Clustered κB sites with wide range of affinities are present in promoters of NF-κB target genes and they collectively dictate RelA recruitment to DNA
We examined the relationship between ChIP-seq scores of RelA, representing its occupancy to DNA in the cell, and sequence of κB sites within the promoters of over 100 NF-κB-regulated genes that are stimulated by TNFα within a 30-minute stimulation period, as reported by Ngo et al (Ngo et al., 2020). The sites are designated as κB sites if they matched any of the 1800 sites with a measurable Z-score derived from PBM analysis (Siggers et al., 2011). These κB sites display a maximum deviation at three positions from the 10-bp consensus. Notably, we observed the presence of multiple weak κB sites surrounding a strong κB site within a 500 bp window centered around the ChIP-seq peak for most TNFα-responsive genes (Fig. 1A). These promoters can be separated based on their cumulative RelA Z-scores. The top and bottom promoters of ranked genes show a significant difference in terms of the number of κB sites (Fig. 1B). The ChIP-seq peak aligned to a strong κB site in nearly all cases. Intriguingly, we observed only a weak correlation between the ChIP-seq score and Z-score of the strong κB site (Supplementary Fig. 1A). This suggests that the affinity of an individual strong site is insufficient in dictating occupancy of RelA dimers to promoter regions in cells. However, we observed a stronger correlation of ChIP-seq strength to the number of probable κB sites surrounding the ChIP-seq peak site (Fig. 1C).
Although we analyzed a window of 500 bp around the peak, clusters of κB sites spanned mostly about 200 bp, which is similar to the span of open chromatin observed within promoters and enhancers during transcription (Fig. 1C and Supplementary Fig. 1A). We further analyzed the arrangement of κB sites in promoters of Cxcl1, Cxcl2, Nfkbid, and Tnfaip3 (Fig. 1D and Supplementary Fig. 1B). Both Cxcl1 and Cxcl2 share an identical strong site (GGGAAATTCC) but the Cxcl2 promoter displayed a lower overall ChIP score compared to Cxcl1. This may be a reflection of the Cxcl1 promoter containing a higher number of weak κB sites than Cxcl2 (13 vs 8), and the cumulative Z-score of its weak sites for RelA dimers is also higher than that of Cxcl2 (70.6 vs 58.5). Some of these weak sites exhibiting minimal resemblance to the κB consensus, such as GGATTTTTT and GGGAATTTTA (deviation from consensus underlined), are found near the strong site in Cxcl1 and Tnfaip3 promoters. To assess if these putative weak sites engage to RelA dimers, we measured affinities of selected κB sites for both NF-κB RelA homodimer and p50:RelA heterodimer (of wild-type full-length proteins) at a near-physiological salt concentration by biolayer interferometry (BLI) assays. Both NF-κB dimers bound these putative sites with a much-reduced affinity when compared with binding to IFN-a: GGGAAATTCC, and IFN-g: GGGAAGTTCC sites as references of strong and weak κB sites. For most weak sites, the Kd could not be determined accurately, although minor binding was detected at higher protein concentrations. Two sites did not display any affinity even at the highest protein concentration (Fig. 1E and Supplementary Fig. 1C) and estimated to have Kd > 1500 nM. Since the nuclear concentration of RelA upon TNFα stimulation is around 230 nM as determined here (Supplementary Fig. 1D), only about 6% of these dimers are expected to bind at a Kd of 3000 nM. These findings raised questions on the authenticity and functional relevance of these putative sites in vivo. Another important aspect is the mismatch between binding constants measured with native proteins at near physiological ionic strengths using solution-based methods and truncated proteins under non-physiological ionic strength using microarray-based assay. Furthermore, we qualitatively examined the binding strengths of weak and strong sites within Cxcl1 and Cxcl2 promoters using EMSA (Supplementary Fig. 1E). EMSA with weak and strong sites within Cxcl1 and Cxcl2 promoters indicated that mutation of both κB sites (M5) abrogated binding of NF-κB, while sequence- specific binding at weak κB sites (M4) was retained upon mutation of strong sites. These findings led us to conclude that these two weak sites are indeed putative κB sites as they both displayed weak binding under low salt (i.e., EMSA) conditions.
To evaluate the authenticity of specific binding by NF-κB dimers to weak κB sites, we crystallized the homodimer of RelA in complex with a weak κB site (-5G-4G-3G-2C-1T0T + 1T + 2T + 3C + 4C) present in the promoter of the Cxcl2 gene (the underlined bp are non-cognate and 0T bp denotes the pseudo-dyad axis separating the two DNA half sites). Remarkably, the crystal structure revealed a similar binding interaction as observed with consensus κB sites (Fig. 1F). The Arg33 and Arg35 engaged the − 4G-3G of half-sites as predicted. However, the presence of non-cognate − 2C-1T compared to -2A-1A (cognate bp for the RelA subunit) influences the global DNA architecture such that Arg187 is slightly differently positioned, but more importantly, the R41 could not engage with − 5G of the weak site. This altered protein-DNA contact corroborated weaker affinity, but nonetheless indicates specific binding. In summary, these results reveal specific functional involvement of weak κB sites within clusters found in promoters of NF-κB regulated genes.
Weak sites within κB clusters drive NF-κB-induced transcription
We investigated the potential of weak κB sites in transcriptional activation of target genes in cellular milieu. Using CRISPR, we selectively deleted strong and its neighboring weak κB sites, independently, in the promoters of five NF-κB target genes — Cxcl1, Cxcl2, Nfkbid, Map3k8, and Tnfaip3 (Fig. 2A). A control was also established by deleting a random segment away from any κB site. Sequencing of bulk populations revealed deletions averaging around 4–10 bp (Supplementary Fig. 2A). These pooled lines were grown, and the transcript levels of target genes were measured by quantitative PCR (qPCR) upon treatment with TNFa (10 ng/ml) for 1 h. We observed a significantly reduced level of transcript in all tested genes upon deletion of both the strong and weak sites (Fig. 2B and Supplementary Fig. 2B). The weak sites of different genes were different in sequence, orientation, and relative distance from the strong site. Chromatin immunoprecipitation (ChIP) assays using a specific antibody against RelA with the wild-type and mutated Cxcl2 promoters indicated a significant reduction of RelA ChIP signal upon deletion of κB sites, weak or strong, compared to the control site deletion (Fig. 2C). These results support the critical function of weak sites in NF-κB RelA recruitment and transcription.
To further validate the authenticity and role of these weak sites in transcription, we constructed luciferase reporters containing promoter segments of Cxcl1 and Cxcl2 genes (as used in EMSA). Consistent with what was observed in the native context, mutation of either the strong or the weak site led to a drastic reduction in reporter expression. This indicates that the presence of both κB sites of Cxcl1 or Cxcl2 promoters is critical for optimal reporter expression i.e., a functional synergy between the sites in activating transcription (Fig. 2D). To explore the synergy between multiple κB sites further, we assessed reporter expression from synthetic promoters containing combinations of strong and weak κB sites with different spacing and orientations. We not only observed two sites to act synergistically but also observed a strikingly robust reporter activity from two weak sites as observed with two strong sites (Fig. 2E). Reporter activity decreased progressively as the spacing between the sites increased to 100 bp. Overall, these observations suggest a strong activating influence of clustered κB sites, even if weak, in gene activation – outperforming a single strong site.
A multitude of nuclear proteins associate with signal induced NF-κB bound to κB DNA
The apparent weakness of κB sites found in promoter regions of RelA regulated genes raises the question not only about their authenticity but also about the presence of other hidden cooperative or synergistic processes leading to effective recruitment of RelA dimers to these sites in vivo. The abundance of a multitude of nuclear factors congregating with the nucleosome indicates the possibility of additional nuclear factors influencing recruitment of RelA in cells. We explored this with a simplistic strategy of pulldown of factors from nuclear extracts of TNFa-stimulated MEF cells with Cxcl2 promoter DNA fragment as described earlier for EMSA and reporter expression assay. First, we confirmed the specificity of pull-down by testing binding of RelA to the wild-type (WT) but not to the mutant (M5) promoter fragment via Western blotting (Supplementary Fig. 3A). Subsequently, we incubated with DNA with MEF cell nuclear extract post- 0.5 and 3 h stimulation with TNFa, and performed mass- spectrometry analysis of associated proteins. The difference between WT and M5 (both κB sites are deleted) DNA served in selecting RelA-κB-site specific proteins, and we identified several nuclear proteins (Fig. 3A, Supplementary Table 1) that included other transcription factors (TFs), DNA repair proteins, RNA-binding proteins, and metabolic enzymes. Among the TFs, various members of the NFAT family were prominent, and members of TEAD, SMAD, and CUX families were also represented. RNA binding proteins belonging to the SR family splicing factors and hnRNPs were also abundant.
To investigate the possible uniqueness of nuclear factors associating with RelA bound to either specific promoters or at different times post-induction, we performed pulldown experiments also with Ccl2 promoter DNA fragments using nuclear extracts of MEF cells 0.5 and 3 h post-TNFa stimulation (Fig. 3A). A largely similar set of proteins engaged to Cxcl2 and Ccl2 DNA fragments, with the exception of a few promoter-specific factors. However, factors associated were markedly different at 0.5 h vs 3 h post-TNFa stimulation, suggesting temporal specificity. Pathway analysis of pulled down proteins revealed transcriptional regulation, NF-κB signaling, and NFAT signaling among the most significantly enriched pathways (Fig. 3B).
Further investigation using bone marrow-derived macrophage (BMDM) cells treated or
untreated with LPS for 1 h corroborated these findings. Notably, NFAT factors pulled down with NF-κB in TNFa-treated MEFs were also identified in LPS-treated BMDM pulldowns. Three main protein classes were consistently identified: DNA-binding TFs, RNA-binding proteins, and various enzymes (Fig. 3C and Supplementary Table 1). Additionally, we observe the NF-κB and NFAT family members identified under different pulldown conditions (Fig. 3D). In all pulldown experiments, we also identified a large number of proteins bound to all DNAs under all conditions, indicating they were bound non-specifically.
We focused on five TFs, Nfatc1, Nfat5, Tead3, Cux1, and Smad4, that have their own transcriptional programs by binding to specific DNA sequences. In this case, we set out to investigate if these factors play a role in RelA’s recruitment to specific promoters and induction of transcription. We first tested if these TFs interact with RelA. To elucidate their interaction between RelA, we co- expressed them as FLAG-fusion proteins with HA-tagged RelA, followed by co-immunoprecipitation in HEK 293T (Supplementary Fig. 3B). Results indicated association between RelA and all five factors to various extents, suggesting their potential roles in sequestering RelA in the nucleus or supporting its DNA binding. Among these proteins, Nfatc1 interacts with RelA strongest and is able to pulldown endogenous RelA efficiently (Supplementary Fig. 3C). Moreover, Nfatc1 appeared to localize in the nucleus along with RelA in response to TNFa (Supplementary Fig. 3D). Using purified p50:RelA heterodimer, Flag-Nfatc1, and Flag-Tead3, we also observe that Nfatc1 and Tead3 can bind with sequence specificity to the Cxcl2 promoter DNA at overlapping regions of p50:RelA, supporting the potential for their association with NF-κB on DNA (Supplemental Fig. 3E). We term these RelA-interacting TFs as "TF-cofactors (TF-CoFs)," and propose that they, along with other identified factors, collectively facilitate RelA recruitment to promoters containing κB sites. In essence, we identified numerous cofactors temporally associating with RelA and potentially other NF-κB factors on promoter DNA fragments harboring κB sites.
TF-cofactors support the promoter recruitment of NF-κB RelA
To assess the role of cofactors in regulating DNA binding by NF-κB RelA, we investigated whether RelA’s binding to the κB sites is influenced by these five TF-cofactors: Nfatc1, Nfat5, Tead3, Cux1 and Smad4. Knockdown (KD) cell lines for Nfatc1, Nfat5, Tead3, Cux1, and Smad4 were generated using shRNA directed towards each cofactor in MEFs. Scrambled KD (scr-KD) MEF cells were used as the WT control. KD efficiency was validated by measuring protein levels, showing a reduction of approximately 60–80% compared to scr-KD control cells (Supplementary Fig. 4A).
To examine how the TF-CoFs affect the genome-wide binding pattern of RelA, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) with a RelA-specific antibody at TNF stimulation of 0 and 30 min in scr-KD and five TF-CoF KD MEFs (Fig. 4A). Bioinformatic analysis of the replicate datasets (p-value < 1E-5) identified 9096 RelA peaks that were TNF-inducible at 30 m in these samples (Fig. 4B, left). Consistent with previous reports, most of the RelA binding sites in scr-KD cells were in intergenic regions (40.9%) or gene bodies, with only a small percentage (~ 11%) in the promoter region (defined as -1 kb to + 100 bp of the TSS) (Fig. 4B, right). In each KD MEF line, defects in RelA binding were observed in many promoters, but binding was also enhanced in some cases. Analysis of overall RelA binding at RelA-induced peaks revealed that Nfat5 and Cux1 depletion had the greatest defects in RelA recruitment, followed by Nfatc1 and Smad4 (Fig. 4C, Supplemental Fig. 4B). RelA binding was reduced only to a few promoters upon Tead3 depletion. Interestingly, to a significant fraction, Tead3 depletion enhanced RelA recruitment suggesting Tead3 acts as a repressor of RelA. Nfatc1 and Nfat5 depletion exhibited a common pattern of RelA recruitment change suggesting they work together. Normalized RelA ChIP-seq density over all identified peaks clearly showed Tead3 had minimal effect, Cux1 had maximal effect, and the rest fell in between. Overall, a significant portion of RelA binding sites were affected in at least one of the KD cells, with only a few promoters affected by all five KDs. Genome browser tracks of two known NF-κB target genes, Rgs19 and Ccl8, present peaks that were high in scr-KD cells that become reduced in both Nfatc1 and Nfat5 KD cells, but not in Cux1 KD cells. However, we observe reduced binding at CD44 promoter upon Cux1 KD (Supplementary Fig. 4C). Further, we observe that NF-κB from TNF- stimulated nuclear extract of Nfatc1-KD cells binds less to Cxcl2 promoter DNA relative to control scr- KD nuclear extract (Supplemental Fig. 4D).
Since the chosen cofactors are all DNA-binding TFs, we investigated if they influence RelA's DNA binding by binding to their own DNA response elements near or overlapping a κB site. Motif analysis revealed that consensus κB sites were the primary TF binding sites (log p-value = -2.8E4) in control scr-KD MEFs (Fig. 4D). Except for Tead3, the consensus sequences of other cofactors were not enriched at cross-linked sites. However, given the overlap between Tead3, Nfats, and NF-κB binding sites, it's challenging to ascertain if these factors bind directly to κB sites. Hence, we focused on Cux1 and Smad4 binding sites and found no enrichment of their DNA binding sites near RelA/NF- κB binding sites. This suggests that cofactors primarily, if not exclusively, act through protein-protein interactions rather than direct DNA binding. Overall, these results suggest that each of the tested cofactors supports RelA recruitment only to a select set of promoters, but also impacts negatively on other promoters.
TF-cofactors regulate gene expression by RelA
To determine if reduced levels of these five TF-cofactors and altered RelA recruitment to promoters correlate with transcript levels of genes that are regulated by RelA, we treated all five cell types used in the ChIP experiments, as well as scramble-KD control MEFs with TNFa for 1 hr, followed by genome- wide bulk RNA sequencing. The RNA isolated from untreated cells served as controls. RNA-seq analysis of triplicate samples identified hundreds of genes that were differentially expressed in these five cell types when compared to control cells (p-value < 0.05; Fig. 5A). The heatmaps display transcript levels in scr-KD (WT) and five KD cells, with genes arranged from highest to lowest expression in induced scr-KD cells (Fig. 5A). Unlike the shared RelA ChIP pattern between the Nfat family members, expression patterns did not strongly match. Interestingly, Smad4-KD cells exhibited expression defects in the largest number of genes, despite their unremarkable role in RelA recruitment compared to the other cofactors. Tead3 KD cells showed higher transcript levels for many genes with higher RelA ChIP scores. Examples of affected genes by knockdown of all five factors are shown in Fig. 5B. Expression of most of the cytokine genes was impacted by one or more of these TF-CoFs 8
tested. IL6 expression was affected by all five KD cells, whereas Nfkbia was not by any of them. Each KD affected distinct, common, and combinatorial sets of genes, suggesting that RelA-regulated promoters follow distinct rules.
We then sought to understand the correlation between transcript levels and ChIP-seq scores for each KD. Excluding Tead3, we found that reduced ChIP scores generally correlated with reduced transcript levels in the other four cases (Fig. 5C and Supplementary Fig. 5A). As shown in Fig. 5C, a greater number of genes (105/209) with reduced ChIP-seq score also had lower transcript levels in Nfatc1 KD cells. However, some genes showed higher transcript levels despite higher (20/209) or lower (15/209) ChIP-seq scores. Nfatc1 KD also enhanced ChIP scores of several genes (89/206), of which 20/89 showed higher expression. In some of these cases, the lack of correlation may be due to direct binding of the TF-CoFs to promoters acting as legitimate TFs by antagonizing RelA binding and imposing their own transcriptional regulation patterns. This was evident in Tead3-KD MEFs where more promoters exhibited higher RelA ChIP scores and transcript levels consistent with its repressor activity by binding to cognate DNA motifs (Fig. 5C). We next examined whether KD of these cofactors affected gene clusters with common biological functions but found no specific biological functions for these cofactors. Overall, we found a lack of correlation between RelA ChIP and transcript levels in control or any KD cells except for a set of genes under KD conditions, which is consistent with previous reports as RNA levels are strongly influenced by other factors post-transcriptionally (Ngo et al., 2020). These results imply a complex relationship between TF-CoFs and RelA in terms of RelA recruitment, subsequent transcription, and perhaps post-transcription.
Clustered κB sites in the Cxcl1 promoter and enhancer and TF-cofactors facilitate transcription
We next tested how clustered weak κB sites and TF-CoFs support NF-κB binding to DNA, using the regulatory region Cxcl1 gene as a model. We looked for the presence of clustered κB sites within the 500 bp region described in Fig. 1A. A 150 bp segment contained at least five κB sites including the strong and a weak site tested earlier with the strong κB site located 57 bp upstream of TSS. In addition to this promoter of Cxcl1, RelA ChIP revealed its binding to a region approximately 15 kb upstream of the Cxcl1 TSS. We assumed that this segment of DNA might be an enhancer of Cxcl1 (Supplementary Fig. 6A). This segment also contains four weak κB sites and a strong site (GGGATTTCCC; Z-score 8.0) within a ~ 150 bp segment (Supplementary Fig. 6A).
To determine if these weak κB sites in the promoter and putative enhancer of the Cxcl1 gene are functional, we used the luciferase reporter assay on wild type (WT) and several mutant constructs, as depicted in Fig. 6A where the enhancer (E) and the promoter (P) were placed with 0.7 kb spacing. The WT construct, designated as EWT-PWT-TSS exhibited ~ 20-30-fold higher reporter activity compared to the construct where all κB site were mutated (EMT-PMT-TSS) (Fig. 6B). The removal of all putative κB sites only within the enhancer (EMT-PWT-TSS) reduced expression by 70%, whereas the removal of all κB sites within the promoter (EWT-PMT-TSS) reduced reporter expression by 90%. We also tested the effect of a DNA fragment that contains a single strong κB site in the promoter (EMT-PSSS-TSS) and a single strong site both in the promoter and enhancer (ESSS-PSSS-TSS). The single strong site in the promoter (EMT-PSSS-TSS) resulted only in a 2-fold luciferase activity, whereas strong sites both in the enhancer and promoter (ESSS-PSSS-TSS) led to a 3.5x increase, which is about 10% of the activity of the EWT-PWT-TSS DNA. These results suggest that the distal RelA binding region is likely an enhancer and that the weak κB sites potentiate gene activation by efficiently recruiting RelA (Fig. 6B).
To verify the role of TF-cofactors in transcription activation, we verified reporter expression in the presence of each of the five TF-cofactor by overexpressing them. Overexpression of Cux1 exhibited dramatic induction of reporter expression. Overexpression of Nfat5 also showed higher reporter expression. However, Nfatc1, Smad4 and Tead3 showed no effect in reporter expression (Fig. 6C). RelA ChIP score in control and TF-cofactor KD HeLa cells showed reduction of ChIP score in all but one (Tead3) compared to the control. It is possible that we used truncated Cux1 and Nfatc1 constructs which possibly act as a potent activator or repressor, respectively, due to the removal of specific domains. Collectively, these results support the notion that specific TF-cofactors act on specific enhancer/promoter.
TF-cofactors and clustered κB sites are essential to form transcriptional condensates
Visualization of transcriptional complex formation, both in vitro and in vivo, has provided important insights into the roles of various factors in the assembly process (Cho et al., 2018; Hnisz et al., 2017; Sabari et al., 2018; Shrinivas et al., 2019). We performed phase-separated condensate formation assays in vitro to assess the role of weak κB sites and TF-CoFs in the assembly of transcription complexes mediated by RelA at the transcription site. We PCR amplified the DNA fragment encompassing the enhancer, promoter and transcription start sites (EWT-PWT-TSS) as present in the luciferase reporter constructs with a Cy3 labeled primer, generating ~ 1200 bp labeled fragments. Additionally, we amplified mutant DNA with all κB sites mutated (EMT-PMT-TSS) and DNA containing only the strong enhancer and promoter κB sites (ESSS-PSSS-TSS) (Fig. 6A). Recombinant GFP- RelAFL was mixed with HeLa nuclear extract to monitor condensate formation with Cy3-DNA. We verified that GFP-RelAFL binds DNA, albeit weakly compared to wild-type RelA (Supplementary Fig. 7A), and found that as high as 2 µM purified recombinant RelA-GFP failed to condense in the absence or presence of DNA under the buffer condition used (150 mM NaCl, 25 mM Tris-HCl 7.5, 100 ng/µL sonicated salmon sperm DNA, 1 mM DTT, 100 ng/µL sonicated salmon sperm DNA, and 2.5 mM MgCl2) (Fig. 7A & Supplementary Fig. 7B). However, 100 nM RelA-GFP condensed with DNA (30 nM) in the presence of 1.0 mg/mL TNFa activated HeLa nuclear extract (NE) in this buffer. The reaction mixture was imaged for 20 min with 30 sec image intervals. We looked at the condensates both along the z-axis at the middle of the drop and at the bottom on the cover slip. We noted fewer condensates in the middle of the drop above the cover slip in all cases compared to those on the cover clip (Supplementary Figs. 7C and 7D). It is likely that condensates form throughout the drop but settle on the cover slip. In both cases, however, we observed increasing condensate intensity over time and found DNA (red) and RelA-GFP (green) were merged in > 90% of these condensates (Fig. 7A and Supplementary 7C). A few condensates were found in control (EMT-PMT-TSS) and the number was enhanced in ESSS-PSSS-TSS DNA. Qualitative estimation of condensate abundance at 20 or 12 10
min showed that significantly higher number of condensates were formed with the WT DNA (EWT-PWT- TSS DNA) than with the mutant DNAs (EMT-PMT-TSS and ESSS-PSSS-TSS) (Fig. 7B and Supplementary Fig. 7D). These results indicate that weak κB sites are crucial for recruiting RelA to the enhancer-promoter-TSS DNA, forming the transcription complex. We further verified if DNA condensates from the nuclear extracts containing endogenously activated RelA in the absence of exogenously added RelA-GFP. Indeed, EWT-PWT-TSS DNA, but not EMT-PMT-TSS DNA, formed condensates (Supplementary Fig. 7E).
We determined RelA-GFP concentration in condensates relative to its concentration in the bulk based on the fluorescence intensity. From a standard curve of increasing concentrations of pure RelA- GFP, the concentration of RelA-GFP in the condensate was estimated to be around 450 nM which is 4.5-fold over the bulk concentration of 100 nM (Supplementary Fig. 7F). It should be noted that the concentration of unlabeled endogenous RelA is also enhanced. The combined concentration of RelA could be as high as 1 µM in the condensates. This modest enhancement of RelA concentration in the condensates likely results in RelA binding to most κB sites at greater occupancy. Altogether, these findings strongly suggest that cofactors and clustered κB sites in enhancers and promoters cooperate to recruit RelA to the transcription sites of target genes by enhancing RelA’s concentrations as depicted in Fig. 6D.
We further investigated the formation and dynamism of stimulus-induced nuclear RelA condensates in live primary MEF cells expressing mVenus-RelA fusion protein. These cells are derived from the mVenus-RelA reporter mouse (RelaV/V) in which an mVenus-RelA fusion is knocked in to the endogenous Rela locus (Adelaja et al., 2021; Cheng et al., 2021). These cells were treated with LPS for 1 hr followed by imaging (Fig. 7C). We found nuclear puncta only after stimulation (Fig. 7C). We performed FRAP (fluorescence recovery after photobleaching) experiments to determine the residence time of RelA in these puncta in MEFs treated with LPS for 1 hr followed by bleaching a cluster of puncta, followed by measuring the recovery rate by collecting images at 0.2 s intervals (Fig. 7D and 7E). As control, we monitored fluorescence change of another spot that was not bleached (Fig. 7F). We found the residence time to be ~ 3.6 s (Fig. 7D and 7E), which is close to the values determined previously where RelA was overexpressed and two different methods, FRAP and single molecule tracking (SRT) (Bosisio et al., 2006; Callegari et al., 2019). A host of other TF was also investigated for their residence time, and the values were mostly found to be ~ 10 s (Lu and Lionnet, 2021). These observations suggest that RelA forms dynamic transcriptional condensates likely at the transcription sites in cells. The mechanisms of the dynamic RelA assembly remain to be determined.