Protein Binding of First-Line Antituberculosis Drugs

ABSTRACT

The 4-drug regimen of rifampin, isoniazid, pyrazinamide, and ethambutol is an inexpensive, reliable option for treating patients with drug-susceptible tuberculosis (TB). Its efficacy could be further improved by determining the free drug concentrations in plasma, knowing that only the unbound drug can freely penetrate to the tissues. Using an ultrafiltration technique, we determined the protein binding (PB) extent and variability of the first-line anti-TB drugs when given simultaneously to TB patients, representing a real-life case scenario. We used clinical samples routinely received by our laboratory. Plasma proteins were also measured. A protein-free medium was used to determine the nonspecific binding. Plasma samples from 22 patients were included, of which plasma proteins were measured for 18 patients. The median PB was determined for rifampin (88%; range, 72 to 91%), isoniazid (14%; range, 0 to 34%), pyrazinamide (1%; range, 0 to 7%), and ethambutol (12%; range, 4 to 24%). Plasma proteins were not found to be significant predictors for the PB of first-line anti-TB drugs. Rifampin PB was positively correlated with its plasma concentration (P value = 0.0051). Conversely, isoniazid PB was negatively correlated with its plasma concentration (P value = 0.0417). Age was found to have a significant effect on isoniazid PB (P value = 0.0376). No correlations were observed in pyrazinamide or ethambutol. In conclusion, we have determined variable PB of rifampin, isoniazid, pyrazinamide, and ethambutol in patient plasma samples, with median values of 88, 14, 1, and 12%, respectively. In this small study, PB of rifampin and that of isoniazid are dependent on their plasma concentrations.

INTRODUCTION

Culminating with the British Thoracic Society publication of 1984, the 4-drug regimen of rifampin, isoniazid, pyrazinamide, and ethambutol has been proven very effective in the treatment of drug-susceptible tuberculosis (TB) (1). Currently, this regimen is an inexpensive, reliable option for treating patients with drug-susceptible TB (2). However, the efficacy of this regimen could be further improved from the dose optimization perspective. One way is to optimize the free drug concentration of these drugs, knowing that only the unbound drug can freely penetrate to the tissues and bind to the target to produce its effect (3).

Significant research in this area can be found in the literature (4–14), as summarized in Table 1. However, to our knowledge, none of the previous studies have determined the extent of protein binding of all four drugs simultaneously by using plasma samples taken from patients with TB. Therefore, we sought to design a unique study to represent a real-life case scenario. Our study's aim was to determine the protein binding extent and variability of the four first-line anti-TB drugs when given simultaneously (i.e., during the intensive phase) to TB patients and to verify whether it is associated with commonly measured plasma proteins, such as total protein and albumin. Additionally, the results of this study could also be of great use for future pharmacokinetic modeling, in order to accurately describe the drug effect and determine exposure targets based on the unbound drug concentration, as opposed to total drug concentration.

TABLE 1.

Literature review of protein binding of first-line anti-TB drugs in humans^a

Drug	Method	No. of subjects	Source of sample	Protein binding (%)	Reference
RIF	UC			75	4
	UC			75–80	5
	ED	37	TB pts (n = 3)	TB pts: 17–38	6
			Unknown status (n = 34)	Others: 8–41
	ED	21	HS (n = 11)	HS: 87–91	7
			TB pts (n = 10)	TB pts: 84–88
	ED	3	Spiked plasma	70–75	11
	ED	10	WN subjects	78	8
		8	UN subjects	65
		10	UN with TB	43
	ED		Spiked plasma	40–80	14
	ED		Spiked plasma	70–80	9
	UF	36	TB patients (WN and UN)	80–94	10
INH	ED	3	Spiked plasma	0	11
	ED		Spiked plasma	4–30	14
	ED		Spiked plasma	20	9
PZA	ED		Spiked plasma	15–40	9
EMB	ED	13	Non-TB pts (n = 10)	8–40	13
			TB pts (n = 3)
	ED, UF	6	HS	20–30	12
	ED	3	Spiked plasma	6	11
	ED		Spiked plasma	<5	14

^aED, equilibrium dialysis; EMB, ethambutol; HS, healthy subjects; INH, isoniazid; pts, patients; PZA, pyrazinamide; RIF, rifampin; TB, tuberculosis; UC, ultracentrifugation; UF, ultrafiltration; UN: undernourished; WN, well nourished.

RESULTS

We included 22 patients with TB who were being treated with first-line TB drugs during the intensive phase. The median age was 50 years, with the majority of them being males (Table 2). Plasma proteins were measured in 18 patients; the remaining 4 patients did not have sufficient plasma volume after the protein binding experiment. The medians of total protein, albumin, and prealbumin were within the normal ranges, with a few patients having albumin and prealbumin levels lower than the typical ranges. Plasma (total) median concentrations were determined for rifampin (10.5 μg/ml), isoniazid (3.0 μg/ml), pyrazinamide (33.2 μg/ml), and ethambutol (2.7 μg/ml).

TABLE 2.

Patient demographics (n = 22)

Patient characteristic	Median (range), %
Age, yrs	50.0 (32.6–84.9)
Sex, male	86.4
Plasma proteinsa
Total protein, g/dl	7.3 (6.4–8.2)
Albumin, g/dl	4.2 (3.0–4.9)
Prealbumin, mg/dl	20.0 (7.0–33.0)
Plasma (total) concn
Rifampin, μg/ml	10.5 (1.4–19.7)
Isoniazid, μg/ml	3.0 (1.6–8.5)
Pyrazinamide, μg/ml	33.2 (24.8–52.2)
Ethambutol, μg/ml	2.7 (0.8–6.5)

^aEighteen patients were included; 4 patients did not have enough volume to measure plasma proteins.

Isoniazid, pyrazinamide, and ethambutol did not show significant nonspecific binding (NSB) (less than 4%, 6%, and 2%, respectively). Rifampin showed inverse linear concentration-dependent binding (3% to 14%) to the Centrifree device. This was corrected for by plotting the preultrafiltrate of protein-free medium (PFM) against the postultrafiltrate of PFM and fitting a regression line, which was later used to correct for the free concentration of rifampin. The median protein binding was then determined for rifampin (88%), isoniazid (14%), pyrazinamide (1%), and ethambutol (12%) (Table 3). A large variability in protein binding among patients was observed for rifampin (range, 72% to 91%) and isoniazid (range, 0% to 34%).

TABLE 3.

Protein binding and free drug concentrations of the first-line anti-TB drugs

Drug	Free drug concn in μg/ml, median (range)	% protein binding, median (range)
Rifampin	1.3 (0.4–2.4)	88.2 (71.7–91.0)a
Isoniazid	2.6 (1.1–8.2)	13.8 (0.0–34.1)b
Pyrazinamide	31.8 (23.9–54.0)	1.0 (0–7.2)
Ethambutol	2.3 (0.8–6.0)	12.1 (3.8–23.7)

^aAfter the exclusion of a single outlier, the median (range) becomes 88.4% (80.6 to 91.0).

^bAfter the exclusion of a single outlier, the median (range) becomes 13.6% (0.0 to 25.8).

Using bivariate linear regressions, albumin and prealbumin were not found to be significant predictors for protein binding (P values = 0.2886 and 0.8133, respectively). However, rifampin protein binding was positively correlated with the drug plasma (total) concentration (beta = 0.5, r² = 0.33, and P value = 0.0051) (Fig. 1). Unlike for rifampin, isoniazid protein binding was negatively correlated with its plasma (total) concentration (beta = −1.9, r² = 0.19, and P value = 0.0417) (Fig. 1). An inverse correlation was observed between isoniazid protein binding and age (beta = −0.2, r² = 0.15, and P value = 0.0735). When protein binding was adjusted for plasma concentration using multiple linear regression, age became a significant predictor (P value = 0.0376), indicating that both plasma concentration and age are significant predictors for isoniazid protein binding. No correlations were observed between the protein binding of pyrazinamide or ethambutol with their plasma concentrations or plasma proteins.

FIG 1.

Rifampin and isoniazid plasma concentration versus their protein binding. (a) Positive correlation between rifampin plasma concentration and its protein binding (beta = 0.5, r² = 0.33, and P = 0.0051). (b) Negative correlation between isoniazid plasma concentration and its protein binding (beta = −1.9, r² = 0.19, and P = 0.0417).

Due to the high variability in protein binding of rifampin and isoniazid and their dependence on their plasma concentrations, we plotted observed versus predicted free concentrations (Fig. 2), using regression equations from Fig. 1 to obtain the predicted values. The predicted rifampin free concentrations at high concentrations were lower than the observed concentrations, while the predicted free concentrations of isoniazid were consistent along the entire range of the observed concentrations.

FIG 2.

Observed versus predicted free concentrations of rifampin and isoniazid. For predictability of the free concentration of rifampin, r² = 0.71 and P < 0.0001; for predictability of the free concentration of isoniazid, r² = 0.98 and P < 0.0001.

DISCUSSION

When pharmacokinetic/pharmacodynamic (PK/PD) models are developed using the total concentration of the drug, the PD effect could be overestimated for drugs that are known to have high protein binding, such as rifampin. This also means that drugs with high protein binding will be underdosed, hence increasing the risk of acquired resistance. To our knowledge, none of the previous studies have determined the protein binding from samples of the four first-line drugs given simultaneously to TB patients, which represents a real-case scenario during the intensive phase of TB treatment. In our experiment, we used the ultrafiltration technique, which has become the most popular approach for determining protein binding due to the method's simplicity, time efficiency, and accuracy of results (15, 16).

Additionally, assessing protein binding from plasma samples collected from patients after administering the drug is preferred over obtaining pooled blank plasma samples that are spiked with the drug in vitro (17). Using this approach allows the determination of protein binding in the presence of endogenous compounds and metabolites that could inhibit or compete with binding sites. Most importantly, this approach allows us to explore interpatient variability. In our study, we included drug plasma samples from 22 patients with TB, which allowed us to determine that protein binding did vary among patients (Table 3), with the exception of pyrazinamide, which had a slightly narrower range than the other drugs due its very low protein binding.

Among the first-line agents, rifampin had the highest drug protein binding, as expected. Its protein binding ranged from 71% to 91%, with an outlier at the lower end of the range. Excluding the outlier, the range becomes 81% to 91%. This emphasizes the importance of including the protein binding parameter in PK/PD models, as only about 9 to 19% of the total concentration can freely penetrate to the site of infection (Fig. 3). Rifampin protein binding seems to correlate with the drug plasma concentration, meaning that the higher the concentration, the more drug will be bound to proteins (Fig. 1). Previous in vitro studies have shown that about 30% to 41% of rifampin binds to albumin (7, 11). In our analysis, albumin was not a significant predictor of the protein binding of rifampin. This could be due to the inclusion of patients with relatively normal levels of albumin.

FIG 3.

Rifampin and isoniazid free fraction variability. (a) The median (range) of rifampin free fraction is 11.6% (9.0 to 19.4%). (b) The median (range) of isoniazid free fraction is 86.4% (74.2 to 100%).

To our knowledge, all the previous isoniazid protein binding studies used plasma spiked with isoniazid (9, 11, 14). Our results, using samples from TB patients who were administered the drug orally, showed a range (0 to 34%) similar to those in previous studies. This relatively wide range can be problematic in clinical settings, in terms of predicting where the patient might fall in that range (Fig. 3). The clinical decision on dosing may be informed by the inverse correlation between protein binding and age, although this correlation was not strong (beta = −0.2 and r² = 0.15). An opposite trend to rifampin was observed with isoniazid—the higher the plasma concentration, the lower the protein binding. Although this could potentially be beneficial, it could also increase the free drug concentrations to supratherapeutic levels.

A previous study has shown a range of protein binding for pyrazinamide of 15 to 40%, using plasma samples spiked with pyrazinamide (9). Other review papers have reported lower ranges, 5 to 10% (18, 19). However, in our study, we found that the protein binding for pyrazinamide is actually much lower than has been previously reported (0 to 7.2%), with 50% of the patients (n = 11) having less than 1% of the drug protein bound. Due to the lower protein binding of pyrazinamide, no correlations with plasma proteins were observed. Similar to the case with pyrazinamide, previous studies on ethambutol have shown a wide spectrum of the percentage of bound drug, ranging from a low of 5% to a high of 40% (13, 14). Our results show a relatively narrow range, from 4% to 24%. No correlations were observed between ethambutol and plasma proteins or patient characteristics.

We acknowledge that our study had a few limitations. First, the low number of females precluded us from investigating the effect of sex on protein binding (if any). Another limitation was that the majority of patients had normal plasma proteins levels; this potentially could have precluded us from investigating the effect of plasma protein on drug protein binding in patients on both extremes (i.e., very high or very low levels of, for example, albumin). Finally, the design of this study does not allow one to investigate drug displacement from protein binding (if any) when multiple drugs are given at the same time, which could be of interest during the continuation phase of TB treatment.

Conclusion.

We have determined the protein binding of first-line anti-TB agents in patients' blood samples. We found that the levels of protein binding are 88, 14, 1, and 12% for rifampin, isoniazid, pyrazinamide, and ethambutol, respectively. Rifampin plasma concentration seems to positively correlate with its protein binding, while isoniazid plasma concentration has a negative correlation with its protein binding. Also, age seemed to have an inverse correlation with isoniazid protein binding. Pyrazinamide and ethambutol did not show any correlations with plasma proteins or patient characteristics.

MATERIALS AND METHODS

Sample preparation and ultrafiltration procedure.

The Infectious Disease Pharmacokinetics Laboratory (IDPL) at the University of Florida (Gainesville, FL) frequently receives clinical samples from local and international sites to measure drug concentrations in patients' plasma. Institutional review board (IRB) approval was obtained (IRB201500777) in order to start collecting clinical patient samples for this experiment. Two-hour samples were collected from 22 patients with TB who were taking the four first-line drugs at the same time (during the intensive phase). Patient information was deidentified and coded to be used for research purposes. Protein binding was determined using an ultrafiltration technique, as follows: (i) patient plasma samples were placed in a water bath (N-EVAP model 116; Organomation Associates Inc.) at 37°C for 30 min; (ii) during that time, the centrifuge (Labnet Hermle Z-400-K) was preheated to 40°C; (iii) 1 ml of the plasma sample was slowly transferred to the ultrafiltration tube (Centrifree Ultrafiltration Devices; no. 4104) in one flow to avoid air bubbles, with the pipette tip touching the reservoir wall; (iv) the tubes then were centrifuged for 30 min at 37 ± 5°C and 3,000 rpm (1,872 × g); and (v) finally, the ultrafiltrates were transferred to 1.5-ml microcentrifuge tubes to be assayed. Total protein, albumin, and prealbumin were also measured in the plasma samples if the remaining plasma volume was sufficient (i.e., >100 μl).

Nonspecific binding.

A protein-free medium (PFM; saline with 10% methanol to keep rifampin in solution) was used to determine whether the drugs bound to the ultrafiltrate membrane/device (i.e., nonspecific binding [NSB]). The same steps as mentioned in the previous section were performed using PFM. The NSB was calculated for each drug, using equation 1. If a drug was found to bind to the Centrifree device, a correction factor (equation 2) was used to calculate the corrected ultrafiltrate concentration (i.e., free concentration) (20).

$$\text{NSB\hspace{0.17em}}(\%)=\left(\frac{\text{prefiltration\hspace{0.17em}concentration}-\text{postfiltration\hspace{0.17em}concentration}}{\text{prefiltration\hspace{0.17em}concentration}}\right)\times 100$$ (1)

$$\text{Corrected\hspace{0.17em}free\hspace{0.17em}concentration}=\text{ultrafiltrate\hspace{0.17em}concentration\hspace{0.17em}}\times \text{correction\hspace{0.17em}factor}$$ (2)

Assay procedure.

The IDPL has already established an ultrahigh-performance liquid chromatography (UHPLC) method coupled with tandem mass spectrometry (MS/MS) to simultaneously quantify first-line anti-TB drugs in plasma. The method was revalidated in saline to mimic the filtrate matrix. In short, the analysis was performed using Thermo Scientific Endura triple quadrupole MS, supplied with a DIONEX UltiMate 3000 RS autosampler and pump. Multiple-reaction monitoring was used to detect and quantify the first line anti-TB agents and their internal standards, which are rifampin-d₃, isoniazid-d₄, pyrazinamide-N¹⁵-d₃, and ethambutol-d₄. Eight and six standard concentrations were used for the calibration curve in plasma and saline, respectively. The ranges of the plasma standard curve are 0.25 to 50 μg/ml for rifampin, 0.15 to 30 μg/ml for isoniazid, 0.5 to 100 μg/ml for pyrazinamide, and 0.05 to 10 μg/ml for ethambutol. The ranges for the saline standard curves are similar to those for plasma, with the exception that the upper limits of the range were 25, 15, 50, and 5 μg/ml, respectively.

Statistical analysis.

The change in concentrations between the plasma and the ultrafiltrate represents the amount of drug bound to plasma proteins. The protein binding for each patient was calculated using equation 3:

$$\text{Protein\hspace{0.17em}binding}\hspace{0.17em}(\%)=(1-\frac{\text{drug\hspace{0.17em}ultrafiltrate\hspace{0.17em}concentration}}{\text{drug\hspace{0.17em}plasma\hspace{0.17em}concentration}})\times 100$$ (3)

Bivariate linear regressions were used initially to investigate the correlation among patient characteristics, drug concentrations, plasma protein concentrations (total protein, albumin, and prealbumin), and protein binding. Multiple linear regression was also used to adjust for any potential confounders. All statistical analyses were performed using JMP Pro version 13 (SAS Institute Inc., Cary, NC).