Study design
INTIMET was a single-site investigator-initiated phase 3 randomized placebo-controlled trial that was approved by the St Vincent’s Hospital Human Research and Ethics Committee (Sydney) and was conducted in accordance with the principles of the Declaration of Helsinki. The trial was prospectively registered on 17th October 2019 prior to the start of participant enrolment (Australian New Zealand Clinical Trials Registry, ACTRN12619001440112). The trial did not involve a data safety monitoring board. Participants provided written informed consent and were not compensated. The trial protocol was previously published, and discrepancies from the study as original planned are as stated48. Assessment of energy expenditure using indirect calorimetry was planned for in our original protocol but was aborted during the COVID-19 pandemic due to concerns regarding adequate infection control. Therefore, energy expenditure and oxidative metabolism data were not available.
Participants
Participants were enroled between November 2019 and December 2021. All metabolic measurements occurred within the Clinical Research Facility (CRF) of the Garvan Institute of Medical Research, Darlinghurst, New South Wales Australia, and data were collected, recorded and stored within the Garvan REDCap database. The study was conducted in 40 participants with type 1 diabetes who met the following criteria, and 20 participants without diabetes were studied at baseline only.
Study eligibility criteria
Inclusion
1.
Age 20–55 years.
2.
Pre-menopausal.
And if type 1 diabetes
3.
At least 10 years since diagnosis of type 1 diabetes.
4.
Fasting c-peptide <0.3 nmol/L.
5.
HbA1c ≤ 9.5% (≤80 mmol/mol).
Exclusion
1.
Current smoking.
2.
Medications affecting glucose metabolism (including, but not limited to glucocorticoids, anti-psychotic medications, immunosuppressants, glucagon-like peptide-1 receptor agonists, sodium-glucose cotransporter inhibitors).
3.
Exposure to metformin within the last 30 days (a wash out period was permitted).
4.
Alcohol intake greater than 40 g/day in men or greater than 20 g/day in women
5.
History of significant weight change (defined as greater than 5%) in the last 3 months, or history of weight loss surgery.
6.
Pregnancy or breastfeeding.
7.
Major organ dysfunction (including eGFR <60 ml/min/1.73 m2, liver disease transaminases greater than 3 times the upper limit of normal, cardiac event within the last 6 months, current cancer or uncontrolled thyroid dysfunction).
8.
Diabetic ketoacidosis or severe hypoglycemia (defined as hypoglycemia requiring third-party assistance) in the last 6 months.
Intervention
After completing baseline studies, participants with type 1 diabetes were randomized to metformin extended release, or identical appearing placebo, as 500 mg tablets, titrated from 500 mg for 7 days, then 1000 mg for 7 days then 1500 mg for the remaining 26-week treatment period. Dosage could be adjusted to accommodate adverse effects, and a submaximal dose was permitted. Pill count was used to assess medication adherence. Randomization, performed by an independent clinician, was 1:1 for metformin to placebo using a computerized minimization procedure to stratify for BMI, HbA1c, sex and age (40%, 20%, 20% and 20% weighting, respectively)49. Sex was determined by self-report at enrolment. Participants, outcome assessors and data analysts were blinded to treatment allocation.
Study procedures
At baseline for all participants with and without diabetes, and at 26 weeks for randomized participants with type 1 diabetes, comprehensive metabolic assessments were performed including blood pressure, BMI and waist-to-hip ratio measurement. Body composition was assessed using dual-energy x-ray absorptiometry (Lunar Prodigy GE-Lunar, CoreScan™ algorithm), to determine fat and fat-free mass (FFM) and visceral adipose tissue (VAT) mass, Magnetic Resonance Imaging (MRI) (GE Architect 3.0-T, GE Healthcare, software version DV28 r5, Chicago, Illinois, image analysis by AMRA® Researcher AMRA Medical AB, Linköping), and transient elastography (Fibroscan® 502 Touch, FibroScan Expert 630®, Echosens, Paris France) to quantify liver and muscle fat. We assessed arterial stiffness using radial artery applanation tonometry fasted and during the basal period of the clamp procedure during euglycemia (SphygmoCor., AtCor Medical, Australia) to compute the augmentation index (AIx), a validated surrogate of cardiovascular disease19.
In participants with type 1 diabetes, insulin doses were recorded for up to 14 days in the pre-treatment and end of treatment period. Records were kept manually by participants treated with insulin injections or downloaded using manufacturer-specific software if treated with an insulin pump. Participants with type 1 diabetes wore blinded continuous glucose monitoring (CGM). The first 20 enrolled participants were provided the Medtronic iPro ® Enlite system, (changed after 7 days, in total 2 devices worn per participant) (Medtronic Northridge, CA) and the final twenty participants were provided the Dexcom G6 monitor (Dexcom, San Diego, CA) due to discontinued supply of the Medtronic system mid-trial. These systems were inserted subcutaneously as per manufacturer specifications to measure interstitial glucose, providing measurements approximately every 5 minutes. The recorded CGM outcomes measures included mean sensor glucose, % time in range, % time <3.0 mmol/L, % time <3.9 mmol/L, % time >10 mmol/L, % time >13.9 mmol/L, glucose variability (coefficient of variation; CV%, defined as standard deviation glucose divided by mean glucose*100), mean overnight (0000 − 0559AM) and mean daytime (0600-2359PM) glucose.
To compute these metrics, valid time-stamped sensor glucose data were manually sorted according to glucose value and time of day. At least valid 3 days of CGM data per sensor period were required for analysis. A 24-h period was defined as 0000–2359 h. A day was defined as eligible if ≤20% of the datapoints were missing during that period. Missing sensor data were not imputed.
Diet records were collected for up to 14 days (Easy Diet Diary, Xyris. Brisbane Australia) and analysed using Foodworks 10 Professional (Xyris Pty Ltd, 2019). Retinal photographs were graded using the International Clinical Diabetic Retinopathy (ICDR) severity scale50. Urine albumin-creatinine ratio (ACR) data was collated from the preceding 12 months and albuminuria was coded as present or absent based on the single sample.
Participants withheld caffeine, alcohol and strenuous physical activity for 48 h prior to their planned clamp procedure. Women were studied in the follicular phase of the menstrual cycle. Participants with type 1 diabetes underwent washout of their subcutaneous insulin under the supervision of the study Endocrinologist, per published study protocol48.
Participants arrived fasted for a 6-h, three-stage hyperinsulinemic-euglycemic clamp48. The first stage (basal-phase) was a 2-h tracer equilibration period (6,6-2H2-glucose, Cambridge Isotope Laboratories), started by a priming bolus (5 mg/kg; administered over 5 min) then continuous infusion of the 6,6-2H2- glucose (Cambridge Isotope Laboratories) (0.05 mg/kg/min), which was continued for at least 2 h, aiming for a 2% enrichment by the time of the steady state basal sampling phase. Participants with type 1 diabetes also received a variable insulin infusion (Actrapid, Novo Nordisk) to achieve a target glucose of 5.5 mmol/L during subcutaneous insulin washout48. In the second stage, a fixed insulin infusion was administered at 20 mU/m2/min for 2 h (low-dose phase), with reduction of the tracer infusion dose to 0.025 mg/kg/min. The final high-dose phase was initiated with an insulin bolus (240 mU/m2/min over 2 min, then 120 mU/m2/min over 2 min), then continued at 60 mU/m2/min for 2 h. Body surface area (BSA) was calculated according to the Du Bois formula51.
Hepatic and adipose insulin sensitivity were defined by EGP and NEFA levels respectively, after partial suppression by insulin during the steady state of the low-dose phase. EGP was calculated using Steele non–steady-state equations, modified for use with stable isotope-labelled glucose52. We used a spiked dextrose infusion (25% dextrose with 2% 6,6-2H2-glucose enrichment) to maintain euglycemia and stable tracer enrichment during the low- and high-dose phases. Muscle insulin sensitivity was defined as the high-dose phase GIR required to achieve the target glucose level. EGP and GIR data were normalized to fat-free mass. For clamp analysis, we took arterialized samples once steady-state conditions were achieved. Steady-state was defined as minimally fluctuating glucose levels for unchanging GIR and occurred within the last 30 min of each clamp phase. Glucose tracer enrichment was analysed via positive chemical ionization gas chromatography-mass spectrometry using the glucose methyloxime pentaproprionate derivatization strategy, per published methods53.
Fasting plasma and serum samples were collected for metabolite assessment on pre- and post-treatment clamp study days prior to clamp commencement, including serum for GDF15 (analysed using an in-house ELISA method per previous descriptions), insulin, NEFA and SHBG, and plasma for other blood measures including lipids, uric acid, LFT, IGF-1, adiponectin, IL-6, sE-selectin, sICAM-1 and glucagon54,55. Any samples below the sensitivity of the assay were assigned a value equivalent to the limit of quantification (LoQ) divided by the square root of 2 (LoQ/√2).
Trial endpoints
The primary endpoint was change in liver insulin sensitivity after 26 weeks. Pre-specified secondary endpoints included the change in muscle insulin sensitivity, change in adipose insulin sensitivity, insulin dose, HbA1c and continuous glucose monitoring parameters, GDF15 and other vascular and metabolic measurements (reported in Supplementary Table S4). Glucagon analyses were exploratory. Microbiome analyses are ongoing and not reported in this article. Liver fibrosis measures were not reported, as all values fell within the lower normal reference range in our cohort. CRP analysis was not performed as the available assay did not meet the sensitivity requirements for this study.
Statistics and reproducibility
Randomized control trial
The primary endpoint was assessed by the time-by-treatment group interaction using a generalized linear mixed model with low-dose EGP at baseline and 26 weeks as the response variable. Group allocation, time, and group-by-time interaction terms were included as fixed effects, and a random intercept was assigned for each participant. If a baseline difference was present, a group-by-time interaction model without treatment group as a fixed effect was used56. All analyses were adjusted for baseline value. Model distributions and link were selected based on residual plots.
In the initial protocol, basal Ra was listed with low-dose EGP as a co-primary endpoint. Basal Ra was removed after trial commencement but before trial completion due to concerns regarding the variable insulin infusion during the basal clamp period impacting the validity of the basal Ra measure. Since participants with type 1 diabetes tended to arrive with glucose levels above the glucose target, variable insulin concentrations were infused in the type 1 diabetes participants during the basal phase to achieve euglycemia. Higher insulin concentrations would then excessively suppress Ra in the type 1 diabetes participants compared to control participants, and variable insulin concentrations during the basal phase would invalidate the basal Ra measures in type 1 diabetes as each participant would be subjected to a different insulin concentration depending on their arrival glucose. Further, insulin concentrations measured in type 1 diabetes represent exogenously infused insulin into the peripheral circulation. Unlike type 1 diabetes, in control participants without diabetes, plasma insulin concentrations measured during the basal phase represent endogenous insulin secreted from the pancreas to the portal circulation, with hepatic clearance before reaching the peripheral circulation. Thus, the basal phase served the purpose of tracer equilibration, and achievement of euglycemia only.
To detect an end-of-treatment difference between groups in EGP of 0.3 mg/kg/min with 80% power (presumed 0.3 mg/kg/min standard deviation; equivalent to 2.4 µmol/kgFFM/min, alpha level 0.05), we needed 17 participants in each group, increased to 20 to accommodate up to 15% participant attrition.
Data were analysed by intention-to-treat principles without imputation. All assessments used two-sided tests and a p-value ≤ 0.05 was considered statistically significant. For secondary analyses, a Bonferroni corrected p-value was applied to account for multiple comparisons. Normally distributed data were presented as mean and standard deviation (SD) and analysed using parametric tests. Skewed data were presented as median and interquartile range (IQR) then analysed with non-parametric tests or as log-transformed data analysed with parametric tests. Categorical
data were presented as counts and percentages. We did not perform specific sex-based analyses, as the study was not powered for subgroup analyses.
Baseline cross-sectional study
Two-sample t-tests, Mann–Whitney U or Chi-square tests were used to compare baseline characteristics between participants with and without type 1 diabetes. We determined relationships between insulin resistance and cardiometabolic variables using Pearson, Spearman’s, and regression models to determine predictors of insulin resistance. As exploratory analyses, these data are presented without correction for multiple testing.
Excluded data points
A single outlying GDF15 value from the type 1 diabetes group was excluded from the analysis. This datapoint was from an individual who was subsequently diagnosed with a medical condition deemed likely to falsely elevate serum GDF15 levels. Hormonal contraceptive users were excluded for all analyses involving SHBG.
Statistical analyses were performed using SPSS version 28.0.1.0 (IBM Corp. Released 2021. IBM SPSS Statistics for Macintosh, Version 28.0. Armonk, NY: IBM Corp). Due to the nature of the data, Rstudio (Rstudio 2023.06.1 + 524 “Mountain Hydrangea”) using the glmmTMB package was used for continuous glucose monitoring data as SPSS did not have the beta-binomial distribution option within its generalized linear mixed modelling framework. A p-value ≤ 0.05 was considered statistically significant.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.