Structural basis of a nucleosome containing histone H2A.B/H2A.Bbd that transiently associates with reorganized chromatin

Human histone H2A.B (formerly H2A.Bbd), a non-allelic H2A variant, exchanges rapidly as compared to canonical H2A, and preferentially associates with actively transcribed genes. We found that H2A.B transiently accumulated at DNA replication and repair foci in living cells. To explore the biochemical function of H2A.B, we performed nucleosome reconstitution analyses using various lengths of DNA. Two types of H2A.B nucleosomes, octasome and hexasome, were formed with 116, 124, or 130 base pairs (bp) of DNA, and only the octasome was formed with 136 or 146 bp DNA. In contrast, only hexasome formation was observed by canonical H2A with 116 or 124 bp DNA. A small-angle X-ray scattering analysis revealed that the H2A.B octasome is more extended, due to the flexible detachment of the DNA regions at the entry/exit sites from the histone surface. These results suggested that H2A.B rapidly and transiently forms nucleosomes with short DNA segments during chromatin reorganization.

coupled with genomic microarray or next generation sequencing analyses revealed that H2A.B is preferentially localized on the bodies of actively transcribed genes 26,27 . Biochemical studies demonstrated that H2A.B is efficiently incorporated into nucleosomes, and the DNA segments at the entry/exit sites of the H2A.B nucleosome are more susceptible to endonuclease 28,29 . Consistently, electron microscopic and atomic force microscopic analyses indicated that the DNA entry/exit sites are more flexible in the H2A.B nucleosome, as compared to the canonical H2A nucleosome 29,30 . The H2A.B nucleosome has a propensity to form open chromatin in the nucleosome array 31 . Based on these studies, H2A.B has been implicated in the nucleosome dynamics associated with chromatin opening.
In the present study, we further analyzed the properties of H2A.B in de novo assembly. In living cells, GFP-H2A.B transiently accumulated at DNA replication and repair foci. In a reconstituted system, we found that H2A.B forms two types of nucleosomes (i.e., octasome and hexasome) with shorter DNA fragments, such as a 124 bp DNA, which allows only hexasome formation with the canonical H2A. A small-angle X-ray scattering analysis revealed that the H2A.B octasome structure contained extra volumes on both sides of the nucleosome, as compared to the canonical H2A octasome, probably reflecting the flexible DNA segments at the entry/exit sites. These data suggested that H2A.B may form an intermediate nucleosome with a shorter DNA segment, when chromatin is newly formed or reorganized after DNA replication, repair, and transcription.

Results
Behavior of GFP-H2A.B during the cell cycle. We first established HeLa cells stably expressing GFP-H2A.B. GFP-H2A.B was localized to the chromosomes in mitotic cells (Fig. 1a, arrow) and distributed in euchromatin in interphase nuclei (Fig. 1a), as previously shown 23 and suggested by chromatin immunoprecipitation analyses 26,27 . Within a cell population, however, cells exhibiting GFP-H2A.B concentrated in discrete foci were also observed (Fig. 1a, arrowheads). Since these foci resembled DNA replication foci 32 , we investigated the localization of GFP-H2A.B during the cell cycle. We established cells expressing both GFP-H2A.B and PCNA-mCherry, as a marker of DNA replication complexes formed during S phase 32 . Time-lapse microscopy revealed that GFP-H2A.B became more concentrated in PCNA foci during S phase ( Fig. 1b; Supplementary Movie S1). These results suggested that H2A.B is assembled into chromatin just after DNA replication, and is soon replaced with the canonical H2A. This is consistent with the lower stability of nucleosomal H2A.B in living cells, as previously detected by fluorescence recovery after photobleaching (FRAP) 25 .
We also confirmed the rapid recovery of GFP-H2A.B, as compared to the canonical GFP-H2A, after photobleaching with a 488nm laser line ( Fig. 2a and b). After bleaching one-half of the nucleus, the exchange of GFP-H2A.B was almost complete (i.e., the intensities of the bleached and unbleached areas reached the same level) within 20 min, while only subtle recovery in the bleached area was observed with GFP-H2A (Fig. 2a). Although the expression levels of these GFP-tagged histones varied from cell to cell, their FRAP recovery kinetics did not depend on the original fluorescence intensity ( Supplementary Fig. S1), probably because the GFP-tagged versions are equilibrated with the endogenous histone molecules that are present in great excess 4 . The FRAP data also revealed the difference in the amounts of chromatin-free, diffusible fractions between GFP-H2A.B and GFP-H2A. If free molecules are present, they diffuse in and out of the bleached area during bleaching, and this causes the intensity of the unbleached area to decrease from the original level, in a phenomenon called fluorescence loss in photobleaching (4; and references therein). In addition, under the slow acquisition conditions used, free unbleached molecules can diffuse into the bleached area before the acquisition of the first post-bleach image. Under long (,39 s) bleaching conditions (Fig. 2a), the post-bleach intensity of GFP-H2A.B in the unbleached area dropped to 0.83, while that of GFP-H2A was close to the original level (0.98). Consistently, the post-bleach intensity of GFP-H2A.B in the bleached area was higher than that of GFP-H2A (0.38 vs. 0.27; Fig. 2a). These data suggested the presence of a large pool of chromatin-free GFP-H2A.B, in contrast to GFP-H2A. The rapid recovery and the larger free pool of GFP-H2A.B were also observed in short-term FRAP experiments (Fig. 2b). Under these rapid (110 ms) strip-bleach conditions, the post-bleach intensity in the unbleached region was not affected, but the intensity of GFP-H2A.B in the bleached area was higher than that of GFP-H2A.
As GFP-H2A.B is concentrated in replication foci during S phase, the FRAP recovery kinetics could depend on the cell cycle. To clearly distinguish S-phase cells from non-S phase (G1 or G2) cells, we used HeLa cells expressing both GFP-H2A.B and PCNA-mCherry, a definitive indicator of replication foci 32 . GFP-H2A.B in non-S phase cells exhibited higher recovery than in S phase cells, but this difference was due to the different amounts in the free pools (i.e., the level at the first post-bleach point), and the curve shapes (exchange rates) were very similar ( Supplementary Fig. S2). The lower level of the chromatin-free fraction in S phase is likely to be a consequence of the massive assembly of GFP-H2A.B into replicated chromatin in addition to the basal level of incorporation into transcribed chromatin 26,27 . As the FRAP curve shapes were similar between S and non-S phase cells, the dissociation rates are apparently comparable even in replication-coupled assembly.
Interestingly, when cells were microirradiated using a 405-nm laser line that can induce DNA damage, GFP-H2A.B accumulated in the irradiated area within 2 min (Fig. 2c). In contrast, the canonical GFP-H2A remained relatively immobile under the same conditions ( Fig. 2c), similar to the results from photobleaching using a 488-nm laser line (Fig. 2b). Once again, the curve shapes were similar between S and non-S phase cells ( Supplementary Fig. S3). These results suggested that H2A.B is specifically assembled into chromatin during DNA damage repair. This may simply be due to the presence of more chromatin-free H2A.B molecules than H2A; however, specific mechanisms might also be involved. Although the deposition complex of H2A.B has not been characterized yet, a specific chaperone may facilitate the assembly of H2A.B to the damaged chromatin. Alternatively, H2A.B incorporation might be coupled with H2AX evicion, which occurs after DNA damage in a ubiquitylation-dependent manner 33 .
Taken together with previous data 26,27 , H2A.B appears to be transiently assembled into chromatin after DNA replication, DNA repair, and transcription, which all involve chromatin reorganization, including H2A-H2B dimer exchange 4,5,8,12 . To obtain mechanistic insights into the transient H2A.B assembly, we performed structural and biochemical analyses using a reconstituted system.
De novo nucleosome formation by H2A.B with 116, 124, and 130 bp DNAs. Previous reports found that, in the H2A.B nucleosome, about 118-130 bp of DNA are protected from digestion by micrococcal nuclease (MNase) 28,29 . This abbreviated contact of DNA with histones may explain the instability of H2A.B once it is assembled within a nucleosome; however, it remained uncertain whether H2A.B could form a nucleosome with a short DNA segment de novo. To determine whether the H2A.B nucleosome is actually assembled with short DNA fragments, we performed nucleosome reconstitutions with various lengths of DNA fragments (116, 124, 130, 136, and 146 bp DNAs) and human histones purified as bacterially expressed recombinant proteins (Fig. 3a). The nucleosome reconstitution was performed with H2A.B-H2B (or H2A-H2B) and H3-H4 by the salt-dialysis method (Fig. 3b). As expected, both H2A and H2A.B formed a nucleosome with 146 bp DNA (Fig. 3c, lanes 5 and 10). However, we found that H2A.B formed two types of nucleosomes, which migrated differently on native PAGE when a 116, 124 or 130 bp DNA fragment was used as the substrate for nucleosome reconstitution (Fig. 3c, lanes 6-8). In contrast, H2A only formed one type of nucleosome when a 116 or 124 bp DNA fragment was used (Fig. 3c, lanes 1 and 2). When the nucleosome reconstitution was performed with H2A in the presence of a 130 or 136 bp DNA fragment, extra bands, with slower migration than the major (lower) band, were detected (Fig. 3c,  lanes 3 and 4). These extra bands may correspond to the tetrasome containing H3-H4 and/or the complex formed by improper histone-DNA binding (Supplementary Fig. S4).
H2A.B forms both an octasome and a hexasome de novo with a 124 bp DNA fragment. To determine the histone compositions of the H2A and H2A.B nucleosomes reconstituted with a 124 bp DNA fragment, we electrophoretically purified the upper and lower bands of the reconstituted H2A.B nucleosome (Fig. 4a). The histone compositions were analyzed by SDS-PAGE and Coomassie Brilliant Blue staining (Fig. 4b). Consistent with our previous study 34 , the H2A nucleosome reconstituted with a 124 bp DNA fragment contained about one-half the amount of H2A-H2B, as compared to that of H3-H4 (Fig. 4b, lane 3, and c), indicating hexasome formation. Similarly, the lower band of the H2A.B nucleosome with a 124 bp DNA fragment contained a reduced amount (about 50%) of H2A.B-H2B (Fig. 4b, lane 5, and c). In contrast, the upper band contained similar amounts of H2A.B-H2B (Fig. 4b, lane 6, and c) relative to the control H2A.B octasome (Fig. 4b, lane 4, and c). These results indicated that H2A.B forms both   hexasomes and octasomes de novo, but H2A only forms a hexasome with a shorter DNA fragment, such as 124 bp.
Solution structure of the H2A.B nucleosome. To gain further insights into the structural differences between the H2A.B nucleosome and the canonical H2A nucleosome, we performed dynamic light scattering (DLS) measurements. In this analysis, we used the 145 bp 601 DNA fragment, which reportedly forms a stably positioned nucleosome 3 . Consistent with previous electron and atomic force microscopic analyses 29 , the mean particle size of the H2A.B octasome with the 145 bp 601 DNA fragment was clearly larger than that of the canonical H2A octasome (Fig. 5a and b). These DLS data suggested that the DNA segments may be unwrapped at the entry/exit sites of the H2A.B octasome. We then performed a small-angle X-ray scattering (SAXS) analysis of the H2A.B octasome with the 145 bp 601 DNA fragment (Fig. 5c, d, and e). The radius of gyration (R g 5 49.4 6 0.4 Å ) and the maximum diameter (D max 5 195 Å ) of the H2A.B octasome were significantly larger than those of the canonical H2A octasome (R g 5 43.4 6 0.3 Å , D max 5 125 Å ). The shape of the distance distribution function (P(r)) of the H2A.B octasome has a long tail, suggesting that the structure of the H2A.B octasome is not compact. Based on these data, we constructed a dummy atom model of the H2A.B octasome structure (Fig. 5e). In this model, extra volumes were observed on both sides of the H2A.B octasome, as compared to the canonical H2A octasome model. These extra volumes may reflect the flexible DNA segments at the entry/exit sites of the H2A.B octasome. It should be noted that nucleic acids scatter more strongly than proteins. Despite the differences in X-ray scattering by nucleic acids and proteins, the known crystal structure of the canonical nucleosome and the SAXS envelope are consistent with each other (Fig. 5e, right panels) 35 . This indicated that the SAXS data could be used to build low-resolution structural models of nucleosomes.

Discussion
H2A.B (originally named H2A.Bbd) was found as a distant H2A variant that is excluded from the inactive X-chromosome (Barr body) 23 . A genome-wide analysis revealed that H2A.B is preferentially associated with actively transcribed genes 26,27 . It was previously reported that 118-130 bp DNAs are protected in the H2A.B nucleosome from  MNase digestion 28,29 . In the present study, we found that H2A.B intrinsically forms both a hexasome and an octasome with shorter DNA segments, such as 116-130 bp. In contrast, the canonical H2A did not form an octasome with a 124 bp DNA fragment. Therefore, the octasome formation with shorter DNA segments may be a specific characteristic of the H2A.B nucleosome. This property of H2A.B might facilitate rapid and transient nucleosome reformation with shorter DNA segments available after transcription, replication, and repair, before more stable nucleosomes are formed with the canonical H2A. The transient formation of H2A.B nucleosomes may protect DNA, by preventing the access of specific or non-specific DNA binding proteins that can alter the chromatin structure and/or epigenetic status. The alteration of the linker DNA orientation, which is caused by DNA flexibility at the entry/exit regions of the nucleosome, may also affect linker histone binding. Indeed, the H2A.B nucleosome is reportedly defective in linker histone H1 binding 30 . Histone H1 binds to nucleosomal and linker DNAs, and forms a stem-like structure 36 to establish a higher-order chromatin structure. H2A.B nucleosomes may prevent histone H1 binding and maintain the open chromatin structure after temporary nucleosome disruption by transcription, replication, and repair, although H2A.B may play more specific roles 27 .
H2A.B is reportedly expressed in spermatogenic cells and the nucleosomal chromatin fraction of human sperm 37 . H2A.B does not form a stable histone octamer without DNA 28 , and nucleosomal H2A.B exchanges rapidly in nuclei 25 . Consistently, H2A.B confers a more flexible nucleosome structure, as compared to the other mammalian histone H2A variants, H2A.X, H2A.Z, macroH2A, and canonical H2A 38 . Intriguingly, these characteristics are common to human histone H3T 14 , which is also highly expressed in testis [39][40][41] . Therefore, the unstable nature of the testis-specific nucleosomes may play an essential role in the formation of the specific chromatin structure packaged within the sperm nucleus.
For live cell analysis, cells were grown on a glass-bottom dish (Mat-Tek) without synchronization. When necessary, the PCNA-mCherry pattern was used to judge the cell cycle point of each cell. The dish was placed on a confocal microscope (FV-1000; Olympus), equipped with a culture system (Tokai Hit) at 37uC under a 5% CO 2 atmosphere. For time-lapse imaging, confocal images were collected every 5 min for 14 h, using a PlanApoN 603 OSC (NA 5 1.4) oil-immersion lens (512 3 512 pixels; 4 ms/pixel; 4 line Kalman; pinhole 110 mm; line sequential scanning with 488-and 543-nm lasers). Fluorescence recovery after photobleaching (FRAP) and laser microirradiation were performed using a confocal microscope (FV-1000; Olympus) with a PlanApoN 603 OSC (NA 5 1.4) oil-immersion lens. For Fig. 2a, after 2 images were obtained every 30 s (0.3% 488-nm laser transmission; 4 ms/pixel; 512 3 512 pixels; pinhole 800 mm; 1.53 zoom), one-half of each nucleus was bleached (100% 488-nm laser transmission; 20 ms/pixel; 2 iterations), and images were obtained using the original settings. For Fig. 2b and 2c, after 5 images were obtained (0.1% 488-nm laser transmission; 4 ms/pixel; 512 3 512 pixels; pinhole 800 mm; 83 zoom), a 2 mm width strip was bleached (100% 488-nm laser transmission with 4 ms/pixel for FRAP, or 100% 405-nm laser transmission with 200 ms/pixel for micro-irradiation), and 95 more images were obtained using the original settings. Fluorescence intensity measurements were performed using Image J version 1.45 s (http://rsb.info.nih.gov/ij/). The net intensities of the bleached and unbleached areas were obtained by subtracting the background intensity outside nuclei in each time frame. To obtain relative intensities to the initial intensity of the same area, the net intensities were normalized to the average intensity of pre-bleach images.
Purification of recombinant human H2A, H2B, H3.1, and H4. Human histones H2A, H2B, H3.1 and H4 were expressed as the N-terminally His 6 -tagged proteins in Escherichia coli cells 44 , and purified according to the method described previously [14][15][16] . Briefly, the His 6 -tagged histones were recovered from the insoluble fraction in 50 mM Tris-HCl buffer (pH 8.0), containing 7 M guanidine hydrochloride, 500 mM NaCl, and 5% glycerol, and purified by Ni-NTA agarose chromatography (Qiagen) under denaturing conditions. The His 6 -tag was then removed by cleavage with thrombin protease (1 unit/mg of histones; GE Healthcare), and further purified by Mono S column chromatography (GE Healthcare). The purified histones were dialyzed against water, freeze-dried, and stored at 4uC.
Preparation of the H2A.B-H2B complex. The human H2A.B gene was amplified from a human testis cDNA pool (Clontech) by polymerase chain reaction (PCR). The resulting DNA fragment was ligated into the same vector used for the H2A expression 44 . The recombinant H2A.B was expressed in Escherichia coli BL21(DE3) codon(1)RIL cells (Stratagene), as the N-terminally His 6 -tagged protein. The His 6tagged H2A.B was recovered from the insoluble fraction in 50 mM Tris-HCl buffer (pH 8.0), containing 7 M guanidine hydrochloride, 500 mM NaCl, and 5% glycerol, and purified by Ni-NTA agarose chromatography (Qiagen) under denaturing conditions. The purified His 6 -tagged H2A.B was mixed with purified H2B (1 mg protein/ml) at a 151 stoichiometry, and the sample was dialyzed against 20 mM Tris-HCl buffer (pH 7.5), containing 7 M guanidine-HCl and 20 mM 2-mercaptoethanol, for 4 hours, followed by overnight dialysis against 10 mM Tris-HCl buffer (pH 7.5), containing 2 M NaCl, 1 mM EDTA, and 2 mM 2-mercaptoethanol. The sample was then dialyzed against the buffers containing 1 M NaCl for 4 hours, 0.5 M NaCl for 4 hours, and 0.1 M NaCl overnight. The His 6 -tag was removed from H2A.B by cleavage with thrombin protease (1 unit/mg of His 6 -tagged H2A.B). The H2A.B-H2B complex was further purified by Superdex200 (GE Healthcare Biosciences) gel filtration chromatography with 10 mM Tris-HCl buffer (pH 7.5), containing 2 M NaCl, 1 mM EDTA and 2 mM 2-mercaptoethanol.
Preparation of 116-146 bp DNA fragments for nucleosome reconstitution. The 116 bp, 124 bp, 130 bp, 136 bp, and 146 bp DNA fragments, containing the asatellite sequence, were prepared by self-ligation of the 56 bp, 60 bp, 63 bp, 66 bp, and 71 bp DNA segments containing an additional four-base overhang at one 59 end (proximal end), which is located at the center of the DNA fragment after self-ligation. The 56 bp, 60 bp, 63 bp, and 66 bp DNAs were produced by deleting 15 bp, 11 bp, 8 bp, and 5 bp from the distal end of the 71 bp DNA, respectively. These DNA fragments were bacterially produced, and were purified by the method described previously 45 . Sixteen repeats of each of these DNA segments were inserted into the pGEM-T Easy Vector (Promega). The plasmids were amplified in E. coli cells. The DNA fragments containing the a-satellite sequence were isolated from the plasmid vector by digestion with EcoRV. The ends of the EcoRV fragments were dephosphorylated with calf intestine alkaline phosphatase (CIAP; Nippon Gene) treatment. After the CIAP treatment, the proximal end was digested with EcoRI, and the four-base overhang was created. Each of these DNA fragments was self-ligated, and the 116 bp, 124 bp, 130 bp, 136 bp, and 146 bp DNA fragments were prepared.
The 145 bp 601 DNA was bacterially produced, and was purified by the method described previously 45,46 . Eight 145 bp 601 DNA fragments were tandemly ligated into the pGEM-T Easy vector. The purified plasmid containing eight 601 DNA repeats was digested with EcoRV, and the resulting 145 bp 601 DNA fragment was treated with CIAP.
Reconstitution of nucleosomes. The nucleosomes were reconstituted by the salt dialysis method, as previously described 14  sample-to-detector distance of 2040.060 mm, which was calibrated by the powder diffraction from silver docosanoate. The nucleosome samples were reconstituted with the 145 bp 601 DNA. Circular averaging of the SAXS intensity data was then performed to obtain the one-dimensional intensity data I(q) as a function of q (q 5 4psinh/l, where 2h is the scattering angle and the X-ray wavelength l 5 1.488 Å ). To correct the inter-particle interference effect, I(q) data were collected at three nucleosome concentrations (0.5, 1.0, and 3.0 mg/mL), and were extrapolated to zero concentration. A SAXS intensity measurement of the buffer solution (20 mM Tris-HCl buffer, pH 7.5, containing 1 mM EDTA and 1 mM dithiothreitol) for background subtraction was also performed, using the same conditions and procedure as those used for the nucleosome solutions.
The radius of gyration, R g , was estimated by fitting the I(q) data using the Guinier approximation I(q) 5 I(0) exp(2q 2 R g 2 /3), where I(0) is the forward scattering at the zero scattering angle, in the smaller angle region of qR g , 1.3. The error of R g was estimated from the least-squares fitting. The distance distribution function P(r) and its error were calculated by the program GNOM 47 . The maximum dimension, D max , was estimated from the P(r) function as the distance r, where P(r) 5 0, and its error was estimated from the errors of the P(r) values around P(r) 5 0. All data were processed and analyzed using the software applications embedded in the ATSAS package 48 .
Construction of dummy atom models of nucleosomes containing H2A.B or H2A. The low-resolution dummy atom models of the nucleosomes containing H2A.B or H2A were produced by DAMMIN, using the scattering data 49 . Ten independently calculated, low-resolution dummy atom models were averaged with DAMAVER for each nucleosome 50 .