Role of Ectopic Gene Conversion in the Evolution of a Candida krusei Pleiotropic Drug Resistance Transporter Family

Abstract

Gene duplications enable the evolution of novel gene function, but strong positive selection is required to preserve advantageous mutations in a population. This is because frequent ectopic gene conversions (EGCs) between highly similar, tandem-duplicated, sequences, can rapidly remove fate-determining mutations by replacing them with the neighboring parent gene sequences. Unfortunately, the high sequence similarities between tandem-duplicated genes severely hamper empirical studies of this important evolutionary process, because deciphering their correct sequences is challenging. In this study, we employed the eukaryotic model organism Saccharomyces cerevisiae to clone and functionally characterize all 30 alleles of an important pair of tandem-duplicated multidrug efflux pump genes, ABC1 and ABC11, from seven strains of the diploid pathogenic yeast Candida krusei. Discovery and functional characterization of their closest ancestor, C. krusei ABC12, helped elucidate the evolutionary history of the entire gene family. Our data support the proposal that the pleiotropic drug resistance (PDR) transporters Abc1p and Abc11p have evolved by concerted evolution for ∼134 MY. While >90% of their sequences remained identical, very strong purifying selection protected six short DNA patches encoding just 18 core amino acid (aa) differences in particular trans membrane span (TMS) regions causing two distinct efflux pump functions. A proline-kink change at the bottom of Abc11p TMS3 was possibly fate determining. Our data also enabled the first empirical estimates for key parameters of eukaryotic gene evolution, they provided rare examples of intron loss, and PDR transporter phylogeny confirmed that C. krusei belongs to a novel, yet unnamed, third major Saccharomycotina lineage.

GENE duplication is one of the major driving forces for the evolution of novel gene function. It enables organisms to rapidly adapt to changing environments (Ohno 1970; Li 1997), and it is also important for speciation (Lynch and Conery 2003). Large proportions (20–60%) of genes in all kingdoms of life are generated by gene duplication (Rubin et al. 2000; Li et al. 2001; Zhang 2002). The birth rate of duplicated genes in eukaryotes has been estimated to be ∼1 gene⁻¹ 100 MY⁻¹ (Lynch and Conery 2000), which is similar to their silent-site nucleotide substitution rate of ∼0.25–1.6 nt⁻¹ 100 MY⁻¹ (Li 1997; Lynch and Conery 2000). However, most duplicated genes are inactivated relatively quickly (Walsh 1995). Saccharomyces cerevisiae contains duplicated genes that arose from an ancient whole genome duplication event (Wolfe and Shields 1997; Kellis et al. 2004), which was recently identified as an ancient interspecies hybridization event (Marcet-Houben and Gabaldon 2015). However, almost 90% of all duplicated genes were lost within a few million years at an estimated half-life of ∼1 MY (Lynch and Conery 2000, 2003). The fate of duplicated genes depends on a number of factors. Davis and Petrov (2004) reported preferential duplication of conserved genes, while genes encoding proteins of multi-protein complexes or dosage-sensitive genes were less likely to generate stable duplicates (Papp et al. 2003a). A comprehensive study by Wapinski et al. (2007) found that duplicated genes typically diverge with respect to regulatory control, but rarely in their biochemical function (Wapinski et al. 2007). The authors concluded that gene duplication may eventually simplify, rather than increase, the complexity of a system by partial “division of labor.” That is, “subfunctionalization” (SF) enables the optimization of overlapping functions of an ancestral protein (Wapinski et al. 2007). However, in smaller populations, mutation and genetic drift can increase functional complexity without any necessary adaptive benefit to the organism (Lynch 2007). The SF of a component of the fungal V-ATPase proton pump is a case in point (Finnigan et al. 2012).

Gene conversion is a “copy and paste” function between highly similar (>90%) sequences. It can occur between alleles, typically during meiosis (allelic gene conversion), but also occurs between paralogous, duplicated, genes during mitosis [ectopic gene conversion, EGC (Fawcett and Innan 2011)]. EGC was first recognized as important for maintaining sequence identity between repeat copies of genes within the large rRNA gene cluster, prompting the generation of a model for the evolution of multi-gene families termed “concerted evolution” (Brown et al. 1972). Many examples of multi-gene families (e.g., histones, GPCRs, ubiquitins, immunoglobulins, and MHC genes) supposedly shaped by concerted evolution followed. However, this model does not explain how members of multi-gene families evolve novel functions, and why there are so many inactive pseudogenes (∼2–5% of yeast genes (Lafontaine and Dujon 2010) and >10,000 in humans (Zheng et al. 2007)). This prompted the formulation of the “birth-and-death” model, which suggested that new genes are created by gene duplication and some stay in the genome for a very long time, whereas most are inactivated or deleted from the genome (Nei and Rooney 2005). The controversy over which of the two models best describes the evolution of multi-gene families continues (Arguello and Connallon 2011), partly because neither model explains the biological evidence satisfactorily.

Many factors affect the gene conversion rate such as genome location, the distance and similarity between similar tracts, and the minimum tract length or minimum efficient processing segment (MEPS)—the minimum length of highly similar repeat sequences required for homologous recombination. Most studies estimate the gene conversion rate to be ∼10–100 times higher than the synonymous nucleotide substitution rate (Mansai et al. 2011). The outcome of gene conversion can be either beneficial (e.g., rRNA gene family) or harmful, with an increasing number of genetic disorders being ascribed to EGC (Chen et al. 2010).

But how can duplicated genes evolve a novel function (i.e., neo-functionalize) and fix an advantageous mutation(s) in a population when EGC is active? Population genetic theory indicates that strong selection against EGC is required for the stable maintenance of a mutation (i.e., a fate determining mutation, as it provides an evolutionary advantage to the organism) ultimately resulting in the evolution of a novel function [i.e., neo-functionalization (NF)] of a duplicated gene copy (Innan and Kondrashov 2010). Gao and Innan noticed that the battle between these two opposing evolutionary forces (i.e., maintenance of an advantageous mutation vs. frequent EGCs between tandem-duplicated genes) continues for a relatively long time (Gao and Innan 2004). This, ultimately, leads to a local peak of sequence divergence between these tandem-duplicates with low divergence in regions away from the mutation, and the outcome of NF in the face of EGC is a recognizable pattern of highly similar sequences interspersed by highly divergent sequences (Osada and Innan 2008; Innan and Kondrashov 2010). A prominent example is the human red and green opsin genes that have arisen by gene duplication after the split of New and Old World monkeys ∼30–40 MYA with only two aa variations of exon 5 accounting for >80% of the functional differences between the red and green eye pigments (Nathans et al. 1986; Neitz et al. 1991; Winderickx et al. 1993; Shyue et al. 1994).

Although EGC, natural mutation, and positive Darwinian selection are important factors shaping the evolution of gene duplicates, and complex mathematical models can predict possible outcomes, more experimental data are needed to improve existing models (Innan and Kondrashov 2010). Such data are hard to obtain because sequencing duplicated genes from a representative sample of individuals from higher eukaryotes is extremely difficult due to their large size and, often, complex gene arrangements, which makes it is almost impossible to correctly resolve such repeat sequences (Blow 2015). The discovery of an ancient pair of tandem-duplicated genes (ABC1 and ABC11), encoding drug efflux pumps, undergoing prolonged concerted evolution in Candida krusei provided a rare opportunity to extract actual values for important evolutionary factors shaping the evolution of duplicated genes. Extensive polymorphism data from geographically diverse isolates gave novel insights into the population structure of C. krusei. They enabled direct estimates for the gene conversion rate and the size of the MEPS, and they confirmed “GC-bias” for gene conversion (Galtier 2003; Marais 2003; Noonan et al. 2004) in C. krusei. Intramolecular gene deletions and intermolecular nonhomologous mitotic recombination between tandem-duplicates also caused copy number variations (CNV), and the formation of ABC11-1 or ABC1-11 chimeras with novel pump functions. The discovery of a third close homolog, C. krusei multidrug efflux pump ABC12, and discovery of ABC11 and ABC1 orthologs in the closest sequenced relative, Pichia membranifaciens, were crucial to tentatively infer the evolutionary history of all three C. krusei family members, and provided clues about the amino acid changes in Abc1p and Abc11p that were most likely fate-determining, and caused the evolution of two functionally distinct efflux pumps.

Materials and Methods

Strains and culture conditions

C. krusei isolates and yeast strains used in this study are listed in Supplemental Material, Table S1 in File S1. All S. cerevisiae strains created in this study are based on AD∆ (Lamping et al. 2007). Fungal strains were grown in yeast extract, peptone and glucose (YPD) medium: 1% (w/v) Bacto-yeast extract (Difco Laboratories, Detroit, MI), 2% (w/v) Bacto-peptone (Difco) and 2% (w/v) glucose. Yeast transformants were selected on plates containing 0.077% (w/v) complete supplement mixture without uracil (CSM-ura; Bio 101, Vista, CA), 0.67% (w/v) yeast nitrogen base without amino acids (Difco), 2% (w/v) glucose, and 2% (w/v) agar (Difco). Plasmids were maintained in Escherichia coli strain DH5α. E. coli cells were grown in Luria-Bertani (LB) medium, to which ampicillin was added (100 µg/ml) as required.

Materials

Molecular biology reagents, restriction and modifying enzymes were from New England Biolabs (Beverly, MA) or from Roche Diagnostics N.Z. (Auckland, New Zealand). Lyophilized desalted DNA oligonucleotides (Table S2 in File S1) were purchased from Sigma-Aldrich Pty. (Sydney, Australia). PCR-amplified DNA fragments were purified using kits from Qiagen Pty. (Clifton Hill, Victoria, Australia). Genomic DNA (gDNA) was isolated from yeast using the Y-DER Yeast DNA Extraction Reagent Kit from Pierce (Rockford, IL), or the DNeasy Tissue Kit from Qiagen. Yeast were transformed using the alkali-cation yeast transformation kit from Bio 101 (Lamping et al. 2007). All plasmids and yeast transformants were verified by DNA-sequencing using the DYEnamic ET Terminator Cycle Sequencing kit v 3.1 (GE Healthcare UK, Little Chalfont, UK), and analyzed at the Micromon DNA Sequencing Facility (Monash University, Melbourne, Australia). PCR reactions used the high fidelity KOD⁺ DNA polymerase (Toyobo, Osaka, Japan, or Novagen, San Diego, CA).

Compounds

The chemicals and antifungal agents were purchased from the following sources: fluconazole (FLC; Diflucan; Pfizer Laboratories, Auckland, New Zealand), itraconazole (ITC; Janssen Research Foundation, Beerse, Belgium), aureobasidin A (AUR; Takara Bio, Shiga, Japan), clotrimazole (CLT; Bayer, Osaka, Japan), miconazole (MCZ), voriconazole (VRZ), rhodamine 6G (R6G), rhodamine 123 (R123), cycloheximide (CHX), nigericin (NIG), trifluoperazine (TFP), doxorubicin (DOX), daunorubicin (DAU), enniatin (ENI), and oligomycin (OLI) (Sigma-Aldrich New Zealand, Auckland, New Zealand). FK506 was a gift from Astellas Pharma (Tokyo, Japan), and the milbemycins α11, α20, β9, and β11 were a gift from Sankyo (Tokyo, Japan).

Multilocus sequence typing (MLST)

To determine the diploid sequence type (DST) for each isolate, we used the publicly available C. krusei MLST database (http://pubmlst.org/ckrusei/; Jacobsen et al. 2007). Using the recommended DNA primers, we PCR-amplified parts of the ORFs of six housekeeping genes (HIS3, LEU2, NMT1, TRP1, ADE2, and LYS2) from gDNA extracted from individual C. krusei strains, and sequenced the fragments in both directions with the same primers. Genotypes that were included in the database were numbered accordingly, and new genotypes were given new numbers by extending the count of identified genotypes.

Genetic manipulations

Plasmid pBluescriptIISK(+) (Stratagene, La Jolla, CA) was used to generate ABC1 and ABC11 subclones. Inverse PCR fragments were amplified from gDNA extracted from different C. krusei isolates as previously described (Lamping et al. 2009). In short, ∼1 µg of gDNA was digested to completion with restriction enzymes, purified by phenol extraction followed by ethanol precipitation, and redissolved in 100 µl of sterile milliQ water. The fully digested, purified, gDNA (∼50 ng) was then ligated (20 µl reaction) to form circular DNA molecules. Portions of the ligated gDNA mix (1 µl) were used as templates for inverse PCR reactions (50 µl). Inverse forward and reverse PCR primer pairs specific for ABC11 (11-P1, 11-P2, and 11-P3), or ABC1 (1-P3; Figure S1B) were designed to be close to, but pointing away from, each other. The following PCR cycling protocol was used: 94° for 5 min, (94° for 20 sec, 55° for 10 sec and 68° for 1 min/kb) repeated 34 times, and with a final extension time of 10 min at 68°. The inverse PCR fragments were gel-purified and further characterized. Some were used to create restriction maps to ascertain which inverse PCR products covered the largest portions of the uncharacterized genome sequence, while others were sequenced directly.

Verification of an 88 bp intron in ABC11

The ABC11 intron was verified experimentally by amplifying the ABC11 cDNA ORF from total RNA isolated from B2399 cells grown in YPD medium and harvested at late logarithmic growth. First strand ABC11 cDNA was amplified from total RNA with SuperscriptIII Reverse Transcriptase (Invitrogen New Zealand, Auckland, NZ), and the ABC11-specific primer ABC11-Stop. A portion of this reaction was then used as DNA-template to PCR-amplify the entire ABC11 cDNA ORF, and DNA sequencing confirmed the 88 bp intron.

Yeast strains overexpressing individual ABC1 and ABC11 alleles with or without an intron

In order to compare the function of C. krusei Abc11p with Abc1p and Abc12p, the proteins were overexpressed in S. cerevisiae AD∆. The construction of ABC1 overexpressing yeast strains AD∆/CkABC1g and AD∆/CkABC1c was described previously (Lamping et al. 2007). Because ABC1 was toxic to E. coli, and it was impossible to clone and propagate plasmid pABC3-CkABC1 in E. coli, an alternative cloning strategy was developed (Lamping et al. 2007). However, this strategy required multiple PCR steps (>60 PCR cycles) to amplify the entire transformation cassette (see top of Figure S2), which caused a high mutation rate. We therefore developed a new cloning strategy that is fast, highly reliable, and accurate, and could also be useful for cloning and sequencing individual alleles of other large genes. Rather than using multiple PCR steps to create the entire transformation cassette in one piece, we directly transformed AD∆ with separate PCR fragments that overlapped by 25 bp (center and bottom of Figure S2), and let the individual pieces fuse by homologous recombination. This required four or five separate homologous recombination events, which are exceptionally accurate and efficient in S. cerevisiae. Instead of using the ABC11-cDNA from the above experiment to create AD∆/CkABC11c, ABC11c was PCR-amplified from gDNA as two separate DNA fragments that overlapped by 25 bp, with the intron eliminated by primer design (see Figure S2). Equimolar amounts (total amounts were ∼0.6–1.2 µg) of these three or four (see center and bottom of Figure S2), gel-purified, PCR fragments were used to transform AD∆. All Ura⁺ transformants (∼10–30) were confirmed by colony PCR, and >80% had the correct DNA sequence. The transformation rate using either three or four separate overlapping PCR fragments was only ∼5–10 times lower than the rates achieved for transformation with a single cassette.

Identification of individual alleles of the entire ABC11-ABC1 locus of C. krusei 89021, V2b, 90147, and 89221

To ascertain which allele of ABC11 of strain 89021 was adjacent to which allele of ABC1, the 1.3 kb XhoI/BglII fragment (see Figure S1B) was subcloned, and six individual clones were sequenced. gDNA sequences for the ABC11-ABC1 locus and sequences of individual ABC1 alleles for V2b, 90147 and 89221 were also determined.

Functional analysis of wild-type and recombinant yeast strains

The susceptibility of yeast to antifungal agents was measured by a microdilution assay as described previously (Niimi et al. 2004; Holmes et al. 2006). Agarose diffusion assays were performed to test the susceptibility of S. cerevisiae strains overexpressing C. krusei Abc1p or Abc11p to different xenobiotics. In brief, a 10 ml YPD overnight culture of each test strain was diluted 1:20 into 3 ml CSM medium, and grown at 30° for another 4 h to mid-logarithmic growth phase (OD₆₀₀ ∼1; ∼10⁷ cells/ml). The cells were diluted to OD₆₀₀ = 0.008 in 5 ml of molten CSM medium containing 0.6% agarose (45°), and overlaid on 20 ml CSM agar base medium. Whatman 3MM paper disks (5 mm diameter) containing xenobiotics were placed onto the solidified top agarose medium, and the plates were incubated at 30° for 48 h, or until growth inhibitory zones became clearly visible. The amounts of xenobiotics used are shown in the figure legends.

Chemosensitization assays

The chemosensitization of yeast strains overexpressing C. krusei Abc11p or Abc1p to FLC was carried out as described previously (Niimi et al. 2004). In brief, cells were cultured in CSM medium as for agarose diffusion assays. Each test strain was diluted to OD₆₀₀ = 0.008 in 10 ml of molten CSM containing 0.6% agarose (45°) with either no FLC (control to determine the toxicity of each drug) or FLC at 0.13× the MIC of FLC (MIC_FLC). The cell suspension was poured into a rectangular Omnitray plate (126 × 86 × 19 mm; Nunc, Roskilde, Denmark) that contained 20 ml of CSM solidified with 0.6% agarose either without (control), or with FLC. Whatman 3MM paper disks containing potential drug pump inhibitors (1 µg of milbemycins α11, α20, β9, and β11, 0.2 µg ENI, 1 µg beauvericin, 5 µg FK506, and 25 nmol OLI) were placed on the overlay, and the plates were incubated at 30° for 48 h.

Analysis of phylogenetic relationships and identification of potential yeast transcription factor binding sites (TFBS)

Phylemon 2.0 (Sanchez et al. 2011), a publicly available web-tool, was used for sequence alignments, tree reconstruction, and the estimation of synonymous and nonsynonymous distances. YeTFaSCo (de Boer and Hughes 2012), another web-tool, was used to search for potential yeast TFBSs upstream of ABC1, ABC11, and ABC12. Sequence alignments were performed with ClustalW v2.0.10 (Thompson et al. 1994) and viewed, and, if necessary, edited, with Jalview14.0 (Waterhouse et al. 2009). FigTree v1.4.0 (http://tree.bio.ed.ac.uk/software/figtree/) was used as a graphical viewer of phylogenetic trees and to produce publication-ready figures. Phylogenetic relationships were calculated by maximum likelihood analysis using PhyML v3.0 (Guindon and Gascuel 2003; Anisimova and Gascuel 2006). Evolutionary tests for the synonymous and nonsynonymous substitution rates between pairs of genes were performed with CodeML and yn00 (Yang and Nielsen 2000) from PAML v4.4c (Yang 2007).

Helical wheel projections and determination of lipid facing surfaces (LIPS) of individual TMS

The surfaces of individual TMSs that most likely face the lipid bilayer were predicted with the LIPS server (http://tanto.bioengr.uic.edu/lips/; Adamian and Liang 2006) using the sequences of the TMSs of 42 cluster A PDR transporters as input [a list of these transporters and their alignment, with the predicted TMS boundaries highlighted, can be found in Lamping et al. (2010)]. The helical wheel projections of individual TMSs were created online (http://rzlab.ucr.edu/scripts/wheel/wheel.cgi).

GenBank accession numbers

ABC11 and ABC1 sequences: 16.4 kb ABC11-ABC1 gDNA of 89021 (JX679710); 13.9 kb of the ABC11-ABC1 gDNA of 89221 (JX679711) and 90147 (JX679712); 7.7 kb ABC11, including 3.1 kb upstream, gDNA of V2b (JX679713) and 4.9 kb ABC1, including ∼100 bp upstream and downstream, gDNA of V2b (JX679714); 10 kb ABC11-ABC1 locus (from ABC11 start to ABC1 stop) of the A allele (FJ445766) and the B allele (FJ445767) of 89021; 10 kb ABC11-ABC1 locus of IFO0011 (KT716056) and B2399 (KT716053); 15.3 kb ABC11-ABC1-11-chimera1-ABC1 locus of 89102 (KT716058) and 15.3 kb ABC11-ABC1-11-chimera2-ABC1 locus of 89102 (KT716059); 5.3 or 4.9 kb ABC1 A and B alleles (including the promoter and terminator) of 89221, 90147, and V2b (FJ445760-5); 6.3 kb HindIII/BamHI fragment and ABC1 cDNA of B2399 [DQ903906-7 (Lamping et al. 2009)]; 4654 bp ABC11-1 chimera of IFO0011 (KT716057), ABC11-1 chimera1 (KT716054) and ABC11-1 chimera2 (KT716055) of B2399, and ABC11-1 chimera of 89102 (KT716060). ERG11 sequences: ERG11 A (FJ445752) and ERG11 B (FJ445753) alleles of 89102.

Data availability

Strains are available upon request. Sequence data are available at GenBank and the accession numbers are listed above in the Materials and Methods section. File S1 contains Tables S1–S4, and detailed descriptions of all Supplemental Material (i.e., legends for Figure S1, Figure S2, Figure S3, Figure S4, Figure S5, Figure S6, Figure S7, Figure S8, Figure S9, Figure S10, Figure S11, and Figure S12).

Results

Characterization of a representative set of C. krusei isolates

Although C. krusei, also known as Issatchenkia orientalis or Pichia kudriavzevii, is a relatively poorly studied ascomycetous yeast, it is important for natural food fermentation (Chan et al. 2012), especially of cocoa beans for chocolate production. It is also an opportunistic fungal pathogen of humans that has reduced susceptibility to azole antifungals due, in part, to pump-mediated drug efflux (Lamping et al. 2009). Odds and colleagues developed a MLST database for clinical C. krusei isolates from all parts of the world (Jacobsen et al. 2007). Their data indicated that C. krusei is diploid (there was no mention of aneuploid or triploid strains), and there was no geographical association for particular subtypes. The high proportion (77%) of singletons (i.e., unique diploid sequence types; DSTs), despite almost saturated levels of haplotypes, was taken as a sign that sexual recombination may play an important part in the C. krusei life cycle (Jacobsen et al. 2007).

The seven isolates used in the present study comprise a diverse range of isolates from humans, animals, and the environment; isolates from three different continents, different body sites, and different years of isolation (see Table S1 in File S1). MLST confirmed that strain 89021 was closely related to 90147 and V2b, and, to a lesser extent, 89221, whereas the genotypes of 89102, IFO0011, and B2399 were somewhat different (see Table S3 in File S1). As expected from Odds and colleagues’ study (Jacobsen et al. 2007), only six (out of 42; ∼14%) new haplotypes were identified, yet six of the seven isolates (∼86%) had unique DSTs (see Table S3 in File S1). Interestingly, New Zealand isolate 90147 had a DST (63) that was identical to an isolate from Italy (see Table S3 in File S1). The peak height ratios of single nucleotide polymorphisms (SNPs) from sequencing chromatograms easily distinguished a diallelic (1:1 ratio) from a triallelic (1:2 ratio) sequence. All SNPs of 89021, 89221, 90147, V2b and IFO0011 had 1:1 peak ratios. However, all B2399 SNPs had 1:2 peak ratios, whereas 89102 exhibited 1:2 peak ratios for ERG11 and ABC1, but 1:1 peak ratios for all housekeeping genes (see Table S4 in File S1). This implied that C. krusei B2399 was triploid, 89102 was aneuploid with a trisomy in the ERG11- and ABC1-containing chromosome, and 89021, 89221, 90147, V2b, and IFO0011 were diploid, which was confirmed by measuring the fluorescence of cells stained with the DNA intercalating agent SYBR green (see Figure S3).

Discovery of the tandem-duplicated multidrug efflux transporter pair ABC11-ABC1 of C. krusei 89021

We previously characterized 6.3 kb (between the dashed vertical arrows; Figure 1A) of C. krusei B2399 gDNA containing the major multidrug efflux pump ABC1 (Lamping et al. 2009). Overexpression of Abc1p in S. cerevisiae AD∆ caused multidrug resistance that could be reversed with known efflux pump inhibitors (Lamping et al. 2007, 2009). But, although C. krusei B2399 was triploid, averaging one SNP/400 nt for the ∼10 kb ERG11 locus (Lamping et al. 2009), ABC1 of C. krusei B2399 had no detectable SNPs (Lamping et al. 2009). Southern blot results for C. krusei B2399 gDNA, however, gave two unexpected EcoRI bands, indicating allelic variation or a recent tandem-duplication of ABC1 (Lamping et al. 2009). This led ultimately to the discovery of the tandem-duplicated ABC11-ABC1 locus of C. krusei 89021 (Figure 1A). Both ends of the 16.4 kb sequence contained partial ORFs (996 and 973 bp, respectively) of unknown function, and an additional 1512 bp ORF with high homology to the MATE (Multidrug And Toxic compound Extrusion) family of drug/sodium antiporters ∼2 kb upstream of ABC11 (Figure 1A). ABC11 and ABC1 of 89021 had 19 and 16 SNPs, respectively (i.e., an average SNP frequency of one SNP/532 nt; Figure 1B). The SNP frequency for the noncoding regions was ∼2.3 times higher (one SNP/244 nt), similar to the previously reported ∼threefold difference between coding and noncoding regions for the 10 kb ERG11 locus of B2399 (Lamping et al. 2009). The average SNP frequency of one SNP/437 nt for the entire 16.4 kb was comparable with diploid Candida CTG clade pathogens such as C. albicans (one SNP/330–390 nt) or C. tropicalis (one SNP/576 nt; (Butler et al. 2009)). Interestingly, ABC1 and ABC11 also shared one identical intron (gray box; Figure 1A).

Figure 1.

Discovery of the tandem-duplicated multidrug efflux pump pair ABC11-ABC1, and polymorphism data for the ABC11 and ABC1 alleles of C. krusei 89021. ABC11 and ABC1 gene sequences are blue and green, respectively, and short and tall vertical lines are synonymous (sSNPs) and non-synonymous SNPs (nSNPs) between the alleles of each gene, respectively. The two vertical dashed arrows in (A) indicate previously characterized sequence (Lamping et al. 2009). (A) The 16.4 kb locus surrounding the C. krusei 89021 tandem-duplicated gene-pair ABC11-ABC1. ORFs and their orientation are indicated; gray boxes within ABC1 and ABC11 represent the common 88 bp intron. (B) A SNP map of ABC11 and ABC1 for the two chromosomes (Chr) of the diploid C. krusei 89021. Allelic differences between the ORF sequences are shown as magenta (Chr A) and purple (Chr B) SNPs, respectively. The map underneath shows the polymorphism pattern for the diploid sequence with red SNPs indicating polymorphisms that are the same in both ABC11 and ABC1 and black SNPs representing unique polymorphisms.

Sequence identity interspersed by four short stretches of sequence divergence distinguishes ABC1 from ABC11

We determined the sequences of 22 ABC1 and ABC11 alleles for the seven C. krusei strains (Figure 2) in order to identify the minimum (core) differences between ABC1 and ABC11. Strains B2399, IFO0011 and strain 89102 had only one ABC11 and one ABC1 allele each (Figure 2B). However, the remaining four diploid C. krusei strains 89021, V2b, 90147, and 89221 each had two ABC1 and two ABC11 alleles (Figure 2A). In fact, all their ABC11-A, ABC1-A, and ABC1-B alleles were identical, apart from 89221 ABC1-B, which had one additional synonymous SNP (sSNP) in patch I (Figure 2A), and strain 89221, which had the two SNPs immediately upstream of patch II of ABC11-A on Chr A homogenized with ABC1 sequence by a very recent EGC event (black arrow; Figure 2A). In contrast, the four ABC11-B alleles differed slightly from each other (Figure 2A). Overall, there were 92 core nucleotide differences (plus one extra codon near the stop codon of ABC11; see Figure S4) between ABC11 and ABC1 limited to six short (415 nt or 8.9% of the ORF) DNA stretches I–VI colored blue and green, respectively, in Figure 2. The five “fixed” SNP differences (blue and green SNPs; Figure 2) in both ABC11 and ABC1 near patches II, IV, and VI found in most strains were excluded from the core differences because they were homogenized by four EGCs in strains 89221, IFO0011 and B2399, respectively (all in the direction from ABC1 to ABC11; black arrows in Figure 2). The six DNA patches I–VI that encoded the core differences between ABC11 and ABC1 were conserved in all but one of the 22 ABC11 and ABC1 alleles, with ABC11 of strain 89102 having adopted ABC1-specific patches III and V by two separate EGCs from the neighboring ABC1 allele (Figure 2B). It would appear that ABC11 patches III and V are possibly less important for the biologically preserved Abc11p function.

Figure 2.

Sequence comparisons of ABC11 with physically linked ABC1 alleles of seven C. krusei strains. Within ORFs, the thick blue and green vertical lines represent the core sequence differences between ABC11 (blue) and ABC1 (green) found in all ABC11 and ABC1 alleles (i.e., patches I–VI), while the thin blue and green lines outside these patches are SNPs that are found in almost all ABC11 (blue) or ABC1 (green) alleles. Orange SNPs represent SNPs unique to either ABC11 or ABC1 found in that particular allele. (A) Sequence differences between physically linked ABC11 and ABC1 alleles of Chr A and Chr B of the diploid C. krusei strains 89021, V2b, 89221 and 90147 are shown in the top and bottom panels, respectively. One EGC that homogenized sequences between patches I–II of Chr A of 89221 in the direction ABC1 to ABC11 (black arrows) is illustrated. This caused the loss of two blue “fixed” SNPs immediately upstream of ABC11 patch II. (B) Map of sequence differences between physically linked ABC11 and ABC1 alleles of C. krusei B2399, IFO0011 and 89102; 89102 contained additional ABC1-11 chimeras between ABC11 and ABC1 (Figure 4A). Five EGCs homogenized sequences between patches IV–V (B2399 and IFO0011), V–VI (IFO0011), and they converted patches III (89102) and V (89102) of ABC11 to ABC1 in 89102. They were all in the direction ABC1 to ABC11 (black arrows).

ABC1 and ABC11 are under very strong purifying selection and have evolved by concerted evolution for ∼134 MY

The vast majority of SNPs between individual alleles of ABC1 and of ABC11 were synonymous. Strain 89021, for example, had 16 ABC1 and 19 ABC11 SNPs, 15 and 16 of which were sSNPs, respectively (short vertical lines; Figure 1B). The nonsynonymous to synonymous nucleotide substitution ratio dN/dS (ω) of 0.052 for ABC1 and 0.053 for ABC11 of 89021 indicated that both genes were under very strong purifying selection. Even the ω value for the concatenated 252 nt of patches II–IV was not much higher (0.101; Table 1), suggesting that Abc1p and Abc11p subfunctionalized rather rapidly after duplication, and have since been under very strong purifying selection. The ω value for the concatenated 159 nt of patches I and VI, encoding the N- and C-termini, respectively, was much higher (0.347), suggesting that these 53 aa were of little functional consequence and, thus, under reduced purifying selection. ABC11 was 97.8% identical to ABC1, with a calculated dS of 0.05 for the entire ORF, suggesting a rather recent duplication event ∼3.2 MYA (Table 1). However, dS within “protected” patches II–IV was much higher (1.18; Table 1), indicating that the duplicates were possibly much older (∼71 MY; Table 1).

Table 1. Silent-site nucleotide substitution rates dS of C. krusei 89021 and P. membranifaciens ABC1, ABC11, and ABC12, and estimated times of duplication.

Sequences Compared	NBDs-TMDs			Patches II–IV			Ta (MY)
	dS	dN	dN/dS	dS	dN	dN/dS
CkABC1 - CkABC11	0.05	0.009	0.191	—	—	—	3.2
	—	—	—	1.18	0.119	0.101	71
PmfABC1 - PmfABC11	0.41	0.061	0.150	—	—	—	26
	0.75	0.104	0.139	—	—	—	46b
	0.15	0.017	0.116	—	—	—	10c
CkABC1 - PmfABC1	2.13	0.114	0.054	—	—	—	137
CkABC11 - PmfABC11	2.01	0.121	0.060	—	—	—	130
CkABC1 - CkABC12	2.42	0.168	0.069	—	—	—	156
PmABC1 - CkABC12	2.30	0.184	0.080	—	—	—	148
CkABC11 - CkABC12	2.48	0.172	0.070	—	—	—	160
PmABC11 - CkABC12	2.03	0.183	0.090	—	—	—	131
ScPDR5 - ScPDR15	1.55	0.152	0.098	—	—	—	100
	—	—	—	1.67	0.073	0.044	100

^aEstimated time since duplication or speciation (when comparing ORFs of C. krusei with P. membranifaciens). ScPDR5 and ScPDR15 are ohnologs of a whole genome duplication event ∼100 MYA (Wolfe and Shields 1997; Kellis et al. 2004). The times of retrotransposition of CkABC12 (∼149 MYA; i.e., from between 131 to 156 MYA), and duplication of C. krusei and P. membranifaciens ABC1/11 (∼134 MYA; i.e., from between 130 to 137 MYA) from their respective ancestors was estimated by comparing the silent-site substitution rates dS for the concatenated sequences of the conserved regions encoding NBD1-TMD1, NBD2-TMD2, or both NBDs and TMDs (NBDs-TMDs; the concatenated sequences included regions encoding amino acids Q99-E823 and N847-S1437 of CkABC1 and CkABC11, amino acids D89-E813 and D848-M1433 of PmfABC1, and E95-E819 and D854-M1439 PmfABC11, E61-Q785 and S811-S1401 of CkABC12, D92-F808 and I868-K1503 of ScPDR5 and N102-F818 and I883-K1518 of ScPDR15) of the indicated ORFs, and dS for the sequences encoded by patches II–IV of CkABC1/CkABC11 (1.18) with those of ScPDR5/ScPDR15 (i.e., 1.55 and 1.67, respectively), assuming S. cerevisiae, C. krusei, and P. membranifaciens genes experience a similar dS/gene/MY.

^bCompare conserved NBD1-TMD1 sequences.

^cCompare conserved NBD2-TMD2 sequences.

This rather high estimate of ∼71 MY for a 97.8% identical tandem-duplicated pair of paralogs had to be further revised with the discovery of the P. membranifaciens ABC11-ABC1 tandem-duplicate. The synteny of the ABC11-ABC1 locus between C. krusei and P. membranifaciens was well conserved, which strongly suggested common ancestry (Figure 3A). The P. membranifaciens ORFs were also closely linked (separated by just 548 nt), and PmfAbc1p and PmfAbc11p each were 79% identical, and 87 and 89% homologous to CkAbc1p and CkAbc11p, respectively. Sequence comparison of PmfABC1 (391367–395992 of scaffold2-contig42) with PmfABC11 (386144–390818 of scaffold2-contig42; Figure 3B) clearly showed how parts of these two paralogs are, like C. krusei ABC1 and ABC11, still evolving by concerted evolution, but in a different fashion (cf. Figure 3, B and C). PmfABC1 and PmfABC11 had a few shorter identical sequence patches in regions encoding the less conserved nucleotide-binding domain 1 (NBD1), there were no patches of sequence identity in ∼1/4 of the ORF encoding transmembrane domain 1 (TMD1), but alignments of the 3′ halves of PmfABC1 with PmfABC11 encoding the highly conserved NBD2 and the TMD2 regions showed a somewhat similar pattern to CkABC1/CkABC11 (Figure 3, B and C). As with C. krusei, the estimated age for this pair of paralogs varied dramatically depending on which dS values were used for the calculations. Using dS for the entire ORF (0.41) gave an estimated age of 26 MY, the dS for the C-terminal half (0.15) gave an even lower estimate of ∼10 MY, but dS for the N-terminal half (0.75), which has almost completely escaped concerted evolution, gave an age of 46 MY (Table 1). The most accurate age-estimate for the duplication event of this pair of paralogs was obtained, however, when we compared the two pairs of orthologs with each other (Figure 3, D and E), because they had evolved independently from each other after speciation. There was surprisingly low (∼76%) sequence homology across the entire ORF (Figure 3, D and E) reflected in surprisingly high dS values for the ABC1 (2.13) and ABC11 (2.01) orthologs, providing an average age estimate of 134 MY since speciation (Table 1).

Figure 3.

Physical maps of the ABC11-ABC1 loci of C. krusei and P. membranifaciens, and sequence comparisons between the two pairs of paralogs and orthologs. (A) Physical maps of the ABC11-ABC1 loci of C. krusei (top) and P. membranifaciens (bottom). ORFs are named according to orthology with the S. cerevisiae genome database (SGD) where possible. Assembly of the C. krusei ABC11-ABC1 consensus is described in Figure S7. The P. membranifaciens ABC11-ABC1 locus was contained in the ∼459 kb scaffold2-contig42 (from 372000 to 404000). Synteny between C. krusei and P. membranifaciens was remarkably well preserved. Numbers within the C. krusei ORFs show their % GC3 content (see text for further details). (B–D) Representations of sequence alignments between indicated ORFs. Regions of identity are in blue and individual nucleotide differences are shown as vertical white lines. Light blue regions are gaps in the alignment. Each lane is 140 nt long. The alignments in (B and C) demonstrate how the ABC11-ABC1 paralogs of P. membranifaciens and C. krusei both are still experiencing concerted evolution with regions of sequence identity interspersed by regions of sequence divergence that are protected from EGCs by strong purifying selection. However, the sequence comparisons between the “orthologs” of C. krusei and P. membranifaciens [i.e., CkABC1 with PmfABC1 (D) and CkABC11 with PmfABC11 (E)] show that these orthologs are actually much older (∼134 MY; Table 1), exhibiting limited sequence identity (∼75%) randomly distributed across their entire ORFs.

Thus, we conclude that the majority of C. krusei ABC1 and ABC11 evolved by concerted evolution for possibly >134 MY; however, patches II–V of C. krusei ABC11–ABC1 were strongly protected against frequent EGCs for ∼70 MY in order to preserve important functional differences between the two gene products. This hypothesis was supported by the fact that the majority of SNPs between patches I–VI of ABC1 and ABC11 were shared polymorphisms (10 red SNPs; Figure 1B)—a characteristic feature for the concerted evolution of duplicated genes (Innan 2003; Osada and Innan 2008; Arguello and Connallon 2011).

Intramolecular and intermolecular (non)homologous recombination events caused gene deletions and CNVs, and generated a number of ABC11-1 chimeras

The chromosomal arrangements for ABC1 and ABC11 of B2399, IFO0011 and 89102 were difficult to unravel because of apparent CNVs. After careful consideration, we were able to successfully amplify, clone, sequence, and functionally express the “hidden” gene copies in S. cerevisiae AD∆. These copies comprised two ABC1–11 and four ABC11–1 chimeras (Figure 4A), which is why these “hidden” genes could not be PCR amplified with ABC1- or ABC11-specific primer pairs. ABC11 and ABC1 of IFO0011 and B2399 were very similar to the A alleles on Chr A of all other diploid strains (Figure 2). However, both strains had lost one gene copy each, probably via intramolecular homologous recombination between one of three different DNA stretches of sequence identities between ABC11 and ABC1 on IFO0011 ChrB and B2399 ChrA’ and ChrB (Figure 4, A and B). This generated three different ABC11-1 chimeras, either possessing the majority of the 10 shared A or the 10 shared B allele-specific SNPs of strain 89021 (see Figure S5), with all three ABC11-1 chimeras being put under the control of the ABC11 promoter (Figure 4A). A different, intermolecular, nonhomologous mitotic recombination event between patches I and II of ABC1 on ChrA and ABC11 on ChrB in the ancestor of 89102 (Figure 4C) is likely to have generated a single ABC11-1 chimera on ChrA, and an additional ABC1-11 chimera between ABC11 and ABC1 on ChrB (Figure 4A). A more detailed description of how 89102 possibly obtained its complex chromosomal arrangement since that initial intermolecular, nonhomologous, recombination event can be found in Figure S6B.

Figure 4.

C. krusei IFO0011, B2399, and 89102 possess copy number variations (CNVs) for ABC1 and ABC11. (A) Chromosomal arrangement of ABC1 and ABC11 alleles found in C. krusei 89021 (the same arrangement was found in C. krusei 89221, 90147 and V2b), IFO0011, B2399, and 89102. (B) Three separate gene deletions caused by intramolecular homologous recombination between homogenized DNA tracts 1, 2, and 3 of ABC11 and ABC1 can explain the loss of gene copies and the resulting chimeric alleles found in three chromosomes of strains IFO0011 (Chr B) and B2399 (Chr A’ and Chr B). (C) The ancestor of 89102 likely experienced an intermolecular nonhomologous mitotic recombination between patches I and II of ABC1 on Chr A and ABC11 on Chr B (a detailed evolutionary history of the unusual chromosomal arrangement of strain 89102 is proposed in Figure S6B).

The fact that all physically linked ABC11 and ABC1 ORFs grouped into either A or B allele types (see Figure S6C) clearly demonstrated that intermolecular meiotic crossovers and allelic gene conversions between neighboring copies of the ∼10 kb ABC11-ABC1 locus were much rarer than EGCs; otherwise, we would not have found such clear demarcations of almost exclusively shared (red) or allele-specific (black; Figure 1B) SNPs between individual tracts of sequence divergence, and, most importantly, we would have expected a significant reduction in the SNP frequency in parts that had experienced a recent allelic gene conversion, which clearly was not the case—random SNP distribution between individual alleles of ABC11 and ABC1 is evident in Figure 1B.

First direct estimates for the gene conversion rate and the size of the MEPS

Polymorphism-based methods can overcome the methodological limitations that preclude precise estimates for the rate of EGC (Arguello and Connallon 2011). Therefore, we exploited the sequence information we had collected from a sample of strains representative for the C. krusei population to obtain the first “direct,” empirical, estimates for the gene conversion rate without relying on computer modeling.

The polymorphism patterns in Figure 1B suggested that ten recent EGCs (Figure 5A) between individual patches of sequence divergence between physically linked ABC11 and ABC1 gene copies of C. krusei 89021 homogenized an entire tract in one event rather than in multiple smaller EGC events. This conclusion is based on the observation that tracts exposed to frequent EGCs between strongly protected patches of sequence divergence contained either almost exclusively shared polymorphisms or allele-specific SNPs randomly distributed across entire tracts (Figure 1B). If, for example, the most recent EGC homogenized only half a tract between ABC11 and ABC1, then one half would be expected to contain shared polymorphisms only (red SNPs), and the other half a number of allele-specific SNPs (black SNPs), but that clearly was not the case. In addition, as mentioned above, there was also no noticeable interference with the observed polymorphism patterns by allelic gene conversions or meiotic crossover events.

Figure 5.

The ten most recent EGCs between patches I–VI of C. krusei 89021 ABC11 and ABC1 enable empirical estimates for important gene conversion parameters. (A) Ten separate recent EGCs, five (EGC 1–5) between “protected” patches I–VI of the A alleles and five (EGC 6–10) between “protected” patches I–VI of the B alleles of ABC11 and ABC1, of unknown direction (black arrows), caused almost identical sequences for the majority of the two physically linked ORFs. (B) Relationship between the estimated average age of the five most recent EGCs between the six patches of core sequence differences of ABC11 and ABC1 of C. krusei 89021, and the tract length between these individual patches of sequence difference. (C) Relationship between the calculated gene conversion rate per gene per 100 MY and the tract length between these patches of sequence difference. MEPS = x when y = 0 (106 nt; arrow pointing to MEPS). (D) Percent GC content in codon position 3 (GC3) for 100 bp segments of ABC1 (green) and ABC11 (blue). Patches I–VI are indicated with vertical gray bars, and the average GC3 content within and between patches I–VI is indicated with horizontal gray and dotted lines, respectively.

Given the clarity of these results, we applied a molecular clock (assuming an estimated nucleotide substitution rate of 0.8 per silent-site per 100 MY for fungi (Lynch and Conery 2000), and one, rather than multiple, gene conversions per tract) to estimate the age of these 10 most recent EGC events (Table 2). The age of the 10 most recent EGC events was calculated by counting the number of sSNPs that had accumulated between individual patches since the last EGC and calculating dS. However, these age estimates could be up to ∼8 times higher or ∼2 times lower, because the nucleotide substitution rates can range anywhere from ∼0.1 to ∼1.6 per silent-site per 100 MY in different organisms (Lynch and Conery 2000). The paucity of new SNPs in EGC tracts reflects the higher rate of EGC than natural mutation. With these limitations in mind our best estimates for the average age of the four recent EGCs between the two largest tracts (1.2 and 1.7 kb; i.e., EGCs 3 and 8 and EGCs 1 and 6; Figure 5A) were 0.43 and 0.13 MY, respectively, while the calculated average age for the three smaller tracts (339–399 bp; i.e., EGCs 2 and 7, EGCs 4 and 9, and EGCs 5 and 10; Figure 5A) ranged from 0.53 to 1.68 MY (Table 2). Assuming a constant gene conversion rate, these equated to 30–385 gene conversions per gene per 100 MY (Table 2). Because all five tracts between the “protected” patches of sequence divergence of the two paralogs, ABC1 and ABC11, were separated by the same distance (∼5.3 kb), there appeared to be little interference with EGC by allelic gene conversion, and, because the effect of genome location could be ignored (as all events occurred in the same location), we could tentatively investigate the influence of tract length on the EGC rate in C. krusei. Despite the limited number of EGC events there was an inverse correlation between the average age of the most recent EGC and its tract length (Figure 5B), and a direct linear correlation between the gene conversion rate and its tract length (Figure 5C). Extrapolating the graph in Figure 5C to y = 0 gave a direct estimate for the MEPS of 106 nt for duplicated loci separated by 5.3 kb in C. krusei.

Table 2. Estimation of the gene conversion rate between patches of individual alleles of the ABC11-ABC1 locus of C. krusei 89021.

DNA Between	Tract Length (bp)	NSa	sSNPs	sSNPs	Y (10−6) Since Last Gene Conversionc		Average Time Since Last Gene Conversion (MY)	Gene Conversion Rate/Gene/100 MY
			Ab	Bb	A	B
I–II	1650	492	0	0	<0.13	<0.13	0.13	385
III–IV	1209	360	0	4	<0.17	0.69	0.43	116
V–VI	399	117	1	0	0.53	<0.53	0.53	94
IV–V	375	112	0	5	<0.56	2.79	1.68	30
II–III	339	93	1	0	0.67	<0.67	0.67	75

The 88 bp intron was removed for this analysis.

^aN_S is the calculated number of possible synonymous sites.

^bThese are the numbers of sSNPs in the A or B alleles between individual patches of ABC11 and ABC1.

^cThe time since the last gene conversion event was calculated by using the synonymous nucleotide substitution rate (dS) as a molecular clock, assuming an estimated dS of 0.8 per synonymous site per 100 MY for fungi (Lynch and Conery 2000) (N_S is twice the above listed value because one gene conversion event “homogenizes” the sequences of both paralogs, ABC1 and ABC11).

Gene conversion in C. krusei shows GC-bias

There is substantial evidence that frequent EGCs lead to increased GC-content, especially at the third positions in codons (GC3) that are under little selective constraint, due to GC-bias in mismatch repair (Galtier 2003; Noonan et al. 2004). Others, however, have found little evidence for GC-bias due to mismatch repair, and argued that GC-content may favor EGC rather than EGC being GC-biased (Casola et al. 2010; Arguello and Connallon 2011). The average GC3-content of those parts of ABC1 and ABC11 that were exposed to frequent EGCs was significantly higher (∼50%) than the GC3-content (∼35%) for patches I–VI that were strongly protected against frequent EGCs (Figure 5D). The GC3-content of patches I–VI was similar to the GC-content (38.3%) for the entire C. krusei M12 genome (Chan et al. 2012), and, most importantly, the GC3-content of all neighboring genes (Figure 3A), and all C. krusei PDR transporters including ABC12, which shares a common ancestor with ABC1 and ABC11 (see below), were equally low (∼38%) (data not shown). Our data therefore favor a model of GC-bias of mismatch repair for EGC in C. krusei.

Elucidation of the ABC11-ABC1 locus

ABC1 and ABC11 belong to cluster A of the nine clusters (A-H2) of fungal PDR transporters (Lamping et al. 2010). We searched recently released draft genome sequences of C. krusei M12 (Chan et al. 2012), SD108 and NBRC 1279T, environmental isolates from Malaysia, the United States and Russia, respectively (see Table S1 in File S1), for all possible PDR transporter homologs, and identified three cluster A (ABC1, ABC11 and ABC12), six cluster D (two S. cerevisiae Pdr12p and four S. cerevisiae Snq2p homologs), and one cluster F member (data not shown). No ABC1 or ABC11 ORF sequences were found in NBRC 1279T, although contig 614 contained ∼9 kb DNA upstream of ABC11, and included ABC11 patch I, and contig 294 contained ∼6 kb DNA downstream of ABC1, and included ABC1 patch VI (see Figure S7). However, five M12 contigs and five SD108 scaffolds contained ∼50% of ABC1 and ABC11 including ∼14 kb (contig 327) and ∼113 kb (scaffold 21) of ABC11 upstream DNA and ∼6 kb (contig 351) and ∼41 kb (scaffold 21) of ABC1 downstream DNA (see Figure S7). SD108 scaffold 21, the longest contig (161 kb), contained a predicted ABC11-1 chimera with an ∼600 bp N-terminal insertion following ABC11 patch I (see Figure S7). Despite some unresolved regions of M12, the available ABC1 and ABC11 ORF sequences matched the A alleles of 89021 (100% identity), whereas the available SD108 sequences matched the B alleles, with only 8 nt differences to the B alleles of 89021, all of which were also detected in some B alleles of the other six C. krusei strains (data not shown). Scaffold 21 also contained the C. krusei azole drug target ERG11 just 59.9 kb upstream of the ABC11-1 chimera, confirming our observation that ERG11, ABC1, and ABC11 are closely linked on the same chromosome.

Discovery of ABC12

We named the third cluster A PDR transporter discovered in another part of the C. krusei genome Abc12p, because it showed highest homology (84%) to Abc1p and Abc11p. All three recent C. krusei genome databases agreed on the chromosomal assembly for ABC12 encoded on M12 contig 30, NBRC 1279T contig 223, and SD108 scaffold 33. ABC12 of C. krusei M12 had only four or eight SNPs compared to ABC12 of SD108 or NBRC 1279T, respectively, with just 1 aa difference in Abc12p between M12 and SD108 (H1424R) or NBRC 1279T (K781E). Interestingly, P. membranifaciens had only two cluster A PDR transporters, PmfABC1 and PmfABC11, indicating it had lost its ABC12 ortholog. Comparing a map for the C. krusei ABC12 locus with a map of the same chromosomal region of P. membranifaciens (Figure 6A) indicated how P. membranifaciens may have lost ABC12. There was an inversion of ∼20 kb (gray; Figure 6A) immediately upstream of C. krusei ABC12 (pink; Figure 6A), and ABC12 may have been lost during this inversion event (PmfABC12 is the only missing ORF at this locus; Figure 6A). Interestingly, three of the seven C. krusei strains (89021, 89221, and 90147) had only one intact copy of ABC12, with the second allele being an inactive pseudogene [i.e., a C2275T mutation caused Q759 (CAA codon) to become a TAA stop codon].

Figure 6.

Physical maps of the ABC12 loci of C. krusei and P. membranifaciens, and a topology map of aa differences between multidrug efflux pumps C. krusei Abc1p and Abc11p. (A) Physical maps of the ABC12 loci of C. krusei (top) and P. membranifaciens (bottom). ORFs are named according to orthology with the S. cerevisiae genome database (SGD) where possible. The C. krusei ABC12 assembly is a consensus between M12 contig 30 (ABC12 = 56555–61015; Abc12p = 1486 aa), SD108 scaffold 33 (ABC12 = 62859–67319) and NBRC 1279T contigs 223 (ABC12 = 39451–43911) and 259 (3′ end from DST1). The P. membranifaciens locus was within the ∼153 kb scaffold2-contig136 contig (from 56000 to 87000). There was no ABC12 homolog in P. membranifaciens. It appears that P. membranifaciens lost ABC12 (pink ORF in C. krusei) during an inversion of the sequence (∼18 kb) highlighted in gray just upstream of C. krusei ABC12. (B) Maximum likelihood tree for C. krusei and P. membranifaciens Abc1p and Abc11p and C. krusei Abc12p and their phylogenetic relationship to the cluster A PDR transporters of representative species of the three major Saccharomycotina lineages: Saccharomycetaceae (green) species S. cerevisiae (Sc) and C. glabrata (Cg), Candida CTG (magenta) species C. albicans (Ca) and Pichia stipitis (Ps), and the third major Saccharomycotina lineage (blue) species H. parapolymorpha (Hp), P. membranifaciens and C. krusei. Y. lipolytica cluster B PDR transporter YliQ6CET9 was used as the outgroup. The bootstrap support values of 100 replicates are shown above individual branches, and branches with 100% bootstrap support are marked with an asterisk. (C) Predicted topology of the Abc1p and Abc11p TMDs with the location of the 18 core aa differences within patches II, III, IV, and V highlighted. The gray background represents the membrane bilayer. Open boxes represent TMS1–12 (from left to right), with ELs and ILs connecting individual TMSs in yellow and blue, respectively. The regions corresponding to patches II–V are in gray boxes, and amino acids that differed between Abc1p and Abc11p of C. krusei B2399 are shown as circles. Sequence comparison with their common ancestor Abc12p revealed which differences between Abc1p and Abc11p were due to changes in Abc1p (two green circles), Abc11p (13 blue circles), or in both Abc1p and Abc11p (three red circles). Amino acids with thick circles are unique changes when compared with all other fungal cluster A PDR transporters.

C. krusei belongs to a recently discovered third major Saccharomycotina lineage

Although yeast comprise a broad phylogenetic range of species (Kurtzman 2011), research within the large Saccharomycotina family has primarily focused on members of the Saccharomycetaceae and the Candida CTG lineages (Tsui et al. 2008; Dujon 2010). This is why the third major Saccharomycotina lineage has only recently been discovered (Morales et al. 2013; Hittinger et al. 2015; Riley et al. 2016). It includes the methylotrophic Pichia pastoris (De Schutter et al. 2009) and Hansenula polymorpha (Ravin et al. 2013), the nitrate-assimilating yeasts H. polymorpha (Ravin et al. 2013), Dekkera bruxellensis (Curtin et al. 2012), and Kuraishia capsulata (Morales et al. 2013), and also C. krusei and P. membranifaciens (Riley et al. 2016). It is now clear that the Saccharomycetaceae lineage is the oldest branch, and the Candida CTG clade and the third major Saccharomycotina lineage separated later (Morales et al. 2013; Riley et al. 2016). A ClustalW-alignment of all cluster A PDR transporters of five Candida CTG, seven Saccharomycetaceae and two species (H. polymorpha and C. krusei) of the third Saccharomycotina lineage is shown in Figure S8. Abc1p, Abc11p, and Abc12p, as well as H. polymorpha HpCdr2p and HpCdr4p, had two characteristic 3-aa deletions at the center of extracellular loop EL3 and the beginning of EL6 (see Figure S8 and Figure S9). A maximum likelihood tree of all cluster A PDR transporters from two or three species of each of the three Saccharomycotina lineages matched their species tree (Morales et al. 2013; Riley et al. 2016) with good bootstrap support (Figure 6B). These findings confirmed that C. krusei and P. membranifaciens belong to a third, only recently discovered, major Saccharomycotina lineage.

Intron loss in ABC12 and deintronization in ABC1 and ABC11

The ClustalW alignment of ABC sequences (see Figure S8) highlighted a 30-aa N-terminal insertion in C. krusei Abc1p and Abc11p, absent in all other homologs (Figure 7A shows an alignment of this region for C. krusei and P. membranifaciens Abc1p, Abc11p, and C. krusei Abc12p). We hypothesized that the common ancestor of ABC1, ABC11, and ABC12 had at least two introns, which were lost in ABC12 by retrotransposition into another part of the genome, while ABC1 and ABC11 converted (i.e., deintronized) one intron into coding sequence, but kept the second intron intact. To test this hypothesis, we searched the DNA sequences encoding the 15–30 aa insertion in ABC1 and ABC11 for remnants of the conserved yeast splice site (ss) signals [i.e., 5′ ss, 3′ ss, and branch site (BS) signal; (Kupfer et al. 2004)]. There was reasonable homology between ABC1/ABC11 and ABC12 before and after the predicted intron boundaries, particularly for C. krusei ABC1 and ABC11, and splice site remnants were still visible (Figure 7B). The predicted 5′ ss GTACG was changed to ACACG, the BS TACTAAC was changed to TTCAAAC, and the potential 3′ ss TTNNTAG was changed to ATNNCAG in C. krusei ABC1 and ABC11 (Figure 7B). The two most significant changes, each of which would abolish splicing, are the GT to AC changes of the 5′ ss, and the T to A change of the BS. Deintronization of the inferred 92 bp intron may have caused a frameshift and a temporary pseudogene, which was possibly recovered by the inferred 10 bp insertion just 21 nt downstream (Figure 7B). Because all four Abc1p/Abc11p paralogs share this unusual insertion (Figure 7A), it would seem that, after the proposed initial deintronization event (possibly not long before speciation; ∼134 MYA) P. membranifaciens ABC1 and ABC11 responded quite differently to the functional disturbance caused by the ∼30-aa insertion; PmfABC1 had a 45- and PmfABC11 had an 18-nt deletion (Figure 7B). However, without a cluster A PDR transporter sequence that still contains this “ancient” intron, the proposed deintronization could have also been a random insertion event instead.

Figure 7.

Evidence for a possible deintronization event in the ancestor of C. krusei and P. membranifaciens ABC1 and ABC11, and evidence for intron loss by retrotransposition in C. krusei ABC12. Identical sequences are highlighted in dark blue and residues that are >80% conserved in (A) are in light blue. (A) Alignment of the N-termini of Abc1p, Abc11p, and Abc12p revealed a 30-aa insertion (the area highlighted in gray) in CkAbc1p and CkAbc11p, and slightly shorter 15 and 24 aa insertions in PmfAbc1p and PmfAbc11p, respectively. (B) Alignment of the DNA sequences encoding the amino acids boxed in (A). Remnants of the 5′ and 3′ splice site (ss; dark gray) and branch site (BS) signals (underlined) were clearly visible in CkABC1 and CkABC11; PmfABC1 and PmfABC11 may have experienced further deletions (i.e., 45 and 18 nt, respectively) in this region to accommodate the possibly detrimental functional effects of the initial deintronization event. Changes to the conserved ss signals in CkABC1 and CkABC11 are highlighted with bold-type letters underneath the alignment. Deintronization of the “ancient” 92 bp intron may have caused the formation of a pseudogene, the function of which was possibly recovered by the 10 bp insertion (marked in gray) just 21 nt downstream. (C) Sequence alignment of C. krusei and P. membranifaciens ABC1 and ABC11 with C. krusei ABC12 surrounding the ABC1/ABC11 intron [the position in the protein where the intron is inserted is marked with an arrow in (A)] confirmed the loss of this “ancient” intron in ABC12. Conserved 5′ ss, 3′ ss and BS signals are marked in dark gray (5′ and 3′ ss) or are underlined (BS). Concerted evolution kept identical 88 bp introns in CkABC1 and CkABC11, but the 77 bp PmfABC1 and the 109 bp PmfABC11 introns escaped such concerted evolution. Not only are these two introns completely different in size and sequence, but each has also gained an additional 11 bp (PmfABC1) and 16 bp (PmfABC11) polypyrimidine tract (dark gray).

An alignment of C. krusei and P. membranifaciens ABC1 and ABC11 with C. krusei ABC12, inferred from protein alignments surrounding the confirmed 88 bp intron (Figure 7A), revealed conserved splice site signals and intron boundaries (Figure 7C), strongly supporting ABC12 intron loss by retrotransposition. Interestingly, PmfABC1 and PmfABC11 introns had conserved splice site signals, but, unlike the 88 bp CkABC1/11 intron, they were completely different from each other, both in sequence and length (78 bp vs. 109 bp; Figure 7C).

Functional characterization of Abc1p, Abc11p, and Abc12p

To investigate functional differences between Abc1p, Abc11p, and Abc12p, we overexpressed each in the host S. cerevisiae AD∆, a strain that is deleted in seven ABC pumps, and thus exquisitely sensitive to many xenobiotics, and which achieves exceptionally high expression of ABC proteins (Lamping et al. 2007). Both gDNA and cDNA ORFs encoding Abc1p and Abc11p of B2399 were cloned separately to see if the presence of introns affected the yeast phenotype. The presence of the intron did not alter drug resistance conferred by Abc1p or Abc11p significantly [see Figure S10 and Lamping et al. (2009)]. Strains overexpressing Abc1p or Abc11p were found to be resistant to a number of xenobiotics: i.e., resistant to azole antifungals (FLC, CLT, ITC, MCZ, and VRZ), fluorescent substrates (R6G and R123), large ionophores (NIG), translation inhibitors (CHX and ANI), and anticancer drugs (DOX and DAU; see Figure 8A and Figure S10). The antifungal AUR and the antipsychotic drug TFP, however, were not substrates of either pump (see Figure S10). Abc12p had the same substrate specificities as Abc1p and Abc11p (the seven Abc1/11p substrates tested in Figure 8A were also Abc12p substrates). Although all three multidrug efflux-pumps transported the same range of substrates, their transport activities for individual compounds differed significantly (Figure 8A). Abc1p appeared to pump the selected test compounds best overall. However, Abc12p was the best transporter of the smaller test compounds [i.e., low molecular weight (MW); ANI, CHX, FLC, and CLT], and Abc11p was the best transporter of the larger (i.e., high MW) test compounds (R6G and NIG). Perhaps surprisingly, Abc11p of 89102 (pump strain A in Figure 8A), a chimera with two Abc1p-specific amino acids in TMS6 (patch III) and one in TMS11 (patch V), was an even better multidrug efflux pump (i.e., it was more resistant to most test compounds) than Abc1p. Additional variations in substrate specificities were observed for the remaining ABC11-1 and ABC1-11 chimeras (Figure 8A, pump strains B–E and F–G, respectively). Although all chimeras were clearly functional, some effluxed certain compounds even more efficiently than wild-type Abc11p or Abc1p (e.g., FLC efflux by Abc1-11p chimeras; Figure 8A, pump strains F and G) whereas some Abc11-1p chimeras showed severely reduced drug efflux, especially of ANI, KTC, or NIG (Figure 8A, pump strains B, C, and E).

Figure 8.

Substrate transport and inhibitor sensitivity profiles conferred by Abc1p, Abc11p, and Abc12p. (A) The drug susceptibilities of the sensitive control strain AD∆ (0.25, 0.015, 1, 0.002, 0.5, 0.008, and 0.25 mg/l for ANI, CHX, FLC, CLT, R6G, KTC, and NIG, respectively) were set to 1. The MWs of the seven test drugs are listed under their names. The bar charts for the seven individual test drugs show the fold-increased drug susceptibilities (note that the y-axis is logarithmic) of AD∆ strains overexpressing C. krusei ABC11 (11; blue), ABC1 (1; green), and ABC12 (12; orange), and of AD∆ strains overexpressing all other chimera genes; i.e., ABC11 of strain 89102 (A; light blue); ABC11-1 chimeras of B2399 (B and E), IFO0011 (C), and 89102 (D); and ABC1-11 chimeras of 89102 (F and G). ABC11-1 chimeras A–E are in different shades of blue, and ABC1-11 chimeras F–G are in different shades of green. All ORFs contained the N-terminal intron and a C-terminal GFP tag. The individual ABC1, ABC11 and ABC1-ABC11 chimera ORFs are indicated below the bar charts. The values represent duplicate measurements of three independent transformants (there was no variation between individual MIC values of the same construct). (B) Inhibitor sensitivities of S. cerevisiae AD∆ overexpressing Abc1p (top right) or Abc11p (top left) from B2399 or Abc11p from 89102 (bottom left) were analyzed by agarose diffusion assays on CSM plates containing FLC at 0.13 × MIC_FLC. Paper disks contained: 1 µg each of milbemycins α11; α26; β9; and β11; 5 µg FK506 (FK); 0.5 µg ENI; 1 µg beauvericin (B), or 50 nmol OLI. Plates were incubated at 30° for 72 h or until growth inhibitory zones became clearly visible. The plates are representative of at least three independent experiments. Drugs on their own were not toxic to the cells (data not shown).

Differences were also observed in relative sensitivities conferred by Abc1p and Abc11p to known efflux pump inhibitors. Milbemycins (see Figure S11) are acaricides, insecticides, and anthelmintics widely used in agriculture and veterinary medicine. FK506 is a commonly used immunosuppressor, ENI and beauvericin are ionophoric depsipeptides with antibiotic and insecticidal activities, and OLI is an ATPase inhibitor. Abc1p was more sensitive than Abc11p to all eight efflux pump inhibitors tested, especially milbemycin β11 (Figure 8B), and, as expected, Abc11p of 89102 conferred inhibitor sensitivities that were between those for Abc1p and Abc11p of B2399 (Figure 8B) because it contains elements of Abc1p (pump strain A in Figure 8A).

SF demonstrated by Abc1p and Abc11p

The 18 core aa differences between Abc1p and Abc11p in TMS3, -4, -6, -7 and -11 (Figure 6C) clearly provided C. krusei with two distinct biological functions. V582I (IL1) and T795A (TMS6; Figure 6C) were allelic variations. Because ABC12 shared a common ancestor with ABC1 and ABC11 (Figure 6B), we compared the 18 core aa differences between C. krusei Abc1p and Abc11p with Abc12p to infer which had likely changed from their common ancestor. In addition, to ascertain which of these changes were possibly fate determining, we assessed the severity of changes by comparing them with 53 fungal homologs [see Figure S8 and Lamping et al. (2010)]. It appeared that Abc1p experienced only 5 aa (green and red circles on Figure 6C) and Abc11p 16 aa (blue and red circles on Figure 6C) changes compared to their common ancestor, three of which were changes to the same position in both transporters (red circles; Figure 6C). The most significant changes, not found in any of the other 53 fungal homologs, have thick circles on Figure 6C. Abc11p experienced the majority of changes, which is also reflected in the maximum likelihood tree for the concatenated sequences of patches II–V (see Figure S9B). Compared to 53 fungal homologs, Abc11p had a unique V616 at the bottom of TMS3, and a unique V1201 at the bottom of TMS7, and Abc1p had two unique amino acids, M628 and G1209, near the top of TMS3 and the center of TMS7 (thick circles; Figure 6C). Most changes were in, or near, the predicted lipid-facing surfaces (LIPS) of individual TMSs (see Figure S12). Four of the five changes predicted to face the substrate-binding pocket in Abc11p were either too high (TMS3, -4 and -6) or too low (TMS7) to contribute directly to the substrate-binding pocket, leaving S1355, at the center of TMS11, the only candidate likely to directly interact with substrates.

Sub-NF of the regulatory control of ABC1, ABC11, and ABC12

A genome-wide study of the evolution of cis-regulatory elements in duplicated genes in yeast found that the number of shared motifs rapidly decreased with time, whereas the total number of cis-regulatory motifs remained constant (Papp et al. 2003b). This suggested that NF played an important role in the evolution of duplicated genes in yeast (Papp et al. 2003b). Similar studies in yeast and humans later showed that most duplicated genes evolved by rapid SF followed by prolonged NF, a model termed sub-neo-functionalization (SNF) (He and Zhang 2005).

To test whether the regulatory control of ABC1 and ABC11 also evolved by SNF, and to confirm possible NF of ABC12, we searched their promoters with the YeTFaSCo database (de Boer and Hughes 2012; Hughes and de Boer 2013) for possible yeast transcription factor binding sites (TFBSs; Figure 9). As expected, we found similar numbers of regulatory motifs in each promoter (13 in ABC1, 14 in ABC11, 16 in ABC12; Table 3 and Figure 9). The ABC1 promoter shared only four TFBSs (HCM1, GTS1, and two GCR1) with the ABC11 promoter, and three TFBSs (PDR3 and two STB5) with the ABC12 promoters, and the ABC11 promoter shared five TFBSs (YAP3, AFT2, GAT1, and two MSN2/4) with the ABC12 promoter, but no TFBS were shared between all three promoters. The ABC1 promoter had five (PDR1, AZF1, UME6, and two SUM1), the ABC11 promoter two (HAA1 and ARO80), and the ABC12 promoter eight (PDR8, YRR1, CRZ1, RIM101, ARR1/YAP8, STB3, DAT1, and STE12) unique TFBSs (Table 3). Regulation of expression of PDR transporters involves a complex network of transcription factors (TFs) belonging to the PDR network (i.e., PDR1, PDR3, PDR8, YRR1, and STB5), and often involves TFs of the general stress response (i.e., MSN2, MSN4, YAP3, and CRZ1; (Sipos and Kuchler 2006)). Interestingly, ABC12 had evolved five TFBSs of the PDR network and four TFBSs of the general stress response network, whereas regulation by these two networks was clearly divided between ABC1 (five for PDR) and ABC11 (six for general stress response; Table 3 and Figure 9).

Figure 9.

Yeast transcription factor binding sites (TFBSs) identified in ABC1 (A), ABC11 (B), and ABC12 (C) promoters. Start and stop codons, predicted TATA-boxes and the poly-A binding site for the ABC11 terminator are in bold-type italics. The YeTFaSCo database was used to search for S. cerevisiae TFBSs in the upstream regions of ABC11 (1 kb), ABC1 (662 bp), and ABC12 (1 kb) (de Boer and Hughes 2012; Hughes and de Boer 2013). Sequences were searched for “Expert Curated no-dubious sorts” of TFBSs using a 90% similarity cut-off. Only “meaningful” TFBSs (i.e., TFBSs that are unlikely to occur randomly; binding sites >6 nt) were considered. TFBSs are highlighted in bold-type and marked, above, with colored bars with their orientation indicated. TFBSs involved in pleiotropic drug resistance are in different shades of red. TFBSs involved in the general stress response are in different shades of green. TFBSs involved in response to different environmental conditions [i.e., resistance to arsenic (ARR1), iron homeostasis and resistance to oxidative stress (AFT2), acetic acid stress (HAA1), and adaptation to alkaline stress (RIM101)] are in shades of blue. TFBSs involved in the diauxic shift and hypoxia are in shades of green-yellow. TFBSs involved in glycolysis and nitrogen metabolism are in yellow-orange-brown, and TFBSs involved in mitosis, meiosis, mating, and/or sporulation are in shades of gray or black.

Table 3. List of S. cerevisiae TFBSs identified in the promoters of C. krusei ABC1, ABC11, and ABC12.

	Groups of TFBSs
	PDR Network					General Stress Response			Weak Acid or Alkaline Stress, Arsenic Resistance, Iron Homeostasis and Quiescence					Diauxic Shift and Hypoxia			Glycolysis and Nitrogen Metabolism			Mitosis, Meiosis, Mating, Sporulation
TFBSsa	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23
ABC1b	1	2			2										1	1	2			1	2		1
ABC11						2		4	1			1				1	2	1	1				1
ABC12		1	1	1	2	2	1	1		1	1	1	1	1				1				1

^a1=PDR1, 2=PDR3, 3=PDR8, 4=YRR1, 5=STB5, 6=MSN2/4, 7=CRZ1, 8=YAP3, 9=HAA1, 10=RIM101, 11=ARR1/YAP8, 12=AFT2, 13=STB3, 14=DAT1, 15=AZF1, 16=HCM1, 17=GCR1, 18=GAT1, 19=ARO80, 20=UME6, 21=SUM1, 22=STE12, 23=GTS1.

^b662 bp ABC1 and 1000 bp ABC11 and ABC12 upstream sequences were searched for potentially meaningful TFBSs (numbers indicate how many times that site was detected; also see Figure 9).

Discussion

The C. krusei population is surprisingly homogenous

This study provides important insights into the population structure of C. krusei. There was surprisingly little sequence difference between isolates from vastly different locations, and, importantly, our data suggest that clinical (i.e., 89021, 89221, 89102, 90147, and V2b) C. krusei isolates may be of environmental origin. Although generally considered diploid, one C. krusei strain (89102) was aneuploid and B2399 was triploid, which is not uncommon in yeast and possibly a response to environmental stress. B2399 may have gained an additional set of chromosomes in the same way as some triploid D. bruxellensis hybrid strains (Borneman et al. 2014). D. bruxellensis is a close relative of C. krusei, and a major contaminant of industrial fermentations in biofuel production and the Australian wine industry (Borneman et al. 2014). It is also possible that B2399 is triploid as a result of a failure in cell division, but that would have had to happen in the distant past given its three distinct sets of chromosomes with typical SNP frequencies.

We have previously shown that the constitutive expression of Abc1p may tip the balance in favor of innate azole resistance for C. krusei (Lamping et al. 2009). In this report, we show that there may yet be other factors contributing to the innate azole resistance of C. krusei, namely the expression of additional multidrug efflux transporters Abc11p and Abc12p.

First direct estimates for the EGC rate and the MEPS and evidence for GC-bias of EGC in C. krusei

ABC1 and ABC11 are an exemplary pair of tandem-duplicated genes that enabled a detailed analysis of the major forces shaping their evolution. The striking pattern of identical sequences interspersed by four short (158, 19, 73, and 1 nt), highly divergent (22.2% dissimilarity), and strongly protected, sequences is a predicted outcome for the concerted evolution of tandem-duplicated paralogs of distinct function (Innan and Kondrashov 2010). Interestingly, ABC1 and ABC11 initially appeared to have experienced concerted evolution for >70 MY, which is significantly longer than the previous estimate of 25 MY for the average period of concerted evolution in yeast (Gao and Innan 2004). However, the discovery of the “same” pair of tandem-duplicated paralogs in P. membranifaciens extended that estimate to >134 MY.

Because the five DNA tracts that were homogenized by frequent EGCs varied from 0.3 to 1.7 kb in size (Table 2), and they were separated by an equal distance (5.3 kb; Figure 5A), and because we obtained polymorphism data for a set of C. krusei alleles representative of the entire C. krusei population [e.g., no additional ABC1- or ABC11-specific SNPs were detected in any of the three C. krusei whole genome shotgun (WGS) sequenced strains] it was possible to obtain, to our knowledge for the first time, a direct empirical estimate for the C. krusei gene conversion rate and the MEPS. The gene conversion rate increased proportionately with the tract length, and was ∼38–480× higher than the estimated silent-site nucleotide substitution rate for yeast. However, depending on the true nucleotide substitution rate of C. krusei, these estimates could be up to ∼8× lower (i.e., ∼5–60) or ∼2× higher (i.e., ∼76–960). The calculated MEPS for C. krusei was 106 bp. These results were in reasonable agreement with previous “indirect” estimates of 10–100× higher gene conversion than nucleotide substitution rates, and a MEPS of ∼200 bp for yeast (reviewed in Mansai et al. 2011), although our data suggest that previous estimates may have been slightly conservative. The GC-bias of gene conversion was also confirmed.

Gene deletions and CNVs combined with frequent EGCs between tandem-duplicated genes provide additional means for the evolution of novel gene function

Three out of seven strains possessed gene deletions and CNVs that, together with frequent EGCs between individual gene copies, created additional ABC11-1 and ABC1-11 chimeras with significantly altered functions. We previously characterized 28 different TMD chimeras between C. albicans Cdr1p and Cdr2p, two closely related paralogs (Figure 6B) (Holmes et al. 2006), and, in that study as well, many chimeras were severely altered in their substrate specificity and inhibitor sensitivities (Tanabe et al. 2011). It would appear that creating chimeras, and/or exchanging regulatory elements between duplicated paralogs, by intramolecular or intermolecular (non)homologous recombination and/or EGC are important evolutionary forces creating novel functionality.

Model for the evolution of C. krusei ABC11, ABC1, and ABC12

The discovery of ABC12 and P. membranifaciens ABC11-ABC1 was critical for unraveling the evolutionary history of C. krusei ABC1 and ABC11 (Figure 10). ABC12 appeared to be the product of gene-duplication by retrotransposition (step 1) of their common ancestor ∼149 MYA (Table 1), causing the loss of at least two inferred introns. The ancestor of ABC1 or ABC11 deintronized (step 4) one of these introns possibly soon after duplication (step 3), and before speciation of the C. krusei and P. membranifaciens lineages ∼134 MYA, which may have caused a temporary pseudogene. A functional ORF was recovered (step 5) not long after by a 10 bp insertion, and the deintronized version was then copied by EGC (step 6) onto its neighboring duplicate. The ABC1 and ABC11 ancestors optimized their efflux pump functions by SF of their common ancestor ABC1/11′. While the majority of their ORFs have evolved by concerted evolution for ∼134 MY, four small stretches of C. krusei ABC1 and ABC11 have been under very strong purifying selection (step 7), which led to an extremely clear pattern of sequence identity interspersed by very short peaks of sequence divergence. ABC12, on the other hand, is likely to have evolved by NF of both its regulatory control and pump function (step 2) since its retrotransposition into a completely different part of the genome. The experimental evidence indicates that, although all three genes are under very strong purifying selection, ABC1, ABC11, and ABC12 are not dosage sensitive genes, and one gene copy appears to be enough for the survival of C. krusei (triploid B2399 and diploid IFO0011 each had only one intact ABC1 and ABC11 gene copy, and diploid strains 89021, 89221, and 90147 each had only one intact ABC12 gene copy).

Figure 10.

Model for the evolution of C. krusei ABC1, ABC11, and ABC12. The common ancestor (ancestors are indicated with apostrophes after their names) ABC1/11/12′ contained at least two N-terminal introns (gray). ABC12′ duplicated ∼156 MYA by retrotransposition (1) into another part of the genome, and evolved its own unique function and expression pattern (2). Not long after, ∼134 MYA, the ABC1/11′ ancestor tandem-duplicated (3). Soon after, ABC11 deintronized the N-terminal intron (4), which may have caused a temporary pseudogene. Function was possibly recovered by a 10 bp insertion downstream (5). The ABC1′ ancestor then adopted the deintronized copy of ABC11′ by gene conversion (6) possibly not long after recovery of the ABC11′ ORF. The order of individual events may differ from the one outlined here. A critical proline change in TMS3 of ABC11′ (P616V) was possibly fate-determining, and caused SF (7) of Abc1p and Abc11p, but the majority of their ORFs evolved by continuing concerted evolution. The regulatory regions, on the other hand, evolved by SNF with no visible homologies remaining in the 5′ and 3′ proximal regions.

P. membranifaciens ABC1 and ABC11 followed a different evolutionary trajectory after speciation from C. krusei, maybe to compensate for the loss of the third multidrug efflux pump ABC12. Like C. krusei ABC1 and ABC11, however, they too experienced concerted evolution for ∼134 MY, and exhibit a unique pattern of sequence identities interspersed with stretches of sequence divergence. Interestingly, the nitrate assimilating Saccharomycotina yeast Candida utilis also has similar PDR multidrug efflux pump tandem-duplicates that were difficult to assemble because of the complex DNA patterns caused by concerted evolution (Watanasrisin et al. 2016). It would seem that prolonged concerted evolution of tandem-duplicates is perhaps more common than generally assumed, at least in the case of PDR transporters.

Different types of intron loss in C. krusei PDR transporters

Because all Saccharomycotina cluster A PDR transporters share a common ancestor with ABC1, ABC11, and ABC12 (Figure 6B), it is likely that these too contained at least two introns, which they have lost by a number of independent retrotransposition events. We also identified one N-terminal intron each in both C. krusei homologs of S. cerevisiae PDR12 encoding a weak organic acid transporter (Piper et al. 1998). Both predicted proteins were highly homologous to S. cerevisiae Pdr12p (75 and 77% identity, respectively), and 86% homologous to each other. We therefore named them PDR12 (SD108 scaffold 20: 24,998–29,583) and PDR121 (i.e., PDR12 one; SD108 scaffold 7: 101,770–106,331). Although the ORF sequences were almost identical, their introns varied in size from 103–104 (PDR12) to 66–80 (PDR121) nucleotides between the three WGS sequences. This was in contrast to the C. krusei ABC1 and ABC11 intron, which was identical for all A and B alleles, demonstrating the overpowering force of EGC between closely linked and highly similar stretches of DNA that are under no or little positive selection. Thus, similar to the ABC1/11/12′ ancestor, the common ancestor of the PDR12 homologs had at least two introns, because the PDR12 and PDR121 introns clearly had distinct origins (i.e., they were inserted 49 aa apart, just after K209 and K258, respectively—a highly conserved region of NBD1 between the Q-loop and the ABC signature motif; see Figure S8). Because PDR121 lacked the PDR12 intron just after K209, PDR121 likely lost this ancestral intron by homologous recombination between a cDNA copy of its mRNA and the region surrounding that intron. However, unlike PDR12 and PDR121, that each had two polypyrimidine tracts, one before and one after the BS signal, and P. membranifaciens ABC11 and ABC1, that each had one polypyrimidine tract after the BS signal (Figure 7C), the conserved C. krusei ABC1-ABC11 intron had no polypyrimidine tract. However, the 31 nt between its BS and 3′ ss signal was well within the maximum 45 nt required for efficient mRNA splicing in yeast (Meyer et al. 2011). Although yeast genes contain few introns [S. cerevisiae contains, on average, 0.05 per gene (Hirschman et al. 2006) compared to 8.1 introns per gene for H. sapiens (International Human Genome Sequencing Consortium 2004)], it is now commonly accepted that the fungus-animal ancestor had an intron density comparable to the highest known modern intron densities, underscoring the importance of intron loss during eukaryotic evolution (Stajich et al. 2007). It seems that ABC1 and ABC11, as well as PDR12 and PDR121, are “ancient” transporters that have not yet lost all their introns.

Gene duplication, a possibly fate determining proline change, and (S)NF of the regulatory control enabled the evolution of distinct biological functions

Since 2006, a number of structures for bacterial (Dawson and Locher 2006), lower eukaryotic (Jin et al. 2012; Kodan et al. 2014), and mouse (Aller et al. 2009) ABCB-type multidrug efflux transporters, functional homologs of PDR transporters, have been solved. The structures revealed a centrally located substrate-binding cavity (Dawson and Locher 2006; Aller et al. 2009; Jin et al. 2012) into which substrates enter from the inner leaflet of the bilayer via a gate formed by a “flexible” TMS4 caused by strategically positioned G and P residues (Kodan et al. 2014). A high frequency of helix kinks, often caused by P residues, is one of the hallmarks of membrane protein structure (Yohannan et al. 2004). Unique G and P residues were also involved in the evolution of Abc1p and Abc11p. It is possible that the unique P change at the bottom of TMS3 of Abc11p (P616V) may have been fate-determining, and induced the SF of Abc1p and Abc11p. Changing the conserved proline-kink at the bottom of TMS3 of Abc11p by an angle of ∼100° possibly caused significant TMS packing disorder, which was compensated for by changes to TMS3, TMS4, IL2, TMS6, TMS7, and TMS11 (Figure 6C), the majority of which were along one face of TMS3 (four changes) and TMS7 (four changes; see Figure S12). The involvement of the TMS7 surface, which includes G1209 as well as V1201, V1205, V1209, and I1213 of Abc11p, in critical helical rearrangements (i.e., helix–helix contacts) during the transport cycle of PDR transporters was previously noted by Kolaczkowski et al. (2013), who found that an A1208V mutation of C. albicans Cdr1p (equivalent to S1212 of Abc11p, which is also part of that surface; see Figure S12) dramatically altered its substrate transport activities (Kolaczkowski et al. 2013).

As a consequence of the changes to Abc11p, Abc1p function may have been optimized by modulation of amino acids in TMS3, TMS6, and TMS7, with M628 and G1209 possibly critically important for that process (Figure 6C). There is also strong evidence that S1355N, the patch V Abc11p/Abc1p amino acid at the center of TMS11 and possibly facing the substrate-binding pocket (see Figure S12), is important for substrate transport and inhibitor sensitivity. Mutation of the equivalent amino acid in S. cerevisiae Pdr5p (S1360A/F) and C. albicans Cdr1p (T1351A/F) severely affected substrate transport, and also eliminated the pumps’ sensitivities to FK506 (Egner et al. 2000; Saini et al. 2005).

In the absence of better alternatives, the YeTFaSCo database provides a valuable tool for the search of TFs that possibly regulate gene expression in members of the Saccharomycotina family. Our discovery of functionally related TFBSs was consistent with the SNF of the regulatory controls of ABC1 and ABC11, and the NF of the regulatory control of ABC12.

Summary

C. krusei is an innately azole resistant diploid fungal pathogen with a tendency to create aneuploid strains or triploid hybrids when exposed to environmental stress. This study confirms that it belongs to a third major Saccharomycotina lineage that has yet to be classified. A detailed study of the tandem-duplicated multidrug efflux pump genes ABC11 and ABC1 (a) provided the first empirical estimates for the gene conversion rate and the size of the MEPS for any eukaryotic organism; (b) confirmed GC-bias for EGC in C. krusei; (c) revealed concerted evolution for the majority of the two ORFs lasting for >134 MY; and (d) highlighted the importance of gene deletions, nonhomologous mitotic recombination, and CNVs combined with frequent EGCs between tandem-duplicated genes, for the evolution of novel functionality. We also provided first examples of intron loss by retrotransposition, deintronization, and, possibly, also by homologous recombination for any fungal PDR transporter. Discovery of a third C. krusei multidrug efflux transporter, ABC12, and the discovery of the tandem-duplicated orthologs P. membranifaciens ABC1 and ABC11, helped unravel the evolutionary history of an entire ABC transporter multi-gene family. It revealed a unique proline-kink change at the bottom of TMS3 of C. krusei ABC11 that possibly sealed the fate of Abc11p and induced the SF of ABC1 and ABC11. The regulatory control of ABC1 and ABC11, however, “escaped” concerted evolution and appeared to have evolved by SNF. We propose that, given the experimental difficulties in recognizing and correctly assembling highly similar, tandem-duplicated, gene sequences, their abundance, and, in turn, the impact of EGC on gene duplication and evolution, may be significantly underestimated by the scientific community.

Supplementary Material

Supplemental material is available online at www.genetics.org/lookup/suppl/doi:10.1534/genetics.116.194811/-/DC1.