Skip to contents

Model and source

Iruplinalkib (WX-0593) is a selective oral ALK/ROS1 tyrosine kinase inhibitor. Yang 2025 (Clin Transl Sci e70287; called “Yang 2025b” here to separate it from the same group’s four-study population PK analysis, Yang_2025_iruplinalkib) describes how the recommended phase 2 dose of 180 mg once daily was chosen. It combines three model layers, each packaged as its own model file:

Model Layer Fitted to
Yang_2025b_iruplinalkib_mouse_ros1 Mouse PK + modified Simeoni tumor growth inhibition ROS1-positive xenografts
Yang_2025b_iruplinalkib_mouse_alk Same structure, ALK-specific tumor parameters ALK-positive xenografts
Yang_2025b_iruplinalkib_human Human PK projected from animal data No human data (Wajima scaling)
Yang_2025b_iruplinalkib Clinical population PK WX-0593-001 patients + WX-0593-002 healthy volunteers

The mouse tumor models give a tumor static concentration (TSC), which is converted to a human TSC through plasma protein binding. Each human PK model is then used to find the dose whose steady-state trough clears that TSC.

#> ℹ parameter labels from comments will be replaced by 'label()'
Model Description
Yang_2025b_iruplinalkib_mouse_ros1 Preclinical (mouse; female BALB/c-nu bearing subcutaneous SLC34A2-ROS1 Ba/F3 cell-line-derived xenografts). Iruplinalkib PK-PD tumor growth inhibition model: a two-compartment PK model with first-order oral absorption, fitted first to the plasma data of the same mice and then held fixed, drives a modified Simeoni model in which exponential-then- linear tumor growth (switch sharpness psi = 20) is opposed by an Emax kill of proliferating cells (Kmax, KC50) followed by three damaged-cell transit compartments. Typical-value model.
Yang_2025b_iruplinalkib_mouse_alk Preclinical (mouse; female BALB/c-nu bearing subcutaneous ALK-addicted NCI-H3122 cell-line-derived xenografts). Iruplinalkib PK-PD tumor growth inhibition model: a two-compartment PK model with first-order oral absorption, fitted to plasma data from the companion ROS1-xenograft mice and held fixed (no PK was sampled in the ALK-xenograft mice), drives a modified Simeoni model in which exponential-then- linear tumor growth (switch sharpness psi = 20) is opposed by an Emax kill of proliferating cells (Kmax, KC50) followed by three damaged-cell transit compartments. Typical-value model.
Yang_2025b_iruplinalkib_human Animal-scaled human projection (not fitted to human data). Two-compartment PK model with first-order absorption and first-order elimination for oral iruplinalkib (WX-0593, a selective ALK/ROS1 tyrosine kinase inhibitor), built before the first clinical data were available to predict the human effective dose. Mouse, rat, dog and monkey intravenous profiles were superimposed with Wajima’s Css-MRT normalisation and rescaled to a 60 kg human using predicted Vss (7.9 L/kg) and CL (311 mL/min) averaged over several interspecies methods; a two-compartment model was fitted to the rescaled profile and combined with the mean animal oral bioavailability (59.2%) and the monkey Ka (0.341 1/h) to give apparent oral parameters. Typical-value model with no variability.
Yang_2025b_iruplinalkib Two-compartment population PK model with first-order absorption and first-order elimination for oral iruplinalkib (WX-0593, a selective ALK/ROS1 tyrosine kinase inhibitor), fitted to the early clinical data used to select the recommended phase 2 dose: the phase 1 dose-escalation and expansion trial WX-0593-001 in patients with ALK- or ROS1-positive advanced non-small-cell lung cancer and the single-dose 120 mg food-effect study WX-0593-002 in healthy volunteers. Apparent oral clearance carries power effects of baseline body weight and serum albumin, and the apparent central volume a power effect of baseline body weight. Dosing in the fed state multiplies the absorption rate constant by 0.796 and adds a 0.479 h absorption lag. Residual error is proportional, with a separate magnitude for each of the two studies.

Population

Mouse. Female BALB/c-nu mice were implanted subcutaneously with SLC34A2-ROS1-overexpressing Ba/F3 cells or ALK-addicted NCI-H3122 cells. Once tumors reached 100-300 mm^3 the mice (n = 9 per group) received oral iruplinalkib once daily: 0, 5, 10 or 20 mg/kg for 14 days (ROS1) or 0, 2.5, 5 or 10 mg/kg for 21 days (ALK). Plasma was sampled before dosing and 1, 2, 4, 8 and 24 h after the final Day-14 dose in the ROS1 mice only (Methods “Mouse Xenograft Study” and “Mouse PKPD Modeling”).

Human projection. No subjects. Mouse, rat, dog and monkey PK were scaled to a 60 kg human (Methods “Prediction of Human PK Parameters” and Methods S1).

Clinical. The PopPK dataset pooled the first-in-human dose-escalation and expansion trial WX-0593-001 in Chinese patients with ALK- or ROS1-positive advanced non-small-cell lung cancer and the healthy-volunteer food-effect study WX-0593-002 (a single 120 mg dose). The paper prints no demographic table and does not give the number of subjects in the dataset. The steady-state simulations used the 54 WX-0593-001 patients dosed at 120, 180 or 240 mg once daily.

The same information is available programmatically via, e.g., readModelDb("Yang_2025b_iruplinalkib")()$population.

Source trace

All four files carry an in-file comment naming the source of every ini() value. They are collected here.

Model Parameter Value Source
mouse (both) lka, lcl, lvc, lq, lvp 0.366 1/h, 2.48 L/h/kg, 3.59 L/kg, 0.0843 L/h/kg, 1.65 L/kg (all fixed) Table S1
mouse (both) propSd 0.46 (fixed) Table S1 proportional residual 46%
mouse ROS1 ltumorExpGrowth, ltumorLinGrowth 0.00851 1/h, 4.45 mm^3/h Table 1, ROS1 column (lambda0, lambda1)
mouse ROS1 lkmax, lkc50 0.0739 1/h, 1221 ng/mL Table 1, ROS1 column
mouse ROS1 ldamageTransit, lrbase_tumor 0.168 1/h, 221 mm^3 Table 1, ROS1 column (K1, E0)
mouse ALK ltumorExpGrowth, ltumorLinGrowth 0.006515 1/h, 3.17 mm^3/h Table 1, ALK column
mouse ALK lkmax, lkc50 0.092 1/h, 1674 ng/mL Table 1, ALK column
mouse ALK ldamageTransit, lrbase_tumor 0.0524 1/h, 200 mm^3 Table 1, ALK column
mouse (both) psi 20 (fixed) Table 1 (FIX)
mouse (both) propSd_tumor_vol 0 (fixed) Not reported
mouse (both) tumor ODEs n/a Methods S1 Eq1-7; Figure 1B
human projection lka 0.341 1/h (fixed) Table 2 (FIX); Methods S1 monkey Ka
human projection lcl, lvc, lq, lvp 33.7 L/h, 379 L, 134 L/h, 239 L Table 2
clinical lka, lcl, lvc, lq, lvp 1.11 1/h, 20.5 L/h, 342 L, 14.8 L/h, 336 L Table 3
clinical ltlag 0.479 h (fed only) Table 3 ‘ALAG for fed subjects’
clinical e_fed_ka 0.796 Table 3 ‘Food on Ka’
clinical e_alb_cl, e_wt_base_cl, e_wt_base_vc 1.23, 0.713, 2.02 Table 3; footnote equations (ALB/42), (BBWT/63)
clinical etalka, etalcl, etalvc 0.513, 0.0940, 0.163 Table 3 IIV 81.9%, 31.4%, 42.1%; omega^2 = log(1 + CV^2)
clinical propSdPatient, propSdHv 0.326, 0.547 Table 3 residual errors for WX-0593-001 and -002

The modified Simeoni model of Methods S1 is

dE1dt=λ0E1[1+(λ0λ1E)ψ]1/ψ−KmaxCKC50+CE1,E=E1+E2+E3+E4,\frac{dE_1}{dt} = \frac{\lambda_0 E_1}{\left[1 + \left(\frac{\lambda_0}{\lambda_1} E\right)^{\psi}\right]^{1/\psi}} - \frac{K_{max} C}{KC_{50} + C} E_1, \qquad E = E_1 + E_2 + E_3 + E_4,

with three transit compartments at rate k1k_1 carrying damaged cells to death (E1E_1 = cycling_cells, E2E_2-E4E_4 = damaged_cells1-damaged_cells3). CC is the plasma concentration in ng/mL, and E1(0)=E0E_1(0) = E_0.

Mouse PK (Figure S1A)

The mouse PK parameters were fitted once to the ROS1-xenograft plasma data and then held fixed in both tumor fits. Both mouse files carry the same PK. The paper gives no between-animal variability for either stage, so these are typical-value simulations.

# Observation rows nominate an endpoint through dvid (1 = Cc, 2 = tumor_vol);
# cmt stays NA on observation rows, so no observable is used as a compartment.
mouse_events <- function(dose, n_doses, obs_times, dvid = 1L) {
  dplyr::bind_rows(
    data.frame(
      time = (seq_len(n_doses) - 1) * 24, amt = dose, cmt = "depot",
      evid = 1L, dvid = NA_integer_
    ),
    data.frame(
      time = obs_times, amt = NA_real_, cmt = NA_character_,
      evid = 0L, dvid = dvid
    )
  ) |>
    dplyr::arrange(time, dplyr::desc(evid))
}

# The mouse models carry no etas, so a plain solve is the typical-value
# prediction (Cc and tumor_vol are the noise-free predictions).
solve_typical <- function(mod, events) {
  as.data.frame(rxode2::rxSolve(mod, events, returnType = "data.frame"))
}

mod_ros1 <- readModelDb("Yang_2025b_iruplinalkib_mouse_ros1")
mod_alk <- readModelDb("Yang_2025b_iruplinalkib_mouse_alk")
last_dose <- 13 * 24
pk_grid <- last_dose + c(seq(0, 2, by = 0.1), seq(2.25, 24, by = 0.25))

mouse_pk <- lapply(c(5, 10, 20), function(d) {
  solve_typical(mod_ros1, mouse_events(d, 14, pk_grid)) |>
    dplyr::mutate(dose = d, tad = time - last_dose)
}) |>
  dplyr::bind_rows()

# Group means read off Figure S1A by the maintainers (ng/mL); the paper does
# not tabulate them.
s1a <- tibble::tribble(
  ~dose, ~tad, ~Cc,
  5, 2, 390, 5, 4, 195, 5, 8, 55,
  10, 2, 1230, 10, 4, 520, 10, 8, 170,
  20, 2, 1450, 20, 4, 830, 20, 8, 490
)

ggplot(mouse_pk, aes(tad, Cc, colour = factor(dose))) +
  geom_line() +
  geom_point(data = s1a, shape = 1, size = 2.5) +
  labs(
    x = "Time after the Day-14 dose (h)", y = "Iruplinalkib (ng/mL)",
    colour = "Dose (mg/kg)",
    caption = "Lines: typical-value model. Circles: Figure S1A group means (digitised)."
  )

The model sits on the observed means at 5 and 20 mg/kg (2 h peaks 355 vs 390 and 1421 vs 1450 ng/mL). At 10 mg/kg the observed peak (about 1230 ng/mL) is close to the 20 mg/kg peak, so exposure in these mice was not dose-proportional. A linear model fitted to all three doses cannot follow that, and it under-predicts the 10 mg/kg group.

A PKNCA check of the steady-state interval confirms the disposition: for a linear model at steady state, AUC over one dosing interval equals dose / (CL/F).

mouse_nca_in <- mouse_pk |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::mutate(treatment = paste(dose, "mg/kg"), time = tad, id = 1L) |>
  dplyr::select(id, treatment, time, Cc)
mouse_dose <- mouse_nca_in |>
  dplyr::distinct(id, treatment) |>
  dplyr::mutate(time = 0, amt = as.numeric(sub(" mg/kg", "", treatment)))

mouse_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(mouse_nca_in, Cc ~ time | treatment + id),
  PKNCA::PKNCAdose(mouse_dose, amt ~ time | treatment + id),
  intervals = data.frame(start = 0, end = 24, cmax = TRUE, tmax = TRUE, auclast = TRUE),
  options = list(auc.method = "linear")
))

mouse_auc <- as.data.frame(mouse_nca$result) |>
  dplyr::filter(PPTESTCD == "auclast") |>
  dplyr::mutate(
    dose = as.numeric(sub(" mg/kg", "", treatment)),
    expected = dose / 2.48 * 1000,
    pct_diff = 100 * (PPORRES - expected) / expected
  )
mouse_auc |>
  dplyr::select(treatment, PPORRES, expected, pct_diff) |>
  dplyr::rename(
    "Dose" = treatment, "Simulated AUC0-24 (ng*h/mL)" = PPORRES,
    "Dose / (CL/F) (ng*h/mL)" = expected, "% difference" = pct_diff
  ) |>
  knitr::kable(digits = 1)
Dose Simulated AUC0-24 (ng*h/mL) Dose / (CL/F) (ng*h/mL) % difference
10 mg/kg 4031.5 4032.3 0
20 mg/kg 8062.9 8064.5 0
5 mg/kg 2015.7 2016.1 0

# Same drawn parameters on both sides (typical values); the residual is
# trapezoid error on a 0.1-0.25 h grid, well under 1%.
stopifnot(all(abs(mouse_auc$pct_diff) < 1))

Mouse tumor growth inhibition (Figures S1B, S1C)

tgi_sim <- function(mod, doses, n_doses) {
  t_end <- n_doses * 24
  lapply(doses, function(d) {
    solve_typical(mod, mouse_events(d, n_doses, seq(0, t_end, by = 2), dvid = 2L)) |>
      dplyr::mutate(dose = d)
  }) |>
    dplyr::bind_rows()
}

tgi_ros1 <- tgi_sim(mod_ros1, c(0, 5, 10, 20), 14) |> dplyr::mutate(model = "ROS1")
tgi_alk <- tgi_sim(mod_alk, c(0, 2.5, 5, 10), 21) |> dplyr::mutate(model = "ALK")
tgi <- dplyr::bind_rows(tgi_ros1, tgi_alk)

# Group means read off Figures S1B and S1C by the maintainers (mm^3). Day 1 is
# the first dose, i.e. model time 0.
s1bc <- dplyr::bind_rows(
  tibble::tibble(
    model = "ROS1", day = rep(c(1, 5, 8, 12, 15), 4),
    dose = rep(c(0, 5, 10, 20), each = 5),
    tumor_vol = c(
      230, 500, 740, 1280, 1600,
      230, 440, 630, 860, 970,
      230, 280, 380, 530, 540,
      230, 140, 230, 270, 270
    )
  ),
  tibble::tibble(
    model = "ALK", day = rep(c(1, 4, 7, 10, 14, 17, 21), 4),
    dose = rep(c(0, 2.5, 5, 10), each = 7),
    tumor_vol = c(
      180, 330, 500, 780, 1010, 1220, 1600,
      180, 290, 450, 600, 690, 890, 975,
      180, 260, 370, 430, 530, 620, 730,
      180, 200, 200, 230, 240, 220, 230
    )
  )
) |>
  dplyr::mutate(time = (day - 1) * 24)

ggplot(tgi, aes(time / 24 + 1, tumor_vol, colour = factor(dose))) +
  geom_line() +
  geom_point(data = s1bc, shape = 1, size = 2.5) +
  facet_wrap(~model, scales = "free_x") +
  labs(
    x = "Study day", y = "Tumor volume (mm^3)", colour = "Dose (mg/kg)",
    caption = "Lines: typical-value model. Circles: Figures S1B/S1C group means (digitised)."
  )

The paper reports the tumor growth inhibition (TGI) range for each xenograft (46.1-97.1% ROS1, 44.1-95.8% ALK, Results) but not its formula. The usual definition, 1 - (change in treated volume) / (change in control volume) at the end of treatment, gives 46%, 97%, 44% and 96% from the digitised Figure S1 means. That agreement supports reading the reported values with this definition.

tgi_end <- tgi |>
  dplyr::group_by(model) |>
  dplyr::filter(time == max(time)) |>
  dplyr::mutate(
    v0 = ifelse(model == "ROS1", 221, 200),
    ctrl = tumor_vol[dose == 0],
    TGI = 100 * (1 - (tumor_vol - v0) / (ctrl - v0))
  ) |>
  dplyr::ungroup()

tgi_ref <- tibble::tribble(
  ~model, ~dose, ~TGI_paper,
  "ROS1", 5, 46.1, "ROS1", 20, 97.1,
  "ALK", 2.5, 44.1, "ALK", 10, 95.8
)
tgi_cmp <- tgi_end |>
  dplyr::filter(dose > 0) |>
  dplyr::left_join(tgi_ref, by = c("model", "dose")) |>
  dplyr::select(model, dose, tumor_vol, TGI, TGI_paper)
tgi_cmp |>
  dplyr::rename(
    "Xenograft" = model, "Dose (mg/kg)" = dose,
    "Simulated end volume (mm^3)" = tumor_vol, "Simulated TGI (%)" = TGI,
    "Reported TGI (%)" = TGI_paper
  ) |>
  knitr::kable(digits = 1)
Xenograft Dose (mg/kg) Simulated end volume (mm^3) Simulated TGI (%) Reported TGI (%)
ROS1 5.0 753.1 60.4 46.1
ROS1 10.0 333.8 91.6 NA
ROS1 20.0 74.5 110.9 97.1
ALK 2.5 979.6 46.2 44.1
ALK 5.0 655.1 68.6 NA
ALK 10.0 183.4 101.1 95.8

ctrl_end <- tgi_end |> dplyr::filter(dose == 0)
stopifnot(
  # Untreated growth: lambda0, lambda1, psi and E0 alone reproduce the control
  # arm (digitised end volume about 1600 mm^3 in both xenografts).
  all(abs(ctrl_end$tumor_vol / 1600 - 1) < 0.10),
  # Typical-value model, deterministic: realised gaps are 14.3, 13.8, 2.1 and
  # 5.3 percentage points. A mis-transcribed Kmax or KC50 moves these by tens
  # of points.
  all(abs(tgi_cmp$TGI[!is.na(tgi_cmp$TGI_paper)] - tgi_cmp$TGI_paper[!is.na(tgi_cmp$TGI_paper)]) < 20)
)

The ALK model matches the reported TGI within about 5 points at both ends of the dose range (46 vs 44% at 2.5 mg/kg, 101 vs 96% at 10 mg/kg). The ROS1 model gives more inhibition than the observed means at every dose (60 vs 46% at 5 mg/kg). At 20 mg/kg it shrinks the tumor (TGI above 100%), whereas the observed mean shrank to about 140 mm^3 by Day 5 and then grew back to about 270 mm^3. A Simeoni model with linear PK cannot regrow under constant dosing. The gap is plausibly the difference between a typical-value curve and a mean of animals: the fit was population-based (FOCE-ELS), but its between-animal variability is not reported, so the mean prediction cannot be rebuilt.

Tumor static concentration

At the tumor static concentration, kill balances exponential-phase growth: TSC = lambda0 * KC50 / (Kmax - lambda0). The human TSC rescales it by the unbound fraction, mouse 15% (85% bound) and human 24.3% (75.7% bound; Results and Discussion).

tsc_tab <- lapply(list(ROS1 = mod_ros1, ALK = mod_alk), function(m) {
  ini <- rxode2::rxode(m)$theta
  l0 <- exp(ini[["ltumorExpGrowth"]])
  kc50 <- exp(ini[["lkc50"]])
  kmax <- exp(ini[["lkmax"]])
  tsc <- l0 * kc50 / (kmax - l0)
  data.frame(tsc_mouse = tsc, tsc_human = tsc * 0.15 / 0.243)
}) |>
  dplyr::bind_rows(.id = "model") |>
  dplyr::mutate(
    tsc_mouse_paper = c(159, 127),
    tsc_human_paper = c(98, 78)
  ) |>
  dplyr::relocate(tsc_mouse_paper, .after = tsc_mouse)
tsc_tab |>
  dplyr::rename(
    "Xenograft" = model, "Mouse TSC (ng/mL)" = tsc_mouse,
    "Paper mouse TSC" = tsc_mouse_paper, "Human TSC (ng/mL)" = tsc_human,
    "Paper human TSC" = tsc_human_paper
  ) |>
  knitr::kable(digits = 1)
Xenograft Mouse TSC (ng/mL) Paper mouse TSC Human TSC (ng/mL) Paper human TSC
ROS1 158.9 159 98.1 98
ALK 127.6 127 78.8 78

stopifnot(
  all(abs(tsc_tab$tsc_mouse / tsc_tab$tsc_mouse_paper - 1) < 0.01),
  all(abs(tsc_tab$tsc_human / tsc_tab$tsc_human_paper - 1) < 0.01)
)

Human PK projected from animals (Figure 3C, Table S2)

Table 2 labels its parameters as apparent oral values (CL/F, V1/F, Q/F, V2/F), so the model applies no separate bioavailability. Two checks show the 59.2% is already divided in. Simulating 180 mg once daily reproduces the paper’s preclinical projection in Table S2 exactly. And the earliest points of the 1 mg/kg (60 mg) intravenous profile in Figure 3B, about 200-330 ng/mL, bracket dose / V1 = 268 ng/mL for V1 = 379 x 0.592 = 224 L; V1 = 379 L would give 158 ng/mL.

mod_hp <- readModelDb("Yang_2025b_iruplinalkib_human")
hp_events <- function(dose_mg, n_doses = 30, obs = seq(0, n_doses * 24, by = 0.25)) {
  rxode2::et(amt = dose_mg, ii = 24, addl = n_doses - 1, cmt = "depot") |>
    rxode2::et(obs, cmt = "central")
}

fig3c <- lapply(c(1, 2.4, 3), function(mgkg) {
  as.data.frame(rxode2::rxSolve(mod_hp, hp_events(mgkg * 60, 6, seq(0, 144, by = 0.5)))) |>
    dplyr::mutate(regimen = paste(mgkg, "mg/kg QD"))
}) |>
  dplyr::bind_rows()

ggplot(fig3c, aes(time, Cc * 1000, colour = regimen)) +
  geom_line() +
  geom_hline(yintercept = 98, linetype = "dashed") +
  labs(
    x = "Time (h)", y = "Iruplinalkib (ng/mL)", colour = NULL,
    caption = "Replicates Figure 3C of Yang 2025 (60 kg; dashed line = human ROS1 TSC 98 ng/mL)."
  )


hp_ss <- as.data.frame(rxode2::rxSolve(mod_hp, hp_events(180))) |>
  dplyr::filter(time >= 29 * 24)
hp_cmp <- tibble::tibble(
  quantity = c("Ctrough,ss", "Cmax,ss"),
  simulated = c(1000 * hp_ss$Cc[hp_ss$time == 30 * 24], 1000 * max(hp_ss$Cc)),
  table_s2 = c(126, 305)
)
knitr::kable(
  hp_cmp |> dplyr::rename(
    "180 mg QD" = quantity, "Simulated (ng/mL)" = simulated,
    "Table S2 preclinical (ng/mL)" = table_s2
  ),
  digits = 1
)
180 mg QD Simulated (ng/mL) Table S2 preclinical (ng/mL)
Ctrough,ss 125.9 126
Cmax,ss 305.4 305
stopifnot(all(abs(hp_cmp$simulated / hp_cmp$table_s2 - 1) < 0.03))

The paper’s projected effective dose, 2.4 mg/kg (144 mg) once daily, gives a steady-state trough just above the 98 ng/mL ROS1 TSC, as in Figure 3C.

Clinical population PK (Figure 4, Table S2)

Food effect

Food slows absorption (Ka x 0.796, plus a 0.479 h lag) but leaves CL/F unchanged. So a fed dose must give the same AUC as a fasted one, equal to dose / (CL/F) for the reference subject.

mod_clin <- readModelDb("Yang_2025b_iruplinalkib")
food_ev <- dplyr::bind_rows(
  data.frame(id = 1L, time = 0, amt = 120, cmt = "depot", evid = 1L, FED = 0),
  data.frame(id = 1L, time = c(seq(0, 12, by = 0.05), seq(12.5, 480, by = 0.5)), amt = NA, cmt = "central", evid = 0L, FED = 0),
  data.frame(id = 2L, time = 0, amt = 120, cmt = "depot", evid = 1L, FED = 1),
  data.frame(id = 2L, time = c(seq(0, 12, by = 0.05), seq(12.5, 480, by = 0.5)), amt = NA, cmt = "central", evid = 0L, FED = 1)
) |>
  dplyr::mutate(WT_BASE = 63, ALB = 42, STUDY_WX0593_002 = 1) |>
  dplyr::arrange(id, time, dplyr::desc(evid))

# zeroRe() warns that the typical-value model has no omega or sigma left;
# that is the intent here.
food_sim <- suppressWarnings(
  as.data.frame(rxode2::rxSolve(rxode2::zeroRe(mod_clin), food_ev, keep = "FED"))
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalka', 'etalcl', 'etalvc'
ggplot(dplyr::filter(food_sim, time <= 24), aes(time, Cc * 1000, colour = factor(FED, labels = c("Fasted", "Fed")))) +
  geom_line() +
  labs(x = "Time after a single 120 mg dose (h)", y = "Iruplinalkib (ng/mL)", colour = NULL)


food_nca_in <- food_sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::mutate(treatment = ifelse(FED == 1, "Fed", "Fasted"), Cc = Cc * 1000) |>
  dplyr::select(id, treatment, time, Cc)
food_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(food_nca_in, Cc ~ time | treatment + id),
  PKNCA::PKNCAdose(
    food_nca_in |> dplyr::distinct(id, treatment) |> dplyr::mutate(time = 0, amt = 120),
    amt ~ time | treatment + id
  ),
  intervals = data.frame(start = 0, end = Inf, cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE, half.life = TRUE)
))
food_res <- as.data.frame(food_nca$result) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "tmax", "aucinf.obs")) |>
  dplyr::select(treatment, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES)
food_res |>
  dplyr::rename(
    "State" = treatment, "Cmax (ng/mL)" = cmax, "Tmax (h)" = tmax,
    "AUC0-inf (ng*h/mL)" = aucinf.obs
  ) |>
  knitr::kable(digits = 1)
State Cmax (ng/mL) Tmax (h) AUC0-inf (ng*h/mL)
Fasted 275.9 2.4 5853.7
Fed 265.2 3.2 5853.8

auc_expected <- 120 / 20.5 * 1000
stopifnot(
  # Same drawn (typical) parameters on both sides: only extrapolation error.
  all(abs(food_res$aucinf.obs / auc_expected - 1) < 0.02),
  food_res$cmax[food_res$treatment == "Fed"] < food_res$cmax[food_res$treatment == "Fasted"]
)

Steady-state exposure at 120, 180 and 240 mg once daily

The paper overlaid its simulations on the 54 WX-0593-001 patients but does not report their body weight or albumin. The virtual cohort below assumes body weight log-normal around the 63 kg reference (log-scale SD 0.16) and albumin normal around the 42 g/L reference (SD 4 g/L), both truncated to plausible adult ranges. All patients are dosed fasted and carry the WX-0593-001 residual error.

rxode2::rxSetSeed(20250707)
set.seed(20250707)
n_per_arm <- 200
doses <- c(120, 180, 240)

make_arm <- function(dose, id_offset) {
  subj <- tibble::tibble(
    id = id_offset + seq_len(n_per_arm),
    WT_BASE = pmin(pmax(exp(rnorm(n_per_arm, log(63), 0.16)), 40), 95),
    ALB = pmin(pmax(rnorm(n_per_arm, 42, 4), 28), 54),
    treatment = paste(dose, "mg QD")
  )
  dose_rows <- subj |>
    tidyr::crossing(time = seq(0, 20 * 24, by = 24)) |>
    dplyr::mutate(amt = dose, cmt = "depot", evid = 1L)
  obs_rows <- subj |>
    tidyr::crossing(time = 20 * 24 + c(seq(0, 4, by = 0.25), seq(4.5, 24, by = 0.5))) |>
    dplyr::mutate(amt = NA_real_, cmt = "central", evid = 0L)
  dplyr::bind_rows(dose_rows, obs_rows)
}

clin_events <- dplyr::bind_rows(lapply(seq_along(doses), function(i) {
  make_arm(doses[i], (i - 1) * n_per_arm)
})) |>
  dplyr::mutate(FED = 0, STUDY_WX0593_002 = 0) |>
  dplyr::arrange(id, time, dplyr::desc(evid))
stopifnot(!anyDuplicated(unique(clin_events[, c("id", "time", "evid")])))

clin_sim <- as.data.frame(rxode2::rxSolve(mod_clin, clin_events, keep = "treatment")) |>
  dplyr::mutate(Cc_ng = Cc * 1000, tad = time - 20 * 24)
#> ℹ parameter labels from comments will be replaced by 'label()'
clin_sim |>
  dplyr::group_by(treatment, tad) |>
  dplyr::summarise(
    Q05 = quantile(Cc_ng, 0.05), Q50 = median(Cc_ng), Q95 = quantile(Cc_ng, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(tad, Q50, colour = treatment, fill = treatment)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.2, colour = NA) +
  geom_line(linewidth = 1) +
  scale_y_log10() +
  labs(
    x = "Time after dose at steady state (h)", y = "Iruplinalkib (ng/mL)",
    colour = NULL, fill = NULL,
    caption = "Replicates Figure 4 of Yang 2025: median and 90% prediction interval."
  )

The paper states its steady-state troughs twice. The Results give the median (10th, 90th percentile) as 151 (71.4, 293), 232 (109, 449) and 301 (142, 588) ng/mL for 120, 180 and 240 mg. Table S2 gives a 180 mg median (5th-95th) of 210 (82-444) ng/mL for the trough and 635 (389-1024) ng/mL for Cmax. These are two simulations of the same model whose cohorts are not described.

clin_nca_in <- clin_sim |>
  dplyr::filter(!is.na(Cc_ng)) |>
  dplyr::transmute(id, treatment, time = tad, Cc = Cc_ng)
clin_dose <- clin_nca_in |>
  dplyr::distinct(id, treatment) |>
  dplyr::mutate(time = 0, amt = as.numeric(sub(" mg QD", "", treatment)))
clin_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(clin_nca_in, Cc ~ time | treatment + id),
  PKNCA::PKNCAdose(clin_dose, amt ~ time | treatment + id),
  intervals = data.frame(start = 0, end = 24, cmax = TRUE, cmin = TRUE, auclast = TRUE),
  options = list(auc.method = "linear")
))

# At steady state with no lag (fasted), the interval minimum is the trough.
clin_ind <- as.data.frame(clin_nca$result) |>
  dplyr::filter(PPTESTCD %in% c("cmin", "cmax")) |>
  dplyr::select(id, treatment, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES)

clin_tab <- clin_ind |>
  dplyr::group_by(treatment) |>
  dplyr::summarise(
    trough_p10 = quantile(cmin, 0.10), trough_p50 = median(cmin),
    trough_p90 = quantile(cmin, 0.90),
    pct_above_tsc = 100 * mean(cmin > 98),
    .groups = "drop"
  ) |>
  dplyr::mutate(
    paper_p10 = c(71.4, 109, 142), paper_p50 = c(151, 232, 301),
    paper_p90 = c(293, 449, 588)
  )
clin_tab |>
  dplyr::rename(
    "Regimen" = treatment, "Sim P10" = trough_p10, "Sim median" = trough_p50,
    "Sim P90" = trough_p90, "Paper P10" = paper_p10, "Paper median" = paper_p50,
    "Paper P90" = paper_p90, "% above 98 ng/mL" = pct_above_tsc
  ) |>
  dplyr::relocate("Paper P10", .after = "Sim P10") |>
  dplyr::relocate("Paper median", .after = "Sim median") |>
  dplyr::relocate("Paper P90", .after = "Sim P90") |>
  knitr::kable(digits = 1, caption = "Steady-state trough (ng/mL) vs Yang 2025 Results.")
Steady-state trough (ng/mL) vs Yang 2025 Results.
Regimen Sim P10 Paper P10 Sim median Paper median Sim P90 Paper P90 % above 98 ng/mL
120 mg QD 70.4 71.4 155.6 151 253.5 293 77.5
180 mg QD 99.7 109.0 206.6 232 404.5 449 91.0
240 mg QD 118.3 142.0 274.4 301 514.5 588 93.0

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = clin_nca,
  reference = tibble::tibble(treatment = "180 mg QD", cmax = 635, cmin = 210),
  by = "treatment",
  params = c("cmax", "cmin"),
  units = c(cmax = "ng/mL", cmin = "ng/mL"),
  tolerance_pct = 20
)
knitr::kable(cmp, caption = "180 mg QD steady state vs Table S2 (clinical) medians. * differs by >20%.")
180 mg QD steady state vs Table S2 (clinical) medians. * differs by >20%.
NCA parameter treatment Reference Simulated % diff
Cmax (ng/mL) 180 mg QD 635 623 -1.9%
Cmin (ng/mL) 180 mg QD 210 207 -1.6%

stopifnot(
  # Centre: a wrong CL/F, unit or dose moves the median by tens of percent.
  all(abs(clin_tab$trough_p50 / clin_tab$paper_p50 - 1) < 0.20),
  # The paper's dose-selection claim: at 180 mg the 10th percentile trough
  # clears the 98 ng/mL human TSC. Gate on the robust quantile, with headroom.
  clin_tab$pct_above_tsc[clin_tab$treatment == "180 mg QD"] > 80
)

The simulated median troughs are within about 10% of the Results values (+3%, -11% and -9% at 120, 180 and 240 mg). At 180 mg the medians are within 2% of Table S2 for both trough and Cmax. The 180 mg regimen keeps about 90% of the virtual patients above the 98 ng/mL ROS1 TSC, which is the paper’s dose-selection claim; 120 mg does not (about 78%). The simulated 90th percentiles are 10-14% below the paper’s. The spread depends on the assumed weight and albumin distributions, which the paper does not report, so it is not gated.

Figure S3 / Table S2 also compare the two human models at 180 mg. The projection from animals (trough 126, Cmax 305 ng/mL) under-predicts the clinical model. The paper attributes this mainly to the much faster human absorption (clinical Ka 1.11 vs 0.341 1/h).

Assumptions and deviations

  • V1/F 342 L, not 348 L. Table 3 gives V1/F = 342 L (RSE 7%). The Table 3 footnote equation prints V1/F = 348 x (BBWT/63)^2.02, and every other number in that equation matches the table. 348 L is the V1/F of the same group’s later four-study analysis (Yang_2025_iruplinalkib), so the footnote looks carried over from it. The maintainers ship the tabulated 342 L. The difference is 1.7%, and no printed result can tell the two apart.
  • Fed Ka factor 0.796, not 0.582. Table 3 gives ‘Food on Ka’ = 0.796 (RSE 27%, fasting reference). The Results text says fed Ka was “58.2% of that under fasting”. No reparameterisation of 0.796 gives 0.582. The same sentence’s 0.479 h lag matches Table 3. The maintainers ship the tabulated 0.796. The 58.2% is close to the later analysis’s 0.588 and may come from another model run.
  • Covariates follow Table 3, not the Results text. The Results say time-varying albumin also affected V1/F. Table 3 and its footnote equations both put albumin on CL/F only, so the model follows them.
  • Albumin units. The reference value 42 has no printed unit and is read as g/L; 42 g/dL is not physiological.
  • IIV as CV%. Table 3 IIV percentages are taken as log-normal CVs (omega^2 = log(1 + CV^2)), as the table’s abbreviation list defines CV. The same group used this convention in the later analysis.
  • Mouse models are typical-value only. The paper reports no between-animal variability for either mouse stage and no residual error for tumor volume. The tumor residual is encoded as fixed(0), and no etas are included.
  • ALK model reuses the ROS1-mouse PK. Plasma was sampled only in the ROS1-xenograft mice. The paper describes a single “previously estimated PopPK” stage fixed in both TGI fits, so both mouse files carry Table S1.
  • TGI definition. The paper does not define TGI. The end-of-treatment 1 - dT/dC is used here; it reproduces the reported ranges from the digitised Figure S1 means.
  • Digitised values. Figure S1 group means were read off the published figure by the maintainers for overlay only; no parameter came from a figure.
  • Projected human model has no bioavailability parameter. The Table 2 values are apparent oral parameters, as their labels say. Table S2 and Figure 3B both confirm this (see above).
  • Clinical virtual cohort. Body weight and albumin distributions are assumptions centred on the model’s reference values. Albumin is held constant over time, although the model treats it as time-varying. All patients are dosed fasted.
  • No erratum or correction notice was found for this article (Europe PMC, checked 2026-11-10).