Multiple comparisons analysis of serological data from an area of low Plasmodium falciparum transmission

The survey protocol was approved by CDC’s internal review board and the Haiti ethical review committee in 2012 at the Ministry of Health. All persons participating in the survey were assigned a six-digit identification number that could not be traced back to the individual.

Samples from persons assumed to have never been infected with malaria were gathered from blood donated to a community blood bank in Memphis, TN, USA. All blood units were from persons that had screened negative for HIV and hepatitis B viruses and had no reported history of international travel in the six previous months.

For the area of low P. falciparum endemicity, a tracking results continuously (TRaC) survey was conducted in 2012 as part of Global Fund’s Round 8 activities that builds on the Strategic Plan against Malaria in Haiti. A 2011 nationwide survey reported Haiti to have a low national parasite prevalence of less than 1 % as estimated by PCR [16]. The 2012 survey was cross-sectional and implemented by Population Services International (PSI) utilizing a community-based household design in collaboration with Haiti’s Ministry of Health (MSPP) and the Centers for Disease Control and Prevention (CDC). Dried blood spots (DBS) were obtained for each individual consenting to participate by spotting finger-prick blood onto Whatman 903 Protein Saver Cards or Whatman 1 circular filter paper (GE Healthcare, Piscataqway, NJ, USA), and air drying for at least 2 h. Samples were individually packaged into plastic bags with desiccant and stored at 4 °C prior to analysis. For this study, 580 samples were randomly selected which represented 12 % of the TRaC 2012 sample population. In line with previous national estimates, this subset of samples was found to have a parasite prevalence of under 1 %.

A 6 mm circular punch was taken from the centre of each blood spot, corresponding to 14 μL whole blood, for elution. Samples were shaken overnight at room temperature in 140 μL protein elution buffer containing: PBS (pH 7.2), 0.05 % Tween-20, 0.05 % sodium azide, and stored at 4 °C until analysis. Elution from blood spots provided an initial 1:10 dilution, and samples were further diluted 1:40 in ELISA or Luminex sample diluent for a final whole blood dilution of 1:400, corresponding to a serum dilution of approximately 1:800 with the assumption of 50 % haematocrit in whole blood. Three P. falciparum merozoite surface protein 1 (MSP-1) fragments were employed, including two allelic forms of the 42kD fragment of MSP-1: MSP-1p42(D) and MSP-1p42(F) from the 3D7 and FVO strains, respectively, and one form of the 19 kD fragment (MSP-1p19) fused to glutathione S-transferase (GST) cloned from P. falciparum isolate 3D7. The sequence encoding the MSP-1p19 fragment previously described by Burghaus and Holder [17] was PCR amplified from 3D7 parasite strain genomic DNA using the following forward and reverse deoxyoligonucleotides: 5′-CGC GGA TCC AAC ATT TCA CAA CAC CAA TGC GTA-3′ and 5′-GCG GAA TTC TTA GTT AGA GGA ACT GCA GAA AAT ACC-3′, respectively. The BamHI and EcoRI restriction endonuclease sites used for directional cloning into pGEX 4T-2 (GE Healthcare) are underlined in the sequences, and an in-frame stop codon (italics) was included in the reverse primer. Standard protocols for PCR amplification, cloning into BL21 strain Escherichia coli (Stratagene, LaJolla, CA), expression, and purification of a GST fusion protein product have previously been reported [18, 19]. Fusion protein eluted from the glutathione Sepharose 4B affinity column (GE Healthcare) was dialyzed against 25 mM Tris buffer at pH 7.5 as previously described [20] and loaded onto a MonoQ HR 5/5 strong anion exchange column (GE Healthcare) previously equilibrated with 25 mM Tris buffer at pH 7.5. The column was washed up to 0.2 M NaCl in 25 mM Tris buffer at a flow rate of 1 mL/min, and then the bound GST-MSP-1p19 fusion protein was eluted with a 5 min linear gradient from 0.2 to 0.5 M NaCl in Tris buffer. The fractions that included a large absorbance peak centered at ~0.3 M NaCl were combined, dialyzed overnight at 4 °C against 500 volumes of buffer containing 0.85 % NaCl and 10 mM Na₂HPO₄ at pH 7.2 (PBS) (Spectra/Por 3 dialysis membrane, 3500 Da cutoff, Spectrum Laboratories, Rancho Dominguez, CA, USA), and concentrated using a Centricon-10 centrifugal filter device (Millipore Corporation, Bedford, MA, USA). The final concentration of the GST-MSP-1p19 fusion protein was >0.5 mg/mL (BCA microassay, Pierce, Rockford, IL, USA) with a yield of approximately 0.25 mg/L of E. coli culture. The external domain of Apical Membrane Antigen-1 (AMA-1) and MSP-1p42 antigens were produced at Walter Reed Army Institute of Research (WRAIR) under previously published conditions [21].

Before analysis, ELISA assays were optimized for antigen coating concentration and dilution of horseradish peroxidase (HRP) conjugate to reduce background signal while still providing strong detection of the specified IgG. After optimization, all samples were run under the same conditions using a standard in-house protocol. Briefly, Immulon 2 HB plates (ThermoFisher Scientific, Waltham, MA, USA) were coated with 100 μL of 0.25 μg/mL antigen in PBS overnight at 4 °C and then blocked for 2 h the next day with 5 % non-fat milk in PBS, 0.05 % Tween-20 (PBS-T, Sigma-Aldrich, St. Louis, MO, USA). Samples were diluted in blocking buffer, added in duplicate at 100 μL/well, and incubated for 2 h with gentle shaking. An HRP-conjugated secondary antibody (goat anti-human IgG) was diluted 1:12,000 in PBS-T, and 100 μL added for 1 h. For each well, 100 μL peroxidase substrate (KPL, West Chester, PA, USA) was added and color allowed to develop for 10 min. The reaction was stopped by adding 100 μL 0.2 M H₂SO₄ and plates read on a spectrophotometer (Molecular Devices SpectraMAX, Sunnyvale, CA, USA) at 450 nm. Positive and negative control wells were included on each plate to ensure consistency between readings. At the end of the study, any particular plate was re-run if its controls showed 25 % or more variation among all plates performed for the study.

The same DBS elutions were also assayed using bead-based multiplex technology. Antigens were coupled to BioPlex^® COOH beads (BioRad, Hercules, CA, USA) according to manufacturer’s protocol in the presence of 50 mM 2-(4-morpholino)-ethane sulfonic acid, 0.85 % NaCl at pH 5.0 and an antigen concentration of 20 μg/mL for all antigens. Sulfo-NHS was purchased from ThermoFisher and EDC from Sigma-Aldrich. As a control to test for any serum IgG against GST for the GST-MSP-1p19 fusion protein, a bead was included in the panel which was coupled to GST. Samples were diluted in blocking buffer containing 0.5 % Polyvinyl alcohol (Sigma), 0.8 % Polyvinylpyrrolidine (Sigma), 0.1 % casein (ThermoFisher), 0.5 % BSA (Millipore), 0.3 % Tween-20, 0.1 % sodium azide, and 0.01 % E. coli extract to prevent non-specific binding. Reagent diluent (Buffer C) consisted of PBS-T plus 0.5 % BSA, 0.02 % sodium azide. Filter bottom plates (Multiscreen 1.2 μm, Millipore) were pre-wetted with PBS-T and 1500 beads/analyte incubated with sample in duplicate for 1.5 h under gentle shaking. Secondary antibodies tagged with biotin (1:500 anti-human IgG_1–3, Southern Biotech, Birmingham, AL, USA; 1:2500 anti-human IgG₄, Sigma) were incubated for 45 min, and subsequent incubation with streptavidin–phycoerythrin (1:200, Invitrogen) for 30 min. Plates had a final wash incubation with reagent diluent for 30 min and were read on a Bio-Plex 200 machine by generating the median fluorescence signal for 50 beads/analyte and then the mean fluorescence intensity (MFI) between the duplicate wells. Final MFI was reported for a sample after subtracting MFI values from blank background beads that were included on each plate.

Data were fitted to parametric distributions by statistical software programs SAS 9.3, StataSE 13, and R 3.1.2 and means and standard deviations estimated. Statistical appropriateness of fitted distributions was performed in the SAS software package with the PROC UNIVARIATE procedure and HISTOGRAM statement. The null hypothesis of the Anderson–Darling, Cramér–von Mises, and Kolmogorov–Smirnov tests is that the data of interest are appropriate for the distribution of interest, so a P value of >0.05 would indicate a statistically-acceptable fit. The analysis relaxed the stringency to a P value of >0.01 in order to be inclusive for more tests for the purpose of comparisons between immunoassays and among antigens. When considering malaria-naive US residents as a reference population, extreme outliers could significantly affect the statistical nature of distributions, so any optical density value (OD) >0.3 or MFI >250 was eliminated from the ‘true negative’ analysis. The number of removals for each antigen and each immunoassay is listed in Additional file 1. No individual had more than one value removed for readings on all antigens.

Finite mixture regression models were fitted to the dataset by StataSE’s fmm and R’s flexmix command, as well as SAS’s PROC FMM. All use maximum likelihood estimation to fit an unweighted two-component model of ‘seronegatives’ and ‘seropositives’. The mean and standard deviation were collected from the seronegative component and used for further statistical analysis. Seroprevalence curves and estimates for seroconversion and seroreversion rates were based on fitting data to a reversible catalytic model as described previously [6].

Continuous data were fitted to penalized B-splines for both ELISA and Luminex assays. A B-spline allows a defined number of connected polynomial functions to be fitted to a continuous variable. A cubic function (degree of 3) with 50 knots was used to generate each spline and a very high penalty function (λ = 1,000,000) was imposed to prevent high variation between knots and allow curve smoothing.