Skip to contents

Model and source

mod <- rxode2::rxode(readModelDb("Fratoni_2023_funobactam_mouse"))

Preclinical (mouse). One-compartment pharmacokinetic model for the beta-lactamase inhibitor funobactam (formerly XNW4107) in the neutropenic murine thigh infection model, with first-order absorption from a subcutaneous depot and first-order elimination (Fratoni 2023). Funobactam was given subcutaneously at 1, 10 and 20 mg/kg together with an imipenem human-simulated regimen in uranyl-nitrate-treated mice (renal impairment induced to slow clearance), and a single best-fit average parameter set was estimated in Phoenix WinNonlin from destructive-sampling mean profiles. Cc is the total plasma concentration that was assayed; Cu is the unbound concentration (mean mouse plasma protein binding 3.53%) that the authors used for every %fT>MIC, %fT>CT, fAUC and fCmax exposure index. The apparent volume is 0.462 L/kg, one tenth of the 4.62 L/kg printed in the Results: only 0.462 L/kg reproduces the authors’ own simulated exposures in Table S1 and Table 2 (see the vignette). No between-subject variability or residual error was reported, so the model has no etas and the residual SD is fixed(0).

Full citation: Fratoni AJ, Berry AV, Liu X, Chen X, Wu Y, Nicolau DP, Abdelraouf K. Imipenem/funobactam (formerly XNW4107) in vivo pharmacodynamics against serine carbapenemase-producing Gram-negative bacteria: a novel modelling approach for time-dependent killing. J Antimicrob Chemother. 2023;78(9):2343-2353. doi:10.1093/jac/dkad242. PK parameters from Results (‘PK studies’); protein binding from Results (‘Ex vivo plasma protein-binding’); simulated exposures used to pin the volume from Supplementary Table S1 and Table 2.

The paper is a pharmacodynamic study of the beta-lactamase inhibitor funobactam given with a human-simulated imipenem regimen. Its only fitted time-course model is the one-compartment murine plasma PK model packaged here; every exposure index the paper analyses (fAUC0-24, fCmax, %fT>MIC, %fT>CT) was simulated from that model.

Population

Specific-pathogen-free female ICR mice weighing 20-22 g were rendered neutropenic with cyclophosphamide (150 mg/kg four days and 100 mg/kg one day before inoculation), given uranyl nitrate 5 mg/kg three days before inoculation to reduce renal clearance, and inoculated intramuscularly in one thigh with 0.1 mL of a ~10^7 cfu/mL suspension two hours before treatment (Methods, “Animals” and “Neutropenic murine thigh infection model”). For the PK study, funobactam 1, 10 or 20 mg/kg was given subcutaneously together with the imipenem human-simulated regimen to 36 infected mice per dose (108 in total), sacrificed in groups of six at predefined times for cardiac-puncture plasma (Methods, “Funobactam single-dose studies”). Total funobactam was assayed by LC-MS/MS and the mean profiles were fitted in Phoenix WinNonlin 8.1.

str(mod$population)
#> List of 8
#>  $ species       : chr "mouse (ICR, female, neutropenic thigh infection model)"
#>  $ n_subjects    : num 108
#>  $ n_studies     : num 1
#>  $ weight_range  : chr "20-22 g"
#>  $ sex_female_pct: num 100
#>  $ disease_state : chr "Neutropenic murine thigh infection. Mice were rendered neutropenic with cyclophosphamide (150 mg/kg day -4, 100"| __truncated__
#>  $ dose_range    : chr "funobactam 1, 10 and 20 mg/kg subcutaneously (single dose) with the imipenem human-simulated regimen (8, 2.5, 1"| __truncated__
#>  $ notes         : chr "Methods, 'Funobactam single-dose studies': 36 infected mice per funobactam dose level (108 total), sacrificed i"| __truncated__

Source trace

Equation / parameter Value Source location
lka (ka) 7.21 1/h Results, “PK studies”: “first-order absorption rate constant, 7.21 h-1”
lvc (V/F) 0.462 L/kg Results, “PK studies” prints 4.62 L/kg; 0.462 L/kg reproduces Table S1 and Table 2 (see Errata)
lkel (kel) 0.41 1/h Results, “PK studies”: “overall elimination rate constant, 0.41 h-1”
fu 0.9647 Results, “Ex vivo plasma protein-binding”: overall average binding 3.53%; 1 - 0.0353
propSd fixed(0) not reported (see Errata)
d/dt(depot), d/dt(central) n/a Results, “PK studies”: “PK was best described using a one-compartment model”; subcutaneous dosing with first-order absorption
Cu <- fu * Cc n/a Methods, “Human-simulated exposure PK studies” and Table 2 / Table S1: all indices use free plasma concentrations

Virtual cohort

The model has no between-subject variability: each dose level was a single naive-pooled mean profile from destructively sampled mice. The replication is therefore deterministic, with one simulated mouse per regimen. There are no etas, so neither zeroRe() nor omega = NA is needed (and omega = NA would error on a model with no random effects).

# Supplementary Table S1 (dose fractionation) and Table 2 (dose ranging).
# Published free exposures over the first 24 h of dosing.
regimens <- tibble::tribble(
  ~treatment,          ~dose, ~tau, ~study,             ~auclast, ~cmax,
  "0.9 mg/kg q6h",     0.9,   6,    "Table S1",          17.8,     1.7,
  "1.8 mg/kg q12h",    1.8,   12,   "Table S1",          18.2,     3.2,
  "3.6 mg/kg q24h",    3.6,   24,   "Table S1",          18.3,     6.3,
  "1.2 mg/kg q4h",     1.2,   4,    "Table S1",          35.0,     2.6,
  "1.8 mg/kg q6h",     1.8,   6,    "Table S1",          35.7,     3.5,
  "3.6 mg/kg q12h",    3.6,   12,   "Table S1",          36.4,     6.4,
  "7.2 mg/kg q24h",    7.2,   24,   "Table S1",          36.6,     12.6,
  "3.6 mg/kg q6h",     3.6,   6,    "Table 2 (interp.)", 71.36,    6.95,
  "7.2 mg/kg q6h",     7.2,   6,    "Table 2 (obs.)",    142.71,   13.90,
  "14.4 mg/kg q6h",    14.4,  6,    "Table 2 (obs.)",    363.70,   36.15
)
stopifnot(!anyDuplicated(regimens$treatment))

make_events <- function(dose, tau, id, treatment) {
  addl <- 24 / tau - 1
  # A once-daily regimen is a single dose within the 24 h window; rxode2
  # resets a non-zero `ii` with `addl = 0` and warns, so pass `ii = 0` there.
  ii <- if (addl > 0) tau else 0
  rxode2::et(amt = dose, ii = ii, addl = addl, cmt = "depot") |>
    rxode2::et(seq(0, 24, by = 0.01), cmt = "central") |>
    as.data.frame() |>
    dplyr::mutate(id = id, treatment = treatment)
}
events <- dplyr::bind_rows(lapply(seq_len(nrow(regimens)), function(i) {
  make_events(regimens$dose[i], regimens$tau[i], i, regimens$treatment[i])
}))

Simulation

sim <- rxode2::rxSolve(mod, events, keep = "treatment",
                       returnType = "data.frame")
#> Warning: multi-subject simulation without without 'omega'
stopifnot(all(c("Cc", "Cu") %in% names(sim)), nrow(sim) > 0)

Replicate published figures

Figure 1: free plasma concentration over the first dosing interval

Figure 1 overlays the free funobactam profile of each dose-ranging regimen (0.9, 1.8, 3.6, 7.2 and 14.4 mg/kg q6h) over one 6 h interval. The lines below are the packaged typical-value model; the paper’s 14.4 mg/kg line was drawn from that dose’s own confirmatory PK study, which ran about 25% above dose proportionality (see the NCA comparison below).

sim |>
  dplyr::filter(grepl("q6h$", treatment), time <= 6) |>
  dplyr::mutate(treatment = factor(
    treatment,
    levels = c("14.4 mg/kg q6h", "7.2 mg/kg q6h", "3.6 mg/kg q6h",
               "1.8 mg/kg q6h", "0.9 mg/kg q6h")
  )) |>
  ggplot(aes(time, Cu, colour = treatment)) +
  geom_line(linewidth = 0.8) +
  scale_y_log10(limits = c(0.01, 100)) +
  labs(x = "Time (h)", y = "Free plasma funobactam (mg/L)", colour = "Regimen",
       title = "Replicates Figure 1 of Fratoni 2023",
       caption = "Typical-value model; observed means are not reproduced.") +
  theme_bw()
#> Warning in scale_y_log10(limits = c(0.01, 100)): log-10 transformation
#> introduced infinite values.

Supplementary Table S1: %fT>MIC of the dose-fractionation regimens

Table S1 was generated by the authors by simulating each dose-fractionation regimen “using parameters from single dose PK studies”, so it is a direct check of the packaged parameters. %fT>MIC is the percentage of the first 24 h with free concentration above each MIC.

mics <- c(0.25, 0.5, 1, 2, 4, 8)
published_ft <- tibble::tribble(
  ~treatment,       ~`0.25`, ~`0.5`, ~`1`, ~`2`, ~`4`, ~`8`,
  "0.9 mg/kg q6h",  86,      57,     29,   0,    0,    0,
  "1.8 mg/kg q12h", 56,      42,     28,   13,   0,    0,
  "3.6 mg/kg q24h", 35,      28,     21,   14,   6,    0,
  "1.2 mg/kg q4h",  100,     100,    69,   24,   0,    0,
  "1.8 mg/kg q6h",  100,     86,     57,   29,   6,    0,
  "3.6 mg/kg q12h", 70,      56,     42,   28,   13,   0,
  "7.2 mg/kg q24h", 42,      35,     28,   21,   14,   6
) |>
  tidyr::pivot_longer(-treatment, names_to = "MIC", values_to = "published") |>
  dplyr::mutate(MIC = as.numeric(MIC))

sim_ft <- sim |>
  dplyr::filter(time < 24) |>
  tidyr::crossing(MIC = mics) |>
  dplyr::group_by(treatment, MIC) |>
  dplyr::summarise(simulated = 100 * mean(Cu > MIC), .groups = "drop")

ft_cmp <- published_ft |>
  dplyr::inner_join(sim_ft, by = c("treatment", "MIC"))
stopifnot(nrow(ft_cmp) == nrow(published_ft))

ft_cmp |>
  dplyr::mutate(simulated = round(simulated, 1),
                diff = round(simulated - published, 1)) |>
  dplyr::rename("Regimen" = treatment, "MIC (mg/L)" = MIC,
                "Published %fT>MIC" = published,
                "Simulated %fT>MIC" = simulated,
                "Difference (points)" = diff) |>
  knitr::kable(caption = "Supplementary Table S1 %fT>MIC replication.")
Supplementary Table S1 %fT>MIC replication.
Regimen MIC (mg/L) Published %fT>MIC Simulated %fT>MIC Difference (points)
0.9 mg/kg q6h 0.25 86 86.9 0.9
0.9 mg/kg q6h 0.50 57 58.3 1.3
0.9 mg/kg q6h 1.00 29 29.2 0.2
0.9 mg/kg q6h 2.00 0 0.0 0.0
0.9 mg/kg q6h 4.00 0 0.0 0.0
0.9 mg/kg q6h 8.00 0 0.0 0.0
1.8 mg/kg q12h 0.25 56 56.3 0.3
1.8 mg/kg q12h 0.50 42 42.2 0.2
1.8 mg/kg q12h 1.00 28 27.8 -0.2
1.8 mg/kg q12h 2.00 13 13.2 0.2
1.8 mg/kg q12h 4.00 0 0.0 0.0
1.8 mg/kg q12h 8.00 0 0.0 0.0
3.6 mg/kg q24h 0.25 35 35.2 0.2
3.6 mg/kg q24h 0.50 28 28.1 0.1
3.6 mg/kg q24h 1.00 21 21.0 0.0
3.6 mg/kg q24h 2.00 14 13.9 -0.1
3.6 mg/kg q24h 4.00 6 6.6 0.6
3.6 mg/kg q24h 8.00 0 0.0 0.0
1.2 mg/kg q4h 0.25 100 99.9 -0.1
1.2 mg/kg q4h 0.50 100 99.8 -0.2
1.2 mg/kg q4h 1.00 69 69.2 0.2
1.2 mg/kg q4h 2.00 24 24.0 0.0
1.2 mg/kg q4h 4.00 0 0.0 0.0
1.2 mg/kg q4h 8.00 0 0.0 0.0
1.8 mg/kg q6h 0.25 100 100.0 0.0
1.8 mg/kg q6h 0.50 86 86.9 0.9
1.8 mg/kg q6h 1.00 57 58.3 1.3
1.8 mg/kg q6h 2.00 29 29.2 0.2
1.8 mg/kg q6h 4.00 6 0.0 -6.0
1.8 mg/kg q6h 8.00 0 0.0 0.0
3.6 mg/kg q12h 0.25 70 70.4 0.4
3.6 mg/kg q12h 0.50 56 56.3 0.3
3.6 mg/kg q12h 1.00 42 42.2 0.2
3.6 mg/kg q12h 2.00 28 27.8 -0.2
3.6 mg/kg q12h 4.00 13 13.2 0.2
3.6 mg/kg q12h 8.00 0 0.0 0.0
7.2 mg/kg q24h 0.25 42 42.2 0.2
7.2 mg/kg q24h 0.50 35 35.2 0.2
7.2 mg/kg q24h 1.00 28 28.1 0.1
7.2 mg/kg q24h 2.00 21 21.0 0.0
7.2 mg/kg q24h 4.00 14 13.9 -0.1
7.2 mg/kg q24h 8.00 6 6.6 0.6

# Same parameters on both sides: the only difference is the authors' rounding
# and time grid, so a tight bound is appropriate. The 1.8 mg/kg q6h cell at
# MIC 4 is excluded: Table S1 prints 6%, but the same row's fCmax is 3.5 mg/L
# (below 4) and Table 2 prints 0% for the identical regimen.
slip <- ft_cmp$treatment == "1.8 mg/kg q6h" & ft_cmp$MIC == 4
stopifnot(sum(slip) == 1, ft_cmp$simulated[slip] == 0)
stopifnot(max(abs(ft_cmp$simulated[!slip] - ft_cmp$published[!slip])) < 2.5)

41 of the 42 cells agree within 1.3 percentage points. The remaining cell, 1.8 mg/kg q6h at an MIC of 4 mg/L, is printed as 6% in Table S1. That is inconsistent with the paper itself: the row’s own fCmax is 3.5 mg/L, so the free concentration never reaches 4 mg/L, and Table 2 gives 0% for the same regimen at the same MIC. The simulated 0% agrees with Table 2.

PKNCA validation

nca_conc <- sim |>
  dplyr::filter(!is.na(Cu)) |>
  dplyr::select(id, time, Cu, treatment)
# Guarantee a time-zero row (pre-dose free concentration is zero).
nca_conc <- dplyr::bind_rows(
  nca_conc,
  nca_conc |> dplyr::distinct(id, treatment) |> dplyr::mutate(time = 0, Cu = 0)
) |>
  dplyr::distinct(id, treatment, time, .keep_all = TRUE) |>
  dplyr::arrange(id, time)

nca_dose <- dplyr::bind_rows(lapply(seq_len(nrow(regimens)), function(i) {
  data.frame(id = i, treatment = regimens$treatment[i],
             amt = regimens$dose[i],
             time = seq(0, 24 - regimens$tau[i], by = regimens$tau[i]))
}))

conc_obj <- PKNCA::PKNCAconc(nca_conc, Cu ~ time | treatment + id,
                             concu = "mg/L", timeu = "h")
dose_obj <- PKNCA::PKNCAdose(nca_dose, amt ~ time | treatment + id,
                             doseu = "mg/kg")
intervals <- data.frame(start = 0, end = 24, cmax = TRUE, tmax = TRUE,
                        auclast = TRUE)
nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj,
                                          intervals = intervals))

Comparison against published NCA

Table S1 rows and the Table 2 row marked “interp.” were computed by the authors from the single-dose model; the Table 2 rows marked “obs.” are from separate confirmatory PK studies at those doses.

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = nca_res,
  reference = regimens |> dplyr::select(treatment, auclast, cmax),
  by = "treatment",
  units = c(auclast = "mg*h/L", cmax = "mg/L"),
  tolerance_pct = 20
)
knitr::kable(cmp, caption = paste(
  "Simulated vs published free funobactam fAUC0-24 and fCmax.",
  "* differs by >20%."
))
Simulated vs published free funobactam fAUC0-24 and fCmax. * differs by >20%.
NCA parameter treatment Reference Simulated % diff
Cmax (mg/L) 0.9 mg/kg q6h 1.7 1.74 +2.2%
Cmax (mg/L) 1.8 mg/kg q12h 3.2 3.19 -0.4%
Cmax (mg/L) 3.6 mg/kg q24h 6.3 6.32 +0.4%
Cmax (mg/L) 1.2 mg/kg q4h 2.6 2.65 +1.9%
Cmax (mg/L) 1.8 mg/kg q6h 3.5 3.48 -0.7%
Cmax (mg/L) 3.6 mg/kg q12h 6.4 6.37 -0.4%
Cmax (mg/L) 7.2 mg/kg q24h 12.6 12.6 +0.4%
Cmax (mg/L) 3.6 mg/kg q6h 6.95 6.95 +0.0%
Cmax (mg/L) 7.2 mg/kg q6h 13.9 13.9 +0.0%
Cmax (mg/L) 14.4 mg/kg q6h 36.2 27.8 -23.1%*
AUClast (mg*h/L) 0.9 mg/kg q6h 17.8 17.9 +0.4%
AUClast (mg*h/L) 1.8 mg/kg q12h 18.2 18.3 +0.3%
AUClast (mg*h/L) 3.6 mg/kg q24h 18.3 18.3 +0.2%
AUClast (mg*h/L) 1.2 mg/kg q4h 35 35.1 +0.3%
AUClast (mg*h/L) 1.8 mg/kg q6h 35.7 35.8 +0.2%
AUClast (mg*h/L) 3.6 mg/kg q12h 36.4 36.5 +0.3%
AUClast (mg*h/L) 7.2 mg/kg q24h 36.6 36.7 +0.2%
AUClast (mg*h/L) 3.6 mg/kg q6h 71.4 71.5 +0.2%
AUClast (mg*h/L) 7.2 mg/kg q6h 143 143 +0.2%
AUClast (mg*h/L) 14.4 mg/kg q6h 364 286 -21.3%*
attr(cmp, "footnote")
#> [1] "* differs from reference by more than ±20%."

nca_wide <- as.data.frame(nca_res$result) |>
  dplyr::filter(PPTESTCD %in% c("auclast", "cmax")) |>
  dplyr::select(treatment, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES) |>
  dplyr::inner_join(
    regimens |> dplyr::select(treatment, study, pub_auc = auclast,
                              pub_cmax = cmax),
    by = "treatment"
  )
stopifnot(nrow(nca_wide) == nrow(regimens))

# Model-derived rows (Table S1 and the interpolated Table 2 row) must match to
# the authors' rounding.
model_rows <- nca_wide |> dplyr::filter(study != "Table 2 (obs.)")
stopifnot(
  max(abs(model_rows$auclast / model_rows$pub_auc - 1)) < 0.02,
  max(abs(model_rows$cmax - model_rows$pub_cmax)) < 0.06
)

All model-derived rows agree with the published values to their printed rounding. The 7.2 mg/kg q6h confirmatory study also agrees within 1%. The 14.4 mg/kg q6h confirmatory study is about 21-23% above the model. The paper does not discuss the excess; it used that study’s observed exposures directly rather than the model, so the starred row reflects the confirmatory data and not the extraction.

The printed volume

The Results print the apparent volume as 4.62 L/kg. Simulating the same regimens with that value shows it cannot be what the authors used:

sim_printed <- rxode2::rxSolve(mod, events, keep = "treatment",
                               params = c(lvc = log(4.62)),
                               returnType = "data.frame")
#> Warning: multi-subject simulation without without 'omega'
printed_auc <- sim_printed |>
  dplyr::filter(time <= 24) |>
  dplyr::group_by(treatment) |>
  dplyr::summarise(auc_printed = sum(diff(time) *
    (head(Cu, -1) + tail(Cu, -1)) / 2), .groups = "drop")

nca_wide |>
  dplyr::inner_join(printed_auc, by = "treatment") |>
  dplyr::filter(study == "Table S1") |>
  dplyr::transmute(treatment, pub_auc,
                   auc_0462 = round(auclast, 2),
                   auc_462 = round(auc_printed, 2)) |>
  dplyr::rename("Regimen" = treatment,
                "Table S1 fAUC0-24" = pub_auc,
                "Simulated, V/F = 0.462 L/kg" = auc_0462,
                "Simulated, V/F = 4.62 L/kg" = auc_462) |>
  knitr::kable(caption = "Table S1 fAUC0-24 (mg*h/L) under both readings.")
Table S1 fAUC0-24 (mg*h/L) under both readings.
Regimen Table S1 fAUC0-24 Simulated, V/F = 0.462 L/kg Simulated, V/F = 4.62 L/kg
0.9 mg/kg q6h 17.8 17.88 1.79
1.2 mg/kg q4h 35.0 35.11 3.51
1.8 mg/kg q12h 18.2 18.26 1.83
1.8 mg/kg q6h 35.7 35.76 3.58
3.6 mg/kg q12h 36.4 36.53 3.65
3.6 mg/kg q24h 18.3 18.33 1.83
7.2 mg/kg q24h 36.6 36.67 3.67

Assumptions and deviations

  • Volume of distribution. The Results print V = 4.62 L/kg. The packaged value is 0.462 L/kg. The authors’ own simulated exposures (Supplementary Table S1, seven regimens, each with fAUC0-24, fCmax and six %fT>MIC values), the interpolated Table 2 rows, the 7.2 mg/kg q6h confirmatory study and the Figure 1 curves are all reproduced to their rounding with 0.462 L/kg, and are all ten-fold higher than 4.62 L/kg gives. Since AUC = dose / (V * kel), the gap could otherwise only come from kel, but a ten-fold smaller kel would also change the Cmax and the %fT>MIC durations, which 0.462 L/kg reproduces. The printed value is therefore treated as a decimal-place slip.
  • Table S1 cell. Supplementary Table S1 prints %fT>MIC = 6% for 1.8 mg/kg q6h at an MIC of 4 mg/L. The same row’s fCmax (3.5 mg/L) and Table 2 (0% for the identical regimen) both contradict it, so it is treated as a typesetting slip and excluded from the numerical check.
  • Free fraction. fu = 0.9647 is one minus the overall mean protein binding of 3.53%. The authors state binding was low and not concentration-dependent over 1-20 mg/kg; fu is held fixed.
  • No variability. The PK study was destructive sampling with six mice per time point, fitted to mean profiles, so no between-subject variability exists to report. The paper reports no residual error model either; propSd is fixed(0) and the model is suitable for typical-value simulation only.
  • Bioavailability. Dosing was subcutaneous only, so bioavailability is absorbed into the apparent volume V/F.
  • Imipenem not modelled. The imipenem human-simulated regimen was designed from murine imipenem PK parameters published earlier by the same group and not reported in this paper. The paper shows funobactam did not alter imipenem exposure (Figure 2), so the funobactam model stands alone.
  • Exposure-response not extracted. The paper’s sigmoid Emax fits relate a 24 h summary index (fAUC/MIC, fCmax/MIC, %fT>MIC, %fT>CT, (%fT>CT)/MIC) to the 24 h change in log10 cfu/thigh. They have no time course, and the paper reports only the derived stasis / 1-log / 2-log targets and R^2 (Tables 4, 5 and S2), not E0, Emax, EC50 or the Hill coefficient. The Cu output is what a user needs to compute any of these indices for a new regimen.