Engineering Kluyveromyces marxianus as a Robust Synthetic Biology Platform Host

The yeast Kluyveromyces marxianus grows at high temperatures and on a wide range of carbon sources, making it a promising host for industrial biotechnology to produce renewable chemicals from plant biomass feedstocks. However, major genetic engineering limitations have kept this yeast from replacing the commonly used yeast Saccharomyces cerevisiae in industrial applications. Here, we describe genetic tools for genome editing and breeding K. marxianus strains, which we use to create a new thermotolerant strain with promising fatty acid production. These results open the door to using K. marxianus as a versatile synthetic biology platform organism for industrial applications.

IMPORTANCE The yeast Kluyveromyces marxianus grows at high temperatures and on a wide range of carbon sources, making it a promising host for industrial biotechnology to produce renewable chemicals from plant biomass feedstocks. However, major genetic engineering limitations have kept this yeast from replacing the commonly used yeast Saccharomyces cerevisiae in industrial applications. Here, we describe genetic tools for genome editing and breeding K. marxianus strains, which we use to create a new thermotolerant strain with promising fatty acid production.

RESULTS
CRISPR-Cas9 system in K. marxianus. We established robust genome editing in K. marxianus by adapting the plasmid-based CRISPR-Cas9 (CRISPRm) system we previously developed for S. cerevisiae (13). We first identified a K. marxianus-specific origin of replication and K. marxianus-specific promoters and terminators for expressing Cas9 (see Materials and Methods). We used the S. cerevisiae gene for tRNA Phe as an RNA polymerase III promoter to express single-guide RNAs (sgRNAs) from the same plasmid (Fig. 1A). To test the effectiveness of the redesigned CRISPR-Cas9 system, we used a wild strain we isolated from a sugarcane bagasse pile (Km1 [Table 1]), and a Km1derived MATa heterothallic strain (Km30 [see Table S1 in the supplemental material]). We transformed the pKCas plasmid (G418 R ) carrying an sgRNA targeting the URA3 gene into Km1, plated on G418 plates, and then selected for 5-fluoroorotic acid (5-FOA) resistance by replica plating to identify ura3colonies. The efficiency of NHEJ-based Cas9 editing (CRISPR NHEJ )-the number of ura3colonies divided by the number of total G418 R transformants-was near 90% and was confirmed by sequencing the URA3 locus. Targeting other genes across the genome resulted in around 75% efficiency (see Fig. S1 in the supplemental material). To test the ability of the K. marxianus CRISPR system to insert exogenous DNA at a defined locus, we also cotransformed into strain Km30 the pKCas plasmid encoding a guide RNA targeting the URA3 gene along with a double-stranded DNA repair template comprised of a linear nourseothricin resistance cassette flanked by K. marxianus URA3 homology sequences adjacent to the Cas9 target site (NatMX flanked by 0.9-kb homology arms). Using replica plating of G418 R trans- Synthetic Biology of Kluyveromyces marxianus ® formants onto two selection plates-5-FOA to detect ura3alleles and Nat R to detect HDR events-we found 100% of the colonies to be ura3and ϳ97% (189/195) to be Nat R , indicating that repair of Cas9-induced double-strand breaks allowed highly efficient HDR-mediated gene integration (CRISPR HDR ). We used colony PCR on select Nat R colonies (targeting outside the 0.9-kb homology arms of the NatMX cassette) to confirm NatMX cassette integration at the URA3 locus.
Engineering mating-competent heterothallic K. marxianus strains. We used CRISPR NHEJ to make stable K. marxianus laboratory strains with defined ploidy and mating type to enable the use of classical yeast genetics. Most naturally isolated K. marxianus strains are homothallic: i.e., they change their mating type spontaneously by "mating-type switching" to create mixed populations of MATa, MAT␣, and MATa/MAT␣ cells (4,5). The K. marxianus mating-type switching mechanism is not genetically conserved with the well-characterized HO endonuclease mechanism employed by S. cerevisiae. Notably, a two-component switching mechanism has been identified in Kluyveromyces lactis (14,15), which uses two transposases (Kat1 and ␣3) for MAT switching. The ␣3 transposase switches MAT␣ type cells to MATa type, and Kat1 switches MATa to MAT␣ type (Fig. 1B).
We identified the K. marxianus orthologs of the K. lactis KAT1 and ALPHA3 genes using reciprocal BLASTp against predicted open reading frames (ORFs) from the whole-genome sequence of Km1 (see Table S2 in the supplemental material) (16). Using CRISPR NHEJ , we targeted both transposase genes (Table S2) to create frameshift mutation loss-of-function alleles. We then isolated several of these double-transposaseinactivated Km1 ␣3 -kat1strains that had small base pair insertions or deletions near the Cas9 cut site (Fig. S1). To identify MATa haploid isolates, we used a pheromone morphological response assay. Yeast mating is initiated by the secretion of small peptide pheromones a-factor and ␣-factor by MATa and MAT␣ cells, respectively. The pheromones, derived from a-pheromone and ␣-pheromone precursor proteins, mating factor a (MFA1 and MFA2) and mating factor ␣ (MF␣1), are detected by their cognate cell surface recognition proteins and lead to polar morphogenesis or the formation of mating projections ("shmoo") that can be used to deduce a strain's mating type. We identified two putative K. marxianus MFA genes (KmMFA1 and KmMFA2) as well as the MF␣ gene (KmMF␣1, encoding two isotypes, KmMF␣1 and KmMF␣2) in the K. marxianus genome by reciprocal BLASTp using the S. cerevisiae and K. lactis protein sequences as queries (16,17) (Fig. 1C; see Fig. S2A in the supplemental material). Incubation of K. marxianus strain Km1 ␣3 -kat1cells with synthetic KmMF␣1 and KmMF␣2 peptide pheromones resulted in isolates that responded to both ␣-factors (Fig. 1D), indicating these are MATa ␣3 -kat1haploids. We categorized unresponsive strains as either MAT␣ or diploid strains, using sequencing of the MAT locus (Fig. S2B).
To further validate the role of ␣3 and Kat1 in switching mating types, we performed complementation assays by constitutively expressing these transposases from plasmids. Plasmids encoding Kat1 or ␣3 were transformed into Km1 ␣3 -kat1 -leu2strains to revert the controlled mating phenotype and promote homothallism. Transformants were then tested for mating-type switching by crossing them with stable heterothallic reference strains (␣3 -kat1 -trp1 -) of either MATa or MAT␣ mating type. Using the auxotrophic mating assay, these experiments showed that complementing stable MATa mutants with Kat1 overexpression plasmids or stable MAT␣ mutants with ␣3-expressing plasmids induced mating-type switching. Kat1 caused MATa isolates to switch and mate with a MATa reference strain, and ␣3 caused MAT␣ isolates to switch and mate with a MAT␣ reference strain (see Fig. S3 in the supplemental material).
Synthetic Biology of Kluyveromyces marxianus ® inactivation strains (␣3 -kat1 -leu2or ␣3 -kat1 -trp1 -) were successfully isolated from 10 of the isolates. We assayed these strains for mating type by crossing them with heterothallic Km1 strains as a reference, using the auxotrophic mating assay described above. Heterothallic haploids (MATa and/or MAT␣) were isolated from 10 of the triple-inactivation strains (Fig. 2B). For strain Km18, which was difficult to transform with plasmid DNA, stable heterothallic strains could be isolated from a cross between a homothallic Km18 strain first made trpusing UV mutagenesis and Km1 heterothallic strains. The Km18 trp -ϫ Km1 diploids were sporulated and germinated and then back-crossed with Km1 haploid reference strains to establish their mating type (Fig. 2C).
K. marxianus strains engineered for higher levels of lipogenesis. To explore the industrial potential of K. marxianus compared to S. cerevisiae (i.e., thermotolerance and Crabtree-negative growth, preferring respiration over fermentation [19]), we tested lipid production in K. marxianus under aerobic conditions. We first screened 11 wildtype K. marxianus isolates for levels of lipogenesis using a lipophilic fluorescent dye (Nile red), combined with flow cytometry and cell sorting. Nile red localizes to lipid droplets in yeast and exhibits increased red fluorescence proportional to the total amount of lipid in the cell (20,21). The K. marxianus strains were grown in 8% glucose or 8% cellobiose lipogenesis medium at 30 and 42°C, and time point samples were collected every 24 h to be analyzed by flow cytometry. The highest fluorescence was observed in strains fed 8% glucose at 42°C for 24 h, with large strain-to-strain differences spanning an ϳ20-fold change in fluorescence (Fig. 3A). A few strains also produced significant amounts of lipid in cellobiose at 42°C, compared to their production of lipid in glucose (i.e., strains Km2 and Km17 [ Table 1]) (see Fig. S4A in the supplemental material). All strains produced much lower levels of lipid when grown at 30°C.
We used fluorescence microscopy to examine the cell morphology of the strains with the highest lipid titers. When Nile red fluorescence was overlaid with differential interfer- ence contrast (DIC) images of K. marxianus isolates Km19, Km6, and Km18 (Table 1) after 24 or 48 h of growth in 8% glucose at 42°C, large lipid droplets encompassed a large fraction of the cell volume (Fig. 3B). Km19 produced the highest levels of lipids as measured by Nile red fluorescence, which peaked after only 24 h (Fig. 3A), at which point Km19 had accumulated lipids at ϳ10% dry cell weight (Fig. S4B). Thin-layer chromatography (TLC) revealed that the majority of the lipid in Km19 accumulated as free fatty acids (FFAs) (Fig. 3C).
Strain engineering for higher lipid production. Lipogenesis in oleaginous yeasts such as Yarrowia lipolytica results in the synthesis and storage of lipid droplets within the cytoplasm (22). Lipid biosynthesis is largely dependent upon the enzymes AMP deaminase (AMPD), ATP-citrate lyase (ACL), acetyl coenzyme A (acetyl-CoA) carboxylase (ACC), and malic enzyme (MAE) (23). Collectively, these enzymes promote the accumulation of acetyl-CoA via citrate. Interestingly, although ACL is thought to be crucial for lipogenesis in oleaginous yeasts (24), we did not identify the genes ACL1 and ACL2 in the K. marxianus reference strain, Km1. ACC1 then converts acetyl-CoA into malonyl-CoA, and the malic enzyme provides NADPH, the reduced cofactor necessary for the production of lipids. Total lipid accumulation is a balance between lipid synthesis and catabolism through ␤-oxidation in the peroxisome (Fig. 4A). Engineered strains of Y. lipolytica with reduced ␤-oxidation (pex10Δ) and peroxisome biogenesis (mfe1Δ) combined with overexpression of lipogenesis enzymes can store up to 80 to 90% dry cell weight as lipid compared to only ϳ10 to 15% lipid content for wild-type cells (23). Synthetic Biology of Kluyveromyces marxianus ® However, the high yield of lipogenesis in Y. lipolytica often takes up to 5 days to reach its peak (25) and requires temperatures of ϳ30°C, due to the lack of thermotolerance in this yeast (23).
We tested whether inactivation or overexpression of genes previously shown in Y. lipolytica to contribute to high lipid production (23) would affect the ability of K. marxianus to produce lipids. Although wild-type Km19 produced the most lipids of the wild-type strains we tested, Km19 is very difficult to transform with plasmids. Therefore, we chose to use strain Km6 (Table 1), since it has similar lipid content and is easily transformed with plasmid DNA, allowing the facile use of plasmid-based CRISPR-Cas9 to inactivate genes or plasmid-based overexpression.
Unlike Y. lipolytica, fermentation of glucose to ethanol and esterification of acetate to ethyl acetate are likely to compete with lipogenesis in K. marxianus. Therefore, we inactivated genes by CRISPR NHEJ to decrease ethanol fermentation (ADH genes) (26) and ethyl acetate production (ATF genes) (7), as well as ester biosynthesis (EHT1) (7,27) and glycerol biosynthesis (GPP1). We also inactivated genes involved in ␤-oxidation (PEX10 and MFE1) (23) (Fig. 4A). For the overexpression experiments, we cloned genes known to be involved in the accumulation of lipids (DGA1, ACC1, and the dimer ACL1/ACL2) into overexpression plasmids constructed using strong promoters (KmTDH3 or KmPGK1) to drive the expression of these genes. The plasmids were individually transformed into wild-type Km6 or strains in which CRISPR NHEJ had been used for targeted gene inactivation. The fatty acid content of each of these engineered strains was measured by gas chromatography and calculated as the percentage of dry cell weight. Although none of the inactivated genes had an appreciable effect on the accumulation of fatty acids (Fig. 4B), overexpression of DGA1 increased the levels of fatty acids across all strains tested, more than doubling the fatty acid content in the wild-type strain (Fig. 4C). No appreciable differences were found in terms of the fatty acid composition, except for the Km6 eht1strain bearing the ACC1 plasmid, which had more than 80% of its total fatty acids comprised of stearic acid (18:0), compared to ϳ10 to 20% in the other strains (see Fig. S5 in the supplemental material).
Breeding to isolate high-producing, thermotolerant, and transformable strains. The ability to cross phenotypically diverse and stable haploid K. marxianus strains should enable combining several beneficial traits into a single strain. For example, strain Km19 is the best lipid-producing strain we identified (Fig. 3A), but it is neither easily transformed nor thermotolerant compared to other K. marxianus strains. On the other hand, strain Km17 transforms easily and is thermotolerant (growing at 45°C) but is only moderately oleaginous (Fig. 3A). We therefore crossed these two strains to combine their beneficial traits into single isolates. We first engineered Km19 ␣3 -kat1 -trp1using CRISPR NHEJ and crossed a MATa isolate with an engineered stable haploid Km17 MAT␣ ␣3 -kat1 -leu2strain. The resulting diploids were sporulated and then germinated at high temperature (44°C) to select for thermotolerant segregants, and 91 haploid progeny were picked to screen for lipid production and plasmid transformability (Fig. 5A).
Single segregants isolated from the above temperature selection were individually scored in terms of lipid production using Nile red staining and flow cytometry and displayed high variability in lipid production (see Fig. S6 in the supplemental material). A few segregants performed better than the parental Km19 haploid strain, but a number of these had high fluorescence due to aggregation, as determined by light microscopy, and were therefore excluded from further analysis. Strains that did not aggregate had their fatty acid percentage in dry cell weight and composition measured using gas chromatography-flame ionization detection (GC-FID) (Fig. 5B). Notably, three of these isolates produced lipids as well as the parent Km19 strain, while inheriting the thermotolerance and transformability of Km17 ( Fig. 5C and D; see Fig. S7 in the supplemental material). These strains therefore combined all three beneficial traits of the parental strains.

DISCUSSION
Common metabolic engineering techniques are not ideal when dealing with complex phenotypes such as thermotolerance, productivity, and robustness (28). Therefore, agnostic approaches for combining complex traits into model yeast species are of high value, including directed evolution (29), genome-wide transcriptome engineering (30), and genome shuffling (31). However, these cannot substitute for classical sexual crossing for combining strain-specific traits. It is known that sexual reproduction enables adaptation to stressful industrial environments due to the faster unlinking of deleterious allelic pairs compared to clonal populations (10). Here we establish K. marxianus as a platform for synthetic biology by engineering stable heterothallic haploid strains that can be crossed to combine complex, unmapped multigenic traits into one strain.
In S. cerevisiae, inactivation of a single gene encoding HO endonuclease makes this yeast heterothallic and is sufficient to gain laboratory control of its mating cycle (32). While inactivation of the single transposase ␣3 can be used to cross K. marxianus strains (8), the resulting strains possess unstable mating types due to the presence of Kat1 and could randomly switch from MATa to MAT␣ (Fig. 1B and Fig. 2C; Fig. S3). To establish stable crossing in K. marxianus and abolish self-mating, we used CRISPR NHEJ to inactivate both transposases that are responsible for mating-type switching (␣3 and Kat1) (Fig. 1B). By simultaneously inactivating ALPHA3 and KAT1, we created stable heterothallic ␣3 -kat1strains that cannot switch mating type and, therefore, can be mated in a controlled manner with strains of the opposite mating type ( Fig. 2A; Fig. S3). Single segregants were isolated and tested for lipid production, transformability, and high-temperature growth. (B) Fatty acid percentage in dry cell weight (DCW) for several segregants from the Km17 ϫ Km19 cross and the parental strains. Three segregants have similar profiles to the more lipogenic parental strain (Km19). Experiments are from biological triplicates with mean and standard deviation shown. (C) Growth curves at 45°C for the segregants 4B, 5E, and 2G, as well as parental strains Km17 and Km19, in biological triplicate. Km19 is unable to grow at this temperature. Growth curves for parental strains at 30, 37, and 42°C can be found in the supplemental material (Fig. S7A), as well as for segregants at 30 and 37°C (Fig. S7B). (D) Transformation efficiency for several segregants normalized by Km17 transformation efficiency. Experiments are from 2 to 4 biological replicates with normalized mean and standard deviations shown.
Synthetic Biology of Kluyveromyces marxianus ® Alternative methods for creating stable haploids by deleting the silenced MAT loci have been used in yeast (33), but this strategy creates sterile strains and can be lethal if the endonuclease that initiates the double-strand break required for mating-type switching is not inactivated as well (34). Our strategy of inactivating both the ␣3 and Kat1 transposases preserves the ability to mate K. marxianus strains, an essential tool for synthetic biology, and to take advantage of this yeast's remarkable phenotypic diversity. We successfully isolated heterothallic haploid strains from 12 wild-type isolates. These strains readily mate with each other, resulting in sporulation-competent diploids that segregate to viable haploid spores. Combined with CRISPR-Cas9 genome editing ( Fig. 1) (6-9), these results establish a full set of tools for use of K. marxianus as a synthetic biology host and for future exploration of its biology on a genome-wide scale.
Using both sets of synthetic biology tools described here, we sought to exploit the diversity of K. marxianus as a thermotolerant, fast-growing, Crabtree-negative yeast. Screening 11 of the wild-type isolates for high levels of lipogenesis, we found high strain-to-strain variability in lipid production. Notably, strain Km19 produced ϳ10% lipid by dry cell weight after 24 h at 42°C, a considerably shorter time than the 120 h required by wild-type Y. lipolytica to accumulate a similar amount of lipid (25). We find Km19 stores the vast majority of lipid as free fatty acids (FFAs) (Fig. 3C), in contrast to Y. lipolytica, which stores lipids as triacylglycerols (22,35). FFAs are particularly suitable for the production of alkanes/alkenes and fatty alcohols, two types of high-value chemicals (36,37). Although Km19 produces lipids at a high rate, it is not thermotolerant compared to other K. marxianus isolates (Fig. 5C) and is difficult to transform with plasmid DNA (Fig. 5D). To eliminate these barriers to conducting genetic engineering in the oleaginous strain Km19, we crossed it with Km17, which transforms well and grows well at 45°C. Notably, some progeny of this cross isolated as single segregants produced lipids to the same level of the parent Km19 strain while retaining thermotolerance and transformability of Km17 and were also not prone to aggregation. The power of combining multiple complex and valuable traits using stable heterothallic strains, together with CRISPR-Cas9 genome engineering, opens a new frontier to use K. marxianus as both a thermotolerant model species and an industrially relevant host.

MATERIALS AND METHODS
Strains, media, and culture conditions. The K. marxianus strains used in this study were purchased from ATCC (American Type Culture Collection) or CBS (The Dutch Centraalbureau voor Schimmelcultures, Fungal Biodiversity Centre) or were obtained from an in-house collection. A complete list of all the wild-type strains is given in Table 1. Strains Km1, Km20, and Km21 have internal identity codes YST31, 1S300000, and 1S1600000, respectively. Strains were stored at Ϫ80°C in 25% glycerol. All experiments began by inoculation of a 12-ml culture tube containing 3 ml yeast extract-peptone-dextrose (YPD) or synthetic complete dextrose (SCD) medium with a single colony grown on a YPD medium agar plate. Cultures were shaken at 250 rpm. YPD agar consisted of 10 g/liter yeast extract, 20 g/liter peptone, 20 g/liter glucose, and 20 g/liter agar. SCD consisted of 2 g/liter yeast nitrogen base (YNB) without amino acids or ammonium sulfate, 1 g/liter complete supplement medium (CSM), and 5 g/liter (NH 4 ) 2 SO 4 . Five percent malt extract medium was made by mixing 30 g of malt extract with 20 g of agar and bringing the volume to 1 liter with H 2 O and then was sterilized by autoclaving at 10 lb/in 2 for 15 min. Sporulation (SPO) medium was made with 10 g/liter potassium acetate, 1 g/liter Bacto yeast extract, and 0.5 g/liter glucose. 5-FOA plates contained 2 g/liter yeast nitrogen base without amino acids or ammonium sulfate, 5 g/liter (NH 4 ) 2 SO 4 , 1 g/liter complete CSM, 20 g/liter glucose, 20 g/liter agar, and 1 g/liter 5-fluoroorotic acid (5-FOA). Lipogenesis medium contained 2 g/liter YNB without amino acids and ammonium sulfate, 1 g/liter ammonium sulfate or monosodium glutamate, and 8% glucose or cellobiose.
Genome sequencing and annotation. A single-colony isolate of strain YST31 (Km1 [Table 1]) grown on a YPD plate was used to inoculate a YPD liquid culture and prepare genomic DNA using the YeaStar genomic DNA kit (Zymo Research). We submitted ϳ5 mg of genomic DNA for small insert library preparation (ϳ250 bp) and Illumina sequencing. Library preparation and genome sequencing (Illumina HiSeq 2500) were performed by the UC Davis Genome Center DNA Technologies Core (http://dnatech .genomecenter.ucdavis.edu/). For YST31, we obtained 14,790,917 PE100 paired-end reads, and after trimming, we assembled reads into 116 scaffolds using CLC Genomics Workbench version 7.5.1. Default settings were used for quality trimming and de novo assembly. The median coverage was 250-fold, and the total genome assembly was 10,784,526 bp. Genome annotation was performed using an automated software pipeline, FGENESHϩϩ (http://www.softberry.com) version 3.1.1. Genes were first predicted ab initio using FGENESH and then refined based on protein homology (38,39). A custom BLAST database based on GenBank nr (downloaded 31 October 2014) was used for homology refinement of gene models. Gene prediction parameters were obtained from Softberry and were based on Saccharomyces cerevisiae gene models as the training set. The resulting annotation output files were renumbered and converted into GenBank format using custom scripts provided by Softberry.
Cas9 plasmid construction. To manipulate Kluyveromyces marxianus, we created a plasmid that can replicate in both Escherichia coli and K. marxianus and confers resistance to kanamycin and Geneticin, respectively. We used plasmid pOR1.1 (13), which can replicate in E. coli and S. cerevisiae, as a backbone for further manipulation. We identified and cloned an autonomous replicating sequence (ARS) from commercially available K. marxianus strain ATCC 36907 (Km11 [Table 1]) as follows. Using a YeaStar genomic DNA extraction kit (Zymo Research, D2002), genomic DNA was extracted from K. marxianus ATCC 36907. One microgram of genomic DNA was incubated with restriction enzyme EcoRI (NEB, R0101S) to fragment the DNA. In parallel, the S. cerevisiae 2 origin of replication was replaced with an EcoRI digestion site in pOR1.1. The plasmid was then linearized with EcoRI and treated with shrimp alkaline phosphatase (Affymetrix, 78390) to dephosphorylate the DNA ends and prevent religation of the vector. The genomic DNA fragment pool was ligated with the linearized plasmid using T4 DNA ligase (Invitrogen, 15224017), transformed into One Shot TOP10 competent E. coli (C404003), and plated on kanamycin selection plates. All growing colonies were pooled, and the plasmids were extracted using the QIAprep spin miniprep kit (Qiagen, 27106). Two micrograms of the resultant plasmid pool was transformed into ATCC 36907 and plated on Geneticin selection plates. Many colonies were picked, and plasmid extraction was performed for each using the Zymo Research yeast plasmid extraction kit (D2001). The plasmids were individually transformed back into One Shot TOP10 competent E. coli, and the plasmids were extracted once more and digested with EcoRI. The digests were run on a 1% agarose gel with TAE (Tris-acetate-EDTA) buffer (40), and the clone with the smallest insert was chosen. The insert was sequenced and then systematically trimmed to a 232-bp functional region that still conferred the ability of the plasmid to replicate in K. marxianus (Table S2).
The resulting plasmid with a K. marxianus ARS was then modified to express Cas9 and a single-guide RNA (sgRNA) cassette using transcription promoters and terminators from K. marxianus. The S. cerevisiae promoter and terminator for Cas9 as used in pCas (13) were replaced with those for homologous genes in K. marxianus. In the new plasmid, Cas9 expression was driven by the promoter region of the gene KmRNR2-a mild-strength promoter-and terminated by the strong KmCYC1 terminator. S. cerevisiae tRNAs were used as promoters to drive sgRNA expression (13) and terminated by the S. cerevisiae SNR52 (ScSNR52) terminator. Between the promoter and the sgRNA, there is a hepatitis ␦-ribozyme sequence that cleaves off the 5' leader sequence, liberating the tRNA from the sgRNA body that binds to Cas9 protein. The released transcript contains the ␦-ribozyme, the protospacer sequence that targets Cas9 to the desired sequence, and the scaffold sgRNA (13).
Overexpression plasmid construction. Four genes found to be involved in lipogenesis in other yeasts were cloned into overexpression plasmid backbones using the In-Fusion cloning kit (Takara). The cloning reaction mixtures contained 25 to 50 ng of vector, 3 times molar excess of PCR-generated insert, and 0.5 l of In-Fusion in a final volume of 2.5 l. K. marxianus ACC1 and DGA1 coding sequences were amplified from Km6 gDNA using Phusion polymerase and cloned into two different linearized backbones, while Yarrowia lipolytica ACL1 and ACL2 coding sequences were cloned into the same plasmid. The vector backbone was the same used for pKCas9 construction, containing a K. marxianus ARS isolated as described above, a Geneticin resistance marker, and the pUC bacterial origin of replication. ACC1 and ACL1 were controlled by the K. marxianus GK1 promoter, while DGA1 and ACL2 were under the control of the TDH3 promoter (Table S2). All ORFs were terminated by the K. marxianus CYC1 terminator sequence. The resulting plasmids were transformed into wild-type K. marxianus strain Km6 (Table 1), as well as into 10 knockout mutant strains derived from Km6 that were constructed using CRISPR NHEJ : adh5 -, adh6 -, adh4 -, gpp1 -, atf1 -, eht1 -, mfe1 -gpp1 -, mfe1 -adh5 -, pex10 -gpp1 -, and pex10 -adh5 - (Table S1).
High-efficiency DNA transformation. We established a high-efficiency transformation protocol for K. marxianus as follows. A single colony was inoculated in 1.5 ml of YPD medium and incubated at 30ºC overnight, then 180 l of this culture was transferred to 5 ml of fresh YPD medium and incubated at 30ºC until the optical density at 600 nm (OD 600 ) reached 1.0 to 1.2 (ϳ5 to 6 h). Then, 1.4 ml of the culture was aliquoted into microcentrifuge tubes and spun down at 3,000 ϫ g for 5 min. The supernatant was removed, and the pellet was resuspended in 50 mM lithium acetate, followed by incubation at room temperature for 15 min. The cells were spun down, the supernatant was discarded, and the cells were used for subsequent transformation reactions.
Single-stranded DNA (ssDNA) was previously prepared as follows (41): 2 g/l of ssDNA from Sigma (D1626-250mg) was agitated with a stir bar overnight in TE buffer (10 mM Tris [pH 8.0] and 1 mM EDTA) at 4ºC and then concentrated to 10 g/l by isopropanol precipitation, resuspended in water, and quantified (NanoDrop 1000; Thermo Scientific). Prior to each transformation, aliquots of ssDNA were boiled for 5 min and then placed in an ice bath for 5 min. Keeping all the reagents on ice, 66.7 l of 60% polyethylene glycol (PEG) 2050, 12.5 l of 2 M lithium acetate, and water in a final volume of 100 l were added to a sterile microcentrifuge tube. Then 2 l of 1 M dithiothreitol (DTT) and 25 g of ssDNA were added, followed by 0.1 to 5 g of pKCas plasmid DNA. The transformation mixture was briefly vortexed, and 100 l was added to the cells. The transformation reaction mixture was then incubated at 42ºC for 40 min. The reaction was spun down for 5 min at 3,000 ϫ g, the supernatant was removed, and 500 l of fresh YPD was added. The cells were allowed to recover for 2 h at 37ºC and 250 rpm. Then 10% of the volume was spread on a YPD G418 selection plate, and the remaining volume was spread on a second G418 plate. CRISPR NHEJ and CRISPR HDR . We transformed the pKCas plasmid into K. marxianus strains in the absence of donor repair DNA to determine the efficiency of NHEJ repair of the double-strand break. We cloned the guide sequence for the sgRNA in pKCas to target URA3 (Table S2) with 5-fluoroorotic acid (5-FOA) plates, which select for ura3colonies. Approximately 1 g of pKCas plasmid was transformed into K. marxianus, and the efficiency of editing was calculated by counting the number of ura3colonies divided by the number of G418 R transformants. Sequencing of the targeted region revealed small insertions or deletions (indels) around the Cas9 cleavage site typical of NHEJ, resulting in premature stop codons within the URA3 ORF (Fig. S1). We find that this system can be used to create inactive alleles in different genes with efficiencies of ϳ75% (Fig. S1). We modified the high-efficiency transformation protocol to enable cotransformation of the pKCas plasmid and a linear repair DNA template (donor DNA), for HDR-mediated genome editing. For tests of HDR-mediated gene insertion, the donor DNA targeted for genome integration contained the NatMX cassette conferring resistance to nourseothricin and 0.9 kb of flanking homology to the target site in the K. marxianus genome. Donor DNA was generated by PCR and concentrated by isopropanol precipitation (42). The best ratio of plasmid to donor DNA was 0.2 g of pKCas plasmid and 5 g of linear donor DNA, with 0.9 kb of homology to the Cas9 targeting site. Tests with higher concentrations of both pKCas and donor DNA were not as efficient. The transformation reaction was carried out as described above, except that cells were allowed to recover for 1 h at 37ºC in YPD without drug at 250 rpm, after which G418 was added and the cells were allowed to recover overnight. For the nourseothricin gene insertion experiments, cells were plated on YPD G418 plates, incubated at 37°C overnight, and then replica plated on either 5-FOA or nourseothricin plates, to identify Nat R colonies with correct insertion in the URA3 locus. Colony PCR performed on select Nat R colonies (targeting outside the 0.9-kb homology arms of the NatMX cassette) confirmed NatMX cassette integration at the URA3 locus.
Mating-competent heterothallic strains. ALPHA3 and KAT1 double-inactivation strains (␣3 -kat1 -) were constructed using CRISPR NHEJ (Table S1). K. marxianus was transformed with either KAT1-or ALPHA3-targeting pKCas plasmids. Genomic DNA was isolated from G418 R colonies, the KAT1 or ALPHA3 regions were PCR amplified, and the PCR products were sequenced. Colonies with sequences containing indels that generated early stop codons were chosen and were saved as glycerol stocks. Singleinactivation strains were subjected to a second round of CRISPR NHEJ to inactivate the second transposase, creating the ␣3 -kat1double mutants. These double mutants were then subjected to a third round of CRISPR NHEJ targeting the LEU2 or TRP1 genes to create auxotrophic strains. The resulting strains were tested for mating type and heterothallic status by using a pheromone assay described below or by crossing them with reference heterothallic haploid strains, allowing haploid heterothallic MATa or MAT␣ strains to be successfully isolated. Some double-transposase-inactivated strains did not mate with the reference strains, possibly because these are stable diploids or triploids (12) or possibly due to chromosomal rearrangements (43). Although we performed the gene inactivations individually, we later tested and verified that simultaneous targeting of both transposases with CRISPR NHEJ works efficiently in K. marxianus.
Mating pheromone response assay. We used reciprocal BLASTp using S. cerevisiae and K. lactis protein sequences as queries to identify two putative MFA genes (KmMFA1 and KmMFA2), as well as the MF␣ gene (16,17). The sequences in S. cerevisiae are encoded by genes YPL187W and YGL089C in the Saccharomyces Genome Database (44), and those in K. lactis are encoded by GenBank entry CAG99901.1 (RefSeq ID XP_454814.1) and as described in reference 17. To the best of our knowledge, these pheromones had not been previously identified in K. marxianus. We found two putative MFA genes (KmMFA1 and KmMFA2) as well as the MF␣ gene (KmMF␣1, which encodes 2 isotypes, KmMF␣1 and KmMF␣2) by reciprocal BLASTp using the S. cerevisiae and K. lactis protein sequences as queries (Table S2). Interestingly, the deduced a-factor amino acid sequences of KmMFa1 and KmMFa2 are not completely conserved, differing by 1 amino acid (Fig. 1C). Sequencing of KmMFA1, KmMFA2, and KmMF␣ from 8 unique strains shows full strain-to-strain conservation. Notably, KmMFa1 is completely conserved with the respective K. lactis sequence, suggesting a relatively conserved sexual cycle.
Synthetic a-and ␣-pheromones were obtained from Genemed Synthesis, Inc., and resuspended in water. To verify if the putative ␣-factors (KmMf␣1 and KmMf␣2) are viable and induce mating projections ("shmoo") in engineered heterothallic cells, we performed a pheromone morphological response assay. Cells were grown in YPD medium to a density of 10 6 cells/100 ml, and both pheromones were added to a final concentration of 25 mg/ml. Cells were then examined under the light microscope at different times. Mating projections were observed after 6 h for the mature ␣-factors KmMf␣1 and KmMf␣2, while no morphological differences were seen in the absence of synthetic pheromone. Strains sensitive to the ␣-pheromones were classified as putative MATa, and strains that exhibited no mating projections in the presence of either ␣-pheromone for 12 h were presumed to be stable MAT␣ haploids or MATa/MAT␣ diploid strains. The mating type of heterothallic strains was verified by sequencing of the MAT loci, by PCR amplification of the MAT loci with primers flanking the MAT loci, and by MAT-specific primers (Table S2 and Fig. S2). The same procedure was done using the synthesized a-factors, but they failed to induce mating projections in the strains tested, possibly because a-factors are reported to be heavily posttranslationally modified, unlike ␣-factors (45).
K. marxianus auxotrophic mating. Auxotrophic double mutant strains (␣3 -kat1 -) of either mating type were grown up overnight from a single colony in 5 ml SCD medium. Strains were pelleted at 3,500 ϫ g for 5 min, followed by washing with 1 ml of sterile water. Strains were pelleted again and resuspended in 50 l of water. Four microliters of the cell suspension was dispensed as a single drop (patched) onto 2% glucose plates or MA5 plates and allowed to dry. Strains to be mated containing a complementary auxotrophic marker (either leu2or trp1 -) were dispensed on top of previous dried spots and allowed to mate at room temperature for 24 to 48 h. Mating plates were replica plated onto SCD agar plates minus both leucine and tryptophan. Diploid cells were grown at 30°C for 48 h. Freezer stocks were made by scraping off the diploid patch, resuspending in 25% glycerol, and freezing at Ϫ80°C. lipogenesis conditions for 24 h. The strains were ranked by Nile red mean fluorescence (Fig. S6), and those prone to flocculation as determined by light microscopy were eliminated from further analysis.

Determination of fatty acids by gravimetric analysis and GC-FID.
To measure the percentage of fatty acids in dry cell weight, we extracted lipids from lyophilized cells and performed gravimetric analysis. Briefly, 250 l of culture was pelleted at 4,000 ϫ g for 10 min in preweighed 1.5-ml microcentrifuge tubes (Metter Toledo Excellence XS205DU balance). After the supernatant was removed, the cell pellet was suspended in 0.5 ml of water and pelleted again at 3,000 ϫ g for 10 min, after which the pellet was resuspended in 100 l of water. The samples were frozen and stored at Ϫ80ºC. Frozen samples were then lyophilized overnight and weighed to calculate dry cell weight.
To measure the percentage of fatty acids in the different strains, we extracted lipids from lyophilized cells prepared as described above and analyzed them by gas chromatography. Fatty acids were extracted and transesterified into fatty acid methyl esters (FAMEs) with methanol in the presence of an acid catalyst. The dry cell pellet was transferred to 15-ml glass conical screw-top centrifuge tubes, and 1 ml of methanolic HCl (3 N concentration) with 2% chloroform was added to the pellet. To ensure complete transesterification, an additional 2 ml of methanolic HCl (3 N) plus 2% chloroform was added. Approximately 100 g of an internal standard (methyl tridecanoate) with its exact mass recorded was prepared in methanol and was added to the tube. The tube was sealed with a Teflon-coated screw cap and heated at 85°C for 1.5 h with vortexing every 15 min. The mixture was then cooled to room temperature, and the resulting FAMEs were extracted by the addition of 1 ml of hexanes followed by 30 s of vortexing. An organic top layer was obtained by centrifugation of the sample at 3,000 ϫ g for 10 min. The top layer was carefully collected and transferred to a GC vial. One microliter was injected in split mode (1:10) onto an SP2330 capillary column (30 m by 0.25 mm by 0.2 m; Supelco). An Agilent 7890A gas chromatograph equipped with a flame ionization detector was used for analysis with the following instrumental settings: injector temperature, 250°C; carrier gas, helium at 1 ml/min; and temperature program, 140°C, 3 min isocratic, 10°C/min to 220°C, 40°C/min to 240°C, and 2.5 min isocratic.
Total oil extraction from dry yeast cells for thin-layer chromatography. Total yeast oil was extracted following the protocol described by Folch et al (47) for analysis by thin-layer chromatography. Approximately 20 mg dry cell weight and 500 mg silica beads were weighed into a 2-ml centrifuge tube. One milliliter MeOH was added to the tube, and it was vortexed. Then, the tube was put into an aluminum block and bead-beaten 4 times for 30 s with 30-s resting intervals in between. The contents were transferred into a conical 15-ml glass centrifuge tube, and 0.25 ml MeOH was used to rinse the residuals on the small centrifuge tube. CHCl 3 (2.5 ml) was added, and the tube was briefly vortexed. The tube was shaken for 1 h at 235 rpm. Two hundred fifty microliters of CHCl 3 -MeOH (2:1) and 1 ml of MgCl 2 (aqueous) (0.034%) were added into the mixture, and the tube was shaken for 10 min. The solution was vortexed for 30 s and centrifuged at 3,000 rpm for 5 min, and the upper aqueous layer was removed. The resulting organic layer was washed with 1 ml 2 N KCl-methanol (4:1 vol/vol), vortexed, and centrifuged in the same way. The aqueous upper layer was removed, and the resulting organic layer was introduced with the artificial upper phase (chloroform-methanol-water at 3:48:47). The resulting mixture was vortexed and centrifuged, and the upper layer was aspirated. This step involving the artificial upper phase was repeated until the white layer at the interface completely disappeared.
Thin-layer chromatography analysis of lipid composition. TLC plates (7 by 10 cm) were preheated in a 120˚C oven for at least 10 min. A piece of paper towel/filter paper (7 by 10 cm) was added into a 600-ml beaker for saturation, and solvent system 1 (SS1: petroleum ether-Et 2 O-AcOH [70:30:2]) was added until the solvent reached about 0.5 to 1 cm in height. The beaker was covered with aluminum foil and Parafilm, and the resulting setup was left alone for at least 10 min. Compounds were spotted on the preheated TLC plate, and the plate ran in SS1 until the solvent front was halfway up the plate. The plate was dried at room temperature for 15 min. The plate was run using solvent system 2 (SS2: petroleum ether-Et 2 O [98:2]) until the solvent front nearly reached the top of the plate. The resulting TLC plate was dried under a fume hood for 30 min before being immersed in MnCl 2 charring solution (0.63 g MnCl 2 ·4H 2 O, 60 ml H 2 O, 60 ml MeOH, 4 ml concentrated H 2 SO 4 ) for 10 s. The stained plate was developed in a 120˚C oven for approximately 20 min or until dark spots were observed.
Data availability. The raw fastq reads have been deposited in the NCBI SRA (accession no. SRP158013) and the scaffolds and annotation in Genbank under BioSample accession no. SAMN09839046.