Effectiveness of a serological tool to predict malaria transmission intensity in an elimination setting

The present study was conducted during December 2013–August 2014 in two previously malaria endemic districts in Sri Lanka, namely Kurunegala (7.75°N, 80.25°E) and Moneragala (6.66°N, 81.33°E) located in the dry zone lowlands about 127 kilometers apart (Fig. 2). The monthly temperature of the Kurunegala district averages above 30 °C while that of Moneragala is relatively low between 25–30 °C. The mean monthly rainfall in the Kurunegala district varies from 62 to 400 mm with two peaks in April–June (150–250 mm) and in October–November (300–350 mm). Mean monthly rainfall of the Moneragala district varies from 110 – 152 mm with over 80% of rain received during the two rainy seasons i.e. October – January and March – May. Malaria transmission rates were high in both these districts until the year 2005 or so [18] when the API declined drastically (Table 1).

Kurunegala district is considered “a more developed/urbanized area” with more municipal and urban areas compared to the Moneragala district. Furthermore, 1.9% of the population of the Kurunegala district is categorized as “urban” while the entire population of the Moneragala district is categorized under either “rural” (98.1%) or “estate” (1.9%) population, the latter being the resident labourers of areas with plantations (Table 2). By 2012 Moneragala had 21.6% of its population migrated from other districts (i.e. mainly from Badulla and Hambantota districts, which are considered as malaria endemic zones). In Kurunegala district however, the migrant population (12.1%) was from Kandy, Gampaha and Puttalam districts where malaria was not abundant (except in Puttalam). In summary the socio-demographic characteristics differ between the two districts, with the Kurunegala district having a more urbanized population with a higher population density and a considerable percentage of the community engaged in services and industries when compared to the Moneragala district, where the population is more of a rural nature engaged in agricultural activities (Table 2).

All the District Secretariat (DS) divisions in the two districts (i.e. 30 from Kurunegala and 11 from Moneragala) were initially included in the sampling frame. Multi stage cluster sampling was used in a non- random basis (to identify the initial clusters) and random basis (to identify second stage clusters). Four DS divisions (out of the 11 DS divisions) from Moneragala district and 12 DS divisions (out of 30) from the Kurunegala district were selected on the basis of the presence of District/Base hospitals or Medical Officer of Health offices (Fig. 2). Second stage clusters were determined after selecting the initial clusters. Two Grama Niladhari (GN) divisions were randomly selected from each selected DS division by drawing lots (i.e. 8 and 24 GN divisions from Moneragala and Kurunegala respectively). Identification of locations was done within the secondary clusters, after obtaining the relevant area maps from the Division of Survey and Mapping of the Department of Census and Statistics, Sri Lanka [20]. Non random sampling (quota sampling) was carried out to recruit individuals for specific age groups.

The base sample size was manually calculated (based on the district prevalence) at a prevalence of 83% (recorded sero-prevalence in Moneragala district being 83% [15]), and at 95% confidence level and with a 5% margin of error. The final sample (n = 500 for each district) was calculated after correcting the base sample size for differences in design, contingencies, laboratory errors etc. An approximate sample size of 500 also warrants a 50–75% relative length of the 95% confidence interval for estimating sero-conversion rate while studying Southeast Asia populations with entomological inoculation rate (EIR) equal to 0.01 [21]. Furthermore, a sample size of ~500 subjects per district seems sufficient to detect sudden drops in transmission intensity occurred at least 10 years ago under different epidemiological settings [22]. However, the selected sample size does not seem to detect abrupt changes that may have occurred in the short-term with high statistical power. It is therefore, expected that the sample sizes used in this study would be unable to detect more than two points of changes (i.e., both short-term and long-term changes) in the data.

The number of individuals to be included from each DS division was calculated based on probability proportionate to the size of the populations of each DS division.

Ethical clearance for this study was granted by the Ethics Review Committee, Faculty of Medicine, University of Colombo, Sri Lanka.

Blood samples were collected by house to house visits, from people attending communal gathering places and local health facilities. Prior approval to collect blood samples from persons attending hospitals or from the community was obtained from the relevant administrative bodies. The required numbers of samples were collected from these selected individuals after obtaining informed consent. Blood samples from children under 12 years were collected from those who attended the local hospitals and already bled for other tests. Consent for minors was obtained from their parents/guardians.

Every individual who was recruited for the study was given a unique identification number. All personal identifiers (e.g. name, address, and contact numbers) were entered in a data collection book. Other information i.e. age, gender, history of past malaria attacks (during their lifetime) as recalled by the individual and socio-economic data was entered to the data collection sheets prepared under each individual’s unique identification number.

Two milliliters (mL) of blood was drawn by a trained technician or a nurse. This was divided in to two tubes; (1 mL each) one coated with EDTA (Ethylene Di-amine Tetra-acetic Acid) and one plain tube, labeled with each individual’s identification number. The samples in EDTA tubes were stored in freezers at the end of blood collection sessions until transfer to the laboratory in Colombo, where they were stored at −70 °C for future use. Blood in the plain tube was allowed to clot for 4–5 h at 4 °C and was centrifuged at 12,000 r.p.m. (rounds per minute) to separate the serum. Serum was carefully extracted to a separate eppendorf tube labeled with the identification number and was sealed with parafilm. These serum samples were stored in the freezers until transfer to Colombo in frozen condition, in ice packs and were stored at −20 °C until further use.

A standard ELISA (Enzyme Linked Immuno-Sorbent Assay) protocol described elsewhere [15] was used with minor adjustments as follows.

Four selected malaria antigens which are described in literature [5, 10, 17] were used in this ELISA i.e. MSP1₋₁₉-Pf, MSP1₋₁₉-Pv (School of Life Science, University of Edinburgh, United Kingdom); AMA1-Pf or AMA1-Pv (Biomedical Primate Research Centre, The Netherlands). ELISA plates (96 micro-well plates/Nunclon, Germany) were coated with 50 micro liters (μL) of antigen with a concentration of 0.5 μg/mL. Plates were incubated at 4 °C overnight and were washed three times with Phosphate Buffered Saline with 0.05% Tween 20 (PBS/T). To each well 100 μL of diluted serum samples (1: 200 for MSP1₋₁₉ and 1: 400 for AMA1 in 2% skimmed milk in PBS/T) were added. Each ELISA plate contained 46 test samples in duplicate, a known positive control in duplicate, a known negative control and a blank well. The plates were incubated at 4 °C overnight. They were washed three times in PBS/T and 50 μL of horse radish peroxidase-conjugated rabbit anti-human IgG (Sigma-Aldrich/Cat#A8792) diluted in 1: 2000 in PBS/T was added to each well. This was incubated for 3 h at room temperature before washing three times in PBS/T. To each well 50 μL of o-phenylenediamine dihydrochloride (OPD) substrate solution (Sigma-Aldrich/Cat#P1526) (0.4 mg/mL in solvent) was added. This was left for 15 min for the colour to develop. The reaction was stopped by adding 25 μL of 2 M H₂SO₄ per well. Then the plates were read at 492 nm using an ELISA micro-plate reader. The Optical Density (OD) value for each well was recorded separately. Standard procedures were carried out to maintain the validity of the protocols. The same positive control was included in every ELISA plate and a chart was maintained to identify any significant deviations (+/− 2 standard deviations (SD)) of the OD value of the particular sample for normalization of the data. Plots were obtained to check cross reactivity (Additional files 1 and 2).

The baseline (cut-off values) for each antibody was determined with OD values obtained with serum samples of 16 non-exposed, non-immune apparently healthy individuals from Matara and Colombo districts (malaria non-endemic areas) and the cut-off values were determined (as mean) OD + 3SD i.e. 0.1115 for MSP1-Pf, 0.168 for MSP1-Pv, 0.0411 for AMA1-PF and 0.054 for AMA1-Pv) for positive and negative responses to each antigen according to previously used methods [10].

The OD values for each test serum sample were entered into an Excel data sheet together with other information collected on the sample (e.g. location, age, gender, previous malaria exposure). The samples with an OD value exceeding the cut-off value for each antibody was considered as sero-positive for the relevant antibody. The total sero-prevalence, as well as gender and age specific sero-prevalence was determined for each district, DS division and GN division and was compared using Chi squared test with SPSS V15.0 statistical software. Reversible catalytic models were fitted to each sero-positivity data using a Binomial-product distribution for the sampling distribution and estimation methods based on maximum likelihood principle [5, 10, 23]. In each analysis, sero-conversion and sero-reversion rates (SCR and SRR, respectively) were estimated for each district first assuming constant transmission intensity over time (Model 1) and then assuming a change of transmission intensity (Model 2). In the latter, the past and present SCR and the time of change of transmission intensity were estimated using the profile likelihood method. Since models one and two are nested on each other, the comparison of these models was performed using the log-likelihood ratio test. A significance level of 5% was used in this test. Finally, all statistical analysis was conducted in the software package R (3.1.0); the respective scripts are available from the authors upon request.

A total of 1,186 individuals were recruited from the two districts (637 from Kurunegala and 549 from Moneragala). The number of females (n = 469) recruited from Kurunegala district was over three times higher than the male participants (n = 150) but the number of males (n = 263) was comparable to the number of females (n = 286) in the Moneragala district. Gender of 18 participants from the Kurunegala district has not been recorded (Table 3). The proportion of males to females recruited from the two districts was significantly different (Chi = 96.54, p < 0.001).

Age of the study subjects ranged between 1–84 years (mean = 43.31 year, median = 46 years) in Kurunegala and 1–85 years (mean = 26.04 years, median = 23 years) in Moneragala with mean and median ages of subjects in Kurunegala being significantly higher than in Moneragala (t = 15.74, df = 1184, p < 0.01 and Mann–Whitney U = 88597.5, p < 0.001). Mean age varied between 40.2–49.07 years in the DS divisions of Kurunegala, while in the DS divisions of Moneragala it ranged between 23.8–27.3 years. Study participants were divided into 4 age groups as follows: 1–5 years, 6–10 years, 11–20 years and age higher than 20 years. Although the number recruited for the fourth age group (age >20 years) was higher when compared to the other age groups in all the DS divisions in both districts, the deviation was not significant between any of the DS divisions. Numbers of individuals recruited in all age groups (except >20 years) were more in the Moneragala district, when compared to the Kurunegala district, (Table 3; Chi = 148.499, p < 0.001).

In the Kurunegala district 18.8% of the study population recalled (by memory) having had one or more attacks of malaria infections (either P. vivax or P. falciparum) in the past, while in the Moneragala district the percentage of a previous history of malaria was in 21.9%. Previous malaria exposure was comparable between the two districts, with relevant data not available from 102 individuals from the Kurunegala district and 56 from Moneragala district (Table 3; Chi = 0.523, p = 0.507).

The number of individuals with previous malaria exposure was significantly higher in the 11–20 years and >20 years categories in the Kurunegala district (Chi = 10.358, p = 0.016) when compared to the other age groups, but this pattern was not observed among the participants from the Moneragala district. Previous exposure to malaria was not significantly different between males and females in either district (for Moneragala Chi = 1.133, p = 0.287, for Kurunegala Chi = 7.53 p = 0.110).

Sero-prevalence of each antibody was calculated for each district for each antibody. Over 60% of the population was sero-positive for one or more anti-malaria antibodies in both districts (Table 4). Higher proportion of the population was sero-positive for AMA1 antigen in both malaria species than to MSP1₋₁₉ antigen (Table 4). The proportion of sero-positive individuals for each of the four tested anti-malaria antibodies was significantly higher in Moneragala with the odds of being sero-positive in Moneragala being1.4 – >5 times higher when compared to Kurunegala (p < 0.05) (Table 4).

Within each district, sero-prevalence did not significantly differ between any of the DS divisions in either district. Sero – prevalence of males and females in Moneragala districts were comparable. However, the sero prevalence between sexes in the Kurunegala district showed a higher prevalence in males, the differences of which were marginally significant for AMA1-Pf (p = 0.031) and for any Pv (p = 0.044)/any Pf (p = 0.045) antibodies (Chi square Test).

The sero-prevalence of the subjects in age groups 1–5 years and 6–10 years were significantly lower, when compared to the values of the individuals age > 11 years (11–20 year age group and > 20 year age group) for all antibodies in Kurunegala district (p < 0.01, Chi square test). In Moneragala district, this difference was even more prominent with the sero-prevalence rates being markedly low in younger age groups up to 20 years (p < 0.01, Chi square test).

Over 50% with a history of previous exposure to malaria were sero-positive for AMA1 antibodies in both districts (54.2% for P. vivax, 55.8% for P. falciparum in Kurunegala district and 59.2% for P. vivax, 56.7% for P. falciparum in Moneragala district). However, the sero – prevalence rate for MSP1₋₁₉ antigen was relatively lower among individuals with history of previous exposure in Kurunegala district (i.e. 25.8% for P. vivax, 31.7% for P. falciparum), when compared to those in Moneragala district (53.3% for P. vivax, 51.7% for P. falciparum). Though the antibody prevalence for either vivax or falciparum antigen, was high (>55%) in individuals with history of previous exposure in both districts, the association between sero-positivity and history of malaria exposure by recall was relatively poor and was not statistically significant for either district (Chi square test/Phi-Cramer’s V statistics).

Changes in transmission intensity were identified for the two districts, with regard to MSP1₋₁₉ -Pf, MSP1₋₁₉-Pv, AMA1-Pf and AMA1-Pv antibodies separately (Table 5). The population was sub-divided in to age groups with intervals of 20 years i.e. 1–20 years, 21–40 years, 41–60 years and 61–80 years. The data was tested using two models based on assumptions (a) presence of a single force of infection (Model 1) and (b) in the presence of changes in transmission (Model 2).

The results obtained for the Kurunegala district using the reversible catalytic model were with several discrepancies especially with Model two that assumed the presence of a change in transmission. The results were incompatible with the observed data in contrast to the outcome of Model one which indicated a single force of infection with a sero-conversion rate (SCR) of 0.0079^-yr and a sero-reversion rate (SRR) of 0.0048^-yr for MSP1₋₁₉ and a SCR of 0.0785and a SRR of 0.0608 for AMA1. Although model two indicated a significant change in transmission when P. vivax data were considered, the probable time point of such a change failed to overlap, (with time point estimated as 30 years whenMSP1-Pv data was used and 5 years with AMA1-Pv data) (Table 5).

The change of transmission for Moneragala district appears to have occurred for P. falciparum approximately 10–14 years and approximately 15 years ago for P. vivax (Fig. 3). The annual SCR decreased from 0.0192 – 0.0011 per year according to MSP1₋₁₉ data and from 0.1749 – 0.0231according to AMA1 data for falciparum malaria and from 0.1312 – 0.0022 per year according to MSP1₋₁₉ data and from 0.1998 – 0.0256 according to AMA1 data for vivax malaria (Table 5).

Sero-conversion was also calculated using the sero-positivity for any falciparum/any vivax anti-malaria antibody. This confirmed the time of change in transmission in Moneragala district, to have occurred 10–14 years and ~15 years prior to serum collection for falciparum and vivax malaria respectively (Table 5). However, data obtained for any P. vivax/any P. falciparum. antibody for Kurunegala district failed to detect a changing point in transmission.