A Structural Model of the Endogenous Human BAF Complex Informs Disease Mechanisms

Abstract

Mammalian SWI/SNF complexes are ATP-dependent chromatin remodeling complexes that regulate genomic architecture. Here, we present a structural model of the endogenously purified human canonical BAF complex bound to the nucleosome, generated using cryoelectron microscopy (cryo-EM), cross-linking mass spectrometry, and homology modeling. BAF complexes bilaterally engage the nucleosome H2A/H2B acidic patch regions through the SMARCB1 C-terminal α-helix and the SMARCA4/2 C-terminal SnAc/post-SnAc regions, with disease-associated mutations in either causing attenuated chromatin remodeling activities. Further, we define changes in BAF complex architecture upon nucleosome engagement and compare the structural model of endogenous BAF to those of related SWI/SNF-family complexes. Finally, we assign and experimentally interrogate cancer-associated hot-spot mutations localizing within the endogenous human BAF complex, identifying those that disrupt BAF subunit-subunit and subunit-nucleosome interfaces in the nucleosome-bound conformation. Taken together, this integrative structural approach provides important biophysical foundations for understanding the mechanisms of BAF complex function in normal and disease states.

Keywords: ATP-dependent chromatin remodeling; BAF complex; cancer; cross-linking mass spectrometry; cryoelectron microscopy; homology modeling; mammalian SWI/SNF complexes; mutations.

Figure 1.. Model for the endogenous human BAF complex bound to a nucleosome.

A-B. Model of the human BAF-NCP complex. Canonical BAF complex subunits are organized into the ATPase, ARP and core modules, and labeled by color. SMARCC1^a/b are shown to represent both SMARCC1 and SMARCC2, SMARCD2 to represent SMARCD1 and SMARCD2, ARID1A to represent both ARID1A and ARID1B found within endogenously-purified complexes. Unaccounted densities in the map of the endogenous BAF complex are indicated. C. Top view of the BAF complex showing the interaction of the ATPase module with the NCP. D. Side view of the BAF complex focusing on the ARP module (ACTL6A, ACTB and the SMARCA4 HSA domain). E. The interaction of the SMARCA4/2 ATPase subunit with the BAF core module. F. Cross-links between all BAF subunits and NCP histones. Subunits are arranged by module (colored blocks). The SMARCA4 ATPase (red asterisk) cross-links to all BAF subunits and two histones. G. A helical bundle, consisting of the SMARCC, SMARCD, and SMARCE1 subunits, forms the basis of the core module of the BAF complex. H. SMARCB1 and its interactions with the NCP and components of the core module. See also Figures S1-S3, Tables S1-S2.

Figure 2.. Assignment of human-specific BAF subunits and domains to densities in the cryo-EM map of the endogenous human BAF complex.

A. Two views of the cryo-EM maps of the endogenous human BAF–NCP complex (transparent gray surface) with the model (colored ribbon representation). Unaccounted densities to which endogenous human-specific BAF subunits could be putatively assigned based on CX-MS data are shown as colored surfaces and labeled. B. (Top) CX-MS localizes the N-terminus of BCL7A. (Bottom) Residues of subunits included in the structural model of the BAF complex that form cross-links with BCL7A. C. Overlay of cryo-EM maps of the endogenous human BAF complex in the presence of AMP-PNP (this study, yellow) and that of the recombinant human BAF complex in the ADP-bound state (EMD-0974, green), low-pass-filtered to 10 Å. Arrows indicate the displacement of the ATPase and ARP modules relative to the recombinant complex. D. CX-MS localizes the SSXT domain of the human-specific subunit SS18 close to BAF subunits SMARCA4 (N-term) and SMARCD1 within the core module. E. Zoomed-in view of the core module (ribbon representation) highlighting the N-termini of SMARCC1/SMARCC2 and ARID1A on the right, which localize to the blue and green unaccounted densities shown in (A), respectively. Regions of SMARCD1 N-terminus are shown on the upper left. Mappable SMARCE1 residue to which SS18 cross-links is shown at the bottom of the core module nearest the SMARCA4 ATPase N-terminus. See also Figure S4.

Figure 3.. Comparison of the endogenous human BAF complex to recombinant human and yeast complexes.

A-B. Bar graphs depicting the percentage of total cBAF amino acids included in the structural models, by (A) complex and (B) subunit. C. (Top, middle) Two views of the cryo-EM maps of the NCP-bound forms of endogenous, recombinant and yeast complexes, showing the interaction of the ATPase/ARP modules with the NCP. (Bottom) Cross-links of the SMARCA4 SnAc and post-SnAc domains to histone H2B of the NCP. D. SMARCB1 winged helix (WH) domain placement in the models of the endogenous human BAF complex using CX-MS and the recombinant human BAF complex (PDB: 6LTJ). E. Overlay of the model of the endogenous human BAF complex (colored by subunit) with the model of the recombinant human BAF complex (gray, PDB: 6LTJ). F. Overlay of the models of endogenous and recombinant yeast and human SMARCA4/Snf2/Sth1 ATPase subunits aligned based on NCP position. G. Immunoblot for BAF components including DPF2 in WT or SMARCB1 KO HEK-293T cells. H. The DPF2 Req domain that interacts with the SMARCB1 Rpt2 domain is required for SMARCB1-BAF complex binding. Human cancer-associated mutations (COSMIC database) mapped onto the DPF2 N-terminal Req domain. See also Figures S4-S6.

Figure 4.. Interaction of the endogenous human BAF complex with the NCP.

A. Cryo-EM maps of the BAF complex alone (left panel) and in complex with NCP (right panel) with the docked models of the BAF subunits and the NCP shown in ribbon representation. The dashed lines highlight the ATPase module. B. Cross-links that are gained (red) or lost (blue) in the BAF ATPase module subunits upon NCP binding. Subunit regions are defined in Figure S3E, Table S2. C. The two largest motions of the ATPase and ARP modules in the free BAF complex identified by multi-body refinement. Two maps at the extreme positions for each motion are shown in blue and red, with motions indicated by black arrows. See also Supplemental Movie S4. D. The two largest motions in the BAF–NCP complex identified by multi-body refinement. A map in the intermediate position (solid gray surface) is shown for each motion and superimposed with two maps at the extreme positions (transparent blue and red surfaces). The motions are indicated with black arrows. See also Supplemental Movie S5. E. Cross-links identified between the ATPase module (SMARCA4/SMARCA2) and histones of the NCP. Subunit regions are defined in Figure S3E, Table S2. F. Scatter plot reflecting the number of cross-links versus the number of cancer-associated mutations within ATPase and ARP module subunits. Selected points are labeled with subunit and/or NCP region cross-linked. G. Schematic reflecting model for nucleosome engagement dynamics of fully-formed canonical BAF complexes. See also Figure S6.

Figure 5.. Bilateral engagement of the NCP acidic patches by SMARCA4 and SMARCB1 is required for BAF-mediated chromatin remodeling.

A. The H2A/H2B nucleosome acidic patch regions are bound by the SMARCB1 C-terminal α-helix (aa 351-385) on one NCP face and the SMARCA4 SnAc/post-SnAC domain (aa 1307-1418) on the other. BAF subunits are shown as colored ribbons, NCP histones as electrostatic surface and NCP DNA as orange ladder. B. (Left) Cancer-associated mutations (COSMIC database) mapped onto the model of the SMARCB1–NCP interface. Number of mutations shown in parentheses, bold= driver mutation (Bailey et al. 2018). (Right) Scatter plot reflecting the number of SMARCB1–NCP cross-links versus the number of SMARCB1 mutations . Selected points are labeled with subunit and/or NCP region cross-linked. C. (Left and middle) Two views showing the mapping of cancer-associated mutations (COSMIC database) onto the model of the SMARCA4 (SnAC/post-SnAC)–NCP interface. Bold= driver mutation. (Right) Scatter plot reflecting the number of SMARCA4–NCP cross-links versus the number of SMARCA4 mutations. Selected points are labeled with subunit and/or NCP region cross-linked. D. Immunoblot showing the expression of WT SMARCA4 and acidic patch-binding region mutant variants in SMARCA4/2-deficient HEK-293T cells. HA-SMARCA4 variants visualized by HA immunoblot; SMARCC1 shown for loading control. E. REAA for BAF-mediated chromatin remodeling performed on HA-purified complexes isolated from SMARCA4/2 double KO HEK-293T cells that were infected with empty vector or SMARCA4 variants. 0.3 mM ATP, 0.01 U/μL DpnII enzyme, 30°C, 90 minutes. DNA bands were visualized on TapeStation platform. F. Quantification of REAA shown in panel E. Average of n=3 independent experiments, with error bars representing S.D. **= p-value <0.01; ***= p-value <0.001 G-I. REAA performed on BaF complexes isolated from HEK-293T SMARCB1-KO cells, captured by either HA-WT SMARCA4, HA-SMARCA4 del1405-1415, or HA-WT SMARCA4 with rescued V5-SMARCB1. G, Immunoblot showing the variant expression. H, REAA DNA fragment analysis (TapeStation). I, Quantification of REAA results from n=3 independent experiments, with error bars representing S.D.

Figure 6.. SMARCA4 hotspot mutations in the BAF core module attenuate the chromatin-remodeling activity.

A. Frequency of missense mutations across full-length SMARCA4. Putative driver missense mutations are in bold (Bailey et al., 2018). Red circles highlight mutations that were evaluated in functional assays. B. The interface between the SMARCA4 HSA domain and the Arm (core-binding) region of ARID1A, with the mutation frequencies in SMARCA4 and ARID1A in red and blue tones, respectively. Number of mutations are indicated in parentheses. C. Mutations in the post-HSA and anchor regions of SMARCA4 that were selected for functional studies. D-F. Chromatin-remodeling activity of WT SMARCA4 and post-HSA/anchor region mutant variants. D, Imunoblot confirming expression of HA-tagged variants. E, Chromatin-remodeling assays (REAA) performed with endogenous complexes; TapeStation was used for fragment analysis. F, Quantification of REAA assays performed in E. Average of n=3 experimental replicates; S.D. is shown. ***= p-value <0.001.

Figure 7.. Model of the endogenous human BAF–NCP complex enables comprehensive dissection of disease-associated mutations.

A. Pan-subunit human BAF complex missense mutational hotspots. Red-gray heatmap indicates the number of missense mutations according to legend. Bold= driver mutations. B. (Left) % of human BAF somatic cancer mutations (COSMIC database) mappable in various models. (Right) Fraction of missense driver mutations (Bailey et al., 2018) of the total number of missense driver mutations in ARID1A, SMARCA4, and SMARCB1 that occur in regions included in the models of the endogenous and recombinant BAF complex. C. Scatter plot reflecting the number of cross-links (by BAF-NCP CX-MS) versus the number of mutations (COSMIC database). Subunit–subunit (red) and subunit–NCP (blue) interactions are indicated. Subunit regions described in Figure S3E, Table S2. D. Bar graphs showing somatic missense mutations that localize to subunit–subunit and module–module interfaces defined by CX-MS. Mutations within five residues of cross-links (window size = 11 aa) that represent an interface between BAF subunits or modules were summed and plotted. E. Mutations localizing to subunit–NCP interfaces defined by CX-MS (window size= 11 aa). F. Missense mutations (within 5 residues of CX site) in human cancer that localize to core subunit–subunit, subunit–NCP core and subunit–NCP tail interfaces, or that are predicted to affect catalytic and non-catalytic ATPase/ARP module subunit–subunit interfaces. See also Figure S7.