Abstract

   Recombinant adeno-associated viral vectors (rAAV) are a powerful tool
   for gene delivery but have a limited DNA carrying capacity. Efforts to
   expand this genetic payload have focused on engineering the vector
   components, such as dual trans-splicing vectors which double the
   delivery size by exploiting the natural concatenation of rAAV genomes
   in host nuclei. We hypothesized that inefficient dual vector
   transduction could be improved by modulating host factors which affect
   concatenation. Since factors mediating concatenation are not well
   defined, we performed a genome-wide screen to identify host cell
   regulators. We discover that Homologous Recombination (HR) is
   inhibitory to dual vector transduction. We demonstrate that depletion
   or inhibition of HR factors BRCA1 and Rad51 significantly increase
   reconstitution of a large split transgene by increasing both
   concatenation and expression from rAAVs. Our results define roles for
   DNA damage repair in rAAV transduction and highlight the potential for
   pharmacological intervention to increase genetic payload of rAAV
   vectors.

   Subject terms: Homologous recombination, Targeted gene repair
     __________________________________________________________________

   AAV vectors are a promising gene delivery vehicle but have a limited
   DNA packaging capacity. Here, the authors identify cellular restriction
   factors whose temporary inhibition significantly enhances transduction
   efficiency of oversized transgenes split into dual AAV vectors.

Introduction

   Recombinant adeno-associated viral vectors (rAAV) are a leading
   platform for clinical DNA delivery. This vector system is derived from
   a human parvovirus and has many inherent features that make it
   attractive for human gene therapy applications^[43]1. It also has some
   inherent limitations, the major one of which is the small genetic
   payload. The unenveloped 25 nm capsid can package no more than 5 kb of
   single-stranded DNA (ssDNA) into its small lumen^[44]2,[45]3, imposing
   a size limit to the coding and regulatory elements that can be included
   in the transgene cassette. For gene replacement therapies, current
   strategies employ cDNA versions of the gene of interest, with
   expression driven by a short, strong promoter such as CMV^[46]4. While
   this accommodates many therapeutically relevant gene sizes, it excludes
   other larger genes, such as dystrophin or the most accurate Cas9
   versions and other base editors. Moreover, CMV and similar promoters
   are ubiquitously expressed at very high levels, leading to off-target
   expression in non-therapeutically relevant tissues, and potentially
   toxic levels of transgene expression. This can lead to immune responses
   and other failures of the therapy. A larger payload size is therefore
   highly desirable for expressing large proteins and moreover for
   enabling fine-tuned spatiotemporal control over expression and levels
   by inclusion of more diverse, efficient regulatory sequences.

   The wild-type (wt) AAV virus and the recombinant AAV vector have the
   same overall structure, consisting only of the 60mer icosahedral capsid
   with one packaged ssDNA vector genome (VG). Viral proteins are required
   during production for replication and packaging, but no viral enzymes
   are packaged within the virion^[47]5. VG architecture requires flanking
   the desired delivery DNA with the viral inverted terminal repeat (ITR)
   sequences at both ends^[48]6. The ITR contains sequence elements that
   are required during viral replication and also direct packaging into
   the preformed capsids for vector production^[49]7. In the transduction
   process (Fig. [50]1A), the capsid traffics the VG to the host/target
   cell nucleus, where the capsid is shed (termed uncoating)^[51]8,[52]9.
   The ssDNA VG is converted to dsDNA through unclear mechanisms which
   enables transgene expression. Individual VGs can also circularize
   through intramolecular ITR-ITR joining and concatenate through
   intermolecular ITR-ITR joining^[53]10. In its recombinant form, where a
   capsid of choice is packaged with an ITR-flanked transgene, rAAV VGs
   cannot replicate; but instead multiple VGs enter the nucleus and can
   concatenate into large episomal species^[54]11. The absence of
   delivered enzymes implies that the post-uncoating nuclear steps of rAAV
   transduction are orchestrated by host cellular factors. Decades of
   observation have articulated the nuclear steps of
   transduction—uncoating, conversion to dsDNA, concatenation,
   expression—yet they remain observations, with little mechanistic detail
   and knowledge of facilitating host factors^[55]12,[56]13.

Fig. 1. Dual Vectors are inefficient.

   [57]Fig. 1
   [58]Open in a new tab

   A Schematic of recombinant adeno-associated viral (rAAV) vector genome
   (VG) processing. Single-stranded DNA VGs are converted to
   double-stranded (ds)DNA and VGs circularize and concatenate by inverted
   terminal repeat (ITR)-ITR joining, through unclear mechanisms. VGs
   associate with histones, but when this occurs during these processes is
   unknown. B The dual vector approach can double payload size by
   exploiting the natural concatenation process. A large gene is split
   into separate Left (L) and Right (R) VGs; inclusion of a splice donor
   (SD)/acceptor (SA) excludes the ITR junction from the final transcript.
   C Reporter constructs to assay split transgene reconstitution. LacZ is
   split between L and R VGs in a trans-splicing (TS) pair that requires
   ITR-ITR concatenation for expression, or a Hybrid (HB) trans-splicing
   plus Alkaline Phosphatase (AP) partial sequence previously demonstrated
   to increase split transgene reconstitution efficiency. An unsplit LacZ
   VG is used as a control. CMV = cytomegalovirus. BSD = Blasticidin
   resistance. pA = poly A signal. F2A, P2A, T2A = 2A self-cleaving
   peptides. D LacZ Activated Fluorescence Assay (LAFA) schematic:
   cultured cells are transduced with equimolar doses of L and R vectors.
   After 48 h, split transgene reconstitution is assayed by
   Beta-galactosidase (LacZ) activity cleaving
   4-Methylumbelliferyl-beta-D-galactopyranoside (MUG), a fluorogenic beta
   galactose analog. E U2-OS cells were treated with 1e4 VG/cell of the
   split pairs or unsplit LacZ diagrammed in (C). LAFA was performed at
   48 hours post transduction and quantified by flow cytometry: L + R
   pairs (magenta, purple), unsplit vector (blue, green), or empty capsid
   (black) negative control are plotted as a histogram. All capsids and
   ITRs are AAV2.

   The absence of viral enzymes in the rAAV particle implies that the
   processes of VG circularization and concatenation are orchestrated by
   host cellular factors. ITR-ITR joining is analogous to host DNA repair
   mechanisms, and dsDNA VGs could conceivably resemble a chromosomal
   break. The host cell has many ways of repairing DNA damage, including
   two main pathways to repair double-strand breaks (DSB): the
   high-fidelity but restricted activity pathway called Homologous
   Recombination (HR) or Homology Directed Repair (HDR), and the
   constitutively active but error-prone pathway of Non-Homologous End
   Joining (NHEJ)^[59]14. The MRE11/RAD50/NBS1 (MRN) complex, which
   recognizes free DNA ends to facilitate repair upstream of HDR versus
   NHEJ pathway choice, recognizes ITRs but inhibits
   concatenation^[60]15,[61]16. The NHEJ pathway has been suggested to
   play a role in concatenation since depletion or inhibition of the NHEJ
   factor DNA-PKcs reduces VG concatenation^[62]17–[63]21. However, the
   effect of HDR pathway on concatenation has not been directly examined.

   The propensity of VGs to concatenate can be exploited to increase the
   final genetic payload size^[64]22. In this “dual vector” approach
   (Fig. [65]1A, B), a large transgene cassette is split into two halves,
   packaged into separate capsids, and co-delivered to target cells.
   Inclusion of a splice donor in the “Left” half VG (L) and a splice
   acceptor in the “Right” half VG (R) allows the ITR junction to be
   spliced out during RNA processing of transcription, and the large gene
   is subsequently reconstituted in the transcript. This technique was
   first developed in the early 2000s by multiple groups
   independently^[66]23–[67]25. To be effective it requires correct,
   directional, and stoichiometric concatenation, followed by correct
   expression. By nature of the sequence, any ITR can in theory join any
   ITR, leading to many permutations that do not all lead to correct split
   transgene reconstitution. The approach also currently relies on very
   high vector doses. To increase the chances of L and R genomes joining
   in the correct orientation, a short fragment of the Alkaline
   Phosphatase (AP) gene serendipitously found to be “highly
   recombinogenic” in the dual vector context^[68]26 can be added between
   the splice signal and the ITR^[69]27. It is presumed that the AP
   fragment promotes annealing of + and – sense VGs which then become a
   reconstituted large gene through HR. However, HR has never been
   directly tested as the mechanism of concatenation. Indeed, AP inclusion
   significantly increases efficiency and is now commonly used in dual
   vectors, but like general mechanisms of concatenation, little is known
   about the mechanisms underpinning the AP phenomenon.

   Here, we aim to improve dual vector transduction by an alternative
   approach that focuses on the host cell. We first identify cellular
   pathways inhibitory to split transgene reconstitution by a genome-wide
   screen. We provide evidence for early chromatinization of VGs, and
   epigenetic recognition as a double-strand break. Surprisingly, blocking
   HR increases both concatenation of and expression from rAAV VGs, which
   dramatically increases dual vector transduction efficiencies. Our
   findings suggest that HR is the “first responder” to the DSB-flagged
   VGs, but that NHEJ is a more efficient mechanism for VG concatenation.
   While the majority of efforts to improve rAAV gene therapies have
   focused on modifying the vector and dose, our findings emphasize the
   importance of understanding host responses in the target cell. Our
   results highlight the potential for pharmacological approaches to
   enhance efficiency and increase deliverable transgene size in rAAV gene
   therapies.

Results

A genome-wide screen to identify cellular inhibitors of split transgene
reconstitution

   To identify cellular factors whose modulation could increase dual
   vector transduction, we established a platform amenable to pooled
   genome-wide screens by building upon dual vector systems previously
   engineered by others in which vectors enable trans-splicing (TS) or
   contain the additional AP segments in Hybrid (HB)
   vectors^[70]23–[71]25,[72]27–[73]29 (Fig. [74]1C). Correct directional
   concatenation and successful expression will yield a multi-cistronic
   cassette containing several reporter genes. The LacZ gene is split to
   span the ITR junction between Left (L) and right (R) vectors and is
   therefore the most stringent reporter of concatenation and subsequent
   expression. After transducing cultured cells, incubation with the
   fluorogenic Β-galactose analog
   4-Methylumbelliferyl-beta-D-galactopyranoside (MUG) provides an
   enzymatic readout of successful split transgene reconstitution. This
   LacZ activated fluorescence assay (LAFA) can be performed on cell
   lysates in microwell plates or in flow cytometry based approaches and
   enables FACS isolation of transduced cells (Fig. [75]1D). We transduced
   U2-OS cells with unsplit LacZ as a positive control, with either the TS
   or HB pair of dual vectors, or an empty capsid as a negative control,
   and used LAFA and flow cytometry to assay expression of LacZ. Flow
   cytometry of treated cells (Fig. [76]1E) is consistent with previous
   observations by others^[77]29: although HB dual vectors (purple)
   achieve higher transduction than TS (magenta) at a moderate dose of 1e4
   total vg/cell (5e3 VG/cell of L vector plus 5e3 VG/cell of R vector),
   both the TS and HB dual vectors transduced at efficiencies two orders
   of magnitude lower than cells receiving the same dose of an unsplit
   LacZ vector (green). The inefficiency of dual vector transduction is
   further demonstrated by comparing 1e4 total VG/cell dual vectors
   (purple) to 1e3 VG/cell unsplit LacZ (blue); a 10-fold lower dose of
   unsplit transgene achieves 10-fold higher transduction than dual
   vectors. To ensure that LacZ activity measured after dual vector
   transduction is from VG-VG recombination and not transcripts
   recombining, we transfected plasmids containing the split vector
   genomes and unsplit controls (Supplementary Figs. [78]1 A–F), which do
   not have free ITRs and thus don’t concatenate as capsid-delivered VGs
   do, but actively express from their promoters. We do not observe LacZ
   reconstitution in this setting (Supplementary Fig. [79]1G), suggesting
   that concatenation must occur at the DNA level for a spliced transcript
   to reconstitute a split gene.

   To identify opportunities for host factor modulation that could lead to
   increases in split transgene reconstitution, we reasoned that (1) most
   pharmacological intervention presents inhibitors rather than
   activators, and (2) as a virally derived entity, it is likely that rAAV
   concatenation and expression are targets of negative regulation as part
   of an innate defense. We therefore designed a genome-wide knockout (KO)
   screen to identify repressors of dual vector transduction. Lentiviral
   Brunello libraries carrying a puromycin resistance cassette, Cas9, and
   an average of four guides per human gene were used to generate
   genome-wide KO libraries of U2-OS cells. After Puro selection, the
   cellular KO libraries were co-transduced with 5e3 VG/cell of each of
   the HB-L and HB-R vectors. Cells were collected at 48 hours
   post-transduction (hpt), stained with MUG, and LacZ expressing cells
   isolated by FACS. Genetic perturbations in these cells were identified
   by Next Generation Sequencing (NGS), and statistical analysis on guide
   overrepresentation was performed to generate a list of candidate
   inhibitors of dual vector transduction (Fig. [80]2A). Pathway analysis
   of the top 5% of significantly enriched perturbations in the
   MUG-positive pool (Supplementary Data [81]1) revealed
   overrepresentation in DNA synthesis and repair, particularly
   double-strand break (DSB) repair. Of the two major DSB repair pathways,
   Homology Directed Repair (HDR) and Non-Homologous End Joining (NHEJ),
   only HDR was overrepresented (Fig. [82]2B). Since the NHEJ pathway has
   been suggested to positively regulate concatenation, we hypothesized
   that the HDR DSB repair pathway could function as a possible negative
   regulator of VG concatenation. We explored this in our subsequent hit
   validation.

Fig. 2. A genome-wide screen to identify cellular inhibitors of split
transgene reconstitution.

   [83]Fig. 2
   [84]Open in a new tab

   A Screen schematic: a Brunello knock-out Library of U2-OS cells (500x
   coverage) was treated with 5e3 VG/cell of the HB dual vectors. Cells
   were collected at 48 hours post transduction, stained with
   4-Methylumbelliferyl-beta-D-galactopyranoside (MUG), and MUG negative
   (NEG) and positive (POS) cells separated by FACS. B Reactome pathway
   analysis results on the top 5% of genes in the positive population
   (candidate negative regulators of split transgene reconstitution),
   sorted by significance. Bubble size indicates the number of genes in
   each Reactome category. HDR = homology directed repair. C
   Guide-enrichment score (sgRNAs enriched in the MUG-positive population)
   sorted in ascending order and plotted as a waterfall plot. Rad51 and
   BRCA1 are highlighted. D -log10(p-value) of guide enrichment scores
   sorted and plotted in ascending order. Rad51 and BRCA1 are highlighted.
   See Methods for analysis details and plotting code.

Vector genomes are epigenetically marked as double-strand breaks

   Since chromosomal DSBs are marked by phosphorylation of the histone
   variant H2AX (γH2AX) which coordinates recruitment of repair factors,
   we examined whether VGs are subject to this epigenetic mark. Previous
   studies have observed an increase in nuclear γH2AX when cells are
   infected with wt AAV^[85]30–[86]32, but this epigenetic mark has not
   yet been examined on AAV viral or vector genomes themselves. To detect
   individual VGs, we adapted a previously established
   approach^[87]15,[88]33–[89]36 in which the rAAV genome is visually
   detected through recognition of lacO binding sites by a fluorescent
   LacI fusion protein. We generated U2-OS cells stably expressing the
   LacI-mNeonGreen fusion protein; untransduced cells have no LacI-mNG
   foci (Supplementary Fig. [90]2A) and normal γH2AX immunostaining
   patterns, with foci distributed throughout nuclei (Supplementary
   Fig. [91]2B). Etoposide was used as a control for γH2AX staining under
   DNA damage conditions. We transduced these cells with rAAV vectors
   containing VGs consisting of 64 LacO repeats. After uncoating and
   conversion to a dsDNA species, the VG recruits many fluorophores and
   can be visualized in live or fixed cells (Fig. [92]3A). We observed VG
   foci distributed throughout the nucleus, fusing with other foci, and
   increasing in size and intensity over time, indicating concatenation
   events (Supplementary Movies [93]1-[94]6). We first asked whether VG
   foci are marked as DSBs by immunostaining for γH2AX and observed
   colocalization as early as 4 hpt (Fig. [95]3B). High-content imaging
   analysis on 1139 VG foci at 4 hpt revealed 64.97% colocalization with
   γH2AX foci. To verify γH2AX association with VGs, and nucleosome
   incorporation in general, we performed ChIP-qPCR for γH2AX and H2B,
   respectively (Fig. [96]3C). At the earliest timepoint for which ChIP
   signal was detectable above background (12 hpt), both γH2AX and H2B
   precipitated VGs, corroborating our imaging observations. To date, this
   is the earliest observed histone association with rAAV-delivered DNA
   and suggests that VGs are both chromatinized and flagged as DSBs
   quickly after uncoating.

Fig. 3. rAAV vector genomes are recognized as double strand breaks.

   [97]Fig. 3
   [98]Open in a new tab

   A Vector genome (VG) visualization schematic. Capsids are packaged with
   array VGs comprised of 64 lacO repeats. After uncoating and becoming
   double-stranded (ds) in LacI-mNeonGreen expressing cells, VGs can be
   visualized as green foci. U2-OS^LacI-mNeonGreen cells were transduced
   with 1e5 VG/cell of AAV2/2.LacO.64 in all panels. B Cells were fixed at
   4 hours post transduction (hpt) and immunostained for γH2AX. Images are
   representative of biological triplicate experiments. C Cells were
   collected at 12 or 40 hpt and chromatin immunoprecipitation (ChIP) was
   performed with antibodies on x-axis in biological triplicate with
   individual replicates plotted as open circles. qPCR to quantify VGs is
   reported as a percentage of input material; bars represent mean and
   error bars = SD. IgG versus histone conditions compared by unpaired,
   two-tailed t-test, ** p < 0.01; **** p < 0.0001 (D) Cells were fixed at
   4 hpt and immunostained for BRCA1 and 53BP1, then (E) foci overlap
   quantified in 957 nuclei. Images and quantification are representative
   of biological duplicate experiment. Source data for plots are provided
   as a Source Data file.

   We next used immunostaining to ask whether VGs recruit DSB repair
   factors. We used antibodies to BRCA1 as a representative HR factor
   which was enriched in our screen, and antibodies to 53BP1 as a
   representative NHEJ factor (Fig. [99]3D). Colocalization of both repair
   factors were observed with VG foci, with BRCA1 more often than 53BP1.
   We quantified overlap of these foci in 957 nuclei (Fig. [100]3E); BRCA1
   colocalizes with 72.4% of LacI mNG foci, whereas 53BP1 is 43.2%. 36.9%
   of all VG foci colocalize with BRCA1 but not with 53BP1, whereas 53BP1
   colocalizes without BRCA1 on only 7.4% of VGs. This suggests that DSB
   repair factors are recruited to most newly uncoated VGs, and that HDR
   machinery may be preferentially recruited over NHEJ. These
   anti-correlated staining patterns are in line with our opposing screen
   results for NHEJ and HDR, and the known competition between these DNA
   repair pathways on DSBs in the cellular genome^[101]14.

HDR deficiency increases concatenation and expression from rAAV vectors

   Successful trans-splicing dual vector transduction requires two major
   steps: concatenation of the L and R co-infected VGs followed by
   successful expression of the split cassette. To examine whether our
   candidate factors affect concatenation or expression, we added an
   mScarlet expression cassette downstream of the lacO array
   (lacO.64.CMV.mScarlet; Fig. [102]4A). This enables simultaneous and
   quantitative visual readouts of both concatenation (mNeonGreen foci)
   and expression (mScarlet intensity) which are not interdependent, such
   as with dual vector transduction and LacZ reconstitution. To test
   effects of HDR loss on concatenation and expression, we introduced
   BRCA1 or scrambled guides by nucleofection into U2-OS^LacI-mNG,Cas9
   cells, waited 72 h for editing, and then transduced with
   rAAV2/2.lacO.64.CMV.mScarlet vectors. Immunostaining for BRCA1 showed a
   significant decrease in cells receiving BRCA1 guides compared to
   scrambled guides (Supplementary Fig. [103]3A). We observed a 2.3-fold
   increase in the average number of VG foci per cell (Supplementary
   Fig. [104]3B) and a significant increase in the percentage of cells
   with any VG foci in cells receiving BRCA1 guides compared to Scramble
   (Supplementary Fig. [105]3C). We next quantified expression by mScarlet
   intensity in the high-content images. Expression of the mScarlet
   transgene is 10% higher in cells receiving BRCA1 guides, a small but
   statistically significant increase (Supplementary Fig. [106]3D).
   Together these observations suggest that BRCA1 inhibits concatenation
   and de novo episome formation, with a small impact on expression.

Fig. 4. HDR deficiency increases concatenation and expression from rAAVs.

   [107]Fig. 4
   [108]Open in a new tab

   A VG schematic. B–D U2-OS^LacI-mNeonGreen cells were transfected with
   siRNA against RAD51 or BRCA1, or nontargeting (NEG) control (x-axis),
   then transduced 26 h later with 1e5 VG/cell AAV2/2.lacO.CMV.mScarlet.
   High-content imaging was performed at 48 hours post transduction (hpt),
   and quantitative analysis used to define (B) individual cells plotted
   as points by number of LacI foci per nucleus, and (C) the percentage of
   all imaged cells with any LacI foci. D Images were segmented by cell
   and transgene expression levels assessed by average mScarlet pixel
   intensity, then plotted as individual points. B, C, & D are plotting
   590 cells per condition, and is representative of biological triplicate
   experiments. E Western blot to examine protein levels in cellular
   lysates 6–36 h after siRNA or drug treatment. Blot is representative of
   biological triplicate experiments. (F–I) U2-OS^LacI-mNeonGreen cells
   were pretreated with drug or dimethyl sulfoxide (DMSO) (legend),
   transduced with AAV2/2.lacO.CMV.mScarlet at dose on x-axis, then
   high-content imaged at 48 hpt. F Primary image example. G Individual
   cells are plotted as points by number of foci per nucleus. H Percentage
   of all cells in (G) with any LacI foci. I Individual cells are plotted
   as points according to mScarlet intensity. G, H, & I are plotting 1757
   cells per condition, and is representative of biological triplicate
   experiments. J Schematic of lacO.MS2 system to quantify
   transcriptionally active VGs. K U2-OS^LacI-mNeonGreen, MCP-mCherry
   cells were treated with 1e5 VG/cell AAV2/2.lacO.CMV.MS2 and imaged
   at 48 hpt on an LSM900 Airyscan2. Arrowheads indicate multiple actively
   transcribing VGs in a single episome. L U2-OS^LacI-mNeonGreen,
   MCP-mCherry cells were treated with 1e5 VG/cell AAV2/2.lacO.CMV.MS2 and
   high-content imaged at 48 hpt. The percentage of LacI foci with at
   least one overlapping MCP focus in triplicate wells and in biological
   duplicate are plotted as individual points; each point represents a
   minimum of 902 imaged nuclei. Error bars = SD. Statistical significance
   was determined by Kruskal-Wallis Test with Dunn’s multiple comparisons
   test in (B, D, G, & I), unpaired, two-tailed T tests in (L) with
   p = 0.0012, and by two-sided Fisher’s exact test in (C & H), with
   *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001. Numbers above
   graphs indicate fold change over control; black lines = mean values.
   Source data for plots are provided as a Source Data file. MS2 =
   Bacteriophage MS2. MCP = MS2 coat protein. mNG mNeonGreen.

   To orthogonally validate this result, we next depleted BRCA1 with siRNA
   and observed nearly complete knockdown by 26 h after transfection
   (Fig. [109]4E). We transduced BRCA1 knockdown cells with 1e5 VG/cell of
   rAAV2/2.lacO.64.CMV.mScarlet at this timepoint and observed significant
   increases in LacI foci per nucleus (Fig. [110]4B), percentage of cells
   with any VG foci (Fig. [111]4C), and mScarlet expression
   (Fig. [112]4D). This exceeded the increases observed in the CRISPR/Cas9
   acute KO. In parallel, to interrogate whether the increased foci
   formation is specific to BRCA1 loss or is an effect of non-functional
   HDR, we tested the effects of Rad51 siRNA knockdown, another HDR factor
   enriched in our screen. At 26 h, the knockdown was not complete, but
   Rad51 levels did not deplete further upon longer incubation
   (Fig. [113]4E). Cells were therefore transduced at 26 h post-siRNA
   transfection (Fig. [114]4B–D) and we observed similar increases to
   those observed with BRCA1 loss across all parameters. As an orthogonal
   method of examining Rad51 loss in transgene expression and
   concatenation and to test for dose-dependent effects, we inhibited
   Rad51 with the drug B02 by pretreating U2-OS^LacI-mNG cells with
   multiple doses of drug, then transduced with multiple doses of
   rAAV2/2.lacO.64.CMV.mScarlet. We observed that B02 increased the number
   of VG foci per cell (Fig. [115]4F, G) and the percentage of cells
   forming any VG foci (Fig. [116]4H), in a dose-dependent fashion. In
   live imaging of VG foci, we observe more fusion events in B02 treated
   cells (Supplementary Movies [117]1-[118]3) than DMSO treated cells
   (Supplementary Movies [119]4-[120]6). Additionally, high-dose Rad51
   inhibitor resulted in approximately doubled mScarlet expression,
   whether vector was at moderate (1e4 VG/cell) or high doses (1e5
   VG/cell) (Fig. [121]4I). B02 does not affect Rad51 protein levels
   (Fig. [122]4E), suggesting that it is the ATPase activity of Rad51
   contributing to the inhibitory effects on rAAV. These results suggest
   that, similar to BRCA1, Rad51 inhibits concatenation, de novo episome
   formation, and transgene expression. Together these results implicate
   repressive roles for HDR in rAAV transduction.

   To examine whether the observed increases in VG foci and transgene
   expression were due to increased total number of VGs present in cells,
   for example due to increased nuclear import or increased dsDNA
   conversion leading to increased stability/decreased degradation in HDR
   deficient cells rather than increased concatenation or expression
   activity, we compared VG copy number in DMSO and B02 treated cells 48
   hpt. Cells treated with B02 have no significant difference in VG copy
   number compared to cells treated with DMSO and receiving the same
   vector dose (Supplementary Fig. [123]3E). Taken together, these
   observations confirm that HDR inhibits both concatenation and
   expression of rAAV VGs.

   Because DDR inhibition can affect the cell cycle, we examined these
   effects by treating cells with drug or siRNA and examining DNA content
   48 h later by flow cytometry (Supplementary Fig. [124]4). BRCA1 siRNA
   does not significantly affect the proportion of cells in G1, S, or G2/M
   compared to the negative control siRNA (maximum of 3.5% difference,
   Supplementary Figs. [125]4F, H), whereas RAD51 siRNA has a significant
   effect (maximum 9.9% difference, Supplementary Figs. [126]4F, G).
   Comparing B02 to DMSO conditions, however, demonstrates pharmacological
   Rad51 inhibition does not recapitulate the effect on cell cycle of
   siRNA (Supplementary Figs. [127]4A, B). Since we observe the strongest
   increases in VG foci formation under BRCA1 depletion and this
   perturbation has the smallest effect on cell cycle, it is unlikely that
   cell cycle differences induced by HDR loss are indirectly driving the
   observed transduction effects, although we have not completely ruled
   this out. Because HR factors are upregulated during S-phase, we also
   examined whether BRCA1 preferentially colocalized with VG in S-phase by
   EdU labeling and immunostaining (Supplementary Fig. [128]4 C-H).
   Comparing cells in S-phase to those outside of S-phase, there were no
   obvious differences in BRCA1/VG colocalization, as well as VG foci per
   cell and percentages of cells with any foci.

ATM and ATR inhibition increase transgene expression

   The histone variant H2AX can be phosphorylated by ATM upon DSB
   detection, or by ATR upon replication stress. To examine this, we
   pharmacologically inhibited these kinases (Supplementary Fig. [129]3
   F-J). We observe a decrease in γH2AX/VG colocalization under ATM
   inhibition, whereas ATR inhibition has no effect (Supplementary
   Fig. [130]3F). This suggests that ATM phosphorylates H2AX associated
   with VGs. It should be noted that ATM inhibition decreases global γH2AX
   staining and ATR inhibition does not (Supplementary Fig. [131]3G).
   Consistent with previous reports^[132]37,[133]38, compared to DMSO
   controls we observed an increase in transgene expression under ATM
   inhibition (Supplementary Fig. [134]3H). ATR is activated upon wt
   AAV2/Adenoviral coinfection^[135]31 and UV-inactivated AAV2
   monoinfection^[136]39, and ATR inhibition has been shown to decrease wt
   AAV DNA replication^[137]31, but ATR is not well-studied in the rAAV
   transduction setting. We observed a large increase in transgene
   expression under ATR inhibition (Supplementary Fig. [138]3H).
   Conversely, these drugs have little effect on the number of LacI foci
   per nucleus (Supplementary Fig. [139]3I), and ATR inhibition, but not
   ATM inhibition, significantly increases the percentage of cells with
   any LacI foci (Supplementary Fig. [140]3J). The large effects of ATR
   inhibition on transduction may not depend on γH2AX but may be due to
   proteomic changes that affect HR in these conditions^[141]40.

Rad51 inhibition increases the proportion of transcriptionally active vector
genomes

   We next asked whether the increase in expression we observe under Rad51
   inhibition is a result of increased transcriptional activity from each
   active VG or an increase in the proportion of VGs that are
   transcriptionally active. To observe transcriptional activity in situ
   with each VG, we replaced the mScarlet gene with an MS2 array which
   forms stem-loops when transcribed. In MCP-mCherry expressing cells the
   nascent transcript can be visualized as an mCherry focus overlapping
   with an mNeonGreen (VG) focus (Fig. [142]4J). Transducing
   U2-OS^LacI-mNG, MCP-mCherry cells with rAAV2/2.lacO.64.CMV.MS2 array
   vectors, we observed large VG foci often with multiple actively
   transcribing units within the concatemer (Fig. [143]4K). After
   pretreatment with B02 or DMSO, high-content imaging at 48 hpt revealed
   a significant increase in the percentage of VG foci with at least one
   overlapping MS2 focus in B02 treated cells (Fig. [144]4L). This
   suggests that Rad51 inhibition increases expression by increasing the
   proportion of transcriptionally active genomes.

   Given the inhibitory role we observed for HDR in concatenation and
   expression independently, we next tested the effects of B02 on dual
   vector transduction, which requires both processes. We pretreated cells
   with multiple doses of B02, transduced cells with multiple doses of the
   Trans-Splicing (TS) L + R split LacZ vector (Fig. [145]1C), and used
   LAFA (Fig. [146]1D) in a 96-well format to readout split transgene
   reconstitution at 48 hpt (Fig. [147]5A). At all dose combinations, B02
   significantly increased dual vector expression. The most dramatic
   effect was seen at the lowest vector doses, at which 10 mM B02
   increased split LacZ expression 27-fold over DMSO treated cells
   receiving the same vector dose. This B02-induced increase in expression
   at low vector doses is comparable to expression levels achieved by
   100-fold higher vector doses not treated with B02 (Fig. [148]5A, gray
   dashed line). This suggests that pharmacologically induced improvements
   in transduction will enable significantly lower doses of vector to be
   used for gene replacement approaches. These effects are not limited to
   osteosarcoma cells, as we observed a similar trend in trophoblast
   cells, albeit to a lower magnitude (Supplementary Fig. [149]5).

Fig. 5. Rad51 inhibition increases dual vector transduction.

   [150]Fig. 5
   [151]Open in a new tab

   The effects of (A–C) Rad51 or (D–F) DNA-PK inhibition on split
   transgene reconstitution were tested in U2-OS cells by LacZ
   activated fluorescence assay (LAFA) 48 hours post-transduction by the
   dual (or single) vector indicated above each plot. Bars represent the
   mean of 5 (A–C) or 4 (D–F) biological replicates; individual values are
   plotted as dots, bars represent SD. A 2-way ANOVA was used, with
   *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001; ****p ≤ 0.0001. G
   U2-OSlacI-mNeonGreen cells were pretreated with DMSO or B02, transduced
   with 1e5 VG/cell AAV2/2.lacO.64, and immunostained for Rad51, 53BP1, or
   BRCA1 as listed at top of panel using an Alexa-647 secondary antibody
   at 48 hpt. H LacI and 647 foci colocalization (center of LacI focus
   overlapping with 647 spot) was quantified by high content imaging in
   biological duplicate and reported as a percentage of total foci imaged
   with a minimum of 931 foci imaged per data point (bar represents
   average of individual plotted points). Source data for plots are
   provided as a Source Data file. DMSO = dimethyl sulfoxide.

The AP fragment increases concatenation by an HR-independent pathway

   While the TS dual vectors serve to model the natural concatenation
   process and best represent our high-content imaging findings, our
   original screen was performed with the more efficient hybrid (HB) dual
   vectors (Fig. [152]1C), for reasons of experimental practicality.
   Originally developed by Ghosh and colleagues^[153]27, this pair
   contains a “highly recombinogenic” overlapping sequence fragment from
   the alkaline phosphatase (AP) gene that bridges L and R vectors,
   providing an additional mode of reconstitution beyond the ITR-ITR
   concatenation in the traditional TS dual vectors (Fig. [154]1B & C).
   The mechanism through which AP increases efficiency of dual vector
   transduction is not well understood; it is suggested in the literature
   that the AP overlapping sequence drives split transgene reconstitution
   through homologous recombination^[155]26,[156]27, however HR or HDR
   have not yet been directly examined in this context. Surprisingly, B02
   had even more potent effects increasing HB dual vector transduction
   than TS (Fig. [157]5B), suggesting that the classical HR pathway is not
   the main mechanism through which hybrid dual vectors reconstitute a
   split transgene. Rad51 inhibition also increased unsplit LacZ vector
   transduction (Fig. [158]5C), but to a much lower degree than dual
   vectors; at 1e4 VG/cell, unsplit LacZ expression is increased 4-fold by
   high dose B02, whereas L + R vectors at the same dose are increased up
   to 34 fold with B02 (Fig. [159]5A–C). This further corroborates our
   findings that HDR loss increases expression from VGs, and suggests that
   the concomitant increase in concatenation synergistically contributes
   to a dramatic increase in dual vector transduction.

   The observation that HDR inhibits VG concatenation is counterintuitive,
   since concatenation could be considered as analogous to repairing a
   chromosomal DSB. It is not clear what VG structures are flagged as DSBs
   or other forms of damaged DNA, but possibilities include the blunt ends
   after ds-conversion^[160]41 (among other conformations, diagrammed in
   Fig. [161]6). HDR is a less active repair mechanism than its
   error-prone counterpart Non-Homologous End Joining (NHEJ), which
   coordinates DSB repair through factors distinct from HDR including
   53BP1 and the DNA-dependent protein kinase catalytic subunit
   (DNA-PKcs). Previous studies recovered fewer circularized VGs from
   transduced muscle in DNA-PKcs deficient mice than in
   controls^[162]18,[163]42, suggesting an important role for NHEJ in VG
   concatenation. We next tested how NHEJ affects split transgene
   reconstitution by pharmacologically inhibiting DNA-PK with the drug
   Nu7441 (Fig. [164]5D–F). We observed a significant reduction in dual
   vector (Fig. [165]5D, E) or single vector transduction (Fig. [166]5F)
   under DNA-PK inhibition. While Nu7441 has effects on cell cycle
   (Supplementary Fig. [167]4A), these effects are too small to be major
   drivers of the observed transduction effects. This suggests NHEJ
   positively regulates ITR-ITR joining.

Fig. 6. Model for VG processing.

   [168]Fig. 6
   [169]Open in a new tab

   After uncoating, vector genomes (VG) undergo double-strand conversion
   (dsVG) via two possible pathways that result in at least three inverted
   terminal repeat (ITR) conformations^[170]41. ITR-primed
   second-strand-synthesis results in a T-shaped hairpin at the priming
   end like ITR 1 and a blunt opposite end like ITR 2, whereas VG-VG
   anealing produces two blunt (ITR 2) ends, that can also form a double-T
   structure (ITR 3) if ITRs anneal intramolecularly rather than pairing
   to the complementary strand as in ITR 2. A rAAV vector genomes quickly
   associate with histones and are marked as double-strand breaks (DSB)
   via H2AX phosphorylation by the host cell. B Homology-directed repair
   (HDR) machinery is preferentially recruited, which may block successful
   recruitment of non-homologous end joining (NHEJ) factors. C Inhibition
   or loss of HDR machinery promotes increased expression and
   concatenation, that synergize to dramatically increase dual vector
   transduction.

Blocking HDR increases NHEJ-driven concatenation

   Considering the opposing effects on split transgene reconstitution we
   observed when inhibiting these two main pathways of DSB repair, we
   hypothesized that after uncoating, HDR factors are preferentially
   recruited to VGs but are less efficient at concatenation than NHEJ.
   This suggests that blocking HDR allows NHEJ-mediated concatenation to
   predominate and thus there is an increase in ITR-ITR joining. To test
   this hypothesis, U2-OS^LacI-mNG cells were pretreated with B02 or DMSO
   then transduced with rAAV2/2.lacO.64 array vectors, fixed at 48 hpt and
   immunostained for Rad51 (HDR) and 53BP1 (NHEJ). High-content imaging
   revealed Rad51 colocalization with nearly 100% of VG foci imaged in
   DMSO treated cells, and a dramatic loss of this colocalization in B02
   treated cells (Fig. [171]5G). Concomitantly, B02 treatment increased
   53BP1 colocalization from ~10% to >60% of VGs (Fig. [172]5G).

Discussion

   Effectively overcoming the packaging size constraint would resolve a
   major limitation for gene delivery with rAAV. The inefficiency of
   trans-splicing dual vectors has been partially solved by split inteins
   to circumvent the protein coding limitation, but these still employ
   strong ubiquitous promoters that offer little control over
   expression^[173]43. Elements such as tissue-specific promoters,
   inducible promoters, enhancer arrays, UTRs/miRNA binding cassettes and
   other strategies for fine-tuning expression levels and spatiotemporal
   control must be in cis with the ORF they are controlling. Leveraging
   these approaches to control gene expression therefore necessitates
   concatenation of multiple VGs. In addition to payload expansion,
   concatenation is presumed to be an important mechanism for persistence
   of rAAV delivered DNA and transgene expression^[174]44.

   Significantly advancing rAAVs as versatile gene delivery vectors will
   require a better understanding of the mechanistic underpinnings of
   rAAV/host interactions. By investigating mechanisms of concatenation
   and expression from the point of view of the host cell, here we
   identify druggable targets for increasing dual vector transduction.
   Taken together, our results propose a model for host-vector
   interactions and pathway decisions that govern concatenation and
   expression of dual vectors (Fig. [175]6). Shortly after uncoating,
   dsDNA VGs associate with histones and are flagged by the host cell
   (Fig. [176]6A). HDR factors such as Rad51 and BRCA1 are preferentially
   recruited to VGs, which blocks NHEJ factor recruitment, and HDR is the
   slower concatenation mechanism (Fig. [177]6B). Blocking HDR allows the
   more efficient NHEJ to concatenate VGs, resulting in an increase in
   split transgene reconstitution (Fig. [178]6C). Under transient Rad51
   inhibition, we achieve dual vector transduction levels comparable to
   100-fold higher vector doses without drug. Lowering vector doses is a
   key step toward ensuring safety of gene therapies. Although prolonged
   use of DDR inhibitors is dangerous, our results suggest that a single,
   transient dose of drug preceding vector administration is sufficient to
   increase transduction, which may be safer to the patient than the toxic
   responses seen with very high vector doses. Our results also suggest
   that diseases of deregulated HDR, such as BRCA deficient cancers, may
   potentially confer an advantage for rAAV clinical applications in some
   settings.

   It is widely believed that the AP fragment in HB dual vectors improves
   efficiency by promoting HR/HDR between VGs^[179]26,[180]27,[181]45,
   however this has not been formally tested in this context. Here, we
   present evidence that HR/HDR inhibits split transgene reconstitution,
   regardless of AP inclusion. This suggests that AP does not primarily
   function through HR/HDR. We observed similar effects on TS or HB dual
   vector transduction when we inhibit HDR or NHEJ (Fig. [182]5),
   suggesting that these DSB repair pathways function similarly in ITR-ITR
   joining or AP overlapping. Upstream of both HDR and NHEJ, the MRN
   complex recognizes DSB free ends, and has been implicated as inhibitory
   to wt AAV infection/replication and to rAAV transduction, including
   rAAV VG foci formation^[183]15–[184]17,[185]35. The MRN complex may
   function in multiple ways in response to rAAV VGs, as it does in the
   response to cellular DSBs.

   The vector tools introduced in this study enable us to investigate
   mechanisms of transduction at individual VG resolution and therefore
   open new avenues for future studies. Additional research is needed to
   clarify exactly how Rad51 and HDR impede concatenation of VGs. The
   canonical role for Rad51 in HR/HDR is to coat ssDNA after resection at
   the DSB end and promote homology search and strand invasion^[186]46.
   One possibility is that the majority of ITRs are in the “T”
   conformation and are technically dsDNA, and not coated by Rad51. Such
   ITRs cannot perform efficient homology searches for neighboring ITRs
   with which they can concatenate. Perhaps this dead-end Rad51 binding
   impedes NHEJ factor recruitment and by extension concatenation through
   non-HDR pathways. It is also unclear specifically how Rad51 inhibition
   increases the proportion of transcriptionally active VGs. Highly
   transcribed genes are known sites of genomic instability, and how the
   cell balances expression and repair is an actively researched
   topic^[187]47,[188]48. One possibility is that Rad51 is involved in the
   cellular response to R-loops formed at the strong CMV promoter in our
   LacO.MS2 array reporter VGs. This rAAV vector system therefore presents
   a tool for studying fundamental mechanisms of gene regulation and DNA
   repair.

   The LacI-mNeonGreen expressing cell lines have excellent signal to
   background ratio at VG foci due to low level expression of LacI-mNG.
   While this enables detection of very small foci early in transduction
   and robust image analysis at all timepoints, it unfortunately limits
   our ability to quantify terminal sizes of episomes reliably by imaging.
   The pool of free LacI-mNG in the nucleus becomes depleted as
   transduction progresses, and it is unclear which, if any, VG arrays are
   saturated; only saturated VGs can be used to quantify size. Future
   studies may incorporate clonal lines with mid-range expression levels
   to enable size quantification under DDR or other pathway perturbations.

   Chromatinization and epigenetic regulation are just beginning to be
   appreciated by the field as major determinants of gene therapy
   outcomes^[189]49,[190]50. Epigenetic regulation is a well-known
   orchestrator of mammalian gene expression or silencing, as well as
   genome maintenance and repair, but has been only minimally examined in
   the context of AAV viral and vector genomes contexts.
   Active^[191]51,[192]52 and repressive^[193]52–[194]54 histone marks
   have been observed on VGs at late timepoints in transduced cultured
   cells and in mouse and NHP models. In NHPs, VGs extracted years post
   injection exist episomally as monomers and high molecular weight
   concatemers with nucleosomal structure^[195]55. It is unclear which VGs
   are expressing versus silenced, but moreover, how and when epigenetic
   control is exerted on VGs remains understudied. Here, we provide
   evidence for VG chromatinization very early in transduction in cell
   culture, and suggest a previously underappreciated role for epigenetic
   regulation in coordinating concatenation through DSB repair pathways. A
   limitation of our approach is that it cannot distinguish whether VGs
   are recruited to host genomic sites of active DNA damage repair and
   then hijack it, or whether VGs independently recruit these factors. The
   former has been suggested as the mechanism for wt AAV viral
   infections^[196]32, but the viral Rep proteins implicated in these
   studies are absent in the recombinant AAV setting. In wt AAV
   infections, viral DNA is observed to localize to sites of host damage,
   where it may capitalize on active repair machinery, however, this
   behavior is not seen in rAAV VGs^[197]32. Another limitation is that we
   cannot determine whether VGs enter and are immediately active or are
   repressed as the default state, which would require derepression and
   priming for activation. This is an interesting topic of future study.

   rAAV VGs remain predominantly episomal, but integrations can occur at
   low levels and are of concern for clinical applications. While the
   present study does not examine integration of VGs, the effects of
   HDR/NHEJ manipulation on integration are an exciting avenue for future
   investigation, particularly in the context of rAAV-delivered HDR donor
   templates for CRISPR-mediated editing or knock-in strategies. The tools
   introduced in this study will be useful for studying vector-host
   interactions driving many aspects of rAAV biology as well as
   fundamental cellular and molecular biology.

Methods

Plasmids

   Trans rep-cap pAAV2/2 (#104963), cis pAAV.CMV.Luc.IRES.EGFP.SV40
   (#105533) and Helper plasmids pAdDeltaF6 (#112867) were obtained from
   Addgene. ITR2.LacO.64 array cis plasmids were cloned from a pENTR 16.4
   LacOR Tjian Lab vector (array of 16 lacO sites concatenated 4 times)
   into the ITR2-containing backbone from pAAV.CMV.EGFP.Luc. To create
   ITR2.LacO.64.CMV.mScarlet, The CMV promoter and mScarlet cassette were
   PCR amplified adding a peroxisome targeting sequence (SKL) to the
   C-terminus; the amplicon was ligated into RE digested ITR2.lacO.64.
   ITR2.LacO.64.CMV.MS2 was generated from RE digest to remove mScarlet
   from ITR2.LacO.64.CMV.mScarlet and RE digest of pMBSV5 (Singer Lab), to
   yield a fragment containing 24 repeats of the MS2 stem-loop sequence,
   which was ligated into the ITR2.LacO.64.CMV backbone. All plasmid
   stocks were propagated in StbL2 competent cells, with some intermediate
   cloning steps carried out in DH5alpha.

Vector production and purification

   Vectors were produced by triple-plasmid transfection in adherent
   HEK293T cells. For large-scale production and purification, at least
   ten 15 cm dishes of HEK293T were transfected with the appropriate
   plasmids and cells + media were collected 72 h later and subjected to 3
   freeze/thaw cycles. Lysate was clarified by centrifugation at 15,000 x
   g, Benzonase treated for 30 min at 37 C, then lysate was 22 μ filtered
   and purified by AVB Sepharose Hi-Trap prepacked columns following the
   manufacturer’s instructions on an AKTA pure FPLC system. All titers
   were assayed by TaqMan qPCR using primers and probes within the
   transgene cassette, or in the case of LacO.64 which consists only of
   the repeat array, the ITRs.

Vector genome quantification in transduced cells

   Transduced cells were trypsinized and pelleted, then nuclei were
   isolated by resuspension in 15 mM Tris HCl, pH 7.5, 300 mM sucrose,
   60 mM KCl, 15 mM NaCl, 3 mM MgCl2, 0.5 mM DTT, 1x cOmplete mini
   protease inhibitor cocktail, and 0.5% NP-40. Cells were allowed to
   swell on ice, then dounced 10 times with a tight pestle. Nuclei were
   pelleted at 500 g for 5 min, washed once with the same buffer, then
   resuspended in PBS + 0.3% SDS and 1 mM EDTA. Lysed Nuclei were
   incubated with 200 ug RNase A at 37 C for 1 h, then with 40 ug
   proteinase K at 55 C overnight. Total DNA was phenol-chloroform
   extracted, and ethanol precipitated, then resuspended in an appropriate
   volume of 10 mM Tris pH 8. VGs were quantified in equal quantities of
   total DNA per sample by TaqMan qPCR with primers and probes against the
   CMV promoter within the transgene cassette and a standard curve of
   linearized plasmid containing this VG.

Chromatin Immunoprecipitation

   ChIP was performed as described with few modifications^[198]56. U2-OS
   cells were transduced with 1e5 VG/cell AAV2/2.Lac0.64.CMV.mScarlet or
   Empty vector. 10-12 or 40-48 hours after transduction cells were
   cross-linked 6’ at room temperature with 1% formaldehyde-containing
   FBS-free medium; cross-linking was stopped by adding PBS-glycine
   (0.125 M final). Cells were washed twice with ice-cold PBS, scraped,
   centrifuged for 10’ at 1000 x g and pellets were flash-frozen. Cell
   pellets were thawed and resuspended in cell lysis buffer (5 mM PIPES,
   pH 8.0, 85 mM KCl, and 0.5% NP-40, 1 ml/15-cm plate) w/ protease
   inhibitors and incubated for 10’ on ice. Lysates were centrifuged for
   10’ at 500 x g and nuclear pellets resuspended in 6 volumes of
   sonication buffer (50 mM Tris-HCl, pH 8.1, 10 mM EDTA, 0.1% SDS) w/
   protease inhibitors, incubated on ice for 10’, and sonicated to obtain
   DNA fragments around 500 bp in length (Covaris S220 sonicator, 20% Duty
   factor, 200 cycles/burst, 150 peak incident power, 10-12 cycles 30” on
   and 30” off). Sonicated lysates were cleared by centrifugation at
   10,000 x g and chromatin (75-250 μg per antibody) was diluted in RIPA
   buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 1% Triton
   X-100, 0.1% SDS, 0.1% Na-deoxycholate, 140 mM NaCl) w/ protease
   inhibitors to a final concentration of 0.8 μg/μl, precleared with
   Protein G dynabeads (Invitrogen) for 2 hours at 4 °C and
   immunoprecipitated overnight with 0.75-2.5 μg of specific antibodies.
   About 4% of the precleared chromatin was saved as input.
   Immunoprecipitated DNA was purified with the Qiagen QIAquick PCR
   Purification Kit, eluted in water and analyzed by qPCR together with
   2.5% of the input chromatin (SYBR Select Master Mix for CFX,
   ThermoFisher Scientific). ChIP-qPCR primer sequences for rAAV or human
   gDNA controls were as follows:

   hGenome_Chr17_for: GCTTTTCGCGAGGACAATTC

   hGenome_Chr17_rev: GTGCTTGAACCACCACTTCA

   AM_qPCR_CMV_F: CATCTACGTATTAGTCATCGCTATTACCA

   AM_qPCR_CMV_R: GAAATCCCCGTGAGTCAAACC

   Enrichment of specific antibodies was calculated as ΔCt over input
   signal.

Cell lines and KO experiments

   HEK293T cells used for vector production were obtained from the UC
   Berkeley BDS Cell Culture Facility (original source ATCC CRL-3216). To
   generate U2-OS^LacI-mNeonGreen, U2-OS cells (UC Berkeley BDS Cell
   Culture Facility cat# 608) cultured in DMEM (1 g/L glucose, 10% FBS)
   were transfected with a modified PiggyBac Transposase plasmid (System
   Biosciences) and a PGK.mNeonGreen-LacI.NLS.IRES.Puro integration
   plasmid using Fugene6 transfection reagent. After 48 h, cells were
   subjected to Puromycin selection (1µg/mL) for 96 h. Survivors were
   expanded for an additional 96 h at maintenance (0.2µg/mL) Puromycin
   concentration. Individual cells were then plated in 96 well plates
   using a FACS Aria. After 18 days, individual colonies were expanded to
   larger wells and to replicate CellCarrier Ultra Plates (Perkin Elmer)
   for initial screening on an OperaPhenix high-content imager. Live
   colonies were imaged with a 488 filter set, and those expressing
   mNeonGreen within an ideal threshold (nuclear signal between 200-2000)
   were passaged to replicate Falcon and CellCarrier Ultra plates for
   additional screening (48 clones total). Chosen colonies were then
   transduced with 1e9 vg/well rAAV2/2.lacO.64 crude vector preparation
   and live imaged at 48 hpt. The three colonies with best signal to
   background ratio at LacI-mNG foci were expanded and used for further
   experiments. To generate U2-OS^LacI-mNeonGreen.MCP-mCherry cells, one
   clonal line was transfected with PiggyBac Transposase and a
   UBC.tandemMCP2-mCherry plasmid with a neomycin resistance marker
   integration plasmid. Selection under 1 mg/mL G418 was carried out as
   above, and clonal expansion and screening performed as above, imaged
   with mCherry filter sets after transduction with rAAV2/2.
   lacO.64.CMV.MS2.24. To generate Cas9-expressing lines,
   U2-OS^LacI-mNeonGreen or U2-OS^LacI-mNeonGreen.MCP-mCherry cells were
   infected with a Cs9/Blasticidin Lentivirus, then cells were put under
   selection (10ug/mL Blasticidin) for one week, and plated as individual
   cells in 96well plates. Expanded clones were validated for Cas9
   presence and activity, and highly active clonal lines chosen and
   expanded. For BRCA1 KO experiments, U2-OS^LacI-mNeonGreen::Cas9 cells
   were transfected in an Amaxa Nucleofector to deliver BRCA1 or Scrambled
   sgRNA expressing plasmids also carrying a Zeocin resistance cassette.
   Cells were then selected for 72 h under 1 mg/mL Zeocin, then transduced
   with rAAV, fixed at 48 hpt, and stained and imaged as described below.
   The guide sequences are as follows (BRCA1 guide plasmids were pooled at
   an equimolar ratio):

   Scramble: GCACTACCAGAGCTAACTCA

   BRCA1:
    1. AGAAACCTACAACTCATGGA
    2. AGGAAACATGTAATGATAGG
    3. GTTTCTATCATCCAAAGTAT

LacZ-activated fluorescence assay (LAFA)

   U2OS or BeWo cells were seeded in DMEM (1 g/L glucose and 10% FBS) at a
   density of 2,500 cells/well in a 96-well plate. The following day,
   cells were pretreated with the pharmacological inhibitors B02 (5 M or
   10 M) or Nu7441 (1.75 M or 2.75 M) in DMEM (high glucose and 10% FBS)
   for 4-16 h. Spent media was replaced with DMEM containing the
   appropriate dose of unsplit or split LacZ vector, and B02 (5 M or
   10 M), Nu7441 (1.75 M or 2.75 M), or DMSO, and 5% FBS. Cells were
   transduced for a total of 48 hours without further media changes.
   Following transduction, cells were washed with 1x PBS and lysed in
   plate using 50uL cold lysis buffer (2 mM DTT, 25 mM Tris-phosphate (pH
   7.8), 2 mM 1,2-diaminocyclohexane-N,N,N´,N´-tetraacetic acid,
   1.25 mg/ml lysozyme, 2.5 mg/ml BSA, 10% glycerol, 1% Triton® X-100). To
   facilitate lysis, cells underwent two freeze/thaw cycles at −80 C and
   37 C, respectively. 4-Methylumbelliferyl-beta-D-galactopyranoside (MUG,
   10 mM in DMSO) was diluted 1:10 in LAFA buffer (100 mM Sodium Phosphate
   Buffer (pH 7.0), 1 mM MgCl2, 10 mM 2-mercaptoethanol, 0.1% Triton®
   X-100) to a final concentration of 1 mM. In a 96-well plate, 20uL of
   the cell lysate was added to 100uL of the MUG-LAFA buffer mixture,
   protected from light, and gently rotated at room temperature for
   90 minutes. Fluorescence signal was read at 365 nm/460 nm (Ex/Em) using
   a Tecan Infinite® M1000. To control for collecting LAFA signal from
   equivalent numbers of cells, ATP was quantified using a CellTiter-Glo®
   kit according to the manufacturer’s protocol, and this was used to
   normalize per well fluorescence.

siRNA knockdown experiments

   siGENOME SMARTPool siRNAs were purchased from Horizon Discovery for
   Human RAD51 (cat# M-003530-04-0010), Human BRCA1 (cat#
   M-003461-02-0010), and Non-targeting siRNA control pool #2 (cat#
   D-001206-14-20). For transfection into cells, Lipofectamine® RNAiMAX
   was used according to the manufacturer’s instructions. At appropriate
   time post-transfection, cells were either collected for western blot or
   flow cytometry, or transduced with rAAV in fresh media for imaging
   experiments.

Western blotting

   Cells in 6-well plates were drug or siRNA treated as indicated. Wells
   were washed with PBS, 250 μL RIPA buffer added, plate was freeze/thawed
   once at −80, wells were scraped on ice, and lysate collected and
   clarified at 15000 g for 10 min at 4c. Total protein was quantified by
   Pierce™ 660 nm Protein Assay kit according to mfr instructions. 12 ug
   total protein was diluted to 1x in SDS PAGE buffer and loaded in each
   well of a NuPAGE 4-12% Bis-Tris gel and electrophoresed proteins
   transferred to a nitrocellulose membrane. The membrane was blocked in
   PBS + 0.1% Tween 20 (PBST) + 5% nonfat dry milk for 1 h at RT, then
   primary antibody was diluted in PBST + 0.5% milk and incubated
   overnight at 4 C. Membrane was washed twice in PBST at RT, secondary
   antibody diluted in PBST + 0.5% milk and rocked for 1 h at RT, then
   washed twice in PBST and imaged on a ChemiDoc.

Screen and pathway-enrichment analysis

   A lentiviral Brunello/Cas9 library was purchased from the Broad
   Institute Genetic Perturbation Platform (GPP) and used to infect 400 M
   U2-OS cells at an MOI of 0.5. Puromycin selection began 48 h later and
   continued for 96 h. The surviving cell library was passaged once into
   15 cm plates and 40 M cells were transduced with 5e3 vg/cell rAAV2/2.HB
   L and R vectors in biological duplicate. At 48 hpt, cells were
   collected and stained with MarkerGene™ (a commercially available MUG
   analog kit) according to the manufacturer’s protocol, sorted on a FACS
   Aria (BD), and gDNA extracted from each population with the Qiagen
   DNeasy Blood and Tissue kit according to manufacturer’s protocol. NGS
   was performed at the Broad GPP core facility.

   Log-normalized sgRNA barcode counts from FACS-sorted MUG-negative cells
   were subtracted from log-normalized sgRNA barcode counts from
   FACS-sorted MUG-positive cells. This difference was used to model a
   hypergeometric distribution using the Broad GPP Web Portal
   ([199]https://portals.broadinstitute.org/gpp/public/). From this
   analysis, we recovered enrichment scores and p-values associated with
   individual sgRNA barcodes, which we mapped to gene perturbations. Genes
   with fewer than 2 unambiguous sgRNA perturbations were culled from
   subsequent analyzes. Plots based on these statistics were made using
   the pandas and matplotlib libraries in Python.

   The most significant 5% of positive regulators (i.e., genes that, when
   perturbed, have increased dual vector transduction) were submitted as a
   gene list for Reactome pathway analysis^[200]57. Pathways significantly
   overrepresented within this gene list, and their respective p-values,
   were used to make matplotlib plots and inform hit validation.

   Genes in the top 5% of negative regulators grouped by their Reactome
   pathway(s) (multiple pathways can contain the same gene) are tabulated
   with their fold-enrichment scores, and their p-values (Supplementary
   Data [201]1). Code used for operations on tables, to make supplementary
   tables, and to plot analyzes can be found at 10.5061/dryad.7h44j1053
   and are included as a Jupyter notebook (Supplementary Data [202]2).

High-content imaging and analysis

   Cells were plated in CellCarrier Ultra 96 well plates (Perkin Elmer)
   and transduced/treated with the appropriate drugs and vectors. Cells
   were fixed and stained with Hoechst to facilitate nuclear segmentation,
   and immunostained where indicated with the below primary antibodies
   overnight at 4 C for detection by an Alexa647-conjugated secondary
   antibody. ROIs comprising up to 50% of the well surface were imaged
   with 11 z-planes at 0.5 micron steps using the appropriate filter sets
   on an Opera Phenix automated imager (Perkin Elmer). MIPs were processed
   and quantified using Harmony (v5.1) image analysis software using an
   appropriate analysis pipeline (Supplementary Data [203]2). Well and
   object results were imported into Graphpad Prism (v10.1.0) for data
   visualization.

Cell cycle analyzes

   For analyses by flow cytometry, cells were seeded in 12-well plates and
   treated with drug or transfected with siRNA 24 h later. At 40 or 36 h
   post treatment, cells were trypsinized and pelleted, resuspended in
   Phenol Red-free DMEM + 0.01% digitonin + 2.5 μg/mL DAPI, and 5000 cells
   per condition were analyzed by flow cytometry on a LSR Fortessa and
   FACSDiva software (v 9.0) from BD Biosciences. Gating for G1, S, and
   G2/M phases was performed on DMSO treated samples. Data were analyzed
   using FlowJo software (v 10.9.0).

   For imaging studies of S-phase, cells were transduced with 1e5 vg/cell
   of AAV2/2.LacO.64.CMV.mScarlet for 48 h, then media changed to
   DMEM + 10 μM EdU + 10% FBS for 15 minutes. Cells were then fixed and
   EdU detected using the Click-iT™ EdU Cell Proliferation kit (Alexa 647)
   according to the mfrs instructions. Cells were then immunostained and
   imaged as described in High-Content Imaging and Analysis.

Antibodies and dilutions

Immunofluorescence

   Rb anti-53BP1: Abcam ab175188, lot# GR135905-4, 1:10,000. Rb
   Phospho-Histone H2A.X: Cell Signaling Technology 25775, lot# 14,
   1:2500. Ms anti-BRCA1: Santa Cruz sc-6954, lot# J1821, 1:2000. Rb
   anti-Rad51: Axxora CAC-BAM-70-001-EX, lot# 04, 1:4000, and ab133534,
   lot# GR3270300-18, 1:2500.

ChIP-qPCR

   Normal rabbit IgGs: Abcam #ab46540, 0.75 μg per ChIP. Rabbit
   anti-Phospho-Histone H2A.X (Ser139): Cell Signaling Technology #2577,
   lot# 14, 2.5 μg per ChIP. Rabbit anti-Histone H2B: Invitrogen #
   MA5-24697, 1.5 μg per ChIP.

Western blotting

   Rb anti-Rad51: ab133534, lot# GR3270300-18, 1:2000. Ms anti-BRCA1:
   Santa Cruz sc-6954, lot# J1821, 1:1000. Ms anti-TBP: Recombinant
   Anti-TATA binding protein TBP antibody [mAbcam51841] (ab300656), lot#
   GR135905-4, 1:2500.

Live imaging experiments

   Cells were seeded in ibidi Culture-Insert 4 Well µ-Dishes (cat# 80466),
   pretreated with DMSO or B02 for 8 h, then 1e6 vg/cell was added to
   wells and imaging commenced 4 h later on a Nikon W1 SoRa spinning disk
   with humidified CO2 incubation chamber and the NIS Elements software (v
   9.0). A 10-micron stack of images were acquired every 6 minutes for 48
   consecutive hours. Maximum Intensity Projections (MIP) were generated
   from these stacks, and MIP movies were generated in ImageJ (v 2.14.0).
   After appropriate cropping, substacks were generated and processing to
   stabilize nuclei in the field of view was accomplished using the
   StackReg plugin on the Rigid Body setting.

Reporting summary

   Further information on research design is available in the [204]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [205]Supplementary Figs.^ (15MB, pdf)
   [206]41467_2025_56738_MOESM2_ESM.pdf^ (15.2KB, pdf)

   Description of Additional Supplementary Files
   [207]Supplementary Movie 1^ (7.3MB, avi)
   [208]Supplementary Movie 2^ (5.1MB, avi)
   [209]Supplementary Movie 3^ (5.6MB, avi)
   [210]Supplementary Movie 4^ (6.9MB, avi)
   [211]Supplementary Movie 5^ (7.3MB, avi)
   [212]Supplementary Movie 6^ (2.6MB, avi)
   [213]Supplementary Data 1^ (343.3KB, csv)
   [214]Supplementary Data 2^ (1.3MB, txt)
   [215]Reporting Summary^ (292.3KB, pdf)
   [216]Transparent Peer Review file^ (279.4KB, pdf)

Source data

   [217]Source Data^ (7.2MB, xlsx)

Acknowledgements