Multilocus Microsatellite Genotyping Array for Investigation of Genetic Epidemiology of Pneumocystis jirovecii

Abstract

Pneumocystis jirovecii is a symbiotic respiratory fungus that causes pneumonia (PcP) in immunosuppressed patients. Because P. jirovecii cannot be reliably cultured in vitro, it has proven difficult to study and gaps in our understanding of the organism persist. The release of a draft genome for the organism opens the door for the development of new genotyping approaches for studying its molecular epidemiology and global population structure. We identified and validated 8 putatively neutral microsatellite markers and 1 microsatellite marker linked to the dihydropteroate synthase gene (dhps), the enzymatic target of sulfa drugs used for PcP prevention and treatment. Using these tools, we analyzed P. jirovecii isolates from HIV-infected patients from three geographically distant populations: Uganda, the United States, and Spain. Among the 8 neutral markers, we observed high levels of allelic heterozygosity (average H_e, 0.586 to 0.842). Consistent with past reports, we observed limited global population structuring, with only the Ugandan isolates showing minor differentiation from the other two populations. In Ugandan isolates that harbored mutations in dhps, the microsatellite locus linked to dhps demonstrated a depressed H_e, consistent with positive directional selection for sulfa resistance mutations. Using a subset of these microsatellites, analyses of individual and paired samples from infections in San Francisco, CA, showed reliable typeability within a single infection and high discriminatory power between infections. These features suggest that this novel microsatellite typing approach will be an effective tool for molecular-epidemiological investigations into P. jirovecii population structure, transmission, and drug resistance.

INTRODUCTION

Pneumocystis jirovecii is a symbiotic respiratory fungus that causes pneumonia (PcP) in immunocompromised patients, including those with AIDS. In the United States, the incidence of PcP in HIV-infected populations has significantly declined owing to PcP prophylaxis and initiation of combined antiretroviral therapy (cART) for HIV infection (1–3). However, PcP remains a leading cause of death among those who do not receive cART or PcP prophylaxis (4, 5). Among patients with HIV infection, mortality rates for PcP remain high even in the cART era, ranging from 10% to nearly 40% (6–8). Concerningly, trends have suggested that PcP is an increasingly important disease among HIV-infected patients in low-income countries, where the majority of persons with HIV infection reside (9). Despite advances in preventing PcP, Pneumocystis is still cause for ample concern because (i) prophylaxis and treatment with sulfa drugs may be selecting for resistance mutations in key P. jirovecii metabolic enzymes, including dihydrofolate reductase (encoded by dhfr) and dihydropteroate synthase (encoded by dhps) (10–12); (ii) PcP incidence is increasing in some parts of the world (13); and (iii) PcP outbreaks periodically occur in immunocompromised patient populations, causing significant morbidity and mortality (13–20).

Because Pneumocystis cannot be reliably cultured in vitro (21), fundamental questions about P. jirovecii biology, ecology, and epidemiology remain unanswered. For example, we do not possess a clear understanding of the Pneumocystis life cycle, including sexual and clonal reproduction, transmission dynamics, the propensity for Pneumocystis strains to persist in unaffected, immunocompetent carriers, or the true risk of putatively sulfa drug-resistant Pneumocystis strains (reviewed in reference 22). Molecular epidemiology studies involving strain typing can help elucidate many of these issues and, in particular, population structure, the evolution of drug resistance, and transmission dynamics.

The current P. jirovecii strain-typing options have revealed much about Pneumocystis population structure and outbreak dynamics (23, 24). These typing methods include single- and multilocus genotyping using karyotypes (25), multilocus enzyme electrophoresis (25), Sanger sequence analysis (25, 26), type-specific oligonucleotide hybridization (25), single-strand conformation polymorphism (25), and multiplex-PCR/single-base extension (MPCR/SBE) (27, 28). More recently, a four-locus scheme has emerged, which includes internal transcribed spacer 1 (ITS1), 26S, mt26S, and beta-tubulin (β-TUB) (29, 30). Though each of these approaches has contributed significantly to our understanding of Pneumocystis biology, many questions remain unresolved (27, 31). A genotyping approach with neutrally evolving markers, a high discriminatory index, and a stepwise model of allele adaptation would be a useful addition for studying P. jirovecii.

The publication of the draft P. jirovecii genome (32) enables the development of such an approach. Using this information, we have identified and validated assays for 8 putatively neutral microsatellites in order to develop a more robust multilocus strain typing tool. Additionally, we have identified 1 microsatellite linked to dhps, a gene in which mutations are thought to confer reduced drug susceptibility to sulfa medications. Microsatellites are short tandem repeats in coding and noncoding regions of the genome that vary in length between strains and offer reproducible genotype calling for individual strains (typeability) and high resolution to distinguish between strains (discriminatory power) (33). This type of marker has been used extensively and has proven to be robust in humans and in pathogens such as Plasmodium falciparum for “molecular fingerprinting” and for studying the evolution of drug resistance (34–36). Here, we describe our multilocus microsatellite genotyping method and use it to investigate the global population structure of P. jirovecii, to study the evolution of sulfa antibiotic resistance, and to evaluate typeability and discriminatory power within one cohort.

MATERIALS AND METHODS

Sample collection and ethics statement.

Molecular analyses were performed on deidentified clinical specimens from Uganda (2008 to 2009, from Mulago Hospital, Makerere University), the United States (2005 to 2011, from San Francisco General Hospital, University of California, San Francisco), and Spain (Barcelona, 2001 to 2004). Patients from Uganda and San Francisco were enrolled on the basis of HIV infection and suspected PcP, as described before (12). Patients from Spain were enrolled on the basis of slide-positive microscopy, as described before (37). Among the samples from Spain, 47% of cases were late presenters and 94% of cases had not received previous sulfa or sulfone prophylaxis. Sample collection and molecular analysis of isolates from each site were approved by the appropriate Institutional Review Boards (IRBs) as described previously (12, 37, 38). As part of these prior studies, all specimens had been genotyped for dihydropteroate synthase gene (dhps) mutations (12, 37, 39). Molecular analyses detailed in this study were approved by the University of North Carolina at Chapel Hill IRB (study no. 12-1783).

Identifying and validating microsatellites.

In order to identify microsatellites, we scanned the published P. jirovecii genome (32) using Tandem Repeat Finder (40), which identified approximately 150 di- or trinucleotide tandem repeats with a repeat size of ≥8. For these, we were able to design primers to amplify 50 tandem repeats using WebSat (41, 42). PCRs were optimized on five total isolates from Kampala, Uganda, San Francisco, CA, and Chapel Hill, NC, and amplicons were sequenced using ABI BigDye Terminator Chemistry (Applied Biosystems, Grand Island, NY). Chromatograms were analyzed in Sequencher v5.0 (Gene Codes Corp., Ann Arbor, MI) to determine repeat unit counts, and sequence data were deposited in GenBank (see below). Loci that showed evidence of ≥1 repeat-unit variation in pairwise comparisons among sequenced isolates were deemed true microsatellites.

We assessed the functional genic context (i.e., intergenic, intragenic, exonic, and intronic) of each true microsatellite using the annotations at the European Nucleotide Archive (accession no. PRJEA68827), which were created using Maker v2.10 (32, 43). This annotation pipeline utilized de novo, homology, protein, and transcriptomic evidence (from 5 × 10⁷ RNA-Seq reads) to predict the genic features. Physical linkage with the dhps locus was determined using BLAST alignment (44) of published P. jirovecii genome contigs against P. murina supercontigs (Liang Ma, personal communication, December, 2012; genome downloaded from the Broad Institute [http://www.broadinstitute.org/annotation/genome/Pneumocystis_group.2/MultiHome.html]).

Amplification and fragment sizing.

Fluorescently tagged forward primers for each variable microsatellite locus were designed (Applied Biosystems, Foster City, CA). Reverse primers carried a 5′ CTGTCTT “pigtail” to promote full adenylation and reduce stutter peaks (Table 1) (45). PCR was performed using HotStarTaq (Qiagen, Valencia, CA) according to the following generalized cycling parameters: 95°C for 15 min; 40 cycles of 95°C for 45 s, primer-specific annealing temperature (T_a) for 45 s, 72°C extension temperature for 60 s; 72°C extension temperature for 10 min; 12°C hold. To increase throughput, thermocycling was performed on Bio-Rad T100 (Bio-Rad, Hercules, CA) or ABI 2720 (Applied Biosystems, Foster City, CA) machines. For primer-specific sequences and T_a, see Table 1.

TABLE 1.

Primers and primer annealing temperatures for the P. jirovecii microsatellite loci

Primer name	Primer sequence (5′→3′)		Ta (°C)b
	Forward	Reversea
PjMS1	AAGATGACAACGAGAATTGGCT	CTGTCTTAAAAGGCGATAAATGTGTGTCC	53.3
PjMS2	TCATATACCGATCCTTTGGGAG	CTGTCTTGCCTGGACAATCCTGTCTCTAT	53.3
PjMS3	AATAGGCGGAATCTCACTAGCA	CTGTCTTGACGGCAAAGAGTTGTTTCCT	53.3
PjMS4	ATCGTAGAAGGATGGAAAGAAG	CTGTCTTGCAAATCGCATGTTCAAT	50.5
PjMS5	ACTGTACCTAATCTTTCATCGG	CTGTCTTAAGGTTTTGGACGTTTGA	50.5
PjMS6	TGCTCGAATTGCAGTAGAGATT	CTGTCTTCATCAGCAAGACGCTTAACTTG	55.2
PjMS7	GATCTGGGTTGAATATAAGCGT	CTGTCTTTCTTTGGTTTCACAACAGCA	55.2
PjMS8	CTTTGATTGCTCACGATATGGA	CTGTCTTATACGTTCACGGGCATAAGAGT	53.3
PjMS9	GAACTTTGTCTTTAACAGGACG	CTGTCTTCTCAGTAGTAGCACCAATTCTTT	47.0

^aReverse primers have the 5′ addition of a CTGTCTT “pigtail” to decrease stutter.

^bT_a, temperature of primer annealing.

PCR fragment length was determined on an ABI 310 genetic analyzer and analyzed using ABI GeneMapper v4.1 software. A genomic DNA (gDNA) standard from a P. jirovecii-positive clinical specimen isolated at the University of North Carolina was used as an interrun standard to adjust for batch variability in fragment sizes. Capillary electrophoresis was performed using a denaturing polymer (POP-4; Applied Biosystems) in a 40-mm capillary at 60°C.

Excepting the possibility of diploidy, multiple microsatellite peaks likely indicate a multiclone infection (22). In the case of multiple peaks, the 2nd, 3rd, and 4th highest peaks were recorded if they were greater than one-third of the maximum peak height and also greater than 100 fluorescence units. For analyses, haplotypes were built from dominant peaks at each marker. This method of haplotype construction is reasonable, as peak height is relatively quantitative, and this has become a standard approach in other organisms in which polyclonal infections are common (46, 47). However, there is the potential that this method does not necessarily produce the single correct haplotype due to biases such as overamplification of shorter fragments during PCR. A further limitation to our method, which is common to all multilocus genotyping schemes, is the difficulty of assigning the correct phase of typed alleles in mixed populations. Fragment sizes were binned to the nearest repeat multiple. For most analyses, data points 3 standard deviations beyond the overall mean repeat size for that marker were considered missing data, and any individual with four or more missing data points was excluded from analysis.

Data analysis.

Summary statistics and genetic linkage were determined per the following methods. Heterozygosity (H_e) was determined using GenAlEx v6.5 for Microsoft Excel (48). H_e, a commonly used measure to quantify the diversity at a genetic locus in haploid organisms, is defined as $1-{\mathrm{\Sigma }}_{i=1}^k\hspace{0.17em}{\mathit{P}}_i^2$, where p_i is the frequency of the ith of k alleles. Linkage between genetic loci was determined in Arlequin v3.11 using an exact test of linkage disequilibrium (49), in which haplotypic data were arranged into contingency tables, and 10⁵ Markov chains were used to explore simulated data with 10⁴ burn-in (dememorization) steps. Bonferroni corrections were used to adjust significance levels for multiple comparisons.

Geographic population structure was determined per the following methods. Distance matrices and phylogenetic trees were generated in Populations v1.2.31 (http://bioinformatics.org/populations/) using the Cavalli-Sforza and Edwards chord distance (51). R_ST (52) and F_ST (53) measures of genetic differentiation between populations were calculated in SPAGeDi v1.4 (54) as follows: R_ST = 1 − (S_W/S), where S is twice the estimated total allele size variance and S_W is twice the estimated allele size variance within subpopulations; and F_ST = 1 − (H_S/H_T), where H_T is total heterozygosity and H_S is the mean heterozygosity within subpopulations. Ninety-five percent confidence intervals were generated in SPAGeDi using jackknifing over loci with 1,000 permutations.

To further investigate population structure, we used a combination of statistical and visual approaches. Fast UniFrac, a program for microbial population genetic analysis, was used to measure the reproducibility of the estimates of genetic distance between geographic populations (55). The reproducibility of genetic splits between populations was assessed using Monte Carlo simulations and jackknifing with 1,000 replicates. Principal coordinates analysis (PCoA), a multidimensional scaling analysis, was carried out in the Fast UniFrac browser window. PCoA identifies the main axes through a matrix using an eigenanalysis to quantitate dissimilarity between populations. A median-joining network was calculated and visualized in Network v4.611 (Fluxus Engineering, Suffolk, England). These networks allow for a visual representation of the mutational paths that may have led to the observed data and assume that mutations are more likely to derive from a more frequent haplotype and proceed to a less frequent haplotype.

In order to determine an optimal set of microsatellite loci for use in transmission studies, we used Simpson's index of diversity (D) to seek the minimum number of microsatellite loci necessary to fully explain the diversity of our data set (56). A Perl script was used to calculate Simpson's D for combinations of the microsatellite markers, using the following equation: $\mathrm{D}\mathrm{=}\mathrm{1}\mathrm{-}\frac{\mathrm{1}}{\mathrm{N}\mathrm{(}\mathrm{N}\mathrm{-}\mathrm{1}\mathrm{)}}\mathrm{\hspace{0.17em}}{\mathrm{\Sigma }}_{\mathrm{j}}^{\mathrm{S}}{\mathrm{n}}_{\mathrm{j}}\mathrm{(}{\mathrm{n}}_{\mathrm{j}}\mathrm{-}\mathrm{1}\mathrm{)}$, where N is the total number of parasite multilocus genotypes, s is the number of unique multilocus genotypes, and n_j is the count frequency of the jth multilocus genotype variant (56). We added microsatellites to our scheme until D was ≥0.999 for our data set.

A set of six microsatellites was then used to investigate the relatedness between clinical isolates and transmission using two approaches. A neighbor-joining phylogeny (57) was generated in Populations v1.2.31, as outlined above, and visualized in the APE package for R (58). All samples from San Francisco with genotype calls for at least one-half of the markers (n = 40) were used in this analysis. In addition, hypotheses of evolutionary descent and transmission were generated in eBURST v3 (59). In this analysis, only samples with complete haplotypes for the selected markers were used due to the design of the program (n = 28). These analyses were performed in the San Francisco cohort alone, as this was the only sample set that contained paired samples from the same individuals.

Nucleotide sequence accession numbers.

Data for the sequences determined in this study were deposited in GenBank under accession numbers KF499042 to KF499075).

RESULTS

Of the 50 putative microsatellites identified and tested for interstrain length variability in the P. jirovecii genome, 9 were variable and thus carried forward into analyses (Table 1). Based upon predicted genome annotations, one microsatellite locus was intergenic, seven loci were intragenic noncoding, and one was intragenic coding (see Table S1 in the supplemental material). P. jirovecii isolates from 91 clinical specimens from Uganda, San Francisco, and Spain were typed at all 9 loci, and complete haplotypes were obtained for 63 (70%) isolates (Table 2). Amplification efficacy for each marker ranged from 78.0% for MS8 to 97.8% for MS1 (see Table S1 in the supplemental material). Each of the 9 markers was polyclonal in at least one sample. Similar to what was observed in other studies (60, 61), a high proportion of the clinical specimens were multiclonal: of the 91 BAL specimens, 63 (70%) contained a minimum of two P. jirovecii strains (at least 1 microsatellite with two peaks) and 14 isolates (15%) contained a minimum of three P. jirovecii strains (at least 1 microsatellite with three peaks).

TABLE 2.

Per-population summary statistics for P. jirovecii isolates

Population source	No. of isolates	No. of complete haplotypesa	No. of paired samplesb	% of isolates with dhps mutant
Uganda	13	10	0	100 (13/13)
San Francisco	49	28	5	65.3 (32/49)
Spain	29	25	0	10.3 (3/29)
All sources	91	63	5	52.7 (48/91)

^aComplete haplotypes are isolates for which fragment length was determined for all nine microsatellite markers.

^bPaired samples represent two bronchoalveolar lavage samples taken from a single patient, during either the same or different PcP episodes.

We assessed the variability of microsatellite repeats in order to quantify which loci were most informative. To do this, we computed summary statistics including heterozygosity (H_e) for each marker, both overall and on a per-population basis (Fig. 1; see Table S1 in the supplemental material). MS1 through MS8 had high mean H_e values (>0.5), indicating that they are diverse and likely informative. In contrast, MS9 had a reduced H_e value (<0.5). Interestingly, MS9 showed highly depressed H_e values relative to MS1 to 8 in Uganda but had similar H_e values relative to MS1 to 8 in Spain. Notably, all 13 Uganda samples contained mutations in dhps (3 Thr55Ala/Pro57Ser double mutants and 10 Pro57Ser single mutants), but very few of the Spain samples (3/29) contained mutations at dhps. Upon performing a scaffolded reconstruction of the P. jirovecii dhps region using Pneumocystis murina supercontigs, we found that MS9 was located ca. 50 kb upstream of the dhps locus. In pairwise tests of linkage disequilibrium, dhps was in significant linkage disequilibrium with MS9 (exact test, P < 0.05); no other markers were near the dhps locus.

FIG 1.

Per-population diversity index for each microsatellite studied. Nine microsatellites were typed in multiple samples from Uganda (triangles), San Francisco (circles), and Spain (squares), and heterozygosity (H_e) values were calculated for each population. Mean H_e values are indicated by horizontal bars. A higher H_e value indicates that a marker is more variable and thus more informative for population and transmission studies. Marker MS9 (shaded) was determined to be in linkage disequilibrium with the dhps locus.

The fact that the dhps locus and MS9 are in linkage disequilibrium with one another enables studies to investigate the genetic background of dhps mutations and to acquire evidence for a selective sweep of dhps mutant alleles. Eleven dhps mutant samples from Uganda and 14 dhps mutant samples from San Francisco with unambiguous MS9 genotype calls were used to construct dhps-MS9 haplotypes (see Table S2 in the supplemental material). In Uganda samples, we observed two unique dhps mutant genotypes (Thr-Arg-Ser and Ala-Arg-Ser) and two MS9 alleles, (TA)₉ and (TA)₁₀. Of the four possible combinations of dhps and MS9 alleles, we identified three with an excess of Thr-Arg-Ser/(TA)₁₀ haplotypes. In San Francisco samples, we observed three unique dhps mutant genotypes (Ala-Arg-Pro, Thr-Arg-Ser, and Ala-Arg-Ser) and 5 MS9 alleles, (TA)₈ to (TA)₁₂. Of the 15 possible combinations of dhps and MS9 alleles, we identified 7 with an excess of Ala-Arg-Ser/(TA)₁₀.

Intercontinental population structure.

In order to minimize bias in analyses of population structure, we tested for linkage between microsatellite loci (62). Because the Pneumocystis jirovecii draft genome is in 356 contigs (32, 63) rather than in chromosomes, we could not rely on genetic maps and instead tested linkage disequilibrium between markers. Bonferroni-corrected per-population pairwise tests of linkage disequilibrium showed no significant linkage disequilibrium between any locus pairs in any of the three populations (results not shown).

Using the 8 unlinked microsatellites, we examined the extent of genetic differentiation between P. jirovecii populations from Uganda, San Francisco, and Spain. We employed UniFrac, a widely used method for probing the genetic diversity between microbial communities, which can draw conclusions about the relative distances between multiple populations. In this analysis, P. jirovecii isolates from Uganda were genetically distinct from those in San Francisco and Spain (Fig. 2A). Principal coordinates analysis revealed that the first coordinate explained approximately one-half (53.8%) of the observed genetic difference between Uganda isolates and the other populations (Fig. 2B). Finally, we constructed a median-joining network analysis charting potential mutational intermediates between observed P. jirovecii isolates. This showed little structure between populations, with the exception being that Uganda occupied a restricted range of genetic diversity (Fig. 2C).

FIG 2.

Limited population structure and genetic differentiation between disease-causing P. jirovecii populations. (A) UniFrac revealed a distinct genetic division between Uganda and San Francisco/Spain but little genetic difference between P. jirovecii isolates from San Francisco and Spain. (B) Principal coordinates analysis demonstrated that approximately one-half of the genetic difference between population clusters was associated with geographic differences. (C) A median-joining network predicts potential mutational paths between haplotypes. The samples from Uganda (black circles) show clustering on one side of the network, likely leading to the differences seen in the other analyses.

We also evaluated genetic distances between these populations using R_ST (52) and F_ST (53), two classic measures of population demographic history (64). Similar to our previous analyses, R_ST and F_ST suggested that while Ugandan samples were significantly divergent from the San Francisco and Spain populations, there was limited genetic differentiation between the three populations overall (Table 3) (65). Specifically, R_ST, which assumes a stepwise mutation model of microsatellite evolution, revealed no genetic distance between the populations from San Francisco and Spain and limited distances in the Uganda-San Francisco and Uganda-Spain comparisons (R_ST range, −0.017 to 0.125). F_ST also revealed limited genetic distances between populations (F_ST range, 0.022 to 0.067).

TABLE 3.

Pairwise R_ST and F_ST statistics for between-population pairwise comparisons

Source	RST or FST (95% CI)a
	Uganda	San Francisco	Spain
Uganda		0.125 (0.074–0.220)	0.115 (0.060–0.233)
San Francisco	0.048 (0.020–0.090)		−0.017 (−0.044–0.044)
Spain	0.067 (0.036–0.109)	0.022 (0.006–0.043)

^aData from all 8 neutral microsatellite loci were used to calculate R_ST (above the diagonal) and F_ST (below the diagonal) in SPAGeDi v1.4. In general, an F_ST or R_ST value of 0 to 0.05 indicates little genetic differentiation, 0.05 to 0.15 indicates moderate differentiation, 0.15 to 0.25 indicates substantial differentiation, and 0.25 to 1.00 indicates very great genetic differentiation. CI, confidence interval.

Relatedness between isolates within individuals and within a single population.

To maximize the efficiency of a potential microsatellite genotyping scheme, we determined the minimum number of microsatellite loci necessary for discriminating between strains, using both the loci with the greatest heterozygosity and those with the greatest amplification efficiency (see Table S3 in the supplemental material). We achieved high resolution (Simpson's D, ≥0.999) across all specimens tested by using the 6 most heterozygous markers (MS6, MS8, MS7, MS1, MS3, and MS5) (see Table S3 in the supplemental material). This extensive diversity is similar to the diversity described in both Pneumocystis (30) and other pathogens (66–69), making this typing method a powerful tool for P. jirovecii population genetic studies (56).

In order to evaluate the suitability of this array to study the transmission of P. jirovecii in humans, we investigated paired specimens collected from the same patient. This analysis was restricted to isolates from San Francisco, which included five isolate pairs (10 isolates) in which two samples were taken from the same individual. The genotypes of each pair are shown in Table S4 in the supplemental material. In a neighbor-joining phylogeny of all San Francisco samples, all paired isolates grouped closely together, with differing extents of genetic similarity (Fig. 3). The extent of genetic similarity between paired isolates was reflective of the time elapsed between collection of the two isolates (see Table S4 in the supplemental material).

FIG 3.

Neighbor-joining phylogenetic tree of P. jirovecii clinical isolates from San Francisco. Repeat samples from the same clinical episode produced an identical genotype (SF45-1 and -2 and SF46-1 and -2). The other paired samples were genetically similar between samples but not identical. The other points represent isolates from patients from which a single sample is available. Very few samples had identical genotypes, suggesting that the six-microsatellite array will have a high discriminatory power between isolates.

To more rigorously investigate this finding, we employed eBurst, a tool developed to study outbreaks and explore pathogen transmission dynamics (59, 70). eBurst visualizes relationships (edges) between individual parasite isolates (nodes) to generate hypotheses about pathogen relatedness. We investigated all San Francisco isolates from the years 2008 to 2011 for which there were complete haplotypic data at the 6 most informative markers (including three isolate pairs). eBurst analysis showed that the two isolate pairs that were temporally closest to one another (SF45 and SF46 in Table S4 in the supplemental material) were identical at these six markers, while the third isolate pair (SF47) was closely related, differing at only one locus (Fig. 4). The other paired isolates (SF48 and SF49) contained missing alleles and could not be included in this analysis (see Table S4 in the supplemental material).

FIG 4.

Related P. jirovecii genotypes in the San Francisco population. eBURST outbreak analysis on San Francisco samples in which a complete six-microsatellite haplotype was available. Paired samples are highlighted in gray. Two pairs (SF45-1 and -2 and SF46-1 and -2) were identical to one another and occurred in the same clinical episode of PcP. Another pair of isolates (SF47-1 and -2) were related, with only one microsatellite difference.

DISCUSSION

Understanding the population genetics and transmission dynamics of fungal pathogens is key for understanding their life history and for designing new prevention and treatment strategies. Appropriate strain typing methods are crucial to achieve this understanding (71). Here, we introduce the first multilocus microsatellite genotyping scheme for Pneumocystis jirovecii. In contrast to past genotyping schemes, which rely in large part on genic coding regions, including potential drug resistance (DHFR, DHPS, and SOD) genes and housekeeping (mtLSU-rRNA and mt26S) genes (25, 27, 31), we employ microsatellites, most of which are noncoding (see Table S1 in the supplemental material). Thus, this genotyping method carries a high discriminatory index that may be relatively free of selective pressure or functional constraints, making it a useful technique for studying population structure, sulfonamide resistance, and transmission dynamics. Microsatellites are widely employed for genotyping individuals owing to their robustness, scalability, and high information content. Given these advantages, this genotyping approach will enable studies that improve our understanding of P. jirovecii biology.

Using these microsatellite markers, we observed limited genetic differentiation between P. jirovecii populations from Uganda, the United States, and Spain. This observation was supported by an ecological clustering algorithm (Fig. 2A) and classical measures of genetic distance (Table 3), both of which indicate a slight divergence of the Ugandan isolates but relatively little divergence overall. Principal coordinates analysis revealed that approximately one-half of the genetic differences between all three populations could be explained by geography (Fig. 2B), with the maximum distance occurring in coordinate one, which separated Uganda from San Francisco and Spain. Furthermore, a median-joining network showed that isolates from all three sites were thoroughly admixed, with the only notable trend being that Ugandan isolates clustered loosely in one part of the network (Fig. 2C).

The relative lack of genetic divergence that we observed between sites, and over time, is similar to what has been reported before (72–74). This has several potential explanations, including the following: (i) P. jirovecii underwent a recent global spread, (ii) P. jirovecii has a low mutation rate, or (iii) intercontinental gene flow between P. jirovecii populations occurs frequently. Our understanding of Pneumocystis populations will be further enhanced as the P. jirovecii genome takes a more complete form and as additional samples from other regions are analyzed.

Interestingly, we identified one microsatellite linked to the dhps gene, which encodes the enzymatic target of sulfonamide antibiotics that are used to treat and prevent PcP. Polymorphisms in dhps are associated with exposure to antifolate drugs (10, 75) and, in some studies, clinical outcomes (76–79). All 13 isolates from Uganda harbored mutations in dhps: three with both the Thr55Ala and Pro57Ser substitutions and 10 with the single Pro57Ser substitution. These mutant haplotypes of dhps likely reflect directional selection by sulfonamide antimicrobials used to treat infections; consistent with this hypothesis, we observed very low heterogeneity in the microsatellite locus linked to dhps in Ugandan isolates (H_e = 0.282). In contrast, the isolates from Spain, which were largely sulfa naive and which harbored primarily wild-type dhps haplotypes, demonstrated a higher heterozygosity at this dhps-linked locus (H_e = 0.632). Taken together with evidence that antibiotic pressure causes changes in dhps mutant genotype frequency (73), this pattern is consistent with a selective sweep occurring around the dhps locus due to drug pressure, as has been observed in pathogens such as Plasmodium falciparum (36). In order to confirm this, further studies will require additional markers near the dhps locus and greater numbers of isolates bearing wild-type and mutant dhps haplotypes.

In addition to understanding population structure and drug selection, microsatellite analysis has potential as a tool for molecular epidemiology studies evaluating transmission. The currently used multilocus genotyping schemes have revealed that (i) specific P. jirovecii strains are associated with failure of prophylaxis (80) and can cause more-virulent PcP episodes (81), (ii) specific strains persist in hospitals for weeks at a time (80), and (iii) P. jirovecii strains can be acquired later in life by immunocompromised and immunocompetent patients (22, 82). Despite this wealth of knowledge, the current multilocus genotyping strategies have left many questions regarding transmission of P. jirovecii unanswered. For example, while it is clear that infants and immunocompetent adults are environmental reservoirs for P. jirovecii (83–87), there is not yet molecular evidence that these reservoirs are an important source of disease-causing P. jirovecii transmission to immunocompromised individuals. Because there is compelling evidence that transiently colonized immunocompetent mice can transmit P. jirovecii to immunosuppressed mice and vice versa (88, 89), studies of P. jirovecii transmission among immunocompetent and immunosuppressed humans may help uncover a mechanism for disease prevention. A highly discriminating genotyping method with a stepwise mutation model would prove useful in addressing this and other questions.

The microsatellite genotyping method described here could help reveal details of P. jirovecii reservoirs and transmission. By assessing five paired clinical samples (10 isolates), we showed that this approach has high discriminatory power, making it ideal for molecular epidemiological studies (56). Among our five paired isolates, we observed two cases (SF45 and SF46) in which the two samples in the pair had identical genotypes in a clinical scenario consistent with a single infection. In the case of SF45, both samples were collected by a single bronchoscopy in two different lobes of the lung. In the case of SF46, the first sample was collected by bronchoscopy upon initial presentation, and the second sample was collected 15 days later after clinical deterioration despite antifolate therapy. The last three cases (SF47, SF48, and SF49) all represent PcP patients who received PcP therapy but experienced a recurrence weeks to months later. In all three cases, the pairs were genetically close, but not identical, similar to past observations (61). This finding could be explained in multiple ways, including the following: (i) these patients may have been infected with multiple P. jirovecii strains, which were differentially detected at different time points (90), or (ii) after clinical resolution of their initial PcP case, these patients returned home and were subsequently colonized by genetically related—though not identical—organisms. Importantly, these data show that this method provides high discriminatory power between infections: other than those isolate pairs expected to be related, there were no identical isolates and only two related isolates (Fig. 3 and 4). This observation suggests that this genotyping method will prove useful for future studies of P. jirovecii outbreaks and transmission dynamics. Additionally, work from others suggests that PCR-based P. jirovecii genotyping approaches may prove successful even in immunocompetent individuals (91), enabling future studies of transmission between immunocompromised patients and immunocompetent health care workers or close contacts.

Pneumocystis jirovecii remains an important opportunistic pathogen and causes significant disease in immunocompromised individuals. Many questions about the biology of the organism remain unanswered, and the PcP disease burden is incompletely understood. Molecular epidemiology studies have the potential to provide critical information concerning organismal biology, and insights from these studies could impact health policy and influence medication use in patients. While high-resolution genotyping methods, such as this one, should be applied with care to some study types (61, 73), our approach provides a stepwise mutational model, high typeability, and high discriminatory power that will be valuable for future molecular epidemiological studies of P. jirovecii.