A conserved kinase-phosphatase switch in transcription
In fission yeast, Cdk9 phosphorylates the Spt5 CTD33 and the inhibitory Thr316 residue of PP1 isoform Dis227. As Pol II traverses the CPS, Spt5-CTD phosphorylation decreases dependent on Dis2 activity, and pSer2-containing Pol II accumulates with Spt5 in a 3′-paused complex poised for termination27,29. We asked if this switch is conserved in human cells, where two PP1 catalytic-subunit isoforms were identified in a chemical-genetic screen for direct Cdk9 substrates9. We validated PP1γ-Thr311 as a Cdk9-dependent phosphorylation site by two approaches. First, we treated green fluorescent protein (GFP)-tagged PP1γ, expressed in HCT116 cells and immobilized with anti-GFP antibodies, with purified Cdk9/cyclin T1, followed by immunoblotting with an antibody specific for PP1 isoforms phosphorylated on their carboxy-terminal inhibitory sites. Increased signal after Cdk9 treatment of wild-type PP1γ but not PP1γT311A suggests that P-TEFb can indeed phosphorylate this residue in vitro (Fig. 1a).
Next we asked if phosphorylation of this site depends on Cdk9 in vivo. Treatment of HCT116 cells with the Cdk9-selective inhibitor NVP-234 diminished reactivity of immunoprecipitated PP1γ with the phospho-PP1 antibody to about the same extent as did a Cdk1-selective inhibitor (RO-3306), whereas combined treatment with NVP-2 and RO-3306 nearly abolished the signal (Fig. 1b), suggesting roughly equal contributions of the two CDKs to negative regulation of PP1γ in vivo. We surmise that PP1γ, like fission yeast Dis227,35, is a regulatory component shared between the cell-division and transcription machineries.
CTR1 of human Spt5 contains multiple repeats of consensus sequence G-S-Q/R-T-P, including Thr806—a Cdk9 target site detected in our screen9—and is analogous to the CTD of the fission yeast protein (Supplementary Fig. 1a). After a 1-h treatment with 10–50 nM NVP-2, pThr806 was diminished, whereas pSer2 was refractory to the Cdk9 inhibitor at 20- to 100-fold higher doses (Fig. 1c). This is consistent with results in fission yeast, where Cdk9 is not a major contributor to pSer2 in vivo21,36, and with the apparent preference of human Cdk9 for phosphorylating Ser5 in CTD-derived peptides in vitro37. Pol II-pSer2 was also relatively refractory to Cdk9 depletion by short hairpin RNA (shRNA) in HCT116 cells9. In fission yeast, chemical-genetic inhibition of Cdk9 led to rapid, nearly complete dephosphorylation of the Spt5 CTD (T1/2 ~20 s); the rate of decay decreased ~4-fold in dis2 mutant strains, suggesting that the fast kinetics in dis2+ cells were partly due to the concomitant activation of Dis2 (PP1) when Cdk9 is inactivated27. In HCT116 cells, both pThr806 and a phosphorylation outside the CTRs, pSer666, were lost rapidly upon treatment with 250 nM NVP-2 (T1/2 ~10 min), consistent with a similar, reinforcing effect of kinase inhibition and phosphatase activation (Fig. 1d).
To complete the potential Cdk9-PP1-Spt5 circuit in human cells, we sought to validate Spt5 as a target of PP1. Purified, recombinant PP1 was able to dephosphorylate a CTR1-derived peptide phosphorylated on the position equivalent to Thr806 in the intact protein, but was inert toward a pSer666-containing peptide derived from the KOW4–KOW5 linker (Fig. 1e). The pSer666 substrate was efficiently dephosphorylated by λ phosphatase, suggesting that this resistance was indeed due to restricted substrate specificity of PP1. We obtained similar results in assays of immunoprecipitated GFP-PP1 isoforms expressed in human cells (Supplementary Fig. 1b, c). Next we asked if Spt5 phosphorylation was sensitive to the loss of PP1 function in vivo. Depletion of all three PP1 catalytic subunits with small interfering RNA (siRNA) increased steady-state levels of pThr806 in extracts (Fig. 1f), whereas knockdown of PP1 isoforms individually or in pairwise combinations had negligible effects on Spt5 phosphorylation (Supplementary Fig. 1d), suggesting redundancy or compensation. In contrast, even the triple knockdown had no effect on pSer666 (Fig. 1f, Supplementary Fig. 1e), consistent with the insensitivity of this modification to PP1 in vitro. Thus, (1) Cdk9 phosphorylates both Spt5 and PP1γ in vivo, (2) PP1 can dephosphorylate an Spt5 CTR1-derived peptide (but not a pSer666-containing peptide) in vitro, and (3) levels of pThr806 (but not pSer666) are limited by PP1 activity in vivo. Taken together, these results indicate that the enzymatic elements of an elongation–termination switch defined in fission yeast are conserved in human cells, but suggest that a different phosphatase might target pSer666 (and possibly other sites in the elongation complex), perhaps to support a different function.
Spt5 CTR1 phosphorylation is low at the 3′ pause
We next asked if the output of Cdk9-PP1 signaling is similar in yeast and human cells. We first confirmed that an antibody raised against pThr8069 recognized CTR1 phosphorylated by Cdk9 in vitro (Supplementary Fig. 2a) but not unphosphorylated CTR1, or CTR2, another carboxy-terminal block of Thr-Pro-containing repeats in Spt538,39. The antibody is likely to recognize multiple repeats within CTR1, given that reactivity with Spt5 overexpressed in human cells was diminished by ~60% but not abolished by mutation of Thr806 to Ala (Supplementary Fig. 2b). In chromatin immunoprecipitation sequencing (ChIP-seq) analysis in HCT116 cells (Supplementary Fig. 2c), the distribution of total Spt5 closely matched that of transcribing Pol II, with peaks near the TSS and downstream of the CPS, indicating promoter–proximal and 3′ pausing, respectively (Fig. 2a, b). This is consistent with the tight association of DSIF with Pol II in elongation complexes40,41. In contrast, pThr806 (pCTR1) and pSer2—which have both been interpreted as markers of elongating Pol II—had different distributions. This was most evident downstream of the CPS, where total Spt5 accumulated together with Pol II; pSer2 peaked in this region, whereas pThr806 signals were diminished—a divergence evident both in metagene plots (Fig. 2a) and browser tracks from individual genes (Fig. 2b). Comparison of metagene plots of pThr806:Spt5 and pSer2:Pol II ratios revealed an inverse relationship (Fig. 2c): pThr806:Spt5 began to drop just upstream of the CPS and reached a minimum at the 3′ pause, whereas pSer2:Pol II increased at the CPS and peaked at the pause. There was little correlation between the modifications at either the TSS or termination zone (TZ), despite the high correlation between Pol II and Spt5 (Supplementary Fig. 2d, e). The loss of pCTR1, relative to total Spt5, was also evident in TZs of nonpolyadenylated, replication-dependent histone genes and small nuclear RNA (snRNA) and small nucleolar RNA (snoRNA) genes (Supplementary Fig. 3a, b).
A reduction in pCTR1 that precedes Pol II pausing and a pSer2 peak is consistent with a sitting-duck model, whereby slowing of elongation triggers pSer2, recruitment of CPFs and termination24,25. We detected a similar, reciprocal relationship between Spt5-CTD phosphorylation and pSer2 in fission yeast27, in which Pol II undergoes a metazoan-like pause downstream of the CPS42. In both human and fission yeast cells, Cdk9 inhibition slows elongation in gene bodies;21,22 to ask if this had the predicted, opposite effects on pThr806 and pSer2, we treated HCT116 cells with NVP-2 and performed ChIP-qPCR analysis. A 1-h treatment with 250 nM NVP-2 caused Pol II depletion from the gene body—as expected if promoter–proximal pause release was impeded—and near-complete loss of pThr806 on both MYC and GAPDH (Fig. 3a–c, Supplementary Fig. 4a–c). Absolute pSer2 levels were also diminished by NVP-2 treatment, but the pSer2:Pol II ratio was elevated two to threefold in gene bodies, suggesting ectopically increased pSer2 due to slowed elongation.
Spt5 phosphorylations have distinct chromatin distributions
The metagene plots of pSer2:Pol II and pThr806:Spt5 ratios (Fig. 2c) also diverged at 5′ ends of genes, where the former had a deep trough—presumably reflecting pausing of Pol II with high Ser5 phosphorylation (pSer5) but low pSer22—whereas the latter peaked, suggesting that paused complexes can contain high levels of pCTR1. Both pThr806 and pSer666 were among the many residues phosphorylated—in Spt5 and other components of the transcription machinery—when paused Pol II complexes were converted to elongating complexes by treatment with P-TEFb18. We therefore asked if pSer666 was enriched in complexes that had escaped the pause. In contrast to the anti-pThr806 antibody, anti-pSer666 was unable to recognize CTR1 or CTR2, but did react with full-length Spt5, dependent on preincubation with Cdk9 and ATP (Supplementary Fig. 5a). Moreover, immunoblot reactivity of Spt5 in cell extracts required an intact Ser666 residue (Supplementary Fig. 5b), suggesting that the antibody is specific for a single modification, and possibly explaining the lower signals it produced, relative to anti-pThr806, in ChIP-seq analysis (Fig. 2b). Despite lower signals, pSer666 had a distribution similar to that of total Spt5, including a peak downstream of the CPS (Figs. 2b, 4a, b, Supplementary Figs. 3a and 5c), suggesting that, unlike pThr806, pSer666 was largely retained in the 3′-paused complex. Moreover, a metagene analysis comparing pSer666:Spt5 and pThr806:Spt5 ratios revealed a trough, rather than a peak, of pSer666:Spt5 near the TSS, followed by an increase upon entering the gene body (Fig. 4c)
The relative increases in pSer666 downstream of the TSS were also detectable by ChIP-qPCR analysis on MYC and GAPDH genes, and sensitive to Cdk9 inhibition (Fig. 4d, Supplementary Fig. 5d). To ask if transcriptional induction triggers increased pSer666, we performed ChIP-qPCR at the p53-responsive, pause-regulated CDKN1A gene43. Upon p53 stabilization by nutlin-344, mRNA levels of both CDKN1A (encoding p21) and MDM2 increase in time-dependent fashion (Supplementary Fig. 6a). By 2 h, when mRNA induction was approximately half-maximal, Pol II was redistributed from the promoter–proximal pause site into the body of the CDKN1A gene (Supplementary Fig. 6b). Both total Spt5 and pThr806 signals increased roughly proportionally over the TSS and gene body, whereas pSer666 peaked ~0.5 kb downstream of the TSS (Fig. 4e). As was the case at constitutively expressed MYC and GAPDH, inhibiting Cdk9 diminished both pSer666 and pThr806 on the induced CDKN1A gene, but increased the pSer2:Pol II ratio in the ~2-kb region downstream of the TSS (Fig. 4e, Supplementary Fig. 6b).
ChIP-seq analysis comparing nutlin-3- to mock-treated HCT116 cells revealed differential distributions of pSer666 and pThr806 on p53-responsive genes. Browser tracks of two representative p53 targets, CDKN1A and GDF15 (Fig. 5a), as well as metagene plots of nutlin-3-induced genes (Fig. 5b, Supplementary Fig. 7a), showed (1) increased pSer666 but not pThr806, relative to total Spt5, in the region just downstream of the TSS; and (2) retention of pSer666 but diminution of pThr806 at the 3′ pause downstream of the CPS, where pSer2 signals were maximal. These changes were specific to p53-regulated genes; at p53-unresponsive genes selected for high Pol II occupancy in mock-treated cells, pThr806 signals were reduced, whereas total Spt5 and pSer666 were relatively unchanged, by nutlin-3 treatment (Fig. 5c). Quantification of reads downstream of the TSS revealed more significant increases in pSer666 than in pThr806 or pSer2, with the largest gains occurring on nonpause-regulated p53 target genes (Fig. 5d, Supplementary Fig. 7b–d). Downstream of the CPS there was a significant increase in pSer666 but not pThr806 in response to p53 induction (Supplementary Fig. 7e). Given the dependence of both pThr806 and pSer666 on Cdk9, their differential distributions might reflect removal by different phosphatases, a possibility we explore in the next section.
PP4 is a Cdk9-regulated Spt5 phosphatase
We reasoned that a pSer666 phosphatase might also be a target of negative regulation by Cdk9, based on the similar kinetics of pThr806 and pSer666 dephosphorylation after Cdk9 inhibition (Fig. 1d). Among the sites labeled by Cdk9 in HCT116 whole-cell extracts was Thr173 of the PP4 regulatory subunit PP4R29. This residue and several others in the PP4 complex were previously shown to be phosphorylated, by a CDK, to inhibit PP4 activity in response to mitotic-spindle poisons45. An anti-phospho-Thr173 antibody failed to detect this modification in cell extracts or in untreated anti-PP4R2 immunoprecipitates, but recognized immobilized PP4R2 after treatment with Cdk9 in vitro, validating PP4R2 as a potential P-TEFb substrate (Fig. 6a). A phosphatase precipitated with either anti-PP4R2 or anti-PP4C (catalytic subunit) antibodies was active toward both pSer666- and pThr806-containing phosphopeptides (Fig. 6b), in contrast to PP1, which only worked on the latter (Fig. 1e). Pretreatment of HCT116 cells with either NVP-2 or RO-3306 increased the phosphatase activity of anti-PP4R2 or -PP4C immunoprecipitates (Fig. 6b) without affecting immunoprecipitation efficiency or complex integrity (Supplementary Fig. 8a), suggesting negative regulation of PP4 activity by Cdk9 or Cdk1—similar to the situation with PP1γ (Fig. 1b). Moreover, incubation of anti-PP4R2 immunoprecipitates with purified Cdk9 and ATP prior to a phosphatase assay reduced activity ~3-fold, indicating that PP4 complexes were sensitive to direct inhibition by P-TEFb (Fig. 6c, Supplementary Fig. 8b).
To test whether PP4 regulated Spt5 phosphorylation in vivo, we depleted PP4 by infection with lentivirus vectors expressing shRNA targeting PP4C. Three different shRNAs each diminished PP4C levels by ~70–80%, and increased pSer666:Spt5 and pThr806:Spt5 signal ratios in immunoblots of chromatin extracts (Fig. 6d, Supplementary Fig. 8c). This was in contrast to effects of PP1 depletion, which preferentially affected pThr806 levels (Fig. 1f), but consistent with the substrate specificity of immunoprecipitated PP4 complexes in vitro (Fig. 6b).
PP4 supports promoter-proximal pausing
To ask if differential localization of PP4 and PP1 might contribute to the different spatial distributions of Spt5 phospho-isoforms on chromatin, we performed ChIP-qPCR analysis of PP4C, PP4R2, and PP1γ on the MYC, GAPDH, and CDKN1A genes (Fig. 7a, b, Supplementary Fig. 9a). Both PP4 subunits cross-linked predominantly between the TSS and ~2–3 kb downstream, and were present at low or undetectable levels near the 3′ ends of genes. PP1γ had nearly the opposite distribution, cross-linking at near-background levels between the TSS and +2 kb before peaking close to the CPS, consistent with its residence in the CPF28,46,47. We also performed ChIP-qPCR analysis in cells exposed to NVP-2 (250 nM, 1 h); this treatment had minimal effects on chromatin association of total PP4C (Supplementary Fig. 9b) or PP4R2 (Fig. 7c, top panel), but decreased the signals obtained with the anti-phospho-PP4R2-Thr173 antibody to near-baseline levels (Fig. 7c, middle and bottom panels), consistent with negative regulation of PP4 by P-TEFb on chromatin.
Finally, we asked if we could mimic effects of P-TEFb activity on Pol II distribution by decreasing cellular levels of PP4R2. Depletion of PP4R2 with siRNA nearly abolished both total PP4R2 and PP4R2-pThr173 ChIP signals (Supplementary Fig. 9c), increased both pSer666 and pThr806 in extracts (Fig. 7d) and shifted the distribution of Pol II into the bodies of the pause-regulated MYC, GAPDH, and CDKN1A genes (Fig. 7e, Supplementary Fig. 9d–f). This suggests a role of PP4 in imposing a barrier to elongation at the promoter–proximal pause; this barrier can apparently be lowered artificially by depletion of PP4R2 or, we surmise, physiologically by PP4-inhibitory phosphorylation catalyzed by Cdk9 (although we cannot formally exclude the possibility that PP4R2 depletion increases Pol II cross-linking to gene bodies by lowering its elongation rate). Taken together, our results suggest that two distinct Cdk9-phosphatase circuits operate at the beginning and end of the elongation phase in the Pol II transcription cycle (Fig. 7f).