Validation and Implementation of a Diagnostic Algorithm for DNA Detection of Bordetella pertussis, B. parapertussis, and B. holmesii in a Pediatric Referral Hospital in Barcelona, Spain

This study aimed to validate a comprehensive diagnostic protocol based on real-time PCR for the rapid detection and identification of Bordetella pertussis, Bordetella parapertussis, and Bordetella holmesii, as well as its implementation in the diagnostic routine of a reference children’s hospital. The new algorithm included a triplex quantitative PCR (qPCR) targeting IS481 gene (in B. pertussis, B. holmesii, and some Bordetella bronchiseptica strains), pIS1001 (B. parapertussis-specific) and rnase P as the human internal control.

the disease. An accurate identification at the species level is not only important from a clinical point of view to select the most appropriate antibiotic treatment, but also for health public purposes, since misdiagnosis of Bordetella species can lead to an incorrect assessment of pertussis vaccine effectiveness (6).
For this purpose, a number of microbiological techniques are readily available, such as culture, serology, and nucleic acid amplification tests (NAATs) (7). Although culture remains the gold standard, it has low sensitivity (8,9). Serology is not an appropriate method to diagnose pertussis in pediatric populations, since it provides results that are difficult to interpret in immunized individuals and requires measuring antibody titers in the acute and convalescent phases of the disease, thus delaying time to result (10). Rapid, sensitive, and specific NAATs are being increasingly implemented to overcome the limitations of culture and serology (9,11).
NAATs targeting IS481, at high copy number in the genome of B. pertussis, and IS1001 for B. parapertussis are commonly used (12,13). However, both targets are also present in B. holmesii and some B. bronchiseptica strains (14). Several algorithms combine nonspecific targets (IS481, IS1001 or IS1002) and may include one or up to two specific targets for B. pertussis and B. holmesii (see Table S1). To our knowledge, only two published methods have reported the use of specific targets for the three most relevant Bordetella species (15,16).
The present study aimed to adapt, optimize, and validate a diagnostic algorithm for the rapid detection and identification of B. pertussis, B. parapertussis, and B. holmesii. In addition, we sought to assess the disease burden caused by these species in our region by implementing the algorithm in pediatric patients suspected of pertussis.
(This work was presented in part at the 27th European Congress of Clinical Microbiology and Infectious Diseases, Vienna, Austria, 2017 [17].)

MATERIALS AND METHODS
Study design and setting. Nasopharyngeal aspirates (NPAs) were prospectively collected from children and adolescents Ͻ18 years with clinical suspicion of whooping cough (according to CDC criteria), that were attended in Hospital Sant Joan de Deu (HSJD) between May 2016 and April 2017. This is a pediatric referral hospital that provides medical care services to more than 300,000 children in Catalonia (Spain). Information on age and sex variables of the patients was recorded for epidemiological purposes.
Sample collection and DNA extraction. Nasopharyngeal aspirates were processed according to the protocol established at the clinical laboratory of the study site (18). Specimens showing poor quality (rnase P Ͼ 35 cycle thresholds [C T ]) or weak IS481 positivity (40 Ͼ C T Ͼ 35) were subjected to additional DNA extraction using NucliSENS easyMag (bioMérieux, France), from an initial volume of 200 l eluted into 25 l. qPCR reference method. The standardized laboratory method of Hospital Sant Joan de Déu (HSJD) consisted of a duplex quantitative PCR (qPCR) that included hydrolysis probes (Roche Diagnostics GmbH, Germany) targeting IS481 and the human rnase P gene as a positive internal control for testing sample quality, as described in Brotons et al. (19) (Table 1). Delta Rn (dRn) C T values were manually set at 0.2 for both targets.
Samples yielding an IS481 C T of Ͻ35 were considered probable B. pertussis isolates, inferred from the high copy numbers of such targets in this species (estimated in 50 to 200) (20). Samples with a C T value of 35 to 40 were reported as Bordetella spp., and as negative if C T was Ͼ40.
New multiplex qPCR and diagnostic algorithm. Previously published protocols for diagnosis of whooping cough were reviewed for designing the proposed algorithm (see Table S1). The method by Tatti et al. (15) was selected as the most complete and accurate basis for designing the new diagnostic algorithm, in relation to the number of species covered and the use of the human rnase P gene as the positive control (although not multiplexed). Its confirmatory target for B. pertussis, ptxS1, however, has been reported as cross-reactive with some strains of B. bronchiseptica (8,21), so it was replaced by ptxA-Pr (11). The final design of the algorithm included three sequential qPCR assays for the specific identification of B. pertussis, B. parapertussis, and B. holmesii. The first triplex qPCR included the targets IS481, pIS1001 (B. parapertussis-specific) (13,15), and the human rnase P (Table 1). If IS481 was positive, two confirmatory singleplex qPCRs were performed, ptxA-Pr for B. pertussis identification (21) and hIS1001 for B. holmesii ( Table 2) (22).
Composition of the qPCR reactions only varied from the reference method in the concentrations and sequences of oligonucleotides, probes, and the reagents used for rnase P detection ( Table 1). The ptxA-Pr probe was adapted by TIB-Molbiol (Berlin, Germany) to universal amplification conditions. dRn C T values were set at 0.2 for 6-carboxyfluorescein (6-FAM) or Yakima Yellow (YAK) probes and at 0. DNA standards were freshly prepared, by 10-fold dilutions ranging from 10 6 to 10°GE/ml of sample (10 4 to 10 Ϫ2 GE/reaction and 10 5 to 10 Ϫ1 fg DNA/reaction). Linear range and intra-assay variability were estimated by testing each dilution in triplicate on the same day, and consensus curves were used for calculating the efficiencies. Interassay variability was estimated with two additional replicates on successive days. Precision was acceptable if the mean coefficients of variation (CV) were Յ3% and Յ5% for intra-assay and interassay, respectively.
In addition, a reference panel with clinical samples previously processed by the reference method was also tested by the new triplex qPCR. The purpose of this validation was to compare the performance of the two techniques targeting IS481 (the gene shared by the two methods) in nasopharyngeal matrices. NPAs were collected between December 2015 and April 2016 from children Ͻ1 year old with suspected pertussis. All IS481-positive samples by the reference duplex qPCR (n ϭ 22), in addition to 22 negative samples gathered during this period, were subsequently analyzed by the proposed algorithm. Samples were stored frozen at Ϫ80°C between both analyses.
Statistical analysis. Equations for the multiplex and singleplex qPCRs were calculated by plotting log10-transformed GE versus C T values. Lower limits of detection (LLOD) were estimated using probit  regression analysis at 95% probability (see Table S2). Diagnostic sensitivity and specificity values were calculated as reported elsewhere (23). Continuous variables were described as mean (standard deviation [SD]) or median values (interquartile range [IQR]). Parametric and nonparametric analyses were performed using a t test and the Mann-Whitney U test, respectively. Categorical data were analyzed by 2 or Fisher's exact test. Confidence intervals (CI) were set at 95% and significance at a two-sided P value of Ͻ0.05. All statistical analyses were performed with SPSS v.22 software (IBM Corp., USA), except for diagnostic sensitivity and specificity values, which were calculated with MedCalc Statistical Software v.17.6 (MedCalc Software Bvba, Belgium).
Precision estimates (CV) were Ͻ3% for all reactions, with 0.32% to 0.94% and 0.58% to 1.75% values for intra-assay and interassay, respectively ( Table 3). Results of the specificity panel showed the four Bordetella targets to be genus specific, and the confirmatory targets pIS1001, ptxA-Pr, and hIS1001 were species specific.
The triplex qPCR correctly diagnosed all positive and negative samples of the reference panel of clinical samples. Sensitivity and specificity values for IS481 were both 100% (95% CI, 84.6 to 100.0% and 83.2 to 100.0%, respectively). Mean C T for IS481 in the reference and multiplex reactions were similar despite a freeze-thaw cycle between tests, with a mean difference between paired samples of 0.94 C T (95% CI, Ϫ0.40 to 2.78, P value ϭ 0.159). B. pertussis, B. parapertussis, and B. holmesii. During the study period, 578 NPAs were collected, from which 9 were excluded, as they were either processed by a different technique (n ϭ 2), invalid samples (n ϭ 1), or did not meet age inclusion criteria (n ϭ 6). In addition, 3 samples showed poor quality (rnase P Ͼ 35 C T ) and were also disregarded. A total of 566 samples was finally included in the study. Of them, 484 (85.5%) were negative for the targets IS481 or pIS1001. A seasonal distribution of positive samples was observed, showing a higher incidence during warmer months (P value Ͻ 0.001) (Fig. 1), 62.2% of them identified within  May to July (n ϭ 51). B. pertussis was the most frequently detected species, while B. holmesii and B. parapertussis only circulated during the seasonal peak. The median age of patients was 1.3 years (IQR, 0.24 to 5.85), and ages ranged from 7 days to 17.5 years. The distribution of B. pertussis was homogeneous across age groups, whereas B. holmesii was only detected in five children aged between 4.6 and 9.9 years, and B. parapertussis was only detected in an infant that was two months old (Fig. 2).

Burden of disease caused by
A remarkable difference in the positive rate was observed by age, with the infants younger than 1 year group showing the lowest positivity rate (7.2%). This rate increased with the age of patients (Fig. 2).
No differences in positivity rates were observed by gender, with proportions of 14.8% and 14.2% in males and females, respectively (P value ϭ 0.857). In contrast, B. holmesii seemed to show a differential distribution, being only found in females (n ϭ 5, P value ϭ 0.06). B. parapertussis was only detected in a specimen taken from a male patient.

DISCUSSION
The present study proposed a rapid and easy-to-use protocol based on three qPCR assays for specific DNA detection of B. pertussis, B. holmesii, and B. parapertussis. The multicopy targets IS481, pIS1001, and hIS1001 were shown to be very sensitive, with LLOD values lower than 70 GE/ml of sample. The single-copy target ptxA-Pr had moderate but acceptable sensitivity, being able to detect Ͻ10 3 GE/ml of sample. qPCR efficiencies were above 90.0% for all targets except for hIS1001 (86%), all of them within the acceptability limits for qualitative methods (80.0% to 120.0%) (23). This simple and rapid multiplexed algorithm allowed the specific identification of the three Bordetella spp. in less than five hours.
The validated algorithm was implemented during 12 months at the study site. The  most predominant species was B. pertussis, followed by B. holmesii and B. (27,28). In addition, a study in France described a very high prevalence, up to 20%, of B. holmesii in adolescents and adults with pertussis-like symptoms (29). Our findings, in agreement with Mir-Cros et al. (28), confirm that B. holmesii is currently circulating in our region, and they could denote its increase as a causative agent of pertussis-like disease.
The incidence of pertussis in Spain has increased in all age groups despite the high levels of vaccination coverage (2). In our study, B. pertussis was shown to be evenly distributed among all ages, while B. holmesii was only detected in children Ͼ4 years old. This result is concordant with the higher prevalence of B. holmesii in symptomatic adolescents and adults that was previously reported in several studies (16,29,30).
A seasonal distribution of B. pertussis was observed, with higher occurrence in spring and summer, in line with the epidemiological trends of pertussis in Spain and Europe (31,32). Although data of B. parapertussis and B. holmesii incidence rates presented in this study is limited, it appears to agree with literature suggesting cocirculation with B. pertussis, also supported by a remarkable number of B. pertussis-B. holmesii coinfections reported (16,33,34).
Bordetella infections were equally frequent in both sexes, as described by others (31). Interestingly, B. holmesii infection was only detected in females, although the low number of cases registered does not allow us to reach further conclusions on this potential association.
One study limitation was the lack of identification to species level in 15.9% of IS481-positive samples. Those Bordetella spp. likely corresponded to B. pertussis, in which ptxA-Pr qPCR results were negative due to low bacterial load or, less probably but possibly, to B. bronchiseptica strains for which no specific gene was investigated.
In conclusion, the new algorithm allowed improvement of accuracy of microbiological diagnosis of whooping cough at the study site by enhancing specificity while maintaining high sensitivity levels. In addition, we assessed the incidence of Bordetella spp. among the pediatric population of the geographical region of Catalonia during the algorithm implementation period. According to our data, the circulation of nonpertussis Bordetella species in this region seems to be minor and associated with seasonal increase of B. pertussis. Despite not representing a significant contribution to pertussis disease burden, it is essential to monitor the epidemiological patterns of these species to conduct an appropriate surveillance of the disease in our region. Nevertheless, our local data may not necessarily reflect the epidemiological status of Bordetella species in other areas, since differences in the species circulation can be highly influenced according to geographical and time variations (35). Therefore, local evaluations of diagnostic algorithms based on species-specific primers should carefully be undertaken before their implementation in any particular region.
Ethical considerations. This study was approved by the ethics committee of HSJD, in conformity with the Helsinki Declaration of 1975 (revised in 2000); the Spanish Organic Law 15/1999, on December 13th, on data protection; and law 14/2007, on July 3rd, on biomedical research. For the present study, no informed consent was requested, as this is a population-based study in which there were no activities that could compromise laboratory performance, and samples were duly anonymized.

ACKNOWLEDGMENTS
A.V.-R. reports a grant received from Spanish Ministry of Economy and Competitiveness and the European Regional Development Fund (FEDER, Una manera de hacer Europa) (PI16/00247), and personal fees from European Centre for Disease Prevention and Control ("Pertinent" project). She also received financial support from bioMérieux S.A. for presenting part of this work at the 27th ECCMID Congress in Vienna (Austria), in 2017. C.M.-A. reports grants from European Centre for Disease Prevention and Control ("Pertinent" project), from Liliana Godia Foundation, and from CIBER of Epidemiology and Public Health (CIBERESP) during the conduct of the study. D.H., L.A., M.J., I.J., and P.G. have nothing to disclose.
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We acknowledge the Department of Molecular Microbiology of HSJD for its significant cooperation and input into the implementation of the new diagnostic protocol. It is worth mentioning the valuable contributions of the "Pertinent" team (ECDC, EpiConcept and participant sites) during the project discussions when designing the diagnostic protocol. We also acknowledge Pedro Brotons for writing assistance and proofreading of the final text of the manuscript.