Live-cell single-molecule dynamics of eukaryotic RNA polymerase machineries

Abstract

Eukaryotic gene expression is orchestrated by RNA polymerases (RNAPI, II, and III) and associated factors, yet their real-time dynamics remain obscure. Using single-molecule tracking in living yeast, we quantified the kinetics of 58 proteins encompassing three RNAP machineries. RNAPI and RNAPIII pre-initiation complexes (PICs) engage in long-lived chromatin interactions, contrasting with transient RNAPII PIC. We further report kinetics of RNAPII-associated factors for elongation, histone modification, C-terminal domain (CTD) modification, RNA processing, and termination. Many elongation factors show brief rather than persistent association, suggesting dynamic interactions with factor exchange, allowing a potential repertoire of regulatory events. CTD truncation reduces U1 snRNP residence time and intron retention in ribosomal protein genes, providing insights into co-transcriptional splicing. Our findings establish a framework of dynamic interactions of RNAP machineries.

Structured Abstract

INTRODUCTION

To study how gene expression is controlled in real time, we need to understand the binding kinetics of transcription proteins on chromatin. Steady-state and ensemble measurements have provided important insights, but rapid and heterogeneous behaviors could be masked. Using single-molecule tracking (SMT) in living yeast, we measured chromatin-binding kinetics for 58 representative proteins spanning the three RNA polymerase machineries (RNAPI, II, and III). We also examined how the C-terminal domain (CTD) of RNAPII affects residence times of CTD-associated factors and influences co-transcriptional splicing and histone modifications.

RATIONALE

SMT allows direct measurement of residence times for the duration of protein engagement on chromatin. Bi-exponential fits of the survival probability allow extraction of long-lived, likely functional binding from short-lived events, including non-specific binding and premature termination. However, the observed residence times are underestimated due to imaging artifacts such as photobleaching and out-of-focus motion. These artifacts are normally corrected with an exponential fit to the decay of stably bound histone H2B, which we have refined using a stretched-exponential model. We constructed strains with endogenously Halo-tagged components of RNAPI, II, and III pre-initiation complexes (PICs) and factors representing the entire RNAPII transcription cycle (upstream regulation, elongation, histone modification, CTD modification, RNA processing, and termination) for a total of 58 proteins. The corrected residence times for these factors allow us to reconstruct a kinetic framework for the temporal control of the three RNAP machineries and the multi-step RNAPII transcription cycle in vivo.

RESULTS

We revealed distinct kinetics for the three RNAP PICs. RNAPI and RNAPIII PICs show prolonged chromatin engagement, with stable components lasting tens to hundreds of seconds, consistent with their roles in constitutive rRNA and tRNA synthesis. In contrast, most RNAPII PIC components bind for only 2–5 s. Many RNAPII elongation factors inferred from genomic occupancy data to be persistently associated with elongating polymerase, instead display brief engagement (2–7 s), suggesting rapid factor exchange. Stable elongation factors (DSIF, Spt6, Spn1, and Paf1 complex) associate with chromatin for ~20–25 s, consistent with the expected RNAPII elongation rate. Termination factors bind and dissociate in ~3 s, contrasting with the 50-s residence time of termination roadblock Reb1. We next examined how CTD length impacts transcription protein kinetics using a truncation mutant (CTD9, retaining 9 of 26 heptad repeats). CTD9 reduces the bound fractions of major RNAPII PIC components while residence times (1/k_off) remain largely unchanged, indicating that CTD length primarily affects the on-rate (k_on) rather than the off-rate (k_off) for PIC assembly. Beyond initiation, CTD9 reduces U1 snRNP residence time more than two-fold (from ~5 to ~2 seconds), correlating with decreased intron retention for ribosomal protein genes, linking CTD length to co-transcriptional splicing. The residence time of Set2 histone methyltransferase decreases ~2-fold in the CTD9 mutant, yielding a nearly 10-fold reduction in H3K36me3 levels.

CONCLUSION

Our single-molecule measurements in living yeast reveal distinct binding kinetics of the three RNA polymerase machineries, complementing and refining insights from steady-state observations. Our data indicate how RNAP-specific dynamics, rapid factor kinetics, and CTD length regulate eukaryotic transcription and correlate factor residence times to functional outcomes in cotranscriptional splicing and histone modification.

Graphical Abstract

Live-cell SMT resolves binding kinetics of RNAP machineries

SMT quantifies chromatin-binding kinetics for 58 transcription proteins across RNAPI, II, and III machineries in living yeast (top). Residence time measurements reveal polymerase-specific PIC dynamics and transient binding of elongation and termination factors that escape detection by ensemble and steady-state approaches (bottom). POI, protein of interest.

Eukaryotic transcription is accomplished by three nuclear RNA polymerases (RNAPs): RNAPI, II, and III (1). RNAPI synthesizes most ribosomal RNAs (rRNAs), which are essential components of the ribosome. RNAPII transcribes protein-coding genes into messenger RNAs (mRNAs) and non-coding RNAs that regulate cellular processes (2, 3). RNAPIII transcribes small RNAs, such as 5S rRNA and transfer RNAs, which are critical for protein synthesis. Each RNA polymerase operates with a specific set of associated factors that modulate its activity throughout the transcription cycle, from initiation through elongation to termination (4). Despite decades of extensive biochemical, genomic, and structural investigations, as well as recent super-resolution imaging studies, the temporal control of the three RNAP machineries and their regulatory factors on chromatin in living cells is far from complete. Important models, such as the formation of partial transcription pre-initiation complexes (PICs) (5, 6), RNAP recycling for repeated rounds of transcription (7, 8), and gene promoter-terminator looping (9), are still being explored. Moreover, outstanding questions remain regarding which elongation factors interact transiently or bind persistently to the processively elongating RNAP (10), and whether termination factors are recruited in an ad hoc manner or act continuously as pre-engaged transcription roadblocks (11). In this study, we employ single-molecule tracking (SMT) in living yeast to systematically quantify the global physiological residence times of over 50 representative transcription proteins or protein complexes, covering key components of the multi-step transcription cycle for the three RNAPs in the nucleus, and establish a fundamental framework for the temporal dynamics of eukaryotic transcription. Our findings reveal distinct dynamics for each of the three RNAP PICs relevant to their specific functions, and further highlight a spectrum of long-lived and transient interactions for transcription elongation and termination factors.

The intrinsically disordered C-terminal domain (CTD) of RPB1 (RNA polymerase II subunit B1; Rpo21 in yeast), unique to the largest subunit of RNAPII, features 26 heptad repeats (YSPTSPS) in yeast and 52 repeats in mammals. The CTD is functionally modified mainly by phosphorylation (12–14), resulting in a diverse array of binding sites for multiple CTD-interacting factors for regulation of transcription and co-transcriptional activities (15–18). CTD truncation of RPB1 leads to transcription defects (18–29) and reduces the size of liquid droplets or condensates formed by purified CTD in vitro (30, 31). In yeast, the CTD maintains diffusion confinement of RNAPII around active genes, facilitating its search for chromatin targets (32). Here, we compared the binding dynamics of CTD-associating factors in a viable but growth-impaired truncation mutant with 9 heptads (CTD9) (32). We found minimal changes in residence times for most associating factors, but a sizeable reduction in the residence time of the U1 snRNP and Set2 histone methyltransferase, with a functional decrease in intron retention for ribosomal protein genes and reduced H3K36 methylation. These results reveal a kinetic link between CTD length, factor residence time, and co-transcriptional processes.

SMT and refined modeling reveal residence times of RNAPI, II, and III

We used SMT to measure the dynamics of representative subunits of RNAPI (Rpa190), RNAPII (Rpo21), and RNAPIII (Ret1) fused to HaloTag (Fig. 1, A to C). A ‘fast tracking’ imaging regime at high laser power and a short frame rate (10 ms/frame) captures both freely diffusing and chromatin-bound molecules (Fig. 1, A to C; and movie S1). We classify particle trajectories as free or bound (Fig. 1C) based on diffusive characteristics, including the apparent diffusion coefficient and the anomalous diffusion exponent (32). The three nuclear RNA polymerases show bound trajectories in their expected subnuclear regions, demonstrating their spatially specific functions (32) (Fig. 1C and fig. S3A). As shown in Figure 1C and corroborated by our previous study (32), free RNAPII is predominantly concentrated in the nucleoplasm, with marked exclusion from the nucleolus. Similarly, free RNAPIII is also excluded from the nucleolus, albeit to a lesser extent. A detectable amount of free RNAPI is observed in the nucleoplasm, but its bound trajectories are mainly located in the nucleolus.

Fig. 1. RNAPI and RNAPIII PICs exhibit prolonged stability.

(A) Example of bound and free single-molecule trajectories. Scale bar: 500 nm. (B) Protein of interest (POI) fused with HaloTag and labeled with JF552 for single-molecule tracking. Scale bar: 500 nm. (C) Bound and free trajectories of RNAPII, I, and III in the nucleus captured by fast tracking. 100 random trajectories are shown for each polymerase. Nucleolus highlighted in grey. Scale bar: 500 nm. (D) Schematic of RNAPI PIC. UE, upstream element; CE, core element. Transcription start site (TSS) is indicated by an arrow. (E) Survival probability of H2B-corrected residence times for RNAPI PIC (n = number of trajectories; mean ± s.e.m.). (F) Mean residence time of the stably bound fraction for RNAPI PIC (n = 5,000 resamplings; mean ± s.e.m.). CF^Rrn6 exhibits long residence time close to the H2B control limit, resamplings were thus constrained to yield a non-negative decay constant for an approximated residence time. (G) Transiently and stably bound fraction of RNAPI PIC (n = 5,000 resamplings; mean ± s.e.m.). (H) Schematic of RNAPIII PIC. A, A box; B, B box; IE, intermediate element; C, C box. TSS is indicated by an arrow. (I) Survival probability of H2B-corrected residence times for RNAPIII PIC (n = number of trajectories; mean ± s.e.m.). (J) Mean residence time of the stably bound fraction for RNAPIII PIC (n = 5,000 resamplings; mean ± s.e.m.). (K) Transiently and stably bound fraction of RNAPIII PIC (n = 5,000 resamplings; mean ± s.e.m.).

In addition to fast tracking, we utilized a ‘slow tracking’ regime that employs low laser power to minimize photobleaching, and a long frame rate (250 ms/frame) to allow motion blur. This approach captures the binding kinetics of bound molecules, with minimal interference from the freely diffusing population (fig. S1F and movie S2). We fit the survival probability of chromatin-bound molecules to a two-exponential decay model to dissect the fraction of transient and stable chromatin binding (fig. S2, D, I to P), and estimated the global mean residence time of the stably bound molecules (hereafter ‘residence time’; fig. S2, D, I to N). Because the temporal resolution of slow tracking is limited by photobleaching, nuclear and chromatin movement, microscopic drift, and dye photophysics, we developed a modeling approach by correcting the two-exponential decay curve against the stretched exponential decay recorded for long-lived histone H2B (2-exp_StrCORR; fig. S2, A to H; see Materials and Methods). This refined correction improves the accuracy and precision of the two-exponential fit for the three RNAPs (fig. S2, I to M), revealing residence times of 62.8 ± 23.8 s for RNAPI^Rpa190, 43.4 ± 6.6 s for RNAPII^Rpo21, and 3.8 ± 0.2 s for RNAPIII^Ret1 (Fig. 1, F and J; Fig. 2C; and fig. S2, I to N). These residence times, which reflect the global average duration of RNAPs binding on chromatin during initiation, elongation, and termination, correlate with the geometric means of their respective gene lengths (fig. S1, C to E), suggesting correspondence with functional activity.

Fig. 2. Short-lived RNAPII PIC with diverse dynamics of upstream co-factors.

(A) Schematic of RNAPII PIC. CAK, CDK activating kinase; UAS, upstream activation sequence. (B) Survival probability of H2B-corrected residence times for RNAPII PIC (n = number of trajectories; mean ± s.e.m.). (C) Mean residence time of the stably bound fraction for RNAPII PIC (n = 5,000 resamplings; mean ± s.e.m.). Given TBP’s involvement in all three nuclear RNAP PICs (Fig. 1, D and H; and Fig. 2A), we dissected its binding dynamics in different subnuclear compartments (fig. S3A); TBP^Spt15 exhibits a shorter residence time in the nucleoplasm (27.1 ± 4.1 s) than in the nucleolus (40.5 ± 10.4 s) (fig. S3, C and D). This aligns with chromatin immunoprecipitation experiments demonstrating rapid TBP turnover at RNAPII promoters, moderate stability at RNAPIII promoters, and the slowest exchange at the RNAPI 35S promoter (115). (D) Transiently and stably bound fraction of RNAPII PIC (n = 5,000 resamplings; mean ± s.e.m.). (E) Survival probability of H2B-corrected residence times for upstream co-factors (n = number of trajectories; mean ± s.e.m.). (F) Mean residence time of the stably bound fraction for upstream co-factors (n = 5,000 resamplings; mean ± s.e.m.). (G) Transiently and stably bound fraction of upstream co-factors (n = 5,000 resamplings; mean ± s.e.m.).

RNAPI and III PICs are long-lived, while RNAPII PIC is transient

RNAPI initiation in yeast begins with the binding of the 6-subunit upstream activation factor (UAF) to the upstream element (UE) of the 35S rRNA gene (33). UAF mediates the recruitment of the TATA-binding protein (TBP), a component common to all three nuclear RNAP PICs (Fig. 1, D and H; and Fig. 2A), which acts as a bridge to position the 3-subunit core factor (CF) to the core element (CE), located downstream of UE (33). The binding of CF to CE allows subsequent recruitment of the Rrn3-bound RNAPI to form the complete RNAPI PIC (Fig. 1D), where Rrn3 facilitates polymerase binding to the PIC components (4). Upon initiation, RNAPI dissociates from Rrn3, CF, and UAF (34), and the rest of the PIC potentially remains at the promoter (35–37). We find a residence time of 43.2 ± 12.9 s for UAF^Rrn9, 17.2 ± 2.2 s for Rrn3, and on the order of minutes for CF^Rrn6 (Fig. 1, E to G). These residence times of RNAPI PIC components are far longer than those observed in previous SMT studies of yeast RNAPII PIC components (38) and other chromatin-binding factors, including chromatin remodelers (fig. S9) (39) and transcription factors (32, 40, 41), which typically last for only several seconds. In yeast, the rRNA gene can be transcribed simultaneously by ~50 RNAPI molecules (42). Mechanistically, the observed long binding of the RNAPI PIC, especially for UAF and CF, supports a model in which these factors stay bound on the promoter, allowing efficient and persistent promoter priming for recruitment of multiple RNAPI complexes to yield a continuous chain of elongating polymerases (43). In vitro studies suggest that only CF and Rrn3 are required for RNAPI basal transcription (44), highlighting the importance of the long-lived CF for reinitiation of 35S rRNA synthesis.

RNAPIII transcription initiation at tRNA gene promoters (RNAPIII-type II promoters) requires TFIIIB and TFIIIC (Fig. 1H; top), with a third factor, TFIIIA, additionally required for 5S rRNA gene promoters (RNAPIII-type I promoters) (45) (Fig. 1H; bottom). At tRNA promoters, the two DNA-binding subcomplexes of TFIIIC, τ_A and τ_B, bind respectively to the A and B box sequences, downstream of the transcription start site (TSS). At 5S rRNA gene promoters lacking the B box, recruitment of TFIIIC requires prior binding of TFIIIA to the A box, intermediate element (IE), and C box (Fig. 1H; bottom). TFIIIC binding assists in the recruitment and positioning of TFIIIB (TBP, Bdp1, and Brf1) to a region upstream of the TSS. Although TFIIIB can bind independently to the TATA element present upstream of some RNAPIII genes in vitro (46), its engagement with chromatin in vivo requires TFIIIC (47) (Fig. 1H). Structural studies have suggested that TFIIIB recruitment leads to the dissociation of TFIIIC, with the τB subcomplex being displaced (48). τA may remain in contact with TFIIIB, followed by the recruitment of RNAPIII for transcription. Once TFIIIB binds stably on the promoter, it supports multiple rounds of RNAPIII transcription in vitro (49).

Our live-cell SMT data extend the biochemical and structural perspective of RNAPIII transcription initiation, providing additional insights into the physiological timing of these processes (Fig. 1, I to K). Among all RNAPIII PIC components, TFIIIB^Bdp1 has the longest residence time of 43.7 ± 6.6 s (Fig. 1J), consolidating its key role in directing RNAPIII reinitiation (49). In line with the distinct dissociation of TFIIIC (48), the residence time of TFIIIC_τB^Tfc3 is half as long as TFIIIC_τA^Tfc4 (16.6 ± 1.8 s versus 33.4 ± 5.8 s) (Fig. 1J). In contrast, TFIIIA^Pzf1 has a residence time of 7.5 ± 1 s (Fig. 1J), suggesting a more transient role in the assembly of the PIC for 5S rRNA genes. In addition to stable TFIIIB binding, recycling of the same RNAPIII molecule for multiple rounds of transcription has been observed in vitro to enhance transcription efficiency (50). Assuming elongation dominates the observed 3.8-second residence time of RNAPIII^Ret1 (Fig. 1J and fig. S1E), we estimate an elongation speed of 1.45 kb/min. This speed, which matches the in vitro estimates of 1.74 kb/min (51), does not align with a recycling model derived from in vitro biochemical conditions (50). If polymerase recycling were a dominant mechanism for RNAPIII transcription, a substantially longer in vivo residence time to accommodate the additional rounds of transcription should be observed.

In contrast to RNAPI and III, RNAPII PIC components, also known as general transcription factors (GTFs) (Fig. 2A), exhibit substantially shorter residence times (Fig. 2, B to D). We refined the previous estimates (38) by applying the stretched exponential correction described above (2-exp_StrCORR; fig. S2); GTFs generally have residence times ranging from 2–5 s, consistent with our earlier study (38) and with recent competition-chromatin immunoprecipitation data (52). TBP^Spt15 and TFIIA^Toa1 show longer residence times of 28.5 ± 8.9 s and 8.5 ± 0.9 s, respectively (Fig. 2C), consistent with their roles in stabilizing the early stages of hierarchical PIC assembly (53). Although TFIIB was previously proposed to work in concert with TFIIA to stabilize the TBP-DNA interaction during the initial phase of PIC formation (54), recent in vitro single-molecule imaging with purified human proteins showed unexpected dynamic binding behavior of TFIIB (55), in line with our observation of a brief 2.1 ± 0.1 s binding duration (Fig. 2C). Overall, we observe transient binding activity for the RNAPII PIC, contrasting with long-lived RNAPI and III PICs (Fig. 1, D to K). This suggests that most RNAPII PICs rapidly disassemble and reassemble on promoters for each transcription event (56, 57). However, the presence of a minor long-lived population of partial RNAPII PICs facilitating reinitiation (5, 52), similar to the behavior of RNAPI and III PICs, cannot be excluded.

Diverse binding dynamics of RNAPII upstream co-factors

We next explored the single-molecule dynamics of upstream co-factors in the RNAPII transcription pathway (Fig. 2, E to G). These factors exhibit a wide range of chromatin-binding behaviors: TBP inhibitor Mot1 ATPase, co-repressor Tup1-Cyc8^Cyc8, and general regulatory factor Reb1 (also a termination roadblock (11)) show longer residence times of tens of seconds (Fig. 2F). Reb1, in particular, exhibits a notable residence time of 50.0 ± 12.9 s (Fig. 2F), highlighting the long-binding activity of its function as a nucleosome-displacement or ‘pioneering’ factor as often observed across organisms (58–60). The TBP inhibitor NC2^Ncb2 and the SAGA^{Spt7, Spt8} complex involved in histone acetylation, bind transiently for 3–4 s (Ncb2: 4.0 ± 0.4 s; Spt7: 3.2 ± 0.3 s; Spt8: 3.4 ± 0.4 s; Fig. 2F).

Residence time analysis reveals stable and transient RNAPII elongation factors

Many RNAPII elongation-associated factors, previously assumed to be integral to the elongation complex based on gene-wide occupancies, display diverse binding dynamics (Fig. 3G). Long binding times were observed for DSIF^Spt4, Spt6, Spn1, and Paf1 complex (Paf1C^Ctr9), in the range of 20–25 s (Spt4: 25.4 ± 3.7 s; Spt6: 25.0 ± 2.6 s; Spn1: 24.4 ± 1.6 s; Ctr9: 21.2 ± 2.1 s), consistent with their roles as core components of the transcription elongation complex (61). Capping enzymes CE^Cet1 and Abd1 bind for 3.8 ± 0.5 s and 3.8 ± 0.6 s, respectively, while the cap-binding complex (CBC^Sto1) has a long residence time of 20.8 ± 1.8 s, which aligns with its function in protecting nascent RNA throughout elongation. In contrast, R-loop resolving factors TFIIS^Dst1 and the THO complex^Hpr1 exhibit dynamic binding of less than 2 s (Dst1: 1.5 ± 0.1 s; Hpr1: 1.7 ± 0.1 s). Histone chaperone FACT^Spt16 binds for 6.3 ± 0.7 s. Among splicing factor U1 snRNP^{Prp40, Prp39} bind for ~5 s (Prp40: 4.7 ± 0.5 s; Prp39: 4.5 ± 0.5 s), while U2 snRNP^Prp11, U4 snRNP^Prp3, U5 snRNP^Snu114, and NineTeen Complex (NTC^Cef1) bind for 2–3 s (Prp11: 2.9 ± 0.2 s; Prp3: 2.2 ± 0.2 s; Snu114: 2.2 ± 0.2 s; Cef1: 3.3 ± 0.4 s). Npl3, a multifaceted CTD and RNA-binding protein involved in RNA export and splicing, binds for 4.9 ± 0.8 s. CTD modifiers (CTDK-1^Ctk1, BUR kinase^Sgv1, and Ess1) have residence times of 4–6 s (Ctk1: 4.6 ± 0.6 s; Sgv1: 3.6 ± 0.6 s; Ess1: 5.6 ± 0.6 s). Histone modifiers (histone methyltransferases COMPASS^Swd1 and Set2; and histone deacetylase Set3 complex^Set3 and Rpd3S^Rco1) have residence times ranging from 3–7 s (Swd1: 4.2 ± 0.3 s; Set2: 5.1 ± 0.8 s; Set3: 6.8 ± 0.6 s; Rco1: 3.2 ± 0.5 s).

Fig. 3. Transient binding of RNAPII elongation factors suggests dynamic interactions over constant association.

(A) Schematic of RNAPII elongation-associated factors. EF, elongation factor. (B to F) Survival probability of H2B-corrected residence times for transient and stable elongation factors (B), capping and RNA binding factors (C), splicing factors (D), CTD modifiers (E), and histone modifiers (F) (n = number of trajectories; mean ± s.e.m.). (G) Mean residence time of the stably bound fraction for elongation-associated factors (n = 5,000 resamplings; mean ± s.e.m.). DSIF and Paf1C participate in both RNAPII and I elongation (116, 117). Both factors show residence times of ~25 s in the nucleoplasm (DSIF^Spt4: 24.4 ± 1.2 s; Paf1C^Ctr9: 26.5 ± 2.0 s) and ~8 s in the nucleolus (DSIF^Spt4: 8.1 ± 0.9 s; Paf1C^Ctr9: 8.5 ± 0.7 s),corresponding to their RNAPII- and RNAPI-associated activities, respectively (fig. S3, B, D and F). (H) Transiently and stably bound fraction of elongation-associated factors (n = 5,000 resamplings; mean ± s.e.m.).

Our findings suggest that many elongation factors interact transiently instead of persistently with elongating RNAPII (Fig. 3). These global dynamics, likely dominated by highly expressed genes, provide insights into the elongation process: (i) gradual changes in the steady-state CTD phosphorylation pattern during elongation, described by genomic studies (62), should be the consequence of multiple, transient enzyme-substrate interactions; (ii) co-transcriptional capping (63) and spliceosome recruitment (64) should be highly efficient, occurring within a 5-second timescale; (iii) co-transcriptional histone modification on the whole gene body likely requires repeated enzymatic interactions with RNAPII and histone tails, or multiple rounds of transcription; (iv) the binding time observed for stable elongation factors DSIF^Spt4, Spt6, Spn1, and Paf1C^Ctr9 should reflect the average duration for RNAPII elongation (Fig. 3G; and fig. S3). For highly expressed genes of 1 kb average length (fig. S1D), we estimate the in vivo elongation speed of RNAPII in yeast to be 2.5 kb/min, well within the 1–3 kb/min range documented on chromatin templates (65).

Contrasting kinetics of RNAPII termination factors and chromatin roadblocks

We expanded our analysis to factors involved in protein-coding gene termination (Fig. 4, A to E), including CF1A^Pcf11, CF1B^Hrp1, CPF^{Cft1, Ssu72}, and Rat1 complex^Rtt103 (Fig. 4A), as well as the NNS complex^Nrd1 for non-coding RNA gene termination (66) (Fig. 4B). All examined factors exhibit short residence times of ~3 s (Pcf11: 2.5 ± 0.3 s; Hrp1: 3.3 ± 0.6 s; Cft1: 2.4 ± 0.2 s; Ssu72: 2.1 ± 0.2 s; Rtt103: 2.9 ± 0.3 s; Nrd1: 2.6 ± 0.2 s; Fig. 4D), indicating transient recruitment and binding before dissociation. In contrast, Reb1, aforementioned as a general regulatory factor (59) (Fig. 2, E to G), also acts as a termination roadblock (11), showing a long residence time of 50.0 ± 12.9 s (Fig. 2F). Nsi1, a Reb1 homolog involved in RNAPI termination (67), also exhibits a relatively stable residence time of 30.3 ± 7.8 s (fig. S3, B to E), supporting its similar role as a roadblock for 35S gene termination (68). Overall, these findings underscore the substantial differences in physiological timescales arising from factors involved in distinct termination mechanisms—either through dynamic interactions with the RNAPII enzyme and nascent RNA, or by functioning as stable chromatin roadblocks.

Fig. 4. Brief engagement of RNAPII termination factors suggests rapid termination process.

(A and B) Schematic of RNAPII termination factors. (A) For protein-coding genes, the polyadenylation signal on nascent RNA consists of the AU-rich and A-rich elements upstream of the cleavage site, and the U-rich elements flanking it. These elements are bound by cleavage factor 1B (CF1B), cleavage factor 1A (CF1A), and the cleavage and polyadenylation factor (CPF), respectively. The Rat1 complex (Rat1C) contains RNA exonuclease activity to facilitate dissociation of polymerase from DNA (118). (B) Termination of non-coding RNAs is mediated by the NNS complex (Nrd1, Nab3, and Sen1), facilitated by binding to the Nrd1-binding motif GUAA/G and Nab3-binding motif UCUU(G) on the nascent transcript (119). (C) Survival probability of H2B-corrected residence times for termination factors (n = number of trajectories; mean ± s.e.m.). (D) Mean residence time of the stably bound fraction for termination factors (n = 5,000 resamplings; mean ± s.e.m.). (E) Transiently and stably bound fraction of termination factors (n = 5,000 resamplings; mean ± s.e.m.). (F) Model of RNAPII transcription cycle binding dynamics characterized by residence time analysis.

System-wide SMT of RNAPII machinery in a CTD truncation background

The RNAPII CTD acts as a central hub for interactions with transcription-associated proteins throughout the transcription cycle (12–14). Integrating and extending data from our previous work (32), we demonstrate that CTD truncation from 26 to 9 heptad repeats (CTD9) (Fig. 5A) impairs PIC formation, substantially reducing the stably bound fraction of RNAPII^Rpo21 and PIC components TFIIA^Toa1, TFIIB^Sua7, TFIIE^Tfa1, TFIIF^Tfg1, and TFIIH^{Tfb4, Kin28} (fig. S4). In contrast, upstream co-factors and early PIC components TBP^Spt15 and TFIID^Taf1 exhibit minimal changes in the stably bound fraction (fig. S4). As expected, many downstream interacting factors involved in stages beyond initiation show reduced stably bound fractions (fig. S4). Residence time for the PIC and many other factors remains largely unchanged (Fig. 5B), indicating that the primary impact of reduced CTD length, at least for the PIC, is on the on-rate (association rate), rather than the off-rate (dissociation rate). This observation aligns with a recent study in human cells, where RNAPII with only 5 CTD heptads associates with DNA less frequently, but once initiated, remains pervasive for transcription (69).

Fig. 5. RNAPII CTD truncation alters CTD phosphorylation, histone modification, and ribosomal protein gene splicing.

(A) Schematic of the disordered C-terminal domain (CTD) of RPB1 (Rpo21), the largest subunit of RNAPII. WT CTD comprises 26 heptad repeats (CTD26) with the consensus sequence YSPTSPS. (B) Log₂ fold change (FC) in mean residence time of the stably bound population of RNAPII and associated proteins between CTD9 and WT (n = 5,000 resamplings; mean ± s.e.m.). The reduction of RNAPII^Rpo21 residence time in CTD truncation likely reflects compromised pre-initiation complex formation and engagement with the Mediator, preventing most bound RNAPII molecules from progressing to productive elongation. The resulting residence time is thus dominated by shorter initiation events rather than longer elongation phases (32). In contrast, stable elongation factor residence times reflect the duration of RNAPII on chromatin during elongation, which remains similar between WT and CTD9. For TBP^Spt15, DSIF^Spt4, and Paf1C^Ctr9, only nucleoplasmic trajectories were analyzed, excluding nucleolar events to enrich for RNAPII-associated activities (fig. S3). (C and D) CTD phosphorylation levels for FLAG-tagged Rpo21 in CTD26 (WT) and CTD9. Representative western blot (C) and quantification using dilution series with normalization to total Rpo21 and correction for the ratio of phosphorylatable sites (D). Bars indicate means, dots represent biological replicates (n = 3); Unpaired Student’s t-test, * P < 0.05, ** P < 0.01, *** P < 0.001. (E and F) Histone H3 modification in CTD26 (WT) and CTD9. Representative western blot (E) and quantification normalized to H3 (F). Bars indicate means, dots represent biological replicates (n = 3); Unpaired Student’s t-test, ** P < 0.01, *** P < 0.001, ns = not significant. (G) Mean residence time of the stably bound fraction for U1 snRNP^Prp40 in CTD truncation mutants (n = 5,000 resamplings; mean ± s.e.m.; Unpaired Student’s t-test, * P < 0.05). (H) Fold change in intron retention (IR) ratio for ribosomal protein (RP) genes and other intron-containing (Non-RP) genes between CTD9 and WT (n = 4 biological replicates; median values shown; Mann-Whitney Test, *** P = 2.01 × 10⁻²²). (I and J) Fold change in IR ratio against expression level (FPKM, fragments per kilobase million) (I) or intron length (J) for all intron-containing genes between CTD9 and WT. Linear regression with 95% confidence interval shown. n = 4 biological replicates.

A handful of elongation-associated factors, including U1 snRNP^{Prp40, Prp39}, histone chaperone FACT^Spt16, histone H3K36 methyltransferase Set2, and histone deacetylase SET3 complex^Set3, show sizeable reductions in residence time in the CTD9 mutant (Fig. 5B). CTD truncation is accompanied by reduced CTD phosphorylation (before and after correcting for the number of phosphorylatable target sites; Fig. 5, C and D), which could contribute to the altered factor dynamics through direct or indirect recruitment mechanisms that depend on phosphorylated CTD (70–79). Additionally, the reduced H3K36me2/me3 and H3K4me3 levels in the CTD9 mutant (Fig. 5, E and F) suggest a broad epigenetic alteration.

Notably, two subunits of the U1 snRNP, Prp40 and Prp39, but not the other key spliceosomal components (U2, U4, U5 snRNPs and NTC), show reduced residence times in the CTD9 mutant (Fig. 5B and fig. S8), indicating a specific effect of the CTD length on U1 snRNP activity. In particular, progressively shorter U1 snRNP residence times are observed across a series of CTD truncation mutants (Fig. 5G). U1 snRNP is shown to interact with the CTD biochemically, mainly through the subunit Prp40 (70, 80). To assess the functional consequence of reduced U1 snRNP residence time, we examined splicing outcomes in the CTD9 mutant by RNA-sequencing. While 95% of yeast genes are intronless, two-thirds of ribosomal protein (RP) genes contain introns and are predominantly spliced co-transcriptionally (64). We found that CTD truncation leads to a decrease in intron retention (IR) (Fig. 5H), i.e., more mature mRNA molecules, specifically for RP genes, without significant change of their expression levels (fig. S5, A to C). We observe a negative correlation between the IR ratio with gene expression level (Fig. 5I) and intron length (Fig. 5J), although this pattern is largely due to the unique properties of RP genes, which typically exhibit higher expression levels and longer introns than other intron-containing genes (Fig. 5, I and J; and fig. S5, F and G). Overall, our systematic SMT analysis of RNAPII machinery in the CTD9 background reveals altered U1 snRNP dynamics and suggests CTD length can influence pre-mRNA splicing of RP genes.

Discussion

In this work, we present a systematic quantitative analysis of dynamic chromatin interactions for RNAPI, II, and III transcription machineries in living yeast, and perform a system-wide analysis for the effect of CTD truncation on the residence time of RNAPII-associated factors from initiation to elongation, RNA processing, and termination. We consider residence time as a proxy for binding stability once engaged, whereas the bound fraction can be confounded by downstream effects of compromised initiation in the CTD truncation mutant (32).

In yeast, a minimum of 8 CTD heptads is required for viability, but with substantial growth retardation (20). Our study utilizes a CTD9 mutant (Fig. 5A), which maintains reasonable fitness (fig. S10B) (20) for large-scale protein tagging and SMT. Yeast RP genes are primarily spliced co-transcriptionally soon after the 3’ splice site is transcribed (64, 81). The interaction between the CTD and U1 snRNP (70) is proposed to facilitate the rapid search for splice sites during nascent RNA transcription and promote efficient co-transcriptional splicing (82). For the CTD9 mutant, we found a ~2-fold reduction in U1 snRNP^Prp40 residence time (Fig. 5, B and G) and splicing of RP genes is unexpectedly enhanced rather than diminished (Fig. 5H), suggesting a gain-of-function phenotype. In this regard, truncation of human RNAPII CTD displays mixed behavior: shortening the CTD from 52 to 5 heptads results in splicing inhibition and lethality (83), while a viable mutant with 25 heptads shows minimal effects on pre-mRNA splicing (26), and another mutant with 26 heptads exhibits perturbation of alternative splicing (84).

One possibility for our findings in yeast is that a truncated CTD might constrain the local search process for U1 snRNP, enabling it to locate splice sites even more efficiently. This would account for the reduced U1 snRNP residence time (Fig. 5, B and H; and fig. S8B) and decreased intron retention in RP genes (Fig. 5H). U1 snRNP activity might additionally be affected by the altered phosphorylation level on truncated CTD heptads (Fig. 5, C and D). Changes in RNAPII elongation speed can also influence splicing efficiency (81), but this does not seem to be altered in the CTD9 mutant, as stable elongation factors DSIF^Spt4, Spt6, Spn1, and Paf1C^Ctr9 exhibit residence times comparable to wild type (WT) (Fig. 5B). Analysis of two representative ribosomal proteins (Rps16a and Rpl43b) showed no clear correlation between intron retention and protein abundance (fig. S5, A to E). The broader functional consequences of CTD truncation on RNA processing warrant future investigation. Further mechanistic insights into the highly transient nature of U1 snRNP interactions (below 5 seconds) will benefit from emerging single-molecule techniques that capture protein-protein interactions in real-time—including in vitro TIRF imaging (6, 85) and proximity-assisted photoactivation (PAPA) in living cells (86).

PICs of RNAPI and III are characterized by prolonged chromatin engagement, with the most stable components ranging from tens to hundreds of seconds (Fig. 1, F and J), in line with their roles in maintaining a ‘default ON’ state for rRNA and tRNA gene transcription. In contrast, most RNAPII PIC components, except for TBP, are marked by brief chromatin binding of several seconds (Fig. 2C), reflecting flexibility for rapid, inducible regulation of mRNA and non-coding RNA expression. Our supplementary kinetic modeling of temporal occupancy, defined as the percentage of time a chromatin target is occupied by a factor, indicates that target sites of long-binding RNAPI PIC components CF^Rrn6 and UAF^Rrn9, and RNAPIII PIC component TFIIIB^Bdp1, exhibit occupancy close to 100% (fig. S6). This supports the model where RNAPI and III promoters are constantly bound by their PIC components and are constitutively accessible. In contrast, chromatin target sites associated with RNAPII PIC components have a much lower occupancy of ~10%, as shown by a previous study from our laboratory (38).

The prolonged binding of the RNAPI PIC components CF and UAF observed in live yeast (Fig. 1F) differs from in vitro experiments using yeast whole-cell extracts, where CF dissociates upon transcription initiation while UAF remains bound (36). However, experiments with HeLa cell nuclear extracts show that the human homologs of CF (SL1) and UAF (UBF) remain bound to the promoter during multiple rounds of transcription (37). The basis of these differences is not known and requires further in vitro studies to reconcile with the observations in living cells.

Our dynamic model of elongation-associated factor binding (Fig. 4F) could explain how the RNAPII CTD enables numerous interactions, linking histone modification, CTD modification, and RNA processing to transcription elongation. Transient association would maintain CTD binding site availability, allowing a greater number of CTD-interacting macromolecular complexes to bind over time without steric hindrance, especially as effective CTD binding typically requires two consecutive heptad repeats (87). Furthermore, transient interactions may expand the repertoire of biological controls over elongation rates and co-transcriptional activities, allowing more adaptable regulation compared to a continuously bound set of elongation factors. We also note that the CTD maintains diffusion confinement of two transient CTD-binding proteins, CTDK-1^Ctk1 and BUR kinase^Sgv1, around a subnuclear region containing active genes (fig. S7), extending our previous finding that CTD drives spatiotemporal confinement of RNAPII to enhance its search for chromatin targets (32).

Previous studies have demonstrated that CTD truncation or disruption of CTD-Set2 interactions reduces H3K36 methylation (71–76, 78). Here, we further show that Set2 exhibits a ~2-fold reduction in residence time in the CTD9 mutant (5.1 ± 0.8 s to 2.3 ± 0.4 s; Fig. 5B), and this corresponds to a ~10-fold decrease in H3K36me3 levels (Fig. 5, E and F). This non-linear relationship between binding kinetics and the final enzymatic product provides an insightful quantitative relationship of how factor residence time translates to functional outcomes.

The binding duration of transcription termination factors in vivo is brief, averaging only 3 s (Fig. 4D). This contrasts with an early model of termination for the long MDN1 gene, based on fluctuation analysis of fluorescent nascent RNA in live yeast, which suggested a termination time of approximately 70 s (88). The rapid turnover implied by our measurements suggests that assembly and disassembly of termination complexes should occur much faster than previously considered. The different findings could be due to MDN1 gene-specific effects compared to the global average, or to the nascent RNA reporter lingering on-site longer than the actual termination event.

The dynamics of transcription proteins revealed in our live-cell study likely reflect activities that are predominantly associated with highly expressed genes. While these factors are crucial in transcription and exhibit considerable occupancy on active genes (89), it is important to note that they may also be involved in other cellular functions beyond transcription, such as cell cycle regulation and DNA repair. It would be of interest to measure single-molecule dynamics of multiple factors simultaneously within specific gene loci in vivo (41, 90, 91) and employ in vitro single-molecule imaging on native or engineered gene templates (6) to elucidate temporal relationships and interactions of the transcription machinery. A recent study on single-molecule tracking in mouse embryonic stem cells revealed residence times of tens of seconds for RNAPII PIC components (92)—an order of magnitude longer than our < 10-second yeast timescales. Notably, higher eukaryotes have evolved more complex transcription mechanisms that are absent from budding yeast, such as promoter-proximal pausing (93) and long-range enhancer-promoter interactions (94). As high-throughput single-molecule tracking platforms emerge in mammalian systems (95), a direct comparison of the full spectrum of transcriptional dynamics across eukaryotic lineages will become feasible.

Materials and Methods

Yeast construction

All Saccharomyces cerevisiae strains in this study (table S1) are isogenic derivatives of BY4741 with pdr5 deletion for enhanced HaloTag ligand labeling, and with GFP fused to endogenous loci of ER (Elo3) and nucleolar (Gar1) markers. Strains with pdr5Δ and GFP markers exhibit growth identical to the parental BY4741 strain (32). The CTD9 mutation was engineered via knock-in at the endogenous locus and retains the tip domain and non-consensus heptads (32). Yeast strains with protein of interest fused with HaloTag exhibit growth rates comparable to WT (fig. S10B). Expression of full-length Halo fusion proteins was validated by in-gel fluorescence (fig. S10A). Selected Halo-tagged constructs were functionally validated for protein interactions by PICT (Protein Interactions from Imaging of Complexes after Translocation (96); fig. S11) and enzymatic activities by western blotting (fig. S12).

Microscope setup

The microscope setup was previously described (32). Briefly, imaging was conducted using a custom-built Zeiss Axio Observer Z1 microscope (Zeiss) with a Plan-Apochromat 150X/1.35 glycerin immersion objective (Zeiss). Data were recorded using an EM-CCD camera (C9100–13, Hamamatsu Photonics) with a 16 μm physical pixel size and a 107 nm pixel size in recorded images, controlled by HCImage (v5.0.1). Laser excitation wavelengths were 488 nm for GFP, 555 nm for JF552, and 637 nm for miRFP670nano3. Laser power and alignment were routinely measured and adjusted throughout imaging sessions to ensure consistency.

Single-molecule tracking

Yeast cultures in synthetic complete (SC) medium were incubated with JF552-HaloTag ligand during early logarithmic growth (OD₆₀₀ = 0.2–0.3) for 4 hours. HaloTag fusion proteins were labeled with adjusted dye concentrations to ensure sparse labeling. After at least six washes, cells were attached to concanavalin A-coated coverslips in Attofluor^™ cell chambers (Invitrogen) and imaged at room temperature. Initial illumination with a 555 nm laser produced a strong nuclear signal, followed by a transient shift of fluorophores into a metastable dark state. JF552 spontaneously and randomly returned to a fluorescent state, allowing for sparse single-molecule detection. G1 cells were selected for analysis using GFP markers to minimize variability in transcriptional activity arising from cell cycle progression (32) (fig. S1A). Each imaging session lasted up to two hours, equivalent to one cell cycle in SC at room temperature, typically covering 10–40 fields of view. For robust statistics, all datasets included at least 1,000 trajectories from at least two independent imaging sessions. Laser exposure during imaging showed no substantial effect on cell cycle progression, as previously demonstrated (32). Single-molecule localization and trajectory linking were performed using Diatrack (v3.05) (97), with a maximum jump set at 6 pixels (642 nm) for fast tracking and 3 pixels (321 nm) for slow tracking (32). Nuclear masks were manually drawn to exclude cytoplasmic localizations. Single-step disappearances of bound molecules, observed in both fast and slow tracking from our previous study, suggest accurate identification of single-molecule signals (32).

Fast tracking

Fast tracking was conducted at 10 ms/frame with continuous 555 nm laser irradiation at ~1 kW/cm² for analyzing bound and free molecule dynamics. Static localization error was ~20 nm (32). To minimize the inclusion of cytoplasmic diffusing trajectories, we exclusively selected cells exhibiting minimal nuclear drift during imaging (32). Transitioning trajectories identified by vbSPT (98) were separated into single-state segments. Each trajectory was then classified as chromatin-bound for the slow population or freely diffusing for the fast population based on diffusivity, geometry, and angular orientation using a multi-parameter approach described previously (32). Freely diffusing trajectories of CTDK-1^Ctk1 and BUR kinase^Sgv1 were further analyzed for spatiotemporal confinement using spatiotemporal mapping and angular movement analysis as described in our previous work (32) (fig. S7).

We functionally attribute the slow trajectories to chromatin binding, either directly or indirectly, as evidenced by their enrichment in subnuclear compartments with specific binding sites (32) (Fig. 1C). These trajectories exhibit a mean apparent diffusion coefficient similar to that of histone H2B, yet they are typically an order of magnitude slower than the fast, freely diffusing population (32) (fig. S13). Notably, this chromatin-associated population was absent in the HaloTag control experiments (32). Biological perturbations affecting factor binding typically result in a reduced fraction of this population (32, 38, 39).

We acknowledge that protein factors may exist independently or as part of various large complexes, resulting in freely diffusing trajectories with intermediate apparent diffusion coefficients. Bound molecules could also exhibit a spectrum of dynamics on compact chromatin that may be functionally relevant but challenging to resolve by live-cell SMT, such as hopping or sliding; these local kinetics are more effectively captured by in vitro techniques such as TIRF imaging (6) or optical trapping (99) using extended linear DNA or chromatin templates. We recognize this as a limitation of our current approach. In this study, we adhere to a simplified 2-state model to classify the principal states as free and bound, as this provides a tractable framework for analyzing the factors’ behavior under our experimental conditions systematically.

Slow tracking

Slow tracking was conducted at 250 ms/frame using continuous 555 nm laser irradiation at a reduced power of ~0.035 kW/cm² to minimize photobleaching and improve residence time resolution. We spiked the imaging culture with Halo-H2B cells labeled with the nuclear marker Pus1-miRFP670nano3 and co-imaged the two strains in the same field of view (32) (fig. S1B). To account for blinking or transient defocalization of bound molecules, we allowed gaps of up to one frame between localizations and linked them as a single trajectory using the same 3-pixel linking criteria as mentioned above. This 3-pixel threshold was chosen based on our previous work estimated that 98% of bound yeast RNAPII, H2B, H3, and H2A.Z single-frame displacements fall within this limit, while smaller thresholds (2 pixels) capture only ~90% on average (32). As chromatin movement in yeast is estimated to be 5-fold more dynamic than in mammalian cells (32, 100), a larger linking distance is required to avoid under-connecting.

A survival curve (1-CDF) against time (t) was plotted and fitted with a two-exponential decay model, incorporating a stretched exponential decay of H2B to account for experimental factors such as photobleaching, nuclear and chromatin movement, microscope drift, and dye photophysics (2-exp_StrCORR; fig. S2D):

$$y=f_{tb}{e}^{-k_{tb}t}+f_{sb}{e}^{-\left(k_{sb}t+{\left(k_{2}t\right)}^{\beta }\right)}$$

where $k_{tb}$ and $k_{sb}$ are the dissociation constants for the transient and stable binding populations, respectively. $k_2$ is the dissociation constant for stable binding population of H 2 B fitted with a stretched exponential decay, and $\beta$ is the stretching exponent ranges between 0 and 1. The fractions of the two components, $f_{lb}$ and $f_{sb}$, add up to 1 (fig. S2D). The transient population, with sub-second residence times, is minimally affected by the experimental factors mentioned above; thus its dissociation rate is not corrected with H2B decay to avoid over-fitting. The H2B correction in this model assumes no histone exchange during the observation window, which is justified as single molecules are observed for much shorter periods (99% disappear within 30 seconds; fig. S2B) than the expected timescales of histone exchange (minutes to hours in human cells (101)). For data visualization, we present the survival probability plots using a linear y-axis and logarithmic x-axis to facilitate interpretation of the two-component model. This scaling reveals the biphasic kinetics clearly with the transition point between transient and stable binding populations, which allows intuitive estimation of both fraction values. Alternatively, we present these plots in a logarithmic y-axis and linear x-axis in fig. S14 for comparison.

The overall stably bound fraction was determined by multiplying the total bound fraction from fast tracking with the stably bound fraction ($f_{sb}$) from slow tracking. One limitation of this approach is the differing exposure times used for fast and slow tracking (10 ms/frame versus 250 ms/frame), as the longer frame rate in slow tracking may not effectively capture highly transient binding events. However, the estimated stably bound fraction still provides a valuable metric for assessing changes in binding kinetics. Stable binding is interpreted as protein activity at specific sites (32, 38–41, 102), while we regard transient binding as a neutral description of duration in two-component fitting, rather than as merely non-specific site binding (32).

Residence time correction with stretched exponential decay of H2B

Traditionally, the survival curve of chromatin-binding proteins is fitted with a two-exponential decay model as stable and transient populations (fig. S2, O and P). The observed residence time of the stable population is then corrected for experimental factors like photobleaching, nuclear and chromatin movement, microscopic drift, and dye photophysics (32), using the slow decay constant from a separate two-exponential decay fit of H2B (2-exp_ExpCORR) (fig. S2, A to D). For optimal accuracy, the Halo-H2B sample should be imaged on the same day (102) or added as a spike-in control, as in this study (32) (fig. S1B). However, studies in mammalian cells suggest that a power-law model may more accurately represent the chromatin binding behavior of transcription factors (103, 104). This model proposes that transcription factors bind across a continuum of affinities at multiple sites and cannot be distinctly categorized into transient and stable populations; thus, specific residence times cannot be derived. In this approach, the survival curve of the protein of interest is first adjusted using the slowest exponential decay constant from a three-exponential decay fit of H2B, followed by fitting to a power-law distribution (103, 104).

Given that classic kinetic parameters like rate constant and residence time are not obtainable with power-law fitting (105), we explored an alternative model to account for the continuum-like feature of the survival curve, using the three nuclear RNAPs in yeast as test cases. Similar to findings with mammalian transcription factors, we found that a two-component model with standard H2B correction (2-exp_ExpCORR) does not accurately capture the decay curve for the three RNAPs (fig. S2, I to L), consistent with previous findings for yeast RNAPII (106). We asked whether refining the modeling of the H2B decay for correction could enhance the curve fitting of the RNAPs. Our analysis revealed that fitting the H2B decay with a combination of one exponential and one stretched exponential decay (1e1s) is superior to a simple two-exponential decay (2-exp) (fig. S2, B to D). This revised model describes the experimentally observed bound H2B as two populations: a transient population described by standard exponential decay, and a stable population characterized by a complex ‘stretched decay’ profile, which better accounts for the experimental factors described above—including photobleaching, chromatin movement, nuclear motions, and dye photophysics—that create a spectrum of apparent decay rates. The stretched exponential decay naturally captures this continuum in a mathematically compact form. Notably, stretched exponential decay has also been adapted for modeling fluorescence decay in complex systems (107, 108).

To provide physical justification, we developed computational simulations showing that chromatin dynamics, causing bound particles to diffuse in and out of the detection plane, account for the stretched-exponential behavior observed experimentally. We modeled chromatin movement as fractional Brownian motion with Hurst exponent H (109), together with two levels of confinement: a harmonic potential that gently pulls particles back toward their initial positions to model nucleosomal tethering with relaxation rate κ, and a spherical nuclear boundary to mimic the yeast nuclear radius of 1 μm. Particles are subject to photobleaching with exponential decay rate k_bleach. Positions are recorded when particles are within the depth of field of our microscope (0.5638 μm) and projected to a 2D plane for analysis (fig. S2E).

We extracted simulation parameters from the experimental slow-tracking H2B MSD (mean squared displacement) curve: Hurst exponent H = 0.36, generalized diffusion coefficient D = 0.00365 μm²·s^−2H, relaxation rate κ = 0.026 s⁻¹, and photobleaching rate k_bleach = 0.095 s⁻¹ (fig. S2F). Photobleaching introduces survival bias into MSD measurements; as molecules bleach over time, the data at longer time lags are systematically enriched with longer-lived molecules, altering the measured displacement statistics. This relationship between molecular survival and apparent motion over time allows extraction of photobleaching rate from the experimental MSD data. Using these parameters, we simulated H2B residence times with both chromatin dynamics and photobleaching. The resulting survival probability of residence time was better fit by a stretched exponential rather than a single exponential. In contrast, simulation with photobleaching alone yielded standard exponential decay (fig. S2, G and H). This simulation demonstrates that chromatin dynamics can contribute to the stretched-exponential behavior observed in our system.

Importantly, applying the stretched decay of H2B for correction (2-exp_StrCORR) refines the two-exponential fitting of the three nuclear RNAPs in yeast, with marked improvement in residence time estimation and resolution, particularly for RNAPI and RNAPII, which have longer residence times (fig. S2, I to N). Reanalysis of a published slow tracking dataset for the yeast chromatin remodelers RSC and INO80 also showed similar improvement (39) (fig. S9). Further research is needed to evaluate the applicability of this approach in organisms beyond yeast, and for different experimental conditions and imaging setups.

Western blotting

Proteins were extracted using the LiAc/NaOH method (110), followed by boiling in Laemmli buffer at 95°C for 5 minutes. After separation by SDS-PAGE, the proteins were transferred to a polyvinylidene difluoride membrane using a Trans-Blot SD Semi-Dry Transfer Cell (Bio-Rad). The blots were blocked with 5% non-fat milk or 3% BSA in Tris-buffered saline with 0.05% Tween-20 (TBST) and then incubated with primary antibody diluted in blocking buffer at 4°C overnight. After washing with TBST, blots were incubated with secondary antibody diluted in TBST for 1 hour at room temperature. The blots were developed using Amersham ECL Prime or ECL Select Western Blotting Detection Reagent (Cytiva) and imaged with a ChemiDoc Imaging System (Bio-Rad). Primary antibodies used: CTD Tyr1P (3D12; MilliporeSigma), CTD Ser2P (3E10; MilliporeSigma), CTD Thr4P (6D7; Active Motif), CTD Ser5P (3E8; MilliporeSigma), CTD Ser7P (4E12; MilliporeSigma), CTD (for total Rpo21 in fig. S12A; 8WG16; MilliporeSigma), FLAG (D6W5B; Cell Signaling Technology), H3K4me1 (61781; Active Motif), H3K4me2 (07–030; MilliporeSigma), H3K4me3 (04–745; MilliporeSigma), H3K9ac (07–352; MilliporeSigma), H3K9/14ac (9677; Cell Signaling Technology), H3K36me1 (Ab9048; Abcam), H3K36me2 (39255; Active Motif), H3K36me3 (61101; Active Motif), H3 (ab1791; Abcam), Rpb3 (1Y26; BioLegend), PGK1 (22C5D8; Thermo Fisher Scientific), and HA (12CA5; Thermo Fisher Scientific). Secondary antibody used: sheep anti-mouse immunoglobulin G-horseradish peroxidase (NA931; Cytiva), donkey anti-rabbit immunoglobulin G-horseradish peroxidase (NA934; Cytiva), and goat anti-rat immunoglobulin G-horseradish peroxidase (NA935; Cytiva).

RNA-sequencing

Yeast cells grown in SC medium were harvested and homogenized with bead beating. Total RNA was isolated using the SPLIT RNA Extraction Kit (Lexogen) and treated with DNase I (NEB) to eliminate genomic DNA. rRNA was depleted using the Ribocop for Yeast kit (Lexogen). Spike-in RNA variant (SIRV-Set 3; Lexogen) was added as quality control measures between samples (fig. S5I), and sequencing libraries were prepared with the CORALL Total RNA-Seq V2 kit (Lexogen). Libraries were sequenced on the Element AVITI^™ System (Element Biosciences, USA) (paired-end 2×75 bp). Four biological replicates were performed, yielding approximately 100M reads each. Adapter sequences were trimmed using Cutadapt (v1.18), and reads were aligned to the S. cerevisiae genome (Genome Assembly R64) using STAR (v2.6.1a). PCR duplicates were removed by collapsing reads with identical mapping coordinates and UMI (unique molecular identifier) sequences. Expression levels were calculated using Mix² (v1.4.0.2; Lexogen) and normalization was performed using DESeq2’s median-of-ratios method (fig. S5H). Intron retention (IR) ratio was quantified using IRFinder-S (v2.0) (111).

In-gel fluorescence

Yeast cultures grown in SC medium were incubated with JFX650-HaloTag ligand for 4 hours. After washing with water, proteins were extracted and separated by SDS-PAGE as described above. Fluorescence signals of the labeled HaloTag proteins were detected using an Amersham Typhoon 5 Biomolecular Imager (Cytiva). Gels were subsequently stained with InstantBlue Coomassie Protein Stain (Abcam) for total protein visualization.

Protein interactions from imaging of complexes after translocation (PICT)

PICT detects protein-protein interactions in living cells by monitoring co-translocation of interacting partners based on the rapamycin-induced dimerization of FK506-binding protein (FKBP) and FKBP-rapamycin binding (FRB) domains (96, 112). Our system uses three engineered components: (i) an anchor protein (Rpl13a tagged with FKBP12) that shuttles from nucleus to cytoplasm during ribosome biogenesis, (ii) nuclear bait proteins fused to FRB and near-infrared fluorescent protein miRFP670nano3, and (iii) nuclear prey proteins fused to HaloTag. Upon rapamycin addition, the FRB-tagged bait proteins relocate from the nucleus to the cytoplasm through heterodimerization with the FKBP12-tagged anchor. If the bait proteins interact with Halo-tagged prey proteins, both co-translocate to the cytoplasm, providing a spatial readout for protein-protein interactions (fig. S11A).

For PICT assays, yeast grown in SC medium were incubated with JF552-HaloTag ligand for HaloTag labeling and biliverdin to enhance miRFP670nano3 fluorescence for 4 hours. Rapamycin (10 μg/ml; +RAP) or DMSO (−RAP) was added during the final hour before imaging. GFP markers on Elo3 and Gar1 define nuclear boundaries to facilitate identification of translocation (fig. S1A). NLS-HaloTag serves as a negative control. All strains harbor the tor1–1 mutation and FPR1 deletion to eliminate rapamycin sensitivity.

Statistics and reproducibility

Statistical analyses were performed using GraphPad Prism (v9.4.0) or RStudio (v2021.09.1+372; R (v4.1.2)). Values are mean ± s.e.m. unless otherwise stated. Statistical tests and P values are reported in figure legends. For reliable fitting and statistical evaluation, data obtained from at least two independent imaging sessions were resampled by bootstrapping. Representative microscopic images and immunoblots from at least two independent experiments are shown in the figures.

Supplementary Material

Supplementary Materials

Fig. S1. Single-molecule tracking of eukaryotic nuclear RNA polymerases. (A) Strain with ER (Elo3) and nucleolar (Gar1) markers fused with GFP for cell cycle identification (32). Scale bar: 4.0 μm. (B) Halo-H2B spike-in control identified with Pus1-miRFP670nano3 nuclear marker. Cell membranes are indicated by dashed outlines, and nuclear envelopes as solid outlines. Scale bar: 2.0 μm. (C to E) Geometric mean of RNAPI (C), II (D), and III (E) genes. (C) ETS, external transcribed spacer; ITS, internal transcribed spacer. (D) Length of RNAPII gene coding sequence (CDS) with untranslated region (UTR) obtained from Tuller et al., 2009 (120). Top 200 mRNA genes for yeast in SC medium obtained from Miura et al., 2008 (121). (E) 275 tRNA genes, RPR1, SNR52, SCR1, SNR6, RNA170, and 150 copies of 5S rRNA gene (RDN5) with a length of 120 bp were set for calculation (122, 123). (F) Kymograph of slow tracking trajectories for the three nuclear RNAPs.

Fig. S2. Improved residence time correction using stretched exponential decay of H2B. (A) Labeling one copy of H2B (Htb1), both copies of H2B (Htb1 and Htb2), or one copy of H3 (Hht1) resulted in essentially identical survival probabilities (n = number of trajectories; mean ± s.d.). (B) Fitting of Halo-H2B survival probability plot with one-exponential decay (1-exp), two-exponential decay (2-exp), and one exponential plus one stretched exponential decay (1e1s) (n = number of trajectories). (C) BIC difference of fittings in (B) relative to 1-exp. (D) Equations of H2B decay modeling for residence time correction. f, fraction; k, decay constant; β, stretching exponent. (E) Illustration of chromatin movement simulation in slow tracking experiment. Blue region corresponds to the depth of field. Trajectories are detected when they are inside the detection range. (F) Empirical MSD (mean squared displacement) curve derived from chromatin-bound H2B trajectories in slow-tracking experiments (mean ± s.d.), compared with simulated MSD from a fractional Brownian motion model in a harmonic potential parameterized using the experimental data. (G) Fitting of simulated H2B survival probability with and without chromatin movement using exponential decay (1-exp) and stretched exponential decay (Stretched). (H) BIC difference of fittings in (G) relative to 1-exp. (I to K) Survival probability plot of RNAPII (I), I (J), and III (K) fitted with two-exponential decay, and corrected by either a standard exponential decay (2-expExpCORR) or a stretched exponential decay (2-expStrCORR) from stable population of H2B (n = number of trajectories; mean ± s.d.). (L) BIC difference for 2-expExpCORR versus 2-expStrCORR. (M) Mean residence time of nuclear RNAPs calculated using either 2-expExpCORR or 2-expStrCORR (n = 5,000 resamplings; mean ± s.e.m.). (N) Survival probability of H2B-corrected residence times for H2B, RNAPII, I, and III (n = number of trajectories; mean ± s.e.m.). (O) Mean residence time of exponential decay (τ) can be estimated graphically through linear approximation (approx.) by taking the slope of the tangent at the initial point, with the x-intercept representing 1/k. (P) The x-axis (time) of the survival probability plot on a log10 scale enhances visualization of the transient and stable populations. Figure legend as in (O).

Fig. S3. RNAPI-associated dynamics for TBPSpt15, DSIFSpt4, Paf1CCtr9, and Nsi1. (A) Spatial distribution of 35S rRNA, 5S rRNA, tRNA genes, and mRNA genes in nucleoplasm and nucleolus. (B) Bound and free trajectories in the nucleus captured by fast tracking. 100 random trajectories are shown for each protein except Nsi1, where only 48 trajectories are shown due to low copy number. Scale bar: 500 nm. (C to E) Survival probability of H2B-corrected residence times for TBPSpt15, DSIFSpt4, and Paf1CCtr9 in nucleoplasm (NPL) and nucleolus (NOL), and Nsi1 (n = number of trajectories; mean ± s.e.m.). (D) Mean residence time of the stably bound fraction (n = 5,000 resamplings; mean ± s.e.m.). (E) Transiently and stably bound fraction (n = 5,000 resamplings; mean ± s.e.m.).

Fig. S4. Single-molecule dynamics of RNAPII and associated factors in WT and CTD9 mutant. (A) Two-component Gaussian fit of apparent diffusion coefficient (D) histograms (n = number of trajectories). (B) Stably bound fraction of RNAPII and associated factors in WT and CTD9 mutant (n = 5,000 resamplings; mean ± s.e.m.). While CTD truncation impairs PIC formation, as observed in this figure and in our previous study (32), changes in chromatin-binding fraction for some factors downstream of initiation, such as CTDK-1Ctk1 and CF1BHrp1, are uncorrelated with the anticipated reduction in transcription. These varied responses to CTD truncation might reflect underlying cellular adaptations or complex regulatory mechanisms that are not yet fully understood and will require further targeted analysis.

Fig. S5. RNA-seq analysis of intron-containing genes and ribosomal protein expression in WT and CTD9 mutant. (A) Same as Fig. 5H, with RPS16A (small ribosomal subunit) and RPL43B (large ribosomal subunit) highlighted. (B) Fold change in expression level (FPKM, fragments per kilobase million) for ribosomal protein (RP) genes and other intron-containing (Non-RP) genes (n = 4 biological replicates; median values shown). (C) Fold change in intron retention (IR) ratio versus expression level (FPKM) for RP genes between CTD9 and WT. Lines connect the same genes. (D and E) Protein expression of Rps16a and Rpl43b in CTD26 (WT) and CTD9. Representative western blot (D) and quantification normalized to Pgk1 (E). Bars indicate means, dots represent biological replicates (n = 3); Unpaired Student’s t-test, ** P < 0.01, ns = not significant. (F and G) Fold change in IR ratio against expression level (FPKM) (F) or intron length (G) for intron-containing non-RP genes. Linear regression with 95% confidence interval shown. n = 4 biological replicates. (H) MA plot of RNA-seq data for CTD9 versus CTD26. Common housekeeping genes ACT1, PGK1, TDH3 highlighted. (I) Relative abundance of SIRV (spike-in RNA variant) isoforms in RNA-seq experiment for CTD26 and CTD9. Bars indicate means, dots represent biological replicates (n = 4).

Fig. S6. High temporal occupancy by PIC components suggests persistent accessibility of RNAPI and RNAPIII promoters. (A) Simplified model for the diffusive search of a protein factor between two specific chromatin target sites in the nucleus. POI, protein of interest. (B) Temporal occupancy measures the percentage of time a chromatin target is occupied by a factor, while search time calculates the average duration a protein molecule spends between stable binding events (mean ± s.e.m.). The model integrates data on binding dynamics from fast and slow tracking, along with the estimated number of molecules (124) and target sites per cell (122, 123). Detailed calculations are available in prior studies (38, 58). It is important to note several limitations: slow tracking at 250 ms/frame may not capture exceedingly brief binding events reported by fast tracking at 10 ms/frame; molecule counts from Ho et al., 2018 (124), are estimates from multiple datasets and may differ in our conditions; for simplicity and consistency with the SMT data, only the molecule count of one subunit of a complex is used; target sites are based on estimated counts for known RNAPI and RNAPIII genes (122, 123), potentially missing unconventional binding sites; additionally, factors might sample more frequently at sites on gene promoters with higher firing rates. Despite these considerations, this modeling approach provides valuable insights but should be viewed as supplementary and interpreted with caution.

Fig. S7. CTDK-1 and BUR kinase show diffusion confinement. (A) Spatiotemporal mapping of CTDK-1Ctk1 and BUR kinaseSgv1 free trajectories showing high probability density at a subnuclear region previously shown to be enriched with active genes (32). Scale bar: 0.5 μm; Error bar = centroid of nucleolus ± s.d.; n = number of nuclei. (B) Angular movement (f180/0) in WT and CTD9 mutant (n = 100 resamplings; mean ± s.d.). (C) Free diffusion coefficients (Dfree) in nucleoplasm (n = 100 resamplings; mean ± s.d.; ns = not significant). Both factors show reduced diffusion confinement in the CTD9 mutant (A and B), yet Dfree remain unchanged (C), consistent with a model of transient confinement rather than stable sequestration (32).

Fig. S8. Residence time of spliceosomal components in CTD9 mutant. (A) Schematic of mRNA splicing. (B) Mean residence time of the stably bound fraction for spliceosomal components (n = 5,000 resamplings; mean ± s.e.m.; Unpaired Student’s t-test, * P < 0.05, ns = not significant).

Fig. S9. Analysis of published chromatin remodelers data using improved residence time correction. (A) Chromatin remodelers RSC and INO80 survival probability (1-CDF) data adapted from Kim et al., 2021 (39). (B) Fitting of RSC and INO80 survival probability (1-CDF) is improved with 2-expStrCORR. (C) Mean residence time of the stably bound fraction (n = 5,000 resamplings; mean ± s.e.m.).

Fig. S10. Validation of Halo-tagged protein constructs used in this study. (A) In-gel fluorescence for 61 Halo-tagged proteins stained with JFX650 showing full-length expression. Representative gel images demonstrate intact fusion proteins with minimal degradation products. Coomassie staining serves as loading control. Fluorescence band intensities reflect relative protein levels but should not be over-interpreted due to variables affecting signal intensity, including protein turnover rates, labeling efficiency, and variability in extraction efficiency (39). Asterisks indicate bands with unexpected migration based on predicted molecular weight (M.W.), consistent with previous observations for these proteins (118, 125–129). (B) Growth assays of all Halo-tagged strains compared to untagged WT. The bottom rightmost panel shows untagged controls: CTD26 (WT) and CTD9 mutant.

Fig. S11. Validation of protein interactions for Halo-tagged proteins using PICT (Protein Interactions from Imaging of Complexes after Translocation) in living cells. (A) Schematic of PICT for studying protein-protein interactions. (B) Rapamycin-induced co-translocation of thirteen representative Halo-tagged proteins (at least one from each functional category studied) with their interacting partners. Prey protein (Halo-tagged, JF552 labeled) and bait protein (FRB-miRFP670nano3 fusions) pairs are selected based on known direct contacts from structural studies. NLS-Halo serves as negative control. Nucleolus, nuclear envelope, and plasma membrane are outlined in dashed white. Scale bar: 1.0 μm.

Fig. S12. Western blot validation of Halo-tagged CTD- and histone modifier activities. (A) Comparable CTD phosphorylation between tagged and untagged controls. Ser5P for Tfb4-Halo and Kin28-Halo; Ser2P for Ctk1-Halo and Sgv1-Halo. Rpb3 serves as internal control. (B) Comparable histone H3 modifications between tagged and untagged controls. H3K4 methylation for Swd1-Halo; H3K36 methylation for Set2-Halo; H3 acetylation for Set3-Halo and Rco1-Halo. Set1-Halo shows impaired H3K4 methylation activity (130) and was excluded from the analysis. Total H3 serves as internal control.

Fig. S13. Apparent diffusion coefficient histograms and survival probability of factors examined. Superscripts indicate the tagged representative subunits of complexes or yeast-specific nomenclature. For survival probability (1-CDF) plots, lines indicate 2-expStrCORR fit and error bars indicate s.d. of raw values. Fast tracking data (apparent diffusion coefficient histograms; D) of RNAPII PIC components, SAGASpt7, Spt8, RNAPIRpa190, RNAPIIIRet1, and H2BHtb1 were adapted from Ling et al., 2024 (32) (n = number of trajectories).

Fig. S14. Survival probability of factors examined. (A to J) Corrected residence times for RNAPI PIC (A), RNAPIII PIC (B), RNAPII PIC (C), RNAPII co-factors (D), transient and stable elongation factors (E), capping and RNA-binding factors (F), splicing factors (G), CTD modifiers (H), histone modifiers (I), and termination factors (J). As in Fig. 1, E and I; Fig. 2, B and E; Fig. 3, B to F; and Fig. 4C, but with logarithmic y-axis and linear x-axis.

Table S1. Yeast strains for this study.

Movie S1

Example video of fast tracking. Fast tracking of RNAPIIRpo21, RNAPIRpa190, and RNAPIIIRet1. Nucleolus and nuclear envelope are outlined in solid white, plasma membrane in dashed white. Asterisks indicate nucleoli. Scale bar: 1.0 μm. Recorded frame rate: 10 ms per frame. Playback speed: 20 ms per frame.

Movie S2

Example video of slow tracking. Slow tracking of RNAPIIRpo21 with H2BHtb1 spike-in control. Nucleolus and nuclear envelope are outlined in solid white, plasma membrane in dashed white. Asterisks indicate nucleoli. Scale bar: 1.0 μm. Recorded frame rate: 250 ms per frame. Playback speed: 100 ms per frame.

Supplementary Materials

The PDF file includes:

Figs. S1 to S14

Table S1

Captions for Movies S1 to S2

References (120–130)

Other Supplementary Materials for this manuscript include the following:

Movies S1 to S2

Data and materials availability:

SMT data are available at Mendeley (113). Custom code for SMT analysis is available at Zenodo (114). RNA-sequencing data were deposited in the NCBI SRA database (BioProject PRJNA1124391). Materials generated in this study are available from the corresponding authors on request.