High and Heterogeneous Prevalence of Asymptomatic and Sub-microscopic Malaria Infections on Islands in Lake Victoria, Kenya

Kenya is intensifying its national efforts in malaria control to achieve malaria elimination. Detailed characterization of malaria infection among populations living in the areas where the disease is endemic in Kenya is a crucial priority, especially for planning and evaluating future malaria elimination strategy. This study aimed to investigate the distribution and extent of malaria infection on islands in Lake Victoria of Kenya to aid in designing new interventions for malaria elimination. Five cross-sectional surveys were conducted between January 2012 and August 2014 on four islands (Mfangano, Takawiri, Kibuogi and Ngodhe) in Lake Victoria and a coastal mainland (Ungoye). Malaria prevalence varied significantly among settings: highest in Ungoye, followed by the large island of Mfangano and lowest in the three remaining small islands. Of the 3867 malaria infections detected by PCR, 91.8% were asymptomatic, 50.3% were sub-microscopic, of which 94% were also asymptomatic. We observed geographical differences and age dependency in both proportion of sub-microscopic infections and asymptomatic parasite carriage. Our findings highlighted the local heterogeneity in malaria prevalence on islands and a coastal area in Lake Victoria, and provided support for the inclusion of mass drug administration as a component of the intervention package to eliminate malaria on islands.

Prevalence of malaria infection. The prevalence of malaria microscopy and PCR in the five surveys is shown in Table 1. Across surveys, parasite rates were 14.6-24.1% by microscopy and 30.1-44.2% by PCR. The overall prevalence was significantly higher (P < 0.05) in male than in female participants by both methods of detection. Parasite rates differed significantly (P < 0.001) by age groups: highest in the 6-10 group by microscopy and 11-15 group by PCR, and lowest in the > 30 group by both methods. Malaria prevalence also differed significantly (P < 0.001) among the five studied areas: highest in Ungoye, followed by Mfangano, and lowest on small islands of Takawiri, Kibuogi, and Ngodhe. Geographic heterogeneity in malaria prevalence was consistent among the five surveys. Figure 1 shows detailed results of age-specific prevalence of malaria infection and parasite count among participants in five different geographical areas, and enlarged spleen in five surveys. The peaks in malaria prevalence among the 6-10 and 11-15 groups were most apparent in Ungoye and to a lesser extent Mfangano, but were observed only occasionally on the small islands. Parasite densities were highest among children aged ≤ 5 years (geometric mean 2023 parasite/μ l; 95% CI: 1736.3-2357.3) and decreased significantly (P < 0.001) in older age groups (Supplementary Table S2). Nonetheless, overall parasite densities were similar among geographical areas (P = 0.589) (Supplementary Table S2). Similar to parasite rates, spleen rates among children were consistently highest in Ungoye. Of the 2835 malaria-infected children detected by PCR, 64.3% (95% CI: 62.5-66.1) had an enlarged spleen compared to 19.5% (95% CI: 18.1-20.8) uninfected children (P < 0.001). Furthermore, of the 3865 malaria-infected individuals detected by PCR, 24.8% (95% CI: 23.4-26.2) had anaemia compared to 20.3% (95% CI: 19.4-21.3) of uninfected individuals (P < 0.001).
Univariable and multivariable associations with malaria infection. As the more sensitive diagnostic method, PCR results were used as the primary outcome of Plasmodium infection (with any species) for the risk factor analysis (Table 4). Using the full statistical model, the univariate analyses showed that including season, gender, age group, geographical area, fever case, enlarged spleen and anaemia case significantly improved the fit of the model (all P < 0.001). After adjusting for all other factors in the multivariate model, being male, having fever, enlarged spleen and anaemia were all associated with excess odds of infection. Higher odds of infection were observed in all age groups when compared to adults above 30 years, with the highest odd in the 11-15 age group (adjusted odd ratio, AOR = 3.38, P < 0.001). Moreover, living in the coastal mainland (Ungoye) and large island (Mfangano Island) were found to be significantly associated with malaria infection. The odds of malaria    infection were almost nine times higher (AOR = 8.55, P < 0.001) for residents of Ungoye and almost four times higher (AOR = 3.46, P < 0.001) for residents of Mfangano when compared to those of Ngodhe. Interestingly, the odds of malaria infections were not significantly different among small islands.

Discussion
The present study is the first large-scale description of malaria prevalence on islands in Lake Victoria in Kenya based on both microscopy and PCR methods. Our findings indicate significant local variation in malaria prevalence in the study area, a small region of altitudinal and vegetational homogeneity with stable perennial transmission. The relative abundance of vector species and suitable larval habitats likely contribute to this variation in malaria prevalence between neighbouring islands and mainland in Lake Victoria. Entomological studies have revealed that the dominant malaria vector species in all studied islands is Anopheles gambiae s.s., but in the past decade it has been replaced by An. arabiensis in the coastal mainland of Lake Victoria 8 . On islands, the grazing area for livestock is limited, which may reduce the availability of livestock for the zoophilic An. arabiensis 12 , thus suppressing the number of this species. Minakawa et al. 13 reported that availability of Anopheles larval habitats found in the late rainy season in Lake Victoria was strongly affected by human activities such as man-made holes and roadside ditches created by vehicles and irrigation. In Ungoye unpaved road and extensive irrigation for farming provide ample favourable Anopheles larval habitats, which likely contribute to the high malaria prevalence in this area. Nonetheless, other factors underlying the variation of malaria endemicity on our study area are not fully understood, but may include geomorphological or geophysical formation 14 , household structure features, use of protective measures, variation in distance to the nearest mosquito breeding site 15 , and both human behavioural and genetic factors that may also result in differential attractiveness to mosquitoes and resistance to malaria parasites.  Local variation in malaria prevalence was also observed among communities on the same island. Our study showed that malaria prevalence by PCR in the eight catchment areas on Mfangano Island ranged from 9% to 59%. For the first time, we reported malaria prevalence by PCR on Ringiti Island, a small inhabited satellite island to the west of Mfangano Island. Our results indicate that even among lowland areas along the shore of Mfangano, parasite rates were higher in some communities (Wakula and Wakinga) than others (Mrongo and Ugina). In the highland areas, parasite rates were lower in Kangungu than Gulwe. The latter is connected by a small access road to Wakula, where prevalence was high, suggesting that human movement might yet be another contributing factor to the micro-geographical variation in malaria endemicity.
Both parasite rates and parasitaemia were strongly age-dependent, although in different ways. While mean malaria parasitaemia was highest among children under five years and decreased with age, prevalence of malaria infections rose throughout childhood and only started decreasing in adolescents and adults. These different patterns may be explained by local mosquito biting behaviour and acquisition of immunity. In addition, the reported use of bed net among school-age children was considerably low (33%) in the lakeside region 16 , mostly because these children did not sleep in beds 17 . The increase in parasite prevalence among school-age children likely reflects an increase in exposure to infective mosquito bites. It is now well established that the main determinant of the age distribution of malaria is the development of anti-parasite immunity that restricts the density of asexual parasitaemia which peaks in children less than 5 years old and subsequently declines in an age-dependant manner 18 . Nonetheless, the mechanisms mediating anti-parasite and anti-disease immunity are complex but are thought to include both innate and adaptive immune responses that limit the blood stage of the parasite cycle in the human host 19 .
Convenience sampling approaches are efficient and cost effective, but are more likely to be flawed by selection bias. In this study, surveys often took place in primary schools, meaning that children of school age (6 to 15 years) were disproportionately represented -the age distribution varied slightly depending on the setting (Supplementary Table S1). As these school-aged children had the highest prevalence by both microscopy and PCR (Table 1), over-representation of this age group among our samples likely overestimated the overall malaria prevalence in the study area.
No P. vivax infection was found in any of the surveys by either microscopy or PCR, consistent with the absence in many sub-Saharan populations of the Duffy blood-group-antigen, which is required for red blood cell invasion by P. vivax 20 . The predominant species in the study area was P. falciparum, consistent with previous studies conducted in Mbita sub-county 10,21,22 . In this study P. malariae was the second most prevalent species. By both microscopy and PCR, the majority of P. malariae infections were double co-infections with P. falciparum. Triple co-infections of P. falciparum, P. malariae, and P. ovale were more common than P. malariae mono-infection ( Table 2). Previous studies conducted in Mbita sub-county also reported the presence of P. malariae in mono-and mixed species infections by microscopy 10,22 . P. ovale was recorded as a minor species and reported cases caused by this species were rare as a result of under-diagnosis or low transmission rates 23,24 . P. ovale is known for its low densities 25,26 which require more sensitive and specific diagnosis. Consistent with this observation, P. ovale-specific prevalence by PCR was 13 times higher than that by microscopy in this study. The prevalence of P. malariae and P. ovale was higher in Ungoye and Mfangano Island than in the small islands. Apart from difference in wide range of ecological factors such as climate and vector-human interactions, these geographical variations could indicate the existence of various species-specific geographical niches. Differential species burden patterns observed among the study areas call for the need to have diagnostic tools or approaches that can facilitate the detection of such non-falciparum malaria cases 27 .
The high proportions of P. falciparum gametocyte carriers were observed among P. falciparum-infected children from all study areas during the study periods. Previous study by Churcher et al. found evidence that within a narrow age band-6 months to 10 years-there may be a positive association between age and the likelihood of transmission 28 . This has been associated with naturally acquired human immune responses against gametocyte antigens that may reduce or prevent the transmission of malaria parasites to mosquitoes 29,30 and may be highly prevalent in young age groups 29 . In contrast with immune responses to the asexual stages of P. falciparum, where effective immunity matures with age, the higher gametocyte densities in children induce higher levels of transmission reducing activity 29 . These studies also suggest that the underlying gametocyte prevalence in different  Table 3. Proportion of populations with gametocytes in all surveys detected by microscopy. N = total number of participants with P. falciparum infections by microscopy; n = number of individuals positive for gametocytes parasite; CI = confidence interval.
settings can vary widely and has an impact on the size of the infectious reservoir 31,32 . Nevertheless, as microscopy is not a very sensitive tool for detecting all relevant densities of gametocytes 32 , we expect the true gametocyte prevalence to be much higher than reported here. Indeed, gametocytes are the currency of transmission from human to mosquito for maintaining the malaria cycle. The presence and infectiousness of gametocytes in circulation determines the success of transmission from humans to mosquitoes 33 . This study also uncovered the extent of asymptomatic malaria in the Lake Victoria basin in Kenya. The vast majority (91.8% by PCR) of malaria infections in the study area were asymptomatic i.e. not accompanied by febrile symptoms. While these asymptomatic infections were most prevalent in children 15 years and under, they were not uncommon in the adult population. As potential gametocyte carriers, this asymptomatic adult population represents an important reservoir for malaria transmission. Many of these asymptomatic infections are present at densities below the limit for microscopic detection and therefore, the use of microscopy as the sole diagnostic method likely leads to an underestimation of the malaria burden. Our study showed that microscopy detected significantly fewer asymptomatic infections when compared to PCR, consistent with several previous findings that molecular-based assays were more sensitive and time-efficient methods for asymptomatic malaria infections 34,35 . Our study found 10.2% of the asymptomatic P. falciparum infections detected by microscopy were also carrying gametocytes. Previous studies have shown that a proportion of these individuals may play an important role in malaria transmission [36][37][38] . In moderate and high transmission settings in Africa, long term persistence of infection and prevalent gametocyte carriage have been observed among those with asymptomatic malaria from weeks to months 39,40 . This phenomenon has been explained by a form of acquired immunity that suppresses parasitaemia in low level without achieving complete clearance 18 .
The increase in the proportion of sub-microscopic infections by age was contrasted by the decreases in infection prevalence and parasite densities in the population. Parasite density is mediated by host acquired immunity 37 , in that individuals with higher levels of immunity (i.e. older ages) would maintain parasites at lower densities 41 . Okell et al. 42 reported that the proportion of sub-microscopic infections is inversely correlated with slide prevalence and parasite density on the global level 42 . The majority of sub-microscopic malaria infections in the Lake Victoria area were asymptomatic. The importance of the sub-microscopic asymptomatic parasite pool rests on the understanding that sub-microscopic infections can transmit malaria [43][44][45] , although the minimum parasite density necessary for transmission is unknown. In the Lake Victoria area, asymptomatic and sub-microscopic infections in older children and adults may be an important source of the local transmission.
Conventional PCR is typically able to detect infections down to approximately one parasite/μ l using dried blood spots 43 . Presumably there are many infections below this threshold that are undetectable. Whether to screen and treat using a more sensitive tool or to institute mass drug administration (MDA) without screening requires a better understanding of the minimum density of malaria parasites that results in human-mosquito transmission, as well as the comparative costs and operational ease of different approaches 43 .
Islands present a unique opportunity to interrupt malaria transmission due to the relative geographical isolation and confined population. MDA was previously recognized as an important method of malaria control 46 , and it is thought that malaria can be eliminated on isolated islands using MDA and vector control if there is a high  level of community commitment 47 . In the Southwest Pacific, malaria was successfully eliminated by MDA from small islands such as Aneityum in Vanuatu 47 and Nissan in Papua New Guinea 48 , where population movement was closely monitored and ports of entry were controlled to minimize parasite re-importation. The main challenges for MDA include the need to achieve high treatment coverage, and the necessity to repeat the intervention or combine it with other interventions such as vector control in all endemic settings for long-term effect 49 . As most infections in our study area were asymptomatic and sub-microscopic, more sensitive and field-friendly diagnostic tools are needed to inform timely adjustment to MDA protocols. Furthermore, robust surveillance systems to detect malaria resurgence after successful MDA or elimination may be crucial. A number of studies have indicated that while semi-immunity to malaria can be acquired in highly endemic areas by the age of 5 years, this immunity wanes rapidly without ongoing parasite exposure 18,50 . In the event of resurgence, children and adolescents might be more susceptible to reinfection and development of clinical and severe malaria due to their shorter duration of previous exposure to parasites when compared to adults. In summary, this study highlighted the heterogeneity in malaria prevalence and importance of asymptomatic and sub-microscopic infections on islands and a coastal area in Lake Victoria. We observed significant geographical differences and age-dependency in malaria parasite carriage. The unique findings of malaria epidemiology in our study area may have important implications for the development and implementation of malaria elimination strategy. Taking into account the high and heterogeneous prevalence of asymptomatic and sub-microscopic malaria infections on islands in Lake Victoria, the use MDA to eliminate malaria may be justified. An appropriate package that complements MDA with effective vector control and rapid detection and treatment of new cases has the potential to achieve long-term malaria elimination on islands in Lake Victoria.

Methods
Ethics statement. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Kenyatta National Hospital/University of Nairobi-Ethics and Research Committee in Kenya (No. P7/1/2012) and the Committee on the Ethics of Human Research of Karolinska Institutet in Sweden (Dnr 201271239-31/4). Survey subjects were informed by local interpreters of the purposes and procedures of the study, and a written informed consent was obtained from each adult participant at study registration. In the case of children, written informed consents were obtained from their parents or legal guardians.
Study area and population. This study was conducted in Homa Bay County, Kenya and included one large island (Mfangano) and three small islands (Ngodhe, Kibuogi and Takawiri) in Lake Victoria, and one coastal village (Ungoye) on the mainland (Fig. 6). The dominant ethnic group in the study area is Luo; Dholuo is primarily spoken, as well as the national language of Kiswahili.
In western Kenya a bimodal pattern of rainfall is generally observed, with the long rainy season from March to June and the short rainy season from November to December, but the periods vary each year. Annual rainfall ranges from 700 mm to 1,200 mm. The mean temperature is 25 °C while the maximum temperature is 30 °C, and humidity is relatively high. Malaria prevalence peaks usually lag 1-2 months after the rainy seasons. In the early 1990 s, annual entomological inoculation rates (EIRs) reported between 60 and 300 infectious bites per person per year 51,52 . Nevertheless, following the distribution of insecticide-treated net (ITN), the EIR equivalents estimated from serological markers were reduced to fewer than 50 infectious bites per person per year 53 . The important malaria vectors in this region are An. gambiae s.s., An. arabiensis and An. funestus 13 . An. gambiae s.s. has experienced a reduction in density of nearly 95% in the Lake Victoria basin over 10 years, and has been replaced by An. arabiensis as the dominant malaria vector species in the mainland part of this region. Nevertheless, An. gambiae s.s is still the predominant malaria vector on islands in Lake Victoria 8 and its abundance can increase up to eight-fold during the long rainy season 14 . HIV and other parasitic infections such as schistosomiasis and filariasis are highly endemic in this area 54 .
Though there are many development initiatives in these areas, poverty is still a major challenge. Almost half of the population in Homa Bay County survives on below the Kenyan poverty line of Kenya Shilling (KSh) 1,562 (USD 15.6) per month 55 . Farming and fishing are the major economic activities for the island and lakeshore communities. Farming is mostly subsistence-based, with major crops including maize, millet and beans as well as small-scale animal husbandry (cattle, goats and chickens). Fishing is done primarily by men using small, unpowered wooden boats. Other fishing related activities such as registration of fishermen and fishing vessels, and sale and purchase of fish are coordinated by community-based beach management units (BMU). Houses are typically made of mud walls with thatched or corrugated iron roofs.
Large Island: Mfangano. Lying just east of the border with Uganda, Mfangano ( Fig. 6; 66 km 2 ) has a population of approximately 18,600 (HDSS). The island's terrain is rocky and hilly. The island is easily recognisable from the mainland by the imposing green-clad cliffs of Mount Kwitutu, which rises 562 m above the lake surface from the island's centre. Mfangano Island is predominantly rural. Nevertheless, it is well connected to Mbita (16 km) in the mainland by a popular ferry service and numerous private, scheduled and unscheduled motorized boat trips. Residents of Mfangano Island rely on the neighbouring Mbita for regional transportation and access to services such as banks. There are six government health facilities (one health centre and five dispensaries) across the island. In contrast to other study sites, on Mfangano there is a sizable Suba speaking population which is becoming rarer due to the intermarriage between the Suba and Luo ethnic tribes. In addition to farming and fishing, tourism is also growing on this island as the connection to and from Mbita is being improved.
There Island in the west (HDSS, population 1936). These catchment points were selected to represent the different local environments found on the island, from the lakeshore lowland areas to the central highland areas.
Small Islands: Ngodhe, Kibuogi and Takawiri. Each of these three small islands (Fig. 6) has a population of approximately 700 (HDSS) distributed between two main settlements or boat landing beaches: Bonde and Luanda on Ngodhe Island, Kibuogi A and Kibuogi B on Kibuogi Island, and Kamarach and Kongata on Takawiri Island. The islands' elevation is up to 70 m from the lake level. Takawiri and Kibuogi are close to Mfangano and the boat traffic connecting among these islands is frequent. The distances between Mbita in the mainland and Takawiri and Kibuogi are approximately 14 km and 16 km, respectively. On the other hand, Ngodhe Island is situated to the north of Rusinga Island, which is connected to Mbita via a causeway. These islands have been extensively deforested and shrubs constitute the main vegetation today. Takawiri and Ngodhe are each served by a dispensary, but no public health facility is available on Kibuogi.
Coastal mainland: Ungoye. Ungoye (Fig. 6) is a small village (19.3 km 2 ) in Nyangwethe area with population of 3,471 (2009 census). The village is connected by an unpaved roadway to the nearby small towns of Sindo (17.2 km) and Mbita Point (31.2 km), each with a sub-district hospital. There is one small public health facility in this area. Ungoye was included for comparative purposes due to its similarity to the islands in environmental characteristics, infrastructure, and access to health services. A recent survey in the Gembe East area revealed a high P. falciparum prevalence of 66.5% by PCR 56 . Study design and data collection. Five cross-sectional surveys were conducted between 2012 and 2014: two after the short rainy season in January-February 2012 and March 2014 and three after the long rainy season in July-August 2012, August 2013, and August 2014. Peak malaria transmission occurs 1-2 months after the rainy season and the mean monthly anopheline vector abundance has been reported to increase by 6-to 8-fold in the rainy season compared to the dry season in the Lake Victoria basin 14 ; thus, surveys were conducted approximately 2 months after the rainy seasons to coincide with the periods of heaviest malaria burden. All sites were included in each cross-sectional survey, with the exception that Mfangano and Takawiri were not included in the March 2014 survey. This study used the convenience sampling strategy, whereby residents were asked to come to selected survey points such as BMU community hall, school or market place for study participation. During this three-year period, we did not see any major changes or events (e.g. mass distribution of ITNs and indoor residual spraying in previously uncovered areas) which might an impact on our study findings.
Field and laboratory methods. Sex and age were recorded for each participant. Axillary body temperature was determined using a digital thermometer (Terumo, New Jersey, US). Fever was defined as a temperature exceeding 37.5 °C. Hemoglobin level was measured with the HemoCue Hb 201 analyzer (HemoCue, Angelholm, Sweden). A measurement below 11 g/dL was classified as anaemic. Children aged 12 years and below were examined for enlarged spleen by AK according to Hackett's method, regardless of fever or malaria status. Measuring spleen sizes in children has been shown as a useful variable to estimate the burden of malaria transmission in other island settings 57 . Thin and thick blood smears were prepared on site, stored in slide boxes and transported daily to the main laboratory in Mbita, where thin blood smears were fixed with methanol, and all smears were stained with 3% Giemsa solution for 30 minutes and examined under oil emersion (10 × 100 magnification) by experienced microscopists. Blood smears were defined as negative if no parasites were found after examining 100 high power microscopy fields. For all positive samples, malaria species were identified and asexual parasite forms were counted against 200 leukocytes. P. falciparum gametocytes counts were separately recorded. Parasite density was estimated from parasite counts, assuming that there were 8,000 leukocytes per μ l of blood. All slides were independently re-examined by two experienced microscopists blinded to the first microscopy reading results. Discrepancies between the two readings were resolved by a third experienced microscopist.
For detection by PCR, blood samples (70 μ l) withdrawn by 75 mm Micro-hematocrit capillary tube (Thermo Fisher Scientific, MA, USA) were spotted on Whatman ET31 Chr filter papers (Whatman International, Maidstone, UK) and dried thoroughly at ambient temperature. The dried blood spots were put in individual zipped plastic bags and stored at − 20 °C. DNA was extracted from a quartered blood spot (17.5 μ l) using the QIAamp Blood Mini Kit (QIAGEN, Germantown, USA) according to the manufacturer's instructions. Plasmodium infection was detected by a nested PCR protocol that can distinguish the four major human malaria species (except P. knowlesi) 58 .
Data analysis. All survey data were double entered into Microsoft Excel spreadsheets and cross-checked for errors. Data were processed and analysed using the STATA/SE 13.1 statistical software package (StataCorp, TX, USA). Differences in participant characteristics and prevalence among age groups and geographical areas were assessed using Chi-squared (χ 2 ) or Fisher's exact tests. The average enlarged spleen (AES) index for each survey was calculated as the sum of the number of children in each spleen size class multiplied by the class number (0-4), divided by the total number of palpable spleen. Interactions between survey, individual settings, age group and prevalence were examined using conditional logistic regression. Univariate logistic regression was performed to identify risk factors for the outcome of Plasmodium infection as determined by PCR. Sex, age group, geographical area, fever, enlarged spleen and anaemia were considered as individual variables. All variables with a P-value of less than 0.05 from a likelihood ratio test in univariate analyses were entered into a multivariate logistic regression model and stepwise backwards elimination was used to identify the main risk factors for infection.