The Cryptococcus neoformans Transcriptome at the Site of Human Meningitis

ABSTRACT

Cryptococcus neoformans is the leading cause of fungal meningitis worldwide. Previous studies have characterized the cryptococcal transcriptome under various stress conditions, but a comprehensive profile of the C. neoformans transcriptome in the human host has not been attempted. Here, we extracted RNA from yeast cells taken directly from the cerebrospinal fluid (CSF) of two AIDS patients with cryptococcal meningitis prior to antifungal therapy. The patients were infected with strains of C. neoformans var. grubii of molecular type VNI and VNII. Using RNA-seq, we compared the transcriptional profiles of these strains under three environmental conditions (in vivo CSF, ex vivo CSF, and yeast extract-peptone-dextrose [YPD]). Although we identified a number of differentially expressed genes, single nucleotide variants, and novel genes that were unique to each strain, the overall expression patterns of the two strains were similar under the same environmental conditions. Specifically, yeast cells obtained directly from each patient’s CSF were more metabolically active than cells that were incubated ex vivo in CSF. Compared with growth in YPD, some genes were identified as significantly upregulated in both in vivo and ex vivo CSF, and they were associated with genes previously recognized for contributing to pathogenicity. For example, genes with known stress response functions, such as RIM101, ENA1, and CFO1, were regulated similarly in the two clinical strains. Conversely, many genes that were differentially regulated between the two strains appeared to be transporters. These findings establish a platform for further studies of how this yeast survives and produces disease.

IMPORTANCE

Cryptococcus neoformans, an environmental, opportunistic yeast, is annually responsible for an estimated million cases of meningitis and over 600,000 deaths, mostly among HIV-infected patients in sub-Saharan Africa and Asia. Using RNA-seq, we analyzed the gene expression of two strains of C. neoformans obtained from the cerebrospinal fluid (CSF) of infected patients, thus creating a comprehensive snapshot of the yeasts’ genetic responses within the human body. By comparing the gene expression of each clinical strain under three conditions (in vivo CSF, ex vivo CSF, and laboratory culture), we identified genes and pathways that were uniquely regulated by exposure to CSF and likely crucial for the survival of C. neoformans in the central nervous system. Further analyses revealed genetic diversity between the strains, providing evidence for cryptococcal evolution and strain specificity. This ability to characterize transcription in vivo enables the elucidation of specific genetic responses that promote disease production and progression.

INTRODUCTION

Cryptococcus neoformans is an environmental, encapsulated yeast and major opportunistic, neurotropic pathogen. Patients with low levels of CD4⁺ lymphocytes are particularly susceptible. In sub-Saharan Africa, the epicenter of the AIDS pandemic, C. neoformans is annually responsible for an estimated million cases of meningoencephalitis and approximately 600,000 deaths (1). Over the past 30 years, many molecular and phenotypic studies have identified a cohort of C. neoformans genes that clearly enhance but are not necessarily sufficient for virulence, such as the capsular polysaccharide, the ability to grow at 37°C, and the production of melanin, urease, phospholipase, and other factors (2). With the availability of genomic sequences, more recent studies have begun to analyze the transcriptome of C. neoformans under conditions that pertain to its pathogenicity (3–7). Using well-characterized laboratory strains of C. neoformans, reports have documented the transcriptional responses to high-temperature (8, 9), nitric oxide (9), iron (10), capsule-inducing conditions (11, 12), antifungal drugs (13), and survival within macrophages (14) and murine lungs (15). However, the transcriptional responses to stresses are dynamic and react to a variety of signals. To investigate the signals, transcription factors, and genes that enable C. neoformans to cause disease, it is critical to identify the genes that are transcribed by C. neoformans in the central nervous system (CNS). In addition, since most molecular transcriptional studies today have focused only on a few laboratory strains, such as H99, it is critical to investigate the transcriptional responses of other wild-type strains (16).

The ability to investigate the genetic responses of a pathogenic microbe within its host offers a powerful opportunity to elucidate the adaptive strategies that are essential for the microbe to survive the hostile host environment. We propose that the gene expression profiles for yeasts in the host are both site and time specific. For instance, we hypothesize that human cryptococcal meningitis involves at least six stages: (i) initiation of infection in the lungs following the inhalation of yeasts or spores; (ii) yeast survival and proliferation within the lung; (iii) dormancy of yeast cells in the host tissue; (iv) reactivation of latent infection with renewed yeast growth; (v) dissemination of the yeasts via the blood, reticuloendothelial, and lymphatic systems, bridging the blood-brain barrier; and (vi) proliferation of yeasts in brain tissue and the subarachnoid space. At these various sites and durations of infection, the transcriptional responses of C. neoformans will vary in response to the unique host environment.

With this understanding, we have taken a very focused approach to identify regulated genes, networks, and signature markers that enable C. neoformans to survive and develop disease within the subarachnoid space. Using an experimental immunocompromised rabbit model of cryptococcal meningitis and the well-studied H99 strain of C. neoformans, we previously identified genes that were highly upregulated in the subarachnoid space of the CNS, such as isocitrate lyase (ICL1). Nevertheless, additional studies determined that despite its upregulation, ICL1 was not essential for disease production at this site (17). Conversely, a gene in the trehalose pathway, the trehalose-6-phosphate synthase gene (TPS1), was also significantly upregulated in the CNS (18), and this pathway was shown to be critical for disease production (19, 20).

In this study, we examined and compared the cryptococcal transcriptome from two HIV-infected patients with high burdens of cryptococcal cells in the cerebrospinal fluid (CSF) at a single time point during infection. The patients resided in different geographical locations (Uganda and United States), and their strains represented the two global molecular types, VNI and VNII, of the dominant variety, C. neoformans var. grubii. We extracted RNA from the yeast cells taken directly from CSF and used Illumina-based RNA-seq technology to analyze their transcriptomes. These in vivo transcriptomes were compared with each other and with the transcriptomes of each strain after incubation in pooled human CSF (ex vivo) or in vitro growth in YPD broth. These conditions replicated simple exposure to CSF and late-logarithmic-phase growth in nutritionally replete medium.

RESULTS

Analysis of two clinical C. neoformans var. grubii isolates.

Isolates of C. neoformans var. grubii were obtained from two untreated patients with cryptococcal infections and AIDS. Strain G0 was obtained from a patient in Uganda participating in the COAT trial (http://clinicaltrials.gov/ct2/show/nct01075152), and strain HC1 was isolated from a patient in the United States. Both strains possess the capsular A serotype and the α mating-type allele. Eight unlinked multilocus sequence typing (MLST) loci (21) were used to determine molecular genotypes of the two strains by comparison with 30 representative strains from the global population of C. neoformans (see Table S1 in the supplemental material). Previously, we demonstrated that the global population of C. neoformans var. grubii strains can be categorized within one of three genetically isolated subpopulations or molecular types, identified as VNI, VNII, and VNB (22). Maximum parsimony analysis determined that the G0 strain has molecular type VNI, and the HC1 strain is molecular type VNII (see Fig. S1) (16, 21, 22). More specifically, strain G0 has the same genotype as the globally prevalent reference VNI strain UG2471 (see Fig. S1) and the MLST-designated genotype M3b (see Table S1) (16). The MLST genotype of strain HC1 matches that of the VNII reference strain C45 (see Fig. S1), which was previously designated genotype M7c (see Table S1) (16). These two genotypes are globally distributed: other VNI isolates with the M3b genotype have been reported from clinical cases in Belgium, Brazil, Botswana, India, South Africa, and Uganda, and isolates of VNII with the M7c genotype have been found in South Africa and the United States (16, 21, 22).

Six cDNA libraries were created for high-throughput Illumina sequencing. Two libraries were made from yeast cells that were directly isolated from the two patients. The remaining four libraries were made from cultures obtained by inoculating the cryptococcal isolates in either YPD or ex vivo CSF (see Materials and Methods). In total, more than 350 million reads were generated, representing on average more than 100× coverage of the C. neoformans genome length for each sample (Table 1).

TABLE 1 .

Number of RNA-seq reads obtained for each sample and mapping results

Characteristic
	G0			HC1
	YPD	Ex vivo CSF	In vivo CSF	YPD	Ex vivo CSF	In vivo CSF
Sequencing type	Single end	Paired end	Single end	Single end	Paired end	Single end
Read length	50	100	50	36	100	36
Total no. of reads	96,331,255	46,748,370	74,481,942	41,283,763	52,103,334	41,270,151
No. of mapped reads	92,513,698	41,423,212	70,860,588	39,989,804	36,610,869	35,340,075
Mapping (%)	96.04	88.61	95.13	96.87	70.27	85.63

We aligned all the sequence reads to the C. neoformans var. grubii H99 genome using Bowtie (23) and TopHat (24). For most samples, more than 85% of the reads mapped to the genome, and there were no major differences among the RNA obtained from the patients’ yeast cells (in vivo CSF) and the in vitro cultures incubated in CSF or YPD (Table 1). We measured the expression levels in fragments per kilobase of exon model per million mapped reads (FPKM) (25), and the gene expression level was defined as the sum of the FPKM values of its isoforms. Due to the high sensitivity of the RNA-seq technique, almost the entire gene set of the C. neoformans genome was detected as expressed (FPKM ≥ 1) under all three conditions. Of the 6,976 genes predicted by the H99 genome, 6,778 (97.16%), 6,647 (95.28%), and 6,809 (97.60%) genes were expressed in G0 under YPD, ex vivo CSF, and in vivo conditions, respectively, and in HC1, 6,849 (98.18%), 6,620 (94.89%), and 6,778 (97.16%) genes were expressed under the same three conditions. A total of 6,363 (91.21%) genes were expressed in both strains under all conditions. In contrast, 207 or 190 genes were exclusively expressed in either G0 or HC1, respectively, under one or more of the three conditions.

Analysis and evaluation of differential gene expression.

In order to compare gene expression levels among different environmental conditions, we generated the whole gene expression profile using all the RNA-seq data by HT-Seq count (http://www-huber.embl.de/users/anders/HTSeq) and normalized using the DESeq package (26) in R (27). Hierarchical clustering indicated that the expression profiles of the two strains growing under the same conditions were more similar to each other than the patterns of transcription of each strain under different conditions (Fig. 1A). This similarity is notable because the duration of CNS infection with each strain was unknown but likely differed. In addition, strains G0 and HC1 represent divergent molecular types of C. neoformans var. grubii, VNI and VNII, respectively (22, 28, 29). These observations demonstrate the possibility that in vivo transcriptomes of additional strains might reveal more conserved in vivo genetic signatures.

FIG 1 .

(A) Hierarchical cluster analysis of gene expression based on log ratio of normalized read count per gene. The cluster displays expression patterns for six RNA-seq samples. The color key represents log₂-normalized transformed counts. (B) Number of genes that are significantly up/downregulated (FDR < 0.2) between two different conditions in both strains. Pairwise comparisons were made between three conditions: ex vivo CSF versus YPD, in vivo CSF versus YPD, and in vivo CSF versus ex vivo CSF. The numerical designation indicates the number of genes that were up/downregulated in the first condition compared to that in the second condition. See Table S2 in the supplemental material for a list of these genes.

Although serial in vivo CSF specimens were not available for our study, the similarity of gene expression patterns of the two strains under different conditions allowed us to regard the two strains under the same conditions as biological replicates. This approach allowed us to identify common transcriptional responses that are relevant for both genotypes. A modified Fisher’s exact test with data fit to a negative binomial distribution of the DESeq package (26) was used to identify the differentially expressed (DE) genes. We made pairwise comparisons between different conditions, and the numbers of DE genes are as follows (Fig. 1B): ex vivo CSF versus YPD, 129 (see Table S2A); in vivo versus YPD, 45 (see Table S2B); in vivo versus ex vivo CSF, 256 (see Table S2C). These comparisons indicate that the transcriptional profiles of the in vivo CSF and YPD samples were actually more similar to each other than to ex vivo CSF samples. These results are also consistent with the hierarchical clustering of the transcriptional profiles in the dendrogram display (Fig. 1A). This finding might be explained by the potentially more active yeast cell growth in YPD and the host than is present during ex vivo CSF exposure. Gene ontology (GO) analysis was used to identify functional categories overrepresented in the DE genes. Compared to yeast cells that were incubated ex vivo in CSF, yeasts from in vivo CSF upregulated the expression of genes that were enriched in GO terms that related to cell metabolism, such as cellular biosynthetic processes (GO:0044249, P = 1e−12), gene expression (GO:0010467, P = 5.4e−19), and structural constituents of ribosome (GO:0003735, P < 1e−30). The increased expression of ribosomal protein genes indicated that the yeast cells were more biosynthetically active within the human brain and subarachnoid space, where they were confronted by host inflammatory cells and processes. Compared to nutrient-rich YPD, human (in vivo and ex vivo) CSF is a nutritionally depleted environment. We hypothesized that the genes upregulated in the CSF (in vivo or ex vivo) might be more related to biological survival and/or fitness of C. neoformans in the human subarachnoid space. Indeed, 20 genes were identified to be significantly upregulated in in vivo and ex vivo CSF conditions compared with that in YPD (Fig. 2A; see also Table S3A in the supplemental material). As predicted, several of these identified genes have been reported as putative virulence or fitness genes in C. neoformans, such as CFO1 (30), ENA1 (31, 32), and RIM101 (33). However, six of the 20 genes had no functional annotation; two of them (CNAG_00456 and CNAG_05159) were identified as the putative target genes of Gat201 (3), which is known as a regulator of virulence (34).

FIG 2 .

Heatmaps showing the expression data of 20 genes that were significantly upregulated in ex vivo CSF and in vivo CSF compared to YPD in both strains (A) and the expression data of six genes that were significantly upregulated in in vivo CSF compared to in ex vivo CSF and YPD in both strains (B). (C) Hierarchical cluster analysis of the 100 most divergently expressed genes between G0 and HC1.

To analyze the specificity of the transcriptional responses of C. neoformans in the human body, six genes were significantly upregulated in the in vivo CSF compared to the other two conditions (Fig. 2B; see Table S3B in the supplemental material). Among these genes, the sulfiredoxin gene (SRX1) has been reported to have a critical role in the resistance of yeasts and higher eukaryotes to oxidative stress (35, 36). In addition, the high expression of SRX1 might be critical for survival of the yeasts in the presence of monocytes/macrophages in CSF or microglial cells in brain parenchyma. Furthermore, another upregulated gene, SIT1, has been reported to be essential for growth, melanin formation, and cell wall density of C. neoformans under low-iron conditions (37) and for invasion of epithelial cells by Candida albicans (38). The expression of SIT1 as well as CFO1 indicates the potential importance of iron at this site.

To verify the potential relevance of genes that were shown by RNA-seq to be upregulated in CSF, four of the DE genes were selected for quantification of their expression by real-time reverse transcription-PCR (RT-PCR). Due to the low quantity of in vivo RNA sample of the G0 strain, cDNAs of five different samples were used as the templates. Transcription of ACT1 was used to confirm and normalize the concentration of mRNA among different samples. The comparisons between RNA-seq and RT-PCR of these genes are shown in Fig. S2 in the supplemental material. Although some differences were shown in the HC1 YPD sample, most of the data were consistent between the two.

Strain-specific differentially expressed genes.

As strains G0 and HC1 belonged to different genetic MLST subpopulations (see Table S1 and Fig. S1 in the supplemental material), we investigated their divergently expressed genes. A dissimilarity score was used to estimate the diversity of gene expression between the two strains. The 100 most divergently expressed genes were identified based on this dissimilarity score (Fig. 2C; see also Table S4). Based on the expression patterns, these genes can be organized into two groups: 69 genes (group 1) are expressed significantly more in G0 than in HC1 under all conditions, and 31 genes (group 2) are expressed significantly more in HC1 than in G0 under all conditions. Gene ontology analysis revealed that these 100 most divergently expressed genes evinced an enrichment for transporters (P = 1.54e−4).

Substantial genomic variation exists among G0, HC1, and H99.

RNA-seq technology was developed primarily to analyze global gene expressions. However, the high coverage and good quality of the data provided us with an efficient way to assess the genetic diversity among clinical strains. Compared to the standard H99 clinical isolate from the United States, 50,155 single-nucleotide variants (SNVs) and 156,880 SNVs were identified in G0 and HC1, respectively (see Fig. S3 and Table S5 in the supplemental material). Strains H99 and G0 are prevalent VNI molecular types with respective MLST genotypes of M1b and M3b, and HC1 is a global VNII strain with the M7c genotype (see Table S1). In G0, 21,059 SNVs were dispersed in 5,185 genes, and in HC1, 104,027 SNVs were dispersed in 6,728 genes. The SNVs in the exon regions were further classified according to the open reading frame (ORF). In G0, 19,997 SNVs were detected in coding regions, of which 8,626 were nonsynonymous. In HC1, 99,542 SNVs were detected in coding regions, and 35,674 were nonsynonymous. The ORF regions contained 70 SNVs and 130 SNVs in G0 and HC1, respectively, which can cause changes between amino acids and stop codons. In these SNVs, 32 SNVs in G0 (see Table S6A) and 45 SNVs in HC1 were regarded as readthrough SNVs (see Table S6B), and the others were regarded as nonsense SNVs (see Table S6C and D). Due to changes in the stop codon, SNVs may affect protein translation and/or stability and cause phenotypic changes between the strains (39, 40). For example, one of the readthrough mutations has specifically occurred in G0 OGG1 (CNAG_03795), which encodes a DNA glycosylase that is putatively involved in the repair of oxidative DNA damage (41). Therefore, in a phenotypic screen, we tested whether this gene variation had a phenotypic consequence on the three yeast strains. A higher level of resistance to H₂O₂ was observed in G0 than in H99 and HC1 (see Fig. S4). Although further studies will be required to determine whether this phenotypic change was specifically caused by this single nucleotide mutation, this observation underscores the ability to identify areas of potential genomic differences that translate into specific and sometimes subtle differences in cryptococcal strain phenotypes.

Gene gain and loss during evolution may also be critical for the functional differentiation between species and strains (42, 43). Therefore, to identify novel genes in both strains compared with a reference genome, we performed de novo assembly of the RNA-seq data using Trinity (44). A total of 18,260 and 24,664 contigs were assembled for G0 and HC1, respectively, which belonged to 8,996 and 10,268 unigenes (the nonredundant set of the contigs). We first aligned all the contigs against H99 genome and mitochondrial DNA sequences, but 53 unigenes in G0 and 171 unigenes of the more divergent HC1 could not be aligned to them. We then used BLAST to identify the homologs of these unaligned unigenes in the NCBI nonredundant nucleotide (nt) and nonredundant protein (nr) databases. Forty-four unigenes in G0 and 118 unigenes in HC1 were identified to homologous genes in the four fully sequenced Cryptococcus genomes (Fig. 3A and B). To investigate the putative functions of these novel genes, we used BLAST2GO (45) to annotate them. Nineteen genes in G0 and 42 genes in HC1 were assigned corresponding gene ontology terms, and they were enriched in functions related to transport, localization, and membrane constitution in both G0 and HC1 compared with the standard whole genome composition of H99 (Fig. 3C).

FIG 3 .

The proportions of the novel genes of G0 (A) and HC1 (B) that mapped to different species. (C) GO classification of novel genes of G0 and HC1 compared to those of H99. Percentages of genes annotated with certain GO terms compared to total number of annotated genes are shown in the figure. A hypergeometric test was used to test the enrichment of G0 and HC1 novel genes compared with H99 gene composition (*, P < 0.05; **, P < 0.01).

DISCUSSION

In this study, we generated genome-wide transcriptional profiles of C. neoformans var. grubii from two untreated AIDS patients. To our knowledge, this is the first report of RNA-seq data generated from yeast cells taken directly from human CSF. When we aligned the sequencing data to the reference genome of strain H99, the in vivo data from strain G0 received an extremely high ratio of hits (>95%), which indicated that RNA recovered from the in vivo samples was of high quality with very limited contamination. Analysis of in vivo transcription is especially challenging, because each in vivo sample is unique and cannot be duplicated, which is not ideal for applying classical strategies for quantitative analysis of transcriptional profiles (46, 47). To address this situation, we treated the samples obtained from different patients as replicates in the analyses. The results demonstrated good reproducibility of the gene expression patterns between the two different clinical strains. Using this approach, we were able to compare and identify genes whose expression is associated with in vivo survival and growth. This strategy can be used in future analyses of in vivo samples.

To identify genes whose expression is associated with in vivo growth, we assessed transcriptional profiles of the two strains under two defined in vitro conditions, YPD and ex vivo CSF, and compared them with expression in the in vivo environment. We hypothesized that because of low nutrients in human CSF, the in vivo gene expression profile would be more similar to that observed in ex vivo CSF than YPD, which is nutritionally replete. This hypothesis was supported by our previous data that the ability of C. neoformans mutants to survive in vitro in human CSF was closely correlated with their production of disease in a rabbit meningitis model (48). Contrary to our expectations, in this study we observed that the in vivo gene expression profiles of C. neoformans strains in human CSF were generally more similar to those in YPD than to ex vivo CSF (Fig. 1). Gene ontology enrichment analysis revealed that the downregulated genes in ex vivo CSF samples compared to those in in vivo human samples were significantly enriched in metabolic and cellular processes. These results suggest that the yeast cells are inhibited in their basic metabolic machinery in the ex vivo CSF; however, inside the host (in vivo CSF), yeast cells maintain an active metabolic gene network and actively proliferate. The results are also consistent with the observations that during human disease, the number of yeast CFUs can increase to more than 1 million CFU/ml in an HIV-infected patient (49), and in the immunosuppressed rabbit model, abundant yeast growth occurs in the subarachnoid space (50). There are several nonexclusive explanations of why C. neoformans growth measured by viable quantitative yeast counts is inhibited in ex vivo CSF but active in vivo. First, in humans, CSF is constantly being renewed, and therefore in vivo CSF has a much higher concentration of nutrients than ex vivo CSF. Second, ex vivo CSF has a much more alkaline pH compared to that of in vivo CSF, which can affect transcription. Third, C. neoformans cells colonize human CNS after going through several stages of infection, which may extend for months or even years. This slow progression through the infection process can cause genetic and epigenetic adaptations and subsequent changes in the gene expression patterns, which are detectible in yeast cells obtained directly from humans but lost in vitro because the yeast cells exposed to ex vivo CSF were propagated in culture prior to CSF exposure.

Although the growth state of yeast cells in ex vivo CSF and human subarachnoid space (in vivo) may not be the same, there are stresses from this specific fluid environment that are similar and unique. Therefore, the 20 genes that we identified as significantly upregulated in the two CSF conditions compared to YPD might specifically represent the response of C. neoformans to certain CSF stresses or signals. Based on the annotation information in FungiDB (51), most of these genes are categorized into three groups: catalytic activity-related genes (6/20), transporters (5/20), and genes of unknown function (6/20). Two of these genes have been reported as necessary survival genes for C. neoformans. We previously demonstrated that the ATPase transporter gene ENA1 is essential for survival of this yeast in CSF (48). Jung et al. showed that the ferroxidase, CFO1, is required for the utilization of the biological protein, transferrin, which is an important iron source for C. neoformans during infection (30). We did not observe any evidence for differential regulation of any of the classical virulence factors (capsule, melanin, high-temperature growth, urease, phospholipase) at this advanced clinical state of infection, most likely because these classical virulence factors are expressed during the early stages of infection in the lungs and/or during dissemination. Conversely, our data suggest that basic metabolic pathways and stress response genes are essential for survival and successful propagation within the subarachnoid space.

As we investigate in vivo gene expression data between humans and animal models and compare those with the data obtained for other fungal infections, it is possible to recognize conserved patterns of gene regulation. For example, the isocitrate lyase gene (ICL1) and the glyoxylate pathway are upregulated in both human and experimental animal models of cryptococcal meningitis. In addition, recent work by Cheng et al. demonstrated the upregulation of RIM101 in peritoneal Candida infections in mice (52), which is similar to our observation of CNS infection with Cryptococcus in humans. With the analyses of further cases, we expect to find common conserved pathways that are characteristic for development of the disease.

To investigate the six differentially regulated genes with unknown functions, we searched the annotated fungal genomes available in FungiDB and found that four of them (CNAG_02118, CNAG_04837, CNAG_05632, and CNAG_06493) are restricted to Cryptococcus lineage. Such lineage-specific genes are often important for species evolution. For example, it has been demonstrated that in primates, lineage-specific genes are essential for human brain evolution (53). We hypothesize that in Cryptococcus, lineage-specific genes may be important for adaptation of Cryptococcus in its survival within certain human body sites. More studies are required to test this hypothesis, and we have now begun to identify the genes of interest.

Transporters are groups of genes that play important roles in yeast biology. In our study, transporter genes were frequently identified in many different analyses. Based on our results, we divided the transporter genes into two groups. One group includes the transporters that are conserved among different strains or even distantly related species, and they are essential for Cryptococcus. Several of these transporters have been identified and studied previously (54–57). Three genes from this group of transporters with previously defined function, CFO1, ENA1, and SIT1, were significantly upregulated in both in vivo and ex vivo CSF compared to in YPD. These genes are important for virulence, drug resistance, starvation response, intracellular survival, and other basic functions in Cryptococcus (30, 31, 37, 48, 58). The second group of transporters includes rapidly evolving and variable genes that may not be particularly essential for the yeast’s pathobiology. These transporters have different expression patterns or even different genetic compositions among genetically related strains. In our study, these transporter genes were enriched among the most divergently expressed and novel genes. Most of these genes have been annotated as carbohydrate transporters, such as a sugar transporter, monosaccharide transporter, galactose transporter, and hexose transporter. Although the exact function of these transporters is unknown, they might be involved in the biosynthesis of the polysaccharide capsule of Cryptococcus, which is one of its main virulence factors (59). In addition to carbohydrate transporters, we also observed several novel genes that belonged to the major facilitator superfamily (MFS) transporters and ATP-binding cassette (ABC) transporters, which have been shown to play a role in multidrug resistance of fungi (60). We hypothesize that this group of transporters may contribute to variable strain-specific properties and even explain the differences in disease manifestations and treatment outcomes between strains. Indeed, when the two strains were compared for the most divergently expressed genes, transporter genes were commonly identified, and these findings support the hypothesis that the microevolution of these strain-specific properties frequently involves transporter functions. However, more studies are necessary to evaluate the potential role of the second group of transporters in the pathogenesis of C. neoformans and how genetic changes in these transporters influence a strain’s unique phenotype.

This first analysis of cryptococcal transcriptomes during infection in the CNS has demonstrated a series of important principles. First, the transcriptomes of cryptococcal cells from the human host can be captured and analyzed. Second, despite genetic differences between strains and the duration of their infection, the transcriptomes from these strains are remarkably similar and suggest that a specific consensus pattern of gene expression may be associated with CNS infection. Third, C. neoformans is metabolically active in the human CSF. Fourth, specific genes with known virulence or survival properties are identified using in vivo transcription profiling, which validates this method. Moreover, genes and pathways essential for survival of yeasts in the human body are not limited to classical virulence pathways or phenotypes. Fifth, through the combination of RNA-seq and genome analysis, identification of SNVs and other genetic diversities may predict differences in gene expression and/or function in individual strains and identify the potential impact of these mutations on yeast microevolution. Additional strategies are needed to determine the functions of nonconserved genes that lack homology to annotated sequences, as they may be essential for the pathogenicity of C. neoformans. Comparative analyses of transcriptomes may identify the potential importance of these unrecognized genes. This study is just the beginning, but it illustrates the potential to characterize fungal transcriptomes and then relate them to the progressive states of disease. They may provide insights into how a strain(s) produces disease and how some strains might produce a genetic signature that could even predict their infection outcome or response to therapy.

MATERIALS AND METHODS

DNA manipulations and phylogenetic analyses.

Genomic DNA was isolated using a MasterPure yeast DNA purification kit (Epicentre Biotechnologies, Madison, WI). As previously described, mating types were determined by a PCR-based method using primers for the a or α allele of the STE20 gene for the strains (61). For MLST, eight previously described loci (CAP59, GPD1, IGS1, LAC1, PLB1, SOD1, URA5, and TEF1) were used in this study for PCR amplification (22, 62). Amplicons were sequenced with Sanger sequencing, and the reads were edited manually. To identify the genotypes of G0 and HC1, their MLST sequences were compared with a global set of reference strains of C. neoformans (see Table S1). The sequences of all the loci were aligned by ClustalW (63). The maximum parsimony (MP) tree was identified with heuristic searches based on 500 random sequence additions using PAUP (64) and was rooted using C. neoformans strain JEC21 (serotype D).

RNA-seq sample preparation.

Yeast cells (approximately 10⁶ to 10⁸ CFUs) were pelleted in CSF in a microfuge tube containing 20 to 30 µl of 1-mm glass beads, and the pellet was stored at −80°C. The frozen pellet was lyophilized and vortexed to powder, and RNA was extracted using a modified Trizol/Qiagen procedure as follows: yeast cells were lysed in 700 µl Trizol and incubated for 5 min at room temperature. After 140 µl of chloroform was added, the tube was shaken for 20 s and incubated at room temperature for 3 min. The sample was centrifuged at 10,000 rpm for 15 min at room temperature. The aqueous phase was separated and mixed with equal volumes of 80% ethanol and immediately applied to a Qiagen RNeasy minicolumn (Qiagen; catalog number 74014) and centrifuged at 13,000 rpm for 1 min, and RNA was isolated according to the manufacturer’s protocol. The yield of total RNA ranged between 1 and 25 mg/ml.

The CSF samples were part of the Duke IRB-approved database and Specimen Repository for Infectious Disease Related Studies (PR0005314) in which patients are deidentified and clinical information is limited. The HC1 strain came from a patient in the United States, and strain G0 was from a Uganda patient. YPD broth (1% yeast extract, 1% Bacto peptone, 2% dextrose) and sterile human CSF (pool of 10 to 20 individuals) were prepared as previously described (48) and used for in vitro or ex vivo incubation of the strains. The strains were grown in YPD broth for 16 h at 37°C and then harvested. Yeast cells in stationary phase after culture in YPD overnight at 37°C were exposed to human CSF for 9 h, during which the CSF was replenished every 3 h, and the cells were then harvested. All harvested cells were snap-frozen and lyophilized for total RNA isolation.

RNA sequencing and quantitative analysis.

Total RNAs from the two strains (HC1 and G0) under the various conditions were extracted using Trizol (Invitrogen) according to the manufacturer’s instructions. The mRNA samples for RNA-seq analysis were performed using a TruSeq RNA sample preparation kit (Illumina, San Diego, CA). The cDNA libraries were sequenced on the Illumina GAII and the Illumina HiSeq 2000 (Illumina, San Diego, CA) instruments. The C. neoformans var. grubii H99 genome with annotations (2012 release) were downloaded from the Broad Institute (http://www.broadinstitute.org) and used as a reference. Sequencing reads of each sample were mapped to a reference genome using TopHat 2.0.0 (24). Subsequently, we used the HT-Seq count (http://www-huber.embl.de/users/anders/HTSeq) to convert the mapped reads to read counts per gene. In the quantitative analysis, the two isolates exposed to the same conditions were treated as biological duplicates. We evaluated the expression differences using a test based on a negative binomial distribution, which was implemented in the R package DESeq (26). Three comparisons were made among the different conditions: (i) in vivo human versus YPD; (ii) in vivo CSF versus ex vivo CSF exposed; (iii) ex vivo CSF exposed versus YPD. The false discovery rate (FDR) was calculated by the “p.adjust” function in R using the Benjamini and Hochberg (65) method and controlled at 20%. All selected FDR-adjusted P values corresponded to raw P values below 0.01. The number of fragments per kilobase of exon model per million mapped reads (FPKM) was calculated according to Cufflinks (25). A dissimilarity score was defined as follows:

$$\text{dissimilarity score}=1-\sum _i\frac{2P_i\left(\text{A}\right)P_i\left(\text{B}\right)}{P_i^2\left(\text{A}\right)+P_i^2\left(\text{B}\right)}$$

where P_i(A) and P_i(B) represent the normalized expression value of gene i in strain A and strain B under the same condition.

Quantitative real-time RT-PCR validation of RNA-seq data.

Total RNA of each sample was treated with DNase (Turbo DNA-free kit; Ambion) to avoid genomic DNA contamination. Reverse transcription was performed using the RETROscript kit (Ambion). Four genes were selected for validation of RNA-seq results using RT-PCR. Primer pairs were designed to span exon-exon junctions using Primer3, and ACT1 was used as an internal control. All the primer sequences are as follows (5′ to 3′): ACT1, CCACACTGTCCCCATTTACGA (forward) and CAGCAAGATCGATACGGAGGAT (reverse); CNAG_05431, AAGCCCCCTGAGAGACCTAG (forward) and GGAGCTCCAACAGACTCGA (reverse); CNAG_06493, AACAGACGGCATCGAAGGTT (forward) and GTTACAGAATGCTGCGCTCG (reverse); CNAG_00654, TGTCAGAACCAGCATGAGGC (forward) and CCGCCCTACCTTCCGTAATC (reverse); CNAG_00815, ACCTGGCATCGATGGGTTTT (forward) and TGGAGGTTGCGGTCACAATA (reverse). The amplifications were conducted in a total volume of 20 µl, containing 1× SYBR green (iTaq universal SYBR green supermix; Bio-Rad), 300 nmol/liter of both primers, and 1 µl of diluted cDNA. The amplification was conducted as follows: 5 min at 95°C, followed by 35 cycles consisting of 30 s at 95°C, 30 s at 54°C, and 30 s at 72°C. Finally, melting curve analysis was performed from 60°C to 95°C, with increments of 0.5°C per 10 s. Amplification, melting curve analysis, and detection were conducted with the MyiQ single-color, real-time PCR detection system (Bio-Rad).

RNA-seq variant calling.

BAM files that were generated by TopHat were used as inputs. The genome analysis toolkit (GATK version 2.4.9) (66) was used to perform variant calling. To solve the incompatible scores between TopHat and GATK, all the mapping scores with 255 in TopHat were reassigned to 60 in GATK. The filter used in GATK was “-stand_callconf 50 -stand_emitconf 10 -dcov 500.” All the SNVs were annotated using VCFannotator (http://vcfannotator.sourceforge.net) relative to the H99 genome.

Detection and characterization of novel genes.

To obtain high-quality assemblies for the transcriptomes, we used only 100-bp paired-end reads for de novo assembly. Reads that contained more than 10% of the bases with phred quality scores below 20 were removed using FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit). De novo assembly of the filtered reads was performed by Trinity (44). To detect the unigenes that either do not exist or were highly divergent from the reference genome, we aligned all the contigs to the C. neoformans var. grubii H99 genome and the mitochondrial DNA sequence using BLAT (67). A novel gene was defined as one for which all the contigs could not be aligned to the reference sequences. All the unigenes that met this criterion were aligned to the NCBI nonredundant nucleotide (nt) and protein database (nr) using BLAST. All the BLAST hits with E values less than 1e−5 were kept for further analysis.

RNA-seq data accession number.

RNA-seq data have been made publicly available at GEO (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE51573.

SUPPLEMENTAL MATERIAL

Table S1

Strains of C. neoformans used in phylogenetic analysis in Fig. S1 in the supplemental material (16, 22).

Table S2

List of genes significantly up/downregulated between ex vivo CSF and YPD (A), in vivo CSF and YPD (B), and in vivo CSF and ex vivo CSF (C).

Table S3

List of 20 genes significantly upregulated in ex vivo CSF and in vivo CSF compared to YPD in both strains (A) and six genes significantly upregulated in in vivo CSF compared to ex vivo CSF and YPD in both strains (B).

Table S4

List of 100 most divergently expressed genes between G0 and HC1.

Table S5

SNVs of strains G0 and HC1 strains compared to strain H99.

Table S6

List of readthrough mutation genes in strains G0 (A) and HC1 (B) compared to strain H99, and nonsense mutation genes in strains G0 (C) and HC1 (D) compared to strain H99.

Figure S1

Genetic relationships among strains of C. neoformans based on eight MLST markers visualized by the maximum parsimony dendrogram. Four molecular types (VNI, VNII, VNB, and VNIV) are indicated. Numbers on each branch indicate the bootstrap values based on 500 replications. Download

Figure S2

Correlation between quantitative real-time reverse transcription (RT)-PCR and RNA-seq. The expression values were normalized by the expression level of ACT1 and then transformed by log₂. Download

Figure S3

Extensive SNVs are observed in G0 and HC1 compared with those in H99. Concentric rings depict the total numbers of SNVs in each 10-kb nonoverlapping sliding window on the reference genome. The outside ring displays the SNVs of G0, and the inside ring displays the SNVs of HC1. The range of axes for each ring is from 0 to 150. This figure was created using the Circos program (www.circos.ca). Many more SNVs were identified in HC1 than in G0. Download

Figure S4

Strains HC1, G0, and H99 differ in their rates of growth at 37°C and after exposure to hydrogen peroxide. Growth assays at 30°C and human body temperature of 37°C on YPD and YPD with 4 Mm H₂O₂. HC1 was more sensitive to H₂O₂ and exhibited less growth at 37°C. G0 displays similar growth to H99 at 30°C and 37°C, but it was more resistant to H₂O₂ at 37°C. Download