Improved methods and optimized design for CRISPR Cas9 and Cas12a homology-directed repair

CRISPR-Cas proteins are used to introduce double-stranded breaks (DSBs) at targeted genomic loci. DSBs are repaired by endogenous cellular pathways such as non-homologous end joining (NHEJ) and homology-directed repair (HDR). Providing a DNA template during repair allows for precise introduction of a desired mutation via the HDR pathway. However, rates of repair by HDR are often slow compared to the more rapid but less accurate NHEJ-mediated repair. Here, we describe comprehensive design considerations and optimized methods for highly efficient HDR using single-stranded oligodeoxynucleotide (ssODN) donor templates for several CRISPR-Cas systems including S.p. Cas9, S.p. Cas9 D10A nickase, and A.s. Cas12a delivered as ribonucleoprotein complexes with synthetic guide RNAs. Features relating to guide RNA selection, donor strand preference, and incorporation of blocking mutations in the donor template to prevent re-cleavage were investigated and were implemented in a novel online tool for HDR donor template design. Additionally, we employ chemically modified HDR donor templates in combination with a small molecule to boost HDR efficiency up to 10-fold. These findings allow for high frequencies of precise repair utilizing HDR in multiple mammalian cell lines. Tool availability: www.idtdna.com/HDR


25
CRISPR-Cas systems have revolutionized genomics by enabling efficient and precise genome editing 26 in a wide variety of biological systems, including eukaryotic cells. 1-5 These systems require an RNA-27 guided DNA endonuclease and a target-specific guide RNA (gRNA) to generate a double-stranded 28 break (DSB) at a desired genomic location, which must be flanked by a short protospacer adjacent PAM, as illustrated in Figure 1A. ssODNs were delivered to Jurkat and HAP1 cells along with their 122 respective Cas9 RNP complex by nucleofection, and the editing frequencies were assessed by next 123 generation sequencing (NGS). Perfect HDR, defined as the precise insertion of the EcoRI sequence 124 at the canonical cut site and otherwise maintaining the WT sequence, was quantified and 125 comparisons were made between donor templates consisting of either the targeting strand (T), which 126 is complementary to the CRISPR-Cas9 gRNA, or the non-targeting strand (NT), which contains the 127 'NGG' PAM sequence. In Jurkat cells there was no statistical difference (p>0.05, paired t-test) in total 128 editing when either the T or NT strand was used. However, a significant difference in editing efficiency 129 (p<0.0001, paired t-test) was observed in HAP1 cells where the mean editing was 80.2% when the 130 NT strand was used and 67.8% when the T strand was used, indicating that the T strand may bind to 131 the Cas9 RNP complex and reduce overall editing efficiency within the cellular environment, as 132 suggested by others (Supplemental Figure 1A). 36 As demonstrated in the top two panels of Figure 1B, 133 the strand that leads to higher frequencies of HDR varies depending on the genomic locus and cell 134 type being used. HAP1 cells had HDR frequencies ranging from 0 to 51.1% with a significantly higher 135 mean HDR frequency when the NT strand was used (20.6%) than the T strand (15.2%) (p<0.0001, 136 paired t-test), likely due to the reduced total editing when the T strand was used. In contrast, we 137 observed significantly higher mean HDR frequencies in Jurkat cells when the T strand was used than 138 when the NT strand was used (11.3% vs 7.5%, respectively) (p<0.0001, paired t-test). Overall, HDR 139 efficiency in HAP1 cells was higher than in Jurkat cells, with mean HDR frequencies of 17.9% and 140 9.4%, respectively. In addition, the bottom two panels of Figure 1B show that although efficient total 141 editing is required for HDR to occur, high editing does not always lead to high HDR insertion at each 142 site tested. For example, even though 53% of the sites tested in HAP1 cells and 74% of the sites 143 tested in Jurkat cells had >90% total editing, there is a broad range of HDR frequencies which varied 144 from 0 to 60% among these highly edited loci for both the NT and T strands. This emphasizes the 145 value of testing multiple guides to determine which have the highest potential HDR frequency prior to 146 any experiment where precise genome modification by HDR is desired. 147 mutation to the cut site, we selected 13 guides flanking the stop codon of GAPDH to determine which These guides had cut sites that ranged from 2 to 22 bases from the desired insertion position ( Figure   151 1C). The available guides in the nearby region included PAMs on both strands of the genomic DNA, 152 and ssODNs for both the targeting and non-targeting strand were designed and tested in K562 and 153 HEK293 cells for their ability to mediate HDR. As shown in Figure 1D in K562 cells, guides with low 154 editing efficiency yielded low HDR insertion, even if the cut site was close to the desired insertion 155 location. For example, the guide that cuts two bases from the desired insertion (-2) had 32.1% total 156 editing of which 12.6% was HDR insertion (NT strand). Similarly, the guide that cuts five bases from 157 the desired insertion (-5) had 10.7% total editing and only 1.1% HDR insertion. In contrast, the guides 158 that cut 14 and 6 bases from the desired insertion (-14, +6) had 96.0% total editing of which 40.7% 159 was HDR insertion (NT strand) and 87.8% total editing of which 39.7% was HDR insertion (NT 160 strand), respectively. As determined by NGS, these guides had the highest total editing and HDR 161 insertion rates, even though they were further from the desired insertion. This case study indicates 162 that guide efficiency is a critical factor for efficient HDR, and guide selection that is as close as 163 possible to the desired HDR mutation is a secondary consideration. This was also observed in 164 HEK293 cells, where a guide that cuts 6 bases from the insertion (+6: 97% total editing, 34% HDR) 165 led to higher HDR than guides 2 or 5 bases from the insertion (-5: 34.7% total editing, 12.7% HDR; -2: 166 62.2% total editing, 22.4% HDR). This effect was less prominent using guides further from the desired 167 insertion site (e.g. -14) in HEK293 cells, which may be due to differences in the available repair 168 machinery and capacity for HDR in each cell type (Supplemental Figure 1B).

169
We performed a similar experiment at a second genomic locus (TNPO3) in HEK293 cells to further 170 examine factors influencing HDR (Supplemental Figure 1C). The total editing was high for nearly all 171 guides tested in this experiment. However, for a guide with a cut site 9 bases from the desired 172 insertion location (-9) the total editing was 92.6%, and this site yielded reduced HDR efficiency of 173 7.0% compared to the guide that cut one base further from the desired insertion (-10) with an 174 increased 98.5% total editing that also gave an increased HDR insertion frequency of 24.2% with the 175 NT strand, further supporting that guide activity can be more impactful on HDR efficiency than optimal 176 positioning with respect to the cut site.

186
Consistent with other reports utilizing Cas9 nickase variants expressed from a plasmid, 17,21 we found 187 a higher rate of indel formation when D10A and H840A nickases were designed in a PAM-out 188 orientation, and the nickases must be placed with optimal spacing between the nick sites to mediate 189 efficient editing (Supplemental Figure S2B) spaced 51-nt apart, and this was reduced to 34.6% when the distance between the nicks was 199 decreased to 46-nt. We have investigated distances smaller than 40-nt between the nicks in PAM-out 200 orientation for Cas9 D10A and found that editing was poor for spacing <35-nt, likely due to steric 201 hindrance between the two RNP molecules (data not shown).

202
When Cas9 D10A nickase RNP complexes targeting both strands in the PAM-out orientation nick the 203 genomic DNA, a DSB with 5' overhangs is generated. Because both strands are targeted by one of 204 the two gRNAs, there is no canonical 'targeting' and 'non-targeting' strand in nickase experiments; 205 thus, they are referred to as top and bottom strands. The paired-guide double nicking strategy doesn't 206 generate a blunt-ended cut like WT Cas9, so we further explored the possibility of using Cas9 D10A 207 8 to insert exogenous sequences between flanking nick sites at a location that would be otherwise 208 considered sub-optimal for WT Cas9 designs using WT Cas9 and either gRNA on its own. We

225
In order to verify that the above observations are not site specific, we conducted a similar experiment 226 at a different locus (AAVS1, PAM-out design with 46-nt spacing). In addition to 5 insert positions at or 227 between the two cleavage sites, we also included two positions 12-nt upstream or downstream to the 228 left or right cleavage sites, respectively (Supplementary Figure S2D). Consistent with the results 229 described above, Cas9 D10A outperformed WT Cas9 at the position centered between the cleavage 230 sites (position "D") in HEK293 cells (Supplementary Figure S2E)  if there is a preference for transversions (e.g. G-to-C purine to pyrimidine conversion), or transitions 246 (e.g. G-to-A purine to purine conversion) in the blocking mutations used. We aimed to further 247 investigate this to define a ruleset for the placement and number of blocking mutation(s) required to 248 maximize HDR efficiency. First, we designed an experiment to determine the effect of a single 249 blocking mutation within the PAM or the seed region of Cas9, which is defined as the PAM-proximal

280
We next wanted to investigate this effect when a larger HDR mutation is inserted, such as an EcoRI

303
This data represents a subset of HDR donor designs that we tested to fully elucidate a ruleset for the 304 placement and number of blocking mutations required for various HDR mutation types. Using HDR 305 efficiency results from Figure 3B, we generated relative HDR efficiencies (i.e., HDR efficiency with    across three biological replicates for each of the three genomic loci tested is shown in Figure 4B (for 344 each ssODN design n = 9). For PAM-distal insertions, the NT strand had an average HDR of 12.7% 345 with repair track mutations compared to 1.6% when the T strand was used. In contrast, for PAM-346 proximal insertions, the T strand containing repair track and PAM mutations gave higher HDR than 347 the NT strand with the same mutations (8.6% vs 0.8%, respectively). For PAM-proximal insertions, the 348 repair track mutations marginally improved the HDR efficiency above incorporating a PAM mutation 349 alone, increasing the HDR from 7.5% to 8.6%. However, for PAM-distal insertions, the repair track 350 mutations significantly improved the frequency of HDR 3.4-fold over having only a PAM mutation (p 351 <0.01, paired t-test).

352
To further investigate strand preference when HDR mutations are placed at suboptimal distances 353 (>15-nt) away from the Cas9 cleavage site, 12 loci from the set of 254 targets presented in Figure 1B 354 were selected as a subset of gRNAs to carry out this experiment. These gRNAs were selected as 355 sites for HDR because they demonstrated one of three characteristics: no strand preference, an 356 obvious strand preference for the T strand, or an obvious strand preference for the NT strand in either 357 optimal insertion frequencies, a single PAM mutation alone was only tested for PAM-proximal 361 insertions. In addition, repair track mutations were incorporated every 3-7 nt between the Cas9 362 cleavage site and the desired HDR mutation ( Figure 4C). These ssODN donor templates were 363 delivered to Jurkat and HeLa cells along with their respective Cas9 RNP complexes by nucleofection, 364 and the frequency of perfect HDR was determined by NGS with the mean HDR rate for each ssODN 365 across the 12 genomic loci shown in Figure 4D. Across all 12 sites tested, the NT strand gave higher 366 HDR than the T strand for PAM-distal insertions, and the T strand gave higher HDR than the NT

377
Cas12a is a type II CRISPR-Cas nuclease with several distinct differences to Cas9. Cas12a 378 generates a DSB with 5' overhangs, requires a 'TTTV' PAM, and enables editing in AT-rich 379 genomes. 4 We designed experiments to characterize HDR design rules for Cas12a in a manner 380 similar to what was done with Cas9. First, the optimal placement of an insertion was determined by 381 designing donor templates for five sites in the HPRT1 gene. These donor templates placed an EcoRI 382 restriction digest recognition site at varying positions relative to the PAM and guide sequence ( Figure   383 5A), ranging from 9 bases away in the 5' direction from the first base of the guide to 45 bases 3' of the 384 first base of the guide. The optimal HDR activity for this insert is not centered around the two Cas12a 385 cleavage sites, canonically positioned 18 and 23 bases from the PAM, as was the case for Cas9. other indels from NHEJ repair. To investigate this possibility, we performed NGS analysis of one of 391 the five sites from Figure 5B to examine the frequency of perfect HDR insertion relative to imperfect 392 HDR insertion. At position 24, while the amount of EcoRI insertion was 6.4% by EcoRI cleavage 393 (Supplemental Figure 5A), the amount of perfect HDR when measured by NGS is <1% and the 394 imperfect HDR, which includes HDR insertion of an EcoRI site plus subsequent indels from NHEJ due 395 to Cas12a re-cleavage, was 5.9% (Supplemental Figure 5B). Thus, we confirmed by NGS that the 396 optimal position for Cas12a-mediated HDR is between positions 12-16 of the guide and moving an 397 insertion outside of the protospacer can give the desired insertion, but also allows for additional 398 undesired editing.

399
To investigate if Cas12a demonstrates a universal strand preference when an EcoRI insertion was 400 optimally placed, a set of 15 Cas12a guide RNAs was selected and donor templates were designed to 401 insert an EcoRI restriction digest recognition site 16 bases 3' of the PAM. Both the T and NT strand 402 ssODN donor templates were delivered with their respective RNP complexes to Jurkat and HAP1 403 cells by nucleofection, and NGS was used to measure the frequency of total editing and perfect HDR.

404
The combined results from the fifteen sites comparing T and NT strand donors in two cell lines is 405 shown in Figure 5C. Although there are differences in total editing across the 15 sites tested (varying 406 from 30% to >95% total editing which indicates inherent, guide or locus-dependent editing outcomes), 407 universally the total editing was lower when the T strand was used. This is shown in figure 5C, top 408 panel by the data points generally clustering below the line through the origin or showing increased 409 total editing when delivered with the NT strand. The reference line through the origin is included as a 410 benchmark in both panels for 5C to indicate the point at which T strand total editing or HDR is 411 equivalent to NT strand total editing or HDR, respectively. As a result of the discrepancy observed in 412 favor of the NT strand mediating increased total editing (top panel of 5C), the frequency of HDR was 413 also lower when the T strand was used as the donor template than when the NT strand was used.

414
These results demonstrate a statistically significant preference for the use of the NT strand as the 415 donor template to achieve optimal results in HDR experiments using Cas12a nuclease.

462
Finally, we investigated if the use Alt-R HDR Enhancer V1 in combination with modified donor 463 templates improved HDR further than if just one of these reagents was used. In this experiment Alt-R 464 HDR Enhancer V1 was used, although we have observed similar results with Alt-R HDR Enhancer V2 465 (data not shown). We found that the maximal HDR efficiency was achieved when Alt-R modified HDR 466 donor templates were used and cells were incubated with Alt-R HDR Enhancer V1 ( Figure 6C)

506
The availability of efficient Cas9 guides near a desired mutation is a significant limitation for many 507 HDR experiments. In many cases, the guide or guides closest to the desired HDR mutation are sub-508 optimal in terms of cleavage efficiency or proximity. While using a paired-guide nickase strategy is 509 viable, the requirement of having two guides with optimal spacing, activity, and orientation limits the    provides the option to add silent blocking mutations using our empirically defined ruleset. In our study, 532 we identified no bias in which alternate base was used as the blocking mutation to prevent Cas9 re-533 cleavage, indicating that there is flexibility in designing appropriate silent blocking mutations so as to specific and with a larger data set potential differences between alternate bases used for silent 536 blocking mutations could be resolved. Further investigation into the optimized number and placement 537 of blocking mutations with Cas12a is underway with the expectation that this will be built into a tool for 538 Cas12a HDR donor template design.

539
After a CRISPR-Cas system and gRNA(s) have been selected and the donor template has been 540 designed, the next consideration for HDR experiments is the selection of homology arm lengths. In 541 previous work investigating HDR improvements in the cell lines mentioned above, asymmetric 542 homology arms did not improve HDR beyond symmetric homology arms when arm length was ≥30-nt 543 from both the mutation location and the Cas9 cleavage site (data not shown). As such, the standard 544 approach we employ is to design ssODN donor templates with 40-nt homology arms. The Alt-R HDR 545 Design Tool allows for custom homology arm lengths to accommodate asymmetric designs, if desired.

546
A final donor template design consideration that we investigated was strand preference for the donor 547 template. Cas9 D10A nickase did not demonstrate a strong strand preference, so testing both strands 548 to determine which results in the highest HDR frequency may be prudent. However, for WT Cas9 the 549 preferred strand is strongly dependent upon where the desired HDR mutation is, relative to the Cas9 550 gRNA. Previous reports have demonstrated that when using ssODN donor templates with Cas9 551 nuclease the SDSA mechanism of repair is preferentially utilized, which consists of two steps. 38 After 552 a DSB is generated, the ends are resected, generating 3' overhangs which are then available for base 553 pairing with the donor DNA. This donor DNA then serves as a template for 5' to 3' DNA synthesis.

554
Although we observed no universal donor strand preference in the experiment outlined in Figure 1B  DSB repair using ssODN donor templates. For mutations directly at the cut site, we provide some 561 evidence that the use of the T strand may reduce total editing with Cas9 which negatively impacts 562 HDR, but this was not the case for both cell types tested. Using Cas12a, we observed a reduction in 563 total editing rates when the T strand was used universally. We hypothesize that the donor template 564 20 acts as a sponge for RNP, reducing the concentration available for genome editing within cells, or 565 activates the non-specific ssDNase activity of Cas12a. The NT strand conferred increased HDR for 566 experiments with Cas12a over the T strand. However, the effect of HDR insertion placement has not 567 been thoroughly investigated for Cas12a to determine if the T strand will be advantageous over the 568 NT strand for PAM-proximal mutations in a manner similar to Cas9 and further experimentation is

668
On-target editing and HDR efficiencies were also measured by NGS. Libraries were prepared using 669 an amplification-based method as described previously 53 . In short, the first round of PCR was

684
The data collected from experiments were analysed on Graph PadPrism 8 using two-tailed unpaired t-685 test to evaluate significance (*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001).     Nuclease complexed with Alt-R CRISPR-Cas9 crRNA and tracrRNA) targeting 3 genomic loci along with 0.5 µM HDR donor template and 2 µM Alt-R Cas9 Electroporation Enhancer by nucleofection. Donor templates were unmodified, PS modified, or Alt-R modified. Immediately after electroporation, cells were plated in media with or without 30 µM Alt-R HDR Enhancer (V1) and media was changed after 24 hours. HDR efficiency was measured by NGS. Data are represented as means ± S.E.M of three biological replicates. Figure 1 (A) EcoRI restriction digest recognition site (GAATTC) was inserted at the Cas9 cleavage site of 254 genomic loci in Jurkat and 239 genomic loci in HAP1 cells using either the targeting (T) or nontargeting (NT) strand as the donor template. RNP complexes (Alt-R S.p. Cas9 Nuclease complexed with Alt-R CRISPR-Cas9 crRNA and tracrRNA) were delivered at 4 µM along with 4 µM Alt-R Cas9 Electroporation Enhancer and 3 µM donor template by nucleofection. Total editing was assessed via NGS. (B) Insertion of an EcoRI site before the stop codon of GAPDH in HEK293 cells using guides around the desired HDR insertion location. The cleavage sites and associated distance to the desired insertion location for each guide are indicated above the sequence shown. Both the T and NT strand were tested. RNP complexes (Alt-R S.p. Cas9 Nuclease complexed with Alt-R CRISPR-Cas9 crRNA and tracrRNA) were delivered at 2 µM along with 2 µM Alt-R Cas9 Electroporation Enhancer and 2 µM donor template by nucleofection. HDR and total editing were assessed via NGS. Data are represented as means ± S.E.M. of three technical replicates. (C) Insertion of an EcoRI site at the TNPO3 locus in HEK293 cells using guides around the desired HDR insertion location. The distance from each cleavage site to the desired insertion location for each guide are indicated on the x-axis. Two pairs of guides cut at the same location, but on opposite strands. The strand containing the guide is indicated as top or bottom (btm). Both the T and NT strand were tested. RNP complexes (Alt-R S.p. Cas9 Nuclease complexed with Alt-R CRISPR-Cas9 crRNA and tracrRNA) were delivered at 2 µM along with 2 µM Alt-R Cas9 Electroporation Enhancer and 2 µM donor template by nucleofection. HDR and total editing were assessed via NGS.