Introduction
The incidence of multidrug-resistant (MDR) gram-negative bacterial (GNB) infections has increased dramatically over the last decade, and such infections have emerged as a major challenge in global public health. The mortality rate of MDR-GNB infections is 40% or higher [
1‐
3]. Novel antibiotic agents for the treatment of MDR-GNB infections are limited. Owing to its potential activity against MDR-GNB infection, the “old” drug polymyxin has been repurposed and recommended as a last-resort therapy for MDR-GNB infections [
4,
5].
Polymyxin B (PMB) and colistin, which have similar pharmacologically active moieties, are two different agents in the polymyxin class. PMB is a fermented mixture of more than 30 components from
Paenibacillus polymyxa [
6]. Nephrotoxicity is the major dose-limiting factor impacting the clinical use of PMB. Previous studies have indicated that the incidence of nephrotoxicity is 20–60% after intravenous administration of PMB [
6‐
11]. Colistin is administered intravenously as a prodrug colistimethate sodium, affected by the ratio of prodrug lost to renal elimination prior to activation. Some studies showed the incidence of AKI of colistin is 30–76.1% [
12‐
14]. Azad et al. found that PMB can noticeably accumulate in the renal cortex, especially in proximal tubular cells, and induce renal tubular epithelial cell apoptosis or necrosis [
15], which indicates that the nephrotoxicity of PMB might be related to drug exposure in vivo. For critically ill patients, there are noticeable individual differences in the pharmacokinetics of polymyxin B [
7,
8,
16], and the dose of PMB might not represent the actual amount of the drug to which the patient is exposed. Polymyxin B1 (PMB1) and polymyxin B2 (PMB2) are the two major components of polymyxin B (PMB), accounting for more than 60% of the total weight, and have been used to characterize the effects of PMB exposure in vivo for therapeutic drug monitoring (TDM) [
16,
17].
Previous studies have reported some risk factors for PMB-induced nephrotoxicity, such as age, baseline SCr (serum creatinine) level, body mass index (BMI), concomitant use of vasoactive drugs and vancomycin, infection site, duration of therapy, and a daily dose of 200 mg or more [
7,
8,
18‐
22]; however, PMB exposure in vivo was not assessed in these previous studies. According to the international consensus [
23], the magnitude of polymyxin (i.e. PMB and colistin) exposure is the most important risk factor for polymyxin-associated acute kidney injury. Previously, for PMB, only Han et al. [
22] and Wang et al. [
24] found that the plasma trough concentration (
Cmin) of PMB and an AUC
ss,24h of at least 100 mg·h/L of PMB were independent risk factors for PMB-induced nephrotoxicity, respectively. Notably, the comorbidity variables included in their studies were limited. Evidence for the relationship between PMB exposure and nephrotoxicity is still insufficient with a limited sample size. Furthermore, whether PMB1 and PMB2, the primary constituents of PMB, are associated with nephrotoxicity has not yet been reported.
On the other hand, only the unbound fraction (
fu) of the drug is pharmacologically active, as the protein-bound drug cannot reach the site of infection. The reported protein binding rate of PMB ranges from 58% to 92.4% [
25,
26]. The variation in protein binding can affect the volume of distribution and clearance of PMB and, thus, its efficacy and toxicity [
27]. Given that pharmacological activity depends on the unbound protein concentration rather than the total plasma concentration, determination of the free protein concentration is essential. However, currently available reports on TDM of PMB did not directly assess the free drug concentration. The relationship between the free drug concentration and renal toxicity is still unknown.
Therefore, the present study explores the relationship between exposure to total and free PMB (including PMB1 and PMB2) in the plasma and nephrotoxicity, and investigates the risk factors for PMB-induced nephrotoxicity in critically ill patients, providing possible reference data for clinical TDM.
Discussion
The present study demonstrated that the Cmax (B1), hypertension and baseline BUN levels were independent risk factors for AKI development during PMB treatment in critically ill patients. To our knowledge, this is the first report showing that Cmax (B1) is related to PMB-induced nephrotoxicity. Furthermore, we were the first to explore the correlation between the free concentration of PMB (PMB1, PMB2) and PMB-induced nephrotoxicity in critically ill patients, and we found that the free Cmax (B) and free Cmax (B1) in the AKI group were significantly higher than those in the non-AKI group.
The ultrafiltration tube method and equilibrium dialysis are the two commonly used methods for determining free concentrations. The results of the ultrafiltration tube method are affected by multiple factors, such as temperature, centrifugal force, pH and the ultrafiltration membrane [
37,
38]. Equilibrium dialysis is based on drug diffusion across a semipermeable membrane that separates the sample to be investigated from a buffer solution and is still considered the gold standard for monitoring free drug concentrations [
39]. Equilibrium dialysis was adopted in our study, and the results showed that the median protein binding rates of PMB, PMB1 and PMB2 were 91.7% (range 65.7–96.0%), 95.2% (range 61.7–98.2%) and 82.7% (range 59.2–95.5%), respectively. It was reported that the plasma protein binding rate of PMB in eight critically ill patients (range 78.5–92.4%) was higher than that in healthy humans (55.9% ± 4.7%) [
26], which is consistent with our study. However, Sandri and colleagues found that the median protein binding rate in 23 critically ill patients was 58% (36–74%) [
25], which is quite different from that observed in our study. Special pathophysiological and iatrogenic factors in critically ill patients can affect protein concentrations by altering synthesis and catabolism or promoting protein movement from the plasma to extravascular sites [
40]. Alterations in protein levels vary significantly among individuals and might result in variability in the free concentration of PMB in critically ill patients. Total concentrations and published protein binding values are usually used to predict the unbound drug concentration in clinical practice in general. However, the measured total concentration is not an adequate surrogate for the free concentration in some antibiotics studies [
41,
42]. In the present study, a positive correlation between the total and free concentrations was found (Fig.
2). We attempted to use the clinical data of patients to develop a multivariate linear regression equation (stepwise regression) to predict the free concentration. However, these variables were not included in the final equation. Furthermore, although the free
Cmax (B1) was not independently associated with AKI in this study, the significant difference between the two groups in the univariate analysis (Table
3) suggests that the free
Cmax (B) and
Cmax (B1) may also be associated with AKI. Our results indicated that monitoring of free drug concentrations should be considered in the management of critically ill patients administered PMB.
In the present study, we preliminarily analysed the in vivo PMB1/PMB2 ratio. The PMB1/PMB2
Cmin ratio and PMB1/PMB2
Cmax ratio in patients ranged from 1.53 to 6.76 and from 1.48 to 4.44, respectively. The ratio of PMB1/PMB2 was less than 80% of that in the preparation in most patients and exhibited noticeable individual differences (Fig.
3). The PMB1/PMB2 ratio in the plasma was related not only to the ratio in the preparation but also to PK parameters such as volume of distribution (
Vd) and
t1/2. Reports on the PK parameters of PMB1 and PMB2 in patients are limited. Wang et al. reported the PK parameters of PMB1 and PMB2 in 15 patients, and a significant interindividual difference in
Vd was observed [
16].
Vd is influenced by plasma protein binding [
43]. For PMB, there were significant individual differences in the plasma protein binding rate among patients [
25,
26]. We speculated that interindividual differences in the PK parameters of PMB1 and PMB2 in different populations could be a reasonable explanation for the discrepancy. Considering that
Cmax (B1) was independently related to AKI, TDM of PMB1 and PMB2 should be warranted.
The clinical risk factors for PMB-induced nephrotoxicity reported in previous studies include age, baseline SCr levels, body mass index (BMI), concomitant use of medications such as vasoactive drugs and vancomycin, and the infection site [
18‐
22]. Notably, the identification of risk factors had yielded mixed results from different studies. For example, Mendes et al. [
19] and Han et al. [
22] found that the baseline SCr level is a risk factor for PMB-induced nephrotoxicity. However, three other reports [
21,
44,
45] involving critically ill patients indicated that the baseline SCr level was not associated with PMB-induced nephrotoxicity. This difference might be related to differences in the severity of illness in the patients included in the different studies. For instance, the in-hospital mortality rate (61.4%) of patients in the study by Mendes et al. [
19] was higher than that (23–42%) of patients in the other three studies [
21,
44,
45] that reported that the SCr level is not associated with PMB-induced nephrotoxicity. CrCL is calculated according to age, weight and sex and may be a more accurate index for evaluating renal function than the SCr level. However, it was not identified as a risk factor in our research or two other studies [
20,
43]. The BUN level is another indicator of renal function and was found to be associated with kidney injury induced by other antibiotics, such as vancomycin-related nephrotoxicity [
46]. It should be noted that BUN may be too nonspecific for kidney injury [
47], and elevated BUN may be caused by non-renal factors such as protein intake, catabolic state, upper gastrointestinal bleeding, volume status and therapy with high-dose steroids [
48‐
51]. In the present study, although BUN eventually entered a multivariate logistic regression model, the ROC curves (Fig.
5) showed the predictive power of BUN is lower than that of the multivariate logistic regression model, which indicated that the predictive power of BUN may be limited when it was used as a single indicator for predicting drug-related AKI.
The kidney is an essential organ for blood pressure regulation and one of the main target organs damaged by hypertension [
52]. A previous study has reported that hypertension may be a potential risk factor for colistin-induced nephrotoxicity [
53]. In our study, hypertension was identified as another risk factor for PMB-induced nephrotoxicity. The mechanism of this damage may be related to oxidative stress and haemodynamics [
54]. In addition, previous studies have found that kidney injury molecule 1 (KIM-1), a factor associated with diabetic nephropathy, may be associated with colistin-induced nephrotoxicity [
55‐
57]. In our study, we found that the incidence rates of type 2 diabetes and chronic renal dysfunction were different between the two groups in the univariate analysis (Table
2). Although the difference became weak in the multivariate analysis, it did not increase the discriminative ability of the final logistic regression model. The meta-analysis [
58] reported that underlying diabetes mellitus was a risk factor for polymyxin-induced nephrotoxicity. Although this meta-analysis focused on polymyxins, and polymyxin B and colistin were not analysed separately, the possible risk of AKI in patients with diabetes should be considered.
In our study, the onset of nephrotoxicity in patients ranged from day 3 to day 12. Four patients (16%) experienced nephrotoxicity on day 3. Considering that early PMB-induced nephrotoxicity on day 3 is a predictive factor for later nephrotoxicity [
44], early monitoring of renal function during PMB treatment is necessary.
The area under the ROC curve (AUC) of the final multivariate logistic regression model reached 0.799, which is similar to the AUC (0.813) of Han et al.’s combined predictor (
Cmin (B) and baseline SCr level) [
22]. Sorlí and colleagues also found that trough plasma level is an independent risk factor for colistin-induced AKI [
33]. The optimal cut-off trough concentrations for predicting PMB-related and colistin-related nephrotoxicity in these two studies were 3.55 mg/L and 3.33 mg/L, respectively. However,
Cmin (B) was not included in the final logistic model in our study. This may have been in part related to the relatively low
Cmin (B) in patients included in our study. Considering that the
Cmin (B) showed a significant difference in univariate analysis (Table
3), we attempted to develop a univariate logistic regression model to observe the correlation between
Cmin (B) and the probability of AKI development. The results showed that the predicted risk of AKI reached 50% when the
Cmin (B) was 3.63 mg/L or higher. However, when the
Cmin (B) was used alone to distinguish patients with AKI, the area under the ROC curve was normal (0.666). The predictive value of the
Cmin (B) for AKI should also be considered in future research. The multivariate model (Eq.
3) is convenient for calculating the predicted risk probability of AKI according to the
Cmax (B1), presence of hypertension and baseline BUN levels. When data such as baseline BUN levels are missing, Eq.
4 can also provide a preliminary method for predicting AKI risk. Because samples of
Cmax (B) were collected 10 min after completion of PMB infusion and
Cmin (B) was obtained immediately before infusion in our study, we contend that
Cmax (B) and
Cmax (B1) may be helpful for the early prediction of PMB-induced AKI in critically ill patients. Overall, our model might provide a helpful method for early identification of patients with a high risk of AKI and formulate corresponding intervention strategies.
The study had several limitations. First, this was a retrospective study. We could not analyse other possible risk factors, such as more sensitive indicators reflecting early renal function injury. Second, we were unable to determine the PK parameters in patients to assess the relationship between the AUC and AKI risk of PMB, and the difference in PK parameters between PMB1 and PMB2 was not further assessed. Third, although the enrolled patients came from three medical centres, the sample size was relatively small, limiting the generalizability of our results. Fourth, although the TDM of PMB was recommended by the guideline, the quantification method of PMB1 has not yet been performed in numerous laboratories, which might limit the application of our findings in clinical routine practice.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.