Phased Diploid Genome Assemblies for Three Strains of Candida albicans from Oak Trees

Although normally a harmless commensal, Candida albicans, it is also one of the most common causes of bloodstream infections in the U.S. Candida albicans has long been considered an obligate commensal, however, recent studies suggest it can live outside animal hosts. Here, we have generated PacBio sequences and phased genome assemblies for three C. albicans strains from oak trees (NCYC 4144, NCYC 4145, and NCYC 4146). PacBio datasets are high depth (over 400 fold coverage) and more than half of the sequencing data are contained in reads longer than 15 kb. Primary assemblies showed high contiguity with several chromosomes for each strain recovered as single contigs, and greater than half of the alternative haplotype sequence was assembled in haplotigs at least 174 kb long. Using these assemblies we were able to identify structural polymorphisms, including a polymorphic inversion over 100 kb in length. These results show that phased de novo diploid assemblies for C. albicans can enable the study of genomic variation within and among strains of an important fungal pathogen.

Genetic and genomic studies have confirmed that the C. albicans genome typically exists in a diploid heterozygous state (Barnett 2008;Hickman et al. 2013;Hirakawa et al. 2015;Ropars et al. 2018). The heterozygous nature of the C. albicans genome makes it particularly hard to assemble with current methods that assume a haploid or highly inbred genome (Jones et al. 2004;Nantel 2006). The initial effort to assemble the genome of the diploid C. albicans reference strain SC5314 performed a haploid assembly of Sanger shotgun sequences followed by an effort to separate (or "phase") haplotypes in the subset of the genome where homologous chromosomes had sufficient divergence to be assembled as distinct contigs (Jones et al. 2004). Subsequently, van het Hoog et al. (2007) integrated contigs from Jones et al. (2004) with optical maps and other data to generate a haploid representation of all eight C. albicans chromosomes. Muzzey et al. (2013) then mapped short read sequences from SC5314 and a panel of diploid strains with loss of heterozygosity (LOH) for particular chromosomes (Legrand et al. 2008) to the haploid assembly from van het Hoog et al. (2007) to phase SNPs and short indel variants that distinguish the two haplotypes in SC5314. More recently, longread sequencing has been used to generate a de novo haploid assembly of a pathogenic strain of C. albicans (Panthee et al. 2018). However, no phased de novo diploid genome assembly has currently been reported for pathogenic or environmental strains of C. albicans.
Phasing haplotypes is important for understanding the relationship between DNA sequence and phenotype in diploid organisms (reviewed in Tewhey et al. 2011). Phased genomes can also expose misassemblies in reference genomes (Korlach et al. 2017), help identify point mutations and large structural variation between haplotypes (Zhou et al. 2019b), and allow detection of allele-specific expression and epigenetic modifications (Zhou et al. 2019a). In a predominantly asexual species such as C. albicans (Birky 1996;Barnett 2008;Bougnoux et al. 2008;Hirakawa et al. 2015), haplotype phasing of alleles is needed for correct inference of phylogenetic relationships (Birky 1996), and for detecting rare mating events that have potential implications for understanding adaptation to antifungal drugs (Berman and Hadany 2012) and disease emergence (Ropars et al. 2018).
Despite their utility, phased diploid representations of genomes are not commonly obtained during de novo genome assembly since phasing parental haplotypes is technically challenging. As of May 2019, there were only 120 assemblies with some level of haplotype resolution in the NCBI Assembly database, a small number compared to well over 300,000 haploid assemblies (Kitts et al. 2016). Since for many organisms it is often not feasible or practical to construct strains with single haplotypes to circumvent the challenges of diploid assembly, different computational strategies have been developed to directly assemble diploid genomes. One such method is called FALCON-Unzip, which involves using heterozygous variation captured by PacBio Single-Molecule Real-Time (SMRT) long-read sequencing to generate a de novo diploid assembly with haplotypes phased in heterozygous regions (Chin et al. 2016).
Here, we generated whole genome shotgun sequences using the PacBio SMRT technology (Eid et al. 2009) and applied the FALCON/FALCON-Unzip (Chin et al. 2016) assembly and phasing pipeline to three strains of C. albicans collected from oak trees in the UK (Bensasson et al. 2019). Our results demonstrate the feasibility of generating essentially-complete de novo diploid assemblies of C. albicans genomes using long-read sequencing and provide resources for future analysis of genome evolution and function in this important human fungal pathogen.

MATERIALS AND METHODS
Yeast strains DNA was extracted from three strains of C. albicans, NCYC 4144 (FRI10b.1), NCYC 4145 (FRI11a.1) and NCYC 4146 (FRI5d.SM), which were isolated from oak tree bark (Robinson et al. 2016). The strains analyzed here were previously characterized using Illumina short-read sequencing, are all phylogenetically distinct and show unusually high heterozygosity (Bensasson et al. 2019). More specifically, comparison between the genome-wide SNPs of these oak strains and representatives from the known C. albicans clades and singletons described in Ropars et al. (2018) showed that NCYC 4146 belongs to clade 4, NCYC 4144 to clade 18 and NCYC 4145 does not resemble any known genome (Bensasson et al. 2019).
Extraction of high molecular weight genomic DNA For extraction of high molecular weight DNA, we used the Promega Wizard Genomic DNA Purification Kit (A1125), and modified it to include an extra RNA digestion step, ethanol precipitation, longer centrifugation steps at higher centrifugal force and more reactions per sample (general recommendations are detailed in Bensasson 2018). For each strain, a single colony was used to inoculate 15 ml of YPD media and grown for 20-24 hr at 3 ∘ C. Cells were pelleted in two minutes at a relative centrifugal force of 16,162 g in 1.4 ml volumes in thirteen (NCYC 4145), fourteen (NCYC 4146) or eleven (NCYC 4144) 1.5 ml tubes. Cell walls were digested overnight at 37 ∘ C using 100 units lyticase (Sigma, L2524-50KU). Cells in each tube were pelleted (2 min at 16,128 g), resuspended in 300 ml Promega Nuclei Lysis solution and 100 ml Promega Protein Precipitation solution, cooled on ice for 5-30 min and cell debris was pelleted at 16,162 g for 10 min. The DNA in the supernatant was precipitated in 300 ml isopropanol and pelleted at 16,162 g for 10 min, washed with 300 ml of 70% ethanol at room temperature, pelleted at 16,162 g for 5 min, and air dried for 15 min. Pellets were resuspended in Promega DNA rehydration solution (1x TE buffer). RNA was digested using 1.5 ml (5.25 units) Promega RNase solution at 37 ∘ C for 1-2 hr, then at room temperature overnight. For each tube, 1 ml was visualized alongside a high molecular weight ladder on an agarose gel in 1x TAE buffer, and extracts showing clear bands were pooled into 4 tubes per strain. DNA extracts were cooled on ice, then digested for a second time with 21 units Promega RNase solution at 37 ∘ C for 1 hr then cooled on ice. DNA was precipitated using 0.5 volumes Promega Protein Precipitation Solution and 2 volumes of cold 96-100% ethanol on ice for 15 min, then pelleted at 16,162 g for 10 min. Pellets were washed with 3 volumes 70% ethanol, repelleted at 16,162 g for 3 min, and air dried for 15 min. DNA was resuspended in Promega DNA rehydration solution by incubation at 37 ∘ C for 1-1.5 hr. DNA extracts were spun at 16,128 g for 10 min, and supernatants from multiple tubes for the same strain were pooled to produce 15-30mg genomic DNA per strain. The quality and quantity of DNA was assessed by agarose gel electrophoresis (0.8% agarose in 1x TAE buffer) alongside a high molecular weight ladder (e.g., GeneRuler High Range DNA Ladder, SM1351), by NanoDrop, Qubit fluorometer and Fragment Analyzer Automated CE System, and showed average DNA fragment lengths over 43 kb for NCYC 4144, and over 34 kb for NCYC 4145 and NCYC 4146.

Long-read genome sequencing
Long-read genome sequencing was performed on the PacBio Sequel sequencing platform by the Georgia Genomics and Bioinformatics Core facility at the University of Georgia. A large insert library was constructed for each strain using the SMRTbell Template Prep Kit following the PacBio's instructions for .20 kb Template Preparation using BluePippin Size-Selection System for Sequel Systems. Fragment Analyzer analyses showed the final size of the insert libraries was 37 kb for NCYC 4144, and 22-24 kb for NCYC 4145 and NCYC 4146. Primers were annealed to the templates, then templates were bound to polymerase using the Sequel Binding Kit 2.0. The resultant polymerase bound complexes were purified using SMRTbell Clean Up Columns for Sequel, and were then bound to MagBeads for loading. Genomic DNA for each strain was loaded onto a single SMRTcell v1 and run on the PacBio Sequel System with a movie time of 10 hr.

De novo genome assembly and scaffolding
We performed de novo assembly with FALCON (v1.2.3) and phasing with FALCON-Unzip (v1.1.3; Chin et al. 2016). The assembly/ phasing/polishing software were downloaded as part of the pb-assembly metapackage (v0.0.1) from bioconda (https://github.com/ PacificBiosciences/pb-assembly). FALCON assembly was performed with a genome size parameter of 14.2 Mb, seed coverage of 30x, and a length cutoff for corrected reads of 1,000 bp. Consensus sequences were obtained after phasing using the Arrow polishing algorithm (v2.3.2) in FALCON-Unzip. Minimum coverage required for Arrow polishing was set to 5 and the positions of regions that could not be polished were summarized using a script (fastaLC2n.pl; https:// github.com/bensassonlab/scripts) and are provided in Files S1 to S6. Software versions and parameters used for assembly are provided in Files S7 to S9.
Reference-based scaffolding of primary contigs was performed using RaGOO (v1.01; Alonge et al. 2019) aligned to the SC5314 reference genome (Assembly 22, A haplotype, GCF_000182965.3). RaGOO scaffolding was only used to assign primary contigs to chromosomes, and subsequent QC analysis was performed in the raw, non-scaffolded assemblies. Haplotig placement relative to the primary assembly of each strain was done by aligning haplotigs to primary contigs with minimap2 (v2.16-r922; Li 2018) using the '-x asm59 alignment preset. The longest aligned segment for each haplotig was used to determine its placement in the primary assembly, allowing unaligned sequence from either end of the haplotig to be reported as alignment "tails".
To obtain a visual summary of long-range assembly contiguity and correctness we performed whole genome alignments between each diploid assembly and the reference genome using minimap2 (v2.16-r922; Li 2018) with the assembly alignment preset '-x asm10', and visualized them with dotPlotly (https://github.com/tpoorten/dotPlotly/). Only alignments at least 20 kb in length and with a mapping quality of 60 were visualized. For detailed analysis of structural variants in chromosome 3, we performed alignments of the diploid assemblies to the reference genome using nucmer from the MUMmer package (v4.0.0beta2, Marçais et al. 2018) with the settings: -maxmatch -l 80 -c 100, then visualized nucmer alignments of at least 1,000 bp using dotPlotly.
We assessed the phasing status across the genome by aligning contigs from diploid assemblies to the haploid reference and analyzing the depth of contig coverage. Alignments were obtained as described above, converted to BED format, and the depth of contig coverage along the genome was calculated using BEDTools genomecov (v2.28.0; Quinlan and Hall 2010). A visual summary of depth of contig coverage across the genome was obtained with the kpPlotCoverage function from the karyoploteR package (v1.10.2; Gel and Serra 2017). For regions of the genome where phasing was successful, we expect a depth of contig coverage of two, whereas for regions where the haplotypes were not phased, we expect a depth of coverage of one. Since phasing by FALCON-Unzip can only be done in regions were heterozygous variants are present, we also contrasted the extent of unphased regions in our assemblies (depth of contig coverage = 1) to the LOH regions detected by Bensasson et al. (2019) based on Illumina short-read sequencing.
For independent confirmation of phased base calls and LOH regions in each strain, we mapped Illumina HiSeq short-read data (PRJEB27862, Bensasson et al. 2019) to primary assemblies using bwa mem (v0.7.17; Li and Durbin 2009), produced a sorted bam file using SAMtools (v1.9; ), then called variants using bcftools mpileup (-d 10,000) and bcftools call (-c) (v1.9; Li 2011). Because of the very high depth of Illumina sequencing for NCYC 4146, we only used data from one run (ERR2708456) for variant calling in this strain. We used vcf2allelePlot.pl (Bensasson et al. 2019) running R (version 3.5.0) to generate allele ratio plots for high quality base calls (phred-scaled quality .40) in order to annotate LOH regions, identify regions within the primary assemblies that had unexpectedly low heterozygosity (below 0.1% using 100 kb non-overlapping sliding windows), and summarize levels of genomewide heterozygosity for each strain. Sites were counted as heterozygous if allele ratios were 0.5; more specifically between 0.2 and 0.8.

Annotation of telomeres and centromeres
In order to check whether the primary genome assemblies extend into telomeres, we extracted 5 kb from the termini of every chromosome in the primary assemblies and identified repeats using Tandem Repeats Finder (TRF, v.4.09, Benson 1999) with the following command: 'trf ,fasta. 2 5 7 80 10 50 500 -d -m -h'. We then searched the TRF results for the presence of terminal arrays of the 23 bp telomeric tandem repeats described for C. albicans (McEachern and Hicks 1993). We also obtained the centromeric sequences for each chromosome from the SC5314 reference genome (Assembly 22, haplotype A) and used them as queries against the primary contigs of each of our assemblies. Alignments were performed with minimap2 (v2.16-r922, Li 2018) on default settings.

Data availability
Raw PacBio reads for the three Candida strains are available at the NCBI short-read archive (SRA) under BioProject PRJNA533645. The phased diploid assemblies for the three oak strains are associated with the overall BioProject PRJNA543321. Individual GenBank accession numbers for primary contigs and haplotigs, respectively, for each strain are as follows: NCYC 4144: GCA_005890765.1 and GCA_005890695.1; NCYC 4145: GCA_005890775.1 and GCA_005890685.1; NCYC 4146: GCA_005890745.1 and GCA_005890705.1. Coordinates of haplotigs relative to their respective primary assembly are available in Files S10, S11, and S12. Annotations of positions of centromeres, telomeric repeats, confirmed LOH regions, assembly gaps, uncertain regions with unexpectedly low heterozygosity, and regions that were not polished by FALCON-Unzip are available for primary assemblies in Files S1, S2, and S3. Annotations of unpolished regions for alternative haplotig assemblies are provided in Files S4, S5 and S6. A full description of software version numbers for phased assembly is provided in File S7 and the configuration files used to run FALCON and FALCON-Unzip are provided in Files S8 and S9. Supplemental material available at FigShare: https://doi.org/ 10.25387/g3.9835208.

RESULTS AND DISCUSSION
High coverage long-read datasets for C. albicans oak strains We generated PacBio DNA sequencing data for three strains of C. albicans isolated from oak trees. Each sample was independently sequenced in a single SMRT cell on the PacBio Sequel instrument. A total of 703,252, 689,752, and 863,237 subreads with average lengths of 9,082, 8,899, and 9,576 bp were obtained for strains n■ Long reads allow Near-complete primary assemblies and extensive haplotype phasing in C. albicans We used the FALCON/FALCON-Unzip assembly and phasing pipeline (Chin et al. 2016) to obtain phased diploid assemblies for the three C. albicans oak strains. For each strain, the assembly is composed of two sets of contigs; a primary contigs set, and an alternate haplotigs set. The primary contigs provide a pseudohaploid representation of the genome, (i.e., they may include haplotype switches), and the haplotigs represent alternate haplotypes in the regions for which phasing was achieved. Phasing interruption and haplotype switches may occur when FALCON-Unzip is no longer able to establish linkage between variants (e.g., in LOH regions). We note that FALCON-Unzip removes non-linear contigs with high copy number, and as a consequence there is no mtDNA represented in the assemblies reported here. Primary assembly size varied from 14.7 Mb in NCYC 4144 to 15.5 Mb in NCYC 4145 and NCYC 4146 ( Table 2). The current reference assembly for strain SC5314 spans 14.2 Mb, whereas the predicted genome size for C. albicans obtained from physical and optical maps is 14.8-14.9 Mb (Jones et al. 2004;van het Hoog et al. 2007). Thus, the primary assembly for NCYC 4144 is very close to the predicted haploid reference genome size for C. albicans, while NCYC 4145 and NCYC 4146 have an additional 500 kb of sequence in their primary assemblies. Haplotig assembly size varied from 12.5-13.7 Mb (Table 2), indicating that the vast majority (84-92%) of the genome was phased in each assembly.
All primary assemblies achieved high contiguity, with primary contig N50 ranging from 1.2 Mb in NCYC 4146, to 1.6 Mb in NCYC 4145, and 1.9 Mb in NCYC 4144 (Table 2). Several chromosomes were assembled as contiguous gapless sequences spanning the entire chromosome, reflecting the low fragmentation of these assemblies. For strain NCYC 4144, chromosomes 1 through 7 were recovered as single contigs, while chromosome R was split in two contigs ( Figure 2A). In strain NCYC 4145, chromosomes 1, 3, 4, and 6 were recovered as single contigs whereas the remaining chromosomes were split into either 3, 5, or 9 contigs ( Figure 2B). In strain NCYC 4146, chromosomes 3, 4, and 5 were recovered as single contigs whereas the remaining chromosomes were split into either 2, 4, or 5 contigs ( Figure 2C). These highly contiguous and near complete primary assemblies are a consequence of the long length and high depth of the PacBio reads used for the assembly. Most read lengths (Figure 1) exceed the length of known repeats in C. albicans, such as Major Repeat Sequences, tandemly arrayed telomeric repeat units and retroelements (Todd et al. 2019).
Whole-genome alignments and dot plots indicate large-scale genome colinearity between the reference strain SC5314 and both primary contigs or haplotigs for the oak strains (Figure 2), suggesting no major missassemblies in our assemblies or the SC5314 reference genome. C. albicans centromeres are composed of discrete loci with unique sequences on each chromosome (Sanyal et al. 2004). Mapping the centromeric sequences annotated in the SC5314 reference genome (Assembly 22) to our primary assemblies revealed the presence of a unique match on each chromosome at the expected positions. The only exception was centromere 6 of strain NCYC 4146, which maps to two chromosome 6 locations on either side of a sequencing gap, indicating either a duplication or potential misassembly (Files S1-S3). Likewise, the majority of chromosome ends in primary assemblies (38 out of 48) contain an array of the 23 bp telomeric repeats described for C. albicans (McEachern and Hicks 1993). The presence Figure 1 Read length distribution for the PacBio data sets generated for three C. albicans oak strains. Vertical lines indicate read N50 length.
n■ of telomeric repeats at the ends of most chromosome sequences (Files S1-S3) indicate that the primary assemblies often reached the chromosome termini and are thus nearly complete.
Haplotig N50 (the N50 length of phased blocks) ranged from 174 kb in NCYC 4145 to 254 kb in NCYC 4144 (Table 2). Although FALCON-Unzip produces many haplotigs per chromosome instead of fully phased chromosomes, the phase block length achieved in these assemblies is more than two orders of magnitude larger than the average gene length in C. albicans (1,439 bp, Braun et al. 2005). BUSCO analysis of highly conserved single-copy genes in the primary contig sets revealed the presence of 97.1%, 96.2%, and 96.9% complete BUSCOs for NCYC 4144, NCYC 4145, and NCYC 4146, respectively (Table 3). We observed higher levels of completely duplicated BUSCOs in the primary assemblies for NCYC 4145 and NCYC 4146 relative to NCYC 4144, which suggests that the extra sequence in primary assemblies for these strains may be associated with duplications in the assembly. Analysis of the same single-copy genes in the haplotig sets revealed that 79.2%, 88.2%, and 84.3% of their alleles are complete in the alternative haplotype of NCYC 4144, NCYC 4145, and NCYC 4146, respectively. These results indicate that our primary assemblies are near complete in terms of gene content, and that both alleles for the majority of single copy genes are phased.

Assembled PacBio genome sequences have high quality base calls
We verified the quality of base calls in the primary assemblies by mapping independently-generated Illumina sequence (Bensasson et al. 2019) to the corresponding primary PacBio assembly for each strain. This resulted in high quality Illumina base calls for over 14 million invariant nucleotide sites in each strain where base calls from Illumina data matched the primary PacBio assembly. Furthermore, we observed over 70,000 sites with high quality SNP variants for each strain, which almost exclusively had the allele ratio of 0.5 that is expected when mapping a heterozygous diploid strain to one of its parental haplotypes. All strains had fewer than 270 high quality variants in Illumina data with allele ratios of .0.95; we expect such allele ratios of 1 if the base call in the primary PacBio assembly is incorrect. The rate of these probable errors is lower than one every 50 kb or one every 300 SNPs and confirms that the base quality of the majority of sites in the assemblies is high. For the strain with the most complete primary assembly (NCYC 4144), an unexpectedly large proportion of these errors (66 out of 169 errors; 39%) occur in the repetitive fraction of the primary assembly (21%) that NCBI present in lower case in the assemblies (Binomial exact test, p-value = 1 · 10 27 ). For the other strains, the proportion of errors in repetitive regions is close to expectations; 69 out of 269 (26%) for NCYC 4145 and 24 out of 143 for NCYC 4146 (17%). For these other two strains, a larger proportion (.29%) of errors (allele ratio 1) occur in the fraction of the primary assembly (1%) that could not be polished using FALCON-Unzip (Binomial exact test, p-value , 2 · 10 216 ). The positions of the unpolished regions in all primary and alternate assemblies are provided in Files S1 to S6.
Heterozygous SNPs identified by mapping Illumina reads to primary PacBio assemblies also allowed us to confirm previouslyidentified LOH regions and estimates of genome-wide heterozygosity for each strain (Bensasson et al. 2019). However, patterns of SNP variation also revealed that the primary assemblies for strains NCYC 4145 and NCYC 4146 contained six regions each that showed unexpectedly low heterozygosity at (or near) breaks in the assembly not previously identified as LOH regions in Bensasson et al. (2019). These "uncertain" regions could represent poorly resolved structural variants or assembly errors and may explain the excess sequence and increased levels of duplication observed in NCYC 4145 and NCYC 4146, including the extra copy of centromere 6 seen for NCYC 4146 (File S6). Figure 2 Diploid assemblies of C. albicans generated in this study (y-axis) aligned to the reference strain SC5314 (x-axis). For each chromosome that is labeled on the x-axis the primary contigs and haplotigs are stacked, with the primary contigs always placed first in the lower portion of each plot. Darker vertical grid lines demarcate chromosome limits, and horizontal lines show gaps between contigs. Note that the primary assemblies (lower diagonals for each chromosome) are nearly gapless. These dot plots display only alignments that are at least 20 kb in length and aligned segments are colored by percent identity with SC5314.
The DNA for these two strains was of lower molecular weight, yielded insert libraries with shorter inserts (22-24 kb) than for NCYC 4144 (37 kb) as well as shorter reads (Figure 1). Further long read data for these two strains would show whether the extra sequence in their assemblies are errors or genuine duplications. Uncertain regions span 1 Mb in the primary assemblies of NCYC 4145 and NCYC 4146 and are described more fully in the annotations for their primary assemblies (Files S2 and S3) along with the positions of centromeres, telomeres, confirmed LOH regions, gaps and unpolished regions for all three strains (Files S1-S3).
Phasing efficiency and loss of heterozygosity Whole genome alignments indicate that some regions of the genome for each strain are not phased, such as chromosome 5 in strain NCYC 4144 (Figure 2A). The phasing efficiency of FALCON-Unzip depends on the presence of heterozygous variants captured by long reads. Accordingly, variants separated by long tracts of homozygosity cannot be phased by this approach (Chin et al. 2016). Since the strains analyzed here were previously shown to contain segmental and whole-chromosome loss of heterozygosity (Bensasson et al. 2019), FALCON-Unzip is not expected to phase the entirety of these genomes.
Approximately 16.1%, 4.9% and 7.7% of the genomes of NCYC 4144, NCYC 4145, and NCYC 4146, respectively, were defined as having LOH based on variants detected in short read data relative to the SC5314 reference genome (Bensasson et al. 2019). As expected if LOH regions impact FALCON-Unzip phasing, the strain with the highest amount of LOH would have the lowest number of genes phased in haplotigs, and vice versa, which is demonstrated by the complete BUSCO score for haplotigs in NCYC 4144 compared to NCYC 4145 and NCYC 4146 (Table 3). To more directly analyze the correspondence between the LOH regions defined previously and the phasing efficiency of FALCON-Unzip, we mapped all contigs (primary and haplotigs) from diploid assemblies to the haploid SC5314 reference genome and analyzed the resulting depth of contig coverage profiles. While the average depth of contig coverage is approximately 2 in the non-LOH regions for the three strains, this value drops to an average of 1. 31,1.58,and 1.39 in LOH regions of NCYC 4144,NCYC 4145,and NCYC 4146,respectively (Figure 3). The difference in average depth of coverage between LOH and non-LOH regions is significant at a 99% confidence level (Welch two sample t-test, p-value , 0.001). These results confirm the expectation that homozygous regions disrupt phasing by FALCON-Unzip.
Although largely matching the previously known LOH regions, contig coverage profiles sometimes reveal windows for which the observed depth is .2, or where the depth is not as expected given the presence or absence of LOH (Figure 3). Since contig coverage is calculated after aligning the de novo assemblies to the haploid reference, duplications in either haplotype that align to the same location in the reference genome could result in contig depth .2 (see subsection below). Additionally, regions defined as LOH by short-read variants that have the expected diploid contig coverage could represent regions where haplotype divergence at the nucleotide level is low, but more substantial at the structural level. Indeed, FALCON-Unzip also uses heterozygous structural variants to phase assemblies, which could potentially be used even in the absence of significant nucleotide divergence (Chin et al. 2016).

Phased diploid assemblies reveal structural variants
The evolution of C. albicans is characterized by frequent genomic rearrangements often involving open reading frames (ORFs) that are flanked by inverted repeats (Todd et al. 2019). For example, there are two groups of ORFs in neighboring regions of chromosome 3 which appear to have undergone duplication followed by inversion, generating inverted repeats (van het Hoog et al. 2007;Todd et al. 2019), and these can lead to further large-scale rearrangements (Todd et al. 2019). Close inspection of whole-genome alignment dot plots between the oak strains' diploid assemblies relative to the SC5314 reference genome in this region of chromosome 3 reveals that oak strains have experienced inversions of the sequence flanked by these inverted repeats ( Figure 4A). The inversions are polymorphic among the three strains, with each strain displaying a unique genotype. Strains NCYC 4146 and NCYC 4144 are heterozygous for the first and second inversions, respectively, whereas strain NCYC 4145 is homozygous for both inversions. These results highlight the utility of long-read sequencing in detecting structural variants, and indicate n■ a The Saccharomycetales ortholog gene set v9 was used in this analysis and comprises a total of 1,711 genes.

Figure 3
Depth of contig coverage for three diploid C. albicans assemblies mapped to the haploid reference genome (strain SC5314). Black boxes below each coverage plot highlight the regions defined as loss of heterozygosity in Bensasson et al. (2019) and generally correspond to regions of contig depth of coverage = 1.
that this approach should prove useful in gauging how much of C. albicans heterozygosity comes from structural variants as opposed to single nucleotide polymorphisms (Sedlazeck et al. 2018).

CONCLUSIONS
Here we show that PacBio SMRT sequencing is an affordable and efficient approach to obtain phased diploid assemblies for the model yeast species C. albicans. Our results suggest that SMRT sequencing and FALCON-Unzip assembly of the SC5314 reference strain is now merited and could provide an end to the "long hard road" to generate a complete diploid C. albicans reference genome (Nantel 2006). Our work also demonstrates that phased diploid assemblies generated using PacBio long-read data can provide detailed insights into genome structure and evolution in C. albicans that are not possible to obtain using short read sequencing. This advance is especially important for an asexual diploid species such as C. albicans with unusually high rates of structural rearrangements (Birky 1996;van het Hoog et al. 2007;Todd et al. 2019) that are associated with gene family evolution and pathogenicity (Butler et al. 2009;Todd et al. 2019).

ACKNOWLEDGMENTS
This study was supported in part by resources and technical expertise from the Georgia Advanced Computing Resource Center. We thank Katherine Sandlin, Roger Nilsen and Magdy Alabady at the Georgia Genomics and Bioinformatics Core for performing some of the DNA extractions and for providing the PacBio sequencing service. This work was funded by the University of Georgia.