4.1 Genotyping of P. jiroveci

Genetic diversity in human pneumocystosis was first observed in the large subunit mitochondrial large subunit rRNA operon in which single base polymorphisms were detected at three nucleotide positions (105, 111, 183). Since then different sequence types of P. jiroveci have been identified in a number of genetic loci ( Table 2 ). As shown in Table 2, the discriminatory power of typing based on nucleotide sequence variations of genes encoding the small and large subunit rRNA (mtSSU and mtLSU rRNA), arom, and DHPS are relatively limited with, respectively, two, five, four, and five types described. Because of the relatively low discriminatory power of these loci, alternatives were sought. Since DNA sequences that encode rRNA have been central to the study of genomic variations and taxonomic relationships in fungi and other species, this locus was chosen for investigation (27, 111, 113) . Like other organisms, Pneumocystis has three species of nuclear rRNA transcripts: 18S, 5.8S and 26S. Between these genes are two regions called internal transcribed spacer (ITS) regions ITS1 and ITS2 ( Figure 3 ).

In many parasites and fungi, primary ITS sequence analysis are in accordance with known taxonomic subfamilies and have improved distinction between separate strains (65, 66, 168). In contrast to most other organisms, Pneumocystis has only a single copy of the nuclear rRNA genes including ITS (154, 197). Studies, which have compared ITS genotyping of specimens to other loci, such as mtLSU, mtSSU and DHPS have found that this locus is the most informative as more sequence types are detected by ITS (100, 200, VI).

4.2 ITS typing of P. jiroveci

ITS typing of P. jiroveci was first reported by Lee and co-workers in 1993 (110, 113). Based on single nucleotide polymorphisms or deletions at several positions of the ITS1 and ITS2 genes, the initial typing scheme designated ITS1 types as A, B and C and ITS2 sequence types a, b and c, with four different combination types (Ac, Bb, Ba and Bc) observed among 15 samples. Subsequently, Wakefield and co-workers reported 10 different ITS types in samples obtained from 24 episodes of AIDS related pneumonia, designating these types as subtypes according to proximity to the original types reported by Lee (e.g. A 1 b 2 ) (201) ( Table 3 ). When more ITS types were found, this typing scheme was extended to include 11 ITS1 and 17 ITS2 types (Wakefield, personal communication, Table 3 and Table 4 ). However, an alternative nomenclature of typing was proposed when several new ITS sequence variations were found in a comprehensive study of samples from 9 countries (104). In this revised typing scheme, ITS1 is classified into 15 genotypes, designated as type A through O, and ITS2 is classified in 14 different types, designated a through n (Table 3 and 4). Additional ITS1 genotypes and ITS 2 genotypes have since then been reported in specimens from US, Italy and Japan, which brings the total number of reported genotypes to approximately 27 different ITS 1 genotypes and 35 different ITS2 genotypes (77, 159).

ITS genotyping is performed by single or nested PCR amplification of the rRNA operon with primers spanning the ITS1, 5.8S and ITS2 region (II, V). Subsequently the nucleotide sequence can be analyzed by hybridization and/or DNA sequencing. In most early studies the genotypes were determined by direct sequencing; however this method may preclude the detection of multiple types in specimens with more than one genotype. In order to detect multiple ITS types in a specimen, the primary PCR product is usually cloned and the clones sequenced.

In our studies, mixed infections (more that a single ITS genotype) have been detected in 23% of all specimens, in other reports mixed infection have accounted for 17-77% of samples (17, 72, 100, 201, V). Because ITS in Pneumocystis is a single copy gene, mixed infection indicates that a specimen contains more than one genotype.

In our first study of ITS typing we used a combination of hybridization and cloning to analyze ITS sequences (V). The hybridization was done by type-specific probing with oligonucleotide probes ITS1A, ITS1E, ITS2b, ITS2e, and ITS2m to determine whether the specimen contained a single or multiple types of P. jiroveci (114). The specimens that contained a single type of P. jiroveci were sequenced and those that contained multiple types were cloned and then sequenced. Screening of recombinant clones that contained the correct insert was achieved by colony hybridization (86). A coinfection was assumed if, for one locus, the PCR product from a specimen hybridized to more than one oligonucleotide. During this work, we observed that probing or hybridization was ambiguous in a large part of the samples and in the end most samples had to be analyzed by cloning of PCR products and DNA sequencing. Following the recognition of an increasing number of new ITS sequence types and coinfections with multiple different ITS types, we have since this report performed ITS typing by cloning and DNA sequencing (VI).

Although potential ITS1 and ITS2 combinations are numerous, the number of ITS combination types observed is more limited. Presently, approximately 65 different ITS genotypes have been reported worldwide (Table 2) and in Denmark 52 different types (V, VI). Table 3-4 illustrate the difficulties in assigning an observed genotype to both classifications, because certain genotypes have only been reported by a single nomenclature system and because the classification by Wakefield has fewer nucleotide scoring positions than the Lee classification.

4.3 Other typing methods: SSCP and UCS repeats

An alternative typing method is single strand conformation polymorphism (SSCP) analysis. In SSCP, nucleotide polymorphisms in PCR fragments are detected by the resulting changes in secondary structure by nondenaturing gel electrophoresis and silver staining or radiolabelling. SSCP analysis of P. jiroveci has primarily been investigated by Hauser and co-workers (72, 73, 153). Four different genes are analyzed by SSCP (Table 2), with identification of different SSCP patterns for each gene. A total of 35 different SSCP patterns were reported in samples obtained from 212 European PCP patients, with 77% of isolates being infected with two or more and 23% with a single SSCP type (72). The suggested advantages of the SSCP method are a high discriminatory power and the ability to detect a high number of coinfections. Limitations of the method are that only relatively short PCR regions can be assessed, that the sensitivity of detecting single nucleotide polymorphisms is lower than sequencing, and that the SSCP patterns generated are dependant on the individual setup (e.g. temperature) in each laboratory. So far, the use of SSCP by other research groups have been limited (125).

More recently a novel typing method has been described based on variation in tandem repeats in upstream conserved sequence (UCS) of the MSG expression site (128). Using this typing method 5 different pattern of UCS repeats was reported among 147 isolates, of which 43% of isolates were coinfected with multiple UCS repeat patterns (Table 2).

5.1 Epidemiology of P. jiroveci genotypes

The finding of geographical clustering and differences in frequency of PCP between different countries has been debated. Differences in reported PCP incidences are difficult to interpret because of geographic variations in access to diagnostic procedures (e.g. BAL verified PCP versus tentative diagnosis), the number of patients at risk (i.e. CD4+ lymphocyte count <200 cells/ μ L) and the epidemiology of other opportunistic infections (i.e. TB). In a European cohort study of more than 5000 patients with AIDS, the adjusted relative risk of developing PCP was lower (RR: 0.77, 95% CI: 0.6-0.99) in patients from southern Europe compared to Northern Europe and in intravenous drug abusers compared with homosexuals (RR: 0.71, 95% CI: 0.6-0.9) (122). In two retrospective studies, differences in the risk of PCP were found in different zip code areas of San Francisco and Cincinnati, suggesting geographical clustering (48, 151). In a study of Pneumocystis specimens obtained from five US cities, the genotype patterns of mtLSU and DHPS genotypes varied by patients' place of diagnosis but not by place of birth (17). In contrast, only low genetic diversity was observed in a study of mtLSU genotypes among samples obtained from United Kingdom, United States, Zimbabwe and Brazil (210).

To investigate the molecular epidemiology of infection we genotyped samples obtained from Danish HIV-1 infected patients during a 7 year period (V). ITS 1 and 2 genotypes were determined in 162 bronchoalveolar lavage samples obtained from 130 patients between 1989 and 1996. By type specific hybridization and sequencing of subclones, we observed a remarkably complex picture of ITS types. In 23% of samples mixed infection with more than one ITS subtype was present and as many as 7 different types were detected in a single sample. A total of 49 different ITS genotypes was detected, however 4 types (Eg, Ne, Eb and Ai) accounted for almost half of the sequence types detected ( Table 5 , 6). We found no specific overall changes in occurrence of ITS genotypes during the study period, no evidence of clustering of specific genotypes, and no link between ITS genotypes and demographic variables or season of the year (V).

Quite similar patterns of genotype distributions have been found in most studies published regardless of geographical origin ( Table 6 ). The majority have reported that the same genotypes, in particular ITS type Eg, Ne and Eb are common, whereas the remaining ITS types are detected more infrequent (Table 6). In most studies, 5 or fewer ITS type's accounts for more than half of all subtypes. These findings suggest that certain strains of P. jiroveci across the world may be relatively conserved. Furthermore, there is additional evidence, which support an absence of genotype changes over time: Very similar genotypes were found at the ITS, mtLSU rRNA, arom, and the mitochondrial small subunit rRNA loci, when archival PCP specimens obtained before the HIV pandemic from 1968 to 1981 were compared with current specimens from AIDS patients (200). A number of the infrequently detected ITS types have only been reported in single studies and could be specific for geographical location. However, in view of the high number of different ITS types reported and the problems of interpretation of mixed infections it remains to be established whether certain sequence types are characteristic of a certain geographical location or just coincidental.

Our findings are in line with the largest study of SSCP genotypes. Hauser et al studied SSCP genotypes in 212 European patients and found no association between SSCP type or number of co-infecting types per patient and geographical location, year of collection, sex, age, or HIV status (72). In agreement with our findings only few patients were infected with the same genotype during a 6 month period.

5.2 ITS genotypes and prognosis

The association between strains and morbidity of PCP has been studied by ITS genotyping of P. jiroveci . A connection between ITS genotypes and severity of PCP was initially suggested by Miller and Wakefield (145). They compared 11 HIV-positive PCP patients with mild disease with 15 patients with moderate or severe disease based on a baseline PaO 2 cut-off value of 9.3 kPa. The ITS sequence type A2c1 (corresponding to type Ai) was more often found among patients with moderate to severe disease whereas B2a1 (type Ne) was more prevalent among patients with mild disease. Further type B2a1 was the most frequently type identified in samples from second and third episodes of PCP, whereas B1a3 (type Eg) was found in first episodes but never in subsequent episodes of PCP. This result suggested that B2a1 (type Ne) is more persistent or transmissible.

To investigate the possibility of virulence associated with specific ITS genotypes, we analyzed clinical data in relation to ITS genotyping in our cohort of 130 patients (V). In contrast to the findings summarized above, we were unable to show any correlation between specific ITS types and clinical outcome, specifically there was no association between ITS type and the severity of PCP by using the same grouping based on PaO 2 . We found no evidence of a link between specific ITS types and primary or recurrent episodes of PCP.

5.3 Mode of infection: Reactivation
vs. reinfection

For many years, it was believed that clinical disease was a result of reactivation of latent infection carried from childhood. This view was primarily based on serologic data. The majority of children develop anti- pneumocystis antibodies during childhood, and seroconvert before four years of age, with an uniformly high prevalence of antibodies to the infection among adult populations in different countries (118, 120, 166, 169, 214).

From the early nineties, however, several data opposing this theory emerged. In animal experiments, airborne transmission of P. carinii was demonstrated in the rat model of the infection with susceptible, immunosuppressed animals acquiring the infection from infected animals, either housed in the same cage or in an adjacent cage in the same isolator unit (80). In addition, immunocompetent mice were transiently colonized with P. carinii after close contact with P. carinii -infected SCID mice and were then able to transmit the infection to P. carinii -free SCID mice that develop PCP (49). In both rats and SCID mice with PCP there was no persistence of latent organisms after treatment of pneumonia and immunreconstitution (34, 203).

Among children with perinatally acquired HIV, the incidence of PCP is highest between 3 and 6 months of age (174, 182). In contrast, there is little evidence of clinical illness or specific symptoms in connection with the primary infection in immunocompetent children. In a prospective study of 74 immunocompetent children less than two years old, Vargas et al were able to detect P. jiroveci DNA in nasopharyngeal aspirates of 32% by nested PCR, even though most of the children were healthy at time of detection (203). In this study, the maximum length in which DNA could be detected was 6 months, with 85% of children developing antibodies at the age of 20 month. Autopsy studies suggest that the clearance of organisms is complete, with no detectable organisms by microscopy or PCR in post-mortem lungs from immunocompetent patients (142, 167).

In HIV-infected PCP patients, the clearance of organisms is slow but appears also to be complete with proper treatment and chemoprophylaxis. Residual cysts can be detected by microscopy in 63-90% at the completion of 3 weeks of therapy, but has no correlation with subsequent risk of relapse or death in patients receiving chemoprophylaxis after treatment (160, 173, 181). At four to six weeks after therapy organisms can be detected in 25% of patients, but are eventually cleared 3-4 months after the completion of therapy (56). Further, discontinuation of secondary PCP chemoprophylaxis in patients who responds to HAART therapy is safe, without evidence of recurrent PCP (103).

5.4 Transmission of PCP

Although animal studies suggest that the principal infection route is by air (32,80), the exact mode of transmission remains unknown. Two major modes have been proposed, either transmission from (a yet unknown) environmental source by airborne spores or transmission directly person to person.

A possible environmental source of infection is supported by detection of P. jiroveci DNA in air filters from hospital, urban and rural settings (14, 162, 208) and in pond water (30). In one case-control study, recent gardening, hiking or camping was associated with a greater risk of PCP (157), which hypothetically could support soil as a source of infection, equivalent to other airborne fungal infections from soil like Coccioides immitis, Histoplasma capsulatum or Blastomycosis hominis . Further support comes from other studies. 1) In a P. carinii rat model, airfilters carried an infective form of the organism, which retained infectivity after several months at -80°C to room temperature, suggesting the presence of a long term viable spore-like form (88). 2) During the eighties, the relative proportion of AIDS patients with PCP remained relatively constant whereas the overall number of AIDS patients increased rapidly (123). Further, no significant monthly variation in cases of PCP occurred in our cohort from 1989-2002 ( Figure 4 ). 3) No difference in risk of contacting PCP has been found associated with living in cities compared to rural areas (123). These observations all suggest an environmental source of infection, because a more epidemic spread of PCP would have been expected if direct transmission between patients were the predominant mode of infection.

5.5 Case clusters of PCP

One observation that may support a direct mode of transmission is reports of PCP "outbreaks". Apparent clusters of PCP were first observed in malnourished children during World War II (63). Since then there has been sporadic observations of mini-epidemics over the last 30 years. Most reports of possible outbreaks have been reported in immunosuppressed children with malignancies and transplant recipients with the number of patients in each cluster ranging from 2 to 19 (II).

We studied three apparent outbreaks of PCP among patients with haematological malignancies and with HIV-infection by genotyping of the ITS locus (II). Eight patients were diagnosed with PCP during a three-month period at a haematological unit in Copenhagen. Six patients were from two independent clusters of PCP observed in a HIV/AIDS unit in London. Although a common ITS sequence type was observed in two patients who shared a room in the haematological unit and also in two of the HIV patients, there was no indication of a particular ITS type being responsible for either of the clusters. Nine different ITS sequence types were observed in samples from the patients with haematological malignancies, and two and three different sequence types, respectively, were found among the patients with HIV. In four patients more than one ITS genotype was detected. Similar observations were made in a Swedish study of three PCP outbreaks, in which no evidence of transmission of mtLSU genotypes were found and in a French study of three different couples diagnosed with PCP (102, 161). In contrast, similar genotypes at the mtLSU, DHPS and ITS locus was observed in a case report of a HIV-1 infected mother and 5 week old infant who developed PCP at the same time (143).

The interpretation of case-clusters is not straightforward. First, there are inherent difficulties in assessing exposure retrospectively. Second, the incubation period of infection is not definitively known. In most case-clusters the presumed incubation period of infection have ranged from two to 12 weeks, which is in accordance with observations in rodent models (23, 32). However, recent findings in SIV infected monkeys (22) suggest that the time of incubation may be much longer. Third, there is a risk of publication bias. Although the majority of PCP cases are seen in HIV infected patients, almost all case-clusters have been described in transplant centres among non-HIV patients in which the "outbreaks" may merely represent random fluctuations in number of PCP cases.

Indeed, a recent case control study among HIV patients found no association between exposure to persons with PCP and subsequent development of PCP (222). In conclusion, although apparent outbreaks have been described, this evidence of direct transmission is so far not conclusive.

5.6 Exposure to PCP in health care settings

The possibility of person to person transmission has been examined by serological and PCR studies of health care workers (HCW) exposed to patients with PCP. Serological studies are conflicting; although Leigh and colleagues observed higher antibody titres in HCW exposed to AIDS patients compared to workers without such exposure, no evidence of elevated titres or of detectable DNA was found in a similar study of HCW by Lundgren et al (107, 117). In contrast, recent PCR studies do provide some evidence of person to person transmission. Vargas et al identified P. jiroveci DNA in three healthy contacts to a child with PCP (204) and Miller et al's observed an elevated an elevated risk of P. jiroveci DNA carriage in HCW with close contact to patients with PCP (144).

5.7 Recurrent PCP

The possibility that reinfection rather than reactivation of latent disease is the cause of PCP has been supported by a number of studies on recurrent PCP ( Table 7 ). They observed a change of genotype in approximately half of patients with recurrent separate episodes of PCP. We found a similar rate of genotype change in our study of ITS types (6 of 10 patients, who had recurrent PCP, Table 7). However, we also found almost the same rate of genotype switch when two consecutive BAL specimens from the same episode were compared (10 of 19 patients, 53%). The high rate of genotype change during single episodes observed in our study is in contrast to previous reports. Keely and colleagues (91-93) found the same mtLSU rRNA genotypes in four patients, from whom specimens were taken at the initiation and completion of therapy. ITS genotyping for these specimens was not reported. Latouche et al (100, 101) observed the same mtLSU and ITS genotype of P. jiroveci in 10 patients, in whom sequential samples from the same episode of pneumonia were typed by direct sequencing. The reason for the discrepancy between the reported results is unclear, but might be explained by differences in sample size or related to differences in methods.

The significance of genotype switches in patients who suffer multiple episodes of PCP is therefore unclear. One possibility is that patients are eradicated of P. jiroveci , but are re-infected. An alternative explanation is that treatment does not eliminate the infection, but that P. jiroveci remains in a latent state, emerging later to cause disease when the immune system or prophylactic treatment fails. The principal evidence for re-infection is the observation of different genotypes during different disease episodes. Although on the face of it, such genotype switching would seem to indicate re-infection; several factors could conspire to generate apparent switches in a patient whose infection was caused by the activation of latent organisms. Among these factors, the prerequisite and dominant one is the presence of multiple genotypes in a patient. If mixed infections occur, different genotypes might be detected during different episodes as a result of sampling from different parts of the lung during different episodes (VI).

5.8 Heterogeneity and compartmentalization of P. jiroveci genotypes

To investigate the influence of mixed infections and variability of genotypes we undertook a study of genotype heterogeneity in autopsy lungs (VI). Polymorphisms at three genetic loci were investigated by analysis of DNA obtained from different segments of the lungs. Of these loci, mtLSU and ITS were chosen to provide high discrimination. The third locus, DHPS, was chosen because of its association with sulpha resistance (III). From each lung segment five recombinant PCR clones were sequenced to detect coinfections with different strains.

In two of three patients a mixture of different mtLSU and ITS genotypes was observed in BAL samples and in the autopsy subclones. In the lungs, compartmentalization of specific ITS and mtLSU sequence types were observed in different lung segments. Although the most predominant genotypes detected in the diagnostic respiratory samples were also the most commonly detected in the lung segments, minor populations of ITS genotypes were detected in the lung but not in the respiratory samples of two patients.

The results of this study were controversial when published. In previous studies, a tacit assumption was that genotypes detected in respiratory samples reflected the global infection within the lung. However, recent data have confirmed the existence of genotype heterogeneity. In samples taken from different site of a P. jiroveci infected human lung the sterol profiles had large variations in pneumocysterol (8). These intrapersonal variations were similar to differences observed between different patients, suggesting distinct phenotypic populations. Uneven distribution of two ITS types by analysis of different segments from a child autopsy lung was observed by Ambrose et al (144). They were unable to detect both genotypes in every lung segments. Ma et al observed genotype differences in UCS repeat patterns between sputum and BAL samples as well as between right and left BAL samples obtained from the same patient (128). Furthermore, studies which have employed rigorous and detailed subcloning of loci with high variability like ITS or used a combination of typing methods, have generally reported a higher incidence of mixed infection compared to studies using direct sequencing of relatively invariant loci.

5.9 Interpretation of molecular typing

The demonstration of cluster cases and genotypic switching in recurrent episodes of PCP can be interpreted to support the hypothesis of nosocomial transmission. Based on these findings, some authors have recommended prevention of PCP by isolation of PCP patients to limit the exposure of immunocompromised patients (31, 91, 216). Presently, however the exact mode of transmission remains unknown. In our genotype study of two presumed outbreaks of PCP, we found no evidence of direct transmission being a primary cause of infection. Although experimental and epidemiological observations suggest that pneumonia may sometimes be caused by transmission from one patient to another, this mode of transmission does not seem to play a definite part in the majority of cases. Genotype data supports that infection with P. jiroveci is not clonal, that coinfections with multiple genotypes may be common and that in some cases distinct populations of organisms may reside within the lung. These observations have important implications for interpretation of genotyping and suggest that single locus typing cannot reliably distinguish between reactivation of latent infection and de-novo infection.

The interpretation of genotype data and in particular ITS sequence types, should be cautious since genotyping performed on respiratory samples cannot a priori be assumed to represent all genotypes present within the lung (as discussed in section 5.8). The broad diversity of types and the absence of genotype clusters could seem to indicate that P. jiroveci is ubiquitous and that multiple sources of the pathogen may co-exist. In view of the specificity of P. jiroveci for humans, the data suggest a complex picture of transmission. It is likely that healthy carriers, in particular infants, may constitute an important natural reservoir of P. jiroveci organisms. Transmission may happen both as a result of transmission of airborne organisms between immunocompromised patients and as a result of transmission from transient low level asymptomatic carriage in immunocompetent hosts. Further, both reactivation and re-infection remain possible sources of infection. The empirical rational for isolating PCP patients is therefore limited. To interpret DNA typing of the organism properly, more knowledge of the life cycle, infectivity and reproductive mode is required. In future, the development of new genotyping methods will have to examine the degree of genotype variation within human lungs and should preferably be based on multilocus sequencing in order to reliably distinguish strain differences.

6.1 Conventional methods

The standard method for diagnosis of PCP relies on the microscopic visualization of P. jiroveci organisms in respiratory samples. Because very few organisms are present within the upper respiratory tract, lower respiratory secretions are usually necessary for definitive diagnosis. Bronchoalveolar lavage (BAL) combined with colorimetric and an immunofluorescent stain of BAL fluid is considered the method of choice with sensitivity and specificity in excess of 95%. An alternative is examination of material obtained by induced sputum. However, the sensitivity of this method is more dependent on the experience of the personnel performing the procedure and evaluating the samples, with high variation in the diagnostic sensitivity reported (between 50 to 90%) (55, 79, 97, 141). Staining of lung biopsy specimens is sometimes needed to definitively diagnose PCP pneumonia in non-HIV infected immunocompromised patients, due to the fact that these patients in contrast to AIDS patients have smaller number of organisms.

6.2 PCR methods

Because BAL is an invasive and uncomfortable procedure, which is not without risk, alternative non-invasive test have been evaluated. PCR diagnosis of P. jiroveci was introduced experimentally in 1990 when Wakefield et al first described DNA amplification of the multicopy mitochondrial ribosomal RNA gene (211,213). Since then, several studies have investigated PCR detection of Pneumocystis DNA in respiratory samples from HIV-1 positive and negative patients. To increase the diagnostic sensitivity of non-BAL specimens, different PCR methods have been evaluated, including different gene targets and the use of nested PCR, in which a second round of PCR reamplifies the primary PCR product.

6.3 Non-invasive diagnosis

Upper respiratory specimens usually present insufficient numbers of organisms to enable diagnosis by microscopy, but the sensitivity of PCR analysis permit the detection of organisms below the threshold of microscopy. Even oral washes, which are easily obtained by rinsing and gargling the mouth with normal saline, are suitable for PCR diagnosis. Wakefield et al were the first to study the diagnostic ability of oral wash combined with PCR; A diagnostic sensitivity and specificity of 78% and 100% was observed by using single round PCR in a study of 31 HIV-1 infected patients (185, 212).

We tested an improved PCR method for diagnosis on oral wash and BAL fluid in a prospective study among HIV-positive patients (I). PCR on oral wash specimens, PCR on BAL and microscopic examination of BAL was compared by the use of two different PCR protocols. The principal objective of the study was to develop a routine PCR method that could generate a diagnostic answer within one working day. A single-round Touchdown PCR (TD-PCR) protocol with the ability to detect PCR inhibition was developed. The TD-PCR was evaluated in a routine diagnostic laboratory and was compared to a previous PCR protocol, run in a research laboratory (211).

Sixty-one paired oral wash and BAL specimens were evaluated by TD-PCR. To detect PCR-inhibition, which often is a problem in respiratory specimens, a lambda DNA internal process control was added to the same reaction tube as the clinical specimen. The main findings are shown in Table 8 . Using the same method, we found equivalent diagnostic accuracy in a study of 26 patients with haematological malignancy (76). Oral wash specimens were easy to obtain even among severely ill patients with high risk of haemorrhage and Pneumocystis DNA could be detected in all 8 patients with PCP and in none without PCP.

Since then other investigators have investigated oral wash for PCP diagnosis (section 6.4, Table 13 page 29). In the largest published study, in which 175 paired samples were investigated, sensitivity and specificity similar to our results were reported with a positive and negative likelihood ratio of 14.4 and 0.1, respectively (60).

6.4 PCR diagnosis of PCP, systematic review

More than 20 studies which investigated the diagnostic use of PCR have now been published. Comparisons of these studies are difficult because of differences between studies in patient groups, PCR methods and study design.

To summarize the current literature a systematic review of diagnostic accuracy was done. The results are shown in Figure 5 , Figure 6 , Figure 7 , Figure 8 , Table 9 and Table 10 (unpublished). Studies were identified through Medline and data extracted to diagnostic 2¥2 tables (47). When possible, the diagnostic accuracy was calculated according to patients with clinical PCP and a systematic review of likelihood ratios was performed (25, 43). Likelihood ratios were chosen for comparisons as these values have advantages to traditional predictive values (Box 1) (64). They are defined as the likelihood that a particular test result would be found in a patient with the target disorder, relative to the likelihood of the same result occurring in a patient without the target disorder. As shown, differences were observed between the studies according to respiratory specimen used for diagnosis (BAL vs. sputum/oral wash) and PCR methods (single PCR vs. nested PCR). In most studies, which assessed PCR analysis of BAL fluid, sensitivity and specificity ranged from 90 to 100%. In studies using BAL fluid, although there is significant heterogeneity among studies, the overall pooled estimate (random effects model using the method of DerSimonian and Laird with a continuity correction of 0.5) of the positive likelihood ratio of a positive test is 9.9 (95% CI, 6.4-15.1) and the overall likelihood ratio of a negative test 0.06 (95% CI, 0.03-0.11).

In comparison, even greater variations in diagnostic accuracy have been reported when upper respiratory specimens are used for PCR diagnosis. Although the likelihood ratio of a positive test PCR when used on upper respiratory samples is overall acceptable (13.9, 95% CI: 8.2-23.7) and has a sensitivity comparable to microscopy of induced sputum (87%, 95% CI: 81-94%), the overall likelihood of a negative PCR test in these samples is not optimal (0.17, 95% CI: 0.10-0.28). The significant heterogeneity among studies (range: 0.03-0.53), reflect a relatively high rate of false positive results in some studies, in particularly those in which nested PCR were used. Similar conclusions can be drawn if only PCR studies examining oral wash are analysed. For these studies, the overall likelihood of a positive test is 14 (95% CI 8.2-23.7) and the overall likelihood of a negative test is 0.23 (95% CI 0.13-0.41).

6.5 Sources of heterogeneity in diagnostic PCR studies

Figure 5 to 8 visualises the remarkable heterogeneity in the reported diagnostic performance of PCR for diagnosis of PCP. There may be several reasons for the variability between studies, but from the literature and our own experience it is apparent that important sources of heterogeneity are:

Characteristics of patient population:

1. Higher diagnostic value of PCR studies among HIV-positive compared to HIV-negative patients.

2. Higher prevalence of P. jiroveci carriage among immunosuppressed patients, in particular HIV-negative, compared to HIV-1 infected and immunocompetent individuals.

3. Chronic pulmonary comorbidity e.g. COPD may enhance carriage of P. jiroveci .

4. Differences in use of PCP prophylaxis or treatment before obtaining the diagnostic specimens, in one study 82% of patients with discrepancy between PCR and staining had received prophylaxis prior to the diagnostic procedure (171).

PCR method performance:

1. Quantity of DNA recovered. Less P. jiroveci DNA is present in upper compared to lower airway samples.

2. PCR inhibition. Respiratory specimens often contain inhibitors of PCR, in a Portuguese study of oral wash, PCR inhibition was noted in 17% of PCR negative samples (136).

3. PCR target. Higher sensitivity of using multicopy compared to single copy genes (60).

4. Single versus nested PCR. Overall, nested PCR has higher sensitivity, but also higher risk of false positives findings (185).

Study methodology:

1. Differences in the definition of PCP. Although most studies have used microscopy to define PCP, the definition of "gold standard" was in some studies more ambiguous and included clinical signs or response to treatment. Such discrepant analysis may be biased with overestimation of sensitivity and specificity (146).

6.6 False positive PCR results: colonisation?

Discrepant results between PCR and microscopy with "false positive" PCR results were already noted in the first studies investigating PCR for diagnosis of PCP. At this period of time it was believed that asymptomatic colonisation was rare. Later, several PCR based studies confirmed the existence of subclinical infection in both immunosuppressed and immunocompetent individuals. However, the extent of colonisation has been debated with marked variation in the rates of P. jiroveci carriage reported. In a study preceding the use of effective PCP prophylaxis, P. jiroveci DNA was detected in induced sputum by nested PCR from eight asymptomatic HIV-infected patients with CD4 counts less than 200 cells/ μ l, of which five later developed PCP 164 to 352 days after (54). Nevez et al. reported that 20% of 169 HIV-negative patients with moderate or severe immunosuppression were positive by nested PCR; none of these patients developed PCP (158). Olsson et al and Takahashi et al reported a PCR positive rates of 28% and 20%, respectively, among HIV-negative immunosuppressed patients without definite or probable PCP (163, 194). Armbruster et al noted that 6% of 77 immunocompetent HIV-negative patients were colonized (10) and Sing et al. found by nested PCR that 17-20% of immunocompetent patients with primary pulmonary disorders were nested PCR positive (185). None of these patients developed PCP (184). These observations suggest that colonization with P. jiroveci is more common than previously thought, even among immunocompetent individuals.

In view of the conflicting reports regarding the rate of subclinical infections, we wished to evaluate the detection of P. jiroveci DNA in patients not primarily suspected of PCP (VII). In a case control study, respiratory samples from 367 patients were analyzed by PCR. For each PCR-positive case, four PCR-negative controls, randomly chosen from the PCR-negative patients, were matched by sex and date of birth. P. jiroveci DNA was detected in 16 (4.4%) of patients. Main findings were a higher rate of chronic or severe concomitant illness (94% vs. 50%, OR: 15, 95% CI: 2-651) in the PCR positive patients compared to controls and a higher rate of corticosteroid exposure in cases compared to controls (75% vs. 13%, OR: 21, 95% CI: 5-106). Detection of Pneumocystis -DNA was associated with a worse prognosis. Seven (44%) of the patients with positive PCR died within one month compared to nine (14%) of the controls (p=0.01). None of the nine PCR-positive patients who survived had developed PCP at one year of follow-up. However, because of the retrospective study design, in which autopsies were not done, it is unknown if the detection of P. jiroveci is an independent predictor of death or merely a marker of severe illness.

Whether the differences in reported rates of P. jiroveci detection result from population based differences in P. jiroveci carrier rates or stems from differences in PCR methods is unknown. PCR is extremely sensitive and will detect both living and dead organisms. In our study, the risk of false positive reactions was reduced by a strict definition of PCR positives. Samples were only considered true PCR positive if three different P. jiroveci gene targets could be amplified. Other studies (158, 184, 185) only requested amplification of a single gene with nested PCR, which may have increased the risk of false positive results.

The low rate of PCR positivity we observed is comparable to other observation studies. Visconti et al (205) reported that 2.5% of 78 immunocompetent patients were positive by nested PCR. In a Danish study, 4.7% of 1762 consecutive lung biopsies at autopsy were microscopically positive for P. jiroveci (180). Oz et al (164) found no P. jiroveci DNA by single round PCR in any of 258 upper respiratory tract specimens from 86 immunocompetent individuals in the absence of microscopy verified PCP. Furthermore, Lidman et al's (108) found no evidence of P. jiroveci DNA in IS samples from 19 health care workers.

6.7 Interpretation of PCR results

Similar to other diagnostic tests, it is essential that the diagnostic accuracy of an applied PCR test is well defined not only in terms of technical performance but also in terms of diagnostic performance in different patient populations. The available data from several PCR studies suggest that while a negative PCR test on both invasive and non-invasive samples has reasonable power to rule out PCP, a positive PCR test must be carefully interpreted in the context of clinical findings. Single round PCR is to be preferred for diagnostic purposes, since nested PCR has increased risk of false positives results and only a limited increase in diagnostic sensitivity. As shown, comparison of the data from different clinical studies is complicated due to differences in patient populations, and PCR methods. In addition, test performance obtained in research studies may not be achieved in real-life routine settings. In the absence of clinical features typical of the pneumonia, the presence of P. jiroveci DNA in upper respiratory tract specimens may indicate harmless colonization of the respiratory tract rather than PCP. A single PCR result must always be viewed with caution and interpreted in the context of the patient's symptoms, the clinical findings and the results of other procedures for detection of other pathogens. Continuous education of clinicians in the proper use and interpretation of PCR-generated results, as well as in the limitations of the tests, is therefore of importance. Currently, our PCR method (I) is available as a routine test at Statens Serum Institute, Copenhagen.

A potential and serious drawback of PCR diagnosis is the risk of missing the diagnosis of other respiratory diseases, if BAL is not performed. In many immunocompromised patients a wide range of infectious or malignant diseases present with symptoms and signs, which are indistinguishable from PCP. Although clinical and laboratory findings such as exertional hypoxemia, interstitial infiltrates, oral candidiasis or elevated LDH may suggest PCP in an HIV infected patient, the specificity of these signs is low as the same signs may result from other infections. Differential diagnosis therefore usually demands microscopic examination and culture of respiratory specimens. Importantly, even when PCP is confirmed, concomitant pulmonary infections may be present. In our cohort; bacterial coinfections were detected in 20% of AIDS-associated PCP (19). Hence, PCR diagnosis should be viewed as a supplement and not an alternative to standard tests.

Oral washes are quickly and noninvasively obtained, which is of particularly importance in patients unable to sustain BAL. Compared to nasopharyngeal aspirates or induced sputum, oral washes have equivalent diagnostic sensitivity and are much less unpleasant. Although similar sensitivity may be achieved by experienced microscopic testing of induced sputum, PCR test on oral washes offer a sensitive test to hospitals without immediate access to experienced histological evaluation. Furthermore, PCR may be used to confirm a diagnosis of PCP in cases of ambiguous microscopic findings.

The development of sensitive and specific, standardized, and commercially available amplified nucleic acid tests for diagnosis of PCP would be highly desirable. Further developments are needed to distinguish between colonisation and clinical disease. Potentially promising methods are real-time quantitative PCR and detection of P. jiroveci mRNA. Recent research suggest that organism load as assessed by quantitative PCR is higher in cases of PCP compared to cases of colonisation (99). If these findings are confirmed and the method further developed, quantitative real-time PCR could well be a rapid and specific test for PCP. In addition, viability of P. jiroveci organism may be tested by study of Pneumocystis mRNA, which could offer a potential method for monitoring of treatment (130).

7.1 Survival

Untreated PCP among HIV patients is almost always fatal. In the beginning of the HIV epidemic the mortality rate of PCP was reported to be 30-40% (26, 87), increasing to 70% among patients with progression to respiratory failure (152). Following improvements in the care of PCP patients, including earlier recognition of the infection and in 1990 the introduction of adjuvant corticosteroids, mortality rates fell to 10-20% (16, 21, 36, 57, 121). In the US, an 18% one-month mortality was observed among 4412 PCP cases from 1992 to 1998 (50). From 1989 to 2002, we observed an overall one-month mortality of 15% among 241 HIV infected patients (unpublished results). By multivariate Cox regression analysis, low PaO 2 at admission, age, initial antimicrobial therapy other than TMP-SMZ, and a positive BAL CMV culture were independent factors associated with a decreased 3-month survival (19,74). In addition, the degree of lung inflammation is an important prognostic factor with BAL neutrophilia and elevated immune markers such as IL-8 associated with an increased risk of death (18, 20).

7.2 Treatment of PCP

Three weeks of sulfamethoxazole in combination with trimethoprim (cotrimoxazole, TMP-SMX) was introduced for treatment of PCP in 1978 and has since remained the standard of treatment (75, 82). Although toxicity from this regimen is common, comparisons with alternatives such as pentamidine, dapsone-trimethoprim, clindamycin-primaquine or atovaquone have shown superior or equivalent effect with equivalent or less toxicity (75). Among AIDS patients, failure of responding to primary treatment with cotrimoxazole has been observed in 10 to 40% of cases (24, 70, 96). In our cohort, primary anti-PCP treatment was changed in 60 (24%) of 256 consecutive episodes of AIDS-associated PCP between 1989 and June 2002 (74). Toxicity was the cause of treatment change in 27 (11%) patients, which occurred after a median of 10 (range 1-17) days. Suspected treatment failure was the reason for change in 33 (13%) patients, with treatment change after a median of 18 (range 5-34) days. Although variations in the reported rates of treatment failures may reflect differences in the definition of treatment failure or threshold for changing therapy, it is evident that a substantial number of PCP cases fail to improve, even under optimal therapy. Indeed, the mortality of PCP related respiratory failure which require ventilatory support has not changed during the last 10 years (13).

7.3 PCP prophylaxis and failure of prophylaxis

In the beginning of the nineties, chemoprophylaxis with low dosage of cotrimoxazole was shown effective in reducing the risk of PCP among patients at risk (177). Among both HIV-infected and HIV-negative patients the risk of developing PCP is closely correlated to level of CD4 count, with most cases of HIV-associated PCP occurring at CD4 counts less than 200 cells/ μ l, and non-HIV associated PCP cases at less than 300 cells/ μ l (131). After a episode of PCP, the rate of relapse is 50% within sex months if no prophylaxis is instituted (148). Primary and secondary PCP prophylaxis is recommended for all HIV-1 infected patients with CD4+ lymphocyte counts less than 200 cells/μL and can be safely interrupted if stable CD4 counts over this level are reached as a result of combination antiretroviral therapy (94,103). Use of cotrimoxazole is associated with a number of adverse effects (24, 71, 176, 177). Before combination antiretroviral therapy became available, it was noticed that after 13 months of cotrimoxazole prophylaxis half of the patients that had started would have experienced an adverse event, in particular rash, and after 3 years half would have switched to other types of prophylaxis (24). For patients who are intolerant to cotrimoxazole, dapsone, another sulfone drug, is a common choice of therapy (170). The major source of prophylaxis failure is non-compliance. Although there has been considerable variation in the reported rates of cotrimoxazole failures, breakthrough in a patient who is compliant with cotrimoxazole prophylaxis is uncommon (95, 115, 177). However, documented cases of breakthroughs have been reported in patients apparently compliant with cotrimoxazole prophylaxis (7, 53, 58). In a recent African study, 33% of infants developed PCP in spite of prescribed prophylaxis (129). In addition, failure of non-co-trimoxazole prophylaxis is well documented, even in compliant patients. In open-label, randomized studies among HIV patients failure rates for dapsone and inhaled pentamidine ranged from 0.32 cases to 2 per 100 patient months of prophylaxis (68, 81, 134, 188, 219). Further, among bone marrow transplant recipients the relative risk of developing PCP was 18.8 with dapsone compared to cotrimoxazole (188).

Failure of prescribed PCP prophylaxis may occur because of poor adherence to treatment, insufficient drug levels or possibly as a consequence of sulfonamide resistance. In view of the high number of HIV-patients continuously exposed to low dose cotrimoxazole, there have been concerns that sulphonamide resistance may have developed in P. jiroveci .

7.4 Sulfonamide resistance and DHPS mutations

The action of cotrimoxazole is mediated by inhibition of two key enzymes in folic acid synthesis ( Figure 9 ). Sulfamethoxazole (SMX) and dapsone targets the enzyme dihydropteroate synthase (DHPS), which catalyzes the formation of dihydrofolate from para-aminobenzoic acid. Sulfonamide is a structural analog of p-aminobenzoic acid (PABA), the substrate of the DHPS enzyme, and inhibits it competitively. Trimethoprim inhibits dihydrofolate reductase (DHFR), which catalyses the formation of tetrahydrofolate from dihydrofolate. The combination of trimethoprim and sulfamethoxazole is thought to have a synergistic effect on bacteria, however in experimental models of Pneumocystis the effect is almost entirely due to sulfamethoxazole (217).

Resistance to trimethoprim-sulfamethoxazole has been extensively documented in bacterial and parasitic infections (84). In San Francisco, the use of cotrimoxazole PCP prophylaxis increased cotrimoxazole resistance among all among isolates of Staphylococcus aureus and 7 genera of Enterobacteriaceae from 6.3% in 1988 to 53% in 1995 among HIV infected patients (133). Similarly, widespread use of sulfa drugs for malaria and bacterial infection in Africa have produced high rates of resistance in P. falciparum and many bacterial species (59).

In E. coli, N. meningitidis, S. pneumonia and P. falciparum, point mutations in conserved regions of the DHPS gene confer sulfonamide resistance by decreasing the affinity for sulphonamides and sulfones (42, 193, 199). In Pneumocystis , DHPS is encoded by the folic acid synthesis (fas) gene, a multifunctional gene that encodes dihydroneopterin aldolase and hydroxymethyl-dihydropterin pyrophosphokinase (207). Lane et al were the first to report sequence polymorhisms in the DHPS gene of P. jiroveci in 1997 (98). They detected several sequence mutations which produced nonsynonymous amino acid changes. The mutations at position 55 and 57 are located in a region of the DHPS protein, which is highly conserved among different microbial organisms and different Pneumocystis species (124). Mutations at codon 55 to 60 are implicated in sulfa resistance in E. coli. S. pneumoniae and P. falciparum (4, 112, 199). Soon after the study by Lane et al, two small studies suggested an association of DHPS mutations with previous exposure to cotrimoxazole prophylaxis (90) and prophylaxis failure (139).

7.4.1 Methods for detection of DHPS mutations

In most studies, mutations in the DHPS gene of P. jiroveci have been detected by direct sequencing of PCR amplified DHPS obtained from clinical samples. However, the use of DNA sequencing for finding these mutations is rather expensive and time-consuming. Since DHPS mutations may be a potential cause of sulfa-resistance in P. jiroveci , the development of a simple method for the detection of these mutations was warranted. To this mean we investigated the use of a restriction fragment length polymorphism assay (RFLP) assay to detect the most common types of DHPS mutations (IV). By use of the restriction enzymes Acc I and Hae III, mutations in codon 55 and 57 can be detected. These enzymes cut in wildtype DHPS; in case of mutation at codon 55 or 57, there is no cleavage by Acc I or Hae III, respectively (Figure 9). We found that the RFLP assay correctly identified mutations when compared to direct sequencing of the DHPS gene in 27 isolates. The total turnaround time for the RFLP assay is approximately 8 h, which compares favourably with the 2 days required for sequencing. The method has been applied to study PCP samples in Portuguese HIV-patients (38), and a similar use of RFLP has been described by Roux and colleagues (175).

An alternative method for detection of DHPS mutations is SSCP. Ma et al investigated nested PCR combined with the SSCP method to detect mutations at codon 55 and 57 (126). This method had a turnaround time of 10 hours and was under optimized conditions able to detect the minority population (either wild type or mutant) if present in 10% of the total population, which suggest that SSCP is more sensitive than direct sequencing for detection of a minority population. A main limitation to the RFLP and SSCP methods are the inability to detect mutations other than those at codon 55 and 57. However, at present, mutations at codon 55 and 57 seem to account for the large majority of DHPS polymorphisms.

7.5 DHPS mutations, epidemiology and correlation to exposure

To explore the clinical importance of DHPS mutations, we undertook a study in HIV-1 infected PCP patients (III). In 152 episodes of PCP in the period 1989-1999, we analyzed DHPS by PCR amplification and subsequent direct DNA sequencing. The prognostic factors assessed included CD4 count, age, pAO 2 at admission, duration of time with AIDS and lactate dehydrogenase (LDH) levels.

Four different mutation patterns with non-synonymous nucleotide changes at codon 55, 57 and 60 of DHPS were identified. In our study, a significant increase in the rate of DHPS mutations was seen during the study period until 1996, followed by a decrease in 1997-1999 ( Figure 10 ). These temporal changes correlated with changes in the rate of previous or current exposure to sulfonamide drugs. Exposure to sulfa drugs and DHPS mutations were highly correlated, 18 of 29 patients who were exposed carried DHPS mutations compared to 13 of 123 patients not exposed (OR: 13.8, 95% CI: 4.9-39.7) ( Table 11 ).

A number of studies have investigated the occurrence of DHPS mutations in different countries (Table 11). Overall the association between exposure to sulfa drugs and DHPS has been confirmed, with a strong association of mutations at codon 55 and 57 to previous exposure to cotrimoxazole or dapsone prophylaxis (17, 78, 89, 90, 123, 175). Large variations in frequencies of DHPS mutations have been reported from different geographical areas, ranging from 20 to 69% of isolates. The highest levels have been observed in San Francisco where more than 80% of patients were infected with mutant strains (156). These rates are considerably higher than in Europe and Japan (195).

7.6 Mechanisms of emergence of DHPS mutations

The recent increase in DHPS mutations coincides with more widespread use of PCP prophylaxis in HIV patients. DHPS mutations seems to have arisen independently in multiple isolates, rather than as a result of dissemination of mutant isolates which developed in single or limited number of P. jiroveci organisms, since no associations is found between DHPS mutations and specific SSCP genotypes or ITS genotypes (125) (Helweg-Larsen, unpublished findings). Additional support for the hypothesis of sulfa drug induced DHPS mutations can be derived from observations of patients with recurrent episodes of PCP. In collaboration with Hauser we recently analyzed 13 European HIV patients with recurrent episodes of PCP (155). In five of seven cases in which both episodes were caused by the same P. jiroveci SSCP type, a switch to mutant DHPS strain from either wild type DHPS or a mixed wild type/mutant DHPS was observed between first and second episode of PCP. The two remaining patients had a mutant strain already at the first episode. All patients had received treatment or maintenance therapy with cotrimoxazole or dapsone. These findings suggests that DHPS mutants may be selected in vivo (within a given patient) under the pressure of co-trimoxazole or dapsone (155). The emergence of DHPS mutations appears to be specific for P. jiroveci , because only wild-type Pneumocystis DHPS have been found in other primate species (46).

7.7 DHPS mutations and outcome

In our cohort, mutations in the P. jiroveci DHPS gene were associated with significant lower 3-month survival after diagnosis of PCP ( Figure 11 ) (III). Among the 24 patients who died within three months after the diagnosis of PCP, ten patients had DHPS mutations (p=0.002), of these seven were treated with co-trimoxazole and three with pentamidine. After adjustment in a Cox model for the presence of a previous AIDS-defining illness, calendar year, age, baseline CD4-cell count, and PaO 2 , the presence of P. jiroveci DHPS mutations remained a strong predictor of survival with a adjusted multivariate hazard ratio of 3.1 (95% CI 1.2-8.1). The survival difference remained if survival analysis was restricted to patients treated with co-trimoxazole, if the analysis was restricted to first episodes of PCP and if only samples with unambiguous mutations were considered. However, DHPS mutations were not invariably associated with failure of sulfa-drug therapy, since standard treatment with co-trimoxazole was successful in 12 of 19 PCP episodes with mutant DHPS strains.

The connection between DHPS mutations and treatment failure or mortality has been studied in other reports ( Figure 12 ). Kazanjian et al observed significantly more treatment failure of sulfone therapy among 97 PCP patients with mutant genotypes compared to wildtype, but found no difference in death rates. Navin and colleagues reported that DHPS mutations had no effect on response to treatment or survival among 130 HIV-infected patients (156). In this study an overall 71% rate of DHPS mutations were observed among 130 HIV-1 PCP patients followed from 1995 to 1999 at two hospitals in San Francisco and Atlanta. A higher rate of mutations (85% of 71 patients) was observed at San Francisco General Hospital compared to Atlanta (45% of 130) with no evidence for a connection between treatment failure and mutations. On the contrary, a trend for lower death rate was found in patients treated with cotrimoxazole who carried DHPS mutation compared to wildtype (16/74 vs. 9/36). Post-hoc subanalysis even suggested an increased virulence from wildtype in patients older than 40 years. However, a potential limitation of this study is the possibility of selection bias. Within the time frame of the study more than 260 PCP patients were diagnosed with PCP at San Francisco General Hospital (150), but only 71 patients enrolled in the study. Further, only 11% of patients discharged with a diagnosis of PCP from the Hospital in Atlanta had their diagnosis verified by microscopy during the study period. A key reason was the exclusion of outpatients and patients unable to provide informed consent (Laurence Huang, personal communication).

7.8 Conclusion

In summary, DHPS mutations in P. jiroveci may have emerged recently as a consequence of selective pressure by widespread sulfonamide use. Although previous exposure to sulfone drugs have been implicated in the majority of cases described, mutant DHPS strains is now increasingly being detected in patients seemingly unexposed to sulfone drugs, particularly in the US, where remarkable high rates of DHPS mutations are being reported. These observations suggest transmission of mutant DHPS P. jiroveci from treated patients to others at risk. Available evidence suggests that DHPS mutations are associated with failure of low-dose sulfone prophylaxis, in particular dapsone prophylaxis. We have shown that in Danish HIV-patients, DHPS mutations are associated with a poor prognosis, however conflicting results from other studies have been published and presently it remains unclear whether the presence of Pneumocystis DHPS gene mutations confers clinical resistance to high dose cotrimoxazole or dapsone plus trimethoprim for PCP treatment. Several reasons for the discrepant results are possible. Among these are geographical or time dependent differences between patients in the studies and the existence of - yet undefined - drug resistance related to the DHPS mutations. Since our study on DHPS mutations only comprised 144 patients from a 10 year calendar period we cannot exclude residual confounding such as differences in severity of PCP or changes in the management of PCP during the study period. As some patients who carry mutant genotypes can be treated successfully with cotrimoxazole, it is likely that DHPS mutations are markers of other sulfa-resistance factors. In E. coli, DHPS mutations confer only low-level resistance to sulfathiazole, and secondary mutations outside DHPS are required for high-level resistance. Presently, resistance to trimethoprim appears to be an unlikely cause; the efficacy of trimethoprim monotherapy is limited (218) and only wildtype DHFR gene sequences were found in one study, in which 43% of isolates carried DHPS mutations (123).

Whether DHPS mutations will impact the overall clinical efficacy of sulfonamides is unknown. As a consequence of combination antiretroviral therapy and the marked decline in AIDS associated PCP, fewer HIV patients is now exposed to sulfone drugs in Europe and USA. Thus there may now be less selective pressure on P. jiroveci . The situation in the developed world is different. The overwhelming majority of HIV-infected persons worldwide reside in sub-Saharan Africa, Southeast Asia, and Latin America, places where access to antiretroviral therapy is currently limited. PCP is increasingly being recognized as a significant cause of morbidity in these regions where treatment is often limited to cotrimoxazole with no alternatives (5, 61). In Africa, cotrimoxazole is widely used as first line therapy for childhood infections and is increasing being used after provisional WHO guidelines have recommended its general use to prevent bacterial and opportunistic infections in HIV infected patients. Because of the huge HIV epidemic and low cost of cotrimoxazole, there is ample risk for development of clinical sulfa resistance in P. jiroveci . Hence further surveillance of DHPS mutations and investigations into other possible mechanisms of sulfa resistance is warranted. Presently, however, there remains no clear clinical use for the detection of DHPS mutations, since presence of mutations does not clearly predict response to therapy.