Disrupting the RNA polymerase II transcription cycle through CDK7 inhibition ameliorates inflammatory arthritis

Abstract

Macrophages are key drivers of inflammation and tissue damage in autoimmune diseases including rheumatoid arthritis. The rate-limiting step for transcription of more than 70% of inducible genes in macrophages is RNA polymerase II (Pol II) promoter-proximal pause release; however, the specific role of Pol II early elongation control in inflammation, and whether it can be modulated therapeutically, is unknown. Genetic ablation of a pause-stabilizing negative elongation factor (NELF) in macrophages did not affect baseline Pol II occupancy but enhanced the transcriptional response of paused anti-inflammatory genes to lipopolysaccharide followed by secondary attenuation of inflammatory signaling in vitro and in the K/BxN serum transfer mouse model of arthritis. To pharmacologically disrupt the Pol II transcription cycle, we used two covalent inhibitors of the transcription factor II H-associated cyclin-dependent kinase 7 (CDK7), THZ1 and YKL-5-124. Both reduced Pol II pausing in murine and human macrophages, broadly suppressed induction of pro- but not anti-inflammatory genes, and rapidly reversed preestablished inflammatory macrophage polarization. In mice, CDK7 inhibition ameliorated both acute and chronic progressive inflammatory arthritis. Lastly, CDK7 inhibition down-regulated a pathogenic gene expression signature in synovial explants from patients with rheumatoid arthritis. We propose that interfering with Pol II early elongation by targeting CDK7 represents a therapeutic opportunity for rheumatoid arthritis and other inflammatory diseases.

INTRODUCTION

Inflammation is a fundamental response to tissue stress, injury, or infection, which is geared toward eliminating infectious agents, repairing affected tissues, and restoring homeostasis (1). The complexity of the inflammatory response requires that its regulatory programs be tightly controlled and well coordinated. This is achieved through multilevel regulatory mechanisms that include alterations in the composition of immune cell populations in tissues, changes in cellular responses to inflammatory stimuli, modulation of signaling pathways, and changes in gene expression in both immune and stromal cells. Failure to establish a proper inflammatory response can compromise host defense and pathogen clearance.

Poorly controlled or chronic inflammation, however, is associated with tissue damage, which is a hallmark of numerous pathologies, including autoimmune conditions such as rheumatoid arthritis (RA), metabolic syndrome, and neurodegenerative diseases (2–4). A key cell type that drives inflammation and its resolution is macrophages (MΦ). In RA, a progressive debilitating disease of the joints that affects millions of people worldwide, MΦ infiltrate the synovium of affected joints, where they produce numerous cytokines, chemokines, and tissue-degrading enzymes, leading to cartilage and bone erosion (5). Although inflammation is regulated at multiple levels, the tightly coordinated transcription of hundreds of genes is a key regulatory mechanism ensuring a robust response that has evolved to help resolve infection but can contribute to tissue injury (6).

The transcription cycle of RNA polymerase II (Pol II) includes three sequential phases: initiation, elongation, and termination. Initiation involves recruitment of the general transcription factors TFIIA, TFIIB, TFIID, TFIIE, TFIIF, TFIIH [which includes the cyclin-dependent kinase 7 (CDK7)/cyclin H/Mat1 complex], and Pol II to the promoter (7). Shortly after initiation, the Pol II carboxyterminal domain (CTD) undergoes phosphorylation on S5 of its heptad repeats Y₁S₂P₃T₄S₅P₆S₇ by CDK7. Pol II recruitment and transcription initiation were historically considered the rate-limiting steps for signal-dependent activation of most genes. More recently, however, it has been recognized that transcriptionally engaged Pol II often pauses less than or around 100 nucleotides (nt) downstream of the transcription start site (TSS), effectively blocking new initiation, and that release from this pause is signal-dependent and rate-limiting for productive transcription of a majority of inducible genes in mammalian cells (8). Pol II promoter-proximal pausing is facilitated by the heterodimeric 5,6-dichloro-1-β-d-ribofuranosyl benzimidazole (DRB)–sensitivity inducing factor (DSIF) and the four-subunit negative elongation factor (NELF) complex (9–11), whose assembly into stably paused Pol II elongation complexes depends on the activity of CDK7 (12–15). In response to a stimulus, positive transcription elongation factor b (P-TEFb; a CDK9/cyclin T1 complex) is recruited and phosphorylated by CDK7 in its activation loop (12). The activated P-TEFb phosphorylates DSIF, NELF, Pol II, and other components of the elongation complex, leading to NELF dissociation, conversion of DSIF to a positive elongation factor, and entry of Pol II into productive elongation (16–18).

Pol II pausing is prevalent in transcriptional programs that acutely respond to signals, including developmental, environmental, or other triggers (19–21). In MΦ, Pol II recruitment to the TSS and transcription initiation appear rate limiting for some genes such as Il1a, Il1b, Cxcl10, Il6, and Ccl5 (22, 23). However, at numerous genes, including proinflammatory Tnf, Isg15, and Irf1; anti-inflammatory Zfp36 (whose product degrades mRNAs of inflammatory genes); nuclear factor κB (NF-κB) inhibitor Nfkbia; and Cited2 (encoding a CBP/p300-interacting transcription cofactor), Pol II is transcriptionally engaged and paused under resting conditions (22–24). On the basis of Pol II and NELF occupancy profiles, up to 80% of the inflammation-inducible transcriptome in MΦ is pause regulated (23, 24), a number generally consistent with that reported in other systems (25, 26). Despite the widespread Pol II pausing observed across cell types and biological processes, the specific contribution of postinitiation transcriptional control to inflammation in immune regulation and disease is unknown.

Here, we assessed the genome-wide transcriptional consequences of disrupting promoter-proximal pausing and early elongation through genetic and pharmacological means, by conditionally deleting NELF or inhibiting CDK7 in both mouse and human MΦ. We then evaluated the effect of CDK7 blockade in vivo using acute and chronic-progressive models of inflammatory arthritis in mice. Last, we investigated the impact of disrupting the Pol II transcription cycle through CDK7 inhibition in cells derived from synovial tissue samples of patients with active RA ex vivo.

RESULTS

CDK7 inhibition blocks the assembly of Pol II pausing complexes in mouse MΦ

To investigate the role of Pol II pausing complexes in inducible gene expression in MΦ, we first analyzed transcriptomes of murine bone marrow–derived MΦ (BMMΦ) from wild-type (WT) mice or those lacking the functional NELF complex in the myeloid lineage [NELF–conditional knockout (cKO)] (23, 24). Differential gene expression analysis revealed that lipopolysaccharide (LPS) treatment for 0.5 hours significantly up-regulated 54 genes in WT BMMΦ [false discovery rate (FDR) < 0.05, fold change (FC) from baseline >1.5; data file S1]; 36 of them were “superinduced” in NELF-deficient MΦ relative to WT MΦ (fig. S1, A and B). Although proteins encoded by these genes were functionally diverse, we noted an enrichment of those with anti-inflammatory activities, including Zfp36, Cited2, Nfkbia, phosphatase regulatory protein Ppp1r15a, transcription regulators Nr4a1 and Egr1, immediate early response gene and apoptosis/necrosis inhibitor Ier3, phosphatases Dusp1/2, inhibitor of cytokine signaling Socs3, and Tnfaip3/A20 (fig. S1A). Using our previously published chromatin immunoprecipitation with sequencing (ChIP-seq) dataset of Pol II from BMMΦ (23, 24), we ranked 36 superinduced genes by the pausing index, which quantifies the relative amount of Pol II at promoter versus gene body in untreated WT MΦ, and found that 8 of the 11 anti-inflammatory genes were paused (fig. S1A, pausing index > 1). Absolute expression of these genes was also significantly higher in NELF-cKO BMMΦ compared with WT BMMΦ (fig. S1C; logFC > 0.6, unadjusted P < 0.05). Extending LPS treatment to 1 hour led to further accumulation of these anti-inflammatory transcripts but largely erased the difference between the two genotypes (fig. S1, A and B).

Because steady-state transcription can reflect changes at multiple steps in gene regulation ranging from transcription initiation to mRNA processing and stability, we then directly assessed the impact of NELF loss on Pol II distribution in resting WT and NELF-deficient BMMΦ or those activated with LPS for 0.5 or 1 hour. At baseline, we observed no genotypic difference in Pol II occupancy over promoters or bodies of superinduced anti-inflammatory genes, including Zfp36, Dusp2, and Nfkbia (fig. S1D), suggesting that Pol II remained paused in NELF-cKO BMMΦ. Macrophage activation with LPS altered the Pol II distribution pattern in alignment with the RNA sequencing (RNA-seq) results: At 0.5 hour, Pol II occupancy throughout these genes was higher in NELF-cKO BMMΦ compared with WT BMMΦ, whereas, by 1 hour, it was again comparable between genotypes (fig. S1D). Quantification of the Pol II read density distribution for all 36 genes encoding superinduced transcripts revealed a similar pattern with comparable occupancy at baseline and more Pol II in both promoter-proximal regions and gene bodies in NELF-cKO BMMΦ compared with WT BMMΦ after 0.5-hour exposure to LPS (fig. S1E). Occupancy was again comparable after 1 hour of LPS exposure.

We reasoned that the transcriptional superinduction of anti-inflammatory genes in NELF-depleted BMMΦ may have longer-term consequences for downstream inflammatory signaling. By 6 hours, the concentration of the key inflammatory cytokines tumor necrosis factor (TNF) and interleukin-6 (IL-6) in the cKO culture medium was modestly but reproducibly reduced compared with that produced by WT BMMΦ (fig. S1F). To assess whether NELF deficiency in MΦ affects inflammation in vivo, we used the K/BxN–serum transfer (ST) model in which MΦ are known to play a critical role (27). In this model, sera isolated from K/BxN arthritic mice contain circulating anti–glucose phosphate isomerase antibodies that can be transferred to other strains to trigger a passive, self-limiting inflammatory arthritis (28). We found that disease severity, assessed by ankle thickness and arthritis clinical scores, was attenuated in NELF-cKO mice compared with WT mice (fig. S1G).

Given that NELF ablation in MΦ did not affect Pol II distribution at baseline, we next sought a way to instead completely block Pol II pausing complex assembly by using pharmacological means. Previous work uncovered a requirement for CDK7 activity in recruiting DSIF and NELF to establish the promoter-proximal pause in human cancer cells (12, 14, 29). Moreover, CDK7 has recently emerged as a promising therapeutic target in cancer, which prompted the development and characterization of several small-molecule covalent inhibitors of CDK7. Among them, THZ1 was first shown to inhibit CDK7 by docking in the adenosine 5′-triphosphate–binding site and covalently binding to the nearby C312 residue of the kinase domain (13, 30). THZ1, however, was reported to have off-target activity toward the related transcriptional kinases CDK12 and CDK13 (13). More recently, a more selective inhibitor, YKL-5-124, was developed, which retains C312 covalent binding ability but gains higher specificity by adopting the pyrrolidinopyrazole core from a series of CDK7 inhibitory compounds (31). The availability of CDK7 inhibitors provided a unique opportunity to decipher the role of Pol II pausing in inflammation beyond genetic manipulation of the NELF complex and to assess their potential therapeutic utility in inflammatory processes.

Exposing BMMΦ to either THZ1 or YKL-5-124 for 4 to 6 hours reduced occupancy of both NELF-E (by >90%; left) and Pol II (by 60 to 70%; middle), as assessed by ChIP–quantitative polymerase chain reaction (qPCR) at the TSS of several paused genes, namely, Tnf, Nfkbia, Cited2, and Zfp36, indicating reduced Pol II pausing (the relative amount of NELF to Pol II at each TSS is shown on the right panel) (Fig. 1A). We performed NELF-E ChIP-seq in BMMΦ to assess the effect of CDK7 inhibition on pausing genome-wide. Mapped relative to known genomic features, the majority of all NELF-E peaks in dimethyl sulfoxide (DMSO)–treated control BMMΦ (n = 10,655) localized near promoters [TSS defined as −500 to +200 base pairs (bp)] (Fig. 1B and data file S2). YKL-5-124 treatment resulted in an ~80% decrease in NELF-E peaks, with the majority of promoter-proximal peaks lost (Fig. 1B); these included NELF peaks at the anti-inflammatory genes Zfp36, Zfp36l2, Cited2, Ier3, Ppp1r15a, Dusp1, and Nfkbia as well as the proinflammatory gene Tnf (Fig. 1, C and D, and fig. S2A).

Fig. 1. CDK7 inhibition attenuates the assembly of Pol II promoter-proximal pausing complexes in BMMΦ.

(A) BMMΦ were treated with THZ1 or YKL-5-124 (250 nM each for 4 to 6 hours, as shown), and occupancy of NELF-E (left) and Pol II (middle) at the TSS of the indicated genes was assessed by ChIP-qPCR using signal with nonspecific IgG as background. NELF-E or Pol II signal in DMSO vehicle-treated BMMΦ was set to 1, and the NELF-E/Pol II ratio of means is shown on the right. n = 3; error bars are SEM; data were analyzed by Student’s t test; **P < 0.01; ***P < 0.001; ****P < 0.0001. (B to D) NELF-E ChIP-seq (n = 3) was performed in BMMΦ treated with YKL-5-124 or DMSO for 6 hours. (B) The upset plot shows total numbers of unfiltered shared (left column) or unique (middle and right columns) peaks (bottom) and their distribution (%) relative to the indicated genomic elements (top). Utr5, 5′ untranslated region. (C) The volcano plot shows the filtered NELF-E peaks down-regulated (left; logFC < −0.6, P < 0.05; n = 471) or up-regulated (right; logFC > 0.6, P < 0.05; n = 867) in YKL-treated BMMΦ. Promoter-proximal peaks (−500 to +200 nt relative to the TSS) are highlighted in red. Labeled in blue are peaks found near promoters of representative paused inflammation-related genes (gene_name location relative_to_TSS). (D) The read density distribution is shown for the indicated genes. (E) BMMΦ were treated with YKL-5-124 or DMSO for 6 hours with or without LPS for the last 0.5 hours, as indicated; occupancy of TFIIE (left) and Pol II (middle) at the TSS of the indicated genes was analyzed by ChIP-qPCR (n = 3) as in (A). Data are shown as means ± SEM; data were analyzed by Student’s t test; *P < 0.05; **P < 0.01; ***P < 0.001.

During establishment of the promoter-proximal Pol II pause, DSIF displaces TFIIE from their common binding site on Pol II (7, 12). Therefore, we reasoned that the diminished recruitment of NELF that we observed upon YKL-5-124 treatment might be due to the impaired exchange of TFIIE for DSIF when CDK7 activity is inhibited. To test this possibility, we evaluated the impact of CDK7 inhibition on TFIIE occupancy at the TSS of paused genes in resting BMMΦ or those activated with LPS for 0.5 hour. We detected TFIIE at the TSS of Tnf, Nfkbia, and Zfp36 in unstimulated BMMΦ, along with Pol II (Fig. 1E). Upon YKL-5-124 treatment, Pol II occupancy at these promoters declined by more than 50%, whereas the ratio of TFIIE to Pol II increased two- to fivefold (Fig. 1E), consistent with retention of TFIIE in a trend reciprocal to that observed for NELF (Fig. 1A). A brief exposure to LPS led to an approximately twofold increase in the occupancy of both Pol II and TFIIE at the TSS of paused genes, with little or no change in their relative amounts, compared with resting BMMΦ. However, pretreatment with YKL-5-124 attenuated additional Pol II binding while increasing occupancy of TFIIE, yielding a three- to sixfold increase in the ratio of TFIIE to Pol II (Fig. 1E). Furthermore, Pol II ChIP-seq analysis revealed a reduction of basal and LPS-stimulated promoter-proximal Pol II occupancy by YKL-5-124 across multiple paused genes (fig. S2B). Combined, our analysis of NELF, Pol II, and TFIIE occupancy suggests that CDK7 inhibition impairs establishment of the promoter proximal pause of Pol II in BMMΦ.

Blocking CDK7 reprograms the inflammatory transcriptome in mouse BMMΦ

Next, we assessed the longer-term impact of CDK7 inhibition on subsequent gene regulation by inflammatory stimuli. First, we pre-treated BMMΦ with THZ1 for 4 hours, followed by a 6-hour LPS treatment, and then assessed gene expression. LPS induced proinflammatory genes Tnf, Il1a, and Il6, as well as anti-inflammatory Nfkbia, Zfp36, and Il10. The induction of pro- but not anti-inflammatory genes was selectively suppressed by THZ1 pretreatment (Fig. 2A), resembling the effect of NELF deletion.

Fig. 2. CDK7 inhibition in BMMΦ precludes LPS-mediated up-regulation of proinflammatory genes and down-regulation of homeostatic and anti-inflammatory genes.

(A) BMMΦ were pretreated with THZ1 or vehicle for 4 hours followed by incubation with LPS for 6 hours (Tnf and Il1a) or 12 hours (the rest), where indicated. Expression of proinflammatory (left) and anti-inflammatory (right) genes was assessed by RT-qPCR and presented as an FC over that in unstimulated MΦ. n = 3 to 7; data are shown as means ± SEM. Data were analyzed by Student’s t test; *P < 0.05; **P < 0.01; ***P < 0.001. (B) BMMΦ were treated with LPS for 6 hours after a 6-hour pretreatment with YKL-5-124 or vehicle, and gene expression was analyzed by RNA-seq (n = 3). A volcano plot shows genes up-regulated (right; logFC > 0.6, unadjusted P < 0.05, n = 1482) or down-regulated (left; n = 897) by YKL-5-124 pretreatment. Highlighted in blue are representative DEGs involved in pro- or anti-inflammatory functions. (C) Shown is a heatmap of mean-centered and row-scaled log-transformed expression values (in counts per million) for genes differentially expressed at 6 hours of LPS treatment ± YKL-5-124 pretreatment [as in (B)] across conditions: v, vehicle; L, LPS; Y, YKL-5-124; YL, LPS and YKL-5-124; with highlighted genes from (B) marked. Hierarchical clustering was performed using Euclidean distance and complete linkage clustering algorithm. Violin plots show cluster-wide distribution of z-score–transformed gene expression stratified by treatment. (D) QuSAGE of YKL-5-124–up-regulated or YKL-5-124–down-regulated pathways (unadjusted P < 0.05, n = 38 each) from (B) ranked by logFC with the size of the circle proportional to the number of genes in the pathway and color coded according to P value. Representative inflammation-related pathways are highlighted. (E and F) Examples of down-regulated (E) or up-regulated (F) pathways from (D); means ± SD of expression of individual pathway-defining genes are shown.

We then evaluated the genome-wide effect of the more selective CDK7 inhibitor YKL-5-124 on the LPS-induced (1, 6, or 12 hours) transcriptome of MΦ by RNA-seq. The analysis of differentially expressed genes (DEGs) in BMMΦ treated with LPS for 6 hours revealed that YKL-5-124 pretreatment suppressed the expression of 897 genes (FC < −1.5; data file S3), broadly remodeling the inflammatory transcriptome including dozens of cytokines, chemokines, and other inflammatory mediators (Fig. 2, B and C). Conversely, the relative expression of 1482 genes was higher in YKL-5-124–pretreated BMMΦ compared with those exposed to LPS alone (FC >1.5; data file S3). Many of these genes, including Cpt2, Amacr, Abcd1, Abcd2, Mlycd, Acsl3, Cited2, and Klf4 (Fig. 2, B and C), are associated with macrophage tissue homeostasis and repair or anti-inflammatory phenotypes. An assessment of the effect of LPS alone on the transcriptome of resting BMMΦ revealed that numerous “YKL-5-124–down” proinflammatory genes are the ones that are induced by LPS, whereas the “YKL-5-124–up” genes are otherwise down-regulated by LPS (Fig. 2C and fig. S3A). Thus, inflammatory transcriptome remodeling upon CDK7 inhibition comprises both down-regulation of proinflammatory and maintained expression of anti-inflammatory genes, a pattern that phenocopies genetic loss of NELF with respect to gene specificity.

To better understand the functional impact of YKL-5-124 on the response of MΦ to LPS, we performed QuSAGE (quantitative set analysis of gene expression), which revealed a widespread attenuation of pathways related to inflammation, including the NLRP3 inflammasome, reactive oxygen and nitrogen species production, interferon regulatory factor 3/7 (IRF3/7) activation by TBK1 and IKKε, matrix metalloproteinase, and the IL-12 and IL-23 pathways (Fig. 2D and data file S4). Moreover, top regulators in these pathways, e.g., Ly96, Irf7, Ikbke, and Tbk1, in the IRF3/7 activation pathway, were downregulated with YKL-5-124 pretreatment (Fig. 2E). Conversely, anti-inflammatory pathways such as peroxisomal lipid metabolism, mitochondrial long-chain fatty acid β-oxidation, and autophagy were up-regulated with YKL-5-124 pretreatment relative to LPS alone (Fig. 2, D and F, and data file S5). QuSAGE of the transcriptome in LPS-treated versus control BMMΦ corroborated differential gene expression analysis: Pathways down-regulated by YKL-5-124 were those activated by LPS and vice versa (fig. S3, B to D), further demonstrating that CDK7 inhibition prevented LPS-dependent acquisition of an inflammatory phenotype in BMMΦ. Such reciprocal effects of YKL-5-124 and LPS on gene expression were noticeable as early as 1 hour of LPS exposure (fig. S3, E and F) and were maintained at 12 hours (fig. S3, G and H).

CDK7 inhibition impairs the Pol II transcription cycle at baseline and after inflammatory polarization of hMΦ

We next evaluated the effects of CDK7 inhibition on human MΦ (hMΦ), which were differentiated in vitro from CD14⁺ monocytes purified from human peripheral blood mononuclear cells (PBMCs) (32). We first found that a 3- to 5-hour exposure of freshly differentiated hMΦ to either THZ1 or YKL-5-124 resulted in 80 to 90% reduction, relative to DMSO-treated controls, of NELF-E occupancy at the TSS of candidate paused genes ZFP36, CITED2, TNF, and IRF1 (Fig. 3A). To determine the effect of CDK7 inhibition on the human NELF complex genome-wide, we performed NELF-E ChIP-seq in hMΦ incubated for 3 hours with or without CDK7 inhibitors. A total of 3595 NELF-E peaks were called in DMSO-treated control hMΦ before data filtering (see Materials and Methods), of which 1739 appear in control hMΦ only and are down-regulated upon YKL-5-124 treatment, and more than half of those are in promoter-proximal regions (fig. S4A). In contrast, among 2350 and 2254 peaks called in THZ1- or YKL-5-124–treated hMΦ, respectively, nearly none were near the TSS, and 1798 (80%) were shared between the two conditions (fig. S4A). Upon filtering, we identified 2649 NELF-E peaks in control hMΦ, of which 488 peaks were promoter-proximal and all but 12 were down-regulated by THZ1 and YKL-5-124 (Fig. 3B, fig. S4B, and data file S6). These inhibitor-sensitive peaks included those at the TSS of the paused anti-inflammatory genes ZFP36, CITED2, DUSP1, NFKBIA, and ZFP36L2 [encoding an RNA binding protein that modulates transcriptional effects of glucocorticoids; (33)] and proinflammatory genes TNF and IRF1 (Fig. 3, B and C, and fig. S4, B and C). Thus, both CDK7 inhibitors broadly and markedly attenuated NELF recruitment to Pol II elongation complexes in hMΦ.

Fig. 3. CDK7 inhibition attenuates the Pol II transcription cycle of hMΦ at baseline and after inflammatory polarization.

(A) hMΦ were treated with THZ1 or YKL-5-124, and NELF-E occupancy at the TSS of the indicated paused genes was assessed by ChIP-qPCR with the average signal with nonspecific IgG as background and the signal in DMSO-treated hMΦ set to 1 (n = 3 or 4). Data are shown as means ± SEM; data were analyzed by Student’s t test; ****P < 0.0001. (B and C) NELF-E ChIP-seq was performed in hMΦ treated with THZ1, YKL-5-124, or DMSO for 3 hours. (B) Shown is a volcano plot for filtered NELF-E peaks up-regulated (right) or down-regulated (left) by YKL-5-124 with promoter-proximal peaks shown in red (n = 4; logFC > 0.6, FDR < 0.05). Highlighted are peaks associated with the TSS of representative paused genes. Inset shows the % and number of peaks significantly down-regulated (FDR < 0.05) by YKL-5-124 stratified by genomic location. (C) Read density distribution for the indicated genes is shown. (D to F) hMΦ were treated ± YKL-5-124 for 5 hours or polarized with TNF and IFN-γ for 24 hours with or without YKL-5-124 added for the last 5 hours of treatment. (D) Pol II and pS5 Pol II occupancy at the TSS of the indicated genes was analyzed by ChIP-qPCR as in (A) with the signal in DMSO-treated hMΦ set to 1 (n = 4 or 5). Data are shown as means ± SEM; data were analyzed by Student’s t test; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; n.s., not significant. (E) Genome-wide Pol II and pS5 Pol II occupancy was evaluated by ChIP-seq and is plotted as heatmaps anchored on YKL-5-124–sensitive promoter-proximal NELF-E peaks (n = 476) from (B) with the average read density shown on top. (F) Read density distribution for the indicated genes is shown.

We then assessed Pol II occupancy in freshly differentiated hMΦ by ChIP-qPCR. We observed substantial Pol II accumulation at the TSS of the paused genes ZFP36, TNF, and IRF1 at baseline, with the 5-hour exposure to YKL-5-124, reducing Pol II occupancy three- to fivefold (Fig. 3D). Furthermore, ChIP-seq revealed that Pol II occupancy at all 476 promoters that lost NELF peaks upon YKL-5-124 exposure was also attenuated (Fig. 3E). These included promoters for ZFP36, IRF1, and TNF, whereas the nonpaused CXCL8 and CXCL10 genes lacked detectable Pol II at their promoters in resting hMΦ (Fig. 3, D and F, and fig. S4D). S5 Pol II CTD phosphorylation is catalyzed by CDK7 during initiation and promoter escape and is enriched in promoter-proximally paused Pol II (7). We readily detected ChIP signal corresponding to S5-phosphorylated Pol II (pS5 Pol II) at the TSS of ZFP36, TNF, and IRF1 in unstimulated hMΦ, and it was reduced to near the immunoglobulin G (IgG) baseline by YKL-5-124 (Fig. 3D). Moreover, recapitulating NELF-E and total Pol II patterns, ChIP-seq showed global reduction in pS5 Pol II by YKL-5-124 at paused promoters (Fig. 3E), including peaks near the TSS of ZFP36, IRF1, and TNF (Fig. 3F and fig. S4D). Together with NELF-E occupancy results, these data suggest that CDK7 inhibition destabilizes or impairs Pol II pausing in resting hMΦ.

Thus far, we have shown that CDK7 inhibition attenuates pausing in resting murine BMMΦ and hMΦ and, at least in BMMΦ, reduces pro- but not anti-inflammatory signaling in response to subsequent inflammatory stimulus, thereby biasing the transcriptome toward an anti-inflammatory state. We therefore asked whether administration of CDK7 inhibitor to preactivated hMΦ can alter their established inflammatory phenotype. We took advantage of a recent single-cell transcriptomic study that described a pathogenic subset of hMΦ present in the bronchoalveolar lavage of patients with severe COVID-19, synovial samples from individuals with flaring RA, ileal tissue from individuals with Crohn’s disease, and colonic epithelium from patients with ulcerative colitis (34). A shared transcriptomic signature of these hMΦ was recapitulated in vitro by polarizing PBMC-derived hMΦ with TNF and interferon-γ (IFN-γ) for 24 hours (34), the conditions that we chose for our study. As expected, at both paused (TNF1 and IRF1) and nonpaused (CXCL8 and CXCL10) proinflammatory genes, a 24-hour TNF + IFN-γ incubation led to increased occupancy of total and pS5 Pol II both at the TSS and in the gene body (Fig. 3, D and F, and fig. S4D). A simultaneous reduction in Pol II and pS5 Pol II at the TSS and especially the gene body of anti-inflammatory ZFP36 (Fig. 3, D and F) was consistent with inflammatory human macrophage polarization. Adding YKL-5-124 to the culture for the last 5 hours of polarization reduced total and pS5 Pol II occupancies at the TSS and gene bodies of proinflammatory paused TNF and IRF1 and nonpaused CXCL8 and CXCL10 to values seen in YKL-5-124–treated unpolarized hMΦ (Fig. 3, D and F, and fig. S4D). Thus, YKL-5-124 attenuated pausing and reduced Pol II occupancy on proinflammatory genes even when administered 19 hours after stimulation with TNF + IFN-γ.

YKL-5-124 reverses preestablished inflammatory polarization in hMΦ

To understand the transcriptional consequences of CDK7 inhibition in inflammatory hMΦ, we evaluated steady-state gene expression in hMΦ polarized with TNF + IFN-γ for 24 hours with or without YKL-5-124 added to the culture for the last 3 or 6 hours of polarization. In agreement with Pol II occupancy findings, TNF + IFN-γ potently upregulated proinflammatory TNF, IL1b, and CXCL8 and reduced the expression of anti-inflammatory ZFP36 (Fig. 4A). As little as 3 hours of YKL-5-124 cotreatment was sufficient to reverse the prior 21-hour induction of TNF, IL1b, and CXCL8 by TNF + IFN-γ and return expression values to the baseline of unstimulated hMΦ (Fig. 4A).

Fig. 4. YKL-5-124 reverses the established inflammatory transcriptome in hMΦ.

(A) hMΦ were cultured with or without TNF + IFN-γ for 24 hours with YKL-5-124 added for the last 3 or 6 hours, as shown. The expression of the indicated genes was assessed by RT-qPCR and presented relative to that in untreated hMΦ (set at 1). n = 4; data are shown as means ± SEM and were analyzed by one-way analysis of variance (ANOVA) with Dunnett’s multiple comparisons test; **P < 0.01; ***P < 0.001; ****P < 0.0001. (B) hMΦ were cultured as in (A). RNA-seq shows DEGs up-regulated (right; n = 486) or down-regulated (left; n = 1822) by 6-hour YKL-5-124 treatment with representative inflammation-related genes highlighted (logFC > 0.6, unadjusted P < 0.05). (C) QuSAGE (unadjusted P < 0.05) of pathways from (B) up-regulated (top; n = 47) or down-regulated (bottom; n = 408) by YKL-5-124, plotted as described in Fig. 2D. (D) Examples of pathways from (C) with key DEGs highlighted; means ± SD of expression of individual pathway-defining genes are shown.

We performed RNA-seq analysis of hMΦ polarized with TNF + IFN-γ with or without addition of YKL-5-124 for the last 6 hours of inflammatory polarization. Differential gene expression analysis revealed that YKL-5-124 broadly altered the transcriptional program elicited by TNF + IFN-γ, suppressing the expression of 1822 genes (FC < −1.5; data file S7) encoding numerous cytokines, chemokines, and other mediators of inflammation (Fig. 4B) including IL1b; TRAF5; TLR6; TNF; CXCL2, 3, 8, and 10; and CCL2, 3, 4, and 8. Conversely, mRNAs encoded by 486 genes that were enriched after YKL-5-124 treatment (Fig. 4B) included anti-inflammatory genes such as the fatty acid metabolic enzyme MCEE, the complement inhibitor CFH, and the receptor tyrosine kinase ligand PROS1 (35, 36). QuSAGE assessing the functional impact of YKL-5-124 in TNF + IFN-γ–polarized hMΦ revealed attenuation of proinflammatory pathways including cytokine-cytokine receptor interactions, type I IFN network including IRF3/7, TBK/IKK pathways and interferon-stimulated genes (ISGs), mitogen-activated protein kinase and chemokine signaling, and glucose transport (Fig. 4C and data file S8). Top regulators of these pathways, such as IRF3, MX2, STAT1, JAK1, and IFIT1, were down-regulated by YKL-5-124 (Fig. 4D, left). Conversely, anti-inflammatory pathways, including mitochondrial long-chain fatty acid β-oxidation, oxidative phosphorylation, and RAR/RXR pathways, and top regulators from the individual pathways, such as MCEE, HADHB, and HADHA, were spared or enriched (Fig. 4, C and D, right). We conclude that CDK7 inhibition with YKL-5-124 can disrupt Pol II pausing and rapidly reverse an established proinflammatory transcription program in hMΦ.

CDK7 inhibition attenuates acute and chronic progressive inflammatory arthritis in vivo

Given the efficacy of CDK7 inhibition in suppressing inflammatory gene expression in vitro, we evaluated the consequences of CDK7 inhibition in mouse models of inflammatory diseases. In the K/BxN-ST model of acute macrophage-dependent inflammatory arthritis, injection of YKL-5-124 [10 mg/kg intraperitoneally (ip) daily] markedly reduced disease severity, assessed by ankle thickness and arthritis clinical scores. Moreover, mice dosed with YKL-5-124 achieved remission nearly a week faster than vehicle-treated control mice (Fig. 5A). Consistently, histological analysis of the hind paws at the peak of the disease revealed less tendonitis and synovial hyperplasia and attenuated macrophage infiltration in YKL-5-124–treated mice compared with vehicle-treated controls (fig. S5A).

Fig. 5. CDK7 inhibition alleviates acute and chronic progressive inflammatory arthritis in mice.

(A) After the induction of K/BxN-ST arthritis, mice were dosed daily with YKL-5-124 (10 mg/kg) or vehicle (Veh); n = 10 per group. The results of a locally estimated scatterplot smoothing (LOESS) regression show the relationship between ankle thickness changes or clinical scores and time (span = 0.75; degree = 2), stratified by treatments. Data from two independent experiments are pooled. The regression curves represent the center of the predictive values, and the ribbon bands indicate 95% confidence interval. (B to G) WT (gray) or hTNFtg mice dosed with THZ1 (10 mg/kg, teal) or Veh (red) three times per week starting at 8 weeks of age. Ankle thickness FC and clinical scores were measured three times per week. (B) The results of LOESS regression show the relationship between ankle thickness FC relative to the first measurement (= 1; left) or clinical scores (right) and time, stratified by treatments and analyzed as in (A) (n = 4). (C) Shown are 3D models of micro–computed tomography (μCT) top and side view images of lower limbs at 20 weeks of age with osteophytes (yellow arrows) and sites of bone erosion (teal arrows) indicated. (D) μCT images were quantified as ankle surface area/volume (1/mm) ratio (n = 3). (E) Representative images of H&E-stained sections of ankles at 20 weeks of age. Scale bars, 100 μm. Orange bars indicate tenosynovial hyperplasia; immune cell infiltration is marked by green arrows. [(F) and (G)] Synovial inflammation (F) and immune cell infiltration (G) scoring is shown as described in Materials and Methods. n = 6. Data in [(D), (F), and (G)] are shown as means ± SEM. Data were analyzed by one-way ANOVA with Šidák’s multiple comparisons test (D) or Student’s t test [(F) and (G)]; **P < 0.01; ***P < 0.001.

We next used human TNF transgenic (hTNFtg) mice to assess the effects of transcriptional CDK7 inhibition in a chronic inflammatory disease. In these mice, body-wide integration of the hTNFtg confers low but constitutive expression of hTNF, which results in progressive inflammatory arthritis, ultimately leading to bone erosion and gradual destruction and ankylosis of the joints, replicating features of untreated human RA (37, 38). We started dosing hTNFtg mice with THZ1 (10 mg/kg ip, three times per week) at 8 weeks of age and monitored arthritis progression for 10 to 12 weeks. THZ1 markedly slowed disease progression compared with vehicle-treated transgenics as evidenced by coordinate reduction of both ankle thickness measurements and arthritis clinical scores across the duration of the experiment (Fig. 5B and fig. S5B).

End point x-ray imaging of lower limbs of the vehicle-treated 20-week-old hTNFtg mice revealed ankylosis and inability to spread their digits; in contrast, the paws of age-matched THZ1-treated transgenic mice resembled those of WT mice (fig. S5C). We used three-dimensional (3D) micro–computed tomography (μCT) to image the end point morphological changes in the ankles at high resolution. In addition to confirming x-ray findings, μCT (Fig. 5C and fig. S5D) showed roughening of bone surface with multiple osteophytes (yellow arrows) and sites of bone erosion (teal arrows) in vehicle-treated hT-NFtg mice but not in either WT or THZ1-treated hTNFtg mice. The side view reveals complete ankylosis and deformity of the tibiotalar joint in untreated hTNFtg mice (Fig. 5C). We quantified osteophytes and bone erosion by determining the relative ratio of bone surface area to bone volume in the ankle joints. Compared with WT, hTNFtg mice displayed a sevenfold higher ratio, consistent with bone erosion, and this increase was suppressed more than twofold by THZ1 treatment (Fig. 5D). Thus, x-ray and μCT assessments both demonstrate a robust inhibition of the chronic inflammation-driven joint destruction in THZ1-treated hTNFtg mice.

Lastly, we examined histopathological features of synovial tissues by hematoxylin and eosin (H&E) staining of ankle sections. We reasoned that, although end point measurements in this chronic model are past the peak of inflammatory responses, we might still be able to capture differences in the long-term changes imparted by the treatment. H&E staining revealed marked thickening and hyperplasia (orange bars) of tenosynovial membrane and the tendon and immune cell infiltration (green arrows) in vehicle-treated hTNFtg mice; both histological parameters of inflammation were reduced by THZ1 (Fig. 5, E to G). We conclude that two compounds that inhibit CDK7, namely, THZ1 and YKL-5-124, attenuated severity of inflammatory arthritis in both self-limiting and chronic models.

YKL-5-124 suppresses an inflammatory signature in synovial explants from patients with RA

Given the therapeutic efficacy of CDK7 inhibition in mouse models of inflammatory arthritis, we sought to examine consequences of CDK7 inhibition in synovial tissue samples of patients with RA ex vivo (34, 39). Cryo-stored synovial tissue samples can be placed in culture and dissociated into single-cell suspension with a robust recovery of cell types and transcription activity within each cell type (39, 40). We used this approach to assess the impact of CDK7 inhibition on gene expression in synovial samples. We prepared synovial single-cell suspensions from highly inflamed RA synovial tissue as in (40), treated cells with DMSO or YKL-5-124 for 3 or 6 hours, purified total RNA, and subjected it to bulk RNA-seq (n = 4 to 8 samples). Differential gene expression analysis revealed that a 6-hour YKL-5-124 treatment down-regulated hundreds of genes, many of which encoded cytokines, chemokines, and other mediators of inflammation including CCL2, 4, 7, and 22; CXCL1, 5, and 8; IL1B, 6, and 24; costimulatory receptor TNFRSF9; tissue-degrading MMP9; and multiple components of the IFN-α/β signature including transcription factors IRF5, 7, and 9, and OAS1/3, ISG15, IFIT3, and RSAD2 (Fig. 6, A and B, and data file S9). RSAD2; IDO1; CXCL1, 5, and 8; TNFRSF9; IL6; and IL1B were significantly down-regulated as early as 3 hours of YKL-5-124 exposure [fig. S6A; (logFC > 0.6, FDR < 0.05) also marked with an * in Fig. 6B]. At the same time, among the genes whose relative expression was increased by YKL-5-124 treatment (Fig. 6A), we noted MRC1 and CD163, established markers of the homeostatic “M2” macrophage phenotype (41).

Fig. 6. YKL-5-124 reduces inflammatory gene expression in synovial cells derived from patients with RA ex vivo.

Cell suspensions were prepared from synovial tissue samples collected from patients with RA, placed in culture, and treated with YKL-5-124 or DMSO for 3 (n = 4) or 6 (n = 8) hours. Total RNA was prepared and analyzed by RNA-seq. (A) The volcano plot shows DEGs up-regulated (right; logFC > 0.6, FDR < 0.05, n = 1008) or down-regulated (left; logFC < −0.6, FDR < 0.05, n = 932) by YKL-5-124 at 6 hours. Highlighted are representative DEGs involved in pro- or anti-inflammatory functions. (B) The heatmap shows mean-centered and row-scaled log-transformed expression values (in counts per million) for genes differentially expressed in (A) across conditions (v, vehicle; and Y, YKL-5-124 for 3 or 6 hours, as indicated) with genes highlighted in (A) marked. The heatmap was generated as in Fig. 2C with violin plots showing cluster-wide distribution of z-score–transformed gene expression stratified by treatment. (C) QuSAGE of up-regulated (top; unadjusted P < 0.05, n = 219) or down-regulated (n = 799) pathways at 6 hours of YKL-5-124 treatment. (D to F) Examples of down-regulated pathways with key genes highlighted. Pathways shown include chemokine receptors and chemokines (D), IFN-α/β signaling (E), and inflammasomes (F). Error bars, SD. (G) The effect of the 6-hour YKL-5-124 treatment on the expression of IL6 (n = 6), IL1B (n = 6), and MRC1 (n = 4) in synovial samples was validated by RT-qPCR; data are presented as means ± SEM and were analyzed by Student’s t test; *P < 0.05; **P < 0.001.

The genome-wide impact of YKL-5-124 on gene expression in synovial cells was corroborated by QuSAGE. Pathways relevant to inflammation, including cytosolic DNA and innate immune sensors, IL-12 and IFN-α/β signaling, inflammasome, noncanonical NF-κB pathway, and cytokine and chemokine receptors and interleukins, were broadly attenuated (Fig. 6C and data file S10). Top regulators of these pathways were also down-regulated. For example, the expression of RSAD2, IFIT3, PTPN6, OAS1/3, and IRF5, 7, and 9 in the IFN-α/β signaling pathway; MEFV, HMOX1, and TXN in the inflammasome pathway; and CCL2, 3, 4, 7, and 22, CXCL5, 6, and 8, and CCR1 in the chemokine and chemokine receptors pathway were suppressed by YKL-5-124 (Fig. 6, D to F). Conversely, anti-inflammatory pathways, or those known to be specifically protective in arthritis such as glycosphingolipid and sphingolipid metabolism, including their top regulators ARSD, GALC, and SUMF1/2, were spared by YKL-5-124 treatment (fig. S6B). We further independently validated the down-regulation of IL6 and IL1B and established pathogenic players in RA (42–44) and up-regulation of the homeostatic macrophage marker CD163 in YKL-5-124–treated cells by qPCR (Fig. 6G). Thus, our transcriptomic analysis of synovial cells isolated from patients with RA provides strong evidence of an anti-inflammatory effect of CDK7 inhibition in human disease samples.

DISCUSSION

Inflammation is a tightly regulated sequence of events that involves multiple cell types with distinct functions, and failure to control these events is associated with numerous inflammatory diseases. MΦ play a vital role in this process: They secrete cytokines and chemokines to either promote or alleviate inflammation, bridge innate and adaptive immunity by presenting antigens to lymphocytes, clear debris, and preserve tissue homeostasis through their phagocytic activities. A key to the rapid adaptation and versatile response of MΦ to dynamic changes in the inflammatory process is regulation of gene expression at the level of transcription. Among the different stages of the transcription cycle, release of Pol II from the promoter-proximal pause has emerged as a widespread rate-limiting step in gene activation in MΦ (22, 23, 45).

Pol II pausing includes two steps, establishment and release, gated by CDK7/cyclin H and CDK9/cyclin T1 (P-TEFb), respectively. As a pause release factor, CDK9 is under intense investigation in cancer, with many CDK9 inhibitors developed and tested since the initial discovery of flavopiridol in the 1990s (46, 47). These efforts, however, have been plagued by off-target effects in vitro and toxicity in patients (48).

Our strategy to focus on pause establishment instead was prompted by our genetic studies in which loss of the pausing factor NELF, an integral component of the Pol II pausing complex, yielded a timed and selective change in macrophage gene expression. In particular, the early response to activation by LPS displayed an enrichment of transcripts encoding inflammation inhibitors. Because the impact of NELF deletion on Pol II occupancy in MΦ had not been directly evaluated, we interrogated Pol II distribution on chromatin in WT and NELF-deficient BMMΦ. Superinduction of anti-inflammatory genes appeared to be a direct transcriptional effect of NELF deficiency, with enhanced occupancy of elongating Pol II along gene bodies compared with WT BMMΦ. Although this augmented Pol II density in NELF-cKO BMMΦ was short-lived, the delayed consequence of the transient enrichment of anti-inflammatory mediators was a subdued inflammatory response in MΦ in vitro and in the K/BxN-ST model for inflammatory arthritis in vivo. It was, therefore, reasonable to hypothesize that targeting pausing may represent a strategy to modulate inflammatory responses.

We chose to target CDK7 because its inhibition had been shown to impede formation of a stably paused complex of Pol II with NELF in promoter-proximal regions (12–14). Both THZ1 and YKL-5-124 triggered a loss of TSS-proximal NELF, even when normalized to residual occupancy of Pol II; conversely, the amount of bound TFIIE increased relative to Pol II, consistent with failure to exchange initiation factors for elongation factors and establish a stable pause. The functional consequence of CDK7 inhibition was rapid and widespread rewiring of the inflammatory transcriptome seen across all experimental platforms tested: mouse BMMΦ, hMΦ, and RA patient-derived synovial cells. Moreover, these changes occurred not only under conditions of pretreatment with CDK7 inhibitor before administering an inflammatory stimulus but also when CDK7 was inhibited after long-term human macrophage polarization toward an inflammatory phenotype by exposure to TNF and IFN-γ. In each case, CDK7 inhibition largely erased the inflammatory signature encompassing dozens of cytokine- and chemokine-encoding genes. In contrast, the paused anti-inflammatory genes that were superinduced in the NELF-cKO BMMΦ were resistant to CDK7 inhibitor treatment in the face of inflammatory signals. Furthermore, genome-wide, among the YKL-5-124–up DEGs were enzymes involved in lipid metabolism, fatty acid β-oxidation, autophagy, inhibition of complement pathways, and enhancement of phagocytosis. Considering that these were the same genes that were otherwise repressed by inflammatory signals, the overall impact of CDK7 inhibition on MΦ was a failure to undergo metabolic inflammatory reprogramming and acquire or maintain an M1-like phenotype. In that regard, it is interesting that our in silico transcription factor network analysis in BMMΦ pointed to Kruppel-like factor 4 (KLF4) as the top activator for YKL-5-124 refractory genes. KLF4 is a well-established master regulator for the homeostatic, anti-inflammatory M2 phenotype (49), whose potential role under conditions of CDK7 inhibition may point to an unexplored link between transcriptional kinases and macrophage programming.

One notable outcome of our transcriptomic studies in MΦ was the absence of cell cycle signatures among either up- or down-regulated pathways. CDK7 has two distinct cellular functions: TFIIH-associated CDK7 phosphorylates the Pol II CTD to facilitate the transition from transcription initiation to elongation, whereas free CDK7/cyclin H/MAT1 is the CDK-activating kinase needed for activating phosphorylation of CDK1, CDK2, CDK4, and CDK6 during cell cycle progression (50–52). In cancer models, potential therapeutic effects of highly selective CDK7 inhibitors, such as YKL-5-124, have been suggested to stem in part from cell cycle blockade (31). Our results argue, however, that the anti-inflammatory consequences of CDK7 inhibition are due entirely to effects on transcription, on the basis of the following considerations: (i) In MΦ, both THZ1 and YKL-5-124 ablated Pol II occupancy at promoters of proinflammatory genes and markedly reduced their transcript abundance; (ii) the reversal of inflammatory gene expression was observed within as little as 3 hours, an interval much shorter than the doubling time of even the most rapidly dividing mammalian cells; (iii) MΦ are often noncycling cells that retain minimal proliferative capacity; and (iv) QuSAGE of YKL-5-124–regulated macrophage transcriptomes across species failed to reveal cell cycle pathways (also consistent with low proliferative capacity).

Because it is difficult to extrapolate from analysis on a short timescale (24 hours) in vitro to inflammation-driven disease in vivo, we chose two very different models of inflammatory arthritis to interrogate the biological consequences of CDK7 inhibition. It is notable that, in both the acute self-limiting K/BxN-ST and the chronic progressive hTNFtg models, CDK7 inhibition robustly reduced disease severity. Although inhibition of the related CDK12 and CDK13 might contribute to the anti-inflammatory effects of THZ1 in the hTNFtg model, the results with the CDK7-selective YKL-5-124 in the K/BxN-ST model, combined with similar effects of the two inhibitors in MΦ in vitro, suggest that CDK7 is the relevant therapeutic target in both models. Our dosing strategy of THZ1 (three times per week), which achieved therapeutic effect in the hTNFtg mice, was much less aggressive than the twice daily dosage used, for example, in a xenograft model of T cell acute lymphoblastic leukemia (13). Of note, a noncovalent CDK7 inhibitor, samuraciclib, has been well tolerated in phase I clinical trials in patients with cancer (53).

An important consideration in evaluating the impact of a CDK7 inhibitor in vivo is that it will trigger a response in multiple cell types in addition to MΦ. In RA and, to varying degrees, murine arthritis models, synovial fibroblasts critically contribute to disease pathogenesis (40, 54, 55). Targeting proliferating fibroblasts using broader spectrum CDK inhibitors has been considered as a possible avenue in RA treatments (56, 57). Moreover, in the context of the tissue microenvironment, the responses of MΦ themselves are distinct from those observed in pure culture (39, 40). Thus, the ability of a CDK7 inhibitor to target nonmyeloid cells present in the joint may contribute to the more pronounced attenuation of inflammatory arthritis by YKL-5-124 than the myeloid cell–specific depletion of NELF. Synovial explants derived from the joints of patients with RA reflected this complexity. Along with MΦ, which comprise on average ~25% of cells depending on the individual, these explants contain many cell types including T and B cells in the immune compartment as well as fibroblasts and some endothelial cells (58, 59). Whereas several genes undergoing down-regulation (e.g., IL1B, IDO1, MMP9, CXCL5, and CCL7) or up-regulation (e.g., MRC1 and CD163) in response to CDK7 inhibition in our study are expressed predominantly in MΦ, others (e.g., TNFRSF9 and ISG15) are more broadly expressed (60). Future interrogation of the consequences of CDK7 targeting on the transcriptional states and phenotypes of individual cell populations within mixed cultures would undoubtedly be revealing.

In addition to certain study limitations discussed above, we note that we have not directly assessed Pol II pausing in the in vivo or synovial explant experiments. This type of analysis is not feasible today but should become possible with further technology development. In addition, although previous studies exploring CDK7 roles in pausing led us to test whether CDK7 inhibitors could phenocopy NELF ablation, in principle, CDK7 inhibition may exert its protective effects through both NELF-dependent and NELF-independent mechanisms. Lastly, larger patient cohorts would enable stratifying patients with RA by prior treatments, something that we could not do with our current patient numbers. Along with single-cell analyses, such studies would provide a more granular view of the effect of CDK7 inhibitors on different subsets of patients.

In summary, we demonstrate here that CDK7 inhibition represents a promising avenue for treating inflammatory arthritis. This strategy can yield anti-inflammatory effects and metabolic reprogramming in both MΦ and, potentially, other cell types. Notably, the potential therapeutic benefit of CDK7 inhibition may be applicable in a wide range of inflammatory diseases beyond RA.

MATERIALS AND METHODS

Study design

This study assessed the effect of CDK7 inhibition on Pol II transcription cycle and inflammatory gene expression in mouse BMMΦ and hMΦ, in murine models of RA, and in synovial explants from patients with RA. Sample size for individual experiments was not predetermined using statistical tests but was chosen on the basis of prior experience of the investigators using similar experimental systems. The exact n values used in each experiment are specified in the respective figure legends. For in vivo inflammatory arthritis experiments, the investigators were blinded to the genotype of mice (Fig. 5 and figs. S1 and S5) and treatment groups (Fig. 5 and fig. S5) during scoring. The investigators were not blinded during data analysis. No randomizing was performed; for the experiments with hMΦ, we used all donor PBMC “buffy coats” provided to us by the blood bank. The criteria for choosing specific patients with RA from the larger cohort at Hospital for Special Surgery (HSS) are described below. We used all patient samples made available to us.

Reagents

Cell culture treatments were LPS (Sigma-Aldrich) at 10 ng/ml, THZ1 (Selleck Chemicals, catalog no. S7549), and YKL-5-124 [covered by patent number US10870651B2; (31)] diluted to 250 μM (1000×) in 100% DMSO and used at 250 nM final concentration. Human TNF (Abcam, ab259410) and IFN-γ (ab259377) were diluted in phosphate-buffered saline (PBS) and used at 20 and 5 ng/ml, respectively. K/BxN serum was collected from K/BxN mice bred in HSS animal facility or acquired commercially.

Mouse BMMΦ

BMMΦ were prepared from 8- to 12-week-old male mice as previously described (22). In brief, bone marrow from tibiae and femora was flushed and incubated for 5 days in Dulbecco’s modified Eagle medium (DMEM; Corning, glucose at 1 g/liter) supplemented with 20% fetal bovine serum (FBS) and 20% l-cell conditioned medium. Adherent cells were then scraped, seeded into 150-mm or six-well plates in DMEM with 20% FBS, and treated the following day as described in the figure legends.

Human PBMC-derived MΦ

Deidentified human buffy coats were purchased from the New York Blood Center. PBMCs were isolated by gradient centrifugation of buffy coats with Lymphoprep (STEMCELL Technologies, 07851). Human monocytes were purified from PBMCs with anti-CD14 magnetic beads (Miltenyi Biotec, 130-050-201) as recommended by the manufacturer. Purified human monocytes were seeded into 150-mm plates (1.5 × 10⁷ cells per plate) or six-well plates (2 × 10⁶ cells per well) and cultured for 5 days in DMEM supplemented with 20% heat-inactivated FBS, penicillin/streptomycin, l-glutamine, with human macrophage colony-stimulating factor (20 ng/ml) added fresh every other day.

Mice

C57BL/6 mice (National Cancer Institute, Charles River Laboratories; strain code 556) and their transgenic derivatives were maintained in the Weill Cornell Barrier Animal Facility in full compliance with the protocol approved by the Institutional Animal Care and Use Committee (Institutional Animal Care and Use Committee, no. 2015–0050). Homozygous Cre only (LysM^cre/cre) and NELF-B cKO (LysM^cre/cre-Nelb^fl/fl) mice were previously described (23). hTNFtg mice were acquired from Taconic Biosciences (strain code 1006).

Synovial cells derived from patients with RA

Synovial tissue was obtained from patients consented into the HSS FLARE study (Institutional Review Board, no. 2014–233). In brief, tissue was collected from patients with RA who were older than 18 years of age, satisfied American College of Rheumatology/European Alliance of Associations for Rheumatology (ACR/EULAR) 2010 classification criteria or the 1987 RA diagnostic criteria, and were undergoing primary total hip, knee, elbow, or shoulder replacement surgery or synovectomy of the hip, knee, elbow, shoulder, wrist, ankle, or finger. We selected “highly inflammatory” patients who had a lymphocytic infiltration (SLI) score of 3 or 4. De-identified 3-mm³ synovial tissue specimens were collected and cryopreserved in CryoStor CS10 (Sigma-Aldrich).

Synovial cell treatments derived from patients with RA

Synovial cell suspensions were prepared as reported (39, 40). In brief, cryopreserved synovial tissues were digested with Liberase TL (100 μg/ml, Roche) and deoxyribonuclease I [100 μg/ml, New England Biolabs (NEB)] for 30 min in RPMI 1640. Single-cell suspensions were then generated using 70-μm cell strainers and plated in RPMI 1640 with 10% FBS and 1% glutamine in 96-well plates at 2 × 10⁵ viable cells per well. Viable synovial cells were then treated with DMSO vehicle or YKL-5-124 for 3 or 6 hours. Supernatants were removed, and cells were stored in RLT Plus Buffer for RNA isolation.

ChIP and ChIP-seq

Mouse BMMΦ and hMΦ were cross-linked with 1% formaldehyde, scraped, PBS washed, and lysed as previously described (61). Crude nuclei were prepared, lysed (61), and sonicated for six cycles (hMΦ) or eight cycles (BMMΦ) of 30 s on/30 s off in Diagenode Bioruptor Pico system. Sonicated samples were cleared with protein A/G agarose beads (Santa Cruz Biotechnology; 30 μl per sample for 30 min) and incubated with bovine serum albumin–blocked protein A/G agarose beads (40 μl per sample) and antibody overnight. For the control, rabbit IgG (Thermo Fisher Scientific) was used with a concentration matching other antibodies: 2 μg of total Pol II (Bethyl, A304–405A), 2 μg of pS5 Pol II (Abcam, Ab5131), 4 μg of NELF-E (Santa Cruz Biotechnology, sc-377052), or 2 μg of TFIIE (Abcam, Ab28177). Samples were then washed twice with Szac radioimmunoprecipitation assay (RIPA) buffer [150 mM NaCl, 1% v/v NP-40, 0.5% w/v Na deoxycholate, 0.1% w/v SDS, 50 mM tris-Cl (pH 8.0), and 5 mM EDTA (pH 8.0)], four times with Szac IP wash buffer [100 mM tris-Cl (pH 8.5), 500 mM LiCl, 1% v/v NP-40, and 1% w/v Na deoxycholate], twice with Szac RIPA buffer, and twice with ChIP-TE buffer [100 mM tris-Cl (pH 7.5), 100 mM EDTA, and 500 mM NaCl]. Washed samples were next incubated with de–cross-linking buffer [0.5% SDS and 0.1% v/v proteinase K in Tris-EDTA (TE) buffer] for 2 hours at 55°C followed by overnight incubation at 65°C. DNA fragments were purified using the QIAGEN QIAquick PCR Purification Kit. ChIP samples were analyzed by either qPCR (see data file S11 for primers) or sequencing. For ChIP-seq, libraries were prepared with NEBNext Ultra II DNA Library Prep Kit for Illumina according to the manufacturer’s protocol and multiplexed with index primers (NEB, E7335, E7500, and E7600).

RNA preparation and real-time qPCR

Total RNA was isolated from cells with an RNeasy Plus micro kit (QIAGEN) and reverse transcribed to cDNA with random primers (NEB). Gene expression was analyzed by qPCR with the Maxima SYBR Green/ROX/2x Master Mix (Thermo Fisher Scientific) or the Power SYBR Green Master Mix (Applied Biosystems) on the StepOne Plus Real-Time PCR System (Applied Biosystems) or QuantStudio 6 Pro (Applied Biosystems) using the comparative cycle threshold method and the following amplification protocol: 95°C × 10 min, 1.6°C/s (95°C × 15 s, 1.6°C/s, 60° × 1 min, 1.6°C/s) × 40 cycles, followed by a melt curve (see data file S11 for primers).

Bulk RNA-seq

The integrity of total RNA samples was evaluated with BioAnalyzer 2100 system (Agilent). Sequencing compatible libraries were prepared with NEBNext Ultra II RNA Kits (NEB) and NEBNext Multiplex Oligos for Illumina 96 Unique Dual Index Primer Pairs (NEB) following instructions from the manufacturer. In brief, total RNA was poly(A) enriched, fragmented, and reverse transcribed to cDNA. cDNA were ligated with adaptors and indexes. Sequencing compatible libraries were then quality controlled with BioAnalyzer 2100 system (Agilent) and sequenced (Illumina NovaSeq 6000 system, 50-bp pair-end protocol) at a depth of ~25 million mappable reads per sample at the Genomic Resources Core Facility at Weill Cornell Medicine.

Arthritis models

The K/BxN-ST model was established with two doses of 100 μl of K/BxN serum that were given 1 day apart. YKL-5-124 was dissolved in 5% DMSO in 5DW (5% dextrose water) and administered at 10 mg/kg ip daily. hTNFtg mice develop arthritis without any experimental manipulation. THZ1 was dissolved in 10% DMSO in 5DW and administered at 10 mg/kg ip three times per week. Ankle thickness and clinical scores for K/BxN-ST or hTNFtg were assessed three times per week. Ankle thickness measurements from both ankles were averaged and semiquantitative clinical scores were determined for each of the four limbs using the following scale: 0 = no swelling; 1 = slight erythema, swelling of one digit; 2 = moderate erythema, swelling of >1 digit; 3 = swelling and erythema of the entire paw, digits do not spread; 4 = severe inflammation with joint rigidity; and 5 = complete ankylosis of the paw. The maximal score = 20. At 20 weeks of age, hTNFtg mice were anesthetized and had their lower extremity analyzed with x-ray imaging.

μCT imaging and analysis

Ankles collected from 20- to 22-week-old mice (WT and hTNFtg) were fixed in 4% paraformaldehyde for 12 hours and placed in PBS. Specimens were scanned using Scanco Medical MicroCT 35 (Scanco Medical AG, Switzerland) using the following parameters: 10-μm resolution, 1900-mm slices, 1000 projections at 55 kilovolt peak, and 145 μA. Images were then analyzed with 3D Slicer for quantitation and statistics.

Bone structure of the imaged ankle was modeled with segment editor tools, where an automatic threshold was used to screen out signal from noise. Scissor tools were then used to fine tune the model to make sure that background signals are excluded. Bone surface area (in square millimeters) and volume (in cubic millimeters) were then measured with the segment statistic tool. Bone surface/volume ratio was calculated for each imaged ankle.

Histology

Ankles from K/BxN-ST model mice at peak of disease were collected and immersed into 10 ml of 10% neutral-buffered formalin solution (Sigma-Aldrich) for 8 hours of fixation. Ankles were decalcified following the Webb-Jee protocol (62). They were then paraffin-embedded, sectioned, and stained by H&E at the HSS Research Institute Histopathology Service. Immunohistochemistry for ionized calcium-binding adaptor molecule 1 (IBA1) was performed at the Laboratory of Comparative Pathology at Weill Cornell Medicine using Leica Bond RX automated stainer with Bond reagents (Leica Biosystems), including a polymer detection system (DS9800, Bond Polymer Refine Detection, Leica Biosystems). Antigen retrieval was performed using the Leica Bond RX protocol: HIER (heat-induced antigen retrieval) with ER2 (EDTA, pH 9) for 20 min at 100°C. Slides were stained with antibodies to IBA1 (Abcam, ab5076; 1:2000 dilution) followed by a secondary biotinylated anti-goat IgG made in rabbit (Vector Labs, B-5000; 1:1000 dilution). Stained sections were imaged on the Zeiss Axioscan 7 Slide Scanner.

Ankles from hTNFtg mice were collected and immersed into 10 ml of 10% neutral-buffered formalin solution (Sigma-Aldrich) for overnight fixation. Ankles were then decalcified, paraffin embedded, sectioned, and stained by H&E at the Laboratory of Comparative Pathology at Weill Cornell Medicine. Stained sections were scanned, and histopathological assessment of synovial inflammation was performed by HistoWiz Inc. using the following scoring chart: 0 = healthy, one or two cell layers of synovial membrane; 1 = three to five cell-layered synovial membranes, mild cellular infiltrate into the synovium and exudate in the joint cavity with low cell density; 2 = multilayered synovial membranes, enhanced cellular infiltrates and increased cell density throughout the joints; and 3 = maximally expanded inflammation filling all joint cavities, hyperplastic synovial tissue with high cell density. Immune cell infiltration scoring was performed by HistoWiz Inc. as follows: 0 = none to minimal lymphoplasmacytic infiltrates; 1 = minimal; 2 = mild; and 3 = moderate lymphoplasmacytic infiltrates with larger lymphocytic aggregates.

RNA-seq data analysis

Read quality evaluation and adapter trimming were performed using fastp (63). All reads that passed initial quality filtering were mapped to the mouse genome (mm10) or human genome (hg38), and reads in exons were counted against GENCODE release 27 annotated with the STAR aligner (64) (https://gitlab.com/hssgenomics/pipelines/RNAseq). Batch correction to account for sex and day of the experiment was performed using the surrogate variable analysis (sva) ComBat function in R. Differential gene expression analysis was performed with edgeR (65) using either quasi-likelihood F test or likelihood ratio test. Genes with low expression (<2 counts per million in all groups) were filtered from all downstream analyses. P values were corrected for multiple testing using the Benjamini-Hochberg FDR procedure unless indicated. Genes with unadjusted P < 0.05 and log₂FC > 0.6 were considered differentially expressed. Log-transformed mean-centered and gene-scaled expression values (in counts per million) for DEGs were visualized as a heatmap (66) clustered using the Euclidean distance and complete linkage clustering method. Downstream analyses were performed in R (www.r-project.org) using a Shiny-driven visualization platform (rnaseqDRaMA) developed at the HSS David Z. Rosensweig Genomics Research Center (https://gitlab.com/hssgenomics/Shiny). Differentially regulated pathways were identified using R QuSAGE 2.30.0 package (67) with MSigDB C2 set (curated gene sets c2.cp.v7.1). Gene sets with fewer than 10 genes were excluded. Pathways with unadjusted P < 0.05 were selected for initial analysis (https://gitlab.com/hssgenomics/Shiny).

ChIP-seq data analysis

Paired-end reads were preprocessed using fastp (63), which supports adapter trimming, low-quality base trimming, and calculation of additional quality control metrics. Trimmed reads were aligned to the mm10 or hg38 genome using bowtie2 (68) with --very-sensitive-local -q -p options set. Low-quality, improper, and multimapping alignments and PCR duplicates were filtered using SAMtools (69).

Peaks were called on each sample (replicas are not pooled) with MACS2 (70) using macs2 callpeak --broad --bw 180 --keep-dup all --q 0.05 with either IgG input as a control or a local background subtraction. After peak calling, all peaks from all samples called separately were combined, and adjacent peaks (within 10 bp) were merged using BEDTools as follows: bedtools merge -d 10 -c 4 -o collapse -i (71). ChIP-seq signals were normalized using sum of per-base coverage of reads/fragments and scaled to 1× genome coverage using NCBI BAMscale as follows: BAMscale scale --bam *.aligned.markdup.noM. sorted.bam --binsize 25 --smoothen 1 --normtype base --scale genome --blacklist ign_chr.txt (72). Reads from each sample within the combined set of peaks were then counted using featureCounts (73) to produce an initial reads-per-peak count matrix. The ATACseq/ChIP-seq pipeline is publicly available at https://gitlab.com/hssgenomics/pipelines/ATACseq.

The quality of individual experiments was evaluated by calculating fraction of reads in peaks and analyzing MACS2 peak statistics distributions (peak width, pileup, and enrichment over the background). Furthermore, to ensure consistency read-in-peak pairwise, Spearman correlations were determined within experimental groups. The average within group correlation was larger than 0.9.

The reads-per-peak counts were analyzed for differential peaks using the edgeR quasi-likelihood framework that allows for building complex modeling solutions not attainable by peak calling algorithms at the peak calling step (e.g., with MACS2 directly). Before the analysis, we removed poorly reproducible peaks containing zero reads in more than three samples within a group and peaks with the length of more >1000 bp. All peaks were annotated to the nearest gene and TSS; if the peak overlaps an annotated Fantom5 (74) enhancer, then that enhancer is reported (https://gitlab.com/hssgenomics/Shiny-ATAC). Peaks with FDR < 0.05 were considered differential.

Statistical analysis

Individual-level data for experiments where n < 20 are presented in data file S12. All RNA-seq and ChIP-seq data processing is described in relevant sections in Materials and Methods. All pairwise comparisons were analyzed using Student’s t test. Multiple comparisons testing was done using one-way or two-way analysis of variance (ANOVA) with post hoc analysis as described in specific figure legends. Scoring in mouse arthritis models was analyzed using locally estimated scatterplot smoothing regression (span = 0.75; degree = 2) with ribbon bands showing the 95% confidence interval.

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. Sequencing data generated in this study are deposited at NCBI GenBank under accession numbers GSE255108 (RNA-seq) and GSE255372 (ChIP-seq). No new code was generated in this study. The code used to perform data analysis is publicly available at https://gitlab.com/hssgenomics/Shiny, https://gitlab.com/hssgenomics/pipelines/RNAseq, https://gitlab.com/hssgenomics/Shiny-ATAC, and https://gitlab.com/hssgenomics/pipelines/ATACseq.