Management of Mendelian Traits in Breeding Programs by Gene Editing: A Simulation Study

Background Genotypes based on high-density single nucleotide polymorphisms have recently been used to identify a number of novel recessive mutations that adversely affect fertility in dairy cattle as well as to track conditions such as polledness. The use of sequential mate allocation strategies that account for increases in genomic inbreeding and the economic impact of affected matings may result in faster allele frequency changes than strategies that do not consider inbreeding and monetary losses. However, the effect of gene editing on selection programs also should be considered because gene editing has the potential to dramatically change allele frequencies in livestock populations. Methods A simulation program developed to evaluate dairy cattle breeding schemes was extended to include the use of clustered regularly interspaced short palindromic repeat (CRISPR), transcription activator-like effector nuclease (TALEN), and zinc finger nuclease (ZFN) technologies for gene editing. A hypothetical technology with a perfect success rate was used to establish an upper limit on attainable progress, and a scenario with no editing served as a baseline for comparison. Results The technologies differed in the rate of success of gene editing as well as the success rate of embryo transfer based on literature estimates. The number of edited alleles was assumed to have no effect on success rate. The two scenarios evaluated considered only the horned locus or 12 recessive alleles that currently are segregating in the U.S. Holstein population. The top 1, 5, or 10% of bulls were edited each generation, and either no cows or the top 1% of cows were edited. Inefficient editing technologies produced less cumulative genetic gain and lower inbreeding than efficient ones. Gene editing was very effective at reducing the frequency of the horned haplotype (increasing the frequency of polled animals in the population), and allele frequencies of the 12 recessives segregating in the U.S. Holstein population decreased faster with editing than without. Conclusions Gene editing can be an effective tool for reducing the rate of harmful alleles in a dairy cattle population even if only a small proportion of elite animals are modified.

Proportions of bulls and cows edited in different scenarios 0%, 1%, 10% (bulls); 0%, 1% (cows) edit_type 3 Technologies used for gene editing C, P, T, Z Time system clock time when the simulation is submitted, PID process identification reported by the operating system, C clustered regularly interspaced short palindromic repeats, T transcription activator-like effector nuclease, P hypothetical technology with perfect success rate, Z zinc finger nuclease Mate allocation 77 The modified Pryce scheme accounting for recessive alleles described by Cole [5] was used to 78 allocate bulls to cows in all scenarios. The selection criterion was the 2014 revision of the 79 lifetime net merit (NM$) genetic-economic index used in the United States [15]. For each herd, 80 20% of the bulls were randomly selected from a list of live bulls, and the top 50 bulls from that 81 group were selected for use as herd sires based on true breeding value (TBV). This produced 82 different sire portfolios for each herd and is similar to the approach of Pryce et al. [6].  The number of trials required to produce a live, gene-edited calf was determined for each of the 135 four editing technologies (Table 3) by computing the number of draws needed from a geometric 136 distribution to have a 99% probability of obtaining a success using the editing failure and 137 embryonic death rates as the probability of success. The total number of trials was the product of 138 the number of trials required for a successful edit and the number of trials needed for a 139 successful ET. Producing a calf of the desired sex was assumed to be possible through the use of 140 sexed semen, selection among the embryos in a flush, or other assisted reproductive technology.

141
Expected allele frequencies 142 The results for each scenario were averaged over 10 replicates. Observed changes in allele 143 frequency were compared against expectations, and expected allele frequencies in each 144 generation for lethal defects were calculated as in [16]: where pt is the frequency of the major allele at time t, qt is the MAF at time t, and t ranges from 1 148 to 20 years. The MAF at time 0 was used in each scenario for each recessive locus (

154
For each recessive locus in each scenario, observed allele frequencies were regressed on birth

155
year using the Python module Statsmodels version 0.6.1 ([18,19]) using the model: where yt is the frequency of a recessive locus at time t, b0 is the intercept, b1 is the regression ; therefore, success rates might be improved 173 through more rigorous ET protocols for edited embryos even when editing technologies differ.

174
Although the cost of producing gene-edited animals decreases as the technology becomes more 175 efficient, this study did not examine those differences because no data on actual costs of 176 production were publicly available.

178
The numbers of trials required to ensure a 99% chance of successfully editing embryos (Step 4) 179 and of getting a live calf on the ground following ET (Step 5) are in Table 3. Of the existing 180 technologies, CRISPR was the most efficient by a factor of ~7, requiring only 100 trials to 181 produce a live calf. ZFN was only a quarter as efficient, requiring 2240 trials to produce a live 182 calf. Although determining the actual cost of producing a gene-edited calf is difficult, $10,000 183 per animal seems reasonable [24]. Production costs would then range from $1 million (CRISPR) 0.1428** 0.006 0.0044** 0.000 0.1418** 0.006 0.0044** 0.000 Regression was for five different editing technologies over 20 years in scenarios where either the top 1% of bulls and 0% cows were edited or the top 10% of bulls and top 1% of cows were edited.CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect hypothetical technology with a perfect success rate, ZFN zinc finger nuclease; b regression coefficient; SE standard error a t test significance for the hypotheses that |blinear| = 0 and |bquadratic| = 0: *P < 0.01, **P < 0.001; bold terms did not differ from 0. b The software used for analysis displays only three digits, SE < 0.001 were truncated to 0.000, but actual values were not exactly 0.  The polled (hornless) state is dominant to the horned state. This discussion is focused on horned, 231 the recessive allele, to mirror the results and discussion for HH3 as well as findings of Cole [5].

232
Previous studies on breeding strategies for decreasing the frequency of the recessive (horned) 233 allele in dairy cattle (e.g., [5,28,29]) suggested that rates of change would be very slow, and a 234 number of authors have instead proposed selection directly on the polled locus or linked markers 235 (e.g., [30,31,32,33,34]). Long-term progress can be improved slightly by putting more weight on 236 favorable minor alleles in selection programs [35], but progress would be much faster using gene 237 edits for the favorable allele.

238
In this simulation, a single locus was assumed to control polledness, but in reality the polled 239 locus is more complex than HH3 and has at least two mutations on BTA1 that result in hornless 240 cattle [36,37]. All gene-editing methods resulted in significant rates of allele frequency change 241 (Table 4), with rates of change increasing with the efficiency of the technology (Fig. 6).

242
Regression coefficients were similar regardless of the proportion of bulls and cows edited. In 243 contrast to the results obtained for HH3, observed trends were greater than expected trends for 244 every editing technology (Fig. 7). Differences between methods were much greater than locus (not shown) were similar to those observed for HH3, again supporting that higher failure 250 rates will result in sampling more diverse pedigrees than would otherwise be the case.

260
The pattern for cumulative genetic gain was similar to that for rates of inbreeding. The Perfect 261 technology did not differ from no gene editing for either 1% bulls and no cows edited or 10% of 262 bulls and 1% of cows edited. However, CRISPR, TALEN, and ZFN all showed significantly 263 lower cumulative genetic gains (P < 0.01), with larger differences for less efficient technologies.

264
Similarly, the scenarios with higher rates of editing also had no differences for no gene editing 265 and the Perfect technology as well as significantly lower rates of gain for CRISPR, TALEN, and 266 ZFN (P < 0.01). As previously discussed, when many embryos die during ET, fewer elite 267 animals are available to become parents in the next generation. This resulted in lower rates of 268 genetic gain that were proportional to the ET failure rate over the course of the simulation. in a scenario had no effect on the rates of allele frequency change. This is expected because gene 275 edits are modelled as independent events, and few animals are carriers of more than one 276 recessive.

278
The proportion of embryos that died in each birth year because they were homozygous for lethal 279 conditions (Fig 8) also decreased rapidly when only 1% of bulls and no cows were edited using 280 CRISPR and TALEN. When the editing procedure is highly efficient, fewer affected embryos 281 are produced, even when the number of edited parents is few. The Perfect technology produced 282 rapid decreases in both scenarios as expected. Terms are nested within one another from left to right (e.g., the proportion of bulls edited is nested within the recessive scenario). CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease   Terms are nested within one another from left to right (e.g., the proportion of bulls edited is nested within the recessive scenario). CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease Figure 3 Observed minor allele frequency of Holstein recessive locus HH3 for five different gene-editing technologies over 20 years a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease Figure 4 Observed versus expected changes in allele frequencies of Holstein recessive locus HH3 for five different gene-editing technologies over 20 years a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease Figure 5 Average inbreeding rate in a simulation of 12 Holstein recessive loci for five different gene-editing technologies over 20 years a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease Figure 6 Observed minor allele frequency of the Holstein recessive locus horned for five different gene-editing technologies over 20 years a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease Figure 7 Observed versus expected changes in allele frequencies of the Holstein recessive locus horned for five different gene-editing technologies over 20 years a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease

Figure 8 Proportion of embryos that died each year because of the effects of recessive genotypes for five different gene-editing technologies over 20 years
a Top 1% of bulls and 0% of cows were edited. b Top 10% of bulls and top 1% of cows were edited. CRISPR clustered regularly interspaced short palindromic repeats, TALEN transcription activator-like effector nuclease, Perfect = hypothetical technology with a perfect success rate, ZFN zinc finger nuclease

Figure 9 Proportions of genotyped Holsteins that have known recessives by birth year
Bulls and cows were born from 2000 through 2015. Carrier status was either not a carrier of a known recessive or a carrier of one known recessive or more.