Study populations and serum samples
Serum samples from Day 0 (before vaccination) and Day 84 (4 weeks after the last of three vaccine administrations) were collected from two clinical trials of GMZ2. Details of the volunteers and vaccination schedules are described elsewhere [7, 8]. In brief, two double-blind, randomized phase Ib clinical trials of GMZ2 were performed in Lambaréné, Gabon; one enrolled adults , the other pre-school children . The trial involving healthy Gabonese adults took place between July 2007 and August 2008. Twenty participants received 100 μg GMZ2 adjuvanted with aluminium hydroxide (alum) subcutaneously on Days 0, 28 and 56, whereas the 20 participants in the control group received rabies vaccine intramuscularly at the same time points (Days 0, 28, and 56). The pediatric trial took place from September 2008 to October 2009 and involved 30 healthy pre-school children aged 1 to 5 years. The children received three doses of either rabies control vaccine (n = 10), 30 μg GMZ2 (n = 10) or 100 μg GMZ2 (n = 10). The 3 doses were administered one month apart (Days 0, 28 and 56) by intramuscular injection.
Both studies were reviewed by the regional ethics committee (Comité d‘Ethique Régional Indépendant de Lambaréné; CERIL) and followed Good Clinical Practice guidelines as defined by the International Conference on Harmonization. All studies were conducted according to the principles of the Declaration of Helsinki in its 5th revision.
Plasmodium falciparum culture, synchronization and enrichment for late stages
The laboratory-adapted P. falciparum strain 3D7A, obtained from the Malaria Research and Reference Reagent Resource (ATCC, Virginia, USA) was cultured in complete medium (RPMI 1640, 25 mM HEPES, 2.4 mM L-glutamine, 50 μg/mL gentamicin and 0.5% w/v Albumax). Confirmatory experiments were done using the P. falciparum strain Dd2 obtained from the same source. All cultures were maintained at 37°C in an atmosphere of 5% CO2 and 5% O2, with daily changes of medium at 5% haematocrit and dilution with red blood cells when the parasitaemia exceeded 5%.
Parasite cultures were synchronized at early ring stage by treatment with 5% D-sorbitol (Sigma, St. Louis, USA) for 10 min at 37°C. Isolation of synchronized P. falciparum parasites (late trophozoite and schizont) was performed using LD-MACS magnetic columns (Miltenyi Biotec, Gladbach, Germany), as described previously, at a parasitaemia of about 5% . Following enrichment, the purity of the parasite preparation was verified by light microscopy and by flow cytometry after DNA staining with Hoechst 33342. In later experiments, Vybrant DyeCycle violet stain (Invitrogen, Germany) replaced Hoechst 33342.
Flow cytometry-based immunofluorescence assay to detect anti-plasmodial antibodies
Preparation of parasites for cytometry was based on a previously described fixation protocol . Briefly, P. falciparum culture enriched for late developmental parasite stages were washed once in phosphate buffered saline (PBS) and fixed by incubation in a combination of PBS with 4% EM grade paraformaldehyde (Merck, Germany) and 0.0075% EM grade glutaraldehyde (Sigma-Aldrich, Germany) for 30 min. Fixed cells were washed again in PBS and permeabilized for 10 min in PBS/0.1% Triton-X-100 (TX100) (Sigma-Aldrich, Germany). After another PBS wash step, free aldehyde groups were reduced by incubating cells for 10 min in PBS with 0.1 mg/ml sodium borohydride (Merck, Germany). The preparation was washed again with PBS and cells blocked in PBS/3% BSA. The cells were counted using a haemocytometer (Neubauer–counting chamber) and the pellet reconstituted in PBS to standardize the number of cells used in the assay. As a modification of the original protocol, all subsequent handling of cells in 1.5 ml sample tubes (Eppendorf, Hamburg, Germany) was performed in 96-well round-bottom plates (Corning, NY, USA) instead. To detect parasite-specific immunoglobulin G (IgG), parasite suspension (2 μl of approx. 5.0 x 107 cells per ml) was added into each well of the 96-well plate resulting in a total volume of 100 μl of test sera and control samples (each diluted in PBS/3%BSA) and allowed to bind for 1 h at RT on a plate shaker. After incubation, the cells were washed thrice with 150 μl of PBS to remove excess unbound primary antibody. Subsequently, pellets were resuspended in 100 μl AlexaFluor 488 goat anti-human IgG (Molecular Probes, Germany), diluted in PBS/3%BSA, and incubated in the dark for 1 hour. Following three washes with PBS, cells were stored at 4°C in the dark prior to cytometric analysis.
Antibody dilutions of both primary and secondary antibodies used in the assay were pre-determined through checkerboard titration experiments. The combination of antibody dilutions that gave the best separation between negative and positive fluorescent parasites was selected and used in subsequent experiments. Furthermore, different dilutions of three second-step AlexaFluor-conjugated goat anti-human IgG antibodies as well as a non-conjugated anti-histidine rich protein 2 (HRP2) monoclonal IgM (used as positive control) were tested. In addition, the shelf-life of parasite preparations was estimated by re-assaying at Days 0, 3, 7, and 14, since measurements from large clinical trials may take more than one day and it would be preferable to be able to use one parasite batch for such extended analyses.
Parasites stained i) without primary Ab and ii) with serum from malaria naïve donors followed by the fluorescently labelled secondary antibody were used as negative controls. Positive control serum came from a pool of serum from malaria-exposed semi-immune adults living in Lambaréné, Gabon. As an additional positive control, infected RBCs were stained for HRP2 with a mouse monoclonal Ab (55A, anti-PfHRP2; Immunology Consultants Laboratories, Newberg, USA) at a 20 μg/ml concentration. Detection was performed using a 1/3,000 dilution of AlexaFluor 633 goat anti-mouse IgM (Invitrogen, Germany). Before analysing the cells with a flow cytometer, fluorescence microscopy was done to verify the effectiveness of the fluorescence stains and to verify the cellular localization of Ab-bound parasite proteins.
Flow cytometry data acquisition and analysis
Parasite-infected cells were measured on a Becton Dickinson FACS Canto II flow cytometer equipped with the FACSDiva software version 6.1.2 (BD Biosciences, San Jose, USA) and an attached Carousel loader in high throughput mode. Relative fluorescence intensity of each event was analysed using FACSDiva software version 6.1.2 (BD Biosciences, San Jose, USA). Ab-reactivity was expressed as percentage of positive fluorescent cells (PPFC) and mean fluorescent intensity (MFI). Data acquisition was stopped after 50,000 events for each serum sample tested.
Model-based analysis of flow cytometry data
Several model-based algorithms have been developed to automate the gating process thereby directly addressing several inherent limitations in gating-based analysis . Some of these methods, including two popular model-based approaches, k-means  and an implementation of the Expectation Maximization algorithm (EM)  were tested on two experimental datasets. As part of this work, the Overlap Subtraction Algorithm (OSA) was developed and compared with model-based approaches. All described methods were benchmarked using manual gating as a gold standard. The OSA is implemented in the programming language R and is available from the authors.
Design and mode of operation of the overlap subtraction algorithm
The algorithm effectively mimics manual gating whenever the gate is set with respect to an internal control. It detects overlapping areas of two datasets (e.g. between a control and the measurement of interest) in the two-dimensional space and sets a gate at the border of the overlap. Currently, the algorithm is able to process one colour staining, though it can be easily extended to process multicolour staining. The algorithm accepts files in the flow cytometry standard (FCS) 2.0 and 3.0 formats. MFI and PPFC are computed and reported as output.
With flow cytometry typically a fixed number of cells (e.g. 50,000) C are measured and analysed for each sample. Depending on the nature of the experiment, for each measured cell ci ∈ C a vector of attributes a1…an can be assigned, e.g., colour intensities for different dyes, forward scatter (FSC), side scatter (SSC), etc. Generally, each cell is represented by a data point in the two-dimensional space, defined by the attributes a1 and a2.
The algorithm starts by partitioning the whole value range for each attribute ai
of interest in β equidistant intervals, resulting in the vectors A1
of length β. The next step is to define two | A1
| x | A2
| matrices T
for the test and control sample respectively. Then the values for T
are calculated according to:
Each entry in the matrices T
stores the number of data points |c| whose values for the attributes a1
lie within a certain interval defined by the two vectors A1
. Next, the percentage of data points coming from the test sample is determined according to the following formula:
Following this calculation, positive entries are selected, i.e. entries in R
that exceeds a certain threshold λ. To achieve a high specificity, λ is set to 0.99 by default, meaning that 99 percent of the data points that were counted for a particular entry come from the test sample. The correct gate is then set by finding the ωth
occurrence of an entry with:
The parameter ω controls the sensitivity of the method. In practice it is used to fine-tune the gate’s distance to the negative control. By using low values of ω the gate is set close to the border of the negative control sample. Higher values of ω tend to produce gates that have a bigger gap from the control sample. After selection of relevant entries, the final gate is determined by Loess Regression through the selected coordinates.
Statistical analysis of datasets from different populations
To detect differences in the MFI between groups due to vaccination, a linear regression model was used. To account for baseline differences on Day 0, it was included as covariate in the model (see Formula 4). Raw MFI measurements were log10
transformed before use in further analysis.
For PPFC measurements, which cannot be assumed to follow a normal distribution, standard transformations to achieve normality as proposed by Ahrens et al.  did not work for both datasets. Therefore, log2 fold changes between Day 0 and Day 84 were calculated.
Between-group differences in the children dataset were tested by a one-way ANOVA followed by contrast extraction for comparisons of interest. Effects of vaccination within groups were tested by Student’s t-test.
Between-group comparisons and effects of vaccination in the adult dataset were tested using a non-parametric Wilcoxon test because even after transformation or calculation of ratios the data shows deviations from a normal distribution. To compare results derived manually as well as those obtained by automatic gating, Pearson’s correlation coefficients were calculated using log10 transformed Ab data measured as MFI. For PPFC comparisons Spearman’s rank correlation was used. Agreement between the methods was further evaluated with the Bland-Altman method . The 95% confidence intervals for the mean difference are indicated for all Bland-Altman plots. All analyses were done with R v.2.13.0  and statistical significance was defined as a two-sided p<0.05.