Fluorescent in vivo editing reporter (FIVER): A novel multispectral reporter of in vivo genome editing

Advances in genome editing technologies have created opportunities to treat rare genetic diseases, which are often overlooked in terms of therapeutic development. Nonetheless, substantial challenges remain: namely, achieving therapeutically beneficial levels and kinds of editing in the right cell type(s). Here we describe the development of FIVER (fluorescent in vivo editing reporter) — a modular toolkit for in vivo detection of genome editing with distinct fluorescent read-outs for non-homologous end-joining (NHEJ), homology-directed repair (HDR) and homology-independent targeted integration (HITI). We demonstrate that fluorescent outcomes reliably report genetic changes following editing with diverse genome editors in primary cells, organoids and in vivo. We show the potential of FIVER for high-throughput unbiased screens, from small molecule modulators of genome editing outcomes in primary cells through to genome-wide in vivo CRISPR cancer screens. Importantly, we demonstrate its in vivo application in postnatal organ systems of interest for genetic therapies — retina and liver. FIVER will broadly help expedite the development of therapeutic genome surgery for many genetic disorders.


31
The development of ever more precise and efficient genome editing technologies is revolutionising 32 1A-C). 110 To test the system, we generated immortalised MEF lines from FIVER mice and transiently trans-111 fected them with ribonucleoprotein (RNP) comprised of SpCas9 protein complexed with either 112 T1 or T2 gRNAs. Confocal imaging and flow cytometry confirmed transition from mtdTomato to 113 mEGFP, accounting for approximately 30% of events, indicative of NHEJ repair following excision 114 of the tdTomato cassette ( Figure 1B-C). In addition, there was a total loss of fluorescence following 115 CRISPR/Cas activity in approximately 30% of cells, due to larger deletions or imperfect repair which 116 truncated the fluorophore or altered the reading frame. In some instances (particularly in immor-117 talised MEF lines, accounting for approximately 10% of events, but not in vivo) we also observed a 118 tdTomato + /EGFP + population following editing; primarily observed using flow cytometry. As a re-119 sult, we took the total of tdTomato − /EGFP + , tdTomato + /EGFP + and tdTomato − /EGFP − populations 120 to represent overall levels of editing. 121 To assess HDR pathways, we constructed both single-and double-stranded repair templates, 122 containing homology arms of various lengths (35 bp to 780 bp) and have focused on ∼700 bp 123 arms flanking an H2B nuclear localisation signal encoded on a minicircle vector (Figure 1-figure   124 supplement 1A) which initially gave the highest and most consistent rates of repair (Figure 1-fig-125 ure supplement 1C). Following co-delivery of this construct (MC.HDR) with CRISPR/Cas machinery, 126 nEGFP fluorescence could be observed ( Figure 1B). The edited cells were also subjected to flow cy-127 tometric analysis ( Figure 1C). A shift in fluorescent profile was observed following editing, however, 128 nEGFP and mEGFP expression were not distinguishable by intensity using standard flow cytometry 129 (Figure 1-figure supplement 2), necessitating an image analysis-based approach to quantify HDR, 130 as described later. 131 The method of delivering editing machinery can impact editing outcomes and will vary depend-132 ing on application (22). To address this, we have built a toolkit to allow delivery of CRISPR compo-133 nents and repair constructs in various forms (RNP, plasmid or minicircle) either by non-viral meth-134 ods (i.e., nucleofection, lipid nanoparticles and hydrodynamic injection) or virally (i.e., lentivirus 135 and adeno-associated virus). 136 As the FIVER system reports on DSB-repair outcomes, we postulated that any site specific nu-137 clease generating DSBs could be employed. While the bulk of work has focused on SpCas9, differ-138 ences in nuclease size, types of ends generated and availability of specific PAM motifs close to the 139 3 of 33 target may warrant use of a range of genome editors. Therefore, we designed gRNAs for use with 140 Staphylococcus aureus Cas9 (SaCas9) and Acidaminococcus sp. Cas12a (AsCas12a) (previously Cpf1) 141 to target the same conserved region flanking tdTomato. We assayed the activity of AsCas12a and 142 demonstrated the ability of FIVER to report its editing outcomes (Figure 1D-F populations, respectively, aligned to the predicted NHEJ repair product ( Figure 2C). However, these 170 predominantly align across the EGFP gene and not the repair junction ( Figure 2C). This suggests a 171 high accuracy of the mEGFP readout. 172 Using de novo genome assembly, the MinION reads were successfully assembled in order to 173 form the major sequences present within the input. When aligned to the reference, sequences  Schematic of FIVER system. We identified conserved gRNA sites on both the sense (T2; green box) and antisense (T1; purple box) strands flanking the tdTomato cassette within the FIVER locus (PAM sites indicated by orange boxes). Here membrane-tagged tdTomato is expressed by every cell. When CRISPR machinery and either T1 or T2 gRNA are provided, the tdTomato cassette is excised. Without the provision of an exogenous repair template non-homologous end joining (NHEJ) repair is employed to repair the lesion, allowing expression of downstream membrane-tagged EGFP, observed with a shift from membrane tdTomato (mtdTomato) to membrane EGFP fluorescence (mEGFP). Alternatively, asynchronous cleavage and/or larger indels (dotted line) can cause disruption of the tdTomato resulting in loss of all fluorescence. If a template containing homology to the locus is provided, the lesion can be repaired by homology directed repair (HDR), in our system this replaces the membrane tag of the downstream EGFP for a nuclear tag (H2B) resulting in a shift from mtdTomato to nuclear EGFP (nEGFP) fluorescence. Finally, if a homology-independent targeted integration (HITI) repair template is provided, then NHEJ can be employed to knock in a membrane-tagged TagBFP construct, resulting in a shift from mtdTomato to nuclear TagBFP (nTagBFP) fluorescence. m = MARCKS membrane tag, n = H2B nuclear localisation signal. (B) Representative confocal images of mouse embryonic fibroblast (MEF) lines derived from FIVER mice and edited with and without repair constructs. Images are maximum intensity projections from z-stacks. (C) Representative flow cytometry plots following editing in MEF lines. All editing outcomes can be observed, however nEGFP and mEGFP are indistinguishable by this method (see Figure 1-figure supplement 2). FACS was carried out 5 days post transfection. (D) Representative confocal images of MEFs edited using AsCas12a machinery with T3 gRNA. Nuclei are stained with Hoechst. (E) Editing in MEF lines using Cas9 is significantly more efficient than using AsCas12a (p <0.001; one-way ANOVA with Tukey's multiple comparison), n = 10,000 single cells, N = 3 technical replicates. (F) There is no significant difference in the ability of SpCas9 and AsCas12a to drive HDR in MEF lines using minicircle (MC) delivery of repair constructs (p = 0.257; unpaired t-test), n > 6,000 cells, N = 3 technical replicates.

of 33
To investigate HDR, targeted resequencing of the tdTomato − /EGFP + (NHEJ and HDR) popu-190 lation was carried out. Primers spanning the entire locus were used to ensure that the long-191 lived minicircle donor template was not erroneously amplified (23,24) (Figure 2A, PCR 7). Of the 192 tdTomato − /EGFP + population, 19.4% of reads aligned to the predicted HDR sequence containing 193 integrated H2B (Figure 2D). Given the rate of total editing here (Figure 2-figure supplement 1C), 194 this means approximately 1.32% of total cells underwent HDR, consistent with the range of HDR 195 efficiency we have previously observed in MEFs (Figure 1-figure supplement 1B) and a similar 196 proportion to that described in the literature (25)(26)(27). This suggests that observed nEGFP is consis-197 tent with changes at the DNA level.

205
In all edited cells, mtdTomato fluorescence rapidly decreased (magenta line, Figure 3B). In the 206 case of NHEJ, this signal was concurrently replaced with mEGFP fluorescence, increasing gradually 207 in mean intensity over time (yellow line, Figure 3B). In the case of HDR, nEGFP accumulates gradu-208 ally before rapidly increasing in intensity, then plateauing (green line, Figure 3B). Similarly, for HITI 209 editing, nTagBFP accumulates gradually at first, before rapidly increasing then plateauing approx-210 imately 40 hours post transfection (blue line, Figure 3B). In all cases, the switch in fluorescence 211 occurs rapidly and is complete by 48 hours post-transfection (Figure 3-video 1).

213
One of the limitations of genome editing as a therapeutic tool is its dependence on endogenous 214 DNA repair pathways to resolve targeted nicks, cuts and/or breaks generated by the nucleases. The 215 reliance on HDR to generate specific changes in the genomes of mammalian somatic cells, where 216 this is not the dominant DNA repair pathway (28), has led to the search for methods to manipulate 217 repair mechanism choice. This includes the identification of small molecules to bias outcomes 218 towards precise repair by stimulating HDR as well as inhibiting NHEJ. However, it remains largely 219 unknown whether all cell types will respond similarly in resolving genome edited DSBs and whether 220 there are cell-type-specific effects of these small molecules.  has been reported to increase HDR efficiency in response to CRISPR-induced DNA damage. We also 231 tested two molecules identified using a blind screening method for molecules which improved the  Only NU7441 had a significant effect on HDR, increasing it approximately 2-fold (p = 0.03, one-235 way ANOVA with Dunnett's multiple comparison, N = 3, Figure 4B). Surprisingly, NU7441 also signifi-      (Figure 4D), indicative of a reduction in NHEJ-dependent HITI. These results were 240 recapitulated with another DNA-PKcs inhibitor (Nedisertib), which had been shown to be more ef-241 fective than NU7441 (37). While Nedisertib did increase HDR (Figure 4-figure supplement 1D), the 242 increase in HDR was the same as with NU7441 despite increasing total editing, tdTomato − /EGFP + , 243 and tdTomato − /EGFP − populations, whilst decreasing TagBFP + and tdTomato + /EGFP + populations 244 all to a greater extent (Figure 4-figure supplement 1E-I). This demonstrates the ability of FIVER 245 to rapidly and unbiasedly screen for such modulators of DNA editing outcomes.

246
Rapid preclinical screening of delivery methods in vitro 247 Balancing efficacy with safety for delivery tools will be an essential part of the development of a

265
In parallel, we transduced FIVER mTECs with SpCas9, gRNA and an HDR template using a dual 266 viral system. Here, the CRISPR machinery was delivered via lentivirus (with its larger packaging 267 capacity) while the HDR templates were delivered via AAV, as AAV is particularly recombinogenic 268 (41,42). We focused on AAV serotypes previously reported to be efficacious in delivering to airway    Highly efficient templated repair in FIVER early embryos 286 HDR is often more efficient in early embryos than in somatic cells (48,49). Thus, to demonstrate our 287 reporter in an optimal system, we investigated the amount and type of genome editing outcomes 288 in blastocysts following nuclear microinjection of FIVER single cell zygotes; we carried out pronu-        include novel nucleases to explore their efficiencies and the editing outcomes they elicit in vivo, as 395 they are taken forward for preclinical use.

396
Using FIVER, we investigated a range of previously reported small molecule modulators of DSB 397 repair. In our initial screen, only NU7441 significantly increased HDR (Figure 4B). In addition, 398 we also observed a significant reduction in the number of TagBFP + cells, confirming that HITI re-399 sults from NHEJ-mediated knock-in of the repair template ( Figure 4D). Though counter-intuitive, 400 NU7441 treatment also increased the level of overall editing, by increasing both tdTomato − /EGFP − 401 and tdTomato − /EGFP + populations ( Figure 4B and figure supplement 1A and B). The increase 402 in tdTomato − /EGFP + could be accounted for to some extent by the concomitant increase in HDR 403 (nEGFP). However, the tdTomato − /EGFP − population is believed to result from imprecise NHEJ 404 repair such that we would see a reduction of this population following inhibition of DNA-PKcs.

405
As we observed in the NGS data, this population results from larger deletions following excision 406 of the tdTomato cassette which extend into the promoter region or coding sequence of EGFP 407 (Figure 2-figure supplement 1B). Mutations such as these may be the result of alt-NHEJ, specifi-408 cally microhomology-mediated end-joining (MMEJ), which is known to result in larger indels than tdTomato − /EGFP − populations, but in combination with NU7441 it was cytotoxic (Figure 4-figure   414 supplement 2), suggesting inhibition of multiple DSB repair pathways is not tolerated. In addi-415 tion, our NGS data revealed asymmetry in editing between the two gRNA targets sites, with more 416 indels present at the upstream site ( Figure 2B). This implies that editing at the two near identical 417 sites could be asynchronous or that local sequence differences lead to more disruptive repair at 418 the upstream site. Taken together, these suggest that multiple repair pathways may be employed 419 following CRISPR activity and that blocking one or more merely shifts the balance between these 420 competing pathways. 421 We also investigated Nedisertib, reported to be a more potent inhibitor of DNA-PKcs (37). How-422 ever, we found that Nedisertib was less efficacious than NU7441 at increasing HDR after 24 hours 423 of treatment, though was a more potent inhibitor of HITI (Figure 4-figure supplement 1D and F). can be biased towards the desired outcomes. Preclinical studies to explore how best to balance 441 efficacy (i.e., efficient editing) and safety (i.e., high on-target, non-integrating activity) are needed.

442
Using FIVER, we were able to demonstrate that even identical gRNA and Cas9 nuclease complexes 443 elicited very different outcomes in our airway organotypic cultures; with AAV-delivered HDR re-444 pair effecting robust editing and greater HDR compared to nanoparticle delivery at a proliferative 445 stage ( Figure 5A and figure supplement 1A and B). However, these nanoparticle reagents were  Crucially, we were able to demonstrate HITI editing outcomes at a second independent locus 455 of clinical interest ( Figure 8C). HITI editing has great potential as a therapeutic approach in many 456 genetic diseases. Achieving therapeutic levels of perfect repair by HDR is still a substantial hurdle 457 for the development of genome editing-based therapeutics. However, as HITI takes advantage of 458 the more prevalent NHEJ pathway, it can help to bridge the gap between the precise editing of 459 HDR and the variable indels generated by NHEJ, resulting in a more predictable, targeted repair 460 which occurs more efficiently than HDR strategies. HITI is also a more realistic repair strategy in (P2A). The same gRNAs were cloned into pLentiCRISPRv2-iRFP670 following digestion with BsmBI.

501
All gRNA sequences are detailed in Table 1.  tion, these were transfected with a plasmid containing SV40 large T antigen and selected for using 514 puromycin (3 g/mL).

515
Mouse tracheal epithelial cells (mTECs) were derived from tracheas of 5-7 week old FIVER mice. RNPs were carried out using the same Neon conditions, using a total of 1 g of Cas9 protein (Ther-539 moFisher Scientific, USA) per 0.5 x10 5 cells.

542
Complexes were allowed to form at room temperature for 5-10 min prior to use. gRNA was pro-    Table 3. Sample preparation and se- were removed using AMPure XP beads, and barcoded DNA was quantified using a Qubit dsDNA HS 589 assay (ThermoFisher Scientific, USA). Equal quantities of each barcoded amplicon were pooled be-590 fore being end-repaired and adenylated to allow ligation of sequencing adapters and tethers from 591 the Nanopore 1D2 Sequencing Kit (Oxford Nanopore Technologies, UK). Libraries were re-purified 592 and an equimolar stock was prepared and sequenced.

593
For targeted sequencing of HDR samples, PCR amplification of the whole locus was carried out 594 using the following primers: FIVER F4 and FIVER R3 (Table 3). Products were purified using the 595 PureLink quick PCR purification kit (ThermoFisher Scientific, USA) according to the manufacturer's 596 instructions. 4 L of purified product was cloned into the pCR-4 Blunt TOPO vector using the Zero 597 Blunt TOPO PCR Cloning Kit for Sequencing (ThermoFisher Scientific, USA). To identify larger dele-598 tions in the promoter region, PCR amplifications using P7 and P8 or P7 and P9 primers was carried 599 out ( Table 3). Products were purified using the PureLink quick PCR purification kit (ThermoFisher  Table 3. 607 Sequence analysis pipelines 608 Ion Torrent script. The fastq output file was used to align reads to the custom reference se-609 quences, using Bowtie 2 (76). Quality control metrics were provided by BamQC (Simon Andrews, Map), was used to generate read length histograms and calculate mean/median read lengths.

622
For de novo genome assembly, Canu (84) was used to assemble MinION data. Settings were 623 tailored to expect a small, repetitive genome. SnapGene software (from GSL Biotech; available at 624 snapgene.com) was used to visualise the resulting genome assemblies.