Plasmodium falciparum genetic diversity in coincident human and mosquito hosts

Population genetic diversity of P. falciparum antigenic loci is high despite large bottlenecks in population size during the parasite life cycle. The extent of this diversity in human blood-stage infections, following expansion from a small number of liver-stage schizonts, has been well described. However, little is known about parasite genetic diversity in the vector, where a similar bottleneck and expansion occurs following parasite mating and where parasite genotypes from several different human infections may accumulate. We assessed parasite genetic diversity within human and mosquito P. falciparum infections collected from the same households during a 14-month longitudinal cohort study using amplicon deep sequencing of two antigenic gene fragments (ama1 and csp). To a prior set of infected humans (n=1175/2813; 86.2% sequencing success) and mosquito abdomens (n=199/1448; 95.5% sequencing success), we added sequences from infected mosquito heads (n=134/1448; 98.5% sequencing success). Across all sample types we observed 456 ama1 and 289 csp unique haplotypes. While both hosts contained many rare haplotypes, population genetic metrics indicated that the overall and sample-level parasite populations were more diverse in mosquitoes than in humans, and infections were more likely to harbor a dominant haplotype in humans than in mosquitoes (based on relative read abundance). Finally, within a given mosquito there was little overlap in genetic composition of abdomen and head infections, suggesting that infections may be cleared from the abdomen during a mosquito’s lifespan. Taken together, our observations provide evidence for the role of the mosquito vector in maintaining sequence diversity of malaria parasite populations. Significance statement Concurrent infections with multiple strains of Plasmodium falciparum, the leading causative agent of death due to malaria, are common in highly endemic regions. During transitions within and between the parasite’s mosquito and human hosts, population bottlenecks occur, and distinct parasite strains may have differential fitness in the various environments encountered. These bottlenecks and fitness differences may lead to differences in strain prevalence and diversity between hosts. We investigated differences in genetic diversity between P. falciparum parasites in human and mosquito hosts and found that, compared to human parasite populations and infections, mosquito populations and infections were more diverse. This suggests that the mosquito vector may play a role in in maintaining sequence diversity in malaria parasite populations.

unexplored. Several studies have compared markers of drug-resistance loci between hosts, and 86 an early report from Zambia observed very different allele frequencies in humans and 87 mosquitoes (3, 4), suggesting differences in parasite population structure between hosts.

88
However, subsequent reports from other settings using different genetic markers have not 89 consistently observed this phenomenon (5, 6). As these studies used marker genes with few 90 polymorphisms, analyses of individuals with complex co-infections was limited. While

104
We previously reported that parasite multiplicity of infection (MOI) as expressed by either 105 marker was higher in mosquito abdomens harboring recently-ingested parasites than humans 106 harboring blood-stage parasites (1). Here, we examine more carefully the differences between 107 host compartments in haplotype diversity and relative abundance both within a given host and 108 at the population level. Based on our previous observation, as well as the robust immune 109 defenses against P. falciparum in humans (9), we hypothesized that the mosquito P. falciparum 110 haplotype population would be more diverse than that of humans.

114
Data overview and analytic population.

115
Samples were collected over the course of 14 months (June 2017 -July 2018) from 38 116 households in three Kenyan villages. Mosquitoes were aspirated weekly from each household 117 and blood samples from household members were collected monthly. To the previously 118 reported data on humans and mosquito abdomens (1), we added data from mosquito heads.

121
Of these, sequencing of at least one marker was successful in 86.2% (1013/1175) of human,

127
Parasite development within the mosquito host begins in the abdomen following which 128 sporozoites must traverse the midgut wall to reach the salivary glands in the head; however, it is 4 137 We next compared the haplotype compositions of infections in the 89 mosquitoes in which P.

138
falciparum was detected in both the head and the abdomen. We calculated the percentage of 139 ama1 or csp haplotypes found only in the head or the abdomen, or observed in both 140 compartments (i.e. the Jaccard distance; intersect/union) within each mosquito. While some 141 haplotypes were observed in both compartments of a given mosquito (mean for ama1: 12.0%, 142 csp: 23.7%), the majority of haplotypes were either private to the abdomen (mean for ama1: 143 50.7%, csp: 41.5%) or head (mean for ama1: 37.3%, csp: 34.8%) (Figures 1B, S2A-B). Despite 144 this limited overlap, sharing between abdomens and heads from the same mosquito was higher 145 than sharing between random pairs of abdomens and heads ( Figure S2C; Kolmogorov-Smirnov 146 p ≤ 1e-10 for both markers).

148
To determine whether the differences in haplotype composition between abdomen and head 149 infections within a single mosquito corresponded to differences at the host population level, we

160
The P. falciparum population in mosquitoes is more diverse than the population in 161 humans.

162
To investigate signatures of differential bottlenecks or selection during parasite transition 163 between mosquito and human hosts, we compared population-level differences in parasite 164 haplotype prevalence among mosquitoes and humans, where differences in prevalence may 165 indicate differential bottlenecks or selection. Across all infections, we observed high haplotype 166 richness, with 456 ama1 and 298 csp distinct haplotypes. The vast majority of these were low-167 frequency haplotypes, many of which were observed in only one host (Figures S3-4). Among 168 54 distinct haplotypes (both ama1 and csp) with a prevalence above 5% across all samples, we 169 observed 28 haplotypes with differential prevalence across hosts: 19 more common in mosquito

173
We next used haplotype prevalence to quantify population-level diversity across orders of 174 diversity (q) ranging from equal weight to each haplotype (q = 0, equivalent to haplotype 175 richness or the number of distinct haplotypes observed) to downweighting rare haplotypes (q = 176 2, effective number of highly abundant haplotypes) (10). The mosquito parasite population was 177 more diverse than the parasite population in human hosts (Figure 2B; S6A). This trend is

183
indicating that mosquitoes contained a larger relative number of infrequent haplotypes. Even so, 184 higher diversity in the mosquito host is still apparent when downweighing the contribution of 185 these minor haplotypes. Taken together, these results indicate that there may be a greater relative loss in diversity across the transition from mosquitoes to humans than humans to 187 mosquitoes.

189
Dominant haplotypes within infections are more common in humans than mosquitoes.

190
In addition to lower population-level diversity in humans compared to mosquitoes, we also 191 observed lower within-sample diversity ( Figure S5) and proportionately more monoclonal

216
We observed, using multiple metrics, more parasite genetic diversity in mosquitoes compared to 217 humans. This high diversity contrasts with the known marked reduction in parasite biomass 218 during the transition from the human to the mosquito abdomen (11), which might be expected to 219 constrain parasite diversity. One potential explanation for this is the possibility of cryptic 220 genotypes in humans undetected by marker sequencing; this has been reported in experimental 221 studies (7), though the large range of MOIs we observed in humans suggests that these 222 infections were not systematically undersampled. Alternatively, the reduced diversity in humans 223 could result from large reductions in population size and negative selective pressures as the 224 parasite passes from mosquitoes, through the human liver, and into the blood stage.

225
Mosquitoes are the location of parasite sexual recombination and therefore certainly provide a 226 site for genomic diversification, but this seems unsuited to explain the diversity of these short 227 segments in ama1 and csp that do not harbor known recombination hotspots (12). A probable 228 contributor to this high mosquito diversity is multiple or interrupted feeds on infected hosts, 229 which would allow strains to accumulate in the mosquito abdomen. This feeding behavior has   (17). Despite these plausible explanations for constrained diversity in 238 humans and higher diversity in mosquitoes, the mechanism by which mosquitoes maintain such 239 high parasite diversity when their parasite population is necessarily sampled from the less 240 diverse human population remains to be fully elucidated.

242
Within individual infections, we observed higher dominance of haplotypes in human compared 243 to mosquito infections, while on a larger scale, the P. falciparum haplotype population was more 244 evenly distributed among humans than among mosquitoes. These differences may result from 245 the differential selection landscapes between hosts, in particular for the proteins encoded by our 246 gene targets, AMA1 and CSP, which harbor epitopes that are known targets of functional 247 human immunity (18). In humans, the concurrent maintenance in the population of multiple 248 viable alleles due to balancing selection, paired with the removal of deleterious alleles due to 249 negative selection, could produce a relatively high evenness of haplotypes in the human 250 parasite population even as individual infections are shaped by directional selection resulting 251 from individual host immune responses. In contrast, the relative lack of differential fitness in the 252 mosquito host described above may lead to even parasite strain abundances within a mosquito.

254
Comparison of paired abdomens and heads from the same mosquito revealed striking 255 differences between P. falciparum presence and haplotype composition. As expected given the 256 delay between midgut and salivary gland infections, many mosquitoes had haplotypes private to 257 the abdomen that were not present in the head. More surprising was the observation of 258 mosquitoes with haplotypes private to the head that were absent from the abdomen, suggesting 259 that infections do not reliably persist in a mosquito's abdomen throughout its lifespan. While 260 these differences may again be due to cryptic haplotypes, the identification of mosquitoes with 261 infections in the head but not the abdomen using sensitive PCR detection methods (19,20) 262 indicates that cryptic haplotypes likely cannot explain all of the observed differences. Despite 263 these discrepancies between abdomens and heads from a given mosquito, at the population 264 level haplotype composition and diversity were similar between mosquito abdomens and heads,

265
suggesting that the selective pressures for or against certain haplotypes (or lack thereof) may 266 be similar in these two compartments.

268
Our findings highlight the role of the mosquito host in influencing the sequence diversification of

283
This study has limitations. First, the inability to sample parasites from mosquitoes without 284 sacrificing them precludes a comprehensive study of paired mosquito abdomen and head 285 infections over time. Even so, we were still able to identify similarities and differences between 286 the haplotype populations in these two compartments. Additionally, the mosquito and human 287 sampling schemes were different, potentially biasing sampling comprehensiveness between hosts. To mitigate the risk that this potential imbalance influenced our results, we performed 289 comparative population analyses using empirical methods with a fixed coverage threshold (10) 290 and sensitivity analyses. Finally, many of the human and mosquito infections had very low 291 parasite densities, which not only increases the possibility of failing to detect infections, but also 292 increases the possibility of false haplotype discovery (22). To reduce the inclusion of false 293 haplotypes to the greatest extent possible, we performed strict haplotype censoring to remove 294 potential false positives (1) and performed sensitivity analyses on key findings to determine 295 whether haplotype filtering criteria influenced the results.

297
In conclusion, our comparison of P. falciparum haplotypes observed in natural, coincident

313
The study design and sample processing have been described previously (1). Briefly, a

327
We performed parallel analyses of amplicon deep-sequenced segments of the P. falciparum 328 ama1 and csp marker genes. Since ama1 and csp are unlinked markers found on different 329 chromosomes, to some extent these parallel analyses can be considered pseudo-replicates,

330
where similar results for both markers increases confidence in our findings.

334
For each mosquito with a P. falciparum infection in both the abdomen and the head, the Jaccard 335 distance (24) was calculated for the haplotypes in the abdomen-head pair: 336 337 Where Ha is the set of haplotypes in the abdomen and Hh is the set of haplotypes in the head.

357
Diversity and evenness

359
For analyses between humans and mosquitoes, all mosquito abdomen and head samples were

387
We calculated haplotype evenness using the following equation (31): Where H is the haplotype richness, or the number of distinct haplotypes in the population. For q 391 = 0 evenness is defined as 1, and for H = 1, evenness is defined as 0.

394
For each sample, we computed haplotype diversity and evenness using equations 1 and 2. In 395 this case, pi in equation 1 is the relative read abundance of each haplotype, q D is the within-396 host diversity, and H is the MOI of the infection.

398
To compare evenness between human and mosquito hosts, we computed a zero-one inflated 399 Beta regression model using the R package gamlss v5.4-1 (32) with host as the main exposure, 400 evenness as the outcome, log2-transformed haplotype reads as a covariate, and individual as a 401 random effect. To determine whether incorporating information from both markers influenced 402 differences in evenness between hosts, for each sample we selected the highest evenness 403 value (between ama1 and csp) and compared these values between humans and mosquitoes.

404
Finally, to explore if evenness values were biased by the initial enforcement of haplotype 405 quality-filtering criteria that were partially based on within-sample haplotype proportion, we 406 performed a sensitivity analysis using unfiltered haplotypes. These haplotypes were inferred by 407 DADA2 v1.8 (23) from input reads which passed upstream read quality-filtering. Using these 408 unfiltered haplotypes, we used the same methods as above to compute and compare evenness.

412
Comparison across groups was performed using Wilcoxon rank-sum tests, Fisher's exact tests,

413
or Kolmogorov-Smirnov tests. All data analysis and visualization was performed in R v4.

518
Acknowledgements 519 520 We thank the field technicians in Webuye for their engagement with the study participants:        Figure 3B in the main text.