Systematic Complex Haploinsufficiency-Based Genetic Analysis of Candida albicans Transcription Factors: Tools and Applications to Virulence-Associated Phenotypes

Genetic interaction analysis is a powerful approach to the study of complex biological processes that are dependent on multiple genes. Because of the largely diploid nature of the human fungal pathogen Candida albicans, genetic interaction analysis has been limited to a small number of large-scale screens and a handful for gene-by-gene studies. Complex haploinsufficiency, which occurs when a strain containing two heterozygous mutations at distinct loci shows a phenotype that is distinct from either of the corresponding single heterozygous mutants, is an expedient approach to genetic interactions analysis in diploid organisms. Here, we describe the construction of a barcoded-library of 133 heterozygous TF deletion mutants and deletion cassettes for designed to facilitate complex haploinsufficiency-based genetic interaction studies of the TF networks in C. albicans. We have characterized the phenotypes of these heterozygous mutants under a broad range of in vitro conditions using both agar-plate and pooled signature tag-based assays. Consistent with previous studies, haploinsufficiency is relative uncommon. In contrast, a set of 12 TFs enriched in mutants with a role in adhesion were found to have altered competitive fitness at early time points in a murine model of disseminated candidiasis. Finally, we characterized the genetic interactions of a set of biofilm related TFs in the first two steps of biofilm formation, adherence and filamentation of adherent cells. The genetic interaction networks at each stage of biofilm formation are significantly different indicating that the network is not static but dynamic.

cerevisiae (Tong et al. 2004). Through clever application of genetics and high throughput biology, the facile mating cycle of S. cerevisiae has been harnessed to generate genetic interaction networks encompassing 90% of its genes and include 23 million double mutants (Costanzo et al. 2016). Although C. albicans is arguably the most mature genetic system for pathogenic fungi, genetic interaction analysis has remained significantly under-developed.
One of the barriers to genetic interaction analysis in C. albicans is that it is primarily a diploid organism without a traditional sexual cycle (Hickman et al. 2013). Typically, genetic interactions are studied by generating strains that contain null mutants of the two genes of interest. The phenotypes of the resulting double mutants are then compared with the single mutants. For S. cerevisiae and other haploid organisms, the construction of the requisite double null mutant involves two gene deletion events. In the diploid C. albicans, however, four discrete gene deletions are required to generate the mutants at two loci. Not surprisingly, multiple null mutants have been constructed infrequently in C. albicans and, until recently, no systematic approaches to studying large numbers of double mutants have been reported. Of course, the recent applications of CRISPR/Cas9 to C. albicans are likely to increase the number of multiple homozygous gene deletions that are studied (Nguyen et al. 2017).
Double homozygous mutants, however, will not be applicable to, or appropriate for, all systems. An alternative strategy for genetic interaction analysis in diploids is based on the phenomenon of complex haploinsufficiency (CHI). Simple haploinsufficiency occurs when a heterozygous mutation in a diploid organism leads to an observable phenotype (Uhl et al. 2003). CHI, correspondingly, refers to the case when a strain with heterozygous mutations at two different loci has a phenotype that differs from both of the corresponding single heterozygotes (Haarer et al. 2007). CHI was initially referred to as unlinked non-complementation in the context of isolating S. cerevisiae mutants that interacted with essential genes such as tubulin (Stearns and Botstein 1988). With the advent of large-scale genetic interaction analysis methods in S. cerevisiae, the concept of CHI was applied to the identification of genes that interact with actin using a high throughput approach (Haarer et al. 2007).
Our group has previously applied CHI to C. albicans using transposon-based methods to generate double heterozygous mutants from a common parental or "query" strain (Bharucha et al. 2011). Specifically, we used transposon-mediated CHI to identify genes and pathways that interact with Cbk1 (Bharucha et al. 2011;Saputo et al. 2016), the key kinase of the Regulation of Ace2 and Morphogenesis pathway (Saputo et al. 2012). Recently, we reported a CHI-based genetic interaction analysis of a set of biofilm-related transcription factors (Glazier et al. 2017). Here, we describe the construction of a library of strains and plasmids for the systematic analysis of haploinsufficiency and CHI associated with C. albicans transcription factors (TF). Although CRISPR/Cas9-based approaches are likely to be very useful for genetic interaction studies (Vyas et al. 2015), CHI will offer advantages in a number of settings and, thus, should represent be a complementary approach to the analysis of multi-loci homozygous deletion mutants.
For example, there are at least two settings in which CHI will likely be the preferred approach. First, essential genes are not amenable to homozygous deletion analysis. Second, and more generally, genetic interactions can only be observed if the phenotypes of the single mutants are sufficiently mild to allow the detection of additional changes from that baseline. If a homozygous deletion mutant of interest completely abolishes a given cellular function, then creating a double homozygous deletion mutant with a candidate interacting gene will be completely uninformative because no additional changes in phenotype is possible. This is a reasonably frequent situation for some of the most intensely studied processes in C. albicans pathobiology. Homozygous TF deletion mutants that affect biofilm formation (Nobile et al. 2012) and hyphal morphogenesis (Homann et al. 2009) frequently show almost complete loss of function. For example, strains lacking Brg1 (Cleary et al. 2012) or Tec1 (Schweizer et al. 2000) are almost exclusively yeast-form under most filament inducing conditions. As such, it would be difficult to determine whether these two genes interact functionally by studying the double homozygous deletion mutants. As described herein, we have used CHI to show that these two genes, in fact, do genetically interact. A similar situation is encountered in the initial adhesion step of biofilm formation where strains lacking BCR1 adhere at levels barely above background (Fox et al. 2015). Finally, in vivo studies of C. albicans homozygous deletion mutants have identified numerous "avirulent" strains; double homozygous mutants, again, would not be informative in this setting because there is little dynamic range available for measuring additional virulence variation in the double mutants.
We have constructed and characterized the phenotypes of a set of heterozygous TF deletion mutants using in vitro assays and an in vivo model of systemic candidiasis in mice. We also used a set of double heterozygous biofilm-related TF deletion strains (Rob1, Efg1, Ndt80, Bcr1, Brg1, and Tec1) derived from this library to characterize the initial adhesion and filamentation steps in biofilm formation. These genetic interaction studies indicate that Tec1 plays a previously unrecognized role in adhesion to abiotic surfaces. In addition, we have found that the set of TF genetic interactions vary significantly depending on whether the cells are adhered to abiotic surfaces or are in planktonic conditions. Under the latter conditions, we provide evidence that Rob1 plays a role in the initiation of hyphae formation based on its genetic interaction with other TFs known to function during this stage.

Strains and plasmids
All heterozygous and double heterozygous deletion strains of Candida albicans were constructed from SN152 using auxotrophic markers of LEU2 or HIS1 which were flanked with homologous regions of 100-500bp for targeted integration (Noble and Johnson 2005). Deletion cassettes were amplified by PCR using template DNA isolated from C. albicans strains in Homann's TF deletion collection (Homann et al. 2009). The primers contain SbfI restriction sites, to allow release of the cassette from the plasmids. Sequences of primers used to generate deletion cassettes and to check integration are provided in Table S1. The primers amplified both the LEU2 and HIS1 deletion cassettes. Amplicons were cloned into pCR4TOPO (Invitrogen). To differentiate between the two auxotrophic genes, DNA was isolated from E. colitransformants and digested with BglII, a restriction enzyme that cuts in the HIS1 gene but not the LEU2 gene. LEU2-containing plasmids were then sequenced to confirm the identity of the inserts.
To generate heterozygous LEU2-marked deletion mutants, LEU2marked cassettes were excised from the plasmid by digestion with SbfI and transformed into C. albicans SN152 using standard lithium-acetate protocols (Hernday et al. 2010). Leu2+ transformants were screened for integration at the correct genomic locus by PCR using primers internal to the deletion cassette and upstream of the integration site (See Table  S1 for check primers). The heterozygous nature of six strains was confirmed by quantitative PCR comparing the deletion mutant to the parental strain and the expected reduction in gene copy number was observed for each strain (Glazier et al. 2017); we have not performed this for each strain. Two independent isolates of each heterozygous deletion stain were archived in the collection.
The deletion cassettes contain the barcodes originally created by Homann et al. (Homann et al. 2009) and the barcode for each strain is provided in Table S1. The amplification efficiency of each barcode was verified and, in some instances, altered for poor performing barcodes. A reference strain containing a NAT1-marked cassette with a unique barcode was constructed by cloning the barcode sequence into pDUP3 by gap-repair recombinational cloning in S. cerevisiae. The resulting plasmid was integrated to an empty region of the genome as described (Gerami-Nejad et al. 2013).
The strains containing the pTDH1-TEC1 allele have been described previously (Glazier et al. 2017). The libraries will be deposited at the Fungal Genetics Stock Center. Prior to deposit, all strains and plasmids are available upon request from the corresponding author.
Phenotypic analysis of heterozygous deletion strains Candida albicans strains were grown overnight in liquid YPD at 30°with shaking, adjusted to OD 600 of 1.0 in PBS, and five 10-fold serial dilutions generated. The strains were then spotted onto agar plates with the appropriate media. Under a given condition, the phenotype of each heterozygous deletion strain was compared to the other heterozygous deletion strains in the pool. Strains with altered phenotypes on the initial screening plates were confirmed by repeat testing with a second independent isolate and compared to both wild type and the homozygous deletion mutant. Conditions and media for phenotyping were based on those described (Homann et al. 2009): YPD 30°; Spider medium; YPD hypoxia (BD GasPak EZ) 30°; 14mM CuSO 4 in YPD; 100mM fluconazole in YPD; 15mM caffeine in YPD; YP-glycerol (30%); YNB-glycerol (30%); 1.5M NaCl in YPD and 1.5M sorbitol in YPD.
Pooled competitive in vitro and in vivo competitive fitness assays The Insitutional Animal Care and Use Committee at the University of Rochester approved all protocols used in this study. Mice were obtained from Charles River Laboratories. Three pools of 42-44 strains were prepared containing stains with unique barcodes by combining individual strain cultures at equal cell density. The initial inoculum pool was immediately flash frozen and DNA isolated. The pooled cultures were used to inoculate YPD cultures which were grown at 37°to mid-log phase (OD 600 1) over a time period that corresponded to 20 generation. The cultures were harvested and DNA was isolated. For in vivo experiments, the pooled cultures were used to infect 5 CD1 (ICR) mice for each pool at 5 · 10 5 CFU/mL by tail vein injection. The mice were sacraficed 24 hr after inoculation and the kidneys were harvested. Homogenates of the kidneys were plated on YPD+gentamicin+vancomycin plates and incubated for 24 hr at 30°. Plates with 500 colonies were scraped of cells and DNA was isolateed. The DNA from the inoculum and from samples recovered after in vitro or in vivo growth was analyzed by quantitative PCR using primers specific to each barcode. The abundance of each strain in the recovered (R) samples was normalized to the inoculum (I) using the procedure described . The experiments were performed twice and the mean abundances for each strain were calculated. The statistical significance of deviations in R/I from the pool mean for each strain were evaluated by Student's t-test using the Benjimani-Hochberg correction for multiple comparisons; statistical significance defined as P , 0.05.
Two-strain competitive fitness assays The relative, competive fitness between the reference strain and single or double heterozygotes was determined by preparing an inoculum with equal amounts of the reference strain SN250 containing a barcode integrated in a region of the genome without any ORFs and the mutant strain as determined by hemacytometry. A portion of the inoculum was flash frozen and DNA was isolated from the cells. The inoculum was used to infect mice (N= 5 for each competition) as described above for the pooled assays. The ratio of mutant/reference was determined by quantiative PCR with primers to the barcodes of the deletion mutant and reference strain, respectively.
Adherence Under biofilm forming conditions C. albicans strains were grown overnight at 30°with shaking in YPD liquid cultures. Cells were washed three times in PBS and re-suspended to an optical density (OD 600 ) of 0.5 in biofilm-inducing media YET-1% sucrose (Glazier et al. 2017). The suspension of cells (200 mL) of cells was incubated for 90 min at 37°in 96-well, tissue culture treated, polystyrene plates. After 90 min, the media was removed and non-adherent cells were removed by careful washing with PBS. The density of adhered cells was measured by reading the optical density at 600 nM (Fox et al. 2015;Lohse et al. 2017) using a SpectraMax plate reader. At least 3 replicate wells were analyzed for each strain. The density for each mutant strain was normalized to the reference strain and referred to as normalized adherence fitness (AF). Epistasis between two mutants (X and Y) was defined as Ɛ = AF XY -AF X AF Y where AF XY is the AF of the double mutant and AF X and AF Y are the normalized biofilm densities of the two heterozygotes. AF X AF Y represents the expected AF for the double mutant if there was no genetic interaction. Therefore, Ɛ = 0 indicate that the two mutations function independently. Double mutants whose SEM range was completely outside of the standard relative error of the expected values were defined as having an epistatic interaction (Glazier et al. 2017). Ɛ . 1 indicates a buffering interaction and Ɛ , 1 indicates a cooperative interaction.
Measurement of adherent cell filament length C. albicans strains were grown overnight at 30°with shaking in YPD liquid cultures. Cells were washed three times in PBS then re-suspended to an optical density (OD 600 ) of 0.5 in biofilm-inducing media YET-1% sucrose (Glazier et al. 2017). The cell suspensions were used to inoculate chamber slides and placed at 37°for 145 min. The medium and suspended cells were removed and the slides washed with PBS to remove non-adhered cells. The slides were heated at 100°to fix cells. The adhered cells were imaged using bright-field settings on a Nikon i80 microscope with a Photometrics CoolSnap HQ 2 camera. The length of filamented cells was determined by measuring the path-length of a line drawn along the long axis of the cell beginning at the bud neck using the NIS-Elements software package (Nikon). At least 100 cells were measure for each strain and two independent experiments were performed. The mean length was calculated and the statistical significance of the differences between strains were evaluated using the Mann-Whitney U-test (P , 0.05 was defined as statistical significance). Genetic interactions were calculated as described above after the length of each mutant was normalized to the reference strain length.
Quantitative assays of fungal yeast/hyphae morphology using ImageStream X MKII Frozen fungal stocks were steaked onto yeast extract-peptone-dextrose (YPD) agar plates and cells were allowed to grow overnight at 30°. Yeasts from the plates were then inoculated into flasks containing 25 ml of YPD broth and incubated for 16 h at 30°with shaking at 180 rpm. Under these conditions, all C. albicans strains grew as budding yeasts. Cultured yeasts were washed twice with phosphate-buffered saline (PBS) and counted using a hemocytometer. Cell density was adjusted to a final concentration of 2 · 10 7 /ml with RPMI 1640 medium containing L-glutamine, but without phenol red (Gibco). Cells were seeded at 2 · 10 6 (100 ml) per well in a 96-well round bottom plate and incubated for 6 hr at 37°. Cells were washed twice with PBS followed by incubation with a solution containing 2.5 mg/ml FM 4-64FX membrane staining dye (Thermo-Fisher Scientific Cat.# F34653) and 0.4 mg/ml calcofluor white M2R (CFW; Sigma Cat# 6258) in a modified phosphate buffer saline, pH 8.0 for 30 min at 37°. Following staining, wells were washed and resuspended in 30 ml 2% paraformaldehyde solution. Samples were analyzed with an ImageStream X MKII (Millipore) image flow cytometer with a 7-mm core at low flow rate and high sensitivity using INSPIRE software. Cells were analyzed along with 7% Speed Beads (1 mm polystyrene beads) for calibration of the flow and focus of the flow cytometer. Fluorochromes were excited with 405, 561 and 785 nm lasers and light emitted by the fluorescently-labeled cells was collected through a ·60 objective, providing a pixel size of 0.3 m m 2 . Of the twelve channels available, channels 1 and 9 were used for bright-field images, channel 6 for side scatter light intensity and channels 5 and 7 for fluorochrome intensities of FM 4-64FX and CFW, respectively. Single-color reference samples for each fluorochrome were generated by inclusion of cells that had been incubated with each fluorochromes separately. Images were analyzed using IDEAS software version 6.2 (Millipore). A compensation matrix was built with the data from single-color reference samples to allow removal of spectral overlap to adjacent channels from each detection channel. Single-cell images were first gated on the focused cells with a size (area) range between 50-200 pixels (15-60 mm 2 ). A dot plot of bright field area vs. aspect ratio (minor axes/major axes) was used to distinguish yeast from elongated hyphae population that had an aspect ratio less than 0.7 and area greater than 21 mm 2 . The percentages of yeast and hyphae populations were measured and presented as means 6 standard errors.

Data availability
The authors state that all data necessary for confirming the conclusions presented in the article are represented fully within the article.

RESULTS
Construction of a library of C. albicans heterozygous transcription factor deletion strains and a collection of plasmid-borne transcription factor deletion cassettes To characterize the function of TFs regulating C. albicans gene expression, Homann et al. created a deletion library of transcription factor (TF) null mutants (Homann et al. 2009). In order to facilitate the application of CHI to the analysis of C. albicans TF networks, we generated an analogous library of 133 heterozygous deletion mutants as well as a collection of the respective deletion cassettes cloned into E. coli plasmids. To do so, DNA was isolated from the strains in the Homann collection (Homann et al. 2009) and used as templates for the PCR amplification of each TF deletion cassette. The cassettes containing the LEU2 auxotrophic marker ( Figure 1) were cloned into a TOPObased E. coli vector by TA cloning techniques. The primers (Table S1) contain SbfI restriction sites to facilitate release of the cassettes. The deletion cassettes have 100-500 bp sequences homologous to the 59 and 39 UTR regions flanking the TF loci. The cassettes also contain barcoded signature tags adjacent to the auxotrophic markers. Each plasmid was sequenced to verify the barcodes and regions of TF-specific homology.
To construct heterozygous deletion strains (Figure 1), plasmids were digested with SbfI to release the cassettes which, in turn, were used to transform C. albicans strain SN152 (Leu -/His -/Arg -). Two independently derived strains were isolated for each TF heterozygote and the integration sites were confirmed by standard PCR analysis. Thus, we have constructed two libraries: one composed of TF LEU2-marked deletion cassettes in E. coli, and one composed of single heterozygous TF deletion mutants in C. albicans. The deletion collection can serve as a resource from which one can quickly generate systematic sets of double heterozygotes in a convenient C. albicans strain background that has been validated for in vivo studies (Noble and Johnson 2005). Alternatively, one could start from a query heterozygous deletion strain and generate double mutants using the set of deletion cassettes. Both of these resources will be deposited at the Fungal Genetics Stock Center.

Phenotypic characterization of the heterozygous deletion collection
Four large-scale haploinsufficiency-based screens have been performed in C. albicans, previously (Uhl et al. 2003;Xu et al. 2007;Oh et al. 2010;Chaillot et al. 2017). Two were based on transposon-generated collections (Uhl et al. 2003;Oh et al. 2010) while the other two were based on a collection of heterozygous deletion mutants. Our heterozygous transcription factor deletion library corresponds to the previously characterized homozygous transcription factor deletion library and the strains have two remaining auxotrophic markers to facilitate generation of double heterozygotes as well as complementation (Homann et al. 2009). To characterize the collection of TF heterozygotes, we first performed a set of phenotyping experiments similar to those previously reported by Homann et al. for the homozygous TF deletion collection (Homann et al. 2009). This data provides a baseline for future CHI analyses of similar phenotypes and represents a systematic examination of the frequency of haploinsufficiency within C. albicans transcription factors.
We examined the growth of 133 of the 137 strains under the following conditions: YPD 30°, Spider agar, YPD hypoxia, YPD 14mM copper, YPD 100mM fluconazole, YPD 15mM caffeine, YP-Glycerol, YNB-Glycerol, YPD 1.5M NaCl and YPD 1.5M sorbitol. These conditions were not as extensive as that performed for the homozygous deletion set but covered a range of biologically distinct physiological states and stresses (Homann et al. 2009). The results of this phenotyping are summarized in Table 1. For the majority of conditions, we observed no evidence of haploinsufficiency or haploproficiency. Filamentation was the phenotype for which we observed the highest rate of haploinsufficiency/haploproficiency: 13/133 (10% of mutants). Representative examples of haploproficient (TUP1, orf19.6109) and haploinsufficient (ROB1, orf19.4998; ZFU3, orf19.6888) strains are shown in Figure 2A. For all but two of the mutants (ZFU3, orf19.6888) and NDT80, orf19.2119), the filamentation phenotypes were concordant with the corresponding homozygous deletion mutant. Vandeputte et al. have reported that the homozygous deletion of ZFU3 (orf19.6888) is hyperfilamentous (Vandeputte et al. 2011). The homozygous deletion of NDT80 (orf19.2119), on the other hand, is deficient in filamentation. As we have recently reported, the increased filamentation shown by ndt80D (orf19.2119) is dependent on Rob1 (orf19.4998) and Tec1 (orf19.5908) (Glazier et al. 2017). Although we have not investigated the mechanism of this observation, it seems that loss of one allele may trigger a dysregulated compensatory process that is dependent on the remaining allele. When both alleles are missing, the compensatory process is not able to support filamentation.
Two heterozygotes, upc2D and crz1D, were hypersensitive to fluconazole, which is consistent with their well-characterized roles in the cellular response to this drug (Vasicek et al. 2014;Onyewu et al. 2004). Upc2 (orf19.391) is a key regulator of ergosterol biosynthesis genes while Crz1 (orf19.7359) is an important transcriptional regulator under control of the calmodulin-calcineurin axis. Two heterozygotes (GLN3, orf19.3912 and STP3, orf19.5915) were resistant to caffeine ( Figure 2B), a drug that interacts with the TOR signaling pathway (Reinke et al. 2006), while a third (ZCF29, orf19.5133) was more sensitive. The caffeine phenotypes for all three heterozygotes are concordant with their corresponding homozygous deletion mutants (Homann et al. 2009). The TOR pathway is critically involved in the regulation of amino acid and nitrogen metabolism and, consistent with that function, GLN3 (Dabas and Morschhauser 2007) and STP3 (Martinez and Ljungdahl 2005) are also involved in nitrogen metabolism. ZCF29 (orf19.5133) has recently been shown to be important in drug resistance (Shekhar-Guturja et al. 2016). Homozygous deletions of GLN3 (orf19.3912) STP3 (orf19.5915) and ZCF29 (orf19.5133) also show altered susceptibility to rapamycin (Homann et al. 2009).
Low rate of haploinsufficiency in rich media During pooled competitive growth The heterozygous TF mutants each contain a barcode and, therefore, are amenable to signature tag methods to quantitate phenotypic differences in pooled samples . We, therefore, asked if any of the n  TF heterozygotes showed growth defects in rich medium (YPD) at 30°. Pools containing 44-48 mutants were prepared by combining equal numbers of cells. The DNA from a sample of the resulting pool was isolated as the input and the ratio of strains was determined by quantitative PCR using barcode specific primers. The pool was then used to inoculate an YPD culture which was incubated for 20 generations to a final cell density between 0.5-1.0 OD 600 . The cells were harvested, the DNA was isolated, and the abundance of each strain was determined by quantitative PCR. The ratio of each strain in the recovered sample relative to the input sample was determined using the method described by Noble et al. and is described in more detail in the Materials and Methods ).
We defined a significant change in the abundance of a heterozygote strain as a statistically significant (P , 0.0001, Student's t-test with the Benjamani-Hochberg correction for multiple comparisons) twofold increase or decrease (6 1 log 2 ) in abundance relative to the inoculum at 24 hr. Although uga33D showed increased fitness, none of the heterozygous deletion mutants showed a competitive disadvantage under these conditions (see Table S2 for data). These observations are consistent with our plate-based assays in which we saw no apparent growth defect in rich media. The rate of haploinsufficiency in the S. cerevisiae genome has been estimated to be approximately 3% (Deutschbauer et al. 2005). Thus, if these were a random set of mutants, then we would expect 3-4 haploinsufficient strains at most. As such, it is not surprising that we did not observe growth defects. This set of data does, however, serve to validate our primers and barcodes and provides a benchmark for future studies using double heterozygous mutants or heterozygous mutants under other conditions.
Pooled competitive growth assay in a murine model of disseminated candidiasis identifies haploinsufficient transcription factor mutants Previous studies of TF homozygous deletion mutants using signaturetagged approaches in murine models of disseminated candidasis have identified a sub-set that is required for infection (Amorim-Vaz et al. 2015). We, therefore, carried out a similar experiment using pools of heterozygous TF mutants and the signature-tagged approach described above. The pools of mutants were used to infect mice by tail vein injection. The mice were killed at 24hr and 48hr, the kidneys were harvested, and the homogenates plated on YPD plates. After 24hr at 30°, the plated cells were collected and the DNA was isolated. The relative proportions of the pooled strains were determined by quantitative PCR as described for the in vitro experiment. We identified heterozygotes that a showed statistically significant (P , 0.0001, Student's t-test with the Benjamani-Hochberg correction for multiple comparisons), twofold or more increase or decrease (6 1 log 2 ) in abundance relative to the inoculum. Twelve heterozygous mutants met these criteria; eight showed a competitive fitness Figure 2 Representative haploinsufficent and haploproficient strains. Ten-fold dilution series of the indicated strains were spotted on agar plates and incubated at either 37°C (A) or 30°C for 1-3 days and photographed. Two independent isolates of each heterozygous strain were plated as noted by -1 and -2. A. The indicated strains were plated on YPD (yeast peptone dextrose) and SM (Spider medium). B. The strains were plated on YPD and YPD+15 mM caffeine. C. The indicated strains were plated on YPD+1%DMSO and YPD+1% DMSO+fluconazole.
disadvantage while four were more fit. The genes with altered competitive fitness and their annotations are listed in Table 2; the full set of data for the pools are provided in Table S3.
Five of the twelve mutants (ace2D orf19.6124, suc1D orf19.7319, tyr6D orf19.6824, crz2D orf19.2356, and uga33D orf19.7317) were identified by Finkel et al. as being required for adherence in vitro (Finkel et al. 2012). The experiment examined two time points early in the infection process (24 hr and 48 hr), times when differences in endothelial attachment may be quite prominent. Therefore, it seems reasonable that strains with altered adherence would be identified in this experiment. Of these five mutants, uga33D (orf19.7317) and tyr6D (orf19.6824) showed increased fitness while the other three strains showed reduced fitness. The uga33D (orf19.7317) also showed increased fitness during growth in rich medium. It is, therefore, possible that this mutant has a competitive growth advantage in vivo as well.
Although we cannot exclude a proliferation defect or delay for the other mutants, the lack of haploinsufficiency for these strains during growth in rich media reduces the likelihood that general growth defects contribute to the decreased infection fitness. We selected mutants with roles in in vitro adherence (ace2D, orf19.6124 and suc1D, orf19.7319) and morphogenesis (sfl1D, orf19.454 and sfl2D, orf19.3969) to confirm the pooled results with competitive experiments with wild type strains.
Ace2 (orf19.6124), the TF regulated by the RAM signaling pathway (Saptuo et al. 2015), and Suc1 (orf19.7319), a zinc finger TF originally identified as a regulator of sucrose utilization (Kelly and Kwon-Chung 1992), both were identified by Finkel et al. as regulators of adherence (Finkel et al. 2012). The homozygous deletion mutant of ACE2 (orf19.6124) also has been shown to have a virulence defect in disseminated candidiasis (Kelly et al. 2004). A previous screen of pooled homozygous TF deletion mutants without morphology defects found that suc1DD The competitive fitness was calculated as described in Materials and Methods. Competitive fitness data from representative pool of 48 heterozygous strains is shown. Mutants with statistically significant (P , 0.05, Student's t-test with Benjamani-Hochberg correction for multiple comparisions) decreased (red points) or increased (green points) competitive fitness relative to the median pool R/I. B and C. Competitive fitness of the indicated strain relative to a wild type reference strain. Bars indicate means of 3-5 independent mice with error bars indicate standard error of the means.
n Table 2 HeterozygousTF mutants with altered infectivity in murine model of disseminated candidiasis (orf19.7319) has a competitive fitness defect when animals are harvested late in infection (Pérez et al. 2013). Finally, we also tested the haploinsufficiency of sfl1D and sfl2D, two heat shock-type TFs with opposite effects on filamentaton (Znaidi et al. 2013). The homozygous deletion mutant of SFL1 (orf19.454) shows increased filamentation while sfl2D (orf19.3969) has a filamentation defect. As noted above, slf1D is hyperfilamentous while sfl2D was similar to wild type. In addition, SFL1 (orf19.454) is required for virulence in a disseminated candidiasis model (Li et al. 2007) while SFL2 (orf19.3969) was dispensable (Spiering et al. 2010). As shown in Figure 3B, three (ACE2 orf19.6124, SUC1 orf19.7319, and SFL1 orf19.454) of the four heterozygous strains showed significantly reduced fitness relative to the wild type reference strain in head to head competitions at 24 hr. The sfl2D (orf19.3969) showed slightly increased fitness but this is not likely to be biologically significant. These data are consistent with the findings of Spiering et al. which showed that SFL2 (orf19.3969) was not required for infectivity at 3 days (Spiering et al. 2010). Therefore, our pooled screening with 48 strains had a true positive rate of 75%; it is likely that the specificity of the screening experiment could be increased with the use of smaller pools.
Two of the genes that we chose to validate, ACE2 (orf19.6124) and SFL1 (orf19.454) are constitutively filamentous as homozygous deletions (Kelly et al. 2004;Saputo et al. 2012;Znaidi et al. 2013). This morphological phenotype complicates infectivity and virulence studies using tail vein infections. The ace2D (orf19.6124) mutant is not hyperfilamentous in vitro, nor did it have a growth phenotype. Therefore, our data indicate that Ace2 (orf19.6124) may have effects on infectivity that are independent of its role in the regulation of morphology. Since the sfl1D (orf19.454) mutant is hyperfilamentous, it remains possible and, indeed likely, that its apparent infectivity defect is due to this morphological feature.
Ace2 (orf19.6124) is regulated by the RAM network through phosphorylation by Cbk1 (orf19.4909) (Saputo et al. 2012). To our knowledge, the virulence of Cbk1 (orf19.4909) has not been assessed previously. This may be due to the fact that cbk1DD (orf19.4909), like ace2DD (orf19.6124), has a profound cell separation defect that complicates interpretation of in vivo studies (Saputo et al. 2012). Based on the phenotype showed by the ACE2 (orf19.6124) heterozygote, we hypothesized that the CBK1 (orf19.4909) heterozygote may also show Figure 4 Haploinsufficiency and genetic interactions of biofilm TFs during the initial adherence stage of biofilm formation. A. The indicated strains were incubated in microtiter plates in YETS medium at 37°C for 90 min before nonadherent cells were removed by careful washing with PBS. The bars indicate the mean OD 600 for each strain (triplicate) with error bars indicating standard deviation. Strains with a statistically significant (Student's t-test P , 0.05) change from the reference strain are indicated by red bars. B. The expected adherence fitness (AF = Mutant OD 600 /Reference OD 600 ) for a given double mutant (AF mutA X AF mutB ) is plotted on x-axis for each biofilm TF double heterozgote and the observed AF for the double mutant is plotted on the y-axis. Mutants that fall within the zone defined by the diagonal (e = 0 6 estimated error) have no genetic interaction while those above the diagonal (green) display alleviating interactions while those below (red) show aggravating interactions. C. Network of biofilm TF interactions during the adhesion step. Green edges indicate alleviating interactions; red edges indicate aggravating interactions. D. Aggravating interaction network centered on Tec1. E. Alleviating interaction network.
Finally, we generated the double heterozygote of cbk1D ace2D and examined its infectivity relative to each of the single heterozygotes. Surprisingly, the double mutant was restored to WT infectivity ( Figure  3C). One possible explanation for this observation is that we had previously observed that deletion of one ACE2 (orf19.6124) allele in the cbk1D (orf19.4909) background restored filamentation to the haploinsufficient parental cbk1D (orf19.4909) strain (Saputo et al. 2016). The homozygous cbk1DD (orf19.4909) strain is completely unable to form filaments while the ace2DD (orf19.6124) is hyper-filamentous. It is, therefore, possible that the increased filamentation displayed by the double cbk1D ace2D correlates with its improved infectivity relative to the two single heterozygotes.
Complex haploinsufficiency-based analysis reveals role for Tec1 During initial adherence Under biofilm forming conditions As part of a separate study of the genetic interaction network for mature biofilm formation (Glazier et al. 2017), we used our library to construct a complete set of double heterozygous TF deletion mutants corresponding to the six member biofilm TF network (Efg1 (orf19.610), Bcr1 (orf19.723), Tec1 (orf19.5908), Ndt80 (orf19.2119), Rob1 (orf19.4998), and Brg1 (orf19.4056)) elucidated by Nobile et al. (Nobile et al. 2012). Recently, the Johnson group reported the effect of homozygous deletion mutants of these TFs on the initial adherence step of biofilm formation (Fox et al. 2015). Accordingly, we used the corresponding set of double heterozygotes to identify functionally important interactions between the TFs during initial adhesion to abiotic surfaces. We used the same optical density-based assay to evaluate initial adhesion as Fox et al. (Fox et al. 2015); however, we used YETS medium instead of Spider medium because it gave less sample-tosample variation (Fox et al. 2015). As shown in Figure 4A, three homozygous deletion stains showed reduced initial adhesion (EFG1 (orf19.610), ROB1 (orf19.4998), and BCR1 (orf19.723)) while only ROB1 (orf19.4998) showed haploinsufficiency.
To characterize genetic interactions between mutants, we normalized the optical density for each mutant to wild type and will refer to this ratio as the adhesion fitness (AF). The multiplicative model was used to define genetic interactions (Mani et al. 2008); this model is the same as that used by Boone and co-workers for the genome-wide S. cerevisiae experiments (Constanzo et al. 2016). In the multiplicative model, the expected fitness of a double mutant derived from a pair of non-interacting genes is equal to the product of the fitness values for each individual mutant: AF 12 = AF 1 x AF 2 (Mani et al. 2008). This is also called genetic neutrality and indicates that the two genes affect the phenotype independently. If AF 12 , AF 1 x AF 2 , then the phenotype of the double mutant is more severe than predicted by the model and these genes show an aggravating interaction. If, on the other hand, AF 12 . AF 1 x AF 2 , then the two genes show an alleviating or suppressing interaction.
Traditionally, genetic interactions are expressed as Ɛ = AF 12 -(AF 1 x AF 2 ) where Ɛ . 0 is an aggravating interaction, Ɛ , 0 is an alleviating interaction and Ɛ = 0 is no interaction. We have provided Ɛ values for each double mutant pair in Table S2. However, as a more easily evaluated graphical representation of the data, we plotted the expected AF vs. the observed AF for the all double mutants with the estimated error. Mutants that plotted above the diagonal show aggravating interactions while those below the diagonal display alleviating interactions. In addition, we required that a double mutant show a statistically significant difference from the expected value to be classified as an interacting pair of genes (Student's t-test, P , 0.05).
An extensive set of genetic interactions was identified of the adhesion phenotype ( Figure 4B) with only two gene-pairs out of the total 15 gene-pairs showing no interaction (See Table S4 for complete set of adhesion data and e calculations). Three TFs [Tec1 (orf19.5908), Rob1 (orf19.4998), and Efg1 (orf19.610)] interacted with all five of the other TFs ( Figure 4C). Interestingly, Tec1 (orf19.5908) showed an aggravating interaction with all other TFs. Thus, the aggravating interaction network for adhesion is centered on Tec1 (orf19.5908) as a central node ( Figure 4D). This relationship would not have been identified using single gene analysis because the tec1D/D (orf19.5908) mutant does not have a defect in adhesion (Fox et al. 2015). Consequently, it appears that Tec1 (orf19.5908) functions in an ancillary or redundant manner to regulate the initial adhesion step of biofilm formation. In addition, our data indicate that Ndt80 (orf19.2119) and Brg1 (orf19.4056) also contribute to this step but likewise do so in a manner that overlaps with other TFs because their homozygous deletion mutants do not show a phenotype. Furthermore, our data are consistent with the notion that multiple TFs have overlapping functions during this step of biofilm formation.
If TEC1 (orf19.5908) plays a significant but ancillary role in adhesion, then its over-expression could be expected to suppress the adhesion defects displayed by other mutants. Therefore, we placed TEC1 (orf19.5908) under control of the strong, constitutive promoter TDH1 using the methods previously described by Finkel et al. (Finkel et al. 2012). As shown in Figure 5, over-expression of TEC1 (orf19.5908) increases adhesion in WT and tec1D (orf19.5908) strains. Additionally, increased expression of TEC1 (orf19.5908) in all of the adhesiondeficient strains that we tested led to increased adhesion. Taken together, these data suggest that TEC1 (orf19.5908) plays an important but ancillary role in the initial adhesion step of biofilm formation. In this role, it would function cooperatively with other TFs but be dispensable when a full complement of its functional partners is present.
A second interaction network consisting of only alleviating or suppressive interactions also emerges from adhesion data. In this case, a four-node structure connects Efg1 (orf19.610), Brg1 (orf19.4056), Bcr1 (orf19.723), and Rob1 (orf19.4998). Ndt80 (orf19.2119) has an alleviating interaction only with Efg1 (orf19.610). Alleviating interactions between two genes can indicate that the two genes function in a linear pathway and a "diminishing returns" effect limits the severity of two mutations that affect genes in the same pathway (Mani et al. 2008). Based on this interpretation, each of the nodes in the alleviating network would function in parallel pathways where reductions in gene dosage along one pathway are buffered. However, reduction in two genes/nodes that are not linearly-related disrupts both parallel pathways and leads to an aggravating interaction.
The biofilm TF network buffers adherent C. albicans cells From genetic perturbations that affect filament length The second step of biofilm formation is considered to be filamentation of the adherent cells (Desai and Mitchell 2015). To our knowledge, this step of biofilm formation has not been probed genetically. To do so, we developed an assay to directly measure the length of hyphae adhered to glass chamber slides. We inoculated each chamber slide with a given strain in YETS medium. After 2 hr, the chambers were washed to remove non-adherent cells. The adherent cells were heat fixed and then examined by light microscopy. The length of the filament was measured using commercially available software. The filament was defined as the region between the angle of the mother cell-filament junction and the distal tip of the filamentous portion of the cell. We did not distinguish between pseudohyphae or hyphae. Mutants that formed no discernible hyphae were excluded from analysis. A representative set of images are shown for the reference strain, tec1D (orf19.5908), bcr1D (orf19.723), and tec1D bcr1D are shown in Figure 6A.
We first examined adherent cell filament formation using the homozygous biofilm TF deletion mutants. Consistent with previously reported filamentation phenotypes under other conditions (Homann et al. 2009), none homozygous deletion biofilm TF mutants were able to filament with the exception of NDT80 (orf19.2119) which formed shorter filaments than the reference strain. The average reference strain filament length was 92 6 30 mm ( Figure 6A). Next, we determined whether the heterozygous TF deletion mutants were haploinsufficient or haploproficient. The filaments formed by five of the six heterozygous TF deletion mutants were between 11 and 20 mm shorter than the reference strain in a statistically significant manner (Mann-Whitney U-test, P , 0.05). The complete set of filament length data and calculations for e values are provided in Table S5. Although the filaments formed by ndt80D (orf19.2119) were longer than those formed by the reference strain, the difference was not statistically different that wild type.
Next, the filament length of the double heterozygous biofilm TF mutants was measured; the genetic interactions within the double mutants were determined using calculations described earlier. We defined an interaction as a filament length that differed from the expected value by $ 30% based on the standard distribution of the lengths. As shown in Figure 6B, 7 out of 15 double heterozygote mutants showed genetic interactions. A number of double mutants had statistically significantly shorter filaments than either single mutants but e = 0 for these strains, indicating that the genes did not interact and, hence, function independently. For example, the filaments of the tec1 bcr1 double heterozygote are shorter than either single heterozygote but are equal to the length expected if each single mutant contributed independently to the observed phenotype ( Figure 6A/B). All of the double mutants that show genetic interactions have filaments that are longer than predicted by the multiplicative model, indicating that only alleviating or suppressive interactions were observed. Figure 6 The genetic interactions of biofilm TFs during filamentation of adherent cells are distinct from adhesion and mature biofilm formation. A. Representative micrographs used to measure the filament length of the indicated strains. The lengths were determined as described in the Materials and Methods with the mean and standard deviation for the indicated strains derived from the measurement of at least one hunderd cells. B. The normalized filament length (mutant length/WT length) expected for a given double mutant was calculated as the product of the normalized mutant lengths of the two single mutants and is plotted on the y-axis. The observed filament length for a given mutant is plotted on the x-axis. Mutants that fall within the zone defined by the diagonal (e = 0 6 estimated error) have no genetic interaction while those above the diagonal (green) display alleviating interactions while those below (red) show aggravating interactions. C. The genetic interaction network of adherent cell filamentation centered on BCR1. Green indicates alleviating interactions. Figure 6C, BCR1 (orf19.723) interacts with all five of the network TFs. Consequently, BCR1 (orf19.723) forms the central hub of a pair of interlocking three-node motifs involving ROB1-BRG1 and EFG1-NDT80 (Zhang et al. 2005). Because this network is made entirely of alleviating interactions, it is quite distinct from either network observed during initial adhesion. It appears that the interacting TFs buffer the network against genetic perturbation with BCR1 (orf19.723) playing a key role in this effect. These observations, taken together with the adherence analysis, suggest that Tec1 (orf19.5908) and Bcr1 (orf19.723) play important but distinct roles during the initial stages of C. albicans interaction with abiotic surfaces under biofilm inducing conditions. Distinct TF interactions regulate filament length Under planktonic conditions relative to biofilm conditions The extensive literature on C. albicans filamentation suggests that the genetic requirements for this process are dependent on specific stimuli (Saputo et al. 2016;Azadmanesh et al. 2017). We, therefore, were interested in comparing the genetic interactions of the biofilm regulators during planktonic filamentation to those for cells adhered to an abiotic surface. To do so, we took advantage of a flow cytometry-based imaging assay that both distinguishes yeast from filamentous cells and allows the measurement of filament length ( Figure 7A). Among a number of advantages to this approach, it allows the rapid analysis of very large numbers of cells and, thereby, enables the detection of subtle changes in morphologic distribution with high statistical power. This is particularly advantageous for the analysis of simple and complex haploinsufficiency where phenotypes for individual mutants can be modest but informative.

As shown in
The biofilm-inducing medium used in the experiments described above (YETS) is not an efficient inducer of hyphae under planktonic conditions and, therefore, we used RPMI tissue culture medium at 37°f or six hours to induce hyphae. Under these conditions, 90% of the reference strain SN250 formed hyphal cells ( Figure 7A) while the homozygous deletions formed no filaments. All of the heterozygous and double heterozygous deletion mutants formed filaments with measurable lengths. The data for the full set single and double heterozygotes as well as the wild type control are provided in Table S6. Two heterozygous mutants, brg1D (orf19.4056) and rob1D (orf19.4998), showed haploinsufficiency with statistically significant reductions in mean filament length of 27% and 20%, respectively.
Using the definitions for genetic interactions described above for the analysis of cell length of adhered cells, eight aggravating Figure 7 The genetic interaction network of biofilm TFs during planktonic filament formation is distinct from adherent cell filamentation. A. Plot of the distribution of filament lengths for WT and rob1D tec1D derived ImageStream-based analysis along with mean length (mm) 6 SD for both strains. B. The normalized filament length (mutant length/WT length) expected for a given double mutant was calculated as the product of the normalized mutant lengths of the two single mutants and is plotted on the y-axis. The observed filament length for a given mutant is plotted on the x-axis. Mutants that fall within the zone defined by the diagonal (e = 0 6 estimated error) have no genetic interaction while those above the diagonal (green) display alleviating interactions while those below (red) show aggravating interactions. C. Genetic interaction network for alleviating interactions centered on BRG1. D. Genetic interaction network for aggravating interactions centered on ROB1. and five alleviating interactions were identified. The brg1D (orf19.4056) mutant only participated in alleviating interactions and is a hub connecting a network of alleviating interactions as shown in Figure 8B. Indeed, deletion of any other TF in the brg1D background suppressed its defect in filament length. In addition, the BRG1 (orf19.4056)-centered network contains two three-TF modules within the alleviating interaction network. This suggests that reduced copy number in other TFs triggers a compensatory response that overcomes the defects associated with reduced BRG1 (orf19.4056) copy number.
In contrast to the filamention of the adherent cells, five aggravating interactions were observed under planktonic conditions. Specifically, the set of interactions organizes into a highly connected network centered on ROB1 (orf19.4998). ROB1 (orf19.4998) interacts with all of the other TFs with the exception of BRG1 (orf19.4056) in a cooperative manner. In addition, each of the ROB1 (orf19.4998) interacting genes also interacts cooperatively with two other TFs in this network. Thus, Rob1 (orf19.4998) and Brg1 (orf19.4056) have very distinct functional interactions during planktonic filamentation formation. Rob1 (orf19.4998) functions cooperatively with multiple TFs during planktonic filamentation. Brg1 (orf19.4056), on the other hand, does not appear to cooperate with any of the other TFs. However, reduced gene dosage of the TFs that function in the Rob1 (orf19.4998) network appears to lead to a compensatory response that suppresses the effect of decreased Brg1 (orf19.4056) function.
Our results indicate that the network of genetic interactions controlling filament length varies dramatically depending on whether the cells are adherent to a glass surface or are suspended as planktonic cells. We cannot rule out the possibility that differences in culture medium (YETS vs. RMPI) contribute in some way to the reorganization of the TF network interactions. Because YETS medium does not induce planktonic cells to form true hyphae, we feel that signals specific to adherence are likely to make a significant contribution to the transcriptional program initiated during filamentation. Even if these changes are completely due to the medium, our data provide a dramatic example of how external cues reorganzie the specific interactions between a common set of TFs to yield a common morphological outcome.
The network of TF interactions regulating the ratio of yeast to filaments Under planktonic conditions is similar to that for planktonic filament length Although the relative proportion of yeast and filament forms could not be used to assess filamentation under adherent conditions for technical reasons, the Image Stream technology allowed us to examine this phenotype under planktonic conditions ( Figure 8A). As mentioned above, 90% of the reference strain SN250 formed filaments. Of the heterozygous mutants, brg1D (orf19.4056) was the only strain that showed a statistically significant reduction in the proportion of filamentous cells (78% hyphae, P , 0.0001, Fisher's exact test, N = 2720). Similar to the interactions observed for filament length, deletion of one allele of ROB1 (orf19.4998), BCR1 (orf19.723), and EFG1 (orf19.610) in the brg1D (orf19.4056) background suppressed its haploinsufficiency and restored the brg1D double mutants to reference strain levels of filament formation ( Figure 8B). The full set of data are provided in Table S6.
Consistent with the filament length data, ROB1 (orf19.4998) showed aggravating interactions with TEC1 (orf19.5908), EFG1 (orf19.610), BCR1 (orf19.723), and NDT80 (orf19.2119) ( Figure  8B&C). TEC1 (orf19.5908) and EFG1 (orf19.610) also show an aggravating interaction. The most severely affected double mutant is Figure 8 The genetic interaction network of biofilm TFs affecting the proportions of yeast and filamentous cells under planktonic conditions is similar to that for length under planktonic conditions. A. ImageStream gating diagrams showing WT and rob1D tec1D yeast and filamentous cells. B. The expected normalized proportion of filamentous cells (mutant/WT) for the double mutants was calculated as the product of the normalized filament proportion of the two single mutants and is plotted on the y-axis. The observed normalized filament proportion for a given mutant is plotted on the x-axis. Mutants that fall within the zone defined by the diagonal (e = 0 6 estimated error) have no genetic interaction while those above the diagonal (green) display alleviating interactions while those below (red) show aggravating interactions. C. The genetic interaction network for alleviating interactions.
the rob1D tec1D strain (60% hyphae, P , 0.0001, Fisher's exact test, N = 2468). The interaction network for aggravating interactions for this set of TFs is shown in Figure 8C. ROB1 (orf19.4998) appears to function as a local node and as a component of a three TF module with EFG1 (orf19.610) and TEC1 (orf19.5908). The most common function of three-component TF modules is as feed-forward loops (Mangan and Alon 2003;Zhang et al. 2005). Consistent with that formulation, Efg1 (orf19.610) is known to directly bind and regulate the expression of TEC1 (orf19.5908). Nobile et al. has shown that Tec1 (orf19.5908) and Efg1 (orf19.610) bind to the promoter of ROB1 (orf19.4998) (Nobile et al. 2012) and, in other work, our laboratory found that Tec1 (orf19.5908) and Efg1 (orf19.610) cooperate with Rob1 (orf19.4998) to regulate ROB1 (orf19.4998) expression (Glazier et al. 2017). Since all three of these TFs regulate genes that are involved in hyphae morphogenesis, it is likely that cooperative effects on target gene expression also contribute to the observed genetic interactions as well. The similarity in genetic interactions observed for filament length and proportion of filamentous cells during planktonic conditions indicates that these two phenotypes are controlled by very similar transcriptional responses.

DISCUSSION
The genetic analysis of simple and complex heterozygotes can provide information that is complementary to that obtained by the analysis of homozygous deletion strains. In addition, genetic interaction analysis can be problematic in diploid organisms such as C. albicans, particularly if the gene of interest is essential or if the homozygous deletion mutant has a very strong phenotype. Here, we described a set of strains and plasmids designed to facilitate genetic interaction analysis in C.albicans using CHI. We have characterized phenotypes for this set of TF heterozygous deletion strains both in vitro and in vivo as a benchmark for subsequent studies. Consistent with previous studies of the prevalence of haploinsufficiency in other yeast, the phenotype is relatively uncommon and we identified only a handful of heterozygous strains that displayed altered function.
The in vitro phenotype with the most haploinsufficient/proficient mutants was filamentation. Homann et al. found that eight homozy-gous TF deletion mutants were deficient for filamentation on Spider medium at 37°while an additional eight strains were hyperfilamentous (Homann et al. 2009). Of the eight homozygous mutants that showed decreased filamentation, three were also haploinsufficient. In contrast, all of the heterozygous counterparts of hyperfilamentous homozygous TF deletions were also hyperfilamentous. In addition, a ninth mutant, ZFU3 (orf19.6888) showed increased filamentation. Although this strain was not hyperfilamentous in the data set reported by Homann et al. Vandeputte et al. reported that zfu3DD (orf19.6888) was hyperfilamentous (Homann et al. 2009;Vandeputte et al. 2011). The rate of haploinsufficiency among the negative transcriptional regulators of filamentation (100%) is strikingly above the rate for the overall genome and for other phenotypes that we tested with this set of mutants. Based on these data, it seems that C. albicans may be more sensitive to alterations in the gene dosage of negative regulators of hyphal morphogenesis than of positive regulators.
We did not observe haploinsufficient growth phenotypes for the TF heterozygous deletion strains on solid agar plates with rich media and only one heterozygous deletion mutant, UGA33 (orf19.7317), showed increased growth in competitive pooled assays in the same media. Homan et al. found that nine homozygous mutants had growth defects under the same conditions and, therefore, it is not surprising that no haploinsufficiency was observed given the low rate of its occurrence overall (Homann et al. 2009). In contrast, haploinsufficient strains were identified at either 24 or 48 hr during disseminated candidiasis in a mouse model. Although there have been previous competitive infectivity screens using homozygous deletion mutants, none had used the early time points that we employed. Nearly one-half of the TF heterozygotes with altered infectivity at early time points were mutants that had been previously shown to be involved in abiotic adhesion under flow conditions (Finkel et al. 2012). Although we have not determined whether these strains are also haploinsufficient for adherence to abiotic surfaces under flow conditions, it is reasonable that our experiment would identify such mutants since endothelial cell attachment under conditions of flow is a crucial and early step in the establishment of a disseminated infection. Figure 9 The genetic interactions of the biofilm TFs are dynamic and are temporally reorganized during biofilm formation. The genetic interaction networks for BCR1 and TEC1 are shown for adhesion, adherent cell filamentation, and mature biofilm stages of biofilm formation. The interaction networks for the mature biofilm stage are derived from Glazier et al. 2017. We used a set of double heterozygous mutants derived from TFs previously shown to regulate biofilm formation to genetically characterize two early steps in the establishment C. albicans biofilm (Nobile et al. 2012;Fox et al. 2015): adherence to an abiotic surface and filamentation. Previously, we had used the same set of double mutants to study the genetic interactions at the point when a relatively mature biofilm had formed, i.e., 48 hr (Glazier et al. 2017). At the stage of initial adherence, we found that the set of six TFs displayed a very different set of intra-network genetic interactions relative those observed for mature biofilms. Thus, the roles of the TFs vary depending on the specific stage of biofilm formation. This is consistent with the recently reported homozygous deletion study of the same set of mutants (Fox et al. 2015). Combined with those results, our data support a model in which different stages of biofilm formation specify different combinations of TFs for optimal function.
We also found that TEC1 (orf19.5908) plays an important role that was not evident from analysis of the homozygous deletion mutants. Deletion of one allele of TEC1 (orf19.5908) in the context of a heterozgous deletion mutation in any of the other TFs led to an aggravating interaction, indicating that TEC1 (orf19.5908) functions cooperatively with the other TFs. Consequently, Tec1 appears to be required to buffer the cell against reduced gene dosage in other biofilm-related TFs during the initial adhesion step. Further supporting this role for Tec1 (orf19.5908) is the fact that overexpression of TEC1 (orf19.5908) increases adhesion of wild type, single heterozygotes, and double heterozygotes with decreased biofilm adhesion. Tec1 (orf19.5908) is bound by and regulated by all other network TFs during biofilm formation (Nobile et al. 2012) as well as by itself. The genetics of its interaction with other TFs, taken together with its regulation pattern, suggest that TEC1 (orf19.5908) is an important output of the TF network. Furthermore, it interacts with all of the other five TFs as part of three, three-module motifs ( Figure 4D). Within genetic interaction networks, three module TF motifs commonly represent feed-forward loops (Mangan and Alon 2003). Thus, the structure of the adhesion network strongly supports the notion that all of the other TFs contribute to TEC1 (orf19.5908) expression through a feed-forward mechanism. Finally, this overall model is supported by the transcriptional profiling data reported by Fox et al. in that TEC1 expression is elevated in adherent cells as compared to planktonic cells grown at the same temperature (Fox et al. 2015).
One of the most potentially informative features of genetic interaction studies is its ability to provide insights into how sets of genes with common functions work together to regulate a biological process (Costanzo et al. 2016). Taking the genetic interactions for the biofilm-related TFs for the early stages of biofilm formation together with our previous data for the genetic interactions at a much later stage of the biofilm process (48 hr), it is clear that the network is not a static set of interactions but is instead more malleable in nature. Taking the set of Tec1 (orf19.5908) and Bcr1 (orf19.723) interactions as illustrative examples (Figure 9), the variable nature of their interactions with other network TF over the course of biofilm formation becomes quite clear. At first glance, it may seem that the interactions of BCR1 (orf19.723) are relatively stable because it interacts with most of the TFs at each step of the process. However, only one of its thirteen interactions, an alleviating interaction with BCR1 (orf19.723), is consistently maintained across each stage of biofilm formation. The interactions displayed TEC1 (orf19.5908) are drastically different at each stage and none are consistently manifest at each stage. These dynamic reorganizations are characteristic of a temporal network, as opposed to a static network (Li et al. 2017). One of the advantages of temporal networks is that they are able to control a wide range of functions with a relatively small number of nodes (Li et al. 2017). The significant changes in biofilm TF network topology and function also fit well with Luscombe et al. who showed that, in S. cerevisiae, TFs do not necessarily serve as permanent hubs but instead frequently had transient roles leading to significant changes in network topology in response, in particular, to environmental cues (Luscombe et al. 2004). During biofilm formation, the cell's microenvironment alters dramatically as the the process progresses and likely provides the external cues leading to the reorganization of the network topology and the specific alterations in the roles of the specific TFs within the network. Consequently, the genetic interactions of the C. albicans biofilm TFs provide insights into how a relatively small set of TFs is able to orchestrate the multistep process of constructing a biofilm. What remains to be explored are the mechanisms by which the cell is able to reorganize the specific contributions of the individual TFs to execute each step of this complex process.
In summary, we have constructed a set of strains and plasmids for the systematic genetic interaction analysis of C. albicans TFs based on simple and complex haploinsufficiency. In addition to providing these tools and baseline phenotypic characterization for the strains, we have used the library to probe genetics regulation of the early stages of biofilm formation and filamentation. The latter analyses provide genetic support for the notion that the TF interaction networks are quite plastic with specific combinations of TFs working together under specific conditions.