1. Chromosomes and Gene Expression
  2. Structural Biology and Molecular Biophysics
Download icon

The nucleosomal acidic patch relieves auto-inhibition by the ISWI remodeler SNF2h

  1. Nathan Gamarra
  2. Stephanie L Johnson
  3. Michael J Trnka
  4. Alma L Burlingame
  5. Geeta J Narlikar  Is a corresponding author
  1. University of California, San Francisco, United States
Research Article
  • Cited 12
  • Views 2,265
  • Annotations
Cite this article as: eLife 2018;7:e35322 doi: 10.7554/eLife.35322

Abstract

ISWI family chromatin remodeling motors use sophisticated autoinhibition mechanisms to control nucleosome sliding. Yet how the different autoinhibitory domains are regulated is not well understood. Here we show that an acidic patch formed by histones H2A and H2B of the nucleosome relieves the autoinhibition imposed by the AutoN and the NegC regions of the human ISWI remodeler SNF2h. Further, by single molecule FRET we show that the acidic patch helps control the distance travelled per translocation event. We propose a model in which the acidic patch activates SNF2h by providing a landing pad for the NegC and AutoN auto-inhibitory domains. Interestingly, the INO80 complex is also strongly dependent on the acidic patch for nucleosome sliding, indicating that this substrate feature can regulate remodeling enzymes with substantially different mechanisms. We therefore hypothesize that regulating access to the acidic patch of the nucleosome plays a key role in coordinating the activities of different remodelers in the cell.

https://doi.org/10.7554/eLife.35322.001

eLife digest

Every human cell contains nearly two meters of DNA, which is carefully packaged to form a dense structure known as chromatin. The building block of chromatin is the nucleosome, a unit composed of a short section of DNA tightly wound up around a spool-like core of proteins called histones.

The tight structure of the nucleosome prevents the cell from accessing and ‘reading’ the genes in the packaged DNA, effectively switching off these genes. So the exact placement of nucleosomes helps manage which genes are turned on. Changing the position of the nucleosomes can ‘free’ the DNA and make genes available to the cell. Enzymes called chromatin remodelers move nucleosomes around – for example, they can make the histone core slide on the DNA strand. However, it is still unclear how these enzymes recognize nucleosomes. Previous research indicates that many proteins bind to nucleosomes by using a surface on the histone proteins called the acidic patch. Could chromatin remodelers also work by interacting with this acidic patch?

To address this further, Gamarra et al. investigate how a chromatin remodeler enzyme known as SNF2h interacts with a nucleosome. By default, SNF2h is inactive because two of its regions called AutoN and NegC act as brakes. The experiments show that the acidic patch helps to bypass this inactivation and switches on SNF2h. Gamarra et al. propose that, when SNF2h docks on to the nucleosome, the patch provides a landing pad for the AutoN and NegC modules; this interaction activates the enzyme, which can then start remodeling the nucleosome. However, another type of chromatin remodeler also uses the patch to interact with nucleosomes but it does not have the AutoN and NegC regions. This suggests that chromatin remodelers work with the acidic patch in different ways. Overall, the findings deepen our understanding of how DNA is packaged in cells, and how this process may go wrong and cause disease.

https://doi.org/10.7554/eLife.35322.002

Introduction

Eukaryotic genomes are packaged into chromatin, enabling large amounts of DNA to fit into the spatial constraints of the nucleus. This packaging has long been appreciated as a passive barrier to DNA access by nuclear factors. The discovery that chromatin regulators play critical roles in virtually all nuclear processes has informed a more nuanced view of chromatin as a dynamic regulatory platform that coordinates access to the genetic material. The smallest unit of chromatin is the nucleosome, a DNA-protein complex composed of ~150 bp of DNA wrapped around an octamer of histone proteins (Luger et al., 1997). Nucleosomes can further interact with each other and with other factors to form higher-order structures (Luger et al., 2012). Consequently, cells have evolved several sophisticated strategies to regulate chromatin structure at the nucleosome level. These include the covalent modification of histone proteins and DNA, as well as non-covalent changes to the position or composition of nucleosomes at specific genomic loci. Many of the non-covalent transformations, ranging from sliding nucleosomes to the complete disassembly of the histone octamer, are catalyzed by ATP-dependent chromatin remodeling enzymes (Zhou et al., 2016). Underscoring their central role in chromatin regulation, remodeling enzymes play essential roles in many processes including transcription, DNA replication, and DNA repair (Falbo and Shen, 2006; Hota and Bruneau, 2016; Price and D'Andrea, 2013). How a relatively small number of remodeler types carry out such diverse regulatory functions remains an area of active research, not least because much remains unknown regarding remodeler mechanisms for substrate recognition and the coupling of that recognition to activity.

Chromatin remodelers are members of the SF2 superfamily of nucleic acid motors, which catalyze noncovalent changes to nucleic acid substrates (Zhou et al., 2016). Chromatin remodelers, however, are unique in that they specifically mobilize DNA in the context of the nucleosome, where DNA is tightly bound to histone proteins. Remodelers are further classified into families based on the domain architecture of their ATPase subunit. These families differ in their specific biochemical activities (Zhou et al., 2016). Substantial progress in our general understanding of remodeling mechanisms has been made by asking what elements of the nucleosome are important for remodeling in different families. For example, maximal remodeling by ISWI family remodelers, which primarily slide nucleosomes, requires DNA flanking the nucleosome and the N-terminal tail of histone H4 (Clapier et al., 2001; Yang et al., 2006). Conversely, maximal remodeling by SWI/SNF family remodelers, which carry out the most diverse set of changes to nucleosome structure, does not require these nucleosomal epitopes (Guyon et al., 1999). Less is known about how these substrate cues are recognized and mechanistically coupled to remodeling. Some important insights have come from biochemical analyses and structures of remodelers in the absence of nucleosomes, which suggest that in the ground state, chromatin remodelers are held in an inactive conformation by family-specific autoinhibitory motifs (Clapier and Cairns, 2012; Hauk et al., 2010; Xia et al., 2016; Yan et al., 2016). Binding to specific nucleosomal epitopes is thought to relieve this autoinhibition via conformational changes in the remodeler, but the details of this process remain unclear.

At the same time, structures of several nucleosome-protein complexes are revealing that many of these proteins interact with a conserved acidic patch formed by histones H2A and H2B on the top surface of the nucleosome (McGinty and Tan, 2016). These proteins interact with the acidic patch using an ‘arginine anchor’ which nestles into a pocket formed by the α1–2 helices of histone H2A (McGinty and Tan, 2016). It is therefore plausible that chromatin remodelers also recognize the acidic patch. Indeed recent work has indicated that mutating the acidic patch reduces the activity of ISWI, SWI/SNF and some CHD family remodeling enzymes (Dann et al., 2017). To investigate the mechanistic role of the acidic patch in nucleosome remodeling by the ISWI family of enzymes we used a combination of ensemble and single molecule methods in the context of the human ISWI enzyme, SNF2h. We observe that interactions with the acidic patch activate SNF2h by relieving auto-inhibition mediated by two conserved domains of ISWI enzymes. Our results further suggest that contacts with the acidic patch helps control the distance the nucleosome is moved during translocation events. Finally we find that the acidic patch also stimulates the activity of the INO80 complex. Together, these results highlight the broad and essential role the acidic patch plays in chromatin regulation.

Results

The acidic patch of the nucleosome is important for nucleosome sliding by SNF2h

The human ISWI ATPase, SNF2h, preferentially slides nucleosomes toward longer flanking DNA (Yang et al., 2006). As a consequence of this activity, SNF2h slides mononucleosomes towards the center of a DNA strand, an activity that can be detected by a native gel mobility assay (Yang et al., 2006). Recent work has shown that mutating residues in the nucleosome acidic patch reduces the ability of SNF2h to expose nucleosomal DNA to restriction enzymes (Dann et al., 2017). To investigate the mechanism by which such mutations affect the nucleosome sliding activity of SNF2h and to determine if these mutations affected SNF2h binding or catalysis, we first used a mutant H2A in which four key residues in the H2A acidic patch are replaced with alanines (E61A, E64A, D90A, and D92A) (Figure 1A). We call nucleosomes in which all four of these H2A residues have been mutated to alanines ‘APM’ (acidic patch mutant) nucleosomes. Using a native gel mobility assay, we measured remodeling rates of wild type (WT) and APM mononucleosomes containing 60 bp of DNA flanking one end of the nucleosome (0/60) (Figure 1B). APM nucleosomes are centered substantially slower than WT nucleosomes under conditions where SNF2h is in excess of nucleosomes (Figure 1B and C). Since the acidic patch has been shown to be critical for nucleosome binding by several chromatin proteins, we expected the apparent Km, Kmapp, to be increased with APM nucleosomes. Interestingly however, Kmapp is affected substantially less by mutation of the acidic patch than the maximal rate constant for remodeling, kmax (Figure 1D, 61 nM to 280 nM, corresponding to ~4.5 fold increased Kmapp, versus 2.5 min−1 to 0.012 min−1, corresponding to ~200 fold reduced kmax). These results indicate that the acidic patch plays a larger role in regulating maximal nucleosome sliding activity than nucleosome binding. We also measured ATP hydrolysis under conditions where nucleosomes are in excess of SNF2h. At saturating nucleosome concentrations, APM nucleosomes stimulate ATP hydrolysis 4-fold less than WT (Figure 1G, Figure 1—figure supplement 2). Together, these results imply that the acidic patch plays a critical role in coupling ATP hydrolysis to remodeling after the binding step.

Figure 1 with 3 supplements see all
The acidic patch is an important epitope for remodeling post-binding.

(A) Charge profile of the nucleosome (left) and magnification of the acidic patch region (right) (PDBID: 1KX5, charge profile generated using ABPS and UCSF Chimera [Pettersen et al., 2004]). Residues of the acidic patch mutated in this study are shown in red. (B) Example of a Cy5-fluorescent scan of a gel-based remodeling assay with 0/60 WT and acidic patch mutant (APM) nucleosomes (1 µM SNF2h, 20 nM nucleosomes, saturating ATP). (C) Quantification of the gel in B and fits to a single exponential decay. Inset is zoomed to show faster time points with WT nucleosomes. (D) Gel remodeling rate as a function of enzyme concentration, plotted on a log(2) scale. APM nucleosomes are remodeled substantially slower even at saturating concentrations of enzyme. Data were fit to a cooperative binding model (WT nucleosomes: KMapp = 61 nM, kmax = 2.5 min−1, h = 1.5; APM nucleosomes: KMapp = 280 nM, kmax = 0.012 min−1, h = 1.2). (E) Remodeling is inhibited by competition for the acidic patch. WT nucleosomes were remodeled with sub-saturating SNF2h and in the presence of KSHV LANA peptide. Remodeling is inhibited by the peptide but not when the arginine anchor is mutated to alanine. (F) Inhibition curve with the LANA peptide (KI = 1.21 µM). Error bars represent standard errors on the mean for three replicates, except for the no-peptide condition in F, which had two replicates. (G) ATPase activity of WT SNF2h. Nucleosomes were in excess of SNF2h and at saturating concentrations. APM nucleosomes stimulated ATPase activity 4-fold weaker than WT.

https://doi.org/10.7554/eLife.35322.003

To further investigate the importance of the acidic patch, we asked whether binding by another factor to the acidic patch could compete for nucleosome sliding by SNF2h. We used a peptide derived from the latency associated nuclear antigen (LANA) from Kaposi’s sarcoma-associated herpesvirus that has previously been shown to interact directly with the acidic patch via an arginine anchor (Barbera et al., 2006). We carried out these experiments using sub-saturating concentrations of SNF2h. Under these conditions most of the nucleosomes are unbound, allowing direct competition between the LANA peptide and SNF2h for the acidic patch. The presence of the LANA peptide dramatically slows remodeling by SNF2h in a dose-dependent manner (~200 fold reduction in rate constant at 50 µM LANA peptide), while a peptide with a mutant arginine anchor does not have a detectable inhibitory effect (Figure 1E and F). We measured a KI of 1.2 µM for inhibition by the LANA peptide, within 5-fold of its published affinity for the nucleosome (Figure 1F) (Fang et al., 2016). These results indicate that remodeling by SNF2h requires an interaction with the acidic patch that is mutually exclusive with the binding of acidic patch interacting factors such as the LANA peptide. The results are also consistent with recent work showing that the LANA peptide can inhibit the restriction enzyme accessibility activity of the SNF2h containing complex, ACF (Dann et al., 2017).

We next investigated if the acidic patch residues act independently or cooperatively by measuring the effects of individual mutations. Interestingly, except for residue E64, all the single alanine mutants have comparable defects as the four point mutants combined (Figure 1—figure supplement 3). This observation suggests that three of the four acidic patch residues tested here act cooperatively to promote sliding by SNF2h.

The AutoN and NegC regions of SNF2h cooperate with the acidic patch to enable maximal remodeling

We next investigated which regions of SNF2h might functionally interact with the acidic patch. In principle, these would include (i) regions that directly contact the acidic patch and (ii) regions that do not directly contact the acidic patch, but whose function is energetically coupled to the presence of the acidic patch.

To investigate regions that may directly contact the acidic patch we carried out cross-linking mass spectrometry using the zero-length, carbodiimide based reagent EDC. This method catalyzes the formation of new amide bonds between protein carboxylates, such as the side chains of aspartate and glutamate residues, and amino groups and is therefore well suited to probing electrostatic interactors of the acidic patch. Hundreds of high confidence cross-linked residue pairs were identified using this approach (Supplementary file 1). To focus on mechanistically meaningful domain-domain interactions, we employed a semi-quantitative mass spectrometry method to compare the extent of SNF2h-nucleosome cross-linking in the presence of different nucleotides (Figure 2B). In previous work we have shown that ADP•BeFx mimics an activated state of the SNF2h-nucleosome complex (Leonard and Narlikar, 2015; Racki et al., 2014; Sinha et al., 2017). We therefore, focused on domain level interactions that were enriched at least two-fold in the presence of ADP•BeFx relative to ADP.

Figure 2 with 5 supplements see all
Dependence on the nucleosome acidic patch is linked to relief of autoinhibition.

(A) Domain architecture of SNF2h. The two critical arginines (R142, R144) of AutoN are highlighted in blue, while the NegC region replaced with a flexible GGS linker (Leonard and Narlikar, 2015) is highlighted in brown. Intervening sequences with no known domain annotations are numbered as in B. (B) Direct domain-domain interactions of SNF2h nucleosomes probed by crosslinking mass spectrometry. SNF2h mononucleosomes were crosslinked with the zero-length reagent EDC in the presence of either the ATP transition state analog ADP•BeFx or ADP. Residue-residue crosslink data were aggregated into domain-domain level spectral counts. The fold change between ADP•BeFx and ADP conditions is displayed as the log2 of the spectral count ratio. The data were centered to the median value and the color scale depicts the interactions most enriched in ADP•BeFx in magenta, and ADP in blue. Light grey tiles indicate no crosslinks observed in either condition. Domains are listed from N- to C- term within each protein. The distribution of fold changes in domain-domain spectral counts is plotted as a histogram below. It should also be noted that due to the structure of the histone octamer and the 2:1 stoichiometry of the SNF2h-Nucleosome complex, we cannot distinguish between intramolecular and intermolecular SNF2h-SNF2h or Histone-Histone crosslinks. (C and D). Mutation of AutoN or NegC increases remodeling rate with WT and APM nucleosomes. Cy5 fluorescent scans of native gel remodeling assays. (E) Saturating remodeling rates (kmax) for WT and APM nucleosomes with WT, 2RA, and mNegC SNF2h. Mutation of the acidic patch has a substantially lower effect on the kmax for 2RA and mNegC SNF2h than for WT SNF2h (8-fold and 2-fold respectively vs. 200-fold). Error bars represent standard errors on the mean for N = 3 replicates.

https://doi.org/10.7554/eLife.35322.010

Cross-links between the H4 tail and RecA lobe 2 of SNF2h are strongly enhanced in the ADP•BeFx state (Figure 2B). This result is consistent with previous work showing that the H4 tail activates ISWI remodelers and promotes a restricted active site conformation in the presence of ADP•BeFx (Clapier et al., 2001; Hamiche et al., 2001; Racki et al., 2014). We also found that the ADP•BeFx state promotes specific cross-links between the H2A/H2B acidic patch and lysines in the extended AutoN region, RecA lobe 1, the NegC region, and the DNA binding HAND-SANT-SLIDE (HSS) region of SNF2h (Figure 2—figure supplement 4). While cross-links between the acidic patch and the AutoN, HSS and NegC regions are compatible with structural constraints from previous studies, the cross-links with the RecA lobe one are not easily explained. Multiple studies of ISWI remodelers suggest that their RecA lobes bind at a location two DNA helical turns from the nucleosome dyad (SHL ±2), quite far from the H2A/H2B acidic patch (Dang and Bartholomew, 2007; Kagalwala et al., 2004; Schwanbeck et al., 2004; Zofall et al., 2006). We hypothesize that the RecA-acidic patch cross-links arise from a population of higher-order SNF2h-nucleosome aggregates in our mass-spec samples and therefore focus below on cross-links to the remaining three regions.

To test the functional significance of cross-links to AutoN, we mutated the lysines that crosslink to the acidic patch in the ADP•BeFx state. However, none of the mutants significantly altered remodeling by SNF2h (Figure 2—figure supplement 1). These results suggest that the lysines in the AutoN region are in proximity to the acidic patch but do not make mechanistically significant interactions. We speculated, however, that other residues in AutoN might make functional interactions with the acidic patch. Previous work has indicated that many nucleosome binding proteins recognize the acidic patch via arginine residues. Near the AutoN lysines that crosslink to the acidic patch resides a key autoinhibitory motif specific to ISWI family remodelers that contains two arginine residues (Figure 2A, Figure 2—figure supplement 4, R142 and R144). Autoinhibition by these arginine residues is relieved by a basic patch on the H4 tail, a nucleosomal epitope essential for maximal remodeling by ISWI-family remodelers (Clapier and Cairns, 2012; Clapier et al., 2002). We tested whether R142 and R144 functionally cooperate with the acidic patch by generating a mutated version of SNF2h with the two critical arginines of AutoN mutated to alanine (2RA), and measured the remodeling activity of this mutant. Consistent with previous reports, 2RA SNF2h remodels WT nucleosomes ~ 2 fold faster than WT SNF2h (Clapier and Cairns, 2012; Yan et al., 2016). However, 2RA SNF2h remodels APM nucleosomes ~ 50 fold faster than WT SNF2h (Figure 2F). This corresponds to a ~ 25 fold reduced dependency on the acidic patch for remodeling with 2RA SNF2h (Figure 2F). The same trend was also seen by an ensemble FRET remodeling assay (Figure 2—figure supplement 2). Together, these data suggest that the acidic patch contributes to relief of autoinhibition by R142 and R144 in AutoN. Since arginine residues are known to mediate this interaction in other systems, binding of either of the two arginines in AutoN to the acidic patch could provide a physical explanation for acidic patch recognition. However, given that neither carbodiimide chemistry nor any other commonly used cross-linking chemistry labels arginine residues, we cannot determine whether R142 and R144 make direct contacts with the acidic patch based on our mass-spectrometry data. We were unable to observe detectable binding between an AutoN peptide containing the 2R residues and the nucleosomal acidic patch through pull down assays (data not shown), suggesting that either these residues do not physically interact with the acidic patch or the surrounding regions of the SNF2h protein are required for stable binding.

To investigate the functional significance of cross-links to the NegC region of SNF2h, we determined the effect of replacing a stretch of 32 residues in NegC with a flexible serine-glycine linker (mNegC). NegC is another autoinhibitory region of SNF2h that imposes flanking DNA length sensitivity on SNF2h by specifically slowing down remodeling of nucleosomes without flanking DNA (Leonard and Narlikar, 2015). Consistent with previous work, mNegC SNF2h slides WT 0/60 nucleosomes ~ 1.2 fold faster than WT SNF2h (Figure 2E and F) (Leonard and Narlikar, 2015). However, mNegC SNF2h slides APM nucleosomes ~ 100 fold faster than WT SNF2h (Figure 2F). As a result, sliding of APM nucleosomes by mNegC SNF2h is only ~2 fold slower than WT nucleosomes. Thus the mNegC mutation almost completely rescues the defect of the acidic patch mutation. These results suggest that residues in NegC also link activation of nucleosome sliding to acidic patch recognition.

The third category of cross-links entailed lysine residues in the HSS regions. Mutants in these lysines greatly reduced remodeling and, in contrast to the 2RA and mNegC SNF2h mutants, did not rescue the defects caused by the acidic patch mutations (data not shown). It is possible that the defects caused by these lysine mutations reflect direct contacts between the HSS residues and the acidic patch. However, because the HSS also contacts flanking DNA, the reduction in sliding rate could also arise from defects in binding DNA.

To better understand which crosslinks are dependent on H2A acidic patch binding, we also performed SNF2h-nucleosome cross-linking reactions with ADP•BeFx in the presence of the LANA peptide which competes for acidic patch binding (Figure 1E). Addition of the LANA peptide substantially reduced crosslinks between the H2A acidic patch and both AutoN and NegC (Figure 2—figure supplement 5). In contrast, crosslinks between the HSS and the acidic patch were less strongly affected by LANA addition (Figure 2—figure supplement 5). This suggests that HSS positioning near the acidic patch in the ADP•BeFx state is not strictly dependent on direct binding to the region of the H2A acidic patch contacted by the LANA peptide. As a result, we cannot unambiguously interpret HSS-acidic patch crosslinks as reflecting mechanistically significant interactions. Together, these data suggest that acidic patch recognition is strongly linked to relief of autoinhibition by NegC and AutoN for nucleosome sliding by SNF2h. Furthermore, activation of SNF2h involves conformational changes that bring both of these autoinhibitory domains and the HSS in closer proximity to the acidic patch.

The acidic patch is required to promote the translocation phase of the SNF2h reaction

To gain additional insight into which steps in the remodeling cycle involve interaction with the acidic patch, we turned to a single-molecule FRET assay (smFRET; Figure 3A). This assay is analogous to the ensemble FRET remodeling assay described in Figure 2—figure supplement 2, in that the activity of SNF2h in sliding nucleosomes away from DNA ends increases the distance between a donor and acceptor dye pair, and thus decreases the measured FRET efficiency. However, with smFRET we can follow the remodeling of individual, surface-immobilized nucleosomes, and thereby gain insights into the activity of SNF2h reaction steps that are obscured in asynchronous, population-averaged ensemble assays. smFRET has previously been used to study remodeling by several ISWI family members, as well as by remodeling complexes from the SWI/SNF and INO80 families (Blosser et al., 2009; Deindl et al., 2013; Harada et al., 2016; Hwang et al., 2014; Zhou et al., 2018). A key insight revealed by these smFRET studies is that ISWI family remodelers do not slide nucleosomes in a continuous manner, such that FRET decreases continuously over time, but rather in a series of alternating phases: a ‘pause’ phase, in which FRET (and therefore nucleosome position) remain constant, and a ‘translocation’ phase, in which the nucleosome is moved relative to the DNA end. These repeating pause phases, which so far appear to be specific to ISWI, are essential to our understanding of the mechanism of action of ISWI remodelers, because the overall remodeling rates observed in ensemble assays are dominated by the durations of the pause phases, not the translocation rate itself. Moreover, substrate cues such as the H4 tail and flanking DNA have been shown to be sensed in the pause phase, not the translocation phase, for ISWI family remodelers (Hwang et al., 2014). That is, shorter flanking DNA or mutation of the H4 tail increases the durations of the pauses, thereby decreasing the overall remodeling rate, while having no effect on the actual sliding rate of the nucleosome. These results suggest a separation of regulation and activity in ISWI remodelers: translocation events are regularly interrupted by pauses that allow for the periodic interrogation of substrate cues. This pausing behavior also explains the ability of ISWI remodelers such as ACF to keep nucleosomes centered: if the nucleosome is translocated off-center, the interruption of this translocation by a new pause can trigger translocation in the opposite direction, restoring the nucleosome to a centered position.

Figure 3 with 3 supplements see all
The acidic patch interacts antagonistically with AutoN to promote pause exit and persistent translocation.

(A) Schematic of the smFRET setup. Nucleosomes labeled on histone H3 with a Cy3 donor dye and on one end of the DNA with a Cy5 acceptor dye are immobilized on the surface of a slide and imaged with a prism-based TIRF microscope. The end-positioned nucleosomes used here have an initial high FRET efficiency (see also Figure 3—figure supplement 2). As remodeling proceeds and the nucleosome is moved away from the DNA end, the FRET efficiency is reduced. (B) Example time-course of remodeling of a single surface-attached, WT nucleosome in the presence of saturating SNF2h and ATP (400 nM and 1 mM, respectively). Vertical yellow line indicates addition of enzyme and ATP; horizontal pink, green, and black lines are the output of an HMM fit used to quantify pause durations and locations (see Methods). Note the alternating ‘pause’ and ‘translocation’ phases of the remodeling reaction; in keeping with the ISWI literature, we call the first pause the ‘wait’ pause, the second pause ‘p1’, the third ‘p2’, etc. Intensity and FRET data here and in C have been smoothed with a 0.95 s median filter for visualization only. (C) Example time-courses of remodeling of, from top to bottom, WT nucleosomes by SNF2h, E64R nucleosomes by SNF2h, WT nucleosomes by 2RA SNF2h, and E64R nucleosomes by 2RA SNF2h. Saturating enzyme (400 nM enzyme with WT nucleosomes; 2 μM enzyme with E64R nucleosomes) and ATP (1 mM) were used in all cases. Additional examples are shown in Figure 3—figure supplement 2. (D) Quantification of the first three pause durations. *Indicates a lower limit; as shown in Figure 3—figure supplement 2B, remodeling of E64R nucleosomes by SNF2h is too slow to capture by smFRET, due to the competition between photobleaching of the dyes and remodeling by SNF2h. Errors were bootstrapped (see Materials and methods). (E) Empirical cumulative distribution functions (cdfs) of the change in nucleosome position during the first translocation event (top panel) and the second translocation event (bottom panel) for different combinations of nucleosome constructs and enzymes. Roughly 50% of the initial translocation events by SNF2h on WT nucleosomes move the nucleosome 7 bp or fewer (black dashed lines); in contrast, nearly 80% of initial translocation events by SNF2h on E64R nucleosomes move the nucleosome 7 bp or fewer. Similarly, during the second translocation event, SNF2h moves WT nucleosomes roughly 5 bp or fewer 50% of the time, whereas again nearly 80% of second translocation events for SNF2h with E64R nucleosomes result in step sizes 5 bp or fewer. Shaded areas represent bootstrapped error estimates (see Materials and methods). See also Figure 3—figure supplement 6 for additional representations of these data and a further discussion of step sizes.

https://doi.org/10.7554/eLife.35322.020

As shown in Figure 3B, we find that SNF2h alone, like the ACF complex and the yeast ISWI enzymes, shares this alternating pause and translocation behavior at the single-molecule level. We first wished to ascertain whether the acidic patch, like other nucleosomal epitopes recognized by ISWI remodelers, affects the regulatory pause phase. Because the rate of remodeling of APM nucleosomes by SNF2h is significantly slower (on the order of hours) than the rate of dye photobleaching (on the order of minutes; Figure 3—figure supplement 1B), we assembled nucleosomes with a single point mutation in the acidic patch (E64R nucleosomes). We used the E64R mutation rather than the E64A mutation because the defect caused by this mutation was better rescued by the 2RA mutation in SNF2h than the defect in E64A (Figure 1—figure supplement 3). The single-point mutation (E64R) is less deleterious than mutating all 4 acidic residues (APM), and SNF2h remodels E64R nucleosomes ~ 40 fold more slowly than WT nucleosomes as opposed to 200-fold more slowly with APM nucleosomes (Figure 1—figure supplement 3). However, this remodeling rate is still very slow relative to the timescales typically measured by smFRET, which posed two additional challenges: an increase in the number of noise events (dye blinking, intensity fluctuations, etc) per remodeling trajectory, and an increase in the amount of data to be analyzed. These challenges were addressed through the use of custom in-house smFRET analysis software (‘Traces’, https://github.com/stephlj/Traces) (Zhou et al., 2018; Johnson et al., 2018; copy archived at https://github.com/elifesciences-publications/Traces) to streamline the analysis of large data sets, and the adaptation of a computationally fast, versatile, open-source hidden Markov model (HMM) library called pyhsmm to quantify the durations of the pauses (see Materials and methods).

As shown in the example trajectories in Figure 3C, remodeling of E64R nucleosomes by SNF2h proceeds through the same alternating pause and translocation phases as does remodeling of WT nucleosomes. However, the pauses are noticeably longer with E64R nucleosomes, by at least a factor of 2 (Figure 3D), indicating that the acidic patch epitope, like other substrate cues, is indeed sensed during the regulatory pause phase. We note that remodeling of E64R nucleosomes in ensemble assays is significantly slower than the photobleaching rate (Figure 3—figure supplement 1B), so that by smFRET we only detect the fraction of remodeling events that are faster than photobleaching. As a result, the remodeling rate obtained for the E64R nucleosomes by smFRET represents an upper bound on the true remodeling rate (i.e. the E64R nucleosomes appear to remodel faster by smFRET than by ensemble assays (Figure 3—figure supplement 1B)).

Given the ability of the 2RA mutation of the AutoN motif to partially rescue remodeling defects in APM nucleosomes in ensemble assays (Figure 2C), we next asked whether this rescue is due to a restoration of wild-type pause durations. In agreement with our ensemble results (Figure 2C), 2RA SNF2h remodels WT nucleosomes slightly faster than SNF2h by reducing pause durations (Figure 3D). The reduction is minor (~1.3 fold), consistent with a previous smFRET study of the effect of the 2RA mutation on pause durations for the ACF complex (Hwang et al., 2014). Furthermore, 2RA rescues the deleterious effect of the E64R acidic patch mutation, by nearly restoring wild-type pause durations (Figure 3D). The effects of the E64R nucleosomal mutation and the 2RA SNF2h mutation on pause durations therefore mirror the effects on overall remodeling rates of APM nucleosomes: 2RA remodels E64R nucleosomes nearly as fast as SNF2h remodels WT nucleosomes, but not as fast as 2RA remodels WT nucleosomes (Figure 1—figure supplement 3). These results are consistent with a model in which relief of autoinhibition of the AutoN motif of SNF2h through direct or indirect interactions with the acidic patch enable pause exit (that is, promote the translocation phase).

The example traces in Figure 3C suggest an additional defect in remodeling of E64R nucleosomes: a reduction in the distance the nucleosome is moved during each translocation event (which we called the step size). Note that after two translocation events, WT nucleosomes with SNF2h or 2RA are moved from ~0.95 FRET to ~0.4 FRET (Figure 3C top and second from bottom), whereas after two translocation events the nucleosome in the E64R/SNF2h example trace has moved from ~0.95 FRET to ~0.75 FRET. Step size, like pause duration, plays an important role in regulating ISWI remodeler activity: since the pause durations dominate the overall remodeling rate, a smaller step size, which means more pauses per unit distance that the nucleosome is translocated, will mean a significant reduction in the overall remodeling rate (as observed in ensemble assays).

Step sizes can be quantified by converting the change in FRET between subsequent pauses to a change in the number of base pairs of DNA between the Cy5-labeled DNA end and the edge of the nucleosome. We accomplish this conversion by means of a calibration curve, described previously (Zhou et al., 2018). Like other ISWI family remodelers, SNF2h moves WT nucleosomes with an initial large step (~8 bp) followed by a smaller (~5 bp) step (Figure 3E, Figure 3—figure supplement 3; [Blosser et al., 2009; Deindl et al., 2013]). However, SNF2h moves E64R nucleosomes a shorter distance in each translocation phase, as indicated by the leftward shift of the cumulative distributions in the red curves of Figure 3E, relative to the black curves (on average, E64R nucleosomes move about 6 bp in the first translocation and about 4 bp in the second (Figure 3—figure supplement 3)). Mutation of AutoN has little effect on the step size in the context of WT nucleosomes, but largely restores the step size in the context of E64R nucleosomes to that of SNF2h with WT nucleosomes (Figure 3E, magenta and blue curves). Given that ISWI family remodelers have been shown to translocate a nucleosome in elementary steps of 1–2 bp (Deindl et al., 2013), our results suggest that with E64R nucleosomes, SNF2h takes fewer of these elementary steps in succession during the translocation phase, before entering a new pause phase.

The major contributions to the remodeling defects observed with SNF2h and APM nucleosomes can therefore be attributed to two effects: first, an increase in pause duration, and second, a decrease in the distance travelled per translocation event, meaning that there are more (and longer) pauses in APM remodeling reactions per unit distance that the nucleosome is moved. Both of these effects are rescued by the 2RA mutation of the AutoN motif of SNF2h. We propose that the acidic patch is important for relieving autoinhibition by AutoN and thereby promoting exit from the pause phase. Further, we propose that the acidic patch is also involved in keeping AutoN out of the active site until the nucleosome has been translocated to the full extent (~8 bp initially,~5 bp subsequently) and a new pause phase is entered (Figure 5).

The acidic patch is used by both ISWI and INO80 complexes

The ISWI ATPase forms complexes with several accessory proteins that regulate its localization and activity (He et al., 2006; Oppikofer et al., 2017; Tsukiyama et al., 1995; Varga-Weisz et al., 1997). ACF is one of the best studied of these complexes. ACF is a heterodimer of the ISWI ATPase subunit (SNF2h in humans) and the accessory subunit Acf1, and is implicated in gene repression, DNA replication, and DNA repair (Collins et al., 2002; Fyodorov et al., 2004; Lan et al., 2010). Biochemically, human ACF has the same core activity as SNF2h, but displays greater nucleosome affinity, enhanced sliding rates, and better kinetic discrimination of flanking DNA length (He et al., 2006; Yang et al., 2006). Despite these similarities, recent evidence suggests that ACF has some mechanistic differences from SNF2h on its own. For instance, in the context of ACF, AutoN regulates flanking DNA length sensing through interaction of an Acf1-specific domain called WAC (Hwang et al., 2014). SNF2h alone has no comparable domain. Furthermore, recent work has suggested that mutating the nucleosomal acidic patch causes a smaller defect in remodeling by ACF compared to SNF2h (Dann et al., 2017). Given this difference, we asked whether ACF requires the acidic patch for remodeling beyond binding. At saturating concentrations of ACF, where binding does not contribute to the overall remodeling rate, ACF slides APM nucleosomes 10-fold more slowly than WT (Figure 4A), indicating that ACF also uses the acidic patch in a step after binding. However, consistent with previous work, ACF is less dependent on the acidic patch compared to SNF2h alone (Dann et al., 2017).

The acidic patch is used by ACF and INO80.

(A) Gel remodeling assay with human ACF and 0/60 nucleosomes. Saturating concentrations of enzyme and ATP were used. (B) Gel remodeling with the yeast INO80 complex and 0/60 nucleosomes. Reactions were performed with saturating enzyme and ATP. Error bars represent standard error of the mean for N = 3 replicates.

https://doi.org/10.7554/eLife.35322.025

We next asked whether ISWI-family complexes uniquely use the acidic patch, or whether it is also used by other chromatin remodeling enzymes. Recently, it was shown that while some CHD family remodelers slide nucleosomes largely independent of the acidic patch, others are dependent on it (Dann et al., 2017; Levendosky et al., 2016). It has been further shown that SWI/SNF family enzymes also require the acidic patch for maximal activity (Dann et al., 2017). These observations raise the possibility that the acidic patch is a feature used by most remodeling enzymes. To address this issue we determined if the acidic patch is required by yeast INO80, which is in a distinct family from the CHD, ISWI and SWI/SNF families. INO80, like ACF, slides nucleosomes preferentially in the direction of longer flanking DNA and can also create evenly spaced nucleosome arrays (Udugama et al., 2011). This sliding activity is thought to be important for positioning the +1 nucleosome at transcription start sites (Krietenstein et al., 2016). Interestingly, we find that INO80 slides APM nucleosomes ~ 200 fold more slowly than WT nucleosomes at saturating concentrations (Figure 4B). This result indicates that INO80 also uses the acidic patch post-binding, but is more dependent on the acidic patch than the ACF complex.

Discussion

In this study, we investigate the role of the highly conserved H2A acidic patch in chromatin remodeling by ISWI enzymes. We find that the acidic patch is used post-binding in order to activate remodeling by both INO80 and ISWI family remodelers. Furthermore, using a combination of ensemble and single molecule methods, we show that the acidic patch is used by SNF2h to relieve autoinhibition by the conserved AutoN and NegC motifs. Below we explore the mechanistic and regulatory implications of these results.

ATP-dependent chromatin remodeling enzymes carry out specialized reactions on a complex substrate. Understanding how recognition of this substrate is coupled to activity can provide a means to understanding common principles underlying ATP-dependent chromatin remodeling mechanisms. Owing to decades of study, ISWI enzymes provide a useful model system to address this question. On the basis of crosslinking and footprinting studies, the ATPase domain of ISWI enzymes is thought to bind and translocate DNA two helical turns from the nucleosomal dyad (SHL ± 2) (Dang and Bartholomew, 2007; Kagalwala et al., 2004; Schwanbeck et al., 2004; Zofall et al., 2006). Work from several groups has also shown that for ISWI remodelers, recognition of both the H4 tail and flanking DNA enhances remodeling activity post-binding (Clapier et al., 2001; Hamiche et al., 2001; Yang et al., 2006). While the mechanisms by which these nucleosome cues activate remodeling are not well understood, the ISWI domains that recognize these cues are known. The C-terminal HAND-SANT-SLIDE (HSS) domain mediates flanking DNA recognition, while the H4 tail appears to be directly recognized by the second RecA lobe within the ATPase domain (Dang and Bartholomew, 2007; Yan et al., 2016). The acidic patch resides on a surface near SHL ± 6, far from where the ATPase domain engages the nucleosome. How then might the acidic patch be recognized and used by ISWI remodelers?

Our results suggest that two known autoinhibitory regions of ISWI enzymes, the AutoN region and the NegC region, functionally interact with the acidic patch, because mutating these regions dramatically reduces the dependence of SNF2h on the acidic patch for sliding. This suggests that a large role of the acidic patch is to relieve autoinhibition by both AutoN and NegC. While we do not have evidence for a direct interaction between the acidic patch and the two arginines in AutoN, our cross-linking mass spectrometry data suggests that activation of the enzyme places residues within both AutoN and NegC near this location. Our smFRET work here and previous smFRET work suggests that the AutoN region inhibits the transition from the pause phase to the translocation phase (Hwang et al., 2014). It has been shown that flanking DNA and the H4 tail are both sensed in the pause phase (Hwang et al., 2014). Further, previous ensemble work has suggested that the NegC region inhibits the transition between a flanking DNA sensing state of SNF2h and a translocation competent state of SNF2h (Leonard and Narlikar, 2015). We therefore propose that the acidic patch helps promote the translocation competent state of SNF2h by providing an alternative binding site for NegC and AutoN (Figure 5).

Model for nucleosome remodeling by SNF2h.

After binding the nucleosome, SNF2h is in equilibrium between an active and autoinhibited state. In the autoinhibited state AutoN and NegC hold the remodeler in an inactive state. The active state is promoted by AutoN and NegC binding near the acidic patch and by H4 tail binding the ATPase domain resulting in conformational changes that bring HSS in close proximity to the acidic patch. From this active state, SNF2h can translocate DNA around the octamer. Although SNF2h remodels nucleosomes as a dimer at saturating enzyme concentrations (Blosser et al., 2009; Racki et al., 2009), in this model we display only one of the protomers for simplicity.

https://doi.org/10.7554/eLife.35322.027

In addition to an increase in pause durations with the acidic patch mutations, the amount of DNA translocated within a translocation phase is reduced compared to WT nucleosomes. We hypothesize that translocation is interrupted by premature reversion of the enzyme to the autoinhibited state in the absence of stabilizing interactions with the acidic patch (Figure 5). As a result, more pauses are encountered per distance translocated. Our results lead to a model in which the acidic patch provides a binding surface for NegC and AutoN that sequesters these regions from inhibiting SNF2h (Figure 5). Combined with previous work, our results underscore how the strong coupling of relief of autoinhibition to recognition of two conserved nucleosome cues (the H4 tail and the acidic patch) make this motor exquisitely specific for its complex substrate.

The acidic patch increases Kmapp for SNF2h by ~5 fold, suggesting that interactions with the acidic patch also stabilize SNF2h binding. Thus the acidic patch appears to be used in at least two distinct steps of the SNF2h remodeling reaction. What residues in SNF2h might be interacting with the acidic patch in these two steps? In contrast to the rescuing effects of AutoN mutations on maximal activity, AutoN mutations additively increase Kmapp beyond solely mutating the acidic patch (Figure 2—figure supplement 3). This suggests that AutoN and the acidic patch do not cooperate in the ground state. It is thus possible that different SNF2h domains contact the acidic patch in different steps of the remodeling reaction. The different set of cross-links observed between the acidic patch and SNF2h regions in the ADP vs. ADP•BeFx state are consistent with such a possibility.

Based on previous work, we have hypothesized that SNF2h undergoes a large conformational change prior to adopting a translocation competent state that repositions the C-terminus from binding flanking DNA towards binding the nucleosome core (Leonard and Narlikar, 2015). Our crosslinking-MS data provides additional insights into the structural rearrangements that accompany enzyme activation. We find that crosslinks between the H2A/H2B acidic patch and the SANT domain increase in the ADP•BeFx-bound state compared to the ADP-bound state (Figure 2B), consistent with the HSS binding to the nucleosome core. Interestingly, these crosslinks are not substantially reduced by LANA peptide binding, suggesting that the location of the HSS in the ADP•BeFx-bound state is not strongly dependent on direct contacts with the residues contacted by the LANA peptide (Figure 2—figure supplement 5). This result raises the possibility that the SANT domain and LANA peptide occupy adjacent regions on the nucleosome core. The SANT domain may then cross-link to the acidic patch when these sites are transiently exposed due to dynamics in LANA peptide binding. In contrast, NegC and AutoN crosslinks to the acidic patch are substantially reduced by the LANA peptide, suggesting LANA binding directly or indirectly displaces these regions from the nucleosome. Overall our results suggest that the key accessory regions of SNF2h namely, HSS, NegC and AutoN, are all positioned near the acidic patch in the activated state. In agreement with this observation, crosslinks between the C-terminus of SNF2h and NegC increase in the ADP•BeFx state (Figure 2B). The close positioning of multiple SNF2h accessory domains near the acidic patch raises the possibility that contacts between accessory domains may play a role in promoting the translocation-competent state. While substantial future work would be needed to test this possibility, it is analogous to recent observations with the yeast CHD1 remodeling motor that contacts between its N-terminus and the C-terminal DNA binding domain regulate the sliding reaction (Sundaramoorthy et al., 2017).

Most chromatin remodeling ATPases also form large multi-subunit complexes, which regulate the activity and specificity of the remodeling reaction. Here we find that the human ISWI complex, ACF, requires the acidic patch for maximal activity but shows a ~ 20 fold smaller defect upon mutation of the acidic patch than observed with SNF2h alone. This result is qualitatively consistent with recent studies showing that the acidic patch has a smaller role in the activity of ACF vs. SNF2h (Dann et al., 2017). Our results here provide a mechanistic framework for understanding these recent observations. In particular, our results suggest that the accessory protein Acf1 alters the mechanism of relief of autoinhibition by the acidic patch in a manner that makes the reaction less dependent on the acidic patch, perhaps by providing an alternative binding partner for the AutoN and NegC domains. Such a domain would be analogous to the WAC motif of Acf1, which provides an alternative binding partner for the H4 tail (Hwang et al., 2014). In contrast to ACF, we find that the yeast INO80 complex, a large multisubunit complex, is as dependent on the acidic patch for nucleosome sliding as the isolated SNF2h ATPase. Importantly, INO80 family remodelers are insensitive to the presence of the H4 tail and no AutoN-like or NegC like motif has been identified in the ATPase subunit of INO80 (Udugama et al., 2011). The acidic patch must then activate INO80 through a mechanism distinct from ISWI complexes. Determining how the acidic patch is used by INO80 ATPases, and what roles the ATPase and accessory subunits play in this mechanism, are important areas of future study.

Combined with previous results (Dann et al., 2017), our results suggest that several families of chromatin remodelers require the acidic patch for remodeling. However, it is possible that this surface is not a universal requirement for chromatin remodeling, as yeast CHD1 can remodel nucleosomes largely independent of the acidic patch (Levendosky et al., 2016). CHD1 instead uses an unidentified aspect of the histone H2A/H2B dimer to promote remodeling (Levendosky et al., 2016). Yeast CHD1 and ISWI family remodelers have been thought to share a common remodeling mechanism, as these families share biochemical activities, like nucleosome sliding and spacing, and substrate cues required for maximal remodeling activity, such as the H4 tail and flanking DNA (Ferreira et al., 2007; Stockdale et al., 2006). However, the enormous difference in dependence on the acidic patch between yeast CHD1 and ISWI enzymes raise the possibility that these families remodel nucleosomes through distinct mechanisms (Levendosky et al., 2016). Importantly, while yeast CHD1 shares domains with ISWI remodelers such as the SANT-SLIDE domains at the C-terminus and a version of the NegC region, called the C-terminal bridge (Hauk et al., 2010; Ryan et al., 2011), CHD family remodelers do not appear to possess an AutoN motif. Instead, remodelers like CHD1 have an N-terminal double chromodomain which has an analogous role as an autoinhibitiory domain that is relieved by H4 tail binding (Hauk et al., 2010).

At a primary level, the requirement of the acidic patch provides a powerful means for remodelers to sense and respond to chromatin structure and nucleosome content. Thus, nucleosomes lacking histone H2A-H2B dimers or containing a modified acidic patch through histone variants or covalent modifications may be recognized and remodeled differently than canonical nucleosomes by different remodelers (Dann et al., 2017). Consistent with this possibility, nucleosomes containing histone H2Az, which have an extended acidic patch, are remodeled ~2 fold faster by ISWI complexes than canonical nucleosomes (Goldman et al., 2010). Analogously, recent work has shown that INO80 preferentially slides H2AZ nucleosomes over H2A nucleosomes (Brahma et al., 2017). Finally, given the growing list of factors that recognize the acidic patch, it is likely that remodelers and other chromatin binding proteins compete for access to the acidic patch. Indeed, binding by the LANA peptide to the acidic patch competes for remodeling by SNF2h. Sensitivity to the acidic patch could be a general mechanism to regulate the outcome of chromatin remodeling at loci where remodelers and other factors are jostling for access to their chromatin substrates.

Materials and methods

Expression and purification of chromatin remodeling enzymes

Request a detailed protocol

SNF2h was purified from E. coli as described previously with minor modifications (Leonard and Narlikar, 2015). DNA was precipitated following cell lysis by addition of 5% w/v polyethylenimine (P3143, Sigma-Aldrich, St. Louis, MO) pH 7.9 dropwise to a final concentration of 0.1% and clarified by centrifugation. Following cobalt affinity purification, the 6xHis tag was cleaved overnight with TEV protease and dialysed into SEC Buffer. TEV-cleaved SNF2h was then purified by anion exchange chromatography using a HiTrap Q column and size exclusion chromatography (GE Life Sciences, Pittsburgh, MA). SNF2h concentration was determined by SDS-PAGE with BSA protein standards and staining with SYPRO Red (Thermo Fisher, Waltham, MA).

Human ACF was expressed and purified recombinantly from Sf9 insect cells by FLAG immunoaffinity purification as described previously with minor modifications (Aalfs et al., 2001). SNF2h-FLAG and Acf1 were expressed separately via infection with baculovirus. Nuclear extracts from each construct were generated and mixed together at a 10:1 Acf1:SNF2h-FLAG volume ratio to ensure full assembly of the complex. This mixture was then bound to FLAG M2-affinity resin (Sigma-Aldrich, St. Louis, MO), washed with increasing KCl concentrations, and eluted with buffer with 100 mM KCl and 1 mg/mL FLAG Peptide. ACF concentration was determined by SDS-PAGE with BSA standards and based on the intensity of the Acf1 band.

INO80 was purified by FLAG immunoprecipitation based on previously published methods (Shen, 2004; Zhou et al., 2018). Briefly, S. cerevisae with endogenously flag-tagged INO80 was grown in YEPD at 30°C to saturation. Cells were pelleted by centrifugation for 10 min at 5000 rpm, resuspended with buffer H0.3 (25 mM HEPES, pH 7.5, 1 mM EDTA, pH 8.0, 10% glycerol, 0.02% NP-40, 0.3 M KCl), and pelleted again. Pelleted cells were then extruded through a 60 mL syringe into liquid nitrogen to create ‘noodles’. Cell ‘noodles’ were then lysed using a Tissue Lyser II (Qiagen, Hilden Germany) cooled in liquid nitrogen. Frozen lysate powder was resuspended in equal volume of H0.3 and spun in an SW28 rotor for 2 hr at 25,000 rpm at 4°C. Clarified lysate was mixed with equal volume buffer H0.3 and applied to FLAG M2-affinity resin (1 mL bead slurry per 40 mL of cleared lysate) equilibrated with H0.3 and incubated for 3 hr at 4°C. Resin was washed with 3 × 50 mL buffer H0.5 (H0.3 buffer except with 0.5 M KCl) followed by 3 × 10 mL washes with buffer H0.1 (0.1M KCl) and eluted with H0.1 supplemented with 1 mg/mL FLAG peptide. Eluate was concentrated, aliquoted, flash frozen in liquid nitrogen, and stored at −80°C. INO80 concentration was determined by SDS-PAGE with BSA standards, based on the intensity of the Ino80-flag band.

Nucleosome labeling and reconstitution

Request a detailed protocol

Recombinant Xenopus laevis histones were expressed and purified from E. coli as previously described (Luger et al., 1999). Histone H2A E61A, E64A, D90A, D92A expression plasmid was a generous gift from the Tan lab at Penn State. Purified histone H2A E64R was provided by the Wolberger lab. Histone octamer was reconstituted as previously described (Luger et al., 1999; Zhou and Narlikar, 2016), except for smFRET nucleosomes where a 2:1 unlabeled:labeled H3 mixture was used during octamer assembly to generate nucleosomes with mostly one H3 or neither H3 labeled. Histone H3 with a cysteine introduced at position 33 was labeled with either Cy3 (for smFRET) or Cy5 (for ensemble assays) prior to histone octamer assembly via cysteine-maleimide chemistry. Cy3-labeled (for ensemble assays) and Cyanine 5 SE-labeled and biotinylated DNAs (for smFRET) were generated by PCR with HPLC-purified, labeled primers (Cy5 primers: TriLink Biotechnologies, San Diego, CA; Cy3 and biotinylated primers: IDT, Coralville, IA) and purified by PAGE. The strong, synthetic 601 nucleosome positioning sequence (Lowary and Widom, 1998) was used to assemble all nucleosomes in this study, with an arbitrary sequence for DNA flanking the 601 positioning sequence (Figure 1—figure supplement 1). These DNAs were assembled with either wild-type or APM octamers by salt gradient dialysis and purified by glycerol gradient centrifugation (Zhou and Narlikar, 2016).

Native gel remodeling assay

Request a detailed protocol

All remodeling reactions were performed under single turnover conditions (enzyme in excess of nucleosomes). Reactions with SNF2h were performed at 20°C with 20 nM nucleosomes, 12.5 mM HEPES pH 7.5, 2 mM Tris pH 7.5, 70 mM KCl, 5 mM ATP•MgCl2, 3 mM MgCl, 0.02% NP40, and ~3%(v/v) glycerol. Reactions with ACF and INO80 were performed as above at 30°C and with minor changes in buffer composition (ACF: 10 nM nucleosomes, 12.5 mM HEPES pH 7.9, 2 mM Tris pH 7.5, 60 mM KCl, 2 mM ATP•MgCl2, 3 mM MgCl2, 0.02% NP40, 0.3 mg/mL FLAG peptide, and ~9% glycerol; INO80: 10 nM nucleosomes, 40 mM Tris pH 7.5, 60 mM KCl, 2 mM ATP•MgCl2, 1.1 mM MgCl2, 0.02% NP40, 0.5 mg/mL FLAG peptide, and 1% glycerol). Reactions were started with addition of enzyme and time points were quenched with excess ADP and plasmid DNA. Time points were then resolved by native PAGE (6% acrylamide, 0.5XTBE) and scanned on a Typhoon variable mode imager (GE Life Sciences, Pittsburgh, PA) by scanning for fluorescent labels. Gels were then quantified by densitometry using ImageJ. The fraction of nucleosomes end-positioned (i.e. unremodeled) at a given time point was determined by the ratio of fast-migrating nucleosomes to the total nucleosome intensity. This was fit to a single exponential decay using Prism 6 (GraphPad, La Jolla, CA) (Equation 1),

(1) y=(y0p)ekobst+p

where y0 is the initial fraction end-positioned, kobs is the observed rate constant, and p is the fraction end-positioned at plateau. Reactions in a given concentration series were fit constrained to a common y0 and p. Concentration series were fit to a cooperative binding model (Equation 2),

(2) kobs=kmaxX(Kmapp)h+Xh

where X is the concentration of SNF2h, h is the hill coefficient, Kmapp is the apparent Km, and kmax is the saturating rate constant. Competition assays were performed as described above with varying concentrations of LANA peptide and fit to a single exponential decay. This was then fit to a simple competition binding model (Equation 3),

(3) kobs=k01+XK1

where k0 is the rate constant without peptide, X is the concentration of peptide, and KI is the inhibition constant.

ATPase assay

Request a detailed protocol

ATPase reactions were performed under multiple turnover conditions (nucleosomes in excess of enzymes). Reactions were performed with 12.5 nM SNF2h, 12.5 mM HEPES pH 7.5, 2 mM Tris pH 7.5, 70 mM KCl, 7.5 µM ATP•MgCl2, 3 mM MgCl2, 0.02% NP40,~3%(v/v) glycerol, and trace amounts of γ-32P-ATP. Reactions were started with addition of enzyme, and 2.5 µL time points were quenched with an equal volume of 50 mM Tris pH 7.5, 3% SDS, and 100 mM EDTA. Inorganic phosphate was resolved from ATP on a PEI-cellulose TLC plate (Select Scientific) with 0.5 M LiCl/1M formic acid mobile phase. Plates were dried, exposed to a phosphorscreen overnight, and scanned on a Typhoon variable mode imager. Rate constants were determined by fitting a line through the first 10% of inorganic phosphate generated using Prism.

Ensemble FRET remodeling assay

Request a detailed protocol

Ensemble FRET remodeling assays were performed under the same conditions as gel remodeling assays. Reactions were initiated by addition of enzyme and then measured in a K2 fluorometer (ISS) equipped with a 550 nm short pass excitation filter and a 535 nm long pass emission filter. Reactions were excited at 515 nm and emission was measured at 665 nm. The resulting curves were fit to a two-phase exponential decay (Equation 4),

(4) y=(p+(y0p)(ffastekfastt)+(1ffast)ekslowt)

where ffast is the fraction in the fast phase and kfast and kslow are the apparent rate constants of the fast and slow phase respectively.

Crosslinking mass spectrometry

Request a detailed protocol

Crosslinking mass spectrometry samples were prepared by incubating 72 µg of mononucleosomes without flanking DNA at 9 µM final concentration with 24 µM SNF2h in buffer containing either ADP or ADP•BeFx (15 mM HEPES pH 7.5, 140 mM KCl, 0.5 mM ADP•Mg, 0.5 mM MgCl2,±0.5 mM BeFx 1:5 BeCl2:NaF) for 10 min at 30°C. The crosslinking reaction with the LANA peptide was performed the same as the ADP•BeFx condition with 30 µM peptide added. The samples were then reacted with 1 mM EDC and 20 µM N-hydroxysulfosuccinimide (added as a 10x stock in water) for 60 min at room temperature. Crosslinking reactions were then quenched by adding 10 mM Tris pH 7.5 and samples were acetone precipitated and washed once with cold acetone. The pellet was resuspended in 8M Urea, 5 mM TCEP, 100 mM ammonium bicarbonate and heated at 56°C for 25 min, followed by alkylation with 10 mM iodoacetamide for 40 min at room temperature. The sample was diluted 5-fold with 100 mM ammonium bicarbonate and digested with 1:25 trypsin for 4 hr at 37°C followed by addition of a second aliquot of trypsin and overnight digestion.

Crosslinked peptides were desalted using 100 µl OMIX C18 tips (Agilent), fractionated by size-exclusion chromatography (SEC), and analyzed by LC-MS similarly to a previously described method (Zhou et al., 2017). Briefly, trypsin digests were acidified to 0.2% TFA, desalted, and run over a Superdex Peptide PC 3.2/300 s column (GE Healthcare). SEC fractions eluting between 0.9 ml and 1.4 ml were dried and resuspended in 0.1% formic acid for LC-MS. Each fraction was separated over a 15 cm x 75 μm ID PepMap C18 column (Thermo) using a NanoAcquity UPLC system (Waters) and analyzed by a Fusion Lumos mass spectrometer (Thermo). Precursor ions were measured from 375 to 1500 m/z in the Orbitrap analyzer (resolution: 120,000; AGC: 4.0e5). Ions charged 3+ to 8+ were isolated in the quadrupole (selection window: 1.6 m/z units; dynamic exclusion window: 30 s; MIPS Peptide filter enabled), fragmented by HCD (Normalized Collision Energy: 28%) and measured in the Orbitrap (resolution: 30,000; AGC; 5.0e4). The cycle time was set to 3 s.

Peaklists were generated using PAVA (UCSF) and searched for crosslinked peptides with Protein Prospector 5.19.22 (Trnka et al., 2014) against a target database containing human SNF2h plus the four core histone sequences from X. laevis concatenated with a decoy database containing 10 randomized copies of each target sequence (total database size: 55 sequences). Loss of the initiator methionine and carbamidomethylation of cysteine. Methionine oxidation, peptide N-terminal glutamine to pyroglutamate formation, acetylation at the protein N-terminus, and mis-annotation of the monoisotopic peak (1 Da neutral loss) were treated as variable modifications. EDC was designated as a heterobifunctional crosslinking reagent with specificity of aspartate, glutamate, and the protein C-terminus on one side and lysine and the protein N-terminus on the other with a bridge mass corresponding to loss of H2O. A mass modification range of 400–5000 Da was specified on these residues and 85 product ion peaks from the peaklist were used in the search. Precursor and product ion tolerances were 8 and 25 ppm respectively.

Crosslinked spectral matches (CSMs) were initially classified as in (Zhou et al., 2017). The dataset was then aggregated into unique crosslinked residue-pair level data with a corresponding spectral count value. Due to the prevalence of multiple, closely spaced Asp and Glu residues in a typical tryptic peptide, site-localization of EDC crosslinks is more challenging than with homobifunctional lysine-directed reagents. To address this, when the site-localization was judged to be ambiguous, all possible residue-pairs were kept with an annotation noting the ambiguity. When calculating spectral counts, fractional spectral counts were assigned to these ambiguous site localizations so that a given spectrum was awarded exactly one spectral count. For instance, a product ion spectrum matching equally well to both K91.H4-D65.H2B or K91.H4-E68.H2B contributes 0.5 towards the spectral counts of each residue-pair. Decoy CSMs were retained throughout this aggregation and spectral counting process. A linear SVM model, built on five features of the Protein Prospector search output (score difference, % of product ion signals matched, precursor charge, rank of peptide 1, and rank of peptide 2) was constructed to sort crosslinked residue pairs into decoy and target classes. Crosslinked residue-pairs with an SVM score greater than 1, score difference greater than 5, and at least one spectral count are reported. The final residue-pair level data set is reported at specificity of 99.7% corresponding to 0.05% FDR. The number of unique crosslinks in the ADP condition was 974, while 1470 crosslinks were unique to the ADP•BeFx condition and 707 crosslinks were common to both conditions.

To determine which protein domains are involved in SNF2h mediated nucleosome sliding, residue level crosslink spectral counts were aggregated into domain level counts. Each domain pair was assigned a minimum spectral count of 1 to avoid dividing by zero, and the log2 ratio of spectral counts for each domain pair in the ADP•BeFx condition to the ADP condition were calculated. Domain pairs with a Log2 ratio of exactly 0 were treated as NA values. The remaining data were normalized such that the median value was set at 0. Hence, most domain-domain interactions were assumed to not change substantially between conditions.

Histone protein sequences were taken from Xenopus Laevis Uniprot Entries (without the initiator methionine) and domains were defined as follows: H2A N-term tail: 1–16, H2A: 17–43, H2A Acidic Patch: 44–100, H2A C-term tail: 101–129; H2B tail: 1–34, H2B: 35–99, H2B Acidic Patch: 100–122; H3 tail: 1–44, H3: 45–135; H4 tail: 25–102.

The sequence of SNF2h was identical to the Human entry in Uniprot (O60264) with an additional two amino acids at the N-terminus to match the construct used. All residue numbers are therefore shifted from the Uniprot entry by two aa. SNF2h domains were defined as follows: Snf2h1: 1–83, AutoN: 84–160, Snf2h2: 161–183, RecA1: 184–402, RecA2: 403–641, NegC: 642–703, Snf2h4: 704–736, HAND: 737–839, SANT: 840–894, SLIDE: 895–1013, Snf2h5: 1014–1054.

Annotated Mass Spectra are available using MS-Viewer at: http://msviewer.ucsf.edu/prospector/cgi-bin/msform.cgi?form=msviewer

The ADP data set is accessed with search key: 2x0kr2kzq1

The ADP•BeFx data set is accessed with search key: fjamygr8pl

The ADP•BeFx in the presence of LANA peptide is accessed with search key: c5o2mcxwum.

Raw mass spectrometry data is available in the MassIVE repository at UCSD with accession key: MSV000082136

Single molecule FRET measurements

smFRET experiments were performed as previously described in (Zhou et al., 2018) with modifications to the reaction buffers as described below.

Sample preparation and imaging

Request a detailed protocol

Briefly, as in (Zhou et al., 2018), quartz slides were PEGylated and then incubated with neutravidin (A2666, ThermoFisher Scientific, Waltham, MA) to mediate attachment of biotinylated nucleosomes. After removal of unbound neutravidin, biotinylated nucleosomes at 12.5 pM in a modified Wash Buffer (12 mM HEPES-KOH, pH 7.5 at 22°C, 60 mM KCl, 1.4 mM MgCl2, 10% glycerol, 0.1 mM EDTA, 0.02% Igepal, 1% [w/v] glucose, and 0.1 mg/mL acetylated BSA) were added and incubated for 10 min. Unbound nucleosomes were removed by washing with Wash Buffer. All incubations and experiments were performed at 20°C. Nucleosomes were imaged on a custom-built prism-based TIRF setup.

Immediately prior to data acquisition, the sample chamber was flushed with a modified imaging buffer (53 mM HEPES-KOH, pH 7.5 at 22°C, 9.1 mM Tris-acetate, pH 7.5 at 22°C [contributed by the Trolox], 63 mM KCl, 1.41 mM MgCl2, 10% glycerol, 0.1 mM EDTA, 0.02% Igepal, 1% [w/v] glucose, 0.1 mg/mL acetylated BSA, 2 mM Trolox [Sigma 238813, made as an 11 mM stock in Tris-acetate, pH'd to 7.5 with 1 M NaOH, and stored at 4°C], 0.03 mM β-mercaptoethanol, 2 U/μL catalase [=0.2 mg/mL; Sigma E3289], and 0.08 U/μL glucose oxidase [0.8 mg/mL, Sigma G2133; made with the catalase as a 100x stock in SPB, and stored at 4°C for not more than one week]). Images were collected using Micro-Manager (www.micro-manager.org, San Francisco, CA) (Edelstein et al., 2010) at 7.4 Hz, with an exposure time of 100 ms. To start each reaction, saturating enzyme (400 nM for WT nucleosomes; 2 μM for E64R nucleosomes) and saturating ATP (1 mM) in 300 μL imaging buffer were added via an automated syringe pump (J-KEM Scientific, St. Louis, MO).

Data analysis

Request a detailed protocol

The number of remodeling events per smFRET experiment was roughly an order of magnitude lower with E64R nucleosomes compared to WT nucleosomes, necessitating significantly larger data sets for the E64R nucleosomes compared to WT. To streamline the data analysis process with these data sets, as well as to improve the overall quality of the data, we made use of the custom in-house software we have developed for smFRET image analysis, called Traces, available for download at https://github.com/stephlj/Traces (Zhou et al., 2018; Johnson et al., 2018; copy archived at https://github.com/elifesciences-publications/Traces). In addition, the long pauses exhibited by SNF2h remodeling E64R nucleosomes are subject to an increased number of artifacts, such as dye blinking or slight fluctuations in the noise, which can complicate quantification of smFRET trajectories. We therefore used the python-based HMM library pyhsmm (https://github.com/mattjj/pyhsmm), which we adapted for the analysis of smFRET data as part of the Traces package, to quantify pause durations. This particular HMM package, which fits a discrete state HMM to each trajectory generated by Traces, enables a reduction in the likelihood of the HMM identifying artifacts as real transitions, and also reduces analysis time.

As described in Figure 3—figure supplement 1, and consistent with previous smFRET studies with nucleosomes (Blosser et al., 2009; Deindl et al., 2013; Hwang et al., 2014), we observe two predominant clusters of FRET values, at 0.57 and 0.95 FRET, in the absence of remodeler. These FRET states correspond to two of the four populations of nucleosomes that result from mixing unlabeled H3 with Cy3-labeled H3 during octamer formation: some nucleosomes will have a Cy3 label on the H3 proximal to the Cy5-labeled DNA end, resulting in the higher FRET state, and some will have a Cy3 label on the H3 distal to the Cy5-labeled DNA end, resulting in the mid-FRET state. There will also be a population of nucleosomes with both H3 histones unlabeled, which show no FRET; and a population with both copies of H3 labeled, which are distinguishable by two-step photobleaching of the Cy3 dye, and are excluded from all analyses (i.e. both calibration curve data (Zhou et al., 2018) and remodeling data). We also excluded any trajectories to which pyhsmm fit an initial FRET value lower than 0.775 FRET, since nucleosomes with distally labeled H3's do not provide as great a dynamic range for monitoring nucleosome remodeling. We excluded any part of a trajectory including and subsequent to backtracking events (where the nucleosome was moved away from the center, that is, where FRET increased instead of decreasing). Each data set (e.g., E64R/SNF2h) consists of at least 100 trajectories collated from at least 3 (typically 5 to 7) different experiments. All errors were bootstrapped over trajectories (that is, for each data set, the ≥100 trajectories were resampled with replacement, and reported values, such as the means of each pause duration, were recalculated for each bootstrapped sample). Reported errors are standard deviations of the bootstrapped values. A similar procedure was used to obtain the errors on the cdfs in Figure 3E and Figure 3—figure supplement 3; the shaded regions represent ±the standard deviation of the set of bootstrapped cdfs.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
    Chromatin Protocols
    1. K Luger
    2. TJ Rechsteiner
    3. TJ Richmond
    (1999)
    1–16, Expression and purification of recombinant histones and nucleosome reconstitution, Chromatin Protocols, New Jersey, Humana Press, 10.1385/1-59259-681-9:1.
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59

Decision letter

  1. Jerry L Workman
    Reviewing Editor; Stowers Institute for Medical Research, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for submitting your work entitled "An activating role for the nucleosome acidic patch in ATP-dependent chromatin remodeling" for consideration by eLife. Your article has been reviewed by three peer reviewers,, one of whom is a member of our Board of Reviewing Editors and the evaluation has been overseen by a Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Tom Owen-Hughes (Reviewer #3).

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

As you will see when reading the reviews appended below there was considerable enthusiasm amongst the reviewers for various aspects of the manuscript. However, the lack of evidence of a direct interaction between the acidic patch and AutoN dampened our enthusiasm as this would appear to be essential data to support the proposed model. There were also issues with regards to the measurements of the "step sizes" in the reaction as to how those relate to what is in the published literature. As it is unclear how long it would take to acquire the necessary data, the reviewers concluded that the current manuscript should be rejected, rather than caught in what might be a considerable delay. We would welcome a resubmission should additional data become available. Please feel free to communicate with us should you desire any clarification.

Reviewer 1

This is an exciting manuscript that takes powerful biochemical and biophysical approaches to analyze details of nucleosome remodeling by the Snf2h (Iswi) remodeling ATPase. The results implicate a central role of the acidic patch on the H2A H2B dimer as an important epitope in receiving auto-inhibition of remodeling activity. This effect is mediated by increased pause and decreased steps in nucleosome translocations. The authors go on to show that the ACF complex and Ino80 are also dependent on the acidic patch for remodeling. Some points that need to be addressed.

1) This paper needs to include some biochemical characterization of the APM nucleosomes compared to WT. Are they as stable for example at higher salt, etc. While the location of the mutations suggests they should not affect nucleosome stability, this can't be assumed. Moreover, the fact that Ino80 remodeling is also affected by the mutations without having similar Auto inhibitory domains suggest that the effect of the mutations is on the nucleosome substrate rather than through the remodelers.

2) The assumption that the single point mutant is simply a weaker version of the 4-point mutants in the acidic patch may not be true. It may act completely differently. This assumption would be stronger if additional single point mutants were examined.

3) The mechanism would be strengthened if the authors can demonstrate a direct interaction between AutoN and the acidic patch and that this is reduced in the mutants. Simple peptide pulldown assays with Auto N compared to the H4 tails would suffice.

4) The authors need to make the case that studying hSn2h in the absence of its partner proteins is warranted. The fact that ACF (of which hSf2h is a subunit) is much less dependent on the acidic patch than hSn2h alone could be construed as suggesting these are artifacts of studying hSnf2h out of its normal context.

Reviewer 2

The main findings of this work are that the ISWI and INO80 remodelers are sensitive to the acidic patch on histone H2A. The ISWI ATPase subunit, SNF2h, is more extensively studied, where it is shown that nucleosome sliding is dramatically reduced with a quadruple acidic patch mutant. A nice complementary experiment is the competition with the LANA peptide, which is known to bind to the acidic patch, and therefore serves as a confirmation that this region is important for nucleosome sliding by ISWI. By using saturating amounts of remodeler, the authors demonstrate that the defects in sliding are not solely due to binding (though nucleosome interactions are decreased with the mutants), and therefore the defect is catalytic in nature. The authors find that a common mutant of the autoinhibitory "auto-N" motif (2RA) can rescue the remodeling defects, and they hypothesize a direct connection between basic region of the auto-N and acidic patch of H2A.

1) While I agree with the general conclusions, the single molecule experiments presented in Figure 3 are harder to interpret. The authors propose that the E64R mutant increases the pause durations (which would be similar to what was observed for H4∆tail and short DNA linker nucleosomes in smFRET experiments; see Hwang et al., 2014). The author report that nucleosomes are shifted in steps, as expected for ISWI, and also the intriguing possibility that the E64R mutant alters the step size. A significant concern of mine, however, is that the example traces showing steps (e.g. Figure 3B) were rather noisy, and I would like to see more examples of the steps (as a supplement). Also, the HMM step sizes should agree with step sizes that one can obtain from the step finding algorithm published by Kerssemakers et al., (2006). If the authors assemble the steps into histograms based on FRET values at each step, it should be apparent what the preferential steps are and whether they change with AP nucleosomes. Such analysis would be more informative and appropriate than the cumulative probability plots shown in Figure 3E.

2) On a related point, in the Discussion section, the description of sliding the E64R mutant is confusing with regard to increased pausing frequencies. The authors describe an apparent change in the stepsize in nucleosome sliding, which may or may not result from increased frequency of pausing, but no data has been presented for pausing frequency (distinct from pausing duration). Figure 3E is discussed with regard to the distances translocated, not on frequencies of pausing, so this point is unclear.

3) Based on the core thesis of their work, that the acidic patch relieves inhibition by auto-N and therefore accelerates overall sliding, I think it would be worthwhile to investigate sliding behavior of H2AZ nucleosomes, which are known to have a more extensive acidic patch region. The Kingston lab (Goldman et al., 2010) showed that ISWI is more active with H2AZ nucleosomes. They found that the "additional acidic residues" of H2AZ were required for the higher sliding rate, but somewhat unexpectedly, adding just the additional acidic residues of H2AZ to normal H2A did not appear to alter sliding. While such experiments may not lend new insight for ISWI, H2AZ is a natural substrate for INO80 and therefore it would be worthwhile to compare sliding rates for H2A, H2AZ, and H2AZ-NK (acidic patch=normal H2A).

4) Figure 3—figure supplement 3A: The authors state that 2-dye nucleosomes were excluded from analysis, so that the FRET differences from distal and proximal dyes could be properly accounted for. This is fine, but it is unclear how both dyes together affect the resulting initial FRET. Was there some mechanism for ensuring that traces used for the remodeling experiments were also from single-dye nucleosomes?

5) Figure 3—figure supplement 5: why are the heights of the peaks from the two differently positioned dyes so different. Shouldn't these have similar densities? The 12/78 seems somewhat odd with the closer dye yielding a much lower peak than the more distal dye. Can the authors comment on this?

6) The authors state in the Discussion section "…acidic patch mutations increase both the durations and frequencies of pauses." The authors only present the E64R mutant in smFRET experiments, and so should either present data for other mutants or adjust this sentence accordingly.

Reviewer 3

In this manuscript it is shown that mutation of residues that comprise the considered acidic patch on the nucleosome affect the ability of some, chromatin remodeling enzymes to act on nucleosomes bearing these nucleosomes. This is interesting as this region of the nucleosome has previously been shown to act as a binding interface for other types of nucleosome binding proteins. However, the acidic patch is not universally required for remodeling enzymes as the enzyme Chd1 has previously been shown to act on nucleosomes where this interface is altered. Also, when SNF2h is present as part of the ACF complex the effect of the AP region is reduced 20-fold.

The functional significance of this interaction is investigated further by studying the rate at which a more active mutant of the SNF2h enzyme (AutoN) remodels nucleosomes. This mutant enzyme is found to be less sensitive to alterations to the acidic patch on the nucleosome. As a result, a model is proposed in which AutoN interacts with the acidic patch in the active state, but switches back to an inhibitory conformation during cycles of enzyme activity. Support for this is gained from single molecule studies showing that pausing is increased by mutation of the AP surface. Although the AutoN mutation does not affect pausing on normal nucleosomes, it does act to restore normal enzyme action on AP mutant nucleosomes.

It's known that the AutoN loop interacts with ATPase lobe 2 and that the histone H4 competes for engagement with this region (Yan et al. 2016). Here it is proposed that the acidic patch region acts as an alternative location via which both AutoN and the H4 tail can interact. A major weakness of the manuscript is that no evidence is presented to directly support direct association of these regions. Ideally, dynamic association of AutoN and the H4 tail with the AP region would be directly measured during the reaction cycle. The authors have established a single molecule FRET system that could be used to make this type of measurement, it would none the less be a substantial amount of new work to include this type of work.

Some form of evidence should be obtained to indicate that AutoN and the AP region interact directly. There are a number of different approaches that could be used for this including FRET, X-link MS or pull down. This would provide important evidence supporting the model presented in Figure 5.

The single molecule FRET measurements appear to indicate different lengths of DNA transit over the nucleosome, and that there may be differences from previously published data. This is potentially very interesting. However, the noise in the individual traces appears to be high. The differences in the movements and the errors associated with the measurements should be established rigorously. The authors should also discuss whether the differences in DNA movements are likely to involve differences in the rate at which elementary steps are generated, or the time during which SNF2H is committed to single base steps.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "The nucleosomal acidic patch relieves auto-inhibition by the ISWI remodeler SNF2h" for further consideration at eLife. Your revised article has been favorably evaluated by John Kuriyan (Senior editor), a Reviewing editor (Jerry Workman), and three reviewers.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

1) The EDC cross-linking/mass spec experiments provided new and interesting information regarding how Snf2h binds to the nucleosome. Some of those data, however, appeared to be somewhat selectively interpreted, biased toward the authors' model. While cross-linking of the AutoN domain was observed with the acidic patch, cross-links were also observed for other regions of Snf2h, such as the HAND/SANT/SLIDE (HSS) domain and RecA domains. In addition, a significant number of intramolecular cross-links were observed from the AutoN to other parts of Snf2h. Cross-links closest to 2RA are to other parts of Snf2h rather than the acidic patch. The authors should discuss the potential significance of these interactions. It would be helpful if the authors could devise a figure to more precisely indicate locations of cross-links with respect to 2RA and NegC.

2) There were many fewer cross-links in the ADP compared to the ADP-BeFx complex, which makes it unclear whether the ratios shown in Figure 2B are the most appropriate way to present nucleotide-dependent differences. Were those data normalized? If not, could a set of cross-links not expected to change in response to nucleotide (e.g. Snf2h intradomain, or histone-histone) be used as an internal standard for normalizing the ADP and ADP-BeFx data?

3)The authors should consider additional experiments to determine whether the HSS domain might be the element that directly interacts with the acidic patch. One experiment would be to test whether the 2RA mutant of the AutoN can rescue the Lys-to-Ala mutants in the HSS that the authors' found disrupt activity. A rescue would signify that AutoN bypass is downstream of HSS regulation. Alternatively, the HSS mutants that fail to slide nucleosomes could be compared to wt in EDC cross-linking/mass spec experiments. Selective loss of other HSS cross-links with the acidic patch would support a direct interaction. These experiments are not a requirement for publication, but if feasible on a short time scale could provide additional valuable information.

https://doi.org/10.7554/eLife.35322.043

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

Reviewer 1

This is an exciting manuscript that takes powerful biochemical and biophysical approaches to analyze details of nucleosome remodeling by the Snf2h (Iswi) remodeling ATPase. The results implicate a central role of the acidic patch on the H2A H2B dimer as an important epitope in receiving auto-inhibition of remodeling activity. This effect is mediated by increased pause and decreased steps in nucleosome translocations. The authors go on to show that the ACF complex and Ino80 are also dependent on the acidic patch for remodeling. Some points that need to be addressed.

We are glad the reviewer finds the work exciting and thank the reviewer for the helpful comments and suggestions, which we address below.

1) This paper needs to include some biochemical characterization of the APM nucleosomes compared to WT. Are they as stable for example at higher salt, etc. While the location of the mutations suggests they should not affect nucleosome stability, this can't be assumed. Moreover, the fact that Ino80 remodeling is also affected by the mutations without having similar Auto inhibitory domains suggest that the effect of the mutations is on the nucleosome substrate rather than through the remodelers.

The reviewer raises a good point. To test for stability effects we have now assessed salt dependent dissociation of the nucleosome. We find no large differences in the dissociation of WT vs acidic patch mutant (APM) nucleosomes as a function of increasing salt (Author response image 1). In addition, since we can rescue the majority of the APM defect with mutations in AutoN and now NegC, we interpret this to mean that the main effect of the acidic patch mutation is not due to gross defects in the nucleosome substrate. Furthermore, CHD1 can also efficiently remodel APM nucleosomes (Levendosky et al., 2016) suggesting that these mutations do not grossly destabilize nucleosomes. We also recognize the reviewer’s point about the effects on INO80. However, given that many nucleosome binding proteins have been shown to directly contact the acidic patch we speculate that a specific region of the INO80 complex contacts the acidic patch during remodeling. We anticipate that future work will identify this region of INO80.

Author response image 1
Salt-based nucleosome dissociation of WT and APM nucleosomes.

A. 15 nM 0/60 cy3-DNA labeled Nucleosomes were incubated in 25 mM HEPES pH 7.5 and varing concentrations of NaCl at 20ºC and resolved on a 6% native polyacrylamide gel. The gels were scanned for the cy3 label (green symbol on nucleosome and DNA cartoons on left of gel). B. Quantification of the fraction of nucleosomes remaining as a function of salt concentration. Fraction nucleosomes was determined by taking the ratio of the signal for the nucleosome bands to the signal for all the bands in each lane. Values were normalized to unincubated controls. The points and error bars plotted reflect the mean and standard error for n=3 experiments.

https://doi.org/10.7554/eLife.35322.031

2) The assumption that the single point mutant is simply a weaker version of the 4-point mutants in the acidic patch may not be true. It may act completely differently. This assumption would be stronger if additional single point mutants were examined.

We thank the reviewer for this suggestion as the experiment suggested by the reviewer has yielded a new insight. If the acidic patch residues act independently we would expect the individual mutations to have smaller defects than the combination of mutations and expect the individual defects to be additive. In contrast if the acidic patch residues act cooperatively, we would expect the individual mutations to have comparable defects as the combination of mutations. Interestingly, as shown in Figure 1—figure supplement 3, except for E64A, all the single alanine mutants have comparable defects as the four-point mutants combined. This observation suggests that 3 out of the 4 acidic patch residues act cooperatively.

3) The mechanism would be strengthened if the authors can demonstrate a direct interaction between AutoN and the acidic patch and that this is reduced in the mutants. Simple peptide pulldown assays with Auto N compared to the H4 tails would suffice.

The reviewer makes a very valid suggestion. We did attempt to investigate by pull-down assays whether an AutoN peptide containing the 2R residues directly interacts with acidic patch nucleosomes but were unable to observe detectable binding. A crystal structure of the relevant AutoN region has been published showing that the 2R residues are within a loop region (Yan et al., 2016). We speculate that if the 2R residues are directly interacting with the acidic patch, this interaction may require the surrounding structured regions of AutoN for conformationally constraining the 2R residues in the loop.

However, to address the general point made by the reviewer of direct interaction we have carried out a cross-linking mass spec analysis in the presence of ADP and ADP-BeFx. We used the zero-length crosslinking system of EDC-NHS, which catalyzes amide bond formation between acidic residues and lysines. Substantial previous work has suggested that ADP-BeFx mimics an activated state of the SNF2hnucleosome complex. We therefore determined the acidic patch cross-links that were specific to the ADP-BeFx state. This approach identified interactions between the acidic patch and both AutoN and the NegC region of SNF2h as being highly enriched in the ADP•BeFx state (Figure 2B). This comparative cross-linking approach also identified another domain interaction that is enriched in the ADP•BeFx state and is consistent with an activated state, namely the H4-tail binding one of the SNF2h RecA domains.

Additionally, we now show that mutagenesis of NegC rescues the defect of the acidic patch to a greater extent than the 2R residues in AutoN (Figure 2). These results are consistent with the interaction between NegC and the acidic patch playing a mechanistically significant role. This data does not rule out a direct interaction with the 2R residues in AutoN because the cross-linking chemistry cannot generate cross-links between Arginines and acidic residues. However, the new data is consistent with a more complex model than we had previously proposed. We now propose a revised model, in which the acidic patch promotes a relief of auto-inhibition by both the NegC region and the AutoN region.

We are very grateful to the reviewer for suggesting that we look for a direct interaction as the data from this search has led to a more sophisticated mechanistic model.

4) The authors need to make the case that studying hSn2h in the absence of its partner proteins is warranted. The fact that ACF (of which hSf2h is a subunit) is much less dependent on the acidic patch than hSn2h alone could be construed as suggesting these are artifacts of studying hSnf2h out of its normal context.

We acknowledge the reviewer’s skepticism about looking at the SNF2h protein alone. However, we believe that studying how the core ATPase functions can help better understand how additional subunits like Acf1 regulate the intrinsic properties of SNF2h. A study by the Muir group that was published in Nature after we received the first round of eLife reviews also shows smaller effects of the acidic patch with ACF compared to SNF2h. However, they report that with other SNF2h complexes such as RSF and NURF the effect of mutating the acidic patch is comparable to that with SNF2h. Thus, we view the effects on SNF2h as a mechanistically meaningful baseline. We note that the Muir study does not provide a mechanistic explanation for the acidic patch effects. Our revised work directly addresses the mechanistic basis by showing how the acidic patch relieves autoinhibition in SNF2h. The model we propose for SNF2h alone can provide a foundation for future studies to understand why other SNF2hcontaining complexes are more sensitive to acidic patch mutations than the ACF complex.

Reviewer 2

The main findings of this work are that the ISWI and INO80 remodelers are sensitive to the acidic patch on histone H2A. The ISWI ATPase subunit, SNF2h, is more extensively studied, where it is shown that nucleosome sliding is dramatically reduced with a quadruple acidic patch mutant. A nice complementary experiment is the competition with the LANA peptide, which is known to bind to the acidic patch, and therefore serves as a confirmation that this region is important for nucleosome sliding by ISWI. By using saturating amounts of remodeler, the authors demonstrate that the defects in sliding are not solely due to binding (though nucleosome interactions are decreased with the mutants), and therefore the defect is catalytic in nature. The authors find that a common mutant of the autoinhibitory "auto-N" motif (2RA) can rescue the remodeling defects, and they hypothesize a direct connection between basic region of the auto-N and acidic patch of H2A.

1) While I agree with the general conclusions, the single molecule experiments presented in Figure 3 are harder to interpret. The authors propose that the E64R mutant increases the pause durations (which would be similar to what was observed for H4∆tail and short DNA linker nucleosomes in smFRET experiments; see Hwang et al., 2014). The author report that nucleosomes are shifted in steps, as expected for ISWI, and also the intriguing possibility that the E64R mutant alters the step size. A significant concern of mine, however, is that the example traces showing steps (e.g. Figure 3B) were rather noisy, and I would like to see more examples of the steps (as a supplement).

We are glad the reviewer agrees with our general conclusions and address the reviewer’s concerns about the single-molecule experiments below.

As the reviewer suggested, we have added additional example traces in a new supplemental figure, Figure 3—figure supplement 5. We believe one reason the E64R trace in the original version of Figure 3 in particular appears noisy is because of the very long-time scale on the x-axis compared to the other example traces in Figure 3B. In addition to providing more example traces, we have replaced some of the example traces in Figure 3B with ones that span, as much as possible, roughly equivalent durations, so that they have similar x-axes and are more visually comparable.

We note that all of the example traces in Figure 3B and in Figure 3—figure supplement 5 are smoothed for visualization only (not for analysis). The apparently noisiness of example traces are subject to effects of the smoothing filter used, the linewidth used to plot the data, the scaling of the axes, and so forth. We have plotted our example traces with a fairly conservative smoothing filter so as not to obscure features in the data. In addition, it is usually possible to decrease the noise in smFRET data by increasing the power of the imaging laser, with a trade-off in terms of faster photobleaching. However, in our case we kept the laser as low as possible, while still maintaining enough signal to noise that our HMM algorithm could detect pauses (see below), because the extremely slow remodeling with the E64R nucleosomes necessitated keeping the photobleaching rate as low as possible. Even so, as noted in the text, we cannot capture all of the remodeling reaction with E64R, because the FRET dyes often photobleach faster than SNF2h can remodel the E64R nucleosomes out of FRET range. For all of these reasons, our example traces may appear noisier than traces in other published smFRET work.

The reviewer’s concern about the noisiness of the traces may reflect a concern that our HMM fitting routine is unable to robustly detect steps, particularly in the E64R data, where the smaller distances translocated between pauses means the pause states are at closer FRET values to one another. This concern is a valid one. To address the effect of noise on the robustness of our HMM fitting algorithm (pyhsmm), we added synthetic noise to the smFRET data with WT SNF2h and the E64R nucleosomes. As shown in Author response image 2A and B, 50% more noise in the E64R/WT data does not, within error, affect either the durations of the pauses, or their location (as reflected by how far the nucleosome is moved between pauses), as quantified by pyhsmm. When the noise is doubled, as in Author response image 2C and D, the outcome of pyhsmm is still largely comparable to the original data, although some pauses are missed (as indicated by the longer p1 pause duration and larger step size between wait and p1). pyhsmm is thus extremely robust, even with noise levels significantly higher than those in our data.

Author response image 2
The addition of synthetic noise does not significantly change the outcome of our HMM fitting routine.

Synthetic, uncorrelated, Gaussian-distributed noise was added independently to the Cy3 and Cy5 intensities for each E64R/WT trace, and then the pyhsmm analysis was re-run on these noisier data. Pause durations and changes in nucleosome positions (step sizes) were then re-computed and are here plotted as in Figure 3 in the main text. In A and B, the standard deviation of the added Gaussian noise was 0.4; in C and D, the standard deviation of the added noise was 0.6. The standard deviations of the original Cy3 and Cy5 intensity data were on average 0.5 and 0.3 respectively, so the data in A and B have about 50% more noise than the original data, while the data in C and D have about double the noise of the original data.

https://doi.org/10.7554/eLife.35322.032

Also, the HMM step sizes should agree with step sizes that one can obtain from the step finding algorithm published by Kerssemakers et al., (2006).

In Author response image 3 we show the results of the Kerssemakers algorithm for several example traces with WT SNF2h and WT nucleosomes, compared to the outcome of our pyhsmm hidden Markov model analysis. Overall the two algorithms are in good agreement. However, the Kerssemakers algorithm, which was originally designed for optical trapping data (that is, one-dimensional data), works only with 6 the FRET values for each trace. Our HMM jointly fits the Cy3 and Cy5 intensity data, which means it uses more information to find steps, which allows it to more accurately detect real steps and avoid false positives. Where the results of the Kerssemakers algorithm and pyhsmm differ, as in the bottom right trace in Author response image 3, an inspection of the Cy3 and Cy5 intensity traces show that pyhsmm fits the Cy3 and Cy5 intensities well.

We note that all of the fits obtained from our HMM algorithm were visually inspected and confirmed to be reasonable and have added a clarification to the Materials and methods section to this effect.

Author response image 3
Comparison of the fit obtained by our HMM algorithm (pyhsmm) to that obtained by the Kerssemakers algorithm.

Cy5 intensities (red data), Cy3 intensities (green data), and FRET values (blue data) are shown for 4 of the example traces in Figure 3 and Figure 3—figure supplement 5. Data are plotted as in those figures (including a 0.95-second smoothing filter, for visualization only). Dashed red and green lines in top panels show fits generated by pyhsmm to the Cy3 and Cy5 intensity data; dashed gray lines in the bottom panels show pyhsmm results in terms of FRET. The Kerssemakers algorithm acts only on the FRET values, so the outcome of Kerssemakers is shown on the bottom FRET panels only, as solid black lines.

https://doi.org/10.7554/eLife.35322.033

If the authors assemble the steps into histograms based on FRET values at each step, it should be apparent what the preferential steps are and whether they change with AP nucleosomes. Such analysis would be more informative and appropriate than the cumulative probability plots shown in Figure 3E.

We acknowledge the reviewer’s suggestion for generating histograms based on FRET values, but we prefer to display the same information in terms of CDFs based on basepairs moved, for two reasons discussed in more detail below:

Author response image 4
Histograms of FRET values for the p1 and p2 pauses for WT SNF2h with WT nucleosomes.

A. Bin positions as in Blosser et al., 2009 (every 0.04 FRET, starting at 0 FRET). B. Bin positions every 0.05 FRET, starting at 0.04 FRET. Due to the non-linearity of our calibration curve, FRET values of pauses are not representative of preferential step locations. We show histograms of FRET values here simply to illustrate the effect of bin choice on the way the data appear, which we chose to do with reference to previously published histograms of FRET values from Blosser et al.,2009.

https://doi.org/10.7554/eLife.35322.034

1) Due to the non-linearity of our calibration curve, we feel that describing preferential steps in terms of FRET values will be misleading.

2) We prefer cumulative density functions (CDFs) to histograms, because where peaks or clusters of data appear in histograms is highly sensitive to choice of bin width and position (see Author response image 4; for reference, see also the Discussion section at http://www.andata.at/en/software-blog-reader/why-we-love-the-cdfand-do-not-like-histograms-that-much.html). CDFs provide a more quantitative way to compare distributions obtained under different conditions (like WT vs E64R nucleosomes), by avoiding potential artifacts generated by the binning procedure in making histograms.

With regards to analyzing steps in terms of FRET values: in previous work observing ISWI enzymes remodeling single nucleosomes (e.g. Blosser et al., 2009, Deindl et al., 2013), the relationship between FRET and bp was well-described by a linear approximation, as determined by a calibration curve obtained the way ours was (see Figure 3—figure supplement 4). Thus in, for example, Blosser et al., 2009, an initial 7 bp step resulted in a change in FRET that was twice as large as a subsequent 3 bp step; that is, an initial step twice as large in bp resulted in a change in FRET twice as large as well. This linear relationship between FRET and bp explains why the first peak in Blosser et al. 2009 Figure 3b is at ~0.55 FRET (whereas in our system it is at ~0.75 FRET). In contrast, our calibration curve demonstrates that FRET and bp are not linearly related but rather follow the expected R-6 relationship, with our nucleosomes starting outside of the pseudo-linear regime. Nucleosomes in a single data set will not all start at exactly the same FRET value, and since they start outside the pseudo-linear range, even if they are moved the same number of basepairs by SNF2h, they will not all have the same change in FRET. So, to compare nucleosomes both within a dataset and between datasets we converted the steps to bp first. In our system, the initial ~7 bp step and the subsequent ~3 bp step both lead to FRET changes of similar magnitudes, because of how these steps move the nucleosome along the calibration curve. We believe our FRET calibration curve differs from that in Blosser et al., 2009 in part because we use a different labeling scheme where histone H3 is labeled instead of histone H2A and because the chemistry of the dye attachment to DNA differs from Blosser et al.

With regards to histograms versus CDFs: as an example of the sensitivity of histograms to bin width and position, in Figure R4 we compare histograms of FRET values for the p1 and p2 pauses for WT SNF2h with WT nucleosomes, using two different choices for the bins. In A, we plot our data as in Blosser et al., 2009, Figure 3B, in which the initial FRET values (i.e. the FRET values of the wait pause) are excluded, and p1 and p2 pause FRET values are binned at every 0.04 FRET. The data do not clearly fall into Gaussian-like peaks. However, in Author response image 4B, we have chosen bin widths and locations that make two Gaussian-like peaks appear in the data. We do believe these particular peaks represent two real underlying clusters of FRET values in the data, based on the CDF analysis. However, it is very difficult to determine bin sizes and locations that reflect a best guess of the “true” clusters in the data from histogram analyses alone.

Because the step size of ISWI-family remodelers has been previously established in the literature (Blosser et al., 2009, Deindl et al., 2013), in our original manuscript we focused on comparing distributions of step sizes obtained with the different SNF2h and nucleosomal constructs. For such quantitative comparisons between data sets, we turned to the cumulative distribution function, or CDF, which does not suffer from the same binning pitfalls as histograms. A CDF represents the same information as a histogram, but without any smoothing or binning of the data. Peaks or clusters in the data appear as steep slopes in the CDF, whereas dips or separations between clusters appear as flat portions (see Figure 3—figure supplement 4 and the new Figure 3—figure supplement 6).

While the average step size (as well as other nuances of the step size distribution data) can be read off of the CDFs, the CDFs in Figure 3E emphasize differences between distributions of step sizes for different data sets, rather than the average step size for any particular data set. To emphasize the preferred step size for each condition, we have added a new supplemental figure, Figure 3—figure supplement 6, which represents the same step-size data in Figure 3E in three different visualizations: Kernel Density Estimation Plots (KDEs), CDFs and histograms. In that figure we also report the average step size for each data set, with the step size for WT SNF2h and WT nucleosomes in good agreement with step sizes reported for the ACF complex and other ISWI family remodelers.

2) On a related point, in the Discussion section, the description of sliding the E64R mutant is confusing with regard to increased pausing frequencies. The authors describe an apparent change in the stepsize in nucleosome sliding, which may or may not result from increased frequency of pausing, but no data has been presented for pausing frequency (distinct from pausing duration). Figure 3E is discussed with regard to the distances translocated, not on frequencies of pausing, so this point is unclear.

We thank the reviewer for pointing out this inconsistency in our terminology and have clarified the text by removing reference to pausing frequency. We now describe remodeling with the E64R nucleosomes resulting in shorter distances translocated between pauses, resulting in more pauses encountered per distance translocated.

3) Based on the core thesis of their work, that the acidic patch relieves inhibition by auto-N and therefore accelerates overall sliding, I think it would be worthwhile to investigate sliding behavior of H2AZ nucleosomes, which are known to have a more extensive acidic patch region. The Kingston lab (Goldman et al., 2010) showed that ISWI is more active with H2AZ nucleosomes. They found that the "additional acidic residues" of H2AZ were required for the higher sliding rate, but somewhat unexpectedly, adding just the additional acidic residues of H2AZ to normal H2A did not appear to alter sliding. While such experiments may not lend new insight for ISWI, H2AZ is a natural substrate for INO80 and therefore it would be worthwhile to compare sliding rates for H2A, H2AZ, and H2AZ-NK (acidic patch=normal H2A).

The reviewer raises a good point. The work from the Kingston lab is consistent with our current model for how the acidic patch relieves auto-inhibition by SNF2h. The comparable INO80 experiment suggested by the reviewer has recently been carried out by the Bartholomew lab (Brahma et al., 2017). They find that INO80 preferentially slides H2AZ nucleosomes over H2A nucleosomes.

4) Figure 3—figure supplement 3A: The authors state that 2-dye nucleosomes were excluded from analysis, so that the FRET differences from distal and proximal dyes could be properly accounted for. This is fine, but it is unclear how both dyes together affect the resulting initial FRET. Was there some mechanism for ensuring that traces used for the remodeling experiments were also from single-dye nucleosomes?

We thank the reviewer for pointing to an aspect of our analysis that we did not properly clarify. Specifically, the same process by which two-dye nucleosomes were excluded from the FRET kernel density plots, namely, the exclusion of trajectories with two-step photobleaching events, was also used to exclude such nucleosomes from remodeling experiments. All movies of remodeling were of such duration as to ensure that all dyes photobleached before the end of the movie, so that we could be sure to exclude any nucleosomes with two Cy3 dyes. Moreover, in contrast to earlier work (e.g. Blosser et al., 2009), we used a 2:1 unlabeled H3:labeled H3, instead of a 1:1 ratio, resulting in very few doubly-labeled nucleosomes. We have clarified the text accordingly.

5) Figure 3—figure supplement 5: why are the heights of the peaks from the two differently positioned dyes so different. Shouldn't these have similar densities? The 12/78 seems somewhat odd with the closer dye yielding a much lower peak than the more distal dye. Can the authors comment on this?

The reviewer is correct that, in theory, nucleosome assembly should result in roughly equal proportions of proximally labeled and distally labeled nucleosomes. However, in our hands nucleosomes assemble with a preference for the H3 histone label to be away from the DNA label. Unequal assembly preference has also been observed by others (e.g., Qiu et al., 2017).

With regards to the 12/78 calibration data in particular: we removed this point and then repeated the calibration curve fit, and obtained similar fit parameters as when the 12/78 data were included (without 12/78, R0 = 11.3 ± 1.1 nm, d0,prox = 5.9 ± 1.1 nm, θprox = 180.0 ± 19.5°, d0,dist = 11.0 ± 1.0 nm, θdist = 89.1 ± 6.9°, compared to R0 = 10.9 ± 1.1 nm, d0,prox = 5.8 ± 1.0 nm, θprox = 153.8 ± 20.6°, d0,dist = 10.6 ± 1.0 nm, θdist = 87.3 ± 12.0° in the text).

6) The authors state in the Discussion section "…acidic patch mutations increase both the durations and frequencies of pauses." The authors only present the E64R mutant in smFRET experiments, and so should either present data for other mutants or adjust this sentence accordingly.

We thank the reviewer for pointing out this lack of clarity. We now state that we use the smFRET data with E64R to infer the role of the acidic patch.

Reviewer 3

In this manuscript it is shown that mutation of residues that comprise the considered acidic patch on the nucleosome affect the ability of some, chromatin remodeling enzymes to act on nucleosomes bearing these nucleosomes. This is interesting as this region of the nucleosome has previously been shown to act as a binding interface for other types of nucleosome binding proteins. However, the acidic patch is not universally required for remodeling enzymes as the enzyme Chd1 has previously been shown to act on nucleosomes where this interface is altered. Also, when SNF2h is present as part of the ACF complex the effect of the AP region is reduced 20-fold.

The functional significance of this interaction is investigated further by studying the rate at which a more active mutant of the SNF2h enzyme (AutoN) remodels nucleosomes. This mutant enzyme is found to be less sensitive to alterations to the acidic patch on the nucleosome. As a result, a model is proposed in which AutoN interacts with the acidic patch in the active state, but switches back to an inhibitory conformation during cycles of enzyme activity. Support for this is gained from single molecule studies showing that pausing is increased by mutation of the AP surface. Although the AutoN mutation does not affect pausing on normal nucleosomes, it does act to restore normal enzyme action on AP mutant nucleosomes.

It's known that the AutoN loop interacts with ATPase lobe 2 and that the histone H4 competes for engagement with this region (Yan et al. 2016). Here it is proposed that the acidic patch region acts as an alternative location via which both AutoN and the H4 tail can interact. A major weakness of the manuscript is that no evidence is presented to directly support direct association of these regions. Ideally, dynamic association of AutoN and the H4 tail with the AP region would be directly measured during the reaction cycle. The authors have established a single molecule FRET system that could be used to make this type of measurement, it would none the less be a substantial amount of new work to include this type of work.

We thank the reviewer for their comments and acknowledge that evidence for a direct contact with the acidic patch and AutoN would strengthen the manuscript. We describe below how we have addressed these and other concerns.

Some form of evidence should be obtained to indicate that AutoN and the AP region interact directly. There are a number of different approaches that could be used for this including FRET, X-link MS or pull down. This would provide important evidence supporting the model presented in Figure 5.

The reviewer makes a very valid suggestion. We did attempt to investigate by pull-down assays whether an AutoN peptide containing the 2R residues directly interacts with acidic patch nucleosomes but were unable to observe detectable binding. A crystal structure of the relevant AutoN region has been published showing that the 2R residues are within a loop region (Yan et al., 2016). We speculate that if the 2R residues are directly interacting with the acidic patch, this interaction may require the surrounding structured regions of AutoN for conformationally constraining the 2R residues in the loop.

However, to address the general point made by the reviewer of direct interaction we have carried out a cross-linking mass spec analysis in the presence of ADP and ADP-BeFx. We used the zero-length crosslinking system of EDC-NHS, which catalyzes amide bond formation between acidic residues and lysines. Substantial previous work has suggested that ADP-BeFx mimics an activated state of the SNF2-hnucleosome complex. We therefore determined the acidic patch cross-links that were specific to the ADP-BeFx state. This approach identified interactions between the acidic patch and both AutoN and the NegC region of SNF2h as being highly enriched in the ADP•BeFx state (Figure 2B). This comparative cross-linking approach also identified another domain interaction that is enriched in the ADP•BeFx state and is consistent with an activated state, namely the H4-tail binding one of the SNF2h RecA domains.

Additionally, we now show that mutagenesis of NegC rescues the defect of the acidic patch to a greater extent than the 2R residues in AutoN (Figure 2). These results are consistent with the interaction between NegC and the acidic patch playing a mechanistically significant role. This data does not rule out a direct interaction with the 2R residues in AutoN because the cross-linking chemistry cannot generate cross-links between Arginines and acidic residues. However, the new data is consistent with a more complex model than we had previously proposed. We now propose a revised model, in which the acidic patch promotes a relief of auto-inhibition by both the NegC region and the AutoN region.

We are very grateful to the reviewer for suggesting that we look for a direct interaction as the data from this search has led to a more sophisticated mechanistic model.

The single molecule FRET measurements appear to indicate different lengths of DNA transit over the nucleosome, and that there may be differences from previously published data. This is potentially very interesting.

We believe the reviewer is referring to the difference in the distances the nucleosome is moved with each translocation event for E64R versus WT nucleosomes, which we agree is quite interesting. However, we would like to clarify that we find the distance the nucleosome is moved with WT SNF2h and WT nucleosomes to be consistent with previously published data. We have edited the text to make this point more clear and have included a new Figure 3—figure supplement 7 that addresses this point (see also below).

However, the noise in the individual traces appears to be high.

We believe one reason the E64R trace in the original version of Figure 3 in particular appears noisy is because of the very long-time scale on the x-axis compared to the other example traces in Figure 3B. In addition to providing more example traces, we have replaced some of the example traces in Figure 3B with ones that span, as much as possible, roughly equivalent durations, so that they have similar x-axes and are more visually comparable.

We note that all of the example traces in Figure 3B and in Figure 3—figure supplement 6 are smoothed for visualization only (not for analysis). The apparently noisiness of example traces are subject to effects of the smoothing filter used, the linewidth used to plot the data, the scaling of the axes, and so forth. We have plotted our example traces with a fairly conservative smoothing filter so as not to obscure features in the data. In addition, it is usually possible to decrease the noise in smFRET data by increasing the power of the imaging laser, with a trade-off in terms of faster photobleaching. However, in our case we kept the laser as low as possible, while still maintaining enough signal to noise that our HMM algorithm could detect pauses (see below), because the extremely slow remodeling with the E64R nucleosomes necessitated keeping the photobleaching rate as low as possible. Even so, as noted in the text, we cannot capture all of the remodeling reaction with E64R, because the FRET dyes often photobleach faster than SNF2h can remodel the E64R nucleosomes out of FRET range. For all of these reasons, our example traces may appear noisier than traces in other published smFRET work.

The reviewer’s concern about the noisiness of the traces may also reflect a concern that our HMM fitting routine is unable to robustly detect steps, particularly in the E64R data, where the smaller distances translocated between pauses means the pause states are at closer FRET values to one another. This concern is a valid one. To address the effect of noise on the robustness of our HMM fitting algorithm (pyhsmm), we added synthetic noise to the smFRET data with WT SNF2h and the E64R nucleosomes. As shown in Author response image 2A and B, 50% more noise in the E64R/WT data does not, within error, affect either the durations of the pauses, or their location (as reflected by how far the nucleosome is moved between pauses), as quantified by pyhsmm. When the noise is doubled, as in Author response image2C and D, the outcome of pyhsmm is still largely comparable to the original data, although some pauses are missed (as indicated by the longer p1 pause duration and larger step size between wait and p1). pyhsmm is thus extremely robust, even with noise levels significantly higher than those in our data.

The differences in the movements and the errors associated with the measurements should be established rigorously.

We now more clearly explain in the methods how these errors were established. Specifically, the errors associated with our measurement of how far the nucleosome is moved with each translocation event were established by a bootstrapping procedure and are shown as shaded regions in the CDF plots of nucleosome movements (Figure 3E). These shaded reasons provide an estimate of the uncertainty in our measurement of the entire distribution of step sizes for each condition and show that the entire distribution of step sizes for the E64R nucleosomes with WT SNF2h are shifted to smaller step sizes, by a statistically significant amount.

We have also added a new Figure 3—figure supplement 6 which reports the mean step size and a standard error on that mean for each set of conditions. These mean step sizes similarly show that on average, the nucleosome is translocated a shorter distance with E64R nucleosomes and WT SNF2h (5.6 ± 0.3 bp for the first step) compared to WT SNF2h with WT nucleosomes (7.8 ± 0.3 bp). However, as shown in Figure 3-S7, the mean step size values alone do not do justice to how different the distributions of step sizes are with E64R versus WT nucleosomes, which is better reflected in the KDEs in Figure 3—figure supplement 7 or the CDFs in Figure 3E.

As shown in Figure 3—figure supplement 6, the step sizes we measure for WT SNF2h and WT nucleosomes are comparable to those reported in the literature for other ISWI family remodelers (Blosser et al., 2009, Deindl et al., 2013, Hwang et al., 2014). The small differences in measured step size compared to the literature likely reflect differences between our approach to determining step size compared to previous approaches in the literature. Specifically, in Blosser et al., 2009, Deindl et al., 2013 and Hwang et al., 2014, the FRET values for all pauses in all traces were combined into a single histogram. Several Gaussians were then fit to this histogram, and the differences between means of these Gaussians, converted from FRET to bp, were taken to be the sizes of each step. As noted above in a response to reviewer 2, where peaks appear in histograms—and therefore where Gaussians will be fitted—is strongly dependent on the bin sizes and locations used. Moreover, combining all pauses into one histogram means data for each pause may overlap. Here, we have instead separated out each pause (wait, p1, p2), and computed the change in bp between the first and second pauses, and between the second and third pauses. This allows us to report the mean step size as the mean value of each data set, rather than fitting a Gaussian to the decidedly non-Gaussian distributions in Figure 3—figure supplement 6.

The authors should also discuss whether the differences in DNA movements are likely to involve differences in the rate at which elementary steps are generated, or the time during which SNF2H is committed to single base steps.

The reviewer raises a good point which will help us clarify our discussion of the translocation events. Assuming SNF2h, like yeast ISWI remodelers (see Deindl et al., 2013), translocates the nucleosome in elementary steps of 1 or 2 bp each, the differences in DNA movements quantified in Figure 3E reflect how many of these elementary steps SNF2h takes during a translocation phase, before entering a new pause phase. Our data suggest that with E64R nucleosomes, SNF2h takes fewer elementary steps before entering a new pause phase. We have clarified this point in the Results section that discusses the step size. Specifically, we say: “Given that ISWI family remodelers have been shown to translocate a nucleosome in elementary steps of 1-2 bp (Deindl et al., 2013), our results suggest that with E64R nucleosomes, SNF2h takes fewer of these elementary steps in succession during the translocation phase, before entering a new pause phase.”

[Editors’ note: the author responses to the re-review follow.]

1) The EDC cross-linking/mass spec experiments provided new and interesting information regarding how Snf2h binds to the nucleosome. Some of those data, however, appeared to be somewhat selectively interpreted, biased toward the authors' model. While cross-linking of the AutoN domain was observed with the acidic patch, cross-links were also observed for other regions of Snf2h, such as the HAND/SANT/SLIDE (HSS) domain and RecA domains. In addition, a significant number of intramolecular cross-links were observed from the AutoN to other parts of Snf2h. Cross-links closest to 2RA are to other parts of Snf2h rather than the acidic patch. The authors should discuss the potential significance of these interactions. It would be helpful if the authors could devise a figure to more precisely indicate locations of cross-links with respect to 2RA and NegC.

The reviewers raise some excellent points. Before we address these points we’d first like to clarify that the heat map in Figure 2B is a mainly a visual aid to summarize two complex data sets and to highlight the major changes in domain interactions between ADP and ADP-BeFx nucleosome-SNF2h complexes in a way that is qualitatively intuitive. Abstracting and displaying the data this way however introduces some unavoidable distortions. For example, there are 8 crosslinked spectral counts identifying AutoN to H2A-Acidic Patch interaction in ADP-BeFx and 0 with ADP. Because zero values would lead to infinite ratios, we arbitrarily assign 1 count to the ADP-BeFx interaction. Normalization (which was conservative, see response to point 2), and the choice of 4-fold enrichment as the breakpoint for shading the darkest red color results in this interaction appearing less important than other domain-domain interactions which are colored more strongly. However, the underlying ratio is technically infinite. Hence, we also include the figure supplement, which shows a representation that is closer to the raw data with each dot representing a pair of crosslinked residues between the acidic patch and SNF2h.

In terms of the comment pertaining to the cross-links closest to 2RA, the residues nearest to these two arginines (R142 and R144) do actually crosslink directly to the H2A acidic patch. However, we realized that the original figure supplement did not easily allow an assessment of this issue. We have now added red lines to the figure supplement (Figure 2—figure supplement 4) to explicitly show the positions of R142 and R144 within the AutoN region of SNF2h and mentioned this clarification in the main text (subsection “The AutoN and NegC regions of SNF2h cooperate with the acidic patch to enable maximal remodeling”). Comparing the ADP and ADP-BeFx grids now more clearly shows that the residues nearest 2RA also crosslink directly to the H2A acidic patch, and only do so in the present of ADP-BeFx (left two panels of the supplement). We did not highlight NegC in the figure supplement, because the NegC domain is already demarcated by the domain boundaries on the right side of this figure (Figure 2—figure supplement 4).

In terms of the comment about intramolecular cross-links from the AutoN to other parts of Snf2h, it appears that our data representation may have miscommunicated the main point we were trying to make. While the reviewers are correct in that there are intra-molecular interactions suggested for the ADP and ADP-BeFx states separately, when we ask which AutoN interactions are significantly enriched with ADP-BeFx, these are to the acidic patch and to other parts of the nucleosome as shown in Figure 2B. AutoN to SNF2h crosslinks are otherwise light colored in the heatmap. We have now added a clarification sentence about these comparisons in the main text (subsection “The AutoN and NegC regions of SNF2h cooperate with the acidic patch to enable maximal remodeling”, Figure 2).

In terms of other SNF2h regions cross-linking to the acidic patch in the ADP-BeFx state, the reviewers are correct that such interactions are apparent in the data, for instance between SNF2h-HSS and the H2B acidic patch and RecA domains to the H2A acidic patch as well as other histones. As suggested by the reviewers we have expanded our discussion of the crosslinking data to include explanations for the potential significance of these cross-links (subsection “The AutoN and NegC regions of SNF2h cooperate with the acidic patch to enable maximal remodeling”).

2) There were many fewer cross-links in the ADP compared to the ADP-BeFx complex, which makes it unclear whether the ratios shown in Figure 2B are the most appropriate way to present nucleotide-dependent differences. Were those data normalized? If not, could a set of cross-links not expected to change in response to nucleotide (e.g. Snf2h intradomain, or histone-histone) be used as an internal standard for normalizing the ADP and ADP-BeFx data?

The reviewers’ suggestion to normalize based on domain interaction that is not expected to change is a good one in principle. However, in practice it is difficult to identify any such interactions for several reasons. For instance, intra-domain interactions can’t be assumed not to change because there are two protomers of SNF2h present and intra-protein SNF2h-SNF2h crosslinks cannot be resolved from inter-protein SNF2h-SNF2h crosslinks with the mass spectrometry method used here (Figure 2 legend). Furthermore, we note that previous work from our group has shown that distortions to the histone core occur in the ADP-BeFx state with SNF2h (Sinha et al., 2017) suggesting that intra-domain histone crosslinks cannot be assumed to remain constant across ATP states. That said, we note that the vast majority of the intra-domain SNF2h changes are not marked as significant using our method clarified further below (light colored tiles in the Figure 2B heat map).

Because we could not reliably normalize to an interaction not assumed to change, we chose a different approach of displaying the cross-linking data as shown in Figure 2B. This figure, which shows the crosslinking data aggregated into domain level interactions, is normalized such that the median Log2 value of spectral counts observed in ADP-BeFx to ADP is centered on 0; e.g. the assumption being made is that most domain-domain interactions will remain the same between the two conditions, and that the differences in numbers of crosslink counts is primarily due to experimental handling and instrument performance. Note that the un-normalized median of the Log2 spectral count ratio is 1.03, and all that has been done is to center the distribution (histogram in Figure 2B) on a value of 0. Thus, the correction being made is conservative in terms of which domain interactions we identify as significantly enriched in the ADP-BeFx condition.

In terms of the reviewers’ observation of fewer cross-links in the ADP state compared to the ADP-BeFx state, it is possible that this is a result of the slightly lower (~2-fold, Leonard and Narlikar, 2015) binding affinity observed between SNF2h and mononucleosomes in this condition. However, such a difference does not change our interpretation of this figure (Figure 2) significantly. The function of this figure is to serve as a heuristic with which to identify domain interactions that are specific to the two mechanistic states of the complex. The identity of the most and least enriched domain interactions will not change based on the normalization. The tiles of the heat map that we chose to color most intensely (log2 ratio > 2) are sufficiently separated from the bulk of the distribution that we are confident that they reflect real changes in the domain interactions of the assembly

3)The authors should consider additional experiments to determine whether the HSS domain might be the element that directly interacts with the acidic patch. One experiment would be to test whether the 2RA mutant of the AutoN can rescue the Lys-to-Ala mutants in the HSS that the authors' found disrupt activity. A rescue would signify that AutoN bypass is downstream of HSS regulation. Alternatively, the HSS mutants that fail to slide nucleosomes could be compared to wt in EDC cross-linking/mass spec experiments. Selective loss of other HSS cross-links with the acidic patch would support a direct interaction. These experiments are not a requirement for publication, but if feasible on a short time scale could provide additional valuable information.

We agree with the reviewers that it is important to determine whether the HSS may be directly interacting with the H2A acidic patch. While we have not performed the specific experiments suggested by the reviewers, we do present new data from a different experiment that tests which crosslinks are dependent on direct interactions with the H2A acidic patch (Figure 2—figure supplement 5). We find that addition of the LANA peptide to the SNF2h-nucleosome complex in the ADP-BeFx state significantly blocks acidic patch crosslinks to all regions of SNF2h except for the HSS. While this experiment does not rule out a direct HSS-acidic patch interaction, it suggests that the positioning of the HSS in the ADP-BeFx state is not solely determined by direct binding to the H2A acidic patch residues contacted by the LANA peptide.

https://doi.org/10.7554/eLife.35322.044

Article and author information

Author details

  1. Nathan Gamarra

    1. Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, United States
    2. Tetrad Graduate Program, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Resources, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing, Conceived of and conducted the biochemistry experiments and relevant data analysis, Wrote this manuscript
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2430-8662
  2. Stephanie L Johnson

    Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing, Conceived of and conducted the single molecule experiments and relevant data analysis, Helped write this manuscript
    Competing interests
    No competing interests declared
  3. Michael J Trnka

    Department of Pharmaceutical Chemistry, University of California, San Francisco, San Francisco, United States
    Contribution
    Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—review and editing, Designed the crosslinking experiments, Analyzed the mass spectrometry data, Developed analytical methods
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8808-5146
  4. Alma L Burlingame

    Department of Pharmaceutical Chemistry, University of California, San Francisco, San Francisco, United States
    Contribution
    Supervision, Funding acquisition, Project administration, Supervised the mass spectrometry data collection
    Competing interests
    No competing interests declared
  5. Geeta J Narlikar

    Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing, Supervised the overall study and helped write this manuscript
    For correspondence
    Geeta.Narlikar@ucsf.edu
    Competing interests
    Reviewing editor, eLife
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1920-0147

Funding

National Science Foundation (Predoctoral Fellowship)

  • Nathan Gamarra

University of California, San Francisco (Discovery Fellowship)

  • Nathan Gamarra

Leukemia and Lymphoma Society (Career Development Program Fellow Award)

  • Stephanie L Johnson

Dr. Miriam and Sheldon G. Adelson Medical Research Foundation

  • Alma L Burlingame

University of California, San Francisco (Program for Breakthrough Biomedical Research (PBBR))

  • Alma L Burlingame

National Institute of General Medical Sciences (R01GM073767)

  • Geeta J Narlikar

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Matthew Johnson for help in developing Traces, adapting his physmm code to smFRET data analysis, and the statistical analysis of smFRET data. We thank Song Tan for the Histone H2A E61A, E64A, D90A, D92A expression plasmid, and Greg Bowman, Robert Levendosky, and Cynthia Wolberger for purified Histone H2A E64R and H2A D90R/E92A. We thank Sebastian Deindl for providing an implementation of the Kerssemakers step-finding algorithm (Kerssemakers et al., 2006) in Matlab. We also thank Coral Zhou for INO80 purification, Julia Tretyakova for histone purification, Serena Sanuli for help generating some of the nucleosome substrates, and all members of the Narlikar Lab for helpful discussions. This work was supported by a grant from the NIH to GJN (R01GM073767), an NSF predoctoral fellowship and a UCSF discovery fellowship to NG, a Leukemia and Lymphoma Society Career Development Program Fellow award to SJ, and a grant from the Adelson Medical Research Foundation to ALB. The Thermo Scientific Fusion Lumos was funded by the UCSF Program for Breakthrough Biomedical Research (PBBR).

Reviewing Editor

  1. Jerry L Workman, Stowers Institute for Medical Research, United States

Publication history

  1. Received: January 25, 2018
  2. Accepted: April 16, 2018
  3. Accepted Manuscript published: April 17, 2018 (version 1)
  4. Version of Record published: May 30, 2018 (version 2)
  5. Version of Record updated: August 10, 2018 (version 3)

Copyright

© 2018, Gamarra et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,265
    Page views
  • 465
    Downloads
  • 12
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Chromosomes and Gene Expression
    2. Genetics and Genomics
    Teresa W Lee et al.
    Research Article
    1. Chromosomes and Gene Expression
    2. Genetics and Genomics
    Nealia CM House et al.
    Research Article