Challenges and clinical relevance of molecular detection of Bordetella pertussis in South Africa

Background We assessed the utility of a multi-target, real-time PCR assay for Bordetella pertussis detection and diagnosis in patients with severe respiratory illness (SRI), influenza-like illness (ILI), and asymptomatic controls. Methods Real-time PCR detection of IS481, pIS1001, hIS1001 and ptxS1 was performed on nasopharyngeal specimens (SRI, ILI and controls) and induced sputum (SRI) collected from June 2012 to May 2016 through respiratory illness surveillance. Using PCR cycle threshold (Ct) value cut-offs, IS481 positive cases were classified as confirmed (Ct < 35) or possible (Ct 35–39) pertussis disease. Results Among 12,922 samples, 146 (1.1%) were IS481 positive of which 62% (90/146) were classified as confirmed. The attributable fraction (AF) was 92.2% (95% CI, 65.6 to 98.2%) and 90.5% (95% CI, 57.5 to 97.9%) amongst SRI and ILI PCR-confirmed pertussis cases, respectively. Amongst possible pertussis cases, AF was 36.9% (95% CI, − 142.3 to 83.6%) and 67.5% (95% CI, − 30.6 to 91.9%) in the SRI and ILI groups, respectively. Conclusion All IS481 positive specimens could be considered as B. pertussis infection, and potentially pertussis disease with supportive clinical information. Electronic supplementary material The online version of this article (10.1186/s12879-019-3869-7) contains supplementary material, which is available to authorized users.


Background
Pertussis (whooping cough) is caused by the vaccinepreventable, bacterial pathogen Bordetella pertussis [1,2]. Despite high vaccine coverage with the whole-cell or acellular vaccines and an initial substantial decrease following vaccine introduction, the incidence of pertussis has increased globally during the last two decades [3][4][5][6][7]. This has been attributed to increased awareness by clinicians, more sensitive molecular techniques for diagnosis [4,7], serological markers for identification of infection in adolescents and adults who usually present atypically with B. pertussis [3], pathogen adaptation and/or waning vaccine immunity [6,8].
The changing epidemiology of pertussis disease requires new control strategies but establishing the burden of pertussis disease in low-resource settings is challenging due to the absence of effective surveillance systems. As such, pertussis surveillance can potentially be linked to surveillance for other respiratory illnesses, such as influenza or pneumonia surveillance [9]. Differences in case definitions, however, may result in reduced sensitivity for detecting pertussis cases using these platforms.
Different laboratory methods such as bacterial culture, molecular testing and serology may be used to confirm a clinical diagnosis of pertussis, however, the sensitivity of each method is dependent on the stage of disease and the age of the individual. Culture and molecular testing are recommended for diagnosis during the early (catarrhal and paroxysmal) phases of disease, whereas serology is recommended during the convalescent phase [10,11]. The sensitivity of both culture and PCR appear to be highest in young infants and decline in adolescents and adults [12].
Real-time PCR detection of B. pertussis is limited by a lack of validated and specific gene targets. The most commonly used PCR target is the multi-copy insertion sequence gene IS481 (present in 50-250 copies per genome); however, this gene is also found in B. holmesii and B. bronchiseptica [13,14]. Use of a second confirmatory target such as the pertussis toxin subunit gene (ptxS1) improves specificity of B. pertussis detection [7,15,16].
The high copy number of IS481 enables highly sensitive detection of B. pertussis, however, this results in uncertainty regarding the interpretation of results with high PCR cycle threshold (C t ) values that cannot always be confirmed by detection of a less sensitive, single copy ptxS1 assay. Multiple copies of IS481 increase the risk of laboratory contamination and detection of false positives [17,18]. Bacterial load in the host and resulting C t values may be influenced by time between symptom onset and specimen collection, immune status of the host, prior use of antimicrobial therapy or quality of the sample collected [19][20][21]. The interpretation and clinical relevance of high C t values, in a setting where stringent laboratory control measures are available to prevent and detect IS481 contamination, is not clearly understood.
We aimed to evaluate a multi-target real-time PCR assay and the use of C t value cut-offs for B. pertussis detection and diagnosis in patients with mild or severe respiratory illness at two sentinel sites in South Africa.

Surveillance population
We enrolled patients and asymptomatic controls through two surveillance systems (one for severe respiratory illness (SRI) and one for influenza-like illness (ILI)) from June 2012 through May 2016. Patients of all ages, hospitalised with lower respiratory tract infection, irrespective of symptom duration, were enrolled in prospective, active surveillance as part of severe respiratory illness (SRI) surveillance at Edendale Hospital and Klerksdorp-Tshepong Hospital Complex (KTHC) [22]. From August 2014, case investigation forms used for patient enrolment were amended to capture clinical symptoms consistent with pertussis disease, namely, paroxysms of cough, inspiratory whoop, apnea (with or without cyanosis) and posttussive vomiting. Patients with influenza-like illness (ILI), and asymptomatic controls were enrolled at the respective outpatient clinics affiliated to each hospital [Edendale Gateway and Jouberton clinics [23]]. A case of ILI was defined as an outpatient presenting with acute fever (> 38°C) and/or self-reported fever and cough or sore throat within the last 10 days. A control was defined as a person presenting at the same outpatient clinic with no history of fever, respiratory or gastrointestinal symptoms during the 14 days preceding the visit. Controls were not followed-up for development of respiratory symptoms after enrolment. We aimed to enrol one HIV-infected and one HIV-uninfected control every week in each clinic within the following age categories: 0-1, 2-4, 5-14, 15-54 and ≥ 55 years (controls were enrolled by HIV status so that the study was powered sufficiently for HIV-stratified analysis). Demographic and clinical data were collected by surveillance officers through interview and hospital record review. For children aged < 5 years, immunization history (including pertussis immunization) was documented and confirmed from the vaccination card if available.

Specimen collection and processing
Nasopharyngeal (NP) specimens were collected from patients and controls as follows: NP aspirates were collected from children (< 5 years), whereas both NP and oropharyngeal (OP) swabs were collected from older individuals (≥5 years) and stored together in a single vial of a 3 ml solution of Universal Transport Medium (UTM) (Copan, Italia, Brescia, Italy). Induced sputum was collected from hospitalised (SRI) patients only. NPs were stored at 4°C and transported on ice packs and induced sputum was stored at − 20°C and transported on dry ice to the National Institute for Communicable Diseases (NICD) in Johannesburg for testing. From August 2014, all SRI-and ILI-enrolled individuals with pertussis symptoms had a second NP collected and inoculated into Regan-Lowe medium (Media Mage, Johannesburg, South Africa) for culture at the NICD. Induced sputum and Regan-Lowe swabs were cultured on charcoal agar (Diagnostic Media Products, Johannesburg, South Africa). Plates were incubated at 37°C for seven days and inspected for growth on days three and seven. Suspected B. pertussis colonies were confirmed using matrix-assisted laser desorption/ionization time of flight mass spectrometry (MALDI-TOF, Bruker, Massachusetts, United States) and by real-time PCR detection of IS481 and ptxS1 genes [16].

Detection of Bordetella species by real-time PCR
Total nucleic acids were extracted from 200 μl of UTM or digested induced sputum using a Roche MagNa Pure 96 instrument (Roche Diagnostics, Mannheim, Germany) and MagNa Pure 96 DNA and Viral NA SV Kit (Roche Diagnostics) using the Pathogen Universal protocol. Real-time PCR was performed using an Applied Biosystems 7500 Fast instrument (Applied Biosystems, Foster City, United States) targeting IS481, pIS1001, hIS1001 and ptxS1 as previously described [16]. The targets differentiate B. pertussis, B. parapertussis, B. holmesii and B. bronchiseptica. PCR assays were performed in 96-well plates including 16 negative extraction controls and 16 no-template controls distributed throughout the plate to detect IS481 contamination. An internal validation was performed for the assay and 100% sensitivity and specificity was obtained for the detection of B. parapertussis, B. bronchiseptica and B. holmesii; and 95% sensitivity and 100% specificity was obtained for the detection of B. pertussis. All testing was conducted at the NICD, at an accredited reference laboratory using standardised procedures and testing algorithms.

Real-time PCR result interpretation and PCR quality control
A specimen was considered positive for B. pertussis DNA if both IS481 and ptxS1 were detected with C t values ≤39. Additionally, samples that were positive for IS481 only (with C t values 35-39 and without confirmation of ptxS1) and negative for pIS1001 (B. parapertussis) or hIS1001 (B. holmesii) were recorded as B. pertussis positive. Samples with positive PCR results (IS481 and/or ptxS1) were further categorised as confirmed or possible B. pertussis based on IS481 C t value cut-offs of < 35 and 35-39, respectively. The testing algorithm was as follows: specimens that tested positive for IS481 on initial screening underwent repeat testing, namely, new extraction, and PCR in duplicate. Results were recorded as positive if IS481 was detected in at least two replicates, taking into account the result from the primary screen.
Detection of the human ribonuclease P gene (RNase P) served as a control to exclude the presence of PCR inhibitors and confirm specimen quality [24]. To exclude spurious amplification, a 1014-bp fragment of the IS481 gene from a selection of confirmed (n = 13) and possible (n = 14) B. pertussis PCR-positive samples was amplified and sequenced as described previously [16]. Sequences were analysed using DNAStar Lasergene SeqMan Pro software (DNAStar, Wisconsin, United States) and the Basic Local Alignment Search Tool (BLAST).
Higher than expected PCR detection rates of IS481 (based on baseline data) at any point during the study period were investigated to exclude laboratory and/or environmental contamination: (i) all laboratory processes were audited, (ii) environmental samples were collected from appropriate areas at surveillance sites and laboratory surfaces were swabbed and tested by real-time PCR for IS481 and (iii) individuals that tested positive for B. pertussis during this time were retrospectively interviewed using a standardised questionnaire to collect data on pertussis-related symptoms.

Data analysis
The prevalence of B. pertussis by patient group (SRI, ILI and controls) and 95% confidence intervals (CI) was calculated. To determine the optimal specimen type for molecular detection of B. pertussis, the detection rates in NP and induced sputum were compared using the chi-squared test among individuals that had both specimen types collected and tested. We estimated the fraction of B. pertussis detection (IS481-positive PCR results) attributable (attributable fraction [AF]) to severe (SRI) (including only patients that tested positive on NP specimens) or mild (ILI) illness using unconditional logistic regression by comparing the B. pertussis detection rate among SRI or ILI cases to those of controls. All models were adjusted for HIV status and age. The AF was estimated from the odds ratio (OR) obtained from the regression models using the following equation: AF = (OR-1)/(OR) × 100 as previously described [25]. An OR was used for these calculations as the data were analysed considering the case-control design of this study. The AF in this study provides an estimate of the proportion of pertussis-positive patients who have symptomatic illness (either SRI or ILI as defined in this study) resulting from pertussis infection. This analysis was implemented overall as well as among confirmed (NP IS481 C t < 35) and possible (NP IS481 C t 35-39) cases. In addition, we compared the demographic and clinical characteristics of possible vs. confirmed IS481-positive patients with SRI or ILI using univariate unconditional conventional or penalised (for categorical variables with 0 outcome in one or more categories) logistic regression. For patients with SRI, confirmed and possible cases were compared for (i) patients that tested positive for IS481 on either one or both specimen types (where both specimen types were available and tested positive for B. pertussis, the lower of the two C t values was recorded) and (ii) patients that tested positive for IS481 on NP specimens only. Statistical analyses were performed using Stata® version 14 (Statacorp, Texas, United States). Statistical significance was determined at p < 0.05 for all analyses.

Ethics approval
Ethics for SRI (M081042) and ILI (M120133) surveillance and B. pertussis testing were approved by the University of the Witwatersrand (M130260) and the University of KwaZulu Natal (BF081/12), South Africa.

Study population
From June 2012 through May 2016, 14,178 individuals were enrolled. Testing was conducted on 12,922 (91%) specimens: 5634 (44%) from SRI patients, 4933 (38%) from ILI patients and 2355 (18%) from control subjects (Fig. 1). Specimens were not tested or available for 9% of enrolled individuals as participants were either too sick to have specimens taken or did not give consent for testing or were discharged before specimens could be collected.
Of 3104 induced sputum samples, 3 (0.1%) were culture positive for B. pertussis. From August 2014 to May 2016, 0.1% (4/4583) of NPs were culture positive for B. pertussis. All isolates were from IS481-positive samples with C t values < 35.

Patient characteristics among pertussis-confirmed and -possible cases
Among SRI patients, 1.7% (94/5634) were PCR positive for B. pertussis, of which 60% (56/94) were categorised as confirmed B. pertussis cases. Possible cases were more likely to occur among patients aged 45-64 years than among patients aged < 1 year (OR 6.3, 95% CI 1.7 to 23.8) ( Table 2). Restricting the analysis to include cases that were positive on NP specimens only, there were no differences between the confirmed and possible cases (Additional file 1). There were no differences in the demographic and clinical characteristics between the confirmed and possible cases that tested positive on either NP and/or sputum specimens or on NP specimens only.
Among ILI patients, 1% (47/4933) were PCR positive for B. pertussis, of which 68% (32/47) were confirmed cases. There was no difference in the demographic and Patients with previously diagnosed chronic conditions including asthma, chronic lung diseases, cirrhosis/liver failure, chronic renal failure, heart failure, valvular heart disease, coronary heart disease, immunosuppressive therapy, splenectomy, diabetes, burns, kwashiorkor/marasmus, nephrotic syndrome, spinal cord injury, seizure disorder, emphysema, or cancer. All percentages are rounded off clinical characteristics between confirmed and possible cases in this group (Additional file 2).

Laboratory and site investigation
In August 2014, there was a higher than usual PCR detection rate for B. pertussis  (Table 3). Eleven patients (65%) met the clinical case definition for pertussis disease, namely, cough for > 2 weeks (any duration in infants < 1 year) + one of the following: paroxysmal cough, inspiratory whoop, posttussive vomiting, apnea among infants < 1 year. Of these, 78% (7/9) and 50% (4/8) were confirmed and possible cases, respectively. No epidemiological links were identified between these individuals.

Discussion
In the absence of pertussis-specific surveillance in South Africa, we evaluated the utility of real-time PCR detection of B.
pertussis using an existing SRI and ILI surveillance platform. In our setting, laboratory methods were standardised and B. pertussis laboratory contamination could be excluded. As such, all individuals in whom IS481 was detected and B. holmesii, B. parapertussis and B. bronchiseptica were excluded, were considered to have B. pertussis DNA in their respiratory tract at the time of sampling and were therefore considered positive for B. pertussis infection. B. bronchiseptica has been associated with HIV-infected individuals [26][27][28][29]; however, this pathogen was not detected in our study. In confirmed cases, the attributable fraction of B. pertussis detection to respiratory disease was significant for both hospitalised (SRI) and mild (ILI) cases. However, for PCR-possible cases (both SRI and ILI) there was no significant association with B. pertussis disease.
In individuals with high C t values (> 35), low bacterial loads may represent asymptomatic colonisation, or residual DNA post disease, or pertussis disease [19,30]. Therefore, the use of stringent C t value cut-offs may still result in the misdiagnosis of some true pertussis cases given that adults and adolescents have been shown to harbor lower bacterial loads and may present with milder disease or atypical symptoms [19,30,31]. In our study, just over one-third of B. pertussis PCR-positive cases had low bacterial loads (C t > 35) and were defined as possible pertussis cases. This is potentially due to the case definition that was not specific for pertussis and the enrollment of individuals of all ages. We also showed    that in hospitalised patients, pertussis-possible individuals were more likely to be adults (45-64 years of age) than infants (< 1 year of age) compared with pertussis-confirmed individuals.
In our study, a subset of 17 patients that were PCR positive for B. pertussis (confirmed and possible cases) from 2014 were retrospectively interviewed. The majority of confirmed cases reported clinical symptoms consistent with pertussis; however, 50% (4/8) of possible cases also had clinical pertussis. Since this was only a small group of individuals, we are prospectively conducting active case follow-up to understand the association between PCR C t values and pertussis symptoms. In Canada, Public Health Ontario Laboratories determined the association between PCR C t with pertussis severity and symptoms, and showed that the proportion of patients with pertussis symptoms did not differ between individuals with low (C t < 36) and high IS481 C t values [19].
IS481 is a multi-copy target with up to 250 copies per B. pertussis genome which increases assay sensitivity, but also increases the risk of false positives due to environmental or laboratory contamination [17,18,32]. As such, different C t value cut-offs have been recommended when interpreting PCR data and these vary between studies. In a Tunisian study, B. pertussis cases were defined as PCR positive for IS481 and ptxS1 with a C t < 45, or as Bordetella spp. if they were positive for IS481 only with a C t < 45 [7]. The assay currently used by our laboratory detects one genomic equivalent of B. pertussis at an average C t of 33.3 [16]. This, together with variable PCR testing and interpretation algorithms in some settings, supports the recommendation by the authors of using a more stringent IS481 C t cut-off (< 35) [16]. However, the 95% confidence intervals (28.5-38.1) imply that higher C t values could still indicate positivity and, in a setting where testing procedures are standardised, should not be ignored. Also, C t values in clinical specimens may vary due to specimen composition or may be affected during storage or transit, and thus may not directly emulate experimental sensitivity in a controlled laboratory environment.
Sequencing confirmed the presence of the IS481 gene from a subset of PCR-positive samples, ruling out non-specific amplification, primer dimer formation and probe degradation, however, one B. pertussis confirmed, and eight possible samples could not be confirmed by sequencing likely due to insufficient DNA concentration for Sanger sequencing. Although only 7% of possible B. pertussis cases tested positive for the ptxS1 gene due to the lower sensitivity of the assay compared to IS481, we excluded B. holmesii from all IS481 PCR-positive samples, suggesting that these were positive for B. pertussis.
Earlier studies have indicated that NP is the preferred specimen type for laboratory detection of B. pertussis [10,33,34]. The use of sputum for B. pertussis detection has been suggested by WHO [34], however, supporting data are lacking. In our study, testing of induced sputum in addition to NP samples increased the diagnostic yield of B. pertussis 5-fold among hospitalised SRI patients. This is similar to another South African study where the B. pertussis detection rate was 3.7% when testing NP specimens and increased to 7% when induced sputum specimens were additionally included [35].
The World Health Organization has recommended that all countries perform pertussis surveillance and have suggested that for resource-limited settings it can be linked to surveillance for other respiratory illnesses. Due to the lack of pertussis data in South Africa, our aim was to establish pertussis surveillance by linking it to an already established syndromic SRI and ILI surveillance platform. While not pertussis-specific surveillance, this enabled us to obtain pertussis data for the country at a low cost by leveraging systems and staff already in place. An additional benefit of this surveillance is that children and adults with mild and severe respiratory illness that do not necessarily present with typical pertussis symptoms were included, adding to our understanding of pertussis disease in individuals that are not clinically suspected to have pertussis. However, use of syndromic surveillance platforms for pertussis surveillance also has several drawbacks, which need to be acknowledged. As the case definition for ILI cases and SRI adult cases included fever, it is likely that some pertussis cases were missed, and the data likely represents an underestimation of disease prevalence. In addition, the interpretation and clinical relevance of PCR-positive individuals that do not present with typical pertussis symptoms is not well understood. However, despite these limitations, we detected B. pertussis by PCR in 1.1% of patients with either mild or severe respiratory illness, which correlates with another South African study that enrolled an infant cohort from 2012 to 2014 as part of the Drakenstein Child Health Study in the Western Cape and found an overall pertussis prevalence of 2% among infants presenting with pneumonia [36].
The study has limitations that need to be considered. The surveillance case definition was not specific for pertussis therefore the time to swabbing following disease onset could not be determined. Overall, there was a small number of pertussis cases, which limited our statistical power for analyses, and as a result, differences between the characteristics of confirmed and possible cases which may exist, may not have been detected. Pertussis vaccination history was not available for adults, and for children < 5 years of age the collected data were poor, so the impact of vaccination on pertussis disease and C t value could not be determined.
In a controlled setting with standardised procedures and when contamination can conclusively be excluded, we suggest that IS481-positive specimens with C t values < 40 should be considered indicative of B. pertussis infection. Low IS481 C t values (< 35) are more likely to represent true cases of pertussis disease, whereas specimens with high C t values may represent B. pertussis carriage (as B. pertussis was detected in controls with higher C t values), residual B. pertussis DNA following disease, or active disease especially in older individuals who may present atypically. As such, in a respiratory surveillance setting and for public health reporting, low bacterial loads may not reflect true cases of disease. However, for individual patient management, possible cases should be interpreted in the context of other clinical and epidemiological factors including patient symptoms, age, immune status and epidemiological links with laboratory-confirmed pertussis cases. B. pertussis carriage is not well understood, and further research is required to better understand the significance of PCR detection of B. pertussis in asymptomatic individuals and individuals lacking typical pertussis symptoms.

Conclusion
In conclusion, based on the results presented in this manuscript, we believe that in the absence of a pertussis-specific surveillance, leveraging of an already existing respiratory illness surveillance system as recommended by the WHO will allow countries to monitor the trend of infection over time, detect outbreaks and allow further research into cases that present with atypical pertussis symptoms.