Exploration of Clinical Breakpoint of Danofloxacin for Glaesserella parasuis in Plasma and in PELF

Background: In order to establish the clinical breakpoint (CBP) of danofloxacin against G. parasuis, three cutoff values, including epidemiological cutoff value (ECV), pharmacokinetic-pharmacodynamic (PK-PD) cutoff value (COPD) and clinical cutoff value (COCL), were obtained in the present study. Methods: The ECV was calculated using ECOFFinder base on the MIC distribution of danfloxacin against 347 G. parasuis collected from disease pigs. The COPD was established based on in vivo and ex vivo PK-PD modeling of danofloxacin both in plasma and pulmonary epithelial lining fluid (PELF) using Hill formula and Monte Carlo analysis. The COCL was established based on the relationship between the possibility of cure (POC) and MIC in the clinical trials using the “WindoW” approach, nonlinear regression and CART analysis. Results: The MIC50 and MIC90 of danofloxacin against 347 G. parasuis were 2 μg/mL and 8 μg/mL, respectively. The ECV value was set to 8 μg/mL using ECOFFinder. Concentration-time curves of danofloxacin were fitted with a two-compartment PK model. The PK parameters of the maximum concentration (Cmax) and area under concentration-time curves (AUC) in PELF were 3.67 ± 0.25 μg/mL and 24.28 ± 2.70 h·μg/mL, higher than those in plasma (0.67 ± 0.01 μg/mL and 4.47 ± 0.51 h·μg/mL). The peak time (Tmax) in plasma was 0.23 ± 0.07 h, shorter than that in PELF (1.61 ± 0.15 h). The COPD in plasma and PELF were 0.125 μg/mL and 0.5 μg/mL, respectively. The COCL calculated by WindoW approach, nonlinear regression and CART analysis were 0.125–4 μg/mL, 0.428 μg/mL and 0.56 μg/mL, respectively. The 0.5 μg/mL was selected as eligible COCL. The ECV is much higher than the COPD and COCL, and the clinical breakpoint based on data in plasma was largely different from that of PELF. Conclusions: Our study firstly established three cutoff values of danofloxacin against G. parasuis. It suggested that non-wild-type danofloxacin-resistant G. parasuis may lead to ineffective treatment by danofloxacin.


Introduction
Glaesserella parasuis, a gram-negative respiratory pathogen, can colonize the upper respiratory tract in swine and cause Glasser's disease with clinical manifestations such as fibrinous polyserositis, arthritis, meningitis and pneumonia [1]. The serotypes 1, 5, 10, 12, 13 and 14 exhibit higher virulence and pathogenicity [2]. Serotypes 5 and 4 are dominant in China [3]. With the abuse of antibiotics, antimicrobial-resistant G. parasuis emerge in different degrees, which bring serious threat to the global economy and public health [4].
From March to May in 2017, a total of 347 G. parasuis strains were collected from disease animals. Thirty-five G. parasuis strains were isolated from pig lungs provided by Keqian clinical diagnostic center; 8 G. parasuis strains were donated by Xiaojuan Xu from State Key Laboratory of Agricultural Microbiology in Huazhong Agricultural University; 204 G. parasuis strains were isolated from disease pigs by Peng Zhang in China Agricultural University; 100 G. parasuis strains were stored in National Reference Laboratory of Veterinary Drug Residues. All these strains were isolated from the lungs and pericardium of weak or moribund pigs showing respiratory distress or arthritis in different provinces of China. All bacterial isolates were confirmed by PCR amplification of 16S rRNA ( Figure S1) [18]. E. coli (ATCC 25922) was used as the quality control strain (QC).

Animals
Seventy-eight six-week-old healthy crossbred (Duroc × Large × white × Landrace) pigs weighing 20 ± 2 kg were purchased from Huazhong Agricultural University pig breeding farm. Prior to experiments, pigs were raised for 7 days to acclimatize. Sixteentwenty g healthy Balb/c mice were purchased from the Experimental Animal Center of Huazhong Agricultural University. Prior to experiments, mice were raised for 7 days to acclimatize. All the animal experiments were approved by the Animal Ethics Committee of Huazhong Agricultural University (hzauch 2014-003) and the Animal Care Center, Hubei Science and Technology Agency in China (SYXK2013-0044). All animal experiments were conducted according to the committee guidelines for the Laboratory Animal Use and Care Committee in Hubei Science and Technology Agency. All efforts were used to reduce the pain and adverse effect of the animals.
The 18 strains of serotype 5 were selected for the mouse pathogenicity test. The 16-20 g healthy Balb/c mice were randomly divided into 19 groups (5 mice/group) with one blank control group. The mice were inoculated with 1 × 10 9 cfu bacteria by abdominal cavity injection, and the control group was injected with TSB broth. Mice were monitored daily for 7 days post-inoculation (dpi). The pathogenicity of G. parasuis was compared according to the survival time [23].

Pharmacodynamics In Vitro and Ex-Vivo
The MIC and MBC of G. parasuis H80 in broth and pulmonary epithelial lining fluid (PELF) were determined using the broth dilution method according to the CLSI M07-A9 standard with some modification.
The in vitro and ex vivo killing curves of danofloxacin in broth and in PELF were drawn by monitoring the Colony formed unite (CFU) changes during the incubation of G. parasuis H80 under a series concentration of danofloxacin (1/2 to 32 MIC) for a continuous time period (0, 1, 2, 4, 6, 8, 12 and 24 h).

Quantitation Analysis of Danofloxacin by HPLC
Quantitation analyses of danofloxacin in PELF and plasma were conducted using high-performance liquid chromatography (HPLC). Agent SB-Aq reverse-phase column (250 mm, 4.6 mm i.d., 5 mm; Agilent) was used to perform HPLC at 30 • C. The detection wavelength was 280 nm. The mobile phase consisted of 0.05% phosphoric acid (phase A) and acetonitrile (phase B) with gradient elute. The peak time of danofloxacin was 10.64 min. 0.5 mL Plasma and 0.5 mL PELF were extracted with 2 mL acetonitrile twice.
The urea dilution method was used to determine the volume of PELF as described previously [26,27]. The concentration of urea in plasma (Urea PLASMA ) and PELF (Urea PELF ) were determined by using a urea test kit (Urea test kit; Sigma Chemical, St. Louis, MO, USA) and the absorbance values measured by using a spectrophotometer (Agilent 8453, ) was used to calculate the AUC 24 /MIC (AUIC) of danofloxacin at different concentrations, E is the summary PD endpoint, and E 0 is the effect representing the value of the PD endpoint without drug treatment (i.e., the value of the summary endpoint when the PK-PD index is 0). X is one of the three PK-PD indices as defined above, and PD max is the maximum effect (in relation to E 0 ) indicated by the plateau where increased exposures result in no further kill. EC 50 is the magnitude of X that is needed to achieve 50% of PD max , and γ is the sigmoidicity factor. The PD target under different efficiency (E = 0, −3 and −4 (bacteriostasis, bactericidal and eradication)) was determined with Sigmoid E max equation [28,29]. The dosage regimen was derived from the concentration-dependent dosage equation (Dose = MIC×AUIC fu × CL/F) [30][31][32]. In the equation, the CL (mL/h) was the plasma (total) clearance per day, AUIC (h) was the targeted endpoint for optimal efficacy, fu was the free fraction of the drug in PELF (from 0 to 1), and F was the bioavailability factor (from 0 to 1). In this study, fu was 0.8974, which was obtained by measuring the protein binding rate by the equilibrium dialysis method.

Monte Carlo Simulation to Set up CO PD
Crystal Ball v7.2.2 was used to perform the Monte Carlo simulation. The distribution of the PK-PD parameter was assumed to be log-normal. A total of 10,000 subjects were simulated. The PD target was selected to calculate the probability of target attainment (PTA). CO PD was defined as the MIC at which the PTA was ≥90%. Sixty-six healthy weaned piglets (20 ± 2 kg) were divided into 11 groups: 5 groups were the experimental group, 5 groups were the negative control group, and 1 group was the blank control group, with 6 piglets in each group. The 5 experimental groups and 5 negative control groups were challenged with 5 representative strains, H42, H80, H12, H83 and H17, by intranasal inoculation of 1 × 10 10 CFU bacterial suspension twice a day. The blank control group was inoculated with blank TSB broth. The dosage regimens were recommended by the PK-PD therapeutic dosage regimen. After challenging, these pigs were monitored daily for two weeks.

Statistical Analysis for Establishment of CO CL
The probability of cure (POC) was calculated based on the clinical outcomes and bacteriological prognosis. Clinical outcomes included treatment success and failure, and each MIC should have a corresponding clinical outcome. The bacteriological prognosis was to determine the presence or eradication of the bacteria after administration. The data were analyzed by three different analysis methods.
The "WindoW" approach [17] included two parameters: "MaxDiff" and "CAR". "MaxDiff (the method of maximum difference, MaxDiff)" represents the difference between higher and lower POC at a certain MIC. "CAR" was based on the clinical outcome and the corresponding MIC distribution. "CAR" could not be set as the lowest MIC or the highest MIC if "CAR" was gradually increasing with MIC, then the "CAR" should choose the second small "CAR".
Nonlinear regression analysis was a new method based on the formula between EUCAST proposed POC with MIC. Log 2 MIC was the independent variable, and the POC Antibiotics 2021, 10, 808 5 of 13 was the dependent variable. The model with the highest correlation coefficient was selected to simulate its CO CL .
The classification and regression tree (CART) model (Salford Predictive Modeler software) was also used for the establishment of CO CL . MIC was used as the predictive variable, and the POC was the target variable. The Gini coefficient minimization criterion was used to select the MIC node automatically.
Antibiotics 2021, 10, x FOR PEER REVIEW 5 o and the corresponding MIC distribution. "CAR" could not be set as the lowest MIC or highest MIC if "CAR" was gradually increasing with MIC, then the "CAR" should cho the second small "CAR". Nonlinear regression analysis was a new method based on the formula between E CAST proposed POC with MIC. Log2MIC was the independent variable, and the POC w the dependent variable. The model with the highest correlation coefficient was selected simulate its COCL.
The classification and regression tree (CART) model (Salford Predictive Modeler s ware) was also used for the establishment of COCL. MIC was used as the predictive va ble, and the POC was the target variable. The Gini coefficient minimization criterion w used to select the MIC node automatically.

Pharmacodynamics of Danofloxacin against G. parasuis
The MICs of danofloxacin in broth and pulmonary epithelial lining fluid (PELF) were 4 µg/mL and 2 µg/mL, respectively. The MBC in broth and PELF were 8 µg/mL and Antibiotics 2021, 10, 808 6 of 13 4 µg/mL, respectively. The antibacterial activity of danofloxacin in PELF is stronger than that of in broth.
As displayed in Figure 2, the in vitro and ex vivo bactericidal effect of danofloxacin against G. parasuis was similar. The lower concentrations (≤MIC) of danofloxacin exhibited similar antibacterial activity to G. parasuis. However, when danofloxacin concentrations were higher than MIC, the inhibitory efficiency gradually strengthened following the increased drug concentration. The time-killing curve showed that the activity of danofloxacin against G. parasuis was concentration-dependent. The Aera Under Curve/Minimum Inhibitory Concentration (AUC/MIC) was selected as the PK-PD parameter.

Pharmacodynamics of Danofloxacin against G. parasuis
The MICs of danofloxacin in broth and pulmonary epithelial lining fluid (PELF) were 4 μg/mL and 2 μg/mL, respectively. The MBC in broth and PELF were 8 μg/mL and 4 μg/mL, respectively. The antibacterial activity of danofloxacin in PELF is stronger than that of in broth.
As displayed in Figure 2, the in vitro and ex vivo bactericidal effect of danofloxacin against G. parasuis was similar. The lower concentrations (≤MIC) of danofloxacin exhibited similar antibacterial activity to G. parasuis. However, when danofloxacin concentrations were higher than MIC, the inhibitory efficiency gradually strengthened following the increased drug concentration. The time-killing curve showed that the activity of danofloxacin against G. parasuis was concentration-dependent. The Aera Under Curve/Minimum Inhibitory Concentration (AUC/MIC) was selected as the PK-PD parameter.

Sensitivity and Accuracy of HPLC Method for Determination of Danofloxacin
The limit of determination (LOD) was 0.01 μg/mL, and the limit of quantification (LOQ) was 0.025 μg/mL in PELF. The LOD was 0.02 μg/mL, and the LOQ was 0.05 μg/mL in plasma. Standard curves were linear from 0.05 μg/mL to 5 μg/mL in plasma (R 2 = 0.9994) and 0.025 μg/mL to 2.5 μg/mL in PELF (R 2 = 0.9996). The inter-day variation for determination in plasma and PELF ranged from 1.94% to 2.37% and 1.36% to 2.71%, respectively. The recovery of danofloxacin in plasma and PELF ranged from 90.79 ± 2.15 to 94.36 ± 1.83 and 91.91 ± 2.49 to 95.73 ± 1.30, respectively.

PK Characteristics of Danofloxacin in Plasma and PELF
The concentration-time curves in plasma and PELF after administration of danofloxacin at a single dose of 2.5 mg/kg b.w. are shown in Figure 3. Concentrations of danfloxa-

Sensitivity and Accuracy of HPLC Method for Determination of Danofloxacin
The limit of determination (LOD) was 0.01 µg/mL, and the limit of quantification (LOQ) was 0.025 µg/mL in PELF. The LOD was 0.02 µg/mL, and the LOQ was 0.05 µg/mL in plasma. Standard curves were linear from 0.05 µg/mL to 5 µg/mL in plasma (R 2 = 0.9994) and 0.025 µg/mL to 2.5 µg/mL in PELF (R 2 = 0.9996). The inter-day variation for determination in plasma and PELF ranged from 1.94% to 2.37% and 1.36% to 2.71%, respectively. The recovery of danofloxacin in plasma and PELF ranged from 90.79 ± 2.15 to 94.36 ± 1.83 and 91.91 ± 2.49 to 95.73 ± 1.30, respectively.

PK Characteristics of Danofloxacin in Plasma and PELF
The concentration-time curves in plasma and PELF after administration of danofloxacin at a single dose of 2.5 mg/kg b.w. are shown in Figure 3. Concentrations of danfloxacin in plasma and PELF at various time points are shown in Table S2. A striking difference is observed between drug concentrations in plasma and in PELF.

PK Characteristics of Danofloxacin in Plasma and PELF
The concentration-time curves in plasma and PELF after administration of danofloxacin at a single dose of 2.5 mg/kg b.w. are shown in Figure 3. Concentrations of danfloxacin in plasma and PELF at various time points are shown in Table S2. A striking difference is observed between drug concentrations in plasma and in PELF. The estimated pharmacokinetic parameters in plasma and PELF were shown in Table 1. Distribution of danofloxacin in simulated drug time curve in plasma and in PELF were shown in Figures S3 and S4. In plasma, the peak time (T max ) was 0.23 ± 0.07 h, the peak concentration (C max ) was 0.67 ± 0.01 µg/mL, the area under the concentration-time curves (AUC) was 4.47 ± 0.51 h·µg/mL; in PELF, T max was 1.61 ± 0.15 h, C max was 3.67 ± 0.25 µg/mL, AUC was 24.28 ± 2.70 h·µg/mL.  Combined with the killing curve in PELF, the PD target (AUIC in ex vivo) under different efficiency was calculated by Sigmoid E max equation simulation ( Table 2). The values of AUIC (h) at E = 0, −3 and −4 (bacteriostasis, bactericidal and eradication) were 12.73, 28.68 and 44.38, respectively. C vivo is the concentration of danofloxacin in PELF; (AUIC) ex is selected PK-PD parameters; a represented the bacterial colonies lower than the limit of detection (10 CFU/mL).

CO CL of Danofloxacin against G. parasuis
The dosage under different efficiency (bacteriostasis, bactericidal and eradication) were 4.58 mg/kg, 10.32 mg/kg and 15.97 mg/kg. The given dosages were simulated by Mlxplore software (Figure S7). The modified dosage regimen was 12.49 mg/kg danofloxacin Antibiotics 2021, 10, 808 9 of 13 twice a day. Three methods were used to obtain CO CL according to the relationship between POC and MIC distribution (Table 4). Following the "WindoW" method, the parameters of MaxDiff (0.28) and CAR (0.78) was corresponding with the MIC of 0.125 µg/mL and 4 µg/mL, respectively. Therefore, the CO CL selection window range is 0.125 µg/mL to 4 µg/mL. The nonlinear regression model was set up as y = 80.989 − 7.271x + 0.271x 2 + 0.16x 3 with a correlation coefficient of 0.996. When POC was 90%, the recommended CO CL (MIC) was less than 0.428 µg/mL. The CART regression tree indicated that the CO CL was less than 0.56 µg/mL ( Figure S8). Combined with the above three results, the CO CL of danofloxacin against G. parasuis was selected as 0.25 µg/mL.

Discussion
G. parasuis is an important respiratory pathogen in swine. Antimicrobial treatment is the more effective way to control this pathogen due to vaccine deficiency. However, antimicrobial resistance in G. parasuis had been found in Germany [33], the United Kingdom, Spain [34] and China [35][36][37]. In order to rationally use antimicrobials agents to control G. parasuis, some studies have been conducted to establish the ECVs and/or CO PD of marbofloxacin, cefquinome and tilmicosin against G. parasuis [29,38,39]. Danofloxacin is very effective against Actinobacillus pleuropneumoniae [40], Pasteurella multocida [41] and Mannheimia haemolytica [42]. However, the clinical breakpoint of danofloxacin against G. parasuis had not yet been established.
Statistical analysis had been widely used for the determination of ECVs. Turnidge [13] recommends using nonlinear regression to analyze the obtained MIC data and determined the ECVs of various drugs. Kronvall [43] used NRI (Normalized Resistance Interpretation) method to analyze MIC data obtained by E test for the establishment of ECVs. European Commission of Antimicrobial Susceptibility Testing (EUCAST) recommended ECOFFinder software on the basis of Turnidge's nonlinear regression [44]. Van Vliet [45] used NRI and ECOFFinder analysis method to analyze wild-type cutoff values of ampicillin, florfenicol, gentamicin and enrofloxacin. In our study, the ECV of danofloxacin determined by nonlinear regression analysis was the same as that simulated by ECOFFinder software, suggesting that ECOFFinder software is a convenient tool for the establishment of ECVs. In the present study, the MIC distribution of danofloxacin against G. parasuis appeared three peaks (0.008 µg/mL, 0.125 µg/mL and 2 µg/mL), suggesting that some G. parasuis isolates may be resistant to danofloxacin. Zhang et al. [46] examined the resistance of 138 G. parasuis strains against fluoroquinolone drugs and showed that 60.1% of isolates were resistant to enrofloxacin, and 5.8% of isolates were resistant to levofloxacin. It suggested that G. parasuis may also be resistant to danofloxacin due to the cross-resistance between fluoroquinolone drugs.
The CO PD was established based on pharmacokinetic data, MIC distribution and PK-PD target. Our present study establishes the CO PD based on the PK data from healthy animals because of the stability and repeatability of a healthy animal model. Considering the drug concentrations in the target sites were directly correlated with clinical efficacy, the PK data both in plasma and in PELF were included in our study [47]. Similar to previous studies, our results indicated that the concentration and AUC of danofloxacin in PELF (in the lung) was 4-7 times higher than that in plasma [11]. The CO PD of danofloxacin in PELF was subsequently higher than the CO PD in plasma, indicating that the CO PD was different between in the target tissue and in plasma. As danofloxacin can be accumulated at the infection site (lung), the CO PD in plasma may not represent the critical value of the target tissue. It was of great significance to establish the CO PD in target tissue and plasma simultaneously. The differences in pharmacokinetic parameters between different studies may be due to differences in pig breeds or individuals. In this study, the T max of pigs after i.m. administration of danofloxacin at a dose of 2.5 mg/kg b.w. was 0.23 ± 0.07 h, and this result is different from the result reported by Yang [48] at 0.97 ± 0.08 h; C max was 0.67 ± 0.01 µg/mL, which is in good agreement with the previously reported 0.76 ± 0.08 µg/mL; the AUC 24h was 4.47 ± 0.51 h·µg/mL, which is less than 5.25 ± 1.35 h·µg/mL, as reported by Yang et al.
Previously, a study exhibited good clinical outcomes of danofloxacin in the treatment of respiratory disease caused by Haemophilus somnus and Pasteurella multocida in European cattle [49]. The clinical data in our study also showed the good clinical outcome of danofloxacin in the treatment of G. parasuis in pigs because the success rate for treatment of G. parasuis with MIC of 1 µg/mL was still as high as 83.33%. The CO CL was established based on the relationship between MIC and POC under modified therapeutic dosage. Since there was no standard approach for the establishment of CO CL , the CO CL in the present study was established by the combination of the three approaches, which included the "WindoW" approach [17], the nonlinear regression [50] and the CART analysis [51,52]. The "WindoW" approach was recommended by CLSI [17]. The nonlinear regression with the formula of POC = 1/(1 + e -a+bf (MIC) ) was proposed by VetCAST to calculate the relation between the dependent variable of POC and the independent variable of MIC [50]. The CART method was previously used to develop clinical breakpoints of cefepime [53], and this method was recommended by Dr. Cuesta [54] and Prof. Toutain [12] because the CART obtained the best statistical results when it was compared with other four supervised classifiers (J48, the OneR decision rule, the naïve Bayes classifier and simple logistic regression).
A large difference was observed between three cutoff values with ECV higher than CO PD and CO CL . In previous studies, there was data that showed the MIC breakpoint of danofloxacin against Mannheimia haemolytica and Pasteurella multocida was 1 µg/mL [55], while Yang's data showed that the epidemiologic cutoff value of danofloxacin against E. coli was 8 µg/mL [48], which was in accordance with our study. The difference of ECV between different studies may be due to the epidemiological characteristic of a different bacterial in different geography. Additionally, previous data showed that some of G. parasuis isolates exhibited decreased sensitivity to fluoroquinolones [56]. Three peaks of MIC distribution in the present data also suggested that some G. parasuis isolates may be resistant to danofloxacin. The higher MIC of the resistant isolates may contribute to the higher ECV value, and further studies may need to confirm the relationship between MIC phenotype and resistance genotype.

Conclusions
This study firstly established the ECV (8 µg/mL) at 95% confidence intervals, CO PD in PELF (0.5 µg/mL), CO PD in plasma (0.125 µg/mL) and CO CL (0.25 µg/mL) of danofloxacin against G. parasuis. Based on the CLSI decision tree, the final CBP in plasma and PELF was 0.25 µg/mL and 8 µg/mL, respectively ( Figure S9). The ECV value was higher than CO PD and CO CL , indicating that some G. parasuis isolates may be resistant to danofloxacin.

Supplementary Materials:
The following are available online at https://www.mdpi.com/article/10 .3390/antibiotics10070808/s1, Figure S1: Amplification of G. parasuis 16S rRNA with PCR, Figure S2: Results of ERIC-PCR for G. parasuis, Figure S3: Distribution of danofloxacin in simulated drug time curve in plasma, Figure S4: Distribution of danofloxacin in simulated drug time curve in PELF, Figure S5: PTA of danofloxacin against G. parasuis in PELF, Figure S6: PTA of danofloxacin against G. parasuis in plasma, Figure S7: Forecast growth of G. parasuis at different dosage regimens, Figure S8: CART tree showing values of clinical outcome, Figure S9: Susceptibility breakpoint decision tree, Table S1: Epidemiological MIC for danofloxacin against G. parasuis, Table S2: Concentrations of danfloxacin in plasma and PELF at various time points (n = 6).