De Novo Assembly of the Pneumocystis jirovecii Genome from a Single Bronchoalveolar Lavage Fluid Specimen from a Patient

ABSTRACT

Pneumocystis jirovecii is a fungus that causes severe pneumonia in immunocompromised patients. However, its study is hindered by the lack of an in vitro culture method. We report here the genome of P. jirovecii that was obtained from a single bronchoalveolar lavage fluid specimen from a patient. The major challenge was the in silico sorting of the reads from a mixture representing the different organisms of the lung microbiome. This genome lacks virulence factors and most amino acid biosynthesis enzymes and presents reduced GC content and size. Together with epidemiological observations, these features suggest that P. jirovecii is an obligate parasite specialized in the colonization of human lungs, which causes disease only in immune-deficient individuals. This genome sequence will boost research on this deadly pathogen.

IMPORTANCE

Pneumocystis pneumonia is a major cause of mortality in patients with impaired immune systems. The availability of the P. jirovecii genome sequence allows new analyses to be performed which open avenues to solve critical issues for this deadly human disease. The most important ones are (i) identification of nutritional supplements for development of culture in vitro, which is still lacking 100 years after discovery of the pathogen; (ii) identification of new targets for development of new drugs, given the paucity of present treatments and emerging resistance; and (iii) identification of targets for development of vaccines.

Observation

Pneumonia caused by Pneumocystis jirovecii (PCP) marked the onset of the AIDS epidemic and remains today the most frequent AIDS-defining infection as well as a major cause of mortality in immunocompromised patients (1). The poor knowledge of P. jirovecii biology and the absence of genome sequence are due first of all to the fact that no in vitro culture method is presently available. P. jirovecii microorganisms can be obtained only from clinical specimens of patients, thus in very limited amounts and heavily contaminated by human cells and lung flora. Here we report the assembly of the P. jirovecii genome sequence from a single clinical specimen of a single patient. To our knowledge, this is the first eukaryotic genome de novo assembled out of a metagenome.

Whole-genome sequencing requires usually between 2 to 5 µg of pure genomic DNA, which is usually not recoverable from clinical specimens without in vitro culture. To compensate, we used cell immunoprecipitation followed by random DNA amplification. Four bronchoalveolar lavage fluid samples (BALFs) from patients with PCP, who were all fortuitously HIV uninfected, were retained to estimate their percentages of Pneumocystis DNA content, two of them after enrichment in P. jirovecii cells using immunoprecipitation (see Text S1, sections A1 and A4, in the supplemental material). Genomic DNA was extracted, amplified, and sequenced by low-level Roche 454 pyrosequencing. The reads from each BALF were assigned in silico to an organism or to a group of organisms (Homo sapiens, P. jirovecii, fungi, bacteria, or viruses), using a simplified classification pipeline (see Text S1, section A5, in the supplemental material). This rough classification established that the enrichment was effective: the proportion of Pneumocystis DNA in the two enriched specimens was 23% (BALF E6) and 15% (BALF E8), whereas it was only 6% (BALF N1) and 0.7% (BALF N8) in the two nonenriched samples (see Table S1 in the supplemental material).

P. jirovecii has been repetitively reported to be the only Pneumocystis species infecting humans. However, one contribution reported the presence in a single patient of P. jirovecii as well as Pneumocystis carinii (2), the species considered to infect specifically rats. To prevent any coinfections for genome sequencing, we applied an extensive verification to exclude the presence of another Pneumocystis species or fungus than P. jirovecii in the BALFs (see Text S1, sections A6 and A7, in the supplemental material). We found that BALF E6 contained for undetermined reasons multiple Pneumocystis species, whereas all three other BALFs contained exclusively P. jirovecii (see Table S2 in the supplemental material). The rest of this study was conducted on BALF E8 with the highest load of P. jirovecii.

The selected BALF, E8, was from a patient with chronic lymphocytic leukemia. It was enriched in P. jirovecii cells, and the genomic DNA extracted from it was amplified and used for whole-genome sequencing. The major challenge in this study was the in silico sorting and assembly of P. jirovecii reads out of a mixture representing many organisms. We iteratively built stringent assembly of the reads to detect homology with the available genome of P. carinii, as well as with other fungal genomic data. Ideally, reads should be attributed to a single organism before an assembly of its genome is attempted. However, this simple strategy could not be applied here, as most raw reads were too short to be attributed to a taxonomic group through sequence comparison. Moreover, some Pneumocystis-specific reads would certainly be lost or misidentified, as P. carinii assembly is known to be incomplete. Hence, we have established an overall strategy for read filtration and de novo assembly that allowed us to progressively refine and complete the P. jirovecii genome (see Text S1, section A9, and Fig. S1 in the supplemental material). The rRNA unit and the mitochondrial genome were recovered, assembled, and annotated separately. The 18S nuclear rRNA subunit displayed 99% nucleotide identity with the P. jirovecii public sequence, providing a solid assessment of the organism identity. The repeated telomere regions that contain the major surface glycoproteins were also kept apart. The repeated regions kept apart included probably the centromeres, which were not investigated further in the absence of a genetic or physical map. A posteriori, 25% of the reads could be attributed to P. jirovecii (see Table S3 in the supplemental material). In addition, we deep sequenced total RNA isolated from a nonenriched BALF of a patient with B-cell lymphoma, which provided 2.7% of reads attributable to the fungus. The de novo assembly of RNAseq data yielded 2,667 transcripts, which were used for genome annotation and provided a glimpse into the fulminate infection process in the human host. As previously observed for P. carinii (3), the most abundant category of the transcripts was annotated as major surface glycoproteins (7.2%).

The 8.1-Mb genome assembly is made of 356 contigs (Table 1). A total of 3,878 coding sequences were identified and annotated, with a gene density of 481 genes per Mbp. Seventy-seven percent of them were supported by the transcriptome data. We identified 458 single-copy orthologs that were shared by P. jirovecii and several other fungal species, including distantly related ascomycetes and basidiomycetes. A phylogenetic tree was computed from an alignment of these orthologous proteins (Fig. 1). It further documents the taxonomic position of P. jirovecii close to P. carinii and Taphrina deformans.

TABLE 1 .

Statistics of P. jirovecii and P. carinii nuclear and mitochondrial genomes

Characteristic	Result
	P. jirovecii	P. cariniia
Assembly
Assembly size (Mb)	8.1	6.3
Mean 454 read depth	36	NAb
Mean Illumina read depth	1,315	NA
No. of contigs	358	4,278
N50 (kb)	41.6	2.2
Mean GC content (%)	28.4	32.5
Annotation
No. of CDSs	3,898	4,591c
Coding regions (%)	68.9	50.0
No. of KEGG orthologsd	269	252
No. of tRNA genes	77e	36f
Mean gene length (bp)	1,472	891
Mean exon length (nt)	211	223
Mean no. of introns per gene	4.5	2.3
Mean intron length (nt)	61	54
Repeat density (%)	9.86	5.71
Mitochondrial genome
Assembly size (kb)	27	23g
No. of contigs	3	1
GC content in whole genome (%)	29.5	31.1
GC content in coding genes (%)	32.5	30.9
No. of protein coding genes (CDSs)	17	17
No. of rRNA genes	2	2
No. of tRNA genes	12	20

^a The P. carinii assembly was downloaded from the Pneumocystis genome project website (http://pgp.cchmc.org/) and corresponds to the sequences published by Slaven et al. (15). This assembly is known to be incomplete.

^b Not applicable.

^c Some of these peptides were derived from incomplete CDSs located at the extremity of a contig. Hence, the numbers of predicted peptides provide only a rough estimation of the proteome size.

^d The number of KEGG orthologs was computed as described previously (4).

^e Not including pseudo tRNAs.

^f Only complete copies are shown; the total number of tRNAs in the genome may be more important.

^g P. carinii mitochondrion data were computed from the published genome (16).

FIG 1 .

Maximum likelihood phylogeny of P. jirovecii and other fungi from alignment of 458 concatenated orthologs. Rhizopus delemar was used as the outgroup. The same tree topology was obtained using the maximum parsimony method.

The most striking feature of P. jirovecii was the lack of certain metabolic capabilities. As we previously found in P. carinii (4), the category of amino acid metabolism pathways was underrepresented in P. jirovecii (see Table S4 in the supplemental material), and a manual analysis revealed that most enzymes specifically dedicated to the synthesis of amino acids were absent (see Table S5 in the supplemental material). The loss of these pathways is a hallmark of obligate parasites (5). This strongly suggests that P. jirovecii scavenges these compounds from human lungs. Accordingly, an important proportion of its genes (22%) corresponded to transporters (e.g., amino acid permeases). Such transporters are believed to be necessary in P. carinii for scavenging amino acids as well as other compounds, such as host cholesterol and S-adenosylmethionine (reviewed in reference 6). The P. jirovecii genome presents a low GC content (29%) and a smaller size than its free-living relatives T. deformans and Schizosaccharomyces pombe (42% and 36%, 13 Mb and 14 Mb, respectively). In obligate bacterial parasites, the reduction of GC content is associated with the reduction of genome, which in turn is generally due to a loss of the genes responsible for the synthesis of compounds that can be scavenged from the host (7). P. jirovecii also lacks the hallmark enzymes of the glyoxylate cycle, a significant virulence factor of fungal pathogens (8), as well as polyketide synthase clusters, responsible for production of secondary metabolites, such as toxins. Together, these features are compatible with the view that P. jirovecii is an obligate parasite specialized in colonization of human lungs, which causes deadly disease only in immunocompromised individuals. This conclusion is further supported by the mechanism of surface antigen variation (6), the coevolution with hosts (9), and the strict host specificity of Pneumocystis spp. (10). Obligate parasitism also fits epidemiological studies, which failed to identify free-living forms as a source of infection. The host and reservoir of P. jirovecii is likely to be only humans, including immunologically impaired individuals, but also infants experiencing primo-infection (11), elderly people (12), pregnant women (13), and healthy transitory carriers (14). Consequently, the entire life cycle of P. jirovecii most probably takes place only within human lungs.

For an organism of clinical importance that cannot be grown in the laboratory as P. jirovecii, the genome sequence represents a wealth of new information for future research. It allows new analyses to be performed, such as RNA sequencing and comparative genomics, which open avenues to solve critical issues such as (i) the identification of nutritional supplements for culture in vitro development; (ii) the identification of new targets for development of new drugs, an important issue because only antifolates are presently efficient and development of drug resistance has been documented; and (iii) the identification of targets for the development of vaccines. The relevant P. jirovecii genes can now be used in these studies rather than those of P. carinii as models, which is particularly crucial for development of new drugs.

SUPPLEMENTAL MATERIAL

Text S1

Supplemental materials and methods.

Figure S1

Bioinformatics strategy used for the filtration, classification, and assembly of the reads from P. jirovecii.

Table S1

Estimation of the proportion of P. jirovecii DNA in four BALFs.

Table S2

BALF screening based on seven markers.

Table S3

A posteriori characterization of BALF E8 enriched in P. jirovecii cells.

Table S4

Number of Kegg orthologs (KOs) predicted for P. jirovecii, P. carinii, and S. pombe.

Table S5

Number of enzymes dedicated to the biosyntheses of amino acids in P. jirovecii, P. carinii, and S. pombe.

Table S6

Data used to build species-specific hmm profiles.