviruses in the reprogramming experiments. Viral titers were determined by functional titration using GFP by flow cytometry as we previously reported³⁵, which is co-expressed along with all transgenes tested in this study (Fig. 4a). The cells were transduced for 12 to 16 h at 37 °C and 5% CO₂. The next morning, viral particles were removed, and a fresh fibroblast medium was added. Forty-eight hours after transduction, the cells were passaged into Geltrex-coated 12-well or 24-well plates at a density of 5 × 10⁴ cells/cm². At day 3 post transduction, the cells were cultured with the serum free medium E7 (E8 without TGF beta). From day 3 on, the medium was changed daily with fresh E7 medium. From day 15 on, the cells were cultured with the xeno-free medium E8. At day 20, the colony number was determined by ALP or TRA-1-60 staining (all original counts of iPSC colonies are recorded in Supplementary Data 3). For iPSC isolation, compact colonies with clear edge were picked manually and seeded into a Geltrex-coated well of a 24-well plate. HiPSC colonies were subcloned to purify GFP negative cells and establish hiPSC lines with transgenes silenced. The established hiPSC lines were characterized by the expression of pluripotency markers, chromosome stability, transcriptome clustering (for details, see sections of “Immunocytochemistry”, “Flow Cytometry”, and “RNA-seq and Analysis”), and differentiation capacity through teratoma formation assay as described³⁴. Teratomas were generated using NSG immunocompromised mice (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, The Jackson Lab Laboratory, Strain #: 005557). Use of NSG mice was approved by our Institutional Animal Care and Use Committee (IACUC). We used 2 million freshly harvested iPSCs in 30% Matrigel with a total volume of 100 μl for one injection. The iPSCs were injected subcutaneously into one flank of an NSG mouse. Teratomas were harvested around 8 weeks post injection. Harvested teratoma was immediately submerged in 4% formaldehyde for fixation. H&E sections of teratoma were analyzed under a compound microscope.

BET protein quantitation

Total BET protein quantitation was performed using validated sandwich and competitive ELISA kits from MyBioSource (Cat. MBS945281, MBS7245725, MBS7206026). BJ cell protein extracts from 3 different passages were obtained by cell disruption using 1× RIPA buffer plus 1× Halt protease inhibitor cocktail (Thermo Scientific, 78429) followed by sonication. Protein concentration was determined using BCA assay (ThermoFisher Scientific) and diluted to 1 µg/mL. Samples were diluted 1:10 to 1:20 using 1× sample dilution buffer provided by manufacturer. Assay procedure was performed per manufacturer’s guidelines. Standard curve was determined using linear regression from logarithmic transformations of concentration versus OD.

Alkaline phosphatase (ALP) staining

To quantitate the reprogramming efficiency, colonies that emerged at day 20 after transduction (or indicated time) were stained using BCIP/NBT colorimetric staining technique as reported before³⁴. Briefly, ALP staining solution was prepared by diluting 5-bromo-4-chloro-3-indolyl phosphate (BCIP) and nitro blue tetrazolium (NBT) in Tris buffered saline, pH 9–9.5, at a final concentration of 150 μg/mL and 300 μg/mL, respectively. Cells were fixed with cold methanol for 10 min followed by a brief wash with PBS. Then, 500 µl of ALP staining solution was added to each well of a 24- or 12-well plate, followed by an incubation of 10–20 min. ALP staining solution was removed and 500 µl of PBS was added to stop the reaction. Colonies were counted through a phase contrast microscope.

TRA-1-60 staining

Colonies positive for TRA-1-60 were identified through the highly sensitive chromogenic 3,3’ diaminobenzidine assay (Metal-Enhanced DAB Substrate Kit, Thermo Scientific, Cat. 34065) as we reported before³⁴. Briefly, cells at day 20 of reprogramming (or indicated time) after transduction were fixed with ice cold methanol for 10 min and then blocked with 2% BSA in PBS for 30 min. Biotinylated anti TRA-1-60 antibody in blocking solution (eBioscience, Cat. 13886382) was added into the fixed cells and incubated overnight at 4 °C. After antibody removal by 3 washes with PBS, streptavidin-HRP (BD, Cat. 554066) was added and incubated for 30 minutes at 20–25 °C. After washing the cells with PBS, 250 μl of 1× metal enhanced DAB substrate was added to each well and incubated for 10 minutes. After incubation, the substrate solution was removed and 500 μL PBS was added. TRA-1-60 positive colonies were counted using a phase-contrast microscope.

Immunocytochemistry

Immunocytochemistry has been described in detail previously³⁴. Briefly, 8-well chamber slides (Sigma, Cat. C7182) were coated with 1% hESC-qualified Geltrex (Gibco, Cat. A1413302) for 1 h at 37 °C. Human iPS Cells were seeded at 50% confluence and cultured with E8 medium until around 80% confluence. Cells were washed with PBS and fixed with 4% paraformaldehyde for 10 min. Cells were blocked and permeabilized for 30 min, using PBS (Gibco) supplemented with 2% BSA Fraction V (Roche) and 0.3% Triton. Primary antibodies (OCT4, SOX2, NANOG, and LIN28, see Table 1 for antibody information) were diluted in the blocking-permeabilization solution. Cells were incubated with the diluted antibody overnight at 4 °C. After washing, cells were incubated with the secondary antibody for 1 h at 20–25 °C. The cells were washed, and the chamber was removed carefully. The cells on slides were mounted using ProLong^TM Gold Antifade Mountant with DAPI (Invitrogen, P36935).

Table 1.

Antibody used in this study

Antibody names	Vendors	Catalog numbers	Dilutions (1:X)a
Biotinylated TRA-1-60, monoclonal	eBioscienceTM	Cat#13-8863-82; RRID:AB_891594	200 (IHC)
TRA-1-60, PE conjugated	BD PharmingenTM	Cat#560193; RRID:AB_1645539	5 (FC)
TRA-1-81, PE conjugated	BD PharmingenTM	Cat#560161; RRID:AB_1645540	5 (FC)
SSEA3, PE conjugated	BD PharmingenTM	Cat#560237; RRID:AB_1645542	5 (FC)
SSEA4, PE conjugated	BD PharmingenTM	Cat#560128; RRID:AB_1645533	5 (FC)
SSEA1, PE conjugated	BD PharmingenTM	Cat#560142; RRID:AB_1645246	5 (FC)
Anti-human alkaline phosphatase, PE	R&D	Cat#FAB1448P	10 (FC)
Mouse anti-human OCT4	BD PharmingenTM	Cat#561555; RRID:AB_10715577	200 (IF)
Mouse anti-human NANOG	BD PharmingenTM	Cat#560482	200 (IF)
Rabbit anti-human LIN28	Invitrogen	Cat#PA1-096; RRID:AB_2539865	200 (IF)
Most anti-human SOX2	BD PharmingenTM	Cat#561469; RRID: AB_10694256	200 (IF)
Rabbit anti-BRD2	Bethyl - Fortis	Cat#A302-582A	2000 (WB)
Rabbit anti-BRD2	Bethyl - Fortis	Cat#A700-008	1000 (WB)
Rabbit anti-BRD3	Bethyl-Fortis	Cat#A302-368A	2000 (WB)
Rabbit anti-BRD4	Bethyl-Fortis	Cat#A301-985A	10000 (WB)
Mouse IgG, PE conjugated	BD PharmingenTM	Cat#555749; RRID:AB_396091	5 (FC)
Mouse IgM, PE conjugated	BD PharmingenTM	Cat#555584; RRID:AB_395960	5 (FC)
Alexa Fluor 488 goat anti-mouse IgG, IgA, IgM(H + L)	Life Technologies	Cat#A10667; RRID:AB_253405	1000 (IF)
Alexa Fluor 568 donkey anti-rabbit IgG(H + L)	Life Technologies	Cat#A10042; RRID:AB_2534017	1000 (IF)
Goat anti-Rabbit IgG HRP conjugated	R&D	Cat#HAF008	1000 (WB)
H3K27ac Monoclonal Antibody, ChIP-Verified	Invitrogen	Cat# MA5-23516	166 (ChIP; 3 µg)

^aIHC Immunohistochemistry, FC flow cytometry, IF Immunofluorescence, WB western blot, ChIP Chromatin Immunoprecipitation.

Flow cytometry

We harvested iPS cells with TrypLE and resuspended the cells in 1 mL of fluorescence activated cell sorting (FACS) buffer (2% BSA, 2 mM EDTA in PBS). The cell clumps were removed by passing through a cell strainer (70 μm). The individualized cells were collected by centrifugation. Cell pellets were resuspended in FACS buffer. About 2.0 × 10⁵ cells in 100 μL FACS buffer were incubated with a PE-conjugated antibody at 4 °C for 30 min. The cells were then washed three times with 3 mL FACS buffer each wash. Cell pellet was resuspended in 600 μL FACS buffer. Then, 10 μL 7-AAD (20 μg/mL stock) was added before assay with flow cytometer (LSRFortessa; BD Bioscience) equipped with BD FACSDiva™ acquisition software. Antibodies (see Table 1) used are PE-SSEA3 (#560237, BD Pharmingen), PE-SSEA4 (#560128; BD Pharmingen), PE-TRA-1-60 (#560193; BD Pharmingen), PE-TRA-1-81 (#560161, BD Pharmingen), PE-hAlkaline phosphatase (FAB1448P, R&D), PE-SSEA1 (#560142; BD Pharmingen), PE-mouse IgG1 (#555749, BD Pharmingen), and PE-mouse IgM (#555584, BD Pharmingen).

Western blot

Fibroblasts were cultured in 6-well plates at a density of 6.5 × 10⁴ cells/cm². After 24 h, the cells were transduced with lentiviral vectors for the knockdown or the overexpression of BRD2 (Addgene, Cat#, 197880), BRD3, and BRD4 (Addgene Cat#, 197882) as well as with a control vector which targets Luciferase or overexpresses GFP, respectively. Forty-eight hours after transduction, protein was extracted using RIPA buffer supplemented with 1× Halt^TM Protease Inhibitor Cocktail (Thermo Scientific, 78429). The protein concentration was assessed using BCA Protein Assay Kit (Pierce, Cat. 23225). Twenty micrograms of protein per sample were run using traditional Tris-glycine protein polyacrylamide gel system. After electrophoresis, the protein was transferred to PVDF membranes (Millipore, Cat. IPVH00010) using a Mini Trans-blot Cell (Biorad), set at 90 mA constant current for 16 h at 4 °C. Antibodies for Western blots can be found in Table 1. Target protein was visualized using Supersignal West Pico Plus (Thermo Scientific, Cat. 34580). Quantification of Western blots was carried out with ImageJ (version 1.54 g). Each band intensity is normalized to the corresponding β actin band. The normalized band intensity of each condition is then compared to controls (luciferase shRNA, empty vector, or Dox-).

Quantitative reverse transcription PCR (qRT-PCR)

RNA was first harvested from the monolayer cultures using TRIzol™ Reagent (Invitrogen, Cat. 15596026). After precipitation with 70% ethanol, RNA pellet was resuspended in 80 μL of 1× DNAse digestion buffer supplemented with 5 units of DNAseI (Thermo Fisher Scientific, Cat. AM2222). The reaction was incubated for 10 min at RT. The reaction was stopped and cleared using Quick-RNA Miniprep columns (Zymo Research, Cat. R1054). The RNA was eluted with 50 µL Nuclease-free water and stored at −80 °C. RNA concentration was determined using Nanodrop spectrophotometer (Thermo Fisher Scientific). RNA quality was assessed by agarose-gel electrophoresis and high-resolution electrophoresis. The cDNA was prepared from 1 µg of total RNA, using QuantiTec Rev Transcription kit (Qiagen, Cat. 205311). PCR reaction was prepared using 1× Kapa SYBR Fast kit master mix (Sigma-Aldrich, KK4602), 200 nM forward and reverse primers, and 1 ng/µL cDNA. PCR reactions were run on a ViiA 7 QuantStudio (Applied Biosystems) device and analyzed using QuantStudio Software v1.3 (Applied Biosystems). Primers used are: hBRD2f1, AAT GGC ACA AAC GCT GGA AAA; hBRD2r1, CAC TGG TAA CAC TGC CCTG; hBRD3f1, TGC AAG CGA ATG TAT GCA GGA; hBRD3r1, CAT CTG GGC CAC TTT TTG TAG AA; hBRD4f1, GAG CTA CCC ACA GAA GAA ACC; hBRD4r1, GAG TCG ATG CTT GAG TTG TGTT; hBRDTf1, GAT CAC GAA GTT GTG ACA ATG GC; hBRDTr1, CAA CAG GTT CAA TCG GGA TCTT.

RNA-seq and analysis

For RNA-seq on reprogramming cells of various conditions, we harvested total RNA 72 h post initiation of reprogramming (at the time of viral removal, i.e., overnight viral transduction). RNA-seq for BRD2 KO conditions additionally had two more samples at 48 h in addition to two samples at 72 h. Total RNA was isolated from cell cultures using Trizol followed by RNA clean-up using Quick-RNA Miniprep columns (Zymo Research) as described before¹⁸. RNA was eluted in nuclease free water and sent out for sequencing at Novogene. Library was prepared using the NEBNext Ultra II RNA Library Prep by Illumina. Human mRNAs were purified by oligo(dT) beads. Non-stranded protocol was used. The purified mRNAs were fragmented, and cDNA was synthesized using random hexamer primers and the M-MuLV Reverse Transcriptase (RNase H-). The second strand was subsequently generated by dNTPs, DNA polymerase I and RNase H. Double-stranded cDNA molecules were purified by AMPure XP beads and overhanging ends were repaired to blunt ends by exonuclease/polymerase. After 5′ phosphorylation and 3′ adenylation, the cDNAs were ligated with P5/P7 sequencing adapters to prepare for hybridization. To select the insert fragment of 150–200 bp in length, the modified libraries are purified with AMPure XP system (Beckman Coulter, Beverly, USA). Sequencing was conducted on Novaseq 6000 S4 Illumina by Novogene using a paired-end protocol (150 bp).

RNA-seq FASTQ files were mapped to human genome hg38 on LoneStar6 at Texas Advanced Computing Center (TACC) using STAR package (V2.7.11b)⁴⁹ and paired-end unstranded setting. We used human genome gene transfer file (GTF) of ensemble release 114 (Homo_sapiens.GRCh38.111.gtf). Reads were counted using the featureCounts package on LoneStar6. Read counts were normalized and quantified using the DESeq2 (V1.48.1)⁵⁰ R package with postcount algorithm for size factor estimation. We filtered out genes/transcripts that all samples had zero raw counts before executing the DESeq() function. We used default settings for DESeq() function except for the sfType argument, i.e., DESeq(dds, sfType = “poscounts”, test = “Wald”, fitType = “parametric”). We used padj < 0.05 rather than pvalue and ±log2(1.5) as the significant levels for DEGs.

Functional class enrichment analysis of DEG and enrichment visualization

Our functional class enrichment analysis (FunCEA) has been described in detail⁵¹. Briefly, GO and pathway term analyses were performed using clusterProfiler (V 4.16)⁵² and ReactomePA (V 1.52)⁵³ R packages. GO term enrichment analysis was carried out using the enrichGO() function with pvalueCutoff = 0.05 and pAdjustMethod = “BH” using the GO.db (V 3.21) annotation database. Pathway enrichment analysis was conducted with the enrichPathway() function with settings of pvalueCutoff = 0.05 and pAdjustMethod = “BH”, using the reactome.db (V 1.92) knowledgebase. Detailed procedures can be found in our recently published protocol⁵¹. The functional enrichment results are then visualized using enrichplot or GOplot R packages. RNA-seq expression levels are visualized using Pheatmap, Vioplot packages, and the boxplot() function of the base Graphics R packages as detailed before^54,55. Volcano plot was prepared using the R package of EnhancedVolcano.

IGV visualization of RNA-seq

RNA-seq BAM files are sorted and indexed with samtools (v1.20) sort and index command, respectively. The sorted BAM files were then converted into bigwig files using bamCoverage command from deeptools (v3.2). The read counts are normalized to RPKM using the bamCoverage argument of “--normalizeUsing = RPKM”. The resulting bigwig files were used for visualization of read coverages as well as splice junctions with IGV (V2.17.4) in reference to human GRCh38/hg38.

Plasmids and shRNA knockdown

Plasmids for the overexpression of BRD2, BRD3 and BRD4 were generated by cloning blunted, PCR amplified ORFs into our lentiviral vector pLVH-EF1a-AcGFP-2SmaI (Addgene, 258553) ¹⁷. Inducible BRD2 shRNA was designed using DSIR siRNA design tool^56,57 and cloned into an inducible TRIPZ vector (Horizon) shRNA system^58,59. Transient lentiviral shRNA knockdown constructs were purchased from Sigma (Mission® shRNA) or cloned in-house and verified in the authors’ lab (Supplementary Data 2). We purchased only the shRNAs that have been verified by the vendor (Sigma clone IDs, TRCN0000006310, TRCN0000021374, TRCN0000021375, TRCN0000318771, and TRCN0000318773.) We further verified their knockdown in our own cell type (fibroblasts). Cell types for shRNA verification are reported in Supplementary Data 2. The lentiviral control shRNA is MISSION^® pLKO.1-puro Luciferase shRNA Control Plasmid (Sigma-Aldrich, Cat#, SHC007). Gene knockdown by shRNA was achieved in reprogramming cells via lentiviral transduction.

CRISPR-mediated gene disruption

BJ cells were transduced with the pCW-Cas9 doxycycline-inducible lentiviral vector, and positive cells were purified by the addition of 0.5 mg/mL puromycin for 5 days. For disrupting BRD2 gene, combinations of validated sgRNAs were used, BRDN0001487068 (Addgene, Cat. #77983), BRDN0001148966 (Addgene, Cat. 77985), and BRDN0001145548 (Addgene, Cat. #, 77986)⁶⁰ (Supplementary Data 2). Each sgRNA was packaged into lentiviral particles, concentrated, and transduced overnight into Cas9⁺ BJ cells. Twenty-four hours after transduction, medium was replaced and supplemented with 1 μg/mL doxycycline. Doxycycline-mediated Cas9 induction was maintained for 24 h.

ChIP-seq

BRD2 ChIP-seq experiment was performed by Active Motif (FactorPath service) using the anti-BRD2 antibody A302-582A (Bethyl-Fortis) on naïve human fibroblasts. For the histone mark H3K27Ac ChIP-seq, we used antibody MA5-23516 (Invitrogen) and SimpleChip Kit (Cell Signaling Technologies, Cat. #9002) following manufacturer guidelines. Briefly, BJ cells were seeded into six 150-mm dishes and cultured in expansion medium to obtain around 2 × 10⁷ cells per experiment. Cells were treated with formaldehyde at a final concentration of 1% for 9 min. Crosslinking was quenched with glycine at a final concentration of 125 mM. Cells were scraped and pelleted by centrifugation at 750 × g and washed with cold PBS. Cell pellets were processed using SimpleChip Kit (Cell Signaling Technologies, Cat. #9002) following manufacturer guidelines. Chromatin extracted from 5 million cells were used for each ChIP. For reverse-crosslinking, immunoprecipitated samples were incubated at 65 °C overnight in the presence of NaCl and Proteinase K. DNA was isolated using Chip DNA Clean & Concentration (Zymo Research, Cat. D5201). Sample DNA was eluted in 20 µL nuclease-free water and stored at −20 °C until use. Fragment size was assessed using high-performance electrophoresis. All samples were sequenced using Illumina platform and single-end protocol.

ChIP-seq analysis

FASTQ files were mapped to the human genome GRCh38 with Bowtie2 using single end setting with local alignment. BAM files were filtered by read mapping quality (MAPQ > 20). Peak calling was performed using MACS2 package. Overlapping peaks and peak annotations were performed using ChIPpeakAnno package⁶¹. Only overlapping peaks from two independent experimental replicates were used for the analysis. Peaks with a confidence score above the 75 or 95 percentiles were regarded as strong interactions. Peaks were annotated using transcript annotations set for all possible isoforms present in the Ensembl database through EnsDb.Hsapiens.v86 package (v.2.99.0). Peaks within 2 kb upstream and downstream of the TSS were annotated. Annotated peaks were analyzed for pathway and Gene Ontology enrichment using clusterProfiler package (v.4.10.1) with an adjusted p < 0.05. Peak density heatmaps were generated using Deeptools⁶². Peak coverage was visualized using integrated genome browser (IGB)⁶³.

Regulon analysis

The table of regulons was obtained from Dorothea (v.1.14.1) collection, sub-setting for the confidence groups A, B, and C but excluding the lower confidence groups D and E. The list of all human transcription factors (hg38 genome) was obtained from Stein Aerts laboratory’s data resource for the Scenic package⁶⁴. Venn diagrams were performed using ggVennDiagram package (v. 1.5.7). Network visualization was performed using ggraph package (v.2.2.2).

Statistics and reproducibility

Statistical analysis for reprogramming data was conducted assuming normal distribution and equal variance based on Shapiro-Wilk’s test and Levene’s test, respectively. For pairwise comparisons, student’s t test was conducted. For multiple comparisons, ANOVA was used, and adjusted significance scores were calculated based on Tukey’s post-hoc test. At least three biological replicate experiments were performed.

For genomic experiments (RNA-seq, ChIP-seq) at least two biological replicates were used for each treatment. RNA-seq statistics were conducted with DESeq2 R package, which uses negative binomial generalized linear model (GLM) to model the raw count data⁵⁰. ChIP-seq statistics were conducted with MACS2 Python package that uses Poisson distribution to model the background read count and detect enriched genomic regions or peaks⁶⁵. Biological significance of enriched terms obtained from RNA-seq, and ChIP-seq data was calculated using adjusted p-values (Benjamini-Hochberg correction) of less than 0.05.

Reporting summary

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

Author contributions

K.H. conceptualized and supervised the project, acquired funding, designed experiments, analyzed and interpreted the data, conducted bioinformatics and data visualization, wrote the manuscript. R.Z. first demonstrated that BRD2 shRNA knockdown enhanced hiPSC reprogramming. R.R.C. confirmed R.Z.’s finding of BRD2 as a reprogramming barrier and further found that lipid deficiency caused by BRD2 is a reprogramming barrier; designed experiments, analyzed data, conducted bioinformatics and data visualization, and wrote manuscript. S.G.C. provided lab support and wrote manuscript. R.S. conducted experiments, analyzed data and prepared some Figures.

Peer review

Peer review information

Communications Biology thanks Danyun Zhao, Androniki Kretsovali and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Shani Stern and Kaliya Georgieva. A peer review file is available.

Funding

This research has been supported mainly by NIGMS/NIH (R01GM127411 to KH), and partly by American Heart Association (17GRNT33670780 to KH), and UNT Health Science Center (to KH), as well as NINDS/NIH (R01NS118760 to SGC).

Data availability

RNA-seq and ChIP-seq data were deposited at the Gene Expression Omnibus under accession number of GSE304893 without any restriction for access upon publication of this manuscript. There is no clinical patient data in this study. SREBF1 ChIP-seq data were retrieved from GSE31477 (SRR502659; SRR502658; SRR502660; SRR353506)⁶⁶. Source data are available in Supplementary Data 1–3. We have deposited plasmid sequence SnapGene files and maps to Addgene along with plasmids as listed below (plasmid name, Addgene number): pLKO.1-puro-shBRD2, 258555; pTRIPZ-EC-hBRD2shRNA, 258560; pLVH-EF1a-AcGFP-P2A-SCD, 259305; pLVH-EF1a-AcGFP-P2A-HMGCR, 259306; pLVH-EF1a-AcGFP-P2A-FLAG-hBRD2^ET, 259307; pLVH-EF1a.AcGFP-P2A-2SmaI, 258553.

Code availability

We did not develop any specialized software or original codes in this study. We, however, have published detailed tutorials for RNA-seq analysis⁶⁷, generating heat maps⁵⁵ and boxplots⁵⁴, as well as for conducting biological functional class enrichment analysis (FunCEA) and visualization of the FunCEA results⁵¹. Application scripts and commands based on available open-source (R and Linux) software for those purposes have been detailed in tutorials mentioned and cited above.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10691-1.