Improving the eciency of liver targeting rAAV-mediated homologous recombination using ribonucleotide reductase inhibitors

Recombinant adeno-associated viral (rAAV) vectors continue to gain popularity for in vivo therapeutic gene delivery. Homologous recombination-based gene therapy using rAAV (AAV-HR) without nucleases has several advantages over classical gene therapy, especially when targeting the liver in neonatal/pediatric populations due to its potential for permanent sustained transgene expression. However, the low eciency of AAV-HR remains a limitation for some clinical applications. Here, we tested series of small molecule compounds with different mechanisms of action in the context of AAV-HR and identied that ribonucleotide reductase (RNR) inhibitors signicantly enhanced the AAV-HR eciency in mouse and human liver cell lines. Furthermore, short term administration of the RNR inhibitor udarabine increased the in vivo eciency of both non-nuclease and CRISPR/Cas9-mediated AAV-HR in the murine liver, without causing toxicity. Mechanistic experiments showed that udarabine administration induced transient DNA damage signaling in both proliferating and quiescent hepatocytes. Surprisingly, in vivo BrdU labeling implicated that the majority of AAV-HR events occurred in non-proliferating hepatocytes in both udarabine-treated and control mice. These studies suggested that the induction of transient DNA repair signaling in non-dividing hepatocytes was responsible for enhancing the eciency of AAV-HR in mice during RNR inhibition. In total, we show the use of a clinically approved RNR inhibitor can enhance AAV-HR based genome editing therapeutics. 3.6-fold) (Extended Fig.14b), whereas a low dose (10 mg/kg) did not signicantly change hF9 levels. hF9 mRNA expression was increased by DEN administration (2- and 3.9-fold at low and high dose, respectively), whereas albumin mRNA expression was unchanged (Extended Fig.14c-d). On-target HR-derived fusion hF9 mRNA showed ~2-fold increase in DEN treated mice (Extended Fig.14e). These data suggested that some enhancement of hF9 expression was possible with DEN, however, DEN administration resulted in more severe toxicity, which was not observed in udarabine-treated mice.


Introduction
Recombinant adeno-associated virus (rAAV) are increasingly popular vectors for gene transfer having achieved recent FDA approval of several rAAV-based gene therapies 1,2 . Although the liver is an attractive target organ for rAAV gene therapy, it is clear that classical gene replacement approaches are unlikely to be applicable to newborn or pediatric patients. Hepatocyte proliferation and/or turnover results in loss of therapeutic expression from vector episomes [3][4][5] .
As an alternative to gene replacement, we have developed a rAAV-based promotorless gene targeting strategy referred to as AAV-HR or GeneRide TM 6 . This gene editing technology targets the 3' coding region of a highly expressed gene, such as Albumin (Alb), using homology arms surrounding a self-cleaving peptide (P2A) and a transgene. In this context, successful gene targeting via homologous recombination (HR) results in conversion of the native mRNA to a fusion mRNA (Alb-P2A-transgene) and translation of the endogenous and transgene proteins. This technology has shown permanent correction of several disease models 6,7 .
Despite the effectiveness as a therapy, the overall rate of AAV-HR remains low. Furthermore, the molecular mechanism of the gene targeting is not fully understood. Here, we identify a small molecule drug that can be administered transiently and safely in combination with rAAV to enhance gene editing in somatic mammalian tissue.
We next used a promoterless vector targeting the Glyceraldehyde-3-Phosphate Dehydrogenase (GAPDH) locus for HR, designated GAPDH-P2A-GFP, to evaluate AAV-HR e ciency. This vector allows for GFP expression only following HR at the target locus, allowing ow cytometry quanti cation of HR events (Fig.1a). As GAPDH-P2A-GFP lacks a promoter and start codon at the N-terminus of GFP, there is little functional expression from episomal rAAV genomes. When treated with hydroxyurea (HU), cells showed a 3 to 4-fold increase in GFP+ cells (Fig1.b) whereas most of other compounds failed to increase the GFP+ fraction (Extended Fig.1b).
A well-de ned molecular target of HU is the RNR enzyme which generates deoxyribonucleotides from ribonucleotides 18 . Therefore, we examined other RNR inhibitors ( udarabine, triapine, gallium nitrate) and found these drugs also signi cantly enhanced transduction of rAAV, except gallium nitrate (Fig.1c) 18 . In the ow cytometry gene targeting assay, we observed an increase in GFP+ Huh7 cells pre-treated with udarabine and triapine (Fig.1b). To validate RNR inhibition as the mechanism for increased gene targeting, we used siRNA knockdown of RRM1, the main subunit of the RNR complex 18 . Knockdown of RRM1 also increased GFP+ fractions in GAPDH-P2A-GFP transduced cells (Fig.1d-e). These data supported that suppression of RNR can enhance the e ciency of gene targeting in vitro.
To con rm on-target HR, we ampli ed genomic DNA at the left integration junction in Huh7 cells postinfection, using a primer binding to the upstream genomic sequence, past the homology arms, and a primer binding to the P2A sequence (Fig.1a). As such, correctly integrated genomic sequences with the expected size were ampli ed only in transduced cells (Fig.1f).
We next examined these compounds in the mouse hepatoma cell line Hepa1-6, using a vector targeting the mouse Albumin locus, named Alb-P2A-GFP (Fig.1g). Consistent with Huh7 data, HU or udarabine pretreatment, prior to Alb-P2A-GFP infection, increased the GFP+ fractions in Hepa1-6 cells compared to controls (Fig.1g). To con rm targeted integration, we analyzed Albumin-P2A-GFP fusion mRNA levels by qPCR (Fig.1h). qPCR of fusion Albumin-P2A-GFP mRNA showed increased abundance in HU (3.5-fold) and udarabine (2.6-fold) treated cells (Fig.1i). Taken together, the above data show RNR inhibitors increased the e ciency of AAV gene targeting in vitro, independent of species and target locus.

Fludarabine administration increased AAV-mediated gene targeting in vivo
To test RNR inhibitors in vivo, we injected the rAAV8-Alb-P2A-hF9 gene targeting vector, containing human coagulation factor 9 (hF9) with homology arm targeting the mouse Albumin locus, with or without administration of HU or udarabine (Fig.2a) 6 . Strikingly, we observed a sustained 2.7 to 4.6-fold increase of plasma hF9 levels in mice treated with udarabine, but not HU, compared to control mice (Fig. 2b). We speculated the lack of e cacy in HU-treated mice might be due to its rapid liver metabolism 19 . Importantly, udarabine administration did not delay body weight gain compared with PBS controls (Extended Fig.2a). In addition, no severe or irreversible toxicity was observed in udarabine-injected mice based on complete blood counts, blood biochemical examination, and liver histopathological analysis by a trained veterinarian pathologist (Extended Fig.2b, Extended Fig.3, and Table 2). The copy number of rAAV genomes per diploid hepatocytes was similar between groups, regardless of drug treatment, showing increased hF9 expression was not the result of higher vector delivery into the liver (Fig.2c).
To con rm HR, we ampli ed the right-sided integration junction using a genomic-binding primer outside the homology arm (primer Rv) and a primer in hF9 (primer Fw1) (Extended Fig.4a). We observed the ontarget HR-derived amplicon at the expected size of 1.6 kb only in rAAV-injected mouse samples (Extended Fig.4b). A nested PCR showed a single band of expected size (1.5 kb) (primers Fw3 and Rv2), which was TOPO-cloned and 5 clones (one from each mouse) were Sanger sequenced, con rming correct on-target integration without any indels (Extended Fig.4c-d). We next quanti ed the amount of fusion hF9 mRNA derived from on-target integration (Fig.2d). Fusion mRNA levels were signi cantly higher in the udarabine-treated versus control animals (5.1-fold increase) (Fig.2e). qPCR of total hF9 mRNA with vector-internal primers, which includes fusion mRNA from targeted integration, random integration, or episomal-derived expression, showed a 4.4-fold increase in udarabine-treated versus control mice (Fig.2f). Endogenous albumin expression remained unchanged by udarabine treatment (Fig.2g).
To obtain mRNA characteristics in mouse livers on a single cell level, we used RNAScope in situ hybridization to label hF9 and mAlb mRNAs following rAAV and udarabine treatment. Endogenous mAlb mRNA was immensely abundant in all hepatocytes, while hF9 mRNA was expressed in relatively few hepatocytes ( Fig.2h and Extended Fig.5). These infrequent hF9 mRNA positive cells were consistent with the low frequency of gene targeting 7,20 . hF9 mRNA positive cells displaying strong intracellular staining were more abundant in udarabine versus PBS-treated animals. Collectively, the above data demonstrated that udarabine increased hF9 expression through enhanced AAV-HR e ciency.
Next, we examined a udarabine dose response. Five days of udarabine treatment resulted in the highest plasma hF9 levels (6.4-fold vs. PBS) while even a single administration showed a 2-fold increase over controls (Fig2.i and Extended Fig.6a). We observed no signi cant decrease in body weight with longer udarabine treatment (Extended Fig.6b). Furthermore, delayed udarabine administration, four weeks after vector injection, did not enhance hF9 expression, suggesting that the effect of udarabine on gene targeting requires proximity to the timing of rAAV transduction (Extended Fig.7). Lastly, udarabine also enhanced gene targeting to the ApoE locus, demonstrating its effect is not restricted to Alb in mouse livers (Extended Fig.9).
To fully de ne the effect of udarabine on liver transduction, we injected rAAV8-HLP-hAAT expressing human alpha-1-antitrypsin (hAAT) from the liver-speci c HLP promoter, with or without udarabine. Fludarabine transiently increased hAAT mRNA and serum levels 2-fold at three days post rAAV injection but returned to the same level as the controls eight days post-injection, remaining similar to controls for 11-weeks (Extended Fig.10). Furthermore, at 11-weeks, there were no differences in hAAT mRNA levels (Extended Fig.10). Furthermore, we determined neither HU nor udarabine changed the amount of total nuclear or double-stranded rAAV genomes in Huh7 cells (Extended Fig.11). Overall, udarabine may transiently affect rAAV-mediated transgene expression without altering nuclear entry, similar to reports using hydroxyurea in vitro 10 .
Fludarabine transiently suppressed hepatocyte proliferation RNR inhibitors (e.g. HU) are used to induce cell cycle arrest and synchronize cultured cells in S-phase; moreover, canonical HR is most active in the S/G2 phase of the cell cycle [21][22][23][24] . As such, we investigated the proliferation status of liver cells by measuring DNA synthesis via bromodeoxyuridine (BrdU) labeling. Mice were injected with BrdU for three days, with or without udarabine, and liver sections were stained with an anti-BrdU antibody ( Fig.3a and Extended Fig.12a-b). BrdU incorporation averaged ~50 positive nuclei per eld of view in control mice, while BrdU labeled nuclei were nearly absent in udarabine-treated mice, demonstrating a strong inhibitory effect of udarabine on hepatocyte DNA synthesis (Fig.3b-c).
However, BrdU incorporation following udarabine removal was signi cantly greater than during drug treatment, suggesting BrdU positive hepatocytes progressed through S phase after drug washout (Fig.3bc). Together, this data indicated that udarabine treatment showed a transient and substantial inhibition of cell proliferation/DNA synthesis in mouse livers. Thus, increased proliferation in the liver was not the cause of increased gene targeting with udarabine treatment.
To further study the relationship between cell proliferation and HR, we injected the rAAV8-Alb-P2A-GFP vector simultaneously with a week of BrdU, then co-stained hepatocytes which underwent HR (GFP signal) and DNA synthesis (BrdU) (Fig.3d). As before, udarabine signi cantly increased GFP+ cells bỹ 5.4-fold ( Fig.3e-f). qPCR analysis determined that on-target fusion GFP mRNA showed 6 to 7-fold greater levels in udarabine versus PBS-treated animals (Fig.3g). Albumin mRNA expression remained unchanged by udarabine treatment (Fig.3h).
We then examined whether GFP+ cells had progressed through S-phase as determined by BrdU incorporation. Surprisingly, we found BrdU (S-phase marker) colocalization with GFP (HR marker) positive cells was infrequent at 4.0% of all GFP+ cells in PBS-treated mice and 1.76% in the udarabine-treated mice (Fig.3i). Thus, most GFP+ cells had not progressed through S-phase of the cell cycle during the timeframe of the experiment, regardless of drug treatment. Overall, progression through S-phase may not be a primary determinant of rAAV-HR in vivo, as only a small fraction of GFP+ cells had undergone DNA synthesis in the examined timeframe.
Fludarabine induced a transient DNA damage response There are several different effects of udarabine on cells: the drug is incorporated into DNA during replication or DNA repair [25][26][27] , it causes the reduction of the dNTP pool by RNR inhibition, and causes premature transcription termination by incorporation into transcribing mRNA [27][28][29] . Any of these events can trigger DNA repair pathways, which can be linked to increased HR e ciency [30][31][32] . To assay for DNA damage response, we stained for phosphorylated Ser139 H2AX (γH2AX) in liver sections from the mice studied in Fig.3a after udarabine treatment 33 . Minimal levels of γH2AX were detected in untreated or BrdU-only treated livers ( Fig.4a-b). Mice treated with diethylnitrosamine (DEN), a potent mutagen, showed intense staining (Extended Fig.12c-d) 34 . During udarabine treatment, mouse livers possessed widespread pan-nuclear γH2AX foci ( Fig.4a-b). The number of positive nuclei lessened in the three days after udarabine removal ( Fig.4a-b). Western blotting for γH2AX using liver lysates from the same animals displayed similar tendencies to the immunohistochemical staining data (Fig.4c). We also found both siRNA-mediated knockdown of RRM1 as well as udarabine treatment strongly increased γH2AX levels in Huh7 cells (Extended Fig.13). These results showed that udarabine induced temporary DNA repair signaling in mouse liver, which may at least in part explain the mechanism of increased gene targeting e ciency with udarabine.

Enhancement of gene targeting is not speci c to all genotoxic treatments
Genotoxic treatments which induce DNA damage response can increase gene targeting rates in cultured cells 35 (Extended Fig.1b). To determine if genotoxicity is also a governing principle for rAAV gene targeting in vivo, we utilized DEN and Alb-P2A-hF9 co-injections in mice. DEN administration decreased body weight compared to controls in a dose-dependent manner (Extended Fig.14a). The highest dose of DEN (30 mg/kg) signi cantly increased hF9 plasma levels (about 3.6-fold) (Extended Fig.14b), whereas a low dose (10 mg/kg) did not signi cantly change hF9 levels. hF9 mRNA expression was increased by DEN administration (2-and 3.9-fold at low and high dose, respectively), whereas albumin mRNA expression was unchanged (Extended Fig.14c-d). On-target HR-derived fusion hF9 mRNA showed ~2-fold increase in DEN treated mice (Extended Fig.14e). These data suggested that some enhancement of hF9 expression was possible with DEN, however, DEN administration resulted in more severe toxicity, which was not observed in udarabine-treated mice.
Fludarabine increased e ciency of CRISPR/Cas9-mediated gene editing The induction of speci c double-stranded breaks generated by programmable nucleases, such as CRISPR/Cas9 can increase rAAV-HR e ciency, which has been extensively investigated as promising gene editing therapies both in vivo and ex vivo 20,36 . To investigate whether CRISPR/Cas9-mediated HR is also affected by udarabine, mice were injected with two rAAVs: one delivering saCas9 and a gRNA targeting the intron downstream of exon 14 in mouse albumin (AAV-Cas9) and one encoding a promoterless GFP sequence anked by albumin homology arms (AAV-GFP-HDR) (Fig.5a). We examined various ratios of AAV-Cas9 to AAV-GFP-HDR (1:1, 1:5, 1:10) co-injections, with or without udarabine. In all ratios tested, we found udarabine treatment increased the percentage of GFP+ hepatocytes compared to controls (Fig.5b-c). qPCR analysis of the on-target fusion mRNA also showed an increase of of the two-week experiment showed no signi cant changes with udarabine treatment (Extended Fig.15ab). To determine if udarabine affected indels present at the target site, we performed targeted deep sequencing analysis using mouse liver tissues injected with AAV-Cas9 alone. The number of alleles with indels showed 2 to 3-fold increase in udarabine-treated mice (Fig.5e). Additionally, the distribution of the top 12 alleles containing indels determined that udarabine treatment resulted in an absence of larger insertions, compared to control (Fig.5f). SaCas9 mRNA levels at the time of tissue collection were unchanged by udarabine treatment (Extended Fig.15c). Together these results showed that udarabine also enhances CRISPR/Cas9-mediated gene editing in vivo, likely through stimulation of DNA repair and/or altered repair characteristics at the target site.

Discussion
In this study, we determined RNR inhibitors are useful drugs to enhance AAV-HR e ciency. Previous literature has described several compounds to increase gene editing e ciencies, however these approaches were only validated in cultured cells, or limited in vivo context, such as correction of gene mutation [38][39][40][41][42] . Our data showed that a short-term treatment with udarabine, an FDA-approved drug, signi cantly enhanced AAV-HR in vivo without causing overt toxicity. Importantly, analysis of the genomic integration site at the Albumin locus con rmed precise recombination with no unexpected mutations in udarabine-treated mice. In addition, our data showed that udarabine also enhances gene editing with nuclease-based systems such as CRISPR/Cas9, implicating the use of udarabine to increase editing e ciencies across numerous contexts, editing technologies, and genomic targets.
Prior investigations into the mechanisms governing AAV-HR determined that key HR-pathway factors are required for rAAV-HR 42 . In this regard, canonical HR is predominantly active during S/G2 phase of cell cycle when homologous sister chromatids are available as a repair templates during DNA replication [22][23][24] . Contrary to this S-phase centric HR principle, our present data showed that ~96% of hepatocytes undergoing HR in vivo were not co-immunostained with BrdU implicating that S-phase progression may not be a major determinant factor in AAV-HR and may not be the cause of udarabine's enhancing effect. In support of this nding, we have found similar rates of rAAV-mediated gene targeting in neonatal and adult mouse livers, as well as in livers undergoing regeneration 6 (and unpublished data). Our observation that AAV-HR occurred mostly in cells that did not progress through S phase is in stark contrast to previous studies in cell culture [43][44][45] . It should be noted most of in vivo hepatocytes are in G0, which is considerably different from cultured cells and could contribute to the discordance 46 . Recently, it was reported that AAV-HR in cardiomyocytes occurred independently of S phase entry, which is in line with our present data 47 .
This lack of S phase progression in HR+ hepatocytes prompted the hypothesis that udarabine may have effects on non-proliferating hepatocytes. Fludarabine inhibits the catalytic subunit of RNR complex, RRM1, which functions throughout the cell cycle 18 . Thus, udarabine can also function in non-dividing cells by antagonizing basal levels of DNA repair due to decreased intracellular dNTP pools from RRM1 inhibition 25 . This notion is supported by our observation that udarabine altered the repair characteristics at the targeted locus in mice injected with AAV-Cas9. Moreover, udarabine, as a purine analogue, is incorporated into replicating DNA but also into nascent RNA, leading to premature transcription termination 48 . Indeed, by examining γH2AX levels, we con rmed widespread activation of DNA repair signaling in udarabine-treated mouse livers where most of hepatocytes are in G0-phase. gH2AX is a hallmark of various DNA repair pathways and facilitates the recruitment and accessibility of repair factors to damaged DNA, which could directly or indirectly promote HR 49,50 . Moreover, the induction of DNA repair signaling could titrate host DNA damage factors, which process rAAV genomes, away from incoming rAAV particles, in uencing transduction and perhaps recombinogenic forms of rAAV 51 . Therefore, activation of DNA repair can, at least partially, explain how udarabine increased the probability of AAV-HR to occur in BrdU negative hepatocytes. In addition, given that DEN, a potent mutagen, induced higher levels of γH2AX, yet mediated a lesser enhancement of HR than udarabine, it is unclear if DNA damage is required for AAV-HR. More likely is that the type of DNA damage and magnitude, in addition to other factors, may be important determinants for AAV-HR e ciency. Since the precise mechanism of rAAV-mediated gene targeting is not fully understood, particularly in vivo, unveiling the details of such mechanisms will be the focus of our future studies.
Overall, these novel discoveries show that a transient treatment udarabine can safely enhance rAAV gene targeting e cacy, in a clinically relevant manner. Advanced approaches building upon this work will expand the utility of various gene editing technologies and permit lifelong therapeutic gene modi cation after a single administration of rAAV gene targeting vectors.

Methods
Cell culture and compound treatment Huh7 cells and Hepa1-6 cells were purchased from JCRB and ATCC, respectively and were cultured in DMEM with 10% FBS, 2 mM glutamine. For hydroxyurea (Sigma), udarabine (Fisher scienti c), triapine (Fisher scienti c) and Gallium (III) nitrate hydrate (Fisher scienti c), cells were pretreated with each compound for 16 hours at indicated concentration and the cultured medium were replaced with fresh medium without compounds before rAAV transduction. For Torin-1, FK228 (Fisher scienti c) (Fisher scienti c), Trichostatin A, MG132 (Sigma) and Teniposide (Abcam), cells were treated with each compound for 24 hours at indicated concentration and rAAV was transduced 1 hour after compound addition.

Vectors
The AAV vectors containing ITR sequences used in this study are based on AAV type 2 backbone. CAG-Fluc, Albumin-P2A-hF9, HLP-hAAT vectors were prepared as described previously 52 . Albumin-P2A-GFP vector was generated by replacing hF9 coding sequence of the Albumin-P2A-hF9 vector with GFP coding sequence using In-Fusion® HD Cloning Kit (TAKARA). SaCas9-sgRNA8 vector and Albumin-P2A-GFP vectors used for Figure.5 were prepared as described previously 20 . For construction of the GAPDH-P2A-GFP vector, human genomic GAPDH segments were PCR-ampli ed using Fw: 5'-GACTGTACAGGGCTGCTCACATATTCTGG-3' and Rv: 5'-CTGTGTACAGAGTGTATGTGGCTGTGGCCC -3' (both containing BsRG1 sites for cloning) and inserted between AAV2 ITRs into BsrGI restriction sites in a modi ed pTRUF backbone 53 . The genomic segment spans approximately 1.7 Kb upstream and 1.7 kb downstream to the GAPDH stop codon. We then synthesized a 1,359 bp fragment spanning the region at the end of the GAPDH locus between the two SexA1 sites to be cloned in the vector. In this fragment, the GAPDH stop codon was removed and it was inserted an optimized P2A coding sequence preceded by a linker coding sequence (glycine-serine-glycine) and followed by the GFP sequence (without the start codon). For Apoe-P2A-hF9 vector construction, a genomic fragment containing sequences used for both homology arms was ampli ed from mouse genomic DNA. Primers (5'-TCC ACA CCT GCC TAG TCT CG -3') and (5'-GTG CCA GAG GCA GTT GAG TT -3') were used to amplify a 2.9kb fragment and the PCR product was directly cloned into the pCR Blunt II TOPO vector using the Zero Blunt TOPO PCR cloning kit (Invitrogen), sequence veri ed. The left homology arm was ampli ed from the cloned ApoE genomic fragment using primers (5'-ata tca tcg atc gcg atg cat taa tta agc ggc cgA AGA CTG TAG GTC CTG ACC C -3') and (5'-ggt ggc gcc gct tcc TTG ATT CTC CTG GGC CAC -3'), the middle part containing hF9 sequence was ampli ed from Albumin vector using primers (5'-gcc cag gag aat caa GGA AGC GGC GCC ACC AAT -3') and (5'-gga gaa gga tac tca TGT CAG CTT GGT CTT TTC TTT GAT CC -3'), the right homology arm was ampli ed from the cloned ApoE genomic fragment using primers (5'-aag acc aag ctg aca TGA GTA TCC TTC TCC TGT CCT GC -3') and (5'-acg taa cag atc tga tat cac gcg tgt aca cta gtG CCC TGC TGA GTC CCT GAG -3'). Lower case letters indicate the overlapping sequences. Phusion Hot Start Flex (NEB) was used for all ampli cations. Amplicons were assembled using the NEBuilder HiFi DNA Assembly Master Mix (NEB) according to instructions.

Animals
All animal works were performed in accordance to the guidelines for animal care at Stanford University or approved by the ICGEB review board, with full respect to the EU Directive 2010/63/EU for animal experimentation, and by the Italian Health Minister (authorization 996/2017-PR). Wild-type C57BL/6 (B6) mice were purchased from Jackson Laboratory (Bar Harbor, ME). We used 3-4-week-old juvenile mice to test the effect RNR small molecule inhibitors on gene targeting e ciency, as they are still undergoing development and growth, including in the liver.
Mouse AAV injection, drug dosing, bleeding and tissue sampling B6 mice received tail vain injections of rAAV8 packaging each vector at the designated dose and were bled at indicated time points. Body weight of mice was measured using Scout pro portable scale (Ohaus) at indicated time points. Serum samples were obtained by centrifugation at 10,000 rpm for 10 minutes and used for ELISA assay of hF9 or hAAT, ALT enzyme measurements, and complete blood count tests. HU and udarabine were resuspended in PBS and injected intraperitoneally with indicated dose/regimen. For our studies examining the in vivo effects of RNR inhibitors, we chose to use HU and udarabine due to their FDA approval and well-established safety pro les. The drug dosing was based on information from several preclinical studies.
Brie y, an initial dose of 1000 mg/kg injection of HU showed severe toxicity, leading us to reduce the dose to 300 mg/kg/day. For udarabine, an LD 10 for single or 5-daily IV infusions were about 980 and 400 mg/kg per dose, respectively, in male mice according to the product monograph of udarabine. Additionally, ve-day doses of ½ the LD 10 or ~200mg/kg per dose was considered safe and administration of a smaller dose (125 mg/kg) three times a day over three days (1125 mg/kg total dose) produced the greatest activity in a leukemia tumor xenograft model. Using this information, we decided to use a dose of three 125 mg/kg injections per day (375 mg/kg/day) of udarabine for 3 days. All drug administrations were through intraperitoneal injections (i.p.). For BrdU labeling of proliferating mice hepatocytes, BrdU was resuspended in PBS and intraperitoneally injected at 200 mg/kg per day for 3 days or 7 days. Diethylnitrosamine (DEN) solution was prepared using saline. We determined an appropriate dose of DEN after nding administration at 100 mg/kg resulted in severe weight loss requiring euthanasia. Dosing at 10 or 30 mg/kg as a single injection per day for 3 sequential days was well tolerated and used to examine this drug's genotoxic effect on gene targeting. At the end of each experiment, mice were anesthetized with iso urane and perfused transcardially with PBS and then liver tissues were quickly harvested and cut into several pieces. The tissues for mRNA extractions were immediately submerged in RNAlater solution (Sigma) and stored at 4°C until use. For gDNA or protein extraction, tissues were snap-frozen in liquid nitrogen and stored at -80°C until use.
Hematology and liver pathology CBC (complete blood count) and liver histopathological analysis were performed in mice injected with three 125 mg/kg injections of udarabine per day, for 3 days. On the day of last injection and 4 weeks after that, mice were submitted to the Veterinary Service Center (VSC) at Stanford University for standard blood paneling tests and blinded analysis of liver pathology by a skilled veterinary pathologist.
AAV production rAAV vectors were produced as previously described using a triple transfection protocol with Ca3(PO4)2 or PEI 25K, followed by puri cation by CsCl gradient 53 or using AAVpro® Puri cation Kit (All Serotypes) purchased from Takara Bio. Puri ed rAAVs were stored at -80°C until used. rAAV genomes were extracted and puri ed using QIAamp MinElute Virus Spin Kit (QIAGEN) and were titered by qPCR. The sequence information of primers are shown in Table1.

Fire y luciferase assay
Luciferase assays were performed using ONE-Glo™ Luciferase Assay System (Promega) following the manufacturer's instruction. Brie y, at indicated time points after rAAV (CAG-Fluc) transduction, the equal volume of the reconstituted substrate to the cultured medium was added to the cells grown in 96 well plates and incubated for 10 minutes with gentle shaking. Luminescent activity was measured using a plate reader. siRNA transfection ON-TARGETplus siRNA against human RRM1 and POLR2A as well as scramble control siRNA were purchased from Dharmacon and were transfected into Huh7 cells using RNAiMAX (Life Technologies) according to the manufacturer's instructions. Final concentration of siRNA was 20 nM. AAVDJ packaging GAPDH-P2A-GFP vectors were added into the cultured medium at a multiplicity of infection (MOI) of 20,000 vector copies/cells at 48 hours after siRNA transfection. Flow cytometry was performed for detection of GFP positive fractions at the indicated time points.

Flow cytometry
Huh7 or Hepa1-6 cells were harvested and washed with cold PBS and resuspended in cold PBS containing 3% FBS. Cells were kept on ice and protected from light until analyzed. Singlet cells were determined based on FSC/SSC plot and GFP positive fractions were gated based on negative control which was non-transduced cells. The number of GFP expressing cells was evaluated using the BD FACSCalibur™ instrument and data were analyzed using the FlowJo software package.

RNA extraction and cDNA preparation
Total RNA was extracted using RNeasy micro plus kit (QIAGEN) according to the manufacture's protocol with DNase treatment. Liver tissue samples stabilized in RNA later solution (-100mg) were homogenized in RINO 1.5 mL Screw-Cap Tube lled with stainless steel beads and 600 mL of RLT buffer (including bmercaptoethanol) using a bead homogenizer (Next Advance Bullet Blender Storm). Total RNA was extracted from the tissue lysates using RNeasy plus mini kit (QIAGEN) with additional on-column DNase treatment. cDNA was synthesized from 200-500 ng of total RNA using High-Capacity RNA-to-cDNA™ Kit (Life Technologies) according to the manufacturer's instructions.
gDNA extraction Cultured cells were collected by trypsinization and washed with PBS. Total genomic DNA was extracted using QIAamp DNA Mini Kit (QIAGEN) according to the manufacture's protocol with RNase A treatment.
Snap-frozen liver tissue (-100mg) were homogenized in RINO 1.5 mL Screw-Cap Tube lled with stainless steel beads and 600 mL of AL buffer using a bead homogenizer. Total RNA was extracted from the tissue lysates using DNeasy Blood & Tissue Kit (QIAGEN).

PCR and qRT-PCR
The polymerase chain reactions (PCRs) to amplify genomic regions where homologous integrations occurred (junction PCR) were performed using Q5® Hot Start High-Fidelity 2X Master Mix (New England Biolabs). The following cycling conditions were used: Human b-Actin (one cycle of 98°C for 30 sec, 28 cycles of 98°C for 10 sec, 60°C for 15 sec, and 72°C for 10 sec, and one cycle of 72°C for 2 min), GAPDH-P2A junction (one cycle of 98°C for 30 sec, 35 cycles of 98°C for 10 sec, 62°C for 15 sec, and 72°C for 1 min, and one cycle of 72°C for 2 min), Mouse albumin (one cycle of 98°C for 30 sec, 32 cycles of 98°C for 10 sec, 60°C for 10 sec, and 72°C for 2 min, and one cycle of 72°C for 2 min) , hF9-Albumin junction nested PCR (one cycle of 98°C for 30 sec, 20 cycles (1 st PCR) and 25 cycles (2 nd PCR) of 98°C for 10 sec, 62°C for 15 sec, and 72°C for 1 min, and one cycle of 72°C for 2 min). PCR products were analyzed in agarose gels containing Ethidium bromide and visualized using ChemiDoc Imaging Systems (Bio-Dad). Primer sequence information are listed in Table1.
Quantitative PCR was performed in duplicate using Apex qPCR GREEN Master Mix (Genesee Scienti c) and CFX384 Touch Real-Time PCR Detection System (Bio-Rad) using the following cycling conditions: 95°C for 15 min, 45 cycles of 95°C for 10 sec, 60°C for 10 sec and 72°C for 10 sec, and one cycle of 95°C for 10 sec and 65°C for 1 min and 65-97°C (05°C/sec). Standard curves for each primer set were generated using serially diluted linearized plasmid and used for quanti cation. CFX Maestro Software was used for data analysis and relative mRNA expression levels were calculated by normalized against βactin. All sequence information of primers is listed in Table1.

Protein extraction and western blotting
Total cell lysates from cultured cells or mouse liver tissues were prepared using RIPA buffer containing Halt™ Protease and Phosphatase Inhibitor Cocktail (both from Thermo Fisher). Liver tissues were homogenized in RINO 1.5 mL Screw-Cap Tube lled with stainless steel beads and 600 mL of RIPA buffer using a bead homogenizer. Protein concentration were measured using Pierce™ BCA Protein Assay Kit (Thermo Fisher) and the same amount of proteins for each sample were loaded into NuPAGE™ 4-12% Bis-Tris Protein Gels (Thermo Fisher). iBlot2 transfer system (Thermo Fisher) was used for western blotting. Detection of BrdU incorporated DNA was accomplished with heat denaturing in an antigen retrieval buffer (Advanced Cell Diagnostics, Newark, CA), followed by staining with a rat monoclonal anti-BrdU antibody (BU1/75 (ICR1)) (Abcam, Cambridge, UK) and secondary Alexa Fluor 594 antibody (Thermo Scienti c). All IHC slides were mounted with Prolong Diamond Antifade with Dapi (Thermo Scienti c) and imaged on a Zeiss LSM 880 confocal microscope. Speci city of all staining procedures was ensured with appropriate biological controls and control slides stained with secondary antibody only.
For eGFP experiments with Cas9 specimens were frozen in optimal cutting temperature compound (BioOptica, Milano, Italy) and 4 µm slices were obtained in a cryostat. Percentage of eGFP+ cells in liver specimens were detected by natural GFP uorescence, while nuclei were visualized by Hoechst (10 lg/ml) staining. Slides were mounted in Mowiol 4-88 (Sigma). Images were acquired on a Nikon Eclipse E-800 epi-uorescent microscope with a charge-coupled device camera (DMX 1200F; Nikon, Amstelveen, The Netherlands). Digital images were acquired on a Nikon Ti Eclipse inverted uorescence microscope equipped with Intensilight Epi-uorescence Illuminator, a Perfect Focus 3 system and 20× (NA 0.45) objective. Acquisition was performed with a DS-Qi2 16 Mpixel camera (Nikon). NIS Elements microscope imaging software (Nikon, version 4.6) was used to quantify the total number of cells (Hoechst staining) and the GFP positive cells. The settings of Automated Spot Detection were set in order to detect all bright spots with a typical minimum diameter of 8 μm. The function "Detect all objects" was activated and spot detection-output were directly exported to Excel, where the data analysis was performed. For each animal an average 10000 nuclei were counted. Measurements (GFP+ cells/total nuclei) were averaged for each animal, and the results were expressed as mean ± SD for each treatment.

RNAScope in situ hybridization of hF9
Liver tissue was processed for RNA in situ hybridization as described above. Fixed frozen tissue was sectioned into 9 µm thick sections and RNAScope hybridization was performed according to the manufacturer's protocol (Advanced Cell Diagnostics). A custom probe was designed to detect codonoptimized human factor 9 mRNA, while control probes targeted either murine peptidylprolyl isomerase B (PPIB) (positive control for RNA quality) or bacterial 4-hydroxy-tetrahydrodipicolinate reductase (dapB) (negative control). RNA speci city was con rmed using RNase digestion of control tissue sections and slides were counterstained with 50% hematoxylin (Thermo Scienti c). Imaging was performed using a Leica DM2000 bright eld microscope.

Image analysis
All image analysis was performed in a blinded manner. Analysis of BrdU incorporation was performed manually requiring nuclear colocalization of BrdU signal and greater signal intensity over background to be recorded as a positive nucleus. Signal from overtly non-hepatocyte nuclei directly associated with larger liver structures such as central veins or bile ducts were not included in tally. Analysis of BrdU incorporation and GFP colocalization was performed similarly. Scoring of phosphorylated Ser139 γH2AX was performed using ImageJ software.

Enzyme-linked immunosorbent assay (ELISA)
Mice serum samples were used to quantify hF9 or hATT protein expression levels. ELISA for hF9 was performed as previously described 6 with the following antibodies: mouse anti-hF9 IgG primary antibody at 1:1,000 (Sigma Cat#F2645), and polyclonal goat anti-hF9 peroxidase-conjugated IgG secondary antibody at 1:4,000 (Enzyme Research Cat#GAFIX-APHRP). ELISA for hAAT was performed as previously described 52 .

Targeted Illumina Deep Sequencing
Genomic DNA was extracted from liver tissue using phenol/chloroform extraction, following incubation with RNAseA and Proteinase K. Genomic DNA was submitted to the Genome Engineering and IPSC Center (GEiC) at Washington University in St. Louis for targeted deep sequencing of the gRNA target site.

Statistics
GraphPad Prism was used for statistical analysis. Groups of two were analyzed by unpaired t test. More than two groups were compared by one-way ANOVA with Bonferroni correction. Statistical signi cance was assumed with P value <0.05 (*), <0.01(**) and <0.001 (***). Bars in graphs represent standard deviation for each group.
Image analysis was performed using two-tailed t-tests between groups. Normally distributed data, determined by D'Agostino Pearson test or Shapiro-Wilk test, was analyzed with parametric student's t test if groups had equal variance. Groups with unequal variance were analyzed with an unpaired t test with Welch's correction. Signi cance in non-normally distributed data was determined using nonparametric Mann-Whitney U test.

Declarations
Tables 1-2 are not provided in this version of the manuscript.