Measuring changes in Plasmodium falciparum census population size in response to sequential malaria control interventions

Reviewer #1 (Public Review):

Tiedje et al. investigated the transient impact of indoor residual spraying (IRS) followed by seasonal malaria chemoprevention (SMC) on the plasmodium falciparum parasite population in a high transmission setting. The parasite population was characterized by sequencing the highly variable DBL$\alpha$ tag as a proxy for var genes, a method known as varcoding. Varcoding presents a unique opportunity due to the extraordinary diversity observed as well as the extremely low overlap of repertoires between parasite strains. The authors also present a new Bayesian approach to estimating individual multiplicity of infection (MOI) from the measured DBL$\alpha$ repertoire, addressing some of the potential shortcomings of the approach that have been previously discussed. The authors also present a new epidemiological endpoint, the so-called "census population size", to evaluate the impact of interventions.

This study provides a nice example of how varcoding technology can be leveraged, as well as the importance of using diverse genetic markers for characterizing populations, especially in the context of high transmission. The data are robust and clearly show the transient impact of IRS in a high transmission setting, however, some aspects of the analysis are confusing.

1. Approaching MOI estimation with a Bayesian framework is a well-received addition to the varcoding methodology that helps to address the uncertainty associated with not knowing the true repertoire size. It's unfortunate that while the authors clearly explored the ability to estimate the population MOI distribution, they opted to use only MAP estimates. Embracing the Bayesian methodology fully would have been interesting, as the posterior distribution of population MOI could have been better explored.

2. The "census population size" endpoint has unclear utility. It is defined as the sum of MOI across measured samples, making it sensitive to the total number of samples collected and genotyped. This means that the values are not comparable outside of this study, and are only roughly comparable between strata in the context of prevalence where we understand that approximately the same number of samples were collected. In contrast, mean MOI would be insensitive to differences in sample size, why was this not explored? It's also unclear in what way this is a "census". While the sample size is certainly large, it is nowhere near a complete enumeration of the parasite population in question, as evidenced by the extremely low level of pairwise type sharing in the observed data.

3. The extraordinary diversity of DBL$\alpha$ presents challenges to analyzing the data. The authors explore the variability in repertoire richness and frequency over the course of the study, noting that richness rapidly declined following IRS and later rebounded, while the frequency of rare types increased, and then later declined back to baseline levels. The authors attribute this to fundamental changes in population structure. While there may have been some changes to the population, the observed differences in richness as well as frequency before and after IRS may also be compatible with simply sampling fewer cases, and thus fewer DBL$\alpha$ sequences. The shift back to frequency and richness that is similar to pre-IRS also coincides with a similar total number of samples collected. The authors explore this to some degree with their survival analysis, demonstrating that a substantial number of rare sequences did not persist between timepoints and that rarer sequences had a higher probability of dropping out. This might also be explained by the extreme stochasticity of the highly diverse DBL$\alpha$, especially for rare sequences that are observed only once, rather than any fundamental shifts in the population structure.

Reviewer #2 (Public Review):

In this manuscript, Tiedje and colleagues longitudinally track changes in parasite numbers across four time points as a way of assessing the effect of malaria control interventions in Ghana. Some of the study results have been reported previously, and in this publication, the authors focus on age-stratification of the results. Malaria prevalence was lower in all age groups after IRS. Follow-up with SMC, however, maintained lower parasite prevalence in the targeted age group but not the population as a whole. Additionally, they observe that diversity measures rebounds more slowly than prevalence measures. Overall, I found these results clear, convincing, and well-presented. They add to a growing literature that demonstrates the relevance of asymptomatic reservoirs.

There is growing interest in developing an expanded toolkit for genomic epidemiology in malaria, and detecting changes in transmission intensity is one major application. As the authors summarize, there is no one-size-fits-all approach, and the Bayesian MOIvar estimate developed here has the potential to complement currently used methods. I find its extension to a calculation of absolute parasite numbers appealing as this could serve as both a conceptually straightforward and biologically meaningful metric. However, I am not fully convinced the current implementation will be applied meaningfully across additional studies.

1. I find the term "census population size" problematic as the groups being analyzed (hosts grouped by age at a single time point) do not delineate distinct parasite populations. Separate parasite lineages are not moving through time within these host bins. Rather, there is a single parasite population that is stochastically divided across hosts at each time point. I find this distinction important for interpreting the results and remaining mindful that the 2,000 samples at each time point comprise a subsample of the true population. Instead of "census population size", I suggest simplifying it to "census count" or "parasite lineage count".

It would be fascinating to use the obtained results to model absolute parasite numbers at the whole population level (taking into account, for instance, the age structure of the population), and I do hope this group takes that on at some point even if it remains outside the scope of this paper. Such work could enable calculations of absolute---rather than relative---fitness and help us further understand parasite distributions across hosts.

2. I'm uncertain how to contextualize the diversity results without taking into account the total number of samples analyzed in each group. Because of this, I would like a further explanation as to why the authors consider absolute parasite count more relevant than the combined MOI distribution itself (which would have sample count as a denominator). It seems to me that the "per host" component is needed to compare across age groups and time points---let alone different studies.

3. Thinking about the applicability of this approach to other studies, I would be interested in a larger treatment of how overlapping DBLa repertoires would impact MOIvar estimates. Is there a definable upper bound above which the method is unreliable? Alternatively, can repertoire overlap be incorporated into the MOI estimator?

Smaller comments:
- Figure 1 provides confidence intervals for the prevalence estimates, but these aren't carried through on the other plots (and Figure 5 has lost CIs for both metrics). The relationship between prevalence and diversity is one of the interesting points in this paper, and it would be helpful to have CIs for both metrics when they are directly compared.

Reviewer #3 (Public Review):

Summary:
The manuscript coins a term "the census population size" which they define from the diversity of malaria parasites observed in the human community. They use it to explore changes in parasite diversity in more than 2000 people in Ghana following different control interventions.

Strengths:
This is a good demonstration of how genetic information can be used to augment routinely recorded epidemiological and entomological data to understand the dynamics of malaria and how it is controlled. The genetic information does add to our understanding, though by how much is currently unclear (in this setting it says the same thing as age-stratified parasite prevalence), and its relevance moving forward will depend on the practicalities and cost of the data collection and analysis. Nevertheless, this is a great dataset with good analysis and a good attempt to understand more about what is going on in the parasite population.

Weaknesses:
Overall the manuscript is well-written and generally comprehensively explained. Some terms could be clarified to help the reader and I had some issues with a section of the methods and some of the more definitive statements given the evidence supporting them.

Author Response

We are delighted that eLife has assessed our study as a valuable contribution as well as appreciating the importance of working on asymptomatic reservoirs of P. falciparum in high transmission where not just children, but adolescents and adults harbor multiclonal infections. The constructive public reviews will serve to improve our manuscript.

Detailed responses to referees’ comments and a revised manuscript are forthcoming. Here we make a provisional response to three key areas addressed by the referees:

(1) census population size

Referee 1 raises important questions although we respectfully disagree on the terminology we have adopted (of “census”) and on the unclear utility of the proposed quantity.

We consider the quantity a census in that it is a total enumeration or count of the infections in a given population sample and over a given time period. In this sense, it gives us a tangible notion of the size of the parasite population, in an ecological sense, distinct from the formal effective population size used in population genetics. Given the low overlap between var repertoires of parasites (as observed in monoclonal infections), the population size we have calculated translates to a diversity of strains or repertoires. But our focus here is in a measure of population size itself. The distinction between population size in terms of infection counts and effective population size from population genetics has been made before for pathogens (see for example Bedford et al. 2011 for the seasonal influenza virus and for the measles virus) and is a clear one in the ecological literature for non-pathogen populations (Palstra et al. 2012).

Both referees 1 and 2 point out that census population size will be sensitive to sample size. We completely agree with the dependence of our quantity on sample size. We used it for comparisons across time of samples of the same depth, to describe the large population size characteristic of high transmission, and persistent across the IRS intervention. Of course, one would like to be able to use this notion across studies that differ in sampling depth.

Here, referee 1 makes an insightful and useful suggestion. It is true that we can use mean MOI, and indeed there is a simple map between our population size and mean MOI (as we just need to divide or multiply by sample size). We can do even more, as with mean MOI we can presumably extrapolate to the full sample size of the host population, or the population size of another sample in another location. What is needed for this purpose is a stable mean MOI relative to sample size. We can show that indeed in our study mean MOI is stable in that way, by subsampling to different depths of our original sample. We will include in the revision discussion of this point and result, which allows an extrapolation of the census population size to the whole population of hosts in the local area. We’ll also clarify the time denominator, as given the typical duration of infections, we expect our population size to be representative of a per-generation measure.

Referee 2 suggests we adopt the term “census count” but as a census in our mind is a count we prefer to use “census”.

Referee 3 considers the genetic data tracking parasite MOI and census changes gives the same result as prevalence which tracks infected hosts. Respectfully, we disagree and will provide an expanded response.

(2) the importance of lineages (in response to referee 2)

We do not think that lineages moving exclusively through a given type of host or “patch” is a requirement for enumerating the size of the total infections in such a subset. It is true that what we have is a single parasite population, but we are enumerating for the season the respective size in host classes (children and adults). This is akin to enumerating subsets of a population in ecological settings.

We are also not clear on the concept of lineage for these highly recombinant parasites as we struggle to find highly related repertoires. In fact, we see the use of the var fingerprinting methodology as a means to capture changes in strain or var repertoires dynamics as a result of changing transmission conditions.

(3) var methodology

Comments and queries were made by all three referees about aspects of var methodology, including the Bayesian approach. These will be addressed in our full response.

Here we respond to a very good point made by referee 2: “Thinking about the applicability of this approach to other studies, I would be interested in a larger treatment of how overlapping DBLa repertoires would impact MOIvar estimates. Is there a definable upper bound above which the method is unreliable? Alternatively, can repertoire overlap be incorporated into the MOI estimator?”

There is no predefined threshold one can present a priori. Intuitively, the approach to estimate MOI would appear to breakdown as overlap moves away from extremely low, and therefore, for locations with lower transmission intensity. Interestingly, we have observed that this is not the case in our paper by Labbé et al. 2023 where we used model simulations in a gradient of three transmission intensities, from high to low. The original varcoding method performed well across the gradient. This may arise from a nonlinear and fast transition from low overlap to high overlap that is accompanied by the MOI transitioning quickly from primarily multiclonal (MOI > 1) to monoclonal (MOI = 1). This issue needs to be investigated further, including ways to extend the estimation to explicitly include the distribution of DBL repertoire overlap.

References: Bedford T, Cobey S, Pascual, M. 2011. Strength and tempo of selection revealed in viral gene genealogies. BMC Evol Biol 11, 220. https://doi.org/10.1186/1471-2148-11-220

Labbé F, He Q, Zhan Q, Tiedje KE, Argyropoulos DC, Tan MH, Ghansah A, Day KP, Pascual M. 2023. Neutral vs . non-neutral genetic footprints of Plasmodium falciparum multiclonal infections. PLoS Comput Biol 19 :e1010816. doi:doi.org/10.1101/2022.06.27.49780

Palstra FP, Fraser DJ. 2012. Effective/census population size ratio estimation: a compendium and appraisal. Ecol Evol. Sep;2(9):2357-65. doi:10.1002/ece3.329.