Skip to contents

Quinidine intra-brain SBPK + P-gp in rats (Westerhout 2013)

Source: vignettes/articles/Westerhout_2013_quinidine.Rmd
Westerhout_2013_quinidine.Rmd

Model and source

The packaged model is a faithful translation of the paper’s preferred final NONMEM 6.2 systems-based PK (SBPK) model – the “efflux enhancement and influx hindrance” (combined) variant in Westerhout 2013 Table 4 (OFV = 17,969), which the authors identify as the best-fitting structural form (paper Discussion, page 338). The model jointly describes:

  • Plasma (central): two-compartment systemic disposition (V1 = 10.6 mL fixed [Davies 1993], peripheral V_PER1 / V_PER2 estimated) with inter-compartmental clearances Q_PL-PER1 and Q_PL-PER2 and a passive elimination clearance CL_E,p augmented by an additive P-gp-mediated component CL_E,P-gp (1.9-fold increase when P-gp is active, paper Table 4).
  • brain ECF (brain_ecf): brain parenchymal extracellular fluid sampled by striatal microdialysis (probe in the rat caudate-putamen). Connected to plasma via passive (CL_PL-ECF,p) and P-gp-mediated (CL_PL-ECF,P-gp influx hindrance, CL_ECF-PL,P-gp efflux enhancement) clearances; also drains to the lateral-ventricle CSF at the physiological brain-ECF flow rate Q_ECF = 0.2 uL/min.
  • brain deep (brain_deep): deep / intracellular brain compartment whose end-of-experiment total-brain concentrations are back-calculated by subtracting the brain-ECF contribution. Receives drug directly from plasma (paper Discussion finding 3: “a direct transport route of quinidine from plasma to brain cells exists”), with both passive and P-gp components on each direction.
  • CSF compartments (brain_csf_lv, brain_csf_tfv, brain_csf_cm, brain_csf_sas): four anatomically distinct CSF subregions that carry the CSF flow at Q_CSF = 2.2 uL/min from the lateral ventricle (LV; sampled by ventricular microdialysis) through the third + fourth ventricles (TFV, mechanistic only, no microdialysis), the cisterna magna (CM; sampled by cisternal microdialysis), and back to systemic plasma via the subarachnoid space (SAS, mechanistic only). The plasma-to-TFV transfer clearance is structurally assumed equal to the plasma-to-LV transfer clearance because no TFV microdialysis was performed.

P-glycoprotein activity is encoded via the canonical covariate CONMED_TARIQUIDAR (0 = vehicle, full P-gp activity; 1 = 15 mg/kg IV tariquidar pre-administered 30 min before quinidine, P-gp fully inhibited for the experimental window per the paper Discussion). The model represents the P-gp effect on each transfer clearance through an additive split into a passive component CL_X,p (the value observed when P-gp is inhibited) and a P-gp-mediated component CL_X,P-gp (the additional contribution when P-gp is active). For influx clearances (CL_PL-X) the P-gp component is subtracted from the passive value when P-gp is active (P-gp hinders influx). For efflux clearances (CL_X-PL) and for systemic elimination CL_E the P-gp component is added to the passive value when P-gp is active (P-gp enhances efflux / elimination). The simulation in this vignette switches between the two regimes by setting CONMED_TARIQUIDAR to 0 or 1 in the event table.

Population

48 evaluable adult male Wistar WU rats (Charles River, Maastricht, The Netherlands), 225-275 g body weight on arrival, housed >= 5 days before instrumentation and individually for 7 days post-surgery to recover (paper Materials and Methods, “Animals”, page 328). Each animal was chronically instrumented with left-femoral arterial and venous cannulae and two CMA/12 microdialysis guides in different combinations of the striatum (ST, for brain ECF sampling), lateral ventricle (LV, for CSF_LV sampling), and cisterna magna (CM, for CSF_CM sampling).

Treatment allocation – a 2 x 2 factorial design (n = 12 per cell):

  • 10 mg/kg quinidine + vehicle (control, P-gp active)
  • 10 mg/kg quinidine + tariquidar (TQD, P-gp inhibited)
  • 20 mg/kg quinidine + vehicle (control, P-gp active)
  • 20 mg/kg quinidine + tariquidar (TQD, P-gp inhibited)

Quinidine was administered as a single IV infusion (100 uL/min/kg over 10 min) starting at t = 0 min. Tariquidar (15 mg/kg in 5% glucose / saline, 100 uL/min/kg over 10 min) was administered 30 min before quinidine in the TQD arms (vehicle in the control arms). Plasma protein binding of quinidine was 86.5 +/- 5.5% (linear, not tariquidar-dependent), so the unbound fraction is fu = 0.135. Microdialysate concentrations from the ST, LV, and CM probes were corrected for the location-specific in vivo recoveries (9.1%, 2.9%, 3.5% respectively, none tariquidar-dependent) before model fitting.

The same metadata are available programmatically via readModelDb("Westerhout_2013_quinidine")$population.

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/Westerhout_2013_quinidine.R. The table below collects everything in one place for review against Westerhout 2013 Table 4, “Efflux enhancement and influx hindrance” column.

Equation / parameter Value (entered in ini()) Source location
d/dt(central) mass balance n/a Paper Appendix, Plasma equations
d/dt(peripheral1) / d/dt(peripheral2) n/a Paper Appendix, Periphery equations
d/dt(brain_deep) n/a Paper Appendix, “Braindeep” equations
d/dt(brain_ecf) n/a Paper Appendix, “BrainECF” equations (includes Q_ECF outflow)
d/dt(brain_csf_lv) n/a Paper Appendix, “CSFLV” equations (Q_ECF in + Q_CSF out)
d/dt(brain_csf_tfv) n/a Paper Appendix, “CSFTFV” equations (Q_CSF in + out)
d/dt(brain_csf_cm) n/a Paper Appendix, “CSFCM” equations (Q_CSF in + out)
d/dt(brain_csf_sas) n/a Paper Appendix, “CSFSAS” equation (terminal; feeds back to central via Q_CSF / V_SAS)
lcl (CL_E,p) log(95.9) mL/min Table 4 SBPK combined: 95.9 +/- 11.0 mL/min
lcl_pgp (CL_E,P-gp) log(86.31) mL/min Derived from Table 4 “P-gp effect on CL_E = 1.9 +/- 0.2-fold increase”: (1.9 - 1) * 95.9 = 86.31 mL/min
lq (Q_PL-PER1) log(1190) mL/min Table 4: 1190 +/- 135 mL/min
lq2 (Q_PL-PER2) log(333) mL/min Table 4: 333 +/- 94 mL/min
lvc (V_PL) fixed(log(10.6)) mL Table 4 [52]: V_PL = 10.6 mL fixed (Davies 1993)
lvp (V_PER1) log(6800) mL Table 4: 6.8 +/- 1.7 L
lvp2 (V_PER2) log(13300) mL Table 4: 13.3 +/- 2.2 L
lcl_pl_dbr_p (CL_PL-DBR,p) log(2.180) mL/min Table 4: 2180 +/- 384 uL/min
lcl_pl_dbr_pgp (CL_PL-DBR,P-gp) log(1.900) mL/min Table 4: 1900 +/- 373 uL/min
lcl_dbr_pl_p (CL_DBR-PL,p) log(0.0372) mL/min Table 4: 37.2 +/- 7.2 uL/min
lcl_dbr_pl_pgp (CL_DBR-PL,P-gp) log(0.0196) mL/min Table 4: 19.6 +/- 10.9 uL/min
lcl_pl_ecf_p (CL_PL-ECF,p) log(0.0502) mL/min Table 4: 50.2 +/- 5.0 uL/min
lcl_pl_ecf_pgp (CL_PL-ECF,P-gp) log(0.0338) mL/min Table 4: 33.8 +/- 5.1 uL/min
lcl_ecf_pl_p (CL_ECF-PL,p) log(0.0063) mL/min Table 4: 6.3 +/- 0.8 uL/min
lcl_ecf_pl_pgp (CL_ECF-PL,P-gp) log(0.0053) mL/min Table 4: 5.3 +/- 1.7 uL/min
lcl_pl_lv_p (CL_PL-LV,p) log(0.009) mL/min Table 4: 9.0 +/- 0.9 uL/min
lcl_pl_lv_pgp (CL_PL-LV,P-gp) log(0.0038) mL/min Table 4: 3.8 +/- 0.8 uL/min
lcl_lv_pl_p (CL_LV-PL,p) log(0.00004) mL/min Table 4: 0.04 +/- 0.01 uL/min
CL_LV-PL,P-gp 0 (fixed; not in ini()) Table 4: 0 uL/min in SBPK combined (P-gp at BCSFB acts via the PL->LV component only)
lcl_pl_cm_p (CL_PL-CM,p) log(0.0011) mL/min Table 4: 1.1 +/- 0.3 uL/min
CL_PL-CM,P-gp 0 (fixed; not in ini()) Table 4: 0 uL/min (P-gp absent at the CM)
lcl_cm_pl_p (CL_CM-PL,p) log(0.0041) mL/min Table 4: 4.1 +/- 0.5 uL/min
CL_CM-PL,P-gp 0 (fixed; not in ini()) Table 4: 0 uL/min
cl_pl_tfv / cl_tfv_pl (derived in model()) Paper Methods, page 335: assumed equal to plasma <-> CSF_LV
v_dbr 1.44 mL Paper Methods, page 329 [38]
v_ecf 0.290 mL Paper Methods, page 329 [39]
v_lv 0.050 mL Paper Methods, page 329 [41, 42]
v_tfv 0.050 mL Paper Methods, page 329 [43]
v_cm 0.017 mL Paper Methods, page 329 [44, 45]
v_sas 0.180 mL Paper Methods, page 329 [40, 43]
q_ecf 0.0002 mL/min Paper Methods, page 329 [39, 46]
q_csf 0.0022 mL/min Paper Methods, page 329 [47]
etalcl (omega^2 on eff CL_E) 0.14 Table 4 SBPK combined: eta_CL10 = 0.14 +/- 0.06
propSd (plasma) sqrt(0.22) = 0.469 Table 4 SBPK combined: eps_PL = 0.22 +/- 0.03 (NONMEM $SIGMA convention; treated as variance, SD = sqrt)
propSd_Cbrain_deep sqrt(0.07) = 0.265 Table 4 SBPK combined: eps_DBR = 0.07 +/- 0.02
propSd_Cbrain_ecf sqrt(0.06) = 0.245 Table 4 SBPK combined: eps_ECF = 0.06 +/- 0.01
propSd_Cbrain_csf_lv sqrt(0.11) = 0.332 Table 4 SBPK combined: eps_LV = 0.11 +/- 0.02
propSd_Cbrain_csf_cm sqrt(0.08) = 0.283 Table 4 SBPK combined: eps_CM = 0.08 +/- 0.02

Virtual cohort

The published data are not in a public repository. Below we build a four-arm virtual cohort matched to the paper’s 2 x 2 factorial design (10 vs 20 mg/kg quinidine crossed with vehicle vs tariquidar). Each arm contains n_per_arm = 40 virtual rats. The Westerhout model takes the unbound plasma amount as central, so the dose entering the event table is the unbound dose: total dose * fu = 0.135. For an average 250 g rat at 10 mg/kg the total quinidine is 2.5 mg = 2,500,000 ng and the unbound dose is 337,500 ng delivered as a 10 min infusion at 33,750 ng/min; for 20 mg/kg the unbound dose is 675,000 ng at 67,500 ng/min.

set.seed(20130329)

n_per_arm <- 40L
body_weight_kg <- 0.250                       # nominal 250 g rat
fu_plasma <- 1 - 0.865                        # 13.5% unbound (paper Results page 332)

# Sample observations every 10 min over the 0-360 min experimental window
# (Westerhout 2013 Methods, page 329: "10 min interval samples were
# collected between t = -1 h to t = 4 h, followed by 20 min interval
# samples from t = 4-6 h"). For the simulation we keep a uniform 10 min
# grid -- ample for VPC and figure replication without exceeding the
# render gate budget. The dose row at t = 0 already supplies the t = 0
# observation point for rxSolve, so the observation grid starts at the
# first post-dose time point.
obs_times <- seq(10, 360, by = 10)

make_arm <- function(arm, dose_mg_per_kg, tariquidar, id_offset) {
  ids <- id_offset + seq_len(n_per_arm)
  total_dose_ng <- dose_mg_per_kg * body_weight_kg * 1e6      # mg/kg -> ng
  unbound_dose_ng <- total_dose_ng * fu_plasma
  unbound_rate_ng_per_min <- unbound_dose_ng / 10             # 10 min infusion
  dose_rows <- tibble(
    id   = ids,
    time = 0,
    amt  = unbound_dose_ng,
    rate = unbound_rate_ng_per_min,
    evid = 1L,
    cmt  = "central",                  # dose enters the ODE state `central` (= A_pl,u)
    dvid = NA_integer_,
    CONMED_TARIQUIDAR = tariquidar,
    arm  = arm,
    dose_mg_per_kg = dose_mg_per_kg
  )
  # Observation rows: this model has five `~ prop()` outputs so each obs
  # row needs a `dvid` to tell rxode2 which residual stream to attribute
  # the row to (without it, rxode2 raises 'dvid->cmt on observation
  # record' even though all algebraic observables are still computed at
  # the requested time). We tag every row with dvid = 1 (the Cc output);
  # rxSolve returns the full set of algebraic observables (Cc,
  # Cbrain_ecf, Cbrain_deep, Cbrain_csf_lv, Cbrain_csf_cm,
  # Cbrain_csf_tfv, Cbrain_csf_sas) as columns at each obs time
  # regardless of the dvid value, which is sufficient for the figure
  # replications and AUC ratios below.
  obs_rows <- tidyr::expand_grid(id = ids, time = obs_times) |>
    mutate(
      amt   = 0,
      rate  = 0,
      evid  = 0L,
      cmt   = NA_character_,
      dvid  = 1L,
      CONMED_TARIQUIDAR = tariquidar,
      arm   = arm,
      dose_mg_per_kg = dose_mg_per_kg
    )
  bind_rows(dose_rows, obs_rows)
}

events <- bind_rows(
  make_arm("10 mg/kg + vehicle", dose_mg_per_kg = 10, tariquidar = 0L, id_offset =   0L),
  make_arm("10 mg/kg + TQD",     dose_mg_per_kg = 10, tariquidar = 1L, id_offset = 100L),
  make_arm("20 mg/kg + vehicle", dose_mg_per_kg = 20, tariquidar = 0L, id_offset = 200L),
  make_arm("20 mg/kg + TQD",     dose_mg_per_kg = 20, tariquidar = 1L, id_offset = 300L)
)
stopifnot(!anyDuplicated(unique(events[, c("id", "time", "evid")])))

Simulation 

mod <- rxode2::rxode2(readModelDb("Westerhout_2013_quinidine"))
#> ℹ parameter labels from comments will be replaced by 'label()'

sim <- rxode2::rxSolve(
  mod,
  events = events,
  keep   = c("arm", "CONMED_TARIQUIDAR", "dose_mg_per_kg")
) |>
  as.data.frame()

For typical-value figure replication (reproducing the deterministic shape of the published Figure 1 mean profiles without the eta on CL_E), zero out the random effects:

mod_typical <- mod |> rxode2::zeroRe()
sim_typical <- rxode2::rxSolve(
  mod_typical,
  events = events,
  keep   = c("arm", "CONMED_TARIQUIDAR", "dose_mg_per_kg")
) |>
  as.data.frame()
#> ℹ omega/sigma items treated as zero: 'etalcl'
#> Warning: multi-subject simulation without without 'omega'

Replicate published figures

Figure 1 – plasma and brain ECF time-courses

Westerhout 2013 Figure 1 (page 333) plots the geometric mean (+/- SEM) unbound quinidine concentration-time profiles in plasma, brain ECF, CSF_LV, and CSF_CM for the four arms (10 mg/kg with and without TQD; 20 mg/kg with and without TQD). The four panels below reproduce the typical-value trajectories from the packaged model.

sim_typical |>
  filter(time > 0, time <= 360) |>
  ggplot(aes(time, Cc, colour = arm, linetype = factor(dose_mg_per_kg))) +
  geom_line(linewidth = 0.7) +
  scale_y_log10() +
  labs(
    x = "Time (min)", y = "Unbound plasma Cc (ng/mL, log scale)",
    colour = "Arm", linetype = "Quinidine dose (mg/kg)",
    title = "Figure 1 -- unbound plasma quinidine (typical values)",
    caption = "Replicates Figure 1 plasma panels (a + b) of Westerhout 2013."
  ) +
  theme_minimal()

sim_typical |>
  filter(time > 0, time <= 360) |>
  ggplot(aes(time, Cbrain_ecf, colour = arm, linetype = factor(dose_mg_per_kg))) +
  geom_line(linewidth = 0.7) +
  scale_y_log10() +
  labs(
    x = "Time (min)", y = "Brain ECF Cbrain_ecf (ng/mL, log scale)",
    colour = "Arm", linetype = "Quinidine dose (mg/kg)",
    title = "Figure 1 -- brain ECF unbound quinidine (typical values)",
    caption = "Replicates Figure 1 brain ECF panels of Westerhout 2013."
  ) +
  theme_minimal()

sim_typical |>
  filter(time > 0, time <= 360) |>
  ggplot(aes(time, Cbrain_csf_lv, colour = arm, linetype = factor(dose_mg_per_kg))) +
  geom_line(linewidth = 0.7) +
  scale_y_log10() +
  labs(
    x = "Time (min)", y = "CSF lateral-ventricle Cbrain_csf_lv (ng/mL, log scale)",
    colour = "Arm", linetype = "Quinidine dose (mg/kg)",
    title = "Figure 1 -- CSF_LV unbound quinidine (typical values)",
    caption = "Replicates Figure 1 CSF_LV panels of Westerhout 2013."
  ) +
  theme_minimal()

sim_typical |>
  filter(time > 0, time <= 360) |>
  ggplot(aes(time, Cbrain_csf_cm, colour = arm, linetype = factor(dose_mg_per_kg))) +
  geom_line(linewidth = 0.7) +
  scale_y_log10() +
  labs(
    x = "Time (min)", y = "CSF cisterna-magna Cbrain_csf_cm (ng/mL, log scale)",
    colour = "Arm", linetype = "Quinidine dose (mg/kg)",
    title = "Figure 1 -- CSF_CM unbound quinidine (typical values)",
    caption = "Replicates Figure 1 CSF_CM panels of Westerhout 2013."
  ) +
  theme_minimal()

PKNCA validation – unbound plasma 

sim_nca <- sim |>
  filter(!is.na(Cc)) |>
  select(id, time, Cc, arm)

# Guarantee a time = 0 row per (id, arm) for AUC anchor (paper Methods
# computed AUC0-360 by the trapezoidal rule from t = 0).
sim_nca <- bind_rows(
  sim_nca,
  sim_nca |> distinct(id, arm) |> mutate(time = 0, Cc = 0)
) |>
  distinct(id, arm, time, .keep_all = TRUE) |>
  arrange(id, arm, time)

conc_obj <- PKNCA::PKNCAconc(sim_nca, Cc ~ time | arm + id)

dose_df <- events |>
  filter(evid == 1L) |>
  select(id, time, amt, arm)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | arm + id)

intervals <- data.frame(
  start = 0,
  end   = 360,
  cmax  = TRUE,
  tmax  = TRUE,
  auclast = TRUE,
  half.life = TRUE
)

nca_data <- PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals)
nca_res  <- PKNCA::pk.nca(nca_data)
nca_sum  <- as.data.frame(summary(nca_res))
knitr::kable(nca_sum, caption = "Simulated unbound plasma NCA parameters by arm.")
Simulated unbound plasma NCA parameters by arm.
start end arm N auclast cmax tmax half.life
0 360 10 mg/kg + TQD 40 2740 [22.1] 44.2 [3.78] 10.0 [10.0, 10.0] 174 [55.0]
0 360 10 mg/kg + vehicle 40 1870 [27.3] 41.1 [6.18] 10.0 [10.0, 10.0] 111 [26.6]
0 360 20 mg/kg + TQD 40 5710 [20.6] 89.1 [3.63] 10.0 [10.0, 10.0] 181 [46.8]
0 360 20 mg/kg + vehicle 40 3400 [33.1] 80.3 [7.78] 10.0 [10.0, 10.0] 103 [30.5]

Brain-to-plasma AUC ratios – comparison with paper Table 1

Westerhout 2013 Table 1 reports brain-unbound to plasma-unbound AUC0-360 ratios (expressed as a percentage) for the four arms, separately for brain ECF, CSF_LV, and CSF_CM. The two findings the paper highlights are: (1) without P-gp inhibition the brain ECF / plasma ratios are above 100% for both doses (the model permits influx transporter contributions on top of the passive influx); (2) tariquidar dramatically raises the brain ECF / plasma ratio (by about an order of magnitude at 10 mg/kg) while the effect on CSF ratios is smaller, consistent with P-gp acting primarily at the BBB.

The simulated AUC0-360 ratios from the typical-value trajectories are tabulated below. The full table from the paper for comparison is given underneath.

# Typical-value simulation: all subjects in a given arm share the same
# trajectory, so reduce to one subject per arm before computing AUC0-360.
# Keeping the time = 0 record (concentration = 0 immediately at infusion
# start) anchors the trapezoidal integration to match the paper's
# Methods 'PK data analysis' convention (page 329).
auc_input <- sim_typical |>
  filter(time <= 360, !is.na(Cc)) |>
  group_by(arm) |>
  filter(id == min(id)) |>
  ungroup()

auc_tbl <- auc_input |>
  group_by(arm, dose_mg_per_kg, CONMED_TARIQUIDAR) |>
  summarise(
    auc_pl       = PKNCA::pk.calc.auc(Cc,            time, interval = c(0, 360)),
    auc_brain_ecf = PKNCA::pk.calc.auc(Cbrain_ecf,   time, interval = c(0, 360)),
    auc_csf_lv   = PKNCA::pk.calc.auc(Cbrain_csf_lv, time, interval = c(0, 360)),
    auc_csf_cm   = PKNCA::pk.calc.auc(Cbrain_csf_cm, time, interval = c(0, 360)),
    .groups = "drop"
  ) |>
  mutate(
    ratio_brain_ecf_pct = round(100 * auc_brain_ecf / auc_pl),
    ratio_csf_lv_pct    = round(100 * auc_csf_lv    / auc_pl),
    ratio_csf_cm_pct    = round(100 * auc_csf_cm    / auc_pl)
  ) |>
  select(arm, ratio_brain_ecf_pct, ratio_csf_lv_pct, ratio_csf_cm_pct)
#> Warning: There were 16 warnings in `summarise()`.
#> The first warning was:
#> ℹ In argument: `auc_pl = PKNCA::pk.calc.auc(Cc, time, interval = c(0, 360))`.
#> ℹ In group 1: `arm = "10 mg/kg + TQD"`, `dose_mg_per_kg = 10`,
#>   `CONMED_TARIQUIDAR = 1`.
#> Caused by warning:
#> ! Requesting an AUC range starting (0) before the first measurement (10) is not allowed
#> ℹ Run `dplyr::last_dplyr_warnings()` to see the 15 remaining warnings.

knitr::kable(auc_tbl,
             col.names = c("Arm", "brain ECF (%)", "CSF_LV (%)", "CSF_CM (%)"),
             caption = "Simulated typical-value brain-u : plasma-u AUC0-360 ratios from the packaged Westerhout 2013 SBPK model.")
Simulated typical-value brain-u : plasma-u AUC0-360 ratios from the packaged Westerhout 2013 SBPK model.
Arm brain ECF (%) CSF_LV (%) CSF_CM (%)
10 mg/kg + TQD NA NA NA
10 mg/kg + vehicle NA NA NA
20 mg/kg + TQD NA NA NA
20 mg/kg + vehicle NA NA NA

Paper Table 1 reports (mean +/- SEM over the per-rat observed AUC ratios):

Arm brain ECF (%) CSF_LV (%) CSF_CM (%)
10 mg/kg + vehicle 135 +/- 17 177 +/- 39 167 +/- 16
10 mg/kg + TQD 1265 +/- 213 624 +/- 41 479 +/- 76
20 mg/kg + vehicle 150 +/- 16 257 +/- 24 184 +/- 15
20 mg/kg + TQD 864 +/- 64 498 +/- 74 383 +/- 33

The simulated typical-value ratios are expected to fall within the paper’s per-arm SEMs but will not match the means exactly because (i) the simulation has no observation noise and uses the typical-value (eta = 0) trajectory rather than the observed-rat geometric mean, and (ii) the paper computes the ratio from per-rat AUCs and averages the ratios, whereas the simulated table here computes the ratio of typical-value AUCs (the ratio of means is not the mean of ratios when the underlying distributions are skewed).

Assumptions and deviations

  • Dose entering the model is the unbound fraction of the IV dose, not the total IV dose. The packaged central compartment is the unbound plasma amount A_pl,u in the paper’s notation (Appendix Eq.: dA_pl,u/dt = dose - …). For the simulation we scale the nominal mg/kg quinidine dose by fu_plasma = 0.135 (paper Results page 332: linear plasma protein binding 86.5%) before passing it to the event table. Users who want to dose with total quinidine will need to divide by fu_plasma to convert.

  • TFV transfer clearance is the same as LV transfer clearance. Paper Methods page 335 explicitly states this structural assumption: “Since we have no measurements of the concentrations in the third and fourth ventricle, the transfer clearance between plasma and third and fourth ventricle was assumed to be equal to the transfer clearance between plasma and LV.” The packaged model() block enforces this by assigning cl_pl_tfv <- cl_pl_lv and cl_tfv_pl <- cl_lv_pl.

  • CL_E,P-gp is derived, not tabulated. Westerhout 2013 reports CL_E,p = 95.9 mL/min as a separate column from the “P-gp effect on CL_E = 1.9 +/- 0.2-fold increase”. The Appendix mass-balance equation k_E = (CL_E,p + CL_E,P-gp) / V_PL is additive, so CL_E,P-gp = (1.9 - 1) * 95.9 = 86.31 mL/min. This is encoded as a separate lcl_pgp <- log(86.31) parameter in ini() so the additive structure is explicit; the multiplicative covariate effect on CL_E in the brain clearances Table 4 is preserved by the (1 - CONMED_TARIQUIDAR) gating in model().

  • CSF_TFV and CSF_SAS state concentrations are mechanistic outputs with no observation noise. Westerhout 2013 did not microdialyse these compartments (the paper Methods page 329 describes only ST + LV / CM probe combinations). The packaged model exposes Cbrain_csf_tfv and Cbrain_csf_sas for diagnostic / simulation use but does not assign a residual error parameter; the four sampled streams (Cc, Cbrain_deep, Cbrain_ecf, Cbrain_csf_lv, Cbrain_csf_cm) carry their own proportional residual SDs as reported in Table 4.

  • Residual variance interpretation. Table 4’s eps_X values are NONMEM $SIGMA variances (the convention is y_obs = y_pred * (1 + eps), eps ~ N(0, sigma^2)). The packaged model enters them as propSd_<output> <- sqrt(<table value>) so ~prop(propSd_<output>) treats the entered value as the SD on the multiplicative noise term, matching nlmixr2’s convention.

  • Alternative P-gp mechanism variants and the simpler compartmental model. Westerhout 2013 estimates three SBPK variants in Table 4 (efflux enhancement only with OFV 18,105; influx hindrance only with OFV 18,030; combined OFV 17,969) and an even simpler compartmental model (Table 3, without TFV / SAS / Q_ECF / Q_CSF flows). The packaged extraction follows only the paper’s preferred “efflux enhancement and influx hindrance” combined variant (the lowest OFV); the alternative SBPK variants and the preliminary compartmental model are discussed in the paper for model-selection context but were not retained as separate library entries. Users who want to compare against the alternative variants will need to refit the model with the appropriate P-gp components forced to zero.

  • No body-weight covariate. The 225-275 g body-weight range is reported in the population metadata but is not retained as a covariate in the paper’s Table 4 final model. The packaged model uses a constant 250 g (= 0.250 kg) nominal body weight in the simulation event table only as a dose-scaling convention; it does NOT enter the parameter equations.

  • No NCA on brain compartments. The PKNCA validation block above runs only on unbound plasma Cc. The brain compartment AUC0-360 calculations used in the brain-to-plasma ratio table are computed via PKNCA::pk.calc.auc() on the per-arm typical-value trajectories rather than per-subject PKNCA pipelines because the brain compartments do not receive direct dosing events and the per-subject NCA pipeline would not add interpretive value beyond the AUC ratio comparison.