Abstract

   Base editors have been reported to induce off-target mutations in
   cultured cells, mouse embryos and rice, but their long-term effects in
   vivo remain unknown. Here, we develop a Systematic evaluation Approach
   For gene Editing tools by Transgenic mIce (SAFETI), and evaluate the
   off-target effects of BE3, high fidelity version of CBE (YE1-BE3-FNLS)
   and ABE (ABE7.10^F148A) in ~400 transgenic mice over 15 months.
   Whole-genome sequence analysis reveals BE3 expression generated de novo
   mutations in the offspring of transgenic mice. RNA-seq analysis reveals
   both BE3 and YE1-BE3-FNLS induce transcriptome-wide SNVs, and the
   numbers of RNA SNVs are positively correlated with CBE expression
   levels across various tissues. By contrast, ABE7.10^F148A shows no
   detectable off-target DNA or RNA SNVs. Notably, we observe abnormal
   phenotypes including obesity and developmental delay in mice with
   permanent genomic BE3 overexpression during long-time monitoring,
   elucidating a potentially overlooked aspect of side effects of BE3 in
   vivo.

   Subject terms: Genetic engineering, CRISPR-Cas9 genome editing
     __________________________________________________________________

   The potential off-target effects of long-term expression of base
   editors in vivo are unclear. Here the authors report SAFETI, Systematic
   evaluation Approach For gene Editing tools by Transgenic mIce, to
   examine off-target effects of base editors over time in mice, and see
   abnormal side effects.

Introduction

   CRISPR-derived base editing is a genome editing method to introduce
   point mutations on DNA or RNA at the target loci. By fusing
   catalytically dead Cas9 (dCas9) or nickase Cas9 (nCas9) with cytidine
   deaminases or adenosine deaminases, cytosine base editors
   (CBEs)^[52]1,[53]2 and adenine base editors (ABEs)^[54]3 were developed
   to install targeted C-to-T or A-to-G point mutations without generating
   double-strand breaks (DSBs). However, the application of CBEs and ABEs
   was limited by off-target DNA and RNA mutations^[55]4–[56]7. Even
   though high-fidelity base editors were subsequently generated by
   protein engineering^[57]4,[58]8–[59]10, their specificity in vivo
   remains to be explored considering their constant and long-term
   expression through common delivery strategies^[60]11,[61]12.

   Here we generate transgenic mouse lines expressing BE3, high fidelity
   YE1-BE3-FNLS (W90Y and R126E mutations in rAPOEBC1)^[62]8, or
   ABE7.10^F148A (F148A in TadA)^[63]4, and comprehensively evaluate the
   side effects of base editors on DNA and RNA across various tissues. In
   addition, we also monitor the phenotypes of the transgenic mice over
   months to explore the adverse effects of long-term expression of BEs.

Results

Generation of transgenic mice expressing base editors

   We generated transgenic mouse lines expressing BE3^[64]1,
   YE1-BE3-FNLS^[65]8, hA3A-BE3^[66]13, ABE7.10^3, or ABE7.10^F148A4 using
   the piggyBac (PB) transposon integration system^[67]14 (Fig. [68]1a and
   Supplementary Fig. [69]1a). The transposon vectors consisted of a CAG
   promoter followed the sequence encoding one of the base editors that
   was linked (via a self-cleaving P2A peptide) to an enhanced green
   fluorescent protein reporter (eGFP; Supplementary Fig. [70]1a). The
   transposon vectors were injected into zygotes of the C57BL/6 J mice
   together with PB transposase enzyme (PBase) mRNA. The zygotes were
   developed to two-cell embryos in vitro, and then transferred into
   oviducts of pseudopregnant females of the ICR strain. We found that
   birth rates of F0 mice were not severely affected by injecting vectors
   consisting of BE3, YE1-BE3-FNLS or ABE7.10^F148A (Fig. [71]1b).
   Besides, fluorescence detection and PCR genotyping indicated that more
   than 75% of the born mice in these groups were successfully integrated
   with the corresponding base editor (Fig. [72]1c, Supplementary
   Table [73]1). By contrast, mice injected with vectors for BE3-human
   APOBEC3A (hA3A-BE3) or ABE7.10 showed dramatically reduced birth rates
   without successfully integrated pups (Fig. [74]1b, c), suggesting that
   hA3A-BE3 and ABE7.10 were toxic to embryos. F0 mice with the successful
   integration of the various base editors were intercrossed to establish
   transgenic mouse founders, with the cross-mating of founders yielding
   four transgenic mouse lines expressing GFP, BE3, YE1-BE3-FNLS, or
   ABE7.10^F148A (Fig. [75]1d). Analysis of transposon integration sites
   showed the transgenes were integrated into the intergenic or intronic
   regions at different locations across the mouse genome (Supplementary
   Fig. [76]1b, c, Supplementary Table [77]2). The average copy numbers in
   GFP, BE3, YE1-BE3-FNLS and ABE7.10^F148A transgenic mice were 3.3, 4.3,
   3.7 and 2.3, respectively, no statistical difference was observed among
   different groups (Supplementary Fig. [78]1d, e).

Fig. 1. Generation of GFP, BE3, YE1-BE3-FNLS, and ABE7.10^F148A mouse lines.

   [79]Fig. 1
   [80]Open in a new tab

   a Schematic diagram of the Systematic evaluation Approach For gene
   Editing tools by Transgenic mIce (SAFETI). b Birth rate of F0 mice
   injected with GFP, BE3, hA3A-BE3, YE1-BE3-FNLS, ABE7.10 or
   ABE7.10^F148A. Birth rate: the percentage of number of live births
   divided by the population size of transferred embryos. The birth rates
   were calculated for three injections, and the total numbers of
   transferred embryos were 265, 258, 259, 320, 302 and 265 in GFP, BE3,
   hA3A-BE3, YE1-BE3-FNLS, ABE7.10 and ABE7.10^F148A, respectively. c The
   proportion of F0 mice with successful integration of the transgenes
   (positive rate). The positive rates were calculated for three
   transplantations, and the total number of births were 91, 78, 86, 21,
   17 and 72 in GFP, BE3, YE1-BE3-FNLS, hA3A-BE3, ABE7.10 and
   ABE7.10^F148A, respectively. d Bright-field and fluorescence images of
   newborn wild-type (WT) mice and mice expressing GFP, BE3, YE1-BE3-FNLS,
   or ABE7.10^F148A. e The expression levels of transgenes in the
   indicated tissues. FPKM, Fragments Per Kilobase Million. f Schematic
   showing the experimental procedure for delivery of Hpd sgRNA through
   AAV8 into the CBE transgenic mice by tail-vein injection. g
   Representative Sanger sequencing chromatograms of PCR amplicons
   spanning the Hpd gRNA target site are shown for GFP, BE3, and
   YE1-BE3-FNLS transgenic mice. Red arrows mark the targeted cytosine. h
   The C-to-T base editing efficiency at the Hpd target site in GFP, BE3,
   and YE1-BE3-FNLS mice assessed by deep sequencing. i Schematic showing
   the experimental procedure for delivery of Dmd sgRNA through AAV9 into
   the ABE transgenic mice by intramuscular injection. j Representative
   Sanger sequencing chromatograms of PCR amplicons spanning the Dmd
   target site are shown for GFP and ABE7.10^F148A mice. Red arrows mark
   the targeted adenine. k The A-to-G base editing efficiency at the Dmd
   target site (A3, A4, A6) in GFP and ABE7.10^F148A mice assessed by deep
   sequencing. Data are presented as means ± SEM (n = 3 biologically
   independent samples). P values were calculated by two-sided, unpaired
   t-test. Source data are provided as a Source Data file.

   We next performed RNA-seq analyses of eight tissues (brain, lung,
   heart, liver, kidney, ovary, muscle and adipose) from 8-week-old GFP,
   BE3, YE1-BE3-FNLS, and ABE7.10^F148A transgenic mice. The mRNA
   expression of all three base editors were highest in muscles, followed
   by the heart, and low in other tissues, and we noted that YE1-BE3-FNLS
   was expressed at higher levels than BE3 or ABE7.10^F148A across various
   tissues (Fig. [81]1e). Similar expression patterns of these base
   editors across various tissues were observed in immunoblotting of
   extracts from organs (Supplementary Fig. [82]1f). These findings are in
   line with previous reports of particularly high expression of
   transgenes in muscles and in the heart^[83]15,[84]16, and may reflect
   the preferential activation of CAG promoter in these organs. Despite of
   the high expression of base editors, we observed no obvious
   morphological abnormalities in any of the examined organs of the
   8-week-old transgenic mice (Supplementary Fig. [85]2a).

   We next examined the on-target editing performance in vivo in the base
   editor (BE) transgenic mouse lines. For the BE3 and YE1-BE3-FNLS mouse
   lines, we constructed a U6-Hpd sgRNA vector containing an sgRNA for
   C-to-T editing of the Hpd locus (Supplementary Table [86]3), which
   results in a stop codon that can rescue the lethal phenotype of
   hereditary tyrosinemia type 1 in mice^[87]17. This vector was packaged
   into AAV2 serotype eight particles (AAV8) and delivered to 8-week-old
   CBE transgenic mice by tail-vein injection (Fig. [88]1f). Two weeks
   after delivery, we conducted Sanger and targeted deep sequencing to
   examine on-target editing in hepatocytes isolated from transgenic mice
   by FACS (Supplementary Fig. [89]2b) and detected successful C-to-T
   editing of the targeted site in ~20% and ~30% of the BE3 and
   YE1-BE3-FNLS hepatocytes, respectively (Fig. [90]1g, h, Supplementary
   Table [91]4).

   We assessed A-to-G base editing activity of ABE7.10^F148A transgenic
   mice using an sgRNA targeting the Dmd gene (Supplementary Table [92]3),
   whose dysfunction results in Dunchenne muscular dystrophy (DMD)^[93]18.
   Specifically, the U6-Dmd sgRNA was packaged into AAV9 and delivered to
   the tibialis anterior (TA) muscle of 8-week-old ABE7.10^F148A mice by
   localized intramuscular injection^[94]19 (Fig. [95]1i). We collected TA
   muscles from 3 ABE7.10^F148A mice two weeks after delivery and found an
   average of 10.5% A-to-G editing at A6 of the targeted site by Sanger
   and targeted deep sequencing (Fig. [96]1j, k, Supplementary
   Table [97]4). Taken together, these results support that the
   genomically integrated BE3, YE1-BE3-FNLS, and ABE7.10^F148A of our
   transgenic mouse lines can successfully perform sgRNA-directed
   on-target editing in vivo.

Base editors induced substantial genome-wide mutations in vivo

   To evaluate whether the BE3, YE1-BE3-FNLS and ABE7.10^F148A editors
   introduce off-target edits in the transgenic mice, we performed whole
   genome sequencing (WGS) analysis of gDNA extracted from muscles (i.e.,
   an organ with high base editor expression). To eliminate the influence
   of genetic background, we also performed WGS on the same tissues from
   GFP transgenic and wild-type (WT) mice. Mutations were called by three
   algorithms in the transgenic samples using the WT samples as the
   reference. Unlike few off-target mutations revealed by previous in vivo
   studies^[98]19–[99]23, we found an average of 1353 SNVs in muscles of
   BE3 transgenic mice, which is 5 times higher than that in the GFP mice
   (Fig. [100]2a). These mutations were specifically observed in the
   transgenic mice rather than in the WT mice (Supplementary
   Fig. [101]3a). In addition, the numbers of C-to-T or G-to-A mutations
   were highest in the BE3 transgenic mice, and much higher than those in
   the GFP group (Fig. [102]2b, Supplementary Fig. [103]3b). This mutation
   bias was the same as that of cytosine deaminase rAPOBEC1^[104]1,
   suggesting these mutations were induced by BE3 expression. We also
   detected a significantly higher number of indels and structural
   variations (SVs) in the BE3 group than the GFP group (Supplementary
   Fig. [105]3c, d). By contrast, there were no differences in the numbers
   of SNVs or indels between the GFP and BE transgenic mice expressing
   high fidelity YE1-BE3-FNLS or ABE7.10^F148A (Fig. [106]2a, b and
   Supplementary Fig. [107]3c, d).

Fig. 2. Off-target DNA mutations in the transgenic mice expressing base
editors.

   [108]Fig. 2
   [109]Open in a new tab

   a Comparison of the total number of DNA SNVs in muscles of GFP, BE3
   YE1-BE3-FNLS and ABE7.10^F148A transgenic mice. n = 3 for each group. b
   Comparison of the number of DNA SNVs with the indicated mutation types
   in muscles of the GFP, BE3, YE1-BE3-FNLS, and ABE7.10^F148A transgenic
   mice. n = 3 for each group. c Schematic diagram for in vivo base
   editing verification by crossing Tyr sgRNA transgenic mice with GFP,
   BE3, YE1-BE3-FNLS, and ABE7.10^F148A mice. d Representative photos for
   the hair colors of offspring from GFP × Tyr sgRNA, BE3 × Tyr sgRNA,
   YE1-BE3-FNLS × Tyr sgRNA and ABE7.10^F148A × Tyr sgRNA mice. e The
   C-to-T or A-to-G base editing efficiency at Tyr target site in
   offspring of BE3 × Tyr sgRNA (n = 9), YE1-BE3-FNLS × Tyr sgRNA (n = 12)
   or ABE7.10^F148A × Tyr sgRNA (n = 11) breeding pairs. f Comparison of
   the number of de novo SNVs in offspring expressing GFP (n = 4), BE3
   (n = 5), YE1-BE3-FNLS (n = 5) or ABE7.10^F148A (n = 6) with the Tyr
   sgRNA transgenic mice as mother. g Comparison of the number of de novo
   SNVs in offspring expressing GFP (n = 7), BE3 (n = 4), YE1-BE3-FNLS
   (n = 7) or ABE7.10^F148A (n = 5) with the Tyr sgRNA transgenic mice as
   father. n = biologically independent samples. Data are presented as
   mean ± SEM. h Distribution of mutation types of the de novo SNVs in
   offspring of crossing GFP/BE3/YE1-BE3-FNLS/ABE7.10^F148A father and Tyr
   sgRNA mother. The number indicates the percentage of a certain type of
   mutation among all mutations. i Sequence logos derived from the de novo
   SNVs in offspring of crossing BE3/YE1-BE3-FNLS father and Tyr sgRNA
   mother. Source data are provided as a Source Data file.

   To further eliminate the interference of genetic background, we next
   examined the de novo off-target mutations induced by constant
   expression of base editors in 43 parent-offspring trios. We used the PB
   transposition system to generate another transgenic mouse line with a
   genomically integrated sgRNA cassette targeting the Tyrosinase (Tyr)
   gene for pigmentation^[110]5,[111]8 (Fig. [112]2c and Supplementary
   Table [113]3). The targeted C-to-T editing by CBE directed by this
   sgRNA has been shown to introduce a stop codon at the Tyr gene that
   results in an Albino phenotype^[114]5,[115]8. We then established four
   breeding pairs by crossing the GFP and BE transgenic male mice (BE3,
   YE1-BE3-FNLS or ABE7.10^F148A) with Tyr sgRNA transgenic female mice,
   and 4 pairs of the reciprocal cross. The offspring of BE3 or
   YE1-BE3-FNLS × Tyr sgRNA mice showed black-white mosaic hair phenotypes
   (Fig. [116]2d), indicating the efficient editing of CBEs at the Tyr
   gene. We next performed WGS on gDNA extracted from tails of the
   progenies expressing both base editors and Tyr sgRNA, and confirmed the
   on-target editing efficiency of BE3 (26%), YE1-BE3-FNLS (54%), and
   ABE7.10^F148A (31%) at the Tyr locus (Fig. [117]2e).

   To obtain the de novo generated mutations in the progenies, we
   performed WGS analysis on gDNA extracted from tails of 56 mice from 43
   parent-offspring trios and called de novo variants in the offspring by
   filtering those shared with their parents or WT mice. Notably, we found
   that the number of de novo SNVs in the BE3 group was significantly
   higher than that in the GFP group (Fig. [118]2f, g and Supplementary
   Fig. [119]4a). The number of de novo SNVs in the YE1-BE3-FNLS and
   ABE7.10^F148A group was also slightly higher than that of the GFP
   samples when the BE transgenic mice were the mother (Fig. [120]2g).
   Interestingly, we noted that a higher number of de novo SNVs were
   observed in the offspring from crosses in which the BE transgenic mice
   were the father (Fig. [121]2f, g). A similar trend was observed in the
   GFP group, consistent with previous studies showing a paternal bias for
   de novo mutations^[122]24. The de novo mutations in the BE transgenic
   mice were evenly distributed across the chromosomes (Supplementary
   Fig. [123]4b).

   Strikingly, the percentages of C-to-T and G-to-A mutations in the BE3
   and YE1-BE3-FNLS transgenic mice were significantly higher than that of
   the GFP group (Fig. [124]2h and Supplementary Fig. [125]4c). Moreover,
   we found a consensus motif WCW (W = A or T) from the de novo mutations
   identified in the BE3 and YE1-BE3-FNLS transgenic mouse lines
   (Fig. [126]2i), consistent with the mutation preferences of the