Skip to contents

Model and source

  • Citation: Ho BM, Jin J, Sanborn J, Nguyen TD, Singh N, Cheng C, Nasief NN, Carlino-MacDonald U, Russo TA, Smith NM (2025). Pharmacodynamic Evaluation of Adjuvant Targets: Low Molecular Weight PBP7/8 Effects on beta-Lactam Activity Against Carbapenem-Resistant Acinetobacter baumannii. Pharmaceuticals (Basel) 18(6):918.
  • Article: https://doi.org/10.3390/ph18060918

This vignette validates the packaged in-vitro mechanism-based pharmacodynamic model (Ho_2025_betalactam_pbp78_CRAB) against the structural equations (supplementary Figure S5) and the final parameter estimates (Table 1) of the paper. The model has no human pharmacokinetic component: drug exposure is a static broth concentration placed into each drug state at time zero, matching the static-concentration time-kill (SCTK) data the parameters were fit to. PKNCA-style noncompartmental validation therefore does not apply; instead the vignette uses the checks appropriate to a mechanistic bacterial-kill model (growth-control approach to carrying capacity, concentration-dependent killing, a known-answer check of the PBP7/8 covariate on KC50, and combination-therapy behaviour).

The central scientific question of the paper is whether inactivating the low-molecular-weight penicillin-binding protein PBP7/8 (the pbpG knockout) acts as a beta-lactam adjuvant. In the model this is a covariate on KC50: the strain indicator BACT (0 = HUMC1 wild type, 1 = HUMC1 delta-PBP7/8 knockout) lowers the effective KC50 of every drug.

Population

The model was fit simultaneously to SCTK data from two isogenic clinical isolates of Acinetobacter baumannii HUMC1: the carbapenem-resistant wild type (meropenem MIC 32 mg/L, polymyxin B MIC 2 mg/L) and its delta-PBP7/8 (pbpG) knockout. Experiments used cation-adjusted Mueller-Hinton broth at 37 C with a starting inoculum near 10^7 CFU/mL; viable counts were taken at 0, 1, 2, 4, 6, 8 and 24 h, with a limit of detection of 10^2 CFU/mL. Sixteen monotherapy arms and 48 beta-lactam-plus-polymyxin-B combination arms per isolate were modelled. The full record is available via readModelDb(...)$population.

mod_meta <- rxode2::rxode(readModelDb("Ho_2025_betalactam_pbp78_CRAB"))
mod_meta$population[c("species", "organism", "system", "temperature",
                      "starting_inoculum", "mic_values", "duration")]
#> $species
#> [1] "in vitro (Acinetobacter baumannii HUMC1 wild type and isogenic delta-PBP7/8 knockout)"
#> 
#> $organism
#> [1] "Acinetobacter baumannii HUMC1 (carbapenem-resistant clinical isolate) and its isogenic HUMC1 delta-PBP7/8 (pbpG) knockout"
#> 
#> $system
#> [1] "24 h static-concentration time-kill (SCTK) assay in cation-adjusted Mueller-Hinton broth (25 mg/L Ca2+, 12.5 mg/L Mg2+), 37 C under constant shaking; viable counts at 0, 1, 2, 4, 6, 8 and 24 h"
#> 
#> $temperature
#> [1] "37 C"
#> 
#> $starting_inoculum
#> [1] "~10^7 CFU/mL target; the model estimated 10^6.71 CFU/mL"
#> 
#> $mic_values
#> HUMC1 wild type, polymyxin B   HUMC1 wild type, meropenem 
#>                     "2 mg/L"                    "32 mg/L" 
#> 
#> $duration
#> [1] "24 h"

Source trace

Every ini() parameter carries an in-file comment pointing to its source in Ho 2025 Table 1; the ODE structure comes from supplementary Figure S5 (equations 1-21). The table below collects the provenance.

Equation / parameter Value Units Source
LOGCFUmax (log10cfumax) 8.20 log10 CFU/mL Table 1
LOGINOC (log10inoc) 6.71 log10 CFU/mL Table 1
LOGMUTR (log10mutr) 5.03 – Table 1 (MUTR = 10^-LOGMUTR, eq S5-5)
MGTS (mgts) 75.3 min Table 1
MGTR (mgtr) 46.4 min Table 1
KDiv (lkdiv, FIXED) 50 1/h Bulitta 2009 framework (ref 29); not reported in Ho 2025
Kmax,S (MER/IMI/ATM/CEF/PMB) 2.09 / 1.56 / 1.12 / 1.52 / 2.98 1/h Table 1
Kmax,R (MER/IMI/ATM/CEF/PMB) 1.13 / 0.83 / 0.76 / 0.89 / 1.12 1/h Table 1
KC50 (MER/IMI/ATM/CEF/PMB) 47.6 / 16.8 / 78.6 / 76.0 / 1.00 mg/L Table 1
betaKC50 (MER/IMI/ATM/CEF/PMB) -0.80 / -1.10 / -1.02 / -0.89 / -0.50 – Table 1 (PBP7/8-knockout effect)
Hill (MER/IMI/ATM/CEF/PMB) 0.91 / 0.88 / 0.51 / 0.90 / 0.93 – Table 1
Residual error additive on log10 log10 CFU/mL Methods (magnitude not reported)
Life cycle (S1/S2 + R1/R2), RepF n/a – Eqs S5-12,13,14,16,18,20
Hill killing, KC50 covariate n/a – Eqs S5-10,11

Setup

mod <- rxode2::rxode(readModelDb("Ho_2025_betalactam_pbp78_CRAB"))
mod <- rxode2::zeroRe(mod) # paper reports no estimated IIV; simulate at typical value
#> Warning: No omega parameters in the model

The fast division rate KDiv (fixed at 50 /h) makes the life-cycle system a fast transient at the start of each solve. The default LSODA solver can fail on it; dop853 is accurate and robust here, so every solve below passes method = "dop853". Because the paper reports no estimated between-subject variability (interexperimental variability was fixed at a 5% CV on each parameter), the model is simulated at typical value.

solve_tk <- function(bact, cmt = NULL, conc = 0, tmax = 24, by = 0.1) {
  ev <- rxode2::et(seq(0, tmax, by = by))
  if (!is.null(cmt)) ev <- rxode2::et(amt = conc, cmt = cmt, time = 0) |>
    rxode2::et(seq(0, tmax, by = by))
  rxode2::rxSolve(mod, events = ev, params = c(BACT = bact), method = "dop853")
}

1. Growth control: approach to carrying capacity

With no antibiotic, the total viable count grows from the inoculum (10^6.71 CFU/mL) toward the carrying capacity CFUmax = 10^8.20 CFU/mL. At 24 h the population should sit essentially at CFUmax.

g_wt <- solve_tk(0)
cat(sprintf("log10 inoculum (t=0)   = %.2f  (LOGINOC = 6.71)\n", g_wt$log_cfu[1]))
#> log10 inoculum (t=0)   = 6.71  (LOGINOC = 6.71)
cat(sprintf("log10 CFU at 24 h      = %.2f  (LOGCFUmax = 8.20)\n", tail(g_wt$log_cfu, 1)))
#> log10 CFU at 24 h      = 8.20  (LOGCFUmax = 8.20)
stopifnot(
  abs(g_wt$log_cfu[1] - 6.71) < 0.02,
  abs(tail(g_wt$log_cfu, 1) - 8.20) < 0.05
)

ggplot(g_wt, aes(time, log_cfu)) +
  geom_line(linewidth = 1) +
  geom_hline(yintercept = 8.20, linetype = "dashed", colour = "grey50") +
  annotate("text", x = 22, y = 8.20, label = "CFUmax = 8.20",
           vjust = -0.4, hjust = 1, colour = "grey30", size = 3.2) +
  scale_x_continuous(breaks = seq(0, 24, by = 4)) +
  labs(x = "Time (h)", y = expression(log[10] ~ "CFU/mL"),
       title = "Growth control (no antibiotic): approach to carrying capacity")

2. Concentration-dependent monotherapy killing

The paper reports concentration-dependent killing for both isolates (Results, Figure 1A). The chunk below reproduces meropenem monotherapy against the wild type at the three studied concentrations (0.5x / 1x / 2x the predicted free Cmax: 25 / 50 / 100 mg/L). Higher concentrations produce a deeper nadir.

mer_conc <- c(0, 25, 50, 100)
sim_mer <- do.call(rbind, lapply(mer_conc, function(d) {
  o <- solve_tk(0, cmt = "cmem", conc = d)
  o$conc <- d
  as.data.frame(o)[, c("time", "log_cfu", "conc")]
}))
sim_mer$conc_lab <- factor(sim_mer$conc, levels = mer_conc,
                           labels = paste0(mer_conc, " mg/L"))

# Deeper nadir with higher concentration (monotonic in dose).
nadir_by_conc <- tapply(sim_mer$log_cfu, sim_mer$conc, min)
cat("Meropenem nadir log10 CFU/mL by concentration (WT):\n")
#> Meropenem nadir log10 CFU/mL by concentration (WT):
print(round(nadir_by_conc, 2))
#>    0   25   50  100 
#> 6.71 6.64 5.33 4.14
stopifnot(all(diff(nadir_by_conc) < 0)) # strictly deeper killing as dose rises

ggplot(sim_mer, aes(time, log_cfu, colour = conc_lab)) +
  geom_line(linewidth = 0.8) +
  scale_x_continuous(breaks = seq(0, 24, by = 4)) +
  labs(x = "Time (h)", y = expression(log[10] ~ "CFU/mL"), colour = "Meropenem",
       title = "Meropenem monotherapy, HUMC1 wild type",
       caption = "Concentration-dependent killing; 24 h regrowth is driven by the resistant subpopulation.")

3. PBP7/8 knockout lowers KC50 (the adjuvant effect)

This is the paper’s central result. The knockout enters each drug’s effective KC50 as kc50_eff = kc50 * exp(betaKC50 * BACT). The chunk below reproduces the “calculated KC50 in PBP7/8 knockout” column of Table 1 as a known-answer check.

kc50 <- c(MER = 47.6, IMI = 16.8, ATM = 78.6, CEF = 76.0, PMB = 1.00)
beta <- c(MER = -0.80, IMI = -1.10, ATM = -1.02, CEF = -0.89, PMB = -0.50)
paper_ko <- c(MER = 21.4, IMI = 6.6, ATM = 28.3, CEF = 31.2, PMB = 0.61)

pred_ko <- kc50 * exp(beta)
kc50_tab <- data.frame(
  Drug = names(kc50), KC50_WT = kc50, betaKC50 = beta,
  KC50_KO_model = round(pred_ko, 2), KC50_KO_paper = paper_ko
)
knitr::kable(kc50_tab, row.names = FALSE,
             caption = "Effective KC50 under PBP7/8 knockout vs Table 1 of Ho 2025.")
Effective KC50 under PBP7/8 knockout vs Table 1 of Ho 2025.
Drug KC50_WT betaKC50 KC50_KO_model KC50_KO_paper
MER 47.6 -0.80 21.39 21.40
IMI 16.8 -1.10 5.59 6.60
ATM 78.6 -1.02 28.34 28.30
CEF 76.0 -0.89 31.21 31.20
PMB 1.0 -0.50 0.61 0.61

# Four of five drugs reproduce the paper's calculated KO KC50 exactly; imipenem
# differs (see Errata). Assert the matching four.
ok <- c("MER", "ATM", "CEF", "PMB")
stopifnot(all(abs(pred_ko[ok] - paper_ko[ok]) < 0.1))

# Average KC50 decline (mechanistic exp(beta) form).
bl <- c("MER", "IMI", "ATM", "CEF")
cat(sprintf("Average beta-lactam KC50 decline: %.1f%% (paper reports 59.7%%, using its KO column)\n",
            100 * mean(1 - exp(beta[bl]))))
#> Average beta-lactam KC50 decline: 61.2% (paper reports 59.7%, using its KO column)
cat(sprintf("Polymyxin B KC50 decline:         %.1f%% (paper reports 39.0%%)\n",
            100 * (1 - exp(beta["PMB"]))))
#> Polymyxin B KC50 decline:         39.3% (paper reports 39.0%)

The lower KC50 translates to deeper killing. Comparing wild type with knockout at the highest studied monotherapy concentration of each drug, the knockout reaches a lower nadir for every antibiotic.

drugs <- list(
  MER = c(cmt = "cmem", conc = 100),
  IMI = c(cmt = "cipm", conc = 30.6),
  ATM = c(cmt = "catm", conc = 132),
  CEF = c(cmt = "ccef", conc = 146),
  PMB = c(cmt = "cpmb", conc = 6)
)
nadir_cmp <- do.call(rbind, lapply(names(drugs), function(nm) {
  d <- drugs[[nm]]
  wt <- min(solve_tk(0, cmt = d["cmt"], conc = as.numeric(d["conc"]))$log_cfu)
  ko <- min(solve_tk(1, cmt = d["cmt"], conc = as.numeric(d["conc"]))$log_cfu)
  data.frame(Drug = nm, WT_nadir = round(wt, 2), KO_nadir = round(ko, 2),
             Extra_kill = round(wt - ko, 2))
}))
knitr::kable(nadir_cmp, row.names = FALSE,
             caption = "Nadir log10 CFU/mL: PBP7/8 knockout vs wild type (deeper kill = adjuvant effect).")
Nadir log10 CFU/mL: PBP7/8 knockout vs wild type (deeper kill = adjuvant effect).
Drug WT_nadir KO_nadir Extra_kill
MER 4.14 3.22 0.93
IMI 5.70 4.56 1.14
ATM 6.71 6.54 0.17
CEF 5.65 4.63 1.03
PMB 2.55 2.35 0.20
stopifnot(all(nadir_cmp$KO_nadir < nadir_cmp$WT_nadir)) # knockout always kills more
mer_cmp <- rbind(
  transform(as.data.frame(solve_tk(0, cmt = "cmem", conc = 100))[, c("time", "log_cfu")], Strain = "Wild type"),
  transform(as.data.frame(solve_tk(1, cmt = "cmem", conc = 100))[, c("time", "log_cfu")], Strain = "delta-PBP7/8")
)
ggplot(mer_cmp, aes(time, log_cfu, colour = Strain)) +
  geom_line(linewidth = 0.9) +
  scale_x_continuous(breaks = seq(0, 24, by = 4)) +
  labs(x = "Time (h)", y = expression(log[10] ~ "CFU/mL"), colour = NULL,
       title = "Meropenem 100 mg/L: PBP7/8 knockout deepens killing",
       caption = "The knockout lowers meropenem KC50 from 47.6 to 21.4 mg/L (Table 1).")

4. Combination therapy drives the population below the limit of detection

The paper reports that many beta-lactam-plus-polymyxin-B combination arms declined below the limit of quantification. Combining meropenem with polymyxin B against the knockout drives the modelled population far below the 10^2 CFU/mL limit of detection at the nadir.

ev_combo <- rxode2::et(amt = 100, cmt = "cmem", time = 0) |>
  rxode2::et(amt = 6, cmt = "cpmb", time = 0) |>
  rxode2::et(seq(0, 24, by = 0.1))
combo <- rxode2::rxSolve(mod, events = ev_combo, params = c(BACT = 1), method = "dop853")
mer_only <- solve_tk(1, cmt = "cmem", conc = 100)
cat(sprintf("KO meropenem 100 alone, nadir     = %.2f log10 CFU/mL\n", min(mer_only$log_cfu)))
#> KO meropenem 100 alone, nadir     = 3.22 log10 CFU/mL
cat(sprintf("KO meropenem 100 + polymyxin B 6  = %.2f log10 CFU/mL (nadir)\n", min(combo$log_cfu)))
#> KO meropenem 100 + polymyxin B 6  = -5.30 log10 CFU/mL (nadir)
stopifnot(min(combo$log_cfu) < min(mer_only$log_cfu)) # combination kills more
stopifnot(min(combo$log_cfu) < 2)                     # below 10^2 CFU/mL LOD

Assumptions and deviations

  • Fast division rate KDiv (not reported). The ODEs (supplementary equations S5-14/16/18/20) use a fast state-2 -> state-1 division rate KDiv, but it is not listed in Table 1. It is a fixed structural constant of the Bulitta 2009 life-cycle framework that Ho 2025 cites (reference 29), where the fast doubling step is fixed to 50 /h as not rate-limiting. The same convention is used in this package by Bulitta_2009_ceftazidime and the Mahadevan_2026 siblings. The model dynamics are governed by the slow generation step (60 / MGT), so the result is insensitive to the exact KDiv value; it is encoded as lkdiv <- fixed(log(50)).
  • Imipenem calculated KO KC50. Table 1 lists the “calculated KC50 of imipenem in PBP7/8 knockout” as 6.6 mg/L, but 16.8 * exp(-1.10) = 5.59. The other four drugs reproduce their calculated KO KC50 exactly from KC50 * exp(betaKC50), so the 6.6 appears to be a transcription slip in the table’s derived column. The model uses the estimated base KC50 (16.8) and betaKC50 (-1.10), giving an effective KO KC50 of 5.59 mg/L. The paper’s reported 59.7% average beta-lactam KC50 decline is computed from its KO column (including the 6.6); the mechanistic exp(betaKC50) form gives 61.2%.
  • Residual error magnitude. The paper states an additive residual error on the log10 scale (equivalent to proportional error on the natural scale) but does not report its value in Table 1, so addSd is encoded as fixed(0) and the model is intended for typical-value simulation. The reported interexperimental variability (a fixed 5% CV on every parameter) is likewise not carried as estimated IIV.
  • Static-concentration scope. The packaged model represents the SCTK fit: the drug states are held constant (d/dt = 0). The paper additionally used this PD model with external hollow-fiber (HFIM) pharmacokinetics to simulate dynamic regimens (Figures 2 and 3); reproducing those requires an external PK layer supplying time-varying concentrations and is outside the scope of the packaged static-concentration model.
  • Solver. The fast division transient can defeat the default LSODA solver; dop853 is used throughout this vignette.
  • BACT coding. The isogenic strain pair is carried in the BACT strain indicator (0 = wild type, 1 = delta-PBP7/8 knockout); here it feeds a covariate-effect coefficient on KC50 rather than switching a whole parameter set, so it must be coded 0/1 with 0 as the wild-type reference.