Antibacterial and Sterilizing Effect of Benzylpenicillin in Tuberculosis

ABSTRACT The modern chemotherapy era started with Fleming's discovery of benzylpenicillin. He demonstrated that benzylpenicillin did not kill Mycobacterium tuberculosis. In this study, we found that >64 mg/liter of static benzylpenicillin concentrations killed 1.16 to 1.43 log10 CFU/ml below starting inoculum of extracellular and intracellular M. tuberculosis over 7 days. When we added the β-lactamase inhibitor avibactam, benzylpenicillin maximal kill (Emax) of extracellular log-phase-growth M. tuberculosis was 6.80 ± 0.45 log10 CFU/ml at a 50% effective concentration (EC50) of 15.11 ± 2.31 mg/liter, while for intracellular M. tuberculosis it was 2.42 ± 0.14 log10 CFU/ml at an EC50 of 6.70 ± 0.56 mg/liter. The median penicillin (plus avibactam) MIC against South African clinical M. tuberculosis strains (80% either multidrug or extensively drug resistant) was 2 mg/liter. We mimicked human-like benzylpenicillin and avibactam concentration-time profiles in the hollow-fiber model of tuberculosis (HFS-TB). The percent time above the MIC was linked to effect, with an optimal exposure of ≥65%. At optimal exposure in the HFS-TB, the bactericidal activity in log-phase-growth M. tuberculosis was 1.44 log10 CFU/ml/day, while 3.28 log10 CFU/ml of intracellular M. tuberculosis was killed over 3 weeks. In an 8-week HFS-TB study of nonreplicating persistent M. tuberculosis, penicillin-avibactam alone and the drug combination of isoniazid, rifampin, and pyrazinamide both killed >7.0 log10 CFU/ml. Monte Carlo simulations of 10,000 preterm infants with disseminated disease identified an optimal dose of 10,000 U/kg (of body weight)/h, while for pregnant women or nonpregnant adults with pulmonary tuberculosis the optimal dose was 25,000 U/kg/h, by continuous intravenous infusion. Penicillin-avibactam should be examined for effect in pregnant women and infants with drug-resistant tuberculosis, to replace injectable ototoxic and teratogenic second-line drugs.


RESULTS
Benzylpenicillin plus avibactam is highly bactericidal in 12-well plates. On the suggestion from K. Dheda to minimize bias and improve confidence of findings, all investigators and laboratory personnel (except D. Deshpande and T. Gumbo) were blinded to the identity of penicillin in all our experiments. Penicillin was designated drug X until the final result was reported. We coincubated log-phase-growth M. tuberculosis H37Ra with 7 different concentrations of penicillin alone in Middlebrook 7H9 broth with 10% oleic acid, albumin, dextrose, and catalase (OADC), with results shown in Fig. 1A. Penicillin did not kill M. tuberculosis until around 64 mg/liter, when it killed the bacteria 1.16 log 10 CFU/ml below the level present on day 0. An absence of effect at low concentrations but some effect at higher concentrations suggests that penicillin's effect is inhibited by a saturable process that can be overcome at high concentrations. Next, we repeated the same experiment but this time examined all penicillin concentrations in the presence of 15 mg/liter of avibactam, leading to results shown in Fig. 1B. Penicillin maximal kill was 6.80 Ϯ 0.45 log 10 CFU/ml, with a 50% effective concentration (EC 50 ) of 15.11 Ϯ 2.31 mg/liter (r 2 ϭ 0.988). This maximal kill equals that of rifampin but exceeds that of either isoniazid or pyrazinamide in the same assay (17). The EC 50 is easily achieved with standard benzylpenicillin dosing: injection or short infusion of 5 million international units (MIU) achieves a peak penicillin concentration of 400 mg/liter in serum (600 mg ϭ 1 MIU). Next, we performed the same experiment but this time with adherent THP-1 monocytes infected with M. tuberculosis H37Ra in 12-well plates. Maximal microbial kill by the penicillin alone was 1.43 Ϯ 0.29 log 10 CFU/ml (Fig. 1C), with an EC 50 of 22.26 Ϯ 6.22 mg/liter. With addition of avibactam, the maximal kill increased to 2.42 Ϯ 0.14 log 10 CFU/ml and the EC 50 decreased to 6.70 Ϯ 0.56 mg/liter, which means that avibactam improved both efficacy (maximal kill) and potency (EC 50 ). Thus, ␤-lactamase inhibition alone leads to dramatic microbial kill by penicillin.
Benzylpenicillin is efficacious against intracellular M. tuberculosis in the HFS-TB and generates little resistance. Next, we performed dose-response studies against M. tuberculosis H37Ra-infected THP-1 human-derived monocytes in the HFS-TB. In the presence of avibactam, the penicillin MIC for M. tuberculosis H37Ra was 8 mg/liter. We wanted to establish the effect of benzylpenicillin-avibactam against intracellular M. tuberculosis, as well as to identify the best dosing schedule and optimal penicillin exposure at the same time. Thus, we examined three penicillin dosing schedules: once-a-day doses infused over 1 h, twice-a-day doses infused over 1 h every 12 h, and two continuous infusion doses over 24 h. We measured the concentration-time profiles in each HFS-TB, with results shown in Fig. 2A. The penicillin concentrations shown in Fig. 2A are well within those achieved in patients treated either with a 5-MIU injection or a 24-MIU continuous infusion over 24 h. Pharmacokinetic modeling of the measured concentrations revealed an elimination rate constant (mean Ϯ standard deviation) of shown against intended curves, demonstrating that intended pharmacokinetics were achieved. Doses were administered as a 1-h infusion. Regimen 1 achieved an extracellular T MIC of 28% from a human-equivalent dose of 0.41 MIU administered once a day, regimen 2 a T MIC of 43% from a human-equivalent dose of 0.9 MIU administered once a day, and regimen 3 a T MIC of 88% from a human-equivalent dose of 0.8 MIU administered every 12 h; regimens 4 to 6 achieved a T MIC of 100% from human-equivalent doses of 2.4 MIU administered every 12 h (regimen 4), 5.3 MIU administered every 12 h (regimen 5), and 10.2 MIU administered every 12 h (regimen 6). (B) The observed versus model predicted slope was 1.06 Ϯ 0.03 (r 2 ϭ 0.972), indicating minimal bias. (C) Measured intracellular concentrations within infected monocytes were much lower than extracellular concentrations. Regimens 1 to 3 have an intracellular T MIC of 0%, regimen 4 has a T MIC of 30%, regimen 5 has a T MIC of 58%, and regimen 6 has a T MIC of 90%. (D) Time-kill curves demonstrate that all penicillin doses (except nontreated) killed below the day 0 burden (stasis), with no rebound growth. (E) A similar pattern was observed using TTP. (F) The relationship between bacterial burden and extracellular T MIC is shown for each sampling day as M. tuberculosis in nontreated HFS-TB replicates continued to grow. Maximal kill (E max ) is the span between nontreated controls and the maximal-effect portion of each curve. (G) The size of the drug-resistant subpopulation was highest in nontreated controls, which decreased with increasing T MIC . (H) However, when measured as a percentage of the total population, the drug-resistant subpopulation percentage versus exposure was best described as an inverted "U" curve with amplification of the penicillin-resistant subpopulation at a T MIC of 30 to 40%. 0.227 Ϯ 0.007 per hour (Fig. 2B). We also measured the concentrations of penicillin inside the M. tuberculosis-infected monocytes, which revealed the intracellular concentration-time profiles shown in Fig. 2C. The intracellular penicillin elimination rate constant of 0.213 Ϯ 0.008 per hour differed slightly from the extracellular elimination rate, which means that the lower intracellular concentrations were likely driven by poor penetration into the human cells and not increased intracellular elimination. Avibactam extracellular concentrations were constant throughout at 13.54 Ϯ 0.26 mg/liter, per design (slope ϭ 0; P ϭ 0.297). Figure 2D and E show that all penicillin regimens killed the intracellular M. tuberculosis to a level below the day 0 bacterial burden. Maximal kill was 3.28 log 10 CFU/ml in 20 days (Fig. 2D). We also simultaneously examined bacterial burden based on time to positivity (TTP) using the mycobacterial growth indicator tube assay (MGIT), a more sensitive assay than CFU. Figure 2E shows that the largest increase in TTP was 2.54-fold higher than on day 0, in the same range as with the three-drug combination first-line therapy in the past (31). Table 1 shows that microbial kill was highest when the proportion of time that concentration persisted above the MIC (T MIC ) was highest, and that this was unrelated to peak concentration. Figure 2F shows that the T MIC associated with maximal kill was Ն65% of 24 h (r 2 ϭ 0.967). This means that when penicillin doses are administered, the aim should be to exceed Ն65% of the day with concentrations above the MIC for maximal effect in TB.
We also captured the M. tuberculosis subpopulation resistant to 3 times the MIC in each HFS-TB unit, with results shown in Fig. 2G. Figure 2H shows the penicillin-resistant subpopulation as percentage of the total bacterial burden versus T MIC , which was a typical inverted "U" shaped curve shape that we have identified for first-line and second-line anti-TB drugs in the past (23,24,32,33). Thus, the penicillin exposureversus-resistance relationship is the same shape as for standard anti-TB drugs. A T MIC of 100% completely suppressed the penicillin-resistant subpopulation (r 2 ϭ 0.975). However, since the penicillin-resistant subpopulation was below 1%, considered clinically significant, penicillin monotherapy was much less prone to resistance emergence than each of the four first-line drugs.
Penicillin has a dramatic bactericidal effect in the hollow-fiber model. Next, we examined the efficacy of penicillin in the HFS-TB with log-phase-growth M. tuberculosis H37Ra using continuous infusion of penicillin equivalent to 1 MIU/h over 24 h (plus avibactam) to achieve 100% T MIC over 4 weeks. Figure 3A shows the concentration-time profile achieved in duplicate HFS-TB, easily achieved in patients: as an example, the same 24 MIU of penicillin administered by continuous infusion over 24 h achieved a steady concentration of 80 mg/liter in serum. Figure 3B shows a surprising M. tuberculosis response curve with this regimen. First, there was a 2.88-log 10 CFU/ml decline in only the first 2 days of therapy, or an EBA of 1.44 log 10 CFU/ml/day. This EBA is 2.4 times that of isoniazid, the first-line drug with the highest EBA (24). Figure 3C demonstrates similar findings based on TTP, which is inversely proportional to bacterial burden. There was a 2.6-fold change in TTP on penicillin monotherapy in the HFS-TB. Second, Fig. 3B and C demonstrate a lack of rebound growth on monotherapy, not seen until now with any first-line drugs (24,32,34). This confirms that at a T MIC of 100%, there was suppression of both penicillin-resistant and penicillin-tolerant M. tuberculosis over 4 weeks of therapy. Penicillin has a sterilizing effect against nonreplicating bacilli. The traditional dogma is that penicillin microbial kill stops in stationary-phase bacteria and is best with rapidly multiplying bacteria; thus, penicillin should not be effective against NRP or SDB (35). Moreover, penicillin is labile at low pH, and some extracellular SDB in lung cavities are at low pH (36)(37)(38). We compared the effect of penicillin (at a T MIC of 100%) plus avibactam versus the standard three-drug regimen of isoniazid, rifampin, and pyrazinamide in the HFS-TB NRP model based on the streptomycin-dependent auxotroph M. tuberculosis SS18b in acidified (pH ϭ 5.8) Middlebrook broth (23,39). The M. tuberculosis strain penicillin MIC was 16 mg/liter, the rifampin MIC was 0.0625 mg/liter, the isoniazid MIC was 0.0156 mg/liter, and the pyrazinamide MIC was 12.5 mg/liter. M. tuberculosis was inoculated into triplicate HFS-TB after Ͼ10 days of streptomycin deprivation; the HFS-TB broth lacked streptomycin throughout the experiment. We measured the antibiotic concentrations achieved in each HFS-TB, which revealed exposures shown in Table 2. Figure 4A show the dramatic effect of penicillin-avibactam as a single agent, which achieved kill below limits of detection (in CFU per milliliter) at the same rate as combination standard therapy. Thus, penicillin killed Ͼ7 log 10 CFU/ml of nonreplicating bacteria at acidic pH in the presence of avibactam, which was higher than the concentration for log-phase-growth bacteria. However, when we used the more sensitive TTP readout shown in Fig. 4B, the combination therapy reached negative cultures faster than penicillin-avibactam.
Penicillin is additive and antagonistic to ceftazidime-avibactam. Avibactam is currently coformulated with ceftazidime; thus, ceftazidime-avibactam is the most im- The decline in CFU/milliliter in the nontreated controls was because of the volumes removed from the systems and number of times which we sampled the peripheral compartment over 8 weeks of the experiment, which reduces the bacterial burden in a bacterial population that is not replicating or growing. Based on the changes in CFU/milliliter changes, the effect of penicillin plus avibactam was similar to that of the standard 3-drug regimen. (B) Based on TTP, the standard regimen reached culture negativity on day 28, which was faster than penicillin-avibactam. mediate source of avibactam for clinical use. However, ceftazidime-avibactam itself has very good anti-TB activity (17). We performed a HFS-TB with intracellular M. tuberculosis H37Ra to determine if the combination kills more than each agent alone. Figure 5A shows the bacterial burden as TTP over the 28 days of treatment. The nontreated controls had progressively declining TTP with time, indicating growth, while the combination of ceftazidime-avibactam plus penicillin had the highest TTPs, indicating the lowest bacterial burden. Effect or kill was defined as TTP on each sampling day minus that on day 0 (pretreatment TTP). The TTP effect of ceftazidime-avibactam alone on each day was added to that of penicillin-avibactam alone on that day to calculate the expected additivity, which was compared to the TTP effect in HFS-TB replicates treated with ceftazidime-avibactam plus penicillin (observed effect), based on Bliss independence definitions (40). Figure 5B shows that expected additivity was similar to observed effect in the first 14 days, since 95% confidence intervals overlapped. However, Fig. 5B shows that antagonism was encountered after day 21, since the observed combination effect TTP difference was higher than the expected additivity TTP difference. Surprisingly, the pattern changed again on day 28, so that expected additivity and observed effect were virtually similar. This means that additivity versus antagonism versus synergy definitions depended on when the sampling was done and that these change with duration of therapy. Nevertheless, Fig. 5B shows that the combination of ceftazidime-avibactam plus penicillin resulted in higher kill than did either drug alone. where error bars are not visible, they were smaller than the symbols used. Data are expressed as TTP, which is a more sensitive assay than CFU. The nontreated controls grew well, as shown by decreasing TTP as duration of therapy increased. The combination of penicillin-ceftazidime-avibactam had the highest TTP on each sampling day, indicating the lowest bacterial burden. (B) Error bars are 95% confidence intervals. The bacterial burden on day of sampling minus that on day 0 gives the TTP change or effect. The effect observed with the combination was compared to that for ceftazidime-avibactam and penicillin-avibactam added together (expected additivity), which revealed overlapping 95% confidence intervals during the first 2 weeks. On day 21, the observed effect was less than expected additivity, as shown by nonoverlapping 95% confidence intervals for the triplicate HFS-TB. This means antagonism. However, the effect narrowed again on day 28.

MDR-TB and incurable M. tuberculosis clinical strains are susceptible to penicillin.
It could be that the penicillin effect is confined to M. tuberculosis laboratory strains and that penicillin would have no effect on clinical multidrug-resistant M. tuberculosis (MDR-TB) strains. Therefore, we sought to identify MICs for 25 clinical strains, a mixture of 20% drug-susceptible strains and 80% MDR-TB/extremely drugresistant M. tuberculosis (XDR-TB) strains from South African patients. We examined this using two assays, the reference broth microdilution test and the MGIT. The penicillin MICs of the clinical strains are shown in Fig. 6A and B. The median MIC by either assay was 2 mg/liter, and the MIC 90 was 64 mg/liter. Since penicillin concentrations higher than the MIC 90 can be achieved in patients, these results mean that penicillin plus avibactam could be effective against a large proportion of MDR-TB and XDR-TB strains.
The standard dose of penicillin would be effective in treating TB in the clinic. Given this MIC distribution, how well would different doses of penicillin do in adult patients with pulmonary TB, in pregnant women in the third trimester, or in preterm neonates with disseminated TB in the clinic? Since TB clinical outcomes are mostly driven by pharmacokinetic and MIC variability (41)(42)(43)(44)(45)(46)(47)(48)(49), we utilized the penicillin population pharmacokinetic parameter estimates and interindividual variability identified by others (shown in Table 3) in 10,000-subject Monte Carlo experiments (MCE) in each of five doses (50)(51)(52). Our aim was to identify the dose of penicillin that would achieve the exposure of time above the MIC Ͼ65% in preterm babies with disseminated TB, pregnant women with pulmonary TB, and nonpregnant adults with pulmonary TB, treated using continuous infusion. Table 3 shows that the simulated population pharmacokinetic parameters for 5 doses each (total of 50,000 simulated subjects for each patient group) were similar to those in the initial domain of input, which is an internal validation step. Figure 6C shows the performance of each dosing regimen to achieve or exceed a T MIC Ͼ65% in infants of a gestational age of Ͻ32 weeks as the MIC increased. The probability of target attainment (PTA) remained above 90% until an MIC of 64 mg/liter in the highest dose in the children; even at an MIC of 128 mg/liter, the PTA for the highest dose fell slightly below 90%. Figure 6D shows results for pulmonary TB in pregnant women in the third trimester, which were similar to those in nonpregnant adults (Fig. 6E). Figure 6F is a summation which shows the proportion of patients who would achieve the T MIC target of Ͼ65% when all MICs were taken into account. Based on Fig. 6F, the optimal dose in preterm children was 10,000 U/kg/h, while that in both pregnant and nonpregnant adults was 25,000 U/kg/h. Therefore, these are the doses to be administered in treatment of TB in these patient groups. In a sensitivity analysis, we determined the optimal doses in adults if penicillin penetration into lung lesions was only 50% (see Materials and Methods). The adult dose of 25,000 U/kg/h would still achieve or exceed the target exposure in Ͼ80% of adults. The performance of avibactam has been published before in our work on ceftazidime-avibactam (17).

DISCUSSION
Benzylpenicillin and penicillin V are arguably the safest antibiotics used in the treatment of all infectious diseases. Billions of doses have been administered since the 1940s in diverse locales. The findings presented here suggest that penicillin-avibactam alone or penicillin in combination with ceftazidime-avibactam has the potential to improve both bactericidal and sterilizing effects and should be examined for the ability to shorten treatment duration as part of multidrug TB regimens in patients. Treatment shortening with reduction in relapse rates will impact transmission and disease burden and allow more efficient resource allocation; it is thus the research "holy grail" of TB treatment. The advent and spread of XDR-TB and incurable TB, and emerging treatment failures even with newer agents such as bedaquiline and delamanid, mean that newer and repurposed agents are urgently required to provide therapeutic options for these patients (53-55). MDR-TB and XDR-TB clinical isolates demonstrated good MICs; indeed, the true MICs may even be lower given that penicillin degrades at a rate of about 10%

Value for parameter in domain of input (% IIV)
Output in >50,000-subject simulation (% IIV)  (56)(57)(58). Because benzylpenicillin was never patented, and production costs have fallen over the decades, it costs very little and is readily available across the world. It represents one of the least expensive repurposed drugs for TB. On the other hand, continuous infusion imposes logistical and monetary costs on patients and TB programs, which could counterbalance the low costs. Moreover, avibactam may further increase the cost of this combination. There has been recent community advocacy calling for stakeholders to design treatments for pregnant women with TB, given the dire need (59). TB in pregnant women is associated with a 2-fold increase in premature birth and low birth weight and a 6-fold-higher perinatal death rate (60). TB is responsible for 6% to 10% of all maternal mortality in low-HIV-prevalence settings and 15 to 34% in high-HIV-prevalence settings (60). There is currently no specific safe TB therapy targeted for this neglected and vulnerable population. Though the first-line drugs are considered to be fairly safe in pregnancy, there have been reports of teratogenicity from rifampin (61,62). Even though current second-line drugs are known to adversely affect embryofetal development, the CDC still recommends them for treatment of pregnant women, after counseling the women, on the grounds that gains overwhelmingly outweigh the risks. Benzylpenicillin-avibactam could help cure the TB in these women, with reduced fetal toxicity risks. While penicillin crosses the placenta, it is not teratogenic, which has made it the "go to" drug for pregnant women with different infections. While patient studies are lacking for avibactam, the FDA label indicates that reproductive studies performed during pregnancy in animal showed no adverse effects on embryofetal development (63). Nevertheless, fetal toxicity from long-term administration of avibactam in people has not been adequately studied. Thus, penicillin-avibactam could represent a breakthrough for pregnant women suffering from TB.

Total clearance (liters/h/kg) Central vol (liters/kg) Total clearance (liters/h/kg) Central vol (liters/kg) Preterm infants
Closely related is that congenital TB is an increasingly recognized problem, with the children often born preterm (64). In the case of MDR-TB in children Ͻ6 years old, the rates of ototoxicity with aminoglycosides are extremely high and devastating (65). Indeed, even in adults with MDR-TB, the ototoxicity rates are up to 70% with amikacin (66). Our finding that the use of the BlaC inhibitor avibactam allows penicillin to kill M. tuberculosis opens the door to immediately test penicillin-avibactam efficacy in children. Penicillin is also already the go-to drug in infectious diseases of neonates, infants, and toddlers. Our findings create opportunities to try combinations of other M. tuberculosis BlaC inhibitors with penicillin.
Next, we wanted to make recommendations for how benzylpenicillin could be used in neonates, pregnant women, and nonpregnant adults with TB, including in outpatient settings. First, penicillin degrades to Ͻ90% in 13 h after reconstitution in normal saline or sterile water for injection at room temperature, or when loaded onto a portable pump reservoir and infused at 37°C for 1 day (67). However, when benzylpenicillin was reconstituted with sodium citrate (a buffer) for the purpose of outpatient parenteral antimicrobial therapy, concentrations remained above 90% when the drug was stored at 3 to 5°C for 7 days and then kept at 37°C for 24 h (67,75,76). The outpatient strategy would be to use this buffer and keep the solution refrigerated at 3 to 5°C for 7 days, with the pump reservoir loaded daily. The recommend doses per day for a neonate, pregnant woman, and nonpregnant adults are shown in Table 4. If normal saline or water for injection is used as the diluent, the solution is refrigerated at 3 to 5°C for 7 days, and then the intravenous therapy would be loaded twice a day at doses shown in Table 4, taking into account the degradation rates, which could impose a hardship and likely necessitate inpatient care. Though the doses recommended are large, they are not far from those used for other severe infections. In children with congenital syphilis, for example, up to 300,000 U/kg/day (i.e., 12,500 U/kg/h) is used, and for adults the penicillin dose we recommend is only 1.5 to 1.75 times that for those with infective endocarditis.
Finally, the combination of ceftazidime-avibactam with penicillin resulted in higher kill rates than benzylpenicillin-avibactam alone. During early therapy the combination was additive, but later on it became antagonistic. This remarkable finding has implications on general pharmacology studies examining drugs for synergy and antagonism. Often, antibiotic pairs are presented as either antagonistic or synergistic. Here, we show that the same concentrations may be additive or antagonistic depending on duration of therapy! This is in addition to our prior findings in the HFS-TB and in patients that anti-TB agents could be antagonistic or synergistic at the same time point in a concentration-dependent fashion (42,45,46,68). This means that definitions of synergy, additivity, and antagonism may need modification. The question of double ␤-lactam coverage for TB that we have proposed is open to debate. For Gram-negative bacilli, double ␤-lactam coverage is sometimes used on the theory that binding of each of the ␤-lactams in the combination is on unique penicillin-binding proteins. It remains to be seen whether this strategy will work for TB.

MATERIALS AND METHODS
Bacterial strains, cell lines, and growth conditions. The following laboratory M. tuberculosis strains were used: H37Ra (ATCC 25177), H37Rv (ATCC 27294), and SS18b (donated by Stewart Cole), and 25 clinical strains from the South African Medical Research Council. Storage and culture conditions for log-phase bacteria, SDB, and intracellular M. tuberculosis were as described in our prior studies (23,68,69). M. tuberculosis SS18b was first cultured in Middlebrook 7H9 broth plus 10% OADC in the presence of 50 mg/liter of streptomycin to achieve log-phase growth, followed by subculture in the same medium but without streptomycin for 14 days. Human-derived THP-1 cells (ATCC TIB-202) grown in RPMI 1640 plus 10% fetal bovine serum FBS (RPMI-FBS) were infected with H37Ra for intracellular experiments in wells and HFS-TB using methods as described previously but with a coincubation period of 3 h.
Materials and chemicals. Penicillin was purchased from Baylor University Medical Center pharmacy (Dallas, TX). Avibactam was synthesized by BOC Sciences, Shirley, NY. Penicillin G-d7 was purchased from Toronto Research Chemicals (Ontario, Canada). Hollow-fiber cartridges were obtained from FiberCell (Frederick, MD).
Identification of MICs. MICs were identified on two different occasions each, using the standard broth macrodilution test and the MGIT. We used rifampin as a positive control for the drugs and M. tuberculosis H37Rv as the control for bacterial isolates. The final penicillin concentrations examined were 0, 0. 5,1,2,4,8,16,32,64, and 128 mg/liter, in combination with a final fixed avibactam concentration of 15 mg/liter in triplicate, which is the peak concentration achieved with a 0.5-g clinical dose of avibactam.
Penicillin concentration assay. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis was performed using a Waters Acquity ultraperformance liquid chromatograph (UPLC) coupled with Waters Xevo TQ mass spectrometer. Separation was achieved by injecting 2 l of sample on a Waters Acquity UPLC HSS T3 column (50 ϫ 2.1 mm; 1.8 m) using a binary gradient. Solvents for UPLC were as follows: solvent A, 0.1% aqueous formic acid, and solvent B, 0.1% formic acid in acetonitrile. Samples were diluted 1:20 with internal standard solution containing penicillin G-d7. The transitions used were m/z 335 Ͼ 176 (penicillin G) and m/z 342 Ͼ 183 (penicillin G-d7). The between-day coefficient of variation (CV) for analysis of low and high (in parentheses) quality controls for RPMI-FBS were 2% (2%) and 2% (2%) for saline, and those for Middlebrook 7H9 broth were 4% (2%). The within-day CV for analysis of low and high (in parentheses) quality controls were 4% (2%) for RPMI-FBS, 2% (1%) for Middlebrook 7H9 broth, and 2% (2%) for saline. The lower limits of quantitation were 0.25 mg/liter for Middlebrook 7H9 broth for the nonreplicating persister and log-phase studies and 0.375 mg/liter in RPMI-FBS and 0.05 mg/liter for saline (to measure intracellular concentrations). The avibactam, rifampin, isoniazid, and pyrazinamide assays for concentration measurement have been described in detail previously (17,71). Pharmacokinetic/pharmacodynamic modeling. Compartmental pharmacokinetic modeling was employed for all HFS-TB studies, as described in detail previously (23,24,31,33). Exposure response relationships were examined using the inhibitory sigmoid E max model for microbial kill and the quadratic function for the relationship between drug-resistant subpopulation and exposure (23). All modeling was performed using ADAPT 5 software (Biomedical Simulations Resource, University of Southern California). Results were transferred to GraphPad Prism 7 for graphing purposes.
Monte Carlo experiments. Pharmacokinetic parameter estimates and interindividual variability as percent CV in preterm infants with a gestational age of Յ32 weeks were from Muller et al., those for pregnant women during the third trimester were from Johnson et al., and those for nonpregnant adults were from Komatsu et al., shown in Table 1 (50)(51)(52). In the case of disseminated TB in preterm neonates, the serum concentrations were used for the exposure calculations. Since the HFS-TB studies had been performed using media with serum and albumin, we made no further adjustments for protein binding. In the case of adult pulmonary TB, there are no recent proper pharmacokinetic studies on the epithelial lining fluid or infected lung tissue concentration of penicillin. In 1950, Jensen et al. (77) noted that the benzyl ester of penicillin achieved high concentrations in the lungs, unlike earlier penicillin. Similarly, studies by Ungar and Muggleton using mice and guinea pigs with bacterial pneumonia revealed that unlike normal lung, the concentration-time profiles achieved in infected lung tissue by benzylpenicillin exceeded those in serum Ͼ2-fold (72). Around the same time, Heathcote and Nassau noted that in patients treated with benzypenicillin who had lung resection, the lung tissue concentrations at the time of pulmonary artery ligation were 1-to 2-fold higher than in serum from simultaneous blood draw (73). We took a conservative approach and assumed a 1:1 ratio in effective concentration in lungs of TB patients versus serum, based on up to 2-fold drug penetration and 50% protein binding of penicillin. For sensitivity analysis, we assumed we were 50% overoptimistic in our assumptions based on ELF penetration calculated using compartmental pharmacokinetic approach for the ureidopenicillin, piperacillin (74). For each drug these population pharmacokinetic parameter estimates, and the covariance, were used as the domain of input in subroutine PRIOR of ADAPT. Simulations were run for each dose for 10,000 patients, for each group of patients, so that Ͼ50,000 patient pharmacokinetics were simulated for preterm children alone, pregnant women alone, and adults not presently pregnant (i.e., 160,000 in all). We generated 24-h concentration-time profiles for patients on continuous infusion with each dose, and then at each MIC we calculated the PTA for a T MIC ofϾ65%. In order to calculate the proportion of 10,000 patients who achieved this target when treated by each dose, we took an expectation over the entire MIC distribution, as detailed previously (17).

ACKNOWLEDGMENTS
This study was funded by Baylor Research Institute. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. T