Modelling the relative abundance of the primary African vectors of malaria before and after the implementation of indoor, insecticide-based vector control

Abstract

Background: Malaria remains a heavy burden across sub-Saharan Africa where transmission is maintained by some of the world's most efficient vectors. Indoor insecticide-based control measures have significantly reduced transmission, yet elimination remains a distant target. Knowing the relative abundance of the primary vector species can provide transmission models with much needed information to guide targeted control measures. Moreover, understanding how existing interventions are impacting on these relative abundances highlights where alternative control (e.g., larval source management) is needed.

Methods: Using the habitat suitability probabilities generated by predictive species distribution models combined with data collated from the literature, a multinomial generalized additive model was applied to produce relative abundance estimates for Anopheles arabiensis, Anopheles funestus and Anopheles gambiae/Anopheles coluzzii. Using pre- and post-intervention abundance data, estimates of the effect of indoor insecticide-based interventions on these relative abundances were made and are illustrated in post-intervention maps.

Results: Conditional effect plots and relative abundance maps illustrate the individual species' predicted habitat suitability and how they interact when in sympatry. Anopheles arabiensis and An. funestus show an affinity in habitat preference at the expense of An. gambiae/An. coluzzii, whereas increasing habitat suitability for An. gambiae/An. coluzzii is conversely less suitable for An. arabiensis but has little effect on An. funestus. Indoor insecticide-based interventions had a negative impact on the relative abundance of An. funestus, and a lesser effect on An. arabiensis. Indoor residual spraying had the greatest impact on the relative abundance of An. funestus, and a lesser effect on An. gambiae/An. coluzzii. Insecticide-treated bed nets reduced the relative abundance of both species equally. These results do not indicate changes in the absolute abundance of these species, which may be reduced for all species overall.

Conclusions: The maps presented here highlight the interactions between the primary vector species in sub-Saharan Africa and demonstrate that An. funestus is more susceptible to certain indoor-based insecticide interventions than An. gambiae/An. coluzzii, which in turn, is more susceptible than An. arabiensis. This may provide An. arabiensis with a competitive advantage where it is found in sympatry with other more endophilic vectors, and potentially increase the need for outdoor-based vector interventions to deal with any residual transmission barring the way to malaria elimination.

Fig. 1

Relative abundance of each vector species predicted by the multinomial generalized additive model (no control). Plot a shows An. arabiensis, plot b shows An. funestus and plot c shows An. gambiae/An. coluzzii. The relative abundances for each cell across all three maps sums to one where any vector species is present, or zero where all species are absent. The extent of the predictions are restrained to the probability of presence extents as predicted by the BRT outputs [15], where probability of presence ≥0.5. These, in turn were restricted in the original maps by masking the predictions at a buffer of 1500 km. When interpreting these maps, note they do not show absolute abundance. For example, despite a clear presence of An. funestus on the northern fringes of the Sahel, the maps are showing it is more abundant than both An. arabiensis and An. gambiae but not that it is found in abundance. Conversely, the An. funestus map shows that it is less abundant than An. gambiae in central Africa, not that it does not exist here (An. gambiae and An. funestus are both highly efficient and dangerous vectors in central Africa)

Fig. 2

Conditional effect plots for a multinomial generalized additive model of mosquito relative abundance. Plot a shows An. arabiensis, plot b shows An. funestus and plot c shows An. gambiae/An. coluzzii. Each plot illustrates the relationship between the predicted relative abundance of the three species and predicted habitat suitability (from Sinka et al. [15] on the logit-scale, rescaled to the unit interval) for each vector. In each plot the habitat suitability values for the other two species are held at their mean values in the dataset: An. arabiensis 0.568; An. funestus 0.450; An. gambiae/An. coluzzii 0.606. The predictions are faded out where the suitability scores are outside the 95 % quantiles of the dataset, where the model had less data with which to estimate the curves and the predictions are therefore less certain (i.e. the suitability scores of the species in question for all of the ratio records except the 2.5 % of datapoints with the lowest suitability scores, and the 2.5 % with the highest, all fall within the full-colour region)

Fig. 3

Red–green–blue plots for pre- (a), and post- (b) intervention (indoor residual spraying) relative abundance. Red An. arabiensis, blue An. funestus and green An. gambiae/An. coluzzii with intervening colours indicating the transitioning increasing or decreasing relative abundance between species

Fig. 4

Effect of indoor-based insecticide control on the relative abundance of African vector populations. Plots show the effect of indoor residual spraying (a) and long-lasting insecticidal nets (b) on the relative abundance of An. funestus (blue), An. gambiae/An. coluzzii (green) and An. arabiensis (red), with a hypothetical pre-intervention relative abundance of 0.1 (An. funestus), 0.43 (An. gambiae/An. coluzzii) and 0.47 (An. arabiensis), indicated by vertical dashed lines. Estimates of post-intervention relative abundances are uncertain and are represented by their 95 % confidence bands (coloured rectangles) and maximum likelihood values (vertical solid lines). The maximum likelihood estimate of the intervention effect on relative abundance for each species is indicated by the arrows linking the pre-and post-intervention estimates