Skip to contents

The model

Elias_2025_glimepiride_pbpk is the whole-body physiologically based pharmacokinetic model that Elias and Koenig (2025) built as a “digital twin” of the sulfonylurea glimepiride from 20 curated clinical studies, and distributed as SBML under CC-BY. It tracks the parent drug and its two sequential metabolites: M1 (cyclohexylhydroxymethyl-glimepiride, formed by CYP2C9 and about 30% as active as the parent) and M2 (carboxy-glimepiride).

mod <- rxode2::rxode(readModelDb("Elias_2025_glimepiride_pbpk"))
length(mod$state)
#> [1] 47

The 41 ODEs are organised as a systemic circulation (venous, arterial, portal and hepatic-venous plasma pools, plus lung and a lumped rest-of-body tissue) linking three submodels (paper Figure 1, Supplementary Figure S2):

  • Intestine. Oral glimepiride dissolves from the dose compartment into the gut lumen (ka_dis_gli) and is absorbed first order into gut plasma (gliabs_k). M1 and M2 pass from gut plasma back into the lumen (mreabs_k) and leave in feces (mexc_k).
  • Liver. Glimepiride equilibrates rapidly between liver plasma and tissue (gliim_k), is converted to M1 by CYP2C9 (Michaelis-Menten), and M1 is converted to M2 (first order). Both metabolites are exported back to plasma.
  • Kidney. M1 and M2 are excreted in urine first order from kidney plasma. Unchanged glimepiride is not excreted renally.

Lung and rest-of-body tissue take up all three species through one shared partition coefficient (kp_gli) and one shared distribution rate (ftissue_gli).

A note on units and dosing. Following the source SBML, every plasma and tissue state is a concentration in mmol/L rather than an amount; only the dosing compartments (mg) and the cumulative urine and feces pools (mmol) are amounts. Give oral doses into depot and intravenous doses into depot_iv (glimepiride), depot_iv_m1 or depot_iv_m2, never directly into a plasma or tissue compartment. Model time is in minutes. The outputs Cc, Cc_m1 and Cc_m2 are venous plasma concentrations in umol/L; multiply by the molecular weight (490.616, 506.62 and 520.6 g/mol) to get ng/mL.

Population

pop <- mod$population
data.frame(Field = names(pop), Value = vapply(pop, as.character, character(1))) |>
  knitr::kable(row.names = FALSE)
Field Value
species human
n_subjects NA
n_studies 20
age_range adults; pediatric studies excluded (Methods Section 4.1)
weight_range 75 kg reference individual; 40-170 kg explored in the body-weight scan (Figure 5)
sex_female_pct NA
disease_state Healthy volunteers, patients with type 2 diabetes mellitus, subjects with renal impairment, normal-weight and morbidly obese subjects, and carriers of CYP2C9 1, 2 and *3 alleles (Table 1). Hepatic impairment was simulated from the cirrhosis model with limited clinical comparison data.
dose_range Oral 0.5-8 mg single and multiple dose; intravenous glimepiride 0.25-1.5 mg and intravenous M1 1.5 mg (Table 1).
regions Europe, East Asia, Egypt and the USA across the curated studies (Table 1 and Supplementary Section S6)
notes Deterministic typical-individual model. The paper reports a single optimized parameter set (Supplementary Table S1), used for every simulation without refitting, and no inter-individual variability or residual-error model; the only stochastic element in the paper is the lognormal sampling of CYP2C9 activity in the population analyses. Parameters were fitted by weighted least squares (100 local-optimisation runs) to the healthy, fasted subset of the curated PK-DB dataset (Methods Section 4.3).

The model was fitted to the healthy, fasted subset of 20 curated clinical studies (paper Table 1): healthy volunteers, patients with type 2 diabetes, subjects with renal impairment, normal-weight and morbidly obese subjects, and carriers of CYP2C9 1, 2 and *3 alleles, with oral doses of 0.5-8 mg and intravenous glimepiride (0.25-1.5 mg) or M1 (1.5 mg). It is a deterministic typical-individual model: the authors report one optimized parameter set (Supplementary Table S1) and no between-subject variability or residual error. The reference individual weighs 75 kg.

Source trace

Every value in ini() traces to the manuscript Methods, the Supplementary Materials (equations in Section S3, optimized parameters in Table S1), or the authors’ archived SBML model, which the manuscript names as the definitive model source.

Quantity Parameters Source
Fractional organ and blood-pool volumes fvar, fvgu, fvhv, fvki, fvli, fvlu, fvpo, fvve Archived SBML FV* (not printed in the paper)
Fractional blood flows fqgu, fqh, fqki, fqlu Archived SBML FQ*
Cardiac output, hematocrit, blood fraction covbw, fcardio, hct, fblood Archived SBML COBW, f_cardiac_function, HCT, Fblood
Molecular weights mr_gli, mr_m1, mr_m2 Archived SBML Mr_gli, Mr_m1, Mr_m2
Partition coefficient, tissue distribution kp_gli, ftissue_gli Table S1 (10.02060; 0.00071 L/min)
Dissolution ka_dis_gli Section S3 names it; value 2 /h from the archived SBML
Absorption, food scaling gliabs_k, f_absorption Table S1 (0.01590 /min); f_absorption = 1 (fasted)
Intestinal metabolite handling mreabs_k, mexc_k Table S1 (0.01592; 0.00017 /min)
Hepatic import gliim_k Section S3 names it; value 100 /min from the archived SBML (not optimized)
CYP2C9 conversion GLI -> M1 gli2m1_vmax, gli2m1_km_gli Table S1 (Vmax 0.00005 mmol/min/L); Km 2.25 umol/L from the archived SBML, literature-based per Methods 4.2
M1 -> M2, metabolite export m12m2_k, m1ex_li_k, m2ex_li_k Table S1 (0.01485; 0.07774; 99.99817 /min)
Urinary excretion m1ex_ki_k, m2ex_ki_k Table S1 (0.14801; 0.09849 /min)
Intravenous release ti_iv Archived SBML ti_gli (10 s)
Scenario covariates WT, RENALFUNC_REL, HEPFUNC_REL, CYP2C9 Methods 4.2 (KDIGO, Child-Turcotte-Pugh and CYP2C9 allele mappings); Supplementary Table S2

The intestine, kidney and liver rate laws are Supplementary Section S3. The whole-body circulation ODEs are not written out anywhere in the manuscript or supplement and were taken from the archived SBML (models/glimepiride_body_flat.md, Zenodo 10.5281/zenodo.15189579, version 0.6.1, the version Methods 4.2 names). That archive’s optimized parameters agree with Supplementary Table S1 to every printed digit.

Simulation helper

mw <- c(gli = 490.616, m1 = 506.62, m2 = 520.6)

# One event table for any number of subjects. Observations sit on the `venous`
# ODE state; rxode2 returns the algebraic outputs (Cc, Cc_m1, Cc_m2,
# Aurine_m1m2, ...) as columns on those rows. Time is in minutes.
sim_gli <- function(subjects, hours = 48, by_min = 15) {
  tt <- seq(0, hours * 60, by = by_min)
  ev <- bind_rows(lapply(seq_len(nrow(subjects)), function(i) {
    s <- subjects[i, ]
    bind_rows(
      data.frame(id = i, time = 0, amt = s$amt, evid = 1L, cmt = s$route),
      data.frame(id = i, time = tt, amt = 0, evid = 0L, cmt = "venous")
    ) |>
      mutate(WT = s$WT, RENALFUNC_REL = s$RENALFUNC_REL,
             HEPFUNC_REL = s$HEPFUNC_REL, CYP2C9 = s$CYP2C9)
  }))
  # The model has no etas, so rxode2 warns that a multi-subject solve has no
  # omega; every subject is a deterministic scenario, which is intended.
  out <- withCallingHandlers(
    rxode2::rxSolve(mod, ev, returnType = "data.frame",
                    atol = 1e-12, rtol = 1e-10),
    warning = function(w) {
      if (grepl("without 'omega'", conditionMessage(w), fixed = TRUE)) {
        invokeRestart("muffleWarning")
      }
    }
  )
  # rxode2 drops the id column when only one subject is solved.
  if (!"id" %in% names(out)) out$id <- 1L
  out |>
    left_join(mutate(subjects, id = seq_len(n())), by = "id",
              suffix = c("", ".cov")) |>
    mutate(hour = time / 60)
}

subject <- function(amt, route = "depot", WT = 75, RENALFUNC_REL = 1,
                    HEPFUNC_REL = 1, CYP2C9 = 1, label = NA_character_) {
  data.frame(amt = amt, route = route, WT = WT, RENALFUNC_REL = RENALFUNC_REL,
             HEPFUNC_REL = HEPFUNC_REL, CYP2C9 = CYP2C9, label = label)
}

auc_trap <- function(t, y) sum(diff(t) * (head(y, -1) + tail(y, -1)) / 2)

Agreement with the authors’ SBML

The decisive check is that this translation reproduces the authors’ own model. The maintainers simulated the archived SBML (glimepiride_body_flat.xml, v0.6.1) in libroadrunner 2.11.0 for nine scenarios covering every dosing route and every scenario covariate, and recorded the 0-48 h maximum venous concentration of each species (on a 15-minute grid) and the 48-hour cumulative urinary recovery of M1 + M2. The same scenarios are solved here.

sbml_ref <- tibble::tribble(
  ~label,              ~cmax_gli, ~cmax_m1,  ~cmax_m2,   ~urine48,
  "PO 4 mg",             0.61412,  0.13026,  0.065808,    3.9265,
  "PO 1 mg",             0.14649,  0.036309, 0.017927,    1.0017,
  "PO 8 mg",             1.2877,   0.22697,  0.11761,     7.6442,
  "IV 1 mg",             0.27741,  0.043921, 0.01975,     0.99554,
  "IV M1 1.5 mg",        0,        0.3795,   0.038029,    1.8036,
  "PO 1 mg, CTP C",      0.23333,  0.01294,  0.0038541,   0.54932,
  "PO 3 mg, severe RI",  0.45403,  0.14468,  0.085931,    1.0733,
  "PO 4 mg, *3/*3",      0.80087,  0.039058, 0.021142,    2.2659,
  "PO 8 mg, 120 kg",     0.81921,  0.17192,  0.091591,    8.5573
)

scen <- bind_rows(
  subject(4, label = "PO 4 mg"),
  subject(1, label = "PO 1 mg"),
  subject(8, label = "PO 8 mg"),
  subject(1, route = "depot_iv", label = "IV 1 mg"),
  subject(1.5, route = "depot_iv_m1", label = "IV M1 1.5 mg"),
  subject(1, HEPFUNC_REL = 1 - 0.8127551020408164, label = "PO 1 mg, CTP C"),
  subject(3, RENALFUNC_REL = 0.19, label = "PO 3 mg, severe RI"),
  subject(4, CYP2C9 = 0.23, label = "PO 4 mg, *3/*3"),
  subject(8, WT = 120, label = "PO 8 mg, 120 kg")
)
sim_scen <- sim_gli(scen)

ours <- sim_scen |>
  group_by(label) |>
  summarise(cmax_gli = max(Cc), cmax_m1 = max(Cc_m1), cmax_m2 = max(Cc_m2),
            urine48 = 1000 * Aurine_m1m2[which.max(time)], .groups = "drop")

cmp <- inner_join(sbml_ref, ours, by = "label", suffix = c("_sbml", "_rx"))
rel_diff <- function(a, b) ifelse(b == 0, abs(a), abs(a - b) / b)
cmp <- cmp |>
  mutate(
    d_gli = rel_diff(cmax_gli_rx, cmax_gli_sbml),
    d_m1 = rel_diff(cmax_m1_rx, cmax_m1_sbml),
    d_m2 = rel_diff(cmax_m2_rx, cmax_m2_sbml),
    d_urine = rel_diff(urine48_rx, urine48_sbml)
  )

cmp |>
  select(label, cmax_gli_sbml, cmax_gli_rx, cmax_m1_sbml, cmax_m1_rx,
         urine48_sbml, urine48_rx) |>
  rename("Scenario" = label,
         "GLI Cmax, SBML (umol/L)" = cmax_gli_sbml,
         "GLI Cmax, nlmixr2lib (umol/L)" = cmax_gli_rx,
         "M1 Cmax, SBML (umol/L)" = cmax_m1_sbml,
         "M1 Cmax, nlmixr2lib (umol/L)" = cmax_m1_rx,
         "M1+M2 urine 48 h, SBML (umol)" = urine48_sbml,
         "M1+M2 urine 48 h, nlmixr2lib (umol)" = urine48_rx) |>
  knitr::kable(digits = 4)
Scenario GLI Cmax, SBML (umol/L) GLI Cmax, nlmixr2lib (umol/L) M1 Cmax, SBML (umol/L) M1 Cmax, nlmixr2lib (umol/L) M1+M2 urine 48 h, SBML (umol) M1+M2 urine 48 h, nlmixr2lib (umol)
PO 4 mg 0.6141 0.6141 0.1303 0.1303 3.9265 3.9265
PO 1 mg 0.1465 0.1465 0.0363 0.0363 1.0017 1.0017
PO 8 mg 1.2877 1.2877 0.2270 0.2270 7.6442 7.6443
IV 1 mg 0.2774 0.2774 0.0439 0.0439 0.9955 0.9955
IV M1 1.5 mg 0.0000 0.0000 0.3795 0.3795 1.8036 1.8037
PO 1 mg, CTP C 0.2333 0.2333 0.0129 0.0129 0.5493 0.5493
PO 3 mg, severe RI 0.4540 0.4540 0.1447 0.1447 1.0733 1.0733
PO 4 mg, 3/3 0.8009 0.8009 0.0391 0.0391 2.2659 2.2659
PO 8 mg, 120 kg 0.8192 0.8192 0.1719 0.1719 8.5573 8.5573

# The two sides solve the same equations with the same parameters (here
# rounded to 5 significant digits), so the only difference is numerical; a
# transcription error in any rate law moves at least one of these columns by
# whole percent.
stopifnot(max(cmp$d_gli, cmp$d_m1, cmp$d_m2, cmp$d_urine) < 1e-3)

Dose dependency (Figure 2)

doses <- c(1, 2, 3, 4, 6, 8)
sim_dose <- sim_gli(bind_rows(lapply(doses, function(d) subject(d, label = paste(d, "mg")))),
                    hours = 24, by_min = 5)

sim_dose |>
  tidyr::pivot_longer(c(Cc, Cc_m1, Cc_m2), names_to = "species", values_to = "conc") |>
  mutate(species = recode(species, Cc = "Glimepiride", Cc_m1 = "M1", Cc_m2 = "M2"),
         dose = factor(amt)) |>
  ggplot(aes(hour, conc, colour = dose)) +
  geom_line() +
  facet_wrap(~species, scales = "free_y") +
  labs(x = "Time (h)", y = "Plasma concentration (umol/L)", colour = "Dose (mg)",
       caption = "Replicates Figure 2B/Supplementary Figure S4 of Elias 2025 (typical 75 kg individual).")

dose_tab <- sim_dose |>
  group_by(amt) |>
  summarise(
    cmax_ng = max(Cc) * mw[["gli"]],
    tmax_h = hour[which.max(Cc)],
    auc24_ng = auc_trap(hour, Cc * mw[["gli"]]),
    cmax_m1_ng = max(Cc_m1) * mw[["m1"]],
    cmax_m2_ng = max(Cc_m2) * mw[["m2"]],
    .groups = "drop"
  ) |>
  mutate(cmax_per_mg = cmax_ng / amt, auc_per_mg = auc24_ng / amt)

dose_tab |>
  rename("Dose (mg)" = amt, "GLI Cmax (ng/mL)" = cmax_ng, "GLI tmax (h)" = tmax_h,
         "GLI AUC0-24 (ng*h/mL)" = auc24_ng, "M1 Cmax (ng/mL)" = cmax_m1_ng,
         "M2 Cmax (ng/mL)" = cmax_m2_ng, "Cmax / dose" = cmax_per_mg,
         "AUC0-24 / dose" = auc_per_mg) |>
  knitr::kable(digits = 2)
Dose (mg) GLI Cmax (ng/mL) GLI tmax (h) GLI AUC0-24 (ng*h/mL) M1 Cmax (ng/mL) M2 Cmax (ng/mL) Cmax / dose AUC0-24 / dose
1 71.95 2.17 373.41 18.40 9.33 71.95 373.41
2 146.22 2.17 770.07 35.47 18.14 73.11 385.04
3 222.75 2.25 1189.86 51.33 26.44 74.25 396.62
4 301.29 2.25 1632.55 66.05 34.27 75.32 408.14
6 463.94 2.33 2585.28 92.37 48.56 77.32 430.88
8 632.94 2.33 3624.92 115.08 61.22 79.12 453.11

urine_8mg <- 1000 * sim_gli(subject(8), hours = 48)$Aurine_m1m2 |> max()
urine_8mg
#> [1] 7.644275

Section 2.3 reports Tmax “stable at 2.0-2.5 h”, Cmax and AUC increasing proportionally with dose, and cumulative M1 + M2 urinary excretion “reaching about 7 umol by 48 h for the 8 mg dose”. Figure 2C’s simulated line passes about 640 ng/mL at 8 mg (the text’s “700 ng/mL” is the line’s end beyond the last dose). Glimepiride conversion to M1 is Michaelis-Menten, so exposure is very slightly more than dose-proportional, as Figure 2C also shows.

stopifnot(
  all(dose_tab$tmax_h >= 2.0 & dose_tab$tmax_h <= 2.5),
  # Near-linear: Cmax per mg varies by under 10% across 1-8 mg.
  diff(range(dose_tab$cmax_per_mg)) / mean(dose_tab$cmax_per_mg) < 0.10,
  abs(dose_tab$cmax_ng[dose_tab$amt == 8] / 640 - 1) < 0.05,
  abs(urine_8mg / 7 - 1) < 0.15
)

Renal impairment (Figure 3)

Methods 4.2 maps the KDIGO categories onto RENALFUNC_REL: normal 1.0, mild 0.69, moderate 0.32, severe 0.19. Renal function scales only the urinary excretion of M1 and M2.

renal_levels <- c(Normal = 1, Mild = 0.69, Moderate = 0.32, Severe = 0.19)
sim_renal <- sim_gli(bind_rows(lapply(names(renal_levels), function(n)
  subject(3, RENALFUNC_REL = renal_levels[[n]], label = n))))
sim_renal$label <- factor(sim_renal$label, levels = names(renal_levels))

sim_renal |>
  mutate(`M1 + M2 urine (umol)` = 1000 * Aurine_m1m2,
         `M1 plasma (umol/L)` = Cc_m1) |>
  tidyr::pivot_longer(c(`M1 plasma (umol/L)`, `M1 + M2 urine (umol)`)) |>
  ggplot(aes(hour, value, colour = label)) +
  geom_line() +
  facet_wrap(~name, scales = "free_y") +
  labs(x = "Time (h)", y = NULL, colour = "Renal function",
       caption = "Replicates Figure 3B of Elias 2025 (3 mg oral).")


renal_tab <- sim_renal |>
  group_by(label) |>
  summarise(cmax_gli = max(Cc), auc_m1 = auc_trap(hour, Cc_m1),
            urine48 = 1000 * max(Aurine_m1m2), .groups = "drop")
knitr::kable(renal_tab |>
  rename("Renal function" = label, "GLI Cmax (umol/L)" = cmax_gli,
         "M1 AUC0-48 (umol*h/L)" = auc_m1, "M1+M2 urine 48 h (umol)" = urine48),
  digits = 3)
Renal function GLI Cmax (umol/L) M1 AUC0-48 (umol*h/L) M1+M2 urine 48 h (umol)
Normal 0.454 0.747 2.965
Mild 0.454 0.879 2.505
Moderate 0.454 1.123 1.581
Severe 0.454 1.246 1.073

Section 2.4: parent exposure is unchanged, metabolites accumulate, and the 48-hour urinary recovery falls from about 3 umol (normal) to about 1 umol (severe impairment) after 3 mg.

stopifnot(
  # Renal function never touches the parent: identical to numerical precision.
  diff(range(renal_tab$cmax_gli)) / mean(renal_tab$cmax_gli) < 1e-6,
  all(diff(renal_tab$auc_m1) > 0),
  abs(renal_tab$urine48[1] / 3 - 1) < 0.10,
  abs(renal_tab$urine48[4] / 1 - 1) < 0.15
)

Hepatic impairment (Figure 4)

The cirrhosis severity f_cirrhosis of the source is 1 - HEPFUNC_REL. The authors’ archived scenarios use f_cirrhosis = 0.3995 (CTP A), 0.6980 (CTP B) and 0.8128 (CTP C), following Koeller et al. (2021).

cirr_levels <- c(Control = 0, `CTP A` = 0.3994897959183674,
                 `CTP B` = 0.6979591836734694, `CTP C` = 0.8127551020408164)
sim_hep <- sim_gli(bind_rows(lapply(names(cirr_levels), function(n)
  subject(1, HEPFUNC_REL = 1 - cirr_levels[[n]], label = n))))
sim_hep$label <- factor(sim_hep$label, levels = names(cirr_levels))

sim_hep |>
  tidyr::pivot_longer(c(Cc, Cc_m1)) |>
  mutate(name = recode(name, Cc = "Glimepiride", Cc_m1 = "M1")) |>
  ggplot(aes(hour, value, colour = label)) +
  geom_line() +
  facet_wrap(~name, scales = "free_y") +
  labs(x = "Time (h)", y = "Plasma concentration (umol/L)", colour = NULL,
       caption = "Replicates Figure 4B of Elias 2025 (1 mg oral).")


hep_tab <- sim_hep |>
  group_by(label) |>
  summarise(cmax_ng = max(Cc) * mw[["gli"]], auc48 = auc_trap(hour, Cc * mw[["gli"]]),
            cmax_m1 = max(Cc_m1), .groups = "drop") |>
  mutate(auc_ratio = auc48 / auc48[1])
knitr::kable(hep_tab |>
  rename("Liver function" = label, "GLI Cmax (ng/mL)" = cmax_ng,
         "GLI AUC0-48 (ng*h/mL)" = auc48, "M1 Cmax (umol/L)" = cmax_m1,
         "AUC ratio vs control" = auc_ratio),
  digits = 2)
Liver function GLI Cmax (ng/mL) GLI AUC0-48 (ng*h/mL) M1 Cmax (umol/L) AUC ratio vs control
Control 71.87 388.40 0.04 1.00
CTP A 88.00 584.24 0.03 1.50
CTP B 105.69 916.02 0.02 2.36
CTP C 114.47 1159.27 0.01 2.98

Section 2.5: Cmax “nearly doubled from 75 ng/mL in normal function to 125 ng/mL in severe cirrhosis, while AUC increased … by approximately 3.5-fold”, and metabolite concentrations fall.

stopifnot(
  abs(hep_tab$cmax_ng[1] / 75 - 1) < 0.10,
  abs(hep_tab$cmax_ng[4] / 125 - 1) < 0.15,
  hep_tab$auc_ratio[4] > 2.5, hep_tab$auc_ratio[4] < 4.5,
  all(diff(hep_tab$cmax_m1) < 0)
)

Body weight (Figure 5)

bws <- c(40, 55, 72, 100, 130, 170)
sim_bw <- sim_gli(bind_rows(lapply(bws, function(w) subject(8, WT = w, label = paste(w, "kg")))))
bw_tab <- sim_bw |>
  group_by(WT) |>
  summarise(cmax_ng = max(Cc) * mw[["gli"]], cmax_um = max(Cc),
            auc48 = auc_trap(hour, Cc * mw[["gli"]]),
            tmax_h = hour[which.max(Cc)], .groups = "drop")

ggplot(bw_tab, aes(WT, cmax_ng)) +
  geom_line() + geom_point() +
  labs(x = "Body weight (kg)", y = "Glimepiride Cmax (ng/mL)",
       caption = "Replicates Figure 5C (Cmax panel) of Elias 2025 (8 mg oral).")


knitr::kable(bw_tab |>
  rename("Body weight (kg)" = WT, "GLI Cmax (ng/mL)" = cmax_ng,
         "GLI Cmax (umol/L)" = cmax_um, "GLI AUC0-48 (ng*h/mL)" = auc48,
         "GLI tmax (h)" = tmax_h),
  digits = 2)
Body weight (kg) GLI Cmax (ng/mL) GLI Cmax (umol/L) GLI AUC0-48 (ng*h/mL) GLI tmax (h)
40 1117.44 2.28 6712.57 2.25
55 842.84 1.72 5069.80 2.25
72 656.51 1.34 3928.12 2.25
100 479.96 0.98 2834.12 2.50
130 371.56 0.76 2168.80 2.50
170 285.02 0.58 1645.74 2.50

Section 2.6: Cmax falls from about 1,000 ng/mL at 40 kg to 300 ng/mL at 170 kg and AUC from about 6,000 to 2,000 ng*h/mL, with Tmax stable. For the Shukla 2004 cohorts (72 kg and 130 kg in the authors’ simulation set-up) the text quotes peaks of “1.4” and “0.8”; Figure 5B shows these are umol/L, not ug/mL.

stopifnot(
  abs(bw_tab$cmax_ng[bw_tab$WT == 40] / 1000 - 1) < 0.15,
  abs(bw_tab$cmax_ng[bw_tab$WT == 170] / 300 - 1) < 0.15,
  abs(bw_tab$cmax_um[bw_tab$WT == 72] / 1.4 - 1) < 0.10,
  abs(bw_tab$cmax_um[bw_tab$WT == 130] / 0.8 - 1) < 0.10,
  diff(range(bw_tab$tmax_h)) <= 0.5
)

CYP2C9 genotype (Figure 6)

Allele activities are 1 = 1.00, 2 = 0.68, *3 = 0.23 (Supplementary Table S2), and the diplotype activity is the mean of the two alleles.

allele <- c(`*1` = 1, `*2` = 0.68, `*3` = 0.23)
genotypes <- c("*1/*1", "*1/*2", "*1/*3", "*3/*3")
geno_act <- sapply(strsplit(genotypes, "/"), function(a) mean(allele[a]))
names(geno_act) <- genotypes
geno_act
#> *1/*1 *1/*2 *1/*3 *3/*3 
#> 1.000 0.840 0.615 0.230

sim_cyp <- sim_gli(bind_rows(lapply(genotypes, function(g)
  subject(4, CYP2C9 = geno_act[[g]], label = g))))
sim_cyp$label <- factor(sim_cyp$label, levels = genotypes)

ggplot(sim_cyp, aes(hour, Cc, colour = label)) +
  geom_line() +
  labs(x = "Time (h)", y = "Glimepiride (umol/L)", colour = "CYP2C9",
       caption = "Replicates Figure 6B of Elias 2025 (4 mg oral, fixed activities).")


cyp_tab <- sim_cyp |>
  group_by(label) |>
  summarise(auc48 = auc_trap(hour, Cc * mw[["gli"]]), .groups = "drop") |>
  mutate(ratio = auc48 / auc48[1])
cyp_tab
#> # A tibble: 4 × 3
#>   label auc48 ratio
#>   <fct> <dbl> <dbl>
#> 1 *1/*1 1699.  1   
#> 2 *1/*2 1955.  1.15
#> 3 *1/*3 2473.  1.46
#> 4 *3/*3 4382.  2.58

stopifnot(
  isTRUE(all.equal(unname(geno_act), c(1, 0.84, 0.615, 0.23))),
  all(diff(cyp_tab$auc48) > 0),
  # "approximately two-fold" (Discussion) to "up to 2.5-fold" (Section 2.7);
  # the 0-48 h ratio of the typical individual is about 2.6.
  cyp_tab$ratio[4] > 1.8, cyp_tab$ratio[4] < 2.8
)

Probabilistic CYP2C9 activity

For the population analyses each allele activity was drawn from a lognormal distribution with shape sigma = 0.81 (fitted to diclofenac intrinsic clearance in human liver microsomes, Supplementary Table S4) scaled so that its mean is the allele activity; the diplotype activity is the mean of the two draws. Reproducing the genotype medians of Supplementary Table S5 checks that reading.

set.seed(20251006)
draw_geno <- function(g, n) {
  a <- strsplit(g, "/")[[1]]
  draw <- function(m) rlnorm(n, meanlog = log(m) - 0.81^2 / 2, sdlog = 0.81)
  (draw(allele[[a[1]]]) + draw(allele[[a[2]]])) / 2
}
big <- sapply(genotypes, draw_geno, n = 1e5)
act_tab <- data.frame(genotype = genotypes, mean = colMeans(big),
                      median = apply(big, 2, median))
act_tab$median_tableS5 <- c(0.83, 0.70, 0.48, 0.19)
knitr::kable(act_tab |> rename("Genotype" = genotype, "Mean" = mean,
                               "Median" = median, "Table S5 median" = median_tableS5),
             digits = 3)
Genotype Mean Median Table S5 median
1/1 1/1 1.001 0.829 0.83
1/2 1/2 0.840 0.691 0.70
1/3 1/3 0.612 0.480 0.48
3/3 3/3 0.229 0.190 0.19
stopifnot(all(abs(act_tab$median - act_tab$median_tableS5) < 0.03))

A virtual cohort of 150 subjects per genotype (4 mg oral, 75 kg) then gives the distribution of glimepiride exposure, which the NCA below compares with Supplementary Table S6.

n_per <- 150
cohort <- bind_rows(lapply(genotypes, function(g)
  data.frame(genotype = g, CYP2C9 = draw_geno(g, n_per))))
sim_pop <- sim_gli(
  bind_rows(lapply(seq_len(nrow(cohort)), function(i)
    subject(4, CYP2C9 = cohort$CYP2C9[i], label = cohort$genotype[i]))),
  hours = 24, by_min = 15
)
sim_pop$genotype <- factor(sim_pop$label, levels = genotypes)

sim_pop |>
  group_by(id, genotype) |>
  summarise(auc24 = auc_trap(hour, Cc * mw[["gli"]]), .groups = "drop") |>
  ggplot(aes(genotype, auc24, fill = genotype)) +
  geom_boxplot(show.legend = FALSE) +
  labs(x = "CYP2C9 genotype", y = "Glimepiride AUC0-24 (ng*h/mL)",
       caption = "Replicates Figure 6D (AUC panel) of Elias 2025.")

Non-compartmental analysis

PKNCA on the probabilistic cohort, grouped by genotype, against the genotype medians of Supplementary Table S6 (4 mg oral; Cmax converted from umol/L to ng/mL). Table S6 does not state its AUC interval. Its AUC values match the 0-24 h AUC of this cohort – medians, upper quartiles and maxima alike, the maxima levelling off near 5,000 ng*h/mL for every genotype as a truncated AUC does – and fall 20-30% short of AUC0-inf for the low-activity genotypes, whose terminal phase is longest, so the comparison uses AUC0-24.

conc <- sim_pop |>
  filter(!is.na(Cc)) |>
  transmute(id, genotype, hour, conc = Cc * mw[["gli"]])
dose_df <- conc |>
  distinct(id, genotype) |>
  mutate(hour = 0, dose = 4)

o_conc <- PKNCA::PKNCAconc(as.data.frame(conc), conc ~ hour | genotype + id)
o_dose <- PKNCA::PKNCAdose(as.data.frame(dose_df), dose ~ hour | genotype + id)
nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  o_conc, o_dose,
  intervals = data.frame(start = 0, end = 24, cmax = TRUE, tmax = TRUE,
                         auclast = TRUE)
))

ref_s6 <- data.frame(
  genotype = genotypes,
  cmax = c(0.645, 0.677, 0.724, 0.809) * mw[["gli"]],
  tmax = c(2.35, 2.44, 2.58, 2.88),
  auclast = c(1864.73, 2145.33, 2644.96, 3933.03)
)

nca_cmp <- nlmixr2lib::ncaComparisonTable(
  nca, ref_s6,
  by = "genotype",
  units = c(cmax = "ng/mL", tmax = "h", auclast = "ng*h/mL")
)
knitr::kable(nca_cmp)
NCA parameter genotype Reference Simulated % diff
Cmax (ng/mL) 1/1 316 314 -0.8%
Cmax (ng/mL) 1/2 332 335 +0.9%
Cmax (ng/mL) 1/3 355 364 +2.6%
Cmax (ng/mL) 3/3 397 396 -0.2%
Tmax (h) 1/1 2.35 2.25 -4.3%
Tmax (h) 1/2 2.44 2.5 +2.5%
Tmax (h) 1/3 2.58 2.75 +6.6%
Tmax (h) 3/3 2.88 3 +4.2%
AUClast (ng*h/mL) 1/1 1860 1830 -1.9%
AUClast (ng*h/mL) 1/2 2150 2200 +2.4%
AUClast (ng*h/mL) 1/3 2640 2880 +8.9%
AUClast (ng*h/mL) 3/3 3930 3890 -1.1%
attr(nca_cmp, "footnote")
#> NULL

nca_med <- as.data.frame(nca$result) |>
  group_by(genotype, PPTESTCD) |>
  summarise(med = median(PPORRES), .groups = "drop") |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = med) |>
  mutate(genotype = as.character(genotype)) |>
  inner_join(ref_s6, by = "genotype", suffix = c("_sim", "_ref"))

stopifnot(
  all(abs(nca_med$cmax_sim / nca_med$cmax_ref - 1) < 0.10),
  all(abs(nca_med$tmax_sim - nca_med$tmax_ref) <= 0.5),
  all(abs(nca_med$auclast_sim / nca_med$auclast_ref - 1) < 0.15)
)

Assumptions and deviations / Errata

  • Whole-body ODEs and physiology from the archived SBML. The paper prints the submodel rate laws (Supplementary Section S3, partly as equation images) and the eleven optimized parameters (Table S1), but not the circulation ODEs, fractional organ volumes and blood flows, cardiac output, hematocrit, molecular weights, the dissolution rate (2 /h), the hepatic import rate (100 /min), the CYP2C9 Michaelis constant (2.25 umol/L) or the intravenous release time. These come from the authors’ archived SBML v0.6.1, which Methods 4.2 cites as the model (“version 0.6.1 (Elias and Koenig, 2025a)”). The archive’s optimized values agree with Table S1 to every printed digit.
  • Parameter precision. Table S1 prints three parameters to only one or two significant digits (GLI2M1 Vmax 0.00005, ftissue gli 0.00071, MEXC k 0.00017). The model uses the archive’s values (5.362e-5, 7.0864e-4, 1.7193e-4); the printed 0.00005 would lower the CYP2C9 Vmax by 7%.
  • Deterministic model. No between-subject variability or residual error is reported, so there are no etas and propSd is fixed at zero. The probabilistic CYP2C9 analysis is reproduced in this vignette, outside the model, by drawing the CYP2C9 covariate.
  • Lognormal CYP2C9 draws. The paper states the shape (0.81) and that the scale was “adjusted to match the mean activity of each allele”. Drawing each allele with mean equal to its activity reproduces the genotype means, medians and SDs of Supplementary Table S5 (e.g. 1/1 mean 1.00, median 0.83), confirming the reading. Table S5’s allele-level SD (0.73 for *1) does not match a sigma = 0.81 lognormal (which has CV 0.96); the genotype-level rows are reproduced and were used.
  • Renal scaling values. Methods 4.2 prints f_renal_function = 0.69 / 0.32 / 0.19 for mild / moderate / severe impairment, and those values are used here. The authors’ archived scenario simulations use 0.495 / 0.347 / 0.198 instead (50, 35 and 20 of a 101 mL/min reference), and their Rosenkranz 1996 study simulation uses 0.5 / 0.35 / 0.2. Only metabolite exposure depends on this choice.
  • Child-Turcotte-Pugh mapping. The paper does not print the f_cirrhosis values; 0.3995 / 0.6980 / 0.8128 for CTP A / B / C are the authors’ archived scenario values (following Koeller et al. 2021).
  • Body-weight peak units. Section 2.6 quotes Shukla 2004 peaks of “1.4 ug/mL” and “0.8 ug/mL”; Figure 5B shows them in umol/L, and the model reproduces them in umol/L.
  • Half-life. Rest-of-body tissue equilibrates slowly (ftissue_gli 0.00071 L/min against a tissue volume of about 70 L with Kp 10), so the venous profile is multi-phasic and an NCA half-life depends on the fitting window. The paper’s “approximately 4 h” refers to the early elimination phase; it is not used as a gate here.
  • Omitted SBML bookkeeping. The source carries states with zero derivatives (Afeces_gli, GU__gli_stomach), cumulative-infusion counters fed by zero-valued infusion-rate parameters (Ri_*, cum_dose_*) and a display-only creatinine-clearance output (KI__crcl, which needs height). None affects any concentration; they are omitted. Intravenous infusions can be given with an rxode2 rate on the depot_iv* compartments.
  • Food effect. f_absorption scales the absorption rate; the paper gives no fed-state value, so it is held at 1 (fasted, the condition of the fitting data).
  • Table S6 AUC interval. The supplement does not say which AUC Table S6 reports. The 0-24 h AUC reproduces its genotype medians, quartiles and maxima; AUC0-inf does not for the 1/3 and 3/3 genotypes, so AUC0-24 is used in the comparison.
  • Biogeographical populations (Figure 7) apply the same probabilistic CYP2C9 sampling with published genotype frequencies; they are not repeated here.