Human ISWI chromatin-remodeling complexes sample nucleosomes via transient binding reactions and become immobilized at active sites

Abstract

Chromatin remodeling complexes can translocate nucleosomes along the DNA in an ATP-dependent manner. Here, we studied autofluorescent protein constructs of the human ISWI family members Snf2H, Snf2L, the catalytically inactive Snf2L+13 splice variant, and the accessory Acf1 subunit in living human and mouse cells by fluorescence microscopy/spectroscopy. Except for Snf2L, which was not detected in the U2OS cell line, the endogenous ISWI proteins were abundant at nuclear concentrations between 0.14 and 0.83 μM. A protein interaction analysis showed the association of multimeric Snf2H and Acf1 into a heterotetramer or higher-order ACF complex. During the G1/2 cell cycle phase, Snf2H and Snf2L displayed average residence times <150 ms in the chromatin-bound state. The comparison of active and inactive Snf2H/Snf2L indicated that an immobilized fraction potentially involved in active chromatin remodeling comprised only 1–3%. This fraction was largely increased at replication foci in S phase or at DNA repair sites. To rationalize these findings we propose that ISWI remodelers operate via a “continuous sampling” mechanism: The propensity of nucleosomes to be translocated is continuously tested in transient binding reactions. Most of these encounters are unproductive and efficient remodeling requires an increased binding affinity to chromatin. Due to the relatively high intranuclear remodeler concentrations cellular response times for repositioning a given nucleosome were calculated to be in the range of tens of seconds to minutes.

Chromatin structure is a key determinant of gene regulation. The wrapping of the DNA around the histone octamer protein core in the nucleosome impedes the access of transcription factors to the DNA sequence information. Thus, changes of nucleosome positions at promoter and enhancer regions may directly affect gene expression (1–4). To control nucleosome positions (and with it the access to the associated DNA), a complex chromatin remodeling machinery operates in the eukaryotic cell nucleus. It comprises numerous different types of molecular machines that can translocate nucleosomes in an ATP-dependent manner. Chromatin remodelers can associate with different subunits to form remodeling complexes with distinct biological functions. Due to the high combinatorial complexity it is estimated that several hundreds of different chromatin remodeling complexes exist in humans. These comprise several groups of ATPases classified into the Snf2, ISWI, Mi-2, Chd1, Ino80, ERCC6, ALC1, CHD7, Swr1, RAD54, and Lsh subfamilies (5–7). The associated subunits are responsible for the targeting of the remodeling complexes as well as for the regulation of the remodeling activity (6, 8). One of the best conserved ATPase families involved in chromatin remodeling is the ISWI family (9). It consists of two ATPases, Snf2H and Snf2L in humans. All ISWI proteins contain a conserved ATPase domain that belongs to the superfamily of DEAD/H-helicases (10, 11), located in the N-terminal half of the proteins. Substrate recognition and catalysis are coupled via a SANT and a SLIDE domain near the C terminus that mediate binding to DNA and histone tails (12, 13). ISWI family complexes have various functions, including chromatin assembly and nucleosome spacing (ACF, CHRAC, and RSF), replication (WICH), transcriptional repression (NoRC), and transcriptional activation (NURF and CERF) (14–17). Thus, it seems likely that the accessory proteins present in a given complex regulate its function and are important for target recognition.

Although the ISWI-type chromatin remodelers have been extensively studied in vitro, an analysis in living cells is currently missing. Here, we present a study on the mobility and interactions of chromatin remodeling complexes in living cells. Autofluorescent protein constructs of human Snf2H, Snf2L, Snf2L+13, and Acf1 were investigated in human osteosarcoma and murine fibroblast cell lines. Their intracellular localization as well as their dynamic behavior was analyzed using a multiscale fluorescence fluctuation microscopy approach that comprised noninvasive fluorescence recovery after photobleaching (FRAP), continuous photobleaching (CP), and fluorescence (cross-)correlation spectroscopy (FCS/FCCS)-based experiments (18–21). The results from our quantitative analysis are rationalized in terms of a “continuous sampling” mechanism: In G1/2 phase of the cell cycle, all nucleosomes are probed by a given remodeling complex within minutes via transient binding reactions but only a small fraction of these binding events leads to repositioning. In contrast, extensive repositioning occurs at replication foci in S phase or at DNA repair sites.

Results

Snf2H-GFP and GFP-Snf2L Are Catalytically Active and Bind to Chromatin.

GFP-tagged Snf2H, Snf2L, and Acf1 proteins were transfected into human U2OS cells and murine NIH 3T3 cells. As expected, the proteins were localized to the nucleus in both cell lines (Fig. 1, Fig. S1). As a cell cycle marker the proliferating cell nuclear antigen fused to a red fluorescent protein domain (PCNA-RFP) was cotransfected. PCNA is a central component of the replication machinery and shows a characteristic punctuate pattern during S phase (22). In mouse NIH 3T3 cells, which display distinct DNA dense pericentric heterochromatin foci, the remodeler concentration increased with the DNA density. This increase visualizes the chromatin binding of the GFP-tagged remodelers, in agreement with previous findings from Varga-Weisz and colleagues who characterized the intranuclear localization of Snf2H and its associated Acf1 subunit by immunostaining in mouse cells (23). The ratio between the DAPI or histone H2A-RFP fluorescence signal and GFP-tagged Snf2H/Snf2L/Snf2L+13/Acf1 was similar throughout the cell nuclei in the NIH 3T3 cell line (Fig. S1). This observation indicates a rather homogenous distribution of bound remodeling complexes on the nucleosome chain at the given resolution during G1/2 phase. The inactive variant GFP-Snf2L+13, which contains exon 13 disrupting the enzyme's ATPase cassette (24), displayed the same localization as GFP-Snf2L, demonstrating that chromatin binding was independent of ATPase activity. Interestingly, the ratio of remodeler to chromatin concentration was reduced at some heterochromatin foci in NIH 3T3 cells (arrows in Fig. S1). This finding points to the partial exclusion of remodelers from these sites, consistent with our previous investigations by high-resolution fluorescence microscopy of fixed U2OS cells (25).

Fig. 1.

Confocal images of Snf2H-GFP, GFP-Snf2L, and GFP-Snf2L+13 in U2OS cells. For each protein G1/2 phase (Top), S phase (Middle), and the case of DNA damage (Bottom) are shown. Immobilization at replication foci in S phase and at DNA repair sites is evaluated from the decay of the CP curves depicted in red compared with the black curves measured at regions of low PCNA density.

Stable U2OS lines expressing one GFP-tagged remodeler variant were used to measure the expression level of the endogenous remodelers. The concentrations of GFP-tagged remodelers were determined by FCS experiments (Table S1). Subsequently, quantitative Western blotting was used to measure the ratio of endogenous remodelers relative to the GFP-tagged ones (Fig. S2). The expression levels of Snf2H-GFP and GFP-Snf2L+13 were about twofold higher than those of the corresponding endogenous proteins, whereas endogenous Snf2L was not detectable in the U2OS cell line. From the combined FCS/Western blot analysis endogenous protein concentrations of 0.83 ± 0.13 μM (Snf2H), >0.15 ± 0.03 μM (hACF complex formed by Snf2H and Acf1), <0.05 μM (Snf2L), and 0.14 ± 0.04 μM (Snf2L+13) were determined.

The remodeling activity of the GFP-tagged proteins was evaluated in in vitro experiments (Fig. S3). As expected, both Snf2H-GFP and GFP-Snf2L were capable of translocating nucleosomes whereas Snf2L+13 lacked any activity. In comparison with freely mobile recombinant His-tagged Snf2H and Snf2L, affinity-purified GFP-tagged remodelers tethered to agarose beads showed a somewhat lower translocation activity (Fig. S3 B and C). Immobilizing His-tagged recombinant proteins reduced their activity to similar levels, indicating that the GFP tag did not impair the catalytic function.

Snf2H-GFP and GFP-Snf2L Are Recruited to Replication Foci and DNA Repair Sites.

Because Snf2H and Acf1 have been shown to play a role in replication (23), we studied the localization of Snf2H-GFP, GFP-Snf2L, and inactive GFP-Snf2L+13 in the S phase of the cell cycle (Fig. 1). Both Snf2H-GFP and GFP-Snf2L were enriched at replication foci as detected by colocalization with PCNA-RFP. Moreover, 40–70% fractions of the remodelers were immobilized at these sites as determined from CP curves acquired at the replication foci. Notably, no difference between active Snf2L and inactive Snf2L+13 was apparent. Besides replication, Snf2H has been shown to be involved in the DNA damage response via the Snf2H-WSTF complex (26). To investigate this process DNA damage was locally induced by microirradiation with a UV laser. Subsequent recruitment of PCNA-RFP and Snf2H-GFP to the damage site was observed. Interestingly, GFP-Snf2L and GFP-Snf2L+13 also accumulated at the damage site, suggesting a potential function of Snf2L in the DNA damage response. In contrast, no accumulation was found for GFP alone (Fig. S4). Bleaching experiments demonstrated that Snf2H-GFP and GFP-Snf2L as well as GFP-Snf2L+13 were immobilized at the repair sites (Fig. 1). The accumulation of inactive Snf2L+13 at replication foci and repair sites demonstrates that the recruitment occurs independent of ATPase activity. In summary, Snf2H-GFP reproduced the localization behavior reported previously for endogenous Snf2H (23, 26). In addition, Snf2H-GFP with GFP at the C terminus and N-terminally tagged RFP-Snf2H showed no differences regarding intracellular localization or mobility (Fig. S5). Thus, with respect to the experiments conducted here no influence of the autofluorescent protein domain on the mobility and interactions of the remodeler was detected.

Multimeric Snf2H and Acf1 Associate into a Heterotetramer or Higher-Order Complex.

Acf1 is one of the noncatalytic subunits that can associate with the Snf2H/ISWI motor protein in a complex termed ACF (27, 28). To probe multimerization of Snf2H and Acf1 in living U2OS cells, their complexes containing proteins labeled in two different colors were evaluated by fluorescence cross-correlation spectroscopy (21, 29, 30). The FCCS analysis of cells transfected with either Snf2H-GFP and RFP-Snf2H or Acf1-GFP and Acf1-RFP revealed a high cross-correlation signal, which was indicative of the formation of Snf2H and Acf1 homodimers or higher-order complexes (Fig. 2, Table 1). From the amplitudes of the auto- and cross-correlation functions, the fraction of the total amount of protein incorporated into these complexes was calculated to be at least 82 ± 8% for Snf2H and 94 ± 6% for Acf1. In addition, the interaction of Acf1-RFP and Snf2H-GFP was probed by transfecting Acf1-RFP into U2OS cells stably expressing Snf2H-GFP. The normalized degree of cross-correlation was r_x = 0.22 ± 0.04 (Fig. 2, Table 1), which shows that both proteins interact. The simplest model to explain the FCCS results is the association of Snf2H and Acf1 homodimers into a heterotetrameric ACF complex (larger complexes would also be compatible with the data). For an ACF heterotetramer the fraction of the total cellular Snf2H pool in this complex is >18 ± 4%. Considering the contribution of unlabeled endogenous protein would increase this value.

Fig. 2.

Interactions of Snf2H and Acf1 probed by FCCS in the U2OS cell nucleus. The value of the ratio r_x (Table 1) calculated from the amplitudes of the cross-correlation and autocorrelation functions reveals the presence of complexes containing Snf2H-GFP and Acf1-RFP (A), Acf1-GFP and Acf1-RFP (B), and Snf2H-GFP and RFP-Snf2H (C). Auto- (AC) and cross-correlation (XC) functions of the two color channels are shown. (D) Control measurements in U2OS cells expressing a GFP–RFP fusion protein (XC positive) or GFP and RFP alone (XC negative).

Table 1.

FCCS measurements of Snf2H and Acf1 interactis in the U2OS cell nucleus

Complex	Dnuc (μm2·s−1)	Ratio G	rx	Complex, %
Snf2H-GFP·RFP-Snf2H	3.4 ± 0.6	0.35 ± 0.02	0.29 ± 0.02	>82 ± 8
Snf2H-GFP·Acf1-RFP	3.8 ± 1.5	0.29 ± 0.04	0.22 ± 0.04	>18 ± 4
Acf1-GFP·Acf1-RFP	2.9 ± 0.9	0.37 ± 0.03	0.32 ± 0.04	>94 ± 6

D_nuc is the diffusion coefficient of the complex determined from the cross-correlation analysis. The degree of cross-correlation r_x reflects the amount of complexes formed that carry both a GFP and an RFP label. For details on the FCCS analysis, see SI Materials and Methods.

Snf2H-GFP, GFP-Snf2L, and Acf1-GFP Are Only Transiently Associated with Chromatin in G1/2 Phase.

The dynamics of Snf2H, Snf2L, and Snf2L+13 in living U2OS cells during G1/2 and S phase were characterized by a combination of FRAP and FCS experiments (Fig. 3, Tables 2 and 3). For all proteins the FRAP recovery curves could be fitted well with a diffusion model, in which the contribution of transient binding is included into an effective diffusion coefficient (19). Recovery in the FRAP experiments during G1 phase was almost complete with immobile fractions of only 1–4% (Table 2). During S phase the immobile fraction increased as observed in FRAP and CP experiments (Table 2, Fig. 1). The FRAP value of ∼10% was lower than that of 40–70% estimated from the CP analysis because the bleached area in the FRAP experiments contained both replication foci and regions of low PCNA density. From an analysis of the FRAP data according to a reaction–diffusion model the lower limit of the dissociation rate in G1/2 phase was determined to be k_off > 14 s⁻¹ (t_res = 70 ms) for Snf2H-GFP and k_off > 8 s⁻¹ (t_res = 130 ms) for GFP-Snf2L (SI Materials and Methods). An upper limit for k_off could not be determined reliably in the FRAP analysis because the average residence time in the chromatin-bound state was too short. This question was addressed by FCS experiments that have a much better time resolution and yield the diffusion time τ in the nucleus as a direct readout from the intensity fluctuations in the focal volume. In addition, the remodelers’ mobility can be measured also in the cytosol by FCS where both Snf2H and Snf2L are freely mobile and present at nanomolar concentration with D = 13 μm²·s⁻¹ or τ = ∼0.6 ms (Table 3). Corresponding values in the nucleus were 2.0 ± 0.3 ms (Snf2H-GFP) and 1.2 ± 0.2 ms (GFP-Snf2L) for 70–80% of the molecules (Table 3). Furthermore, an additional slow population with D_slow = 0.03–0.10 μm²·s⁻¹ (τ = 100–600 ms) was present in the nucleus, which reflects chromatin translocations that manifest themselves as slow intensity fluctuations of bound fluorescently labeled proteins. This explanation is inferred from the value of the anomaly parameter α > 1, which is indicative of a confined diffusion behavior in the FCS analysis, and the observation that this contribution is a general feature of proteins that interact transiently with chromatin like heterochromatin protein 1 (19). This interpretation is corroborated by the absence of a corresponding mobility fraction in the FRAP experiments for Snf2H/Snf2L, where a <150 ms residence time and D values of 0.9 ± 0.1 and 1.0 ± 0.1 μm²·s⁻¹ were measured (Table 2). Taken together, the FCS data clearly point to an average residence time of a few milliseconds for the vast majority of the remodeler fraction detectable by FCS (Table 3). Because immobilized molecules are typically bleached after 1 s, FCS provides no information about larger residence times. However, no remodelers bound on the timescale of seconds were found in the FRAP analysis and the remodeler fraction immobilized on the timescale of ≥1 min comprised only 1–4%. Thus, the short millisecond residence time applies for almost the whole remodeler pool during G1/2 phase.

Fig. 3.

Mobility of Snf2H-GFP, GFP-Snf2L, GFP-Snf2L+13, and Acf1-GFP. All proteins displayed a high mobility and only transient binding to chromatin during G1/2 phase (Tables 2 and 3) but an increased immobilized fraction was observed for Snf2H and Snf2L during S phase in U2OS cells. A–D show averaged FRAP curves, and E–H display single FCS experiments for the indicated proteins.

Table 2.

FRAP measurements in U2OS cells

	Deff, μm2·s−1			Immobile, %
Protein	G1/2	S	koff, s−1 G1/2	G1/2	S
Snf2H-GFP	0.9 ± 0.1	0.8 ± 0.1	>13.6 ± 2.3	4 ± 1	9 ± 2
Snf2H-GFP, no ATP	1.2 ± 0.1	0.8 ± 0.1	>12.3 ± 2.3	1 ± 0	9 ± 2
GFP-Snf2L	1.0 ± 0.1	0.6 ± 0.1	>8.3 ± 1.5	3 ± 1	9 ± 1
GFP-Snf2L, no ATP	1.2 ± 0.2	0.6 ± 0.2	>3.3 ± 0.4	2 ± 1	12 ± 3
GFP-Snf2L+13	1.6 ± 0.1	1.5 ± 0.7	>2.0 ± 0.4	1 ± 1	n.d.
GFP-Snf2L+13, no ATP	1.4 ± 0.2	n.d.	>7.7 ± 1.5	1 ± 0	n.d.
Acf1-GFP	0.9 ± 0.3	n.d.	>2.3 ± 0.8	3 ± 2	n.d.
Acf1-GFP, no ATP	0.7 ± 0.1	n.d.	>2.3 ± 0.5	0	n.d.

Measurements in S phase represent an average over several replication foci and regions of low PCNA density within the bleached region. n.d., not determined.

Table 3.

FCS measurements in U2OS cells

Protein	Dcyt, μm2·s−1	Dnuc, μm2·s−1	fslow, %	Dslow, μm2·s−1	αslow	K*bin
Snf2H-GFP	13.0 ± 1.0	3.9 ± 0.4	22 ± 3	0.03 ± 0.01	2.1 ± 0.2	2.3 ± 0.7
GFP-Snf2L	13.3 ± 0.8	6.6 ± 0.4	29 ± 3	0.11 ± 0.01	1.5 ± 0.1	1.0 ± 0.3
GFP-Snf2L+13	13.0 ± 1.4	5.2 ± 0.4	25 ± 4	0.10 ± 0.01	1.3 ± 0.2	1.5 ± 0.5
Acf1-GFP	10.4 ± 0.5	3.5 ± 0.5	33 ± 5	0.07 ± 0.02	1.4 ± 0.3	2.0 ± 0.6

From the comparison of FCS measurements of diffusion coefficients in the cytosol (D_cyt) and in the nucleus (D_nuc) the pseudoequilibrium binding constant K*_bin (K_bin multiplied with the binding site concentration) was determined. The larger value for Snf2H-GFP compared with GFP-Snf2L reflects either a higher binding affinity to chromatin or the presence of more binding sites than for GFP-Snf2L. The slow mobility fractions f_slow displayed a behavior indicative of a slowly, confinedly moving lattice that is characterized by α > 1 in the FCS analysis.

To compare the mobility of the remodeler ATPases to the mobility of a noncatalytic complex subunit, U2OS cells were transiently transfected with Acf1-GFP. The FRAP and FCS analysis yielded effective diffusion coefficients and immobile fractions that were very similar to that of Snf2H-GFP (Fig. 3, Tables 2 and 3). These results indicate that there are no large differences between the dynamics and the interactions of Snf2H-GFP and Acf1-GFP.

Lack of ATPase Activity Leads to a Small Increase in Snf2H and Snf2L Mobility.

To identify the contribution of the remodeling reaction to the mobility of Snf2H and Snf2L, the naturally occurring inactive splice variant Snf2L+13 was included in the analysis. In Snf2L+13 the exon 13 is spliced into the ATPase cassette of Snf2L, resulting in a noncatalytic protein. Because Snf2L+13 has been found in the same complexes as Snf2L, it could function as a dominant negative variant (24). Whereas the overall dynamic behavior of Snf2L+13 was similar to that of Snf2L and Snf2H, a small but significant increase in the mobility was apparent. The immobile fraction (1 ± 1%) was virtually absent for Snf2L+13 and a larger effective diffusion coefficient of 1.6 ± 0.1 μm²·s⁻¹ was determined as compared with Snf2L with 3 ± 1% immobile protein and D_eff = 1.0 ± 0.1 μm²·s⁻¹ (Fig. 3 B and C, Table 2). Similar results were obtained with transiently transfected HEK 293T cells (Fig. S6), which, in contrast to U2OS cells, express active Snf2L endogenously (31). Here, values of 2 ± 1% and 2.0 ± 0.4 μm²·s⁻¹ (Snf2L) in comparison with 1 ± 1% and 2.7 ± 0.4 μm²·s⁻¹ (Snf2L+13) for the effective diffusion coefficient and the immobile fraction were measured.

To further investigate differences between the active and the inactive state of Snf2L and Snf2H, U2OS cells were depleted of ATP by addition of sodium azide (Fig. 3 A–D). Because the remodeling reaction is ATP dependent, no remodeling is expected to occur under these conditions. The mobility of Snf2H-GFP and GFP-Snf2L was measured using FRAP and FCS as described above. The results were the same as those obtained in the comparison of Snf2L and Snf2L+13: A small reduction by 1–3% in the immobile fraction and a slight change of the effective diffusion coefficient toward a higher mobility in the ATP-depleted cells (Fig. 3 A–D, Table 2). Thus, the mobility differences between the active and the inactive (Snf2L+13, ATP-depleted Snf2H and Snf2L) case were significant but small. This result suggests that only a fraction of a few percent of the remodelers is actively involved in chromatin remodeling during G1/2 phase and displays an increased residence time in the chromatin-bound state.

Discussion

The translocation of nucleosomes by chromatin remodeling complexes was investigated previously in numerous in vitro studies (e.g., refs. 6–8, 13, 16, 17, 27, 28, 32, and references therein). However, details on their mode of operation in living cells are missing. Here, we analyzed the dynamics and interaction behavior of Snf2H, Snf2L, and Snf2L+13 as well as Acf1 in living cells with a multiscale approach that combined FRAP, CP, and FC(C)S experiments (18–21). In the case of FRAP, the shape of the recovery curve is used to extract dynamic parameters such as the diffusion coefficient and the kinetic rate constants. Furthermore, immobilized proteins are identified reliably from an incomplete recovery. CP has the advantage that immobilized protein fractions at loci as small as the optical resolution can be evaluated. FCS on the other hand yields information on the mobile protein fraction with high spatial and temporal resolution that is complementary to FRAP and CP.

The FRAP, CP, and FCS experiments revealed that all proteins studied here were highly dynamic in the nucleus during G1/2 phase and bound only transiently to chromatin with upper limits for the residence times of <150 ms for Snf2H/L and <500 ms for Snf2L+13 and Acf1. These residence times are at the low end of values reported for other chromatin interacting proteins but similar to values for some transcription factors (18, 33). From a comparison of Snf2L and the inactive Snf2L+13 splice variant in conjunction with ATP depletion experiments we infer that only a small fraction of a few percent of a given ISWI type ATPase is involved in actively translocating nucleosomes during G1/2 phase. However, in replication foci during S phase or at DNA repair sites the fraction of tightly bound remodelers with residence times of at least several seconds is dramatically increased up to 40–70%. Because an increase of immobile Snf2L+13 was observed, too, binding to these sites occurs also independently of ATPase activity. This observation suggests a mechanism in which increasing the binding affinity of a given remodeler to its nucleosome substrate promotes nucleosome translocation as proposed previously on the basis of in vitro studies (6).

In the FCCS experiments the mutual homo- and heteromeric interactions of Snf2H and Acf1 were evaluated and demonstrated the simultaneous presence of Snf2H–Snf2H, Acf1–Acf1, and Snf2H–Acf1 interactions. Previous cryoelectron microscopy and FCCS in vitro work reported that dACF/hCHRAC is a heterotetramer composed of two Snf2H/ISWI and two Acf1 subunits (30, 34). Together with our FCCS results it is concluded that in the human cell nucleus hACF/hCHRAC heterotetramer complexes are formed via the equilibrium association of Snf2H and Acf1 dimers. The quantitative analysis of the FCCS data in terms of a homodimer–heterotetramer equilibrium results in a fraction of >18 ± 4% of the Snf2H pool being associated with Acf1. This value is the lower limit that does not consider the presence of unlabeled endogenous Snf2H and Acf1, which reduce the cross-correlation signal. Due to the presence of other Snf2H-containing complexes like NoRC or WSTF not the whole Snf2H pool is present in Acf1-containing complexes (8, 9, 26). Because the monomeric Snf2H fraction was determined to equal <18% (Table 1), these Snf2H-containing complexes are likely to comprise Snf2H multimers as well.

Using a combination of FCS and quantitative Western blots we were able to measure endogenous remodeler concentrations of 0.83 ± 0.13 μM (Snf2H), >0.15 ± 0.03 μM (hACF complex formed by Snf2H and Acf1), <0.05 μM (Snf2L), and 0.14 ± 0.04 μM (Snf2L+13) in the human U2OS cell line. To our knowledge these quantifications of remodeler concentrations in vertebrates/higher eukaryotes have not been described elsewhere. They point to a distinct pattern of different remodeling activities. Whereas the cell line-specific Snf2L protein is not detected in U2OS cells, its inactive splice variant was found at concentrations of 140 nM in the nucleus. In comparison, a specific type of active chromatin remodeler complex like hACF would be typically present at a concentration >100 nM. Extrapolating this value to the total concentration of all remodeling complexes and including complexes of the other ATPase families (Snf2, Mi-2, Chd1, Ino80, ERCC6, ALC1, CHD7, Swr1, RAD54, and Lsh), the total chromatin remodeler concentration in the human cell nucleus is estimated to be in the 10 μM range with a nucleosome concentration of 140 μM (35). Thus, the resulting remodeler/nucleosome ratio is similar to that of yeast with one remodeler per ∼10 nucleosomes (36, 37). Because of their presence in high concentrations and the capability of complexes like ACF to bind >1 nucleosome, it has been proposed that ISWI chromatin remodelers might represent a stable structural component of chromatin (38). In the light of the high mobility of these complexes and the absence of a significant immobile fraction during G1/2 phase, such an architectural function during the whole cell cycle seems unlikely. Furthermore, several other conclusions on the mechanism by which chromatin remodeling complexes operate in living cells can be made (Fig. 4). The very short residence time in the chromatin-bound state during G1/2 phase reveals either that the remodeling reaction is too fast to be resolved or that only a very small remodeler fraction (<5%) is active at a given point of time. From in vitro data obtained with SWI/SNF remodelers, the velocity of chromatin remodeling can be estimated to equal ∼13 bp·s⁻¹ (39). Ensemble FRET studies with purified mononucleosomes and Snf2H resulted in even slower rates of 17 bp in a few seconds (40). Thus, the residence times measured here would be compatible with no or only a very small translocation of a few base pairs for the vast majority of the remodeling complexes. It is noted that such a small step size would be in contradiction to the experimentally observed remodeling behavior of ISWI remodelers, for which translocation steps of ∼10–50 bp were reported (30). In addition, the ACF complex has been shown to be committed to its substrate for several minutes in vitro, further supporting the notion that the remodeling reaction does not occur within <150 ms (41, 42). Thus, we conclude that the 96–97% mobile remodeler fraction found here is not engaged in chromatin remodeling under “housekeeping” conditions in G1/2 phase. In contrast, the remodeling activity is high at replication foci during S phase or at DNA repair sites, and a large remodeler fraction immobilized for seconds to minutes was detected. Accordingly, active remodeling is associated with binding times at least on the second timescale. These binding times include contributions from the ATPase independent interactions with recruiting protein factors and the nucleosome translocation reaction itself. At these active sites, immobilized ISWI remodelers could also play a role for chromatin architecture.

Fig. 4.

ISWI chromatin remodeler interactions with chromatin. (A) During G1/2 phase chromatin remodelers bind only transiently to nucleosomes with residence times in the 10-ms regime and operate via a “continuous sampling mechanism.” Most encounters are not productive and efficient nucleosome translocation requires additional signals that increase the binding affinity. These signals are present at only a small subset of nucleosomes during housekeeping conditions. (B) In S phase at replication foci or at sites of DNA repair remodeling “hotspots” exist. These hotspots are formed by increasing the binding of remodelers to the respective genomic loci to residence times in the range of seconds to minutes. Identification of a remodeling hotspot could be mediated by histone modifications, chromatin-associated proteins, or structural features of the chromatin environment.

In summary we propose a model, in which the chromatin remodelers continuously sample a large number of nucleosomes by transiently binding and dissociating without nucleosome translocation in G1/2 phase (Fig. 4A). In the default state the nucleosomes would be stably bound to their positions, as opposed to a “fluid” chromatin model, in which constant movements of the majority of nucleosomes would render most of the DNA accessible to interacting protein factors. Only upon introducing additional signals like, for example, core histone modifications at certain nucleosomes, the phosphorylation of the H2A.X variant as a DNA damage signal, or decoration with interacting proteins as during S phase or DNA repair (Fig. 4B), would these nucleosomes be marked “to be translocated” by making them high-affinity substrates. This view is supported by the results of several reports that demonstrate a connection between remodeling activity and histone modifications as reviewed in ref. 7. From the parameters measured here for the isolated Snf2H ATPase and its complex with Acf1 (endogenous protein concentration 0.15–0.83 μM, average residence time 75–430 ms) we estimate that nucleosomes are sampled at a rate of 0.1–3 min⁻¹ by a specific ISWI-type complex. To probe 99% of all genomic nucleosomes, average sampling times of ∼80 s (all Snf2H containing remodelers) and ∼45 min (ACF) are calculated with the values from the FRAP analysis (SI Materials and Methods and Table S2). These results have to be regarded as upper limits because the average residence times in the order of 10 ms from the FCS analysis would decrease the corresponding sampling times to 11–64 s. Thus, after setting an appropriate signal a nucleosome can be translocated within tens of seconds or minutes to its new position. For this mechanism a relatively large concentration of remodelers is required to sample the ensemble of all nucleosomes in the genome at an acceptable speed. If the cell contained a 100-fold lower concentration of ISWI-type remodelers, i.e., a concentration of a few nanomolar, the sampling time would increase to hours. This would introduce a significant lag time for the readout of external signals on the timescale of cellular transitions. From this point of view a large pool of remodelers with only a small fraction of simultaneously active complexes seems mandatory to respond quickly to changes of the chromatin state. It is noted that these features of the continuous sampling mechanism are in excellent agreement with findings on cyclically occurring epigenetic processes that initiate and terminate remodeling of chromatin by SWI/SNF and NuRD complexes (33, 43, 44). In contrast, a large fraction of continuously active remodelers would constantly expose all parts of the DNA sequence to the surrounding nucleoplasm at much higher energy costs, because the remodelers’ ATPase activity is stimulated by the bound nucleosome substrate (32). Thus, our findings point to a tight regulation of chromatin accessibility, realized by nucleosomes as switches that are continuously sampled by chromatin remodelers and triggered upon distinct signals.

Materials and Methods

Experiments were conducted with GFP and RFP constructs of human Snf2H, Snf2L, Snf2L+13, and Acf1 in human U2OS osteosarcoma, human embryonic kidney HEK293T, and mouse NIH 3T3 cell lines as described in SI Materials and Methods. For Snf2H-GFP, GFP-Snf2L, and GFP-Snf2L+13 stably transfected U2OS clones were derived. Other constructs were introduced via transient transfection. FRAP, CP, and F(C)CS experiments were conducted as described previously (19) and in SI Materials and Methods. At least 20 cells were measured for a given experimental condition. For FRAP the time evolution of the intensity integrated over the bleach spot was fitted to a diffusion model, to a binding model, or to a reaction–diffusion model that incorporates both diffusion and binding processes. In the CP experiments the decay of the fluorescence signal due to the dynamic equilibrium of photobleaching, diffusion, and chromatin dissociation/association of GFP-tagged proteins was used to measure the immobilized fraction with the resolution given by the confocal excitation volume. The FCS data were fitted to a one- or two-component anomalous diffusion model, which is characterized by a nonlinear time dependency of the mean squared particle displacement given by the anomaly parameter α.