Structural insights into Cullin4-RING ubiquitin ligase remodelling by Vpr from simian immunodeficiency viruses

Viruses have evolved means to manipulate the host’s ubiquitin-proteasome system, in order to down-regulate antiviral host factors. The Vpx/Vpr family of lentiviral accessory proteins usurp the substrate receptor DCAF1 of host Cullin4-RING ligases (CRL4), a family of modular ubiquitin ligases involved in DNA replication, DNA repair and cell cycle regulation. CRL4DCAF1 specificity modulation by Vpx and Vpr from certain simian immunodeficiency viruses (SIV) leads to recruitment, poly-ubiquitylation and subsequent proteasomal degradation of the host restriction factor SAMHD1, resulting in enhanced virus replication in differentiated cells. To unravel the mechanism of SIV Vpr-induced SAMHD1 ubiquitylation, we conducted integrative biochemical and structural analyses of the Vpr protein from SIVs infecting Cercopithecus cephus (SIVmus). X-ray crystallography reveals commonalities between SIVmus Vpr and other members of the Vpx/Vpr family with regard to DCAF1 interaction, while cryo-electron microscopy and cross-linking mass spectrometry highlight a divergent molecular mechanism of SAMHD1 recruitment. In addition, these studies demonstrate how SIVmus Vpr exploits the dynamic architecture of the multi-subunit CRL4DCAF1 assembly to optimise SAMHD1 ubiquitylation. Together, the present work provides detailed molecular insight into variability and species-specificity of the evolutionary arms race between host SAMHD1 restriction and lentiviral counteraction through Vpx/Vpr proteins.


Introduction
the ubiquitin transfer mechanism. 116 117

SAMHD1-CtD is necessary and sufficient for Vpr mus -binding and ubiquitylation in vitro 119
To investigate the molecular interactions between Vpr mus , the neo-substrate SAMHD1 from rhesus 120 macaque and CRL4 subunits DDB1/DCAF1 C-terminal domain (DCAF1-CtD), protein complexes were 121 reconstituted in vitro from purified components and analysed by gel filtration (GF) chromatography. 122 The different protein constructs that were employed are shown schematically in S1A Fig. In the absence 123 of additional binding partners, Vpr mus is insoluble after removal of the GST affinity purification tag (S1B 124 DDB1/DCAF1B and SAMHD1 followed by GF resulted in co-elution of all three components ( Fig 1A). 129 Together, these results show that Vpr mus forms stable binary and ternary protein complexes with 130 DDB1/DCAF1-CtD and/or SAMHD1 in vitro. Furthermore, incubation with any of these interaction 131 7 system, constructs containing SAMHD1-CtD fused to T4 lysozyme (T4L-SAMHD1-CtD), or 135 containing only the N-terminal domains of SAMHD1, and lacking SAMHD1-CtD (SAMHD1-ΔCtD), 136 were incubated with Vpr mus and DDB1/DCAF1-CtD, and complex formation was assessed by GF 137 chromatography. Analysis of the resulting chromatograms by SDS-PAGE shows that SAMHD1-ΔCtD 138 did not co-elute with DDB1/DCAF1-CtD/Vpr mus (Fig 1B). By contrast, T4L-SAMHD1-CtD 139 accumulated in the same elution peak as DDB1/DCAF1-CtD and Vpr mus (Fig 1C). These results confirm 140 that SAMHD1-CtD is necessary for stable association with DDB1/DCAF1-CtD/Vpr mus in vitro, and 141 demonstrate that SAMHD1-CtD is sufficient for Vpr mus -mediated recruitment of the T4L-SAMHD1-142 CtD fusion construct to DDB1/DCAF1-CtD. 143 To correlate these data with enzymatic activity, in vitro ubiquitylation assays were conducted by 144 incubating SAMHD1, SAMHD1-ΔCtD or T4L-SAMHD1-CtD with purified CRL4 DCAF1-CtD , E1 145 after 15 min incubation (Figs 1E and S2E). In agreement with the analytical GF data, SAMHD1-ΔCtD 151 was not ubiquitylated in the presence of Vpr mus (Figs 1F and S2F). By contrast, T4L-SAMHD1-CtD was 152 efficiently ubiquitylated, resulting in >90% loss of the band corresponding to unmodified T4L-153 SAMHD1-CtD after 15 min (Figs 1G and S2F). Again, these data substantiate the functional importance 154 of SAMHD1-CtD for Vpr mus -mediated recruitment to the CRL4 DCAF1 ubiquitin ligase. To obtain structural information regarding Vpr mus and its mode of binding to the CRL4 substrate receptor 159 DCAF1, the X-ray crystal structures of a DDB1/DCAF1-CtD complex, and DDB1/DCAF1-CtD/T4L-160 Vpr mus (residues 1-92) fusion protein ternary complex were determined. The structures were solved 161 using molecular replacement and refined to resolutions of 3.1 Å and 2.5 Å respectively (S1 Table). 162 8 Vpr mus adopts a three-helix bundle fold, stabilised by coordination of a zinc ion by His and Cys residues 163 on Helix-1 and at the C-terminus (Fig 2A). Superposition of Vpr mus with previously determined Vpx sm 164 [50], Vpx mnd2 [51,52], and Vpr HIV-1 [54] structures reveals a conserved three-helix bundle fold, and 165 similar position of the helix bundles on DCAF1-CtD (S3A Fig). In addition, the majority of side chains 166 involved in DCAF1-interaction are type-conserved in all Vpx and Vpr proteins (Figs S3B-G and S6A), 167 strongly suggesting a common molecular mechanism of host CRL4-DCAF1 hijacking by the Vpx/Vpr 168 family of accessory proteins. However, there are also significant differences in helix length and register 169 as well as conformational variation in the loop region N-terminal of Helix-1, at the start of  in the loop between Helices-2 and -3 (S3A Fig).  residues P1329, F1330, F1355, N1371, L1378, M1380 and T1382 with Vpr mus side chains of T79, R83, 183 R86 and E87 in Helix-3 result in the stabilisation of the sequence stretch that connect BP blades 6 and 184 7 ("C-terminal loop", Figs 2B and S3F). Moreover, side chain electrostatic interactions of Vpr mus 185 residues R15, R75 and R76 with DCAF1 E1088, E1091 and E1093 lock the conformation of an "acidic 186 loop" upstream of BP blade 1, which is also unstructured and flexible in the absence of Vpr mus (Figs 2B,187 C and S3D,F). 188 Notably, in previously determined structures of Vpx/DCAF1/SAMHD1 complexes the "acidic loop" is 189 a central point of ternary contact, providing a binding platform for positively charged amino acid side 9 chains in either the SAMHD1 N-or C-terminus [50][51][52]. For example, Vpx sm positions  in such a way, that SAMHD1 K622 engages in electrostatic interaction with the DCAF1 "acidic loop" 192 residue D1092 (Fig 2C, left panel). However, in the Vpr mus crystal structure the bound Vpr mus now blocks 193 access to the corresponding SAMHD1-CtD binding pocket, in particular by the positioning of an 194 extended N-terminal loop that precedes Helix-1. Additionally, Vpr mus side chains R15, R75 and R76 195 neutralise the DCAF1 "acidic loop", precluding the formation of further salt bridges to basic residues in 196 SAMHD1-CtD ( Fig 2C, right panel). 197 To validate the importance of Vpr mus residues R15 and R75 for DCAF1-CtD-binding, charge reversal 198 mutations to glutamates were generated by site-directed mutagenesis. The circular dichroism (CD) 199 spectrum of the Vpr mus R15E R75E double mutant GST-fusion protein was identical to the wild type, 200 indicating similar secondary structure content and thus no major structural disturbances caused by the 201 amino acid substitutions (S3H Fig). The effect of the Vpr mus R15E R75E double mutant on complex 202 assembly was then analysed by GF chromatography. SDS-PAGE analysis of the resulting 203 chromatographic profile shows an almost complete loss of the DDB1/DCAF1-CtD/Vpr mus /SAMHD1 204 complex peak (Fig 2D, fraction 6), when compared to the wild type, concomitant with enrichment of (i) 205 some proportion of Vpr mus R15E R75E-bound DDB1/DCAF1-CtD ( Fig 2D, fraction 7), (ii) free 206 DDB1/DCAF1-CtD (fraction 7-8), and of (iii) Vpr mus R15E R75E/SAMHD1 binary complex (Fig 2D,. This suggests that charge reversal of Vpr mus side chains R15 and R75 weakens the strong 208 association with DCAF1 observed in wild type Vpr mus , due to loss of electrostatic interaction with the 209 "acidic loop", in accordance with the crystal structure. Consequently, some proportion of Vpr-bound 210 SAMHD1 dissociates. This notion is further supported by GF analysis of binary combinations of the 211 Vpr mus R15E R75E double mutant with either SAMHD1 or DDB1/DCAF1-CtD. Incubation of Vpr mus 212 R15E R75E with SAMHD1 followed by GF leads to co-elution of both proteins, concomitant with a 213 shift of the elution peak towards higher apparent molecular weight, compared to SAMHD1 alone ( To obtain structural insight into Vpr mus in the context of a complete CRL4 assembly, and to understand 223 the SAMHD1 recruitment mechanism, we initiated cryo-EM analyses of the 224 CRL4 DCAF1-CtD /Vpr mus /SAMHD1 assembly. In these studies, the small ubiquitin-like protein NEDD8 225 was enzymatically attached to the CUL4 subunit, in order to obtain its active form (  Two consecutive rounds of 3D classification yielded three particle populations, resulting in 3D 231 reconstructions at 8-10 Å resolution, which contained both the Vpr mus -bound CRL4 core and the stalk 232 Alignment of 3D volumes from states-1, -2 and -3 shows that core densities representing DDB1 BPA, 242 BPC, DCAF1-CtD and Vpr mus superimpose well, indicating that these components do not undergo major 243 conformational fluctuations and thus form a rigid platform for substrate binding and attachment of the 244 CRL4 stalk (Fig 3). However, rotation of DDB1 BPB around a hinge connecting it to BPC results in 245 three different orientations of state-1, -2 and -3 stalk regions relative to the core. BPB rotation angles 246 were measured as 69° between state-1 and -2, and 50° between state-2 and -3. 247 These data are in line with previous prediction based on extensive comparative crystal structure 248 analyses, which postulated an approx. 150° rotation of the CRL4 stalk around the core [13, 15, 16, 19, 249 57]. However, the left-and rightmost CUL4 orientations observed here, states-1 and -3 from our cryo-250 EM analysis, indicate a slightly narrower stalk rotation range (119°), when compared to the outermost 251 stalk conformations modelled from previously determined crystal structures (143°) (Fig 3C). A possible 252 explanation for this discrepancy arises from inspection of the cryo-EM densities and fitted models,  BP domains A and C (BPA, BPC), DCAF1-CtD and Vpr mus , derived from our crystal structure (Fig 2), 267 could be fitted as rigid bodies into this cryo-EM volume ( Fig 4A). No obvious electron density was 268 visible for the bulk of SAMHD1. However, close inspection revealed an additional tubular, slightly 269 arcing density feature, approx. 35 Å in length, located on the upper surface of the Vpr mus helix bundle, 270 approximately 17 Å away from and opposite of the Vpr mus /DCAF1-CtD binding interface (Fig 4A, red  271 arrows). One end of the tubular volume contacts the middle of Vpr mus Helix-1, and the other end forms 272 additional contacts to the C-terminus of Helix-2 and the N-terminus of Helix-3 ( Fig 4B). A local 12 resolution of 7.5-8 Å (S5C Fig) precluded the fitting of an atomic model. Considering the biochemical 274 data, showing that SAMHD1-CtD is sufficient for recruitment to DDB1/DCAF1/Vpr mus , we hypothesise 275 that this observed electron density feature corresponds to a region of SAMHD1-CtD which physically 276 interacts with Vpr mus . Given its dimensions, the putative SAMHD1-CtD density could accommodate 277 approx. 10 amino acid residues in a fully extended conformation or up to 23 residues in a kinked helical 278 arrangement. All previous crystal structure analyses [46], as well as secondary structure predictions 279 indicate that SAMHD1 residues C-terminal to the catalytic HD domain and C-terminal lobe (amino 280 acids 599-626) are disordered in the absence of additional binding partners. Accordingly, the N-terminal 281 globular domains of the SAMHD1 molecule might be flexibly linked to the C-terminal tether identified 282 here. In that case, the bulk of SAMHD1 samples a multitude of positions relative to the DDB1/DCAF1-283 CtD/Vpr mus core, and consequently is averaged out in the process of cryo-EM reconstruction. 284 To test this hypothesis, Vpr mus amino acid residues in close proximity to the putative SAMHD1-CtD 285 density were substituted by site-directed mutagenesis. Specifically, Vpr mus W29 was changed to alanine 286 to block a hydrophobic contact with SAMHD1-CtD involving the aromatic side chain, and Vpr mus A66 287 was changed to a bulky tryptophan, in order to introduce a steric clash with SAMHD1-CtD ( Fig 4B). 288 The structural integrity of the Vpr mus W29A A66W double mutant was confirmed by CD spectroscopy 289 mutant with DDB1/DCAF1-CtD or SAMHD1 were also analysed by GF. These data show loss of 296 SAMHD1 interaction (Fig 4D), while the ability to bind DDB1/DCAF1-CtD is retained ( Fig 4E). 297 Together, these biochemical analyses support a location of the SAMHD1-CtD binding site on the upper 298 surface of the Vpr mus helix bundle, as suggested by medium-resolution cryo-EM reconstruction. 299 To obtain additional experimental evidence, the CRL4 DCAF1-CtD /Vpr mus /SAMHD1 assembly was further 300 examined by cross-linking mass spectrometry (CLMS), using the photo-reactive cross-linker sulfo-301 primary amines and hydroxyl groups, while the other end covalently links to any amino acid sidechain 303 within reach upon UV-activation via a carbene intermediate [58]. Accordingly, incubation of proteins 304 or protein complexes with sulfo-SDA, followed by UV-illumination, allows for high-density cross-305 linking of lysine, and to a lesser extent serine, threonine and tyrosine side chains to amino acids within 306 reach of the SDA spacer group, with faster kinetics than pure NHS ester-based cross-linkers, due to the 307 CtD, as observed in cryo-EM (Fig 3). 315 An additional 300 cross-links involved SAMHD1, extending to the C-terminal half of CUL4, to a DDB1 316 sequence stretch comprising amino acid residues 900-1000, to parts of DCAF1-CtD and to Vpr mus (Figs 317 5A, S5E). The CRL4 DCAF1-CtD /Vpr mus residues exhibiting cross-links to SAMHD1 were mapped onto the 318 state-2 model, and showed the presence of a large, yet defined, interaction surface ( Fig 5B). Importantly, 319 cross-links were apparent between the C-terminus of SAMHD1-CtD (residues K622, K626) and a region 320 in Vpr mus Helix-1 (residues 27-36), which forms a part of the putative SAMHD1-CtD binding interface 321 observed in cryo-EM, and which contains Vpr mus W29, one of the residues substituted in the mutagenesis 322 and biochemical analysis presented above ( Fig 5B, purple spheres). In addition, amino acid residues 323 from the N-terminal portion of SAMHD1-CtD (residues K595, K596, T602-S606) cross-linked close to 324 the DCAF1-CtD "acidic loop" (residues 1092-1096), which is immobilised by Vpr near the proposed 325 SAMHD1-CtD binding site, and to the very C-terminus of CUL4 (residues Y744, A759), which is also 326 adjacent to the predicted SAMHD1-CtD binding position ( CtD on the upper surface of the Vpr helix bundle, as indicated by cryo-EM. In addition, the spatial 331 distribution of cross-links involving the SAMHD1 N-terminal domains suggest that these are flexibly 332 connected to SAMHD1-CtD, leading to highly variable positioning relative to CRL4 DCAF1-CtD /Vpr mus and 333 thus offering a multitude of cross-linking opportunities to nearby CRL4 components, again in line with 334 the cryo-EM reconstruction results, especially upon consideration of the positional heterogeneity of the 335 CUL4 stalk (Fig. 3). 336 In order to evaluate the distance information inherent in SAMHD1-CtD cross-links in a more 337 quantitative way, the volume accessible to SAMHD1-CtD for interaction with CRL4 DCAF1-CtD /Vpr mus , 338 consistent with the CLMS distance restraints, was simulated using the DisVis software tool [60,61]. 339 For this analysis, SAMHD1-CtD was modelled as peptide in extended conformation. During the 340 simulation, the state-2 CRL4 DCAF1-CtD /Vpr mus molecular model was kept fixed, and a six-dimensional 341 search of all possible degrees of freedom of rotation and translation for the SAMHD1-CtD model in 342 molecular contact with CRL4 DCAF1-CtD /Vpr mus was computed and ranked according to agreement with 343 CLMS distance restraints. To visualise the output, all possible spatial positions of the centre of mass of 344 SAMHD1-CtD, which satisfy >50% of the CLMS restraints, were plotted as density map on the structure 345 of DCAF1-CtD/Vpr mus ( Fig 5C). In accordance with the cryo-EM reconstruction, this independent 346 computational analysis also locates SAMHD1-CtD on top of the Vpr mus helix bundle. 347 Taken together, the structural, biochemical and CLMS data are consistent with a model where the very 348 C-terminus of SAMHD1 is recruited by Vpr mus , to place the remaining SAMHD1 domains appropriately 349 for access to the catalytic machinery at the distal end of the CRL4 stalk. Like SIV Vpx, "hybrid" Vpr proteins down-regulate the host restriction factor SAMHD1 by recruiting 368 it to CRL4 DCAF1 for ubiquitylation and subsequent proteasomal degradation. However, using a 369 combination of X-ray, cryo-EM and CLMS analyses, we show that the molecular strategy, which Vpr mus 370 evolved to target SAMHD1, is strikingly different from Vpx-containing SIV strains. In the two clades 371 of Vpx proteins, divergent amino acid sequence stretches just upstream of Helix-1 (variable region 372 such a position precludes stabilising ternary interaction with DCAF1-CtD, but still results in robust 383 SAMHD1 ubiquitylation in vitro and SAMHD1 degradation in cell-based assays [24]. 384 The catalytic dNTPase activity of SAMHD1 depends on nucleotide-dependent oligomerisation, 385 mediated by two allosteric nucleotide-binding sites, where guanine-based nucleotides in the first site 386 induce dimer formation, and dNTP-binding to the second site leads to assembly of the catalytically 387 active tetramer. SAMHD1-CtD is essential for tetramer formation by contributing critical molecular 388 contacts to neighbouring protomers. Furthermore, tetramer destabilisation by CDK1/2-cyclinA-389 dependent phosphorylation of T592 in SAMHD1-CtD endogenously attenuates SAMHD1 activity in 390 cycling cells [46,[63][64][65]. Accordingly, it is conceivable that by sequestering SAMHD1-CtD, Vpr mus 391 destabilises the SAMHD1 tetramer, and in this way abrogates SAMHD1 activity, prior to inducing its 392 proteasomal degradation, in accordance with previous observation of Vpx HIV-2 -mediated SAMHD1 393 tetramer disassembly and inhibition of dNTPase activity [62]. 394 Predictions regarding the molecular mechanism of SAMHD1-binding by other "hybrid" Vpr 395 orthologues are difficult due to sequence divergence. Even in Vpr deb , the closest relative to Vpr mus , only 396 approximately 50% of amino acid side chains lining the putative SAMHD1-CtD binding pocket are 397 Vpr agm.GRI sub-type involves molecular recognition of both SAMHD1-NtD and -CtD [49,53]. In 407 conclusion, recurring rounds of evolutionary lentiviral adaptation to the host SAMHD1 restriction 408 factor, followed by host re-adaptation, resulted in highly species-specific, diverse molecular modes of 409 Vpr-SAMHD1 interaction. Similar molecular arms races between cell-intrinsic antiviral host factors and 410 viral antagonists shaped the species-specific lentiviral antagonism of e.g. host restriction factors of the 411 APOBEC3 family and tetherin, through induction of their degradation by the respective viral antagonists 412 Vif or Nef/Vpu [67][68][69]. Furthermore, viral re-adaptation to certain simian and human variants of these 413 restriction factors, following cross-species transmission, took part in the emergence of pandemic HIV 414 strains, thus highlighting the importance of structural insight into these processes [9]. In addition to the 415 instance presented here, further structural characterisation of SAMHD1-Vpr complexes will be 416 necessary to fully define outcomes of this particular virus-host molecular arms race. 417 Previous structural investigation of DDB1/DCAF1/Vpr HIV-1 in complex with the neo-substrate UNG2 418 demonstrated that Vpr HIV-1 engages UNG2 by mimicking the DNA phosphate backbone. More precisely, 419 UNG2 residues, which project into the major groove of its endogenous DNA substrate, insert into a 420 hydrophobic cleft formed by Vpr HIV-1 Helices-1, -2 and the N-terminal half of Helix-3 [54]. This 421 mechanism might rationalise Vpr HIV-1 's extraordinary binding promiscuity, since the list of potential 422 Vpr HIV-1 degradation substrates is significantly enriched in DNA-and RNA-binding proteins [27]. 423 Moreover, promiscuous Vpr HIV-1 -induced degradation of host factors with DNA-or RNA-binding 424 activity has been proposed to induce cell cycle arrest at the G2/M phase border, which is the most 425 thoroughly described phenotype of Vpr proteins so far [26,27,70]. In Vpr mus , the N-terminal half of 426 Helix-1 as well as the bulky amino acid residue W48, which is also conserved in Vpr agm and Vpx, 427 characterise these determinants will further extend our understanding of how the Vpx/Vpr helical 435 scaffold binds, and in this way adapts to a multitude of neo-substrate epitopes. 436 Our cryo-EM reconstructions of CRL4 DCAF1-CtD /Vpr mus /SAMHD1, complemented by CLMS, also 437 provide insights into the structural dynamics of CRL4 assemblies prior to ubiquitin transfer. The data 438 confirm previously described rotational movement of the CRL4 stalk, in the absence of constraints 439 imposed by a crystal lattice, creating a ubiquitylation zone around the Vpr mus -modified substrate receptor 440 (Figs 3 and 7A) [13,15,16,19,57]. Missing density for the neddylated CUL4 WHB domain and for 441 the catalytic ROC1 RING domain indicates that these distal stalk elements are highly mobile and likely 442 sample a multitude of orientations relative to the CUL4 scaffold ( Fig 7B). Vpx/Vpr-family accessory proteins. By tethering either SAMHD1-CtD or -NtD to DCAF1, and in this 460 way flexibly recruiting the bulk of SAMHD1, the accessibility of lysine side chains both tether-proximal 461 and on the SAMHD1 globular domains to the CRL4 catalytic assembly might be further improved (Fig  462   7C, D). This ensures efficient Vpx/Vpr-mediated SAMHD1 priming, poly-ubiquitylation and 463 proteasomal degradation to stimulate virus replication. In addition, structural insight into this 464 evolutionary optimised, highly specific protein degradation machinery might inform the positioning of 465 novel CRL4 DCAF1 -based synthetic modalities for targeted protein degradation, e.g. in the form of 466 proteolysis-targeting chimera-(PROTAC-) or molecular glue-type compounds [74,75]. 467 Lastly, while current highly active antiretroviral therapy (HAART) regimens are able to control HIV-1 468 replication in infected patients [76], they cannot eradicate the virus due to viral rebound after treatment 469 cessation, and they lead to emergence of resistant virus variants [77]. Accordingly, identification and 470 inhibition of novel targets, in addition to those already covered by HAART, are of high interest. In this 471 context, HIV accessory proteins have for a long time been regarded as promising drug targets [78,79]. Constructs were PCR-amplified from cDNA templates and inserted into the indicated expression 482 plasmids using standard restriction enzyme methods (S2 Table). pAcGHLT-B-DDB1 (plasmid #48638) 483 and pET28-UBA1 (plasmid #32534) were obtained from Addgene. The pOPC-UBA3-GST-APPBP1 484 co-expression plasmid, and the pGex6P2-UBC12 plasmid were obtained from MRC-PPU Reagents and 485 Services (clones 32498, 3879). Bovine erythrocyte ubiquitin and recombinant hsNEDD8 were 486 purchased from Sigma-Aldrich (U6253) and BostonBiochem (UL-812) respectively. Point mutations 487 were introduced by site-directed mutagenesis using KOD polymerase (Novagen). All constructs and 488 variants are summarised S3 Table.  489 Proteins expressed from vectors pAcGHLT-B, pGex6P1/2, pOPC and pET49b contained an N-terminal 490 GST-His-tag; pHisSUMO -N-terminal His-SUMO-tag; pET28, pRSF-Duet-1 -N-terminal His-tag; 491 pTri-Ex-6 -C-terminal His-tag. Constructs in vectors pAcGHLT-B and pTri-Ex-6 were expressed in 492 Sf9 cells, and constructs in vectors pET28, pET49b, pGex6P1/2, pRSF-Duet-1, and pHisSUMO in E. 493 coli Rosetta 2(DE3). 494 20 Recombinant baculoviruses (Autographa californica nucleopolyhedrovirus clone C6) were generated 495 as described previously [84].

Sf9 cells were cultured in Insect-XPRESS medium (Lonza) at 28°C in an 496
Innova 42R incubator shaker (New Brunswick) at a shaking speed of 180 rpm. In a typical preparation, 497 1 L of Sf9 cells at 3×10 6 cells/mL were co-infected with 4 mL of high titre DDB1 virus and 4 mL of 498 high titre DCAF1-CtD virus for 72 h. 499 For a typical E. coli Rosetta 2 (DE3) expression, 2 L of LB medium was inoculated with 20 mL of an 500 overnight culture and grown in a Multitron HT incubator shaker (Infors) at 37°C, 150 rpm until OD 600 501 reached 0.7. At that point, temperature was reduced to 18°C, protein expression was induced by addition 502 of 0.2 mM IPTG, and cultures were grown for further 20 h. During co-expression of CUL4 and ROC1 503 from pRSF-Duet, 50 µM zinc sulfate was added to the growth medium before induction. 504 Sf9 cells were pelleted by centrifugation at 1000 rpm, 4°C for 30 min using a JLA 9.1000 centrifuge 505 rotor (Beckman). E. coli cells were pelleted by centrifugation at 4000 rpm, 4°C for 15 min using the 506 same rotor. Cell pellets were resuspended in buffer containing 50 mM Tris, pH 7.8, 500 mM NaCl, 4 507 mM MgCl2, 0.5 mM tris-(2-carboxyethyl)-phosphine (TCEP), mini-complete protease inhibitors (1 508 tablet per 50 mL) and 20 mM imidazole (for His-tagged proteins only). 100 mL of lysis buffer was used 509 for resuspension of a pellet from 1 L Sf9 culture, and 35 mL lysis buffer per pellet from 1 L E. coli 510 culture. Before resuspension of CUL4/ROC1 co-expression pellets, the buffer pH was adjusted to 8.5. fractions were pooled and reduced to 5 mL using centrifugal filter devices (Vivaspin). If applicable, 100 521 µg GST-3C protease, or 50 µg thrombin, per mg total protein, was added and the sample was incubated 522 for 12 h on ice to cleave off affinity tags. As second purification step, gel filtration chromatography 523 (GF) was performed on an Äkta prime plus FPLC (GE), with Superdex 200 16/600 columns (GE), 524 equilibrated in 10 mM Tris-HCl pH 7.8, 150 mM NaCl, 4 mM MgCl 2 , 0.5 mM TCEP buffer, at a flow 525 rate of 1 mL/min. For purification of the CUL4/ROC1 complex, the pH of all purification buffers was 526 adjusted to 8.5. Peak fractions were analysed by SDS-PAGE, appropriate fractions were pooled and 527 concentrated to approx. 20 mg/mL, flash-frozen in liquid nitrogen in small aliquots and stored at -80°C. 528 Protein concentrations were determined with a NanoDrop spectrophotometer (ND 1000, Peqlab), using 529 theoretical absorption coefficients calculated based upon the amino acid sequence by ProtParam on the 530 ExPASy webserver [85]. 531 532

533
Prior to gel filtration analysis affinity tags were removed by incubation of 30 µg GST-3C protease with 534 6 µM of each protein component in a volume of 150 µL wash buffer, followed by incubation on ice for 535 12 h. In order to remove the cleaved GST-tag and GST-3C protease, 20 μL GSH-Sepharose FF beads 536 (GE) were added and the sample was rotated at 4 °C for one hour. GSH-Sepharose beads were removed 537 by centrifugation at 4°C, 3500 rpm for 5 min, and 120 µL of the supernatant was loaded on an analytical 538 GF column (Superdex 200 10/300 GL, GE), equilibrated in 10 mM Tris-HCl pH 7.8, 150 mM NaCl, 539 4 mM MgCl 2 , 0.5 mM TCEP, at a flow rate of 0.5 mL/min. 1 mL fractions were collected and analysed 540 by SDS-PAGE. For the following control samples, 120 µl of purified protein was applied directly to the 541 GF column, because no purification tag had to be removed by cleavage: SAMHD1, T4L-SAMHD1-542 CtD, SAMHD1-ΔCtD. For these samples, the concentration was adjusted to 18 µM, 37 µM and 30 µM, 543 respectively, to account for the lower extinction coefficient of these isolated protein components, in 544 order to allow for better visualisation of the elution peak. The GF column was calibrated using the high reservoir solution containing 100 mM Tri-Na citrate pH 5.5, 18% PEG 1000 and suspending over a 500 577 µl reservoir. Crystals grew over night at 18°C. Crystals were cryo-protected in reservoir solution 578 supplemented with 20% glycerol and cryo-cooled in liquid nitrogen. A data set from a single crystal was 579 collected at Diamond Light Source (Didcot, UK) at a wavelength of 0.92819 Å. Data were processed 580 using XDS [86] (S1 Table), and the structure was solved using molecular replacement with the program 581  Table). In the model, 584 94.5 % of residues have backbone dihedral angles in the favoured region of the Ramachandran plot, the 585 remainder fall in the allowed regions, and none are outliers. Details of data collection and refinement 586 statistics are presented in S1 Table. Coordinates and structure factors have been deposited in the PDB, 587 accession number 6zue. 588 DDB1/DCAF1-CtD/T4L-Vpr mus (1-92) complex. The DDB1/DCAF1-CtD/Vpr mus complex was 589 assembled by incubation of purified DDB1/DCAF1-CtD and HisSUMO-T4L-Vpr mus (residues 1-92), at 590 a 1:1 molar ratio, in a buffer containing 50 mM Bis-tris propane pH 8.5, 0.5 M NaCl, 4 mM MgCl 2 , 0.5 591 mM TCEP, containing 1 mg of HRV-3C protease for HisSUMO-tag removal. After incubation on ice 592 for 12 h, the sample was loaded onto a Superdex 200 16/600 GF column (GE), with a 1 mL GSH-593 Sepharose FF column (GE) connected in line. The column was equilibrated with 10 mM Bis-tris propane 594 pH 8.5, 150 mM NaCl, 4 mM MgCl2, and 0.5 mM TCEP. The column flow rate was 1 mL/min. GF 595 fractions were analysed by SDS-PAGE, appropriate fractions were pooled and concentrated to 4.5 596 mg/mL. CtD/Vpr mus /SAMHD1 complex and 2 µM UBCH5C-ubiquitin conjugate were applied to the grid, 631 incubated for 45 s, blotted with a Vitrobot Mark II device (FEI, Thermo Fisher Scientific) for 1-2 s at 632 8°C and 80% humidity, and plunged in liquid ethane. Grids were stored in liquid nitrogen until imaging.

25
Cryo-EM data collection. Initial negative stain and cryo-EM datasets were collected automatically 634 for sample quality control and low-resolution reconstructions on a 120 kV Tecnai Spirit cryo-EM (FEI, 635 Thermo Fisher Scientific) equipped with a F416 CMOS camera (TVIPS) using Leginon [94,95]. 636 Particle images were then analysed by 2D classification and initial model reconstruction using SPHIRE 637 [96], cisTEM [97] and Relion 3.07 [98]. These data revealed the presence of the complexes containing 638 both DDB1/DCAF1-CtD/Vpr mus (core) and CUL4/ROC1 (stalk). High-resolution data was collected on 639 a 300 kV Tecnai Polara cryo-EM (FEI, Thermo Fisher Scientific) equipped with a K2summit direct 640 electron detector (Gatan) at a nominal magnification of 31000x, with a pixel size of 0.625 Å/px on the 641 object scale. In total, 3644 movie stacks were collected in super-resolution mode using Leginon [94,95]  manually inspected to exclude images with substantial contaminants (typically large protein aggregates 649 or ice contaminations) or grid artefacts. Power spectra were manually inspected to exclude images with 650 astigmatic, weak, or poorly defined spectra. After these quality control steps the dataset included 2322 651 micrographs (63% of total). At this stage, the data set was picked twice and processed separately, to 652 yield reconstructions of states-1, -2 and -3 (analysis 1) and of the core (analysis 2). 653 For analysis 1, particle positions were determined using cisTEMs Gaussian picking routine, yielding 654 959,155 particle images in total. After two rounds of 2D-classification, 227,529 particle images were 655 selected for further processing (S4C, D Fig). Using this data, an initial model was created using Relion 656 3.07. The resulting map yielded strong signal for the core but only fragmented stalk density, indicating 657 a large heterogeneity in the stalk-region within the data set. This large degree of compositional (+/-658 stalk) and conformational heterogeneity (movement of the stalk relative to the core) made the 659 classification challenging. Accordingly, alignment and classification were carried out simultaneously. 660 The first objective was to separate the data set into three categories: "junk", "core" and "core+stalk". 661 Therefore, the stalk was deleted from the initial model using the "Eraser"-tool in Chimera [101]. This 662 core-map was used as an initial model for the Tier 1 3D-classification with Relion 3.07 at a decimated 663 pixel size of 2.5 Å/px. The following parameters were used: number of classes K=6, T=10, global step 664 search=7.5°, number of iterations= 25. The classification yielded two classes containing the stalk 665 (classes 3 and 5 containing 23% and 22% of the particle images, respectively) (  For analysis 2, particle positions were determined using template matching with a filtered map 672 comprising core and stalk using the software Gautomatch (https://www2.mrc-673 lmb.cam.ac.uk/research/locally-developed-software/zhang-software/). 712,485 particle images were 674 found, extracted with Relion 3.07 and subsequently 2D-classified using cryoSPARC [102], resulting in 675 505,342 particle images after selection (S5A, B Fig). These particle images were separated into two 676 equally sized subsets and Tier 1 3D-classification was performed using Relion 3.07 on both of them to 677 reduce computational burden (S5B Fig). The following parameters were used: initial model="core", 678 number of classes K=4, T=10, global step search=7.5°, number of iterations=25, pixel size 3.75 Å/px. 679 From these, the ones possessing both core and stalk were selected. Classes depicting a similar stalk 680 orientation relative to the core were pooled and directed into Tier 2 as three different subpopulations 681 containing 143,172,193,059 and 167,666 particle images,respectively (S5B Fig). 682 For Tier 2, each subpopulation was classified separately into 4 classes each. From these 12 classes, all 683 particle images exhibiting well-defined densities for core and stalk were pooled and labelled 684 "core+stalk", resulting in 310,801 particle images in total. 193,096 particle images representing classes 685 containing only the core were pooled and labelled "core" (S5B Fig)  686 For Tier 3, the "core" particle subset was separated into 4 classes which yielded uninterpretable 687 reconstructions lacking medium-or high-resolution features. The "core+stalk" subset was separated into Mobile phase A consisted of 0.1% (v/v) formic acid and mobile phase B of 80% (v/v) acetonitrile with 733 0.1% (v/v) formic acid. Flow rates were 0.3 μL/min using gradients optimized for each chromatographic 734 fraction from offline fractionation, ranging from 2% mobile phase B to 55% mobile phase B over 735 90 min. MS data were acquired in data-dependent mode using the top-speed setting with a 3 s cycle 736 time. For every cycle, the full scan mass spectrum was recorded using the Orbitrap at a resolution of 737 120,000 in the range of 400 to 1,500 m/z. Ions with a precursor charge state between 3+ and 7+ were 738 isolated and fragmented. Analyte fragmentation was achieved by Higher-Energy Collisional 739 Dissociation (HCD) [109] and fragmentation spectra were then recorded in the Orbitrap with a resolution 740 of 50,000. Dynamic exclusion was enabled with single repeat count and 60 s exclusion duration. 741 CLMS processing. A recalibration of the precursor m/z was conducted based on high-confidence 742 (<1% false discovery rate (FDR)) linear peptide identifications. The re-calibrated peak lists were 743 searched against the sequences and the reversed sequences (as decoys) of cross-linked peptides using 744 the Xi software suite (v.1.7.5.1) for identification [110]. Final crosslink lists were compiled using the 745 identified candidates filtered to <1% FDR on link level with xiFDR v.2.0 [111] imposing a minimum 746 of 20% sequence coverage and 4 observed fragments per peptide. 747 CLMS analysis. In order to sample the accessible interaction volume of the SAMHD1-CtD consistent 748 with CLMS data, a model for SAMHD1 was generated using I-TASSER [112]. The SAMHD1-CtD, 749 which adopted a random coil configuration, was extracted from the model. In order to map all crosslinks, 750 missing loops in the complex structure were generated using MODELLER [113]. An interaction volume 751 search was then submitted to the DisVis webserver [61] with an allowed distance between 1.5 Å and 22 752 Å for each restraint using the "complete scanning" option. The rotational sampling interval was set to 753 9.72° and the grid voxel spacing to 1Å. The accessible interaction volume was visualised using UCSF 754