MainGene expression under the control of sequence-specific transcription factors (TFs) provides the cornerstone of transcriptional programs directing general and cell-type-specific functions. Over 100 TFs in budding yeast and more than 1,000 mammalian TFs harbor DNA-binding domains (DBDs) that enable recognition of short cis-regulatory DNA elements throughout the genome. In the search for functional targets, TFs must navigate a vast excess of genomic DNA. For example, GCN4, the bZIP activator regulating budding yeast genes for amino acid biosynthesis1 and the yeast functional homolog of human AP1, binds to 546 targets comprising 1,000 off-target GCN4-binding sites on yeast genomic DNA with low in vivo occupancy1.To test the hypothesis that chromatin suppresses off-target GCN4 binding, we introduced a 3.1-kb MC carrying the HIS3 gene (TATO7LO6–HIS3: TRP1–ARS1–TetO7–LacO6–HIS3) (Fig. 1a) into a HTB2–GFP knock-in yeast strain to enable detection of the purified MC (Extended Data Fig. 1a). The HIS3 MC is organized in an array of nucleosomes (Fig. 1b), displaying a well-positioned +1 nucleosome with an upstream border at −29 bp relative to the transcription start site (TSS), a −1 fuzzy nucleosome with a downstream border at ~−150 bp and an NFR of ~120 bp spanning the HIS3 promoter harboring poly(dT) sequences, TATA boxes and a single UAS (Fig. 1c). The chromatin organization of the HIS3 MC, typical of yeast promoters, is consistent with previous studies of genomic and minichromosomal HIS3 (refs. 29,34,35). In addition to the HIS3 promoter NFR, we observed NFRs at ARS1, the HIS3 3′ end (Fig. 1c) and the TRP1 promoter, as well as TetR footprints over tandem operators because of affinity purification (Extended Data Fig. 1c).Fig. 1: GCN4 single-molecule binding kinetics on purified, immobilized MCs.Full size imagea, Schematics of purified 3.1-kb ccDNA and MC immobilized on PEGylated TIRF slides. Right, spheres represent H2B–GFP-labeled nucleosomes. SBP, streptavidin-binding peptide. b, MNase digestion patterns of purified MC, analyzed on an Agilent Bioanalyzer. All MC samples used in this work show similar band patterns. c, Normalized nucleosome occupancy (filtered for 125–175-bp reads) across HIS3 from purified MC determined by MNase-seq. Gray ovals indicate deduced nucleosome positions. The bar diagram shows cis elements: ACS, ARS consensus sequence (ORC binding); ARS1, autonomously replicating sequence 1; dT, poly(dT) sequence; UAS, GCN4-binding motif. d, Representative TIRF video frames and TAMRA–GCN4 fluorescence trace over time. In the H2B–GFP panel, MC spots are boxed; the example GCN4 trace is generated from the bolded box. The TAMRA–GCN4 panels show the same field of view at indicated times (arrows). Asterisks indicate additional GCN4-binding events. The bar diagram shows GCN4-bound (dark red) and unbound (gray) states. e, Rastergrams of 100 randomly selected GCN4 traces on ccDNA or MC templates arranged in order of first GCN4-binding event after flow-in at time 0.Source dataThe MC or its closed circular DNA (ccDNA) appears as bright fluorescent spots upon immobilization for TIRF imaging (Fig. 1d and Extended Data Fig. 1d–e). We measured TAMRA–GCN4 binding under physiological salt conditions to individually resolved MCs or ccDNAs by fluorescence colocalization, which showed single (or multiple) GCN4 molecules binding to the 3.1-kb substrate as indicated by the abrupt increase in TAMRA fluorescence (Fig. 1d and Extended Data Fig. 1e). To minimize multiple GCN4-binding events (Fig. 1d and Extended Data Figs. 1e and 2a,c), we performed measurements at low GCN4 concentration (0.1 nM) and call discrete TAMRA–GCN4 signals above background as the bound state. Inspection of rastergram profiles for GCN4 association and dissociation on individual HIS3 MC or ccDNA molecules (Fig. 1e) revealed qualitatively that chromatin organization regulates GCN4 target search by both lowering the GCN4 association rate (kon) and raising the dissociation rate (koff).Off-target GCN4 binding to naked DNAWe first examined GCN4 kinetics on the 3.1-kb ccDNA template (TATO7LO6–HIS3) devoid of histones. To quantify dissociation rates (koff = 1/τ), we plotted the residence time (τ) histogram for thousands of binding events and applied exponential fitting to extract koff. The τ histogram for the wild-type (WT) HIS3 ccDNA reveals a large GCN4-binding fraction with τ > 100 s (49%, n = 1,021; Fig. 2a and Extended Data Fig. 3a). Deletion of the core sequence ‘TGACTC’ from the UAS (UAS∆ ccDNA) shifted the GCN4 τ peak to 50–100 s, which was still surprisingly stable, and correspondingly decreased the τ > 100 s fraction (23%, n = 1,468; Fig. 2a and Extended Data Fig. 3a). Because a previous study1 identified off-target GCN4-binding sites located beyond the UAS on the naked HIS3 MC DNA sequence, we scrambled the underlying degenerate motifs (sequences mismatched to the cognate motifs) plus additional half-sites (UAS∆scr ccDNA; Extended Data Fig. 3g,j), which resulted in a further histogram peak shift to ~20 s (n = 1,589; Fig. 2b).Fig. 2: GCN4 displays stable ectopic entrapment by degenerate motifs and length-correlated nonspecific binding on naked DNA.Full size imagea,b, GCN4 residence time (τ) histogram (average of replicates, x axis log-binned) for WT and UAS∆ ccDNA (a) and for UAS∆scr (scrambled degenerate motifs) ccDNA (b). Error bars indicate the s.d. among all replicates (c). The last bin includes binding events > 600 s. c, Curves of 1 − CDF for data in a,b. Shadows indicate the s.d. among replicates and thin black curves represent exponential fits. Pie charts represent the three-component exponential fit for GCN4-binding fractions (%) and τ (s) on WT, UAS∆ and UAS∆scr ccDNA. The numbers of GCN4-binding events (n) and biological replicates are labeled. d, Linear DNA fragments for TIRF microscopy with UAS and half-site marked. e,f, GCN4 τ histogram for WT (e) and UAS∆ (f) linear DNA fragments. g, GCN4 τsb (Fsb in brackets) on WT linear DNA fragments (e). Column values indicate the fitting result of original dataset and error bars indicate the 95% confidence intervals (CIs) determined by bootstrapping. P values were calculated using a double-tailed t-test (Methods). NS, not significant. h, GCN4 major τ (dominant fraction in brackets) on UAS∆ (f) and UAS∆scr linear DNA fragments. Column values, error bars and P values are as in g. i, Model of GCN4 1D diffusion on bare DNA. When a specific target (UAS) is present, GCN4 is stably captured and τ is dominated by target-specific binding. On nonspecific DNA of greater length, there are more degenerate motifs and nonspecific DNA along which GCN4 diffuses in 1D, prolonging its overall off-target τ.Source dataExponential fitting of GCN4 residence time (Fig. 2c) discerned three binding populations: stable (sb), intermediate (int) and transient (tb), reflecting different levels of binding stability. For WT ccDNA, the dominant stable fraction (Fsb=70%) displayed τsb of 245 s, while UAS∆ and UAS∆scr ccDNAs gave lower τsb of 90 s (Fsb = 66%), and 35 s (Fsb = 59%), respectively (P ≤ 0.027; Fig. 2c and Extended Data Fig. 3b,c). Note that Fsb and τsb for WT ccDNA may have been underestimated because of partial exclusion of right-censored binding events (Methods). These changes in τsb represent progressively weaker stabilities of GCN4 on the UAS, degenerate sites and other nonspecific sequences. Thus, degenerate motifs and the entire spectrum of nonspecific DNA sequences on the 3.1-kb HIS3 ccDNA contribute to off-target GCN4 residency.Given the stable GCN4 off-target binding on 3.1-kb ccDNA, we considered whether DNA length also has an impact on GCN4 τ. On short WT HIS3 promoter DNA fragments of 55 bp, 156 bp and 286 bp (Fig. 2d), GCN4 displayed τsb of 212 s, 299 s and 287 s (Fsb = 89%, 65% and 86%), respectively, comparable to τsb = 245 s (Fsb = 70%) for the 3.1-kb WT ccDNA (Fig. 2e,g). On UAS-deleted DNA fragments (UAS∆) of 49 bp, 150 bp and 280 bp (Fig. 2d), the dominant GCN4-binding population displayed increasing τ values of 0.46 s, 4.4 s and 20 s (F = 91%, 98% and 83%), respectively, in positive correlation with increasing DNA length (Fig. 2f,h). To eliminate the effect of degenerate motif entrapment on the 280-bp UAS∆ DNA harboring one half-site (Fig. 2d), we scrambled the half-site (280-bp UAS∆scr), leading to a τ decrease to 6.7 s (Fig. 2h and Extended Data Fig. 3e,f), still correlated with DNA length. These observations are consistent with previous findings for murine Sox2 (ref. 36) and support GCN4 1D diffusion on naked DNA. This was further confirmed by three-color fluorescence resonance energy transfer (FRET) analysis6,8 of GCN4 binding to linear promoter DNA harboring the UAS and a half-site (Extended Data Fig. 4). We note that GCN4 1D diffusion displayed salt sensitivity (Extended Data Fig. 4c–e), similar to GAGA factor8, suggesting that 1D hopping has a role in GCN4 1D diffusion. Altogether, both ectopic entrapment at degenerate motifs and promiscuous interactions with nonspecific sequences during GCN4 1D diffusion are responsible for stable off-target residence on naked DNA (Fig. 2i).Chromatin suppresses stable off-target GCN4 binding by blocking DNA accessibilityThe dissociation kinetics of GCN4 on the MC differ greatly in comparison to ccDNA. The GCN4 τ histogram showed a peak shift from >100 s for WT ccDNA (Fig. 2a) to ~10 s for the WT MC (n = 2,738; Fig. 3a). Exponential fitting also revealed a major decrease in Fsb (70% to 9%, P 60 s. c, Curves of 1 − CDF and pie charts representing three-component exponential fit for WT, UAS∆ and UAS∆scr MC (formats as in Fig. 2c). The numbers of binding events (n) and biological replicates are labeled. d, Normalized nucleosome occupancy on UAS∆ MC compared to WT by MNase-seq. e, Top, reconstituted long-linker mononucleosomes for TIRF microscopy. Bottom, AF488–DNA and Cy5–H2B scans of nucleosomes on native PAGE gel (three independent electrophoreses show similar band patterns). Solid lines in the Cy5 panel mark the positions of 275-bp and 269-bp DNA bands. f,g, GCN4 τ histogram for WT (f) and UAS∆ (g) long-linker nucleosomes (with half-site). Gray shading represents τ histograms for 156-bp WT (f) or 150-bp UAS∆ (g) linear DNA.Source dataDeletion of the UAS (UAS∆ MC, n = 6,131) did not alter nucleosome arrangement (Fig. 3d) but reduced τsb (from 143 s to 89 s, P = 0.029), while intermediate and transient fractions and τ values remained the same as the WT MC (Fig. 3a,c and Extended Data Fig. 5b–d). Hence, off-target GCN4 binding on the UAS∆ MC is dominated by intermediate (57%, 12 s) and transient (35%, 2 s) populations, in contrast to UAS∆ ccDNA, where off-target binding is dominated by the stable population (66%, 90 s; Fig. 2c). Although degenerate motif interactions are a major cause of stable off-target binding on ccDNA, scrambling degenerate motifs on the MC (UAS∆scr MC, n = 7,499) did not change the overall τ distribution (Fig. 3b,c and Extended Data Fig. 5e,f), indicating that nucleosome organization sufficiently suppresses off-target GCN4 entrapment.Notably, the GCN4 τ histogram profile for the HIS3 MC was intermediate to the τ profiles for 150-bp and 280-bp linear UAS∆ naked DNA (Extended Data Fig. 5g), consistent with average NFR lengths on the MC (Fig. 1c and Extended Data Fig. 1c). This suggests that nucleosomes limit GCN4 off-target search time by confining DNA exposure to NFRs. We further analyzed GCN4 dissociation kinetics using a reconstituted ‘long-linker’ nucleosome containing a positioned Widom 601 core particle (nucleosomal H2B is Cy5-labeled), with substitution of a half-site between SHL-2 and SHL-2.5 to mimic the HIS3 + 1 nucleosome and a 122-bp extension of linker DNA containing the HIS3 promoter with UAS (Fig. 3e). For this WT long-linker nucleosome (122N6), the GCN4 τ histogram showed a decrease in the predominant >100-s fraction compared to unreconstituted bare DNA (Fig. 3f), with a corresponding decrease in Fsb and τsb (Extended Data Fig. 5i).For the UAS∆ nucleosome (116N6, with half-site), there was also a substantial left-shift in the τ histogram compared to bare DNA (Fig. 3g), with a corresponding reduction in τ from 24 s to 2.5 s—shorter τ than for naked 150-bp UAS∆ DNA (P = 0.003; Extended Data Fig. 5i). The same analysis on UAS∆ 116N6 without the half-site also revealed a reduction in GCN4 τ from 5.4 s on bare DNA to 3.0 s (P = 0.005; Extended Data Fig. 5i). Hence, nucleosomes modulate GCN4 residence time by limiting bare DNA length and restricting 1D diffusion.Promoter NFR length determines the association rate of stable GCN4 bindingTo complement the kinetics of GCN4 dissociation, we also determined the association rate by measuring the interval between GCN4 flow-in and the first binding event (Fig. 4a). The general association rate constant (kon) on naked HIS3 DNA correlated well with template length (Fig. 4b); 55-bp DNA displayed the lowest kon (0.087 nM−1 s−1), increasing progressively by ~5-fold (to 0.41 nM−1 s−1) for the 3.1-kb ccDNA. For the four NFRs on the HIS3 MC with ~500 bp of total bare DNA, the general kon (0.26 nM−1 s−1) was comparable to the value for a 587-bp linear DNA fragment (0.22 nM−1 s−1, P = 0.06), suggesting that the general kon is mainly driven by NFRs. In addition, the kon for a 122N6 long-linker nucleosome and 156-bp naked promoter DNA was almost equivalent (0.11 nM−1 s−1, P = 0.44), indicating that the general kon of GCN4 is primarily determined by the total length of NFR DNA, with little contribution from the nucleosome core particle. In other words, GCN4 association predominantly occurs at NFRs.Fig. 4: Promoter NFR length modulates GCN4 association rate and fraction of stable binding.Full size imagea, Rastergrams of GCN4 traces sorted by first binding event (top) or first stable binding event (>100 s; bottom) to WT MC. Dotted lines represent exponential fitting for GCN4 kon (top) and kon,sb (bottom). b,c, GCN4 kon (b) and kon,sb (c; for binding events > 100 s) for WT templates as indicated (same datasets as Figs. 2 and 3). For linear DNA templates, the x axis is in linear scale with DNA length labeled. Dot and diamond values are fitting results of the original datasets and error bars indicate the 95% CIs determined by bootstrapping. P values were calculated using a double-tailed Wilcoxon rank-sum test. The numbers of analyzed traces (n) were as follows: 55 bp, 232; 156 bp, 279; 286 bp, 203; 587 bp, 294; ccDNA, 396; MC, 266; 122N6, 266. d, Graphics show how kon,sb is limited by 1D diffusion. When the distance from GCN4 landing position to the UAS is too long, GCN4 is more likely to dissociate before reaching UAS (red). e,f, Normalized nucleosome occupancy on promoter∆ (e) and RSC-treated (f) MCs. In e, promoter∆ and WT MCs are aligned at the shared UAS position. g, GCN4 kon,sb for WT, promoter∆ and RSC-treated MCs (formats for statistics as in c). The numbers of analyzed traces (n) were as follows: promoter∆ MC, 245; RSC-treated MC, 169 (same datasets as h). h, Fsb for WT, promoter∆ and RSC-treated MCs. Column values indicate averages of replicates and error bars indicate the s.d. among biological replicates. P values were calculated using a double-tailed t-test. The numbers of binding events (n) and biological replicates are labeled.Source dataFor the stable binding GCN4 subpopulation representing UAS-specific association on the MC, we determined the stable association rate constant (kon,sb) for binding events τ > 100 s (Fig. 4a,c). Unlike general kon, the kon,sb reached a plateau around 0.15 nM−1 s−1 as linear DNA length increased to 286 bp (Fig. 4c), suggesting that kon,sb is limited by the distance range of 1D diffusion before entrapment or dissociation37,38 (Fig. 4d). Of interest, kon,sb apparently exceeded this limit for ccDNA (‘Discussion’). Notably, the WT MC, the 122N6 long-linker nucleosome and the 156-bp naked promoter DNA all showed comparable kon,sb values (0.11, 0.12 and 0.11 nM−1 s−1, respectively, P ≥ 0.20), indicating that NFR length at HIS3 promoter determines the GCN4 association rate for UAS-specific binding.Furthermore, contracting the HIS3 promoter NFR length from ~120 bp to ~30 bp through the deletion of promoter sequences responsible for NFR formation16 but keeping the UAS intact (promoter∆; Fig. 4e) led to a reduction in kon,sb from 0.11 to 0.076 nM−1 s−1 (P = 0.012; Fig. 4g) but no reduction in the general kon (Extended Data Fig. 6d). In contrast, expansion of the promoter NFR to ~200 bp through the action of the ATP-dependent chromatin remodeler RSC (Fig. 4f) resulted in a notable increase in GCN4 Fsb (9% to 16%, P = 0.018; Fig. 4h) while barely raising kon,sb (0.11 nM−1 s−1 to 0.13 nM−1 s−1, P = 0.70; Fig. 4g), likely because of the 1D diffusion constraint. Hence, altering the HIS3 promoter NFR length leads to corresponding changes in the UAS-specific association rate and the fraction of stable GCN4 binding to chromatin.AD accelerates GCN4 dissociation and association to enhance promoter targetingRecent studies have revealed that, in addition to DBDs, IDRs or ADs in TFs contribute to genome-wide chromatin binding26,27. GCN4 possesses a defined AD, which is also an IDR, spanning residues 1–134 at the N terminus39. To examine the role of the GCN4 AD in promoter search, we measured binding of a truncated GCN4 (GCN4∆AD) to the MC (Fig. 5a). Compared to GCN4FL, the τ distribution showed a large increase in stable GCN4∆AD binding to the WT MC (Fsb from 9% to 75%, τsb from 143 s to 293 s, n = 561; Fig. 5b) and a similar increase in stable GCN4∆AD binding to WT ccDNA (Extended Data Fig. 7b). These findings indicate that the AD in GCN4FL facilitates GCN4 dissociation from either chromatin or naked DNA templates.Fig. 5: AD accelerates GCN4 dissociation and association to enhance target selectivity and search efficiency.Full size imagea, GCN4 constructs for TIRF. b,c, GCN4∆ADτ histogram for MC WT (b) and UAS∆ (c) (formats as in Fig. 4e). Gray shading represents the τ histogram of GCN4FL on corresponding templates. Error bars indicate the s.d. among replicates. The numbers of binding events (n) and biological replicates are labeled. d, GCN4∆ADτ histogram for linear 55-bp WT and 49-bp UAS∆ DNA. e, Comparison between GCN4∆AD (same datasets as d) and GCN4FL (same datasets as Fig. 2e,f) τsb on 55-bp WT DNA and major τ on 49-bp UAS∆ DNA (formats for statistics as in Fig. 2g). f, Rastergrams for GCN4∆AD and GCN4FL on WT MC, as in Fig. 1e. g, kon of GCN4∆AD (same datasets as b,c) and GCN4FL (same as Fig. 4b) for different WT templates. Dots represent the fitting values of original datasets and error bars indicate the 95% CIs determined by bootstrapping. The numbers of analyzed traces for GCN4∆AD were as follows: ccDNA, 54; MC, 90. h, GCN4 constructs for live-cell SMT. Representative trajectories shown with the numbers of trajectories (n) and biological replicates are labeled. i, Curves of 1 − CDF (mean of bootstraps) of τ for GCN4FL, GCN4∆AD and H2B in live yeast. Error bars indicate the s.d. among bootstrapped data. P values were calculated using a double-tailed Wilcoxon rank-sum test. j, The τsb (mean of bootstraps) for GCN4FL and GCN4∆AD in live yeast, corrected by H2B τsb. Error bars indicate the s.d. P values were calculated using a double-tailed t-test. k, Kinetic model of GCN4 target search on native chromatin. (1) AD accelerates GCN4 association. (2) Nucleosomes occlude association. (3) GCN4 1D diffusion is constrained within NFR by nucleosome barriers. (4) GCN4 AD undergoes autoinhibition. (5) Potential 3D steric occlusion exists from nearby nucleosomes. (6) Stable GCN4 occupancy at the UAS recruits downstream factors.Source dataNotably, GCN4∆AD also displayed large Fsb (84%) and long τsb (394 s) on the UAS∆ MC (n = 857; Fig. 5c) comparable to (or even higher than) the values for the WT MC with UAS, indicating that the sequence specificity of GCN4∆AD is compromised. Similar weakened specificity was also observed for ccDNA binding (Extended Data Fig. 7b). Such high and indistinguishable binding stabilities on WT and UAS∆ templates document that the GCN4 AD improves sequence specificity for the full-length TF by facilitating greater dissociation40,41 from off-target sequences than from the UAS.However, for short (~50 bp) linear DNA fragments with little extra sequence, GCN4∆AD exhibited ideal specificity for WT over UAS∆ templates (Fig. 5d,e), in full concurrence with the experimental history of sequence-specific DNA binding reported for bZIP DBDs42. Notably, although GCN4FL and GCN4∆AD displayed the same τsb on 55-bp WT DNA, the AD reduced the off-target τ (that is, increase koff) for GCN4 binding to 49-bp UAS∆ DNA (P = 0.024; Fig. 5e). These observations strengthen our conclusion that the GCN4 AD functions as a sequence filter for the cognate motif.In addition, we found that the AD also increased the general kon of GCN4FL for both ccDNA and MC, compared to the rate for GCN4∆AD (P