Metformin Improves Glucose Effectiveness, Not Insulin Sensitivity: Predicting Treatment Response in Women With Polycystic Ovary Syndrome in an Open-Label, Interventional Study

Abstract

Context:

Although metformin is widely used to improve insulin resistance in women with polycystic ovary syndrome (PCOS), its mechanism of action is complex, with inconsistent effects on insulin sensitivity and variability in treatment response.

Objective:

The aim of the study was to delineate the effect of metformin on glucose and insulin parameters, determine additional treatment outcomes, and predict patients with PCOS who will respond to treatment.

Design and Setting:

We conducted an open-label, interventional study at an academic medical center.

Subjects:

Women with PCOS (n = 36) diagnosed by the National Institutes of Health criteria participated in the study.

Interventions:

Subjects underwent fasting blood sampling, an IV glucose tolerance test, dual-energy x-ray absorptiometry scan, transvaginal ultrasound, and measurement of human chorionic gonadotropin-stimulated androgen levels before and after 12 weeks of treatment with metformin extended release 1500 mg/d. Interval visits were performed to monitor anthropometric measurements and menstrual cycle parameters.

Main Outcome Measures:

Changes in glucose and insulin parameters, androgen levels, anthropometric measurements, and ovulatory menstrual cycles were evaluated.

Results:

Insulin sensitivity did not change despite weight loss. Glucose effectiveness (P = .002) and the acute insulin response to glucose (P = .002) increased, and basal glucose levels (P = .001) decreased after metformin treatment. T levels also decreased. Women with improved ovulatory function (61%) had lower baseline T levels and lower baseline and stimulated T and androstenedione levels after metformin treatment (all P < .05).

Conclusions:

Using an IV glucose tolerance test, which distinguishes improvements in glucose effectiveness and insulin sensitivity, metformin does not improve insulin sensitivity in women with PCOS but does improve glucose effectiveness. The improvement in glucose effectiveness may be partially mediated by decreased glucose levels. T levels also decreased with metformin treatment. Ovulation during metformin treatment was associated with lower baseline T levels and greater T and androstenedione decreases during treatment, but not with insulin or LH levels. Thus, the action of metformin in PCOS primarily affects glucose levels and steroidogenesis, which may be mediated by mechanisms that affect both pathways, such as inhibition of mitochondrial complex I.

Polycystic ovary syndrome (PCOS) is the most common endocrinopathy in reproductive-aged women, affecting 7–10% (1–3). Insulin resistance has been recognized as important in the pathogenesis of the disorder in approximately 65% of women with PCOS (4, 5). The consequences of insulin resistance and the additional β-cell dysfunction (6) include an increased prevalence of impaired glucose tolerance and type 2 diabetes compared to body mass index (BMI)-matched controls (7–9). The prevalence of metabolic syndrome is also increased (9–11). The compensatory hyperinsulinemia resulting from insulin resistance drives androgen production from theca cells, decreases SHBG, and may suppress folliculogenesis directly (12).

Metformin is widely used to improve insulin resistance in women with PCOS, although it is well known that its mechanism of action is more complex. Insulin resistance with its compensatory hyperinsulinemia has provided the rationale for off-label use of metformin to treat affected women (13). Unfortunately, not all women with PCOS respond to metformin with improved ovulation or decreased androgen levels (14), and identifying the subsets of patients who will benefit from metformin therapy remains a challenge. Importantly, metformin may not improve insulin sensitivity. Metformin has no effect on insulin sensitivity in the absence of weight loss in persons with type 2 diabetes (15, 16) or in women with PCOS in some studies (17, 18). Therefore, it may not be useful in all women with PCOS and insulin resistance who have been considered the primary candidates for therapy.

The objectives of the present study were threefold. First, the effect of metformin therapy on glucose and insulin parameters was examined to determine its action in women with PCOS using an IV glucose tolerance test (IVGTT). Second, the secondary responses to metformin treatment were assessed, including changes in androgen levels, anthropometric measurements, and ovulation over a 3-month treatment period using physical examinations, serial ultrasounds, hormone levels, and dual-energy x-ray absorptiometry (DEXA). Third, the factors that predicted responses to metformin treatment were delineated. These results provide critical information regarding the mechanism of metformin action, the therapeutic responses to metformin, and the subsets of women with PCOS that are most likely to benefit from metformin therapy. This study also provides insight into the outcome measurements that can be analyzed in pharmacogenetic studies of metformin therapy.

Subjects and Methods

Subjects

Subjects (n = 39) were between the ages of 18 and 40 years and were diagnosed with PCOS according to the National Institutes of Health criteria: 1) irregular menses (< nine menstrual periods/y); and 2) clinical and/or biochemical evidence of hyperandrogenism (19). Clinical hyperandrogenism was defined as a Ferriman-Gallwey score greater than 9, the upper 95% confidence limit for the Boston-based control populations (20). Biochemical hyperandrogenism was defined as an androgen level greater than the 95% confidence limits in control subjects with regular, ovulatory menstrual cycles: T > 63 ng/mL (2.8 nmol/L), dehydroepiandrosterone sulfate > 430 μg/dL (1.16 μmol/L), or androstenedione levels > 3.8 ng/mL (0.13 nmol/L) (9).

All subjects were otherwise healthy nonsmokers with normal thyroid and renal function, normal prolactin levels, no diabetes, and a premenopausal follicular phase FSH level. Nonclassic congenital adrenal hyperplasia was excluded with a follicular phase 17OH progesterone ≤ 300 ng/dL (9.1 nmol/L) (21). Subjects were on no hormonal medication for at least 3 months and no medications that influence insulin, inflammation, or lipid levels for at least 1 month. Pregnancy was excluded, and subjects had no plans for pregnancy during the study period.

The study was approved by the Partners Human Research Committee. All subjects provided written, informed consent.

Protocol

Subjects underwent a baseline ultrasound and blood sampling for estradiol and progesterone at a screening visit and were observed prospectively (average, 41 d) to validate baseline menstrual cycle frequency by history. Subjects were admitted to the Massachusetts General Hospital Human Research Center at 8 am. After a short physical examination, subjects underwent fasting blood sampling. The subjects subsequently underwent an IVGTT, with IV glucose 0.3 g/kg administered at time 0 and regular human insulin 0.03 U/kg injected at 20 minutes (22). At the same visit, subjects underwent a DEXA scan on the Hologic-2000 densitometer (Hologic, Inc) and a transvaginal ultrasound (Philips HD11XE, 4–8 MHz convex array transducer). Finally, subjects had a baseline human chorionic gonadotropin (hCG) sample drawn for androgens and SHBG, were given hCG 5000 IU, and returned 24 hours later for a final blood draw for stimulated androgens.

Subsequently, subjects started treatment with metformin extended release (ER) 500 mg/d, with the dose increased to 1000 mg after 2 weeks, then to 1500 mg/d after 2 more weeks, for a total of 12 weeks at the full dose. Subjects returned every 2 weeks for anthropometric measurements, blood sampling (estradiol and progesterone levels), and a pelvic ultrasound to monitor folliculogenesis, and they returned for additional visits if follicle size indicated impending ovulation. Compliance was determined by questioning at the biweekly visits. After 12 weeks of metformin ER 1500 mg/d, subjects were admitted to the Massachusetts General Hospital Clinical Research Center to repeat the study as outlined above.

Assays

Serum LH and FSH were measured using a two-site monoclonal nonisotopic system (Architect; Abbott Laboratories) (23). LH and FSH levels are expressed in international units per liter as equivalents of pituitary standard 92/510 (FSH) and 80/552 (LH). Serum T was measured using a RIA (Coat-a-Count, Diagnostic Products Corporation). Androstenedione and 17OH progesterone were measured by liquid chromatography-tandem mass spectrometry (Mayo Medical Laboratories-New England). SHBG was measured using a chemiluminescent enzyme immunometric assay (Immulite; Diagnostic Products Corp). Insulin was measured using an immunochemiluminescent immunoassay (Immulite 2000; Diagnostic Products Corp), with a lower limit of detection of 2.0 μIU/mL (14.4 pmol/L).

Statistics

MinMod Millenium (24) was used to analyze IVGTT data. Equations are included in the Supplemental Data. The data calculated include: 1) the first phase, acute insulin response to glucose (AIR_g), which is the change in insulin over time within the first 10 minutes of the glucose infusion; 2) the insulin sensitivity index (S_i) or the capacity of insulin to mediate glucose disposal, which is calculated as the fractional transport of insulin from the plasma, ie, the difference in the current insulin measurement and the basal insulin measurement, and the rate that levels of insulin decrease as it enters tissue compartments; 3) the disposition index (DI), which is a measurement of insulin secretion and action, calculated as a product of the AIR_g and the S_i; 4) glucose effectiveness (S_g), or the capacity of glucose to mediate its own disposal, which is calculated as the change in glucose levels from basal to current levels; 5) basal glucose (G_b); 6) basal insulin (I_b); 7) β-cell function (homeostasis model of assessment [HOMA]); and 8) insulin resistance (HOMA for insulin resistance) before and after metformin treatment (24).

Data were subsequently log-normalized for analysis. Pre- and post-metformin data were compared using paired t tests or one-way ANOVA for repeated measures, as appropriate. Two-way ANOVA was used to examine pre- and post-hCG-stimulated changes and responders vs nonresponders before and after metformin treatment. Analyses were performed using SigmaStat (SYSTAT). Data are reported as mean ± SE, except where noted. A P value < .05 was taken as the minimum level of significance.

Results

Baseline characteristics

Three subjects did not complete the study: two subjects became pregnant during metformin treatment, and one subject was lost to follow-up after 1 month. Data from the remaining subjects (n = 36) were used for analysis. Subjects were 28.6 ± 5.2 years of age, with a Ferriman-Gallwey score of 12.2 ± 1.2, and 82% suffered from acne. Nine of the 36 subjects had impaired glucose tolerance (glycated hemoglobin ≥ 5.7%). Ten of the 36 subjects had a baseline HOMA-IR score of > 3, suggesting insulin resistance, with two subjects found in both groups. All subjects tolerated the full dose of metformin ER 1500 mg/d, with no dropouts and no requirement for a dose drop. Most subjects experienced at least one side effect (87%; Supplemental Table 1).

Change after metformin treatment

Anthropometric measures

After the 3-month metformin treatment period, weight, waist and hip circumferences, and diastolic blood pressure decreased (Table 1 and Supplemental Table 2). In addition, calculated lean mass decreased (Table 1). There were no changes in truncal or total fat, bone mineral content or systolic blood pressure during the study period (Table 1 and Supplemental Table 2).

Table 1.

Anthropometric Measurements of Women With PCOS at Baseline and After Metformin 1500 mg/d for 3 Months

	Baseline	16 Weeks	P Value
Weight, kg	86.7 ± 4.5	85.4 ± 4.5	.02
Waist, cm	102.5 ± 3.2	100.4 ± 3.2	.04
Hip, cm	115.1 ± 2.9	113.7 ± 2.9	.04
DEXA
Trunk fat, g	18 228 ± 1584	18 260 ± 1572	.7
Total fat, g	36 570 ± 2882	36 883 ± 2868	.3
Bone mineral content, g	2177 ± 52	2187 ± 54	.1
Lean mass, g	48 874 ± 1657	47 527 ± 1622	<.001
Systolic blood pressure	117 ± 2	114 ± 2	.21
Diastolic blood pressure	75 ± 2	71 ± 2	.03

Measurements were performed in the Massachusetts General Hospital Clinical Research Center.

Glucose homeostasis parameters derived from IVGTT and MinMOD analyses

The changes in the indices of glucose homeostasis determined from MinMOD Millenium analyses of IVGTT data are detailed in Table 2. There was an increase in S_g, the AIR_g, and the DI, along with a decrease in G_b levels after treatment with metformin (Table 2). The change in glucose effectiveness correlated with the change in the AIR_g (r = 0.635; P < .001; Supplemental Table 3). There was no change in S_i, fasting insulin levels (I_b), β-cell function, or insulin resistance during the course of the study. The same findings held true in the subset of women who lost weight (S_i, 3.9 ± 0.6 vs 2.5 ± 0.5 mIU/L⁻¹*min⁻¹; P = 1.0), and there was no correlation between the change in insulin sensitivity and change in weight (r = 0.065; P = .7; Supplemental Table 3).

Table 2.

Glucose and Insulin Parameters in Women With PCOS at Baseline and After 3 Months of Treatment With Metformin

Parameters	Pre-Metformin	Post-Metformin	P Value
Sg, min−1	0.020 ± 0.001	0.026 ± 0.002	.002
AIRg, mIU/L*min	521.5 ± 62.7	746.2 ± 93.9	.002
DI	1554.1 ± 154.6	2670.7 ± 407.4	.01
Gb, mg/dL	81.0 ± 1.2	77.8 ± 1.0	.001
Si, mIU/L−1*min−1	4.0 ± 0.4	4.1 ± 0.4	.8
Ib, mIU/L	10.4 ± 1.2	9.3 ± 1.0	.4
β-Cell function, mIU/mm	270.3 ± 52.9	270.1 ± 35.3	.3
Insulin resistance, mm·mU/L2)	2.1 ± 0.2	1.8 ± 0.2	.2

Data are expressed as mean ± SE. To change to SI units, multiply G_b by 0.0555 (for mmol/L) and I_b by 6.945 (for pmol/L).

There were correlations between changes in fasting insulin levels and waist circumference (r = 0.339; P < .05) and changes in fasting glucose and insulin levels (r = 0.356; P < .05). There was an inverse correlation between changes in S_g and changes in waist circumference (r = −0.330; P < .05; Supplemental Table 3).

When subjects with improved glucose-mediated glucose disposal were compared to those with no improvement, the increase in AIR_g (397.2 ± 90.2 vs −0.67 ± 38.1 mIU/L*min; P = .002; Supplemental Table 4) and DI (1860 ± 580 vs 68 ± 286; P = .02) and the decrease in waist circumference (−2.8 ± 1.1 vs −0.5 ± 1.6 cm; P = .01) were greater in the women with improved S_g, but no other differences were noted.

Lipid profile

Total cholesterol (173.1 ± 6.4 vs 163.9 ± 6.4 mg/dL [4.5 ± 0.2 vs 4.3 ± 0.2 mmol/L]; P = .01) and low-density lipoprotein cholesterol (103 ± 5.8 vs 95.7 ± 6.3 mg/dL [2.7 ± 0.2 vs 2.5 ± 0.2 mmol/L]; P = .006) decreased after metformin treatment. Triglycerides (86.1 ± 8.1 vs 94.0 ± 11.0 mg/dL [0.97 ± 0.09 vs 1.06 ± 0.12 mmol/L]; P = .5) and high-density lipoprotein (HDL) (52.2 ± 2.3 vs 51.0 ± 2.3 mg/dL [1.4 ± 0.1 vs 1.3 ± 0.1 mmol/L]; P = .7) did not improve with metformin.

Androgen levels

There was no difference in androstenedione (164.2 ± 9.5 vs 154.8 ± 8.0 ng/dL [5.7 ± 0.3 vs 5.4 ± 0.3 nmol/L]; P = .1) at 8 am vs at noon after the IVGTT, but 17OH progesterone levels (79.9 ± 7.0 vs 72.4 ± 5.8 ng/dL [2.4 ± 0.2 vs 2.2 ± 0.2 nmol/L]; P = .02) were higher at 8 am. Therefore, the afternoon sample was used for comparison to the levels after hCG stimulation, which were drawn at noon.

T, androstenedione, and 17OH progesterone levels increased after hCG stimulation (Table 3 and Supplemental Table 5). Although subjects were scheduled for the pre- and post-metformin studies in the follicular phase, some subjects were in the luteal phase by chance when assessed retrospectively by progesterone levels and ultrasound evidence of a corpus luteum. T (56.8 ± 5.3 vs 33.9 ± 8.2 ng/dL [1.97 ± 0.18 vs 1.18 ± 0.28 nmol/L]; follicular vs luteal phase; P = .01) and androstenedione levels (172.5 ± 11.4 vs 124.0 ± 8.2 ng/dL [6.02 ± 0.40 vs 4.33 ± 0.29 nmol/L]; P = .05) were higher in the follicular phase, and 17OH progesterone levels (61.0 ± 4.4 vs 131.5 ± 6.0 ng/dL [1.85 ± 0.13 vs 3.98 ± 0.18 nmol/L]; P = .0003) were higher in the luteal phase. Therefore, subjects with at least one study in the luteal phase were removed from analysis (n = 11). The T levels were significantly lower after metformin treatment in subjects studied in the follicular phase (n = 25; Table 3). There was no interaction between metformin treatment and hCG stimulation (all P > .05).

Table 3.

Androgen Levels in Women With PCOS Before and 3 Months After Treatment With Metformin 1500 mg/d and Before and After hCG Stimulation

	Pre Metformin		Post Metformin		hCG P Value	Metformin P Value
	Pre hCG	Post hCG	Pre hCG	Post hCG
T, ng/dL	59.7 ± 6.2	74.9 ± 10.0	45.8 ± 6.9	69.9 ± 10.8	<.05	<.05
Androstenedione, ng/dL	171.4 ± 12.4	259.3 ± 22.6	161.5 ± 14.2	241.5 ± 26.1	<.001	.4
17OH progesterone, ng/dL	57.9 ± 4.3	214.8 ± 38.8	56.8 ± 5.8	196.1 ± 35.9	<.001	.8
SHBG, nmol/L	36.5 ± 3.8	40.7 ± 4.3	33.9 ± 4.0	38.7 ± 4.4	.4	.7

Data are expressed as mean ± SE. Subjects who were studied in the luteal phase were removed from the analysis (final n = 25). To change to SI units, multiply T by 0.0347 (for nmol/L), androstenedione by 0.0349 (for nmol/L), and 17OH progesterone by 0.0303 (for nmol/L).

Women with PCOS whose T levels improved had higher pre-hCG T levels (65.0 ± 7.1 vs 38.2 ± 9.0 ng/dL [2.26 ± 0.25 vs 1.16 ± 0.27 nmol/L]; P = .02), pre-hCG androstenedione levels (184.4 ± 13.5 vs 119.4 ± 18.3 ng/dL [6.44 ± 0.47 vs 3.62 ± 0.64 nmol/L]; P = .02), and post-hCG androstenedione levels (277.1 ± 25.7 vs 188.0 ± 34.3 ng/dL [9.7 ± 0.9 vs 6.6 ± 1.2 nmol/L]; P = .04) before metformin treatment.

There was no difference in baseline or change in weight, glucose or insulin parameters, prevalence of impaired glucose tolerance or insulin resistance, ovarian volume, FSH or LH levels, 17OH progesterone, or SHBG levels in the two groups (Supplemental Table 6).

Ovulation

Overall, 61% of the women with PCOS had an improved ovulatory response after metformin treatment. Patients with lower baseline T (pre-hCG and pre-metformin treatment) were more likely to have an ovulatory response (43.1 ± 5.5 vs 67.1 ± 7.7 ng/dL [1.5 ± 0.19 vs 2.3 ± 0.27 nmol/L]; P < .01). Of note, there were three anovulatory bleeds in three subjects who also had at least one ovulatory cycle, resulting in a 5% rate of anovulatory cycles overall.

After metformin treatment, the pre-hCG stimulation (35.6 ± 6.4 vs 53.6 ± 8.4 ng/dL [1.2 ± 0.22 vs 1.9 ± 0.29 nmol/L]; P = .047), post-hCG stimulation (50.7 ± 8.9 vs 88.3 ± 14.4 ng/dL [1.8 ± 0.3 vs 3.1 ± 0.5 nmol/L]; P = .01), and change in T levels (Δ) resulting from hCG stimulation (15.0 ± 3.1 vs 34.7 ± 8.5 ng/dL [0.52 ± 0.11 vs 1.2 ± 0.29 nmol/L]; P = .02) were lower in those who ovulated. Similarly, the pre-hCG stimulation (128.6 ± 10.5 vs 187.4 ± 19.5 ng/dL [4.5 ± 0.37 vs 6.5 ± 0.68 nmol/L]; P < .01), post-hCG stimulation (178.6 ± 17.2 vs 287.5 ± 37.3 ng/dL [6.2 ± 0.6 vs 10.0 ± 1.3 nmol/L]; P = .004), and change in androstenedione levels (Δ) resulting from hCG stimulation (50.1 ± 8.6 vs 100.1 ± 23.8 ng/dL [1.8 ± 0.3 vs 3.5 ± 0.83 nmol/L]; P = .03) were lower in subjects who responded to metformin with ovulation (Figure 1).

Figure 1.

Predictors of ovulatory response in women with PCOS treated with metformin. T and androstenedione levels before hCG (open bars) and after hCG (closed bars), and before metformin (A and C) and after metformin (B and D) in women with PCOS who responded to metformin treatment with increased ovulation (responders) or did not respond (nonresponders). *, P < .05. To change to SI units, multiply T by 0.0347 (for nmol/L) and androstenedione by 0.0349 (for nmol/L).

There was no difference in weight, glucose or insulin parameters, ovarian volume, prevalence of impaired glucose tolerance or insulin resistance, FSH and LH levels, 17OH progesterone or SHBG levels in the two groups (Supplemental Table 7).

Discussion

The goals of the study were to determine the effect of metformin therapy on glucose and insulin parameters in women with PCOS, to determine the important secondary responses to metformin treatment, and to delineate the factors that predicted those responses. Despite the common belief that metformin improves insulin sensitivity, the current data demonstrate that metformin does not work in this manner. Rather, metformin improved glucose-mediated glucose disposal (S_g), the acute insulin response to glucose (AIR_g), and fasting glucose levels in the absence of changes in the insulin sensitivity index (S_i). These findings were true despite decreases in weight and in hip and waist circumferences. Secondary responses included a decrease in T levels and an improved ovulatory response. Lower baseline T levels predicted the ovulatory response, and baseline and stimulated T and androstenedione levels were lower in women who ovulated during metformin treatment. Thus, the study highlights the importance of direct metformin effects on the ovaries, hepatocytes, and muscle cells to produce these independent outcomes. Effects at these targets may be mediated through the common mechanism of mitochondrial complex I inhibition.

Decreased hepatic glucose output is a well-known primary effect of metformin treatment (25–27). The current study supports this mechanism in women with PCOS by demonstrating improved fasting glucose levels with metformin, as in a previous meta-analysis (14). The lowered hepatic glucose output results from inhibition of electron transport in the mitochondrial respiratory complex I (28, 29). Metformin may also reduce hepatic glucagon-dependent glucose output through decreased cAMP production (30).

Metformin improves glucose effectiveness, a measure of the ability of glucose to restore its own concentration through mass-action effects and suppression of endogenous glucose production (31). The previously accepted mechanism of improved glucose effectiveness, metformin activation of AMP-activated protein kinase (AMPK) promoting glucose uptake and fatty acid oxidation in muscle (32), is now controversial because mouse hepatocytes lacking AMPK exhibited normal metformin-induced inhibition of gluconeogenesis (33). Nevertheless, the increased AMPK occurs in response to lower cell ATP levels and energy stores (34), and ATP levels are lower in muscle after metformin treatment (35). Glucose disposal in muscle is partially insulin-independent, as demonstrated by glucose uptake in isolated human muscle biopsies and culture (36, 37). Glucose disposal increased in humans after metformin treatment, as did muscle glycogen content, and both were associated with decreased ATP and energy stores (35). Taken together, enhanced glucose-mediated glucose disposal may be related to a combination of the improved mass action of glucose and the overall lowering of glucose levels (36) brought about by the action of metformin in the inhibition of mitochondrial complex 1.

Improved glucose effectiveness may be responsible for the increased AIR_g in the current study and others (17). The AIR_g is diminished at glucose levels greater than 100 mg/dL and suppressed completely at 115 mg/dL (38). Diet-induced lowering of glucose levels improves the AIR_g (39). Therefore, metformin may improve the AIR_g by lowering glucose levels, even within the normal range in women with PCOS.

The absence of a change in insulin sensitivity may seem surprising, but may be explained by the different techniques for measuring insulin sensitivity. Using the IVGTT, results in women with PCOS have demonstrated no change in insulin sensitivity after 10 weeks to 3 months of metformin treatment (17, 40). In contrast, studies using a euglycemic hyperinsulinemic clamp suggest that insulin sensitivity improves (18, 41). The reason for the discrepancy relates to the fact that the clamp studies report glucose utilization (M), a steady-state measure of the glucose infusion, which equals the glucose translocation out of the glucose space when endogenous glucose production is suppressed by the insulin infusion (18, 41). Therefore, the M value reflects both the glucose-mediated glucose disposal (glucose effectiveness) and the insulin-mediated glucose disposal (42, 43). In insulin-resistant individuals, the glucose-mediated glucose disposal could constitute a considerable portion of glucose uptake during euglycemia and account for the M value more than the insulin-mediated glucose disposal (42). Women with PCOS are more insulin resistant than their BMI-matched counterparts (18, 41). Therefore, the glucose-mediated glucose disposal may account for a significant portion of the M value in women with PCOS, although this hypothesis was not evaluated in the current study in the absence of a BMI-matched control group.

In addition, it is important for the serum insulin levels achieved to be similar when comparing two hyperinsulinemic-euglycemic clamp studies. However, post-metformin clamp studies have lower hepatic glucose output and fasting insulin levels along with lower insulin levels during the insulin infusion resulting from increased insulin clearance compared to pretreatment studies (42, 44). Error in the M measurement is augmented by the considerable variability in insulin levels between individuals during a constant insulin infusion (18, 41–43) and the fact that insulin levels are not linear at high infusion rates, although required for analysis (42). In studies in which glucose levels are carefully controlled, patients with type 2 diabetes demonstrated no change in insulin-stimulated glucose disposal in the absence of weight loss when examined using a hyperinsulinemic-euglycemic clamp (15, 16). However, glucose utilization improved significantly during the hyperglycemic clamp, suggesting that glucose-mediated glucose disposal improved, whereas insulin sensitivity did not change (15). Taken together with data from the current study, glucose effectiveness, not insulin sensitivity, improves with metformin treatment.

The second objective of the study was to determine additional effects of metformin treatment in women with PCOS and to determine factors that predicted improvement. Along with improved glucose effectiveness, metformin treatment was associated with improved ovulation in 61% of subjects, a rate that is similar to that in previous studies (18, 41, 44, 45). Patients with lower baseline T levels (pre-hCG and pre-metformin) were more likely to have an ovulatory response supporting previous findings (41, 44, 46), although a distinct cutoff was not apparent. After metformin treatment, the T and androstenedione levels were lower in women who ovulated compared to those who did not.

Previous studies suggest that higher insulin levels, insulin resistance, and less severe menstrual abnormalities were predictors of an ovulatory response (44). A meta-analysis suggested no effect of weight (14). In the current study, improved ovulation had no relationship to weight, initial menstrual cycle frequency, FSH or LH levels, insulin levels or insulin resistance, ovarian volume, or follicle number. There was also no relationship between ovulation and increased glucose effectiveness, nor was ovulation predicted by glucose tolerance (M) in previous studies (39). The data support the concept that metformin has a direct ovarian effect on androgen levels that is not mediated through changes in gonadotropin or insulin levels. Taken together, it is possible that metformin induces ovulation by directly decreasing steroidogenesis, thereby reducing the inhibitory effects of androgens on folliculogenesis.

In addition to lower T levels after metformin treatment in women who ovulated, T levels decreased overall. Women with the highest initial T levels had the most significant reductions. The T response to metformin has been variable in previous work (17, 18, 44, 46–50), with lean women exhibiting a greater decrease (14, 49). Nevertheless, there was no relationship between BMI and response to metformin in the current study, requiring further examination of BMI as a predictive factor. The improvement in T level was also unrelated to changes in glucose levels or effectiveness and insulin or LH levels.

The mechanism explaining the ability of metformin to lower androgen levels may also be elucidated through further understanding of mitochondrial complex I. Metformin has been reported to decrease androgen levels in human theca cell cultures (51, 52). Indirect examinations have demonstrated a decrease in ovarian P450c17α activity after metformin treatment (47). Recent studies demonstrate a reduction in both CYP17A1-lyase and 3β-hydroxysteroid dehydrogenase II (3βHSDII) activity when assessed in adrenal cells, which appears to be mediated through inhibition of mitochondrial complex I of the respiratory chain (53). Taken together, metformin may decrease androgen production through inhibition of mitochondrial complex I, resulting in decreased NAD+, the cofactor required for 3βHSDII activity (33, 53).

Other parameters that improved after metformin treatment include weight, hip circumference, diastolic blood pressure, total cholesterol, and low-density lipoprotein. Weight loss has been demonstrated in other studies of metformin treatment (46). The DEXA measurements in the current study suggest that it is not fat or bone mineral mass that decreases with metformin treatment. Although the decrease may represent lean mass, previous studies suggest that metformin protects or increases lean mass in aging men and in adolescent girls (54, 55). Therefore, further studies are needed to determine the reason for the change in body weight and composition. Systolic and diastolic blood pressure decreased with metformin in other studies as well (50). Previous studies have demonstrated that metformin improved lipid profiles with increased HDL in women with PCOS (44) and decreased triglyceride levels in subjects with type 2 diabetes (15), without improvements in cholesterol levels (46). HDL and triglycerides did not improve in the current study, which may relate to lower baseline triglyceride and HDL levels in this study compared to others as well as differences in the study population and metformin dosing. None of these parameters were associated with improvement in glucose effectiveness.

There are limitations to the study. It is not a randomized, placebo-controlled trial. Menstrual cycles were monitored by history and prospectively between the screening visit and the first study visit. Therefore, anovulatory bleeding could be misinterpreted as ovulatory, resulting in a conservative estimate of improved ovulatory frequency. The treatment period is relatively short. Although the effects of metformin can be rapid (56), a more pronounced ovulatory response may have been demonstrated if the treatment duration were extended to 6 months (45, 50). Finally, the study did not test mitochondrial complex 1 function directly, and it is possible that the mechanism of glucose and T reduction may occur through alternate or tissue-specific mechanisms.

This study challenges the rationale of using metformin in subjects with PCOS to improve insulin sensitivity. Metformin improved fasting glucose and glucose effectiveness. It also improved T levels and ovulation in association with lower T levels. However, the improved glucose effectiveness and T levels were not associated with each other. Taken together with evidence that metformin lowers intracellular ATP in hepatocytes, muscle, and adrenal cells, the data point to a coordinated mechanism of metformin action related to its ability to inhibit mitochondrial complex I. Improvement in glucose effectiveness and androgen levels must be examined as separate outcome parameters in women with PCOS on metformin treatment. Importantly, further investigation to delineate the exact site and mechanism of action of metformin on mitochondrial complex I may help explain the variability in treatment response and provide insight into new therapeutic targets for the treatment of PCOS.