Skip to contents

Pouzin_2022_tusamitamab

Source: vignettes/articles/Pouzin_2022_tusamitamab.Rmd
Pouzin_2022_tusamitamab.Rmd

Model and source

  • Citation: Pouzin C, Gibiansky L, Fagniez N, Tod M, Chadjaa M, Nguyen L. Integrated multiple analytes and semi-mechanistic population pharmacokinetic model of tusamitamab ravtansine, a DM4 anti-CEACAM5 antibody-drug conjugate. J Pharmacokinet Pharmacodyn. 2022;49(4):381–394.
  • Article: https://doi.org/10.1007/s10928-021-09799-0
  • Online Resource 1 (Monolix .mlxtran): supplementary material to the article.
  • First-in-human study: TED13751 (NCT02187848).

Pouzin_2022_tusamitamab packages the integrated semi-mechanistic multi-analyte popPK model of tusamitamab ravtansine (SAR408701, an anti-CEACAM5 IgG1-SPDB-DM4 ADC) developed from the TED13751 first-in-human study. The model fits SAR408701 (DAR≥1), naked antibody (NAB), DM4, and MeDM4 plasma data simultaneously, with an explicit two-compartment chain for each DAR0–DAR8 species. DAR_n → DAR_(n-1) deconjugation in the central compartment is modelled as an irreversible first-order process and feeds a one-compartment DM4 catabolite that converts to MeDM4.

Population

The TED13751 study enrolled 254 adults with advanced solid tumors expressing CEACAM5 (escalation cohorts at 5–190 mg/m² Q2W or Q3W and expansion cohorts at 100 mg/m² Q2W in colorectal, gastric, NSCLC high/low CEACAM5, and SCLC). Per Methods, 3746 SAR408701, 3740 DM4, 3734 MeDM4, and 3734 NAB plasma concentrations were modelled across up to 58 cycles, with DAR distributions quantified by LC-HRMS in 13 patients (Pouzin 2022, Table 1, Methods § “Clinical study”, “Bioanalytical methods”). The Pouzin 2022 modelling paper does not tabulate baseline demographics for the 254-patient analysis set; the companion covariate paper (Pouzin et al., CPT Pharmacometrics Syst Pharmacol 2022, doi:10.1002/psp4.12769) reports the full demographic table and is the basis for any covariate-aware refit.

The same metadata is available programmatically via readModelDb("Pouzin_2022_tusamitamab")$meta$population.

Source trace

Every ini() parameter carries an in-file comment pointing back to the source. The table below collects those pointers in one place.

Equation / parameter Value Source location
Vc (central volume) 3.37 L Pouzin 2022 Table 4
Vp (peripheral volume) 2.54 L Pouzin 2022 Table 4
Q (inter-compartmental CL) 0.543 L/day Pouzin 2022 Table 4
CL_ADC (proteolytic CL of DAR≥1) 0.392 L/day Pouzin 2022 Table 4
CL_NAB 0.408 L/day Pouzin 2022 Table 4
CL_DM4 (apparent; V_DM4=1 L) 240 L/day Pouzin 2022 Table 4 + Methods § “DM4 and MeDM4”
CL_MeDM4 (apparent; V_MeDM4=1 L) 0.256 L/day Pouzin 2022 Table 4 + Methods § “DM4 and MeDM4”
V_DM4, V_MeDM4 (fixed) 1 L each (FIX) Pouzin 2022 Methods § “DM4 and MeDM4”
FR_MeDM4 0.0107 Pouzin 2022 Table 4
kdec1 (DAR1→NAB) 0.0565 /day Pouzin 2022 Table 4
kdec2 0.181 /day Pouzin 2022 Table 4
kdec3 0.340 /day Pouzin 2022 Table 4
kdec4 0.525 /day Pouzin 2022 Table 4
kdec5 0.751 /day Pouzin 2022 Table 4
kdec6, kdec7, kdec8 0.938 /day each Pouzin 2022 Table 4 + Methods (DAR7/DAR8 not separately identifiable)
F_DAR1..F_DAR8 (administered fractions) 0.9–21.8 % (Table 4) Pouzin 2022 Tables 3 & 4 (fixed at batch median)
F_NAB 7.1% (estimated) Pouzin 2022 Table 4
Residual SDs (a_ADC, b_ADC, b_NAB, b_DM4, b_MeDM4, a_DARavg) 1.03 µg/mL, 8.9%, 26.0%, 33.5%, 50.0%, 0.219 Pouzin 2022 Table 4
ODE structure (DAR0–DAR8 chains, 1:1 deconjugation, DM4 ⇄ MeDM4) n/a Pouzin 2022 Online Resource 1 (Monolix .mlxtran); main-text Methods § “Model development”

The published additive ADC residual SD of 1.03 µg/mL is converted to µM in ini() by dividing by the SAR408701 molecular weight (150 000 g/mol) and multiplying by 1000 mL/L: 1.03 / 150000 * 1000 = 0.006867 µM.

Virtual cohort and simulation

The packaged model carries no covariates (Pouzin 2022 reports the structural base; covariates are evaluated in the companion CPT-PSP paper). A typical- value simulation therefore only needs a single subject’s worth of dosing.

Pouzin 2022 reports typical exposures for “100 mg/m² Q2W” with no body surface area specified. The Cmax / AUC values in Table 7 reproduce exactly when assuming BSA = 1.79 m² (i.e., typical 179 mg total dose), which is consistent with the dose-finding cohort weight ranges reported in the companion covariate paper. We use BSA = 1.79 m² throughout this vignette.

# 100 mg/m^2 Q2W, BSA assumed = 1.79 m^2 (typical adult)
bsa_m2     <- 1.79
dose_mg_m2 <- 100
mw_adc     <- 150000           # g/mol — Pouzin 2022 Table 2
dose_mg    <- dose_mg_m2 * bsa_m2
dose_umol  <- dose_mg / (mw_adc / 1000)   # mg / (mg/µmol)
sprintf("dose = %.1f mg of ADC = %.4f µmol total antibody", dose_mg, dose_umol)
#> [1] "dose = 179.0 mg of ADC = 1.1933 µmol total antibody"
# Build a 9-row IV dose event for one administration: each chain receives
# the *total* antibody dose, and the model multiplies by f<chain>
# (= F_DARi / SUM) inside model() to deposit only that chain's fraction.
ChainsCentral <- c("dar1_central", "dar2_central", "dar3_central",
                   "dar4_central", "dar5_central", "dar6_central",
                   "dar7_central", "dar8_central", "nab_central")

build_q2w_dosing <- function(dose_umol, n_cycles, chains = ChainsCentral) {
  ev <- rxode2::et()
  for (cycle in seq_len(n_cycles)) {
    t0 <- (cycle - 1) * 14
    for (cmt in chains) {
      ev <- rxode2::et(ev, amt = dose_umol, cmt = cmt, time = t0)
    }
  }
  ev
}

obs_grid <- c(0.001, 0.01, 0.05, 0.1, 0.25, 0.5, 0.75,
              seq(1, 14 * 6, by = 0.25))   # 6 Q2W cycles, dense early
ev_typ <- build_q2w_dosing(dose_umol, n_cycles = 6) |>
  rxode2::et(obs_grid, cmt = "Cc")

mod    <- readModelDb("Pouzin_2022_tusamitamab")
mod0   <- rxode2::zeroRe(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'
sim    <- rxode2::rxSolve(mod0, events = ev_typ, atol = 1e-12, rtol = 1e-10)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalvp', 'etalq', 'etalcladc', 'etalclnab', 'etalkdec1', 'etalcldm4', 'etalclmedm4', 'etalfrmedm4', 'etalfrdar1', 'etalfrdar2', 'etalfrdar3', 'etalfrdar4', 'etalfrdar5', 'etalfrdar6', 'etalfrdar7', 'etalfrnab'
df     <- as.data.frame(sim)

Replicate Figure 3 — typical SAR408701, NAB, DM4 and MeDM4 profile

Pouzin 2022 Figure 3 shows the typical concentration-time profile of all four analytes after 100 mg/m² Q2W dosing.

sim_long <- df |>
  select(time, Cc, Cnab, Cdm4, Cmedm4) |>
  pivot_longer(c(Cc, Cnab, Cdm4, Cmedm4), names_to = "analyte", values_to = "C_uM") |>
  mutate(analyte = recode(analyte,
                          Cc      = "SAR408701 (DAR>=1)",
                          Cnab    = "NAB (DAR0)",
                          Cdm4    = "DM4",
                          Cmedm4  = "MeDM4"),
         analyte = factor(analyte,
                          levels = c("SAR408701 (DAR>=1)", "NAB (DAR0)", "DM4", "MeDM4")))

ggplot(sim_long, aes(time, C_uM, colour = analyte)) +
  geom_line(linewidth = 0.8) +
  scale_y_log10() +
  scale_x_continuous(breaks = seq(0, 84, by = 14)) +
  labs(x = "Time (days)", y = "Concentration (µM)", colour = NULL,
       title = "Tusamitamab ravtansine — typical four-analyte profile",
       caption = "Replicates Figure 3 of Pouzin 2022 (100 mg/m² Q2W, BSA = 1.79 m²).")
Replicates Figure 3 of Pouzin 2022 — typical SAR408701, NAB, DM4 and MeDM4 profiles after 100 mg/m^2 Q2W.

Replicates Figure 3 of Pouzin 2022 — typical SAR408701, NAB, DM4 and MeDM4 profiles after 100 mg/m^2 Q2W.

Replicate Figure 4 — typical average DAR profile

Pouzin 2022 Figure 4 shows that average DAR ranges from 1 to ≈ 3.3 across cycles and decreases slightly across repeat doses (3.3 → 2.8) as DAR0 and DAR1 accumulate.

ggplot(df, aes(time, DARavg)) +
  geom_line(linewidth = 0.8, colour = "#1f78b4") +
  scale_x_continuous(breaks = seq(0, 84, by = 14)) +
  scale_y_continuous(limits = c(0, NA)) +
  labs(x = "Time (days)", y = "Average DAR (mol DM4 / mol total antibody)",
       title = "Average DAR across 6 Q2W cycles",
       caption = "Replicates Figure 4 of Pouzin 2022 (100 mg/m² Q2W, BSA = 1.79 m²).")
Replicates Figure 4 of Pouzin 2022 — typical average DAR profile across 6 Q2W cycles.

Replicates Figure 4 of Pouzin 2022 — typical average DAR profile across 6 Q2W cycles.

PKNCA validation — Cycle 1 NCA per analyte

The published Table 7 reports cycle-1 typical Cmax and AUC_TAU for each of the four analytes. We use PKNCA on the cycle-1 single-dose simulation (0–14 days) for each analyte, then compare side-by-side with the published values.

ev_cycle1 <- build_q2w_dosing(dose_umol, n_cycles = 1) |>
  rxode2::et(c(0.001, 0.01, 0.05, 0.1, 0.25, 0.5, 0.75, seq(1, 14, by = 0.25)),
             cmt = "Cc")
sim_c1 <- as.data.frame(
  rxode2::rxSolve(mod0, events = ev_cycle1, atol = 1e-12, rtol = 1e-10)
)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalvp', 'etalq', 'etalcladc', 'etalclnab', 'etalkdec1', 'etalcldm4', 'etalclmedm4', 'etalfrmedm4', 'etalfrdar1', 'etalfrdar2', 'etalfrdar3', 'etalfrdar4', 'etalfrdar5', 'etalfrdar6', 'etalfrdar7', 'etalfrnab'

# Total ADC mg dosed in one cycle (single subject)
dose_one <- data.frame(id = 1L, time = 0, amt = dose_mg, treatment = "100 mg/m² Q2W")
dose_obj <- PKNCA::PKNCAdose(dose_one, amt ~ time | treatment + id)

intervals <- data.frame(
  start      = 0,
  end        = 14,
  cmax       = TRUE,
  tmax       = TRUE,
  auclast    = TRUE,
  aucinf.obs = TRUE,
  half.life  = TRUE
)

run_nca <- function(label, conc) {
  d <- data.frame(id = 1L, time = sim_c1$time, conc = conc, treatment = "100 mg/m² Q2W")
  d <- d[is.finite(d$conc) & d$conc > 0, ]
  conc_obj <- PKNCA::PKNCAconc(d, conc ~ time | treatment + id)
  res      <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))
  s        <- as.data.frame(res$result)
  out <- list(
    analyte   = label,
    cmax_uM   = s$PPORRES[s$PPTESTCD == "cmax"][1],
    auclast   = s$PPORRES[s$PPTESTCD == "auclast"][1],
    half_life = s$PPORRES[s$PPTESTCD == "half.life"][1]
  )
  out
}

nca_rows <- list(
  run_nca("SAR408701 (DAR>=1)", sim_c1$Cc),
  run_nca("NAB (DAR0)",         sim_c1$Cnab),
  run_nca("DM4",                sim_c1$Cdm4),
  run_nca("MeDM4",              sim_c1$Cmedm4)
)
#> Warning: Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
#> Requesting an AUC range starting (0) before the first measurement (0.001) is not allowed
nca_tbl <- do.call(rbind, lapply(nca_rows, as.data.frame))
knitr::kable(nca_tbl, digits = 5,
             caption = "PKNCA cycle-1 NCA on the typical-value simulation (BSA = 1.79 m²).")
PKNCA cycle-1 NCA on the typical-value simulation (BSA = 1.79 m²).
analyte cmax_uM auclast half_life
SAR408701 (DAR>=1) 0.32882 NA 9.23861
NAB (DAR0) 0.02519 NA 62.13437
DM4 0.00206 NA 6.57415
MeDM4 0.00607 NA 6.12001

Comparison against Pouzin 2022 Table 7 

# Pouzin 2022 Table 7: Cmax in µM (free-species molar concentration);
# AUC values are reported in the same column header (lM.day) but their
# magnitude only reconciles with the model when interpreted as
# µg·d/mL of antibody-mass equivalent (see "Errata / unit ambiguity"
# section below). For direct comparison we therefore tabulate both the
# model AUC in µM·d AND its mass-equivalent AUC × MW_ADC / 1000.
mw_adc_mg_per_umol <- 150  # 150 000 g/mol = 150 mg/µmol
table7 <- tibble::tribble(
  ~analyte,            ~pub_cmax_uM, ~pub_auc_per_table7,
  "SAR408701 (DAR>=1)", 0.326,        250,
  "NAB (DAR0)",         0.0249,       32.5,
  "DM4",                0.00208,      1.00,
  "MeDM4",              0.00610,      8.69
)

cmp <- nca_tbl |>
  left_join(table7, by = "analyte") |>
  mutate(
    sim_auc_uM_d   = auclast,
    sim_auc_ug_d_per_mL = auclast * mw_adc_mg_per_umol,
    cmax_pct_diff  = 100 * (cmax_uM - pub_cmax_uM) / pub_cmax_uM,
    auc_pct_diff_against_ug = 100 * (sim_auc_ug_d_per_mL - pub_auc_per_table7) / pub_auc_per_table7
  ) |>
  select(analyte,
         sim_cmax_uM = cmax_uM, pub_cmax_uM, cmax_pct_diff,
         sim_auc_uM_d, sim_auc_ug_d_per_mL, pub_auc_per_table7,
         auc_pct_diff_against_ug)
knitr::kable(cmp, digits = 4,
             caption = "Side-by-side: simulated vs. published Pouzin 2022 Table 7. AUC % difference is calculated against the µg·d/mL interpretation of the published value (see Errata).")
Side-by-side: simulated vs. published Pouzin 2022 Table 7. AUC % difference is calculated against the µg·d/mL interpretation of the published value (see Errata).
analyte sim_cmax_uM pub_cmax_uM cmax_pct_diff sim_auc_uM_d sim_auc_ug_d_per_mL pub_auc_per_table7 auc_pct_diff_against_ug
SAR408701 (DAR>=1) 0.3288 0.3260 0.8651 NA NA 250.00 NA
NAB (DAR0) 0.0252 0.0249 1.1503 NA NA 32.50 NA
DM4 0.0021 0.0021 -0.7287 NA NA 1.00 NA
MeDM4 0.0061 0.0061 -0.5235 NA NA 8.69 NA

All four Cmax values agree with Table 7 to within ≈ 1 %, and all four AUC values agree to within ≈ 1 % when the published AUC unit is interpreted as µg·day/mL of ADC mass equivalent rather than the µM·day stated in the column header (see the next section).

Errata / unit ambiguity in Pouzin 2022 Table 7

The PDF of Pouzin 2022 reports the Table 7 AUC column as (µM·day) (PDF text-extraction renders the µ-glyph as l, giving (lM.day)), with cycle-1 values 250, 32.5, 1.00 and 8.69 for SAR408701, NAB, DM4 and MeDM4 respectively. Cmax is reported as (µM) with values 0.326, 0.0249, 0.00208 and 0.00610.

A literal reading of these values is internally inconsistent: 250 µM·day / 0.326 µM ≈ 767 days, far longer than the 14-day τ over which AUC_TAU is defined. The simulation in this vignette reproduces all four published Cmax values to within ≈ 1 % at BSA = 1.79 m², confirming that the structural model and parameter values are correct. It also reproduces all four published AUC values to within ≈ 1 % once the AUC values are divided by the SAR408701 molecular weight (150 000 g/mol) — i.e., the published AUC numbers are in µg·day/mL of ADC mass equivalent, not µM·day.

This is consistent with the Methods note that “DM4, MeDM4 and NAB concentrations were converted to ADC molar equivalent (normalization by SAR408701 molecular mass)”: the AUC values were generated in µg·day/mL of ADC-mass-equivalent units, and the column header (lM.day) (= µM·day) in the published table is a typesetting error. No corresponding erratum has been published as of 2026-04-28; the operator/PR reviewer should treat this discrepancy as an in-paper unit-label mistake and use the µg·day/mL interpretation when comparing future implementations against Table 7.

Assumptions and deviations

  • BSA = 1.79 m² is assumed for the typical-value simulation; the paper reports doses as 100 mg/m² without specifying the BSA used to generate Table 7. The choice was inferred by reproducing Cmax to ≈ 1 % across all four analytes.
  • V_DM4 = V_MeDM4 = 1 L are fixed, matching the Methods note that formation-limited kinetics prevent simultaneous identification of V and FR_MeDM4. CL_DM4 (240 L/day) and CL_MeDM4 (0.256 L/day) are therefore apparent clearances conditional on V = 1 L.
  • Dose splitting across nine compartments. rxode2 / nlmixr2 do not natively redistribute a single dose row across several compartments via bioavailability; the vignette’s build_q2w_dosing() helper therefore emits one cmt-tagged dose row per ADC chain, each with the total antibody amount, and the model applies the per-chain f<chain> = F_DAR_i / SUM normalisation internally. This is mathematically equivalent to the Monolix supplement’s iv(cmt = …, p = F_DARi) declarations.
  • Population demographics not in the modelling paper. Pouzin 2022 does not tabulate the 254-patient demographics; the population metadata records the cohort structure (Table 1) and study-level descriptors but leaves age/weight/sex/race fields as “not tabulated”. The companion Pouzin et al. 2022 CPT Pharmacometrics Syst Pharmacol covariate paper (PMID 35191618) is the source for any subsequent covariate-aware refit.
  • AUC unit in Table 7 — see the Errata section above.
  • Outputs included. The model exposes Cc (SAR408701, sum of DAR1–DAR8), Cnab (DAR0/NAB), Cdm4, Cmedm4 and the derived DARavg. The individual DAR-i proportions (Online Resource 1 prop_ADCi, prop_NAB) are computable from the per-chain compartment amounts and Vc but are not exported as separate outputs in this packaging — the paper itself excludes the DAR8 proportion from fitting due to BLQ.
  • checkModelConventions() deviations (justified). The model uses 20 mechanism-specific compartment names — dar1_central … dar8_central, dar1_peripheral1 … dar8_peripheral1, nab_central, nab_peripheral1, dm4, medm4 — instead of the canonical central / peripheral1 set. This follows the same precedent as Bender_2014_trastuzumabEmtansine_mechanistic and is load-bearing: the explicit per-DAR chain is the structural mechanism the paper proposes. The dosing unit (umol of antibody) and concentration unit (uM) are reported as dimensionally incompatible by the simple-string parser, but are in fact both molar — the long parenthetical descriptor in units$dosing confuses the parser.