Phenotypic Variability Correlates with Clinical Outcome in Cryptococcus Isolates Obtained from Botswanan HIV/AIDS Patients

Cryptococcosis results in hundreds of thousands of deaths annually, predominantly in sub-Saharan Africa. Cryptococcus is an encapsulated yeast, and during infection, cells have the capacity for substantial morphological changes, including capsule enlargement and shedding and variations in cell shape and size. In this study, we examined 70 Cryptococcus isolates causing meningitis in HIV/AIDS patients in Botswana in order to look for associations between phenotypic variation and clinical symptoms. Four variant phenotypes were seen across strains: giant cells of ≥15 µm, micro cells of ≤1 µm, shed extracellular capsule, and irregularly shaped cells. We found that “large” and “small” phenotypes were associated with differing disease symptoms, indicating that their production may be important during the disease process. Overall, our study indicates that Cryptococcus strains that can switch on cell types under different situations may be more able to sustain infection and resist the host response.

having the highest level. Evidence of recombination, shown as loops in the networks, was seen in the VNBI and VNBII populations. Induced capsule and cell size differ between species and genotypes. Capsule thickness and yeast cell diameter were measured after growth in Dulbecco's modified Eagle medium (DMEM) broth with 5% CO 2 at 37°C for 5 days, and data were compared across genotypes for mean and variation (Table 1; Fig. 2A to C). On average, C. tetragattii isolates had significantly greater capsule thickness (P ϭ 0.0025) and yeast cell diameter (P Ͻ 0.0001) than C. neoformans isolates. Within C. neoformans genotypes, VNBI isolates had significantly greater capsule thickness than VNI isolates (P ϭ 0.0043) and significantly greater yeast cell diameter than VNI (P ϭ 0.0032) and VNBII (P ϭ 0.0178) isolates. F test analysis was used to compare the variance of the data (Table 1). Despite their limited genetic diversity, C. tetragattii isolates had significantly more variation in capsule thickness than C. neoformans isolates (P ϭ 0.0245). There were no significant differences in variation in capsule thickness among C. neoformans genotypes. Yeast cell diameter measurements did not vary significantly different between any groups.
Giant cells are more frequent in C. tetragattii, while micro cells and extracellular capsule are present only in C. neoformans. Following growth under inducing conditions, the three variant phenotypes seen at differing frequencies across isolates were giant cells (yeast cell diameter of Ն15 m) (Fig. 3A), micro cells (yeast cell diameter of Յ1 m) (Fig. 3B), and shed capsule (Fig. 3C). Elongated and irregularly shaped cells were seen in a number of strains across all genotypes (except VNII where n ϭ 2) ( Fig. S1D and Table S1), but these comprised a small subset (Ͻ5%) of the total population of cells. Cells larger than 15 m have previously been classed as "titan cells"; however, as in vivo titan cells possess additional defining characteristics that were not measured in this study, including altered capsular structure, increased DNA content, and increased vacuolar formation (27), they will be referred to here as giant cells. The Isolate names in black type correspond to patients who were alive at the time of the analysis, while isolate names in white type on darker backgrounds correspond to patients who died. C. neoformans VNI (n ϭ 17), VNII (n ϭ 2), VNBI (n ϭ 25), VNBII (n ϭ 9), C. gattii VGI (n ϭ 1), and C. tetragatiii VGIV (n ϭ 16) are shown. MLST alleles for each strain can be found in Table S1 in the supplemental material.
Phenotypic Variability in Cryptococcus ® number of clinical isolates exhibiting these morphological variants for each genotype is summarized in Table 1. Giant cells were significantly associated with greater capsule thickness (P Ͻ 0.0001) and yeast cell diameter (P Ͻ 0.0001), indicating that isolates with larger cells are more likely to produce giant cells, and Fig. 2B shows that cell size is generally spread along a continuum. Micro cells and shed capsule, however, were not significantly associated with capsule thickness or yeast cell diameter in C. neoformans, indicating that cell size is not related to their production and that micro cells are a distinct cell type rather than the endpoint of a continuum of increasingly smaller cells. The production of giant cells was significantly associated with C. tetragattii (50% of isolates; P ϭ 0.0070) compared to C. neoformans (15% of isolates), while micro cells and shed capsule were seen only in C. neoformans (58 and 81% of isolates, respectively) and were significantly associated with each other (P Ͻ 0.0001). Within C. neoformans, giant cells, micro cells, and shed capsule were observed in all genotypes (except VNII, which is inconclusive as n ϭ 2). To confirm that micro cells and shed capsule are two distinct phenotypes, isolates were stained with 4=,6-diamidino-2-phenylindole (DAPI) (Fig. 3D) and calcofluor white (Fig. 3E) after capsule induction to investigate nuclear and cell wall morphology, respectively. Micro cells displayed fluorescence comparable to regular cells, demonstrating that they have nucleic acid material and intact cell walls. Shed capsule clustered around the cell or released into the medium displayed no fluorescence.
Correlations occur among clinical, phenotypic, and genotypic variables. Correlations were assessed between clinical, phenotypic, and genotypic variables across the isolate collection. Correlation plots were generated showing the direction, strength, and statistical significance of correlations between clinical variables overall (Fig. 4A), between phenotypic variables overall (Fig. 4B), and between clinical and phenotypic variables overall and, for certain clinical variables, for C. neoformans and C. tetragattii species individually (Fig. 4C). All other analyses and individual P values are recorded in Table S2. Principal-component analysis (PCA) was used to investigate the variance and dimensionality of the data set. PCA biplots were generated using the first two components for clinical data accounting for 25.4% of variation (Fig. 4D) and for phenotypic data accounting for 61.9% of variation (Fig. 4E). Phenotypic Variability in Cryptococcus ® Most of the significant associations found between clinical symptoms were as expected, with similar variables being positively correlated, including fever and temperature (P Ͻ 0.0001), cough and chest pain (P ϭ 0.0013), altered mental status and behavioral changes (P Ͻ 0.0001), and nausea and vomiting (P Ͻ 0.0001). Similarly, cognitive symptoms (such as confusion, behavioral changes, and altered mental status), respiratory symptoms (such as cough, shortness of breath, chest pain, and respiratory rate), and intracranial symptoms (such as visual changes, blurred vision, neck pain, and photophobia) grouped closely together in the PCA biplot of clinical variables (Fig. 4D).  Table S2 in the supplemental material. (D and E) PCA biplots of the first two significant dimensions obtained using principal-component analysis of clinical data accounting for 25.4% of variation in the data set (D) and phenotypic data accounting for 61.9% of variation in the data set (E). Large diamonds represent genotype averages, the length and opacity of arrows represent the degree of contribution of that variable to the model, and genotype ellipses represent the 80% confidence interval.

Fernandes et al.
Given the complexity of clinical data that relies on self-reporting by patients that may be quite ill and subsequent interpretation by clinicians, this serves as a as a useful internal control for data quality. Patient death was significantly positively correlated with respiratory rate (P ϭ 0.0101), cerebellar signs (P ϭ 0.0478), and cerebrospinal fluid (CSF) CFU (P ϭ 0.0048).
The capacity for production of "large" and "small" phenotypes correlates with certain clinical symptoms indicative of early and late stage infection, while the capacity for variation is associated with patient death. In the PCA biplot of phenotypic variables (Fig. 4E), properties based on cell and capsule size grouped closely together along with giant cells, while micro cells and shed capsule grouped closely with each other and away from the other properties. These groups were a strong driver of variation in the biplot, indicating genotype-specific differences, while the production of irregular cells was not. When correlated with clinical data, the phenotypic variables within these groups showed similar directions and strengths of association, with "large" phenotypes, including greater capsule thickness, greater yeast cell diameter, and giant cells generally grouping and often opposing the "small" phenotypes of micro cells and shed capsule (Fig. 4C).
CD4 T-cell count is a reliable predictor of host immune status, with counts of Ͻ500 indicating immune suppression, and counts of Ͻ200 in HIV-infected individuals indicating AIDS. In the current study, CD4 counts ranged from 2 to 389 (Table S1). Overall, CD4 count was positively correlated with cell size across the entire data set (P ϭ 0.0250) and negatively correlated with the production of shed capsule (P ϭ 0.0158). Nausea and vomiting are frequent symptoms of increased intracranial pressure during cryptococcosis, and these symptoms were negatively correlated with "large" cell phenotypes: nausea was negatively correlated with yeast cell diameter (P ϭ 0.0135) and giant cells (P ϭ 0.0357), and vomiting was negatively correlated with all "large" phenotypes, including total diameter (P ϭ 0.0007), yeast cell diameter (P ϭ 0.0010), capsule thickness (P ϭ 0.0014), volume ratio (P ϭ 0.0269), and giant cells (P ϭ 0.0041). Vomiting was also positively correlated with the "small" phenotypes, including micro cells (P ϭ 0.0077) and shed capsule (P ϭ 0.0153). Shed capsule was also negatively correlated with visual changes (P ϭ 0.0358) and photophobia (P ϭ 0.0032), symptoms typically attributed to meningeal irritation. A similar pattern was seen with fever and neck stiffness, which are associated with an aggressive inflammatory response. Patient temperature was positively correlated with total diameter (P ϭ 0.0412), capsule thickness (P ϭ 0.0104), and volume ratio (P ϭ 0.0016) in the C. neoformans isolates, while neck stiffness correlated negatively with micro cells across the entire collection (P ϭ 0.0041).
Finally, patient death was significantly positively correlated with the production of all three major morphological variants (P ϭ 0.0165); four patients had isolates that produced giant cells, micro cells, and shed capsule, and all four patients died during the period when clinical data were scored. The production of all three major morphological variants was also positively correlated with the patient having tuberculosis (P ϭ 0.0133) and negatively correlated with systolic blood pressure (P ϭ 0.0474). In contrast, irregular cells were negatively correlated with death (P ϭ 0.0241). These were also more likely to be produced by isolates obtained from patients who had undergone antifungal therapy with either fluconazole or amphotericin B prior to admission (P ϭ 0.0438). All of the above associations were tested for each species alone as well as across the entire data set, and the trends remained largely the same, although some statistical power was lost due to smaller sample sizes, indicating that these results were not being driven by an association with one of the species.
Some correlations show species specificity. Several species-specific correlations were found, indicating that aspects of pathogenesis and host response are distinct between C. neoformans and C. tetragattii isolates. In the PCA biplot of phenotypic variables, the C. tetragattii isolates grouped mostly separately from C. neoformans isolates due to their larger cell and capsule sizes, higher incidence of giant cells, and lack of micro cells and shed capsule (Fig. 4E). C. neoformans isolates were significantly associated with lower CD4 count (P ϭ 0.0103), higher LP opening pressure (P ϭ 0.0208), and vomiting (P ϭ 0.0013) compared to C. tetragattii. Isolates from the C. neoformans genotype VNBI were significantly associated with lower CD4 count (P ϭ 0.0063) and lower diastolic blood pressure (P ϭ 0.0235) compared to all other isolates (Table S2). In the C. neoformans population, shortness of breath was positively correlated with total diameter (P ϭ 0.0430), capsule thickness (P ϭ 0.0366), and volume ratio (P ϭ 0.0367). In C. tetragattii isolates, yeast cell diameter was positively correlated with dizziness (P ϭ 0.0292) and negatively correlated with respiratory rate (P ϭ 0.0335).

DISCUSSION
Our study aimed to correlate phenotypic variation in cryptococcal isolates with pathogenesis and clinical manifestation of disease using in vitro stresses designed to simulate those encountered during human infection. The observed phenotypes were a stable attribute of the isolates, although we cannot be certain that these phenotypes would occur in infected patients in the same way. Furthermore, the clinical data set used in this study contains missing values, and certain parameters rely on selfreporting, which may not be very robust. Despite this, we saw some strong correlations and trends, and while this does not necessarily imply causation, the strength and pattern of the associations indicate that phenotypic plasticity and morphological presentation do play a role in cryptococcal disease.
Phenotypic plasticity is high and not related to genetic diversity. Almost a quarter of the clinical isolates in this study were identified as C. tetragattii, a species which is uncommon globally but has been found to cause a relatively high incidence of disease in HIV/AIDS patients in sub-Saharan Africa, having been identified in Botswana, Malawi, South Africa, and Zimbabwe (10,(28)(29)(30). Relatively little is known about this species; however, our previous study of the C. gattii complex found that C. tetragattii is similar to other species in the complex in terms of capsule production and cell size, but along with C. bacillisporus is more temperature sensitive than virulent genotypes C. gattii and C. deuterogattii, and commonly produces irregular cells. This suggests that C. tetragattii is a "weaker" pathogen and may be infecting immunocompromised hosts due to high environmental presence in Botswana and other sub-Saharan African countries. Environmental studies of the area are needed to determine where its niche lies.
MLST analysis revealed a limited amount of genetic diversity among C. tetragattii isolates, which contrasted with high genetic diversity and evidence of recombination in the C. neoformans genotypes (Table 1; Fig. 1), something that has been seen in other studies of southern African C. neoformans populations (29,(31)(32)(33). Significant phenotypic differences occurred among, between, and within the C. neoformans genotypes and within C. tetragattii, and despite their clonal nature, C. tetragattii isolates had significantly more variation in capsule thickness (P ϭ 0.0245), suggesting that some phenotypic differences may be a result of epigenetic mechanisms. Rhodes et al. (2017) found the most rapidly evolving genes between lineages of C. neoformans to be transcription factors and transferases, suggesting that transcriptional reprogramming and remodelling may be responsible for generating phenotypic diversity, rather than genomic changes (34). Phenotypic plasticity allows rapid adaptation inside the host and is common in many fungal species, such as Candida albicans, which can exhibit various morphological types, including yeast and filamentous forms, and white and opaque forms which influence mating, virulence, and interactions with immune cells in vitro (35). Cryptococcus, like Candida, appears to show a level of pleomorphism with different phenotypes appearing in response to stress.

Clinical markers of early and late infection suggest that cell phenotypes change during the course of infection and may play a role in immune response.
A model of the relationship between the various phenotypic variants, cryptococcal species, and clinical variables based on the significant associations and trends found in this study is presented in Fig. 5. CD4 count, representing the immunity status of the patient, was overall positively correlated with "large" phenotypes, suggesting that larger cells, more capsule, and giant cells are produced during earlier stages of infection, and overall negatively correlated with "small" phenotypes, suggesting that micro cells and shed capsule may be a later response during infection as immune function declines (Fig. 4C and Fig. 5). Nausea and vomiting are symptoms that are typically associated with increased intracranial pressure, as this results in stimulation of the vomiting center of the brain. These were overall negatively correlated with "large" phenotypes, again suggesting that larger cells may be an earlier response, and overall positively correlated with "small" phenotypes. The latter could be expected, as shed capsule in particular has long been implicated in increasing intracranial pressure due to the accumulation of capsular polysaccharide blocking passage of the CSF across arachnoid villi (36). In all, this suggests a transition from "large" to "small" cell phenotypes as infection progresses and intracranial pressure rises.
The "small" phenotypes were also overall negatively correlated with visual changes (significantly with shed capsule [P ϭ 0.0358]) and photophobia (significantly with shed FIG 5 A summary model of the phenotypic variants seen in this study and the direction of their associations with the C. tetragattii or C. neoformans species and clinical variables. "Large" phenotypes include larger yeast cell size, larger capsule size, and giant cells; these are prevalent in C. tetragattii and are generally correlated with symptoms indicating a less suppressed immune system and low intracranial pressure. "Small" phenotypes include micro cells and shed capsule; these are prevalent in C. neoformans and are generally correlated with symptoms indicating a more suppressed immune system, high intracranial pressure, and low inflammation. Overall, a higher capacity for variation may play a role in increased virulence in Cryptococcus. Phenotypic Variability in Cryptococcus ® capsule [P ϭ 0.0032]), symptoms associated with inflammation of the meninges (43). It has been observed that cryptococcal meningitis in HIV patients is characterized by a lack of an active host inflammatory response (36) so it is possible that micro cells and extracellular capsule accumulating in the CSF and causing raised intracranial pressure might also be dampening the host inflammatory response. The major cryptococcal capsular polysaccharide (GXM) has been documented to have numerous immunosuppressive properties, including modulation of cytokine secretion, induction of macrophage apoptosis, and suppression of leukocyte migration (37)(38)(39)(40). Studies in mice have found GXM to markedly dampen the hyperinflammatory response via inhibition of proinflammatory cytokine secretion (41), and the administration of purified GXM reduced the number of immune cells infiltrating the brain (42).
Capsule size in cryptococcal cells was found to differ significantly between genotypes, species, and individual isolates ( Table 1; Fig. 2A to C). Compared to all C. neoformans genotypes, C. tetragattii isolates had significantly thicker capsules (P ϭ 0.0025) and larger yeast cells (P Ͻ 0.0001). To date, studies investigating the relationship between capsule size and virulence have found conflicting results. Robertson et al. (2014) found that highly encapsulated C. neoformans isolates were significantly associated with lower fungal clearance rates, poor inflammatory responses, and increased intracranial pressure (43). In contrast, Pool et al. (2013) showed that hypercapsular C. neoformans isolates possessed the least neurovirulence in a mouse model, while isolates producing less capsule were more virulent and resulted in a higher fungal load in the brain (44). As such, it is unclear whether a large capsular phenotype enhances the overall virulence of Cryptococcus. Our current study found that capsule size strongly correlates with yeast cell size, and it is possible that capsule alone has less importance than the overall size of the cell.
Giant cells are prevalent in C. tetragattii and likely result from a gradual increase in cell size, while micro cells appear exclusive to C. neoformans and are a distinct cell population strongly associated with shed capsule. The presence of giant cells has been recorded in infected tissues during mammalian infection, where they appear most frequently in the extracellular space (21,22,25,45). Giant cells are thought to be important in the establishment and persistence of infection due to their large size preventing phagocytosis and allowing them to remain in the host for long periods of time (16). Giant cells also appear to increase virulence through conferring resistant properties to their normal-size progeny, enhancing their capacity for survival and dissemination and thus increasing the overall virulence of the strain (16,45). The vast morphological changes associated with in vivo giant cell production, including alteration of cell body and organelles, indicate that giant cell production is a developmental transition (27,46). Recent studies on giant cell production found that the capacity to produce giant cells in vitro may be a reliable predictor of their formation in vivo (47) and that regularly sized cells present in the initial inoculum transitioned progressively toward the giant cell phenotype under inducing conditions (48) with the transition occurring at low cell densities (49). The current study found giant cells to be significantly associated with larger average yeast cell diameters (P Ͻ 0.0001) and greater capsule thickness (P Ͻ 0.0001) in both C. neoformans and C. tetragattii (Fig. 3A), which supports the hypothesis of a gradual shift toward a larger phenotype. Although the majority of literature on giant cells relates to C. neoformans, we found that the presence of giant cells was significantly associated with the C. tetragattii complex.
Cryptococcus micro cells are an intriguing and understudied class of cell that has received little attention thus far. These cells are commonly seen in infection and have been speculated to assist in the infection process, with their small size allowing them to cross biological barriers and to disseminate easily to the brain (14,44). The presence of micro cells was strongly (P Ͻ 0.0001), but not always, associated with shed capsule (Table 1; see Table S2 in the supplemental material), suggesting that they are induced by the same or similar processes. Staining with DAPI and calcofluor white confirmed that micro cells are real cells with nuclear material and cell walls and that they are distinct from shed capsule (Fig. 3D and E). Furthermore, unlike giant cells, micro cells appear to be a distinct cell class, as there was no correlation with cell size, indicating that there is no continuum of increasingly smaller cells. Their presence appears to be exclusive to C. neoformans (and seen across genotypes); the current analysis and our previous study investigating 70 C. gattii complex strains, including C. gattii, C. deuterogattii, C. bacillisporus, and C. tetragattii clinical, environmental, and veterinary strains found no evidence of micro cells in any strain (26).
Pleomorphism may have a role in the virulence of Cryptococcus. The production of "large" and "small" phenotypes was largely mutually exclusive; most isolates possessed either giant cells or micro cells/shed capsule but rarely both. It is therefore intriguing that the four isolates that produced all three major morphological size variants (giant cells, micro cells, and shed capsule) resulted in patient death (P ϭ 0.0165) (Fig. 4C). While this must be viewed with the limitation that there were very few isolates within this category, it may indicate that the capacity for variation plays a role in virulence. Three out of four of these isolates belonged to VNBI, which appeared to be the most virulent genotype with the highest percentage of patient deaths, and also with the largest average capsule sizes within C. neoformans. As "large" and "small" variants appear associated with quite different disease symptoms, this could enable greater capacity for infection, immune evasion, and pathogenesis. An alternative hypothesis is that in the severely weakened immune state of late HIV/AIDS patients, different cryptococcal cell types can flourish; however, many isolates caused patient death but did not demonstrate this diversity of cell types. Further work into the factors that induce each of these unique cell types and their role in virulence and disease progression is required to substantiate this preliminary finding.
Conversely, the presence of "irregular cells" with irregularly shaped and elongated cell morphologies was significantly negatively correlated with death (P ϭ 0.0241), suggesting that these are defective cells. Irregular cells were not associated with any particular species or genotype and had no significant correlations with any other phenotypic variable, but they were significantly more likely to occur in isolates obtained from patients who had undergone antifungal therapy prior to admission (P ϭ 0.0438). This suggests that irregular cells may be produced by isolates with reduced drug susceptibility but with a cost of lower resistance to host-imposed stress, making them less able to mount aggressive disease. We have found that in vitro, the production of these cells may be triggered by the stress of nutrient limitation (26). The extent of the elongation of these cells varied from isolate to isolate, but in some cases, they appeared to approach the morphology of pseudohyphal forms. Pseudohyphal forms have been reported in Cryptococcus but are thought to be rare during cryptococcosis and are also associated with reduced virulence (27,50).
This study demonstrates the complex relationship between phenotypic variation and adaptation to the host environment in Cryptococcus, with pleomorphic characters potentially contributing to overall virulence. As different properties may be beneficial at different stages and sites of infection, isolates that are able to produce diverse cells in response to changing situations may be more able to sustain infection and resist the host response.

MATERIALS AND METHODS
Cryptococcus isolates. A collection of 70 Cryptococcus isolates were provided from two major public hospitals in Botswana: the Princess Marina Hospital in Gaborone and the Nyangbwe Referral Hospital in Francistown, as part of an ongoing study into cryptococcosis in Africa. Some of this collection has been reported previously in Chen et al. (2015) (29). Fungal cells were cultured from the cerebrospinal fluid (CSF) of patients with HIV/AIDS and cryptococcal meningitis enrolled within an 18-month period from 2012 to 2013. Cryptococcal meningitis was confirmed by a positive India ink test or CSF culture. Patients received induction therapy with amphotericin B; however, all isolates were obtained before treatment commenced. Any patients where death was not attributable to cryptococcal meningitis were excluded from the data set. Screening on fluconazole plates found no isolates with high-level resistance. All isolates used in this study are referred to by their isolate identifier (ID) and are listed in Table 1, with full details of isolate name and genotypic, phenotypic, and clinical data listed in Table S1 in the supplemental material. Phenotypic data for type strain H99 is also included in this table as a reference.
Culture conditions. Cryptococcus isolates were cultured from -80°C glycerol stocks, streaked for single colonies on Sabouraud dextrose agar (SDA) ( (25). These media included Dulbecco's modified Eagle medium (DMEM) (Life Technologies), and Sabouraud medium in both plate (SDA) and broth (SDB) form diluted 10-fold (CIM-10) or 20-fold (CIM-20) in 50 mM morpholinepropanesulfonic acid (MOPS) (Sigma-Aldrich) (13,26). For culture on agar plates, a single loopful of cells was taken from the overnight culture and streaked for single colonies onto each medium. For broth cultures, cells were collected by centrifugation, washed once with phosphate-buffered saline (PBS) (Oxoid), and counted with a hemocytometer before 10 5 cells were inoculated into 5 ml of media in a 6-well tissue culture plate (BD Falconer). All cultures were incubated with 5% CO 2 at 37°C for 5 days. As a control, isolates were streaked for single colonies on SDA and incubated at 30°C for 5 days. DMEM broth was the most successful induction medium (Fig. S1) and was used for all subsequent analyses.
Staining and microscopy. To visualize capsule, single colonies from plates or 1 ml of culture from broth were suspended or resuspended in 150 l of PBS and counterstained with 20 l of India ink. A 15 l aliquot of this mixture was placed on a glass slide and dried for 10 min under a coverslip. Slides were then photographed using an IS10000 inverted microscope (Luminoptic) and a 40ϫ objective using ISCapture Imaging software (Tucsen Photonics). For each isolate, a minimum of 20 random fields of view were photographed using stage coordinates determined by a random number generator. Additional stains used to visualize nucleus and cell wall morphology, respectively, were (i) DAPI (Sigma-Aldrich) at 1:5,000 incubated for 30 min at room temperature and (ii) calcofluor white (Sigma-Aldrich) at 1 g/liter with one drop of 10% potassium hydroxide incubated for 2 min at room temperature.
Measurement of cell and capsule size. Total diameter (including capsule) (d t ) and yeast cell diameter (d y ) were measured for 100 cells per isolate using ImageJ software (National Institutes of Health). From these measurements, capsule thickness (t c ) was calculated as 1 2 ͑d t Ϫ d y ͒. Total volume (v t ) and yeast cell volume (v y ) were calculated using the formula for the volume of a sphere ( 4 3 r d

3
) . Cells with a d y greater than 15 m or less than 1 m were identified as giant cells or micro cells, respectively, and were noted, along with any morphologically irregular cells. These variants were excluded from all assessments of mean cell size for isolate populations. MLST analysis. Genetic variation was studied using the International Society of Human and Animal Mycology (ISHAM) consensus MLST scheme for the C. neoformans/C. gattii species complex, which uses seven unlinked genetic loci: the housekeeping genes CAP59, GPD1, LAC1, PLB1, SOD1, and URA5, and the noncoding intergenic spacer region IGS1 (51). Sequences were obtained from single nucleotide polymorphisms (SNPs) identified from whole-genome sequences, where available (52). The remaining loci were amplified independently using ISHAM-recommended primers and amplification conditions, and PCR products were commercially purified and sequenced by Macrogen Inc. (Seoul, South Korea). Sequences were edited using Geneious R6 (Biomatters Ltd.). An allele type (AT) was then assigned for each of the seven loci per isolate, and the resulting allelic profile was used to assign a sequence type (ST) according to the ISHAM consensus MLST database (http://mlst.mycologylab.org). A minimum spanning network of the concatenated sequences was generated using the TCS 1.21 software package (http:// darwin.uvigo.es/software/tcs.html) to visualize the relatedness of isolates (53).
Statistical analysis. Significant differences between species or genotypes for phenotypic or clinical data were determined using two-tailed unpaired t tests with Welch's correction. Differences in variance between species and genotypes were assessed by F test analysis. Associations between continuous phenotypic and clinical variables used Spearman rank order correlations, those between continuous and binary variables used Mann-Whitney U tests, and those between binary variables used chi-square tests, or Fisher's exact tests if any expected value was Ͻ5. Correlations were tested across all isolates, across C. neoformans isolates only, and across C. tetragattii isolates only, with all P values listed in Table S2. P values of Ͻ0.05 were considered significant. Error bars represent the means Ϯ 95% confidence intervals. Data were analyzed using Excel (Microsoft Corporation), Prism 5 (GraphPad Inc.), and SPSS Statistics (IBM) software. Correlation plots and principal-component analysis (PCA) biplots were generated in R 3.4.0 (R Core Team) using the packages corrplot for correlation plots, missMDA to impute missing values (54), and FactoMineR and factoextra to generate PCA biplots (55).