An observational study of Plasmodium inter-species interactions during a period of increasing prevalence of Plasmodium ovale in symptomatic individuals seeking treatment

Abstract

Background

The epidemiology and severity of non-falciparum malaria in endemic settings has garnered limited attention. We aimed to characterize the prevalence, interaction, clinical risk factors and temporal trends of nonfalciparum species among symptomatic individuals in endemic settings of Kenya.

Methods

We diagnosed and analyzed infecting malaria species via PCR in 2027 clinical samples collected between 2008 and 2016. Descriptive statistics were used to describe the prevalence and distribution of Plasmodium species. A statistical model was designed and used for estimating the frequency of Plasmodium species and assessing interspecies interactions. Mixed effect linear regression models with random intercepts for each location was used to test for change in prevalence over time.

Findings

72•5% of the samples were P. falciparum single species infections, 25•8% were mixed infections and only l•7% occurred as single non-falciparum species infections. 23•l% were mixed infections containing P. ovale. A likelihood-based model calculation of the population frequency of each species estimated a significant within-host interference between P. falciparum and P. ovale curtisi. Mixed-effect logistic regression models identified a significant increase of P. ovale wallikeri and P. ovale curtisi species over time with reciprocal decrease in P. falciparum single species and P. malariae. The risk of P. falciparum infections presenting with fever was 0•43 times less likely if co-infected with P. malariae.

Interpretation

Findings show higher prevalence of non-falciparum species than expected. The proportion of infections that were positive for infection by P. ovale wallikeri and P. ovale curtisi was observed to significantly increase over the period of study which could be due to attenuated responsiveness to malaria drug treatment on these species. The increase in frequency of P. ovale species in Kenya could threaten malaria control effort in Kenya and pose increased risk of malaria to travelers.

Funding

AFHSB and its GEIS Section

Introduction

Malaria control programmes in sub-Saharan Africa (sSA) have mostly focused on Plasmodium falciparum, the predominant cause of lethal malaria. Although widely distributed in malaria-endemic regions including sSA ^{1, 2} information on the epidemiology of non-falciparum malaria caused by P. ovale spp. and P. malariae is scarce. Further, the severity of the disease caused by these species, which is thought to be milder compared to P. falciparum, is not well studied in the endemic human populations. Most of the clinical data for non-falciparum species has been obtained from travelers returning from malaria endemic areas.^{2, 3}

P. ovale spp. and P. malariae often occur in coinfections with P. falciparum which complicates the ability to accurately detect these infections using light microscopy or malaria rapid diagnostic tests (mRDT).⁴ This is further exacerbated by the fact that these infections often occur at low parasitemia,⁵ which likely contributes to the underestimation on the prevalence of non-falciparum in sSA. However, molecular and serological methods have been shown to be more sensitive, revealing that non-falciparum infections are more prevalent than previously estimated,^{6, 7} with one study measuring seroprevalence of P. ovale spp. and P. malariae at 57% and 45%, respectively in asymptomatic populations in Benin.

Plasmodium species occur sympatrically globally, and as simultaneous infections in host and vectors but their salient biology sustains their species distinctiveness. Studies on mixed Plasmodium species biology suggest that interactions between co-infecting Plasmodium species shape infection dynamics, virulence⁸, species frequency due to altered resource allocation as well as response of each of the species to both human host and vector immunity. Some studies associate these interactions with preventing development of severe malaria disease,¹⁰ and modulating transmission^{8, 11, 12} for selected species in an infection. Mixed Plasmodium species infections pose an intervention challenge due to the variability in case management and transmission prevention. For example, the incubation period for most Pf infections lasts between 9 to 14 days while that of non-falciparum species ranges from 18 days to 85 days for relapsing P. ovale malaria^{10, 13}.

Artemisinin-combination therapies (ACTs) are the first-line treatment for uncomplicated P. falciparum species malaria infections in most malaria endemic countries. However, there is limited in vivo data available on the efficacy of these drugs against non-falciparum species malaria.⁶ Further, most therapeutic efficacy studies exclude mixed infections as part of their enrolment criteria. These studies mostly use microscopy to assess drug efficacy, which is less sensitive compared to PCR.¹⁴ Recent studies have detected persistent non-falciparum parasites after treatment with ACTs when assessed by PCR.^{6, 15–17} There is thus a need to characterize species specific responses to ACTs, which can help inform the efficacy of ACT regimens in treating non-falciparum infections in malaria endemic regions. This would improve both the management of non-falciparum species malaria cases in endemic regions as well as the growing number of imported malaria cases associated due to these species globally.^{3, 18, 19}

Molecular surveillance provides more accurate speciation data as recently demonstrated by a cross-sectional study conducted in the general population via convenience sampling in western Kenya, where the overall prevalence of Plasmodium spp. was estimated to be 37•1% (13•2% non-falciparum species) by PCR versus 19•9% (1•6% non-falciparum species) by microscopy. To further our understanding on the spatial and temporal trends in non-falciparum species, we conducted a longitudinal study in four different malaria endemicity zones in Kenya, and collected clinical and malaria speciation data from symptomatic individuals who presented at care facilities seeking treatment. Comprehensive patient data was collected and blood samples were analyzed using highly sensitive speciating quantitative PCR (qPCR). We investigated the clinical risk factors associated with non-falciparum infections in a malaria endemic population, and developed and applied a novel statistical framework for exploring whether species occur independently of other species. Lastly, we explored the temporal trends in both single and mixed species infection over the study period across the different endemicity zones.

Methods

Ethics statement

This study was conducted under the approval of the Kenya Medical Research Institute (KEMRI), Scientific and Ethics Review Unit (SERU) and Walter Reed Army Institute of Research (WRAIR) institutional review boards, protocol numbers: KEMRI #1330, WRAIR #1384 entitled “Epidemiology of malaria and drug sensitivity patterns in Kenya.”

Study sites and sample collection

Samples were collected between 01-March-2008 and 31-December-2016 from hospitals located in 6 regions that were chosen to span 4 distinct malaria transmission zones across Kenya. ²¹ These comprised 2 sites in the Lake endemic malaria zone at Kisumu and Kombewa hospital, 2 sites in the highland epidemic malaria zone at Kisii and Kericho hospital, one site in the Coast endemic malaria zone at Malindi hospital as well as 1 site in the Semi-arid seasonal malaria zone at Marigat hospital (see Supplementary Figure 1, Table 1). Patients aged six months and above, presenting at outpatient departments with symptoms of malaria and/or testing positive for uncomplicated malaria by rapid diagnostic test (mRDT; Parascreen® (Pan/Pf), Zephyr Biomedicals, Verna Goa, India) were recruited into the study after providing written informed consent or assent. Written informed consent for children aged below 18 years was provided by their parent or legal guardian in accordance with the laws of the government of Kenya.

Individuals were excluded from the study upon refusal to participate or unwillingness to give blood. The study also excluded prisoners, including children attending the Kenyan government’s correction and rehabilitation programme. Additionally, children under age 18 without available parent or legal guardian, volunteers who were previously enrolled in the study during the same calendar year, individuals weighing less than 5 kg and anyone who, in the opinion of the attending medical provider, would be adversely affected by the drawing of 2.5 mL of blood were excluded. For all consenting individuals who were enrolled into the study, Case Report Forms (CSF) were used to collect comprehensive patient information including age, sex, occupation, home of origin, history of malaria infection and treatment, chief complaint, other complaints (such as headache, vomiting, coughing, diarrhea), body temperature and body weight. 2-3 ml of whole blood was collected for mRDT testing and malaria blood smear preparation. About 100 μl of each sample was spotted on FTA filter paper (Whatman Inc., Bound Brook, New Jersey, USA) for DNA extraction and nucleic acid analysis.

The clinician performed diagnosis by assessing for symptoms such as conjunctival pallor, lymphadenopathy and splenomegaly. Final diagnosis results were based on clinical evaluation confirmed by mRDT and/or microscopy. All Plasmodium positive cases were treated with Coartem® which contains artemether-lumefantrine (AL) in accordance with the Kenya Ministry of Health recommended case management guidelines for uncomplicated malaria. After drawing blood, the attending clinician administered and observed the first dose of artemether-lumefantrine 20mg/120mg tablets, with individuals weighing 5 to<15 kg prescribed 1 tablet per dose, 15 to < 25kg 2 tablets per dose,25 to <35kg 3 tablets per dose and individuals greater than 35kg 4 tablets per dose. Each patient was given the remaining 5 doses (i.e 5, 10, 15, 20 and 25 tablets for each weight range) and advised to take the next dose after eight hours, followed by the remaining doses at 12 hourly intervals till completion of treatment. Further, individual were encouraged to return to the hospital should symptoms persist. Individuals with recurrent parasitemia during the study period were treated with ACTs but not re-enrolled in the study.

Molecular diagnosis

Genomic DNA was extracted using the QIAamp DNA mini kit (Qiagen, Valencia, CA, USA) as recommended by the manufacturer. The DNA was used for malaria diagnosis using Genus-specific qPCR assay prior to speciesspecific analyses as previously described for the detection of Plasmodium genus and P. falciparum, which targets 18s rRNA genes For each assay, a negative template control (distilled water) and P. falciparum DNA Nucleic Acid Tests (NAT)/NIBSC genus-specific positive controls were included to ensure no contamination of the PCR and to validate results. The human housekeeping gene Ribonuclease P (RNaseP) was used as an extraction and qPCR assay control.

All samples that were diagnosed as positive for malaria were further characterized for species composition using a separate set of species-specific primers indicated in supplementary Table 1. The assays for characterization of P. falciparum and P. malariae had identical PCR reaction components and conditions as the Genus-specific qPCR assay except for the primers used. Specifically, FAL R, FAL F, and FAL were used for P. falciparum diagnosis while MAL F, MAL R, and MAL PP were used for P. malariae diagnosis. Detection of the two P. ovale species was conducted using a previously described method, with comparable limits of detection (1-5 parasites/uL) for each PCR reaction.^{23, 24} Reaction components and amplification conditions were identical to the Genus-specific qPCR assay. Each experiment included at least one reaction mixture without DNA as a negative control.

Statistical analysis

Estimating the frequency of Plasmodium species and assessing inter-species interaction

A statistical model was designed to assess whether there was a significant difference between the observed frequencies of each infection type and the expected frequencies when assuming independent acquisition of different Plasmodium species (referred to as the “independent model”). The approach jointly estimates the frequency of each Plasmodium species and the mean number of infections per individual, and uses these estimates to generate null distributions for the number of single and multi-species infections. The model was then extended to statistically assess for the presence of between-species interactions (referred to as the “interference model”). Interactions were assumed to either increase or decrease the probability of additional infections being caused by different Plasmodium species dependent on the previous infection. In both models, the use of the Poisson and negative binomial distribution to describe the number of infections was explored and the best fitting model identified using comparisons of sample-size corrected Aikaike information criterion (AICc). The best fitting interaction models were compared against the independent model using log-likelihood ratio tests to statistically test for the presence of between species interactions. Full methodology is described in the Supplementary Material.

Linear modelling of species prevalence and risk factors associated with fever

Summary descriptive statistics were used to describe the prevalence and distribution of Plasmodium species and the occurrence of mixed species infections across Kenya between 2008 and 2016 within symptomatic individuals. To test for a change in the prevalence of Plasmodium species over time, we used a mixed effect linear regression model with random effects for each location using the lme4 R software package.²⁵ Due to low numbers of samples collected prior to 2011 in Kombewa, Marigat, and Malindi, samples were grouped according to the 4 endemic zones: highland epidemic (Kericho and Kisii), coastal endemic (Malindi), semi-arid (Marigat), and lake endemic (Kisumu and Kombewa). The samples were also grouped according to the year and month they were collected in, and the regression was weighted according to the number of samples collected in each year and month. This framework was used to assess for changes in the prevalence of infections within symptomatic individuals caused by P. malariae, P. ovale curtisi, P. ovale wallikeri as well as the prevalence of infections positive for only P. falciparum.

Patient metadata was available for a subset of individuals reporting at clinic, which included the chief symptomatic complaint reported by the patient as well as whether the individual was currently presenting with fever based on temperature measurements on arrival at the hospital. For clinical and demographic data, the mean and bootstrapped 95% CI using the bias-corrected and accelerated (BCa) bootstrap method are presented. We explored the patterns in the chief complaint with respect to the infecting species composition to assess whether the distribution of infecting species by complaint was predicted by the earlier estimated frequency of Plasmodium species. We used the same approach to assess for changes throughout the year in infecting species composition. In both analyses, we report instances in which both more than one occurrence of the infecting species composition was observed per chief complaint or month group and the observed infection composition did not occur within the 95% prediction quantile based on the earlier estimated frequency of Plasmodium species. Lastly, we used a mixed effect logistic regression model to assess the risk factors associated with fever presentation at clinic among individuals infected with P. falciparum. The included covariates were patient age, sex, time of sample collection, number of malaria attacks in the last year and whether the individual was co-infected with P. malariae, P. ovale curtisi or P. ovale wallikeri. All statistical analyses were conducted using R version 4.0.1.

Role of the funding source

The funder of the study played no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

Sample Collection

During the study period, 3120 study participants from six field sites (county and sub-county hospitals) who comprised symptomatic individuals presenting at care facilities, were screened, 3058 met inclusion criteria and were enrolled in the study. Of this, patient metadata was collected from 2719 study participants, with 2027 samples successfully analyzed for all Plasmodium species by speciating qPCR. Table 1 shows the location and metadata for the samples that were qPCR analyzed, study sites location in the regions which they were collected and the epidemiological zones they occur in, with lake and coastal endemic zones grouped separately.

Table 1. Study site demographics and clinical summaries.

Site	Region	Age*	Sex (% female)*	Fever*	Malaria episodes in last 12 months*	N
Kerich	Highland Epidemic	12.25 [10.33-14.56]	46.40% [37.60-55.20]	73.55% [64.46-80.17]	0.73 [0.56-0.92	170
Kisii	Highland Epidemic	8.49 [6.96-10.47]	52.29% [44.44-59.48]	88.89% [82.35-92.81]	0.50 [0.42-0.59]	432
Kisumu	Lake Endemic	8.85 [7.99-9.70]	50.37% [45.76-54.58]	62.31% [57.84-65.67]	0.95 [0.85-1.05]	887
Kombewa	Lake Endemic	8.16 [7.12-9.50]	52.19% [46.03-56.90]	66.44% [60.27-71.23]	1.98 [1.78-2.15]	437
Malindi	Coastal Endemic	14.36 [9.18-22.55]	50.00% [22.73-68.18]	50.00% [27.27-68.18]	0.08 [0.00-0.21]	24
Marigat	Semi-Arid Zone	17.71 [13.30-23.81]	57.14% [28.57-71.43	71.43% [47.62-85.71]	0.10 [0.04-0.19]	77
All		9.27 [8.75-9.76]	50.78% [48.28-53.10]	68.03% [65.68-70.22]	1.02 [0.96-1.08]	2027

^*Mean [95% Bootstrapped Confidence Interval]

Plasmodium species composition

P. falciparum was the most prevalent species in symptomatic individuals presenting at care facilities. P. falciparum was present in 98•0% (1986/2027) of the infections, P. ovale wallikeri at 20•0% (405/2027), P. malariae at 5•0% (102/2027), and P. ovale curtisi at 5•0% (101/2027). 74•2% (1504/2027) of samples carried single parasite species infections while 25•8% (523/2027) had multiple species infections. 0•15% (3/2027) of samples had a mixture of all the four species. The rest of the samples had either two 23•8% (482/2027) or three 1•9% (38/2027) species infections. The most prevalent mixed infections contained P. falciparum and P. ovale wallikeri. The full composition of the infections observed is shown in Figure 1. When analyzed per study site, in symptomatic individuals presenting at care facilities, Kisumu and Marigat had the highest prevalence of P. falciparum single infections at 79•2% (704/887) (95% CI: 76•6% - 81•9%) and 79•2% (61/77) (95% CI: 68•9% - 86•8%), respectively (Supplementary Figure 2). On the other hand, Kombewa (rural, lake endemic) had the lowest P. falciparum single infections at 61•3% (268/437) (95% CI: 56•7% - 65•8%), which was significantly lower than Kisumu and Marigat (p < 0.0001 and p = 0.0019 respectively).

Figure 1. Observed Plasmodium species composition.

Infections caused by only one species accounted for 74•20% of infections i.e. P. falciparum, P. ovale wallikeri, P. ovale curtisi, and P. malariae. The most prevalent multiple species infections were caused by P. falciparum and P. ovale wallikeri.

Plasmodium species frequencies and inter-species interactions

Using the statistical model developed to estimate the frequency of each Plasmodium species under the assumption of independent species acquisition, i.e. no inter-species interactions, a frequency of 89•5%, 7•2%, 1•7%, and 1•7% for P. falciparum, P. ovale wallikeri, P. malariae and P. ovale curtisi, respectively was predicted, with a mean number of infections equal to 2•93 (Supplementary Table 2). These estimates captured the observed frequency distribution of the composition of infecting species, with the 95% bootstrapped quantiles generated containing the observed data for each infection type (Figure 2a). The median estimate from the model predictions varied between infection type, with occurrence of P. ovale wallikeri single species infections, and P. falciparum/P. ovale curtisi infections being over estimated compared to the observed infection composition. Conversely, the model under-predicted the occurrence of P. ovale curtisi single species infections and P falciparum/P. ovale wallikeri infections. The variation in the observed distribution of infection types was better explained using a model that included interactions betweenPlasmodium species (Supplementary Table 2, Figure 2b), with the best fitting model including one interaction term (= 9•24, p = 0•0024), which predicted a significant interference between P. falciparum and P. ovale curtisi (k_1.3 = 0•405) (Supplementary Table 3). In all equivalent models, the use of a Poisson distribution to describe the number of infections yielded more parsimonious models. This finding was explained by the observed flat gradient in the model likelihood with respect to the dispersion parameter r for well-chosen values of μ (Supplementary Figure 3). Full model parameters and likelihoods of each model fitted are shown in Supplementary Table 3.

Figure 2. Predicted infection species composition.

Plots show the estimated distribution for each infection composition, consisting of P. falciparum (pf), P. malariae (pm), P. ovale curtisi (poc), and P. ovale wallikeri (pow). Distributions were estimated using 50,000 sampling repetitions drawn from the best fitting a) independent and b) interference model. Blue regions show the 95% quantile interval, with the median shown in white line. The observed infection composition from the data is shown with the red dashed line. The interference model shown in b) included one additional parameter, which was an interference between P. falciparum and P. ovale curtisi.

Analyses of the longitudinal trends of the infecting species composition showed significant changes occurred over the study period (Figure 3). The mixed-effect linear modeling identified a significant increase in both P. ovale spp. infections over time among symptomatic individuals presenting at care facilities, with P. ovale wallikeri having the largest increase (2.1% per year, p = 0•0217) (Supplementary Table 4). Conversely, there was a decrease in the frequency of single species infections caused by P. falciparum (2.5% per year, p = 0•0065). Lastly, we assessed whether seasonal dynamics impacted the infecting species compositions. We used the frequency of each Plasmodium species estimated earlier to generate the expected species compositions for each month. All species compositions were captured within the 95% prediction quantile for each month except for infection with P. falciparum/P. ovale wallikeri, which was observed more frequently than expected between April-June and less frequently than expected during December (see Supplementary Figure 5).

Figure 3. Frequency of infections containing P. ovale curtisi, P. ovale wallikeri, P.malariae,and only P.

falciparum. Each plot shows the percentage of infections that were positive for each Plasmodium species over timefor the four transmission zones sampled. A mixed-effects linear regression model with a random intercept for each transmission region was fitted to the data and is plotted in red in each plot.

Fever presentation and symptomatic complaints

Overall, 68•0% of symptomatic infections presented with fever (recorded with a temperature above 37•5°C) (Table 1). The proportion of cases presenting with fever at clinics was highest in younger individuals and decreased with increasing age (Supplementary figure 6). Analysis of the risk factors associated with fever presentation in individuals positive for P. falciparum revealed that the infection composition was associated with fever. The odds of presenting with fever at the clinic (adjusted for age, sex, year, and previous malaria attacks) was estimated to be 57% lower if the individual was co-infected with P. malariae (adjusted OR: 0•43, 95% CI: 0•25 -0•74, p = 0•0023) (Figure 4). Additionally, fever presentation increased over time (adjusted OR for an increase of 1 year: 1•08, 95% CI: 1•02 - 1•15), and was less prevalent in older individuals (adjusted OR for an increase of 1 year in age: 0•95, 95% CI 0•94 - 0•96).

Figure 4. Risk factors associated with clinical presentation of fever.

The odds ratio for each predictor assessed is shown with their 95% confidence intervals as whiskers surrounding each point. Odds ratios significantly not equal to 1 are shown in red and were observed for the age, year of sample collection and coinfection with P. malariae.

We investigated the relationship between the reported chief complaint symptoms and the Plasmodium species infection composition of infected individuals upon arrival at clinics. Infections induced by single or mixed nonfalciparum parasites induced similar symptoms as those seen with P. falciparum as single or mixed infections (Supplementary Figure 7). There was one case where a P. ovale wallikeri single infection induced seizure in a 6year old boy, which was also associated with severe malaria. Lastly, we assessed whether the observed infecting species composition for each symptomatic complaint was predicted by the frequency of each Plasmodium species estimated earlier. Overall, the observed infection compositions occurred as expected for each chief complaint in which more one species composition was observed. The only exceptions were for fever, for which both P. falciparum single infections and P. falciparum/P. ovale wallikeri double infections did not occur within the 95% prediction quantile (see Supplementary Figure 8).

Discussion

Numerous studies have explored how P. falciparum populations are being disrupted by improved malaria control and elimination efforts. However, there are a limited number of studies that have investigated the burden, epidemiology and clinical implication of P. ovale spp. and P. malariae in malaria endemic settings. In this study we conducted a longitudinal field study in six sites in Kenya, recording the clinical presentation and infection composition of clinical malaria episodes in symptomatic individuals presenting at care facilities, which allowed the prevalence of non-falciparum malaria in symptomatic individuals to be estimated. Using novel statistical models to estimate the frequency of each Plasmodium species, we estimate a higher frequency of non-falciparum species than previously shown. Over the study period, there was a significant increase in symptomatic infections containing non-falciparum malaria, and a significant decrease in infections containing P. falciparum single infections. These observations may suggest that the decrease in malaria prevalence exhibited over the study period changed the composition of Plasmodium species occurring in symptomatic individuals presenting at care facilities in these settings. Further study is prompted to both understand if the same change has occurred in the community and to address the implications of this disruption in the context of treatment, control and elimination efforts.

There was a wide array of malaria symptoms recorded as the chief complaints when the study participants were interviewed at the clinics. Clinical episodes due to non-falciparum species infections caused similar symptoms as those caused by P. falciparum species, with seizure reported by some, including those infected only with non-falciparum species. Additionally, P. falciparum coinfection with P. malariae was significantly associated with a decreased risk of presenting with fever at the clinic. This is in agreement with previous studies in Nigeria and Ghana that have also shown decreased overt malaria clinical presentation in coinfections of P. falciparum and P. malariae. One possible explanation is that existing, high parasitaemia P. falciparum infections suppress super infection events by new strains through a previously documented parasite density-dependent suppression of new liver stage infections. An additional explanation could be that due to the shared within-host niche between P. malariae and P. falciparum, the presence of one species at a parasite density sufficient to trigger seeking treatment restricts the ability of previous lower density infections of the other species from persisting. ^{26, 27} For example, infection with P. falciparum may be more likely to result in P. malariae parasites from previous infections being cleared, due to being less able to evade P. falciparum triggered immune mechanisms compared to P. ovale spp. that occupies a different niche.

Our findings showed that during the study period, there was a significant increase in the proportion of symptomatic infections carrying P. ovale wallikeri and P. ovale curtisi, but a decrease in infections due to P. falciparum as single species. P. ovale wallikeri was the most prevalent non-falciparum species across all regions and showed the largest proportional increase over the study period. The statistical modelling developed confirmed this, estimating the frequency of P. ovale wallikeri to be 6•7% compared to 2•5% for P. ovale curtisi. These estimates also suggested a significant interference between P. falciparum and P. ovale curtisi, which approximately halves the probability of a successful subsequent infection by P. ovale curtisi after an initial infection by P. falciparum and vice versa. This interaction could also explain the greater increase in P. ovale wallikeri over time compared to P. ovale curtisi. P. ovale wallikeri and P. ovale curtisi have been hypothesized previously to have different within-host adaptations. ^{1, 28} Although our study cannot comment directly on this, the observed significant interaction, altered rate of increase over time, and significant increase in P. falciparum/P. ovale wallikeri coinfections during the Kenyan rainy season in April – June is suggestive of differences in their epidemiology.

An alternative explanation for the increase in P. ovale spp. could be due to the differential impact of AL on P. ovale spp. This may lead to unresolved infections following AL treatment, which have been previously reported.^{6, 15, 16} In five cross-sectional surveys conducted between 1994 and 2016, among all willing villagers of the Nyamisati population in Tanzania, P. falciparum declined over the study period but the prevalence of P. ovale spp. and P. malariae increased 6- and 2-fold, respectively. This study also coincided with the introduction of AL in Tanzania. In our study, we observed an increase in the frequency of P. ovale spp. (but not P. malariae), coinciding with the introduction of AL in Kenya. This might be due to several factors including unresolved infections following AL treatment^{15, 16, 30} and/or relapsing malaria. ^{28, 30} It is of interest that P. ovale wallikeri exhibited a greater increase, with a recent study from Gabon observing no cases of relapsing infection with P. ovale wallikeri within 32 weeks after treatment with AL, whereas P. ovale curtisi did in 23% percent of patients. It is also possible that P. ovale spp. response to AL is becoming differently attenuated over time. Additional studies are required to further investigate how the use of AL and other ACTs might be playing a role in the shift of Plasmodium species composition in Kenya and elsewhere in sSA.

There are a number of limitations in our study. Firstly, the samples analyzed by this study were obtained from symptomatic individuals presenting at care facilities and may not be representative of the malaria prevalence in the population as a whole. Similarly, the estimated frequency of each species may not be reflective of the true frequency, with the frequency of each species in asymptomatic individuals likely to be different. In addition, in the statistical modelling for estimating the frequencies of Plasmodium species, we modelled that each acquired species was sequentially acquired due to successive bites. However, it is possible that multiple species could be passed on within one infectious bite. This could be an alternative explanation for the differences seen between the observed and predicted infection types using the independent model (Figure 3). For example, this could not be due to between species interactions within the host, but due to an increased affinity or differing adaptation to the vector, with P. falciparum, and P. ovale wallikeri able to be passed on within the same infectious bite. Despite this, predictions from the best fitting model were highly accurate and the developed methodology is flexible enough to enable alternative models for the likelihood of any given infection type. Secondly, the study enrolled malaria positive individuals based on the microscopy and mRDTs diagnosis that have inherent limitations for detection that could have missed out on low density parasitemia and non-falciparum species. This enrolment strategy was a source of bias because mRDTs are more sensitive to P. falciparum than other species. This method was however used by the study in order to align with the diagnosis methods that have been approved by the world health organization and adopted for case definition for malaria case management at health facilities in Kenya. Follow up studies using either using ultra-sensitive mRDTs or focusing on community detection independent on mRDT status would be encouraged in order to characterise the size of the bias introduced with the current enrolment strategy. Additionally, from among those that had positive parasitological diagnosis for malaria, the study enrolled and analyzed only those who provided written informed consent/asset. Lastly, the longitudinal coverage of samples from the coastal endemic and semi-arid zone was substantially less than the other regions. Conclusions drawn relating to the rate of change of Plasmodium species over time are thus largely informed by the highland epidemic and lake epidemic zones. The altered endemicity of these zones could subsequently be driving the patterns seen, with the increased seasonality in these epidemic regions altering the presentation of mixed-species infections.

In conclusion, the prevalence of non-falciparum species collected between 2008 and 2016 has changed, with the proportion of symptomatic infections presenting at care facilities that were positive for infection by P. ovale wallikeri and P. ovale curtisi significantly increasing. Further study is required to assess if the same increase is detected within asymptomatic infections and whether the change in non-falciparum species prevalence reflects changes in the wider population or is unique to individuals seeking treatment. Our developed statistical model for estimating the frequency of Plasmodium species and between species interactions predicted a significant interference between P. falciparum and P. ovale curtisi, which would be, to our knowledge, one of the first efforts to statistically show between species interactions. Additionally, the risk of P. falciparum infections presenting with fever was 0•43 time less likely if co-infected with P. malariae. The observed increase in dormant Plasmodium species infections could thus explain the increased observation of traveler malaria originating from Kenya and other malaria endemic areas. Increased surveillance for non-falciparum species infections is recommended within both symptomatic and asymptomatic individuals to monitor the changing risk of malaria infection from non-falciparum species.

Data sharing

Raw data showing sample, period of collection, location and species composition is shared with the manuscript. All software used and analysis is available at https://www.github.com/OJWatson/kenya_non_falciparum.