Effects of Insulin Glargine and Liraglutide Therapy on Liver Fat as Measured by Magnetic Resonance in Patients With Type 2 Diabetes: A Randomized Trial

To reduce the risk of long-term microvascular and macrovascular complications, diabetes therapy should aim for tight glucose control as assessed by glycated hemoglobin (HbA_1c) below 7% (53 mmol/mol) while minimizing hypoglycemia (1,2). However, most patients still do not achieve glycemic targets and thus require treatment intensification (1,3). When metformin, eventually combined with other oral agents, fails to achieve glycemic targets, injectable therapy is recommended. Among these, second-line injectable treatments may have different effects on liver fat.

Insulin is associated with weight gain (4) and may be lipogenic because it increases uptake of glucose in the adipose cell and activates the lipogenic and glycolytic cell’s enzymes (5). Yet, prior studies that have used imaging techniques for fat quantification have shown a reduction in liver fat with insulin therapy (6–9). The mechanisms by which insulin therapy may reduce liver fat include inhibition of lipolysis, which may reduce free fatty acid flux to the liver (7), and decreased production of endogenous insulin passing through the liver to stimulate hepatic lipogenesis (6).

GLP-1 agonists (i.e., liraglutide and exenatide) promote weight loss. Further, GLP-1 agonists may decrease liver fat by increasing fatty acid uptake and VLDL transport (10) and stimulating hepatic lipid β-oxidation (11,12), improving hepatic insulin sensitivity (10) while also reducing hepatic lipogenesis via activation of the AMPK pathway (13).

Therefore, the effect of second-line injectable treatment alternatives (insulin glargine and liraglutide) on hepatic steatosis should be addressed in patients with type 2 diabetes inadequately controlled with oral agents therapy, including metformin.

Improvement in steatosis, as determined by magnetic resonance spectroscopy (MRS), is considered an acceptable alternative to liver biopsy to monitor liver steatosis in short-term phase 1 and 2 trials (14). The MRI-proton density fat fraction (PDFF) is a promising noninvasive biomarker for quantification of hepatic steatosis (15,16).

The purpose of this study was to determine, in patients with type 2 diabetes who were inadequately controlled on metformin therapy alone or combined with another oral antidiabetic medication, the effects of insulin glargine versus liraglutide therapy on the liver fat burden, as measured in vivo by MRS and MRI.

This prospective, open-label, randomized trial was approved by our institutional review boards and compliant with the Declaration of Helsinki (17). Patients provided written informed consent before study-related procedures were initiated. The study sought to recruit patients with type 2 diabetes prescreened by endocrinologists from the two institutions and by advertisements in local newspapers.

Inclusion criteria included age ≥18 years at screening; being ambulatory; having a diagnosis of type 2 diabetes inadequately controlled (HbA_1c ≥6.5% [48 mmol/mol]) under metformin monotherapy or metformin-sulfonylurea, metformin-repaglinide, or metformin-dipeptidyl peptidase-4 inhibitor combined therapy; and having an abdominal girth >94 cm for men and >80 cm for women.

Patients were excluded if they had contraindications for MRI (e.g., metallic implants, pacemakers, or claustrophobia); type 1 diabetes or episodes of ketoacidosis; major debilitating disease, including malignant disorders; or evidence in the last 6 months of significant heart disease or stroke, including myocardial infarction, unstable angina, coronary bypass and/or percutaneous transluminal coronary angioplasty, congestive heart failure (New York Heart Association Functional Classification III-IV), or severe ischemic heart disease. Exclusion criteria also included patients receiving insulin or thiazolidinediones 3 months before screening, a serum creatinine level >150 mmol/L or an estimated glomerular filtration rate <30 mL/min; women planning prospective pregnancy; or an established diagnosis of chronic liver disease other than nonalcoholic fatty liver disease (NAFLD), including hepatitis B (HBsAg) or C (anti-HCV) viral infection, hemochromatosis (transferrin, ferritin), Wilson disease (ceruloplasmin), α-1-antitrypsin deficiency, autoimmune hepatitis (anti-nuclear antibodies, anti-smooth muscle antibody, anti-mitochondrial antibody), current or previous use of oral or injectable corticosteroids, and excessive alcohol intake, defined as a daily limit of 30 g (3 drinks) for men and 20 g (2 drinks) for women. Additional screening tests included aspartate and alanine aminotransferase, albumin, phosphatase alkaline, γ-glutamyltranspeptidase, creatinine, sodium, potassium, and complete blood count.

Eligible patients who gave consent were scheduled for assessment by the participating endocrinologists. Patients were randomized using random permutation and allocated to treatment using sealed envelopes for research nurses, endocrinologists, and patients. Investigators involved in imaging data analysis (i.e., physicist, engineers, image analysts, and radiologists) were blinded to patients' information and the allocation sequence. Although aware of the treatment group, the endocrinologists were blinded to the imaging results until final data analysis.

Patients were randomly assigned to the insulin glargine or liraglutide treatment group. Patients randomized to insulin were started at baseline on glargine (Lantus, Sanofi) 10 IU subcutaneously at bedtime and titrated by 1 unit each day to achieve a fasting plasma glucose <7 mmol/L. Those randomized to liraglutide (Victoza, Novo Nordisk) were started on 0.6 mg subcutaneously per day and increased by weekly forced titration to 1.8 mg or the maximal tolerated dose. If they were randomized to insulin, they received additional education for hypoglycemia treatment and were further advised to reduce the size of portions in their diet. Patients who were randomized to liraglutide were also informed about possible gastrointestinal side effects. Metformin and oral hypoglycemic agents were administered at a constant dose, except for gliptins, which were interrupted if patients were randomized to liraglutide (as recommended), and for sulfonylureas, which could be reduced if patients had hypoglycemia. The patients were treated for 12 weeks.

Patients were seen at the research clinic by the endocrinologist a week before baseline, at baseline (i.e., beginning of treatment), and then monthly for the duration of the study for a total of five visits. Each session consisted of the compilation of the patient’s history, a physical examination, including the weight and BMI, and biochemical tests. The follow-up visits also included documentation of adverse effects and adjustment of the insulin dose as needed for subjects in the insulin group. MRS and MRI were conducted at baseline and the end of the study for noninvasive liver fat quantification.

The primary outcome measure was the change in liver fat content from baseline to the end of the study under liraglutide or insulin treatment, as measured by MRS and MRI. Secondary outcome measures included changes in fasting plasma glucose, HbA_1c, anthropometric measurements, and the calculated BMI.

A priori sample size calculations were based on the ability to detect a 5% absolute clinical difference in fat content before and after the intervention. With a deviation estimate of 5% obtained from similar MRS studies (6,18–20), we estimated that 17 patients in each group would be required, for a total of 34 patients (two-tailed α = 0.05, β = 0.20). To account for a potential dropout rate of 5%, up to 36 patients could be enrolled.

All spectra were acquired on a 3T clinical system (Achieva TX, Philips Healthcare, Best, the Netherlands) using the following sequence: single breath-hold stimulated echo acquisition mode sequence without fat and water saturation with parameters: voxel size = 25 mm × 25 mm × 25 mm; repetition time (TR) = 3,500 ms; echo time (TE) = 10, 15, 20, 25, 30 ms; mixing time = 15 ms; spectral width = 1,250 Hz. The TR was chosen to be sufficiently long to minimize T1-weighting effects, and multiecho data were acquired for correction of T2-weighting effects. Two voxels were acquired for each patient.

All spectroscopy data sets were exported from the MRI system in the SPAR/SDAT format and were analyzed by a single observer (5 years of experience, blinded to treatment) using the AMARES (advanced method for accurate, robust, and efficient spectral fitting) algorithm (21) provided in the jMRUI software. The analyses were performed according to the method described in detail by Hamilton et al. (22). The mean value of the two voxels acquired was reported as the MRS-determined fat fraction.

Multiecho spoiled gradient-recalled echo sequences with seven-echo readout were acquired during a single breath-hold to cover the entire liver. The sequence parameters were flip angle, 10°; field of view, 400 mm; section thickness, 9 mm; 1 mm gap; receiver bandwidth, 1,215 Hz/pixel; sensitivity encoding acceleration factor, 2.6; acquired voxel size, 2.5 × 2.5 × 9 mm; number of averages, 1; TR, 235 ms; the first TE was 1.15 ms, with a ∆TE of 1.15 ms (therefore, the 7 TEs were 1.15, 2.30, 3.45, 4.60, 5.75, 6.90, and 8.05 ms) (23).

The anonymized MRIs were transferred to an offline workstation and analyzed by an image analyst (2 years of experience, blinded to treatment), supervised by an abdominal radiologist (7 years of experience, blinded to treatment). Semiautomated segmentation of all MRI studies was based on an active contour method (or “snakes”) and performed using SliceOmatic 4.3 Rev-11 two-dimensional postprocessing software (TomoVision, Montreal, Canada). The outline of the liver was manually segmented on the middle slice and then propagated automatically on the other slices. Corrections of the liver boundary delineation were performed manually. Large vessels, including the portal vein at the hepatic hilum and the inferior vena cava, were also excluded from the segmented liver volume.

MRI-PDFF maps were calculated pixel-by-pixel with MatLab 8.1 R2013a software (MathWorks, Natick, MA) according to a method described by Yokoo et al. (23). The algorithm simultaneously estimated T2* and PDFF in each pixel on the image by using nonlinear least squares fitting from all seven echoes. A triglyceride model of human liver fat spectra, described by Hamilton et al. (22), was used to incorporate spectral correction of multifrequency interference effects of the protons in the fat. The PDFF in each pixel then was calculated as the ratio of the fat proton density to the total (fat and water) proton density (15).

Total liver fat index (TLFI) was calculated as the product of the liver volume and the liver mean PDFF (24).

Categorical variables are expressed as numbers and percentages. Continuous variables are expressed as mean ± SD. The primary analysis was an intention-to-treat analysis. Comparisons between the groups at baseline were done with the two-tailed unpaired Student t test or the Fisher exact test in case of categorical variables. Comparisons between the groups were done with repeated-measures ANOVA using a linear mixed model to account for the time factor at two levels (before and after treatment), the group factor at two levels (insulin glargine, liraglutide), and their interaction. P values of <0.05 were considered significant. Statistical analyses were performed by a biostatistician with SAS 9.3 statistical software (SAS Institute Inc., Cary, NC).

Between June 2011 and February 2014, 95 patients were assessed for eligibility. Among those, 60 were excluded (51 did not meet the eligibility criteria, 5 refused to participate, 3 had claustrophobia, and 1 had a pacemaker). A total of 35 patients were randomized to one of the two treatment groups—17 were allocated to insulin glargine and 18 to liraglutide—and monitored for 12 weeks. Among those who were allocated to insulin glargine, none withdrew before the end of the study, whereas four of those allocated to liraglutide discontinued prematurely due to adverse effects. Baseline characteristics were similar between the two groups (Table 1). At randomization, the proportion of patients previously receiving metformin only versus combined therapy with metformin was similar between the two groups (P = 1.00).

Among those randomized to the insulin glargine group, the daily dosage was 22.4 ± 15.6 IU at the end of the study. Among those randomized to the liraglutide group who completed the study, the daily dosage was 1.3 ± 0.5 mg.

Insulin glargine treatment resulted in a significant improvement in fasting plasma glucose (8.4 to 6.4 mmol/L, P = 0.001) and HbA_1c (7.9% to 7.2% [62.5 mmol/mol to 55.2 mmol/mol], P = 0.005). Liraglutide treatment also resulted in a significant improvement in fasting plasma glucose (8.7 to 6.3 mmol/L, P < 0.001) and HbA_1c (7.6% to 6.7% [59.8 mmol/mol to 50.2 mmol/mol], P < 0.001). These improvements did not differ between the treatment groups (P = 0.60 and P = 0.52, respectively) (Table 2).

Although insulin glargine treatment over 3 months was associated with a trend toward a reduction in the liver fat fraction as assessed by MRS (12.6% to 9.9%, P = 0.06), liraglutide treatment did not show any change in the liver fat fraction (12.0% to 12.4%, P = 0.80). These changes did not significantly differ between the treatment groups (P = 0.16) (Fig. 1 and Table 2). An MRS and MRI protocol at baseline and at the end of treatment for a representative patient is shown in Supplementary Fig. 1.

Insulin glargine treatment resulted in a significant decrease in the mean liver PDFF, as assessed by MRI (13.8% to 10.6%, P = 0.005), but liraglutide treatment had no effect on the mean liver PDFF (13.8% to 12.2%, P = 0.15). These changes did not significantly differ between the treatment groups (P = 0.34) (Fig. 1 and Table 2).

Although insulin glargine treatment significantly reduced the liver volume (from 2,010.6 to 1,858.7 mL, P = 0.01), liraglutide was without effect (1,848.2 to 1,780.3 mL, P = 0.30). These changes did not significantly differ between the treatment groups (P = 0.34) (Fig. 1 and Table 2).

Finally, insulin glargine treatment was associated with a significant reduction in the calculated TLFI (304.4 vs. 209.3 % ⋅ mL, P = 0.01). However, liraglutide treatment had no effect on TLFI (269.8 vs. 236.9 % ⋅ mL, P = 0.39). These changes did not significantly differ between the treatment groups (P = 0.23) (Fig. 1 and Table 2).

Insulin glargine treatment over 3 months had no effect on body weight (87.1 to 87.1 kg, P = 0.98), but liraglutide treatment was associated with a significant reduction in weight (87.4 to 84.6 kg, P = 0.005). These changes were significantly different between the treatment groups (P = 0.03) (Table 2).

A weak positive correlation was found between changes in body weight and changes in the mean liver PDFF (R = 0.342, P = 0.06) (Fig. 2).

Although insulin glargine treatment over 3 months had no effect on BMI (31.2 to 31.3 kg/m², P = 0.92), liraglutide treatment was associated with a significant reduction in BMI (31.3 to 30.1 kg/m², P = 0.008). These changes were significantly different between the treatment groups (P = 0.04) (Table 2).

Insulin glargine treatment resulted in a significant increase in serum triglycerides (1.4 to 2.9 mmol/L, P = 0.003), whereas liraglutide treatment had no significant effect on serum triglycerides (1.9 to 2.6 mmol/L, P = 0.10). These changes did not significantly differ between the treatment groups (P = 0.21) (Table 2).

Of the 17 patients randomized to insulin, 3 had an episode of nonsevere hypoglycemia, but none discontinued treatment. Of the 18 patients randomized to liraglutide, 4 patients discontinued treatment prematurely: 1 had nausea and diarrhea, 1 had nausea and vomiting, 1 had headaches, and 1 had general discomfort. The rate of adverse effects was similar in the insulin glargine and in the liraglutide group (18% vs. 22%, P = 1.00) (Table 2).

In this randomized trial, MRS and MRI-PDFF were used as surrogate biomarkers to compare the effect on liver steatosis of insulin glargine or liraglutide as second-line therapy in patients with type 2 diabetes inadequately controlled with oral agents therapy, including metformin. We found that MRI-PDFF was able to detect a significant reduction in liver fat under insulin glargine therapy. However, the improvements in the liver fat fraction and glycemic control were not significantly different between the insulin glargine and the liraglutide treatment groups.

The size effect in steatosis reduction observed in the insulin glargine group was larger on MRI-PDFF than on MRS. This minor discrepancy may be explained by the sampling methods. Regions of interest were used to measure the MRS-based fat fraction, whereas the MRI-based liver mean fat fraction was calculated as an average of the PDFF over the entire liver parenchyma. Further, by combining whole liver segmentation with the calculation of the PDFF maps, these MRI-based biomarkers may circumvent the regional variability inherent to regional sampling methods.

In this randomized controlled trial using these surrogate imaging end points, we demonstrated a reduction in the total liver fat burden in response to insulin therapy in patients with type 2 diabetes. The administration of exogenous insulin improved glycemia and HbA_1c by correcting peripheral insulin deficiency. This suggests that the administration of subcutaneous insulin required to induce and maintain near-normoglycemia probably suppresses endogenous insulin secretion known to stimulate hepatic lipogenesis (6). This concept is reassuring and important because many patients with type 2 diabetes and NAFLD have to be treated with insulin to achieve adequate glycemic control.

The reduction in the liver fat fraction with insulin therapy is consistent with the findings of previous studies that have used MRS (6–8) or computed tomography (9) for quantification and monitoring of fractional liver fat. In parallel to liver fat fraction reduction, our study revealed an average concomitant liver volume decrease of 152 mL in the insulin-treated group. These observations are similar to those of earlier human studies that have shown liver volume reduction in parallel to liver fat decrease in the setting of a low-calorie diet (25) or exercise and diet (26). Further, the reduction in the liver fat fraction and liver volume were accompanied by a reduction in the total liver fat burden.

In our study, liraglutide therapy did not significantly change the biomarkers of liver fat burden (MRS-PDFF, liver mean MRI-PDFF, liver volume, and TLFI). Prior studies suggested that GLP-1 receptor agonists, such as liraglutide and exenatide, would improve liver steatosis in animal models (27,28), in human case series (29), and in prospective nonrandomized trials (30,31). However, in the setting of the randomized controlled trial Liraglutide Efficacy and Action in Diabetes (LEAD)-2 substudy, there was a nonsignificant trend toward a reduction in liver steatosis as assessed by computed tomography with the liver-to-spleen attenuation ratio, but only for the 23 patients receiving higher doses (1.8 mg) of liraglutide (11,32). It is possible that the postprandial increase in insulin secretion in the portal circulation and the decrease in glucagon secretion induced by the GLP-1 analogs could favor lipogenesis and therefore explain the lack of effect of liraglutide on hepatic steatosis observed in the current study. Larger and longer studies will be required to confirm the effect of insulin and GLP-1 analogs on liver steatosis.

Considering that liver fat may mirror tissue adiposity, we assessed anthropometric measures to determine if other fat depots were similarly affected by insulin glargine and liraglutide treatment. In patients randomized to insulin therapy, no change was found in weight and BMI despite a reduction in MR-based biomarkers of liver fat burden. This study included patients with mildly elevated HbA_1c at baseline, which may have contributed to weight and BMI stability among patients randomized to insulin glargine. The modest weight gain observed in those patients is consistent with the findings of the largest randomized trial on the early introduction of insulin (33). It is possible that the inclusion of patients with a higher baseline HbA_1c treated with insulin would have led to weight gain due to correction of a catabolic state.

Interestingly, the improvement in BMI in patients randomized to liraglutide therapy was not associated with concomitant changes in the liver fat burden. This indicates that different interventions can have discordant effects on fat deposition. A similar observation was recently made by Jonker et al. (34), who found a discordant effect of exercise on visceral and epicardial adipose tissue in patients with type 2 diabetes.

We unexpectedly observed a significant increase in serum triglyceride discordant with the decrease in liver fat in patients randomized to liraglutide therapy. This finding was inconsistent with prior results of a meta-analysis by Chaudhuri et al. (35), which reported favorable effects of insulin glargine on plasma lipid profiles. As a potential explanation, we hypothesize that improved insulin action in the liver may increase glucose uptake and utilization for lipogenesis while reducing the rate of synthesis and secretion of apolipoprotein B. On one hand, this would promote the generation of fewer but larger more triglyceride-rich VLDL and may explain increased fasting plasma triglyceride with insulin. On the other hand, reduced liver fat may be secondary to reduced hepatic influx of fatty acids due to improved insulin action in white adipose tissue.

Our study has some limitations, including the choice of MR rather than liver biopsy as the reference standard. Considering the exploratory nature of our primary end point in patients with inadequately controlled type 2 diabetes but otherwise not known to have chronic liver disease, performing baseline and end-of-treatment liver biopsies even in a research setting would not have been ethically acceptable (36).

Although changes in liver fat fraction were observed in this study, the length of the intervention was likely too short to expect changes in liver inflammation or fibrosis that would translate into stiffness changes. Furthermore, a longer study would have favored GLP-1 agonist therapy in terms of body weight loss and, possibly, a concomitant reduction in the liver fat fraction. However, the 12-week duration of our study was sufficient to address our primary end point and was justified by the normalization of transaminases, another surrogate marker of treatment response, within this interval in a previous randomized trial evaluating the effect of a thiazolidinedione in patients with type 2 diabetes (18). Further, this length of intervention was sufficient to demonstrate improvements in hepatic triglyceride content as measured by MRS in a prior trial (20).

In addition, our study had a relatively small sample size. However, the sample size was calculated a priori based on estimates obtained from similar MR studies (6,18–20). Further, the continuous measurements and the smaller variances obtained by MR-based methods compared with ordinal grading of liver steatosis by histopathology permit smaller sample sizes to explore treatment trends in short-term trials (14).

Another limitation was the substantial dropout rate in the liraglutide therapy group. Unlike episodes of hypoglycemia due to insulin, major gastrointestinal side effects led patients in the liraglutide group to stop treatment before 12 weeks. Yet, the primary analysis was designed as an intention-to-treat analysis. Furthermore, the rate of gastrointestinal adverse effects observed in this study was similar to those reported in the literature, and the rate of patient withdrawal was similar to that observed in larger randomized clinical trials (11,37,38).

This randomized clinical trial showed that the administration of subcutaneous insulin glargine therapy reduced the liver fat burden, although the improvements in liver fat fraction and glycemia control were not significantly different from those in the liraglutide group. This reduction in liver fat burden on insulin therapy is reassuring and important because many patients with type 2 diabetes and NAFLD have to be treated with insulin to achieve adequate glycemic control. All changes in liver fat biometry were detected by MRI. This observation suggests that these noninvasive biomarkers, which require a combination of liver segmentation and MRI-PDFF maps, may detect early changes secondary to drug therapy. Future studies in patients with type 2 diabetes may use these noninvasive biomarkers to assess the development of liver fat burden, its progression, and its response to treatment in larger clinical trials.

Funding. Funding for this investigator-initiated study was provided by grants from the Radiological Society of North America Research and Education Foundation (RSD1108), Diabète Québec, and Canadian Heads of Academic Radiology-GE Healthcare to A.T. and by Natural Sciences and Engineering Research Council of Canada Discovery grant 107998 to J.d.G. A.T. is supported by a clinical research scholarship from the Fonds de Recherche du Québec en Santé (FRQ-S) and Fondation de l'Association des Radiologistes du Québec (#26993). R.R.-L. is supported by a senior scholarship from the FRQ-S and a J-A DeSève Research Chair. J.d.G. is supported by a Canada Research Chair in 3D Imaging and Biomedical Engineering. G.S. is supported by a national Scientist Award of FRQ-S.

Duality of Interest. A.T. reports honoraria as an invited speaker from Siemens Medical and from Boehringer Ingelheim outside the scope of the submitted work. R.R.-L. reports grants and other support from AstraZeneca and Eli Lilly; grants from Merck, Novo Nordisk, and Sanofi; and other support from Boehringer Ingelheim, Medtronic, and Takeda, outside the submitted work. G.G. reports personal fees from Philips Healthcare as an employee (clinical scientist). G.S. discloses receipt of research funds from Biotronik, Siemens Medical, and Bracco Diagnostic and honoraria as an invited speaker from Bracco Diagnostic, Siemens Medical, and Cook Medical. J.-L.C. reports grants and other support from Novo Nordisk, Sanofi, Eli Lilly, and Merck and other support from Bayer, outside the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.T. contributed to study design, researched data, wrote the manuscript, and reviewed and edited the manuscript. R.R.-L. contributed to discussion and reviewed and edited the manuscript. H.C., C.W.-B., G.G., K.M.-.T., G.C., D.O., and J.d.G. researched data and reviewed and edited the manuscript. A.-S.J. performed the statistical analysis and reviewed and edited the manuscript. G.S. and J.-L.C. contributed to study design, researched data, and reviewed and edited the manuscript. A.T. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.