Molecular xenomonitoring as a complement to human antigen seroprevalence for lymphatic lariasis surveillance in Samoa

Brady McPherson Australian Defence Force Malaria and Infectious Disease Institute Helen J. May eld (  h.may eld@uq.edu.au ) University of Queensland Angus McLure Australian National University Katherine Gass The Task Force for Global Health Take Naseri Samoa Ministry of Health Robert Thomsen Samoa Ministry of Health Steven A. Williams Smith College Nils Pilotte Quinnipiac University Therese Kearns Menzies School of Research Health Patricia M. Graves James Cook University Colleen L. Lau University of Queensland


Introduction
Lymphatic lariasis (LF) is a globally signi cant parasitic disease which can result in severe morbidity and disability, such as lymphedema, elephantiasis or scrotal hydroceles. Elimination programs for vector-borne diseases such as LF, malaria and onchocerciasis represent signi cant global heath investments. Since 2000, the World Health Organization's (WHO) Global Programme to Eliminate LF has facilitated mass drug administration (MDA) of anti-larial drugs to >910 million people in 68 countries. Despite huge successes over >20 years, LF remains endemic in many countries. Most infections are asymptomatic but may still contribute to transmission. Identifying residual infections is therefore critical for programmatic decision-making such as stopping or resuming MDA.
The standard diagnostic test used in LF programmatic surveys detects circulating larial antigen (Ag), produced by adult worms 1 . Although Ag can be detected quickly and inexpensively in the eld using rapid diagnostic tests such as the Alere Filariasis Test Strip 1 , they cannot distinguish between current and recently cleared infections.
After treatment with anti-larial medications, Ag persist for months or years, and therefore may not be su ciently sensitive for determining whether transmission has been interrupted 2,3 . As countries work towards elimination and LF Ag prevalence drops to very low levels, highly sensitive surveillance methods will be required to detect changes in prevalence and monitor progress. An alternative indicator of transmission is the presence of micro laria (Mf, immature parasites) in the blood, which con rms active infection. However, Mf may not be detected even if adult worms are present, e.g., if the worms are too young, too old, or have not mated. Molecular Xenomonitoring (MX) involves testing mosquitoes for larial DNA using the polymerase chain reaction (PCR). Unlike Ag tests, the mosquitoes' short life span means that a positive PCR indicates infectious people were recently nearby 4 . MX may therefore provide a more sensitive signal of transmission than Ag 5 . MX has been used in research settings in American Samoa 6 , Sri Lanka 3 and India 7 , although signi cant knowledge gaps still exist regarding survey design and interpretation of results. Recent studies have highlighted the need for more e cient mosquito collection methods, better understanding of the relationship between results from MX and human surveys 8 , and whether speciation of mosquitoes improves the usefulness of MX results (especially in countries where entomology expertise is limited) 6,9 . LF is endemic in Samoa, transmitted by Aedes mosquitoes including the highly e cient Aedes polynesiensis, with a ight range of ~150m 10,11 . Samoa has conducted LF elimination programs under GPELF since 1998 12 , but programmatic surveys in 2013 and 2017 12 and research projects in 2018 13 showed ongoing transmission. In 2018, Samoa was the rst country to distribute nation-wide triple-drug MDA (ivermectin, albendazole and diethylcarbamazine) 14 . Evaluating the effectiveness of this intervention is therefore of interest globally, but a key challenge is the ability to detect reductions in infection prevalence post-MDA. Considering that Ag could persist for months or years after treatment 15 , MX might provide a more sensitive surveillance strategy in the post-MDA setting.
In this study, we aimed to evaluate the effectiveness of MX for LF surveillance by comparing the results of spatially-representative MX surveys with human Ag prevalence surveys conducted pre-and post-MDA (2018 and 2019) in Samoa. The main objectives were to: a) estimate the prevalence of LF infection in mosquitoes pre-and post-MDA; b) assess the usefulness of MX as an indicator of the presence of Ag-positive humans at primary sampling unit (PSU); and c) compare the sensitivity of MX and human Ag for detecting changes in infection prevalence for post-MDA surveillance. Results were considered in a practical context based on what can realistically be achieved on the ground and how ndings can be used to assist programmatic decision-making.

Human Ag and Mf infection
In 2018, we recruited 1542 participants aged 5-9 years and 1551 participants aged ≥10 years from 30 randomly selected PSUs. In 2019, we recruited 1811 participants aged 5-9 years and 1815 participants aged ≥10 years from the same 30 PSUs. The numbers of participants in 2018 and 2019 in each of the four regions are given in Supplementary Table S7 and Supplementary Table S8 respectively. Nationally, age-adjusted Ag prevalence in the 6-9 months post-MDA was similar in 2018 and 2019 (Fig. 5). Results at the regional level were mixed, with two regions (NWU and SAV) showing an increase from 2018 to 2019 in both age groups. Only one region (AUA) showed a decrease in Ag prevalence in 2019 in both age groups.
In 2018, 13 Mf-positive participants were identi ed in the 30 randomly selected PSUs, and 10 out of these 30 PSUs (

Discussion
This study provided evidence that MX can be useful for LF surveillance, and potentially more sensitive than human Ag prevalence for detecting changes in infection levels immediately following triple-drug MDA. After one round of triple-drug MDA in Samoa, MX detected a decline in LF prevalence in mosquitoes, while there was no signi cant change in human Ag prevalence. We compared MX results for the main vector Ae. polynesiensis and 'all species' combined and found that spatial and temporal trends in mosquito infection prevalence were similar.
Higher numbers of female mosquitoes were caught in 2019 than 2018, partly due to drier conditions in 2018 and additional PSUs and households surveyed in 2019. In 2019, survey teams were more proactive with trap placements to increase catches (i.e. if few mosquitoes were trapped in the rst 24 hours, the trap was moved to another part of the house for the next 24 hours). The increased catch numbers in 2019 resulted in more pools, leading to higher workload and laboratory costs but provided more precise infection prevalence estimates. Mosquito species trapped in Samoa were as expected, with Cx. quinquefasciatus predominating, followed by Ae. The pool PCR-positivity rate and mosquito infection prevalence were three times greater in the purposively selected than the randomly selected PSUs. This was expected since the purposively selected PSUs had historically high LF prevalence but highlights the importance of local knowledge for identifying areas of high transmission. Mosquito infection prevalence in randomly selected PSUs was lower in 2019 than 2018 nationally and in each region ( Fig. 3 and Fig. 8), but this difference was not statistically signi cant for AUA. Even though our study was designed to detect differences at national and regional levels, we also observed statistically signi cant reductions in mosquito infection prevalence between 2018 and 2019 in 10 PSUs, and prevalence did not increase in any PSUs (Supplementary Table S6).
Unsurprisingly, using 'all species' identi ed more positive PSUs than using only Aedes spp. or Ae. polynesiensis. Comparison between mosquito categories showed higher infection prevalence in Aedes spp. than Culex spp. Nevertheless, analyses using 'all species' or only Ae. polynesiensis had concordant spatial and temporal trends, though the larger sample sizes with 'all species' resulted in tighter con dence intervals, illustrating one advantage of using all mosquito samples available. However, sorting mosquitoes by species requires substantial expertise. MX surveys could be greatly simpli ed if sorting was restricted to the genus level. Promisingly, despite the differences in infection prevalence between Culex and Aedes mosquitoes and between Ae. polynesiensis and other Aedes mosquitoes, adjustment for genus or species had little effect on the point estimate or con dence interval for the difference in prevalence between 2018 and 2019 at national, regional or PSU levels (Supplementary Figure S1 and Supplementary Figure S2 ). This nding suggests that, for the purposes of detecting temporal trends, sorting by genus or species in our study made little difference to the results or their interpretation. Whether this observation can be generalised to future surveys in Samoa or elsewhere is uncertain.
If the relative abundance of species captured were substantially different between two time-points or two survey locations to be compared, failing to sort by species may bias comparisons.
The imperfect sensitivity of each surveillance approach suggests that using MX and Ag surveillance together may be bene cial if the aim is to identify (and respond to) as many focal areas of transmission as possible.
Combining these surveillance methods may also assist in understanding transmission as MX may be a better indicator of recent transmission than Ag (assuming PCR testing can be conducted in a reasonable timeframe).
Because evidence of infection (human and/or mosquito) was observed in nearly all PSUs, the small number of negative PSUs (for both human and mosquito infection) made it di cult to assess the predictive accuracy of PCR-positive mosquitoes for the presence of Ag-positive people.
When comparing MX and Ag for post-MDA surveillance, we found a statistically signi cant decrease in infection prevalence in mosquitoes at the country level for 'all species' and Ae. polynesiensis, but no signi cant change overall in Ag prevalence in 5-9 year-olds or in those aged ≥10 years. This could be due to a slower change in Ag compared to mosquito infection. These results demonstrate each surveillance method's sensitivity to changing prevalence, and when used together may provide a more complete picture. MX may provide a more accurate picture of the transmission at speci c time points, whereas Ag, due to its more gradual decline, provides a smoother trend over time. Our results should be considered in light of the study's limitations. In 2018, the sample size of households required to trap su cient mosquitoes was calculated using the best available evidence, but dry weather conditions led to lower-than-expected catches. Operational challenges in 2018 led to simpli ed pooling by reducing the number of mosquito categories. In 2019, protocol changes and wetter conditions led to increased catches. Mosquitoes were pooled between households in 2018, so prevalence estimates could not be adjusted for household-level clustering. Despite these limitations, we were able to identify a statistically signi cant decline in infection prevalence in mosquitoes, which was not observed using human Ag prevalence.
In conclusion, our study suggests that in the immediate post-MDA period, MX might be better for detecting declining prevalence than Ag surveillance in adults or children. However, if the goal is to detect the presence of residual transmission, Ag and MX surveys may provide complementary information. In MX, using both the primary mosquito species and other species increased the sample size and improved the ability to detect residual transmission and changes in prevalence. In our study, adjusting for species made very little difference to estimates of temporal trends, suggesting that the labour-intensive process of sorting mosquitoes into species categories did not in uence the overall results or their interpretation. In countries where the expertise required to trap and identify mosquitoes is limited, omitting speciation could make MX more feasible. Further research is required to determine whether our ndings are generalisable to future surveys in Samoa and to other settings.

Study region
Samoa is a tropical island nation in the South Paci c with approximately 201,316 residents in 2018, 16 the majority of whom live on the islands of Upolu and Savai'I ( Fig. 9). Villages are predominantly rural, with urbanised areas around the capital of Apia and the Savai'i ferry port of Salelolonga. Average rainfall is 3000-6000 mm/year 17 and inland areas remain largely forested. The predominant LF vector species is the day-biting Ae. polynesiensis, with evidence that night-biting Ae. samoanus (included in Ae. (Finlaya) subgenus) also able to transmit W. bancrofti 18 .

Site selection
In 2018, 30 villages were randomly selected from a line list of 338 villages in the 2016 national census. For eight selected villages with population of <600, the next village on the list was added to the PSU. Five additional villages were purposively selected by the Samoa Ministry of Health due to high Ag prevalence in previous surveys 12,19 . Within each PSU, 15 households were selected using a 'virtual walk' method as described previously 13 . If a selected location was not a household, or was uninhabited (destroyed, abandoned, unoccupied), it was replaced by the closest house. If a PSU consisted of two villages, the number of house locations per village was proportionate to each village's population. Spatial data on country, island, region and village boundaries in Samoa were obtained from the Paci c Data Hub (paci cdata.org) and DIVA-GIS (divagis.org). Geographic information systems software ArcMap (v10.6, Environmental Systems Research Institute, Redlands, CA) was used to manage spatial data and produce maps.  Figure S4 for survey timeline). Survey design has been previously described. 13 Data analysis Mosquito abundance, distribution, and prevalence We calculated mosquito abundance for each species and PSU as the number of female mosquitoes tested from all traps over the two-day trapping period in each year. We reported abundance by region, PSU and species.
Mosquitoes not classi ed as Culex or Aedes were excluded.
Prevalence of mosquitoes infected with W. bancrofti was estimated from pool tested results using the R package PoolTestR 26 . When estimating prevalence for a single PSU for a single species, genus, or without any adjustment for mosquito species, the function PoolPrev was used to calculate the maximum likelihood prevalence. For other estimates, the function PoolPrevBayes was used with default uninformative priors to t Bayesian, mixed effect, multivariable logistic regression models modi ed for pooled data with variable pool sizes.
To compare prevalence between randomly and purposively selected PSUs in 2019, we tted a multivariable model with xed/population effects for region, species and selection method (random/purposive) and random/group effects for PSU. The OR and 95% CrI intervals for purposively vs randomly selected villages was used to determine differences and statistical signi cance. To examine the sensitivity of MX for detecting changes in infection rates from MDA, we compared the estimated prevalence in 2018 and 2019. For models comparing prevalence between years, we only included data from the 28 PSUs sampled in both years and collapsed species categories to those used in 2018.
To examine changes over time at the national or PSU level, we tted multivariable models with xed/population effects for vector species and, random/group effects for each PSU at baseline (2018), and random/group effects for the temporal change in each PSU. Region was not included as a covariate as approximate leave-one-out cross validation 27 indicated including it did not improve the model. To examine changes over time by region, we modi ed the model to include xed/population effects for each region in 2018 and xed/population effects for the temporal change. From these models we calculated ORs (which approximate prevalence ratios as prevalence was low) and 95% CrI to quantify differences between years at national, regional and PSU levels. As a sensitivity analysis, we tted alternative models adjusting for genus rather than species, or without adjustment for either. The presence/absence results for PSUs for Ae. polynesiensis and 'all species' were mapped.
Human Ag and Mf prevalence Human Ag prevalence for each year was estimated at national and regional levels, and in the 30 randomly selected PSUs. Statistical analyses were performed using Stata (version 16, Stata Corp, College Station). Estimates were calculated using the proportion command for each age category (5-9 and ≥10 years), adjusting for selection probability at PSU and individual levels, and for clustering at PSU level. Results were standardized by gender and (for those aged ≥10 years) by age group using 5-year categories in the 2016 census. Finite population correction (FPC) was applied since sampling was without replacement. For human Ag, ORs were estimated by region and overall for each age group (5-9 and ≥10 years) using melogit, adjusting for selection probability at PSU and individual levels, and for clustering at PSU. Age, gender and region were included as xed effects. After strati cation by region, OR for the 5-9 year-old group could not be estimated in two regions (AUA and ROU) because there were no Ag-positives in 2019.
Comparing MX and human Ag as surveillance methods We compared MX and human Ag as surveillance methods in terms of presence/absence of positives, and prevalence difference between years. We measured correlation between Ag and MX results using a one-sided Fisher's exact test implemented with sher.test in R 28 . We assessed sensitivity and usefulness of MX for predicting the presence of Ag-positive humans in a PSU by using Ag presence/absence as the 'reference' state. Analyses were conducted for different mosquito categories to determine if the time, effort and expertise required to speciate mosquitoes improved predictive accuracy. We used the same presence/absence data to illustrate the concordance between the two surveillance methods (Supplementary 6), and reported concordance as percentage of PSUs in which the two methods provided the same result (both positive/both negative). Results were mapped to show the ndings for each surveillance method by PSU and year.
To compare each surveillance method pre-and post-MDA, we used prevalence estimates by year in 28 PSUs that were surveyed in both years. We calculated ORs (2019 vs 2018) to determine signi cant changes in prevalence post-MDA. We did not conduct analyses for associations between the presence of PCR-positive pools and Mfpositive persons because Ag is the standard indicator currently used for LF surveillance. Also, the 2018 human survey was conducted immediately post-MDA, which would have affected Mf results.

Data Availability
All relevant data are within the paper. We are unable to provide individual-level antigen prevalence data and demographic data because of the potential for breaching participant con dentiality. The communities in Samoa are very small, and individual-level data such as age, sex, and village of residence could potentially be used to identify speci c persons. For researchers who meet the criteria for access to con dential data, the data are available on request from the Human Ethics O cer at the Australian National University Human Research Ethics Committee, email: human.ethics.o cer@anu.edu.au. Figure 8 Change in prevalence between 2018 and 2019, expressed as an odds ratio (OR), for mosquito infection prevalence (MX for all species and Ae. polynesiensis), and human Ag prevalence (in those aged ³10 years, and 5-9 years). Given the low prevalence, the ORs are approximately equal to prevalence ratios. ORs <1 indicate decrease in infection prevalence in 2019 compared to 2018, ORs >1 indicate an increase, and OR of 1 indicate no change. OR could not be calculated for 5-9 year-olds in AUA and ROU because no antigen-positive cases were detected in these groups in 2019.