Pathogens compete for hosts through patterns of cross-protection conferred by immune responses to antigens, the phenotypic traits whose variation underlies competition. Many pathogens exhibit extreme antigenic diversity; this is particularly the case for Plasmodium falciparum and its major antigenic protein (PfEMP1) encoded by the multi-gene family known as Var. There are about 60 Var genes per parasite genome and tens of thousands of variants in local populations of regions with high malaria transmission. Previous ‘strain theory’ posits that despite high rates of recombination, frequency-dependent competition for hosts (immune selection) plays an important role in structuring parasite populations into coexisting strains with limited overlap of Var repertoires, a pattern akin to emergent niches with limiting similarity in ecology. These earlier models were not able to account for the full complexity of the system, including the large size and the continuous innovation of the variant pool. Furthermore, empirical evidence for strain structure did not rely on comparisons to predictions from neutral models (solely based on demography). We present an ecological extension of strain theory for the hyper-diverse ‘Var’ system that addresses these limitations, and allows us to identify signatures that differentiate underlying processes based on networks of genetic similarity.
Results/Conclusions
We use an individual-based, stochastic epidemiological model to compare outcomes of antigen-specific immune selection to those generated by two neutral scenarios: generalized immunity and neutral genetic drift. We analyze the genetic structure of the parasite population using networks in which edges encode genetic distances between pairs of either Var genes or genomes. We show that these similarity networks exhibit distinctive signatures of specific immune selection. In particular, networks of immune selection are partitioned into clusters of highly similar Var genomes, with reduced genetic overlap when compared to those under neutrality. When inspected across time, strain modules persist longer and with more even frequencies. At the limit of very high transmission, most parasites are highly dissimilar from each other despite high recombination, in a characteristic structure revealed by combined network features. Analyses of Var tag data from deep sampling of local populations in Ghana implemented with machine learning algorithms of network classifications, reveal the resemblance of empirical patterns to those expected under immune selection. Hence, our method enabled us to recover a signature of frequency-dependent competition in empirical data despite enormous strain diversity. We argue that this characteristic genetic structure matters to epidemiology and hence to population dynamics.