RNA-guided gene drives can efficiently bias inheritance in wild yeast

Inheritance-biasing elements known as “gene drives” may be capable of spreading genomic alterations made in laboratory organisms through wild populations. We previously considered the potential for RNA-guided gene drives based on the versatile CRISPR/ Cas9 genome editing system to serve as a general method of altering populations1. Here we report molecularly contained gene drive constructs in the yeast Saccharomyces cerevisiae that are typically copied at rates above 99% when mated to wild yeast. We successfully targeted both non-essential and essential genes and showed that the inheritance of an unrelated “cargo” gene could be biased by an adjacent drive. Our results demonstrate that RNA-guided gene drives are capable of efficiently biasing inheritance when mated to wild-type organisms over successive generations.


Introduction
Gene drives have the potential to address diverse ecological problems by altering the traits of wild populations. As 'selfish' genetic elements, they spread not by improving the reproductive fitness of the organism, but by increasing the odds that they themselves will be inherited. Because this inheritance advantage can counteract the fitness costs associated with the drive itself or with adjacent genes carried along with it, they are theoretically capable of 'driving' unrelated traits through populations over many generations 1 .
Inheritance-biasing is a common strategy in nature 2 . One elegant class of inheritance-biasing genes spread by cutting homologous chromosomes that do not contain them, thereby inducing the cellular repair process to copy them onto the damaged chromosome by homologous recombination (Fig. 1A). This process is known as 'homing' 3 . The best-known homing endonuclease gene is I-SceI, whose product cuts the gene encoding the large rRNA subunit of S. cerevisiae mitochondria. Nearly all natural homing endonucleases are embedded within self-splicing introns or inteins, which help to minimize disruption of the targeted gene. Most are capable of homing with extremely high efficiencies; I-SceI is correctly copied 99% of the time 4 . (A) Homing endonucleases cut competing alleles, inducing the cell to repair the damage by copying the endonuclease gene. (B) By converting heterozygous germline cells into homozygotes containing two copies (teal), gene drives increase the odds that they will be inherited and consequently spread themselves and associated changes through wild populations (grey). Austin Burt first suggested that homing endonucleases might be used to construct synthetic gene drives capable of altering wild populations of multicellular organisms in 2003 (Fig. 1B) 5 . Laboratories subsequently reported that I-SceI endonuclease genes inserted into mosquitoes 6 or fruit flies 7,8 exhibited homing in transgenic laboratory populations with an I-SceI recognition site inserted into the corresponding wild-type locus. However, the versatility of gene drives based on homing endonucleases is constrained by the difficulty of retargeting these enzymes to cleave useful sequences within native chromosomal loci 9 .
The recent development of the CRISPR nuclease Cas9, which cleaves target sequences specified by "guide RNA" molecules, has enabled scientists to edit the genomes of diverse species [10][11][12][13][14][15][16][17][18][19][20] . The question of whether Cas9 can be used to build efficient gene drives is highly relevant due to the potential for such constructs to alter any gene in any population that can be edited with Cas9, a group that may constitute every species tested to date 21 . We previously detailed the theoretical potential for RNA-guided gene drives to alter wild populations, including an evolutionary analysis, novel drive architectures, and containment measures robust to human error 1 . However, these new architectures and control strategies have not yet been validated. Indeed, whether Cas9 can bias inheritance in a eukaryotic organism at all remains unknown, raising the question of whether initiating public discussions and engaging in regulatory reform 22 is truly necessary. We addressed these issues in the yeast S. cerevisiae by building gene drive constructs in haploid cells, mating them to wild-type haploid strains, and measuring the relative abundance of unmodified versus gene drive alleles in the resulting diploids and their sporulated haploid progeny over successive mating events.

Results
The inheritance-biasing efficiency of an endonuclease gene drive is determined by 1) the frequency of cutting and 2) the fraction of repair events that lead to the drive being copied onto the target chromosome. Only repair by homologous recombination (HR) results in drive copying and inheritance bias; the competing non-homologous end-joining (NHEJ) repair pathway ligates the broken ends together, preventing the drive from being copied and often generating mutations at the site of the break that prevent further cutting.
To prevent the accidental escape of our gene drives into wild yeast populations, we employed a method of molecular containment 1 in which we split our Cas9 based gene drive system into two physically separate genetic parts: an episomally encoded Cas9 gene and a drive element encoding the guide RNA. This allowed us to avoid creating a self-sufficient inheritance-biasing cassette while still targeting wild-type yeast strains. Though simple, this form of molecular containment is not vulnerable to human error; even if drive-containing yeast were to escape into the wild, the required Cas9 episomal plasmid would rapidly be segregated away from the drive element, rendering the drive inoperative.
In order to directly measure the efficiency of CRISPR/Cas9 gene drives in yeast, we took advantage of the red color that builds up in yeast lacking functional copies of the ADE2 gene 23 . If red ade2∆ haploids are mated with cream-colored wild-type haploids, the resulting heterozygous diploids inherit one functional copy of ADE2 and are therefore cream-colored. When these diploids undergo meiosis and reproduce via sporulation, half of the resulting haploids inherit the broken copy and are consequently red; the other half inherit the intact copy and are cream-colored ( Fig. 2A).
But if the red haploids encode a functional gene drive, it will cut and replace the intact ADE2 locus inherited from the wild-type parent, yielding red diploids (Fig. 2B). When these sporulate, all of the haploid offspring will inherit one of the two copies of the drive and will also be red. Thus, the cutting efficiency of a gene drive that targets and replaces ADE2 can be assessed by plating yeast cells and counting the fraction of colonies that are red.
We built a Cas9-based gene drive construct targeting ADE2 by placing a guide RNA targeting the wild-type ADE2 gene in place of the endogenous ADE2 locus. We mated these red ade2::sgRNA haploids to wild-type yeast of the opposite mating type in the presence or absence of the Cas9 plasmid and plated on media that selects for diploids. Nearly all diploid colonies were red when the Cas9 plasmid was present, indicating highly efficient cutting of the ADE2 copy inherited from the wild-type parent (Fig. 2C). We did not observe red diploid colonies in the absence of Cas9, demonstrating that the drive only functions in laboratory yeast populations encoding both Cas9 and guide RNA. Mutations in ADE2 generate a red phenotype on adenine-limiting media due to the buildup of red pigments. Mating a red mutant haploid to a wild-type haploid produces cream-colored diploids, which yield 50% red and 50% cream-colored progeny upon sporulation. (B) When haploids with a gene drive targeting ADE2 mate with wild-type haploids in the presence of Cas9, cutting and subsequent replacement or disruption of ADE2 produces red diploids that yield exclusively red progeny. (C) Diploids produced by mating wild-type and ade2::sgRNA gene drive haploids yield cream-colored colonies in the absence of Cas9 or when the target site is removed by recoding but uniformly red colonies when both are present, demonstrating Cas9-dependent disruption of the wild-type ADE2 copy. (D) Spores from 15 dissected tetrads produce uniformly red colonies on adenine-limited plates, confirming disruption of the ADE2 gene inherited from the wild-type parent. In the absence of the target site or Cas9, normal 2:2 segregation is observed.
To verify that the ADE2 alleles from wild-type parents were indeed lost, we sporulated the mated diploids and examined their resultant haploid progeny. Upon dissecting 18 cas9+ diploids, we observed a perfect 4:0 ratio of red:cream haploids, confirming that all copies of the ADE2 locus were disrupted. In contrast, 18 cream-colored cas9-diploids yielded a 2:2 red:cream ratio, indicating normal inheritance of the inactivated drive and the wild-type alleles (Fig. 2D).
To determine whether ADE2 disruptions in red diploids were the result of successful copying of the drive element via HR, we sequenced the 72 haploids derived from dissected cas9+ diploids. All sequenced colonies contained intact drives without additional mutations, indicating that drive mobilization was efficient and occurred at high fidelity. This represents the first example of a synthetic endonuclease gene drive that biases its own inheritance when mated to a wild-type eukaryote.
We next tested whether RNA-guided gene drives could be designed to bias the inheritance of not only the minimal drive element, but also any closely associated "cargo" gene whose spread through an existing population may be desirable. As a proof of principle, we inserted the URA3 gene in cis to the ade2::sgRNA minimal guide element. URA3 allows laboratory modified yeast strains to grow in the absence of uracil supplementation (Fig. 3A). We mated these URA3-containing drive haploids to wild-type haploids in the presence of an episomal Cas9 plasmid, selected diploids (all of which were red), sporulated them, and dissected 18 tetrads. As was the case for the original ADE2 gene drive, all of the sporulated haploid cells formed red colonies. Crucially, all grew normally when replica plated on uracil deficient media, indicating that URA3 was efficiently copied with the drive (Fig. 3B).
While both ADE2-based gene drives are highly efficient at biasing inheritance in the laboratory, they would be unable to cut a chromosome lacking the non-essential ADE2 gene or one with a mutation in the targeted sequence. This is a particular problem for drives seeking to alter entire populations because any aberrant repair by nonhomologous end-joining (NHEJ) following cutting will remove the target sites and thereby generate a driveresistant allele. We previously postulated that gene drives that target and recode an essential gene could avoid drive resistance in large populations because error prone repair events that modify the target site would cause lethality 1 . Non-essential but nonetheless important genes could similarly be edited because mutants created by NHEJ events would still be less fit than the drive itself. The ADE2-targeting gene drive was modified to carry URA3 as a cargo gene. (B) Diploids produced by mating wild-type URA3haploid yeast with haploids encoding the gene drive carrying URA3 were sporulated and tetrads dissected to isolate colonies arising from individual spores. All of these grew when replicaplated onto plates lacking uracil, demonstrating that the drive successfully copied URA3 in all diploids. (C) The ABD1-targeting gene drive cuts and recodes the 3' end of the essential ABD1 gene.
To validate this strategy of essential gene recoding during drive insertion, we constructed a third gene drive targeting the essential ABD1 gene (Fig. 3C) 24 . A haploid strain containing a recoded ABD1 allele upstream of a guide RNA targeting the natural ABD1 coding sequence was mated to wildtype cells in the presence of Cas9. We then selected diploids, sporulated the cells, dissected 18 of them, and sequenced the 72 resulting segregants. As expected, all of them contained the recoded ABD1 locus and the guide RNAs, demonstrating the feasibility of gene drives based on essential gene recoding.
Because the rate at which a gene drive is copied likely depends on a variety of host factors, including the types of repair machinery available to the cell at the time of the cut, the chromatin state of the locus, and the degree of homology flanking the double-strand break generated by the drive, we were curious whether our gene drives could be copied from our laboratory strain into a diverse group of native S. cerevisiae strains. We correspondingly mated ADE2 drive-containing haploids with 6 phylogenetically and phenotypically diverse wild-type strains of haploid S. cerevisiae (Fig. 4A) 25 . To quantitatively measure the efficiency of gene drive copying for each cross, we performed quantitative PCR on populations of all diploids using one set of primers specific to the drive and another set designed to amplify either wild-type or NHEJ-disrupted alleles. depicting the relative abundance of wild-type and drive-containing alleles in diploids arising from matings between SK1 haploids bearing gene drives and diverse wild-type haploid strains. "No Cas9" and "No Target" refer to the haploid cells containing the ADE2 drive element mated to wild-type haploids in the absence of Cas9 or to an otherwise wildtype strain containing Cas9 that also has a mutation in the targeted sequence that blocks cutting. "2nd gen" refers to the haploid progeny of an earlier mating.
The mean fraction of diploid chromosomes containing the ADE2 gene drive was over 99% regardless of wild-type parent (Fig. 4B), attesting to the robustness of the drive in diverse backgrounds. Addition of the URA3 cargo gene did not appreciably change this efficiency. The ABD1 drive was copied at an equivalent rate.
A final question of importance concerns the efficiency and stability of the drive over successive copying events. Because S. cerevisiae reproduces mostly through asexual division and only a subset of the population sporulate in the laboratory even when induced to undergo meiosis 26 , experiments to determine the long-term population-level efficiency of gene drives are difficult to perform and interpret. As a surrogate for the efficiency of the drive element over multiple generations and hence how it may behave on population scales, we sought to measure the efficiency of successive copying events by the same drive. We correspondingly mated several haploid offspring of the first-round ADE2 gene drive diploids to wild-type haploids containing the Cas9-expressing plasmid (Fig. 4B,  right). All of the gene drive constructs biased inheritance at the same efficiency in the second generation as they did in the first, indicating a continued ability to spread through sexually reproducing populations over multiple generations so long as the fitness cost associated with the drive is less than the inheritance benefit in the relevant environment.

Discussion
Our discovery that Cas9 can bias inheritance in diverse wild yeast strains over successive generations at very high efficiency demonstrates that RNA-guided gene drives can function in eukaryotic organisms as predicted. By itself this does not guarantee the success of gene drives in other organisms as the rate of homologous recombination varies between species and is known to be particularly high in yeast. However, the fact that we observed inheritance biasing rates equal to or exceeding that of the natural I-SceI homing endonuclease gene is highly promising. Because a drive based on I-SceI was correctly copied after ~97% of successful genome cutting events in transgenic Anopheles gambiae mosquitoes that contained an I-SceI recognition site 6 and Cas9 typically cuts more efficiently than does I-SceI 7,27 , our findings suggest that RNA-guided gene drives will be highly effective at suppressing 5,28,29 or spreading antimalarial alleles 30,31 through populations of this important malaria vector. Similarly, RNA-guided gene drives are likely to be more effective than those based on I-SceI, zinc-finger nucleases, or TALENs 7,8,32 due to their ability to efficiently cut multiple sequences without requiring highly repetitive elements.
Crucially, our gene drive that successfully targets and recodes the essential ABD1 gene in diploid yeast demonstrates the feasibility of targeting and recoding genes important for fitness, a strategy expected to improve the evolutionary stability of gene drives 1 . We recommend that future efforts seeking to build gene drives intended for eventual release adopt this same approach. Targeting sites adjacent to both homology arms may also promote more efficient drive copying by biasing repair towards homologous recombination 33 and increase the overall frequency of cutting 27 .
Our findings suggest that yeast may prove a useful platform for swiftly testing RNA-guided gene drive architectures before moving them into multicellular organisms. The power of yeast genetics and the ease of genome manipulation will facilitate combinatorial investigations into gene drive optimization. For example, studies might explore how biasing repair pathway choice 34 affects the efficiency of copying for gene drives of various sizes. Because the factors involved in these pathways are broadly conserved, these experiments could guide gene drive optimization in other organisms 35 .
While highly encouraging for potential gene drive applications, our results also sound a note of caution for subsequent experiments. That our drives were readily copied into a variety of yeast strains collected from all over the world underscores the potential for a single gene drive to affect very large populations. Poor flanking homology is not an effective barrier, at least in S. cerevisiae. Moreover, the ADE2 gene drive took only two weeks to design, build, and test, suggesting that many laboratories are capable of building gene drives in yeast. Since yeast reproduce mainly through asexual division, gene drives would need to be considerably less costly to organismal fitness in order to spread in the wild than would a comparably efficient gene drive in an organism that always reproduces via mating. However, natural endonuclease gene drives such as I-SceI do exist within yeast. Whether our gene drives or the typical RNA-guided gene drive will constitute this level of burden is as yet unknown.
It is considerably more difficult to edit the genomes of multicellular model organisms such as Drosophila than the yeast genome, and still more difficult to alter those species for which gene drive applications are most likely to be relevant. However, a growing number of laboratories now make heritable alterations in more than a dozen sexually reproducing species. This confluence of factors demands caution. Because synthetic gene drives would alter the global environmental commons, the decision to deploy such a drive must be made collectively by society. Any accidental release could severely damage public trust in scientists. As demonstrated by numerous containment breaches involving pathogenic viruses and bacteria, physical methods of containment are always susceptible to human error and should not be exclusively relied upon whenever alternatives are available.
All scientists making heritable alterations with Cas9 should therefore employ non-physical containment methods sufficient to prevent the creation of an RNA-guided gene drive capable of spreading in the wild. Even scientists not intending to work with gene drives should consider taking precautions, since any unintended insertion of the cas9 gene and guide RNAs near a targeted site could generate a gene drive. Fortunately, a simple and costless precaution is both available and already utilized for different reasons by many laboratories: avoid delivering the Cas9 gene on a DNA cassette that also encodes a guide RNA. As we have shown, guide RNAs alone cannot bias inheritance in the absence of Cas9 and consequently cannot spread through wild populations (Fig. 4B).
We therefore recommend that future gene drive experiments in yeast should also separate Cas9 from guide RNAs, for example by employing our episomal Cas9 plasmid, or employ another method of molecular containment. Gene drive experiments in species that always reproduce sexually pose greater risks and consequently should employ additional precautions such as targeting sequences not present in wild organisms or performing experiments in geographic areas where the organism in question cannot survive 1 . Working in genetic backgrounds that are less likely to escape the laboratory and mate -such as wingless flies in the case of Drosophila -may also be prudent. Should the scientific community deem it necessary, a voluntary peer review system designed exclusively for at-risk gene drive experiments, ideally conducted at the proposal stage, could further reduce the risk of accidental release.
In conclusion, our demonstration of diverse gene drive architectures enabling Cas9-mediated inheritance biasing in wild S. cerevisiae can guide efforts to build RNA-guided gene drives in other organisms and underscores the urgent need for precautionary control strategies, inclusive public engagement, and regulatory reform 22 in advance of real-world applications.

Methods:
Plasmids and genomic cassettes Gene drive cassettes were synthesized from gBlocks (Integrated DNA Technologies, Coralville, IA) and inserted into SK1 cells via Cas9-mediated genome modification as follows. Guide RNAs for each drive were cloned into p416-Cas9 containing plasmids with expression driven by the SNR52 promoter 18 . 60 base pair homology arms to the target locus were added on both ends of the gene drive cassette via PCR and 5 ug of PCR product was co-transformed with the p416-Cas9-gRNA plasmids. Correctly integrated gene drives were verified by sequencing and p416-Cas9-gRNA plasmids were cured using 5-Fluoroorotic Acid (FOA) selection.
To create the URA3-containing ADE2 gene drive, the ADE2 gene drive was cloned next to the Candida albicans URA3 gene in the pAG60 plasmid. The entire URA3 cassette and gene drive were PCR amplified and inserted using Cas9-mediated genome modification into the ADE2 locus of haploid SK1 cells.
The recoded C-terminus of the ABD1 gene and corresponding gene drive were synthesized as a gBlock to remove homology and generate mutations in the seed sequence via synonymous changes. The TEF1 terminator was inserted at the 3'end of the recoded ABD1 gene between the gene and the gRNA as ABD1 shares a terminator with the VHC1 gene. The entire cassette was integrated into the haploid SK1 genome using Cas9-mediated genome modification.
The p416-Cas9-gRNA plasmid (conferring uracil prototrophy) is a variant of the previously described p414-Cas9-gRNA plasmid (conferring tryptophan prototrophy) 18 (Addgene #43802). One or the other was used in each mating experiment. The pRS413 vector was transformed into select cell types to confer histidine prototrophy as a marker to select for diploid cells. After mating in liquid YPAD and selection for diploids on selection plates, the selection plates were scraped into 10 mL selective media and grown overnight at 30°C. A fresh 5 mL YPAD culture was then inoculated to and OD=0.1 and grown 4-5 hours at 30°C. The entire culture was then washed twice in 10 mL water, inoculated into 2 mL of sporulation media(1% potassium acetate), and incubated at room-temperature for 3 days or until spores were visible. Sporulated cells were suspended in 50 µL of a stock solution of zymolyase (50 µg/mL in 1M sorbitol) and incubated at 30C for 5 minutes, transferred to ice, diluted with 150 µL cold H2O, microdissected using a Zeiss tetrad dissection microscope, and isolated spores grown on YPAD plates.

Selection for URA3 function
Dissected spores were grown in synthetic complete (SC) media and then spotted onto SC medium as well as SC medium without uracil. To enhance red color, all SC solid media used for plate images contained 0.5 X adenine hemisulfate (final concentration of 0.08 mM).

Quantitative PCR
Candidate primer pairs were designed to amplify short regions specific to each drive or the wild-type sequence replaced by the drive, as well as the ACT1 gene as a control. All sequences are included in the supplementary information. Genomic DNA was extracted using Method A as described in Looke et al. 36 KAPA SYBR FAST qPCR Master Mix (2X) was used to perform the qPCR reaction along with 25 ng of genomic DNA. The amplification efficiency and relative specificity of each primer pair were measured by amplifying dilutions of genomic DNA from wild-type and drive haploids, respectively, and the best-performing and well-matched pairs selected for use (see below for all primers used). Quantitative PCR reactions were performed on genomic DNA isolated from each parental haploid as well as from diploids arising from three independent mating events. Three reactions (technical replicates) were performed per sample on a LightCycler 96 machine by Roche.

Calculations:
Results from three technical replicates were averaged for calculations. In order to directly calculate the ratio of alleles before PCR amplification, we first determined the efficiencies of the different primer pairs. Efficiencies were calculated from qPCR runs of serial dilutions (6 orders of magnitude) as: Efficiency=10 -1/slope R 2 values were higher than 0.99 in all cases except for one pair (ade2::URA3+sgRNA).
The allelic ratios were calculated as: xa . Ea Ct,a = xb . Eb Ct,b xa / xb = Eb Ct,b / Ea Ct,a with xa and xb being the initial concentration of drive and wt DNA, Ea and Eb the efficiency of the respective primer pairs and Ct,a and Ct,b the Ct values for each sample. Figure 4B was generated using BoxPlotR 37 .