CRISPR/Cas9-induced gene conversion between ATAD3 paralogs.

Paralogs and pseudogenes are abundant within the human genome, and can mediate non-allelic homologous recombination (NAHR) or gene conversion events. The ATAD3 locus contains three paralogs situated in tandem, and is therefore prone to NAHR-mediated deletions and duplications associated with severe neurological phenotypes. To study this locus further, we aimed to generate biallelic loss-of-function variants in ATAD3A by CRISPR/Cas9 genome editing. Unexpectedly, two of the generated clones underwent gene conversion, as evidenced by replacement of the targeted sequence of ATAD3A by a donor sequence from its paralog ATAD3B. We highlight the complexity of CRISPR/Cas9 design, end-product formation, and recombination repair mechanisms for CRISPR/Cas9 delivery as a nucleic acid molecular therapy when targeting genes that have paralogs or pseudogenes, and advocate meticulous evaluation of resultant clones in model organisms. In addition, we suggest that endogenous gene conversion may be used to repair missense variants in genes with paralogs or pseudogenes.

The human genome contains ∼8,000 to 13,000 pseudogenes, depending on the annotation technology.1, 2, 3, 4, 5 These include processed pseudogenes, derived from retrotransposition of processed mRNAs and thus lacking introns, and unprocessed pseudogenes, derived from segmental duplications also known as low-copy repeats. Processed pseudogenes are the most abundant type, presumably due to a burst of retrotranspositional activity in ancestral primates.6, 7, 8

Pseudogenes have long been considered nonfunctional, and the term paralog is sometimes used to describe genes that are transcribed and translated,; however, increasing evidence suggests that numerous pseudogenes have some form of biological activity. Gene-pseudogene interactions can occur at the DNA, RNA, or protein level. For example, at the DNA level, CYP21A2P causes adrenal hyperplasia by gene conversion with CYP21A2. With regard to human genome rearrangements, glucocorticoid-remediable aldosteronism results from fusing the regulatory sequence from one paralog to another. The most common trait observed in humans, red-green color blindness, is due to recombination between the red and green opsin gene paralogs OPN1LW and OPN1MW. At the RNA level, PTENP1 increases PTEN expression by sequestering microRNA and can act as a tumor suppressor. At the protein level, NOTCH2NL genes activate NOTCH signaling by sequestering the inhibitory ligand DELTA, thus leading to expansion of the cortical progenitor population.,

Paralogs and pseudogenes can mediate non-allelic homologous recombination (NAHR). Three paralogs with extensive sequence homology are located in tandem at the ATAD3 locus on chromosome 1p36.33: ATAD3A, ATAD3B, and ATAD3C, juxtaposed within an ∼85 kb genomic interval (Figure 1A). This genomic architecture predisposes the region to genomic instability and reciprocal copy number variants (CNVs), deletions, and duplications. Biallelic deletions mediated by NAHR, most often spanning ∼38 kb between ATAD3B and ATAD3A and less frequently ∼67 kb between ATAD3C and ATAD3A, lead to an infantile-lethal presentation, including respiratory insufficiency, neonatal seizures, congenital contractures, corneal clouding and/or edema, pontocerebellar hypoplasia, and simplified sulcation and gyration., Deletions between ATAD3B and ATAD3A lead to a fusion transcript under regulation of the weaker ATAD3B promoter, and thus show decreased expression of an ATAD3B/ATAD3A fusion protein that presumably is sufficient for fetal development but apparently cannot sustain life beyond the neonatal period. The reciprocal, NAHR-mediated duplication at this locus, between ATAD3C and ATAD3A, results in a fusion gene encoding a dysfunctional protein.,

Figure 1

The ATAD3 locus and sequences of the clones showing gene conversion. (A) The three paralogs of the ATAD3 family (ATAD3C, ATAD3B, and ATAD3A) are situated in tandem, within an ∼85 kb genomic interval on chromosome 1p36.33. Scale bar indicates 10kb. (B) Upper and lower panels show the sequence of wild-type ATAD3A and ATAD3B at the location targeted for editing (the gRNA sequence and protospacer adjacent motif (PAM) site [CGG] are indicated in the upper panel; arrow points to the cut site). Middle panels show the sequence of clones 38 and 15. The exchange of 13 bp by gene conversion is outlined within the magenta box. A single nucleotide upstream to this region (highlighted by a light blue box) is identical to ATAD3A in clone 38, and to ATAD3B in clone 15, indicating that the gene conversion event occurred independently in these two clones.

In order to pursue in vitro functional studies, we used CRISPR/Cas9 gene editing to engineer knockouts of ATAD3A in HEK293T cells, with a guide specific to ATAD3A that would not be expected to edit ATAD3B or ATAD3C (Figure S1). Transfected cells were seeded into wells at a very low dilution, with the aim of obtaining clonal populations by limiting dilution and clonal expansion. From among 38 wells that showed growth, we sequenced 19 wells. Fifteen of these had mixed populations of at least two sequences. Four wells were clonal: one showed a homozygous wild-type sequence indicating lack of editing or otherwise repair by homologous recombination; another had a single base pair insertion, as expected to result from non-homologous end-joining repair (Figure S2). However, the other two clonal populations were unexpectedly found to have an in-frame 13 base pair (bp) sequence alteration at the cut site. This resulted in two missense variants: ATAD3A (RefSeq:NM_001170535.3): c.805_807delCTTinsGCC; p.(Leu269Ala) and c.811_813delGCCinsACT; p.(Ala271Thr), in addition to two synonymous variants that did not alter amino acids. Evaluation of the alteration revealed that a short sequence from either ATAD3B or ATAD3C had become embedded within ATAD3A, i.e., gene conversion (Figure 1B).

To ensure that the amplicon was amplified specifically from ATAD3A, and did not amplify ATAD3B or ATAD3C, we used two independent pairs of primers (Supplemental methods, Figure S1). We also sequenced ATAD3B and ATAD3C at the altered site in the HEK293T unedited cells, to verify that there were no single nucleotide polymorphisms (SNPs) at the ATAD3B or ATAD3C homologous loci (Figure S3). Based on a SNP 8 bp downstream of the conversion event, which is different in ATAD3C, we implicated ATAD3B rather than ATAD3C as the likely donor, although this origin cannot be proved for certain.

Multiple sequence alignment of the two “converted” clones (clone 15 and clone 38) as compared with the three ATAD3 paralogs, indicated that the two clones had risen independently, as they differed at a position 24 bp upstream of the 13-bp conversion (Figure 2). The flanking regions of the 13-bp nucleotides are identical in the paralogs; therefore, it is not possible to determine the exact length of the conversion tract. For clone 38, the minimum of the donor tract is 13 bp and the maximum is 104 bp (calculated as the distance between the next upstream and downstream nucleotides that are unique to ATAD3A and not shared with ATAD3B). For clone 15, the minimum of the donor tract is 37 bp and the maximum is 181 bp (Figure 2). These ranges are consistent with the accepted mean length of 55 to 290 bp for conversion tracts. Characterization of gene editing events by the Inference of CRISPR Edits (ICE) software, revealed that 14% to 25% of the population had been repaired by gene conversion (Figures 3 and S4). To check whether CRISPR/Cas9 had generated on-target deletions or duplications at the NAHR-prone ATAD3 locus, we subjected the pool of edited cells to an Affymetrix CytoScan HD chromosomal microarray and compared the result with unedited HEK293T cells. No CNVs were observed (Figure S5).

Figure 2

Multiple sequence alignment. The sequence of ATAD3A can be clearly recognized on either side of the gene conversion event. Red stars are unique to ATAD3A; blue stars are unique to ATAD3B/3C. Nucleotides in gold font are unique to either ATAD3B or ATAD3C.

Figure 3

Relative percentages of edited populations. The relative contribution of each sequence to the initial population that underwent CRISPR/Cas9 editing was quantified by the ICE software (Synthego). Note that this software cannot differentiate between clones 15 and 38, such that 16% represents all clones that underwent gene conversion. Other estimates of the ICE software indicated gene conversion of 14% to 25% and are shown in Figure S4. The “N” in the upper sequence indicates a single nucleotide insertion, insT; the Sanger sequence is provided as Figure S2.

CRISPR systems have significant therapeutic potential, and much research is focused on minimizing off-target effects. Reports of on-site unexpected effects are less emphasized. “Bystander” deletions and duplications adjacent to the on-target cut site, complex rearrangements, and loss of the targeted chromosome have all been demonstrated; such events may escape detection by routine targeted genotyping assays.23, 24, 25, 26 In addition, unanticipated pseudogene-directed homology repair has been reported between CD33 and its pseudogene SIGLEC22P. The authors emphasized the need for sequencing to detect CRISPR effects, rather than relying solely on changes in PCR fragment size, gene expression, or splicing patterns.

Here, we show that CRISPR/Cas9 can lead to gene conversion, replacing a minimum of 13 nucleotides at the expected target site and resulting in two amino acid alterations in ATAD3A. The recognized pathogenic mutational spectrum at the ATAD3 locus is largely driven by NAHR-mediated CNVs, generated by the DNA double-strand break repair (DSBR) pathway.,, Gene conversion results from synthesis-dependent strand annealing or from an alternate resolution of the double Holliday junctions by the error-prone break-induced replication (BIR) repair pathway, and may be mediated by template switches. The migrating bubble-like replication fork results in conservative inheritance, with replacement of a short tract of the genome by a copy of the donor tract., Half crossovers can form from interruption of BIR. It is fathomable that gene conversion occurs at the ATAD3A locus, possibly introducing benign variation, such as multisite variants (MSVs) polymorphic across paralogous sequences and paralogous sequence variants., The genomic distance between the ATAD3 paralogs (total of ∼85 kb) is within the cutoff of < 250 kb, which includes most known NAHR events, as has been empirically shown for Alu-Alu rearrangements. In addition to CNVs, missense variants have been identified in ATAD3A, and it would be intriguing if these could be repaired by gene conversion from ATAD3B or ATAD3C.

Intentional NAHR or gene conversion between paralogs can have potential therapeutic benefit. CRISPR/Cas9-induced recombination between paralogs of the nucleotide-binding-site-leucine-rich-repeat (NBS-LRR) gene family in soybeans has been shown to accelerate disease resistance, by leading to novel chimeric paralogs with intact open reading frames. In human cells, efficient gene conversion of SMN2 to SMN1 was shown to be induced by CRISPR/Cpf1 homology-directed repair and single-stranded oligodeoxynucleotides. In addition, CRISPR/Cas9-induced gene conversion between HBD and HBB, encoding human hemoglobin subunits, has been documented; this correlated with HBB insertion and deletion rates, supporting gene conversion rather than PCR-mediated sequence shuffling between highly homologous sequences., Indeed, spontaneous gene conversion between other paralogs of the β-globin cluster was the first instance of gene conversion recognized in the human genome. CRISPR/Cas9 genome editing can also generate micronuclei and chromosome bridges initiating chromothripsis on the targeted chromosome. Fascinatingly, McDermott et al. reported a fortuitous cure of WHIM syndrome by chromothripsis, which resulted in deletion of the disease-causing gain-of-function allele.

Our report highlights the complexity of CRISPR/Cas9 design and recombinational repair mechanisms when considering both delivering nucleic acid therapeutics and when dealing with genes that have paralogs and pseudogenes. The specific mechanism of repair remains to be explored: whether the CRISPR/Cas9-generated event is repaired exclusively by a DSBR/homologous recombination mechanism (as shown in Figure S6), or whether it may also involve post-replication repair of one-ended, double-stranded DNA (oeDNA) generated at a collapsed fork, the latter occurring through a BIR/microhomology-mediated BIR pathway that could result in a half-crossover or CNV.

These data also suggest experimental use of CRISPR/Cas9 in model organisms warrants meticulous evaluation of the resultant clones. On a complementary or reciprocal note, our findings foment interest that pathogenic variants in such genes may be repaired by mediating gene conversion intentionally. Further studies focused on the optimal characteristics of paralogs/pseudogenes for gene conversion are called for, to determine feasibility of such an approach. These include determining the optimal distance between paralogs/pseudogenes for CRISPR/Cas9-mediated gene conversion, whether specific sequence motifs enhance gene conversion rates, if intrachromosomal recombination events are preferred over interchromosomal events, and if altering levels of proteins that mediate recombination or introducing CRISPR/Cas9 editing at different stages of the cell cycle can influence gene conversion efficiency.

Data and code availability

The study did not generate datasets or code.