Microhomology-mediated end joining is the predominant form of DNA repair in the mosquito Aedes aegypti with implications for gene editing, gene drive, and transgene removal

Abstract

Programmable site-specific nucleases have revolutionized the field of genetics, and in the field of mosquito vector control, gene editing by these tools has inspired a new wave of population control approaches that aim to prevent disease transmission. Little is known of how DNA repair is prioritized in mosquitoes, which diverged from the nearest model system (Drosophila) by >200 million years, despite site-specific gene editing now being commonplace. Here, we report a scalable, high-throughput platform for studying DNA double-stranded DNA break (DSB) repair in mosquitoes by delivering CRISPR/Cas9, I-SceI, or other nucleases to Aedes aegypti embryos, capable of measuring single-strand annealing (SSA), non-homologous end joining, and microhomology-mediated end-joining (MMEJ) repair outcomes. We find CRISPR/Cas9 can induce deletions of up to 8.6 kb through SSA repair and is tolerant of resection distances of 3.5 kb. Indel events were insensitive to lig4 knockouts, and across 20 synthetic guide RNAs (sgRNAs) representing 5 locations in 2 transgenic strains were almost exclusively attributed to MMEJ repair, establishing MMEJ as the dominant form of repair in A. aegypti at CRISPR/Cas9 DSBs. This information is critical to our understanding of how DNA repair shapes processes required for genetic control strategies involving gene drive action/resistance as well as transgene stability.

Graphical Abstract

Graphical Abstract.

Introduction

Aedes aegypti mosquitoes are the primary vectors responsible for the transmission of dengue, Zika, yellow fever, and chikungunya viruses, and have been the focus of population control approaches since the beginning of the 20th century [1–3]. Several genetic control approaches incorporating species-specific sterilization-based approaches, dominant-lethal genes, or self-limiting genes that affect fitness and/or reproduction to suppress the target populations have been proposed or developed for this and other mosquito vectors [4, 5]. More recently, genetic control technology known as homing gene drives has been proposed to suppress or modify mosquito populations with the goal of preventing disease transmission [6, 7].

Homing gene drives rely on site-specific gene editing using CRISPR/Cas9 or homing endonucleases to induce DNA double-stranded DNA breaks (DSBs), with host-derived repair pathways controlling the ultimate outcomes [8]. Repair of the cleaved chromosome using homologous recombination (HR) results in gene conversion and hence gene drive [6], while non-homologous end joining (NHEJ) is credited as the main competing mechanism in which target site resistance alleles accumulate to cause gene drive failure [9, 10]. Likewise, we have previously described transgene removal approaches based on DSB induction and single-strand annealing (SSA) in both A. aegypti and Drosophila melanogaster [11–15]. Like HR, SSA requires resection and homology searching, and is in competition with deleterious end-joining processes. Thus, optimizing biotechnologies for both spreading and removing biologically useful transgenes in medically relevant species such as A. aegypti require knowledge of the native DNA break response.

HR, NHEJ, SSA, and microhomology-mediated end joining (MMEJ), are all dependent on factors such as cell cycle stage, sequence homology, nuclease, and the type of DSB end [16–18]. While important HR-related factors involved in homology searching, protein regulation, and DSB end resection such as RAD51, BRCA2, and the MRN complex, among others, have been identified in Aedes, BRCA1 and RAD52 appear to be absent [19]. Despite the apparent absence of BRCA1 and RAD52, HR-based repair still occurs whereas DSB repair in human cells in their absence results in lethality [19]. Factors involved in NHEJ, such as the Ku70, Ku80, DNA Ligase IV (lig4), and DNA-dependent protein kinase catalytic subunit (DNA-PKcs) are present in A. aegypti (unlike Drosophila, where DNA-PKcs is missing) and loss-of-function mutations result in an increase in SSA [13].

In yeast, SSA favors recombination of direct repeats closest to the break site rather repeats further away [20]. In mouse embryonic fibroblasts, SSA rates are six-fold higher when DSBs are next to direct repeats compared to 3.3 kb away, but little difference is seen comparing breaks 3.3 and 9 kb away [21]. Despite the successful use of SSA for removing transgenes in A. aegypti [12, 14], it remains unknown how location relative to direct repeats influences Aedes SSA repair. Also, our previous findings that I-SceI more efficiently biased editing outcomes toward SSA as compared to Y2-I-AniI [14] are consistent with the idea that individual nucleases sway DSB repair outcomes in eukaryotes and some may even be advantageous for SSA repair (reviewed in [22]). Interestingly, despite the popularity of CRISPR/Cas9 in vector mosquitoes, its specific influence (if any) on repair outcomes in mosquito vectors remains unknown. Ultimately, biasing break repair toward desirable gene editing outcomes can benefit the field of genetic vector control, for example, by promoting HR and/or inhibiting deleterious repair outcomes to improve both gene drive efficiency and/or transgene elimination in A. aegypti.

Here, we employ a novel embryo assay for measuring DSB repair outcomes in pre-blastoderm A. aegypti embryos to understand the influence of CRISPR/Cas9 DSB position on the efficiency of SSA and the use of competing EJ-based outcomes. Using this assay, we confirm that DSB repair events following I-SceI treatment recapitulate previous findings based on more onerous germline experiments, and we utilize sgRNAs targeting four regions of a previously developed A. aegypti transgenic strain [12] to show that Cas9-induced DSBs are equally likely to result in SSA when located <1.3 kb from the nearest direct repeat. In contrast, SSA rates fell dramatically at DSB distances of ~3.5 kb, suggesting an upper limit on efficient resection for this pathway. Finally, through analysis of indel variants obtained from each of 20 sgRNAs we provide evidence that MMEJ, and not NHEJ, is the dominant repair pathway competing with SSA in A. aegypti in response to Cas9-induced DSBs. Together, these data better inform transgene designs in A. aegypti to optimize HR (gene drive) or transgene removal (SSA), while opening up future research directions into modifying MMEJ efficiency to further increase the fidelity of these processes.

Materials and methods

Mosquito rearing

kmo^RG [12], 1pc.-kmo^RGB [14], BR-KmoEx4, BR-KmoEx4;Lig4⁻ [12], and kmo⁻ null mutant [23] strains of A. aegypti were reared at 27°C and 70% relative humidity (±10%) with a day/night cycle of 14 h light and 10 h dark. Males and females were fed a 10% sucrose diet. Females were fed defibrinated sheep’s blood using an artificial membrane feeder (Colorado Serum Company, Colorado, USA). To obtain hemizygous G₀ embryos for SSA assay microinjections, 200 kmo^RG or 1pc.-kmo^RGB males were mated to 200 kmo⁻ females. For lig4 NHEJ versus MMEJ assay injections, 200 BR-KmoEx4;lig4^+/+ or BR-KmoEx4;lig4^−/− males and females were self-crossed, respectively. At 72-h post-bloodfeed, females were placed in a dark egg laying chamber to collect pre-blastoderm embryos for CRISPR microinjection.

sgRNA design, synthesis, and activity assays

sgRNAs were designed by selecting PAM sites at varying distances from 684 bp direct repeat sequences. sgRNAs were synthesized using unique forward primers specifying protospacers for each sgRNA and a common reverse primer (Supplementary Table S1) specified by Bassett et al. that included a T7 promoter encoding sequence (5′-AAAAGCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3′), which annealed to create a double-stranded DNA (dsDNA) template for amplification by polymerase chain reaction (PCR) under the cycling conditions [1] 98°C 30 s, [2] 98°C 10 s, [3] 48°C 30 s, [4] 72°C 15 s, [5] repeat [2, 3, 4] 44× [6], 72°C 10 min. Following Zymo DNA Clean and Concentrator PCR cleanup (Zymo, USA), 250 ng of purified dsDNA amplicons were used for T7 in vitro transcription of sgRNA products using the MEGAscript T7 Transcription kit [24] (Invitrogen, California, USA). Subsequent sgRNAs were purified using the MEGAclear Transcription cleanup kit (Invitrogen, California, USA) following 100 µl elution option and ammonium acetate precipitation steps to resuspend final sgRNA in 50 µl DEPC-H₂O. Aliquots for sgRNAs were then stored in single-use, 1 µg quantities. In vitro Cas9/sgRNA digestions were performed following PNA Bio’s in vitro cleavage assay using Cas9 protocol (PNA Bio, California, USA) on amplicons encompassing the sgRNA recognition site (complexed at 37°C, 30 min with 15 ng/µl Cas9 protein, 10 ng/µl sgRNA, 15 ng/µl target DNA). Digested products were run on either 2% (kmo^RG sgRNA assays) or 0.7% (1pc.-kmo^RGB sgRNA assays) agarose gels to visualize cleavage activity.

Embryo microinjection

CRISPR/Cas9 injection mixtures were prepared using 400 ng/µl Cas9 protein (PNA Bio, California), 100 ng/µl sgRNA, and nuclease-free water. Injection mixtures were filtered using MilliporeSigma Millex-LG Sterile Syringe Filter Unit 0.2 µm (5 min, 11 000 × g) and incubated at 37°C for 30 min for Cas9 protein and sgRNA complexing. Mixtures were then centrifuged at 18 000 rpm for at least 45 min before being microinjected into either pre-blastoderm, hemizygous kmo^RG/kmo⁻, 1pc.-kmo^RGB/kmo⁻ for SSA assay or homozygous BR-KmoEx4;lig4^+/+ or BR-KmoEx4;lig4^−/− for lig4 NHEJ versus MMEJ assay. Injection plasmids pSLfa-PUb-I-SceI, pSLfa-PUb-GGCas9, pMOS-attP-U6-sgB2-PUb-eGFP, and pMOS-attP-U6-sgD1-PUb-eGFP were isolated and purified following the Machery–Nagel NucleoBond Xtra Midi Kit for transfection-grade plasmid DNA (Machery–Nagel, Pennsylvania, USA), centrifuged for 1 min, and resuspended in injection mixtures. I-SceI plasmid injection mixture contained 500 ng/µl pSLfa-PUb-I-SceI and was injected into pre-blastoderm, hemizygous kmo^RG/kmo⁻ embryos. Plasmid-supplied CRISPR/Cas9 injection mixtures contained 500 ng/µl pSLfa-PUb-GGCas9 and either 300 ng/µl pMOS-attP-U6-sgB2-PUb-eGFP or 300 ng/µl pMOS-attP-U6-sgD1-PUb-eGFP and was injected into pre-blastoderm, hemizygous kmo^RG/kmo⁻ embryos. Injection needles were made from borosilicate capillary needles beveled at 20°. At 24-h post-injection, embryos were harvested for genomic DNA extraction and purification following the Machery–Nagel Tissue Kit (Machery–Nagel, Pennsylvania, USA).

PCR assays for CRISPR editing outcomes

PCR was performed to amplify SSA and indel sequence edits using Phusion or Q5 High-Fidelity DNA polymerases (New England Biolabs, Massachusetts, USA). SSA assay primers used were Ex5-fwd2 (5′-CTAACCTCAACATAATTATACATG-3′) and In1-rev_SSA (5′-GTTCTWACATTCTRCTTTCAAG-3′) with the cycling steps [1] 98°C 3 min, [2] 98°C 30 s, [3] 60°C 30 s, [4] 72°C 45 s, [5] repeat [2, 3, 4] 39× [6], 72°C 5 min. Indel assay primers for site sites A and B: KmoEx4-F (5′-TGTGAGTAGATTCCTTCGTCGTTGG-3′) and sgRNA-HR-R (5′-GGTGGAGGCCTCCCAGCCCATGG-3′, T_a = 67°C, 72°C extension 1 min), site C: DsREDh-F (5′-TGACCGTGACCCAGGACTCC-3′) and PUb-Pro-R (5′-TCCTGATTTGATCGACAATTTCGG-3′, T_a = 70°C, 72°C extension 45 s), site D: GFP-5F (5′-ATGGTGAGCAAGGGCGAGGAG-3′) and In1-rev2 (5′-cattctgctttcaaggttggctctc-3′, T_a = 70°C, 72°C extension 1 min) following the above PCR cycling conditions with the exception of annealing temperature and extension times. PCR amplicons were purified following Zymo’s DNA Clean and Concentrator kit and sequenced using Oxford Nanopore linear amplicon sequencing (Plasmidsaurus, California, USA).

Bioinformatics Analysis

Two reference sequences were identified that reflected different SNP profiles present in either kmo wild-type or kmo⁻ null mutant A. aegypti strains. For the SSA assay, fastq sequences from Oxford Nanopore deep linear amplicon sequencing (10⁴–10⁵ reads per sample, Supplemental File 2) were aligned to identify best matches to either kmo wild-type or kmo⁻ null mutant reference alleles. We developed the mapping strategy after analyzing the wild-type kmo⁺ sequences (from SSA repair deleting kmo^RG) and the kmo⁻ alleles, during which we noticed 8 SNPs that were largely unique to each allele in addition to a deletion mutation exclusive to the kmo⁻ allele. With this in mind, we then created two, 200 bp reference sequences (one for kmo⁺ and one for kmo⁻) encompassing the 8 SNPs and kmo⁻ deletion site-encompassing region in a single fasta file.

Raw nanopore reads were then aligned to this master fasta file using Minimap2 and its “--secondary=no” command option that forces primary alignments only (best match alignments) and prevents secondary alignments (partial alignments). Alignments were performed using Minimap2 (version 2.24-r1122 [25]) with default alignment parameters other than prohibiting secondary alignments. SAMtools (version 1.13 [26]) was then used to filter out poor quality alignments by setting a map quality score cutoff of 10 on all alignments. To filter short PCR products and incomplete nanopore sequencing reads, BEDTools (version v2.30.0 [27]) intersect command was used to select reads whose alignments mapped 250 bases upstream and downstream of the sgRNA target site. Alignment files were then indexed using SAMtools for Integrative Genomics Viewer (IGV [28]) to count the number of reads aligned to wild-type kmo and kmo⁻ null mutant reference sequences.

For indel quantification, indel assay amplicons were aligned to their respective kmo^RG target site amplicon reference sequence. Alignments and quality filtering was performed as described above using Minimap2, SAMtools, and BEDTools. Aligned bam files were then indexed using SAMtools for indel analysis by IGV and CRISPRessoWGS (version 2.3.0 [29]) to obtain an indel rate among sequenced amplicons.

Results

Previously, we integrated a 3.7-kb transgene containing 3xP3-DsRED and PUb-eGFP gene cassettes within the kynurenine-3-monooxygenase (kmo) locus of A. aegypti (kmo^RG) and demonstrated that DSB induction by the homing endonuclease I-SceI could result in SSA-based repair [12] through the use of 684 bp flanking direct repeats from kmo exons 3-2 (Fig. 1A). With this design, a DSB anywhere between the 684 bp repeats could theoretically trigger SSA, resulting in the removal of 4.4 kb (3.7 kb kmo^RG and the downstream 684 bp direct repeat), and subsequent restoration of the wild-type kmo allele, though our prior studies limited the DSB location to a single site ~300 bp from the proximal repeat (3.4 kb from the distal direct repeat). To determine how factors such as DSB location and endonuclease may influence SSA-based transgene elimination, we developed a DNA repair assay in embryos as this approach could allow for a faster, more scalable platform to study SSA in various contexts.

Figure 1.

Quantifying SSA and indels from kmo^RG transgene mutagenesis. (A) The 3.7-kb kmo^RG transgene with sgRNA cut sites A–D (orange arrows), I-SceI cut site, and engineered 684 bp direct repeat sequence of exons 3 and 2 that can result in transgene elimination by SSA (bottom left arrow) or indel mutations by NHEJ or MMEJ (bottom right arrow) after I-SceI or CRISPR/Cas9 DSBs. (B) Experimental design and timeline of measuring I-SceI or Cas9/sgRNA DNA repair events in hemizygous kmo^RG embryos.

Since prior studies used I-SceI to generate a DSB in the kmo^RG transgene to trigger SSA within the germline [12, 14], we decided to validate our embryo assay using I-SceI as the DSB-inducing nuclease, as it would allow us to compare our new, somatic embryo editing data to previous germline results. To do this, we microinjected a plasmid containing PUb-I-SceI to enable constitutive expression of the I-SceI homing endonuclease into pre-blastoderm A. aegypti embryos hemizygous for the kmo^RG transgene (Fig. 1). Hemizygous embryos were used as we wanted to limit repair events to SSA and the competing NHEJ/MMEJ pathways, whereas in homozygous embryos a DSB on one transgene could also trigger HR-based repair that could compete with all other repair types (Fig. 1A).

To detect each repair outcome, we employed two diagnostic PCR-based assays: an SSA assay to amplify both restored wild-type kmo (from SSA repair of kmo^RG) and kmo⁻ mutant alleles (the opposing allele not targeted by CRISPR/Cas9), and an indel assay amplifying the kmo^RG transgene. Since kmo⁻ null alleles were expected to amplify in all samples, but restored wild-type alleles were dependent on SSA repair, relative SSA rates could be obtained by dividing the number of wild-type kmo sequence reads from SSA by the total number of both kmo⁻ null and wild-type kmo sequence reads following nanopore deep amplicon sequencing and bioinformatics analyses (Fig. 1B). To ensure the validity of this approach, we confirmed in true genetic heterozygotes where one allele was kmo⁻ and one allele was a known germline-based SSA event that each allele was amplified and represented in equal frequencies (Supplementary Fig. S1).

Nanopore sequencing of SSA and indel assay amplicons from four replicates (n = 4) of 100 embryos injected with PUb-I-SceI revealed SSA rates ranging from 18% to 21% and indel rates ranging from 16.7% to 22.3% (Fig. 2). By dividing the SSA rates by the indel rates for each sample, we obtained relative SSA/indel ratios ranging from 0.93 to 1.15, in line with our previously reported germline SSA/indel ratios that were also ~1 [12]. Manual inspection of aligned reads did not reveal any evidence of template switching, and raising the required mapping quality (and hence alignment stringency) from 10 to 50 had no impact on rates of SSA (16.6%–20%) or SSA/indel ratios (0.89–1.1).

Figure 2.

SSA and indels from I-SceI mutagenesis of kmo^RG. (A) Mean frequencies of indel and SSA rates for negative control samples receiving no microinjection (left) and I-SceI injected samples (right) (n = 4). (B) SSA/indel ratio for I-SceI injected embryos. Error bars indicate standard deviation from mean frequencies, each dot represent sequencing data from a single group of 100 pooled embryos analyzed.

As we also showed that I-SceI and I-AniI homing endonuclease-induced DSBs differentially impacted SSA transgene elimination and indel mutation rate [14], we next sought to address how CRISPR/Cas9-induced DSBs and the location of the DSB site relative to the direct repeats could influence repair outcomes. To do this, we designed and synthesized 11 sgRNAs targeting 4 regions of interest along kmo^RG, henceforth referred to as sgRNA target sites A–D (Fig. 1A). Sites A–D differed in distance to the nearest direct repeat, with site A being 16–35 bp away from the proximal, upstream direct repeat (nearly 3.7 kb from the distal direct repeat), site B corresponding to the previously used I-SceI target sequence located 314–334 bp away (3.4 kb from the distal repeat), site C placed as far from each direct repeat as possible roughly 1.3 kb away (nearly 2.4 kb from the distal repeat; the true center of the construct lied in the polyubiquitin promoter controlling eGFP expression; we avoided targeting this region due to the presence of alternative target sites at the native polyubiquitin locus), and site D located 3 bp away from the downstream direct repeat (3.7 kb from the distal repeat). Each of the 11 sgRNAs were validated in vitro by incubating Cas9/sgRNA RNP complexes with a PCR amplicon encompassing their intended sgRNA target and visualizing the digested product using gel electrophoresis (Supplementary Fig. S2).

Nanopore sequencing of amplicons from two replicate experiments confirmed that SSA-based transgene elimination could be triggered by a Cas9-induced DSB (Fig. 3A). Critically, we discovered SSA could be triggered by Cas9/sgRNA complexes targeting site C, implying DNA end resection and homology searching can extend to a direct repeat located 1.3 kb away from a DSB site. While a few sgRNAs displayed almost no activity (A2, C1, C3) in either replicate, the overall editing rates of the majority of sgRNAs were variable between experiments. As NHEJ, MMEJ, and SSA are competing pathways, we calculated the ratio of SSA to indel events for each sgRNA. Indel events were shown to dominate across all sgRNA sites, occurring at frequencies 2–4 times higher than SSA events (Fig. 3B). While SSA/indel ratios varied between experiments, testing the association between SSA and indel frequency by linear correlation analysis revealed a strong correlation between the two rates in both experiments (R² = 0.865, P < .001; R² = 0.673, P = .00107, respectively), suggesting that editing activity measured by indel frequency was positively correlated to rates of SSA transgene elimination (Fig. 3C). We conclude that overall editing rates may depend more on variables that could not be controlled in this experiment such as the timing, developmental fate, and cell cycle stage of edited nuclei, but when editing did occur, both repair outcomes occur proportionally, with indel-based repair representing the majority of events and SSA representing a minority.

Figure 3.

SSA and indels from Cas9/sgRNA mutagenesis of kmo^RG. (A) Percentage of SSA and indel events for each sgRNA used compared to no treatment controls (–). (B) SSA and indel rates represented as a ratio (SSA/indel). (C) Linear correlation analysis of SSA and indel rates with linear equation, R², and P-value. Shaded area in between dotted lines represents 95% confidence interval (CI = 0.95).

In order to better differentiate indel events between NHEJ and MMEJ, we analyzed insertion/deletion sizes at each sgRNA site across both experiments. Indels caused by NHEJ are expected to occur randomly and should occur independently in each experiment. In contrast, MMEJ events should result in predictable and repeatable events dependent on the presence of microhomology surrounding the target sequence. Indel histogram data revealed profiles characterized by the enrichment of specific deletions unique to each sgRNA target site (Fig. 4). Sequence-level interrogation of each target site revealed the presence of small repeats, or microhomology, surrounding each sgRNA target site, which could serve as in cis homology for the error-prone MMEJ repair pathway (Table 1). Interestingly, we saw almost identical indel profiles between samples injected with sgRNAs D1 and D2, which targeted the top and bottom strands, respectively, but shared the same DSB site (Fig. 4). We next wanted to discern whether these MMEJ repair events could be attributed to specific kinetic properties of Cas9/sgRNA RNP injection or whether they were consistent with different CRISPR delivery methods. Injection of plasmid supplied Cas9 and sgRNAs targeting cut sites B2 and D1 revealed identical indel histograms to Cas9/sgRNA RNP experiments with the most common deletions characterized by flanking microhomologies (Supplementary Fig. S7).

Figure 4.

Indel frequencies from Cas9/sgRNA mutagenesis of kmo^RG. Indel events by size from Cas9/sgRNA targeting sites A–D (x-axis) plotted against their frequency among all indel sequencing reads (y-axis). Each graph represents two stacked replicates (R1 in white, R2 in light grey) of sequenced PCR amplicons from 100 pooled embryos injected with Cas9/sgRNA RNP mixture.

Table 1.

Microhomology-mediated deletions after Cas9/sgRNA mutagenesis of kmo^RG

sgRNA	Microhomology surrounding cut site	Mutation
A1	CCGGTGC GGC CGCATA GGC GCGCCTATATA	Δ9
A2	ATTAGACTGTA CG AT CG CCGGCGGGGGATC	Δ4
A3	ATAGGCGCGCCT AT AT AT TAGACTGTACGA	Δ4
A3	ATAGGCGCGCCTA T AT A TTAGACTGTACGA	Δ2
B1	CCATGGGT AGGG ATAAC AGGG TAATTAGGG	Δ9
B2	AGGGATAAC AGGGT AATT AGGGT GCGCTCC	Δ9
B3	CACCGGTCGCCACC A T GGGT A GGGATAACA	Δ6
C1	CGTAGGGGGCCTGC TT AATTAATCAGACTA	*
C2	TAGCGCGC GCGG TAGC GG GGGATCACTATC	Δ6
C2	TAGCGCGCGC GG TAGCG GG GGATCACTATC	Δ7
C3	TAATCAGACTA GCGC G CGC GGTAGCGGGGG	Δ4
C3	TAATCAGACTA GCG CGC GCG GTAGCGGGGG	Δ6
D1, D2	TGTATCTT GCCGCT GAGCCG GCCGCT AAAA	Δ12
D1, D2	TGTATCTT GCCG CTGA GCCG GCCGCTAAAA	Δ8

sgRNA cut site (underlined), predicted microhomology used for MMEJ recombination (bold), and deleted sequence (bold and italics) after Cas9/sgRNA mutagenesis of kmo^RG. * indicates low editing activity and lack of microhomology data.

A linear correlation analysis of indel size frequencies between first and second replicate indel data revealed a significant correlation for all sgRNAs, with R² values highest when editing rates were high (sgRNAs B2, C2, D1, and D2; Fig. 5A). Indel size frequencies between D1 and D2 were also strongly correlated (R² = 0.825, P < .0001, Fig. 5A). As a control, we performed a similar correlation analysis between sgRNAs with distinct target sites (A3 versus B2, B2 versus C2, and C2 versus D1), with no significant correlation found (Fig. 5B). To test whether these indel events were impacted by the absence of NHEJ, we injected Cas9/sgRNA targeting site D1, a target site with notable MMEJ-attributed indels, into a separate transgenic strain with either functional or nonfunctional copies of the NHEJ-promoting DNA ligase IV (lig4). Across three replicates, we found indel mutation profiles were identical between NHEJ competent lig4^+/+ and incompetent lig4^−/− injected samples and strongly correlated (Fig. 6; R² = 0.967, P < .0001). We conclude that indel formation in A. aegypti embryos is strongly influenced by local microhomology surrounding sgRNA target sites independent of NHEJ, and that as a result MMEJ is the dominant form of repair in this mosquito.

Figure 5.

Indel correlation analysis of replicates after Cas9/sgRNA mutagenesis of kmo^RG. (A) Linear correlation analyses of indel frequencies from first replicate (x-axis) and second replicate (y-axis) amplicon sequencing for each respective sgRNA, each displayed with linear equation, R² and P-value. Shaded area in between dotted lines represents 95% confidence interval (CI = 0.95). (B) No correlation control graphs from plotting indel frequencies of two different cut sites against each other.

Figure 6.

Indelfrequencies from Cas9/sgRNA-D1 mutagenesis of BR-KmoEx4 in Lig4⁺ and Lig4⁻ embryos. (A) Schematic representation of BR-KmoEx4. (B) Replicate-stacked histograms for Lig4⁺ and Lig4⁻ injected embryos (R1 in white, R2 in light grey, R3 in dark grey) of sequenced PCR amplicons from 100 pooled embryos injected with Cas9/sgRNA-D1 RNP mixture. Indel events by size (x-axis) are plotted against their frequency among all indel sequencing reads (y-axis). (C) Linear correlation analysis of Lig4⁺ (x-axis) and Lig4⁻ (y-axis) mutation frequencies (mutation frequencies represent average of three replicates for each indel in Lig4⁺ and Lig4⁻ samples, respectively) displayed with linear equation, R² and P-value. Shaded area in between dotted lines represents 95% confidence interval (CI = 0.95).

While we confirmed that SSA could occur from a Cas9-triggered DSB located 1.3 kb from the 684 bp kmo^RG allele direct repeats, we sought to determine whether a DSB further away could still initiate SSA-based transgene elimination. To test this, we made use of a second transgenic strain 1pc.-kmo^RGB, also integrated in the kmo locus [14]. In addition to the same 3xP3-DsRED and PUb-eGFP genes, this insertion also contains nos-I-SceI and 3xP3-mTagBFP2 cassettes, resulting in 7.9 kb of sequence flanked by the same 684 bp direct repeats from exons 3-2 used in kmo^RG experiments (Fig. 7A). While an I-SceI target site was present in the original construct [14], we selected a variant that had acquired an indel at the target site following an I-SceI induced DSB (Fig. 7B). We designed nine sgRNAs, referred to as sgRNAs E1-9, targeting the synthetic 3′ untranslated region (3′UTR#5) as it is located in the middle of the transgene nearly 3.5 kb from the nearest direct repeat (Fig. 7A), and confirmed their activity in vitro (Supplementary Fig. S3).

Figure 7.

The 1pc.-kmo^RGB transgene. (A) Schematic of the 7.9 kb 1pc.-kmo^RGB transgene that includes an inactivated I-SceI recognition site within DsRED and engineered 684 bp direct repeat sequence of exons 3 and 2 that can result in transgene elimination by SSA or indel mutations by NHEJ or MMEJ after CRISPR/Cas9 mutagenesis. Recognition sites for sgRNA E1-11 and sgRNA-D2 are placed 3.5 kb and 3 bp, respectively, from the nearest direct repeat. (B) Wild-type I-SceI recognition sequence (I-SceI⁺, top) compared to mutated recognition sequence used in 1pc.-kmo^RGB (I-SceI, bottom), I-SceI recognition sequence underlined, and insertions/substitutions indicated in red.

SSA assays from Cas9/sgRNA-E injected hemizygous 1pc.-kmo^RGB embryos revealed low SSA rates matched with high indel rates (Fig. 8A). A maximum SSA/indel ratio of 0.25 among the first set of replicates was observed and just 0.18 among the second (Fig. 8B). In contrast to the smaller transgene, no significant correlation between SSA and indel rates were observed from these two replicate experiments (R² = 0.112, P = 0.346; R² = 0.219, P = 0.173, respectively), indicating that SSA is not competitive when resection distances exceed 3 kb (Fig. 8C). Indel histograms at each of the nine sgRNA positions again indicated a strong MMEJ footprint (Supplementary Fig. S4), with a statistically significant correlation in indel size frequencies observed between replicates of the same sgRNA (Supplementary Fig. S5A) but not different sgRNAs (Supplementary Fig. S5B), with deletions again seeming to occur in a sequence-dependent manner likely caused by microhomologies that surround the sgRNA cut site (Supplementary Table S2).

Figure 8.

SSA and indels from Cas9/sgRNA mutagenesis of 1pc.-kmo^RGB. (A) Percentage of SSA and indel events for each sgRNA used compared to no treatment controls (–). (B) SSA and indel rates represented as a ratio (SSA/indel). (C) Linear correlation analysis of SSA and Indel rates with linear equation, R² and P-value. Shaded area in between dotted lines represents 95% confidence interval (CI = 0.95).

We next sought to answer whether the low SSA rates from Cas9/sgRNA-E were a result of the increased distance between target site E and the nearest direct repeat or due to the increased cargo size for SSA to remove. To address this, we injected Cas9/sgRNA-D2 into our hemizygous 1pc.-kmo^RGB embryos, as it was located immediately next to the direct repeat and its protospacer was conserved. Across three replicates, the SSA rates ranged from 1% to 3%, with 20%–50% indel rates, resulting in an average SSA/Indel ratio at 0.06 (Supplementary Fig. S6A and B). Interestingly, where microhomologies were conserved, we observed similar biases in indel frequencies (notably the same 5, 8, and 12 bp deletions observed in previous Cas9/sgRNA-D experiments, though new deletions were also observed as the sequence further outside the cut site was altered; Supplementary Fig. S6C and D). Taken together, no increase in 1pc.-kmo^RGB SSA/indel ratios were observed when sgRNAs were placed closer to direct repeats at site D (5 bp) compared to further away at site E (3.5 kb).

Discussion

Here, we developed and utilized a novel pooled embryo editing assay in A. aegypti to examine DNA break repair choices following cleavage by I-SceI nuclease or CRISPR/Cas9 RNP complexes. We employed PCR, nanopore sequencing, and CRISPResso2 for quantification of editing events—a method that has been reported to be as accurate as Illumina and Sanger-based sequencing methods while overcoming obstacles associated with these more traditional technologies such as amplicon size limitations, throughput capabilities, and higher run costs [30]. This method enabled collection of DNA repair events in as little as 24-h post-microinjection, eliminating the need for hatching, outcrossing, and screening of subsequent generations, a process that can take many months [12, 14]. While these rapid embryo assays include editing events in both somatic and germline cells, ratios of SSA/indels were found to be identical to prior experiments measuring only germline editing [12, 14].

We previously demonstrated SSA in A. aegypti utilizing homing endonucleases [12, 14]; here, we demonstrate that CRISPR/Cas9 can also trigger transgene deletion by SSA repair in vivo. Using CRISPR/Cas9 as an endonuclease to trigger DSB repair, we found SSA can eliminate at least 8.6 kb by removing the 1pc.-kmo^RGB transgene with its direct repeat. When comparing Cas9 to I-SceI, we observed lower SSA rates along with higher rates of indel formation, even where they shared the same target site. These data suggest that the nuclease supplied to induce DSBs influences the repair outcome, similar to our previous findings that Y2-I-AniI DSBs accumulated more indels than I-SceI, however it should be noted that while indels and SSA events are measured by these assays, perfect repair of DSBs are not detected by these methods [14]. One possible explanation for these differences could be attributed to the different types of DSB ends that Cas9 and I-SceI leave after cleavage, since I-SceI induces a 4 bp 3′-ssDNA overhang compared to blunt-end DSBs induced by Cas9 that may be advantageous for SSA by promoting longer resection of the break end. Alternatively, this could be explained by differences in DNA–protein binding kinetics of the two enzymes at their cleavage site, as Cas9 is known to reside at the break site after strand cleavage for an extended time compared to I-SceI, potentially occupying DNA binding sites used by SSA-promoting proteins for processes such as DNA end resection or homology searching of direct repeats [17]. Further, it is also possible that the delivery of nuclease (Cas9 protein versus plasmid-supplied I-SceI) could influence the DNA repair dynamics, however this did not seem to affect MMEJ editing outcomes by our indel analysis. Importantly, in this case transgene removal also restores wild-type alleles in a scarless manner to restore the native expression of the gene it once interrupted. Though 8.6 kb has been the most DNA removed by SSA in Aedes [14], yReMEDE by Chennuri et al. in D. melanogaster reported SSA removing over 17 kb by an I-SceI break 20 bp away from a 250 bp direct repeat [15], demonstrating the ability for SSA to remove much larger gene drive transgenes.

We also considered how DSB location influences the rate of SSA and other competing repair pathways such as NHEJ or MMEJ. Though prior SSA studies in yeast have reported that DSBs closer to direct repeats correlated to higher rates of SSA [20], across 11 sgRNAs that were targeted to various distances (3 bp, 16 bp, 314 bp, and 1.3 kb) from the nearest direct repeat sequences, we found no correlation between DSB distance from direct repeats and the rate of transgene elimination by SSA. Instead, our experiments targeting 1pc.-kmo^RGB implied that breaks induced by Cas9 both ~3.5 kb and 5 bp from the nearest direct repeat were able to trigger SSA in rare cases and were associated with much higher rates of indel formation. These data imply that cargo size, and not DSB site distance, may be the main factor limiting SSA frequency, though we do not know the upper limit of cargo size removable by SSA. Further, this suggests that in A. aegypti, resection distances of 1.3 kb or less are common, but only rarely exceed 3 kb. This contrasts with other organisms, as SSA could be triggered by DSBs 28 kb from the nearest direct repeat in mouse embryonic stem cells [21], 25 kb in S. cerevisiae [31], and up to 3.5 kb D. melanogaster [32]. However, it is not clear whether SSA is efficient at these distances, or simply just a rare outcome, similar to what we observed. With a better understanding of the relationship between target site and direct repeat, arrangements can be chosen that best support intended transgene removal goals.

Equally important to understanding the factors that influence SSA are those that influence competing, error-prone repair pathways such as NHEJ and MMEJ as these mechanisms can contribute to the failure of transgene elimination or the failure of gene drive. Our indel analysis with both RNP and plasmid-supplied Cas9/sgRNAs revealed that the most frequent deletion mutations are driven by target-site-specific repeats and are likely the result of MMEJ as they are independent of NHEJ. Despite the conservation of the MMEJ pathway across eukaryotes [33], this pathway has not been studied in mosquitoes and is often omitted as an important DSB repair mechanism compared to HDR and NHEJ. In human cell lines, higher frequency Cas9-induced indels were found to be flanked by microhomologies as short as 2 nt [34]. In mouse embryonic stem cells, high rates of MMEJ were found to be correlated with efficient Cas9 mutagenesis and larger indel sizes [35, 36]. In Drosophila, MMEJ has also been reported to be a prominent repair outcome at Cas9-induced breaks acting independent of the NHEJ pathway, and is dependent on DNA polymerase Q/θ [37]. As A. aegypti DNA polθ has not been characterized, it is unclear how much of the MMEJ observed in our study can be considered Theta-mediated end joining, and this remains an important future direction. In zebrafish and human cell lines, MMEJ has already been pursued to optimize precision gene editing [38]. Taken together, we suggest these findings can be applied to vector control to either improve site-specific knockout efficiency or optimize gene drive technology by avoiding repeat-rich target sites that can create resistance alleles by MMEJ.

Progress in the field of mosquito gene drive research has given rise to interest in modulating gene drive activity. While many proposed solutions aim to inactivate gene drive transgenes by targeting Cas9 to inhibit “homing” in the mosquito germline (i.e. CATCHA, ERACR, eCHACR, anti-CRISPR proteins) [39–41], these approaches would be employed after an initial gene drive release, which would require the development and approval of a second release of transgenic mosquitoes. In contrast, self-eliminating drives acting through the SSA DNA repair pathway could work through a single-transgenic construct containing both the drive and the self-elimination mechanism—theoretically requiring only a single deployment [11, 12, 14, 15]. Recently, a self-eliminating gene drive known as yReMEDE has been reported in D. melanogaster in which an autonomous homing gene drive targeting the yellow locus was eliminated over generations through SSA repair by inducing DSBs in cis with I-SceI [15]. Additional work is needed to test the performance of SSA-based self-elimination technology in the context of A. aegypti homing gene drives. Our work demonstrates proof of concept that deleting transgenes (at least 8 kb in size) using CRISPR/Cas9 is achievable, suggesting that self-eliminating homing gene drives harboring Cas9 encoding genes (~4 kb) and other gene cassettes are attainable. Our findings on SSA and MMEJ repair mechanisms provide new insights on how gene edits arise from Cas9-induced DSBs. Better understanding DNA repair processes can benefit the field of vector control by helping optimize gene drive technology, gene drive reversal and elimination mechanisms, and more generally, genetic experiments that utilize site-specific gene editing. As we build on what we know about cellular genetics, we can craft more creative, ideal solutions to complex problems in the fields of mosquito vector control and gene editing.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the National Institute of Allergies and Infectious Diseases, National Institutes of Health (AI148787 to Z.N.A. and K.M.M.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Funding to pay the Open Access publication charges for this article was provided by NIAID-funding indicated above.

Data availability

Nanopore sequencing data generated in this manuscript are available from the NIH sequence read archive under accession numbers (PRJNA1249513, PRJNA1250562, PRJNA1250585, and PRJNA1250599).