Peroxisome proliferator activated receptor-γ agonist pioglitazone improves vascular and metabolic dysfunction in systemic lupus erythematosus

Objectives Premature cardiovascular events in systemic lupus erythematosus (SLE) contribute to morbidity and mortality, with no effective preventive strategies described to date. Immune dysregulation and metabolic disturbances appear to play prominent roles in the induction of vascular disease in SLE. The peroxisome proliferator activated receptor-gamma agonist pioglitazone (PGZ suppresses vascular damage and immune dysregulation in murine lupus and improves endothelial dysfunction in other inflammatory diseases. We hypothesised that PGZ could improve vascular dysfunction and cardiometabolic parameters in SLE. Methods Eighty SLE subjects with mild to severe disease activity were randomised to a sequence of PGZ followed by placebo for 3 months, or vice versa, in a double-blind, cross-over design with a 2-month wash-out period. Primary endpoints were parameters of endothelial function and arterial inflammation, measured by multimodal assessments. Additional outcome measures of disease activity, neutrophil dysregulation, metabolic disturbances and gene expression studies were performed. Results Seventy-two subjects completed the study. PGZ was associated with a significant reduction in Cardio-Ankle Vascular Index (a measure of arterial stiffness) compared with placebo. Various metabolic parameters improved with PGZ, including insulin resistance and lipoprotein profiles. Circulating neutrophil extracellular trap levels also significantly decreased with PGZ compared with placebo. Most adverse events experienced while on PGZ were mild and resolved with reduction in PGZ dose. Conclusion PGZ was well tolerated and induced significant improvement in vascular stiffness and cardiometabolic parameters in SLE. The results suggest that PGZ should be further explored as a modulator of cardiovascular disease risk in SLE. Trial registration number NCT02338999.


Supplemental methods:
Measurements of cardiovascular risk factors and vascular function.
Overnight fasting lipid profiles were obtained at the NIH Clinical Center Central Laboratories.
Lipoprotein particle concentration and diameters were measured using automated Nuclear Magnetic Resonance Spectroscopy (NMR) using the LP4.20 algorithm. The HDL cholesterol efflux capacity was measured using published methods (2). Briefly, 3x10 5 J774 cells/well were seeded in 24-well plate and radiolabeled with 2 μCi of 3H-cholesterol/mL in RPMI-1640 media containing 1% FBS for 24-hours. Cells were incubated for 16-hours in RPMI/2% BSA in the presence or absence of 0.3 mmol/L 8-(4-chlorophenylthio)-cAMP to upregulate ATP-binding cassette transporter A1. This was followed by addition of 2.8% apoB-depleted plasma to the efflux medium for 4 hours. A liquid scintillation counter was used to quantify the efflux of radioactive cholesterol from cells using the formula: (μCi of 3H-cholesterol in media containing 2.8% apoBdepleted subject plasma-μCi of 3H-cholesterol in plasma-free media /μCi of 3H-cholesterol in media containing 2.8% apoB-depleted pooled control plasma-μCi of 3H-cholesterol in pooled control plasma-free media). Pooled plasma was obtained from five healthy adult volunteers. All assays were performed in duplicate. LCAT concentration was quantified by ELISA (BioVendor; Ashville, NC).

Non -invasive vascular function studies:
These studies included the cardio-ankle vascular index (CAVI), peripheral arterial tonometry (PAT; reactive hyperemia index (RHI), and pulse wave velocity (PWV) (3). Subjects were asked to fast for at least 6 hours prior to these tests and to refrain from smoking or drinking caffeinated beverages for 24 hours prior to the studies. Subjects were asked to hold vasodilators, BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) anti-hypertensives and statins on the morning of the test. During testing, subjects were placed in a temperature-controlled quiet room in the supine position. 1) CAVI. CAVI was measured using VaSera-1500A (Fukuda Denshi Co. Redmond, WA). After placing blood pressure (BP) cuffs around both arms and ankles and attaching electrocardiogram electrodes to the upper arms, a microphone was placed on the sternal angle to record heart sounds. Measurements were automatically calculated using the VaSera VS-1000 software. The principle underlying CAVI has been discussed previously. (4) 2) PAT. Microvascular endothelial function was evaluated using PAT with an EndoPAT 2000 device (Itamar Medical Ltd. Caesarea, Israel) as previously described. (5) Finger probes were placed on symmetric fingers bilaterally, and a BP cuff was placed on one arm, with the other arm serving as control. PAT was continuously measured for 20 minutes. In between, for 5 minutes, BP cuff was inflated to supra systolic pressure in the test arm. At the end of the occlusion and dilatation periods, reactive hyperemia was captured as an increase in the PAT signal amplitude and compared with the control arm. A postocclusion to preocclusion ratio was calculated by EndoPAT software, providing RHI. Augmentation index (AI) was calculated from PAT pulses at the baseline period. The result was further normalized to heart rate of 75 bpm (AI@75), as previously described. (5) 3)SphygmoCor PWV and velocity system: Central aortic BP and stiffness were quantified using SphygmoCor CP system (AtCor Medical Pty Ltd.; New South Wales, Australia). The central aortic pressure PWV was determined by using the pressure tonometer and an EKG signal was used simultaneously to visualize ventricular-vascular interactions. Standard algorithm and procedures, as described elsewhere, were used to quantify results. (6) BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

F-FDG-PET/CT:
A subset of 30 subjects underwent this test, which was performed following an overnight fast. Images were obtained approximately 60 minutes after administration of 10 mCi of 18 F-FDG.
All scans were completed using a 64-slice scanner (Siemens Biograph) acquiring 1.5 mm axial slices of the aorta. Standard bed positions of 3 minutes each were applied, and whole body scans were obtained for each patient from the vertex of the skull to the toes.. The extent of 18 F-FDG uptake within the aortic wall was measured with dedicated software (OsiriX MD, Pixmeo SARL).
Each arterial region of interest produced 2 measures of metabolic activity: a mean standardized uptake value (SUVmean) and a maximal SUV (SUVmax), which were obtained in the aorta from the aortic outflow tract to the abdominal aorta. Regions of interest were also placed on 10 contiguous slices over the superior vena cava to obtain a single average background blood activity. The SUVmean from each of the superior vena cava slices were then averaged to produce 1 venous value.
To account for background blood activity, SUVmax values from each aortic slice were divided by the average venous SUVmean value to yield taeget/background ratio(TBR), a measure of vascular inflammation as previously described. (9) RNA Isolation and IFN Gene Signature (ISG) quantification by Nanostring: Total RNA was extracted from whole blood using Paxgene blood RNA isolation kit ( PreAnalytiX, Switzerland). RNA concentration was measured by NanoDrop (Thermo Fisher Scientific). The nCounter Element prep kit (NanoString Technologies) was used for Nanostring assay. A NanoString TagSet consisting of fluorescently labeled specific Reporter Tags and a biotinylated universal Capture Tag were supplied by NanoString. There were 6 spike-in positive controls used to determine the hybridization efficiency, and 6 negative controls used to check nonspecific background. A target-specific oligonucleotide probe pairs (synthesized by IDT, Coralville, IA) contained 37 ISGs, previously identified as discriminative of the IFN signature, BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)   Table 2 A,B), after in vitro stimulation with PMA (10 ng/mL) and ionomycin (500 ng/mL) in combination with protein transport inhibitors (Monensin 2uM, Brefeldin A 1 ug/mL) for 4 hours in a 37°C tissue culture incubator.

Data analysis:
Longitudinal changes in either gene expression or flow cytometry populations were evaluated by paired test, with P values corrected for multiple comparisons using the Benjamini-Hochberg method when described, using R-Shiny web tools developed in-house but similar to those previously described (10). For transcriptomic responses differentially expressed genes were used for gene set enrichment analyses performed using tmod and BTM that reflect biologically responding pathways (11,12).

Statistical Analysis:
Sample size was calculated based on previous publications for arterial stiffness and vascular dysfunction in SLE and RA (5,(13)(14)(15)(16). Based on our previous experience with the CV lupus cohort and on published literature on cross-over designs using PGZ to measure CV markers in other populations, dropout rate was estimated between 3-20%. Bring a proof of concept study, at least moderate differences in outcome were considered to be clinically meaningful. The actual analysis used data on all subjects including those who provided partial information. This sample was considered to provide power to detect moderate standardized differences for the other variables measured in this study.
Most of the outcome variables were measured at Day 1 (prior to randomization), month 3 (end of period 1), month 5 (start of period 2) and month 8 (end of period 2). Change score in these continuous outcomes was summarized by mean and standard deviation (SD): M3 -D1 for period 1 and M8 -M5 for period 2, for each sequence respectively. Linear mixed models were used to analyze the change scores in mean CAVI, PWI, and the log-transformed RHI. The models included the fixed effects of baseline value (D1 for period 1, M3 for period 2), treatment group (PGZ or placebo), period (1 or 2), and sequence (AB or BA). The random effect was "subject". Residual plots were used to assess model normality assumptions. For the treatment group difference (assessed by the difference in change scores between PGZ and placebo), the estimate, its associated confidence interval (CI), and p-values were reported from the mixed effects models. Other secondary efficacy endpoints were analyzed in a similar manner. The Wilcoxon rank sum test was used to analyze NETs when normality assumption was violated(43). For continuous variables only measured in period 1, analysis of covariance was used to analyze the data. Efficacy analyses were based on the intention to treat population, which includes all the subjects who were randomized. For the primary endpoints, 0.05/3 was set as the cutoff for statistical significance. For  all secondary efficacy endpoints, p-values and 95% confidence intervals were provided to examine the statistical evidence. All statistical analyses were performed using SAS (Version 9.4, SAS Institute, Cary, NC).

Supplemental Results
As the targeted analysis of inflammation-related genes showed no effect when subjects were treated with pioglitazone, we performed unbiased screening to detect potentially other effects of this drug on immune phenotype. For this we focused on analyzing a subgroup of 24 subjects, which comprised the 12 individuals from each study arm that demonstrated the greatest decreases in CAVI when treated with pioglitazone. Whole blood transcriptomic analysis was performed with changes between consecutive study timepoints used to rank genes for enrichment analyses of blood transcriptional modules (BTM). The number of significantly enriched pathways (FDR<0.05) was no higher in the study arms receiving pioglitazone than placebo for both of 0-3-and 5-8-month treatment periods (Supplemental Table 6A). Further, the pathways that did change with pioglitazone treatment showed no overlap between the two study arms (Supplemental Table 6 B, C). Next, PBMC were analyzed by high-dimensional flow cytometry, using a 27-color panel to quantify 58 populations spanning broad immune lineages in unstimulated cells, and a 34-color panel to quantify 83 populations that more comprehensively characterized T cell populations revealed by transcription factor or cytokine expression after in vitro stimulation with PMA. For both panels no significant changes in any population frequencies were observed between study timepoints after correction for multiple testing. The numbers of populations changing with nominal significance are reported (Supplemental Table 7 A, D), with those populations corresponding to the periods of drug treatment in each study arm also shown (Supplemental Table 7    1A) Antibody clones and fluors are detailed for the 27-color panel.

Specificity
Fluorochrom e Clone vendor cat# Th subsets

GATA3+IL-5+
Th cells   Anti-dsDNA data analysis results: 59 subjects had their anti-dsDNA status unchanged (stayed negative or positive) in both periods, 7 showed improvement (DNA status changed from positive to negative) under PGZ and 3 showed improvement under placebo. All other subjects had mixed results.
Data are mean±SD. Change is defined as the post baseline value minus the baseline value during the period: i.e., M3 -D1 for period 1, M8 -M5 for period 2. SLEDAI 2K = Systemic Lupus Erythematosus Disease Activity Index 2000; Anti-ds-DNA IU/mL= Anti double-stranded DNA antibody international unit/milliliter; C3 mg/dL= Complement protein C3 milligram/deciliter; C4 mg/dL= Complement protein C4 milligram/deciliter; SF36= Short Form 36; PGA= Physician global assessment *: Linear mixed effects models were used to calculate the estimated treatment effect (the treatment group difference in the change score between pioglitazone and the placebo), its 95% CI, and the p-value.