Study area
The study was conducted as part of the National Ethiopian Malaria Indicator Survey in 2015 (EMIS-2015). Ethiopia has successfully implemented three national EMISs in 2007, 2011 and 2015 [10,11,12]. EMIS-2015 was embarked during the peak of malaria transmission season, between September and December 2015. The sampling design provided national and sub-national estimates of major malaria intervention indicators, including malaria prevalence for malarious areas of the country, as well as estimates for areas below 2000 m elevation, areas between 2000 and 2500 m, and for 10 administrative regions.
Study design and sample collection
EMIS-2015 was a cross-sectional survey that used a two-stage cluster sampling approach. A total of 555 enumeration areas (EAs) were selected proportional to the population size of the regions, as estimated by the Ethiopian Central Statistics Agency. In each EA, 25 households were randomly selected following onsite EA household mapping. A total of 53,335 individuals were surveyed in 13,875 selected households. Demographic, socio-economic, use of malaria prevention, and malariometric data were collected in the selected households from each EA. Children under 5 years of age in each selected household and persons of all ages in every fourth household were eligible for biological sample testing. Whole blood from a finger prick from consenting individuals was collected for a malaria RDT (histidine‐rich protein 2 (HRP2) and Plasmodium lactate dehydrogenase (pLDH) antigen CareStart®, AccessBio, US), a malaria blood slide (thick and thin blood film), haemoglobin (Hemocue Hb 201+ , Hemocue AB, Ängelholm, Sweden) measurement, and dried blood spot (DBS) samples. Whatman 903 Protein Saver (GE Healthcare, Pittsburgh, PA, USA) filter paper was used for DBS collection. Filter papers were air dried, individually packed in a plastic bag together with a desiccant and stored at − 20 °C at the Ethiopian Public Health Institute (EPHI) before they were shipped for further laboratory-based testing at the US Centers for Disease Control and Prevention (CDC), Atlanta, GA, USA.
Ethical consideration
The EMIS-2015 protocol received ethical clearance from the National Research Ethics Review Committee of Ethiopia and PATH. The survey protocol underwent human subjects review at CDC and was considered to be a non-research programme evaluation activity. Additional ethical clearance for the present serology study was obtained from the Institutional Review Board of the College of Health Sciences of the Addis Ababa, University (AAUMF 03-008).
Blood spot elution, bead coupling and antigens used in the study
Laboratory analysis was conducted as previously described [15]. Briefly, a 6-mm punch was used from each DBS sample and blood eluted to a 1:20 concentration in blocking buffer (Buffer B: Phosphate Buffered Saline (PBS) containing 0.5% Bovine Serum Albumin (BSA), 0.05% Tween 20, 0.02% sodium azide, 0.5% polyvinyl alcohol, 0.8% polyvinylpyrrolidone and 0.5% w/v Escherichia coli extract) overnight.
Six Plasmodium antigens were used for this study: two P. falciparum antigens (Merozoite Surface Protein-1 19kD (MSP-1) and Apical Merozoite Antigen-1 (AMA-1), two P. vivax antigens (PvMSP-1 and PvAMA-1), one P. malariae (PmMSP-1) and one P. ovale (PoMSP-1). The four Plasmodium MSP-1 19kD antigens were produced as recombinant proteins and purified as described previously [18]. The external domain of P. falciparum AMA-1 antigen was produced at the Walter Reed Army Institute of Research (WRAIR) under previously published conditions [19]. The P. vivax AMA-1 antigen was produced at the London School of Hygiene and Tropical Medicine (LSHTM) under previously published conditions [20]. The Schistosoma japonicum glutathione-S-transferase antigen (GST) was produced recombinantly and served as a generic protein to assess immunoglobulin G (IgG) non-specific binding.
All antigens were coupled to magnetic beads (Luminex Corp, Austin, TX, USA) as per prior studies [18, 21]. Briefly, beads were pulse vortexed, transferred to a microcentrifuge tube and centrifuged for 1.5 min at 13,000g. Supernatant was removed, and the beads were washed with a 0.1 M and pH 6.2 sodium phosphate (NaP) solution. Beads were activated by suspending in NaP with 5 mg/mL of EDC (1-ethyl-3-[3-dimethylaminutesopropyl] carbodiimide hydrochloride) and 5 mg/mL sulfo-NHS (sulfo N-hydroxylsulfosuccinimide) and incubating with rotation for 20 min at room temperature (RT), while protected from light. After a wash with coupling buffer (50 mM 2-(4-morpholino)-ethane sulfonic acid, 0.85% NaCl at pH 5.0), antigens were coupled to beads in the presence of a coupling buffer for 2 h at a concentration of 20 ug/mL for all antigens, except for PvAMA1 and GST at 15 ug/mL. Beads were washed once with PBS and suspended in PBS with 1% bovine serum albumin (BSA) with incubation for 30 min at RT by rotation. Beads were then resuspended in a storage buffer (PBS, 1% BSA, 0.02% sodium azide and 0.05% Tween-20) and stored at 4 °C. Coupled beads were run with a panel of known malaria seronegatives to assure minimal non-specific MFI signal would be given by beads used in this study.
The DBS elution was assayed for IgG antibodies using bead-based multiplex technology, and all wash steps were performed with a handheld magnet. In 5 mL of reagent buffer (Buffer A: PBS, 0.5% BSA, 0.05% Tween-20, 0.02% NaN3), a bead mix was prepared with all coupled bead regions included (approximately 625 beads/antigen per well), and 50 uL bead mix was pipetted into each well of a BioPlex Pro plate (BioRad, Hercules, CA, USA). Beads were washed twice with 100 µL PBS with Tween (PBST), and 50 µL of the reagent mix [in 5 mL Buffer A: 1:500 anti-human IgG (Southen Biotech), 1:625 anti-human IgG4 (Southern Biotech), 1:200 streptavidin-PE (Invitrogen)] was added to each well. Negative control samples and a dilution curve of hyperimmune serum were added to each assay plate to monitor any change in control values over the course of the study. Samples of blood eluted from DBS were added to predefined wells (already containing beads and reagent mix) at 1:50 dilution. Plates were incubated overnight with gentle shaking at RT and protected from light. The next morning (approximately 16 h total incubation time), plates were washed three times, and beads were re-suspended with 100 µL PBS and read on a MAGPIX machine (Luminex Corp, Austin, TX, USA). Mean fluorescence intensity (MFI) signal was generated for a minimum of 50 beads/region, and the background (bg) MFI from wells incubated with Buffer B was subtracted from each sample to give a final value of MFI-bg. Minimal variation was seen in the assay signal for the background or positive hyperimmune serum curve values, so no plates were re-run for the study.
Statistical analysis
Data analysis was done using Stata 13 software (College Station, USA). Samples with GST MFI-bg reads above 1000 MFI value (non-specific binding) were excluded from the analysis (n = 678). To dichotomize seropositivity, log10-transformed MFI-bg values were fitted to a two-component Finite Mixture Model (FMM) by the FMM procedure with normal distribution and maximum likelihood estimation outputs. A seropositivity cutoff value was determined by the mean MFI-bg value of the first (assumed seronegative) component plus three standard deviations [18]. Overall, P. falciparum and P. vivax seropositivity were defined as an individual being positive for either or both of the MSP-1 and AMA-1 antigens for each species. Seropositivity estimates for P. malariae and P. ovale are presented using the seropositivity cut-off method outlined above. However, to minimize cross-reactivity between the species-specific MSP-1 antigens, the current study report an additional conservative two-stage approach that increase specificity, but decrease sensitivity of the specific antibodies to the targeted antigens. For this conservative approach for P. malariae and P. ovale, individual readings for PmMSP-1 or PoMSP-1 response first had to be above the MFI-bg cutoff as described above. Additionally, the PmMSP-1 and/or PoMSP-1 MFI-bg signal for that sample also needed to be above the PfMSP-1 MFI-bg signal for the same sample (ratio to PfMSP-1 greater than 1.0) to be considered PmMSP-1 and/or PoMSP-1 positive.
A reversible catalytic model was fitted to the dichotomized data using maximum likelihood methods to generate a seroconversion rate (SCR or λ) and a seroreversion rate (ρ). The model was used to generate age seroprevalence curves, from which a seroconversion rate (SCR) representing the force of infection for the community was calculated. Evidence for two forces of infection was investigated, if visual inspection of SCR curves indicated such a comparison and was guided by a profile likelihood plots to determine the most likely time (year) of change in transmission [16, 22]. Multiple logistic regression models were employed to determine odds ratios (OR); 95% confidence intervals (CI) for gender, age and elevations. Similar regression models were employed for P. falciparum and P. vivax seropositivity. Adjustments were made by region, elevation and age group.
Empirical bayesian kriging in ArcGIS version 10.5 (ESRI, CA, USA) was used to predict the spatial distribution of seroprevalence as a continuous surface of probability of being above a cut-off. This is an established method that uses a statistical model to predict and interpolate the spatial distribution of a variable from available data [23]. Maps were developed separately for MSP-1 and AMA-1 antigens for both P. falciparum and P. vivax for all individuals and for children under 5 years of age. Seroprevalence by woreda/district was compared with the 2015 annual parasite incidence (API) data. QGIS version 2.1.8 and ArcGIS version 10.5 (ESRI, Redlands, CA) were used to produce maps.
Sampling weights calculated during MIS 2015 were used to ensure the representativeness of the samples tested to the study population.