SAF-A Regulates Interphase Chromosome Structure through Oligomerization with Chromatin-Associated RNAs

1Mammalian interphase chromosomes are organized into topologically constrained chromatin domains, which are responsive to transcription and local gene density. Gene-poor genomic regions have a compact large-scale chromatin structure, while regions rich in genes and transcriptional activity have a more decompacted structure. Previously, we suggested that transcription and topoisomerase activities, that occur at the gene level, alter local topology to form supercoiling domains and these correspond to structures seen by Hi-C. It is unclear, however, how these processes could impact on large-scale chromatin structures.

2In contrast to mitosis, where topoisomerases and condensin play a central role in scaffolding chromatin, the molecular underpinnings of interphase domains are poorly characterized and their functional impact remains unknown. A “nuclear matrix” consisting of insoluble proteins and RNA particles was proposed to organize interphase chromatin architecture, maintain chromosome territories, enhance gene expression, and provide a platform for nuclear processes, but chromatin mobility in vivo and the lack of a stable nucleoprotein structure in live cells undermined the concept. However, the structural contribution of RNA to chromatin organization remains undisputed: a large proportion of chromatin by mass corresponds to RNA, mostly belonging to the loosely termed chromatin-associated RNA (caRNA) class. The functional roles of caRNAs are hinted by the observation that they are stably associated with interphase chromosome territories and their disruption leads to chromatin condensation. The molecular basis for this is unknown, but it is thought heterogeneous ribonucleoprotein particles (hnRNPs) provide a docking platform to associate with nascent transcripts while caRNAs might influence chromatin structure. Recent studies have been unable to find specific species of RNA, similar to XIST, which could regulate large-scale chromatin structure, suggesting instead that diverse caRNAs transiently interact with chromatin, forming a dynamic compartment. However, a refinement of this model would require insight into how the interaction of proteins with caRNAs can regulate chromatin structure.

3Scaffold attachment factor A (SAF-A), also known as heterogeneous ribonucleoprotein U (HNRNP-U), is an abundant protein reported to bind scaffold attachment regions and involved in several cellular processes such as pre-mRNA splicing, accumulation at DNA damage sites and Xist-mediated transcriptional silencing. Structurally, SAF-A contains a low complexity RNA-binding RGG repeat and an ATP-binding AAA+ domain, known to facilitate the assembly and operation of diverse protein and nucleoprotein machines. Other well-characterized AAA+ domain-containing proteins, such as replication factor C (RFC) and DnaA, oligomerize through their AAA domains, often with nucleic acids, to form higher molecular weight structures. We characterized SAF-A activity to understand the relationship between its structure and function in regulating chromatin architecture.

4We are able to demonstrate that SAF-A regulates transcriptionally active large-scale chromatin structures in human cells. Using functional mutants of SAF-A, we dissect the underlying molecular mechanisms to show that SAF-A can cycle from a monomeric to a homo-oligomeric state through ATP binding and caRNAs; concomitantly, SAF-A oligomerization drives chromatin decompaction while monomerization compacts large-scale chromatin organization.

5We suggest that SAF-A interacts with caRNAs to form a chromatin mesh, and unlike the historical concept of a nuclear matrix, is highly responsive to ongoing transcription and can undergo dynamic cycles of assembly and disassembly. Surprisingly, a gross change in chromatin structure has limited effect on transcription indicating that gene-level chromatin structure is important for regulating transcription, rather than large-scale chromatin architecture. Although changes in large-scale chromatin structure do not influence gene expression, an alteration in the chromatin landscape had a dramatic effect on genome stability; loss or mutation of SAF-A triggered a DNA damage response and chromosomal instability.

6To unify the structural and enzymatic aspects of SAF-A function, we speculate that SAF-A/RNA interactions will drive the formation of local chromatin domains or micro-bodies, while its role in regulating large-scale chromatin structures will partition the genome into functionally diverse segments. This process is essential for maintaining genome stability and could provide a constraint to maintain clusters of genes together in the genome during evolution.

7SAF-A’s contribution to chromatin structure is unknown; it is nuclear diffuse (Figure S1A), but excluded from condensed DAPI bright regions such as the gene-poor nuclear periphery. SAF-A’s abundance, localization, and structural characteristics suggest it may play a role in organizing nuclear architecture. To examine whether SAF-A affects large-scale chromatin, we analyzed chromatin compaction using three-dimensional (3D) DNA fluorescence in situ hybridization (FISH) across loci with different epigenetic states and distinct levels of chromatin folding using fosmid probe pairs (Figures 1A and S1C; Table S1) (detailed in ). SAF-A small interfering RNA (siRNA) treatment resulted in a substantial reduction of SAF-A protein (Figure S1B) and caused a significant compaction of gene-rich 11p15.5 and 11p15.1 loci enriched in “open” chromatin, but not the gene-poor 11p14.1 locus (Figure 1B), indicating that SAF-A is involved in decompacting chromatin. Similar changes in chromatin structure were observed for additional gene-rich (2p25.1, 21q22.3, Xq13.1) and gene-poor loci (Xq25, 1p31.2) using fosmid probes positioned 0.1–2.0 Mb apart (Figure S1D). Depletion of other hnRNP members, such as HNRNP A1 and HNRNP C, showed no obvious nuclear effects (Figure 1C), indicating that SAF-A has a unique role among hnRNPs in regulating chromatin structure. To extend our analysis, we targeted labeled oligo probes to gene-rich and gene-poor regions on chromosome 11 (Figure 1A) and analyzed chromatin compaction by FISH and 3D modeling. SAF-A depletion led to significant compaction of the HSA11 gene-rich domain, but not the gene-poor region (Figure 1D), indicating that SAF-A contributes to the partitioning of the genome into segments with distinct structures. Changes in chromatin compaction following SAF-A depletion were observed by 24 hr, while global changes in nuclear area were not apparent until later, indicating that the effect is due to changes in local chromatin structure rather than an alteration in nuclear size (Figure S1F).

8The human genome is partitioned into chromosomes with very different gene densities; we reasoned that if SAF-A predominantly regulated gene-rich chromosome regions, its depletion would have a greater effect on gene-rich, rather than gene-poor chromosomes. To test this, we monitored the chromosome territory of the gene-rich HSA19 and gene-poor HSA18 before and after SAF-A depletion. The two similarly sized chromosomes have different gene densities (45 genes/Mb on HSA18 versus 216 genes/Mb on HSA19) (Figure S1C), and HSA19 has been shown to adopt a more open chromosome territory configuration than HSA18. Upon SAF-A depletion, we observed significant chromosome territory compaction, but HSA19 was compacted significantly more that HSA18, consistent with SAF-A predominantly regulating gene-rich chromosome regions at scales up from 100 kb to a whole chromosome territory (Figure 1E). At the largest nuclear scale, we also observed a significant change in DAPI texture (Figures S1G and S1H) after SAF-A depletion.

9Previously, we reported that transcription also decompacts large scale chromosome structures. To determine whether SAF-A and transcription act independently or in a synergistic manner, we analyzed chromatin compaction in control cells or cells depleted of SAF-A following treatment with the transcription inhibitor α-amanitin. Inhibition of transcription and loss of SAF-A together did not result in a further compaction of chromatin than either of the two factors alone, consistent with the two processes functioning together (Figure 1F).

10We reasoned that if SAF-A and transcription acted together, there could be two potential models for the functional relationship between SAF-A, transcription, and chromatin: either transcription could regulate SAF-A to alter chromatin structure (model 1) or SAF-A could have an impact on transcription, which regulates structure (model 2) (Figure 2A). To discriminate between these possibilities, we investigated the effect of SAF-A depletion on cellular transcription using different approaches. First, we examined global transcription via a short pulse of 5-ethynyl uridine (5-EU) incorporated into newly transcribed RNA (∼80% pol II transcripts) and quantified by fluorescent labeling (Figure 2B). Depletion of SAF-A had little effect on the distribution of transcripts or the apparent levels of transcription. In contrast, inhibition of transcription using α-amanitin had a profound effect on newly synthesized RNA. To confirm these results, [5-3H]uridine was pulsed into cells and measured in newly synthesized RNA; there was no change in transcription after 24 or 48 hr of SAF-A depletion (Figure S2A). This analysis was extended by pulsing cells with 4-thiouridine before analyzing the nucleoside composition of RNA and DNA (Figure S2B). SAF-A depletion had little effect on either nascent RNA or steady-state RNA levels (Figure 2C). Consistently, cytological markers of transcription (Ki67 and SC35) showed no changes in their distribution after SAF-A depletion (Figure S2C). To determine if individual transcripts were affected, we analyzed mRNA by next generation sequencing. At 24 hr following depletion, only SAF-A was differentially expressed compared to control cells, while at 48 hr, there were limited changes in cell-cycle-dependent genes (Figure S2D) consistent with other studies. These results were validated by qRT-PCR (Figure 2D) providing overwhelming evidence that transcription is upstream of SAF-A (model 1). To further support model 1, we reasoned that changes in transcription might be expected to influence SAF-A behavior. In control cells SAF-A is insoluble (Figure 2E), but after transcription, inhibition SAF-A is released into a soluble fraction consistent with transcription being functionally upstream of SAF-A (model 1). These surprising results indicate that while SAF-A exerts a significant effect on large-scale chromatin structures (Figure 1), there is a limited concomitant effect on transcription, suggesting it is small-scale chromatin structures that predominantly impact transcriptional processes.

11To understand the relationship between SAF-A and transcription and how SAF-A mechanistically impacts chromatin, we undertook a structural analysis of the SAF-A protein. SAF-A encodes four conserved domains (Figure 3A): SAP (SAF-A/B, acinus, and PIAS) that has DNA binding activity, SPRY (SPla and the ryanodine receptor) of unknown function, AAA+ (ATPases associated with diverse cellular activities), and a low complexity RGG (arginine glycine-glycine) RNA-binding domain. Globular domain composition and disorder prediction indicate that the SPRY and AAA+ domain are ordered but separated from the SAP and RGG domains (Figure S3A), suggesting the SPRY and AAA+ domains adopt a defined conformation. 3D homology modeling of SAF-A AAA+ domain (Figures S3B and S3C) showed the AAA+ domain has a core αβα nucleotide-binding fold containing highly conserved Walker-A (WA), Walker-B (WB), and a helical subdomain known as the “lid.” The Walker A motif is predicted to be necessary for nucleotide binding, while the Walker B motif is required for nucleotide hydrolysis.

12We directly tested whether SAF-A AAA+ domain could hydrolyze ATP or GTP. Purified SAF-A AAA+ fragment, but not SAF-A Walker A or Walker B mutants, exhibited ATPase activity at 0.15 nmol ATP hydrolyzed per nmol of SAF-A AAA+ fragment per min, but no GTPase activity (Figure 3B), similar or higher than other AAA+ proteins such as DnaA and ORC at 0.017 and 0.27 nmol ATP hydrolyzed per nmol protein per min, respectively. As the AAA+ domain is located adjacent to an RNA-binding domain, we speculated that RNA might modulate the SAF-A ATPase activity. A longer fragment of SAF-A encompassing the AAA+/RGG domains was purified and the ATPase activity measured in the presence or absence of RNA. Total RNA increased the ATPase activity 2.5 times compared to the absence of RNA (Figure 3B).

13To determine whether SAF-A undergoes a conformation change upon ATP-binding, consistent with protein actuation as for other AAA+ domain proteins (e.g., DnaA), we measured the change in fluorescence upon nucleotide binding. His-tagged AAA+ fragments of SAF-A were treated with increasing concentrations of ATP and the quenching of intrinsic fluorescence of tryptophan and tyrosine residues were measured to reveal a pronounced alteration in structure upon nucleoside binding (Figure 3C). Mutation of the Walker A motif abrogated ATP binding but Walker B mutations had no effect; Kd for ATP binding to wild-type SAF-A and Walker B mutant SAF-A were 38 μM and 29 μM, respectively.

14ATP binding triggers a change in AAA+ domain conformation in other proteins, which then stabilizes subunit-subunit interactions. We hypothesized that SAF-A might form an oligomeric complex upon ATP binding. Further analysis of SAF-A revealed an additional predicted α helix (aa 512–521) after canonical β strand 2. This α helix is analogous to the initiator-specific motif (ISM) observed in DnaA and ORC family proteins, which guides neighboring AAA domains into a non-planar arrangement, preventing formation of a flat ring (Figure 3D) and required for folding into a spiral oligomer.

15To investigate whether SAF-A forms oligomeric chains, we reasoned that protein-protein cross-linkers would stabilize oligomeric SAF-A structures that could then be isolated from cells. To minimize the likelihood of purifying protein aggregates, we treated cells with cross-linkers with amino acid specificity (BM(PEG)2 [1,8-bismaleimido-diethyleneglycol], Cys-Cys, 14.7 Å; DSS [disuccinimidyl suberate], Lys-Lys, 11.4 Å; GMBS [N-γ-maleimidobutyryl-oxysuccinimide ester], Cys-Lys, 7.3 Å). In contrast to aggregates, oligomeric structures have geometric regularity, so it is less likely that a defined cross-linker will stabilize an aggregate structure. After protein extraction and denaturation, high molecular weight SAF-A species were observed with all cross-linkers consistent with protein oligomerization (Figures 4A and S4A). Transcription inhibition substantially reduced oligomer formation (Figure 4A) suggesting that SAF-A oligomerization is transcription-dependent. To identify critical domains for SAF-A oligomerization, individual FLAG-tagged fragments were expressed in cells, cross-linked, and isolated. The AAA+/RGG domain efficiently formed oligomers (Figure 4B) while the SPRY domain was refractory to oligomerization; the RGG RNA binding domain was absolutely required (Figure S4B). In the presence of either ATP or the non-hydrolysable analog ATPγS, the SAF-A AAA+/RGG formed high molecular weight oligomeric species indicating that ATP binding maintains the oligomeric form (Figure 4B). In contrast, a AAA+/RGG Walker A mutant that is unable to bind ATP showed much less oligomerization, while Walker B mutant fragment-deficient in ATP hydrolysis function stably formed oligomers. These results support the idea that SAF-A undergoes protein oligomerization and hydrolysis similar to DnaA.

16To investigate the molecular basis for cyclical SAF-A oligomerization and hydrolysis, we sought to identify the components involved. Guided by the known DnaA interaction with DNA, we purified oligomeric SAF-A (Figure 4C, left) to identify whether it bound to RNA. We found that newly synthesized RNA specifically bound to oligomeric SAF-A (Figure 4C, right) supporting the idea that SAF-A and RNA form a complex. To analyze the steps required for SAF-A oligomerization, we pre-treated cells with RNaseA to monomerize SAF-A (Figure 4D). Addition of total RNA before cross-linking promoted SAF-A oligomerization, which could be abrogated by apyrase treatment showing that SAF-A oligomerization requires both RNA and ATP. To test SAF-A de-oligomerization, we purified cross-linked oligomeric SAF-A from cells and reversed the cross-links using DTT (Figure 4E). ATP or ATPγS stabilized the oligomers whereas RNase or apyrase treatment promoted monomerization. Together, our data show that SAF-A can undergo cycles of oligomerization promoted by RNA and ATP binding and ATP hydrolysis (Figure 4F).

17To investigate SAF-A dynamics in live cells, we used three complimentary approaches. First, to assay full-length SAF-A oligomerization in cells, we designed a proximity ligation assay (PLA) to detect SAF-A ↔ SAF-A interactions (Figure S5A). In PLA, proteins are detected with antibodies fused to DNA oligos: close proximity of the proteins allows ligation of the DNA oligos, which can then be amplified and detected by a fluorescent signal. Flp-In T-REx 293 cells were depleted of endogenous SAF-A and then doxycycline treated to express wild-type FLAG-tagged SAF-A or Walker A or Walker B mutants, at a physiological level (Figure S5B), and co-transfected with T7-tagged full-length SAF-A or SAF-A with Walker A or Walker B point mutations. PLA signals indicating SAF-A interactions (Figure 5A) were found in cells expressing wild-type SAF-A or the SAF-A Walker B mutant, while no interactions were observed for the SAF-A Walker A mutant, suggesting that SAF-A oligomerizes in vivo through ATP binding, while ATP hydrolysis is not important for these associations. We next investigated if transcription regulates SAF-A oligomerization in live cells, consistent with its role upstream of SAF-A in model 1 (Figure 2). To directly test this we treated cells with the transcription inhibitor α-amanitin and assayed for SAF-A oligomerization by PLA. After transcription inhibition the number of PLA signals significantly decreased (Figure 5B), concomitant with a reduction in SAF-A oligomeric forms (Figure 4A).

18Second, we analyzed SAF-A interactions using FLIM-FRET (fluorescence lifetime imaging microscopy – Förster resonance energy transfer). FLIM-FRET is a well-established approach to detect interactions at the nanometer scale; we predicted that SAF-A↔SAF-A interactions in oligomers should be detectable by FRET (Figure 5C). Although fluorescent protein tags inhibited the oligomerization of the full-length protein, oligomerization could be observed biochemically for GFP and mCherry-tagged AAA+/RGG SAF-A fragments (Figure S5B). We expressed the GFP-tagged AAA+/RGG fragment in cells, either on its own (donor only) or in combination with a mCherry-tagged version of the same fragment and measured the fluorescence lifetime of GFP. An interaction between the two differentially labeled fragments is expected to result in energy transfer between the GFP and mCherry tags, associated with a decrease in the fluorescence lifetime of GFP. The difference in GFP lifetime between the donor alone (GFP fragment on its own) and the FRET conditions can then be expressed as FRET efficiency. We found a significant increase in FRET efficiency in cells expressing both GFP and mCherry tagged AAA+/RGG fragments, strongly indicative of oligomerization involving the two components (Figure 5D). Consistently with the PLA data, a decrease in FRET efficiency was observed for AAA+/RGG Walker A mutant fragments, confirming the importance of ATP binding for oligomerization (Figures 5E and S5C). Fragments carrying Walker B mutations resulted in an increase in FRET efficiency compared to the wild-type, suggesting a defect in the oligomerization cycle and underscoring the role of ATP hydrolysis in SAF-A monomerization. Abolishing transcription, using two different inhibitors, also impeded oligomerization, as evidenced by a decrease in FRET efficiency (Figures 5E and S5C).

19Finally, as the FLIM-FRET analysis indicated that the Walker B mutant showed excessive oligomerization, we speculated that ATP hydrolysis might be important for the reversal and dynamics of this process. Consistently, when we examined the staining patterns of FLAG-tagged wild-type SAF-A and the two mutants by super-resolution STED microscopy (Figure S5D), the SAF-A Walker B mutant showed a granular staining pattern indicative of large SAF-A oligomers, which was not observed in the wild-type SAF-A or the SAF-A Walker A mutant (Figures 5F and S5E). Quantification of the granules showed that the Walker B mutant had significantly more and brighter granules than for either wild-type or Walker A mutant (Figure S5F), highly indicative of SAF-A Walker B mutant oligomers continuing to undergo oligomerization, forming large structures. These three pieces of evidence support a model of cyclical SAF-A oligomerization and monomerization in response to transcription and regulated by ATP binding and hydrolysis.

20We next tested whether SAF-A oligomerization was required for regulation of large-scale chromatin structures by examining 11p15.5 (Figure 6B) or 2p25.1 (Figure S6A) compaction following rescue experiments with RNAi-resistant wild-type SAF-A, or SAF-A encoding Walker A or Walker B mutants at different time points (Figure 6A, colored triangles). As predicted, wild-type SAF-A was able to rescue the compacted phenotype and efficiently decompact chromatin (Figure 6B, box 3). Expression of the Walker A mutant was unable to rescue the chromatin state (Figure 6B, box 5) showing that SAF-A oligomerization is required for chromatin decompaction. Conversely, the Walker B mutant was able to decompact the gene-rich region (Figure 6B, box 7), indicating that oligomerization, and not the dynamic cycling between oligomeric and monomeric states, is essential for decompacting large-scale chromatin structure. In the opposite experiment, we tested whether SAF-A monomerization was necessary for chromatin compaction, taking advantage of the ability of α-amanitin to compact chromatin (Figure 1F): expression of the SAF-A protein in the presence of α-amanitin (Figure 6B, box 8) was able to drive chromatin compaction (Figure 6B, box 4), however, the Walker B mutant was unable to drive chromatin compaction (Figure 6B, box 8) consistent with ATP-hydrolysis being required to cycle from an oligomeric to monomeric state to compact chromatin. Thus, we conclude that SAF-A oligomer assembly and disassembly regulates large-scale chromatin structure.

21We next addressed how SAF-A interacted with chromatin to regulate large-scale chromatin structures. We hypothesized two potential modes of interaction: SAF-A could either be directly bound to chromatin, or it might bind indirectly, interacting with RNAs (Figure S2B) through the RGG domain (Figure 2A). In pilot experiments, we found that unlike positive controls, DNA was not co-precipitated in chromatin immunoprecipitation (ChIP) assays with antibodies against SAF-A, suggesting the protein was not directly bound to chromatin. In contrast, cross-linking immunoprecipitation sequencing (CLIP-seq) experiments have shown that SAF-A is able to bind to virtually all classes of regulatory RNA.To analyze the mode of SAF-A binding, nuclear samples were treated with RNaseA/T1 or DNaseI before extraction into a soluble and insoluble (chromatin-associated) state; RNase treatment significantly abolished SAF-A chromatin binding, while DNaseI only had a limited effect (Figure 6D). To further explore the mechanics of SAF-A/RNA interaction with chromatin, we used an approach developed by based on the observation that caRNAs, detected by RNA-FISH with a human C0t1 probe in human-hamster hybrids, remain associated with the chromosome territory from where they are transcribed, in an unstable manner (t1/2 = 1 hr, Figure S6E). SAF-A depletion did not affect C0t1 RNA binding to chromosomes (Figure S6C) while transcription inhibition triggered a rapid loss of C0t1 or LINE-1 signal and chromosome territory compaction (Figure S6D), consistent with SAF-A binding to chromatin via caRNAs. We also found that SAF-A bound to newly transcribed caRNA through its RGG domain (Figure S6F) and further tested whether this binding was required for chromatin decompaction using the approach developed in Figure 6A. Re-expression of SAF-A ΔRGG was unable to decompact chromatin structures (Figure 6E) indicating that SAF-A oligomerization regulates large-scale chromatin structures through an interaction with caRNAs.

22Unexpectedly, our data indicates that altering large-scale chromatin structure, by depletion of SAF-A, only has limited impact on transcription (Figures 2 and S2). However, we speculated that interphase chromatin structure disruption would have an effect on genome integrity. We observed diffuse phosphorylated H2AX (γ-H2AX) signals in SAF-A-depleted RPE1 cells (Figure S7A), a response reflective of nuclear stress. Diffuse γ-H2AX signals appeared 48 hr after an alteration in large-scale chromatin structure suggesting that irregular chromatin triggers the response. We also characterized the effect of Walker A or Walker B mutations on genomic stability: only the oligomerization-deficient Walker A mutant promoted the accumulation of diffuse γ-H2AX pattern (Figure 7A) suggesting that inhibition of oligomerization and enforced chromatin compaction triggers a response, indicative of cellular mechanisms that monitor chromatin structure integrity. Spiral-shaped and distorted metaphase chromosomes were also observed in SAF-A-depleted cells, which also showed reduction in viability, G1/S phase arrest (Figures 7B and 7C), chromosomal segregation defects, and the formation of anaphase bridges staining positive for the BLM protein (Figure 7C). To assess the long-term impact of an alteration in chromatin structure on chromosomal instability, we performed chromosome painting FISH for HSA2 and HSA4 (Figure S7D). Control cells predominantly had two copies of HSA2 and HSA4 while 25% and 21%, respectively, have more than three chromosome signals after the depletion of SAF-A, indicating that an alteration in interphase chromatin structure causes chromosomal instability. A similar effect could be observed in cells expressing the SAF-A Walker A mutant, but not the Walker B mutant, indicating that large-scale chromatin structure, regulated by SAF-A oligomerization, is essential for chromosome stability (Figure 7D).

23We have demonstrated a previously unknown role for the abundant nuclear protein SAF-A in regulating interphase chromosome structure via oligomerization with caRNAs. ATP binding through the SAF-A AAA+ domain mediates oligomerization, while the RNA-binding domain of SAF-A directly links protein assembly with caRNAs. Our preliminary data indicates that caRNAs have a short half-life (∼1 hr) in the nucleus, and the caRNA pool is rapidly turned over suggesting that caRNAs should be considered as a dynamic nuclear compartment that is involved in chromatin structure regulation. It is therefore possible that caRNAs, generated by normal transcription, decompact large-scale chromatin structures facilitating their transport from the nucleus.

24Expanding both the classical concept of a nuclear matrix underpinning interphase chromatin domains and more recent observations of RNA as a fundamental chromatin component, we suggest that SAF-A, together with heterogeneous caRNAs forms a dynamic chromatin mesh that regulates chromatin structure. Unlike previously envisioned static models of the nuclear matrix, this process is dynamic and responsive to transcription. From our studies, it is not clear what structure SAF-A adopts. However, the protein shows compelling similarity to DnaA protein, and the presence of an ISM suggests it might be a spiral oligomer. The underlying regulatory mechanisms for triggering SAF-A oligomerization are unclear, but SAF-A has a number of predicted phosphorylation sites and interacts with known kinases. In addition, a SAF-A homolog, HNRNPUL1, shows 60% similarity, but there is no evidence for the presence of a Walker A or Walker B motif raising the possibility that it might function as a regulatory protein to “cap” growing SAF-A chains.

25An insight into the functional interplay between SAF-A and chromatin could be provided by recent simulation experiments suggesting that non-specific binding-induced-attraction is sufficient to generate local chromatin domains or chromatin micro-bodies. Formation is driven by protein-chromatin interactions mediated by RNA that could partition the genome into functionally open or closed domains, observed by chromosome conformation capture techniques, providing an evolutionary constraint for maintaining gene-rich domains together. Surprisingly, this model does not require explicit boundary elements suggesting that evolutionary domains could divide or merge depending on local transcription. It is also tempting to speculate that transcription ripples that occur over 100 kb in mammalian cells could limit the maximum size of functional domains. The regulation of chromatin micro-bodies is unknown but could be regulated by the cyclical ATP-dependent polymerization of SAF-A, while extensive RNA-chromatin interactions will provide a “glue” to stabilize local chromatin structures.

26Previous reports show that SAF-A is essential for mouse embryonic development, and numerous mutations have been reported in the AAA+ domain of SAF-A in the cancer genome sequencing projects with a pronounced cluster in the AAA+ lid domain (Figure S7E). It has also recently been reported that haploinsufficiency for SAF-A causes impaired neurological development in humans. Consistent with these pathological effects, our data shows that maintenance of interphase chromosome structure by SAF-A is necessary to maintain chromosome stability and aberrant chromatin structure triggers a DNA damage response (Figure 7). This suggests the mammalian genome has mechanisms for sensing changes in large-scale chromatin structure. This is signaled by an ATM/ATR-dependent response that leads to H2AX phosphorylation and a rapid cell-cycle arrest. While we show that large-scale chromatin structures are secondary to transcription, our data highlights that maintenance of chromosome organization is necessary for chromosome stability, implicating chromatin architecture as a guardian of the genome.

27Further information and requests for reagents should be directed to and will be fulfilled by the Lead Contact, Nick Gilbert (nick.gilbert@ed.ac.uk).

28RPE1, 293T and FlpIn-T-REx-293 cells (ThermoFisher) were cultured as described previously. GM10253A hybrid cells (Coriell) were cultured in RPMI1640 supplemented with 3 mM L-glutamine, 10% fetal calf serum, penicillin (100 U.ml-1), streptomycin (100 μg.ml-1) and phenol red (8.1 mg.l-1). Transcription was blocked by adding α-amanitin (50 μg.ml−1), flavopiridol (100 μM), actinomycin D (500 nM) or DRB (50 μM) to cells for the times indicated. Flp-In T-REx 293 was used to establish stable cell lines for inducible expression of FLAG-tagged SAF-A and their derivatives by transfecting a modified pcDNA5/FRT Expression vector (ThermoFisher) harboring N-terminal 3 × FLAG tagged SAF-A or its derivatives, and pOG44 (ThermoFisher) encoding Flp recombinase. After selection with 50 μg.ml-1 hygromycin B (ThermoFisher) for 2 weeks, the cells were treated with 1 μg.ml-1 doxycycline for 24 hr for inducing expression of proteins.

29cDNA for human SAF-A was obtained from Genome Cube (IRATp970D1041D). Truncation and point mutations of SAF-A were constructed by a standard PCR cloning strategy and inserted into the corresponding vectors with indicated tags. All plasmids were verified by DNA sequencing. siRNA resistant SAF-A was constructed by introducing silent mutations into the nucleotide sequence (2113-2124 bp) 5′-CCTGGGAATCGT-3′ as follows: 5′- CCaGGaAAcCGa −3′. The SAF-A cDNA and derivatives were inserted into a modified pcDNA5/FRT Expression vector (ThermoFisher) with an N-terminal 3 × FLAG tag for transient expression in 293T and RPE1 cells and for stable expression in FlpIn T-REx 293 cells. For recombinant expression in E. coli, SAF-A derivatives were cloned into a pET32a vector with an N-terminal 6 × His tag. For PLA, T7 SAF-A derivatives were cloned into a modified pCGN vector where the HA tag was replaced by a T7 tag.

30For siRNA treatment, cells (10%–20% confluent) were transfected with 10 nM Stealth RNA (ThermoFisher) using Lipofectamine RNAi MAX (ThermoFisher) for the time indicated. Stealth RNA sequences and the identification numbers are CCUGGGAAUCGTGGCGGATATAATA HSS104917 for SAF-A and GCUUUGCCUUCGUUCAGUAUGUUAA HSS179304 for hnRNP C. The control RNA is Stealth RNAi siRNA Negative Control, Med GC (ThermoFisher).

31For shRNA treatment, RPE1 cells (80% confluent) were transfected with pSuper-puro vector harboring hnRNP A1 target sequence using Lipofectamine 2000 (ThermoFisher). After 48 hr selection with 3 μg.ml-1 puromycin, DNA-FISH and western blotting were performed.

32To analyze Triton X-100 soluble/insoluble SAF-A, cells were treated with CSK buffer containing 100 mM NaCl, 0.1% Triton X-100, 300 mM Sucrose, 1 mM MgCl2, 1 mM EGTA, 10 mM PIPES (pH 6.8), and 100 μM PMSF (phenylmethylsulfonyl fluoride) on ice for 10 min, followed by immuno-staining or divided into the supernatant and pellet fractions by centrifugation, at 5000 rpm, 4°C for 5 min, for western blotting. To analyze DNase or RNase A/T1 extraction of SAF-A, cells were treated with CSK buffer containing 100 units.mL-1 DNase I (ThermoFisher) and 2.5 mM CaCl2 or 0.5/20 units.ml-1 RNase A/T1 (Ambion) for 30 min at R.T. and were divided into the supernatant and pellet by centrifugation, at 5000 rpm, 4°C for 5 min.

331 × 106 293T cells were exposed to disuccinimidyl suberate (DSS; 1 mM, 0.3 mM, 0.1 mM), 1,8-bismaleimido-diethyleneglycol (BM(PEG)2; 1 mM, 0.3 mM, 0.1 mM) or N-γ-maleimidobutyryl-oxysuccinimide ester (GMBS; 1 mM, 0.3 mM, 0.1 mM) crosslinking for 5 min at R.T. in PBS or permeabilised with 100 μL of reaction buffer including 125 mM NaCl, 50 mM Tris pH 7.0 (for BM(PEG)2) or 50 mM PIPES pH7.0 (for DSS or GMBS), 5 mM MgCl2 and 0.1% Triton X-100 at 37°C for 10 min before crosslinking. After quenching DSS with 20 mM Tris pH 7.0, BM(PEG)2 with 20 mM L-cysteine (Sigma-Aldrich) or GMBS with 20 mM Tris pH 7.0 and 20 mM L-cysteine, cells were processed for western blotting. 0.3 mM cross-linker was used for routine experiments while 1 mM was used for SAF-A IPs (Figures 4C and 4E).

34For Figure 4B right, reaction buffer was supplemented with 5 mM ATP or ATPγS. For Figure 4D, after pre-treatment with 3 μg.ml-1 PureLink RNaseA (ThermoFisher) in reaction buffer at 37°C for 10 min, 293T cell lysates were incubated in the presence or absence of 30 μg.ml-1 total RNA from 293T cells and 5 units.ml-1 Apyrase (NEB) at 37°C for 10 min, following by crosslinking. After RNaseA treatment reaction buffer was supplemented with RNasin Plus RNase Inhibitor (Promega) 1000 units.ml-1, 1 mM DTT and 5 mM CaCl2.

35Cells were suspended in NuPAGE LDS sample buffer (ThermoFisher) with 10 mM DTT, incubated at 100°C for 5 min and sonicated briefly. For SAF-A oligomer blots, protein samples were resolved on 5 or 8% bis-tris gels or 7% tris-acetate gels (ThermoFisher) and transferred to Immobilon-P PVDF 0.45 μm membrane (Merck Millipore) by wet transfer. Membranes were probed with antibodies using standard techniques and detected by enhanced chemiluminescence.

36FLAG-tagged SAF-A expressed in 293T cells was cross-linked with 1 mM 1,8-bismaleimido-diethyleneglycol (BM(PEG)2) or 1 mM dithio-bis-maleimidoethane (DTME), was extracted by sonication (10 × 30 s at amplitude 2 μm), immuno-purified using anti-FLAG M2 agarose beads (Sigma-Aldrich) and eluted with 100 μg.ml-1 3 × FLAG peptide (Sigma-Aldrich) suspended in 50 mM NaCl, 50 mM Tris pH 7.0, 5 mM MgCl2 and 0.1% Triton X-100. For Figure 4E, immunoprecipitants were incubated in the presence or absence of 10 mM DTT to reverse crosslinks, 0.5/20 units.ml-1 RNase A/T1, 5 units.ml-1 Apyrase, 1 mM ATP or 1mM ATPγS for 30 min at 37°C. A 4%–16% NativePAGE bis-tris gel (ThermoFisher) was used for native protein fractionation, according to the manufacturer’s instructions. In Figure 4C, 5-ethynyl uridine was conjugated to 0.2 mM biotin azide (PEG4 carboxamide-6-azidohexanyl biotin; ThermoFisher) using click chemistry, transferred to PVDF membrane, followed by incubation with ExtrAvidin−Peroxidase (Sigma-Aldrich) and detected with SuperSignal West Femto (ThermoFisher).

37To obtain SAF-A proteins, E.coli BL21 (DE3) cells transformed with a pET32a vector encoding SAF-A cDNA were grown in 2 × YT medium (16 g.ml-1 Tryptone, 10 g.ml-1 Yeast extract, 5 g.ml-1 NaCl) supplemented with 20 μg.ml-1 ampicillin and 20 μg.ml-1 chloramphenicol. Protein expression was induced overnight at 18°C with 1 mM IPTG after OD600 reached 0.6. Cells were lysed in buffer containing 50 mM Tris-HCl (pH 8.0), 500 mM NaCl, 60 mM imidazole, 10% glycerol, 0.1% Triton X-100 and 1 mM DTT. The SAF-A proteins were affinity-purified by HisTrap HP (GE Healthcare Life Sciences), dialysed into 150 mM NaCl, 30 mM imidazole (pH 7.0) and 1 mM DTT overnight at 4°C and were further purified by CM-Sepharose cation exchange chromatography (Sigma-Aldrich) and eluted using a 150 to 500 mM NaCl gradient in dialysis buffer.

38An ATPase activity assay was performed with 3.0 μm SAF-A protein in reaction buffer (50 mM Tris–HCl (pH 7.5), 125 mM NaCl, 10 mM MgCl2, 5 mM DTT, 1 mM ATP or 1 mM GTP) in a final volume of 50 μL for 30 min at 37°C. The 10 μL reaction was quenched with 90 μL of BIOMOL Green (Enzo Life Science) and free phosphate was detected by absorbance at 620 nm using a Tecan M200Pro. A phosphate standard curve was used to estimate the amount of phosphate released during ATP hydrolysis. Error bars denote standard deviations from three independent samples.

39Fluorescence experiments were performed at 37°C using a LS 50 PerkinElmer luminescence spectrometer. Reactions comprised 4 μm SAF-A in 50 mM Tris–HCl (pH 7.5), 125 mM NaCl, 10 mM MgCl2, 5 mM DTT. Spectra for SAF-A AAA+ fragments were obtained with 290 nm excitation wavelength and fluorescence changes calculated at 360 nm. All fluorescence measurements were performed in duplicate.

40FlpIn T-REx 293 cells carrying an inducible copy of wild-type or mutant SAF-A were transfected with stealth RNAi to knock down endogenous SAF-A using RNAi MAX (ThermoFisher) for 24 hr. Cells were then transfected with plasmids encoding T7-tagged SAF-A or point mutants with Lipofectamine 3000 (ThermoFisher) and the expression of FLAG-tagged-SAF-A was induced with doxycycline at the same time. After 24 hr, PLA was performed according to the manufacturer’s instructions using Duolink in situ (Sigma-Aldrich).

41Nascent transcription was measured as described.

42Cells were treated with 1 mM 5-ethynyl uridine (Base Click) and 1 mM thymidine (Sigma-Aldrich) for the indicated time. In Figure S6C, cells were pretreated with 50 ng.ml-1 actinomycin D. Click reaction was performed with Click-iT EdU Alexa Fluor 647 Imaging Kit or Click-iT Nascent RNA Capture Kit, according to the manufacturer’s instructions (ThermoFisher).

43After 30 min incubation with 0.5 mM 4-thiouridine cells were rinsed with PBS and lysed in DNA lysis buffer (100 mM EDTA, 20 mM Tris pH 7.5, 50 mM NaCl 0.5% SDS and 300 μg.ml-1 protease K). The cell lysate was incubate at 37°C for 30 min and precipitated in 2 volumes of ethanol. The precipitate was suspended in 70% ethanol overnight and resuspended in 50 μL of combined DNA and RNA lysis buffer (1 × DNaseI digestion buffer (NEB) with 1 mM ZnSO4). The nucleic acid was digested with 1 μL Dnase I (NEB) and 2 μL (1 mg.ml-1) nuclease P1 for 24 hr at 37°C. After this the sample was denatured by heat (95°C for 5 min) and rapidly cooled on ice. 2 volumes of 30 mM Na acetate pH 5.2, (1 mM ZnSO4) was added and the sample was digested with fresh DNaseI (1 μl) and Nuclease P1 (2 μl) for a further 24 hr at 37°C. 50 μL of lysate was separated by reverse phase HPLC.

44HPLC was carried out on a Dionex Ultimate 3000 HPLC system equipped with Chromeleon software, a column chiller (8°C), 5 μm APEX ODS C18 column, 50 mM ammonium phosphate (monobasic) mobile phase (1 ml.min-1), and a UV absorbance detector capable of collecting UV absorbance at 5 wavelengths simultaneously. Nucleotides were detected at their peak absorbances: thymidine 5′monophosphate (in DNA) at 268 nm (retention time 22.5 min), guanosine monophosphate (in RNA) at 254 nm (retention time 11.5 min), and 4-thiouridine monophosphate (in RNA) at 323 nm (retention time 14.7 min). Nucleotide quantities were measured by the areas under each peak at the respective peak UV absorbances.

45Total RNA from cells transfected with siRNA for the time indicated were prepared with RNeasy Mini kit (QIAGEN) and treated with ribo-Zero kit (illumina) to remove ribosomal RNA. RNA-seq libraries were prepared with a NEBNext® Ultra RNA Library Prep Kit for Illumina (NEB), according to manufacturer’s instructions, and were quantified using a BioAnalyzer 2100 with a High Sensitivity DNA Kit (Agilent). RNA-seq Libraries were analyzed by Edinburgh Genomics at the University of Edinburgh, UK, for quality control and sequencing (HiSeq2500, 50 base single-end reads). RNA-seq data were processed using a standard TopHat and Cufflinks methodology. Single end reads were aligned using Tophat 2.0.10 and Bowtie 2.1.0 using a human hg19 bowtie2 index. Aligned reads were processed using Cufflinks 2.2.1 with a human hg19 reference annotation and differential gene expression calculated between control and knockdown samples with Cuffdiff using a 0.0005 p value cut off.

463D RNA/DNA FISH was performed as described. Fosmid probes (BacPac resources), human LINE1 cDNA and human C0t1 DNA (ThermoFisher) were labeled in digoxigenin-11-UTP or biotin-16-dUTP. Custom MyTags libraries for Oligo-DNA-FISH were synthesized to cover 20 kb of genomic sequence positioned every 1 Mb across a 10 Mb region of the genome (MYcroarray). Gene rich library probes labeled with ATTO 488 cover chr11: 760000 – 10980000. Gene poor library probes labeled with ATTO 550 cover chr11: 20000000 – 29200000. Each 5 μm probe was hybridized as for DNA-FISH. Chromosome Painting was performed using a chromosome 2 painting probe labeled with Texas Red and a chromosome 4 painting probe labeled with FITC (Cytocell) or a chromosome 18 or chromosome 19 paint (MetaSystem) directly labeled with Spectrum Orange or FITC, respectively.

47Image capture and analysis was performed as described and image color balance was adjusted to improve data visualization in the manuscript.

48Super-resolution images were acquired using a STED microscope. Flp-In T-REx 293 cells depleted of endogenous SAF-A (48 h) and then doxycycline induced for 24 hr to express wild-type or mutant SAF-A were prepared on high precision cover-glass (Zeiss, Germany). STED images were acquired on a Leica SP5 SMD g-STED microscope (Leica Microsystems, Germany) using a 100 × objective lens (HC PL APO 100 × /1.40 Oil STED White) with immersion oil (Leica Type F, refractive index 1.5180). A confocal image was taken using a 488 nm laser for FLAG-SAF-A and 647 nm laser for DRAQ7 stained DNA. The STED image was taken using 488 nm for excitation from a white light, super continuum laser and a pulsed 592 nm laser for depletion. Detection was with a HyD detector with a time gate of 0.5-6 ns. STED image processing and analysis were carried out using the STED module of the LAS AF (Leica Application Suite Advanced Fluorescence). STED resolution was assessed by imaging 20 nm fluorescent beads (ThermoFisher). To measure image granularity nuclei were segmented on DRAQ7 to exclude background and nucleoli using an iVision script. The resulting mask was used to isolate SAF-A image data, background signal was removed and granules were segmented based on intensity script and quantified for each nucleus.

49Förster Resonance Energy Transfer (FRET) is a process whereby the transfer of energy occurs from an excited state fluorophore (donor) to a second chromophore (acceptor) in close proximity and is used to detect molecular interactions in vivo where two molecules are less than 10 nm apart. In FLIM-FRET microscopy, fluorescence lifetime is measured as a quantitative readout of FRET, with interactions causing a decrease in the fluorescence lifetime of the donor molecules. We performed FLIM-FRET experiments with 293T cells transiently transfected with either GFP-AAA+/RGG SAF-A fragment (donor alone) or GFP-AAA+/RGG and mCherry-AAA+/RGG together to investigate the oligomerization of SAF-A in different conditions. To deplete endogenous SAF-A, 293T cells were transfected with stealth RNA using RNAiMAX for 24 hr and transfected with GFP-AAA/+RGG and mCherry-AAA+/RGG wild-type or point mutants with Lipofectamine 3000 for 24 hr. Acquisition was performed on a Leica SP5 SMD (Single Molecule Detection) confocal laser-scanning microscope using a 63 × 1.4NA HCX PL Apo oil immersion objective lens. The GFP donor fluorophore was excited using a tunable white light laser operating at 488 nm and 40 MHz pulse rate. Emission was detected with an external single photon avalanche diode (MicroPhoton Devices) and photon arrival time relative to the laser pulse was determined using a PicoHarp 300 time-correlated single photon counting (TCSPC) module. FLIM measurements were integrated for 120 s with a 512 × 512 pixel format at a maximum photon count rate of 105 counts per second and subsequently binned to a 256 × 256 pixel format to maximize photon counts per pixel for accurate data fitting.

50FLIM analysis was carried out using SymPhoTime v5.4.4 (PicoQuant). The measured fluorescence decay was fitted to a bi-exponential decay model using the maximum-likelihood estimation method. Fluorescence lifetime values were determined by pixel-by-pixel fitting, which was guided by an initial global decay fit. Reported fluorescence lifetimes are an amplitude-weighted average of the two decay components, тamp = (A1т1 + A2т2) / (A1 + A2) (where A1 and A2 are the amplitudes of the decay components and т1 and т2 are the fluorescence lifetimes) and were only calculated for pixels with greater than 300 photon counts. FRET efficiency, EFRET = 1 – тDA/тD, where τDA is the lifetime of the donor (GFP) in the presence of the acceptor (mCherry) and тD is the single pixel median amplitude-weighted fluorescence lifetime reported for the GFP donor-only sample, calculated per pixel from the GFP-mCherry samples. Image maps representing FRET efficiency per pixel were constructed in R using the spatstat package.

51The domain composition of SAF-A was initially observed using SMART and then the most likely domain sequence lengths extended on the basis of homology searches of the Protein Data Bank (PDB) taking into consideration additional conserved secondary structure elements for the SAP (PDB ID: 1ZRJ), SPRY (PDB ID: 3TOJ) and AAA (3ZVL) domains, respectively. The C-terminal RNA binding region was annotated on the basis of functional studies. The disordered regions were predicted using MetaPrDOS by integrating results from 5 disorder prediction methods (see Figure S3A).

52The SAF-A isoform 2 AAA domain sequence (aa 469-653) was analyzed using PHYRE-2 to assess fold compatibility. All returned top hits were AAA superfamily related with 100% confidence. The AAA domain from the crystal structure of the mammalian polynucleotide kinase 3′ phosphatase PDB ID: 3ZVL chain A (1.65 Å) was selected as a template for modeling the SAF-A AAA domain based upon an HHPred search of the PDB70 database. The HMM-HMM target-template alignment was used as input for modeling after manual checking of alignment, secondary structure equivalence, gap positioning and undertaking minor editing. A PsiPred secondary structure prediction identified an additional predicted α -helix (aa 512-521) and an extended α-helix (aa 533-554) when compared with the template located after the canonical β strand 2; these additional predicted secondary structure features were restrained during the model building process. A total of 100 models were built using Modeler 9v12 and the model with the lowest “DOPE” energy was selected as the representative model, and assessed for valid stereochemistry (Ramachandran plot: 98.9% of residues in favored and allowed regions) and packing quality (average Z-score −1.04). PyMol (http://www.pymol.org) was used for 3D visualization, analysis and figure preparation. Pro-origami was used to create the 2-D cartoon topology schematic. The target-template alignment was generated and annotated using EsPript v3.

53The statistical significance of compaction was tested using a nonparametric Mann–Whitney U (Wilcoxon) test (using R programming). p < 0.05 was taken as statistically significant. For oligo probe DNA-FISH analysis, the Woolz image processing system, initially developed for the eMouseAtlas program, was used to analyze the distribution of 3D DNA-FISH spot images. MAPaint, a Woolz based interactive 3D segmentation tool, was used to delineate spot domains. From these domains convex hulls and their volumes were then computed. The open source Woolz image processing system is freely available from https://github.com/ma-tech/Woolz. For analyzing DAPI texture cells were grown overnight on slides, stained with DAPI and mounted. 12 bit images were collected using a 405 nm laser on a SP5 confocal microscope (Leica) using a 100 × objective and were segmented to exclude background and nucleoli using a custom iVision (BioVis) script. Segmented nuclei were sub-sampled 100 × each with an 8 × 8, 12 × 12, 16 × 16, 20 × 20, 24 × 24 and 28 × 28 window (i.e., 600 measurements per nucleus using an iVision script). Sub-sampled images were imported into R and visualized using the spatstat package. To quantify texture images were transformed to a gray level co-occurrence matrix (GLCM) using the radiomics package and second order matrix statistics were calculated.

54The accession number for RNA sequencing reported in this paper is NCBI GEO: GSE98541.

55R.-S.N., L.B., C.N., A.R.D., M.A., B.R., P.C.B., S.K.M., and N.G. performed laboratory experiments. D.C.S., A.B., B.H., R.S.S., and B.R. undertook data analysis. R.-S.N., L.B., R.R.D., S.K.M., and N.G. designed the study. R.-S.N., L.B., and N.G. wrote the manuscript with contributions from all the authors.