Skip to contents

Model and source

  • Citation: Feldmuller M, Jodl SJ, Ploeger B, Wagenfeld A, Wiesinger H, Zollmann FS, Klein S, Zhang R, Rohde B, Hochel J. Zabedosertib, a novel interleukin-1 receptor-associated kinase-4 inhibitor, shows a favorable pharmacokinetic and safety profile across multiple phase 1 studies. Front Pharmacol. 2025;16:1521505. doi:10.3389/fphar.2025.1521505. PMCID PMC12163019. Model structure from Supplementary Section 1.5 (Figure S1 and text); parameter estimates from Supplementary Table S4; target-occupancy inputs from Section 2.11, Section 3.5 and Supplementary Section 1.6.
  • Description: One-compartment population PK model for oral zabedosertib (BAY 1834845, an IRAK4 inhibitor) in healthy adult men (Feldmuller 2025). First-order absorption through one transit compartment (depot -> transit1 -> central, both steps at rate ka), a dose-dependent relative bioavailability that falls exponentially with the administered dose towards a fixed floor of 1%, linear elimination of UNBOUND drug, and capacity-limited binding of unbound drug to a high-affinity plasma binding partner (‘AGP’) in the central compartment with explicit association / dissociation kinetics (Kd fixed to the in vitro AGP affinity, 0.4 uM; dissociation rate and binding capacity estimated). The observed total plasma concentration is unbound plus bound. A target-occupancy sub-model links the model-predicted unbound plasma concentration to IRAK4 binding (Kd taken as the in vitro whole-blood IL-6 IC50, 86 nM; target residence time 5 min). IIV on unbound CL, unbound V, ka and binding capacity; combined proportional plus study-specific additive residual error.
  • Article: https://doi.org/10.3389/fphar.2025.1521505 (open access; the population PK model is in Supplementary Section 1.5)

Zabedosertib (BAY 1834845) is an oral interleukin-1 receptor-associated kinase-4 (IRAK4) inhibitor. Feldmuller 2025 report three phase 1 studies in healthy men. The population PK analysis pooled the single-ascending-dose (SAD, NCT03054402) and multiple-ascending-dose (MAD, NCT03493269) studies and explained two nonlinearities: exposure that rises less than proportionally with dose (a dose-dependent relative bioavailability), and less accumulation than the terminal half-life predicts (capacity-limited binding to a high-affinity plasma binding partner with slow dissociation, assumed to be alpha-1-acid glycoprotein, AGP). The paper then projected IRAK4 target occupancy from the model-predicted unbound concentration.

Population

The model was fitted to 2500 plasma concentrations from 77 healthy men (Supplementary Table S3): 45 SAD participants who took single tablet doses of 15-480 mg fasted, and 32 MAD participants who took 45 or 120 mg once daily (QD) or 60, 120 or 200 mg twice daily (BID) for 10 days after a light meal, plus a single-dose period. SAD data after the liquid formulation or in the fed state were excluded. Participants were 18-50 years old (mean 38 and 35.8 years in the SAD and MAD studies), with BMI 19.4-30.1 kg/m^2, and almost all White (Table 1). No body weight is reported and no covariate analysis was done, so the model has no demographic covariates.

The same information is available programmatically via readModelDb("Feldmuller_2025_zabedosertib")()$population.

Source trace

Equation / parameter Value Source location
Structure: depot -> transit -> unbound central; unbound + binding partner <-> bound, all in the central volume V_unb; elimination from the unbound pool n/a Supplementary Figure S1 and Section 1.5 text
F1 = fmin + (1 - fmin) * exp(-log(2) / d50 * DOSE) n/a Supplementary Section 1.5
kon = koff / Kd n/a Supplementary Section 1.5 (“estimated by dividing the koff by the Kd”)
lcl (Cl_unb) 15.4 L/h Table S4
lvc (V_unb) 7.84 L Table S4
lka (ka) 1.2 1/h Table S4
lkd_pb (Kd) 0.4 uM, fixed Table S4 (fixed to the in vitro AGP affinity)
lkoff_pb (koff) 0.058 1/h Table S4
lbmax_pb (‘AGP’) 14.9 uM Table S4
ld50_fdepot (d50) 204 mg Table S4
fmin_fdepot (fmin) 0.01, fixed Table S4
etalcl, etalvc, etalka, etalbmax_pb 0.293, 0.071, 0.39, 0.124 Table S4 (omega^2)
propSd sqrt(0.037) = 0.192 Table S4 (sigma^2 prop)
addSd, addSd_mad sqrt(0.25), sqrt(2.5) ug/L, fixed Table S4 (sigma^2 add,SAD and add,MAD)
lkd_irak4 0.086 uM, fixed Section 3.5 (IL-6 release IC50 86 nM, taken as the IRAK4 Kd)
lkoff_irak4 12 1/h, fixed Section 2.11 (target residence time about 5 min)
Target binding dTO/dt = kon * Cu * (1 - TO) - koff * TO n/a Section 2.11 (second-order binding, first-order dissociation, drug in excess)
Molecular weight 470.5 g/mol Not in the paper; PubChem CID 121364961 (C20H21F3N4O4S)

Virtual cohorts

All participants were healthy men and the model has no demographic covariates, so the virtual cohorts differ only in their dosing. Each dose group has 100 virtual participants. STUDY_NCT03493269 only selects the additive residual error and does not affect the predictions below.

mod <- readModelDb("Feldmuller_2025_zabedosertib")

# SAD study: sampling times of Supplementary Section 1.3.2 plus a regular grid
sad_doses <- c(15, 30, 60, 120, 240, 480)
sad_obs <- sort(unique(c(
  0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 5, 6, 8, 12, 24, 36, 48,
  seq(0, 336, by = 4)
)))
n_arm <- 100L

make_sad <- function(dose, id_offset) {
  ids <- id_offset + seq_len(n_arm)
  dose_rows <- tibble(id = ids, time = 0, amt = dose, evid = 1L, cmt = "depot")
  obs_rows <- tidyr::expand_grid(id = ids, time = sad_obs) |>
    mutate(amt = 0, evid = 0L, cmt = "central")
  bind_rows(dose_rows, obs_rows) |>
    mutate(treatment = paste(dose, "mg"), STUDY_NCT03493269 = 0L)
}

sad_events <- bind_rows(lapply(seq_along(sad_doses), function(i) {
  make_sad(sad_doses[i], id_offset = (i - 1L) * n_arm)
})) |>
  arrange(id, time, desc(evid))
stopifnot(!anyDuplicated(sad_events[, c("id", "time", "evid")]))

# MAD study: 10 days; on day 10 the BID groups took only the morning dose
mad_regimens <- tibble::tribble(
  ~treatment,   ~dose, ~ii,
  "45 mg QD",     45,  24,
  "120 mg QD",   120,  24,
  "60 mg BID",    60,  12,
  "120 mg BID",  120,  12,
  "200 mg BID",  200,  12
)
mad_obs <- sort(unique(c(seq(0, 360, by = 1), seq(216, 240, by = 0.5))))

make_mad <- function(treatment, dose, ii, id_offset) {
  ids <- id_offset + seq_len(n_arm)
  dose_times <- c(seq(0, 216 - ii, by = ii), 216)
  dose_rows <- tidyr::expand_grid(id = ids, time = dose_times) |>
    mutate(amt = dose, evid = 1L, cmt = "depot")
  obs_rows <- tidyr::expand_grid(id = ids, time = mad_obs) |>
    mutate(amt = 0, evid = 0L, cmt = "central")
  bind_rows(dose_rows, obs_rows) |>
    mutate(treatment = treatment, STUDY_NCT03493269 = 1L)
}

mad_events <- bind_rows(lapply(seq_len(nrow(mad_regimens)), function(i) {
  make_mad(
    mad_regimens$treatment[i], mad_regimens$dose[i], mad_regimens$ii[i],
    id_offset = 1000L + (i - 1L) * n_arm
  )
})) |>
  arrange(id, time, desc(evid))
stopifnot(!anyDuplicated(mad_events[, c("id", "time", "evid")]))

Simulation

rxode2::rxSetSeed(20250530)
sim_sad <- rxode2::rxSolve(
  mod, events = sad_events, keep = "treatment", returnType = "data.frame"
)
#> ℹ parameter labels from comments will be replaced by 'label()'
rxode2::rxSetSeed(20250531)
sim_mad <- rxode2::rxSolve(
  mod, events = mad_events, keep = "treatment", returnType = "data.frame"
)

dose_levels <- paste(sad_doses, "mg")
sim_sad$treatment <- factor(sim_sad$treatment, levels = dose_levels)
sim_mad$treatment <- factor(sim_mad$treatment, levels = mad_regimens$treatment)

Replicate published figures

Single-dose profiles (Figure 2 and Supplementary Figure S2A)

sim_sad |>
  filter(time >= 0.5) |>
  group_by(treatment, time) |>
  summarise(
    Q05 = quantile(Cc, 0.05), Q50 = median(Cc), Q95 = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  facet_wrap(~treatment, nrow = 1) +
  scale_y_log10(limits = c(1e-3, 20)) +
  labs(
    x = "Time after dose (h)", y = "Total zabedosertib (mg/L)",
    title = "Single fasted tablet doses: median and 90% prediction interval",
    caption = "Replicates Figure 2B and Supplementary Figure S2A of Feldmuller 2025."
  )
#> Warning: Removed 234 rows containing missing values or values outside the scale range
#> (`geom_ribbon()`).
#> Warning: Removed 70 rows containing missing values or values outside the scale range
#> (`geom_line()`).

The simulated profiles rise to a broad peak in the first hours and then fall with a terminal half-life of about a day, as in Figure 2. Peak concentrations crowd together above 120 mg: the bioavailability falls with dose and the binding partner saturates.

Multiple-dose profiles (Supplementary Figure S2B and S2C)

sim_mad |>
  filter(time >= 0.5) |>
  group_by(treatment, time) |>
  summarise(
    Q05 = quantile(Cc, 0.05), Q50 = median(Cc), Q95 = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  facet_wrap(~treatment, nrow = 2) +
  scale_y_log10(limits = c(1e-2, 20)) +
  labs(
    x = "Time after first dose (h)", y = "Total zabedosertib (mg/L)",
    title = "10-day dosing: median and 90% prediction interval",
    caption = "Replicates Supplementary Figure S2B-C of Feldmuller 2025 (the paper shows four of the five MAD groups)."
  )
#> Warning: Removed 54 rows containing missing values or values outside the scale range
#> (`geom_ribbon()`).

Dose proportionality (Figure 3 and Supplementary Figure S3A)

sad_auc_obs <- tibble::tribble(
  ~dose, ~auc,
  15,  32.1,
  30,  69.5,
  60,  107,
  120, 119,
  240, 182,
  480, 188
)

sim_sad |>
  group_by(id, treatment) |>
  summarise(auc = sum(diff(time) * (head(Cc, -1) + tail(Cc, -1)) / 2), .groups = "drop") |>
  mutate(dose = as.numeric(sub(" mg", "", treatment))) |>
  group_by(dose) |>
  summarise(gm = exp(mean(log(auc / dose))), .groups = "drop") |>
  ggplot(aes(dose, gm)) +
  geom_line() +
  geom_point() +
  geom_point(data = sad_auc_obs, aes(dose, auc / dose), shape = 4, size = 3, colour = "red") +
  scale_x_log10() +
  labs(
    x = "Dose (mg)", y = "Dose-normalised AUC (h/L)",
    title = "Geometric mean dose-normalised AUC",
    caption = paste(
      "Line: simulated AUC(0-336 h); red crosses: Table 3 AUC.",
      "Replicates Figure 3B and Supplementary Figure S3A of Feldmuller 2025."
    )
  )

PKNCA validation

Single dose (Table 3)

sad_nca_conc <- sim_sad |>
  filter(!is.na(Cc)) |>
  select(id, time, Cc, treatment)
sad_nca_conc <- bind_rows(
  sad_nca_conc,
  sad_nca_conc |> distinct(id, treatment) |> mutate(time = 0, Cc = 0)
) |>
  distinct(id, treatment, time, .keep_all = TRUE) |>
  arrange(id, treatment, time)

sad_dose <- sad_events |>
  filter(evid == 1) |>
  select(id, time, amt, treatment)

sad_intervals <- data.frame(
  start = 0, end = Inf,
  cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE, half.life = TRUE
)

sad_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(sad_nca_conc, Cc ~ time | treatment + id),
  PKNCA::PKNCAdose(sad_dose, amt ~ time | treatment + id),
  intervals = sad_intervals
))

# Table 3: geometric means (tmax: median) after fasted tablet doses, SAD study
sad_published <- tibble::tribble(
  ~treatment, ~cmax, ~tmax, ~aucinf.obs, ~half.life,
  "15 mg",    1.13,  2.00,   32.1,       24.0,
  "30 mg",    1.92,  3.50,   69.5,       26.5,
  "60 mg",    2.91,  3.00,  107,         26.7,
  "120 mg",   3.30,  3.25,  119,         24.4,
  "240 mg",   4.43,  2.50,  182,         24.3,
  "480 mg",   4.65,  1.50,  188,         22.8
)

nlmixr2lib::ncaComparisonTable(
  simulated = sad_nca,
  reference = sad_published,
  by = "treatment",
  units = c(cmax = "mg/L", tmax = "h", aucinf.obs = "mg*h/L", half.life = "h"),
  tolerance_pct = 20
) |>
  dplyr::rename("Dose" = treatment) |>
  knitr::kable(caption = "Simulated (median) vs Table 3 (geometric mean; tmax median). * differs by >20%.")
Simulated (median) vs Table 3 (geometric mean; tmax median). * differs by >20%.
NCA parameter Dose Reference Simulated % diff
Cmax (mg/L) 15 mg 1.13 0.847 -25.1%*
Cmax (mg/L) 30 mg 1.92 1.51 -21.6%*
Cmax (mg/L) 60 mg 2.91 2.67 -8.1%
Cmax (mg/L) 120 mg 3.3 3.9 +18.2%
Cmax (mg/L) 240 mg 4.43 5.13 +15.7%
Cmax (mg/L) 480 mg 4.65 4.69 +0.8%
Tmax (h) 15 mg 2 4 +100.0%*
Tmax (h) 30 mg 3.5 4 +14.3%
Tmax (h) 60 mg 3 3.5 +16.7%
Tmax (h) 120 mg 3.25 3.5 +7.7%
Tmax (h) 240 mg 2.5 3 +20.0%
Tmax (h) 480 mg 1.5 3 +100.0%*
AUC0-∞ (obs) (mg*h/L) 15 mg 32.1 34 +5.8%
AUC0-∞ (obs) (mg*h/L) 30 mg 69.5 65.3 -6.0%
AUC0-∞ (obs) (mg*h/L) 60 mg 107 103 -3.6%
AUC0-∞ (obs) (mg*h/L) 120 mg 119 135 +13.8%
AUC0-∞ (obs) (mg*h/L) 240 mg 182 154 -15.1%
AUC0-∞ (obs) (mg*h/L) 480 mg 188 155 -17.7%
t½ (h) 15 mg 24 25.5 +6.4%
t½ (h) 30 mg 26.5 27.4 +3.6%
t½ (h) 60 mg 26.7 26 -2.7%
t½ (h) 120 mg 24.4 25.1 +3.1%
t½ (h) 240 mg 24.3 25 +3.0%
t½ (h) 480 mg 22.8 25.1 +9.9%
sad_gm <- as.data.frame(sad_nca) |>
  filter(PPTESTCD %in% c("aucinf.obs", "half.life")) |>
  group_by(treatment, PPTESTCD) |>
  summarise(gm = exp(mean(log(PPORRES))), .groups = "drop") |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = gm) |>
  inner_join(sad_published |> select(treatment, auc_pub = aucinf.obs, hl_pub = half.life),
    by = "treatment"
  ) |>
  mutate(
    auc_pct = 100 * (aucinf.obs / auc_pub - 1),
    hl_pct = 100 * (half.life / hl_pub - 1)
  )
knitr::kable(sad_gm, digits = 1, caption = "Geometric-mean AUC and half-life against Table 3.")
Geometric-mean AUC and half-life against Table 3.
treatment aucinf.obs half.life auc_pub hl_pub auc_pct hl_pct
120 mg 137.4 26.2 119.0 24.4 15.5 7.2
15 mg 34.6 26.6 32.1 24.0 7.7 10.7
240 mg 155.1 25.3 182.0 24.3 -14.8 3.9
30 mg 61.1 27.8 69.5 26.5 -12.1 4.8
480 mg 155.6 26.4 188.0 22.8 -17.2 15.8
60 mg 100.5 26.8 107.0 26.7 -6.1 0.6

stopifnot(
  # A mis-transcribed clearance, d50 or binding constant shifts every dose
  # group by tens of percent.
  abs(median(sad_gm$auc_pct)) < 15,
  max(abs(sad_gm$auc_pct)) < 35,
  abs(median(sad_gm$hl_pct)) < 20
)

The model reproduces the single-dose exposure across the 32-fold dose range, including the plateau of AUC above 240 mg. Simulated Cmax is lower than observed at the two lowest doses and higher at 120-240 mg; the paper notes that the profiles stay near a plateau from 2-3 h to about 6 h after dosing, which a first-order absorption model with one transit step reproduces only approximately, so Cmax and tmax are the least well described quantities.

Multiple dose, day 10 (Supplementary Section 4.7)

mad_nca_conc <- sim_mad |>
  filter(!is.na(Cc)) |>
  select(id, time, Cc, treatment)

mad_dose <- mad_events |>
  filter(evid == 1) |>
  select(id, time, amt, treatment)

mad_intervals <- mad_regimens |>
  transmute(
    treatment = factor(treatment, levels = mad_regimens$treatment),
    start = 216, end = 216 + ii,
    cmax = TRUE, auclast = TRUE
  ) |>
  as.data.frame()

mad_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(mad_nca_conc, Cc ~ time | treatment + id),
  PKNCA::PKNCAdose(mad_dose, amt ~ time | treatment + id),
  intervals = mad_intervals
))

# Supplementary 4.7: total zabedosertib on day 10 (geometric means); AUC over
# the dosing interval, AUC(0-24)md for QD and AUC(0-12)md for BID
mad_published <- tibble::tribble(
  ~treatment,   ~cmax, ~auclast,
  "45 mg QD",    2.81,  47.4,
  "120 mg QD",   4.18,  72.0,
  "60 mg BID",   4.29,  46.5,
  "120 mg BID",  6.01,  62.2,
  "200 mg BID",  6.19,  63.5
)

nlmixr2lib::ncaComparisonTable(
  simulated = mad_nca,
  reference = mad_published,
  by = "treatment",
  units = c(cmax = "mg/L", auclast = "mg*h/L"),
  tolerance_pct = 20
) |>
  dplyr::rename("Regimen" = treatment) |>
  knitr::kable(caption = "Simulated (median) vs Supplementary 4.7 (geometric mean), day 10 dosing interval. * differs by >20%.")
Simulated (median) vs Supplementary 4.7 (geometric mean), day 10 dosing interval. * differs by >20%.
NCA parameter Regimen Reference Simulated % diff
Cmax (mg/L) 45 mg QD 2.81 3 +6.6%
Cmax (mg/L) 120 mg QD 4.18 6.42 +53.6%*
Cmax (mg/L) 60 mg BID 4.29 4.71 +9.9%
Cmax (mg/L) 120 mg BID 6.01 6.45 +7.3%
Cmax (mg/L) 200 mg BID 6.19 7.09 +14.6%
AUClast (mg*h/L) 45 mg QD 47.4 52.7 +11.1%
AUClast (mg*h/L) 120 mg QD 72 100 +39.3%*
AUClast (mg*h/L) 60 mg BID 46.5 45.6 -1.9%
AUClast (mg*h/L) 120 mg BID 62.2 63 +1.3%
AUClast (mg*h/L) 200 mg BID 63.5 66.5 +4.8%
mad_gm <- as.data.frame(mad_nca) |>
  filter(PPTESTCD %in% c("cmax", "auclast")) |>
  group_by(treatment, PPTESTCD) |>
  summarise(gm = exp(mean(log(PPORRES))), .groups = "drop") |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = gm) |>
  mutate(treatment = as.character(treatment)) |>
  inner_join(
    mad_published |> select(treatment, cmax_pub = cmax, auc_pub = auclast),
    by = "treatment"
  ) |>
  mutate(
    cmax_pct = 100 * (cmax / cmax_pub - 1),
    auc_pct = 100 * (auclast / auc_pub - 1)
  )
knitr::kable(mad_gm, digits = 2, caption = "Geometric-mean day-10 Cmax and AUCtau against Supplementary 4.7.")
Geometric-mean day-10 Cmax and AUCtau against Supplementary 4.7.
treatment auclast cmax cmax_pub auc_pub cmax_pct auc_pct
120 mg BID 61.38 6.42 6.01 62.2 6.86 -1.32
120 mg QD 95.57 5.96 4.18 72.0 42.51 32.73
200 mg BID 64.84 7.15 6.19 63.5 15.58 2.11
45 mg QD 53.95 3.13 2.81 47.4 11.23 13.82
60 mg BID 46.44 4.78 4.29 46.5 11.52 -0.13

stopifnot(
  abs(median(mad_gm$auc_pct)) < 30,
  abs(median(mad_gm$cmax_pct)) < 30
)

The MAD participants took zabedosertib after a light meal, which raises Cmax by about 1.4-fold without changing AUC (Table 6), whereas the model has no food effect; the paper describes the model as one for fasted tablets. The simulated day-10 AUC and Cmax are within about 15% of the observed geometric means for four of the five regimens. The exception is 120 mg QD, where the model is about a third higher than the observed AUC and about 40% higher than the observed Cmax. The observed 120 mg QD group (n = 8) had a lower daily AUC than 60 mg BID (Table 5, Cav/D ratio 0.817), whereas the model, which applies the dose-dependent bioavailability to each administration, predicts similar daily exposure for the two regimens.

Target occupancy (Figure 8)

The paper projected steady-state IRAK4 occupancy at trough from the unbound plasma concentration predicted by the population PK model. Here each regimen is given for 20 days so that the trough is at steady state.

to_grid <- tidyr::expand_grid(
  daily = c(1, 2.5, 5, 10, 15, 30, 60, 120, 240),
  regimen = c("QD", "BID")
) |>
  mutate(ii = ifelse(regimen == "QD", 24, 12), arm = row_number())
n_to <- 100L

to_events <- bind_rows(lapply(seq_len(nrow(to_grid)), function(i) {
  g <- to_grid[i, ]
  ids <- 5000L + (i - 1L) * n_to + seq_len(n_to)
  n_dose <- 20 * 24 / g$ii
  dose_rows <- tidyr::expand_grid(id = ids, time = seq(0, by = g$ii, length.out = n_dose)) |>
    mutate(amt = g$daily * g$ii / 24, evid = 1L, cmt = "depot")
  obs_rows <- tibble(id = ids, time = n_dose * g$ii, amt = 0, evid = 0L, cmt = "central")
  bind_rows(dose_rows, obs_rows) |>
    mutate(arm = g$arm, STUDY_NCT03493269 = 1L)
})) |>
  arrange(id, time, desc(evid))

rxode2::rxSetSeed(20250601)
sim_to <- rxode2::rxSolve(mod, events = to_events, keep = "arm", returnType = "data.frame")

to_summary <- sim_to |>
  group_by(arm) |>
  summarise(occ = median(occ_irak4), .groups = "drop") |>
  left_join(to_grid, by = "arm")

ggplot(to_summary, aes(daily, 100 * occ, colour = regimen)) +
  geom_line() +
  geom_point() +
  geom_hline(yintercept = c(20, 50, 80, 90), linetype = "dashed", colour = "grey50") +
  annotate("point", x = 240, y = c(79, 67), shape = 4, size = 3) +
  scale_x_log10() +
  labs(
    x = "Daily dose (mg/day)", y = "Projected target occupancy at trough (%)",
    colour = NULL,
    title = "Median steady-state IRAK4 occupancy at trough",
    caption = "Crosses: Section 3.5 values at 240 mg/day. Replicates Figure 8 of Feldmuller 2025."
  )

to_240 <- to_summary |> filter(daily == 240)
knitr::kable(
  to_240 |>
    mutate(published = ifelse(regimen == "BID", 79, 67), simulated = round(100 * occ, 1)) |>
    select(regimen, simulated, published),
  caption = "Median steady-state trough IRAK4 occupancy (%) at 240 mg/day."
)
Median steady-state trough IRAK4 occupancy (%) at 240 mg/day.
regimen simulated published
QD 50.2 67
BID 69.8 79

stopifnot(
  # The projection runs about 10-17 percentage points below the paper (see
  # Assumptions and deviations); these bounds pin the model's own behaviour.
  with(to_240, occ[regimen == "BID"] > occ[regimen == "QD"]),
  with(to_240, occ[regimen == "BID"] > 0.55 & occ[regimen == "BID"] < 0.85)
)

The model reproduces the shape of Figure 8 – occupancy rising over two orders of magnitude of dose and levelling off above 120 mg/day, with BID dosing above QD at every daily dose – but sits about 10-17 percentage points below it (about 70% vs 79% for 120 mg BID, about 50% vs 67% for 240 mg QD). Expressed as occupancy odds, the gap is a nearly constant factor of 1.6-2.0 across daily doses from 15 to 240 mg/day, which points to a scale difference in the concentration or Kd used for the projection rather than to the PK structure (total exposures match Table 3). The paper does not give the unit conversion or the code of the projection, and Figure 8 is labelled as a preliminary analysis, so the difference is recorded here rather than tuned.

Assumptions and deviations

  • Transit step rate. Supplementary Figure S1 shows one transit compartment between the depot and the central compartment, but Table S4 lists a single absorption rate constant. Both steps run at ka, the only reading under which Table S4 is complete.
  • Volume of the bound pool. Figure S1 draws the unbound drug, the binding partner and the bound drug inside one central compartment labelled V_unb, so bound and unbound drug share the volume V_unb and the observed total concentration is (unbound + bound) / V_unb.
  • Molecular weight. Kd and the binding capacity are in uM while doses are in mg. The paper does not print the molecular weight; 470.5 g/mol is the PubChem value for zabedosertib (CID 121364961).
  • Dose in the bioavailability equation. DOSE is taken as the amount of each administration (podo(depot)), so 120 mg BID uses F(120 mg).
  • Additive residual error units. Table S4 gives the fixed additive variances (0.25 SAD, 2.5 MAD) without units. They are read as ug/L, the bioanalytical unit, which matches the lower limits of quantitation of the two studies (1.0 and 10.0 ug/L, Supplementary Table S2), and converted to mg/L. Both are negligible against the observed concentrations.
  • Food. The MAD study was dosed after a light meal; the model has no food covariate (the paper describes it as a model for fasted tablets).
  • Unbound concentration. The model’s unbound concentration Cu is a model state, not the unbound concentrations in Table 4, which the paper derived from total concentrations with an equilibrium binding formula (Supplementary Section 1.4.2). With a dissociation half-life of about 12 h the model’s unbound concentration is far from binding equilibrium after each dose, so the two are not comparable.
  • Target-occupancy projection. The target Kd is the whole-blood IL-6 release IC50 (86 nM) and the dissociation rate constant is the reciprocal of the 5-minute target residence time; the target is assumed to be in excess drug, as stated in Section 2.11. The projection does not reproduce the level of Figure 8 (see above).
  • No correction notice linked to this article was found in Europe PMC as of 2026-11-04.