Amphotericin B Deoxycholate in adults with Cryptococcal Meningitis; a Population Pharmacokinetic Model and Meta-Analysis of Outcomes.

Amphotericin B Deoxycholate in adults with Cryptococcal Meningitis; a Population 1 Pharmacokinetic Model and Meta-Analysis of Outcomes 2 3 Katharine E Stott, Justin Beardsley, Sarah Whalley, Freddie Mukasa Kibengo , Nguyen 4 Thi Hoang Mai, Ruwanthi Kolamunnage-Dona, William Hope, Jeremy Day 5 6 a Antimicrobial Pharmacodynamics and Therapeutics Laboratory, Department of Molecular 7 and Clinical Pharmacology, Institute of Translational Medicine, University of Liverpool, UK 8 b Malawi-Liverpool-Wellcome Trust Clinical Research Programme, Blantyre, Malawi 9 c Oxford University Clinical Research Unit, Ho Chi Minh City, Viet Nam 10 d MRC/UVRI Uganda Research Unit on AIDS, Entebbe, Uganda 11 e Hospital for Tropical Diseases, Ho Chi Minh City, Viet Nam 12 f Department of Biostatistics, Institute of Translational Medicine, University of Liverpool, UK 13 g Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, 14 University of Oxford, UK 15 * Corresponding author: hopew@liverpool.ac.uk 16 17


ABSTRACT 23
There is a limited understanding of the population pharmacokinetics (PK) and 24 pharmacodynamics (PD) of amphotericin B deoxycholate (DAmB) for cryptococcal 25 meningitis (CM). A PK study was conducted in n=42 patients receiving DAmB 1 mg/kg q24h. 26 A 2-compartment PK model was developed. Patient weight influenced clearance and 27 volume in the final structural model. Monte Carlo simulations estimated drug exposure 28 associated with various DAmB dosages. A search was conducted for trials reporting 29 outcomes of CM patients treated with DAmB monotherapy and a meta-analysis was 30 performed. 31 The PK parameter means (standard deviation) were: clearance, 0.03 (0.01) x weight 32 + 0.95 (0.02) litres/hour; volume, 0.89 (0.90) x weight + 1.54 (1.13) litres; first-order rate 33 constant from central to peripheral compartment, 7.12 (6.50) hours -1 ; from peripheral to 34 central compartment, 12.13 (12.50) hours -1 . The meta-analysis suggested that DAmB dosage 35 explained most of the heterogeneity in cerebrospinal fluid (CSF) sterility, but not in 36 mortality outcomes. Simulations of area under concentration-time curve  ) 37 resulted in median (interquartile range) values 5.83 mg.h/litre (4.  Cryptococcal meningitis is a leading infectious cause of morbidity and mortality 48 worldwide, with approximately 223,100 incident cases and 181,100 deaths annually (1). The 49 ten-week mortality for patients receiving the current standard-of-care is 24-31% (2-5). 50 There have been no new antifungal agents developed for use in low-to-middle income 51 countries in the last 3 decades. Given the paucity of new agents, one important strategy for 52 improving clinical outcomes is a better understanding and use of currently available 53

compounds. 54
Amphotericin B (AmB) is a polyene antifungal agent with broad spectrum activity 55 against yeasts and moulds, as well as some parasites. AmB was initially isolated from a 56 streptomycete and described in 1955 (6). AmB was the first therapeutic option for 57 treatment of lethal invasive fungal diseases such as cryptococcal meningitis (7,8). 58 Amphotericin B deoxycholate (DAmB) is the most potent formulation of AmB on a mg-mg 59 basis (9, 10) and is a mainstay for the treatment of cryptococcal meningitis. 60 Clinical studies have progressively examined escalating dosages of 0.4mg/kg q24h 61 (11, 12), 0.7mg/kg q24h (13-15) and 1.0mg/kg q24h (5) of DAmB for cryptococcal 62 meningitis. The primary motivation of these studies was identification of the dosage that 63 induces maximal antifungal activity. A regimen of 0.7 -1.0 mg/kg q24h in combination with 64 flucytosine for two weeks is currently recommended for induction therapy (16). Higher Herein, we describe the development of a population pharmacokinetic model of 71 DAmB. In addition, a meta-analysis of clinical trials of DAmB monotherapy was performed 72 to estimate the contribution of various DAmB dosages to the observed heterogeneity in 73 study outcomes. Finally, Monte Carlo simulations were performed to estimate the mean, 74 median and dispersion of drug exposures that are associated with microbiological and 75 clinical outcomes from DAmB monotherapy. 76 77 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from 1 respectively for each regression (Table 3). The mean parameter values predicted the 126 observed values better than the medians. The measures of population bias and imprecision 127 were comparable between the models, with bias -0.85, -0.34, -0.23 and -0.43 and 128 imprecision 3.13, 2.97, 2.16 and 3.29 for Models 1, 2, 3 and 4 respectively. The more 129 positive log likelihood value and lower Akaike information criterion (AIC) for Model 2 130 implied that the inclusion of weight as a covariate explained a portion of the observed 131 variance. 132 The model that incorporated an exponential term for clearance (Model 3) decreased 133 the log likelihood value and increased the AIC (Table 3). The inclusion of eGFR in Model 4 134 failed to increase the log likelihood value or reduce the AIC further. In addition, there was 135 no statistically significant difference between Model 1 and either Model 3 or Model 4; the 136 latter models were therefore rejected. Model 2 was chosen as the final model. Observed-

Meta-analysis of clinical outcome data 141
Five clinical trials that included a DAmB monotherapy arm were identified. There 142 was one trial in which 63 patients received 0.4 mg/kg q24h (11), 3 in which a combined 208 143 patients received 0.7 mg/kg q24h (13-15) and 1 in which 99 patients received 1.0 mg/kg 144 q24h (5). An additional study that reported clinical outcomes in untreated cryptococcal 145 meningitis patients was also included. The baseline variables and clinical outcomes of these 146 study arms are summarised in table 4. Due to the small number of studies, we were unable 147 to adjust for baseline variables that may have had an impact on outcome measures (age, 148 CD4 cell count, baseline level of consciousness, baseline fungal burden and baseline 149 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from cryptococcal antigenaemia). The forest plots of the dose-adjusted random effects model are 150 shown in Figure 5. The model suggests that dose adjustment accounts for 77% of the 151 heterogeneity in CSF sterility (p=0.007), but did not have a significant impact on the 152 heterogeneity in either 2-or 10-week mortality outcomes (33%, p= 0.139 and 39%, p=0.092 153 respectively). We conducted a PK study in HIV positive adults with cryptococcal meningitis in 168 regions of high disease burden, and developed a population PK model that enabled the 169 extent of interpatient variability to be quantified. We described the PK of DAmB using a 2-170 compartment PK model with i.v. infusion and first-order clearance of drug from the central 171 compartment. Simulated AUCs reveal relatively modest PK variability, suggesting that the 172 frequently poor clinical outcomes are not the result of significant PK variability. The 173 relationship between weight and drug clearance suggests that weight accounts for a portion 174 of the observed variance. Dosage adjustment on the basis of weight is necessary to ensure 175 lighter patients are not over-dosed and heavier patients are not under-dosed. However, the 176 lack of impact of either eGFR or ethnicity on the PK suggests that dosage adjustment for 177 these variables is not necessary to achieve comparable drug exposure across patient 178

populations. 179
The model-simulated median AUC of 10.17 mg.h/L following a regimen of 0.7 mg/kg 180 q24h is consistent with AUCs estimated using non-compartmental techniques. For example, 181 Bekersky et al calculated an AUC 0-24 of 13.9 +/-2 mg.h/L after 0.6 mg/kg i.v. in healthy 182 volunteers (20). However, the simulations following 1mg/kg resulted in a median AUC of 183 14.52 mg.h/L, which is considerably lower than that derived from a non-compartmental 184 analysis (NCA) conducted by Ayestarán et al for the same dose administered to neutropenic 185 patients (28.98 +/-15.46 mg.h/L) (21). The reason for this is not immediately clear but may 186 relate to physiological differences between these two critically unwell patient cohorts (22). 187 Our meta-analysis of clinical outcomes from studies of DAmB monotherapy is limited 188 by the fact that the included studies recorded CSF sterility at diverse time points ranging 189 from 2 weeks (14) to 10 weeks (11). Nevertheless, the meta-analysis suggests that the 190 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from dosage of DAmB has a significant impact on the proportion of patients with sterile CSF and 191 that achieving CSF sterility is dose-dependent up to 1mg/kg q24h. However, DAmB does not 192 have a dose-dependent relationship with mortality at either 2 or 10 weeks. 193 The potential reasons that DAmB dosage has a positive impact on CSF sterilisation, 194 but not mortality are as follows: first, AmB toxicity may contribute to mortality (18,19). nosocomial bacteraemia may occur in up to 15-18% of patients hospitalised for cryptococcal 206 meningitis (26). Third, fungal burden -and therefore conceivably, time to CSF sterility -is 207 just one of multiple clinical variables associated with mortality in cryptococcal meningitis. 208 Older age, altered mental status, low body weight, high peripheral white blood cell count 209 and anaemia are independently associated with mortality at either 2 or 10 weeks (4). 210 Immune reconstitution inflammatory syndrome (IRIS) remains a significant cause of 211 mortality, occurring in 3-49% of cryptococcal meningitis patients surviving to initiation of 212 antiretroviral treatment and carrying a mortality rate of up to 36% (17, 27). Raised 213 intracranial pressure is an additional factor associated with mortality and at least 1 214 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from therapeutic lumbar puncture imparts a relative survival advantage of 69% in the first 10 215 days of treatment (28). Finally, the trial cohorts included in the meta-analysis were from 216 diverse sites in Africa, Asia, Europe and the USA. Factors such as health seeking behaviour 217 and nutritional status may have influenced mortality outcomes. Our meta-analysis did not 218 include any baseline factors besides DAmB dosage and we are therefore unable to identify 219 whether they account for the heterogeneity in mortality that is not explained by DAmB 220

dosage. 221
The discordance between the influence that drug dosage has on CSF sterilisation and 222 mortality is reflective of a growing consensus that CSF sterility is just one of many 223 determinants of mortality in cryptococcal meningitis. A systematic review of 27 clinical trials 224 determined that there was no correlation between CSF sterility at 2 weeks and all-cause 225 mortality at either 2 or 10 weeks (29). The most biologically plausible explanation for this is 226 that fungal burden in the CSF may not reflect the extensive encephalitis that is characteristic 227 of cryptococcal meningitis (which is more accurately termed meningoencephalitis). 228 Histopathological defects are more marked in patients co-infected with HIV; fungi 229 accumulate in perivascular spaces, are deposited [predominantly extracellularly] in brain 230 parenchyma, and form granulomatous cryptococcomas in brain tissue (30, 31). It is 231 conceivable that brain parenchymal damage is a dominant determinant of mortality and 232 that clearance of fungi in CSF is not mirrored by clearance in the cerebrum and other CNS 233 subcompartments. CSF sterility is an imperfect surrogate for the extent to which drug has 234 penetrated to and sterilised the central nervous system. 235 The meta-analysis suggests a strong dose-exposure-response relationship. Higher 236 dosages are likely to be required to achieve efficacious drug exposure at the site of 237 infection. DAmB has a large molecular weight (924 g/mol) and complex binding properties 238 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from (32). It does not readily penetrate the intact blood-brain barrier. Its concentration in 239 meninges and cryptococcomas has been technically difficult to quantify in any finer detail 240 than brain homogenates in preclinical models (33,34). This challenge is compounded by a 241 lack of clarity regarding the DAmB concentration required for therapeutic efficacy at the site 242 of infection. Animal studies estimate that the cerebral concentration of DAmB at which the 243 suppression of growth is half-maximal is 0.02 mg/litre in mice and 0.154 mg/litre in rabbits 244 (33). AmB exposure above the level required to optimise antifungal activity appears only to 245 contribute to toxicity (33, 35). Our simulations suggest that the optimal plasma AUC value in  (39). 254 It may be the case that the maximal antifungal effect of DAmB is achieved with a 255 dose of approximately 0.7 mg/kg, or slightly higher, and that gains made above this dose in 256 terms of CSF sterility are offset by losses in terms of excessive toxicity. This may explain why 257 significant increases in the proportion of patients achieving CSF sterility are not mirrored by 258 reductions in mortality. The present analysis is not sufficient to more precisely define the 259 optimal dosage of DAmB. This is partly due to the lack of consensus regarding DAmB 260 exposure targets. We are unable to propose exposure targets based on our dataset, which 261 does not include site-specific PK or detailed toxicodynamic data. In addition, the 262 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from pharmacodynamic and clinical outcome data presented herein are derived from patient 263 cohorts that are distinct from the patients that provided samples for the PK analysis. DAmB 264 monotherapy at dosages of 0.7 mg/kg q24h and 1.0 mg/kg q24h has not been directly 265 compared in a randomised controlled trial. However, comparison of these dosages in 266 combination with 5FC has been performed. Bicanic et al demonstrated increased early 267 fungicidal activity with 1mg/kg q24h DAmB versus 0.7mg/kg q24h DAmB, both in 268 combination with 5FC 100mg/kg/day in four divided dosages, but this was not reflected in 269 reductions in mortality. A higher percentage of deaths was seen in the higher dose DAmB 270 arm at both 2 weeks (9% versus 3%) and 10 weeks (26% versus 21%) but this was not 271 statistically significant (p= 0.62 and 0.77 at 2 and 10 weeks, respectively) (2). 272 In summary, these analyses suggest that the optimal dosage of DAmB for the 273 treatment of cryptococcal meningitis lies between 0.7-1.0 mg/kg q24h. The precise drug 274 exposure target that optimises clinical outcomes without producing significant toxicity 275 remains to be defined. The extent of inter-individual PK variability in DAmB is modest and 276 unlikely to account for the consistently poor clinical outcomes of cryptococcal meningitis. The study protocols were approved by the relevant institutional review boards and 288 regulatory authorities at each trial site and by the Oxford University Tropical Research Ethics 289

Committee. 290
The protocol for the randomised controlled trial has been described previously (41). 291 Briefly, patients had HIV infection, a syndrome consistent with cryptococcal meningitis, and 292 laboratory evidence of cryptococcal infection. Patients who were pregnant, had renal 293 failure, had gastrointestinal bleeding, had received more than 7 days of anti-cryptococcal 294 antifungal therapy, were already taking corticosteroids, or required corticosteroid therapy 295 for co-existing conditions were excluded. The inclusion and exclusion criteria for the 296 prospective descriptive study were identical to those of the clinical trial. Patients received 1 297 mg/kg DAmB once daily by intravenous infusion over 5-6 hours, as well as 800mg 298 fluconazole per day. Two patients recruited during the clinical trial received dexamethasone 299 according to the following regimen: 0.3mg/kg/day intravenously (IV) for week 1, 300 0.2mg/kg/day IV for week 2, then orally 0.1mg/kg/day for week 3, 3mg/day week 4, 301 2mg/day week 5, 1mg/day week 6, then stop. For the first five patients enrolled, blood 302 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from samples were obtained immediately prior to intravenous DAmB infusion, and then at 1, 2, 4, 303 8, 12, 16, 20 and 24. The results for these patients informed a subsequent sampling strategy 304 defined using optimal design theory such that patients were sampled pre-dose, then at 1, 2, 305 4, 8, 12 and 24 hours after the initiation of infusion. PK sampling occurred on treatment 306 days 1 or 2, and 7. Whenever patients had lumbar punctures performed for other clinical 307 indications such as raised intracranial pressure, paired plasma samples were collected for 308 subsequent PK analysis. Therefore, additional sparse samples were taken up to 17 days after 309 initial dosing. Quantitative fungal counts were determined for each lumbar puncture, as 310 described previously (15). 311 312

Measurement of Amphotericin B Concentrations 313
Amphotericin B concentrations in plasma were measured using high-performance 314 liquid chromatography (HPLC) with a Shimadzu Prominence HPLC system (Shimadzu, Milton 315 Keynes, UK). Amphotericin B was extracted by protein precipitation. A total of 300 µL of 316 methanol that contained piroxicam 2 mg/L (Sigma Aldrich, Dorset, UK) as internal standard 317 was added to 100 µL of matrix. Samples were vortexed for 5 seconds and then centrifuged 318 at 13,000 x g for 3 minutes. 319 One hundred-fifty µL of supernatant was removed and placed in a 96-well plate, to 320 which 50 µL of water was added. A 50µL aliquot was injected onto a Kinetex 5µ XB-C18 321 liquid chromatography column (Phenomenex, Macclesfield, UK). Chromatographic 322 separation was achieved using a gradient with the starting conditions of 75% A:25% B (0.1% 323 formic acid in water as mobile phase A and 0.1% formic acid in acetonitrile as mobile phase 324 B). Mobile phase B was increased to 80% over five minutes and then reduced to starting 325 conditions for two minutes of equilibration. Amphotericin B and internal standard were 326 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from detected using UV detection at wavelengths of 406nm and 385nm; they eluted after 4.1 and 327 4.6 minutes, respectively. 328 The standard curve for amphotericin B encompassed the concentration range 0.05-329 8.0 mg/L and was constructed using blank matrix. The limit of quantitation was 0.05 mg/L. 330 The coefficient of variation was <9.3% over the concentration range 0.05-8 mg/L. The intra-331 and inter-day variation was <7.9%. weight and eGFR with estimated PK parameters. Univariate linear regression was 364 employed, firstly to assess the relationship between patient weight and the Bayesian 365 estimates for both clearance and volume. Since a positive relationship was observed 366 between weight and both PK parameters, the population PK model was re-fitted to the data 367 (Model 2) with incorporation of the following equations to describe (e) clearance (SCL), and 368 To explore whether there were significant differences between the model predicted 393 PK parameters in Vietnamese and Ugandan patients, Bayesian estimates of volume of 394 distribution and clearance from the central compartment were compared using a Mann-395 on July 23, 2018 by guest http://aac.asm.org/ Downloaded from Whitney test and a Student's t-test respectively. Since no significant relationship between 396 ethnicity and DAmB PK was apparent, this variable was not incorporated in the final model. 397 The fit of the model to the data was assessed using a linear regression of observed- demonstrated ability to predict patient mortality were selected a-priori and extracted from 416 the studies; namely altered mental status, patient age and baseline CSF fungal burden (4, 417 29). To aid meaningful trial comparison, baseline fungal burden and baseline CSF 418 cryptococcal antigen titre were extrapolated from one another where they were not 419 explicitly reported in the study, applying a correlation presented by Jarvis et al (4). 420 We collated a variety of clinical trial outcomes based on those that were commonly 421 reported across trials of DAmB monotherapy: documented CSF sterility during trial follow-422 up, mortality at 2 weeks and, where possible, mortality at 10 weeks. Meta-analysis was 423 performed on each outcome using a dose-adjusted random effects model to account for the 424 baseline heterogeneity in the included studies. We included dose as a moderator variable in 425 the model to assess the degree to which it explained heterogeneity in clinical outcome (45). 426 The resulting mixed-effects model took the form: 427 where 0 and 1 are the model parameters intercept and dose respectively; is the 429 dose given in the th study, assuming study-specific random effects; and ~(0, 2 ), 430 where 2 is the amount of residual heterogeneity among the true effects that is not 431 dose that was ultimately administered was determined by the patient's weight. 443 Drug exposure was quantified using the DAmB AUC (9, 10, 46). The simulated AUC 444 for each patient was estimated 144 to 168 hours post therapy initiation. Simulations were 445 performed to estimate the AUC that resulted from dosages administered in clinical trials of 446 DAmB monotherapy for which PD measures were available -specifically, 0.4, 0.7 and 447 1.0mg/kg q24h (5,11,(13)(14)(15)