|Home | About | Journals | Submit | Contact Us | Français|
Evenly spaced nucleosomes directly correlate with condensed chromatin and gene silencing. The ATP-dependent chromatin assembly factor (ACF) forms such structures in vitro and is required for silencing in vivo. ACF generates and maintains nucleosome spacing by constantly moving a nucleosome towards the longer flanking DNA faster than the shorter flanking DNA. But how the enzyme rapidly moves back and forth between both sides of a nucleosome to accomplish bidirectional movement is unknown. We show that nucleosome movement depends cooperatively on two ACF molecules, suggesting that ACF functions as a dimer of ATPases. Further, the nucleotide state determines whether the dimer closely engages one vs. both sides of the nucleosome. Three-dimensional reconstruction by single particle electron microscopy of the ATPase-nucleosome complex in an activated ATP state reveals a dimer architecture in which the two ATPases face each other. Our results suggest a model in which the two ATPases work in a coordinated manner, taking turns to engage either side of a nucleosome, thereby allowing processive bidirectional movement. This novel dimeric motor mechanism differs from that of dimeric motors such as kinesin and dimeric helicases that processively translocate unidirectionally and reflects the unique challenges faced by motors that move nucleosomes.
Chromatin-remodeling motors play essential roles in organizing the chromatin state for regulating eukaryotic genomes, yet how they carry out their myriad activities is poorly understood. Their substrate, the nucleosome, contains 147 bp of DNA wrapped in ~1.5 turns around an octamer of histone proteins. Even the smallest movement of the histone octamer relative to the DNA presumably requires a coordinated process of breaking and reforming the many histone-DNA contacts. The ACF chromatin-remodeling complex exemplifies the task, as it is able to move nucleosomes to create evenly spaced nucleosomal arrays that contain equal DNA on either side of each nucleosome1-10. These evenly spaced arrays are important for packaging the underlying DNA into silent chromatin structures in vivo1-10.
ACF is part of the ISWI family of remodeling complexes. The ATPase subunits of ISWI complexes can move nucleosomes by themselves while the accessory subunits modulate this basic activity11-15. The human ACF complex consists of one ATPase subunit, SNF2h and one accessory subunit, Acf16,7. SNF2h is part of the SF2 family of DExx box proteins that includes helicases and nucleic acid translocases16. The ATPase domain of SNF2h has two RecA-like domains, which are thought to form a cleft within which ATP binds. SNF2h also has an alpha-helical extension comprised of three additional domains, HAND, SANT and SLIDE which are thought to play a role in binding flanking DNA17,18. We showed previously that ACF generates a dynamic equilibrium in which nucleosomes with equal flanking DNA on either side accumulate8. Our data implied that ACF achieves the dynamic equilibrium by constantly sampling either side of the nucleosome. This sampling mechanism raised the question of how ACF efficiently switches back and forth between both sides of a nucleosome. We hypothesized that understanding how the ATP state affects interactions of the enzyme with the nucleosome would provide insight into the sampling process.
Previous work has shown that ISWI enzymes require the H4 N-terminal tail for maximal activity19-23. The role of the H4 tail is not known, but it has been hypothesized that an acidic patch on the ATPase domain of ISWI enzymes may interact with the basic patch, K16R17H18R19, on the H4 tail17,19,20,21,24,25. These previous observations imply that the ATPase subunit contacts the H4 tail and that the contacts may change during the ATPase cycle. We therefore used changes in the mobility of the H4 tail as a handle to follow how changes in the nucleotide state alter interactions between SNF2h and the nucleosome. We used electron paramagnetic resonance (EPR) spectroscopy for these studies26. We covalently attached a maleimide spin label to a cysteine introduced in place of an alanine at position 15 on the H4 tail (A15C-MSL, Supplementary Fig.1b), which is directly adjacent to the basic patch. Thermal fluctuations cause a spin label attached to a protein to undergo motion in a spatial region defined by the adjacent protein surface. The resulting EPR spectrum is a highly sensitive measure of the region accessible to the probe. Conformational changes can thus be detected via changes in probe mobility, and these are monitored as changes in the EPR spectrum. EPR can also resolve and quantify multiple states and is particularly powerful in monitoring transitions between unstructured and structured regions of proteins. The A15C-MSL nucleosomes were assembled using an asymmetric DNA template comprising the 601 positioning sequence with 60 bp of flanking DNA on one side (0-601-60, Fig. 1a and Supplementary Fig. 1a)27. The presence of the probe did not alter the maximal rate of nucleosome remodeling by SNF2h (data not shown).
In the absence of SNF2h, the EPR spectrum of the A15C-MSL probe indicated a highly mobile probe (Fig. 1a, top spectrum). The high mobility of the probe in unbound nucleosomes suggested that the H4 N-terminal tails are largely unstructured. Next, we determined how binding of SNF2h altered the mobility of the H4 tail. When the nucleosomes were saturated with SNF2h in the absence of nucleotide (apo state), the EPR spectrum shows two sets of spectral components as indicated by the arrows (Fig 1a, middle spectrum). The inner spectral components (blue arrows) are indicative of a highly mobile probe whereas the wider set of spectral components (highlighted by the red dashed lines) and the broadening of the central peak are indicative of a second state with more restricted mobility. A given peak height in the left-most immobilized spectral component represents 4.1 times more spins than the same peak height for the mobile component. Deconvolution of the spectra indicated that about half of the H4 tails were in each of the two states (56±5.4% in the immobilized state, see Supplementary Fig. 2 for fitting and quantification method)28. In the presence of ADP, the immobilized subpopulation also constituted half of the probes (spectra not shown). Our attempts to trap the SNF2h-nucleosome complex in the ATP state using ATP analogs were unsuccessful as these analogs either supported low levels of remodeling (AMP-PNP, ATPγS) or did not bind with detectable affinity (AMP-PCP). We were however able to mimic an activated ATP state using the analog, ADP•BeFx. In contrast to the data in the apo state and with ADP, almost all of the probe on the H4 tail became immobilized in the presence of SNF2h and ADP•BeFx (91.5±2.6% of probe in the immobilized peak). This change is shown by the increase in spectral intensity of the immobilized component (left-most peak, Fig. 1a, bottom spectrum). This dramatic increase in the amount of probes immobilized indicated that both H4 tails were bound by SNF2h in the presence of ADP•BeFx. Together, these data indicate that SNF2h induces nucleotide-dependent changes in the H4 tail conformation such that in the apo and ADP states, half the H4 tails are immobilized and in an activated ATP state mimicked by ADP•BeFx, all H4 tails are bound.
The EPR data raised two possibilities for how SNF2h binds the nucleosome in the apo and ADP states: (a) SNF2h symmetrically binds both H4 tails and each H4 tail exists in a two-state equilibrium between mobile and immobile states (with an equilibrium constant of 1) or, (b) SNF2h asymmetrically binds only one of the two H4 tails. For model (a), we expect immobilization to increase upon lowering temperature as the highly mobile state is entropically favored whereas the structured immobile state is enthalpically favored, as seen for docking of the kinesin neck linker29. The fraction of H4 tails immobilized was unchanged, within error, from 23°C to 4°C (Fig. 1d, 54.6% immobilized at 4°C and 23°C). A van't Hoff plot of the equilibrium constant for H4 tail mobility as a function of temperature yields a ΔH of 0.76 kJ/mol, and ΔS of 4.4×10-3 kJ/mol-K (Supplementary Fig. 3), values that are substantially smaller than the favorable ΔH of 50 kJ/mol and unfavorable ΔS of 0.17 kJ/mol-K for docking of the kinesin neck linker29. These data rule out model (a) and provide strong support for the asymmetric binding of model (b). To further test model (b) we used hydroxyl radical footprinting of the same nucleosome construct to follow changes in ACF contacts as a function of nucleotide state (Fig. 1c). In the apo state, ACF binding induces asymmetric protection of nucleosomal DNA: protection is observed in the SHL(−2) region, but not the SHL(+2) region, consistent with other ISWI complexes and with model (b)9,30. In contrast, in the ADP•BeFx state, ACF binding results in significant protection in both SHL(−2) and SHL(+2) regions, consistent with the EPR data.
The asymmetry with respect to H4 tail binding observed in the apo state could arise either (a) due to the presence of asymmetric flanking DNA or (b) due to structural constraints placed by the apo state. To distinguish between these possibilities we repeated the EPR experiment using nucleosomes with 60bp of flanking DNA on both sides (60-601-60 template). Apo-SNF2h still bound only one of the two H4 tails in the context of this symmetric nucleosome (Fig. 1b). These data strongly support a model in which apo-SNF2h can only bind one H4 tail at a time, and the availability of flanking DNA biases which side of the nucleosome the enzyme binds preferentially. Together the above data suggest that the enzyme switches between an asymmetric conformation where it interacts with one H4 tail at a time in the apo state and a more symmetric conformation where it binds both H4 tails in the ADP•BeFx state.
The observation that SNF2h binds both H4 tails in the presence of ADP•BeFx suggests that either (a) one SNF2h molecule bridges both H4 tails or, (b) SNF2h binds as a dimer such that each ATPase contacts an H4 tail. To distinguish between these models we first investigated the oligomeric state of SNF2h alone. Using equilibrium analytical ultra-centrifugation we found that unbound SNF2h is a monomer (data not shown). Because several well-studied dimeric helicases dimerize upon binding their DNA substrates, we next determined if SNF2h dimerizes on nucleosomes31. If dimerization of SNF2h is tightly coupled to nucleosome binding we expected to see cooperative SNF2h binding. We measured the binding to nucleosomes by taking advantage of our observation that the fluorescence of a Cy3 dye attached near the entry site of the DNA increases upon SNF2h binding (Fig. 2a). We find that in the apo state, SNF2h binds to the nucleosome cooperatively, consistent with previous observations of cooperative binding by the Drosophila ISWI protein32. The Hill Coefficient of 1.8 suggests that at least two molecules of SNF2h bind in a manner such that binding of one molecule is strongly coupled to binding of the second (Fig. 2b). The EPR data from Figure 1 and the Hill Coefficient of 1.8, together suggest that in the apo state, SNF2h binds as a dimer but only one of the two SNF2h molecules engages an H4 tail.
To determine whether two SNF2h molecules were necessary to mediate maximal nucleosome remodeling, we measured the dependence of chromatin remodeling activity on SNF2h concentration using a FRET-based method (Fig. 2c). The rate constant of remodeling also depends cooperatively on SNF2h concentration with a Hill Coefficient of 1.8 (Fig. 2d, left panel). We next determined if the entire ACF complex also functions most effectively as a dimer. We analogously saw a cooperative dependence of the remodeling rate constant on ACF concentration with a Hill Coefficient of 1.9 (Fig. 2d, right panel). Together, these data strongly suggest that the predominant functional form of ACF is a dimer of ATPases.
A hallmark of dimeric motors such as kinesin and the E. coli Rep helicase is that they cycle between states in which one motor subunit is engaged with the substrate and states in which both motor subunits are transiently engaged31. By working in coordinated pairs, one motor subunit can serve as an anchor to the other moving motor to prevent dissociation from the substrate. The ATP state helps regulate the affinity of the motor for the substrate. Our EPR results suggest that a SNF2h dimer analogously cycles between at least two states, one in which only one ATPase engages an H4 tail and another in which both ATPases engage the two H4 tails. To determine if these different conformational states reflect states with different affinities, we measured the affinity of SNF2h for the nucleosome in different ATP states (Fig. 2e). In the presence of ADP, SNF2h bound cooperatively but with slightly weaker affinity than in the apo state. In both the apo and ADP states, the high cooperativity of binding indicates that binding of one SNF2h molecule by itself is very weak and requires the presence of another SNF2h molecule to increase its overall affinity. The cooperativity could arise either from direct SNF2h-SNF2h contacts or could be mediated through a conformational change in the nucleosome without direct SNF2h-SNF2h contacts. In the presence of ADP•BeFx, the K1/2 for SNF2h binding is ~3-fold lower than that in the apo state indicating a stronger binding affinity. Further, SNF2h binding in the presence of ADP•BeFx is not cooperative (Hill Coefficient =1). These data suggest that in presence of ADP•BeFx the affinity of each SNF2h molecule is sufficiently high such that binding of one SNF2h molecule is no longer highly dependent on the presence of the other.
Our finding that two ATPases are required for maximal nucleosome remodeling raises the question of how the nucleosome structure accommodates two SNF2h molecules. Other dimeric motors, such as kinesin and helicases, are oriented such that each ATPase subunit can take turns translocating on the polymeric substrate in the same direction31. We used negative stain electron microscopy to visualize the complex of SNF2h with nucleosomes in the presence of ADP•BeFx (Fig. 3 and Supplementary Fig. 4 and Supplemental Methods). Nucleosomes with 60bp of flanking DNA (0-601-60) were incubated with SNF2h concentrations comparable to the K1/2 for nucleosomes in the presence of ADP•BeFx, adsorbed to a glow discharged carbon film, and negatively stained with uranyl formate. The specimen was imaged at tilt angles of 60° and 0° (Supplementary Fig. 5). A total of 10,059 pairs of particles were interactively selected from 100 image pairs. Classification of particles from images of untilted specimen shows three distinct classes (Fig. 3a, b) that can be clearly recognized as a nucleosome by itself and a nucleosome with either one or two SNF2h molecules bound. On average ~70% of the complexes contained two SNF2h molecules bound. The singly bound SNF2h molecules could reflect the use of non-saturating SNF2h, a technical necessity to prevent particle crowding on the grid. In the complexes with two SNF2h molecules bound, the flanking DNA was not clearly visible in the two-dimensional (2D) class averages, possibly because the flanking DNA is flexible and gets averaged out. An alternative possibility is that in most of the complexes, the flanking DNA is rearranged due to interaction with a domain of SNF2h. Consistent with this possibility we do not clearly observe the extended HAND-SANT-SLIDE domain that has been shown to interact with flanking DNA in the apo state17. We hypothesize that there may be a conformational rearrangement of the HAND-SANT-SLIDE domain in the presence of ADP•BeFx. Further, no large region of direct contact between the two SNF2h molecules is apparent, consistent with the Hill Coefficient of 1.0 in this state (Fig. 2e).
In the 2D class averages, the region of each SNF2h monomer that interacts with the nucleosome appears to contain two globular lobes (Fig. 3a). These lobes may represent the two RecA-related ATP binding folds observed in SF2 family motors33,34. The two lobes are also apparent in the 2D class averages of SNF2h alone (Fig. 3c and Supplementary Fig. 6). Three-dimensional (3D) reconstructions of the nucleosome with two (or one) SNF2h bound were calculated using the well-established random conical tilt approach to a resolution of ~ 27 Å without the explicit application of any two-fold symmetry (Fig. 3a and Supplementary Figs 5 and 7)35. The two SNF2h molecules face each other on the nucleosome, and seem to obey the 2-fold symmetry of the nucleosome with one putative ATPase domain at SHL(+2) and one at SHL(−2) (Fig. 3a, 2-fold or opposing symmetry most apparent in the middle panel). Consistent with the EPR and footprinting data in Figure 1, each SNF2h monomer seems to directly contact one H4 N-terminal tail and the SHL(−2)/(+2) regions in the ADP•BeFx state. While the overall architecture appears almost symmetric, given the 27 Å resolution, any local structural asymmetries that may exist between the two SNF2h molecules cannot be resolved.
The above architecture raises the question of how the dimeric partners cooperate rather than compete in a “tug of war.” Our findings suggest an “alternating action” model schematized in Figure 4. In this model, each ATPase takes turns in engaging the flanking DNA on either side and the corresponding H4 tail at SHL(−2) or (+2)9,17. An ability of the two ATPases to take turns, as suggested by our observation that the apo state of the enzyme engages only one H4 tail at a time, would help avoid a “tug of war” situation. The ATPase that engages the longer DNA hydrolyzes ATP faster, as previously shown8,36. This ATPase becomes the leading ATPase and sets the direction of nucleosome movement by translocating on nucleosomal DNA37,38. The leading ATPase generates a DNA loop/wave that can propagate across the histone octamer as suggested previously39-41. The second, subordinate ATPase could then further act as another anchor to stabilize the intermediate while the leading ATPase is translocating (Fig. 4, mimicked by ADP•BeFx). Interaction with the second H4 tail may help in the binding of the subordinate ATPase. In the simplest version of this model, the subordinate ATPase does not bind or hydrolyze ATP once the leading ATPase fires42. This division of labor between identical subunits is analogous to hexameric helicases where occupancy of one ATPase subunit regulates the affinity of an adjacent subunit for nucleotide43. Whether the communication between the two ATPases is direct or through the nucleosome remains an important future question. A variation of this model in which the non-leading ATPase also hydrolyzes ATP is described in Supplementary Figure 8. Successive rounds of sampling and translocation would then equalize the DNA on either side of a nucleosome. A dimer-based mechanism is also indicated by single-molecule data showing that dimeric ACF complexes can switch the direction of nucleosome translocation several times without dissociation (supporting manuscript by Blosser et al.). Our results help explain the significance of previous observations that two Drosophila ACF molecules can bind in the context of DNA and provide a mechanistic explanation for the processive action of ISWI complexes3,44,45.
We hypothesize that in contrast to kinesin and dimeric helicases, whose biological functions require unidirectional translocation along a largely uniform polymeric substrate, the biological functions of chromatin remodeling enzymes like ACF place very different demands on motor architecture. The opposing architecture of the two motors in ACF may enable ACF to rapidly and processively change the direction of nucleosome movement in order to achieve a defined spacing. It will be interesting to investigate whether bidirectional movement via dimerization is a general feature of enzymes that space nucleosomes, and whether remodeling enzymes with other activities use different strategies.
EPR measurements were performed with an EMX EPR spectrometer from Bruker Instruments (Billerica, MA). First derivative, X-band spectra were recorded in a high-sensitivity microwave cavity using 50-s, 100-Gauss wide magnetic field sweeps. EM samples were adsorbed to a glow-discharged copper grid coated with carbon film for 30 seconds followed by conventional negative stain with 0.75% uranyl formate. Images were collected using a Tecnai T12 microscope (FEI company, Hillsboro, OR) and recorded at a magnification of 52,000X with an UltraScan 4096 × 4096 pixel CCD camera (Gatan Inc, USA). Full methods are described in Supplementary Methods.
We thank J. Widom (Northwestern University) for the 601 plasmid. We thank H. Madhani, M.D. Simon, and members of the Narlikar Lab for helpful discussion and comments on the manuscript. We thank R. Howard for help with AUC; W. Ross, J. Lin, S. Hota and B. Bartholomew for advice on footprinting; and C. Cunningham for assistance with nucleosome depiction. This work was supported by grants from the Sandler Family Supporting Foundation (Sandler Opportunity Award and New Technology Award in Basic Science to Y.C, Program for Breakthrough Biomedical Research (PBBR) Award to G.J.N.), UCSF Academic Senate Shared Equipment Grant (to Y.C), grants from the National Institutes of Health (to R.C. and G.J.N). P.P. and J.G.Y. were supported by US National Science Foundation Graduate Research Fellowships. G.J.N is a Leukemia and Lymphoma Society Scholar.
This paper is dedicated to Daniel Herschlag in honor of his 51st birthday.
Supplementary Information accompanies the paper on www.nature.com/nature.