Comprehensive Molecular Serology of Human Chlamydia trachomatis Infections by Peptide Enzyme-Linked Immunosorbent Assays

For detection of anti-Chlamydia trachomatis antibodies by serological assays, use of classical whole-organism chlamydial antigens results in high cross-reactivity. These antigens bind mainly antibodies against the major outer membrane protein (OmpA) and bind antibodies against other immunodominant non-OmpA proteins to a lesser extent, resulting in poor assay sensitivity. The specificity of C. trachomatis serology is also compromised by the high prevalence of cross-reactive anti-C. pneumoniae antibodies in human populations. We previously identified 48 highly specific C. trachomatis B cell epitope peptide antigens of 21 immunodominant proteins. This study validated peptide antigen-based novel ELISAs that provide highly specific and sensitive detection of anti-C. trachomatis antibodies. Compared to four commercial ELISAs that achieved only poor sensitivities (51.5% to 64.8%), the combined signals of 5 to 11 peptides provided high sensitivity (86.5% to 91.8%) at the same 98% specificity. Thus, by using multiple peptide antigens of immunodominant proteins, we created simple ELISAs with specificity and sensitivity superior to standard C. trachomatis serodiagnosis.

combined signals of 5 to 11 peptides provided high sensitivity (86.5% to 91.8%) at the same 98% specificity. Thus, by using multiple peptide antigens of immunodominant proteins, we created simple ELISAs with specificity and sensitivity superior to standard C. trachomatis serodiagnosis. KEYWORDS B cell epitopes, Chlamydia pneumoniae, Chlamydia trachomatis, IgG1, IgG3, IgA1, antibody detection, antibody isotype, cross-reactivity, diagnosis, multipeptide ELISA, peptide antigens, serology, species specific O f 11 Chlamydia spp., C. trachomatis and C. pneumoniae are mainly pathogens for human hosts (1,2). C. trachomatis serovars A to C cause ocular infection, serovars D to K cause genitourinary tract infections, and serovars L1 to L3 cause lymphogranuloma venereum (1,3). In contrast, the single human serovar of C. pneumoniae causes respiratory disease infections such as pharyngitis, bronchitis, and community-acquired pneumonia (2). For understanding and diagnosing C. trachomatis infections and diseases, highly specific serological assays are urgently needed for complementation of pathogen detection assays (4). Anti-C. trachomatis antibodies have the power to indicate recent exposure as well as a history of exposure to C. trachomatis (4)(5)(6)(7). However, cross-reactivity caused by antibodies elicited during highly prevalent C. pneumoniae respiratory infections in children (8,9,76) complicates results of serological studies for C. trachomatis, emphasizing the need for C. trachomatis species-specific serological assays.
The microimmunofluorescence (MIF) test for detection of C. trachomatis-specific antibodies has remained the gold standard since its introduction (4,(10)(11)(12). MIF is performed as an indirect fluorescent antibody technique that enables microscopic observation of captured antibody on fixed chlamydial elementary body (EB) antigens (10)(11)(12). Due to the presence of conserved Chlamydia genus-specific B cell epitopes in the MIF antigens (13,14,76), the principal drawback of the C. trachomatis MIF test is high cross-reactivity of C. trachomatis EB antigens with antibodies against other chlamydial species (15)(16)(17)(18)(19). Additionally, it is a painstaking technique that requires extensive expertise associated with meticulous standardization of antigen preparations and subjective microscopic observation of immunofluorescence, thereby resulting in high interlaboratory variation (4,19,20).
In simple enzyme-linked immunosorbent assay (ELISA) format, C. trachomatis-specific serological assays are problematic due to high cross-reactivity of EBs as ELISA antigens and unavailability of suitable alternative specific antigens for sensitive detection of anti-C. trachomatis antibodies (21). Lipopolysaccharide (LPS) present on EBs is genus specific and provides substantial cross-reactivity with antibodies against other chlamydial species (14). The major constituent antigen of chlamydial EBs is the major outer membrane protein, or outer membrane protein A (MOMP/OmpA), which harbors both Chlamydia species-specific and genus-specific B cell epitopes, again resulting in cross-reactivity (13,14,22,23). Use of other immunodominant protein antigens such as OmcB, HSP60, HSP70, HSP10, MIP, and CrpA also raises concerns about cross-reactivity, given the high sequence conservation of these proteins in Chlamydia spp. (21)(22)(23). The recently preferred Pgp3 C. trachomatis plasmid protein (24)(25)(26)(27)(28) is present and highly conserved in most other Chlamydia spp. infecting animal hosts (29,30) and thus cannot resolve cross-reactivity concerns after human exposure to these animal chlamydiae, while it also may be absent in certain C. trachomatis strains following the loss of the plasmid (31,32).
Several peptide-based ELISAs are commercially available for detection of antibodies against C. trachomatis. However, most of these peptide-based assays use single OmpA peptides or a mixture of C. trachomatis strain-specific peptides derived from polymorphic OmpA regions (33)(34)(35)(36)(37). Although such assays provide highly specific detection of anti-C. trachomatis antibodies when adjusted for maximum stringency, they have not received wide acceptance, presumably due to the resultant low sensitivity (38)(39)(40). Recently, several groups have used peptide antigens from other immunodominant C. trachomatis proteins (41,42). Again, however, these assays suffer from low sensitivities.
Previously, we have identified species-specific immunodominant B cell epitopes of 11 Chlamydia spp. by use of Chlamydia-monospecific mouse hyperimmune sera raised by inoculation of live bacteria (43,77). These peptide antigens are derived from strongly reactive regions of chlamydial immunodominant proteins (44,45). They are sufficiently divergent to eliminate virtually any cross-reactivity with antibodies against noncognate Chlamydia spp. and were used as peptide antigens in ELISAs for species-specific detection of anti-Chlamydia antibodies (43,77). Importantly, we showed that use of 16-mer or longer peptide antigens in an optimized ELISA format vastly increased specific signals compared to use of short peptide antigens (46). In addition to our previous identification of 10 mouse C. trachomatis peptides (43), we recently identified 38 human host-specific peptide antigens from 21 immunodominant C. trachomatis proteins, using sera of women with active C. trachomatis infections (77). Finally, we validated the utility of such peptide antigens for highly similar species-and strainspecific molecular serologies of the 11 chlamydial species spotted on peptide microarrays (47).
Given the stochastic nature of antibody responses against individual proteins (45) or B cell epitopes (43,77), we hypothesized that measurement of combinations of antibody responses against multiple B cell epitopes of different immunodominant proteins would provide higher assay sensitivity than the use of single B cell epitope antigens, e.g., OmpA peptides. Therefore, in this study we aimed to comprehensively evaluate combinations of the top-ranked 11 C. trachomatis peptide antigens for C. trachomatis-specific detection of human antibodies. To this end, we used a wellcharacterized set of positive and negative-control sera and compared the peptide ELISA results for these sera to those obtained with the MIF test and 4 commercial anti-C. trachomatis IgG ELISAs. The results of the comparisons confirm profound sensitivity improvements due to the combined use of multiple highly specific peptide antigens. Thus, we present a simple ELISA methodology with defined synthetic antigens for C. trachomatis antibody detection that has the advantage of simultaneously higher sensitivity and higher specificity than standard assays.

RESULTS
Anti-C. trachomatis IgG antibodies determined by commercial ELISAs in sera of 125 women with active C. trachomatis infection. Initially, the sera of 125 women with nucleic acid amplification test (NAAT)-proven active C. trachomatis infection were tested for anti-C. trachomatis IgG with four commercial ELISAs. These ELISAs use purified elementary bodies or an OmpA protein segment or OmpA peptides. Anti-C. trachomatis IgG was detected in 61% (GenWay), 58% (Serion), 53% (Savyon), and 42% (Medac) of these sera, respectively, at manufacturer-defined ELISA cutoff values (Table 1). Despite the presence of an active infection, 32% of the sera were negative for anti-C. trachomatis IgG in all 4 ELISAs tested. This incongruent outcome may have been caused by an actual absence of an antibody response against C. trachomatis in certain individuals and/or by low sensitivity of the antibody detection assays. Antibodies against C. trachomatis-specific peptide antigens. Next, sera of the 125 C. trachomatis-infected (positive) women and 49 control sera of C. trachomatisnon-infected (negative) women were tested with ELISAs using 11 C. trachomatis peptide antigens (Table 2) and horseradish peroxidase (HRP) conjugates against human IgG1, IgG3, and IgA1 (Fig. 1). IgG1 and IgG3 antibodies dominated the responses to individual peptide antigens (B cell epitopes). However, the response pattern was stochastic for all antibody isotypes (48), and even sera that reacted strongly overall showed holes in the peptide reactivity patterns (Fig. 1).
Anti-C. trachomatis IgG1, IgG3, and IgA1 antibody frequencies in women with active C. trachomatis infection. For each of three isotypes, the average signal (in Log 2 relative light units [RLU]) of the 11 C. trachomatis peptide antigens (Table 2) for the 49 C. trachomatis-negative-control sera was used to define cutoff values. At 91.8% (IgG1), 91.8% (IgG3), and 93.9% (IgA1) specificity cutoff values, 80.8%, 69.6%, and 68.8% of sera from C. trachomatis-infected women were positive for IgG1, IgG3, and IgA1 antibodies, respectively ( Fig. 1). When the 3 antibody isotype-specific signals (IgG1, IgG3, and IgA1) for the 11 C. trachomatis peptides were averaged and 85%, 90%, and 95% specificity cutoff values were chosen, the detection frequencies (assay sensitivity) of anti-C. trachomatis antibodies in C. trachomatis-infected women were 90.4%, 85.6%, and 75.8% ( Table 3). The IgG assay (IgG1 plus IgG3) at the same specificity cutoff values achieved 88.6%, 83.2%, and 72.7% sensitivities (Table 3). These sensitivities were higher by 27.8%, 22.4%, and 11.9%, respectively, than the sensitivity of the GenWay IgG ELISA, the best commercial test ( Performance of individual assays compared to consensus for anti-C. trachomatis antibody status. A binomial consensus antibody status was established for all 174 serum samples, with any serum considered positive if it was positive in any 1 of the 4 commercial ELISAs or for peptide reactivity with any 1 of the 3 antibody isotypes (Fig. 1). Based on this consensus, 126 sera were classified as positive and 48 sera as negative for anti-C. trachomatis antibodies (Fig. 1). Using the consensus as the gold standard for anti-C. trachomatis antibody detection, the sensitivities of individual assays at different specificity cutoff values were calculated from receiver operating characteristic (ROC) curves (Table 4). For example, the sensitivities of commercial ELISAs at 98% specificity cutoff values were 64.8% for GenWay, 52.8% for Serion, 52.9% for Savyon,   (Table 4). Using the combined average of optical density (OD) values of these four assays, the sensitivity at the 98% specificity cutoff increased marginally to 66.9% (Table 4). In contrast, the combined average of IgG1 plus IgG3 plus IgA1 ELISA signals for 11 C. trachomatis peptide antigens achieved 92.9% sensitivity at the same 98% specificity cutoff and 91.8% sensitivity with only IgG1 plus IgG3 conjugates. This level of assay performance is significantly higher than that of any commercial ELISA (P Ͻ 10 Ϫ5 ; Fisher's exact test). However, IgG3 plus IgA1 peptide reactivity achieved only 77.2% sensitivity compared to 91.8% sensitivity achieved by IgG1 plus IgG3. These data indicate that the level of the unique contribution of the long-lived IgG1 isotype to the short-lived IgG3 isotype is higher than that of IgA1, another short-lived isotype (P ϭ 0.002; Fisher's exact test). At a less stringent cutoff such as 90% specificity, the sensitivity of all peptide assays increased. This was most pronounced for IgG3 plus IgA1, which increased in sensitivity by 16% from 77.2% to 93.2% (Table 4), while IgG1 plus IgG3 sensitivity increased by only 3.9% from 91.8% to 95.7%.
Concordance of total peptide reactivity with commercial ELISAs. For each serum, the IgG1 plus IgG3 plus IgA1 signals of all 11 C. trachomatis peptides were averaged and a 91.8% specificity cutoff value was chosen such that 45 of 49 C. trachomatis-negative sera were classified as negative ( Fig. 1). At this cutoff, 110 (88%) of 125 sera of women with C. trachomatis infection were positive for anti-C. trachomatis antibodies. The average concordance between the very strongly peptide-reactive and peptide-negative sera and the four commercial ELISAs was high (89.7% and 98.3%, respectively; Fig. 2), indicating highly specific reactivity of C. trachomatis peptide antigens. In contrast, the majority of discordant results were observed for sera with low peptide reactivity, which were frequently negative in the commercial ELISAs ( Fig. 2), most likely due to poor ELISA sensitivity. Overall, the average levels of seroreactivity with C. trachomatis peptide antigens were 76.5% in concordance for GenWay, 74.7% for Serion, 71.3% for Savyon, and 64.4% for Medac ELISAs. The significantly higher concordance of the EB antigen-based GenWay ELISA than of the OmpA peptide-based Savyon and Medac ELISAs (P Յ 0.047; Fisher's exact test) may have been related to the complexity of the EB antigen, whose reactivity resembles the reactivities of 11 peptide epitopes of 8 immunodominant proteins (Table 2).
Predictive power of single versus combined peptide antigens. Use of the area under the ROC curve (ROC-AUC) as a global assay performance measure (Table 4) corresponding to each antibody isotype are followed by data corresponding to average reactivity, indicated in quartile reactivity ranking (light to dark ochre) or as nonreactive (white). Following the columns indicating background corrected total peptide reactivities are 3 columns in white to dark brown that indicate peptide reactivity ranking at population-dependent cutoff values (91.8% specificity for IgG1 and IgG3, 93.9% specificity for IgA1). The following 4 columns indicate reactivity ranking of 4 commercial ELISAs for IgG antibody determination using manufacturer-recommended cutoff values. Next, consensus anti-C. trachomatis antibody status is shown with data representing classification of any serum as positive that was positive in any of the preceding 7 columns. The last column indicates the MIF titer, if known, with MIF-untested sera indicated by dashes. The final two bar graphs show the average IgG1 plus IgG3 plus IgA1 reactivity of each serum sample with 11 peptide antigens or with the best-performing commercial anti-C. trachomatis IgG ELISA (GenWay). indicated that the assay using the IgG1 plus IgG3 plus IgA1 reactivity average of 11 C. trachomatis peptide antigens (AUC ϭ 0.987) significantly outperformed any of the 4 commercial ELISAs (maximum AUC ϭ 0.906 for GenWay; P ϭ 0.001). (An AUC value of 0.50 indicates that discrimination of positive data from negative data equals random categorization [no discrimination], and an AUC value of 1.0 indicates that prediction is perfect [100% accurate discrimination]). This was, however, achieved by 33 individual tests of a single serum (3 antibody isotypes ϫ 11 peptide antigens). To reduce this extensive number of assays, we used ROC-AUC analyses to identify the minimal combination of peptide antigens that still provided high assay performance. As shown in Table 5, the diagnostic performance of individual peptide antigens varied substantially, but the relative ranking results were similar irrespective of the detected binding antibody isotype. Compared to the best-performing single peptide, addition of even one peptide to achieve a combination of two peptide assays improved performance  (Fig. 1). Peptide reactivity showed significantly higher concordance with the GenWay and Serion ELISAs than with the Medac ELISA (P Յ 0.047; Fisher's exact test).
dramatically. For example, the AUC IgG1ϩIgG3ϩIgA1 of peptide 1 increased from 0.889 to 0.947 when peptide 2 was added. Further stepwise addition of peptide antigens in ROC analyses showed that beyond the combination of the 5 top peptide antigens (AUC IgG1ϩIgG3ϩIgA1 ϭ 0.979), addition of any one or more of the remaining 6 peptides improved the combined assay only marginally (AUC IgG1ϩIgG3ϩIgA1 ϭ 0.987), with similar trends for all detected antibody isotypes (Table 5).
Diagnostic suitability of assays determined by likelihood ratios. Next, we sought to comparatively assess the diagnostic utility of the C. trachomatis peptide IgG assay and the commercial anti-C. trachomatis IgG ELISAs by use of likelihood ratios (LR). For the assay with 5 C. trachomatis peptide antigens and IgG1 plus IgG3 detection (Fig. 4A), any 91% to 96% specificity cutoff value corresponding to~94% to 90% indicates that the reactivities of peptides 1 and 2 were averaged, Љ1-3Љ indicates that the reactivities of peptides 1, 2, and 3 were averaged, etc., for each detection with one to three conjugates. c Individual plasma samples were tested with each of the 11 peptide antigens by use of single IgG1, IgG3, or IgA1 detection conjugates. The reactivities were combined by averaging the reactivities (Log 2 RLU per second) for 2 or 3 conjugates. sensitivities (Fig. 3A) achieved large diagnostic effect sizes for both positive-likelihood ratios (ϩLR ϭ 10.4 to 22.5) and negative likelihood ratios (ϪLR ϭ 0.07 to 0.10). In contrast, for the best-performing GenWay ELISA (Fig. 4A), cutoff values at 70% to 90% specificity with corresponding~89% to 77% sensitivities (Fig. 3B) achieved only small    (Table 5) (Fig. 3A). (A) Positive and negative likelihood ratios are independent of antibody prevalence. Sensitivities were calculated at specificities ranging from 70% (left) to 98% (right). Using sensitivity and specificity data of ROC curves (Fig. 3), positive-likelihood ratios (ϩLR) and negative-likelihood ratios (ϪLR) were calculated. The best performance of an assay is found at high ϩLR and low ϪLR (top right). At ϩLR from 5 to 25, the 0.082 ϪLR average of peptide reactivity differs highly significantly from the 0.279 ϪLR of the GenWay ELISA or the 0.391 ϪLR of the Savyon ELISA (P Ͻ 10 Ϫ3 ; paired Student's t test). (B and C) Positive and negative predictive values in dependence on the prevalence of populations of anti-C. trachomatis antibodies. Sensitivity cutoff values and the resultant specificities in ROC curves (Fig. 3) were selected for high (B) and low (C) prevalence. (B) When equal sensitivity and specificity levels of an assay are chosen from the ROC curve, PPV and NPV of the assay become equal at an assumed 50% antibody prevalence. Importantly, the peptide reactivity assay achieved higher performance (sensitivity ϭ specificity ϭ PPV ϭ NPV ϭ 92.7%) than the GenWay (83.3%) and Savyon (80.3%) ELISAs. The 3-Gy shaded areas in panel B indicate the zones of high, moderate, and poor performance of the assays, corresponding to large, moderate, and small diagnostic effects (both PPV and NPV were Ն90.9%, 83.3%, and 71.4%, respectively). For high performance of the peptide assay, a high (Ն90.9%) PPV is obtained in the 44% to 100% prevalence range and a high (Ն90.9%) NPV in the 0% to 56% range. This creates an overlapping 44% to 56% prevalence range in which the peptide assay delivers high PPV as well as NPV values (high performance) and a 29% to 71% prevalence range that delivers moderate performance. None of the commercial ELISAs delivered high or moderate performance for both PPV and NPV, and GenWay and Savyon delivered only poor performance at prevalence ranges of 34% to 66% (GenWay) and 38% to 62% (Savyon). At 38% to 62% prevalence, the PPV and NPV values of the peptide assay differed highly significantly from those of the GenWay or Savyon ELISAs (P Ͻ 0.001; paired Student's t test). (C) Modeled for lower antibody prevalence, 98% specificity was chosen for all assays. The peptide assay delivers high and moderate performance for prevalence ranges of 19% to 42% and 11% to 59%, respectively. In contrast, the commercial ELISAs at best achieved moderate performance at 14% to 35% prevalence (GenWay) and at 16% to 29% prevalence (Savyon). At 14% to 35% prevalence, the peptide reactivity assay performed highly significantly better than any of the four commercial ELISAs (P Ͻ 0.002; paired Student's t test). diagnostic effect sizes for both ϩLR (3.0 to 9.6) and ϪLR (0.16 to 0.25). At ϩLR from 5 to 25 (Fig. 4A), the ϪLR average of peptide reactivity (0.082) was significantly lower (P Ͻ 10 Ϫ3 ) than the ϪLR of the GenWay ELISA (0.279). The level of performance of the Savyon ELISA (Fig. 4A) as representative of remaining 3 commercial ELISAs was even lower than that of the GenWay ELISA. Diagnostic suitability of assays determined by predictive value modeling in dependence on anti-C. trachomatis prevalence. Finally, we used predictive value modeling to comparatively assess the diagnostic utility of the C. trachomatis peptide IgG assay (IgG1 plus IgG3) and the commercial anti-C. trachomatis IgG ELISAs in populations with different anti-C. trachomatis antibody prevalences. For the C. trachomatis peptide IgG assay set at 92.7% specificity/sensitivity, high performance (positive predictive value [PPV] and negative predictive value [NPV] Ն 90.9%) and moderate performance (PPV and NPV Ն 83.3%) were obtained at prevalence ranges of 44% to 56% and 29% to 71%, respectively (Fig. 4B). For the best commercial ELISAs, only a significantly smaller (P Ͻ 0.001) diagnostic effect size (poor performance; PPV and NPV Ն 71.4%) was obtained for a prevalence range of 34% to 66% (GenWay) or 38% to 62% (Savyon) (Fig. 4B).
For low-prevalence populations, the C. trachomatis peptide IgG assay (IgG1 plus IgG3) achieved high and moderate performance for prevalence ranges of 19% to 42% and 11% to 59%, respectively. In contrast, the commercial ELISAs at best achieved significantly reduced (P Ͻ 0.002), moderate performance at 14% to 35% (GenWay) and 16% to 29% prevalence (Savyon) (Fig. 4C). Note that the low sensitivity of the commercial ELISAs resulted in a precipitous NPV drop when a high-specificity cutoff value (and PPV) was chosen.
Diagnostic usefulness of assays using 11 versus 5 C. trachomatis peptides. Depending on the intended use of anti-C. trachomatis antibody data, a minimal set of the most informative peptide antigens combined with detection of one or two antibody isotypes may reduce the required number of tests substantially without compromising performance (Table 5). Therefore, we compared several minimal assays with the full assay that used all 11 C. trachomatis peptide antigens and conjugates against three antibody isotypes (see Fig. S1 in the supplemental material). ROC curves, ϩLR, ϪLR, PPV, and NPV were evaluated similarly to the methods described for Fig. 3 and 4 to determine suitable assay cutoff values (Fig. S1). As reported in Table 5, any reduction of either peptide antigen numbers or detected antibody isotypes reduced assay performance, particularly at high-stringency cutoff values. For instance, at a 98% specificity cutoff, the IgG1 plus IgG3 plus IgA1 reactivity average of 11 peptides achieved 92.9% sensitivity whereas the IgG1 plus IgG3 reactivity average of 5 peptide antigens achieved only 86.5% sensitivity, a 7.4% sensitivity reduction (Fig. S1A). Thus, the best assay performance was obtained with the maximum number of 11 peptide antigens and detection of all three antibody isotypes, but use of 5 optimal peptides still achieved adequate diagnostic performance (Fig. S1B and C and D). The data shown in Table 5 (see also Fig. S1) will help medical practitioners to choose the combination of peptide antigens and conjugates most appropriate for specific epidemiological settings.

DISCUSSION
We demonstrate here that peptides of C. trachomatis-specific B cell epitopes, in combined use, are suitable ELISA antigens for highly sensitive and specific detection of anti-C. trachomatis antibodies. The reactivity "holes" (false-negative results) of individual peptide antigens (Fig. 1) or whole-protein antigens (45) with even strongly antibody-positive sera corroborate the contention that the use of single peptide and protein antigens would increase false-negative results. Thus, only the combined use of multiple peptide antigens can reliably quantify the host antibody level produced in response to C. trachomatis infections, approximating the quantitative results obtained with complex antigens but typically at higher specificity. For adequate performance with an acceptable workload, we recommend the use of the five strongest C. tracho-matis peptide antigens as an alternative to the maximum performance obtained by use of 11 peptides (Table 5; see also Fig. S1 in the supplemental material). While many commercial anti-C. trachomatis ELISAs are exclusively based on OmpA antigens, we conclude that the higher sensitivity of the peptide assays described here is the result of the use of multiple B cell epitopes chosen from several immunodominant C. trachomatis proteins in addition to OmpA (Table 5). Therefore, multipeptide antigens capture a wide spectrum of antibodies, similarly to complex chlamydial antigens with multiple epitopes (e.g., elementary bodies), but avoid the cross-reactivity associated with conserved epitopes within such complex antigens. Furthermore, the high sensitivity is also the consequence of the peptide antigen design and the ELISA protocol optimized for maximum seroreactivity (43,46,47,77). Thus, the present report offers a simple ELISA with defined synthetic antigens for C. trachomatis antibody detection that has the advantage of simultaneous high sensitivity and high specificity (Table 4).
In the absence of a gold standard, the peptide reactivity assays were evaluated using two composite reference standards (CRS). Such CRS typically consist of several imperfect tests with poor sensitivity but high specificity (49)(50)(51)(52). Excluding the peptide reactivity assays (index tests), the NAAT and four commercial ELISAs were used as component tests to establish a first CRS for C. trachomatis exposure status (Table 3). This evaluation established the superior performance of the peptide assays for determination of C. trachomatis exposure status via use of anti-C. trachomatis antibodies. For subsequent comparative performance evaluation of all individual assays, a second CRS was constructed from 7 component tests, including 4 commercial ELISAs and tests of IgG1, IgG3, and IgA1 reactivity averages of 11 C. trachomatis peptide antigens (Table 4; termed ЉconsensusЉ in Fig. 1). It is well known that increasing the numbers of component tests in CRS reduces the number of false negatives (higher sensitivity) but increases the number of false positives (lower specificity) unless all component tests have perfect specificity (49)(50)(51)(52). For that reason, we minimized the input of peptide antigen data by using averages rather than numerous individual peptide reactivities and applied stringent cutoff values for high specificity (Fig. 1). Supporting the validity of the CRS approach, previous studies published by us as well as by others reported high specificity for C. trachomatis-specific peptide antigens (also used in Savyon, Serion, and Medac ELISAs) but poor sensitivity (41-47, 53, 54, 77). Taken together, these considerations strongly favored the use of CRS over an alternative gold standard determined by latent class analysis (LCA), because the ELISA assays in this investigation (except for NAAT) clearly violate the required assumption of the conditional independence of the LCA component tests (49)(50)(51)(52)55).
For useful diagnostic antibody detection, a highly desirable characteristic of an assay is that it maintains high predictive values, both positive and negative, at a wide range of antibody prevalences. For assays that provide acceptable performance only within a narrow prevalence range, signal cutoff stringency can be adjusted to an anticipated population prevalence level. This entails a tradeoff between assay specificity and sensitivity. ROC curves can be used to identify the suitable specificity/sensitivity points and associated assay signal cutoff values that result in PPV and NPV balanced for the anticipated population prevalence. For example, a stringent cutoff (high background cutoff ϭ high specificity) resulting in high PPV is required for low population prevalences of anti-C. trachomatis antibodies ( Fig. 4C; see also Fig. S1D). Therefore, a tradeoff between specificity and sensitivity must be found that maximizes both PPV and NPV for the antibody prevalence in question. As shown in Table 4 (see also Fig. S1A), for a high-stringency cutoff (98% specificity), the combined use of 5 to 11 peptide antigens still provided high sensitivity (86.5% to 91.8%; IgG1 plus IgG3 conjugates). This resulted in moderate to high performance (PPV and NPV Ն 88%) in a wide anti-C. trachomatis antibody prevalence range of 14% to 49%. In contrast, with 64.8% sensitivity at 98% specificity, the best-performing GenWay ELISA achieved these performance characteristics only within a narrow 19% to 27% prevalence range. The remaining three commercial ELISAs did not provide even that level of performance for any prevalence (Fig. 4C), due to poor assay sensitivity (51.5% to 52.9%) at this high (98%) specificity cutoff (Table 4).
Likelihood ratios are another informative measure of assay performance in diagnostic testing. Unlike predictive values, likelihood ratios measure assay performance independently of antibody prevalence. As determined by the likelihood ratios, the multipeptide assay again outperformed the commercial ELISAs, with large diagnostic effect sizes for both ϩLR ϭ Ն10 and ϪLR ϭ Յ0.10 by the assays using 5 to 11 peptides for IgG (IgG1 plus IgG3) but not with any of the four commercial IgG ELISAs ( Fig. 4A; see also Fig. S1B). Analyses of the area under the ROC curve (ROC-AUC), another cutoffindependent global assay characteristic, again showed higher performance for multipeptide assays than for commercial ELISAs (Table 4).
To capture the history of C. trachomatis exposure, we recommend detection of both long-lived IgG1 and short-lived IgG3 (Table 5; see also Fig. S1). For recent exposure, detection of IgG3 plus IgA1 is more appropriate, due to the immediate response and the short half-life of these antibody isotypes (37,56,57). Importantly, for antibody status determination by use of only two monoclonal conjugates, the IgG3 plus IgA1 combination may have an advantage for detection of recent chlamydial infections (37,56,57), while the IgG1 plus IgG3 combination is optimal for detection of total antibody status. While numerous combinations of peptide antigens showed virtually identical levels of performance with the study sera, we suggest use of the five top-ranked peptides (Table 5), whose performance in combination is closest to that seen with the best combination of all 11 peptides. These five immunodominant B cell epitopes of C. trachomatis OmpA, IncE, PmpD, CT143, and CT442 proteins (Table 5) are highly divergent from those of other Chlamydia spp. but highly conserved within C. trachomatis (Table 2). Combined with our previous studies (43,46,47,77), the result of this study opens the potential for novel Chlamydia species serology with very high specificity at sensitivity equal to or better than what classical assays for serodiagnosis of chlamydial infection offer. The simple synthetic peptide ELISA format is within reach of any diagnostic laboratory and can be readily commercialized similarly to OmpA peptide ELISAs (33)(34)(35)(36)(37)(38)(39)(40).
For specific detection of anti-C. trachomatis antibody, the majority of currently available commercial ELISAs use only C. trachomatis-specific OmpA peptide or recombinant protein segment antigens, thereby avoiding cross-reactivity. Those assays provide specific ELISA signals but, because of the use of the single antigen, also have unacceptably poor assay sensitivity as a consequence of failing to detect antibody responses against other immunodominant proteins of C. trachomatis. Several studies reported that C. trachomatis-positive human sera recognize non-OmpA proteins such as polymorphic outer membrane proteins (Pmp) and inclusion membrane proteins (Inc) equally frequently or even more frequently (44,45). Importantly, using a C. trachomatis proteome-wide microarray, Wang et al. reported that antibodies to OmpA are present in only 58% of anti-C. trachomatis-positive sera, although high levels of antibodies to other C. trachomatis proteins are detectable in the remaining 42% (45). Such falsenegative results may be a consequence of well-known problems with MOMP expression and epitope "masking" in incorrectly folded recombinant proteins or of the randomness of somatic recombination of the antigen recognition regions of immunoglobulin genes (48). Thus, the reactivity "holes" (false-negative results) of even highly antibody-positive sera (Fig. 1) unequivocally imply that use of single antigens in serology inherently compromises sensitivity as well as quantitative accuracy.
Intact chlamydial elementary body antigens serve as antigen in the gold-standard MIF test for anti-C. trachomatis-specific antibody detection, and EB or EB lysates are also commonly used in ELISAs. In these complex antigens, OmpA is the primary reactant for species-specific anti-C. trachomatis antibodies, but genus-specific LPS (14) or conserved B cell epitopes on CT681/OmpA, CT443/OmcB/Omp2, CT442/CrpA, Hsp60, or CT381/ ArtJ frequently cross-react with antibodies against other chlamydial species, including C. pneumoniae (13,22,23,58). A Pgp3 protein-based ELISA that was recently used extensively for C. trachomatis serology showed 11% to 14% higher sensitivity than the OmpA peptide-based Savyon and Medac ELISAs (40). In contrast, the IgG1 plus IgG3 peptide assay reported here showed~40% higher sensitivity than the Medac and Savyon ELISAs, at similar (98%) specificity (Table 4). Therefore, both the inherently low specificity of assays using complex cross-reactive antigens and the inherently low sensitivity of single-protein assays impede their use in control and monitoring programs for trachoma and sexually transmitted C. trachomatis infections (9,40,(59)(60)(61)(62)(63).
In a previous study, we reported 48 C. trachomatis-specific peptide antigens; we comprehensively evaluated the top 11 of those peptides in the present study. Using four of those peptides modified for a microarray, we recently showed that these peptide antigens similarly detect anti-C. trachomatis antibodies in microarray format with higher sensitivity than a commercial ELISA. Serological assays for reliable measurement of antibody levels with high sensitivity require the use of multiple peptide antigens but also a setup using single peptides per microtiter well. Therefore, multiple wells must be used for each individual serum sample, incurring a substantial workload. Microarray chips can incorporate dozens of peptide antigens, thus eliminating the numerical and labor constraints associated with multipeptide microtiter plate ELISAs. Therefore, the logical use of these peptide antigens in real-life implementation of C. trachomatis serology is in multiplexed microarray format (47,64,65). With the possibility of spotting 196 to 784 peptides on a single microarray chip, such microarrays offer unprecedented opportunities (65). Combined multipeptide signals not only would allow specific and accurate quantification of anti-C. trachomatis antibody responses but also might reveal multiepitope reaction patterns (Fig. 1) that may correlate with disease phenotypes and thus may serve as disease biomarkers (61)(62)(63)(66)(67)(68)(69)(70).
Seroepidemiology based on highly specific and sensitive assays can determine past exposure and measure the cumulative risk of C. trachomatis infection over time (4)(5)(6)(7) and thus can help to minimize and control the burden of C. trachomatis infections in populations (71). The high performance of multipeptide ELISAs provides a powerful argument for their use in trachoma control (24,26,27) and in screening for sexually transmitted infections (21,25,28). Such multipeptide ELISAs will increase the effectiveness of trachoma surveillance programs, and they will enable earlier identification and treatment of young women with chlamydial infection in sexually transmitted disease (STD) screening, reducing a woman's likelihood of experiencing highly consequential reproductive health complications. In conclusion, specific and sensitive peptide ELISAs will improve C. trachomatis serology and will thereby help to more accurately diagnose and mitigate C. trachomatis infections and their sequelae (61)(62)(63)(66)(67)(68)(69)(70).

Sera.
In this study, we used sera of 174 women that were collected from three cohorts: (i) 125 women with C. trachomatis infection confirmed by NAAT (53) whose sera were considered C. trachomatis positive; (ii) 18 healthy, low-risk women who had never been diagnosed with C. trachomatis infection (and whose sera were considered C. trachomatis negative); and (iii) 31 healthy women who were normal blood donors and self-reported to be free of infections (sera considered C. trachomatis negative) (BioIVT North America and Asia Pacific, Westbury, NY). Anti-Chlamydia trachomatis MIF titers (antigens A to I, K, and L1 to L3) were determined for 121 of the 125 C. trachomatis NAAT-positive women at the University of Washington. Serum samples of the healthy 18 women with low risk for C. trachomatis exposure tested negative for antibodies against C. trachomatis in an ELISA using C. trachomatis elementary bodies as antigen (54), and the 31 blood donor sera tested negative in four commercial anti-C. trachomatis IgG antibody ELISAs; these 49 sera were not tested by MIF. The ages of all study subjects ranged from 18 to 38 years, with an average of 22 years. Among the cohort of 125 C. trachomatis-positive women, 74 were African American; 28 Caucasian; and 23 of Hispanic, Asian, or mixed racial origin. The 18 low-risk women never diagnosed with C. trachomatis infection were all Caucasian. Of the blood donor cohort, 13 were African American; 10 Caucasian; and 8 of Hispanic, Asian, or mixed racial origin.
Written consent was obtained from all serum donors. Blood donor sera were collected at BioIVT facilities located in the United States, using BioIVT standard operating procedures approved by appropriate regulatory and ethics authorities. The study protocol for the remaining sera was approved by the institutional review boards for human research of the University of Pittsburgh and the University of North Carolina at Chapel Hill. each subsequent step. This approach resulted in determining the maximum predictive power of combinations of 2 through all 11 peptide antigens.
Determination of likelihood ratios in diagnostic testing. For evaluation of diagnostic anti-C. trachomatis antibody assay performance, positive-likelihood ratios (ϩLR) and negative-likelihood ratios (ϪLR) were calculated at 70% to 98% assay specificity cutoff values and the resultant assay sensitivity from ROC curves. The positive-likelihood ratio (ϩLR) is defined as the ratio between the probability of a positive test in subjects with the disease compared to those without the disease (74); i.e., ϩLR ϭ true positive rate/false-positive rate. The negative-likelihood ratio (ϪLR) is defined as the ratio of the probability that a negative result would occur in subjects with the disease to the probability that the same result would occur in subjects without the disease (74); i.e., ϪLR ϭ false-negative rate/truenegative rate. Based on accepted ranges of effect size (73)(74)(75), the performance of an assay was deemed diagnostically useful if both the ϩLR and ϪLR exceeded certain effect sizes as follows: large effect (ϩLR ϭ Ն10, ϪLR ϭ Յ0.10), moderate effect (ϩLR ϭ Ն5, ϪLR ϭ Յ0.15), and small effect (ϩLR ϭ Ն2.5, ϪLR ϭ Յ0.25).
Determination of predictive value in dependence of anti-C. trachomatis antibody prevalence. For population prevalence-dependent evaluation of assay performance, positive and negative predictive values (PPV and NPV) of individual assays were plotted against corresponding antibody prevalences. PPV and NPV for 0% to 100% prevalence were calculated from sensitivities and specificities in ROC curves as described earlier (73)(74)(75). The PPV was defined previously (75) as the probability that the disease is present when the test is positive as follows: PPV ϭ [sensitivity ϫ prevalence] Ϭ [(sensitivity ϫ prevalence) ϩ (1 Ϫ specificity) ϫ (1 Ϫ prevalence)]. The NPV was defined previously (75) as the probability that the disease is absent when the test is negative as follows: NPV ϭ [specificity ϫ (1 Ϫ prevalence)] Ϭ [(1 Ϫ sensitivity) ϫ prevalence ϩ specificity ϫ (1 Ϫ prevalence)]. The performance of an assay was considered high, moderate, or poor if both PPV and NPV exceeded 90.9%, 83.3%, or 71.4%, respectively. For direct quantitative comparisons of the diagnostic utility of the assays at 50% antibody prevalence, the point of equal sensitivities and specificities on the ROC curve was used to calculate PPV and NPV for each of the individual assays. For comparisons of the levels of diagnostic utility at low prevalence, a highly stringent cutoff (98% specificity) and the resultant sensitivity were uniformly chosen for all assays to calculate PPV and NPV.
Statistical analyses. Statistical analyses were performed and graphical outputs generated by use of either of the software packages Microsoft Excel 2016 (Microsoft, Redmond, WA) and Statistica 7.1 (StatSoft, Tulsa, OK). Antibody detection frequencies were compared by two-way Fisher's exact test. For comparisons of ROC-AUC values, corresponding sensitivities were calculated at specificity cutoff values of 98%, 95%, 90%, 85%, and 80%. For comparisons of likelihood ratios, corresponding ϪLR values were calculated for ϩLR values of 5, 10, 15, 20, and 25. For comparisons of predictive values within given anti-C. trachomatis antibody prevalence ranges, 5 PPVs and 5 NPVs of each assay were calculated, corresponding to 5 equidistant prevalences. Means were then compared by one-tailed paired Student's t test.