Skip to contents

DSM265 malaria chemoprophylaxis PK/PD (Cherkaoui-Rbati 2023)

Source: vignettes/articles/CherkaouiRbati_2023_dsm265.Rmd
CherkaouiRbati_2023_dsm265.Rmd

Model and source

  • Citation: Cherkaoui-Rbati MH, Andenmatten N, Burgert L, Egbelowo OF, Fendel R, Fornari C, Gabel M, Ward J, Moehrle JJ, Gobeau N. A pharmacokinetic-pharmacodynamic model for chemoprotective agents against malaria. CPT Pharmacometrics Syst Pharmacol. 2023;12(1):50-61. doi:10.1002/psp4.12875.
  • Description: Semi-mechanistic chemoprophylaxis PK/PD (QSP) model of the antimalarial dihydroorotate-dehydrogenase (DHODH) inhibitor DSM265 against Plasmodium falciparum (Cherkaoui-Rbati 2023). The model mimics the parasite life cycle in two coupled pools: a liver (hepatic-schizont) stage and a blood stage, each growing exponentially and killed by an Emax/EC50 Hill turnover term driven by the DSM265 dried-blood-spot concentration. Merozoites are released from liver to blood through a sigmoidal maturation window centred at T50 ~ 6 days. PK is a two-compartment model with zero-order absorption, a lag time, linear elimination, and a dose-level effect on relative bioavailability; individual PK parameters feed the PD layer as drivers. Parameters were estimated sequentially (induced-blood-stage malaria, sporozoite human challenge, and published placebo-challenge studies) and assembled into the final chemoprophylaxis model. Liver- and blood-stage growth rates and EC50 are assumed 100% correlated within an individual, and liver Emax, Hill and turnover are fixed to the blood-stage values (not separately identifiable). Reproduces the Day-28 protection success rates in Cherkaoui-Rbati 2023 Figure 3.
  • Article: https://doi.org/10.1002/psp4.12875
  • Supplement (Monolix project files for each estimation step, parameter Tables S1-S3): https://doi.org/10.1002/psp4.12875

This is a semi-mechanistic pharmacokinetic-pharmacodynamic (a quantitative systems-pharmacology-style) model of the antimalarial DSM265, a Plasmodium falciparum dihydroorotate-dehydrogenase (DHODH) inhibitor developed as a chemoprophylactic candidate. The model couples a two-compartment PK model for DSM265 to a two-stage parasite-dynamics model (liver / hepatic-schizont stage and blood stage), each growing exponentially and killed by a turnover-lagged Emax / EC50 Hill term driven by the drug concentration. Merozoites move from liver to blood through a sigmoidal maturation window centred at T50 ~ 6 days.

Population

Data were pooled from 15 studies comprising 190 participants (Cherkaoui-Rbati 2023, Table 1): one first-in-human single-ascending-dose study (55 subjects, 25-1200 mg), two induced-blood-stage malaria (IBSM) studies (15), a phase IIa study of uncomplicated P. falciparum malaria patients in Peru (24), two sporozoite human-challenge (spz HuCh) studies (39, including 10 placebo), and 57 placebo subjects from nine further published sporozoite-challenge studies (Coffeng et al. 2017). The chemoprophylaxis arms of the spz HuCh studies dosed 400 mg DSM265 at 1, 3, or 7 days before intravenous inoculation of 3200 sporozoites.

The same information is available programmatically via the model’s population metadata (readModelDb("CherkaouiRbati_2023_dsm265")()$population).

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/CherkaouiRbati_2023_dsm265.R. The structural equations and fixed values were confirmed against the deposited Monolix project files (supplement, Step 1-5 projects). The table below collects the key origins.

Equation / parameter Value Source location
lcl (CL) 0.476 L/h Table 2
lvc (Vc) 8.58 L Table 2
lq (Q1) 37.1 L/h Table 2
lvp (Vp,1) 57.3 L Table 2
ld1 (Tk,0) 2.85 h Table 2
ltlag (Tlag,1) 0.147 h Table 2
lfrel (F) 1 (fixed) Table 2
e_dose_frel -0.102 Table 2 (beta_F,Dose; transform log(dose/400))
lkgrow_liver (GR_L) 0.0716 1/h (fixed) Table 3
lkgrow_blood (GR_B) 0.0624 1/h (fixed) Table 3 / Table S2
lemax_liver/lemax_blood (Emax) 0.205 1/h (fixed) Table 3 / Table S1
lec50_liver (EC50,L) 1.7 ug/mL Table 3 (estimated)
lec50_blood (EC50,B) 1.25 ug/mL (fixed) Table 3 / Table S1
lhill_liver/lhill_blood (h) 10 (fixed) Table 3 / Table S1
lkin (k_in) 0.0771 1/h (fixed) Table 3 / Table S1
lklb (k_LB) 6 1/h (fixed) Table 3
lt50 (T50) 144 h (fixed) Table 3
lsigtliver (sigma_LB) 0.1 h (fixed) Table 3
lfinc (F_inc) 0.00119 (fixed) Table 3 / Table S3
lrhogr / lrhoec50 (correlation scalings) 1 (fixed) Table 3 (rho = 1)
propSd 0.179 Table 2 (proportional error)
addSd 2.66 ln(p/mL) Table 3 (additive error)
d/dt(parasite_liver), d/dt(parasite_blood) n/a Equations 1.1, 1.2
d/dt(kill_liver), d/dt(kill_blood) n/a Equations 2.1, 2.2
trlb (merozoite release) n/a Equation 3

Model conventions and the time origin

The model uses absolute time t with sporozoite infection at t = 0, which drives the merozoite-release window trlb = expit((t - T50) / sigma_LB) (T50 = 144 h). Chemoprophylactic drug doses are therefore administered at negative times relative to infection (e.g. -24 h for a Day -1 dose), matching the convention of the source data. The sporozoite inoculum is administered as a dose into parasite_liver, with bioavailability F_inc applied, so the initial liver burden is F_inc * inoculum (for 3200 sporozoites that is ~4 infected hepatocytes, Cherkaoui-Rbati 2023 Results). Mosquito-bite dosing should supply the inoculum in sporozoites (bites x 640 spz/bite).

PK validation (DSM265 disposition)

A single 400 mg oral dose with no infection reproduces the long terminal half-life (~96-100 h, Cherkaoui-Rbati 2023; “a long elimination half-life in humans of around 96 h”).

rxode2::rxSetSeed(1024)
set.seed(1024)

mod <- readModelDb("CherkaouiRbati_2023_dsm265")

# The model has two residual-error endpoints (Cc and lparasitemia_blood), so
# rxode2 requires every observation row to name one of them in `cmt`; an
# observation row on an ODE state such as "central" is rejected. Both
# observables, and every state, are still returned as columns on every row.
n_pk <- 150L
pk_obs_times <- c(seq(0, 48, by = 1), seq(52, 480, by = 4))
pk_dose <- data.frame(
  id = seq_len(n_pk), time = 0, evid = 1L, amt = 400, cmt = "central"
)
pk_obs <- tidyr::crossing(
  id = seq_len(n_pk), time = pk_obs_times
) |>
  dplyr::mutate(evid = 0L, amt = NA_real_, cmt = "Cc")
pk_events <- dplyr::bind_rows(pk_dose, pk_obs) |>
  dplyr::arrange(id, time, dplyr::desc(evid)) |>
  dplyr::mutate(treatment = "400 mg")

pk_sim <- rxode2::rxSolve(mod, events = pk_events, keep = "treatment") |>
  as.data.frame()
#> ℹ parameter labels from comments will be replaced by 'label()'
pk_sim |>
  dplyr::filter(!is.na(Cc), time > 0) |>
  dplyr::group_by(time) |>
  dplyr::summarise(
    Q05 = quantile(Cc, 0.05, na.rm = TRUE),
    Q50 = quantile(Cc, 0.50, na.rm = TRUE),
    Q95 = quantile(Cc, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  scale_y_log10() +
  labs(
    x = "Time (h)", y = "DSM265 concentration (ug/mL)",
    title = "DSM265 PK after a single 400 mg oral dose",
    caption = "Two-compartment model, Cherkaoui-Rbati 2023 Table 2."
  )

sim_nca <- pk_sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::select(id, time, Cc, treatment)

# Defensive time-zero rows (extravascular pre-dose Cc = 0).
sim_nca <- dplyr::bind_rows(
  sim_nca,
  sim_nca |> dplyr::distinct(id, treatment) |>
    dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::distinct(id, treatment, time, .keep_all = TRUE) |>
  dplyr::arrange(id, treatment, time)

conc_obj <- PKNCA::PKNCAconc(sim_nca, Cc ~ time | treatment + id,
                             concu = "ug/mL", timeu = "h")

dose_df <- pk_events |>
  dplyr::filter(evid == 1) |>
  dplyr::select(id, time, amt, treatment)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | treatment + id, doseu = "mg")

intervals <- data.frame(
  start = 0, end = Inf,
  cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE, half.life = TRUE
)
nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))

nca_summary <- as.data.frame(nca_res$result) |>
  dplyr::group_by(PPTESTCD) |>
  dplyr::summarise(median = median(PPORRES, na.rm = TRUE), .groups = "drop")
nca_summary |>
  dplyr::rename("NCA parameter" = PPTESTCD, "Median (simulated)" = median) |>
  knitr::kable(
    digits = 3,
    caption = "Simulated single-dose NCA (400 mg DSM265)."
  )
Simulated single-dose NCA (400 mg DSM265).
NCA parameter Median (simulated)
adj.r.squared 1.000
aucinf.obs 794.226
clast.obs 0.252
clast.pred 0.252
cmax 6.164
half.life 114.886
lambda.z 0.006
lambda.z.n.points 155.000
lambda.z.time.first 2.000
lambda.z.time.last 480.000
r.squared 1.000
span.ratio 4.161
tlast 480.000
tmax 1.000 
hl <- nca_res$result |>
  dplyr::filter(PPTESTCD == "half.life") |>
  dplyr::pull(PPORRES)

# Typical-value terminal half-life from the Table 2 estimates (the slower
# eigenvalue of the two-compartment system). This is deterministic, so it is
# held tightly against the "around 96 h" the paper quotes.
pk_ini <- rxode2::rxode(mod)$theta
#> ℹ parameter labels from comments will be replaced by 'label()'
k <- exp(pk_ini[["lcl"]]) / exp(pk_ini[["lvc"]])
k12 <- exp(pk_ini[["lq"]]) / exp(pk_ini[["lvc"]])
k21 <- exp(pk_ini[["lq"]]) / exp(pk_ini[["lvp"]])
s <- k + k12 + k21
beta <- (s - sqrt(s^2 - 4 * k * k21)) / 2
t_half_typical <- log(2) / beta
t_half_typical
#> [1] 96.86637

# The cohort median runs longer than the typical value: the very large
# variance on Vc (SD 1.06 on the log scale) skews (Vc + Vp) / CL upwards. The
# cohort check is therefore an envelope only.
median(hl, na.rm = TRUE)
#> [1] 114.8859
stopifnot(
  abs(t_half_typical - 96) < 3,
  median(hl, na.rm = TRUE) > 80,
  median(hl, na.rm = TRUE) < 160
)

Chemoprophylaxis: parasite dynamics

The next simulation reproduces the two-stage parasite dynamics for a single 400 mg dose given 7 days before infection: liver parasites grow and then burst near Day 6, seeding the blood stage, which either grows (treatment failure) or is suppressed (protection).

pd_obs_times <- seq(-168, 28 * 24, by = 6)
one_dose <- data.frame(id = 1, time = -168, evid = 1L, amt = 400, cmt = "central")
one_spz <- data.frame(id = 1, time = 0, evid = 1L, amt = 3200, cmt = "parasite_liver")
one_obs <- data.frame(
  id = 1, time = pd_obs_times, evid = 0L, amt = NA_real_, cmt = "lparasitemia_blood"
)
one_events <- dplyr::bind_rows(one_dose, one_spz, one_obs) |>
  dplyr::arrange(id, time, dplyr::desc(evid))

one_mod <- mod |> rxode2::ini(
  etalcl = 0, etalvc = 0, etalvp = 0, etald1 = 0, etaltlag = 0,
  etalfrel = 0, etalfinc = 0, etalrhogr = 0, etalrhoec50 = 0
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ change initial estimate of `etalcl` to `0`
#> ℹ change initial estimate of `etalvc` to `0`
#> ℹ change initial estimate of `etalvp` to `0`
#> ℹ change initial estimate of `etald1` to `0`
#> ℹ change initial estimate of `etaltlag` to `0`
#> ℹ change initial estimate of `etalfrel` to `0`
#> ℹ change initial estimate of `etalfinc` to `0`
#> ℹ change initial estimate of `etalrhogr` to `0`
#> ℹ change initial estimate of `etalrhoec50` to `0`
one_sim <- rxode2::rxSolve(one_mod, events = one_events) |>
  as.data.frame()
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etald1', 'etaltlag', 'etalfrel', 'etalfinc', 'etalrhogr', 'etalrhoec50'
#> Warning: 
#> with negative times, compartments initialize at first negative observed time
#> with positive times, compartments initialize at time zero
#> use 'rxSetIni0(FALSE)' to initialize at first observed time
#> this warning is displayed once per session
one_sim |>
  dplyr::transmute(
    time_day = time / 24,
    `Liver parasites (count)` = pmax(parasite_liver, 1e-3),
    `Blood parasitemia (p/mL)` = pmax(parasite_blood / 5000, 1e-3)
  ) |>
  tidyr::pivot_longer(-time_day) |>
  ggplot(aes(time_day, value, colour = name)) +
  geom_line() +
  scale_y_log10() +
  labs(
    x = "Time since infection (days)", y = "Parasites",
    colour = NULL,
    title = "Two-stage parasite dynamics (typical individual, 400 mg at Day -7)",
    caption = "Liver burst ~Day 6 seeds the blood stage (Cherkaoui-Rbati 2023 Fig 2a)."
  ) +
  theme(legend.position = "bottom")

Chemoprophylaxis: Day-28 protection success rates (Figure 3)

The paper’s central validation is the Day-28 protection success rate for 400 mg DSM265 given at Day -1, -3, or -7 before infection. “Success” is being malaria-free at Day 28, i.e. blood parasitemia below the 20 p/mL PCR limit of quantification.

n_arm <- 150L
arms <- tibble::tribble(
  ~arm,      ~dose_time,
  "Day -1",  -24,
  "Day -3",  -72,
  "Day -7",  -168
)

make_arm <- function(arm_label, dose_time, id_offset) {
  ids <- id_offset + seq_len(n_arm)
  drug <- data.frame(id = ids, time = dose_time, evid = 1L, amt = 400, cmt = "central")
  spz <- data.frame(id = ids, time = 0, evid = 1L, amt = 3200, cmt = "parasite_liver")
  # One observation at Day 28 (the landmark) plus a coarse grid for stability.
  obs <- tidyr::crossing(id = ids, time = c(seq(0, 28 * 24, by = 24))) |>
    dplyr::mutate(evid = 0L, amt = NA_real_, cmt = "lparasitemia_blood")
  dplyr::bind_rows(drug, spz, obs) |>
    dplyr::mutate(arm = arm_label)
}

success_events <- purrr::pmap_dfr(
  list(arms$arm, arms$dose_time, (seq_len(nrow(arms)) - 1L) * n_arm),
  make_arm
) |>
  dplyr::arrange(id, time, dplyr::desc(evid))
stopifnot(!anyDuplicated(unique(success_events[, c("id", "time", "evid")])))

success_sim <- rxode2::rxSolve(mod, events = success_events, keep = "arm") |>
  as.data.frame()
success_rate <- success_sim |>
  dplyr::filter(abs(time - 28 * 24) < 1e-6) |>
  dplyr::mutate(parasitemia = parasite_blood / 5000) |>
  dplyr::group_by(arm) |>
  dplyr::summarise(
    n = dplyr::n(),
    success_pct = 100 * mean(parasitemia < 20, na.rm = TRUE),
    .groups = "drop"
  )

published <- tibble::tribble(
  ~arm,     ~obs_pct, ~pred_pct,
  "Day -1", 100,      60,
  "Day -3", 33,       33,
  "Day -7", 39,       11
)

success_rate |>
  dplyr::left_join(published, by = "arm") |>
  dplyr::rename(
    "Arm" = arm,
    "N" = n,
    "Simulated success (%)" = success_pct,
    "Observed (%) [Fig 3]" = obs_pct,
    "Published model (%) [Fig 3]" = pred_pct
  ) |>
  knitr::kable(digits = 1, caption = "Day-28 protection success rate vs. Cherkaoui-Rbati 2023 Figure 3.")
Day-28 protection success rate vs. Cherkaoui-Rbati 2023 Figure 3.
Arm N Simulated success (%) Observed (%) [Fig 3] Published model (%) [Fig 3]
Day -1 150 62.0 100 60
Day -3 150 41.3 33 33
Day -7 150 12.7 39 11 
sr <- stats::setNames(success_rate$success_pct, success_rate$arm)
# The paper's estimated-model Day-28 success rates rise from ~11% (Day -7) to
# ~60% (Day -1). Assert each arm lands inside a generous envelope around the
# published 90% CIs, and that earlier-than-infection dosing (Day -7) protects
# far LESS than near-infection dosing (Day -1) -- a ~50-point gap the paper
# reports, which is robust to the drawn cohort. Do NOT tighten these to a single
# run's values (pattern 12 of known-vignette-failure-patterns.md).
stopifnot(
  sr[["Day -7"]] >= 0, sr[["Day -7"]] <= 45,
  sr[["Day -1"]] >= 25, sr[["Day -1"]] <= 100,
  sr[["Day -1"]] > sr[["Day -7"]]
)

The simulated success rates track the published model predictions (about 60%, 33% and 11% for Day -1, -3 and -7; Figure 3, blue) to within a few percentage points. That is within the binomial noise of a 150-subject arm, and well inside the published 90% intervals. Like the published model, the simulation under-predicts the observed rate at Day -1 (5 of 5 protected) and at Day -7 (7 of 18). The authors attribute the Day -1 gap to the continuous (rather than stochastic) description of the few parasites present early in infection. The broader pattern also holds: protection improves as the first dose moves closer to infection, because the later dose keeps drug concentrations above the protective threshold through the Day-6 liver burst. The paper notes that, even at Day -1, protection is incomplete because the MIC is the concentration at which killing equals growth (not a parasiticidal concentration) and because of interindividual PK/PD variability.

Assumptions and deviations

  • The model uses absolute time with infection fixed at t = 0; prophylactic doses are given at negative times. This matches the source data convention and keeps the merozoite-release window T50 = 144 h interpretable as 6 days after infection.
  • The liver-stage Emax, Hill coefficient, and turnover rate were fixed to the blood-stage values, and the liver/blood growth rates and EC50 were taken as 100% correlated within an individual, exactly as reported (Cherkaoui-Rbati 2023 Table 3; these liver parameters were not separately identifiable from the available data). Only the liver-stage EC50 and its variability were estimated in the final model.
  • Final parameter values were taken from Table 2 (PK) and Table 3 (PD). The deposited Monolix project files (supplement) carry the same fixed values to more digits and were used to confirm the ODE structure, the sigmoidal release term, and the dosing of the sporozoite inoculum into the liver pool; the estimated typicals in those project files are initial values, so the Table 2 / Table 3 point estimates are authoritative.
  • The liver-stage EC50 is 1.7 ug/mL in Table 3 but “1.6 ug/ml” in the Results text, which also quotes a liver-stage MIC of 1.5 ug/mL. The Table 3 value is used. By Equation 5, MIC_L = EC50_L * (GR_L / (Emax_L - GR_L))^(1/h_L) = 0.940 * EC50_L. That gives a typical MIC_L of 1.60 ug/mL with the Table 3 EC50 and 1.50 ug/mL with the text’s, so the text’s two numbers agree with each other and the table differs from both. The two values differ by 6%.
  • The PK model’s individual parameters are, in the source, carried into the PD layer as subject-level regressors. Here they are represented by the Table 2 population PK model with its interindividual variability, so the packaged model is self-contained and can simulate PK and PD jointly.
  • The sporozoite inoculum is dosed into parasite_liver in sporozoites, with F_inc as its bioavailability. Mosquito-bite dosing (640 sporozoites per bite) is supported by supplying the inoculum already converted to sporozoites.
  • The additive residual error (2.66 on the natural-log scale of parasitemia) is large because most parasitemia samples were below the limit of quantification and were handled by the M3 likelihood method; it is reported faithfully here.
  • No correction or erratum notice for this article was found as of 2026-10-11. ```