Endemic chikungunya fever in Kenyan children: a prospective cohort study

We conducted the study between March 2014 and October 2018 in two dispensaries, Ngerenya and Pingilikani, both located in Kilifi County along the northern coast of Kenya (Fig. 1). The two dispensaries also lie within the Kilifi Health Demographic Surveillance System (KHDSS) boundaries which covers an area of 891km² with approximately 290,000 residents that are enumerated every 4 months at the household level [24]. Linkage of clinical records at the dispensary and KHDSS surveillance data allowed estimation of disease incidence. The area under surveillance is endemic for malaria transmission and experiences two main rainy seasons; the ‘long rains’ during April to June and the short rains from October to December each year [25]. Residents living within the surveillance area are served by 21 public health facilities operating under the Kenya Ministry of Health guidelines. This includes Ngerenya and Pingilikani dispensaries located in the northern and southern parts of the surveillance area, respectively [24]. The catchment area of the two dispensaries have different rates of mosquito-borne pathogen exposure as inferred from previous malaria studies; Ngerenya and environs experiences low intensity malaria transmission, whereas malaria transmission around Pingilikani is moderate [26].

We drew a random sample of 3500 out of 5669 children, aged < 16 years, who visited the two dispensaries with fever (defined as axillary temperature of ≥37.5 °C) during the study period. Random sampling of individuals was done using the runiform function in STATA/IC version 15.1 (StataCorp College Station, Texas, USA). Finger-prick whole blood samples were collected in EDTA vacutainers from all febrile children for the purpose of malaria rapid diagnostic testing as part of an ongoing malaria surveillance study [27]. The remaining whole blood was stored at − 80 °C until the day of viral RNA isolation (see RT-PCR section below). No other samples were collected. Clinical diagnoses following investigations done at the dispensaries were provided by non-study clinicians.

We seperately conducted a community cross-sectional survey of 435 asymptomatic children aged < 16 years whose residence was within the catchment of the two dispensaries. Serum was obtained from these children and stored at − 80 °C until the day of viral RNA isolation. This cross-sectional community sampling was conducted between April and May 2016 and allowed estimation of the frequency of asymptomatic CHIKV infections in the study population. Ethical approval was provided by the Kenya Medical Research Institute Scientific and Ethics Review Unit and written informed consent obtained from parents or guardians of all study participants (SSC Nos. 3296, 2617 and 3149).

CHIKV infection was defined as detection of CHIKV viral RNA by reverse-transcriptase polymerase chain reaction (RT-PCR) using a published primer-probe set targeting the CHIKV non-structural protein 1 (nsP1) region [28], namely CHIKV 874 (5′-AAAGGGCAAACTCAGCTTCAC-3′), CHIKV 961 (5′-GCCTGGGCTCATCGTTATTC-3′) and CHIKV 899-FAM (5′-FAM-CGCTGTGATACAGTGGTTTCGTGTG-TAMRA-3′). Briefly, 100 μl of finger-prick whole blood samples collected at presentation to the dispensaries was used for total RNA isolation using TRIzol™ Reagent (ThermoFisher) as per manufacturer instructions. Presence of CHIKV viral RNA was then determined using the QuantiFast RT-PCR Kit (Qiagen) in a 25 μl reaction with primers and probe at a final concentration of 100 nM and 20 nM, respectively, and 5 μl total RNA template. RT-PCR assays were done on a 7500 Real-Time PCR System (Applied Biosystems) with cycling conditions as follows: 50 °C for 20 min, 95 °C for 15 min, followed by 45 cycles of 94 °C for 15 s and 60 °C for 1 min. A second confirmatory screen was performed on some of the RT-PCR positive samples using an assay targeting the CHIKV nsP4 region as previously described [29]. For both assays, a positive result was defined as a cycle threshold (Ct) value of < 40. Viral RNA from a cultured CHIKV isolate obtained from a febrile Kenyan patient (GenBank accession: MT526796) was used as a positive control, while RT-PCR mastermix without template was used as a negative control.

RT-PCR positive CHIKV samples were sequenced using the Nanopore MinION technology following PCR amplification using methods described in the PrimalSeq approach [30]. Full details of our sequencing and bioinformatics workflow can be found in the Supplementary Material. In brief, CHIKV-specific multiplex primers were designed using the Primal Scheme software and used for pre-enrichment of CHIKV viral RNA in the samples using a multiplex PCR method. Barcodes and sequencing adapters were ligated into each sample, after which all samples were pooled into a single reaction tube and the library loaded on to a MinION sequencing device for sequencing. The sequences generated in this study were deposited in GenBank, accession numbers: MT526798-MT526807. MUSCLE was used to align the Kilifi CHIKV genome sequences with those available in GenBank from other geographical locations (see Supplementary Material). Maximum Likelihood phylogenies were reconstructed from the alignment using RAxML [31] using the GTR substitution model with 4 gamma categories (GTR + G4) and visualized in FigTree v1.4.4.

Based on previous studies on childhood malaria [32], we expected that some children would present to the dispensaries with fever on multiple occasions during the study period. We therefore restricted our CHIKV RT-PCR screening to the earliest febrile episode (‘index episode’) for each of the 3500 children. However, to identify occurrence of recurrent infections, CHIKV RT-PCR screening was also done on all samples from subsequent febrile illnesses among children that were CHIKV positive at the index febrile episode. We defined recurrent CHIKV infection as occurrence of more than one episode of fever accompanied by a CHIKV RT-PCR positive test in the same individual.

Data from the dispensaries were linked to the larger KHDSS data using unique person identifiers allowing us to estimate disease incidence in the population, and to assess the influence of various demographic variables on CHIKF incidence. We decided, a priori, to restrict the incidence analyses to a 5 km radius around each dispensary which we assumed to be the catchment area. Because we only sampled from all the visits (with febrile illness) made to the dispensaries, we estimated what the expected cases of CHIKF (out of all visits due to febrile illness) would be by multiplying the test positive rate from the sample by the total number of visits due to febrile illness. This was based on the assumption that the test positive rate among the randomly selected would hold true for the remaining unsampled visits due to febrile illness. Incidence rate was then computed as total number of expected CHIKF cases divided by total person years of observation (pyo) and was expressed per 100,000 pyo. Incidence rate ratios (IRRs) comparing incidence across various socio-demographic variables were computed using a univariate negative-binomial regression model. All statistical analyses were conducted using STATA/IC version 15.1 (StataCorp College Station, Texas, USA).

Between March 2014 and October 2018, there were 29,819 visits by children aged less than 16 years to the two dispensaries (6746 in Ngerenya, 23,073 in Pingilikani), of which 13,696 were febrile (Fig. 2). Of the 13,696 fevers, 1256 lacked a unique patient identifier and 732 visits were missing samples. After these exclusions, there were 11,708 febrile visits from a total of 5569 children (median 1 febrile visit per child during the study period, IQR 1–2) eligible for CHIKV RT-PCR analysis (Fig. 2).

The median age of the 3500 children was 3.1 years (IQR 1.3–6.4), with 1701 being resident in Ngerenya and 1799 in Pingilikani, respectively (Fig. 1). Of the 3500 children 443 were RT-PCR positive (12.7, 95% CI 11.60, 13.80; Table 1), with RT-PCR positivity being twice as frequent in Ngerenya (16.7, 95% CI 15.00, 18.54) than in Pingilikani (8.8, 95% CI 7.61, 10.24). CHIKF prevalence showed no statistically significant variation by sex or age, and in no instance was a clinical diagnosis of CHIKF assigned (Table 1). No clinical diagnosis was significantly overrepresented among CHIKF cases (Table 1). CHIKF prevalence was highest in 2016, when an epidemic was reported in Kenya [23], and was significantly lower in the pre- and post-epidemic years where it ranged between 2 to 6% in Ngerenya and 6 to 10% in Pingilikani (Table 1). CHIKF was more prevalent during January to March in Ngerenya (coinciding with the dry season; see Figure S1) but this was less apparent in Pingilikani (Table 1).

CHIKV RT-PCR positivity was rare among asymptomatic children in the community; only 3 out of 435 asymptomatic children sampled during the cross-sectional survey in the same study locations in 2016 were CHIKV RT-PCR positive (0.7, 95% CI 0.22, 2.12). This yielded an overall clinical-to-asymptomatic CHIKV infection ratio of 18:1 during the epidemic year.

We calculated the incidence of CHIKF restricting our analysis to children aged < 16 years whose residence was within the dispensary catchment area. This represented a total of 136,509 person-years of observation (pyo) over the study period at both locations.

The overall CHIKF incidence during the study period was 314 cases/100,000 pyo (95% CI 285, 345; Table 2). An inverse relationship was evident between CHIKF incidence and age, consistent with acquisition of protective immunity against disease due to ongoing CHIKV exposure in this setting (Table 2). CHIKF incidence was 17% higher in Ngerenya than Pingilikani but the difference did not achieve statistical significance (IRR 1.17; 95% CI 0.97, 1.41); furthermore, while excess incidence was observed in Ngerenya during the 2016 CHIKF epidemic, such an increase was not observed in Pingilikani where CHIKF incidence had been gradually declining since 2014 (Table 2). In Pingilikani the highest incidence was observed during the rainy season (April–June) which contrasted with Ngerenya where the highest incidence was during the dry months of January–March (Table 2). We also observed high incidence in populations living closer to the dispensaries which likely reflects access to care (Table 2) as has been previously reported for malaria incidence [33].

We determined whether recurrent CHIKF episodes occurred in our study population. To do this we identified the 443 children whose index febrile episode was CHIKV positive and screened all their subsequent samples collected during febrile episodes over the period 2014–2018. We defined recurrent episodes as CHIKV RT-PCR positivity during these subsequent febrile dispensary visits (see Methods). Of the 443 children, 170 presented to the dispensary with fever on at least one other occasion during the study period, contributing a total of 320 subsequent febrile dispensary visits with varying durations since the corresponding index CHIKF episode (Fig. 2). Of the 170 children, 19 (11.2, 95% CI 6.86, 16.90) were CHIKV RT-PCR positive on at least one subsequent febrile episode (Figs. 2 and 3). The duration between the index and the subsequent RT-PCR positive febrile episodes ranged from 2 to 43 months (Fig. 3, Table S2). Eighteen of the 19 CHIKV RT-PCR positive cases were detected during or after the 2016 CHIKF epidemic year. Neither age, sex nor geographic location were statistically associated with recurrent episodes of CHIKF (Chi2 test p > 0.05 for all) and the RT-PCR Ct values and clinical diagnoses were comparable between the index and recurrent CHIKF episodes (Figure S2 and Table S2).

We attempted to sequence all index and recurrent CHIKF episode samples from the 19 children (Fig. 3). Complete and partial genome coding sequences were generated from 9 samples, which all mapped to the ECSA genotype, albeit distinct from the clade containing the IOL strains that emerged during the 2004 epidemic and the clade including genomes from CHIKF patients sampled in Mandera, northern Kenya, during the 2016 epidemic (Fig. 4).

The E1 A226V mutation, associated with adaptation and increased transmissibility by Aedes albopictus [15], was absent from all sequences sampled in Kenya [23]. Time-resolved phylogenies showed strong temporal clustering of CHIKV sequences from coastal Kenya that may suggest antigenic changes over time driven by acquisition of herd immunity (Fig. 4). Index-recurrent CHIKF episode genome sequence pairs were available for four children with respective time intervals of 2.9, 3.5, 22.3 and 28.3 months between the index and recurrent episodes (Fig. 4). The degree of relatedness of these index-recurrent episode sequence pairs was also time-dependent, with the sequence pair with the longest intervening time interval (28.3 months) being most divergent (Fig. 4). The relatedness of CHIKV genomes from index-recurrent sequence pairs in the same child was similar to the relatedness of CHIKV genomes from different children (Fig. 4), suggesting that these were reinfections rather than relapses.