Transient DNA Binding Induces RNA Polymerase II Compartmentalization During Herpesviral Infection Distinct From Phase Separation

During lytic infection, Herpes Simplex Virus 1 generates replication compartments (RCs) in host nuclei that efficiently recruit protein factors, including host RNA Polymerase II (Pol II). Pol II and other cellular factors form hubs in uninfected cells that are proposed to phase separate via multivalent protein-protein interactions mediated by their intrinsically disordered regions. Using a battery of live cell microscopic techniques, we show that although RCs superficially exhibit many characteristics of phase separation, the recruitment of Pol II instead derives from nonspecific interactions with the viral DNA. We find that the viral genome remains nucleosome-free, profoundly affecting the way Pol II explores RCs by causing it to repetitively visit nearby binding sites, thereby creating local Pol II accumulations. This mechanism, distinct from phase separation, allows viral DNA to outcompete host DNA for cellular proteins. Our work provides new insights into the strategies used to create local molecular hubs in cells.


Introduction
all viral proteins, we identified predicted IDRs longer than 10 amino acids, and used these to 124 estimate what fraction of each protein sequence is unstructured ( Figure 1E, Table S1). When 125 categorized by temporal class, we noted that the immediate early (IE) and viral tegument 126 proteins-the two groups that are presented to the cell first upon infection-had the highest 127 fraction of predicted intrinsic disorder. In fact, when compared to a list of proteins known to 128 undergo LLPS in vitro, the IE and tegument proteins are slightly more disordered ( Figure 1E; 129 Table S2 and citations within). Under the working hypothesis that multivalent interactions 130 between protein IDRs drive phase separation, the similarity in predicted disorder profiles 131 between this curated list and the IE and tegument proteins suggests that IDRs in viral proteins are 132 as likely to drive LLPS as experimentally validated proteins. 133 Based on the above descriptive observations, we hypothesized that Pol II is recruited to RCs 134 through interactions between its CTD and other IDR-containing proteins within the RC. To test 135 this, we measured the FRAP dynamics of Pol II in RCs. We saw a consistent slowing of recovery 136 as infection progressed and RCs got larger ( Figure 1F), which could be interpreted as evidence 137 that RCs act as a separate liquid phase that incorporates Pol II and sequesters it from the rest of 138 the nucleoplasm. Subsequent experiments to directly test this hypothesis, however, cast doubt on 139 this interpretation. 140 We recently reported that hub formation by Pol II in uninfected cells occurs in a manner 141 dependent on the length of the Pol II CTD (Boehning et al., 2018). To test whether the Pol II 142 CTD likewise mediates interaction with RCs, we compared Pol II accumulation in RCs in cells 143 with the wild-type Pol II CTD (with 52 heptad repeats) and cell lines bearing truncated (25 144 repeats) or extended (70 repeats) CTDs. To our surprise, the length of the CTD had no detectable 145 effect on its incorporation into RCs ( Figure 1G), suggesting that Pol II does not require IDR 146 interactions through its CTD to become enriched. 147 As a further test of the role of IDR interactions in Pol II accumulation in RCs, we treated 148 cells with 1,6-hexanediol, which disrupts weak hydrophobic interactions between IDRs that 149 drive LLPS (Boehning et  concentration (10% v/v) of 1,6-hexanediol for one, five, or ten minutes. Five minutes after 152 treatment, the morphology of the nucleus began to change, and by ten minutes it was noticeably 153 deformed, consistent with widespread disruption of cellular organization by 1,6-hexanediol (Lin 154 et al., 2016). Nonetheless, Pol II remained highly enriched in RCs after 1,6-hexanediol treatment 155 ( Figure 1H), implying that formation of RCs does not require interactions between IDRs of Pol II 156 and viral proteins. 157

Pol II diffusion within and across RC boundaries is inconsistent with an LLPS model 158
The data outlined in Figure 1 present a potential contradiction, as RCs exhibit several 159 properties commonly associated with phase separation in vitro, yet Pol II recruitment to RCs was 160 not susceptible to disruption by 1,6-hexanediol or dependent on CTD length. Given these results, 161 we sought to better understand the mechanism driving the enrichment of Pol II in RCs by 162 measuring the behavior of individual Pol II molecules. To accurately capture both immobile and 163 freely diffusing Pol II molecules, we used stroboscopic photo-activatable single particle tracking 164 (spaSPT) to visualize and track molecules ( Figure 2A) (Hansen et al., 2017. We labeled 165 HaloTag-Pol II with equal amounts of JF549 and PA-JF646 (Grimm et al., 2015(Grimm et al., , 2016, which 166 allowed us to monitor its overall distribution using the former dye while tracking individual 167 molecules using the latter dye. We generated masks of the location of the nuclear periphery and 168 individual RCs and used these masks to sort trajectories as either "inside" or "outside" of RCs 169 ( Figure 2B). 170 Quantitative measurements can be made by building histograms of all the displacement 171 distances from the trajectories and fitting to a two-state model in which Pol II can either be freely 172 diffusing ("free"), or immobile and hence presumably bound to DNA ("bound") ( Figure 2C). 173 Such a two-state model gives two important pieces of information: the fraction of "bound" and 174 "free" molecules, and the apparent diffusion coefficient of each population . 175 It is important to note that, because this modeling approach takes the aggregate of many 176 thousands of molecules, these data cannot measure how long a particular molecule remains 177 bound in a given binding event. Here, "bound" refers to both specific DNA binding events-e.g. 178 molecules assembled at a promoter or engaged in mRNA elongation-as well as transient, 179 nonspecific binding interactions. 180 The difference in the behavior of Pol II inside of RCs compared with the rest of the 181 nucleoplasm is immediately apparent from examining the lengths of jumps between consecutive 182 frames ( Figure 2D). Fitting a two-state model to the data, we were surprised to find that the mean 183 apparent diffusion coefficient of the free population was unchanged between trajectories inside 184 of RCs compared with those outside RCs or in uninfected cells. If RCs were a bona fide separate 185 phase, one would expect differences in molecular crowding or intermolecular interactions to 186 predominantly affect free diffusion, resulting in substantially different diffusion coefficients 187 between the populations (Bergeron-Sandoval et al., 2016). Furthermore, the similarity in 188 diffusion coefficients between infected and uninfected cells argues against a separate viral 189 protein-Pol II complex responsible for recruitment to RCs. 190 We verified this result in two ways: First, we performed a fluorescence loss in 191 photobleaching (FLIP) experiment, in which a strong bleaching laser targets the inside of an RC 192 and loss of fluorescence elsewhere in the nucleus is measured to quantify exchange of Pol II 193 between the nucleoplasm and the RC. Consistent with the spaSPT data, we see that Pol II 194 molecules exchange between RCs and the rest of the nucleoplasm as fast, if not faster, than Pol II 195 in an uninfected cell ( Figure 2F). Similar results were obtained by using Pol II tagged with the 196 photo-convertible fluorescent protein Dendra2 (Cisse et al., 2013) and photo-converting, rather 197 than bleaching, molecules in the RC ( Figure S2). Thus, Pol II molecules diffuse out of the RC,198 rather than remaining sequestered within a single compartment. Second, a liquid-liquid phase 199 separation model predicts that a diffusing Pol II molecule within an RC will be more likely to 200 remain within the RC than to exit the RC when it reaches the compartment boundary. To test this 201 prediction, we examined all trajectories for events in which a molecule crosses from inside of the 202 RC to outside, or vice versa, to look for evidence of such a constraint. Comparing the 203 distribution of displacements for a particle going from inside the RC to outside, we see no 204 difference in the distribution of displacements, either entering or leaving RCs, when compared to 205 uninfected cells in which mock RC annotations were randomly imposed in silico ( Figure 2G, 206 Figure S3). With these experiments we cannot detect any evidence of a boundary for molecules 207 entering or leaving RCs, further arguing that this compartment does not consist of a distinct 208 liquid phase. 209 While the two-state model shows no change in diffusion coefficient of molecules inside 210 versus outside RCs, the fraction of molecules in the "bound" state nearly doubles inside RCs, 211 reaching ~70% ( Figure 2H). The diffusion coefficients we measured with the bound populations 212 are still consistent with those of chromatin , indicative that these 213 populations reflect DNA binding ( Figure S4). The increase in the fraction of bound molecules is 214 further supported by the FRAP data ( Figure 1F). We verified this was not an artifact of the 215 masking process by using the same process of artificially imposing RCs randomly in silico 216 ( Figure S3), and we found no difference in the fraction of bound molecules. Such a significant 217 shift in the relative populations of bound and free molecules inside RCs, taken together with the  218  previous data, shows that the mechanism driving Pol II recruitment to RCs is dominated by DNA  219 binding, rather than by IDR-mediated interactions that sequester Pol II away in a separate liquid 220 phase. 221 Pol II recruitment to RCs occurs independent of transcription initiation 222 The above data argue against formation of RCs by LLPS, suggesting that some other 223 mechanism must explain the doubling of DNA-bound Pol II in RCs. One possibility is that 224 increased recruitment of Pol II is explained by high levels of active transcription within RCs. 225 Multiple lines of evidence suggest that transcription derived from the viral genome is activated to 226 a much greater extent than even the most highly transcribed host mRNA (Rutkowski et al., 2015) 227 and so an enriched population of actively elongating Pol II would be expected to increase the 228 "bound" population. 229 To test whether active transcription is necessary for Pol II recruitment to RCs, we treated 230 infected cells with either Flavopiridol or Triptolide, two small molecules that selectively inhibit 231 different stages of transcription initiation ( Figure 3A). replication machinery, so we allowed the infection to progress for four hours before treating with 241 either compound. Cells at this timepoint have well formed RCs, and Pol II binding is already 242 greatly increased ( Figure 2H). We treated these cells with 1 µM Flavopiridol or 1 µM Triptolide 243 for 15, 30, or 45 minutes to allow any elongating polymerases to finish transcribing ( Figure 3B). 244 After treatment, we fixed cells, and we performed RNA fluorescence in situ hybridization 245 (FISH) using a probe against an intronic region to detect nascent transcripts and, in parallel, and 246 immunofluorescence to mark the DNA-binding protein ICP8, a common marker for RCs (Taylor 247 et al., 2003). After 30 minutes of drug treatment, transcription is significantly reduced ( Figure  248 S5). Remarkably, even after 45 minutes of treatment, ~80% of the Pol II signal remains within 249 RCs ( Figure 3C, Figure S5). These data suggest that the recruitment of Pol II to RCs occurs 250 largely independently of transcription, or even stable engagement with gene promoters. 251 We next tested whether treatment with these transcription inhibitors would change the bound 252 fraction measured by spaSPT. In uninfected cells, Triptolide or Flavopiridol treatment both 253 reduce the fraction of bound Pol II by half, to ~15% ( Figure 3D), similar to what others have 254 reported (Boehning et al., 2018;Teves et al., 2018). Surprisingly, inhibition of transcription with 255 Flavopiridol reduced the bound fraction inside of RCs by only ~5% ( Figure 3D). Even treatment 256 with Triptolide, which prevents stable engagement with TSS-proximal DNA only reduced the 257 fraction bound by ~12% ( Figure 3D). We were surprised to see that with either drug treatment, 258 HSV1 infection appears to also confer some resistance to the effects of the drugs on Pol II 259 binding to host chromatin, despite the fact that these concentrations of transcription inhibitors are 260 sufficient to prevent new transcription ( Figure 3D, Figure S5) . Given the inherent limitation of  261  spaSPT for inferring the length of binding events, we wanted to confirm that drug treatment  262  prevented stable Pol II binding. Indeed, FRAP experiments in infected cells treated with  263 Triptolide show a dramatically faster recovery rate ( Figure 2E). in RC formation and function. We therefore sought to measure the amount of DNA in RCs using 280 Oligopaints, a variant of DNA fluorescence in situ hybridization, to target fluorescent probes to 281 two specific regions of the viral genome ( Figure  the two probe sets. These fluorescence intensities were compared to samples that were infected 285 in the presence of phosphonoacetic acid (PAA), a compound that prevents replication of the viral 286 DNA and thus ensures that there is one copy of the viral genome per punctum ( Figure 4B) 287 (Eriksson and Schinazi, 1989). 288 While there is a great deal of expected RC-to-RC heterogeneity, the number of genomes 289 within an RC correlates well with the time post infection ( Figure 4C). There is also a strong 290 correlation between RC size and genome copy number ( Figure S6). Based on these data, we 291 calculate that the average RC at 6 hpi has a DNA concentration of 3.9 x10 4 bp/µm 3 , 292 approximately 240 times less concentrated than average host chromatin. The sum of all RCs in 293 an average infected cell corresponds to just ~0.2% of total DNA in karyotypically normal human 294 nuclei (Table S3). Despite being orders of magnitude lower in DNA content and concentration, 295 inhibition of viral DNA replication with PAA caused a ~20% decrease in the fraction of bound 296 Pol II molecules inside the pre-replication foci as measured by spaSPT, nearly down to the level 297 of host chromatin ( Figure 4D). This, despite the fact that all immediate early and early genes, 298 including many of the proteins known to interact with Pol II, are still highly expressed in PAA-299 treated samples (Lester and DeLuca, 2011;Zhou and Knipe, 2002 Taylor  309 and Knipe, 2004). Moreover, immunofluorescence against histones shows no detectable signal in 310 RCs (Dembowski and DeLuca, 2015). In addition, one function of viral ICP0 is to actively evict 311 histones from DNA, which suggests that the HSV1 genome is maintained largely free of histones 312 (Lee et al., 2016). 313 To test histone occupancy of the viral DNA, and get a measure of its accessibility, we turned 314 to ATAC-seq, which gives signal proportional to the accessibility of the DNA at a given locus 315 (Buenrostro et al., 2013). We infected our HaloTag-Pol II cell line, and performed Tn5 316 transposition reactions at 2, 4, and 6 hpi. We also included a sample that was uninfected, and one 317 infected in the presence of PAA. At all times after infection, the distribution of fragment lengths 318 mapping to the viral genome showed a much faster decay, and no evidence of nucleosomal 319 laddering, in contrast to reads that map to the host genome ( Figure 4E, Figure S7). When we 320 visualized the reads along the viral genome, the profiles were strikingly flat and featureless, 321 suggesting that all regions of the viral genome are equally accessible to Tn5 ( Figure 4F). 322 Based on the amount of viral DNA present in an infected cell, we calculated the fraction of 323 reads one would expect to map to the virus relative to the host. of RCs is two orders of magnitude more accessible, despite its overall lower DNA concentration 327 relative to host DNA (Table S3). 328 In metazoan genomes, active genes can be identified by their high accessibility (Thurman et  329 al., 2012). An average of all annotated human mRNA genes, centered at the TSS, shows a 330 characteristic peak of accessibility at the TSS for reads with a length corresponding to inter-331 nucleosomal distances (<100 bp), and a characteristic trough of mononucleosome sized 332 fragments (180 -250 bp) ( Figure 4G). By contrast, TSS averages mapped to the viral genome 333 for either short or mono-nucleosome fragments show no changes in accessibility. Thus, even 334 averaging over all viral transcripts, it is clear that the entire viral DNA remains equally 335 accessible ( Figure 4H). Taken together, these data indicate that the HSV genome is maintained in 336 a largely nucleosome-free state and thus, highly accessible to DNA binding proteins such as Pol 337 II. 338 Transient DNA-protein interactions drive Pol II hub formation through repetitive 339 exploration of the replication compartment 340 Knowing that the DNA inside RCs is vastly more accessible to nuclear factors than the host 341 chromatin, we next asked what emergent properties of this accessible DNA might help explain 342 Pol II recruitment. We took advantage of a viral strain that is able to incorporate chemically 343 modified nucleotides during replication (Dembowski and DeLuca, 2015), to label newly 344 replicated viral DNA with Alexa Fluor 647, and thus allow DNA in the RCs to be visualized at 345 high resolution using stochastic optical reconstruction microscopy (STORM) ( Figure 5A) (Rust 346 et al., 2006). Unlike host chromatin, whose overall density and compaction scales reproducibly 347 with domain size for active chromatin (Boettiger et al., 2016), viral DNA shows a spatial 348 variability in local density of nearly three orders of magnitude. 349 The greater accessibility and higher variability in local density of viral DNA lend themselves 350 to a possible mechanism by which Pol II becomes enriched. Recent theoretical work has shown 351 that a polymer, like DNA, which has many binding sites in close proximity can induce a particle 352 to revisit the same or adjacent sites repetitively during its exploration of the nucleus (Amitai,353 2018) ( Figure 5B). In such a case, we should be able to see signatures in our spaSPT dataset of 354 Pol II continually revisiting adjacent sites on the viral DNA. To check, we calculated the angle 355 formed by every three consecutive displacements and compiled these angles into a histogram for 356 all particles strictly identified as freely diffusing (see Methods) ( Figure 5C) (Izeddin et al.,357 2014). For particles experiencing ideal Brownian motion, the angular histogram will be isotropic. 358 Anisotropy can arise either by imposing reflective boundaries on the particle, or adding the 359 aforementioned "traps" thereby giving the particle a greater probability of revisiting proximal 360 sites before diffusing away (Amitai, 2018). 361 In uninfected cells, and in infected cells outside of RCs, Pol II displays diffusion that is only 362 mildly anisotropic, consistent with mostly Brownian motion throughout the nucleus. In stark 363 contrast, inside RCs Pol II diffusion is more anisotropic, especially around 180° ( Figure 5D). To 364 compare across samples, we computed the likelihood of a backward translocation (180° ± 30°) 365 relative to the likelihood of a forward translocation (0° ± 30°). Analyzed this way, Pol II in an 366 uninfected nucleus has a 1.3-fold greater chance of moving backward after a given translocation 367 than it has of moving forward ( Figure 5E). As expected, Pol II outside of RCs in infected cells 368 has a nearly identical value to Pol II in an uninfected cell. Inside of an RC, however, that 369 probability increases to 1.7-fold, showing that this effect is unique to RCs ( Figure 5E). In cells 370 treated with Triptolide, we see that when stable binding is inhibited, the effect created by 371 transient binding events is further amplified ( Figure 5E). Under this condition, Pol II inside an 372 RC is 2-fold more likely to have a backward displacement after a forward one ( Figure 5D), 373 which helps explain the dramatic retention of Pol II inside RCs, even 45 minutes after inhibition 374 of transcription ( Figure 3C). Importantly, in uninfected cells where RC annotations have been 375 shuffled in silico, no additional anisotropy is observed ( Figure S3). 376 These data are most consistent with a model in which Pol II repetitively visits the highly 377 accessible viral genome via multiple weak, transient binding events that result in Pol II rapidly 378 hopping along the DNA. The sharp anisotropy of the molecular exploration within the RC means 379 that a given Pol II molecule that enters an RC is more likely to visit the same site, or sites close 380 in proximity, multiple times before it either finds a stable binding site or diffuses away. 381 The heterogeneous distribution of viral DNA within RCs, and the anisotropic way Pol II 382 explores RCs, is also borne out in the distribution of Pol II molecules. We performed 3D increases for all radii between 0 and 1000 nanometers, suggesting that Pol II forms hubs within 393 RCs and that this clustering occurs at multiple length scales. This is consistent with other recent 394 studies of Pol II in uninfected cells (Boehning et al., 2018), and in contrast to a structural protein 395 like the CCCTC-binding factor (CTCF), whose L(r)-r curve shows clusters of a single 396 characteristic size (Hansen et al., 2017). 397

Nonspecific interactions with viral DNA license recruitment of other proteins 398
Seeing that Pol II is recruited to RCs via transient and nonspecific binding to the viral 399 genome made us wonder whether this effect was specific to Pol II, or whether DNA accessibility 400 can generally drive the recruitment of DNA-binding proteins to RCs. Certainly, many other 401 DNA-binding proteins are recruited to RCs (Dembowski and DeLuca, 2015; Taylor and Knipe,  402 2004). To assess whether nonspecific DNA binding could be responsible for accumulation of 403 other proteins within RCs, we looked at enrichment of the tetracycline repressor (TetR). TetR is 404 a sequence-specific transcription factor found in bacteria that binds with high affinity to the 19 405 bp tetO sequence, which is absent in both human and HSV1 genomes (Bolintineanu et al., 2014). 406 Thus, we reasoned that if nonspecific DNA association is the mechanism driving recruitment to 407 the RC, TetR should also be recruited to the RC. 408 We transiently transfected TetR-GFP into the HaloTag-Pol II cell line, then infected them 409 with HSV1 the following day. TetR-GFP, lacking a nuclear localization signal (NLS), 410 ubiquitously occupies both the nucleus and cytoplasm. As predicted by our model, GFP signal 411 was enriched inside of RCs ( Figure 6). Pixel line scans of the matched Pol II and TetR channels 412 show that the level of enrichment is modest (~25% over background for TetR, compared with 413 ~200% for Pol II), but the two signals showed a high Spearman correlation (r > 0.77). A 414 fluorescent protein with only an NLS showed no enrichment at RCs in infected cells ( Figure S8). 415 Thus, even a sequence-specific transcription factor with no cognate binding sites in the viral 416 genome can be recruited to RCs based on its modest affinity for nonspecific DNA sequences. 417 These data suggest a model in which viral Pol II recruitment consists of transient, nonspecific 418 binding/scanning events of the highly exposed viral genome ( Figure 7A). A DNA-binding 419 protein exploring the nucleus (uninfected, or infected but outside of RCs) may encounter some 420 occasions for nonspecific interaction with duplex DNA, but because of the condensed nature of 421 the host chromatin, these binding/scanning events are necessarily short ( Figure 7B). In addition, 422 it may take a protein many thousands of these transient binding events to finally reach a high-423 affinity site . Within RCs, multiple copies of the highly accessible HSV1 424 genome are present, nonspecific events happen more frequently, with fewer and shorter 3D 425 excursions between DNA contacts ( Figure 7C), leading to redundant exploration of the RC and 426 local accumulations of protein. This enrichment becomes even more skewed when aided by a 427 high density of other interactors-for example when transcription is active on the viral DNA, 428 and thus the both specific and general transcription factors are enriched via the same mechanism. 429 Out data strongly indicate, though, that host Pol II accumulation at HSV1 RCs is not dependent 430 on active transcription but instead is largely driven by transient, nonspecific protein-DNA 431 interactions resulting from the high accessibility of the viral DNA. 432

Multiple routes to create high local concentrations 434
Here we have demonstrated that Herpes Simplex Virus type 1 accumulates Pol II in 435 replication compartments through a novel mechanism: its unusually accessible DNA genome 436 provides many potential nonspecific binding sites, which causes a net accumulation of Pol II, 437 acting as a molecular sink even in the absence of transcription. Such a mechanism for locally 438 concentrating proteins is revealing, as it neither requires the formation of a stable 439 macromolecular structure nor produces any behaviors at the single-molecule level suggesting a 440 separate liquid phase. It is particularly striking because, from the macroscopic view, Pol II 441 recruitment to RCs appears to share many of the behaviors commonly attributed to liquid-liquid 442 phase separation-enrichment of proteins that have high intrinsic disorder, spherical, dynamic 443 structures that undergo fusion, and a change in refractive index-and yet RCs are clearly a 444 distinct class of membraneless compartments. 445 Given the high prevalence of IDRs in the viral proteome, it is likely that they have crucial 446 functions in other aspects of the viral lifecycle, such as the assembly of the capsid or packaging 447 of tegument proteins prior to envelopment. It still remains a possibility that these IDRs form 448 some phase-like system inside of RCs. Crucially, our data demonstrate that, even if this is the 449 case, it does not contribute to the enrichment or entrapment of Pol II. Our results prompt the 450 need for a better characterization of bona fide phase separation, with a focus on its functional 451 consequences in vivo. These data underscore the importance of rigorously dissecting the diverse 452 mechanisms driving subcellular compartment formation, and suggest that caution should be 453 exercised before immediately assigning LLPS as the primary assembly mechanism or 454 interpreting the functional role of a phase-separated system solely based on macroscopic 455 behaviors. 456 We shifted the equilibrium in those locations, thereby enhancing the probability of regulatory factors 465 binding at specific, functional sites. As a consequence, assembly of typically inefficient multi-466 protein complexes like the transcription pre-initiation complex (Darzacq et al., 2007), could 467 become more favorable inside of RCs. We speculate that nonspecific protein-DNA interactions 468 could be a general mechanism used by many other viruses. We also note that many RNA-binding 469 proteins have been reported to undergo apparent LLPS (Courchaine et al., 2016) and believe it 470 will be interesting to explore if RNA-binding proteins share a similar mechanism to what we 471 describe here. 472

Nonspecific binding events represent an important part of nuclear exploration 473
Our data also reveal a previously underappreciated aspect of how a DNA binding protein 474 finds its target DNA within the nucleus. It has long been recognized that nonspecific binding to 475 DNA could greatly accelerate the target search process by allowing for sliding in 1D along the 476 DNA, thereby reducing the search space and allowing for faster-than-diffusion association 477 kinetics (Berg et al., 1981). This is the case in bacterial systems where DNA is generally 478 accessible to binding. A number of theoretical studies have addressed various aspects of the 479 problem in eukaryotic systems, where TFs compete with nucleosomes for access to DNA 480 (Mirny, 2010 The data we present here offer a new perspective on the importance of nonspecific and low-487 affinity binding, and competition for nucleosomes inside the nucleus. When the virus begins 488 replicating, we measured the newly synthesized DNA to be ~130 times more accessible to  binding proteins than the surrounding host chromatin (Table S3). Not only is there simply more 490 DNA available to bind in the absence of nucleosomes, but the distance that a protein can scan in 491 1D is greatly increased due to the lack of impeding nucleosomes. Pol II is not a canonical DNA-492 binding protein, and no systematic study has been undertaken to measure its binding affinities 493 against with different substrates. Still, having evolved to be an enzyme that must transcribe 494 highly diverse DNA sequences suggests that its affinity for DNA outside of an assembled PIC 495 may be relatively high. A recent study which explicitly modeled nonspecific DNA binding in the 496 context of 3D genome organization finds that the most effective regime for recruiting a DNA 497 binding protein is exactly what the virus appears to have settled on-that is a region of low 498 overall DNA density that is free of nucleosomes recruiting a protein with high nonspecific 499 affinity (Cortini and Filion, 2018). 500 The virus' strategy may be shared by the host in facilitating enhancer-promoter contacts. 501 Both promoter and enhancer elements are identifiable by their increased DNA accessibility. 502 While some recent reports have suggested that enhancers may phase separate as a mechanism of 503 activating transcription ( We would like to thank James Goodrich, Jennifer Kugel, and Robert Abrisch for providing the 530 HSV1 strain KOS that began this project, and for helpful discussions. were performed on shared instrumentation at the CRL Molecular Imaging Center, supported by 541 The Gordon and Betty Moore Foundation. We would like to thank Holly Aaron and Jen-Yi Lee 542 for their assistance. DNA sequencing in this work used the Vincent J Coates Genomics 543 Sequencing Laboratory at UC Berkeley, supported by NIH 669 S10 Instrumentation Grants 544 S10RR029668 and S10RR027303. 545 affinity, nonspecific interaction. This, in turn, may reduce the distance a molecule might diffuse 673 before its next binding event, and increases both the chances of that molecule remaining in close 674

Figure Legends
proximity and the chances that it will find a high binding energy interaction. 675 Spearman's correlation was used to measure the covariation in the two channels across the line. 748

Supplementary Figure Legends
All scale bars are 10 µm. 749 750 were identified 3 hpi, and followed until they moved out of the focal plane. 765

Experimental Procedures 766
Tissue Culture 767 Human U2OS cells (female, 15 yr old, osteosarcoma) were cultured at 37°C and 5% CO2 in 768 1 g/L glucose DMEM supplemented with 10% Fetal Bovine Serum and 10 U/mL Penicillin-769 Streptomycin, and we subcultivated at a ratio of 1:3 -1:6 every two to four days. Stable cell 770 lines expressing the exogenous gene product a-amanitin resistant HaloTag-RPB1(N792D) or 771 Dendra2-RPB1(N792D) were generated using Fugene 6 following the manufacturer's protocol, 772 and selection with 2 µg/mL a-amanitin. Stable colonies were pooled and maintained under 773 selection with 1 µg/mL a-amanitin to ensure complete replacement of the endogenous RPB1 774 pool, as described previously (Boehning et  Specific protein sequences were placed in a table and this was fed into the script. All protein 818 sequences were downloaded from the reference organism at uniport.org. The resulting traces 819 were smoothed by a rolling mean of 8 residues to remove noise and prevent single low-energy 820 residues from splitting single large IDRs into multiple apparent IDRs. Contiguous substrings of 821 residues with centered-mean IUPred disorder likelihood greater than 0.55 were annotated as 822 "disordered regions" (Fig. 1E), and those contiguous regions larger than 10 amino acids were 823 included in the calculation of "fraction IDR". 824

825
FRAP experiments were performed as previously described, with modifications. HaloTag-826 RPB1 cells labeled with 500 nM JF549 were imaged on an inverted Zeiss LSM 710 AxioObserver 827 confocal microscope with an environment chamber to allow incubation at 37°C and 5% CO2. 828 JF549 was excited with a 561 nm laser, and the microscope was controlled with Zeiss Zen 829 software. Images were acquired with a 63x Oil immersion objective with a 3x optical zoom. 830 1200 total frames were acquired at a rate of 250 msec per frame (4 Hz). Between frames 15 and 831 16, an 11-pixel (0.956 µm) circle was bleached, either in the center of a RC, or in a region of the 832 nucleus far from the nuclear periphery or nucleoli. 833 FRAP movies were analyzed as previously described (Hansen et al., 2017). Briefly, the 834 center of the bleach spot was identified manually, and the nuclear periphery segmented using 835 intensity thresholding that decays exponentially to account for photobleaching across the time of 836 acquisition. We measured the intensity in the bleach spot using a circle with a 10 pixel diameter, 837 to make the measurement more robust to cell movement. The normalized FRAP values were 838 calculated by first internally normalizing the signal to the intensity of the whole nucleus to 839 account for photobleaching, then normalizing to the mean value of the spot in the first 15 frames. 840 We corrected for drift by manually updating a drift-correction vector with the stop drift every 841 ~40 frames. FRAP values from individual cells were averaged across replicates to generate a 842 mean recovery curve, and the error displayed is the standard error of the mean. 843 Fluorescence Loss in Photobleaching (FLIP) 844 FLIP experiments were performed on the same microscope described above for FRAP. 845 Rather than bleach an 11-pixel spot a single time, in FLIP the spot is bleached with a 561 nm 846 laser (or in the case of Dendra2, photoconverted with a 405 nm laser) between each acquisition 847 frame. Movies were collected for 1000 frames at 250 msec per frame (4 Hz), or 1 frame per 848 second (1 Hz) for Dendra2. 849 FLIP movies were analyzed using the same core Matlab code as the FRAP data, except that 850 fluorescence intensities from another 10-pixel circle were recorded to measure the loss of 851 fluorescence elsewhere in the nucleus. This analysis spot was chosen to be well away from the 852 bleach spot, either at a neighboring RC in infected samples or somewhere else in the 853 nucleoplasm far away from both the nuclear periphery and nucleoli. Instead of internally 854 correcting for photobleaching, photobleaching correction was based on an exponential decay 855 function empirically determined to be at a rate of e -0.09 per frame. FLIP data from multiple cells 856 were averaged together to determine the mean and standard error for a given condition. 857

858
RNA FISH was used to measure the transcription output for a given RC. corresponding to the wavelength of dye used. All samples were imaged the same day after 887 hybridaztion and/or incubation with secondary antibody, and all samples to be quantitatively 888 compared across coverslips were imaged on the same day using exactly the same illumination 889 and acquisition settings to minimize coverslip-to-coverslip variation. 890 Single Particle Tracking (spaSPT) 891 Single particle tracking experiments were carried out as previously described (cit SPT data sets were processed in 4 general steps using a custom-written Matlab (Mathworks): 915 1) Masks for RCs were annotated manually, 2) the masks were corrected for drift throughout the 916 sample acquisition, 3) particles were localized and trajectories constructed, and 4) trajectories 917 were sorted as "inside" compartments or "outside". 918 First, the 100 frames at the beginning and the end of each movie were separately extracted 919 and a maximum-intensity projection used to generate "before" and "after" images of the cell or 920 cells in the field of view. These images would be used to correct for movement of the cell as well 921 as the individual RCs. For each cell, the nucleus was annotated in the "before" image, and then 922 again in the "after" image. We assumed that the cell movement over the ~4 minutes of 923 acquisition was approximately linear, and calculated the drift-corrected nuclear boundary for 924 every frame in the stack of SPT images. The same procedure was applied to each of the 925 replication compartments. Particle localization and tracking were implemented based on an 926 adapted version of the Multiple Target Tracking (MTT) algorithm, available at 927 https://gitlab.com/tjian-darzacq-lab/SPT_LocAndTrack. In the first step, particles were identified 928 with the following input parameters: Window = 9 px; Error Rate = 10 -6.25 ; Deflation Loops = 0. 929 Following detection, a mask generated from the drift-corrected nuclear boundary was applied to 930 discard any detections not within the nucleus. Trajectories were reconstructed with the following 931 parameters: Dmax = 10 µm 2 /sec; Search exponent factor = 1.2; Max number of competitors = 3; 932 Number of gaps allowed = 1. 933 Finally, after trajectories have been reconstructed, they were sorted as "inside" RCs or 934 "outside". To minimize the potential for bias in calling trajectories inside of compartments, we 935 only required a single localization in a trajectory to fall within a compartment for that trajectory 936 to be labeled as "inside". As is discussed in the main text, we tested this sorting strategy for 937 implicit bias by computationally generating mock RCs in uninfected or infected samples ( Figure  938 S3). To do this, all of the annotations for RCs from the infected samples (n = 817), as well as the 939 distribution of number of RCs per infected cell, were saved in a separate library. We then took 940 the uninfected cells and, in a similar process as described above, annotated the nuclear boundary 941 and nucleoli. We then randomly sampled from distribution of RCs per cell a number of RCs to 942 place in the nucleus, and then from the library of annotations randomly chose these RCs and 943 placed them in the nucleus by trial-and-error until all of the chosen RCs could be placed in the 944 nucleus without overlapping with each other, a nucleolus, or the nuclear boundary ( Figure S3A). 945 The SPT data were then analyzed as above-drift-correction, followed by localization, building 946 of trajectories, and sorting into compartments-using the exact same parameters. We also 947 followed this same procedure of randomly choosing and placing artificial RCs in infected cells, 948 this time avoiding previously annotated RCs instead of nucleoli ( Figure S3B). 949

Two-state kinetic modeling using Spot-On 950
We employed the Matlab version of Spot-On (available at https://spoton.berkeley.edu) in our 951 analysis, and embedded this code into a custom-written Matlab routine. All data for a given 952 condition were merged, and histograms of displacements were generated for between 1 and 7 Dt. 953 These histograms were fitted to a two-state kinetic model which assumes one immobile 954 population and one freely diffusing population: