PPAR-gamma agonist pioglitazone modifies craving intensity and brain white matter integrity in patients with primary cocaine use disorder: A double-blind randomized controlled pilot trial

Abstract

Background and aims

Pioglitazone (PIO), a potent agonist of PPAR-gamma, is a promising candidate treatment for cocaine use disorder (CUD). We tested the effects of PIO on targeted mechanisms relevant to CUD: cocaine craving and brain white matter (WM) integrity. Feasibility, medication compliance, and tolerability were evaluated.

Design

Two-arm double-blind randomized controlled proof-of-concept pilot trial of PIO or placebo (PLC).

Setting

Single-site outpatient treatment research clinic in Houston, Texas, USA.

Participants

Thirty treatment-seeking adults with CUD. Mean [standard deviation (SD)] age was 47.8 (7.45), education was 12.7 (1.5), with 19.3 (7.8) years of reported cocaine use. Eighteen of the 30 participants (8=PIO; 10=PLC) completed diffusion tensor imaging (DTI) of WM integrity at pre/post-treatment.

Intervention

Study medication was dispensed at thrice weekly visits along with once weekly cognitive behavioral therapy for 12 weeks.

Measurements

Measures of target engagement mechanisms of interest included cocaine craving assessed by the Brief Substance Craving Scale (BSCS), the Obsessive Compulsive Drug Use Scale (OCDUS), a visual analog scale (VAS), and change in WM integrity. Feasibility measures included number completing treatment, medication compliance (riboflavin detection), and tolerability (side effects, serious adverse events).

Findings

Target engagement change in mechanisms of interest, defined as a ≥ 0.75 Bayesian posterior probability of an interaction existing favoring PIO over PLC, was demonstrated on measures of craving (BSCS, VAS) and WM integrity indexed by fractional anisotropy (FA) values. Outcomes indicated greater decrease in craving and greater increase in FA values in the PIO group. Feasibility was demonstrated by high completion rates among those starting treatment (21/26 = 80%) and medication compliance (≥ 80%). There were no reported serious adverse events for PIO.

Conclusions

Compared with placebo, patients receiving pioglitazone show a higher likelihood of reduced cocaine craving and improved brain white matter integrity as a function of time in treatment. Pioglitazone shows good feasibility as a treatment for cocaine use disorder.

INTRODUCTION

Over 3 million people worldwide stand to benefit from the development of safe and effective medications for the treatment of cocaine use disorder (CUD) (1, 2). Left untreated, individuals with CUD are more likely to incur societal and economic costs, especially related to hospitalization and emergency department visits (3). Advances in our understanding of key neurobiological mechanisms associated with chronic cocaine use have moved the field beyond classic neurotransmitter systems toward novel targets for the development of medication treatments (4). One emerging target for the treatment of CUD and other drug use disorders is the peroxisome proliferator-activated receptor (PPAR) pathway (5).

The PPARs constitute a family of nuclear receptor transcription factors that modulate gene expression involved in key cell functions, including metabolism and inflammation (6). Activation of the PPAR gamma (PPAR-γ) isoform is the mechanism of action used in therapy to treat type-2 diabetes (7, 8). Importantly, PPAR-γ agonists exert a wide range of anti-inflammatory and anti-oxidative effects that have been shown to promote neuroprotection in models of neurodegenerative diseases such as Parkinson’s and Alzheimer’s disease (9–11).

Current data suggest at least two mechanisms through which PPAR-γ agonists may affect CUD. First, PPAR-γ activation enhances cognition by modulating extracellular signal-regulated kinases (ERK), a protein involved in hippocampal learning and memory consolidation, including memories for drug cues (12, 13). Recently, the PPAR-γ agonist pioglitazone (PIO) was shown to attenuate cocaine cue reactivity in rats; an effect that was accompanied by normalization of ERK activity in the medial prefrontal cortex (PFC) and hippocampus (14). Second, PPAR-γ activation may confer neuroprotection against drug-induced neurotoxicity. Evidence for cocaine neurotoxicity comes from many different forms of neuroimaging (15, 16), one of which is white matter (WM) integrity as measured by diffusion tensor imaging (DTI). Multiple studies, both preclinical (17, 18) and human (19–21) show impaired WM integrity is associated with functional consequences, including impulsivity (19), decision making (22) and relapse (23). Importantly, emerging evidence implicates WM integrity across a range of psychiatric disorders and cognitive processes relevant to addiction (24, 25). Based on (1) modulation of cocaine cue reactivity in animal models via PPAR-γ agonism, and (2) evidence for neuroprotective effects of PPAR-γ agonism, we sought to identify analogous effects of PIO treatment in individuals with CUD.

Here we define target engagement as the ability of treatment to exact change on key mechanisms of interest underlying a particular disorder. Beyond the prototype measure of receptor occupancy, biobehavioral domains linked to other aspects of treatment efficacy, such as cocaine craving and WM integrity proposed here, have been recognized as targetable mechanisms (26, 27). Consistent with this experimental therapeutics approach (26), the aims of this study were to (1) compare PIO with placebo (PLC) on change in measures of cocaine craving intensity and WM integrity; and (2) compare PIO with PLC on measures of feasibility, including retention, medication compliance, and tolerability. While this project was not designed or powered as a test of clinical efficacy, measures of cocaine use were collected for exploring potential clinical outcomes.

METHODS

Design

We conducted a 12-week, two-arm, double-blind, randomized controlled pilot trial of PIO versus PLC on measures of craving and WM integrity as indexed by fractional anisotropy (FA) value (ClinicalTrials.gov identifier: NCT02774343). In keeping with recommendations of Leon et al (28), sample size for this pilot trial was based primarily on pragmatics of recruitment and the necessities for examining feasibility. We pre-specified and planned to conduct a Bayesian final analysis to obtain probability estimates of effect sizes along with uncertainty. We did not plan to conduct any significance testing and as such, power analysis was not warranted.

Participants

Major inclusion criteria were age (18 – 60 years old) and meeting DSM-IV criteria for cocaine dependence (29). Patients with current dependence on any drug except cocaine, alcohol, nicotine, and cannabis were excluded. Physiological dependence on alcohol requiring medical detoxification was exclusionary. Patients with medical conditions contraindicating PIO pharmacotherapy or using medications that would adversely interact with PIO were excluded, as were medical contraindications to MRI scans (see Appendix S1).

Procedures

Eligible subjects underwent a 1-week baseline period to obtain pretreatment measures of target domains. Randomization was performed by the blinded study coordinator using a standard random number generator as programmed using SAS 9.3. Following randomization, subjects assigned to PIO underwent a two-week dose titration period until reaching the target dose of 45 mg/d by Week 3. Subjects assigned PLC received identical capsules on the same schedule. Thrice weekly attendance during treatment included cognitive-behavioral therapy 1 hour per week and prize-based contingency management for attendance.

Measurement of Target Domains

Craving

On a weekly basis, patients completed the Brief Substance Craving Scale (BSCS)(30) and the Obsessive Compulsive Drug Use Scale (OCDUS) in the treatment clinic (31). Every two weeks, visual analog scale ratings of craving (VAS craving) consisting of 100 mm line, anchored by 0 “not at all” and 100 “extremely,” were used to assess cocaine craving right now, craving on average in the past week, and the worst craving in the past week. Data were analyzed as a total score across the three questions.

White matter integrity

DTI scans were acquired on a Philips Integra 3T magnet (parameters in the Appendix S2). DTI analysis, including segmentation and regions of interest (ROI) determination, followed atlas-based methods and quality assurance protocols for serial stability (32–34). These methods, critical in small-N samples, are ordered, anchored for motion and stability, and alternating to reduce cross terms and maximize signal-to-noise. Due to incidental finding of a brain abnormality (N=1), and scanner maintenance interruption (N=1), only 18 subjects were scanned at both pre- and post-treatment (10 = PLC, 8 = PIO). ROI analyses of FA focused on WM ROI based on several previous studies of WM integrity in cocaine dependence: (1) commissural fibers: the genu and splenium of corpus callosum (CC); (2) projection fibers: the anterior and posterior thalamic radiations; and (3) association fibers: the cingulum and the external capsule (18, 19, 22, 23, 35, 36).

Feasibility Measures

Treatment retention was determined by clinic visit attendance and total number of study weeks completed. Medication adherence was determined by urinary riboflavin levels (37) and self-reports of taking capsules. Medication safety assessments included adverse event (AE) reports and weekly rating scale for side-effects.

Cocaine use Measures

Urine samples collected at each visit were analyzed for benzoylecgonine at a concentration of 300ng/mL. Self-reported cocaine use throughout the study period was assessed using a modified Timeline Followback (TLFB) procedure (38).

Statistical analyses

Descriptive statistics were used to summarize baseline characteristics of the sample. All analyses were conducted on a modified intent-to-treat sample of randomized participants who received at least one capsule (per-protocol; 14 = PLC, 12 = PIO). Multilevel models were used to test the interaction of time and treatment on craving and cocaine use outcomes, while regression models for residual change scores were used for DTI outcomes. Simple effects and the posterior probabilities that these effects existed were performed as follow-up tests for interpreting interactions. Statistical analyses were performed using SAS 9.3, R 3.1.1 (R Core Team, 2016), and OpenBUGS 3.0 (2007).

Bayesian statistical reasoning was used to quantify the evidence for change in mechanisms of interest (target engagement). This approach is recommended over conventional Frequentist statistics in the context of hypothesis testing in decisions to carry forward the development of a candidate medication based on target engagement (39–42). Bayesian statistics provide a direct estimate of the probability of the alternative hypothesis (H1) existing, given the data. Here we defined H1 as a treatment group-by-time effect, favoring PIO over PLC. A posterior probability of ≥ 0.75 for the existence of an effect was considered sufficient evidence to warrant further investigation. This reasoning is consistent with previous probability thresholds stipulated for decision-making in existing medication trials; these thresholds were chosen a priori (43–47). Vague, neutral priors were used to maximize the influence on the posterior estimates (see Appendix S3).

RESULTS

Sample description

Forty-two cocaine users were screened to obtain a sample of 30 subjects who were randomized to treatment (see Appendix S4 for CONSORT). The two groups were similar in demographic and substance use characteristics, as shown in Table 1. Most participants (n = 26, 86%) met diagnostic criteria for alcohol use disorder (AUD) as well.

Table 1.

Sample demographic characteristics; N = 15 per group.

	Placebo		Pioglitazone
Variable	N	%	N	%
Female	4	26.7	4	26.7
Race
Black	7	46.7	12	80.0
White	4	26.7	3	20.0
Hispanic	3	20.0	0	0
Other	1	6.7	0	0
Employed	6	40.0	7	46.7
Alcohol use disorder	12	80.0	14	93.3
	Mean	SD	Mean	SD
Age	47.4	7.8	48.3	7.1
Education	12.8	1.1	12.6	1.9
Cocaine use (past 30 days)	13.9	9.0	18.2	11.8
Lifetime cocaine (years)	17.7	8.0	21.0	7.7
Alcohol use (past 30 days)	13.9	9.9	21.7	8.6
Lifetime alcohol (years)	22.3	12.2	23.9	13.1

Mechanisms of interest

Table 2 provides outcomes of the Bayesian analyses for the primary target variables and cocaine use outcomes (see Appendix S5 and S6 for corresponding tabular and graphic presentation of outcomes). For comparison purposes, Frequentist solutions to these models are provided in Appendix S7.

Table 2.

Summary of results from Bayesian statistical analyses of craving measures, cocaine use outcomes, and DTI outcomes.

Outcomes for Clinical variables	Variable
	BSCS	VAS	OCDUS	Cocaine use (TLFB)	Cocaine Use (UDS)
Post prob: Treatment × Time	0.97	>0.99	0.60	0.99	0.75
Post prob: Simple Effect of Time (PIO)	p(b<0) > 0.99	p(b<0) > 0.99	p(b<0) = 0. 64	p(O.R. < 1) > 0.99	p(O.R. < 1) = 0.38
Estimate: Simple effect of Time (PIO)	b = −0.24 (−0.36–−0.12)	b = −3.84 (−5.22–−2.46)	b = −0.07 (−0.45–0.30)	O.R. = 0.74 (0.66–0.85)	O.R. = 1.05 (0.75–1.47)
Post prob: Simple effect of Time (PLC)	p(b<0) = 0.98	p(b<0) > 0. 99	p(b>0) = 0. 77	p(O.R. < 1) = 0.89	p(O.R. < 1) = 0.10
Estimate: Simple effect of Time (PLC)	b = −0.09 (−0.18–−0.01)	b = −1.34 (−2.19–−0.51)	b = 0.15 (−0.25–0.55)	O.R. = 0.94 (0.84–1.04)	O.R. = 1.24 (0.88−1.78)
Outcomes for DTI variables	gCC FA	sCC FA	ATR FA (hemisphere)	PTR FA (hemisphere)	EXTC FA (hemisphere)	CING FA (hemisphere)
Post Prob: Treatment Controlling for Baseline FA	0.99	0.99	0.93 (L) 0.93 (R)	0.98 (L) 0.97 (R)	0.93 (L) 0.52 (R)	0.68 (L) 0.54 (R)
Estimate of Simple of PIO	b = 25.96 (4.84–45.05)	b = 29.12 (6.52–51.92)	b = 12.59 (L) (−4.84–30.16) b = 13.83 (R) (−5.26–32.99)	b = 15.29 (L) (1.19–29.42) b = 15.96 (R) (−1.06–33.10)	b = 8.86 (L) (−3.09–20.87) b = 0.27 (R) (−9.53–10.06)	b = 4.29 (L) (−14.61–23.42) b = 1.26 (R) (−27.60–30.18)

Note. Post Prob = posterior probability; PIO = pioglitazone; PLC = placebo; p = probability; b = beta coefficient (with 95% credible intervals); O.R. = odds ratio. BSCS = Brief Substance Craving Scale; VAS = visual analog scale; OCDUS = Obsessive Compulsive Drug Use Scale; TLFB = Timeline Followback; UDS = urine drug screen; FA = fractional anisotropy value; gCC = genu of corpus callosum; sCC = splenium of corpus callosum; ATR = anterior thalamic radation; PTR = posterior thalamic radiation; EXTC = External Capsule; CING = Cingulate.

^*Coefficients are in the log-form.

Craving

For the BSCS, there was a 0.97 posterior probability of a treatment by time interaction existing with simple effects indicating a high likelihood that for every additional week in treatment, craving was reduced by a factor of 0.24 for participants receiving PIO compared to 0.09 for participants receiving PLC. For the VAS, there was a >0.99 posterior probability of an interaction, with regression coefficients indicating a decrease in craving by a factor of 3.84 for PIO versus 1.34 for PLC for every additional biweekly period in treatment.

WM integrity

Table 2 provides outcomes of Bayesian analyses on WM integrity as measured by fractional anisotropy (FA) values. Across the specified ROIs, four regions revealed a Bayesian posterior probability > 0.75 bilaterally of PIO after controlling for baseline FA values: the genu of the CC (gCC = 0.99), the splenium of the CC (sCC = 0.99), the anterior thalamic radiation (left ATR = 0.95, right ATR = 0.96), and the posterior thalamic radiation (left PTR = 0.99, right PTR = 0.98). Controlling for baseline FA, gCC, sCC, ATR, and PTR FA values increased from week 0 to week 12 in PIO (positive b parameter estimates) relative to PLC. Probabilities across hemispheres for the cingulum and external capsule (EXTC) did not meet the posterior probability threshold, with the exception of the left hemisphere of the EXTC.

Cocaine use

On the TLFB, there was a 0.99 posterior probability of an interaction, with simple effects indicating a 0.99 posterior probability of a reduction in the odds of reporting cocaine use for every additional day in treatment for PIO (O.R. 0.74, 95% CrI = 0.66–0.85) relative to an 89% posterior probability for PLC (O.R. 0.94, 95% CrI = 0.84–0.1.04). For cocaine positive urines, there was a 0.75 posterior probability of an interaction existing, with a greater chance of having decreased cocaine-positive urines in PIO (Posterior Probability = 0.36, O.R = 1.05, 95% CrI = 0.45–1.47) versus PLC (Posterior Probability = 0.10, O.R = 1.24, 95% CrI = 0.88–1.78).

Feasibility

Regarding retention, 30 subjects enrolled and four (PLA, n = 1; PIO, n = 3) dropped out of the trial before starting medication. Of the 26 subjects who started treatment, 21 (PLC, n = 11; PIO, n = 10) completed all 12 weeks (80%). Regarding medication adherence, mean compliance levels based on riboflavin were 95.4% (PLC) and 96.2% (PIO). Self-reported medication compliance was 86.9% (PLC) and 84.1% (PIO). The most frequently reported side effects were sleep disruption, diarrhea, stomach pain, cough, and increased urination; all rated mild (see Appendix S8).

DISCUSSION

The results of this study provide preliminary evidence of therapeutically targetable mechanisms of PIO. For patients with CUD, there was a > 75% posterior probability that PIO conferred benefit over PLC in reducing cocaine craving over a 12 week treatment period. This effect was found on two commonly used brief self-report measures (BSCS, VAS). On the second target domain of WM integrity, we found evidence suggesting that PIO improved FA value versus PLC by the end of treatment. Findings also provide evidence that PIO can be applied in a CUD population with adequate acceptability and tolerability.

Target domains (craving and WM integrity) in this proof-of-principle study were selected based on prior work suggesting these domains might be modifiable by PIO. Specifically, craving is a well-known trigger of relapse (48, 49), and stress induces heightened reactivity and craving to drug cues thereby increasing relapse risk (50–52). Notably, PIO was effective in reducing drug intake in a preclinical model of stress-induced relapse (53). Thus, one avenue to reduced craving in the PIO group may be via decreases in stress-induced craving. We posit that this will be an important mechanism to examine in future work.

More than a decade of preclinical and human research shows an association between chronic cocaine exposure and alterations in WM integrity. These alterations may have functional consequences associated with impulsivity (19, 54), decision making (22), and time to relapse (23). Notably, WM alterations are emerging as key factors in impaired cognitive functions and are prevalent in several psychiatric and neurological disorders, including addiction (24, 55). Because of the documented anti-inflammatory and potential neuroprotective effects of PPAR-γ agonists, multiple clinical trials are investigating PIO as a neuroprotective agent (56–61). Accordingly, we hypothesize that the change in FA values observed in the PIO group may correspond to improved cognitive and daily life functions. Comprehensive measurement of these variables and their association with changes in WM integrity provide key opportunities for further scientific progress.

This initial project was limited by several factors, including: (1) small sample size; (2) restricted measurement battery with regard to potential key variables (e.g., stress, inflammation, executive cognitive function, overall life/health function); (3) a sample of co-morbid CUD and AUD patients, obscuring the relative role of each substance in the outcomes (see Appendix S9 Alcohol Consumption); (4) examination of single dose (45 mg); (5) for the DTI data, the Bayesian parameter estimates of the variance in the simple effects may be underestimated (62, 63); and (6) the degree to which pioglitazone’s therapeutic potential generalizes to other substances of abuse remains unknown.

Within the restriction of these limitations, the combined potential to attenuate craving and preserve WM integrity in individuals with CUD suggests PIO may have a role as a pharmacotherapy for CUD. One function may be as an adjunctive treatment to reduce risk for relapse during the critical early phase of recovery when craving is prominent, and in the context of preserved cognitive/life functioning, which can impede efforts to maintain drug abstinence.