P. carinii, once thought to be a protozoon but now regarded an ascomycetelike fungus (4,19,20), has been associated with human disease since the 1940s. Despite intense efforts to understand this important disease agent, lack of a suitable means of propagation has complicated study of basic biology and epidemiology. Similarly, latency and reactivation (versus recent acquisition) have not been resolved. PCR and other molecular methods have improved understanding of genetic variability and host specificity (11,12,14,20,21), as well as the cause of recurrent infections in persons with AIDS (8,22-24).

Both an indirect (environmental) source and a direct (person-to-person) source have been proposed as modes of transmission of P. carinii in humans. Three primary observations reported in this study address infection sources and transmission patterns: geographic variation in genotype frequency distribution, which was correlated with the place of diagnosis rather than the place of birth; rate of coinfection with multiple P. carinii genotypes, which was greater in patients with primary rather than secondary PCP; and abundance of the DHPS double mutant, which accounted for 49.5% of all DHPS genotypes, strongly suggesting genetic selection.

Allelic frequency distribution patterns of P. carinii isolates were associated with place of diagnosis rather than place of birth. Place of diagnosis was determined more reliable than place of residence as the best indicator of the most likely place of exposure to P. carinii because place of residence was frequently recorded as permanent or legal residence rather than as place of residence during the 3- to 6-month period before admission. These results suggest that infection in adults is acquired later than the first few months or years of life and that any latency has natural limits. Two independent lines of study offer a context for these observations and suggest that PCP is an actively acquired infection. The first line of study comprises molecular genetic analyses of P. carinii strains from adult AIDS patients with recurrent PCP. These analyses have shown that different P. carinii genotypes are detected on subsequent PCP episodes in a substantial proportion of patients (22,23), which suggests that infections in adults are actively acquired and that subsequent infections do not necessarily represent relapses. The second line of study comprises observations of primary PCP in infants with perinatally acquired HIV infection (25) who had PCP episodes at 3 to 6 months of age; the prevalence of PCP among these infants was very similar to that among adults with AIDS. These observations suggest that P. carinii is common in the environment, consistent with the suggestion that the organism is easily acquired.

While the association with geographic distribution is consistent with recent acquisition of clinical infections, the actual source of the infection is not known. The rat P. carinii model suggests direct (animal-to-animal) transmission (5,26,27); however, if person-to-person transmission does occur, it has not been shown to be of epidemiologic significance (28,29). PCR amplification of P. carinii DNA from spore traps from various sites supports the possibility of an environmental source (30-33), but no specific plant, animal, or soil source has been identified. Identification of a specific environmental reservoir for P. carinii f. sp. hominis, if one exists, could elucidate disease transmission and improve prevention efforts.

The rate of coinfection with multiple P. carinii strains by multilocus typing was 17.8% (34 of 191). Of the 191 patients for whom both genes were typed, 20 (10.5%) were coinfected at the mtlsurRNA locus and 19 (9.9%) at the DHPS locus. Five patients tested positive for multiple strains concurrently at both loci. While the sensitivity to detect coinfections was similar for both genes, concordance between the two genes was very low in detecting coinfections with multiple strains in any particular isolate--only four genotypes can be detected at each individual locus. When the loci are considered together, however, the sensitivity is greatly increased, because the number of possible genotypes increases from 4 to 16, 14 of which were observed in this study.

The 17.8% coinfection rate is within the 10% to 30% range reported by other investigators using DNA sequencing-based approaches (14,22,34). Although coinfection rates as high as 69% have been reported with single-strand conformation polymorphism analysis (35), these differences can probably be resolved by two considerations. The first is the sensitivity of the selected locus, which is a function of the genetic variability or evolutionary rate of the locus (i.e., the more variable the locus, the greater its sensitivity to detect a different genotype). In this study, mtlsurRNA and DHPS displayed a high degree of genetic conservation, with only four alleles detected at each locus, all of which have been observed (11,16,17). Because these are relatively slowly evolving genes, the sensitivity to detect coinfections with multiple strains is expected to be less than in a faster evolving locus. The second consideration is PCP prophylaxis. Correlation between primary PCP and coinfection with multiple strains suggests that patients may be exposed to multiple P. carinii strains over extended periods and harbor short-lived, latent infections. Consequently, patients who have not been treated are more likely to have been exposed (and to harbor) multiple P. carinii strains. On the other hand, patients who have been treated for PCP at least once and are taking secondary prophylaxis are less likely to become reinfected. This hypothesis is consistent with indications that P. carinii is ubiquitous in the environment and exposure in humans is commonplace (22,23). A short latency period has been suggested (36) and is consistent both with coinfections' correlating with primary PCP and allelic frequency distributions' correlating with the place of diagnosis but not the place of birth.

The question of latency and reactivation versus recent acquisition is of great importance as it relates to prevention. If the preponderance of infections in humans results from activated latent infections, chemoprophylaxis is the only method of preventing disease. If, however, infections are actively acquired, identifying specific sources of infection would lead to other methods of prevention and reduce dependence on antimicrobial agents.

The most important clinical implication of the polymorphism observed specifically at the DHPS locus relates to the emergence of possible antimicrobial resistance; however, the observation of different DHPS alleles in P. carinii populations also has implications for transmission routes and patterns and the possibility of person-to-person transmission.

In human P. carinii isolates, four distinct genotypes have been reported at the DHPS locus (16,17). All four of these DNA base changes result in amino acid substitutions at an important structural position, in an otherwise very highly conserved gene (17,37). Genotype 1 in our study corresponds to the "wild-type" genotype, as defined by the only allele observed in P. carinii from any other mammalian species (16,17). Genotype 2 is a point mutation that results in a threonine-to-alanine substitution at amino acid position 55. Genotype 3 is also a point mutation, resulting in a proline-to-serine substitution at amino acid position 57. Genotype 4 is a double mutant, which has alanine at position 55 and serine at position 57. This particular mutation has rarely been associated with failure of both TMP-SMZ treatment (38) and prophylaxis (17). The presence of polymorphism at two different positions, both associated with amino acid substitutions, at a gene locus that otherwise lacks variability, is itself indicative of pronounced selective pressure. Such a degree of selection is not itself an indication of resistance, but certainly suggests that resistance may be emerging. Mutations in this same region of the DHPS molecule have been correlated with specific antimicrobial resistance to sulfa drugs in a number of other microorganisms, including Plasmodium falciparum (39), Streptococcus pneumoniae (40), Streptococcus pyogenes (41), Escherichia coli (42), and Neisseria meningitidis (43).

The two single mutations are uncommon by themselves (5.9% for genotype 2 and 2.3% for genotype 1), yet 50% of all isolates have the double mutant, which suggests that, alone, neither mutation is highly selected, but together they pose a very strong selective advantage. If humans are dead-end hosts for P. carinii, how can such a pronounced degree of polymorphism be explained? Person-to-person transmission may be essential to allow genetic selection to occur, resulting in this polymorphism.

Geographic variation in allelic frequency detected at the mtlsurRNA locus is not unexpected, even though other studies with smaller sample sizes failed to observe differences (11). Mitochondrial DNA sequence data have been highly useful in detecting intraspecific differences between populations of diverse organisms (44,45). Perhaps more unexpected, however, was the detection of geographic variation at the DHPS locus. Unlike mtlsurRNA, because it encodes a gene product that is the target of the primary anti-P. carinii drug TMP-SMZ, the DHPS locus is assumed to be subject to intense selection pressure. Consequently, this selection might be expected to override any potential variation from geographic separation or geographic patterns to reflect TMP-SMZ exposure patterns. The relative frequency of the double mutant genotype was much higher in samples from the West Coast, particularly San Francisco, than in samples from Atlanta (Figure 2). The reason for this difference is not obvious because in preliminary multivariate analysis conducted with samples from Atlanta and San Francisco, when the data are controlled for sulfa exposure, place of diagnosis is still the most significant factor influencing the frequency of the double mutant genotype (data not shown). Explanations for this observation are being evaluated.

An overall frequency of approximately 50% for the DHPS double mutant genotype is somewhat higher than that reported in other recent studies (46,47). The reason for this observation is also unclear. One possible explanation is that the samples were collected more recently (March 1995 to June 1998) and therefore reflect a sulfa-induced increase in frequency of the double mutant. Most patients were from San Francisco and Atlanta, the two populations that showed the greatest difference in genotype frequency at the DHPS locus. Sample collection in Atlanta began in 1995, but in San Francisco the earliest samples were from 1997. When we stratified the patients in Atlanta by date, 1995-96 versus 1998-99, we saw no significant difference in the frequency of mutant genotypes. Similarly, when we stratified data for San Francisco by date, 1997 versus 1998-99, we saw no differences. Consequently the frequency of the mutation did not change during this study period, nor does the variation observed according to geography appear to be confounded by the dates when specimens were collected.

The pattern of allelic variation differed at each of the cities where samples were obtained, and this variation correlated with the place of diagnosis but not with the place of birth. Coinfection with multiple P. carinii genotypes was associated with primary rather than secondary PCP. The position 55/57 double mutant accounted for 50% of all genotypes of the DHPS locus examined. These observations suggest the following possibilities, which contradict much of the current opinion on the epidemiology and transmission of PCP: Most cases of PCP are not a result of infections acquired very early in life; infections are actively acquired from a relatively common source (humans or the environment); and humans, while not necessarily involved in direct infection of other humans, are nevertheless important in the transmission cycle of P. carinii f. sp. hominis.