Female TO mice were treated with phenylhydrazine three days prior to infection by i.p inoculation with Plasmodium berghei constitutively expressing GFP (PbGFPCON - derived from the ANKA line) 13 . To ensure reproducibility and high infectivity, the parasites had been passaged between mice no more than eight times since previously passing through mosquitoes. After four days of infection, exflagellation was tested by spotting tail blood into ookinete medium on slides for 15 mins at 20°C. If >5 exflagellation centres were observed in each of five random fields of view (×40 objective) in which the blood cells form an even monolayer, then the mouse was exsanguinated and the infected blood used in the SMFA.

Two anti-malarial compounds - dehydroepiandrosterone sulphate (DHEA-S) and lumefantrine (LUM), and a known inhibitor of microgametogenesis and infection (positive control) - cycloheximide (CH) 14 , were selected for evaluation of the counting macro. DHEA-S and CH were dissolved in DMSO and LUM was dissolved in dimethylformamide (DMF). All were stored under dry N₂ gas as 30 mM stock solutions. Compounds and solvents were provided blind to the experimenter and only un-blinded when all experiments were complete. All compounds were used in the transmission-blocking assay with no more than a 0.1% final solvent concentration. 0.1% DMSO or DMF were used as 'solvent controls'.

For each of the three replicate experiments, three infected mice (see above) were used per assay. They were anaesthetized, before being rapidly bled by cardiac puncture and their blood pooled. 500 μl of infected blood was mixed with each compound to be tested, or DMSO/DMF as a control, to yield a 10 μM final concentration of compound. The infected, drug-treated blood was then injected into a membrane feeder pre-warmed to 39°C and offered to overnight-starved Anopheles stephensi (strain SDA 500) mosquitoes in groups of ~80, which were then allowed to feed on the blood for 30 min 4 . The next day, mosquitoes that had not fed were removed and the remaining mosquitoes maintained at 19°C and 80%RH, being fed on a fructose/p-aminobenzoic acid (PABA) solution 4 which was replenished every 2-3 days. At 7-9 days post feeding, mosquito midguts were dissected into PBS and the pooled material fixed in PBS containing 4% paraformaldehyde (PFA) for 30 min at room temperature. The guts were then washed with PBS and stored in PBS in the dark at 4°C for no more than 3 days before being counted.

Groups of 20 fixed guts were mounted on glass slides containing a small amount of Vectashield mountant and a coverslip applied. To facilitate imaging, guts were then flattened by gently withdrawing some fluid from the preparation. The slides were observed with a Leica DMR microscope at ×5 magnification using a Leica HCX PL FLUOTAR 5×/0.15 lens (to give maximum optical depth of field combined with adequate spatial resolution) and guts imaged with a Zeiss AxiocamHR camera controlled by AxioVision v4.7.2 software. Images from the same set of feeds were all taken at a resolution of 1388 × 1040 pixels with the same exposure setting and guts were all imaged on the same day under the same conditions. This was done to ensure that there was minimal variation in fluorescence intensity between guts due to the degradation of GFP within the oocysts over the time they were stored.

The acquired images of the infected mosquito midguts can be broken down into two components: 1. Oocysts that are small (~7-15 pixels/14-30 μm diameter), round and show high GFP fluorescence; 2. Midgut tissue that is larger than the oocysts (~600 × 400 pixels/1200 × 800 μm), and has fainter variable and non-uniform yellow-green autofluorescence. Even though oocysts are brighter in greyscale than the surrounding midgut tissue, it is impossible to reliably separate them by simply thresholding the image (Figure 1). A low threshold level will leave traces of the gut in the image and thus obscure the oocysts (Figure 1b), whereas a high threshold level despite giving a cleaner image, does so at the cost of losing oocysts with fainter fluorescence (Figure 1c). Also, as midgut background fluorescence varies between guts, it is impossible to set a standard threshold level that will apply to all images to give consistency in analysis.

An automated oocyst counter must therefore be able to process each image in exactly the same manner to give reproducible data, therefore, background fluorescence must be standardized before an accurate counting analysis can take place. Experimentally this is very difficult to achieve and so a macro for ImageJ was developed that can take any midgut image and process it with a simple set of standard image transformations that remove the background tissue fluorescence whilst preserving the oocyst component of the image (Figures 1d and 1e).

The background fluorescence for each image is minimized by taking a copy of the image, displacing it horizontally by 2 pixels (equivalent to ~1 μm of the specimen), and then subtracting it from the original image. Subtracting a copy of the image slightly out of phase with the original causes large areas of relatively uniform and weak fluorescence intensity to be cancelled out (midgut tissue). Small areas (~7-15 pixels diameter) of bright fluorescence (oocysts), whilst diminishing in size and intensity remain distinct (Figure 1d). A standard threshold can then be applied to all images such that they highlight only the oocysts (Figure 1e). The standard threshold level that correctly identifies the oocysts is determined by visually evaluating three representative images from the dataset, and then is applied to all images in the batch of images for a particular group of feeds. The macro then uses the particle counter function built-in to ImageJ to count the oocysts it has identified. The counts generated are then automatically tabulated, and can be imported into MS Excel or similar programs for analysis.

The ImageJ particle counter function used by the counting macro allows the user to specify the minimum particle size counted. ImageJ defines a particle as an object of undefined shape with an area measured in pixels squared (pixels²). For example, objects of 2 × 2 pixels and 1 × 4 pixels in size are both classified as being 2 pixels². This value was hypothesized to be critical for the accuracy of counting. To test this, gut images from three gametocyte membrane feeds (n = 45, 49 and 51 mosquitoes) were counted by the macro using different minimum particle sizes. To enable the precision of the macro counts to be evaluated, the same midgut images were then recounted and recorded manually by an experienced trained observer. The observer recorded each identified oocyst individually and used their own judgement to discriminate between overlapping oocysts.

It was found that by increasing the minimum particle size the counter detects from 1 to 8 pixels², the number of particles identified by the macro as oocysts decreased - with 1 pixel² notably over-counting (by 897.29%) and 8 pixels² under-counting (65.83% of manual count) (Figure 2a). To investigate this effect, three images were randomly selected from the experiments and the fluorescence intensity of the individual oocysts present in the guts (n = 81 oocysts) measured and related to whether that particular oocyst was counted or not for different minimum particle sizes (Figure 2b). The macro was most insensitive to identifying oocysts that had a low integrated fluorescent intensity (pixel intensity × area). With increasing minimum particle size, the population of oocysts with low integrated intensity that were not identified rose, and also brighter (higher integrated intensity) oocysts began to be missed (Figures 2b and 2d). The over-count of oocysts produced by a minimum particle size of 1 or 2 pixels² (Figure 2a) was found to be due to false identification of random pixel noise that was smaller in size than the oocysts (Figures 2c and 2d).

The effect of particle size on calculating the prevalence of infection was also tested by imaging 100 uninfected guts, and then counting them using the macro set at the different minimum particle sizes (Table 1). Due to high false-positive rates a minimum particle size of 1 or 2 pixels² gave a prevalence of 100%. At 3 pixels² this dropped to 9% and by 6 pixels² had reduced to 2%. The 2 images that the macro incorrectly identified at this size contained either abnormally high background fluorescence in the midgut tissue or contained a single fluorescent artefact that was the same size and shape as an oocyst.

As no setting identified oocysts with 100% accuracy, a minimum particle size of 3 pixels² (recognizing objects in the processed image containing 9 pixels or greater) was selected for use in further experiments with this material as it most closely balanced out the false positives and false negatives produced by the macro (Figure 2d).

To check for reproducibility, the oocyst counts for individual infected midgut images from the three test gametocyte feeds analysed with a minimum particle size of 3 pixels² were plotted on a graph against the corresponding manually determined oocyst count (Figure 3). Trendlines were fitted to the data (not shown) that gave R² values of 0.98, 0.97 and 0.97 respectively (Figure 3), suggesting that although at this setting the macro under-counts the number of oocysts present by 13.47% (Figure 2a, 100% minus 86.53), most importantly it is invariant in the manner that it treats each image.

10 μM concentrations of CH, DHEA-S, and LUM, and DMSO or DMF-only carrier controls were added to parasite infected blood and fed to mosquitoes (n = 28-58) in triplicate experiments. At day 7-9 the guts were dissected, imaged and then counted both by the macro at a minimum particle size of 3 pixels² and manually. As the mean oocyst intensities of the solvent control varied between the three replicates, (for DMSO, between 71.51 and 180.37 oocysts per gut by manual count), feeds were analysed by expressing the oocyst intensity for each individual replicate as a percentage of its respective solvent control, and then taking the mean of the three replicates (Table 2). This was performed with the oocyst intensities generated for either the manual or automated counting methods and then compared. Both counting methods firmly agreed that DHEA-S was a very weak inhibitor of sporogony (Manual = 17.79% reduction in oocyst intensity; Automated = 27.41% reduction in oocyst intensity), whilst LUM (Manual = 89.15% reduction; Automated = 84.75% reduction) strongly inhibited sporogony. The prevalence of infection as determined by the macro count agreed well with the manually determined prevalence (Table 2), with LUM showing a statistically significant decrease in prevalence (P =< 0.004 unpaired Student's T-Test). DHEA-S has been shown to exhibit anti-malarial activity by inducing phagocytosis of asexual ring stage parasites 15 . The data presented here suggests that it exerts little or no effect on mosquito stage parasites (Table 2). LUM has been reported to be schizonticidal 16 , most likely exerting its effects through interacting with the heme detoxification pathway 17 and is most commonly used today in combination with artemether (Coartem^®) for treating malaria 18 . Interestingly, it exhibits a strong inhibition of sporogony in P. berghei (Table 2). The 10 μM concentration we have tested is comparable to serum levels detected in malaria patients treated with LUM 19 . As a consequence, it would be prudent to investigate whether LUM affects P. falciparum in a similar manner.

By manual count, the earlier observations of Toyé et al 14 were confirmed in that the protein synthesis inhibitor CH totally inhibited parasite development, and all guts were free of oocysts. The counting macro nonetheless returned a mean oocyst number of 0.20% for CH compared to the DMSO control, and mean prevalence of 5.61%. When specifically reviewing these images, falsely reported oocysts were identified as containing oocyst-sized fluorescent artefacts. This may be controlled experimentally by imaging and counting a series of uninfected guts at the same time as the experimental guts and taking the calculated prevalence in the uninfected controls as the 0% prevalence baseline for that experiment.

Under the conditions described in this report, researchers were able to perform 20 experiments per week on a sustained basis using the macro. This compares very favourably with manual counting, which permits only four experiments per week to be completed on a sustained basis. The counting macro, therefore, as well as saving time, also reduces the cost of performing SMFAs and increases the reproducibility of their analysis.