Population heterogeneity in Plasmodium vivax relapse risk

Abstract

A key characteristic of Plasmodium vivax parasites is their ability to adopt a latent liver-stage form called hypnozoites, able to cause relapse of infection months or years after a primary infection. Relapses of infection through hypnozoite activation are a major contributor to blood-stage infections in P vivax endemic regions and are thought to be influenced by factors such as febrile infections which may cause temporary changes in hypnozoite activation leading to 'temporal heterogeneity' in reactivation risk. In addition, immunity and variation in exposure to infection may be longer-term characteristics of individuals that lead to 'population heterogeneity' in hypnozoite activation. We analyze data on risk of P vivax in two previously published data sets from Papua New Guinea and the Thailand-Myanmar border region. Modeling different mechanisms of reactivation risk, we find strong evidence for population heterogeneity, with 30% of patients having almost 70% of all P vivax infections. Model fitting and data analysis indicates that individual variation in relapse risk is a primary source of heterogeneity of P vivax risk of recurrences. Trial Registration: ClinicalTrials.gov NCT01640574, NCT01074905, NCT02143934.

Fig 1. Time to first recurrence and weekly incidence per patients at risk in the Thailand-Myanmar and Papua New Guinea data.

(A, B) Time from enrolment to the first recurrence for patients who received blood-stage treatment only (red) and patients who received primaquine and blood-stage treatment (blue). In the Papua New Guinea data, all patients (n = 504) received blood-stage treatment and either a placebo (red) or primaquine (blue). For the time to first recurrence in the Thailand-Myanmar data (n = 1298) by treatment and study see Fig H in S1 Figs (C, D) Weekly incidence per patients at risk for patients treated with primaquine and blood-stage treatment (blue dots) and blood-stage treatment only (red dots). The weekly incidence per patients at risk is the number of patients that had a recurrence within the current week divided by the number of patients who were at risk (i.e., the patients who have not yet had a recurrence) at the beginning of the week. The curves are fitted to the weekly incidence per patients at risk (using splines).

Fig 2. Fitting models of temporal and population heterogeneity to the data.

The left column (A and C) shows the fit of the temporal heterogeneity model and the right column (B and D) shows the fit of the population heterogeneity model. The lines are the models fitted to the data and the shaded areas are the 95% confidence regions from the data. The models were fitted using a maximum likelihood approach (see Methods). (A, B) Fit of the heterogeneity models for each antimalarial treatment and study in the Thailand-Myanmar data. Abbreviations: AS artesunate, CHQ chloroquine, CHQ/PMQ chloroquine and primaquine, DP/PMQ dihydroartemisinin-piperaquine and primaquine, VHX Vivax History study, BPD best Primaquine Dose study. (C, D) Fit of the heterogeneity models to all Papua New Guinea data. For a fit to the Papua New Guinea data grouped by village see Figs C and D in S1 Figs.

Fig 3. Fitting the temporal and population heterogeneity models to the first and second recurrence time in the Thailand-Myanmar data.

Both recurrence times are fitted simultaneously (see Methods and S1 Methods). The lines are the models fitted to the data and the shaded areas are the 95% confidence regions from the data. The left column shows the fit to the first recurrence time and the right column the fit to the time from the first to the second recurrence. The first row shows the temporal heterogeneity model fit and the second row the population heterogeneity model fit. Abbreviations: AS artesunate, CHQ chloroquine, CHQ/PMQ chloroquine and primaquine, DP/PMQ dihydroartemisinin-piperaquine and primaquine, VHX Vivax History study, BPD best Primaquine Dose study.

Fig 4. Association between different recurrence times.

(A) Since we have multiple within patient recurrence times in the Thailand-Myanmar data, we can estimate the correlation between the time to first recurrence and the time to the second recurrence. (B) All individuals in the Thailand-Myanmar data who were treated with artesunate are grouped by their time to first recurrence quartiles, from shortest (green) to longest (red). (C, D) As the recurrence times are correlated, we find that individuals with a shorter time to the first recurrence (green) also have a shorter time from first to second recurrence (the data are shown in bolder lines and darker colors). (C) The temporal heterogeneity model cannot capture this feature in the data. The simulations show that all individuals have a similar time from first to second recurrence, regardless of the time to the first recurrence (simulated data are shown in thinner lines and lighter colors). (D) In the data simulated under the population heterogeneity model, however, the first recurrence time is predictive of the second recurrence time and this correlation agrees well with the data. The simulated data for chloroquine treatment compared with the Thailand-Myanmar data are shown in Fig L in S1 Figs.

Fig 5. Time to infection population heterogeneity model fit and relapse rate vs infection rate for different villages in the Papua New Guinea data.

(A-E) Model fit of the population heterogeneity model to the time to infection data from Papua New Guinea stratified by village. All villages were fit simultaneously with the same drug washout time distribution, the rate of new infections and relapses was allowed to vary between villages. The lines indicate the model fit and the shaded area the 95% confidence region from the data. (F) Relapse rate and infection rate for different villages. For each village, the median relapse rate (dot) and interquartile range (vertical line) of the relapse rate distribution from the population heterogeneity model fit is plotted against the infection rate. The Pearson and Spearman correlation between the log-transformed median relapse rate and the infection rate are 0.97 and 0.9, respectively, with p-values of 0.0075 and 0.083, respectively. For model fits to the Papua New Guinea data by village using the other models see Figs C and D in S1 Figs.