INTRODUCTION

Top Abstract Introduction Materials and Methods Results Discussion References

Controversy surrounds the association of Chlamydia pneumoniae with atherosclerotic heart disease (7, 9, 18), asthma (1, 10), multiple sclerosis (11, 22), and Alzheimer's disease (8, 16, 19), primarily because of the lack of a definitive test for detecting C. pneumoniae. Culture is performed successfully by few laboratories and was much less sensitive than PCR for detection in vascular tissue (13). Serology has been considered the "gold standard" for the diagnosis of infection (12) but did not correlate with the presence of C. pneumoniae DNA or antigen in tissue (5, 13, 17). A superior marker of current or recent infection is required to clarify the clinical importance of C. pneumoniae infection in chronic diseases such as atherosclerosis.

Nucleic acid amplification tests such as PCR enable the detection of low concentrations of organism in clinical specimens. However, great variability of detection has been reported. For atherosclerotic tissue, reports of between 0 and 100% detection have been published, as recently summarized (3, 13). Similarly, the prevalence of C. pneumoniae DNA in peripheral blood mononuclear cells (PBMC) varied between 9% (27) and 59% (4) among patients with proven atherosclerotic heart disease. Some of this discrepancy may be attributable to differences between assays, but sampling variability is an alternative explanation. Whether replicate testing improves sensitivity or reproducibility has not, to our knowledge, been systematically examined for C. pneumoniae nucleic acid amplification tests.

In a previous study, we compared five C. pneumoniae PCRs for both analytical and clinical sensitivity (15), and we noted a major discrepancy between the tests. Despite relatively similar analytical test sensitivities, only the nested PCR based on the ompA gene (23) routinely detected a number of PBMC positives. We hypothesized that sampling variability as well as differences in PCR performance explained the results, and we inferred that clinical specimens had low concentrations of C. pneumoniae DNA.

We test here these hypotheses with probit regression analysis. Specifically, we sought to determine the following. (i) Can replicate C. pneumoniae PCR increase test sensitivity over testing a single time (analytical sensitivity)? (ii) Does replicate testing increase C. pneumoniae detection in clinical specimens (clinical sensitivity)? (iii) Can probit analysis quantitate C. pneumoniae in clinical specimens?

	MATERIALS AND METHODS

Top Abstract Introduction Materials and Methods Results Discussion References

PCR methods. A comparison of five PCRs for C. pneumoniae was previously described (15), and two of these PCRsa nested (23) and a non-nested procedure (6)-were used in this study. Samples (200 µl) of laboratory-cultured strains or clinical specimens (see details below) were extracted using QIAamp DNA Mini-Kits (Qiagen, Mississauga, Ontario, Canada) following a tissue or blood extraction protocol and eluted in 100 µl of buffer. PCR was performed on 2.5-µl purified DNA samples in a total volume of 25 µl. The components of the reaction mixture and thermocycling conditions were previously described by Campbell et al. (6) for the non-nested PCR and by Tong and Sillis (23) for the nested PCR. The non-nested procedure consisted of 40 rounds and amplified a 437-bp cloned PstI fragment. The nested PCR consisted of 40 rounds of amplification of a 333-bp fragment of the ompA gene, followed by 30 rounds of amplification of a 207-bp internal fragment. AmpliTaq Gold (Perkin-Elmer, Branchburg, N.J.) was used for all amplifications. All amplification products were analyzed by 2% (wt/vol) agarose gel electrophoresis followed by ethidium bromide staining. Stringent procedures to minimize or detect contamination included extraction and amplification in separate rooms after changing gloves and lab coats, use of plugged pipette tips and positive displacement pipettors, and insertion of at least one blank every 5 to 10 tubes. Every fifth blank was left open during specimen addition to detect aerosol contamination.

Dilution series of C. pneumoniae. C. pneumoniae ATCC VR-1310 was cultured in U-937 human mononuclear cells for 40 to 48 h and then spiked into a repeatedly negative volunteer-derived PBMC fraction (CPT tube; BD Vacutainer Systems, Franklin Lakes, N.J.). In PCR replicates of two, serial 10-fold dilutions were tested to establish an upper threshold (all tests positive) and lower thresholds (all tests negative) of test sensitivity as 4 inclusion forming units (IFU)/ml (0.01 IFU/2.5-µl PCR) and 0.04 IFU/ml (0.0001 IFU/2.5-µL PCR), respectively. For each of the non-nested and nested PCRs, we performed 10 replicates of six dilutions between 4 and 0.04 IFU/ml and, at a separate time, 10 replicates of 10 dilutions between 4 and 0.008 IFU/ml. The final probit regression model included data from both of these two dilution series (total of 160 tests for each of the nested and non-nested PCRs).

Clinical validation. For the clinical validation set, 26 clinical specimens (each from unique patients) were tested in 10 replicates by both PCR assays (total of 520 PCR tests): 10 PBMC specimens from patients undergoing elective coronary angiography, 6 pediatric and 8 adult nasopharyngeal specimens (NPS) from patients with acute respiratory symptoms, and 2 sputum specimens from adults with chronic airway limitation. A patient's specimen was considered positive if one or more of the 10 individual determinations were positive. All specimens were previously C. pneumoniae positive in at least one of three replicates using the nested PCR and were confirmed by Southern blotting and oligonucleotide hybridization with C. pneumoniae-specific probe.

Statistical methods. The relationship between the proportion positive from each replicate of 10 and the corresponding log concentrations of C. pneumoniae was examined using probit regression analysis (SPSS for Windows 10.0; SPSS, Inc., Chicago, Ill.). Using the probit model, the two tests were compared, the median detection concentration of C. pneumoniae was estimated, and the concentrations corresponding to probits of 0.01 to 0.99 were calculated. Proportions were tested with StatXact version 3.02 (Cytel Software Corp., Cambridge, Mass.). A P value of <0.05 was taken as statistically significant (two tailed). The relationship between replicate number and test sensitivity was calculated in Excel 97 (Microsoft Corp., Redmond, Wash.), using the calculated probit at a given concentration and the following formula: probability of at least one positive in n replicates = 1  (1  probit)ⁿ. To estimate median C. pneumoniae concentration in clinical specimens, the concentration corresponding to the observed proportion positive of 10 replicates was interpolated from the nested PCR probit analysis table.

	RESULTS

Top Abstract Introduction Materials and Methods Results Discussion References

Constructing probit regression models. Separate probit regression curves were constructed for the non-nested and nested PCR assays using data from 10 replicates each of 16 dilutions of culture-grown C. pneumoniae ATCC VR-1310. Figure 1 shows the relationships between the number of positives at each dilution and the concentration of C. pneumoniae for the non-nested and nested PCRs. The circles (non-nested, panel A) or triangles (nested, panel B) represent the number of positive results from the 10 replicates at each concentration of C. pneumoniae, and the solid line represents the fitted probit regression line for the assay. At a concentration of 4 IFU/ml, both PCRs detected all 20 replicates as positive. At a concentration of 1 IFU/ml, the non-nested PCR detected 5 of 10 versus 8 of 10 for the nested PCR; and at a concentration of 0.1 IFU/ml, the non-nested PCR detected 0 of 10 versus 1 of 10 for the nested PCR.

View larger version (20K): [in this window] [in a new window]   FIG. 1.   Number of positives per 10 replicates versus the concentration of C. pneumoniae ATCC VR-1310 for non-nested PCR (A) and nested PCR (B) and a regression curve determined by probit regression analysis (SPSS).

View larger version (24K): [in this window] [in a new window]   FIG. 2.   Comparison of probit regression curves for a non-nested () and nested () C. pneumoniae PCR (SPSS). The probit (predicted proportion of replicates positive) versus the C. pneumoniae ATCC VR-1310 concentration was obtained from 10 replicates of 16 dilutions (see the text).

Relationship between detection and number of replicates. For a given probit and its corresponding C. pneumoniae concentration, the sensitivity of replicate testing and the number of replicates are related. The number of replicates needed for various probits of 0.01 to 0.99, to achieve an overall test sensitivity of 50, 80, 90, or 95%, were calculated using the nested C. pneumoniae PCR data (Table 2). For example, for a C. pneumoniae concentration of 0.35 IFU/ml (probit of 0.50), a single PCR determination has 50% sensitivity, a three-replicate assay would detect at least one positive with 80% sensitivity, a four-replicate assay with 90% sensitivity, and a five-replicate assay with 95% sensitivity. At lower concentrations, more replicates are required for a given overall test sensitivity. At 0.05 IFU/ml (probit 0.10), a single test achieves 10% sensitivity, compared with 7 replicates (50% sensitivity), 16 replicates (80% sensitivity), or 29 replicates (95% sensitivity). Conversely, at >5 IFU/ml, a single PCR determination is 95% sensitive for detecting C. pneumoniae, and replicates would not increase sensitivity further.

View larger version (25K): [in this window] [in a new window]   FIG. 3.   Predicted probability of PCR positive test versus the C. pneumoniae concentration for nested PCR by number of replicates. Replicates of 10, 5, 3, or 2 versus single PCR illustrated from left to right. Single PCR curves obtained from probit regression analysis of 10 replicates of 16 dilutions of C. pneumoniae ATCC VR-1310 (see text) are also shown. The remaining four curves were calculated in the spreadsheet program Microsoft Excel using the following formula: probability (at least 1 positive in n replicates) = 1  (1  probit)n.

Detection of C. pneumoniae in clinical specimens. We next compared the performance of non-nested and nested PCR assays using clinical specimens. For 10 previously positive PBMC specimens from coronary angiography patients, the non-nested PCR detected a single positive only on the tenth repeat of that specimen (data not shown). In the first one, three, or five replicates, no positives were detected. For the 14 nasopharyngeal and 2 sputum specimens, no positives were detected in any of 10 replicates. In summary, the non-nested PCR identified 1 of 26 patient specimens, and 1 of 260 PCR tests, as positive.

Quantitation of C. pneumoniae in clinical specimens. By interpolation from the probit regression analysis for the nested PCR (Table 1), estimates of C. pneumoniae in clinical specimens were made (Table 3). For the 10 PBMC specimens, the median probit was estimated at 0.15, for an interpolated concentration of 0.07 IFU/ml (95% CI = 0.04 to 0.11). For specific patient specimens, probits varied between 0.1 and 0.9 or 0.05 to 2.7 IFU/ml. The single positive sputum was positive in 2 of 10 replicates for a probit of 0.2 (0.09 IFU/ml, 95% CI = 0.06 to 0.14). In four of eight adult NPS specimens, 1 of 10 replicates was positive (probit of 0.10) for an interpolated concentration of 0.05 IFU/ml (95% CI = 0.02 to 0.08). In the remaining four adult and all pediatric NPS specimens none of the 10 replicates detected C. pneumoniae (<0.03 IFU/ml). At a probit of <0.05, the median concentration for the nasopharyngeal specimens was <0.03 IFU/ml (95% CI = 0.00 to 0.05).

	DISCUSSION

Top Abstract Introduction Materials and Methods Results Discussion References

We have demonstrated here that replicate C. pneumoniae PCR markedly increased analytical sensitivity compared with performing a single PCR test. We validated the model by demonstrating that replicate testing increased C. pneumoniae detection in clinical specimens, particularly with the nested PCR, and that the sensitivity levels in analytical and clinical samples were consistent with the probit analysis predictions. We then used probit analysis to quantitate C. pneumoniae in clinical specimens and inferred a higher concentration of C. pneumoniae in PBMC compared with NPS.

The interpretation of replicate testing was facilitated by probit regression analysis, which has been utilized in particular for toxicology studies. In microbiology, probit analysis has been used very rarely: we found only four references in a MEDLINE search of the literature between 1967 and 2000. Vrielink and colleagues used probit analysis to compare the diagnostic sensitivities of enzyme immunoassays for human T-cell leukemia virus types 1 and 2 or hepatitis C virus (24-26), and Saldanha used this regression technique to quantitate the hepatitis C virus genome and compare PCR sensitivities (20).

The sensitivity of a diagnostic test is often considered a constant property, apart from some variation due to laboratory technique or specimen type. However, interpreting the probit as the test sensitivity, we demonstrated that PCR sensitivity varied between 0 and 100%, depending on the C. pneumoniae concentration. The finding that PCR was approximately 100% sensitive above a certain threshold and 0% sensitive below a certain threshold is not surprising. What is surprising is the 100-fold interval of concentrations (between 4 and 0.04 IFU/ml) in which PCR results were intermittently positive. Within this interval, repeat testing and probit modeling could be exploited for detection and quantitation.

Our findings may have important implications for the routine detection of C. pneumoniae in clinical specimens such as blood or respiratory specimens. Conversely, the lack of recognition of these concentrations of intermittently positive values may yield unreliable results.

In an excellent review of the molecular diagnosis of C. pneumoniae, Boman et al. discuss specimen collection, the preparation of nucleic acid from samples, the choice of gene target and primer selection, the optimal amplification conditions, and the detection of the amplification product (3). These authors briefly review sampling variation as a cause of false-negative results and discuss increasing the sample volume as a possible strategy to increase sensitivity, while acknowledging that this strategy may cause an unacceptable increase in the level of PCR inhibitors. We suggest adding the issue of PCR replicates to their list of areas where standardization is required. Readers need to know how many PCR replicates were done by a laboratory and how a positive specimen was operationally defined.

We acknowledge two potentially serious limitations of our study: face validity and feasibility. By face validity, we refer to whether most readers or laboratory directors would have confidence that a single positive in 5 or 10 PCR determinations represented a true positive. At a more stringent requirement of 2 PCR positives per 10 replicates, 0 of 26 clinical specimens were positive by the non-nested PCR, and 6 of 26 specimens were positive by the nested PCR (P = 0.03). A single positive PCR determination may represent contamination, a nonspecific reaction, or a true positive. We demonstrated that lower analytical concentrations were only intermittently PCR positive, and this relationship was predictable from a statistical viewpoint. These results are not likely to be due to contamination, which would not have varied predictably with the concentration. In addition, 0 of 200 negative controls tested with this assay by our laboratory have been positive. To ensure the specificity of the reaction, all first-time positive PCR clinical specimens were confirmed with Southern blotting and oligonucleotide hybridization, and 12 specimens had DNA sequencing of PCR product. The results all confirmed a C. pneumoniae-specific amplification product. If contamination and nonspecificity are ruled out, the results are true positives. Nevertheless, we would not consider a single 1 of 10 samples positive as a "confirmed" C. pneumoniae positive. In our PBMC study, we verified positive specimens by independent re-extraction, followed by PCR in triplicate (21). As more sensitive assays are developed, confirmation by amplifying a different target will be preferable (2).

Regarding feasibility, we acknowledge that a trade-off may be required in determining the optimal number of replicates for different specimen types. For the detection of C. pneumoniae DNA in clinical specimens, replicates of two or three, depending on the specimen type, may be adequate if the laboratory can demonstrate high reproducibility. A larger number of replicates will likely not be feasible, but the laboratory may wish to test a small number of positives in 5 or 10 replicates to examine reproducibility. In a research setting, replicates of up to 10 may be desirable for specimen types in which C. pneumoniae concentration is likely to be low. We currently test blood and respiratory specimens in replicates of three, but we have increased both the concentration (DNA eluted in 50 µl rather than 100 µl) and the sample size (5 µl per 50-µl PCR mixture rather than 2.5 µl per 25-µl PCR mixture) to approximate the same sensitivity as previously achieved with 10 replicates. Nevertheless, we suggest replicate PCR as a "reference standard" only until methods of extraction and detection are improved to the point where single or duplicate PCR will provide comparable sensitivity and reliability. Methods to concentrate target, such as monocyte enrichment using CD14 antibodies (14), or nucleic acid molecule selection using capture probes, may obviate the need for replication altogether.

We conclude that repeat testing of the same specimen markedly increases the sensitivity and reliability of a PCR assay, particularly for clinical specimens with a low C. pneumoniae concentration. Replicate testing may improve the development and comparison of PCRs and provide more precise estimates of organism prevalence in various chronic disease states. Conversely, failure to recognize the low sensitivity of a single PCR determination may cause frustration when positive results cannot be reliably confirmed. Further validation is needed with other C. pneumoniae assays and other clinical specimens.