Does metformin reduce excess birthweight in offspring of obese pregnant women? A randomised controlled trial of efficacy, exploration of mechanisms and evaluation of other pregnancy complications

Primary objective

The primary objective was to determine the efficacy of metformin (up to 2500 mg daily) given to obese pregnant women from 12–16 weeks’ gestation until delivery in reducing gestational age-, parity- and sex-adjusted birthweight centile of the baby.

Secondary objectives

• To determine the pattern of association between insulin resistance and adverse pregnancy outcomes including the incidence of pregnancy-induced hypertension, pre-eclampsia, caesarean section and post-partum haemorrhage, maternal weight gain during pregnancy and the incidence of the admission to the neonatal unit.

• To determine the effect of metformin on maternal body composition.

• To determine the effect of metformin on neonatal body composition.

• To determine the effect of metformin on maternal inflammatory and metabolic variables (measured at 28 and 36 weeks’ gestation) and neonatal inflammatory variables (measured in cord blood at birth).

• To confirm that metformin does not increase the rate of babies with a low birthweight centile.

• To determine the efficacy (as opposed to the effectiveness) of metformin when analysis is restricted to those with pharmacological circulating levels of the drug.

Substudies

A series of nested substudies was also performed with the following objectives:

• to determine the effect of metformin on maternal cortisol levels in obese pregnant women

• to determine the effect of metformin on hepatic and peripheral insulin sensitivity at 36 weeks’ gestation in obese pregnant women

• to determine the effect of metformin on endothelium-dependent flow-mediated dilatation (FMD) in obese pregnant women

• to determine the effect of metformin on maternal subcutaneous and visceral adipose tissue deposition and hepatic and skeletal muscle ectopic fat deposition during pregnancy

• to determine the effect of metformin on fetal liver volume and subcutaneous fat deposition

• to determine the effect of metformin on myometrial contractility and myometrial glycogen storage in obese pregnant women

• to determine the effect of metformin on placental glucocorticoid receptor and 11β-hydroxysteroid dehydrogenase (HSD) type 1 and 2 messenger ribonucleic acid (mRNA) levels.

Screening phase inclusion criteria

• Caucasian obese (BMI of ≥ 30 kg/m²) pregnant women between 12^+ 0 and 16^+ 0 weeks’ (+days) gestation.

• Aged ≥ 16 years.

• Signed informed consent form.

Screening phase exclusion criteria

• Non-Caucasian.

• BMI of < 30 kg/m².

• Gestation > 16 weeks.

• Pre-existing diabetes mellitus.

• GDM in a previous pregnancy.

• Systemic disease at the time of trial entry, with the disease either requiring regular medication or having required treatment with systemic steroids in the past 3 months.

• Previous delivery of a baby < 3rd centile by weight.

• Previous pregnancy complicated by pre-eclampsia prompting delivery before 32 weeks’ gestation.

• Known sensitivity to metformin hydrochloride or any of the known excipients.

• Acute condition at the time of trial entry with the potential to alter renal function, such as dehydration sufficient to require intravenous infusion, severe infection, shock and intravascular administration of contrast agents.

• Acute or chronic diseases that may cause tissue hypoxia such as cardiac or respiratory failure, recent myocardial infarction, hepatic insufficiency, acute alcohol intoxication and alcoholism.

• Lactation.

• Multiple pregnancy.

Randomisation exclusion criteria following screening

• GDM in index pregnancy [diagnosed with 75-g oral glucose tolerance test using World Health Organization (WHO) diagnostic criteria of fasting glucose ≥ 7.0 mmol/l, 2-hour glucose ≥ 7.8 mmol/l31]. Participants were also excluded if glucose tolerance testing was diagnostic of GDM based on the criteria used in the recruiting centre [e.g. International Association of the Diabetes and Pregnancy Study Groups (IADPSG) criteria32].

• Liver or renal dysfunction at the time of trial entry tested prior to randomisation (urea > 6.6 mmol/l, creatinine > 85 mmol/l, sodium > 145 mmol/l, potassium > 5.0 mmol/l, bilirubin > 16 µmol/l, alanine transferase > 60 IU/l) or with abnormal lactate levels (according to local laboratory reference range).

Ineligible and non-recruited participants

No further information was collected on women who were ineligible because of abnormalities in glucose tolerance or liver or renal function, other than the number of such women for trial metrics.

Telephone or face-to-face interviews were carried out to explore the reasons for eligible women declining to participate (see Chapter 5).

Dose changes

Local investigators or participants were allowed to alter the treatment regimen at their discretion as long as the maximum daily dose did not exceed 2500 mg. Changes to the treatment dose were recorded in the electronic case report form as soon as was practicable.

Other medications

Alcohol was prohibited because of the increased risk of lactic acidosis. Iodinated contrast agents may increase the risk of renal failure and, hence, if they were required treatment was discontinued for at least 48 hours from immediately prior to contrast administration until after renal function had been re-evaluated and found to be normal. Clinicians prescribing glucocorticoids, beta-2-adrenoreceptor agonists and angiotensin-converting enzyme inhibitors should have been aware that they might amplify or diminish the hypoglycaemic effect of metformin.

Other analytical methods

Primary outcome

The primary outcome was z-score corresponding to the gestational age-, parity- and sex-adjusted birthweight centile of the baby.

Secondary outcomes

• Maternal insulin resistance at 36 weeks’ gestation, which will be correlated with adverse pregnancy outcomes.

• Maternal anthropometry and body composition at 16 and 36 weeks’ gestation and 3 months post-partum.

• Baby anthropometry and body composition at birth and 3 months of age.

• Maternal inflammatory markers and lipid and fatty acid indices prior to commencing treatment and again at 28 and 36 weeks’ gestation, including CRP, IL-6, leptin, lipid profile, NEFAs, polyunsaturated fatty acids and PAI-1 : PAI-2 ratio.

• Neonatal CRP, glucose, insulin and other inflammatory and metabolic indices as previously described (measured in cord blood at birth).

• Incidence of low birthweight centile.

• Liquid chromatography-mass spectrometry measurement of metformin in maternal plasma to determine adherence.

Secondary outcomes from nested substudies

• Maternal salivary cortisol levels at baseline and 28 and 36 weeks’ gestation.

• Hepatic and peripheral insulin sensitivity at 36 weeks’ gestation as measured by the hyperinsulinaemic–euglycaemic clamp technique.

• Maternal brachial artery endothelium FMD measured at 16 and 36 weeks’ gestation.

• Maternal subcutaneous and visceral adipose tissue deposition and hepatic and skeletal muscle ectopic fat deposition assessed using magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS).

• Fetal liver volume and fetal subcutaneous fat deposition assessed using MRI.

• In vivo measurements of myometrial contractility on myometrial biopsies obtained at the time of caesarean section.

• Placental glucocorticoid receptor and 11β-HSD1 and 11β-HSD2 mRNA levels.

Maternal outcomes

There was no evidence of a reduction in the main secondary outcome of HOMA-IR at 36 weeks’ gestation (Table 6). Mean HOMA-IR in the placebo and metformin groups was 5.98 and 6.30 molar units, respectively (adjusted mean ratio 0.974, 95% CI 0.865 to 1.097). Similarly, there was no evidence of a statistically significant effect of metformin on fasting or 2-hour glucose (after a 75-g oral glucose challenge) or fasting insulin at 36 weeks’ gestation (see Table 6). However, fasting glucose and the HOMA-IR score at 28 weeks’ gestation were lower in the metformin group (adjusted mean difference/ratio –0.105 mmol/l, 95% CI –0.193 to –0.016 mmol/l; and 0.895 molar units, 95% CI 0.803 to 0.998 molar units, respectively) (data not shown).

Metformin had no effect on maternal weight gain during pregnancy (adjusted mean difference in weight gain –0.680 kg, 95% CI –1.863 to 0.503 kg) or maternal weight retention at 3 months post partum (Table 7). We subsequently performed an analysis of weight gain adjusted for baseline weight and again found no significant effect of metformin on weight gain (adjusted mean difference in weight gain –0.637 kg, 95% CI –1.819 to 0.544 kg). There were no differences in the other anthropometric measures of waist, hip, mid-arm and mid-thigh circumference or bicep, tricep and subscapular skinfold thickness (see Table 7).

Plasma IL-6 and CRP concentrations were both lower in the group treated with metformin at 36 weeks’ gestation (adjusted mean ratio 0.847 mmol/l, 95% CI 0.754 to 0.952 mmol/l; and 0.860 mg/l, 95% CI 0.743 to 0.996 mg/l, respectively). Cholesterol, HDL, LDL, triglycerides, leptin, NEFA and PAI1/2 ratio at 36 weeks’ gestation were similar in the two groups. There was a trend towards higher serum cortisol at 36 weeks’ gestation in the metformin group on ITT analysis and this reached statistical significance in the per-protocol analysis (adjusted mean ratio 0.088, 95% CI 0.010 to 0.167; p = 0.0281) (see Table 6).

Metformin did not appear to prevent GDM. The proportion of women fulfilling either the IADPSG or the WHO criteria for GDM at any time in pregnancy was similar in the two groups. Post hoc analysis of the timing of diagnosis of GDM (IADPSG criteria) showed no statistically significant difference between the two groups: in the placebo group 26 women were diagnosed at 28 weeks’ gestation and 10 at 36 weeks’ gestation, whereas in the metformin group 11 women were diagnosed at 28 weeks’ gestation and 15 at 36 weeks’ gestation (p = 0.0718, Mantel–Haenszel chi-square, post hoc analysis); however, the trend was towards a later diagnosis in the metformin-treated group.

There were no differences in outcomes at other time points between the two groups, with the exception of fasting glucose and HOMA-IR score, as mentioned previously.

Further analysis of the data on a per-protocol basis resulted in similar findings with a few exceptions. For CRP at 36 weeks’ gestation and vomiting, the direction of difference was maintained but statistical significance was lost. Two-hour glucose and fasting insulin at 28 weeks’ gestation were lower in the metformin group (estimated mean difference –0.312 mmol/l, 95% CI –0.620 to –0.004 mmol/l; p = 0.0471; and 0.871 µIU/ml, 95% CI 0.778 to 0.976 µIU/ml; p = 0.0173, respectively). Serum cortisol at 36 weeks’ gestation was higher in the metformin-treated group (estimated mean difference 0.088 nmol/l, 95% CI 0.010 to 0.167 nmol/l; p = 0.0281) (summarised in Table 7).

At 36 weeks’ gestation there were no differences in serum B₁₂ or folate levels between the two groups on ITT analysis (adjusted mean difference: B₁₂ 0.952 ng/l, 95% CI 0.879 to 1.032 ng/l; p = 0.2296; folate 1.050 µg/l, 95% CI 0.885 to 1.247 µg/l; p = 0.5737). However, on per-protocol analysis, participants in the metformin group had a lower serum B₁₂ concentration at 36 weeks’ gestation (adjusted mean difference 0.890 ng/l, 95% CI –0.804 to 0.985 ng/l; p = 0.0248). There were no differences between the groups in the proportion of participants with a serum B₁₂ or folate concentration < 5th centile (Table 8).

Neonatal outcomes

The proportion of live-born babies weighing > 90th centile was similar in the two groups [placebo 38/220 (17%), metformin 31/214 (14%)]. Importantly, we also did not see a difference in the proportion of babies weighing < 10th centile [placebo 11/220 (5%), metformin 14/214 (7%)]. There was no significant difference in neonatal ponderal index at birth between the two groups (adjusted mean ratio 1.032 g/cm³, 95% CI 0.996 to 1.069 g/cm³).

Neonatal cord blood glucose, insulin, HOMA-IR and CRP were similar in the two groups (see Table 6).

Introduction

Body mass index is routinely used as a proxy for adiposity. However, a limitation of BMI is that it makes no distinction between fat mass and FFM and indeed other components of total body weight. BMI has been shown to be well correlated with adiposity in pregnant women, although with a wide CI34 and a weakening of correlation closer to term. 35 As fat mass is particularly relevant in the context of insulin sensitivity, we aimed to examine the effect of metformin on fat mass specifically, as well as on overall gestational weight change.

The effect of maternal metformin on the body composition of the infant was also of interest in this study. Body composition at birth, which includes fat mass and FFM as opposed to birthweight alone, is probably a more important predictor for long-term risk of disease such as obesity and metabolic syndrome. Whereas FFM is generally a reflection of genetic effects, fat mass is more variable and susceptible to factors that affect fetal growth such as maternal weight gain in pregnancy, maternal obesity and insulin sensitivity. 10,36,37

There are numerous ways to assess body composition, including underwater weighing, dual-energy X-ray absorptiometry and bioelectrical impedence analysis. These all have limitations, particularly in our two subject groups: pregnant women and infants. Air displacement plethysmography (ADP) is increasingly recognised as the gold standard tool to best assess body composition, particularly in these challenging groups. 38 ADP is based on the principle of Boyle’s law, which states that air compressed will decrease in volume proportional to increasing pressure at a constant temperature. The subject is placed inside a sealed chamber of known volume and any change in pressure and volume is attributable to the volume of the subject. The technique is quick and generally acceptable to all, excepting those with severe claustrophobia.

Methods

Maternal fat mass was measured using ADP at baseline, 36 weeks’ gestation and 3 months post-partum. Infant fat mass was measured using the same technique within 72 hours of birth and at 3 months of age.

Equipment was used in a temperature-controlled room (21–27 °C) and calibrated at the start of each day of use.

Adult ADP tests were performed with the BOD POD [see www.lifemeasurement.com (accessed 21 June 2016)]. Subjects were fasted and had abstained from exercise for at least 2 hours prior to the test. They were asked to empty their bladder and remove all jewellery and glasses. The test was performed with the subject wearing minimal skintight clothing, for example swimwear, and a hair cap. Subjects were in a resting state.

Infant ADP tests were performed with the PEA POD (see www.lifemeasurement.com). The infant’s clothes and nappy were removed and a head cap applied or hair smoothed. The instrument was calibrated to account for the mass of two hospital identification bracelets and an umbilical cord clip for measurements in the newborn babies.

Results

Edinburgh participants only were invited to participate in this substudy. The final cohort numbers following unblinding are provided in Table 10.

There were no significant differences in maternal fat mass at baseline, 36 weeks’ gestation and 3 months post partum. There was no significant difference in percentage change in maternal fat mass between early and late pregnancy and 3 months post-partum between the placebo group and the metformin group.

There was no significant difference in percentage change in neonatal fat mass from birth to 3 months of age between the placebo group and the metformin group.

Discussion

Metformin taken during pregnancy does not appear to have had an effect on maternal total body fat percentage in late pregnancy or at 3 months post-partum. In the study population as a whole, we did not see any differences in total body weight or skinfold thickness, which would suggest that metformin has not had a significant effect on the distribution of body fat, although the sample size was small. This is supported by our MRI data (see Magnetic resonance imaging assessment of maternal and fetal adipose distribution) in which we did not see any significant differences in body fat distribution between the two groups in a smaller cohort of subjects.

There was no significant difference in body fat percentage in the babies, either at birth or at 3 months of age. This is similar to the findings of the metformin compared with insulin for the treatment of gestational diabetes trial,39 in which there were no differences at birth in birthweight, upper arm circumference, tricep and subscapular skinfold thickness and ponderal index between the babies exposed to metformin and those whose mothers received insulin. However, when these children were assessed again at age 2 years, those who were exposed to metformin had similar total body fat (assessed by dual-energy X-ray absorptiometry) but larger skinfold thickness,40 suggesting that they had less central fat and may therefore be more insulin sensitive in the longer term. This highlights the need for follow-up studies in our cohort to assess any longer-term effects that metformin exposure may have had on fat distribution in these children.

Introduction

Normal pregnancy is associated with marked changes in insulin sensitivity, glucose homeostasis and lipid and protein metabolism. In early pregnancy, fasting glucose decreases by 0.11 mmol/l with little further decrease by the end of pregnancy. 41 Insulin secretion increases but insulin sensitivity remains unchanged. 42,43 This promotes lipogenesis to prepare for the rising energy needs of pregnancy and also to allow lipid storage in preparation for the energy demands of lactation. Glucose tolerance is normal at this stage, as is peripheral insulin sensitivity and hepatic basal glucose production. 43–45 By mid-pregnancy, despite the increase in insulin secretion, basal hepatic glucose production also increases, as does total gluconeogenesis, to meet the increasing demands of the fetoplacental unit. 46–48 By late gestation, peripheral insulin sensitivity is markedly decreased such that insulin is unable to suppress lipolysis, allowing an increase in free fatty acids and therefore more energy available for gluconeogenesis. 49 Overall, the insulin sensitivity of late pregnancy is reduced by 50–70% compared with the non-pregnant state. These mechanisms are important to ensure a ready supply of energy substrates for the developing fetus.

In obese pregnant individuals, these mechanisms are disordered. Obesity is associated with a state of diminished insulin sensitivity and so obese women enter pregnancy already resistant to insulin. The reduction in fasting glucose in very early pregnancy is diminished or absent. 41 By late gestation, a physiological reduction in peripheral insulin sensitivity by 15% has been demonstrated. 50 In addition, there is marked hepatic insulin resistance with reduced insulin-mediated glucose disposal and a reduction in insulin-stimulated suppression of endogenous glucose production (EGP). 51 Thus, there may be an excess of free fatty acids and glucose, which are freely transferred across the placenta and may potentially drive fetal overgrowth and programming of later-life insulin resistance. However, our own work52 has demonstrated differences between lean and severely obese pregnant women only at early and mid-gestation, with a convergence in degree of insulin resistance by late pregnancy.

This mechanistic substudy was designed to examine the effect of metformin on insulin resistance in obese pregnancy. We used the gold standard method of assessing insulin sensitivity, the hyperinsulinaemic–euglycaemic clamp, to assess this in a subgroup of women participating in the trial and who were adherent to treatment. To our knowledge, this is the first study to have employed this technique to examine the effect of metformin in pregnant women. The characteristics of the women were similar to those of the women in the study overall. Importantly, those with GDM were excluded.

Methods

Solutions of 6,6–²H₂-glucose (d₂-glucose) and 1,1,2,3,3–²H₅-glycerol (d₅-glycerol) (Cambridge Isotope Laboratories, Inc., Andover, MA, USA) and soluble insulin (Actrapid^®, NovoNordisk, Bagsvaerd, Denmark) were prepared in 0.9% saline.

Participants (n = 21) attended the clinical research facility at 08.00 after fasting overnight for 8–10 hours. A 44-mm 20-gauge cannula was inserted into the superficial vein in the dorsum of one hand and kept patent with a slow infusion of 0.9% saline. This hand was wrapped in an electric heated blanket to arterialise the venous blood for sample collection. A second cannula was placed in the antecubital fossa vein of the contralateral arm for the infusates. We infused d₂-glucose (prime 25 µmol/kg then continuous infusion of 22 µmol/kg/hour) and d₅-glycerol (prime 1.6 µmol/kg then continuous infusion of 6.6 µmol/kg/hour) for 5.5 hours. Four steady-state blood samples were taken at 10-minute intervals at the end of three time periods: (1) 60, 70, 80 and 90 minutes; (2) 180, 190, 200 and 210 minutes; and (3) 300, 310, 320 and 330 minutes with infusion of (1) tracers only (no insulin); (2) 20 mU/m²/minute of insulin (to suppress lipolysis and EGP); and (3) 40 mU/m²/minute of insulin (to stimulate glucose uptake). Following commencement of the insulin infusion at 90 minutes, blood samples were obtained every 5 minutes from the sampling cannula for measurement of whole blood glucose concentration using an Accu-Chek^® blood glucose monitor (Roche Products Ltd, Welwyn Garden City, UK). A solution of 20% dextrose was infused as required to maintain arteriolised blood glucose between 4.5 and 5.5 mmol/l. Additional blood samples were obtained every 30 minutes from 90 minutes onwards using fluoride oxalate anticoagulant for formal enzymatic measurement of plasma glucose. The steady-state blood samples for analysis of the tracers and those for insulin and NEFAs were collected on ice, the plasma separated by centrifugation and plasma aliquots stored at –80 °C.

The volume of dextrose infused during the final 30 minutes of the high-dose hyperinsulinaemic–euglycaemic clamp divided by the corresponding insulin concentration at steady state was used to derive the glucose disposal per unit plasma insulin or M/I.

Steele’s equation for steady state was applied to calculate the rate of appearance (Ra) or rate of disappearance (Rd) of the tracee (d₂-glucose or d₅-glycerol):

where F is the infusion rate of the tracer and TTR is the tracer-to-tracee ratio.

Endogenous glucose production was calculated by subtracting the variable glucose infusion rate from the calculated Ra glucose. Data from glucose infusion studies were corrected for background ¹³C enrichment. No exogenous unlabelled glycerol was infused and the abundance of other isotopic species within the tracer infusion is negligible; therefore, corrections were not applied in the calculation of Ra glycerol.

Results

Effect of metformin on indices of insulin sensitivity for glucose metabolism at 36 weeks’ gestation

Mean plasma glucose and insulin concentrations achieved for the two groups are shown in Figures 3 and 4.

FIGURE 3.

Clamped plasma glucose.

FIGURE 4.

Clamped plasma insulin.

Whole-body glucose disposal (WGD) was calculated in mg of glucose per kg of FFM per minute at steady state during the high-dose clamp. WGD is an indirect measure of whole-body insulin sensitivity, with a greater glucose disposal rate implying greater insulin sensitivity. The M/I was also calculated to correct for slight differences in achieved plasma insulin in each group and expressed in units of mg of glucose per kgFFM per minute.

Glucose disposal per unit plasma insulin but not WGD was higher in the metformin group than in the placebo group (difference between means 0.02 mg/kgFFM/minute, 95% CI 0.001 to 0.03 mg/kgFFM/minute; p = 0.04; and 0.78 mg/kgFFM/minute/µIU/l, 95% CI –0.12 to 1.67 mg/kgFFM/minute/µIU/l; p = 0.08, respectively).

In contrast, HOMA-IR scores for these individuals at baseline were similar in the two groups (difference between means –0.87, 95% CI –3.31 to 1.57; p = 0.46).

In the absence of insulin, EGP (expressed as both milligrams of glucose per kilogram of total body weight per minute and as milligrams of glucose per kilogram of FFM per minute) was greater in the metformin group than in the placebo group (difference between means 0.54 mg/kgFFM/minute, 95% CI 0.08 to 1.00 mg/kgFFM/minute; p = 0.02) (Figure 5). During low-dose insulin infusion, EGP was again higher in the metformin group (difference between means 0.43 mg/kgFFM/minute, 95% CI 0.02 to 0.84 mg/kgFFM/minute; p = 0.04) (see Figure 5). There was no significant difference in the percentage suppression from basal EGP to EGP during low-dose hyperinsulinaemic–euglycaemic clamp between the two groups (difference between means 2.89%, 95% CI –7.65% to 13.42%; p = 0.57) (see Figure 5).

FIGURE 5.

Endogenous glucose production. *p ≤ 0.05.

At baseline, prior to insulin infusion, the Rd of glucose and EGP should be equivalent as they are in steady state. The Rd was also significantly greater in the metformin group than the placebo group during the low-dose insulin phase of the clamp (difference between means 0.36 mg/kgFFM/minute, 95% CI 0.22 to 1.18 mg/kgFFM/minute; p = 0.07). At high-dose insulin infusion, Rd was increased further but there was no significant difference between the treatment groups (difference between means 0.35 mg/kgFFM/minute, 95% CI –0.79 to 1.50 mg/kgFFM/minute; p = 0.52) (Figure 6).

FIGURE 6.

Rate of disappearance of glucose. *p ≤ 0.05; **p ≤ 0.01.

Effect of metformin on insulin sensitivity for lipolysis at 36 weeks’ gestation

Mean plasma glycerol and NEFA concentrations achieved for the two groups are shown in Figures 7 and 8. Glycerol turnover per kgFFM is shown in Figure 9. There was no difference in glycerol turnover between the metformin and placebo groups (difference between means: no insulin 0.03 mg/kgFFM/minute, 95% CI –0.12 to 0.18 mg/kgFFM/minute; p = 0.67; low-dose insulin 0.02 mg/kgFFM/minute, 95% CI –0.06 to 0.10 mg/kgFFM/minute; p = 0.64; high-dose insulin –0.01 mg/kgFFM/minute, 95% CI –0.10 to 0.08 mg/kgFFM/minute; p = 0.87). Low-dose insulin infusion resulted in suppression of the Rd of glycerol in both groups; high-dose insulin resulted in no further suppression of Rd in either group. There was no difference in serum NEFAs between the groups at any time.

FIGURE 7.

Clamped plasma glycerol.

FIGURE 8.

Clamped plasma NEFAs.

FIGURE 9.

Glycerol turnover.

Discussion

We hypothesised that administration of metformin to obese pregnant women would improve insulin sensitivity in the third trimester. These data show that, as expected, subjects taking metformin demonstrated a greater M/I, which is associated with increased insulin sensitivity. The Rd of glucose was enhanced in the metformin-treated group suggesting improved peripheral insulin sensitivity. However, EGP was higher in the metformin-treated subjects, suggesting that, if anything, those on metformin exhibited a reduced ability to suppress hepatic glucose production in response to insulin and enhanced glucose release on fasting. This is perhaps surprising given that metformin is thought to exert its action principally via the liver, inhibiting gluconeogenesis and reducing hepatic glucose production,54 although an effect mediated via peripheral glucose disposal has also been shown. 55,56 Additionally, metformin has recently been shown to stimulate EGP in healthy fasting individuals who are not pregnant. 57 Metformin’s mechanism of action in pregnant women has, to our knowledge, never been studied. These data suggest enhanced glucose flux in the metformin-treated women (i.e. higher liver production and higher peripheral disposal of glucose). Interpretation of this is perhaps complicated by the presence of the fetoplacental unit, which represents a significant proportion of the FFM at 36 weeks’ gestation and must account for uptake of some of the glucose. These data confirm that the hyperinsulinaemic–euglycaemic clamp is a more sensitive measure of insulin resistance as we demonstrated differences in M/I and EGP but not in HOMA-IR score.

The lipolytic pathway was equally sensitive to exogenous insulin in both the metformin- and the placebo-treated groups. Maximal suppression (around 70%) was achieved by low-dose insulin, with no further suppression at the high-dose infusion. Enhanced rates of lipolysis are thought to be important towards the end of normal pregnancy to provide maternal substrates for gluconeogenesis and triglyceride synthesis and spare glucose to facilitate normal fetal growth. Reduced third-trimester lipolysis is associated with fetal growth restriction58 and so these data are perhaps reassuring on the safety of metformin in pregnancy in terms of not increasing the risk of intrauterine growth restriction.

In conclusion, these substudy data confirm that metformin can improve insulin sensitivity in obese pregnant women. However, this may be offset by increased glucose flux and hence there is a lack of effect on fetal nutrition or growth. Additionally, metabolism at 36 weeks’ gestation may not reflect metabolism in mid-gestation, when differences between lean and obese women are greater.

Introduction

Hypertensive disorders are more prevalent in obese populations59 including those who are pregnant. 60 Obese women are at greater risk of chronic hypertension, pregnancy-induced hypertension and pre-eclampsia:61–64 for a woman with a BMI of > 35 kg/m², the risk of pre-eclampsia is twice that of a normal lean woman. 65 The causal mechanisms behind these associations are not clear but insulin resistance may play a role. Women with diabetes mellitus of all types during pregnancy have a heightened risk of pre-eclampsia and women with pre-eclampsia have an increased risk of type 2 diabetes mellitus in later life. 66 Other possible contributory mechanisms include endothelial dysfunction, inflammation, dyslipidaemia and oxidative stress, all of which are associated with both obesity and pre-eclampsia. 67–69

The vascular endothelium is the layer of endothelial cells between the blood vessel wall and the bloodstream. It is a key regulator of vascular homeostasis, acting not only as a barrier but also as an active signal transducer for circulating influences that modify the vessel wall tone and phenotype. 70 Endothelial dysfunction is characterised by a shift of the actions of the endothelium towards reduced vasodilatation and a more pro-inflammatory and pro-thrombotic state. 71 Mechanisms that participate in the reduced vasodilatory responses include reduced nitric oxide bioavailability and oxidative stress. 71 Endothelial dysfunction is associated with most forms of cardiovascular disease, such as hypertension, coronary artery disease, diabetes and chronic renal failure, and often precedes their clinical manifestations. 72 It has also been demonstrated in disease states in the absence of overt cardiovascular complications, such as the metabolic syndrome73 and obesity. 74

Endothelial dysfunction in pregnancy has been most widely studied in the context of pre-eclampsia, in which impaired maternal vascular function has been reported. 75 GDM is associated with increased oxidative stress and overexpression of inflammatory cytokines, both of which contribute to endothelial dysfunction. 76 Obesity in pregnancy shares these features and impaired endothelial function has been demonstrated in obese pregnant women. 67,77 Given that insulin resistance and hyperglycaemia are linked to inflammation, vascular dysfunction and hypertension, these processes are potential mediators for these upstream causative pathways, linking endothelial dysfunction with cardiovascular disease in obese pregnant women.

Metformin might be the ideal agent to address and reverse these abnormalities in obese pregnant women. In addition to its primary function as an insulin-sensitising agent, metformin has beneficial effects on the vascular endothelium, lipid profile and oxidative stress. 78,79 These effects appear to be independent of metformin’s glucose-lowering and insulin-sensitising effects. Metformin also improves vascular function in a variety of clinical syndromes associated with insulin resistance, for example reducing cardiovascular disease risk in patients with type 2 diabetes mellitus80 and improving endothelial function in patients with type 1 diabetes mellitus, metabolic syndrome and PCOS. 79,81–83

Nested within the EMPOWaR trial, this mechanistic substudy was designed to test the hypothesis that, compared with those taking placebo, participants treated with metformin would demonstrate improved endothelial function in late pregnancy.

Methods

We recruited a subset of women participating in the EMPOWaR trial. The characteristics of participants in the substudy were similar to the characteristics of those in the EMPOWaR trial overall. Importantly, those with GDM were excluded. We also included a comparator ‘control’ group of lean pregnant subjects who, with the exception of BMI, were matched for baseline characteristics to the EMPOWaR study population. Participants of the EMPOWaR trial were assessed at 12–16 weeks’ gestation following randomisation but prior to commencing study treatment and at around 36 weeks’ gestation while receiving trial medication. All measurements were performed blind to treatment allocation. Lean control subjects were assessed at both 12–16 and 36 weeks’ gestation.

Endothelial function

We assessed endothelial function by measuring FMD in the brachial artery. Subjects were asked to refrain from eating, smoking or consuming alcohol or caffeine in the preceding 4 hours. Measurements were obtained in a temperature-controlled room with the subject resting in a semi-recumbent position on a bed. Subjects in late pregnancy had a left lateral tilt applied to avoid aortocaval compression.

Measurements were made using ultrasound imaging (CX50 Ultrasound system with a 7-MHz linear array transducer; Philips Medical Systems, Guildford, UK) of the brachial artery, 2–5 cm above the antecubital fossa. A baseline rest image was acquired for a period of 60 seconds. Arterial occlusion was performed using a sphygmomanometric cuff applied below the antecubital fossa, inflated to suprasystolic pressure for 5 minutes and then released to induce hyperaemia. The brachial artery ultrasound was recorded for 30 seconds before and 5 minutes after cuff deflation. Images were acquired with electrocardiogram gating, with measurements made in end-diastole, corresponding to the onset of the R wave. To minimise movement, the scan probe was held in place with a probe-retaining device throughout the period of the study. Images were stored digitally and measurements made using edge-detection software [Vascular Research Tools 5; Medical Imaging Applications LLC, see www.mia-llc.com (accessed 24 June 2016)] (Figure 10).

FIGURE 10.

Representative images from analysis software.

The results were expressed as change in arterial diameter (D) divided by baseline diameter:

⁽²⁾ F M D ( % ) = ( D peak − D baseline D baseline ) × 100 .

The peak diameter (D_peak) was taken as the mean of the diameter measurements taken 5 seconds either side of the peak diameter. The baseline diameter (D_baseline) was taken as the mean of the 60 seconds of baseline recording. We then measured endothelium-independent vasodilatation. After 10 minutes of rest, a second baseline image was obtained for 60 seconds and then a single low dose (25 µg) of sublingual nitroglycerin (GTN) spray was given and measurement recorded for 5 minutes. Again, the baseline diameter was taken as the mean of the first 60 seconds and the peak diameter was taken as the mean of the measurements 5 seconds either side of the peak.

Results

Forty-one eligible women in the EMPOWaR study agreed to participate in the substudy. However, the majority of EMPOWaR participants were unable or unwilling to attend for the two study visits: only one woman in the placebo group and two in the metformin group attended both visits. In contrast, all women in the lean control group attended both visits except for one lean subject who was ineligible at 36 weeks’ gestation because of a preterm birth. Images from nine subjects had to be discarded as they were of insufficient quality for analysis. The final study population (Table 12), therefore, included 28 subjects (n = 6 placebo, n = 12 metformin and n = 10 lean) at 12–16 weeks’ gestation and 26 subjects (n = 8 placebo, n = 9 metformin and n = 9 lean) at 36 weeks’ gestation. Data are presented for 28 subjects at baseline and 25 subjects at 36 weeks’ gestation.

There were no differences in FMD between the placebo, metformin and lean groups at baseline or at 36 weeks’ gestation (one-way ANOVA: baseline, p = 0.88; 36 weeks, p = 0.89) (Figure 11; see also Table 12). There was a decline in endothelial function in late pregnancy compared with early pregnancy across all groups (two-way ANOVA p = 0.03). There were no differences in endothelium-independent vasodilatation between groups or within groups in early and late pregnancy (see Table 12 and Figure 11).

FIGURE 11.

(a) FMD and (b) GTN-mediated dilatation. *p ≤ 0.05.

Discussion

We have used flow-mediated vasodilatation, the gold-standard non-invasive method of assessing endothelial function, in obese and lean pregnant women as part of the EMPOWaR study. Although we demonstrated a selective gestation-associated decline in endothelium-dependent vasodilatation, we saw no differences in endothelial function across obese and lean groups, nor any effect of the treatment intervention metformin. This suggests that pregnancy is associated with altered endothelial function and this is independent of body weight or glucose sensitivity.

There was a decline in endothelial function between early and late pregnancy in participants in all three groups. This is contrary to the finding of one longitudinal study of endothelial function in eight normal-weight women that demonstrated increasing FMD across the three trimesters of pregnancy. 84 However, a much larger study of 157 normal-weight pregnant women demonstrated an increase in FMD in pregnant compared with non-pregnant subjects, apparent from 10 weeks’ gestation, but a progressive decline in FMD to pre-pregnancy levels from around 30 weeks’ gestation, in keeping with our data in both lean and obese participants. 85

To our knowledge, this is the first study of the effect of metformin on endothelial function in pregnant women. It is not clear why metformin does not appear to have had an effect on the vascular endothelium in the study population when this has been demonstrated in other insulin-resistant populations. Pregnancy is associated with major vascular and haemodynamic changes, the primary event probably being a fall in peripheral vascular resistance. 86 This is most likely mediated by endothelium-dependent factors including nitric oxide synthesis upregulation by oestradiol and possibly vasodilatory prostaglandins. 87–89 The consequent fall in systemic vascular resistance leads to a compensatory increase in cardiac output. Larger vessels dilate less than smaller ones. 90 It is possible that the increased vessel diameter of pregnancy, along with stimulated nitric oxide activity of pregnancy, may obscure any effect of metformin on the endothelium. However, we did not see any differences in endothelium-independent vasodilatation, which we would expect to see if this was purely a vessel size effect or related to increased nitric oxide consumption.

Flow-mediated dilatation of the brachial artery is accepted as the gold standard for the non-invasive measurement of endothelial function as it is widely used, well tolerated, low risk and, importantly for our study population, suitable for use in pregnant women. There are, however, some limitations. Most notably for our study population, measurement of the vessel diameter can be technically challenging, particularly in the obese in whom visualisation of the intima is difficult as the ultrasound signal is attenuated by subcutaneous fat. Other limitations include the small sample size, the lack of longitudinal data for subjects other than those in the lean group and the variability of the baseline demographics (e.g. smoking status and parity) within the unblinded groups. This makes it difficult to identify differences within the small sample size.

In the larger EMPOWaR cohort as a whole, we saw no differences in blood pressure or the incidence of hypertensive disorders of pregnancy, which we may have expected if the placebo group had impaired endothelial function compared with the metformin group. This suggests that our finding of no difference between the two groups may be correct rather than a type 2 error. In conclusion, we have not demonstrated any effect on endothelial function of metformin in obese pregnant women.

Introduction

For many years now it has been recognised that the site rather than the total quantity of body fat is an important determinant of insulin sensitivity and morbidity associated with obesity. 91–94 More recently, it has been recognised that deposition of lipid in ‘ectopic’ sites, namely the liver and skeletal muscle, is a major contributor to the development of insulin resistance. 95,96 It remains unclear whether or not increasing adiposity causes the deterioration in insulin sensitivity or vice versa. The ‘portal hypothesis’ of obesity suggests that an increase in central abdominal fat leads to elevated delivery of free fatty acids and inflammatory cytokines to the liver and that, consequently, hepatic insulin resistance develops and drives glucose upwards. 97 The ‘spillover hypothesis’ suggests that, in the context of obesity, the subcutaneous compartment becomes saturated and leads to the accumulation of visceral fat and deposition of lipid in ectopic sites such as the liver and muscle. 98 Clearly, the two hypotheses are not mutually exclusive and it is likely that both are contributory mechanisms. Regardless of the exact cause, there is no doubt that excess lipid accumulation is associated with impaired insulin sensitivity and morbidity such as type 2 diabetes mellitus.

Fat distribution in normal and obese pregnancy is less well studied and the contribution it pays to maternal and fetal outcomes and longer-term health is not clear. However, gestational weight gain is one of the most important predictors of post-partum weight retention99 and thus contributes significantly to the obesity epidemic among young women. 100 Gestational weight gain is highly variable but in lean women the contribution from fat tends to be predominantly in the subcutaneous compartments, largely the trunk and thighs. 101,102 Obese women actually tend to gain less weight than lean women during pregnancy but the fat mass that they do gain tends to be more central and therefore potentially more metabolically harmful. 103

Gestational weight gain and pre-existing obesity also impact on fetal growth. The conventional strategy for measurement of fetal growth is typically by ultrasound. Estimation of fetal weight is based on measurement of the fetal head circumference, femur length and abdominal circumference, with the abdominal circumference being the most individually sensitive predictor of fetal macrosomia. 104,105 The abdominal circumference is predominantly affected by the size of the fetal liver and is positively correlated with hepatic glycogen stores, which increase towards term. 106,107

Glucose is the major substrate that determines fat accumulation in the fetus, with the greater the glucose supply the greater the deposition of fat. 108 There is a linear association between maternal glucose tolerance and neonatal adiposity at birth,109 and a strong positive correlation between degree of maternal insulin resistance and neonatal fat mass at birth. 10 Increasing maternal BMI has also been shown to be associated with increased intrahepatocellular lipid in the newborn, assessed by magnetic resonance proton spectroscopy. 110

Magnetic resonance imaging has been used for many years in pregnant women with no apparent adverse effects on the mother or developing fetus. 111,112 It enables the quantification of the different maternal fat depots, in contrast to other techniques that can measure only total body fat mass (e.g. ADP; see Maternal and neonatal body composition). This substudy also provided the opportunity to develop scanning protocols to assess the body composition of the fetus in utero.

The aim of this mechanistic substudy was to examine the effect of metformin on maternal and fetal fat deposition in the early and late third trimester. We hypothesised that improving insulin sensitivity with metformin would result in less deposition of fat in the more insulin-sensitive sites (i.e. visceral, hepatic and skeletal muscle) and potentially protect the fetus from accumulation of excess fat.

Methods

Participants in the EMPOWaR study were invited to undergo a MRI scan at 28 and 36 weeks’ gestation. We aimed to recruit 40 participants to have a scan at both time points. Whole-body MRI and ¹H-MRS studies were performed on a Siemens MAGNETOM^® Verio 3 Tesla MRI system (Siemens AG, Healthcare Sector, Erlangen, Germany). Participants were positioned in the magnet in a full left lateral position to avoid aortocaval compression. Data from the abdomen and thigh were acquired using a combination of spine and body matrix coils elements. Aural protection was provided by use of earplugs and headphones. Contact between the participant and scanning staff was maintained at all times. Heart rate and oxygen saturation were monitored continuously throughout the scan period; blood pressure was measured at the start of the scan and every 10 minutes throughout the procedure.

Scan sequences

Standard localising images were acquired to confirm organ and fetal position.

¹H-magnetic resonance spectroscopy

For ¹H-MRS measurement of intramyocellular and intrahepatocellular lipid, single-voxel spectra localised to the right quadriceps muscle and the right lobe of the liver were acquired using a point-resolved spectroscopy sequence (repetition time 5000 milliseconds/echo time 30 milliseconds) with and without water suppression and with eight signal averages. The voxel size was 2 cm³ in the muscle and 3 cm³ in the liver. The voxel site was chosen to avoid large blood vessels or subcutaneous adipose tissue. Lipid concentration was calculated from the water-suppressed acquisition using the spectroscopy analysis tool jMRUI (MRUI Consortium, Brno, Czech Republic).

In- and out-of-phase imaging (Dixon method)

In addition to MRS, the Dixon method113–115 was used for calculation of hepatic and skeletal muscle fat fraction in the participant and also fetal hepatic fraction. The lipid signal was calculated by subtraction of in- and out-of-phase images (2.46 milliseconds and 8.61 milliseconds) and T2* decay during this time, corrected using the two in-phase images (2.46 milliseconds and 4.92 milliseconds) according to protocols that are well established for use in the adult liver. 113–115 These protocols are not well established in fetal imaging and method development was required during the study to optimise this technique for the fetal measurements.

Assessment of maternal abdominal adipose tissue

To quantify maternal subcutaneous and visceral fat, a three-dimensional T1-weighted volumetric interpolated breath-hold examination sequence was acquired axially through the liver. Lipid signals were defined using a semiquantitative thresholding technique using the commercial software SliceOmatic™ (TomoVision, Magog, QC, Canada).

Adipose tissue appears bright on T1-weighted images. Regions of interest with attenuation above an investigator-defined threshold were coloured to define visceral adipose tissue and subcutaneous adipose tissue (Figure 12). The areas of the coloured regions were extracted, converted into a unit of volume (mm³) and expressed as a percentage of the total abdominal volume of the region being examined. Volumes were calculated using the extracted areas of the regions of interest (mm²) multiplied by the width of each slice (2 mm).

FIGURE 12.

Maternal adipose tissue images. (a) Uncoloured and (b) coloured axial slices to show subcutaneous (green) and visceral (red) fat distribution.

A multislice approach was used. The left renal pelvis was identified in all subjects and adipose tissue was measured in 20 2-mm slices cranial to this level.

Estimation of fetal liver volume

A T2 half-Fourier acquisition single-shot turbo spin-echo sequence was acquired of the fetus to cover the fetal liver in the axial, sagittal and coronal planes (dependent on degree of fetal movement during the acquisition period).

The fetal liver is identifiable by the investigator on these images using standard anatomical landmarks. The entire liver was coloured as the area of interest on every slice in which it was visible using the same software as was used for the maternal fat measurements (Figure 13). The area of the coloured region was extracted and converted into a unit of volume (mm³) by multiplying the area (mm²) by the width of each slice (2 mm).

FIGURE 13.

Fetal liver images. Uncoloured (a and c) and coloured (b and d) axial (a and b) and sagittal (c and d) slices to show the fetal liver.

Estimation of fetal hepatic fat and fetal subcutaneous fat

A T1-weighted fast low-angle shot (FLASH) sequence was acquired for the in- and out-of-phase fetal liver fat fraction. The slice thickness was 8 mm. For fetal subcutaneous fat, a fat excitation FLASH sequence was used, again with an 8-mm slice width. Shoulder-to-shoulder coverage of the fetus was obtained in the sagittal plane and a single slice at the level of the umbilical cord insertion was used in the axial plane. The subcutaneous fat was coloured on every available slice in the sagittal plane (Figure 14) and on the single slice in the axial plane using the same technique as described for the maternal subcutaneous and visceral fat and fetal liver volume. The amount of fat was expressed as a percentage of the total volume of the area examined. These protocols were subject to method development during the study process.

FIGURE 14.

Fetal subcutaneous fat images. (a) Uncoloured and (b) coloured sagittal slices through the fetus to show the subcutaneous fat.

Results

The demographic characteristics of the participants are provided in Table 13. In total, 37 participants (n = 18 and n = 19 in the placebo and metformin groups, respectively) underwent MRI and ¹H-MRS studies at both 28 and 36 weeks’ gestation. A further 10 participants (n = 6 and n = 4 in the placebo and metformin groups, respectively) underwent MRI at 28 weeks’ gestation only and 10 participants (n = 6 and n = 4 in the placebo and metformin groups, respectively) underwent MRI at 36 weeks’ gestation only.

Reproducibility

Fetal liver volume

Following repeated measurements by the same investigator (M1 and M2; intrarater) and of the same images by two investigators (CC and SS; inter-rater), measurements of fetal liver volume were found to be well correlated in both the sagittal (Figures 15–18) and axial (Figures 19–22) planes, with the majority of points scattered evenly around the line of no difference and within the upper and lower 95% limits of agreement. Measurement in the axial plane showed the best reproducibility. This may be because there is less rotation of the fetus on this axis and more reliable images were obtained.

FIGURE 15.

Sagittal fetal liver volume intrarater reproducibility. (a) Observer 1; and (b) observer 2. M1, measurement 1; M2, measurement 2, in reference to repeated measures by the same observer.

FIGURE 16.

Sagittal fetal liver volume intrarater reproducibility: Bland–Altman analysis. (a) Observer 1. n = 93 pairs, mean difference (lower to upper 95% limits of agreement) –1015 (–3228 to 1198) mm³; and (b) observer 2. n = 99 pairs, mean difference (lower to upper 95% limits of agreement) –495 (–3003 to 2014) mm³.

FIGURE 17.

Sagittal fetal liver volume inter-rater reproducibility. CC, observer 1; SS, observer 2.

FIGURE 18.

Sagittal fetal liver volume inter-rater reproducibility: Bland–Altman analysis. n = 96 pairs, mean difference (lower to upper 95% limits of agreement) –1110 (4284 to 2065) mm³.

FIGURE 19.

Axial fetal liver volume intrarater reproducibility. (a) Observer 1; and (b) observer 2. M1, measurement 1; M2 measurement 2, in reference to repeated measures by the same observer.

FIGURE 20.

Axial fetal liver volume intrarater reproducibility: Bland–Altman analysis. (a) Oberver 1: n = 71 pairs, mean difference (lower to upper 95% limits of agreement) –2965 (–8335 to 2404) mm³; and (b) observer 2: n = 80 pairs, mean difference (lower to upper 95% limits of agreement) –853 (–8575 to 6869) mm³.

FIGURE 21.

Axial fetal liver volume inter-rater reproducibility. CC, observer 1; SS, observer 2: n = 93 pairs, mean difference (lower to upper 95% limits of agreement) –1015 (–3228 to 1198) mm³.

FIGURE 22.

Axial fetal liver volume inter-rater reproducibility: Bland Altman analysis. n = 77 pairs, mean difference (lower-upper 95% limits of agreement) –92.84 (–8255 to 8069) mm³.

Fetal subcutaneous fat, intrarater variability

Repeated measures by the same investigator were performed. Repeated measures for this parameter were less highly correlated, as demonstrated by a wider scatter of points around the line of equality (Figure 23) and between the 95% limits of agreement (Figure 24). This is likely to reflect both the smaller number of suitable images for analysis of fetal subcutaneous fat and the fact that this was a novel technique that required significant method development during the study period.

FIGURE 23.

Fetal subcutaneous fat intrarater reproducibility. M1, measurement 1; M2, measurement 2, in reference to repeated measures by the same observer.

FIGURE 24.

Fetal subcutaneous fat intrarater reproducibility: Bland–Altman analysis. n = 67 pairs, mean difference (lower to upper 95% limits of agreement) 1530 (–5630 to 8691) mm³.

Maternal subcutaneous and visceral fat masses

There was no difference in the subcutaneous fat mass (expressed as a percentage of the abdominal volume examined) between the placebo group and the metformin group at 28 weeks’ gestation (difference between means –3.45%, 95% CI –7.63% to 0.73%; p = 0.10) or at 36 weeks’ gestation (difference between means –3.43%, 95% CI –7.63% to 0.76%; p = 0.11). There was no difference in visceral fat mass (expressed as a percentage of the abdominal volume examined) at 28 weeks’ gestation (difference between means –0.02%, 95% CI –1.93% to 1.89%; p = 0.98) or at 36 weeks’ gestation (difference between means 0.18%, 95% CI –2.17% to 1.82%; p = 0.86) (Figure 25).

FIGURE 25.

Maternal subcutaneous (SC) and visceral (V) fat mass at 28 and 36 weeks’ gestation. **p ≤ 0.01, ***p ≤ 0.001.

Both groups demonstrated a significant loss of subcutaneous fat mass between 28 and 36 weeks’ gestation (paired t-test: placebo difference between means –2.14, 95% CI –3.27% to –1.01%; p = 0.0009; metformin difference between means –2.15, 95% CI –3.38% to –0.923%; p = 0.0018). There was no significant difference in the percentage change in subcutaneous fat mass from 28 to 36 weeks’ gestation between the placebo group and the metformin group (difference between means –0.45, 95% CI –4.71% to 3.82%; p = 0.83) (Figure 26).

FIGURE 26.

Percentage change in (a) subcutaneous and (b) visceral fat mass between 28 and 36 weeks’ gestation.

Neither group demonstrated any change in visceral fat mass between 28 and 36 weeks’ gestation and thus there was no difference in change in visceral fat mass between 28 and 36 weeks’ gestation between the groups (Mann–Whitney test, p = 0.60) (see Figure 26).

Maternal skeletal muscle and hepatic fat fraction

Dixon method

The mean (SD) hepatic fat fraction (%) at 28 weeks’ gestation was 3.93 (1.35) and 4.90 (2.76) in the placebo and metformin groups, respectively. At 36 weeks’ gestation the equivalent figures were 5.16 (2.78) and 5.00 (2.54).

The mean (SD) skeletal muscle fat fraction (%) at 28 weeks’ gestation was 2.84 (0.90) and 3.20 (1.33) in the placebo and metformin groups, respectively. At 36 weeks’ gestation the equivalent figures were 3.43 (1.29) and 3.62 (1.82).

There were no significant differences in fat fraction measured by the Dixon method by gestation or treatment group in either the skeletal muscle (p = 0.55) or the liver (p = 0.50) (Figure 27).

FIGURE 27.

Maternal (a) hepatic and (b) skeletal muscle fat fraction measured by the Dixon method.

¹H-magnetic resonance spectroscopy method

The mean (SD) hepatic fat fraction (%) at 28 weeks’ gestation was 1.24 (1.26) and 0.87 (1.22) in the placebo and metformin groups, respectively. At 36 weeks’ gestation the equivalent figures were 1.11 (1.80) and 0.86 (0.71).

The mean (SD) skeletal muscle fat fraction (%) at 28 weeks’ gestation was 7.13 (3.77) and 7.55 (4.86) in the placebo and metformin groups, respectively. At 36 weeks’ gestation the equivalent figures were 10.99 (9.60) and 9.10 (4.46).

There were no significant differences in fat fraction measured by ¹H-MRS by gestation or treatment group in either the skeletal muscle (p = 0.64) or the liver (p = 0.42) (Figure 28).

FIGURE 28.

Maternal (a) hepatic and (b) skeletal muscle fat fraction measured by ¹H-MRS.

Fetal liver volume, hepatic fat fraction and subcutaneous fat

There was no statistically significant difference in fetal liver volume at 28 or 36 weeks’ gestation between the placebo group and the metformin group. Both groups demonstrated a significant increase in liver volume over time (p < 0.0001), as would be expected, but the percentage increase was not significantly different between the two groups when measured in either plane (Figures 29 and 30). There was no change in the fetal hepatic fat fraction by gestation or treatment group (Figure 31). There was no difference in subcutaneous fat, measured in the sagittal and axial planes (expressed as a percentage of body volume), at 36 weeks’ gestation between the two treatment groups (Figure 32).

FIGURE 29.

Fetal liver volume: axial plane. Placebo: 28 weeks, n = 25; 36 weeks, n = 22; metformin: 28 weeks, n = 22; 36 weeks, n = 22.

FIGURE 30.

Fetal liver volume: sagittal plane. Placebo: 28 weeks, n = 25; 36 weeks, n = 23; metformin: 28 weeks, n = 23; 36 weeks, n = 22.

FIGURE 31.

Fetal hepatic fat fraction. Placebo, n = 17; metformin, n = 16 (paired samples).

FIGURE 32.

Fetal subcutaneous (s/c) fat volume: (a) sagittal plane; and (b) axial plane. Subcutaneous fat sagittal plane: placebo, n = 14; metformin, n = 25; subcutaneous fat axial plane: placebo, n = 8; metformin, n = 5.

Discussion

The purpose of this substudy was to assess whether or not distribution of body fat in obese pregnant women was altered by metformin. We also examined the fetus with the aim of assessing effect on liver volume, hepatic fat content and fetal subcutaneous fat deposits. We aimed to scan 40 participants at both 28 and 36 weeks’ gestation. Ultimately, we scanned 57 participants, with longitudinal maternal data available for 37 participants (those who attended for both scans). We obtained longitudinal fetal hepatic lipid data in 17 and 16 of the participants allocated to the placebo and metformin groups, respectively. Liver volume was successfully measured in the axial plane in 25 and 22 fetuses at 28 and 36 weeks, respectively, in participants allocated to placebo and in 22 fetuses at both time points in participants allocated to metformin. Liver volume was successfully measured in the sagittal plane in 25 and 23 fetuses at 28 and 36 weeks, respectively, in participants allocated to placebo and in 22 fetuses at both time points in participants allocated to metformin. Subcutaneous fat mass was measured in the sagittal plane in 14 and 25 subjects allocated to placebo and metformin, respectively. In the axial plane it was successfully measured in eight and five subjects in the placebo and metformin groups, respectively.

We have demonstrated that the maternal scanning protocols work well in obese pregnant women, with good inter- and intrarater correlation. We have shown that it is possible to obtain the fetal data as we had planned, although not in every subject scanned because of fetal movement and time limitations, with priority being given to acquisition of the maternal data. There was reasonably good inter- and intrarater correlation for the fetal liver volume data. Correlation was less good for the subcutaneous fat measurements but the cohort numbers were small for this study.

In summary, participants in both the placebo and the metformin arms of the study lost subcutaneous fat over the course of pregnancy but there was no difference in the percentage change between the two groups. We saw no differences in the amounts of visceral fat either between treatment groups or by gestation. Ectopic lipid deposition in both the liver and skeletal muscle was also the same in both groups and did not change between 28 and 36 weeks’ gestation.

Mean hepatic fat fraction in both groups was relatively low, particularly when measured by ¹H-MRS. When measured using the Dixon method, it was higher than we have previously seen in a cohort of 10 lean and 10 obese non-diabetic pregnant women116 but still lower than in a cohort of obese non-pregnant women117 and certainly below the diagnostic threshold for non-alcoholic fatty liver disease. 118 Mean skeletal muscle fat fraction measured by ¹H-MRS was similar to that in our previous cohort of obese pregnant women116 and to that of a group of normally glucose-tolerant obese (mean BMI 30 kg/m²) non-pregnant women. 119 This suggests that pregnancy itself, rather than metformin, may be exerting a protective effect on the liver, which deters accumulation of lipid in this site.

The fetal data must be interpreted with a greater degree of caution. The scanning protocols are not well established and were subject to some method development during the study period. Fetal movement during the image acquisition period is an extra challenge. We aimed to limit the scan duration to 60 minutes, which was the limit of acceptability for the participants, and we prioritised the acquisition of maternal data in this time. However, we still acquired a reasonable amount of data suitable for analysis and have not demonstrated any differences in fetal liver volumes or hepatic and subcutaneous fat depots between the placebo group and the metformin group. This is in keeping with the primary outcome of the EMPOWaR trial, which demonstrated no difference in birthweight of the babies, and also the secondary outcomes, for which we saw no difference in neonatal fat mass measured by ADP or in neonatal skinfold thicknesses (see Maternal and neonatal body composition).

In conclusion, we have not demonstrated any effect of metformin in obese pregnant women on maternal or fetal body fat distribution at 28 and 36 weeks’ gestation.

Background

It is now well recognised that the intrauterine environment is a key time for determining not only fetal growth and consequent birthweight but also future life health, a concept known as ‘early life programming’. 120 One of the key mechanisms thought to be responsible for programming is overexposure of the fetus to glucocorticoids, with consequent alterations to the fetal hypothalamic–pituitary–adrenal (HPA) axis. 121,122

The maternal HPA axis undergoes significant activation during pregnancy, resulting in a substantial increase in maternal cortisol levels. 123,124 This is partly because of placental secretion of large quantities of corticotrophin-releasing hormone. 125 These physiological changes in the HPA axis are essential for normal fetal growth and development and promotion of fetal organ maturation and are also thought to have a role in a gestational clock mechanism signalling the appropriate time for the onset of labour. 126 However, over- and possibly underexposure to glucocorticoids in utero can have adverse effects on the fetus that persist into adult life. Glucocorticoids are lipophilic and readily cross the placenta, yet fetal glucocorticoid levels are tenfold lower than maternal levels because of the action of placental 11β-HSD2. 127 11β-HSD2 acts as a placental enzyme barrier that converts active cortisol into inactive cortisone,127 thus protecting the fetus from overexposure to excess glucocorticoid. Both animal and human studies have suggested that the efficiency of placental 11β-HSD2 is variable and may be weakened by factors such as diet, infection, inflammation, hypoxia and stress,121,122 thus allowing greater transplacental passage of cortisol to the fetus. Even modest changes in placental 11β-HSD activity appear to have the potential to significantly alter fetal exposure. 128,129

Glucocorticoids exert their effect through the glucocorticoid receptor (GR), which acts as a transcription factor. The activated GR translocates to the nucleus, binds to GR response elements in the promoters of target genes and influences their transcription. 130 11β-HSD1 is the enzyme that converts cortisone back to cortisol. Alterations in levels of GR could also affect the sensitivity of the placenta to cortisol and alterations in levels of 11β-HSD1 will affect cortisol availability.

The impact of undernutrition in pregnancy on the HPA axis has been extensively studied in both animal models and human cohorts. It is difficult to ascertain the impact of diet alone because of potential confounding from the impact of stress but it would appear that prenatal exposure to undernutrition has an adverse impact on long-term health, mediated in part through overexposure to intrauterine glucocorticoid. 131 The impact of overnutrition or obesity in pregnancy on the HPA axis is not yet known. In the non-pregnant population, obesity is associated with activation of the HPA axis but there is associated increased hepatic metabolism and renal excretion of cortisol results in near-normal levels of circulating cortisol. 132–134 If dysregulation is maintained during pregnancy, fetuses of obese women may be exposed to altered levels of glucocorticoids compared with fetuses of lean women, with consequent impact on birthweight and health in later life. The current data on the impact of obesity are conflicting. One study found that women who were obese at the start of pregnancy had elevated evening salivary cortisol in the third trimester, particularly women who had gestational weight gain greater than that in Institute of Medicine guidelines. 135 Another study demonstrated higher hair cortisol (a measure of longer-term cortisol exposure) in obese women. 136 Contrary to this, our own recently published data suggest that obese women have lower total serum levels of cortisol and lower calculated free levels of cortisol through pregnancy than lean women,137 suggesting that cortisol exposure to the fetus may be lower in obese pregnancy. A better understanding of the impact of obesity on offspring birthweight and the maternal and fetal HPA axis is clearly required.

As the primary outcome of our study was birthweight, and fetal exposure to cortisol is a likely determinant of birthweight, this mechanistic substudy was designed to examine the effect of metformin on maternal salivary cortisol levels (as a measure of cortisol exposure). Furthermore, as placental glucocorticoid metabolism is known to be tightly regulated by changes in inflammation, and in a rodent model metformin altered expression of genes involved in glucocorticoid metabolism in skeletal muscle and adipose tissue, we tested whether or not treatment with metformin had any effect on the placental expression of genes regulating fetal glucocorticoid exposure, including GR, 11β-HSD1 and 11β-HSD2.

Methods

Results

Salivary cortisol

Samples were obtained from 235 participants. Unblinding after data lock revealed final cohort numbers as detailed in Table 14. Baseline demographics were similar between the two groups and are also shown in Table 14.

TABLE 14 - Baseline characteristics of participants in the salivary cortisol substudy

Characteristic	ITT		Per protocol
	Placebo (n = 121)	Metformin (n = 114)	Placebo (n = 77)	Metformin (n = 79)
Age (years), mean (SD)	29.9 (5.0)	29.1 (5.6)	29.8 (5.0)	29.9 (5.5)
Nulliparity, mean (SD)	50 (41.3)	52 (45.6)	30 (39.0)	38 (48.1)
BMI (kg/m2), mean (SD)	36.9 (5.0)	37.5 (5.2)	36.7 (5.0)	37.2 (4.6)
Current smoking, n (%)	9 (7.4)	15 (13.2)	5 (6.5)	8 (10.1)
Highest educational qualification, n (%)
Up to 16 years	41 (33.9)	31 (27.2)	23 (29.9)	16 (20.3)
> 16 years	80 (66.1)	83 (72.8)	54 (70.1)	63 (79.7)

There was no difference in the increment of mean cortisol on waking at each gestation time point between the placebo group and the metformin group (Table 15 and Figure 33).

TABLE 15 - Salivary cortisol (mmol/l)

Time point	ITT							Per protocol
	Placebo		Metformin		Adjusted mean difference	95% CI	p-value	Placebo		Metformin		Adjusted mean difference	95% CI	p-value
	Mean	SD	Mean	SD				Mean	SD	Mean	SD
Baseline
n	107		101					70		72
Bedtime	2.23	8.0	1.99	2.38				1.51	3.7	2.00	2.7
Waking	9.03	24.6	6.4	3.9				10.23	30.2	6.42	3.5
Mean difference	–6.80	22.5	–4.36	5.1	1.023	0.978 to 1.071	0.3111	–8.72	26.7	–4.42	5.0	1.040	0.973 to 1.111	0.2429
28 weeks
n	48		52					36		47
Bedtime	3.17	4.0	2.04	1.6				3.19	4.0	2.04	1.7
Waking	8.68	6.9	9.26	4.3				9.28	7.7	9.26	4.5
Mean difference	–5.51	7.8	–7.22	3.7	0.994	0.984 to 1.004	0.2223	–6.09	8.2	–7.22	3.9	0.997	0.987 to 1.008	0.6153
36 weeks
n	36		41					28		35
Bedtime	3.40	3.15	2.98	1.8				3.63	3.5	2.89	1.7
Waking	8.90	4.2	8.74	4.8				8.99	4.3	9.04	5.1
Mean difference	–5.50	3.7	–5.74	5.6	0.999	0.990 to 1.008	0.7780	–5.35	3.7	–6.15	5.9	0.996	0.986 to 1.007	0.4921

FIGURE 33.

Bedtime and waking salivary cortisol: (a) baseline; (b) 28 weeks; and (c) 36 weeks.

Discussion

We have not demonstrated any differences in bedtime or waking salivary cortisol between obese women taking metformin and those taking placebo. The findings should be interpreted with caution as the sample size was small and there was considerable variation in the waking levels of cortisol (as is well documented in other studies). Nevertheless, these findings accord with the lack of change in fasting morning plasma cortisol between groups and suggest that metformin treatment had no effect on the activity of the maternal HPA axis (see Table 6). In addition, we found no difference in the expression of GR, 11β-HSD1 or 11β-HSD2 in the placentae of women who took metformin during pregnancy compared with those who took placebo once BMI band and mode of delivery were taken into account. In our analyses we adjusted for mode of delivery, which is recognised to alter placental gene expression of genes related to fetal glucocorticoid exposure. We acknowledge that there may be other confounding factors such as time between delivery and placental sample collection that we did not adjust for in the statistical analyses. However, all samples were collected using standard protocols and good-quality RNA was extracted from the tissues used. Although 11β-HSD2 is known to be sensitive to inflammation and we observed changes in circulating inflammatory markers at 36 weeks’ gestation between women taking metformin and those taking placebo, the lack of change observed here could be consistent with a compensatory change in placental metabolism in women taking metformin. Alternatively, as our primary outcome measure of birthweight was the same in the two treatment groups, and fetal glucocorticoid exposure is a key determinant of birthweight, it is perhaps unsurprising that we have not seen an effect on this determinant of birthweight.

Introduction

Maternal obesity is well recognised to be associated with an increased risk of caesarean section. 61,140–143 Although this association is widely reported, little is known about the mechanism behind it. There is a commonly held clinical assumption that this is related to fat dystocia from the maternal tissues and/or dystocia as a result of larger babies. These are indeed likely to be contributory factors but myometrial contractility is known to be impaired in obese women, even after adjustment for birthweight. 144 Additionally, contractility is impaired in women with diabetes mellitus. 145 Of particular note is that the response to oxytocin is reduced in these women. Thus, if the diabetic environment has already reduced contractility, for example because of glycosylation of proteins and fibre damage, leading to a poorly progressing labour in these women, then oxytocin is likely to be needed. The reduced responsiveness to oxytocin would suggest that its efficacy is reduced and this may contribute to the increased rate of caesarean sections in this group of women. In addition, there is also evidence that insulin per se is detrimental to contractility, possibly by causing hyperpolarisation. 146

Metabolic changes in the uterus are a recognised preparation for labour. These include changes in lactate dehydrogenase isoforms to those favouring hypoxic conditions and, of interest to this study, increased glycogen storage from glucose uptake into myometrial cells, along with fatty droplet inclusions. 147,148 These metabolic changes are expected to maintain forceful contraction, necessary for labour, in the face of repetitive, transient ischaemic episodes, consequent to occlusion of the blood vessels within the myometrium with each contraction. It could therefore be suggested that, if insulin sensitivity improvement occurs with metformin, then more glycogen would be stored and labour outcome improved. This suggestion can be tested directly by determining glycogen levels and also by challenging the myometrial tissue with a solution lacking glucose and monitoring contractile activity.

This mechanistic substudy was designed to examine whether or not improving insulin sensitivity with metformin improved myometrial contractility in obese pregnant women and increased myometrial glycogen content.

Methods

At caesarean section, biopsy of lower-segment myometrium was obtained from consenting women participating in the EMPOWaR trial, according to the SOP (see Appendix 8). All biopsies were immediately placed in physiological saline (PSS), stored at 4 °C and used for experimentation within 24 hours of collection for contractility studies or snap frozen at –80 °C for later analysis of glycogen storage.

Results

All sample collection and analysis were performed blind to treatment allocation group. Eight samples were obtained for the contractility studies. Unblinding following data lock revealed final cohort numbers as follows: placebo, n = 2; metformin, n = 6. Twenty-eight samples were obtained for the glycogen storage study with final cohort numbers as follows: placebo, n = 17; metformin, n = 11. Baseline characteristics are shown in Table 18.

Contractility

Responses to oxytocin were calculated and showed the expected increases in amplitude and overall force (AUC, integral). The mean increases in amplitude and force from control values (preceding spontaneous contractions, 100%) for all samples (n = 8) were 186% ± 21% and 216% ± 31% (standard error of the mean), respectively. In the metformin samples (n = 6) these values were 170% ± 12% and 210% ± 26%, respectively. In the placebo samples (n = 2) the averages were 261% and 296.1%, respectively (Figure 34).

FIGURE 34.

Example of myometrium contraction trace. Continuous trace of spontaneous contractions from paired strips of myometrium. Spontaneous contractions were established and then oxytocin (0.5 nM) was superfused throughout the experiment. After 30 minutes of control contractions, test strips were bathed in PSS 0-glucose. Contractility measurements were made from test strips at 30-minute intervals and compared with those of time-matched controls.

We also examined the effects of glucose removal on the above sample (see Figure 34). After 120–150 minutes in PSS 0-glucose, contractions in the eight samples decreased from control values (100%) to an amplitude of 83% ± 6% and overall force (AUC) was 57% ± 9%. In the metformin samples these values were 82% ± 6% and 69% ± 10% and in the placebo samples these values were 94% and 30%, respectively.

Formal statistical analyses to compare the results between the metformin group and the placebo group were not performed because of low sample numbers.

Discussion

It is not possible to draw any robust conclusions from this substudy as the sample size was so small and the distribution of treatment allocation following unblinding was too uneven. The unequal distribution of treatment allocations between the two groups is a consequence of an unavoidable risk when carrying out mechanistic studies blind to treatment allocation. Clearly, a large sample size will minimise the possibility of this being a problem. We had hoped to obtain a much larger sample of myometrial biopsies but the provision of expertise for ‘out-of-hours’ tissue collection was limited. We had anticipated that we would see improved myometrial contractility responsiveness to oxytocin and higher glycogen storage in samples from participants taking metformin. We were unable to demonstrate any differences but, as stated, our sample size was very small. There are few data on the effect of metformin on smooth muscle contractility. One small study of human myometrial biopsies did not demonstrate any differences in in vitro contractility with the addition of metformin. 151 However, there is evidence from animal models that metformin improves contractility in vascular and penile smooth muscle, mediated through activation of adenosine monophosphate-activated protein kinase or inhibition of the production of reactive oxygen intermediates. 152–155 We have previously shown impaired uterine contractility in the presence of increased cholesterol. 156 Increased cholesterol is well recognised to be associated with obesity and we hypothesised that this may be one of the contributory mechanisms behind the excess risk of caesarean delivery in obese women. In non-pregnant populations metformin has a beneficial effect on the cholesterol profile. 79 However, we did not see any effect on cholesterol in our study population and therefore it is not possible to confirm this hypothesis from these data.

Although our sample size in this substudy was insufficient to draw any conclusions, we did not see any differences in rates of caesarean section or post-partum haemorrhage in the main trial cohort, which would perhaps suggest that metformin has not had either a beneficial or an adverse effect on myometrial contractility.