Human RNA polymerase III termination favors decomposition over facilitated recycling Authors: Sun Han, Qianmin Wang, Eunho Song, Changwon Kang, Yanhui Xu, Sungchul Hohng Published: 2026 ABSTRACT AbstractEukaryotic transcription termination remains incompletely understood. Using single-molecule fluorescence assays, we examined intrinsic termination by human RNA polymerase III (RNAP3) and identified mechanistic trifurcation leading to readthrough and two termination pathways with distinctive dissociation kinetics. The dominant pathway, decomposing termination, involves (near-)simultaneous dissociation of RNA and DNA from RNAP3, resulting in complex decomposition. In contrast, recycling termination, where RNA is released while RNAP3 remains DNA-bound and one-dimensionally diffuses to enable facilitated recycling, accounts for 2–14% of transcription, diverging from its predominance in bacterial intrinsic termination. Confirming that robust termination requires a nontemplate-strand T-tract of four or more consecutive thymidines, we uncover that longer T-tracts favor recycling over decomposing termination, reflecting mechanistic distinction between the two pathways. While upstream RNA hairpins enhance termination efficiency, their impact decreases with distance from T-tracts, affecting both pathways similarly. Kinetically, transcript release takes consistently longer in decomposing than recycling termination, aligning with polymerase displacement versus RNA shearing. Furthermore, an RPC10 subunit mutation causes recurrent elongation arrest with differential effects on termination pathways, revealing its selective role in governing this branching process. These findings reinforce evolutionary conservation of termination-pathway duality and highlight diversion of pathway preference in human RNAP3 termination, establishing its fundamental mechanistic principles.Subject terms: Single-molecule biophysics, Transcription Introduction Transcription termination is a tightly regulated essential step in gene expression, ensuring release of correct RNA products. Termination can be modulated via antitermination or attenuation to dynamically regulate downstream gene expression; conversely, its dysregulation can compromise transcript fidelity, disrupt genome organization, and impair transcriptional homeostasis. However, the mechanisms governing transcription termination have been underexplored relative to initiation, particularly in eukaryotic systems. Human RNA polymerase III (RNAP3) transcribes genes for small noncoding RNAs including 5S rRNA, tRNAs, U6 snRNA1, and operates with high efficiency and remarkable autonomy2. Its capacity for intrinsic termination, independent of extrinsic factors, makes it a robust model for dissecting fundamental termination mechanisms. However, the precise biophysical steps of RNAP3 termination remain less defined than those of its bacterial counterparts. Despite mechanistic differences, termination universally involves an elongation pause followed by conformational rearrangements triggering transcript release. In bacteria and bacteriophages, class I intrinsic termination requires two cis-acting elements: a pause-inducing RNA oligo(U)3 and a release-facilitating upstream hairpin4. Alternatively, phage class II termination relies on a conserved DNA sequence followed by a T-rich tract5–7. In contrast, eukaryotic RNAP3 typically utilizes a single cis-element, a nontemplate-strand oligo(T), sufficient to cause both pause and release8–10. Whether upstream RNA hairpins contribute to RNAP3 termination remains debated11–13. This controversy may stem from the weak interactions observed between RNAP3 and the RNA·DNA hybrid14–16. Cryogenic electron microscopy (cryo-EM) studies reveal that the RPC2 subunit interacts with the oligo(T) tract17 and that the hybrid is short, spanning only five to six base pairs16,18,19. These features suggest that a weak oligo(U)·oligo(dA) hybrid alone can destabilize elongation complex enough to trigger transcript release20,21, potentially obviating the need for secondary structures like hairpins. Single-molecule studies have identified two termination pathways with distinct mechanisms and fates22,23. In decomposing termination, RNA and DNA dissociate (near-)simultaneously through polymerase displacement, a process reminiscent of the hypertranslocation model24,25. Predominant in phage and Rho-dependent bacterial termination24,25, this pathway requires 3D search for reinitiation and is considered unfacilitated compared with facilitated 1D diffusion. In contrast, recycling termination involves RNA-only release while the polymerase remains DNA-bound and diffuses bidirectionally along the template to facilitate recycling for nearby reinitiation. This pathway dominates bacterial intrinsic termination22,23,26. Since post-termination DNA retention was recently confirmed for Escherichia coli and Saccharomyces cerevisiae polymerases27–31, human RNAP3 provides an ideal model to test whether this dual-pathway paradigm extends to higher eukaryotes. In this study, we employed single-molecule fluorescence assays to investigate human RNAP3 transcription across five representative genes. By omitting auxiliary factors, we defined core, intrinsic features of RNAP3 termination. We found that RNAP3 utilizes both decomposing and recycling pathways, demonstrating that dual-pathway termination is conserved across phages25, bacteria22–24, yeast31, and humans, though the recycling pathway is vanishingly rare in phages25. Notably, despite its dominance in bacterial intrinsic termination22, the recycling pathway is infrequent for human RNAP3 in the absence of additional factors. T-tract length and RNA hairpins enhance termination. Longer T-tracts favor the minor recycling pathway over the major decomposing pathway, while hairpins influence both pathways similarly. Kinetic analysis revealed that transcript release proceeds more slowly in decomposing termination, likely via more extensive conformational changes, than in recycling termination. Finally, an RPC10 subunit mutation causes elongation arrest and impairs termination. Collectively, these findings illuminate how eukaryotic termination signals are encoded and processed. Results Single-molecule fluorescence of transcription We developed single-molecule assays for human RNAP3 termination using a 5′-Cy3-labeled ApU primer and 23 DNA templates sharing a 17-nt upstream overhang but varying in their downstream segments (Fig. 1). The upstream ends were biotinylated for surface immobilization, and the free downstream ends were Cy5-labeled. Human RNAP3 was labeled with Alexa Fluor 488 (A488) via an N-terminal SpyTag-SpyCatcher fusion on the RPC1 subunit. DNA templates for RNAP3 termination. a T4/A4 template. This oligo(T) variant, also shown in panel c, contains four consecutive nontemplate-strand Ts and complementary template-strand As. The upstream end was biotinylated for surface immobilization, and the free downstream end was labeled with Cy5. Transcription was primed using Cy3-ApU, generating a stalled 23-nt RNA in the absence of UTP and labeling its 5′ end with Cy3. This priming strategy is more efficient in vitro than promoter-dependent initiation and does not require initiation factors. b Native gene terminators. These templates were derived from five human RNAP3-transcribed genes, RNA5S1 (5S rRNA), TRA-AGC1-1 (tRNA-Ala), RNU6-1 (U6 snRNA), RPPH1 (RNase P RNA), and RN7SL1 (7SL RNA). Each contains an oligo(T) tract of 4–8 Ts on the nontemplate strand. c Oligo(T) variants. These templates have defined T-tract lengths (T0/A0 to T9/A9) and strand-swapped controls (T9/T9, A9/A9, and A9/T9). All share a common sequence background with matched oligo(T) positioning. d Hairpin-spacer variants. Templates hp+0 to hp+20 encode an upstream RNA hairpin positioned 0–20 bp upstream of a T4 terminator and were used to assess distance-dependent effects on termination. Stalled transcription complexes were formed by incubating A488-RNAP3, Cy5-DNA, and Cy3-ApU at 37 °C for 20 min with ATP, CTP, and GTP. UTP omission stalled RNAP3 after synthesis of a 23-nt RNA labeled with Cy3 at the 5′ end (Fig. 2a). At this position, the polymerase was spatially proximal to Cy3, provoking protein-induced fluorescence enhancement (PIFE) by hindering its photoisomerization32–36. These three-color stalled complexes, containing Cy3-RNA, Cy5-DNA, and A488-RNAP3, were immobilized. Three-color assay schematic and traces. a Single-molecule assay with three fluorophores. RNA was labeled with Cy3 (green), DNA with Cy5 (red), and human RNAP3 with Alexa Fluor 488 (A488, blue). Stalled RNAP3 complexes, immobilized via biotinylated DNA, resumed elongation upon NTP injection, exhibiting a single-step decline in Cy3 PIFE, which was defined as time zero. The complexes then followed one of three pathways: readthrough-runoff, recycling termination, or decomposing termination. Human RNAP3 is illustrated based on a 17-subunit cryo-EM model19. b Representative fluorescence traces for each pathway. In readthrough-runoff (left), Cy5 PIFE onset precedes Cy3 and A488 loss, indicating that RNAP3 reaches the DNA downstream end and subsequently dissociates together with the runoff transcript. In recycling termination (center), Cy3 loss (RNA release) occurs first, followed by delayed Cy5 PIFE as RNAP3 diffuses to the DNA downstream end. In decomposing termination (right), Cy3 and A488 are lost nearly simultaneously in the absence of Cy5 PIFE, indicating concomitant dissociation of RNA and RNAP3. Traces are truncated at 250 s, as no additional events occurred beyond this window. Control experiments confirmed that Cy5 PIFE was elongation-dependent. Mean photobleaching lifetimes were 124 min for Cy3, 59 min for Cy5, and 21 min for A488 (Supplementary Fig. 1), confirming fluorophore stability over the observation window. Molecule counts per condition are provided in Supplementary Table 2. Elongation was resumed by injecting all four NTPs, after unbound species were washed away. Individual complexes were monitored for 1000 s using alternating 5-s excitation of the three fluorophores; photobleaching was negligible within this 17-min window (Supplementary Fig. 1). As RNAP3 moved away from the 5′ end, Cy3 PIFE declined, marking successful resumption (Fig. 2a). Only three-color complexes exhibiting this single-step decline (time zero) were analyzed. Using the T4/A4 template (Fig. 1a) containing the minimal genomic (T)4 terminator37, we identified termination, defined as transcript release, by Cy3-RNA loss, and estimated efficiency by counting Cy3-losing complexes. A488-RNAP3 was subsequently lost within 4.0 min (Fig. 3a), indicating no polymerase arrest or trapping. Critically, A488 loss never preceded Cy3 loss, establishing the sequence of dissociation. As the polymerase reached the downstream end, Cy5 exhibited PIFE (Fig. 2a). Later, when the polymerase dissociated from the end, Cy5 PIFE declined simultaneously with A488 loss (15-s resolution). Milestones and RNAP3 dissociation kinetics. a Milestones of the three pathways. Relative timing of key fluorescence-defined events was measured from Cy3 PIFE decline (time zero). Cy3 loss (green) indicates RNA release at the termination site or runoff from the free downstream end. A488 loss (black) indicates polymerase dissociation at or near the termination site. Cy5 PIFE onset (red) and decline (blue) indicate polymerase arrival at the downstream end and dissociation from the end, respectively. Time points were determined using the three-color assay, with two-color assay values shown in parentheses. b Post-termination polymerase retention times. The histogram shows delays between RNA release (Cy3 loss) and polymerase dissociation (A488 loss). Decomposing termination (white bars) exhibited rapid dissociation kinetics, well fit by a single-exponential decay (τ = 27 s, R² = 1.0), consistent with single-step dissociation at or near the termination site and negligible diffusion before release. Recycling termination (gray bars) showed prolonged retention at the free downstream end, well fit by a Gaussian function (μ = 155 s, σ = 18 s, R² = 0.91), consistent with multi-step end-dissociation. Readthrough and two termination pathways We observed three fluorescence patterns using T4/A4 (Fig. 2b). The first pattern represented readthrough-runoff (left traces), in which the polymerase bypassed the terminator and rapidly triggered Cy5 PIFE onset upon reaching the downstream end (Fig. 3a). After lingering (mean 1.3 min), Cy5 PIFE declined simultaneously with A488 and Cy3 loss, indicating transcript and polymerase dissociated together upon runoff. Every readthrough culminated in runoff. Thus, readthrough was defined by Cy5 PIFE onset preceding Cy3 loss. The second pattern identified recycling termination (center traces). Here, the polymerase recognized the terminator, causing Cy3 loss. A mean of 1.8 s later, Cy5 PIFE onset occurred, indicating the RNA-free polymerase remained DNA-bound and diffused to the downstream end. This 1D diffusion was rapid on the 99-bp template. After lingering for 2.6 min, the polymerase dissociated, marked by the simultaneous Cy5 PIFE decline and A488 loss. Recycling termination was thus distinguished by Cy5 PIFE onset following Cy3 loss. The third pattern corresponded to decomposing termination (right traces). Transcription terminated with Cy3 loss, but without subsequent Cy5 PIFE. A488 was typically lost concurrently with Cy3, and always within 1.5 min of Cy3 loss. The absence of Cy5 PIFE indicated the polymerase did not diffuse to the end but underwent decomposition with or shortly after transcript release, rather than arresting or trapping. Decomposing termination was thus characterized by the absence of Cy5 PIFE following Cy3 loss. Every complex exhibited one of these three patterns; no other patterns occurred, suggesting the polymerase never dissociated alone. This was consistent with our observation that Cy3-RNA release concurred with or preceded, but never followed, A488-RNAP3 dissociation. The recycling and decomposing termination pathways were differentiated by the presence or absence of Cy5 PIFE following Cy3 loss, reflecting the occurrence or nonoccurrence of 1D diffusion to the end. While the absence of PIFE could theoretically reflect extreme delay, the distinction is more fundamental, as shown below. Distinctive kinetics of polymerase dissociation Post-termination retention of A488-RNAP3 differed between the two pathways, reflecting distinctive dissociation kinetics (Fig. 3b). Recycling termination was characterized by a Gaussian distribution (R² = 0.91), with RNAP3 remaining DNA-bound for a mean of 155 s (σ = 18 s; range 120–240 s) after RNA release. This duration was largely spent at the Cy5-end (mean PIFE: 153 s). Given that 1D diffusion took only 1.8 s, intervening pauses were likely rare. Importantly, each complex that retained A488 following Cy3 loss exhibited Cy5 PIFE, indicating that all polymerases reached the downstream end, as every upward diffusion was reversed at the upstream-end tethering. Decomposing termination, by contrast, followed a single-exponential decay (R² = 1.0). Most A488 signal (71%) was lost within the 15-s resolution of Cy3 loss, and 92% within 30 s, with a mean loss time of 27 s, much earlier than the 155-s end-dissociation in the recycling pathway. The observed range (0–90 s) did not overlap with the recycling window (120–240 s). Given the rapid 1D diffusion rate, these data confirm that decomposing polymerases dissociate (near-)simultaneously with RNA release, presumably at or near the termination site, performing negligible diffusion on DNA. High-resolution assays with two fluorophores In three-color assays, A488 photobleaching was delayed by reducing laser intensity, a compromise requiring 5-s excitations. As every recycling polymerase diffused to the downstream end, its dissociation was directly marked by Cy5 PIFE decline alone without the need for A488 loss. Moreover, because all decomposing polymerases dissociated at or near the termination site without diffusing to the downstream end, their dissociation markable with A488 loss was faithfully represented by Cy5 PIFE absence, rendering the A488 signal redundant. Thus, two-color assays using only Cy3-RNA and Cy5-DNA were sufficient for outcome classification (Fig. 4). Two-color assay schematic and traces. a Single-molecule assay with two fluorophores. The setup and protocol were identical to those used in the three-color assay (Fig. 2), except that A488-labeled RNAP3 was replaced with unlabeled RNAP3 and the excitation cycle was shortened to 1 s, improving temporal resolution 7.5-fold, from 15 s to 2 s. b Representative fluorescence traces for each pathway. Outcome classification followed the same criteria as in the three-color assay. In readthrough-runoff (left), Cy3 loss follows Cy5 PIFE onset, indicating that RNAP3 diffuses to the downstream end before dissociating from it. In recycling termination (center), Cy3 loss precedes Cy5 PIFE onset, indicating that RNAP3 releases RNA before diffusing to the downstream end. In decomposing termination (right), only Cy3 loss is observed in the absence of Cy5 PIFE, indicating that RNAP3 and RNA dissociate nearly simultaneously without diffusion to the downstream end. Molecule counts per condition are provided in Supplementary Table 2. Source data are provided as a Source Data file. Shortening each excitation to 1 s in two-color assays improved temporal resolution 7.5-fold, from 15 to 2 s (Supplementary Fig. 2). Unlabeled RNAP3 exhibited 1.2-fold higher activity than A488-RNAP3, based on Cy3 PIFE decline frequency (Supplementary Table 1). Importantly, termination efficiencies were consistent between methods: 45% versus 45% for decomposing (P = 0.90) and 7.3% versus 6.9% for recycling termination (P = 0.80) on T4/A4. Furthermore, two-color assays reliably detect elongation arrest, as arrested complexes tenaciously retain Cy3-RNA without triggering Cy5 PIFE. While arrest was never observed with wildtype RNAP3, it was identifiable for the RPC10 mutant described later. Five native terminators for RNAP3 Every complex exhibited one of the three fluorescence patterns in the high-resolution two-color assays. No additional patterns were observed across all 23 templates, grouped in three sets. The native terminator set (Fig. 1b) comprised five human gene terminators: RNA5S1 (5S rRNA), TRA-AGC1-1 (tRNA-Ala), RNU6-1 (U6 snRNA), RPPH1 (RNase P RNA), and RN7SL1 (7SL RNA). These templates containing (T)5C(T)3, (T)4C(T)2, (T)5, (T)5, and (T)4 exhibited termination efficiencies of 73%, 74%, 73%, 50%, and 66%, respectively (Fig. 5a; Supplementary Table 1). Termination efficiencies across templates. a Human RNAP3 termination efficiencies measured by high-resolution two-color assays. The 23 templates are grouped into native terminators, oligo(T) variants, and hairpin-spacer variants. Bars represent decomposing termination (open) and recycling termination (filled). Total termination equals the sum of both pathways. Data are presented as means ± standard deviations from independent experiments (n = 5 for RPPH1, n = 4 for RN7SL1, and n = 3 for the remaining 21 templates). Each experiment included at least 30 single molecules. The total number of molecules analyzed per template is provided in Supplementary Table 2. b Recycling fraction for each template. The recycling fraction, defined as the fraction of recycling termination among all termination events, ranged from 0.023 to 0.17. Efficiency values are listed in Supplementary Table 1, and molecule counts per condition are provided in Supplementary Table 2. Source data are provided as a Source Data file. While these efficiencies in the absence of cellular factors slightly fell below in vivo estimates (75–95%)38, decomposing termination occurred at 44–66%, and recycling termination at 1.5–11%. Thus, the decomposing pathway was predominant in human RNAP3 intrinsic termination. The recycling fraction (recycling termination among total termination) ranged from 0.023 to 0.14 (Fig. 5b). This contrasts sharply with the 0.70–0.91 range reported for E. coli intrinsic termination22, reflecting a reversal in pathway preference between human and bacterial systems. The first three templates (5S rRNA, tRNA-Ala, and U6 snRNA) displayed similar profiles: 73–74% overall termination (62–66% decomposing, 7.5–11% recycling) and recycling fractions of 0.10–0.14. The fourth (RNase P RNA) exhibited the lowest overall efficiency (50%), primarily due to a 20% reduction in decomposing termination. By contrast, the fifth template (7SL RNA) showed a decomposing termination (65%) comparable to the first three but the lowest recycling termination efficiency (1.5%), leading to an 8% decrease in overall efficiency. Oligo(T) variants revealing a length threshold A T-rich stretch in the nontemplate strand is essential for RNAP3 termination9,17,37,39,40. To assess how length affects termination, we designed ten templates, designated T0/A0 through T9/A9, containing zero to nine consecutive (T/A)s in the nontemplate/template strands (Fig. 1c). (T4/A4 was used in the three-color assay.) All templates were otherwise identical, with the T-tract positioned to ensure a consistent transcriptional context. Termination efficiencies increased with oligo(T) length, showing a sharp threshold at four Ts. Three or fewer Ts (T0/A0–T3/A3) had low efficiencies (20–24%), whereas four or more Ts (T4/A4–T9/A9) had high 52–65% (Fig. 5a; Supplementary Table 1). This group difference was significant for both decomposing (P = 4.1×10⁻19) and recycling pathways (P = 3.7×10⁻10). Notably, total efficiency jumped 2.5-fold at the threshold, from 21% (T3/A3) to 52% (T4/A4). Thus, robust termination requires at least four consecutive (T/A)s, consistent with human genomic distribution and Xenopus mutation studies37,41. Longer T-tracts preferentially promoted recycling over decomposing termination. Although both pathways increased with oligo(T) length, the relative and absolute gains in the recycling pathway were greater. Recycling termination efficiency rose 6.6-fold from T0/A0 to T9/A9, while decomposing termination increased 2.8-fold (Fig. 6a). Net increases in the recycling pathway consistently exceeded those of the decomposing pathway (Fig. 6b), and the recycling fraction rose with length, reaching 0.13–0.17 for T4/A4 and longer versus 0.057–0.093 for T3/A3 and shorter (Supplementary Table 1). Oligo(T) and hairpin effects on termination. a Fold changes in termination efficiencies relative to T0/A0. Recycling termination (filled) increases more prominently with oligo(T) length than decomposing termination (open), whereas hairpin proximity affects both pathways comparably. b Absolute increases in termination efficiencies relative to T0/A0. Oligo(T) length contributes more strongly to recycling termination than to decomposing termination, whereas hairpins enhance both pathways to a similar extent. Data are presented as means ± standard deviations from independent experiments (n = 5 for RPPH1, n = 4 for RN7SL1, and n = 3 for the remaining 21 templates). Each experiment included at least 30 single molecules. The total number of molecules analyzed per template is provided in Supplementary Table 2. Source data are provided as a Source Data file. To test strand specificity and pairing, we examined three constructs with altered (T)9 and (A)9 placement: A9/T9, T9/T9, and A9/A9 (Fig. 1c). Compared with T9/A9 (64%), mismatched T9/T9 showed a modestly reduced level (56%), while strand-swapped A9/T9 (34%) and T-lacking A9/A9 (32%) exhibited markedly lower efficiencies (Fig. 5a; Supplementary Table 1). These results indicate that a nontemplate T-tract is essential and partially effective even when unpaired, suggesting that RNAP3 primarily recognizes the nontemplate rather than the template strand. Together, these results reinforce that four or more consecutive nontemplate Ts are required for efficient human RNAP3 termination38,41. Longer T-tracts preferentially enhance recycling over decomposing termination. This differential sensitivity supports the pathway distinction. The bias toward recycling termination may arise from an increased RNA release propensity resulting from the higher U content within the RNA·DNA hybrid. Thus, oligo(T) length influences both termination efficiency and pathway choice. Hairpin-spacer variants showing distance dependence To assess the contribution of RNA hairpins to human RNAP3 termination, we designed five templates containing a fixed (T/A)4 terminator and an upstream hairpin at varying distances to systematically evaluate distance-dependent effects on termination profiles: hp+0, hp+5, hp+10, hp+15, and hp+20 had a stable stem-loop42 at 0, 5, 10, 15, or 20 nt upstream of the oligo(T)4, respectively (Fig. 1d). Upstream RNA hairpins generally enhanced termination compared with the no-hairpin control (T4/A4; 52% total efficiency), with similar effects on both termination pathways (Fig. 5; Supplementary Fig. 3; Supplementary Table 1). The strongest enhancement occurred when the hairpin was immediately adjacent to the T-tract: hp+0 yielded the highest efficiency at 73%. Total efficiency declined progressively with distance, dropping to 56% for hp+15 and 46% for hp+20, with proportional decreases in both pathways. Gel validation of single-molecule measurements To validate the transcriptional competence of our templates, we performed denaturing gel electrophoresis of transcripts synthesized in vitro using the same priming strategy across 14 representative templates (Fig. 7). Transcription from eight templates (T4/A4–T9/A9, T9/T9, and hp+0), which exhibited single-molecule termination efficiencies of 52–73%, produced distinct bands at the expected size for terminated RNA. Bulk transcription validation. Denaturing urea-PAGE analysis of transcription products from 14 representative templates, hp+0, T0/A0 through T9/A9, T9/T9, A9/T9, and A9/A9, spanning termination efficiencies from 20% (T0/A0) to 73% (hp+0), closely matching the range of the full 23-template set (20–74%). Eight high-efficiency templates, T4/A4–T9/A9, T9/T9, and hp+0 (52–73%), produced detectable bands near the expected terminated RNA size (T). In contrast, six low-efficiency templates, T0/A0–T3/A3, A9/T9, and A9/A9 (20–34%), produced detectable runoff-sized RNAs (R) but no visible band at the expected terminated size. Cy3-ApU is included as an intensity reference across gels but shows anomalously slow electrophoretic migration. This shift is attributed to the substantial hydrophobicity and relatively large molecular weight of Cy3, which dominate the hydrophobic volume and mass-to-charge ratio of such short 2-nt RNA fragments67,68. RNA size markers are 15 and 30 nt. The experiment was repeated three times independently with similar results. Source data are provided as a Source Data file. By contrast, six templates (T0/A0–T3/A3, A9/T9, and A9/A9) characterized by lower termination efficiencies (20–34%), produced detectable runoff-sized RNAs but no visible bands at the predicted termination site. This absence likely reflects termination levels below the bulk detection limit or suggests that baseline termination is premature or nonspecific rather than T-tract-dependent. In the latter case, the values reported in Fig. 6b would represent the specific oligo(T)-dependent component of termination. Separately, we attempted to enhance RNA yield by using a 15-nucleotide RNA primer (Cy3-AUCCAACCCAGACCG) in place of the Cy3-ApU dinucleotide. However, the longer primer substantially reduced transcription efficiency (Supplementary Fig. 4), possibly due to decreased elongation complex stability. Therefore, while our single-molecule assays provide high-resolution kinetic data for all templates, further optimization of primer design will be required to maximize RNA production for biochemical assays. Kinetic differentiation of termination pathways To investigate termination kinetics, we analyzed individual complex RNA-dwell times at the terminator (Supplementary Table 2). The termination time, the interval between Cy3 PIFE decline and Cy3 loss, was measured for both pathways. For comparison, readthrough timing was quantified by the immediately subsequent end-reaching time, the interval between Cy3 PIFE decline and Cy5 PIFE onset, which marks an almost instantaneous arrival of RNAP3 at the Cy5-end (Fig. 8a). Kinetic profiles of termination pathways. a Readthrough times across all 23 templates. Readthrough timing was approximated by the immediately subsequent Cy5 PIFE onset (downstream end-reaching), measured from Cy3 PIFE decline (elongation resumption, time zero). Data are shown as means ± standard deviation. Data are presented as means ± standard deviations from independent experiments (n = 5 for RPPH1, n = 4 for RN7SL1, and n = 3 for the remaining 21 templates). Each experiment included at least 30 single molecules. The total number of molecules analyzed per template is provided in Supplementary Table 2. b Milestone time distributions for distinct outcomes. Readthrough was timed by Cy5 PIFE onset (proxy), and termination by Cy3 loss (RNA release). Box plots show the median (center line), interquartile range (box limits), 1.5× IQR (whiskers), and minima/maxima (points). Each data point represents the mean value (n = 5 for RPPH1, n = 4 for RN7SL1, and n = 3 for the remaining 21 templates). RNA release in decomposing termination occurs later than RNA release in recycling termination and than downstream end-reaching after readthrough. c Pairwise comparisons of milestone times. Each panel plots one event time against another for the same template. With one exception, most data points fall above the diagonal, indicating that RNA release in both termination pathways generally occurs later than readthrough end-reaching, and that RNA release occurs later in decomposing than in recycling termination. All kinetic data are listed in Supplementary Table 2. d Downstream end-dissociation of RNAP3. Mean end-dwell durations varied widely for RNA-free RNAP3 in recycling termination and for RNA-bound RNAP3 in readthrough-runoff, while individual end-dwell times spanned nearly three orders of magnitude in both (3–830 s and 3–860 s), as measured by Cy5 PIFE durations. Ratios of end-dwell times (recycling termination over readthrough-runoff) are plotted across all templates. Among them, 18 variants (T0/A0 through hp+20) share an identical 40-bp downstream-end sequence. Template-to-template fluctuations in the ratios (0.14–1.9) support a stochastic component of RNAP3 end-dissociation that is not strictly determined by DNA sequence or RNA retention. Source data are provided as a Source Data file. Across all five native terminators, RNA release took consistently longer in decomposing termination than recycling termination. Mean RNA dwell times were 90–120 s for decomposition versus 49–62 s or 14 s for the recycling pathway (Supplementary Table 2). Readthrough was faster still, with proxy end-reaching times of 21–29 s. Thus, RNA release in both termination pathways generally lagged behind readthrough, with the exception that recycling termination (14 s) preceded readthrough (28 s) on the 7SL RNA template. Box plots of dwell-time values further highlighted these kinetic differences (Fig. 8b). Readthrough events clustered at shorter times, consistent with rapid elongation. RNA release occurred at intermediate times for recycling termination, while it was significantly delayed in decomposing termination. Plotting these values against each other for template-matched datasets confirmed this temporal order, as RNA release occurred later in decomposing than recycling termination, and both generally followed readthrough (Fig. 8c). These findings define a temporal hierarchy: readthrough typically occurs first, followed by recycling termination, and finally decomposing termination. The prolonged RNA dwell times suggest RNA release for decomposition involves extensive or slow conformational rearrangements. Conversely, the faster kinetics of recycling termination are consistent with a simpler, more rapid RNA-only release mechanism. Incidentally, the T4/A4 three-color assay showed polymerase end-dwell times were 2.0-fold longer in recycling termination (153 s) than readthrough-runoff (78 s). In higher-resolution two-color assays across 23 templates, this ratio ranged widely from 0.14 (7SL RNA) to 1.9 (T4/A4), with mean dwell times fluctuating both without RNA (21–167 s) and with RNA (57–148 s). This lack of correlation (Fig. 8d) suggests RNAP3 end-dissociation is stochastic and mechanistically independent of transcript carriage. Salt effects on termination We examined how ionic strength affects termination, as noncovalent molecular interactions are salt-sensitive. Using T4/A4, we varied (NH4)2SO4 concentrations from the standard 60 mM to 30, 75, and 165 mM (Fig. 9). Across this range, both decomposing termination (43% to 59%) and recycling termination (6.7% to 9.3%) increased steadily (Fig. 9a) by similar proportions (1.4-fold), while the recycling fraction remained constant at 0.13–0.14 (Fig. 9b). Although each process was slightly prolonged at higher salt levels (Fig. 9c), the temporal order of three outcomes was preserved. Salt effects on RNAP3 termination. a Termination efficiencies at increasing ammonium sulfate concentrations. Total termination increases with ionic strength, with a stronger effect on decomposing termination (open squares) than on recycling termination (filled squares). Data are presented as means ± standard deviations from n = 3 independent biological experiments. Each experiment included at least 30 single molecules. b Recycling fraction across salt conditions. The recycling fraction shows minimal variation, suggesting proportional enhancement of both termination pathways at higher ionic strength. c Dwell times for the three outcomes. Although absolute times increase with ionic strength, the temporal ordering of readthrough (open circles), recycling termination (filled squares), and decomposing termination (open squares) is maintained across all conditions. Data are presented as means ± standard deviations from n = 3 independent biological experiments. Each experiment included at least 30 single molecules. Source data are provided as a Source Data file. RPC10 mutation affecting termination A mutation in the RPC10 C-terminal domain is known to impair termination and RNA cleavage in yeast8,43. To test this in humans, we introduced the equivalent double mutation (p.Asp88Ala;Glu89Ala) and analyzed three templates (Fig. 10). On T0/A0, wildtype and mutant RNAP3 exhibited similarly low efficiencies: 19% versus 19% decomposing and 1.5% versus 1.0% recycling (Fig. 10a). On T4/A4, however, the mutation markedly impaired both decomposing (45% to 19%) and recycling termination (6.9% to 1.7%). RPC10 D88A/E89A mutation. a Termination efficiencies of wildtype (dashed bars) and mutant (solid bars) RNAP3 on T0/A0, T4/A4, and T8/A8 templates. Decomposing (open) and recycling (filled) termination pathways are shown for each condition. The mutant exhibits a substantial reduction in both pathways. Data are presented as means ± standard deviations from n = 3 independent biological experiments, with each data point representing one independent experiment. Each experiment included at least 30 single molecules. The total number of molecules analyzed per condition is provided in Supplementary Table 2. b Representative fluorescence trace of an arrest event observed only with mutant RNAP3. Following Cy3 PIFE decline (elongation resumption), the Cy3 signal persists in the absence of Cy5 PIFE, indicating prolonged stalling without transcript release or polymerase dissociation over the 1000-s (17-min) observation window. c Transcription outcome distributions for wildtype and mutant RNAP3. Stacked bar plots show the proportions of readthrough (hatched), recycling termination (gray), decomposing termination (white), and arrest (black). For each template, wildtype and mutant data are displayed above and below each other. Arrest events were exclusive to the mutant and accounted for 35–39% of active complexes. Source data are provided as a Source Data file. In contrast, on T8/A8, the mutant’s decomposing capacity was largely restored (41%) toward the wildtype level (55%), whereas the recycling pathway remained severely reduced (2.7% vs 9.9%). These differential effects support the distinction between the two pathways, especially because the mutant maintained activity comparable to wildtype (P = 0.58) and showed similar kinetics (Supplementary Table 2). The extended oligo(T) in T8/A8 thus partially compensates for the mutation by specifically promoting decomposition, suggesting the RPC10 C-terminal domain is more critical for recycling termination. Strikingly, a fourth fluorescence pattern was observed exclusively with the mutant (Fig. 10b). This pattern, never detected with wildtype RNAP3 or in our previous studies of bacterial and phage RNA polymerases22,24–26,44, showed a Cy3 PIFE decline without subsequent Cy3 loss throughout the 17-min window. This likely represents elongation arrest8 without termination, occurring in 35–39% of complexes across all three templates (Fig. 10c). Its high prevalence suggests that the RPC10 D88A/E89A mutation compromises elongation processivity, leading to recurrent transcription arrest. Discussion This study demonstrates that intrinsic termination of human RNAP3 transcription proceeds via two pathways, decomposing and recycling termination. The coexistence of dual pathways at a single terminator extends from bacteria to humans, supporting a model in which multiple pathways operate in parallel to maximize efficiency under distinct regulatory constraints and timescales45. Such redundancy likely provides an intrinsic fail-safe mechanism within individual complexes rather than relying on secondary sites or extrinsic factors. Multiple lines of evidence indicate that these pathways are mechanistically distinct rather than alternative manifestations of a single process. First, post-termination polymerases dissociate with different kinetics (Fig. 3). A single-exponential decay (0–90 s) during decomposing termination is consistent with a one-step stochastic process. In contrast, a Gaussian fit (120–240 s) following recycling termination reflects a multi-step kinetic profile of end-dissociation. The nonoverlapping temporal ranges suggest kinetically discrete populations rather than fast and slow variants of a single event. Second, the two pathways proceed on independent timescales across templates (Fig. 8). RNA release in decomposing termination is consistently the slowest compared with recycling termination RNA release and readthrough end-reaching, whereas recycling termination shows template-dependent variability relative to readthrough. The absence of correlated timing shifts argues against a shared rate-limiting intermediate and instead supports distinct pathway-specific determinants. Third, native terminators and T-tract variants display diverse recycling fractions (Fig. 5). Longer T-tracts preferentially enhance recycling over decomposing termination, and native genes exhibit similar biases, such as low fraction in the 7SL RNA gene compared with higher fractions in the other four. If recycling were merely a subset of decomposition, both pathways would be expected to scale proportionally with terminator strength. Instead, the observed sequence-specific modulation supports a distinct and tunable branch. Fourth, the RPC10 mutation differentially affects the two pathways (Fig. 10). The D88A/E89A mutation severely impairs recycling termination, whereas decomposing termination remains partially responsive under external conditions. This differential sensitivity further supports mechanistic divergence, implicating distinct structural determinants underlying the two outcomes. Fifth, the recycling fraction varies little across a salt gradient (Fig. 9). This constancy persists despite changes in overall termination efficiency, suggesting that pathway commitment occurs prior to termination completion. The invariant ratio implies that the elongation complex samples alternative structural states that bias subsequent progression rather than undergoing a simple electrostatic shift. Collectively, discontinuous dissociation kinetics, independent end-dwell behavior, differential sequence sensitivity, selective genetic perturbation, and ion-invariant pathway commitment converge on a common conclusion: intrinsic RNAP3 termination operates as a branched stochastic process of termination-pathway duality. Upon reaching the terminator, the complex encounters a mechanistic trifurcation that directs it toward readthrough, recycling termination or decomposing termination. This framework reconciles both the coexistence and the distinct regulatory properties of the two termination pathways. Beyond these distinctions, the dominant pathway of intrinsic termination is reversed between bacterial and human systems. Whereas the recycling pathway prevails in E. coli hairpin-dependent termination, human RNAP3 primarily follows the decomposing pathway. This scarcity of recycling termination seems initially inconsistent with the high reinitiation characteristic of RNAP343,46. Conceptually, facilitated reinitiation aligns more closely with recycling termination, after which DNA-bound RNAP3 can return to the promoter via 1D diffusion rather than the 3D search required following decomposing termination. However, recycling termination is relatively infrequent in human RNAP3. The recycling fraction ranged from 0.023 to 0.14 across five gene terminators, markedly lower than in bacterial intrinsic termination (0.70–0.91)22 or Sen1-dependent termination by yeast RNA polymerase II ( ~ 0.33)31, yet comparable to Rho-dependent termination (0.11–0.27)24 and higher than phage intrinsic termination (0.0–0.0053)25. This suggests that recycling termination contributes minimally to the overall reinitiation efficiency of RNAP3, implying that additional mechanisms support its high transcriptional output. One possibility is promoter-tethered recycling, in which promoter-bound factors recapture the polymerase47–49. Local RNAP3 concentration may also be elevated by molecular crowding, gene looping, or specific chromatin contacts that bridge the terminator and promoter47,50–53. Such topological constraints could transform decomposition into a localized search, in which TFIIIB and TFIIIC act as physical backstops to funnel the post-termination complex back to the promoter48. In addition, the RPC3·RPC6·RPC7 subunit trimer may act as an internal chaperone that maintains RNAP3 in a reinitiation-competent state16,54. These models of factor-mediated repositioning or direct handoff commonly imply a channeling effect, in which the polymerase is spatially restricted within a confined volume near both the terminator and the promoter. Within a looped gene architecture, a decomposing polymerase can be channeled back to the same promoter via constrained 3D diffusion, potentially rivaling or even exceeding the efficiency of 1D search by a recycling polymerase. Whether a factor-mediated channeling mechanism facilitates RNAP3 reinitiation at the same promoter remains to be determined. Kinetic analyses reveal a prevalent ordering: readthrough is typically followed by recycling termination, with decomposing termination always occurring last (Fig. 8). The exception was the 7SL RNA template, where RNA release in recycling termination occurred faster than readthrough. While similar template-dependent variability has been reported in phage RNA polymerases25, the recycling pathway is consistently faster than readthrough in E. coli Rho-dependent termination24. Thus, while recycling termination is generally slower than readthrough in RNAP3, its kinetics remain highly sensitive to sequence context. Across all tested templates, RNA release is slower in decomposing termination than in recycling termination, showing only partial overlap on the hp+0 template. This mirrors E. coli Rho-dependent termination, where the decomposing pathway (3–9 min) lags behind the recycling pathway (0.4–2 min)24. These results imply that RNA release requires more profound conformational rearrangements in decomposing termination as observed in our E. coli and phage termination studies24,25. Thus, it takes a more active, laborious process to dissociate both RNA and DNA together than to release RNA alone. Recent in vivo high-resolution mapping of human RNAP3 transcription has revealed that termination, rather than initiation, is a major rate-limiting step55. We found that intrinsic RNA release takes 14 s (recycling in 7SL RNA gene) to 120 s (decomposing in 5S RNA gene), which can be slower than initiation or elongation for short RNAs produced by RNAP3. This finding provides a biophysical basis for the in vivo observations, identifying the terminator-associated structural bifurcation as a fundamental kinetic bottleneck that determines total transcriptional output. This conserved kinetic distinction of RNA release supports our model where recycling termination proceeds via RNA shearing, in which the transcript is pulled from the RNA·DNA hybrid within the transcription bubble while the polymerase remains DNA-bound24. In contrast, decomposing termination involves polymerase displacement and forward movement of the transcription bubble, a process reminiscent of, yet distinct from, the hypertranslocation model24,25. The requirement for more extensive structural rearrangements likely underlies the slower kinetics of decomposing termination. Recycling termination may enable rapid and dynamic responses to sequence-specific or cellular signals. In bacteria, it facilitates recycling for reinitiation via 1D diffusion24–27, although its in vivo confirmation remains pending. By contrast, decomposing termination may serve as the robust default pathway under specific regulated or stress conditions. By analogy, human RNAP3 may rely primarily on decomposing termination as a high-fidelity baseline for constitutive expression, while reserving recycling termination for high-frequency, context-dependent regulation. The RPC10 C-terminal domain is critical for intrinsic termination by human RNAP3 (Fig. 10). The D88A/E89A mutation impairs both termination pathways and generates arrest complexes absent in the wildtype. These defects likely reflect impaired RNA release and polymerase dissociation. Loss of conserved acidic residues may compromise each termination process. Partial recovery on T8/A8 suggests that an extended oligo(T) tract can partially compensate for the mutation, particularly by promoting the decomposing pathway through increased hybrid instability. The RPC10 subunit plays a more prominent role in human termination than in yeast. In contrast to the yeast system8,43, where the same mutation has mild effects, human RNAP3 is highly sensitive, especially at short T-tracts. This difference may reflect species-specific tuning and the flexible funnel positioning of the C-terminal domain observed via cryo-EM19. Uniformly delayed kinetics across both pathways suggest RPC10 helps resolve backtracked or stalled states through RNA cleavage at the active site. Its disruption stabilizes arrest complexes, reducing elongation processivity and termination efficiency. A distinguishing feature of RNAP3 termination is its strong dependence on the number and placement of Ts in the nontemplate strand. Robust termination requires at least four consecutive nontemplate Ts, consistent with previous findings on human and Xenopus RNAP317,37,38,40,41 (Fig. 5). This differs from yeast RNAP3, in which at least five or six Ts are required9,39,56. Longer oligo(T) tracts increase both termination efficiency and recycling fraction, likely by facilitating RNA shearing through elevated U content. Thus, the T-tract both induces termination and influences the post-termination fate of RNAP3. Strand-swap and mismatch experiments further demonstrated that the essential T-tract must reside on the nontemplate strand. These findings support a model in which RNAP3 recognizes this strand via direct, base-specific interactions, consistent with cryo-EM data showing RPC2-oligo(T) contacts17. Moreover, the nontemplate oligo(T) retains partial activity even when unpaired, suggesting that recognition depends more on base-amino acid contacts than on DNA duplex stability. We also found that an upstream RNA hairpin enhances both termination pathways, with the strongest effect when positioned immediately adjacent to the T-tract. As the hairpin is moved upstream, this effect gradually diminishes for both pathways. These findings suggest that, as in bacterial hairpin-dependent termination22,26, upstream hairpins weaken the RNA·DNA hybrid to destabilize the complex, promoting spatial precision and termination efficiency. As secondary structures are widespread in RNAP3 transcripts, similar hairpin-mediated effects are likely to contribute to termination in vivo. Transcription events were classified by Cy3-RNA and Cy5-DNA fluorescence, fully matching A488-RNAP3 signals. RNAP3 end-dissociation was represented by Cy5 PIFE decline, mirroring A488 loss. While the three-color assay uniquely times transient polymerase retention during decomposing termination, the two-color setup improves resolution 7.5-fold (2 versus 15 s), avoids the modest activity reduction (17%) observed with A488-RNAP3, and unambiguously distinguishes two termination pathways, readthrough, and arrest. Importantly, termination efficiencies were consistent between the two methods. This study defines key features of RNAP3 intrinsic termination, though several questions remain. First, the experiments used purified RNAP3 with no cellular factors. In vivo, proteins such as Senataxin57, MAF158, or various chromatin regulators59 likely modulate termination efficiency, pathway selection, or both. Second, the five gene templates and synthetic constructs tested here may not capture the full diversity of RNAP3 targets. Third, although our single-molecule assay yields high-resolution kinetics, it lacks direct structural visualization of terminating complex. Future work could reconstitute termination on chromatin or with regulatory factors to better replicate physiological conditions. Combining time-resolved cryo-EM with single-molecule imaging may help visualize the stepwise conformational changes. It is also important to examine other RNA polymerases and determine how pathway usage varies with gene class, promoter architecture, or chromatin context. Ultimately, understanding how intrinsic and extrinsic cues converge at termination may illuminate broader mechanisms of transcriptional regulation and their disruption in disease60–62. In summary, human RNAP3 employs two termination mechanisms at a single terminator, reinforcing dual-pathway conservation across life. Decomposing termination dominates and proceeds more slowly, as polymerase displacement requires more extensive structural rearrangements than the RNA shearing involved in recycling termination. Thus, RNAP3 undergoes a structural trifurcation establishing a temporal hierarchy: readthrough for continued elongation, recycling termination for facilitated reinitiation, and decomposing termination for unfacilitated reinitiation. This underscores the mechanistic diversity and regulatory flexibility of RNAP3 termination. Methods Assembly of stalled transcription complexes Each transcription template was prepared by annealing a 3′-Cy5-labeled nontemplate strand (1.5 µM) with a 3′-biotin-labeled template strand (1.0 µM) (Supplementary Table 3). Oligomers were mixed in 20 mM HEPES and 100 mM NaCl, heated to 95 °C, and cooled to 20 °C at −1 °C/min. Oligonucleotides were synthesized by GENEWIZ (Azenta Life Sciences). Annealed duplexes (50 nM) were incubated at room temperature for 10 min with 210 nM human RNAP3 (wildtype or D88A/E89A mutant RPC10) and 150 µM Cy3-ApU (TriLink) in 20 mM HEPES, 60 mM (NH4)2SO4, 10 mM MgSO4, and 10 mM DTT. ATP, CTP, and GTP (250 µM each; Cytiva) were then added, and reactions were incubated at 37 °C for 20 min to form stalled complexes. Two-fluorophore single-molecule assay Stalled complexes were immobilized at 37 °C in a flow chamber (Supplementary Fig. 5) via biotin-streptavidin-biotin conjugation on PEGylated quartz slides63. Slides were cleaned with piranha solution, silanized with 3-aminopropyltrimethoxysilane (United Chemical Technologies), passivated with a 1:40 mixture of biotin-PEG-SC and mPEG-SVA (Laysan Bio), and incubated with 0.2 g/l streptavidin (Invitrogen). To minimize photobleaching, imaging was performed in oxygen-scavenging buffer containing 5 mM protocatechuic acid (Sigma-Aldrich), 150 nM protocatechuate 3,4-dioxygenase (Oriental Yeast Co.)64, saturated Trolox65, 40 mM HEPES, 60 mM (NH4)2SO4, 10 mM MgSO4, and 10 mM DTT. Imaging began with a 30-s baseline before the injection of 1.25 mM NTPs (time zero) to resume elongation. Fluorescence was recorded using a custom prism-type total internal reflection fluorescence microscope with an Olympus UPlanSApo 60× objective, 532 nm and 640 nm Excelsior lasers (Spectra-Physics), Semrock notch filters (NF03 532E-25 and 633E-25), and a 635dcxr dichroic mirror (Chroma). Fluorescence was split and detected with an iXon DU-897 EMCCD camera (Andor). Cy3 and Cy5 were alternately excited every 1 s. Transcription activity was defined as the number of complexes showing a single step of Cy3 PIFE decline among those colocalizing with Cy5. Approximately 20% of colocalizing complexes were active and classified as readthrough, recycling termination, decomposing termination, or arrest based on Cy3/Cy5 signal patterns (see Results). Termination efficiency was defined as the fraction of active complexes undergoing termination. Three-fluorophore single-molecule assay A488-labeled RNAP3 was generated by fusing SpyTag to RPC1 and conjugating with A488-SpyCatcher. SpyCatcher (engineered with a single cysteine) was reacted with A488 C5-maleimide to achieve 1:1 labeling. The A488-SpyCatcher was stored in PBS with 2 mM DTT and incubated with SpyTag-RNAP3 on ice for 60 min at a 4.3:1 molar ratio. A488-RNAP3 transcription was stalled in 20 mM HEPES, 28 mM NaCl, 0.6× PBS, 7 mM MgCl2, 1.4 µM ZnCl2, and 1.48 mM DTT, then subjected to the two-color protocol with the same imaging buffers. Excitation time was extended to 5 s, compensating for reduced laser intensity used to delay A488 photobleaching. Approximately 50% of active complexes emitted detectable A488 signal under 473 nm Excelsior laser excitation. Bulk transcription assay DNA templates (50 nM) were incubated with 210 nM RNAP3 and 1.25 mM NTPs plus 150 µM Cy3-ApU or 200 nM 15-mer RNA primer in 40 mM HEPES, 60 mM (NH4)2SO4, 10 mM MgSO4, and 10 mM DTT at 37 °C for 17 min. Reactions were terminated by treatment with 3.3 units/µl DNase I (Enzynomics) for 30 min followed by 5 g/l proteinase K (Enzynomics) for 15 min. RNA was ethanol-precipitated twice at −20 °C: initially with 250 mM NaCl and 62.5 µM NTPs, then a second time without NTPs to remove residual nucleotides. Pellets were resuspended in 4 µl quenching buffer (43% formamide, 0.5× TBE, 10 mM EDTA), denatured at 90 °C for 3 min, and separated by 9% urea-PAGE. Gels were run at 200 V for 30 min and analyzed using ImageJ (National Institutes of Health). Due to low signal intensity, gels were imaged on a G:BOX Chemi XT4 (Syngene) for 15 min under green-light excitation using a 605 M emission filter to detect Cy3-RNA. Uncropped scans of gels are provided in the Source Data file. RNAP3 expression and purification Wildtype human RNAP3 was expressed in HEK Expi293F cells (Thermo Fisher Scientific) transfected with six plasmids encoding all 17 subunits (including RPC7α), as previously described54. RPC2 carried an N-terminal protein A tag. Cells were lysed in 20 mM HEPES, pH 8.2, 300 mM NaCl, 0.25% CHAPS, 10 µM ZnCl2, 5 mM ATP, 5 mM MgCl2, 10% glycerol, and 2 mM DTT, with protease inhibitors. Complexes were purified with IgG beads (Smart-Lifesciences) at 4 °C for 3 h and washed five times with lysis buffer containing 200 mM NaCl and 0.05% CHAPS. Protein A tags were cleaved on-column overnight. RNAP3 was eluted in 20 mM HEPES, pH 8.2, 150 mM NaCl, 5 mM MgCl2, 10 µM ZnCl2, 10% glycerol, and 2 mM DTT, and further purified by Mono Q chromatography using a 150–1000 mM NaCl gradient (eluting around 400 mM). The final preparation was dialyzed into 150 mM NaCl, concentrated to 1.8 mg/ml, and stored at −80 °C. The D88A/E89A mutant RNAP3 was expressed using nine plasmids: N-terminal protein A-tagged RPC10 in one vector, and the remaining subunits distributed across eight others (RPC1, RPC2, RPC3/4/5, RPC6/PRABC3/4/5, RPC7, RPC8, RPC9, and RPAC1/2/PRABC1/2). SpyTaq-RNAP3 was expressed using six plasmids: N-terminal protein A-tagged SpyTag-RPC1 in one, and the remaining subunits across five others (RPC2, RPC3/4/5, RPC6/PRABC3/4/5, RPC7/8/9/10, and RPAC1/2/PRABC1/2). Both the mutant and SpyTag RNAP3 were purified from Expi293F cells, as previously described66. Statistical analysis All statistical tests were two-tailed with a significance threshold of α = 0.05. Exact P values (whether significant or not) are reported in the text. Differences between two conditions were evaluated using unpaired two-tailed t-tests, with Welch’s correction applied when variances were unequal. Assumptions of normality and homoscedasticity were assessed by inspecting replicate-level residuals. The unit of analysis was the replicate-level percentage (e.g., decomposing or recycling termination), not individual complex molecules. Data are presented as means ± standard deviations from independent biological experiments (n = 5 for RPPH1, n = 4 for RN7SL1, and n = 3 for the remaining 21 templates). Each independent experiment included at least 30 single molecules. The total number of molecules analyzed per condition is provided in Supplementary Table 2. Statistical analyses were conducted using MATLAB R2021a (MathWorks), IDL 7.0 (ITT), Excel 2016 (Microsoft), and OriginPro 8.5 (OriginLab). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source data Author contributions S. Han designed the experimental scheme, performed all single-molecule and gel experiments and data analysis, and drafted the manuscript. Q.W. expressed and purified all recombinant human proteins, including wildtype, SpyTag, and RPC10-mutant RNAP3. E.S. contributed to the initial design and participated in discussions and interpretation of the results. S. Hong, C.K., and Y.X. supervised the project. S. Hong oversaw the single-molecule experiments and data analysis. C.K. oversaw the biological aspects of the study and led the review and editing of the manuscript. Y.X. supervised protein expression and purification. All authors discussed the results and commented on the manuscript. Peer review Peer review information Nature Communications thanks anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Funding S. Hohng acknowledges support from the National Research Foundation of Korea [2022R1A2C3008746 and RS-2023-00218318]. Y.X. acknowledges support from the New Cornerstone Investigator Program. Q.W. acknowledges support from the National Natural Science Foundation of China [32471248]. C.K., E.S., and S. Han declare no relevant funding. Open access was funded by the KAIST Stem Cell Center and the Seoul National University College of Natural Sciences. Data availability The data supporting the findings of this study are available from the corresponding authors upon request. Source data for the figures and Supplementary Figs. are provided as a Source Data file. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Sun Han, Qianmin Wang. Contributor Information Changwon Kang, Email: ckang@kaist.ac.kr. Yanhui Xu, Email: xuyh@fudan.edu.cn. Sungchul Hohng, Email: shohng@snu.ac.kr. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73393-6. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source data Data Availability Statement The data supporting the findings of this study are available from the corresponding authors upon request. Source data for the figures and Supplementary Figs. are provided as a Source Data file. Source data are provided with this paper.