Predicting the likelihood and intensity of mosquito infection from sex specific Plasmodium falciparum gametocyte density

1Mosquitoes must ingest both male and female mature gametocytes to become infected with malaria. The shape of the relationship between gametocyte density and the probability of mosquito infection is thought to be complex for Plasmodium falciparum; but recent evidence has shown that the most comprehensive characterisation of this relationship was conducted using molecular tools that only quantified female gametocyte specific Pfs25 mRNA. An additional shortcoming of these previous estimates is that female gametocytes were quantified from trendlines with unknown gametocyte sex-ratios, potentially affecting assay precision.

2It is intuitive that gametocyte sex ratio is important in determining transmission efficiency. Gametocyte sex-ratio in P. falciparum is typically female biased but may differ between infections and during infections, and may be influenced by immune responses that differentially affect male and female gametocytes. Fertility assurance explains why under certain conditions, such as a low gametocyte density, gametocyte sex-ratio may become less female biased to ensure all female gametes are fertilised. Despite evidence that gametocyte sex ratio is adjusted in response to malaria developmental bottlenecks, there is no direct evidence for its epidemiological importance. Since mosquito infections may occur from blood with P. falciparum gametocyte densities below the microscopic threshold for detection and gametocyte density itself is an important determinant of sex ratio, sensitive quantification of male and female gametocytes is essential for assessments of the role of gametocyte sex ratio in natural infections and for reliable estimates of total gametocyte density. Here, we use a new molecular target of male P. falciparum gametocytes (Pf3D7_1469900 or PfMGET) to explore the association between the density of female and male gametocytes and mosquito infection prevalence. The relationship between gametocyte density and oocyst density is also investigated as this may show a different relationship with gametocyte density and recent work has highlighted the potential importance of mosquito parasite density in mosquito-to-human transmission. Both the number of infected mosquitoes and their parasite load may thus need to be considered when assessing the human reservoir of infection.

3Gametocyte carriers were included from Mali (n = 71), Burkina Faso (n = 64) and Cameroon (n = 13). P. falciparum parasite prevalence by microscopy ranged from ~40–70% between sites; sampling occurred in different seasons (Table 1). All sites recruited gametocyte carriers prior to antimalarial treatment but used different enrolment criteria. Microscopically detected gametocyte carriers were included in three studies while one site (Balonghin) included submicroscopic gametocyte densities. Total gametocyte densities ranged from 0.04 to 1164 gametocytes/μL; the median proportion of male gametocytes was 25% (interquartile range 13–39%)(Table 1, Figure 1). Gametocyte density estimates by microscopy and qRT-PCR were positively correlated in all sites, but there was variation in the strength of correlation: Ouelessebougou (r = 0.74), Bobo Dioulasso (r = 0.27), Balonghin (r = 0.56) and Yaoundé (r = 0.91) (Figure 1—figure supplement 1). Across 148 successful membrane feeding experiments, 16.7% (1297/7757) of mosquitoes became infected, with considerable variation between gametocyte donors and study sites.

4Female gametocyte densities quantified by Pfs25 quantitative reverse-transcriptase PCR (qRT-PCR) and male gametocyte densities by PfMGET qRT-PCR were positively correlated (Figure 1A; Spearman’s rank correlation coefficient = 0.79, p<0.001). The fact that the relationship is linear on the log scale means that the percentage of gametocytes that were male decreased with increasing total gametocyte densities (Figure 1B, p<0.001 [Kruskal Wallis test]). This association remained apparent and statistically significant if any of the study sites with a smaller number of observations was removed (i.e. with removal of data from Bobo Dioulasso, Balonghin or Yaoundé). Conversely there was no evidence for an association between proportion of male gametocytes with asexual parasite density (Figure 1C, p=0.713).

5The proportion of mosquitoes developing oocysts is best described by a model that incorporates both the density of female and the density of male gametocytes (deviance information criterion, DIC = 451.5). In the best fit model, female gametocyte density explains most of the variability with the proportion of mosquitoes infected increasing rapidly with increasing gametocytaemia before saturating at high female gametocyte densities (Figure 2A). At female gametocyte densities of 200 per µl approximately 30% of mosquitoes are infected. At low gametocyte densities transmission appears to be impeded by a lack of male parasites (Figure 2B). The number of mosquitoes infected is on average lower for hosts with fewer than 50 male gametocytes per µl of blood, with the model predicting that male densities < 10 per µl reduces the proportion of infected mosquitoes by 50% (Figure 2C). Predictions for the proportion of mosquitoes infected according to the density of female and male gametocytes in the blood is given in Figure 2D. Poorer statistical fits were observed for models where the proportion of mosquitoes with oocysts was described by either the density of females alone (DIC = 481.3), the density of males (DIC = 538.8), or total gametocyte density (sum of male and female gametocytes, DIC = 501.7). There were considerable differences in the relative infectivity between sites. Compared to the Mali data, infectivity was similar in the study in Balonghin, Burkina Faso (0.95 times as high; 95% CI 0.1–1.4) but 3.65 times higher in Bobo Dioulasso, Burkina Faso (95% CI 2.3–5.2) and 1.68 times higher in site in Yaoundé, Cameroon (95% CI 1.1–2.6) (Figure 2—figure supplement 1, Figure 2—source data 2.). Other models, including both qRT-PCR and microscopy, showed poorer performance in predicting mosquito infection prevalence (Figure 2—source data 2).

6The distribution of oocysts between mosquitoes is highly over-dispersed, with some mosquitoes harbouring very high oocyst densities. This aggregated distribution is reflected in the relationship between proportion of mosquitoes infected and mean oocyst density (Figure 3A). Mean oocyst density increased with increasing gametocyte density and continued to increase across the range of gametocyte densities observed, without evidence for a plateau being reached (Figure 3B). Total gametocyte density (sum of females and males) was able to predict mean oocyst density better than female gametocyte density alone. The higher complexity of the density model (requiring the fitting of the negative binomial distribution) meant that it was not possible to fit a model analogous to the best fit model on infectivity (i.e. a function dependent on female gametocyte density but with reduced transmission at low male densities). Care should be taken interpreting the relative importance of females and male gametocytes in predicting oocyst density as this result should be validated with a larger dataset.

7We present an improved model to predict mosquito infection from female and male gametocyte density estimates. Previous molecular assessments of P. falciparum gametocyte density quantified female specific Pfs25 mRNA then converted this into a measure of gametocytes per µL of blood by reference to standard curve of mixed-sex gametocytes. These assessments therefore quantified neither female nor total gametocyte density accurately. The current model presents a considerable improvement over previous work, both by separately quantifying male and female gametocytes using sex-specific mRNA markers and standard curves and also by using considerably more accurate estimates of total gametocyte density. Male and female gametocytes were quantified separately using automated extraction of nucleic acids. Manual extraction can result in considerable variation that may have affected the accuracy of our earlier gametocyte estimates. In addition, we used qRT-PCR that has higher precision than the previously used quantitative nucleic acid sequence based amplification, QT-NASBA.

8The analyses show that the relationship between the density of gametocytes and transmission can be adequately described by a simple saturating relationship, and that a more complicated curve with two inflection points (which generates a double plateau) does not improve model fit. The change in shape of the best fit relationship has implications for predicting the effect of transmission blocking interventions such as gametocytocidal drugs. The first plateau of the previous relationship predicted similar infectivity for gametocyte densities between 1 and 200 gametocytes per µl. If this was the case then an intervention that killed 99% of gametocytes would have very little effect on infectivity on a person with a density of 200 gametocytes per µl but impacted transmission considerably below this level. The new relationship presented here suggests an intervention reducing gametocyte density would have an impact on transmission to mosquitoes across the entire range of gametocyte densities observed in the field with, for example, a predicted reduction in mosquito infection prevalence of 32% to 1% when female gametocyte density is reduced from 200 to 1 female gametocytes per µl. The new work indicates that even gametocytocidal drugs with incomplete efficacy would reduce transmission.

9The increased simplicity found here compared to the previously described relationship is likely to be driven by the improved accuracy of gametocyte density estimates as models fit solely to female parasites density had a similar simplified shape. The concurrent quantification of female and male gametocytes leads to an improved estimation of total gametocyte biomass and could explain why models with information on both sexes were more accurate. Similarly, combining data from microscopy and PCR could also improve overall precision when both techniques have relatively high measurement error. Though improved accuracy of gametocyte density estimates through combining imprecise measures cannot be discounted the model fit to total gametocyte density (male plus female) gave a poorer fit than the model where transmission was dependent on female density unless there were very low male parasite densities. This suggests that the number of male gametocytes may become a limiting factor for transmission at low gametocyte densities, although future studies that directly relate infectivity to mosquitoes to female and male gametocyte densities are needed to confirm this association. Such studies may benefit from an assay that targets a transcript with identical expression levels in male and female gametocytes or a multiplex male-female gametocyte assay that avoids uncertainties in quantifying total gametocyte biomass from two separate assays.

10The work confirmed previous evidence that the proportion of male gametocytes is negatively associated with total gametocyte density, which may reflect a strategic investment in male gametocytes to maximize the likelihood of transmission in low-density infections (fertility assurance). The models also indicated that male gametocyte density may become a limiting factor for transmission success at low gametocyte densities. These observations are in line with in vitro findings with P. falciparum, Plasmodium berghei and Plasmodium chabaudi where infections with a higher proportion of male gametocytes gave higher transmission success at low gametocyte densities but reduced success at higher gametocyte densities.

11The presented model considerably improved the prediction of mosquito infection rates compared to our previous manuscript. Nevertheless, some level of uncertainty remains, particularly between study sites with several experiments resulting in considerably lower mosquito infection rates than predicted based on gametocyte density and sex-ratio. There are several plausible reasons for this. Naturally acquired antibody responses to gametocyte antigens may reduce transmission efficiency and form a first explanation why many mosquitoes may fail to become infected when feeding on some hosts with high-density gametocyte infections. Whilst immune responses that completely prevent mosquito infections are only sporadically detected in naturally exposed populations, it is plausible that functional transmission reducing immunity has reduced mosquito infection rates for a proportion of gametocyte carriers in our study. Temporal fluctuations in transmission reducing immunity may also have contributed to the apparent differences in infectivity between study sites that recruited gametocyte carriers at different time-points in the season. The site with the highest infectivity in the current study (Bobo Dioulasso), recruited individuals in the dry season when the impact of transmission reducing immunity may be lowest. The extent of between site heterogeneity in transmission should be interpreted with caution since two of the four sites had <20 study participants. At all sites malaria-infected individuals were recruited prior to malaria treatment, making it unlikely that drug-induced sterilization of circulating gametocytes may have affected our analyses. Variation in the susceptibility of the mosquito colonies to parasite genotypes may provide a second plausible reason for the imperfect model fit. A third hypothesis could be that our quantification of gametocytes by Pfs25 and PfMGET qRT-PCR may not fully reflect gametocyte maturity and infectivity as gametocytes may be detectable in the blood stream before and after peak infectivity is reached. Lastly, technical issues related to RNA degradation may affect gametocyte quantification and thus yield unreliable low gametocyte density estimates. There were no apparent issues in maintaining the essential temperature control in the field, during transportation, or following RNA extraction for any of the samples included in this study, nor did the associations between microscopy and qRT-PCR gametocyte density estimates indicate (site-specific) RNA degradation that may have resulted in underestimations of true gametocyte densities.

12The study shows that though the prevalence of mosquitoes with oocysts plateaus at high gametocyte densities the average number of oocysts in those mosquitoes continues to rise, as previously reported for P. vivax and P. falciparum. There were insufficient data to fit a model where oocyst density was dependent on female density but with a penalty for low male densities which was the best fit model for infectivity. Whether such a model would fit the oocyst density data better that the model presented here is a question for future research.

13Understanding the association between gametocyte density and mosquito infection rates is of immediate relevance for malaria control efforts. Here we show that accurate measures of female and male gametocyte density can better predict human-to-mosquito infection, and could be used to assess the infectiousness of human populations.

14Field samples were collected at four malaria endemic sites. Samples were collected prior to treatment and after written informed consent was obtained from participants or their guardian(s). Ethical clearance was provided by the National Ethics Committee of Cameroon; Ethical Review Committee of the Ministry of Health, Burkina Faso; Ethics Committee of the Malaria Research and Training Centre, Bamako; Ethics review committee Centre MURAZ; University of California, San Francisco, and London School of Hygiene and Tropical Medicine. Procedures for Ouelessebougou, Mali are described elsewhere. From the trial in Ouelessebougou, baseline samples from microscopically detected gametocyte carriers were used. Additional samples were collected from asymptomatic microscopically detected gametocyte carriers aged 5–15 years in Bobo Dioulasso in Burkina Faso and Yaoundé, Cameroon. Lastly, a random selection of samples from a xenodiagnostic study in Balonghin in Burkina Faso was used. Samples were eligible for selection if gametocytes were detected by Pfs25 QT-NASBA that has an estimated lower limit of detection between 0.02–0.1 gametocytes/µL and thus provided data-points at the lower range of gametocyte densities. The same membrane feeding protocol was used at all sites: local Anopheles coluzzii colony mosquitoes (Mali, Cameroon and Bobo Dioulasso, Burkina Faso) or colony mosquitoes comprising a mixture of A. coluzzii, A. gambiae s.s. and hybrid forms (Balonghin, Burkina Faso) were allowed to feed for 15–20 min on heparin blood samples until dissection in 1% mercurochrome at day seven post-feeding and oocyst detection by two independent microscopists. For all sites membrane feeding and sample collection were performed prior to antimalarial treatment

15Female gametocytes were quantified by quantitative reverse transcriptase PCR (qRT-PCR) targeting female Pfs25 mRNA, as described elsewhere in detail based on established protocols. For male gametocytes we used a recently developed qRT-PCR based on PfMGET (male gametocyte enriched transcript, Pf3D7_1469900), a transcript that is highly enriched in male P. falciparum gametocytes. Primer sequences are provided in Table 2. For all qRT-PCR, mRNA was extracted from blood collected in EDTA tubes by venipuncture; 100 μL of whole blood was stored at −80°C in 500 μL RNAprotect (Qiagen; for Burkina Faso and Cameroon samples) or 900 μL L6 buffer (Severn Biotech, Kidderminster, UK; for Mali samples) until automated extraction using a MagNAPure LC (Total Nucleic Acid Isolation Kit–High Performance; Roche Applied Science, Indianapolis, IN, USA). cDNA was synthesised directly from nucleic acids for the PfMGET assay, for which the primers are intron-spanning, and after DNase treatment (RQ1 DNase I Digest Kit, Promega) for the Pfs25 assay, using High Capacity cDNA Reverse Transcription Kits (Applied Biosystems, Foster City, CA). qRT-PCR results were converted to male and female gametocyte densities using standard curves (ten-fold serial dilutions from 106 to 10 gametocytes/ml) of separate male and female gametocyte populations that were generated using a transgenic parasite line expressing a male specific fluorescence marker. The purity of male and female trendlines was previously confirmed by staining of sorted gametocyte populations using female gametocyte specific anti-Pfg377 antibodies. For both Pfs25 and PfMGET qRT-PCR, a threshold for positivity was set at one gametocyte per sample (0.01/μL).

16The statistical methods used here are the same at those in the original paper, which we briefly recapitulate here. qRT-PCR results are in the form of cycle-thresholds (CT, which is the number of cycles it takes for the fluorescence associated with target amplification to exceed a defined threshold. The relationship between CT and gametocyte density is estimated by fitting a linear regression to CT estimates generated using a sample with known gametocyte density (a 10-fold dilution series). Let the observed CT be denoted by y then,where β0 and β1 are regression coefficients estimates, x is the (known) parasite density from the dilution series and ε represents a normally distributed random error (ε ~N(0,s2)). Equation (1) can be rearranged to enable us to estimate gametocyte density from a CT measurement. We use a Bayesian hierarchical model to estimate the coefficients β0 and β1.

17These gametocyte density estimates are used to determine the relationship between gametocyte density and the proportion of mosquitoes developing oocysts. Four functional forms (linear, power, hyperbolic and Gompertz) were each fit four times: on female gametocyte density alone; on male gametocyte density alone; on total gametocyte density; and finally female gametocyte density but multiplied by a function accounting for reduced transmission at low male densities. Terms allowing for different infectivity at each site were incorporated into the models. The algebraic forms of these models are given in Figure 2—source data 2. The model quantifying the uncertainty in gametocyte density estimates was fit at the same time as the regression determining the relationship between gametocyte density and infectivity using Bayesian Markov Chain Monte Carlo methods assuming a Binomial error structure for each feeding experiment. Fitting the models simultaneously enabled the uncertainty in the gametocyte density estimates to be reflected in the uncertainty of the shape of the relationship. The models were compared using the DIC with a lower value indicating the most parsimonious fit.

18The relationship between gametocyte density and oocyst density was examined in an analogous way, with two exceptions: (1) A negative binomial error structure was used to describe oocyst counts and (2) The increased complexity of the model precluded the inclusion of the function accounting for reduced transmission at low male densities, so each functional form was only fit twice; on female gametocyte density and total gametocyte density.