An Anti-Inflammatory Sterol Decreases Obesity-Related Inflammation-Induced Insulin Resistance and Metabolic Dysregulation

Obesity-related inflammation-induced insulin resistance and metabolic dysregulation were investigated in retrospective analysis of placebo hematologic and metabolic laboratory data from trials associated with increasing chronic low-grade inflammation and body mass index. Studies included healthy subjects and those with progressive stages of metabolic dysregulation, including type 2 diabetes mellitus with uncontrolled hemoglobin A1c. Intrasubject variances in erythroid and metabolic values increased with metabolic dysregulation. Random effects were demonstrated in treatment-naïve diabetes for erythroid, glucose, and HbA1c fluctuations. The anti-inflammatory insulin sensitizer, HE3286, was tested for its ability to decrease obesity-related inflammation-induced insulin resistance and metabolic dysregulation in diabetes. HE3286 significantly decreased erythroid and metabolic variances and improved 1,5-anhydroglucitol (a surrogate of postprandial glucose) compared to the placebo group. HE3286 HbA1c decrease correlated with weight loss and inversely with baseline monocyte chemoattractant protein-1 (MCP-1) in metformin-treated diabetics. Normalization of HbA1c to the 84-day average hemoglobin revealed that HE3286 HbA1c decrease correlated with high baseline MCP-1 and MCP-1 decrease in treatment-naïve diabetics. HE3286 decreased insulin resistance, increased the frequency of decreased day 84 HbA1c in metformin-treated subjects, and decreased day 112 HbA1c in treatment-naïve diabetics. HE3286 may be useful to restore metabolic homeostasis in type 2 diabetes.


Introduction and Purpose
HE3286 (17 -ethynylandrost-5-ene-3 ,7 ,17 -triol) is a chemical derivative of the natural mammalian sterol androst-5-ene-3 ,7 ,17 -triol ( AET). AET exhibits anti-in�ammatory activity in rodent models, is elevated in plasma of obese subjects with normal glucose disposal, and may play a compensatory role in preventing development of metabolic syndrome (reviewed in [1]). AET is pharmaceutically unsuitable, due to poor oral bioavailability and its propensity for oxidative inactivation by 17 -hydroxysteroid dehydrogenase [1]. HE3286 is stabilized against oxidation at position 17 and consequently orally bioavailable, does not bind to any known nuclear steroid hormone receptors, and is pharmacologically unrelated to androgens, estrogens, corticosteroids, or peroxisome proliferators [1]. HE3286 has shown broad anti-in�ammatory activity in animal models of rheumatoid arthritis, ulcerative colitis, multiple sclerosis, lung in�ammation, autoimmune type 1 diabetes, and neu-roin�ammation (reviewed in [1]). In these models, nuclear factor kappa B (NF B) activation and proin�ammatory cytokine production were consistently suppressed. Furthermore, HE3286 was not markedly immunosuppressive in rodent models of ovalbumin immunization, Klebsiella pneumoniae or Pseudomonas aeruginosa infection, Coxsackievirus B3 myocarditis, delayed-type hypersensitivity, and mitogeninduced proliferation, or in the human mixed lymphocyte reaction assay (reviewed in [1]).
Obesity induces an insulin-resistant state in adipose tissue [2], liver, and muscle and is a strong risk factor for the development of type 2 diabetes mellitus [3]. In adipose tissue, MCP-1 and tumor necrosis factor alpha (TNF ) play dominant proin�ammatory roles [2]. Adiposity-induced in�ammation-stimulated kinases phosphorylate insulin receptor substrate-1 on serine residues and inhibit insulin signaling [4]. Two recent publications report the activity of HE3286 against in vitro in�ammatory responses and in vivo rodent models of obesity-induced in�ammation and insulin resistance [5,6]. HE3286 suppressed endotoxininduced NF B activation, reporter gene expression, nuclear localization, and p65 phosphorylation in mouse macrophages and decreased phosphorylation of the proin�ammatory extracellular signal-regulated (Erk1/2), IkappaB (Ikk), Jun N-terminal (Jnk), and p38 mitogenactivated protein (p38 Mapk) kinases. HE3286 also attenuated TNF -stimulated in�ammation and TNFinduced adipocyte-stimulated macrophage chemotaxis [5,6]. HE3286 treatment of diabetic db/db mice, insulinresistant diet-induced obese mice, and genetically obese ob/ob mice suppressed progression to hyperglycemia and markedly improved glucose clearance. is effect appeared to be consequent to reduced insulin resistance, since HE3286 lowered blood insulin levels in both db/db and ob/ob mice. In these studies HE3286 suppressed levels of the chemokine monocyte chemoattractant protein-1 (MCP-1), along with its cognate receptor, C-C motif chemokine receptor-2, in white adipose tissue [6]. In Zucker diabetic fatty rats, HE3286 downregulated in�ammatory cytokine and chemokine expression in both liver and adipose tissues and suppressed macrophage migration into adipose tissue. HE3286 normalized fasting and fed glucose levels, improved glucose tolerance, and enhanced skeletal muscle and liver insulin sensitivity, as assessed by hyperinsulinemic, euglycemic clamp studies. In addition, HE3286 reduced liver cholesterol and triglyceride content, leading to a feedback elevation of low-density lipoprotein (LDL) receptor and decreased total serum cholesterol [5]. Recently, we have reported that HE3286 binds to Erk1/2, Lrp1, and Sirt2 [7] and proposed that the HE3286-mediated decrease in hyperactivation of Erk1/2 may be causal for its metabolic [8] and anti-in�ammatory activities.
In a clinical study in obese, impaired glucose tolerance (IGT) subjects, HE3286 signi�cantly increased the frequency of insulin-resistant subjects with improved day 29 insulinstimulated glucose disposal, increased HDL cholesterol, and decreased day 28 CRP compared to placebo-treated subjects [9]. Based on baseline glucose clamp studies, insulin-resistant subjects had elevated in�ammatory biomarkers, with lower adiponectin and higher cytokine secretion in LPS-stimulated PBMC. Aer 28 days of HE3286 treatment, adiponectin levels increased signi�cantly in insulin-resistant subjects, compared to placebo. ese results support our hypothesis that obesity-induced in�ammation is a signi�cant contributor to metabolic dysregulation and that the anti-in�ammatory activity of HE3286 can preferentially bene�t the insulinresistant in�amed subpopulation of obese IGT subjects.
Based on preclinical studies and these foregoing results in IGT subjects, it was conjected that HE3286 might bene�t obese in�amed insulin-resistant individuals with type 2 diabetes mellitus (T2DM). A widely accepted clinical endpoint for T2DM is the change in HbA1c, a surrogate marker for the extent of hyperglycemia an individual experiences over time. Traditionally, erythroid hematology values are considered stable in healthy individuals, and hemoglobin and HbA1c turnover is reported to re�ect the normal red cell half-life of 38-60 days [10]. In T2DM, the life span of red cells can be altered signi�cantly by in�ammation, particularly TNF -induced oxidative stress [11], obese lowgrade systemic in�ammatory response syndrome [12], the presence of elevated levels of advanced glycation endproducts on the surface of red cells [13,14], hypoxia [15], and excessive erythrocytosis [16]. ere are reports of large �uctuations in HbA1c in type 1 diabetes [17], especially in subjects with poor glycemic control [18,19]. is information prompted us to also assess the association of obesity-related chronic low-grade in�ammation with hemoglobin concentration and HbA1c variability in uncontrolled T2DM. We retrospectively analyzed the hematologic and metabolic clinical laboratory data for placebo groups from 10 clinical studies that were conducted between 2001 and 2010. ese studies included both healthy subjects and individuals in progressive stages of metabolic disease that presented with increased chronic low-grade in�ammation coincident with elevated BMI that included dyslipidemic, IGT, and T2DM participants with uncontrolled HbA1c.
With an understanding of the variability associated with progressive adiposity, in�ammation, and metabolic disease, we assessed the activity of HE3286 to decrease obesityinduced in�ammation and insulin resistance in T2DM.  [20]. ese four double-blind, randomized, placebo-controlled, healthy human safety studies were conducted in e Netherlands (Kendle International, Utrecht) and the United States (Parexel International, Baltimore, MD). Two single-dose, dose-escalation studies assured safety and evaluated the pharmacokinetics of androst-5-ene-3 ,17diol (HE2100) (studies 200 and 202). A multidose, dose escalation study was performed to assess safety and pharmacokinetics and potential early activity of HE2100 (study 201). Early activity, de�ned by effects on peripheral blood elements, was con�rmed by a follow-up study that included elderly subjects and an initial study of bone marrow hematology (study 203).

Subjects and Methods
Details of studies 2200-100, -101, -120, and -130 have also been published [21]. Healthy adult and elderly subjects were randomized to receive three consecutive daily subcutaneous injections of placebo, 50, or 100 mg androst-5-ene-3 ,7 ,17 -triol (HE2200), followed by 2 months of periodic observation (trial 2200-100), or to receive placebo, 25, or 100 mg HE2200 transmucosally (buccal administration) once daily for �ve days followed by 2 months of periodic observation (trial 2200-101). Study 2200-120 was a phase II study in healthy hepatitis B-naïve, and elderly (65-85 years old) volunteers, who received hepatitis B vaccine, were randomized to concomitantly receive either 100 mg of HE2200 or placebo equivalent. Subjects received three subcutaneous injections of study drug or placebo prior to the �rst and second doses of hepatitis B vaccine given 28 days apart. e third dose of vaccine was given at 6 months without HE2200 or placebo treatment, and the study terminated 28 days later.
Study 2200-130 was a phase II study in dyslipidemic subjects, ages 18-70 years, with plasma triglyceride concentrations 1.7-2 mmol/L, total cholesterol levels of 5.7-8.3 mmol/L, and HDL levels of ≤1.2 mmol/L for males and ≤1.4 mmol/L for females. Aer informed consent was obtained, subjects initiated a Step II AHA diet and discontinued all lipid lowering agents for a six-week run-in period. Each subject�s lipid pro�le at week four of the diet was used to determine eligibility for the study. At six weeks, quali�ed subjects were randomized to receive 25 or 100 mg of HE2200 or placebo equivalent by buccal administration for 28 days.
HE3286-0102 was a multicenter, double-blind, doseranging phase I study designed with 5 cohorts of obese, impaired glucose tolerance (IGT), but otherwise healthy participants [9]. Subjects were screened for fasting blood glucose level of <7.0 mmol/L and a 7.8-11.1 mmol/L 2-hour postprandial glucose following a 75-gram oral glucose load. Oral placebo or HE3286 doses of 4 (2 BID), 5 (QD), 10 (5 BID), and 20 (10 BID) mg were administered daily for 28 days. One-step hyperinsulinemic, euglycemic clamps were performed on the day before the �rst dose and day 29 in the BID dose groups.

Placebo Data
Analysis. Placebo data from 10 clinical studies was analyzed to obtain coefficients of variation (CV) for red blood cells (RBC), hematocrit (Hct), mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), hemoglobin (Hb), hemoglobin A1c (HbA1c), platelets, lymphocytes, monocytes, white blood cells, fasting glucose, insulin, triglycerides, cholesterol, high-density lipoprotein cholesterol (HDL), and low-density lipoprotein cholesterol (LDL). CVs from each medical condition were compared to those of healthy volunteers for signi�cant differences in magnitude using Welch ANOVA, allowing for unequal differences, and for signi�cant differences in dispersion, using the 2-sided test. Only placebo values were used to avoid any treatment effects from drug administrations. Pearson correlations between individual subject CVs for each parameter and CVs for HbA1c and for insulin were determined to assess commonality in increased variances between parameters. In addition, intravisit changes in HbA1c were determined for each subject with HbA1c data.

HE3286-0401 Data Analysis.
Data for all placebo-and HE3286-treated T2DM subjects were analyzed for the same parameters by cohort (metformin treated and treatment naïve) and in subgroups (strati�ed by baseline monocyte chemoattractant protein-1 (MCP-1) in metformin-treated subjects, and by body mass index (BMI) in the MCP-1 selected treatment-naïve subjects). Baseline characteristics were assessed for balance between groups.
Based on the �ndings of excessive variability in T2DM, including random effects in intravisit Hb, HbA1c values were normalized (nHbA1c) by the 84 day average Hb value for each subject. is was done based on the fact that random effects have a mean of zero, but an excessive dispersion. us the average change in placebo subjects over 84 days would be expected to be zero, and true changes in HbA1c due to treatment could be compared.
Correlations were determined for independent variables of baseline MCP-1, baseline tumor necrosis factor alpha (TNF ), BMI, Hb CV, and day 84 changes in homeostatic model assessment of beta cell function (HOMA2 %B), glucose, weight, Hb, TNF , and MCP-1 with dependent variables Hb CV, HbA1c CV, and changes in HbA1c and 84 day average normalized HbA1c (ΔnHbA1c), in order to understand clinical parameters affecting variability and treatment effects. e magnitude of HE3286 treatment effects compared to placebo was tested in the subgroups of MCP-1 > 40 pg/mL in metformin-treated subjects, and of BMI > 31 in treatment-naïve subjects.
Heteroscedasticity (differences in variances between subgroups) was tested for changes in insulin, C-peptide, fasting glucose, HOMA2 %B and HOMA2 insulin resistance (HOMA2 IR), leptin, HbA1c, insulin, MCP-1, and triglycerides. Subgroup distributions were tested for normality (Shapiro-Wilks test) for HE3286 and placebo treatment. Differences in dispersions between HE3286 and placebo treatment were analyzed using the 2-sided test. administered orally for 12 weeks to adult T2DM patients ( Figure 1(a)). is was an adaptive design to investigate the characteristics of T2DM subjects that respond to HE3286. In cohort 1 of the study, 95 eligible patients, who consented to participate, were randomized 1 : 1 to receive study treatment (HE3286 10 mg/day or placebo) in addition to a stable dose of metformin. Inclusion criteria for cohort 1 included HbA1c ≥ 7.5% and fasting glucose ≤ 12.5 mmol/L. In cohort 2, 69 subjects who consented to participate and who met a revised eligibility criteria as determined by cohort 1 were randomized 1 : 1 to receive study treatment (HE3286 10 mg/day or placebo) as monotherapy. Aer the analysis of data from the �rst stage of the study, the population for cohort 2 was phenotypically enriched by screening for the following:

Analysis of Variance in Erythroid and Metabolic Parameters in Placebo Comparison
Studies. Variance in erythroid parameters was examined by three ways. First, the variances for selected hematologic and metabolic laboratory values, such as the mean coefficient of variation (CV) and the CV range for each individual subject, were determined and compared with those of healthy subjects. Second, the intravisit changes in HbA1c were compared for individual subjects for each condition with those of healthy subjects. ird, intravisit and day 84 changes in HbA1c and other hematology and laboratory parameters were tested for random effects.

Statistical
Analyses. Random effects were tested using Residual Maximum Likelihood (REML) using StatXact, and outliers were examined using Mahalanobis distance (Cytel Soware Corporation, Cambridge, MA) in conjunction with SAS soware (SAS Institute, Cary, NC). Correlations were tested using Spearman or Pearson correlations, and the hypothesis that placebo participants with clinical conditions have higher frequencies of abnormal hematology and laboratory values than healthy subjects was tested using onetailed Fisher's exact test. Heteroscedasticity (tests of different variabilities between subpopulations) was tested for normal distributions (Shapiro-Wilks test), and dispersion was tested using the 2-sided test (Prism Graph Pad, San Diego, CA). If there were signi�cant differences in variances between groups, they were further examined using a -test assuming unequal variances, nonparametric Mann-Whitney test, or Fisher's exact test. Due to the exploratory nature of this hypothesis-testing study, values were not adjusted for multiple comparisons. . rough the course of this analysis it was discovered that the in�ammatory status of the selected patient population created large and rapid changes in the patient's red cell mass that affected the whole body hemoglobin mass and consequently the �delity of the HbA1c metric. In order to investigate HE3286 treatment effects on HbA1c in T2DM patients, HbA1c changes were normalized to the day 84 average Hb for each subject, by averaging Hb values acquired at each clinic visit. is is statistically justi�ed based on the fact that random effects have a mean of zero but are characterized with high variances. Normalized HbA1c (nHbA1c) was applied to correct for the in�ammationinduced variances found in this T2DM study population with uncontrolled in�ammation.

Retrospective Exploration of Increased Variance with
Metabolic Disease Progression. e hypothesis that chronic low-grade in�ammation leads to increased variance in laboratory values was explored by a retrospective review of hematology and metabolic clinical parameters from placebo subjects enrolled in 10 clinical studies conducted by Harbor erapeutics, Inc., since 2001. Only placebo subject data from these studies were used for intercomparisons to exclude study drug effects.

Variances of Hematology and Laboratory Values between
Medical Conditions in Placebo Subjects. Changes in variance (CV means and ranges) for hematologic and metabolic parameters sorted by medical condition are displayed in Figure 2. Dyslipidemic patients showed increased variances in hematocrit, HbA1c, and fasting glucose compared to healthy subjects. Although their lipid parameters were abnormal, their lipid variances were not signi�cantly higher than those of healthy subjects. IGT subjects had signi�cantly higher variances for RBC, hematocrit, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, hemoglobin, and HbA1c. Although they had higher postprandial glucose, their fasting glucose variances were not sig-ni�cantly greater than those of healthy subjects.

Conclusions from the Retrospective Analysis of Placebo Subjects with Medical
Conditions. ere appears to be a progressive increase in metabolic and hematologic laboratory parameter variances with increased BMI and metabolic disease progression that results in random HbA1c changes. Based on this, intervention with the HE3286 anti-in�ammatory agent might confer bene�t to this pathology. However, with random HbA1c effects in the placebo participants, it is difficult to demonstrate signi�cant changes with comparisons to active agents unless a correction is applied to the data. Accordingly, treatment effects were investigated by normalizing HbA1c to the day 84 average total body hemoglobin mass for each patient.
3.6. HE3286-0401. In this study, HE3286 was well tolerated. Seventy-six percent of cohort 1 (metformin treated) and 77% of cohort 2 (treatment naïve) completed the study (Figure 1(b)). Only 1 serious adverse event occurred, a transient asymptomatic elevation from baseline of blood amylase, which resolved on study. is event was considered by the investigator to be possibly related to study medication. ere were no clinically signi�cant abnormalities related to any body system, including hypoglycemia and electrocardiograms, attributable to HE3286 administration. ere were no detectable differences or trends in adverse events between placebo-and HE3286-treated subjects. No patient died while on study. Baseline demographics and characteristics of each group are presented in Table 2.

Correlates of HbA1c and Hemoglobin Changes and
Variances in HE3286 and Placebo Subjects 3.7.1. Cohort 1. Correlations of ΔHbA1c, HbA1c CV, and Hb CV with other parameters are shown in Table 3. e HE3286 cohort 1 group HbA1c change was negatively correlated with baseline MCP-1 ( ) and change in HOMA2 %B ( ) and positively correlated with glucose change ( 9), weight change ( ), and change in Hb ( 3). HbA1c change was positively correlated with fasting glucose change in both HE3286 and placebo. e placebo HbA1c change was negatively correlated with baseline TNF ( ) and positively with change in TNF ( 2). In addition, in the placebo group, intrapatient HbA1c coe�cients of variation (CV) were signi�cantly correlated with baseline TNF ( 2). ese relationships led to the hypothesis that HE3286 decreased HbA1c in the more in�amed (higher MCP-1) patients, in conjunction with increased pancreatic beta cell function and weight loss, but in placebos, in�ammation (TNF ) was primarily contributing to HbA1c changes. e �nding that Hb change was positively correlated with HbA1c change suggested the possibility that in�ammation-induced effects produce random Hb levels that contribute to increased variance in HbA1c changes. High intrapatient CVs were observed for both HbA1c (up to 16%) and Hb (up to 12%) in both HE3286 and placebo groups (data not shown). Residual maximum likelihood (REML) analyses indicated a signi�cant random CV component in the intrapatient Hb values, for both the HE3286 and placebo groups ( < for each, data not shown).

Cohort 2.
Cohort 2 participants were selected using more stringent criteria for MCP-1, BMI, insulin, and Cpeptide. TNF was not measured in this group. Consequently, signi�cant correlations observed in cohort 1 for HbA1c change with baseline MCP-1 and changes in weight and in TNF were not observed in the overall cohort 2 group. Table 3 also shows correlations between ΔHbA1c, HbA1c CV, and Hb CV and other parameters in cohort 2. In this overweight to obese population, ΔHbA1c was correlated negatively with baseline BMI ( ) and with HOMA2 %B change ( ) and positively with day 84 change in fasting plasma glucose ( ) for HE3286, but not placebo. Further, we found that the cohort 2 participants (selected for higher in�ammation) had higher variances in erythroid hematology values than cohort 1 (see Figure 2), and cohort 2 placebo day 84 changes in HbA1c had a signi�cant random component (REML , data not shown). In placebo patients, HbA1c CV was positively correlated with baseline Hb CV ( ) and the ΔMCP-1 in�ammation marker ( ) and negatively correlated with weight change ( ). Hb CV, in turn, was positively correlated with baseline MCP-1 ( ). ese relationships led us to the hypothesis that HE3286 decreased HbA1c in patients with higher obesity (BMI), in conjunction with improved pancreatic beta cell function and decreased fasting glucose, that, for placebo patients, HbA1c change and Hb CV were related to in�ammation status (MCP-1), and that weight loss in placebos might be related to in�ammation effects on malnutrition.
Together, the results from studies in HE3286-0401 T2DM patients suggested that low-grade chronic in�ammation develops during metabolic disease progression in the obese diabetic and contributes to dysregulation of metabolic and hematologic homeostasis. If this is correct, then intervention with an anti-in�ammatory compound such as HE3286 might lead to restoration of homeostasis, normalization of glucose levels, and a decline in weight. Further, the in�ammatory effects on erythropoiesis may be quelled and the utility of ΔHbA1c as a biomarker of glucose control restored.

HE3286-0401 Treatment Effects
3.8.1. Cohort 1. According to our observations in phase I in�amed obese prediabetics, HE3286 should show bene�t in in�amed T2DM individuals. e correlation of baseline MCP-1 in the HE3286 HbA1c response was explored, and signi�cant treatment effects were observed in the more in�amed subjects (baseline serum MCP-1 upper 2 tertiles (>40 pmol/L)). Table 4 displays the HE3286 day 84 treatment effects on clinical parameters in this subgroup. Signi�cant decreases were observed for HOMA2 IR ( ), Cpeptide ( ), Hb ( ), Hct ( ), and RBC ( ) changes in the HE3286 treatment group when compared to the placebo (metformin alone) group.
e effect of HE3286 on ΔnHbA1c in the overall population was not signi�cant. erefore, the day 84 treatment effect on ΔnHbA1c was investigated in the more in�amed MCP-1 subgroup (Table 4). e median magnitude of the ΔnHbA1c was −0.44% Hb (HE3286, −0.34; placebo, +0.1). e HE3286 data was normally distributed, but placebo was signi�cantly abnormal ( , test, data not shown). is situation necessitated the use of nonparametric methods of data analysis. e HE3286 treatment effect was found to signi�cantly decrease nHbA1c from zero ( ). e frequency of HE3286 patients with decreased nHbA1c was signi�cantly greater than placebo (17/22 versus 9/25, ). ere were no signi�cant differences between HE3286 and placebo groups at follow-up day 112.
e HE3286 and placebo patients distributions with MCP-1 > 40 pmol/L are shown for ΔnHbA1c in Figure 4(a) and for ΔHOMA2 IR in Figure 4(b). e majority of HE3286 patients showed decreased nHbA1c and HOMA2 IR, whereas the majority of placebos showed increases. ese results are consistent with inhibition of NF B hyperactivation and consequent restoration of normal insulin signaling, consistent with the preclinical HE3286 observations.

Cohort 2.
e correlation between baseline BMI and change in HbA1c in the HE3286 group was explored by stratifying participants on the median BMI (31 kg/m ). e ΔnHbA1c in the HE3286, but not placebo participants (with BMI > 31 kg/m ), correlated signi�cantly with their baseline MCP-1 ( ) (Table 3). is strengthens the hypothesis that HE3286 bene�ted the obese in�amed subset of T2�M patients. e ΔnHbA1c also correlated signi�cantly with ΔMCP-1 ( ) in HE3286 (Table 3), but not in placebo participants. us the decrease in in�ammation (MCP-1) was associated with the decrease in HbA1c with HE3286 treatment.
e effect of HE3286 on nHbA1c in the overall population was not signi�cant. e obese patients with a BMI (>31 kg/m ), demonstrated a signi�cant treatment effect (test) to decrease nHbA1c by 0.6% Hb compared to placebo, but only aer exclusion of 2 outliers (Mahalanobis distance). e day 84 distribution of the ΔnHbA1c for BMI > 31 is shown in Figure 4(c) (outliers circled). e variances were much higher in the treatment-naïve patients' parameters compared to uncontrolled metformin-treated patients (Figure 3(f)). We speculated that these two outliers were still subject to in�ammation-induced random effects, aer only 84 days of treatment and that additional treatment may be necessary to observe effects in these individuals. Because of the lag in ΔHbA1c following glucose excursions, we tested the treatment effects on follow-up day 112. A signi�cant day 112 treatment effect (with no outliers) was found in the high BMI stratum, both by nonparametric and parametric tests ( Table 4). HE3286 participants had a signi�cant mean change from baseline (−1.0% Hb, ), whereas placebo did not. e mean change compared to placebo was also signi�cant (−0.7% Hb, ). e HE3286 participants also had a signi�cant median change from baseline (−1.2% Hb, ), whereas placebo did not. e magnitude of the response in the HE3286 treatment groups was signi�cant (−1.0% Hb, ) compared to placebo, as was the frequency of subjects with a 0.5% HbA1c decrease (9/12 versus 4/13, ). e day 112 ΔnHbA1c distributions for BMI > 31 kg/m are shown in Figure 4(d).

Postprandial Treatment Effect.
A signi�cant treatment effect that lowered fasting glucose was not found and was attributed to high metabolic parameter variations. Consequently, the possibility that HE3286 decreased HbA1c through action on postprandial glucose was investigated. Serum 1,5-anhydroglucitol (1,5-AH) is a dietary human metabolite that is reabsorbed by a kidney glucose transporter [22]. e 1,5-AH level declines when blood glucose levels are elevated above 10 mmol/L and likewise increases when the blood glucose level declines.
1,5-AH was measured in a subset of 42 participants (19 from stages 1 and 23 from stage 2) that had available day 84 retention samples. Analysis of 18 patients treated with HE3286 demonstrated that their 1,5-AH concentration increased signi�cantly (+10.4 mol/L, ); 24 treated with placebo demonstrated no signi�cant concentration increase (+0.6 mol/L, > ). e distribution of 1,5-AH responses is shown in Figure 4(e). e majority of HE3286 patients signi�cantly increased 1,5-AH, compared to placebos (15/18 versus 11/24, , Fisher's exact test). is outcome indicates that HE3286 had a treatment effect to decrease postprandial glucose excursions compared to placebo, which further supports that it's pharmacologic property is to decrease insulin resistance (see [9] and Figure 4(b)) and lower HbA1c.

HE3286-0401 Heteroscedasticity in HE3286 and Placebo
Groups. Heteroscedasticity (differences in variances between groups) was investigated by analyzing data distributions for normality (Shapiro-Wilks test) and analyzing dispersion (2-sided test). In cohort 1 placebo, but not HE3286, day 84 distributions ( test) were signi�cantly abnormal for changes in insulin, C-peptide, fasting glucose, HOMA2 %B, HOMA2 IR, and leptin in all subjects, and for changes in HbA1c, fasting glucose, and HOMA2 %B for MCP-1 > 40 pmol/L participants. Variances for cohort 1 placebo subjects ( test) were also signi�cantly higher than those of HE3286 subjects for insulin, C-peptide, and HOMA2 IR for all subjects.
Cohort 2 placebo, but not HE3286 distributions were abnormal ( test) for the group as a whole for changes in all the following parameters: day 84 nHbA1c, fasting glucose, MCP-1, and triglycerides and day 112 nHbA1c, fructosamine, and HOMA2 %B. In the BMI > 31 subgroup, abnormal distributions were found for changes in all the following parameters: day 84 HOMA2 %B and day 112 insulin, C-peptide, HOMA2 %B, and HOMA2 IR. Variances for cohort 2 placebo subjects as a whole were signi�cantly higher ( test) for changes in all of the following parameters: day 84 insulin, HOMA2 %B, and triglycerides, and day 112 insulin and HOMA2 %B. Variances in placebo were also higher for the BMI > 31 kg/m 2 subgroup for changes in day 84 MCP-1 and triglycerides and day 112 insulin (Table 5). ese differences in distribution and dispersion between groups were not readily evident until day 84 of treatment (data not shown). Together, these �ndings further support an HE3286 treatment effect that decreases random metabolic effects and restores homeostasis to uncontrolled T2DM patients.

Discussion
4.1. Study HE3286-0401. is initial clinical trial of HE3286 in diabetes was designed to take all eligible patients with uncomplicated T2DM even though HE3286 was only quali-�ed in animal models of obese diabetes and subsequently only demonstrated activity in obese individuals that present with in�ammation-induced insulin resistance. e strategic intent of the study was to identify the responding T2DM population by surveying a broad swath of the constellation of syndromes that are de�ned by the T2DM condition. Based on �ndings in cohort 1, which indicated low BMI individuals were HE3286 nonresponders, and the inclusion criteria in the second cohort of the trial were modi�ed, concentrating the population to elevated weight (BMI) and in�ammatory status (MCP-1). Additional criteria included a requirement for detectable insulin and C-peptide levels. is eliminated the patient population that had progressed to lose signi�cant -cell function and who were no longer able to produce insulin, a population clearly not indicated for treatment with an insulin sensitizer. Notably, these criteria were also imposed on clinical trials with the thiazolidinediones (J. Olefsky, personal communication). In addition treatment-naïve patients were recruited in cohort 2 to remove the potential for metformin to blunt the HE3286 treatment effect and consequently amplify the single agent treatment outcome.
We designed this study to test the hypothesis, based on preclinical data and on molecular studies of HE3286 binding partners, that HE3286 would decrease the hyperactivation of NFkB with consequent restoration of insulin signaling [5,6], dependent on its interaction with extracellular signal regulated kinase (ERK) 1 and 2 [7] in addition to other binding partners. ERK1 is an important mediator of in�ammation-induced insulin resistance [23][24][25], insulin receptor substrate (IRS)-1 serine (inhibitory) phosphorylation, and the inhibitory effect of TNF on insulin signaling [26]. HE3286 does not inhibit insulin-mediated ERK activation, but inhibits LPS-and TNF -stimulated ERK hyperactivation, and IRS-1 serine phosphorylation mediated by IKK and JNK [5,6]. Coincident HE3286-mediated changes in ERK, IKK, JNK, and p38 MAPK signal transduction may explain the preferential responses observed in high adiposity in�amed T2DM patients. Signal transduction pathways in omental fat are altered in obese, compared to lean individuals. In humans, activation of JNK and p38 MAPK was increased in omental fat (compared to paired subcutaneous fat) from obese, but not lean individuals, and this hyperphosphorylation correlated with clinical parameters of hyperglycemia and insulin resistance [27]. It will be important to further clarify the role of ERK in the activity of HE3286.

HE3286
Correlates. Data analysis presented here demonstrated that the cohort 1 day 84 changes in the primary end point HbA1c had a signi�cant relationship with expected changes in beta-cell function, fasting glucose, and weight, and also with baseline in�ammation status (MCP-1). Surprisingly a relationship with hemoglobin was also detected, a biomarker that is presumed stable over several weeks. Of these covariates associated with HbA1c change, only fasting glucose was signi�cant in placebo patients.
In the enriched cohort 2 population, the HbA1c HE3286 treatment response was no longer dependent on MCP-1 but rather BMI with a statistically signi�cant negative Heteroscedasticity describes differences in variances between groups. b Participants with baseline monocyte chemoattractant protein greater than the lowest tertile (40 pmol/L, see results). c Participants with baseline body mass index greater than the median (31 kg/m 2 , see results). d Abbreviations: Δ: change in; HOMA2 %B: homeostatic model assessment of pancreatic beta cell function; HOMA2 IR: homeostatic model assessment of insulin resistance; HbA1c: hemoglobin A1c; nHbA1c: HbA1c normalized to 84 day average hemoglobin mass; MCP-1: monocyte chemoattractant protein-1.
correlation; the higher the BMI the larger the effect on HbA1c decline. Higher BMI subjects presented with higher MCP-1. e cohort 2 outcome remained correlated with expected changes in -cell function and with fasting glucose.
us the general population enrolled in cohort 1 was a very different ensemble of participants than those enrolled in cohort 2. While the relationships of change in HbA1c with changes in -cell function and fasting glucose remained, the relationship to weight loss was not seen in the cohort 2 participants selected with higher BMI inclusion criteria.

Placebo
Correlates. Cohort 1 placebo group HbA1c change was dependent only on baseline in�ammation status (TNF and day 84 TNF change). In cohort 2, there were no placebo correlates to HbA1c change. Importantly, fasting glucose change was not correlated with HbA1c change in this group, indicating that glucose levels were uncoupled from the HbA1c surrogate marker. Rather, placebo HbA1c variance (CV) was correlated strictly with in�ammation in both cohorts. is was evidenced by correlation to baseline TNF in cohort 1 and dependent on both changes in MCP-1 and surprisingly hemoglobin CV in cohort 2. Hb CV was in turn dependent on baseline MCP-1 (TNF was not measured).
In cohort 1 this later dependency on Hb CV was not detected perhaps due to the heterogeneity of the more general patient population (including nonobese and nonin�amed diabetics). In cohort 2, placebo HbA1c CV was negatively correlated with weight change, indicating that higher weight led to increased variance. Cohort 2 placebo weight loss was unexpectedly unrelated to HbA1c and glucose control. Since placebo HbA1c CV was correlated with TNF change in cohort 1, the weight loss associated with higher HbA1c CV in the cohort 2 placebo group is presumed to be related using 1,5-anhydroglucitol [28] to better understand variation in glucose control.

HE3286-0401
Conclusions. e hypotheses tested in this study appear to be borne out in the high adiposity T2DM patient. HE3286 preferentially improved clinical parameters in obese in�amed insulin-resistant T2DM patients. Since in�ammatory changes were driving HbA1c changes in the placebo group for both cohorts 1 and 2, the changes observed with HE3286 treatment appear to be due to its anti-in�ammatory activity (i.e., to break the cycle of in�ammatory kinase-mediated inhibition of insulin receptor signaling). Furthermore, T2DM subjects that lack chronic, low-grade in�ammation lack the speci�c lesion in the insulin receptor signaling pathway that HE3286 was developed to interdict. eir glucose intolerance arises for other reasons, and therefore they are unaffected by HE3286.
Obese type 2 diabetic incidence is increasing at an alarming rate. Regaining glucose control and metabolic regulation and preventing or delaying macrovascular and microvascular complications could help to contain rising health care costs for end-stage diabetes complications. Understanding which patients are to bene�t from a new therapy is now a regulatory consideration. e FDA has published the Critical Path Initiative, with personalized medicine, or the patient-speci�c information to individualize therapy and disease management as a major theme, and published on the importance of clinical validation of personalized medicine selection criteria in diabetes [29]. Based on estimates of obese, in�amed diabetics in the future at approximately 50% (J. Olefsky, personal communication), HE3286 offers a potentially important personalized medicine for these subjects.
HE3286 is active at low (hormonal level) doses and is an anti-in�ammatory insulin sensitizer with a toxicology pro�le conducive for chronic daily use [1]. In the responsive subpopulations HE3286 signi�cantly decreased HbA1c compared to placebo, by day 84 in metformin-treated subjects with high MCP-1, and by day 112 in treatment-naïve subjects with high BMI. e data presented here in uncontrolled T2DM patients make a compelling argument for further testing of HE3286 in the high adiposity, in�amed T2DM patient subset, using oral glucose tolerance testing, 1,5-AH, or continuous glucose monitoring to assess treatment effects. e correlation or lack thereof with the surrogate marker HbA1c should be con�rmed in these uncontrolled patients.