Research Article

DNA loop extrusion by human cohesin

See allHide authors and affiliations

Science  13 Dec 2019:
Vol. 366, Issue 6471, pp. 1338-1345
DOI: 10.1126/science.aaz3418

Cohesin extrudes DNA loops

DNA is folded into loops in eukaryotic cells by a process that depends on a ring-shaped adenosine triphosphatase complex called cohesin. Davidson et al. and Kim et al. now show that in the presence of the NIPBLMAU2 protein complex, the human cohesin complex can function as a molecular motor that extrudes DNA loops with high speed in vitro. In contrast to how it mediates sister chromatid cohesion, cohesin does not appear to entrap DNA topologically during loop extrusion. The results provide direct evidence for the loop extrusion model of chromatin organization and suggest that genome architecture is highly dynamic.

Science, this issue p. 1338, p. 1345

Abstract

Eukaryotic genomes are folded into loops and topologically associating domains, which contribute to chromatin structure, gene regulation, and gene recombination. These structures depend on cohesin, a ring-shaped DNA-entrapping adenosine triphosphatase (ATPase) complex that has been proposed to form loops by extrusion. Such an activity has been observed for condensin, which forms loops in mitosis, but not for cohesin. Using biochemical reconstitution, we found that single human cohesin complexes form DNA loops symmetrically at rates up to 2.1 kilo–base pairs per second. Loop formation and maintenance depend on cohesin’s ATPase activity and on NIPBL-MAU2, but not on topological entrapment of DNA by cohesin. During loop formation, cohesin and NIPBL-MAU2 reside at the base of loops, which indicates that they generate loops by extrusion. Our results show that cohesin and NIPBL-MAU2 form an active holoenzyme that interacts with DNA either pseudo-topologically or non-topologically to extrude genomic interphase DNA into loops.

In eukaryotic interphase cells, DNA is folded over long distances into loops and topologically associating domains (TADs) (13), which contribute to gene regulation and recombination (4, 5). These structures depend on cohesin (69), a ring-shaped ATPase (Fig. 1A) that is also essential for sister chromatid cohesion and belongs to the SMC (structural maintenance of chromosomes) family of protein complexes found in all kingdoms of life [(10); reviewed in (11)].

Fig. 1 Human cohesin extrudes DNA loops in an NIPBL-MAU2– and ATP hydrolysis–dependent manner.

(A) Illustration of cohesin, PDS5A/B, and NIPBL-MAU2. (B) Illustration of a flow cell containing tethered λ-DNA molecules stretched by perpendicular flow. (C) Coomassie staining of HeLa cohesin after SDS-PAGE. JF646-HaloTag ligand was visualized by epi-red excitation. (D) Coomassie staining of recombinant NIPBL-MAU2 after SDS-PAGE. (E) Fraction of DNAs that formed loops (mean ± SD from three independent experiments) in the presence of the indicated components; n = 503, 511, 493, 476 DNAs analyzed (from left to right). (F and G) DNA looping in the presence of the indicated components. DNA was visualized by Sytox Orange staining. Scale bar, 2 μm. (H) As (F), except buffer flow was paused prior to loop extrusion (left). A DNA loop, visible as a bright dot, formed in the absence of buffer flow (center) and was extended by resumption of perpendicular flow (right). (I) As (F), except paraformaldehyde (PFA) was subsequently introduced into the flow cell, which allowed visualization of the “arms” of the DNA loop. (J) Mean loop extrusion rate in the presence of NIPBL-MAU2, ATP, and HeLa or recombinant cohesinSTAG1. Median and quartiles are shown; n = 15, 36 DNAs analyzed (from left to right). (K) Fraction of DNAs (mean ± SD from three independent experiments) that formed loops after sequential introduction of the indicated components; n = 327 DNAs analyzed. (L) Fraction of DNAs (mean ± SD from three independent experiments) that formed loops in the presence of HeLa cohesin, NIPBL-MAU2, and ATP or CTP; n = 426, 366 DNAs analyzed (from left to right). (M) Fraction of DNAs (mean ± SD from three independent experiments) that formed loops in the presence of NIPBL-MAU2, ATP, and wild-type (wt), SMC1 K38A/SMC3 K38A ATP binding–deficient (KA), or SMC1 E1157Q/SMC3 E1144Q ATP hydrolysis–deficient (EQ) recombinant cohesinSTAG1; n = 344, 377, 406 DNAs analyzed (from left to right).

How cohesin forms loops and TADs is unknown, but one hypothesis posits that cohesin extrudes DNA loops until it encounters convergently oriented “loop anchor” DNA sequences that are bound by the zinc finger protein CTCF (12, 13). This hypothesis could explain why cohesin and CTCF colocalize (14, 15) at TAD boundaries (1, 2) and loop anchors (3), why these contain CTCF sites in convergent orientations (3), and why increasing cohesin’s residence time on chromatin causes the formation of longer loops (6, 9, 16) and relocalization of cohesin into axial chromosomal domains called “vermicelli” (13, 17). The loop extrusion hypothesis is also consistent with cohesin being highly mobile in the genome (18, 19) and is supported by the observation that the related SMC complex condensin can extrude loops in vitro (20), a property that can explain DNA folding in mitotic chromosomes (2123). However, there is no evidence that cohesin can form DNA loops by extrusion or that cohesin possesses a motor activity that would enable such a process, and it has been pointed out that loops could also be formed by sporadic encounters of distant DNA sequences (11).

ATP hydrolysis–dependent DNA looping by cohesin and NIPBL-MAU2

We therefore analyzed whether recombinant human cohesinSTAG1 complexes (fig. S1J) (24) can form DNA loops by using an assay (25) in which yeast condensin performs loop extrusion (20). In these experiments, linear 48.5-kbp λ-phage DNA molecules were tethered at both ends to a glass surface in a flow cell, stained with Sytox Orange, stretched by perpendicular buffer flow, and imaged by total internal reflection (TIRF) microscopy in the presence of cohesinSTAG1 and ATP to visualize DNA looping (Fig. 1B). However, under these conditions, no loop formation was observed (0/372 DNAs analyzed; fig. S1A). Because recombinant cohesin could differ from endogenous cohesin, we generated HeLa cells in which all SCC1 alleles were modified to encode SCC1-Halo-FLAG fusion proteins and affinity-purified a mixture of cohesinSTAG1 and cohesinSTAG2 (Fig. 1C), in which cohesinSTAG2 was more abundant than cohesinSTAG1 by a factor of 3 (26). However, this preparation also did not cause loop formation (0/503 DNAs analyzed; Fig. 1E, fig. S1B, and movie S1).

Because condensin formed loops (20) under conditions in which cohesin failed to do so, we next considered differences between cohesin and condensin. Although both complexes share a ring-like architecture in which two elongated SMC proteins are connected via their hinge domains and a kleisin subunit [reviewed in (11)], condensins are pentamers that contain two HAWKs (HEAT repeat proteins associated with kleisins) (27), whereas cohesin complexes are tetramers containing only one HAWK. In somatic human cells, this can either be STAG1 or STAG2 (Fig. 1A) (28). However, cohesin is regulated by three additional HAWKs (Fig. 1A). NIPBL is thought to load cohesin onto DNA (29, 30), whereas PDS5A and PDS5B have multiple functions, including release of cohesin from DNA (31), enabling cohesin acetylation (32), and contributing to TAD boundaries (9). We therefore tested whether cohesin forms loops in the presence of one of these HAWKs.

When we added recombinant PDS5A or PDS5B (fig. S1I) to HeLa cohesin, no DNA looping was observed (0/493 and 0/476 DNAs analyzed, respectively; Fig. 1E and fig. S1, C and D). However, when recombinant NIPBL bound to MAU2 (Fig. 1D) was combined with either HeLa cohesin or recombinant cohesinSTAG1, we observed looping in 238/511 (47%) and 142/344 (42%) of DNA molecules, respectively (Fig. 1, E, F, G, and M, fig. S2A, and movies S2 and S3). Looping was also observed in the absence of buffer flow [227/372 (61%) of DNAs], indicating that loops were formed actively by cohesin rather than passively by stretching DNA (Fig. 1H and fig. S2, B and C). Loops formed at a single position and incorporated DNA symmetrically from both ends (30/30 looping events; Fig. 1F and fig. S3, A and B), unlike loops formed by condensin, which extrudes DNA asymmetrically (20). Occasionally the base of cohesin loops moved along the DNA (Fig. 1, G and H). Similar observations were previously made using condensin mutated in a kleisin “safety belt” that normally encircles DNA (20), raising the possibility that cohesin lacks such a safety belt. Cohesin loops reached maximal size within 45 s from the onset of looping, and 424/447 (95%) of them were maintained during subsequent imaging for ~8 min. The remaining loops were released, returning DNA to its original stretched conformation (fig. S1E). Infrequent spontaneous cleavage of DNA likewise released looped DNA into its original shape (fig. S1F), ruling out the possibility that the observed structures were caused by recombination. The two “arms” of DNA loops were usually closely aligned (Fig. 1F) but occasionally moved apart from each other, revealing that they were only connected at the base of the loop (see figures below). Open DNA loops formed by cohesin could also be observed after paraformaldehyde fixation (Fig. 1I).

HeLa and recombinant cohesin formed loops at mean rates of 1 kbp s–1 and 0.8 kbp s–1, respectively (maximal rate 2.1 kbp s–1; Fig. 1J and fig. S3, C and D), similar to the rate reported for yeast condensin [maximal rate 1.5 kbp s–1 (20)]. The rate correlated inversely with the distance between the tethered DNA ends (Pearson correlation coefficient r = –0.35, P = 0.001; fig. S3E). Looping also occurred more frequently on DNA molecules in which the tethered ends were closely spaced (fig. S3F). These findings suggest that looping by cohesin, like looping by condensin (20), is prevented when the force required to bend DNA exceeds a threshold.

To investigate whether NIPBL-MAU2 functions in looping by first binding DNA and then recruiting cohesin, we introduced NIPBL-MAU2 and cohesin sequentially into a flow cell (Fig. 1K). No looping was observed when NIPBL-MAU2 was introduced (0/327 DNAs; Fig. 1K and fig. S1G), nor when cohesin was added subsequently (0/327 DNAs; Fig. 1K). In contrast, looping occurred on 225/327 (60%) of DNA molecules when cohesin and NIPBL-MAU2 were introduced simultaneously (Fig. 1K), indicating that cohesin can only loop DNA in the presence of NIPBL-MAU2.

Like yeast cohesin (33, 34), HeLa and recombinant cohesinSTAG1 hydrolyzed ATP slowly but were activated by as much as a factor of 30 by NIPBL-MAU2 and further by DNA (Fig. 2). Under these conditions, 2 (HeLa cohesin) or 1.7 (cohesinSTAG1) ATP molecules were hydrolyzed per cohesin complex per second (Fig. 2, C and D). DNA looping was supported by ATP [199/426 (47%) of DNAs analyzed] but not cytidine triphosphate (CTP) (0/366 of DNAs; Fig. 1L and fig. S1H) and did not occur in the presence of ATP binding–deficient or ATP hydrolysis–deficient mutants of cohesinSTAG1 [0/377 and 0/406 of DNAs, respectively, compared to 142/344 (41%) of DNAs in the presence of wild-type cohesinSTAG1; Fig. 1M and fig. S1, J and K]. Cohesin can therefore form DNA loops in a manner that depends on its ATPase activity and on NIPBL-MAU2.

Fig. 2 Cohesin’s ATPase activity is enhanced by NIPBL-MAU2 and DNA.

(A) Example of thin-layer chromatography/autoradiography assay used to measure [γ-32P]ATP hydrolysis in the presence of the indicated components. (B) ATP hydrolysis (mean ± SD from three independent experiments) determined from (A). (C and D) ATPase rates (mean ± SD from three independent experiments) in the presence of the indicated components determined using thin-layer chromatography/autoradiography.

Cohesin and NIPBL-MAU2 are located at the base of DNA loops during extrusion

If cohesin loops DNA by extrusion, it must be present at the base of loops. Indeed, Janelia Fluor 646 (JF646)–labeled HeLa and recombinant cohesinSTAG1 (Fig. 1C and fig. S1J), as well as NIPBL-MAU2 (Fig. 3G), were present at the base of loops and moved against buffer flow during their formation (Fig. 3, A, B, and H, fig. S4, and movies S4 and S5). To determine whether looping is mediated by single or oligomeric cohesin complexes, we measured the fluorescence intensity of HeLa cohesin at the base of loops (n = 62) and compared it to the intensity of cohesin bound to glass (n = 815), where the oligomeric state of cohesin could be analyzed by photobleaching (for technical reasons, we were unable to detect bleaching of cohesin on DNA). For both cohesin populations, fluorescence measurements yielded single peaks with similar mean intensities (217 and 242 a.u., respectively; Fig. 3, C and D), and in the latter case most signals bleached in a single step, indicative of monomeric cohesin [37/50 (74%) of cohesin signals analyzed; Fig. 3E], or did not bleach [8/50 (16%)]. In rare cases, cohesin on glass exhibited a fluorescence intensity of ~500 a.u., indicative of a dimeric state, and bleached in two steps by going through a 250-a.u. intermediate [5/50 (10%) of cohesin signals analyzed; Fig. 3F]. These data support the hypothesis that cohesin loops DNA by extrusion and does so as a monomer.

Fig. 3 Cohesin and NIPBL-MAU2 are located at the base of DNA loops during extrusion.

(A) DNA looping and cohesin localization during flow-in of JF646-labeled recombinant cohesinSTAG1, NIPBL-MAU2, and ATP. DNA was visualized by Sytox Orange staining. Lower panel shows merged images. Scale bar, 2 μm. (B) As (A), except DNA was incubated with JF646-labeled HeLa cohesin. Yellow arrowheads indicate position of cohesin during loop extrusion. (C and D) Fluorescence intensity distribution of HeLa cohesin (C) at the base of DNA loops (n = 62 cohesin signals) and (D) bound nonspecifically to the glass surface (n = 815 cohesin signals). (E) Time trace of cohesin signal binding nonspecifically to the glass surface and bleaching in one step. (F) As (E), except cohesin bleached in two steps. (G) Coomassie staining of recombinant NIPBL-MAU2 after SDS-PAGE. JF646-HaloTag ligand was visualized by epi-red excitation. (H) As (A), except DNA was incubated with unlabeled HeLa cohesin, JF646-labeled NIPBL-MAU2, and ATP.

Cohesin translocates rapidly along DNA against the direction of buffer flow

Yeast condensin can unidirectionally and ATP-dependently translocate along extended DNA molecules at 0.06 kbp s–1 (35). Our flow cells also contained some singly and doubly tethered DNA molecules that were stretched in the direction of buffer flow and not oriented perpendicular to it (fig. S2B, arrowheads). Along these, cohesin frequently translocated against the direction of buffer flow at a mean rate of 0.4 kbp s–1 [164 translocating cohesin signals along 345 DNAs analyzed (0.48 translocating cohesin signals per DNA); Fig. 4, A, B, C, E, and F]. The fluorescence intensity distribution of translocating cohesin (Fig. 4D) closely resembled that of cohesin at the base of loops and on glass (Fig. 3, C and D), indicating that cohesin also translocated as a monomer. We did not observe cohesin translocating in the direction of buffer flow, possibly because such cohesin molecules had reached the ends of DNA molecules before they could be imaged. Like loop extrusion, cohesin translocation was strictly dependent on ATP and NIPBL-MAU2 (Fig. 4, E and F). The relationship between cohesin translocation and loop extrusion is unknown, but we speculate that translocation, which occurs at approximately half the rate of loop formation, represents a partial extrusion reaction in which cohesin translocates on a single DNA rather than along the two arms of a loop.

Fig. 4 Cohesin translocates rapidly along DNA against the direction of buffer flow.

(A and B) Top: Illustrations of DNAs tethered at one or both ends and stretched in the direction of buffer flow. Lower panels: Kymographs of JF646-labeled HeLa cohesin translocating along singly tethered (A) and doubly tethered (B) DNA molecules against the direction of buffer flow in the presence of NIPBL-MAU2 and ATP. DNA was visualized by Sytox Orange staining. (C) Mean rate of JF646-labeled HeLa cohesin translocation against the direction of buffer flow in the presence of NIPBL-MAU2 and ATP. Median and quartiles are shown; n = 16 translocation events analyzed. (D) Fluorescence intensity distribution of cohesin translocating unidirectionally against the direction of buffer flow; n = 17 cohesin signals. (E) Number of JF646-labeled HeLa cohesin translocation events against the direction of buffer flow per DNA molecule (mean ± SD from three independent experiments) in the presence of NIPBL-MAU2 and either ATP or CTP; n = 179, 178 DNAs analyzed (from left to right). (F) Number of JF646-labeled HeLa cohesin translocation events against the direction of buffer flow per DNA molecule (mean ± SD from three independent experiments) in the presence of ATP ± NIPBL-MAU2; n = 177, 183 DNAs analyzed (from left to right).

ATP and NIPBL-MAU2 are required for ongoing loop extrusion by cohesin

The role of cohesin’s ATPase activity and NIPBL-MAU2 in loop extrusion could be indirect or direct (or both) by promoting loading of cohesin onto DNA or by enabling cohesin to extrude DNA, respectively. To distinguish between these possibilities, we first initiated loop extrusion by HeLa cohesin, NIPBL-MAU2, and ATP for 8 min (Fig. 5A, step 1), then replaced the content of the flow cell with different combinations of NIPBL-MAU2, ATP, and CTP and monitored the loops for another 8 min (Fig. 5A, step 2). Throughout step 1, ~95% of loops remained stable (Fig. 5, F to I, step 1). When the content of the flow cell was replaced with NIPBL-MAU2 and ATP, 98/104 (94%) of the remaining loops persisted (Fig. 5, B and F, fig. S5A, and movie S6), indicating that loops were maintained by DNA-bound cohesin that did not exchange with freely diffusing cohesin. When we substituted ATP for CTP (Fig. 5, C and G, fig. S5B, and movie S7) or introduced ATP in the absence of NIPBL-MAU2 (Fig. 5, D and H, fig. S5C, and movie S8), only 2/95 (2%) and 3/92 (3%) of loops were maintained, respectively. ATP and NIPBL-MAU2 are therefore required for the maintenance of loops. These results, and our observations that cohesin can only loop DNA in the presence of NIPBL-MAU2 (Fig. 1K) and that NIPBL-MAU2 is located at the base of loops during their extrusion (Fig. 3H), suggest that NIPBL-MAU2 and cohesin’s ATPase activity are both required for cohesin-mediated loop extrusion per se. Because 97% of DNA loops fell apart within 8 min after removal of unbound NIPBL-MAU2, most NIPBL-MAU2 must interact with cohesin transiently and dissociate from cohesin within this time period.

Fig. 5 ATP and NIPBL-MAU2 are required for ongoing loop extrusion by cohesin.

(A) Experimental setup. Step 1: DNA molecules were stretched by perpendicular flow for 8 min in the presence of the indicated components. Step 2: The content of the flow cell was replaced with the indicated components for 8 min. (B) A DNA loop formed during step 1 and maintained during flow-in of NIPBL-MAU2 and ATP in step 2. DNA was visualized by Sytox Orange staining. Scale bar, 2 μm. (C) As (B), except the content of the flow cell was replaced with NIPBL-MAU2 and CTP in step 2. (D) As (B), except the content of the flow cell was replaced with ATP in step 2. (E) As (B), except the content of the flow cell was replaced with NIPBL-MAU2 and ATP in buffer containing 300 mM NaCl in step 2. The brightness of the step 2 image was enhanced to compensate for the reduced Sytox Orange signal in 300 mM NaCl. (F to I) Fraction of loops (mean ± SD from three independent experiments) maintained in the indicated conditions; n = 108, 104, 97, 138 loops analyzed (from left to right).

The topology of cohesin-DNA interactions during loop extrusion

Cohesin is thought to mediate cohesion by topologically entrapping sister DNA molecules inside its ring structure (36), but cohesin can also entrap single DNAs (37). Thus, topological DNA entrapment could be a prerequisite for loop extrusion. Alternatively, cohesin could extrude loops “pseudo-topologically” by threading a loop of DNA through its ring structure (Fig. 6K), or non-topologically by extruding in a manner that does not involve encircling DNA at all (37). To distinguish between these possibilities, we tested whether loop extrusion is resistant to high-salt treatment [a property reported for topological cohesin-DNA interactions (29)] and whether loop extrusion depends on the integrity of the cohesin ring and on the ability of cohesin’s ring-forming subunits to transiently dissociate from each other, which would be required for topologically entrapping DNA.

Fig. 6 Loop extrusion by cohesin does not require opening of the tripartite ring.

(A) Illustration of single-chain trimer cohesin. (B) Coomassie staining of single-chain trimer cohesin incubated either with BMOE (+) or with BMOE that had been inactivated by preincubation with dithiothreitol (DTT) (–). Mobility shift indicates hinge–cross-linked molecules; percentage of cross-link efficiency is indicated (mean ± SD). (C) Rotary shadowing electron micrographs of single-chain cohesin incubated with BMOE ± DTT as described in (B). Scale bar, 50 nm. (D) ATPase rate (mean ± SD from three independent experiments) in the presence and absence of the indicated components, determined as in Fig. 2. (E) Coomassie staining of recombinant STAG1 after SDS-PAGE. (F) Fraction of DNAs that formed loops (mean ± SD from three independent experiments) in the presence of NIPBL-MAU2, ATP, and single-chain trimer cohesin incubated with BMOE ± DTT as described in (B) and then preincubated with or without STAG1; n = 429, 481, 432, 387 DNAs analyzed (from left to right). (G) Mean loop extrusion rate in the presence of STAG1, NIPBL-MAU2, ATP, and single-chain trimer cohesin incubated with BMOE ± DTT as described in (B). Median and quartiles are shown; n = 17, 17 DNAs analyzed (from left to right). (H and I) DNA looping in the presence of the indicated components. Single-chain trimer was incubated with BMOE ± DTT as described in (B). DNA was visualized by Sytox Orange staining. Scale bar, 2 μm. (J and K) Models of cohesin-dependent loop extrusion. (J) Ongoing loop extrusion is dependent on ATP hydrolysis by cohesin and on NIPBL-MAU2, which interacts with cohesin transiently. (K) Covalent fusion of all ring interfaces does not abolish loop extrusion by cohesin, consistent with models in which cohesin extrudes loops either non-topologically or pseudo-topologically.

To address the first question, we initiated loop formation in the presence of 50 mM NaCl (Fig. 5A, step 1) and then increased the salt concentration to 300 mM (Fig. 5A, step 2). Under these conditions, only 1/133 (<1%) of loops persisted (Fig. 5, E and I, and fig. S5D); such a result is consistent with non-topological DNA entrapment by cohesin, but also with the possibility that interactions between cohesin and NIPBL-MAU2 are salt-sensitive.

We next generated a version of cohesinSTAG1 in which the kleisin SCC1 can be cleaved by tobacco etch virus (TEV) protease (fig. S6, A and B) (24), mimicking cleavage of cohesin by separase in metaphase. TEV cleavage reduced but did not abolish cohesin recruitment to DNA (fig. S6C); however, it prevented loop extrusion [1/473 (<0.3%) DNAs; fig. S6D]. This result is consistent with the possibility that loop extrusion requires topological entrapment of DNA, but likewise with the possibility that SCC1’s integrity is required for cohesin’s extrusion function.

To distinguish between these possibilities, we engineered a form of cohesin in which all interfaces could be covalently linked to prevent topological DNA entrapment. To do this, we generated a 369-kDa in-frame fusion of SMC3, SCC1, and SMC1, which were connected by linker sequences (Fig. 6A). In this “single-chain trimer,” we replaced all cysteine residues with serines and introduced a pair of cysteines in the hinge domains of SMC1 and SMC3 to allow cross-linking by bis-maleimidoethane (BMOE). Cross-linking efficiency was 82 ± 6%, as measured by SDS–polyacrylamide gel electrophoresis (PAGE) mobility shift (Fig. 6B). Rotary shadowing electron microscopy revealed that single-chain trimers formed ring-like structures indistinguishable from the ones formed by HeLa (38) and recombinant cohesinSTAG1 (24), except for the absence of a density corresponding to STAG1 or STAG2 (Fig. 6C; note that these rings appear “open” because SCC1 is difficult to visualize). Like recombinant “wild-type” cohesinSTAG1, single-chain cohesin hydrolyzed ~1.7 molecules of ATP per second in the presence of NIPBL-MAU2 and DNA (Fig. 6D). ATP hydrolysis rates were not reduced by BMOE treatment and, as observed for yeast cohesin (34), were unchanged in the presence of STAG1 (Fig. 6, D and E).

Although non–cross-linked single-chain trimer is an active ATPase, it did not form DNA loops (0/429 DNAs; Fig. 6F). However, when this protein was preincubated with recombinant STAG1, loop extrusion occurred frequently [100/481 (21%) of DNAs; Fig. 6, F and G, and movie S9]. Remarkably, treatment of single-chain trimer with BMOE neither prevented nor even reduced loop extrusion in the presence of STAG1 [98/432 (23%) of DNAs; Fig. 6, F and G, and movie S10]. To test whether residual non–cross-linked cohesin could account for this activity, we diluted non–cross-linked single-chain trimer bound to STAG1 to a concentration corresponding to the 18% that failed to be cross-linked in the previous experiments. This reduced the fraction of DNAs that were extruded from 21% to 0.5% (2/387 DNAs; Fig. 6F), which suggests that the residual amounts of non–cross-linked cohesin are not sufficient to explain loop extrusion after BMOE treatment. Instead, these results indicate that cohesin can extrude loops even when all its interfaces have been covalently linked and that cohesin can therefore perform this process without topologically entrapping DNA. This implies that cohesin performs loop extrusion by interacting with DNA either pseudo-topologically or non-topologically (Fig. 6K). Furthermore, these experiments revealed that STAG1 has an important role in loop extrusion, even though it is dispensable for cohesin’s DNA-dependent ATPase activity.

Discussion and conclusions

Our results show that in the presence of NIPBL-MAU2, single cohesin complexes can form DNA loops by extrusion in a manner that depends on cohesin’s ATPase activity. How cohesin forms loops in the context of chromatin remains unknown. However, the high frequency and speed of loop extrusion in vitro, combined with the fact that the loop extrusion hypothesis can explain cellular phenomena such as the CTCF convergence rule (3) and vermicelli formation (13, 17), strongly support the hypothesis that cohesin also forms loops and TADs by extrusion in vivo, thereby facilitating key processes such as gene regulation (4) and immunoglobulin gene recombination (5). Given the high speed of loop extrusion observed in our experiments (mean rate 1 kbp s–1), current estimates of loop formation rates in HeLa cells [0.375 kbp s–1 (7)], and the number of cohesin complexes that have been identified on chromatin [~160,000 during G1 phase in HeLa cells (26)], an entire human genome could be extruded rapidly, possibly in a few minutes. This implies that loop extrusion is a highly dynamic and therefore regulatable process.

Our observation that NIPBL-MAU2 is not only required to initiate loop extrusion but is essential throughout this process has several important implications. First, it indicates that even though cohesin-NIPBL-MAU2 interactions are short-lived in vitro, the active loop-extruding holoenzyme is cohesin bound to NIPBL-MAU2. Like condensin, cohesin therefore needs two HAWKs for this process: STAG1 or STAG2 and NIPBL. In contrast, and as previously suggested (9, 34), PDS5 proteins may be HAWKs that inhibit loop extrusion, as they can compete with NIPBL for binding to cohesin (34, 39), do not stimulate cohesin’s ATPase activity (34), and are required for TAD boundaries (9).

Second, and consistent with cellular observations (16, 34, 40), cohesin’s ATPase and NIPBL-MAU2 cannot be required for cohesin loading alone but instead must have direct functions in loop extrusion (Fig. 6J). In the case of cohesin’s ATPase activity, one of these functions could be the generation of force to move DNA, and in the case of NIPBL-MAU2 the stimulation of cohesin’s ATPase activity (33, 34) (Fig. 2). Given that cohesin’s ATPase activity and NIPBL-MAU2 function in loop extrusion, it will be worthwhile to reinvestigate whether these factors do in fact load cohesin onto DNA or whether they are merely required to keep cohesin there—for example, by enabling processive loop extrusion. Consistent with this possibility, yeast cohesin’s constitutive HAWK Scc3 (orthologous to human STAG1 and STAG2) is also needed for maintaining cohesin on DNA (41) and its NIPBL ortholog Scc2 is required for retaining cohesin on chromatin in G1 phase (42). Such a scenario could also explain why a loading factor has never been identified for condensin. This could either be because condensin’s HAWKs are performing this function or simply because there is also no bona fide loading complex for cohesin. Our identification of cohesin-NIPBL-MAU2 as a potent loop extrusion factor will enable addressing these questions and the mechanistic basis of cohesin-dependent loop extrusion.

Supplementary Materials

science.sciencemag.org/content/366/6471/1338/suppl/DC1

Materials and Methods

Figs. S1 to S6

Movies S1 to S10

References (4346)

References and Notes

Acknowledgments: We thank C.H. Haering for sharing reagents, M. Madalinski for purifying JF646-HaloTag Ligand, G. Litos for purifying proteins, D. Drechsel for characterizing single-chain trimer cohesin, T. Lendl for image analysis support, and P. Pasierbek and other members of IMP/IMBA Biooptics facility for assistance. Electron microscopy was performed by the EM Facility of the Vienna BioCenter Core Facilities GmbH, member of Vienna BioCenter, Austria. Funding: Research in the laboratory of J.-M.P. is supported by Boehringer Ingelheim, the Austrian Research Promotion Agency (Headquarter grant FFG-852936), the European Research Council under the European Union’s Horizon 2020 research and innovation program GA No 693949, and the Human Frontier Science Program (grant RGP0057/2018). B.B. was supported by long-term fellowships from the European Molecular Biology Organization and the Human Frontier Science Program. Author contributions: I.F.D., B.B., G.W., and J.-M.P. designed experiments; I.F.D. performed experiments, purified proteins, and analyzed data; B.B. generated and purified STAG1 and single-chain cohesin and optimized cohesin purification; D.G. generated NIPBL-MAU2 and performed initial ATPase assays; B.B. and W.T. generated the SCC1-Halo-Flag HeLa cell line; G.W. purified HeLa cohesin; I.F.D. and J.-M.P. wrote the manuscript with input from all authors; J.-M.P. supervised the study. Competing interests: The authors declare no competing interests. Data and materials availability: Materials will be provided upon request and completion of a material transfer agreement. Data described in this manuscript are archived at the Research Institute of Molecular Pathology and will be provided upon request.
View Abstract

Stay Connected to Science

Navigate This Article