ATP dependent DNA transport within cohesin: Scc2 clamps DNA on top of engaged heads while Scc3 promotes entrapment within the SMC-kleisin ring

In addition to extruding DNA loops, cohesin entraps within its SMC-kleisin ring (S-K) individual DNAs during G1 and sister DNAs during S-phase. All three activities require related hook-shaped proteins called Scc2 and Scc3. Using thiol-specific crosslinking we provide rigorous proof of entrapment activity in vitro. Scc2 alone promotes entrapment of DNAs in the E-S and E-K compartments, between ATP-bound engaged heads and the SMC hinge and associated kleisin, respectively. This does not require ATP hydrolysis nor is it accompanied by entrapment within S-K rings, which is a slower process requiring Scc3. Cryo-EM reveals that DNAs transported into E-S/E-K compartments are “clamped” in a sub-compartment created by Scc2’s association with engaged heads whose coiled coils are folded around their elbow. We suggest that clamping may be a recurrent feature of cohesin complexes active in loop extrusion and that this conformation precedes the S-K entrapment required for sister chromatid cohesion.


SUMMARY 23
In addition to extruding DNA loops, cohesin entraps within its SMC-kleisin ring (S-K) 24 individual DNAs during G1 and sister DNAs during S-phase. All three activities require 25 related hook-shaped proteins called Scc2 and Scc3. Using thiol-specific crosslinking we 26 provide rigorous proof of entrapment activity in vitro. Scc2 alone promotes entrapment of 27 DNAs in the E-S and E-K compartments, between ATP-bound engaged heads and the SMC 28 hinge and associated kleisin, respectively. This does not require ATP hydrolysis nor is it 29 accompanied by entrapment within S-K rings, which is a slower process requiring Scc3. 30 Cryo-EM reveals that DNAs transported into E-S/E-K compartments are "clamped" in a sub-31 compartment created by Scc2's association with engaged heads whose coiled coils are 32 folded around their elbow. We suggest that clamping may be a recurrent feature of cohesin 33 complexes active in loop extrusion and that this conformation precedes the S-K entrapment 34 required for sister chromatid cohesion. are entrapped in S-K rings or other closed compartments, for example the E-S and E-K 139 compartments between ATP-bound engaged heads and the SMC hinge and associated 140 kleisin, respectively, or indeed other types of compartment created by multiple contacts 141 between HAWKs and SMC proteins. For these reasons, we describe here the use of thiol-142 specific crosslinking to measure bona fide topological entrapment of DNAs within S-K rings 143 in vitro. Both Scc2 and Scc3 are essential for this process, as are their abilities to bind DNA. 144 The process is dependent on ATP binding and stimulated by its hydrolysis, a feature largely 145 absent from previous assays (Minamino et al., 2018) . 146 147 Remarkably, we find that Scc2 alone promotes the rapid entrapment of DNAs within E-S and 148 E-K compartments in a process that is not accompanied by entrapment within S-K rings, 149 and propose that E-S/E-K entrapment occurs simultaneously through a single mechanism. 150 Because E-S/E-K entrapment is an order of magnitude more rapid than S-K entrapment, we 151 suggest that creation of the former by Scc2 may be a precursor to the latter, a process 152 contingent on the action of Scc3. Electron cryo-microscopy (cryo-EM) of complexes formed 153 between cohesin's SMC-kleisin trimers and linear or circular DNAs in the presence of Scc2 154 suggests that entrapment within E-S/E-K compartments involves transport of DNA between 155 ATPase heads prior to their engagement, whereupon DNAs are "clamped" in a sub 156 compartment formed by Scc2's association with engaged heads in a manner similar to that 157 recently observed in a complex between DNA and both human and Schizosaccharomyces 158 pombe cohesin associated with both Scc2 NIPBL/Mis4 and Scc3 SA2/Psc3 (Higashi et al., 2020; Shi 159 et al., 2020) . Our observations reveal key insights into the biochemical activities of Scc2 and 160 Scc3 and suggest that the recurrent clamping of DNAs by Scc2 NIPBL/Mis4 and engaged heads 161 resulting in E-S/E-K entrapment, followed by their subsequent release, may be an integral 162 aspect of cohesin's ability to load onto and translocate along DNA. 163 individual cohesin rings, the DNA retardation should be much greater. Those DNAs retarded 199 by entrapment within a single ring correspond to the CMs previously observed in vivo 200 (Gligoris et al., 2014). Ladder formation required cysteine pairs at all three interfaces. It was 201 never observed with linear DNA (Fig. S1D) or when just a single cysteine (Fig. 1E) or BMOE 202 ( Fig. 1F) was omitted. Crucially, the ladders were strictly dependent on addition of ATP (Fig.  203   1G). 204 205

Entrapment of DNAs by S-K rings requires Scc3 and is stimulated by Scc2 206
To assess the roles of Scc2 and Scc3, we measured ladder formation at four successive 10 207 min intervals in the presence and absence of the two proteins. Ladders indicative of 208 entrapment increased with time (up to 40 min), suggesting that formation is a slow process, 209 were greatly reduced by omission of Scc2 ( Fig. 2A), and almost completely abolished by 210 omission of Scc3 (Fig. 2B). In the absence of both Scc2 and Scc3, the level of entrapment 211 was comparable to that observed in the presence of Scc2 alone (data not shown). 212 213

Entrapment of DNAs by S-K rings depends on ATP binding to Smc3 and on ATP 214 hydrolysis 215
To address the role of cohesin's ATPase, we mutated Smc3's Walker A site (Smc3K38I) to 216 abolish ATP binding to Smc3. This almost completely abolished entrapment (Fig. 2C). We 217 did not test the effect of mutating the equivalent residue in Smc1 as this has previously been 218 shown to abolish association of Smc1/3 heterodimers with Scc1 (Arumugam et al., 2003) . 219 We next tested the effect of mutating both Walker B sites to residues that permit ATP binding 220 but strongly inhibit hydrolysis (Smc1E1158Q Smc3E1155Q, 'EQEQ') ( Fig. S1B), which 221 caused a more modest, albeit still significant, reduction (Fig. 2D). These data suggest that 222 cohesin's ability to complete the ATP hydrolysis cycle stimulates entrapment but is not strictly 223 necessary. To address whether Smc3's K112 K113 are also important we analysed the effect 224 of substituting them by glutamine (Smc3K112Q K113Q), mutations thought to mimic the 225 acetylated state. This also reduced S-K entrapment (Fig. S1C), an effect that parallels its 226 abrogation of cohesin loading in vivo (Hu et al., 2015). 227 228

DNA binding to Scc3 is required for its entrapment by S-K rings 229
During a search for cohesin domains that bind DNA, we discovered that Scc3's association 230 with a fragment of Scc1 containing residues 269-451 greatly stimulates its association with 231 double stranded DNA, as measured using an electrophoretic mobility shift assay (EMSA) that recently observed in a co-crystal of DNA bound to condensin's Ycg1 HAWK bound to 234 its kleisin partner Brn1 (Kschonsak et al., 2017), we mutated two clusters of positively 235 charged residues (Scc3K224E K225E R226E and Scc3 K423E K513E K520E) on opposite  236 sides of the groove within Scc3 that is equivalent to Ycg1's DNA binding groove (Fig. 3A). 237 Neither triple (3E) mutant eliminated DNA binding (Fig. S2B) nor caused lethality (Fig. S2E). 238 Despite this, both reduced cohesin's association with all genomic sequences except point 239 centromeres (CENs) (Fig. S2C & D). In contrast, combining the two triple mutations (to 240 create 6E) was lethal (Fig. S2E), abolished binding of Scc3 to DNA in the presence of Scc1 241 surrounding CENs). This distribution resembles that of Scc2 in wild type cells and indeed, 247 scc3-6E had little or no effect on Scc2's accumulation with CENs (Fig. 3E). This implies that 248 cohesin containing Scc3-6E forms complexes with Scc2 at CENs but subsequently fails to 249 form a stable association with chromatin or translocate into neighbouring sequences. Our 250 ability to detect such complexes at CENs but not at other loading sites along chromosome 251 arms can be attributed to the fact that Scc2's partner Scc4 binds to the kinetochore protein 252 Ctf19 and this association transiently tethers complexes at CENs while they are attempting 253 to load (Hinshaw et al., 2017). Though accumulation of cohesin bound by Scc2 at CENs 254 does not depend on Scc3's ability to bind DNA, it does still require Scc3 (Fig. S2F) The S. pombe Scc2/4 complex has previously been shown to bind DNA in vitro (Murayama 265 and Uhlmann, 2014) but the physiological significance of this activity has never been investigated. EMSA revealed that S. cerevisiae Scc2 also binds DNA (Fig. 4A), as do Scc2/4 267 complexes with slightly higher affinity (Fig. S3A). Unlike Scc3, whose DNA binding was 268 greatly enhanced by Scc1, DNA binding by Scc2 was reduced by addition of a Scc1 fragment 269 (Scc1 150-298 ) that contains sequences necessary for Scc2-dependent loading in vivo (Fig.  270 4A) (Petela et al., 2018). Interestingly, the inhibitory effect of Scc1 150-298 was not observed in 271 the binding of DNA to full length Scc2/4 (Fig. S3A) 4C) but also S-K entrapment 287 (Fig. 4D). These results suggest that Scc2's ability to bind DNA has a crucial role in 288 entrapping DNA within S-K rings in vitro, an activity also required for loading cohesin onto  ( Fig. S4A), or with simultaneous crosslinking of the hinge (Fig. S4B), to measure entrapment 300 of DNAs in J-S compartments (Fig. 1A ii). J crosslinking alone or in combination with hinge 301 (J-S) or kleisin (J-K) was efficient even in the presence of ATP, DNA, Scc2, and Scc3 (Fig.  302   S4C). In other words, both J-S and J-K circularisation occurred efficiently under conditions 303 that promote efficient entrapment of DNAs inside S-K rings. However, DNAs were never 304 entrapped within J-S compartments and only rarely by J-K ones (Fig. S4D). J-K entrapment 305 was not only much less frequent than S-K entrapment but also independent of Scc2. The 306 fact that J-K entrapment was comparable to S-K entrapment in the absence of Scc2 307 (compare Fig. 2A and S4D) suggests that the low-level entrapment of DNAs in S-K rings 308 induced by Scc3 alone may in fact correspond to DNAs entrapped in J-K compartments. 309 Though J-K circularisation by BMOE is modestly lower than that of S-K, this cannot account 310 for its far lower DNA entrapment. We therefore suggest that most ATPase heads associated 311 with DNA entrapped within S-K rings in vitro are not juxtaposed. They are either fully 312 disengaged, in the E-state, or in some other conformation. 313 314

Rapid DNA entrapment in E-S and E-K compartments 315
We used the same approach to measure entrapment in E-S or E-K compartments ( Fig. 1A  316 iii), in this case replacing J-specific cysteines by a pair specific for the E-state 317 of Scc2 and Scc3 (Fig. 5A). Notably, both processes occurred much more rapidly than S-K 331 entrapment, with significant amounts of DNA entrapped by multiple rings within 2 min (Fig.  332 5B). Because S-K entrapment occurs much more slowly, the efficient entrapment of DNAs In contrast to entrapment within S-K rings, which depends on Scc3, the rapid entrapment of 341 DNA in the E-S/E-K compartments was Scc3 independent (Fig. 5C). However, it was highly 342 dependent on Scc2, both in the presence or absence of Scc3 ( is present, neither DNA nor Scc2 are required for efficient E-state crosslinking of wild type 361 complexes but both are important for Smc3K38I complexes (Fig. S5D). Smc3K38I 362 presumably destabilises head engagement in a manner that can be overcome by the 363 presence of Scc2 and DNA. To explore whether Scc2 and DNA also promote head 364 engagement of otherwise wild type complexes, we tested their effect when ATP's ability to 365 promote head engagement is compromised by omission of Mg 2+ (Fig. S5E). Under these 366 circumstances, addition of both Scc2 and DNA restored efficient head engagement and both ATP-dependent head engagement suggests that DNA binds to a site created by head 369 engagement as well as to Scc2. 370 371 Scc2 causes DNAs to be entrapped in E-S and E-K compartments without entering S-372

K rings 373
Because Scc3 is crucial for S-K entrapment, the rapid entrapment of DNAs within E-S and 374 E-K compartments in the presence of Scc2 alone should be unaccompanied by S-K 375 entrapment. This is indeed the case. In contrast with E-S or E-K entrapment, which is very 376 efficient, few if any DNAs are entrapped in S-K compartments by 2 min (Fig. 5H). Though 377 paradoxical, this striking observation has a very simple explanation. The similarity in kinetics 378 suggests that E-S and E-K entrapments are created simultaneously as part of the same 379 reaction. In other words, a single type of DNA passage followed by head engagement gives 380 rise to both types. We envisage two types of mechanism to explain how this occurs without 381 S-K entrapment. According to the first (and simplest), DNA moves "upwards" between 382 disengaged ATPase heads, and is subsequently trapped in the E-S compartment following 383 ATP-driven head engagement (Fig. 9A). An alternative is that a loop of DNA is inserted into 384 an open S-K ring. If one of the loop's segments were located above the ATPase domains 385 while the other below, then subsequent head engagement would lead to simultaneous 386 entrapment in both E-S/E-K compartments (Fig. 9B). Neither type of DNA movement 387 involves passage through a gate created by opening the S-K ring, hence explaining the lack 388 of S-K entrapment. Entrapment within E-S/E-K, but not S-K compartments, in the presence 389 of Scc2 alone was also observed with relaxed (nicked) DNAs (data not shown).  partially open or closed (Fig. 7E). 500 described for human Scc2 NIPBL (Shi et al., 2020) (Fig. 8A). Scc2 holds DNA by way of a 504 curved basic and polar surface located around the transition between its neck and head 505 regions (Fig. 8B). The surface, which causes the DNA to bend slightly (~9°), is created by 506 the spatial arrangement into a semicircle of a series of residues (e.g. S508, N557, K714, 507 S753, S783, and K1324/25) from the ends of six α-helices and one loop (containing K427) 508 that together engulf the phosphate backbone. This region includes all four positions mutated 509 in Scc2-4E (S717E K721E K788E H789E, Fig. 4), thus neatly explaining why the charge-510 reversing mutations lowered the binding affinity for DNA and inhibited entrapment of DNAs 511 in vitro. 512 513

Scc2 binds both Smc1 and Smc3 514
Clamping involves not only the binding of DNA to Scc2 and to engaged heads (see below) 515 but also entrapment in a novel compartment created by Scc2's association with both Smc1 516 and Smc3. The latter involves multiple binding sites (Fig. 8C) and is therefore much more 517 delocalized than Scc2's DNA binding. Highly prominent is the binding of Scc2 to Smc1 518 through the docking of HEAT repeats 18-24 (residues 1127-1493) onto the F-loop (residues 519 1095-1118) on Smc1 head's C-lobe and the coiled coils that emerge above (Fig. 8C i). This 520 mode of association is highly analogous to the binding of S. cerevisiae condensin Smc4 by 521 the HAWK Ycs4 (Lee et al., 2020) (Fig. S6). Scc2 simultaneously interacts with Smc3's N-522 lobe -thereby providing a mechanism by which it promotes head engagement and subse-523 quent ATP hydrolysis. One key contact in this regard involves residues in an otherwise dis-524 ordered loop (1178-1203) that form a β-strand which docks onto the end of the central β-525 sheet of the Smc3 head (Fig. 8C iii). More conserved is a major contact mediated by salt 526 bridges between a collection of highly conserved aspartate and glutamate residues located 527 within two loops of Scc2 (819-EDEED-823 and 781-DD-782) and two key lysine residues 528 (K112 K113) in Smc3 (Fig. 8C ii). Entrapment of DNA between Scc2 and engaged heads 529 arises because in addition to the above contacts, Scc2 contacts Smc3 through HEAT re-530 peats 1-4 (residues 151-409), which bind to the start of the joint module (coiled coil arm, 531 residues 999-1004) in a manner thatwhen compared to the unbound crystal structure 532 This movement is necessary to accommodate the simultaneous binding of DNA by both 534 Scc2 and the heads while presumably playing a role in stabilising the unzipped conformation 535 of the coiled coils (Fig. 8C iv). Several residues within this interface (for example, Smc3 show that the DNA must be outside the tripartite S-K ring.

Pseudo-symmetric binding of DNA to the engaged Smc1/3 ATPase heads 540
While Scc2 holds the upper half of the DNA's backbone through a spiral of basic and polar 541 residues (Fig. 8B), the engaged heads of Smc1 and Smc3 produce a 2-fold pseudo-sym-542 metrical ABC ATPase heterodimer that binds DNA through two sites that are exactly two 543 turns of the DNA apart and coincide with the major groove in the DNA (Fig. 8D). As expected 544 for DNA-binding proteins that are not sequence specific, neither protein inserts residues into 545 the major or minor grooves and both rely solely on interactions with the DNA backbone. A 546 consequence of this binding mode is that the two-fold symmetry of DNA is matched almost 547 perfectly by the Smc1/3 heterodimer (Fig. 8D top). The two pseudo-symmetrical DNA bind-548 ing sites close to the major groove are formed through basic and polar amino acids in Smc1 549 (e.g. K63, S112, R113, & K124) and Smc3 (e.g. K57, K112, K113, & K125) (Fig. 8D bottom). 550 The DNA is bent slightly (~9°) and it seems likely that without bending the DNA binding sites 551 on Smc1 and Smc3 would be too close together. Overall, DNA binding is linked to head 552 engagement and the ATPase cycle, as the complete binding path for DNA along the heads 553 only arises when both heads come together in the E-state. 554 555

How do Smc3 K112 and K113 affect loading? 556
In yeast, Smc3 K112 and K113 have important roles in loading of cohesin onto chromo-557 somes. We show here that changing KK to QQ reduces S-K entrapment in the presence of 558 Scc2 and Scc3 in vitro (Fig. S1C), recapitulating the adverse effect on genome wide asso-559 ciation in vivo (Hu et al., 2015). The QQ double mutation is thought to mimic acetylation of 560 K112 K113, which takes place as cells undergo S phase and may have a role in altering 561 how cohesin interacts with DNA, principally whether it can associate (de novo) with and 562 translocate along chromosomes. Our high resolution cryo-EM structure (Figs. 8A and C) 563 reveals that K112 K113 belong to the array of residues that create a basic environment for 564 charge-mediated binding to the DNA backbone (Fig. 8E). This raises the possibility that 565 K112 K113 participate directly in the binding of DNA to engaged heads. However, close 566 inspection of their side chains shows that they in fact face towards the two aspartate and 567 glutamate rich loops in Scc2 (819-EDEED-823 and 781-DD-782) (Fig. 8C ii), implying that 568 they engage in ionic interactions between Smc3 and Scc2 as well as or instead of DNA. If 569 so, one consequence of acetylation or replacement by QQ may be disruption of this mode of Scc2-Smc3 binding, a notion consistent with our previous finding that QQ greatly reduces 571 stimulation of cohesin's ATPase activity by Scc2 in the absence of DNA, at least when Scc3 572 is present (Petela et al., 2018). 573 574 Remarkably, the charge reversal substitution Scc2E822K was isolated as a spontaneous 575 mutation that suppresses the lethality of scc4 mutants whose Scc2's activity is greatly com-576 promised (Petela et al., 2018). Because Scc2E822K would be predicted to reduce binding 577 to Smc3 K112 K113, which might have been expected to further reduce not improve the 578 compromised Scc2 activity in of scc4 mutants, we suggest that E822K might loosen but not 579 eliminate the ionic interactions between Scc2 and K112 K113, permitting the latter to make 580 a greater contribution to DNA binding and thereby increase the affinity between engaged 581 heads and DNA. Given the extreme conservation of residues equivalent to Smc3 K112 K113 582 and Scc2 E822 D823, it seems likely that the interface has a similar function in most eukar-583 yotes and yet mutations equivalent to smc3 K112Q K113Q in S. pombe and in human tissue ATPase is necessary for LE as well as for loading and we therefore suggest that QQ muta-587 tions may turn out to compromise LE. with Pds5 in a similar manner to Scc2 or whether their primary role during release is to bind 594 DNA. Acetylation during S phase blocks release and helps to stabilize Pds5's association 595 with chromosomal cohesin. 596 597

Scc1 is bound to both heads and does not engage in DNA binding 598
Though our map (Fig. 7A) shows little to no density for residues of Scc1 known to bind the 599 central cleft of Scc2 (Scc1 residues ~190-290), it shows very clearly that Scc1's two struc-600 Passage of DNA between disengaged heads prior to ATP-driven head engagement would 615 be essential for entrapment simultaneously in E-S and E-K compartments (see Discussion). 616 There must therefore exist a mechanism by which the heads are moved sufficiently far apart 617 to permit DNA passage. To investigate this, we tested the effect of ATP, Scc2, and DNA on 618 crosslinking between the J-state cysteine pair. Conditions that promote efficient E-state 619 crosslinking, namely addition of ATP, Scc2, and DNA, caused a modest ~20% reduction in 620 J-state crosslinking (Fig. S5F), confirming that the E-state is formed at the expense of J. As 621 expected, other combinations of these three factors had less effect. Surprisingly, addition of 622 Scc2 and DNA in the absence of ATP had the greatest effect, causing a ~50% reduction in 623 J crosslinking, an effect that was highly reproducible. Such a marked reduction is 624 presumably caused by the heads adopting a different conformation. Importantly, this does 625 not correspond to the E-state as very little crosslinking takes place between Smc1N1192C 626 and Smc3R1222C under these conditions (Fig. S5A). We therefore suggest that in the 627 absence of ATP, both Scc2 and DNA reduce J-specific crosslinking by driving or indeed 628 holding the ATPase heads apart, a process that could facilitate passage of DNA between 629 them and thereby facilitate its entrapment in E-S compartments when heads engage in the 630 presence of ATP. 631

In vitro reproduction of DNA entrapment within cohesin SMC-kleisin rings 634
In vivo studies have shown that cohesin entraps circular minichromosomes within its S-K 635 ring (Gligoris et al., 2014;Srinivasan et al., 2018) . We demonstrate here that purified cohesin 636 possesses such an activity also in vitro. Unlike previous assays that have merely measured 637 the physical association between cohesin and DNA and investigated its resistance to salt or 638 sensitivity to kleisin cleavage Uhlmann, 2015, 2014), our method measures 639 topological association directly. By covalently circularising the cohesin ring and its 640 component compartments we can make unambiguous deductions about the topology 641 between DNA and cohesin. The entrapment of DNAs within S-K rings measured by this 642 method depends on Scc2, Scc3, and ATP. Importantly, it is also stimulated by ATP 643 hydrolysis, a feature that has been lacking in previous assays but is of paramount 644 importance for entrapment in vivo (Srinivasan et al., 2018). Four other key properties of the 645 in vitro S-K entrapment activity reflect cohesin's behaviour in vivo, namely it depends on the 646 ability of Scc2 and Scc3 to bind DNA, on the ability of ATP to bind Smc3 heads, and on 647 Smc3's K112 K113 residues, whose lack of acetylation is necessary for loading in yeast (Hu 648 et al., 2015). We therefore suggest that the in vitro DNA S-K entrapment described here 649 involves mechanisms similar or identical to those of cohesin operating within cells. If we 650 assume that the efficiency of BMOE induced S-K circulation is around 20%, we estimate 651 that many DNAs are entrapped by 15 or more cohesin rings in our assay after a 40 min 652 incubation. 653 654

Potential mechanisms for Scc2-driven E-S/E-K entrapment 655
An obvious question concerns the state of the Smc1 and Smc3 ATPase heads when DNA 656 is entrapped. They could either be engaged in the presence of ATP, juxtaposed together in 657 the absence of ATP (a state facilitated by extensive association of the Smc1 and Smc3 coiled 658 coils), or fully disengaged. To address this, we used a Smc1-Smc3 cysteine pair specific for 659 engaged heads (Smc1N1192C Smc3 R1222C), which revealed that DNAs are also 660 entrapped efficiently between the hinge and engaged heads (the E-S compartment), 661 between engaged heads and the kleisin subunit associated with them (the E-K 662 compartment), but only rarely between juxtaposed heads and their associated kleisin (the 663 J-K compartment). For obvious reasons, we were not able to address using cysteine-specific 664 crosslinking whether DNAs are also entrapped within S-K rings with fully disengaged heads.
likely that at least some S-K rings that have entrapped DNA in vitro are in this state. Our 667 failure to observe efficient entrapment within J-K compartments was unexpected given that 668 this state has been documented in vivo (Chapard et al., 2019). 669 We were surprised to find that unlike entrapment within S-K rings, which requires Scc3, 671 entrapment within E-S/E-K compartments was entirely Scc3 independent. In other words, 672 entrapment within E-S/E-K compartments in the presence of Scc2 alone is not accompanied 673 by entrapment within S-K rings. The similarity in kinetics suggests that entrapment within E-674 S and E-K compartments driven solely by Scc2 occurs simultaneously as part of the same 675 reaction. The simplest explanation for this is that DNA is transported (upwards) between 676 disengaged ATPase heads and then subsequently trapped in the E-S compartment due to 677 ATP-driven head engagement (Fig. 9A). The simultaneous entrapment within E-K 678 compartments arises naturally from this, as the kleisin polypeptide must be looped 679 "upwards" to accommodate DNA entry in this manner. Crucially, this process would not 680 require opening of the hinge or either SMC-kleisin interface, processes that would be 681 necessary for entrapment within S-K rings. Our cryo-EM structures reveal that entrapment 682 within E-S/E-K compartments is accompanied and probably driven by the binding of DNA to 683 Scc2 and DNA binding sites on the upper surface of Smc1 and Smc3 ATPase heads created 684 upon head engagement. In other words, entrapment within E-S/E-K compartments (in the 685 absence of S-K entrapment) arises from the clamping of DNA between Scc2 and engaged 686 heads. The remarkable similarity between this structure and one formed between DNA, 687 Scc2 NIPBL , and tetrameric human cohesin (PDB 6WG3) (Shi et al., 2020) shows that the 688 clamping of DNA between Scc2 NIPBL and engaged heads not only does not involve Scc3 SA2 689 (whose ortholog SA2 was present in the human structure) but more importantly the same 690 clamping happens even when Scc3 is absent (Fig. S6). 691 692

An alternative is that simultaneous E-S/E-K entrapment arises from insertion of a loop of 693
DNA into an open S-K ring. If one segment of this loop were located above the heads while 694 the other located below them, head engagement would also lead to simultaneous 695 entrapment in E-S and E-K compartments (Fig. 9B). This scenario, which could also involve 696 the clamping of DNA between Scc2 and engaged heads, is not only more complex, but it 697 must somehow explain how DNA is bent prior to insertion, a process which would carry a 698 clear entropic penalty and more importantly is not apparent from cryo-EM images of EQEQ 699 cohesin associated with circular DNAs (Fig. 7B right). Importantly, the states created by the DNAs trapped in E-S/E-K compartments could in principle belong to separate populations 708 (Fig. 9D). According to this scenario, and because E-S/E-K entrapment is not accompanied 709 by S-K entrapment, DNAs entrapped solely within E-S compartments would have to be held 710 by complexes whose kleisin subunit had dissociated from one or both ATPase heads while 711 DNAs entrapped solely within E-K compartments would have to be trapped by complexes 712 whose hinge had opened. There are two arguments against this interpretation. First, it is 713 very unclear why head engagement in the presence of ATP, Scc2, and DNA should be 714 associated with two such different events. An even more compelling argument stems from 715 our cryo-EM structures of DNA clamped by Scc2 and engaged heads. A low resolution 716 structure reveals coiled coils folded around their elbow and a dimerised hinge associated 717 with Smc3's coiled coil (Fig. 7E), while a high resolution structure shows that both N-and C-718 terminal kleisin domains (Scc1-N and -C) are bound to Smc3's neck and the base of Smc1's 719 ATPase respectively (Fig. 8A). In other words, association of DNA with engaged heads in 720 the presence of Scc2 does not appear to be accompanied by opening of any of the S-K 721 ring's three interfaces, ruling out the possibility that E-S and E-K entrapment are 722 independent processes. 723

724
The notion that DNA can be engaged by cohesin rings in a manner that does not require 725 opening of the hinge or either SMC-kleisin interface (Fig. 9A & B)  that circular DNAs are entrapped within E-S/E-K but not S-K compartments under identical neck, must pass over the DNA bound to the engaged heads before its Scc1-C domain binds 736 to the base of Smc1's ATPase (Fig. 9A and molecular equivalent Fig. 8F). This topology is 737 not merely of academic interest as it provides crucial insight into the pathway by which DNAs 738 are clamped by Scc2 and engaged ATPase heads. Because EQEQ cohesin's association 739 with Scc2 and circular DNA does not appear to cause much DNA bending (Fig. 7B right), 740 we favour the notion that E-S/E-K entrapment arises when DNAs pass (in an upwards 741 direction) between heads prior to their engagement in the presence of ATP and not by 742 insertion of a loop into the S-K ring. 743

744
Given that the ATPase heads are frequently associated either in the E-or J-state, there must 745 exist a mechanism to create an opening between them, if only transiently, in order for DNA 746 to pass through before being clamped by their subsequent engagement. Our observation 747 that Scc2 and DNA disrupts the J-state, albeit only in the absence of ATP, may be relevant 748 in this regard (Fig. S5F). This J-state disruption was not caused by adoption of the E-state 749 (Fig. S5A), which requires ATP, and it must therefore involve transition to a state in which 750 Smc1 and Smc3 ATPase heads adopt yet another conformation. This could be a state in 751 which Scc2 and DNA together drive apart the ATPase head domains, thereby enabling DNA 752 to pass between them. Subsequent ATP binding would then cause head engagement and 753 E-state formation, trapping DNA inside both the E-S and E-K compartments. Interestingly, a 754 conformation of this nature has recently been observed in condensin bound to Ysc4 (Lee et 755 al., 2020), where the latter bridges the Smc2 and Smc4 heads, holding them apart by some 756 distance. Although DNA was absent from this structure, the separation of the heads would 757 be sufficient for DNA to pass between them and exposes the HAWK's DNA-binding surface 758 without any impediment. If we assume that Scc2 bridges Smc1 and Smc3 heads in a similar 759 fashion prior to their engagement, then one merely has to propose that DNA initially binds 760 to Scc2 while in the bridged state and remains associated as the Smc3 head pivots around 761 and the Smc3 ATPase head engages with that of Smc1 (Movie 1). We envisage that Scc2's 762 association with Smc1 heads (which strongly resembles that between Ycs4 and Smc4, Fig.  763 S6) remains unaltered during this transition, as it did between Smc4 and Ycs4 in condensin 764 (Lee et al., 2020). 765 Our demonstration that DNAs transported into the sub-compartment created by Scc2's 767 association with engaged ATPase heads results in entrapment in E-S/E-K but not S-K compartments is difficult to reconcile with the proposal that DNAs must first pass through a 769 transiently opened Scc1-Smc3 interface before they enter the clamped state created by 770 head engagement (Higashi et al., 2020) , a process that has been termed DNA "gripping". 771 Passage through a gate created by opening the Scc1-Smc3 interface before being clamped 772 by engaged heads and Scc2 would be accompanied by E-K and S-K entrapment but not by 773 E-S entrapment, which is contrary to what we observe. 774 775 It is also worth pointing out that E-K entrapment would not be possible if Scc1's NTD were 776 dissociated from Smc3's neck upon head engagement, as has been suggested by a cryo-777 EM structure of heads engaged in the absence of both DNA and Scc2 (Muir et al., 2020). 778 The fact that E-K entrapment accompanies E-S entrapment during our Scc2-only reaction we suggest that head engagement may indeed promote Scc1's dissociation from Smc3 but 786 that this process is actively inhibited by Scc2. One of the functions of Pds5 and Wapl in 787 mediating release during G1 when Smc3 is not acetylated may be to replace Scc2 and 788 thereby abrogate this protection mechanism. In this regard, it is interesting that Scc2 789 contacts the joint region within Smc3's coiled coil adjacent to where Scc1's NTD binds to 790 Smc3's neck (Fig. 8C iv), an interaction also observed in the human structure (Shi et  Another idea is that cohesin's hinge provides the second site and that the clamp/release 820 transport cycle is accompanied by changes in the folding of Smc1/3 coiled coils around their 821 elbow region, which could be the key to walking along the DNA. However, this notion is 822 difficult to reconcile with the observation that cohesin's coiled coils can be folded whether its 823 ATPase heads are engaged (Fig. 7E) or disengaged (Bürmann et al., 2019). 824 825

Scc3 catalyses entry of DNA inside the SMC-kleisin ring 826
Our finding that Scc3 is essential for S-K but not E-S/E-K entrapment reveals that Scc3 has 827 a unique role in promoting entry of DNA inside the SMC-kleisin ring as well as being 828 necessary for loop extrusion (Davidson et al., 2019). In principle, Scc3 could catalyse DNA 829 entry either via a gate created by transient hinge opening or through one produced by 830 transient dissociation of one or both SMC-kleisin interfaces (Fig. 9E). Crucially, S-K 831 entrapment in vivo is not abolished by fusing Scc1's NTD to Smc3 or by fusing its CTD to 832 Smc1, implying that DNA must enter either through the hinge or through dissociation of 833 either one of the two SMC-kleisin interfaces (possibly through simultaneous dissociation of 834 both) (Srinivasan et al., 2018). There is little or no direct evidence regarding which 835 mechanism is correct. SMC-kleisin dissociation has been strongly implicated in release and is therefore also a plausible mechanism for entry (Beckouët et al., 2016). Nevertheless, 837 hinge opening is equally plausible, especially in the light of recent findings that folding of 838 cohesin's coiled coils around an elbow brings its hinge domain into close proximity to DNA 839 bound to the heads, and that Scc3 SA2 interacts with a half opened hinge when DNA is bound 840 to human cohesin-Scc2 NIPBL complexes (Bürmann et al., 2019;Shi et al., 2020) . Ascertaining 841 which mechanism is at play will require a method to measure the effect on S-K entrapment 842 of chemically linking interfaces together in a manner that is orthogonal to the BMOE-induced 843 crosslinking. For example, prior crosslinking of both SMC-kleisin interfaces would abolish 844 entrapment via a kleisin gate (Fig. 9E bottom pathway) but not via a hinge gate (Fig. 9E top  845 pathway). CENs as well as along chromosomes arms, which implies that Scc3 has additional functions 863 that do not involve or require its ability to bind DNA. 864

865
The notion that entrapment of DNA within S-K rings is preceded by its prior entrapment 866 within E-K/E-S compartments to Scc2 and SMC heads in a manner observed in our cryo-867 EM structure raises the interesting possibility that DNA is eventually entrapped within the S-868 K ring, not by passing from outside to inside, but instead by being allowed to exit from either 869 the E-S or the E-K compartment by transiently opening one of the S-K ring's three interfaces. pathway) while transient dissociation of one or another, or indeed both, kleisin-head 872 interfaces would permit escape from the E-K compartment (Fig. 9E bottom pathway). In both 873 cases, the subsequent closing of these exit gates would lead to entrapment of DNA within 874 the S-K ring. According to these scenarios, exit via the hinge or via a SMC-kleisin interface 875 without head disengagement would lead, at least initially, to the selective loss of E-S and E-876 K entrapment respectively. If true, clamping of the DNA would provide the opportunity to 877 open gates without losing grip of the DNA while doing so. It is interesting in this regard that 878 whereas E-K entrapment does not increase between 2 and 40 min when both Scc2 and 879 Scc3 are present, E-S entrapment continues to increase in parallel with the rise in S-K 880 entrapment. Whether this asymmetry is a hint that Scc3 promotes entrapment within S-K 881 rings by opening an SMC-kleisin interface will require far more rigorous types of 882 experiments. Though EQEQ mutants reduce S-K entrapment, they do not eliminate it, 883 suggesting that DNA entry can in principle occur without head disengagement, as depicted 884 in Fig. 9E. 885

886
The notion that a key function of Scc3, dependent on its ability to bind DNA, is to facilitate 887 entrapment of DNA within S-K rings has an important corollary. S-K entrapment is thought 888 to be a crucial feature of sister chromatid cohesion. Hitherto, direct evidence for this 889 mechanism has been confined to the observation of small circular minichromosomes 890 entrapped within S-K rings in vivo. We show here that a function of Scc3, not shared by 891 Scc2, is to facilitate entrapment within S-K rings. If this is also an essential function of Scc3 892 in vivo, it follows that S-K entrapment must also be an essential cohesin function and one 893 that applies to proper chromosomes as well as small circular ones. to detect. Future work will be required to address whether E-S/E-K entrapment also occurs 909 inside cells, to elucidate the mechanism of S-K entrapment, and to reveal conditions that 910 promote J-K entrapment, a form that has been detected in vivo but not yet efficiently in vitro. 911

912
We thank all members of the Nasmyth and Löwe groups for valuable discussions and many 914 unseen contributions over the course of this work. We would like to thank Nicolas Jean 915

DECLARATION OF INTERESTS 931
The authors declare no competing interests. 932

Recombinant yeast cohesin complex cloning 935
The S. cerevisiae genes SMC1, SMC3, SCC3, SCC2, SCC1, and SCC4 were codon 936 optimised for expression in Spodoptera frugiperda cells and synthesised using the 937 Genescript Thermo Fisher service. These were then cloned into MultiBac vectors. Tag

Virus generation and protein expression 950
DNAs were first transformed into DH10Bac (Thermo Fisher) cells and bacmids containing 951 the expression vector screened for by blue-white selection. DNA was then extracted and 2 952 µg of bacmid DNA was transfected into 2 ml S. frugiperda Sf9 cells (Thermo Fisher) at a cell 953 density of 1x10 6 cells ml -1 using FuGENE HD reagent (Promega), grown in Sf900 II SFM 954 media (Thermo Fisher). These were then incubated at 27°C for 5 days to create P1 virus. 955 P2 virus was then amplified by infecting 50 ml Sf9 cells at a density of 2x10 6 cells ml -1 with 956 500µl P1 virus and incubating in the dark at 27°C for 3 days with shaking at 100 rpm. P2 957 virus was then harvested by pelleting cells by centrifugation at 4000g and decanting into 5% 958 FBS (Sigma), and then stored in the dark at 4°C. Typically, proteins were then expressed by 959 adding 5 ml P2 virus to 500 ml Sf900 cells at a density of 2x10 6 cells ml -1 and incubating in 960 the dark at 27 o C for 2 days with shaking at 100 rpm. Cells were then harvested by 961 centrifugation at 1000g, washed with PBS, and then frozen in liquid nitrogen and stored at -962 Cells were thawed in Buffer A (50 mM HEPES pH 7.5, 150 mM NaCl, 1 mM TCEP (Thermo 966 Fisher), 5% glycerol) supplemented with 1 Complete Protease Inhibitor (EDTA-free) tablet (Roche), 70 µg RNAse A (Roche), and 100 U ml -1 Supernuclease (Sino Biological) and then 968 lysed by sonication. Following sonication, cell lysate was supplemented with 1 mM PMSF 969 (Sigma). Proteins were then purified via a three strep purification protocol. First, proteins 970 were purified via affinity pulldown of their StrepII tags using a StrepTrap HP column (Fisher 971 Scientific) and eluted into Buffer A supplemented with 2.5mM desthiobiotin (Fisher 972 Scientific). Scc2 constructs were eluted into 50 mM Tris pH 8.0 rather than 50 mM HEPES 973 pH 7.5. Proteins were then further purified by anion exchange chromatography using a 5 ml 974 HiTrap Q HP column (GE Healthcare) across a gradient of 100 mM to 1M NaCl. Scc2 975 constructs were eluted across a gradient of 0 mM to 1 M NaCl. Finally, proteins were purified 976 via size exclusion chromatography using a Superose 6 increase 10/300 GL column (VWR) The DNA was then extracted, and the EtBr removed by washing several times with butanol 996 saturated with 50 mM HEPES pH 7.5 and then the butanol phase discarded. The CsCl2 was 997 then removed by dialysis against 2 L 50 mM HEPES pH 7.5 buffer over 24 hours at 4 o C, 998 with two buffer changes. The DNA was then collected and stored at -20 o C. 999 (buffered with 50 mM HEPES pH 7.5, 50 mM NaCl, 5 mM MgCl2, 1 mM TCEP, 5% glycerol).