Amino acid and fatty acid metabolomic profile during fasting and hyperinsulinemia in girls with polycystic ovarian syndrome

Abstract

Polycystic ovarian syndrome (PCOS) is associated with insulin resistance (IR) and altered muscle mitochondrial oxidative phosphorylation. IR in adults with obesity and diabetes is associated with changes in amino acid, free fatty acid (FFA), and mitochondrial acylcarnitine (AC) metabolism. We sought to determine whether these metabolites are associated with IR and/or androgens in youth-onset PCOS. We enrolled obese girls with PCOS [n = 15, 14.5 yr (SD 1.6), %BMI 98.5 (SD 1.0)] and without PCOS [n = 6, 13.2 yr (SD 1.2), %BMI 98.0 (SD 1.1)]. Insulin sensitivity was assessed by hyperinsulinemic euglycemic clamp. Untargeted metabolomics of plasma was performed while fasting and during hyperinsulinemia. Fasting arginine, glutamine, histidine, lysine, phenylalanine, and tyrosine were higher (P < 0.04 for all but P < 0.001 for valine), as were glutamine and histidine during hyperinsulinemia (P < 0.03). Higher valine during hyperinsulinemia was associated with IR (r = 0.59, P = 0.006). Surprisingly, end-clamp AC C4 was lower in PCOS, and lower C4 was associated with IR (r = −0.44, P = 0.04). End-clamp FFAs of C14:0, C16:1, and C18:1 were higher in PCOS girls, and C16:1 and C18:1 strongly associated with IR (r = 0.73 and 0.53, P < 0.01). Free androgen index related negatively to short-, medium-, and long-chain AC (r = −0.41 to −0.71, P < 0.01) but not FFA or amino acids. Obese girls with PCOS have a distinct metabolic signature during fasting and hyperinsulinemia. As in diabetes, IR related to valine and FFAs, with an unexpected relationship with AC C4, suggesting unique metabolism in obese girls with PCOS.

INTRODUCTION

Polycystic ovarian syndrome (PCOS) is one of the leading causes of metabolic disease in the US of reproductive-aged women, affecting 6–10% of women, and its prevalence is rising in pace with obesity (28, 29). Women with PCOS accompanied by obesity have an increased prevalence of insulin resistance (IR), nonalcoholic fatty liver disease (NAFLD), and type 2 diabetes mellitus (T2DM) (2, 12, 13, 18). PCOS develops in early post-reproductive life and can be diagnosed in adolescence. At an early age, muscular, adipose, and hepatic IR are already characteristic of the syndrome, independent of BMI, but to a greater extent in obese adolescents (1, 5, 11, 30). Utilizing a multistep hyperinsulinemic euglycemic clamp with isotope tracers, we have demonstrated in our androgens and insulin resistance study (AIRS) that obese adolescents with PCOS have mild adipose and significant hepatic and muscle IR relative to obese girls without PCOS (5). PCOS-associated comorbidities are increasing in prevalence; however, an optimal understanding of the underlying pathophysiology of PCOS in adolescents is lacking, limiting effective therapeutic options.

Several metabolic alterations associated with IR are thought to relate to altered muscle substrate oxidation, particularly increased serum concentrations of branched-chain amino acids (BCAA), long-chain free fatty acids (FFA), and acylcarnitines (AC). Elevated BCAA concentrations have been found in individuals with obesity and IR (40, 44), and elevated BCAA predict future development of IR in youth (34) and T2DM in adults (51). Increases in serum BCAA are thought to be secondary to decreased oxidation in muscle with intermediates impairing insulin signaling. AC species, in particular short-chain species such as AC C4, the primary breakdown product of valine, have also been associated with IR in obesity (38). In adults with T2DM and obesity, long-chain FFA are likewise elevated secondary to decreased or incomplete fat oxidation as well as increased release from excess adipose mass and adipose IR in obesity (45).

Studies of plasma BCAA and lipid species in the PCOS population are more limited despite the fact that IR is common, especially in obese women. Combined nontargeted and targeted metabolomics approaches in fasting samples showed that all BCAA (valine, isoleucine, and leucine) were elevated in obese women with PCOS compared with obese controls (4), whereas only valine was elevated in normal-weight women with PCOS compared with BMI-matched controls (57). However, a comprehensive untargeted metabolic profile has not been investigated in obese women with PCOS. How BCAA and lipid species relate to IR and PCOS has also not been well investigated in an adolescent PCOS population, which has significant IR and is at high risk for developing T2DM (5, 10). We demonstrated that obese adolescents with PCOS have a decreased postexercise oxidative phosphorylation, suggesting decreased muscular mitochondrial function, which was associated with decreased peripheral insulin sensitivity (10). Moreover, adolescent mitochondrial function and overall substrate metabolism may be different from that of adults, as adolescents are at an earlier stage of the disease and in a different hormonal environment. Thus, the aim of the present study was to quantitate the differences in amino acids (AA), lipids, and their breakdown products fasted and in response to hyperinsulinemia in obese youth with and without PCOS. Second, we aimed to assess the associations between insulin-stimulated FFA, AA, and acylcarnitines (AC) with IR of adipose tissue, liver, and muscle in obese adolescents.

METHODS

Participant population.

Participants were selected from the obese PCOS and control groups of the AIRS (5, 10, 11). AIRS inclusion criteria included female sex, obesity (body mass index ≥95th percentile for age and sex), age 12–20 yr, and physical inactivity (<3 h/wk habitual exercise). Exclusion criteria included diabetes mellitus, alanine aminotransferase >80 IU/ml, blood pressure >140/90 mmHg, hemoglobin <9 mg/dl, serum creatinine >1.5 mg/dl, smoking, medications affecting insulin sensitivity or to treat hypertension or lipids, pregnancy, and breastfeeding. PCOS was diagnosed using the National Institutes of Health (NIH) criteria of oligomenorrhea, being ≥18 mo postmenarche and having hyperandrogenism (29, 53). This study was approved by the University of Colorado Anschutz Institutional Review Board in addition to Children’s Hospital Colorado’s Scientific Advisory Review Committee. All participants provided written, informed consent. For participants younger than 18 yr old of age, parents and participants provided consent and assent, respectively. Samples were selected for metabolomics analysis from AIRS participants with complete insulin clamp data and samples available for analysis. They were then matched on BMI between PCOS and controls, with a ratio of two girls with PCOS for each control due to limited availability of stored control samples.

Overall study design.

The protocol included two study visits, as previously described (10). Briefly, the screening visit included consent, screening laboratories, a physical examination, and an optional oral glucose tolerance test. The clamp visit was comprised of dual-energy X-ray absorptiometry (DEXA) for body composition (26) and ³¹P-MR spectroscopy of the calf during exercise for study of postexercise muscle mitochondrial function on day 1, followed by fasting plasma samples and a hyperinsulinemic euglycemic clamp on the subsequent day. Participants were in the follicular phase of the menstrual cycle if possible, verified by morning serum progesterone concentrations.

Diet.

Participants were provided a 3-day standardized, isocaloric, weight maintenance diet (55% carbohydrate, 30% fat, and 15% protein) from the University of Colorado Clinical Translational Research Center (CTRC) metabolic kitchen before the visit.

Physical activity.

All participants were asked to avoid exercising 3 days before the visit. Participants wore an ambulatory GT3x accelerometer (Actigraph, Pensacola, FL) for 7 days before the visit. The data were corrected for wear time and categorized into six activity levels (16). In addition, a 3-day physical activity recall questionnaire was used to quantify physical activity in metabolic equivalents of tasks (52).

Hyperinsulinemic euglycemic clamp with stable isotope tracers.

Fasting laboratories were drawn before the start of the clamp at 6 AM after 12 h of monitored inpatient fasting. The clamp consisted of four phases, as previously described (5, 11). The basal phase went from 6 AM until the start of insulin infusion 2 h later. For the subsequent three phases, a priming dose of insulin was given, followed by a continuous infusion of insulin for 90 min at each of the following concentrations: 10, 16, and 80 mU·m²·min⁻¹. Blood glucose was measured at the bedside every 5 min using a YSI analyzer (YSI instruments, Yellow Springs, OH) and was maintained at ≈95 mg/dl via a variable 20% dextrose infusion. Additionally, a 4.5 mg/kg [6,6-²H₂]glucose (Isotec, Miamisburg, OH) prime was administered, followed by a continuous infusion of 0.04 mg·kg⁻¹·min⁻¹ [6,6-²H₂]glucose paired with a 1.6 µmol/kg ²H₅-glycerol prime, followed by a continuous infusion of 0.11 µmol·kg⁻¹·min⁻¹ (5). Samples for isotope enrichment were drawn 10 min apart for the last 30 min of each phase. Plasma samples after fasting and the last stage of the clamp were used for metabolomics analysis.

All isotopic measurements were corrected for background enrichment. The glucose and glycerol rate of appearance (R_a), rate of disappearance (R_d), and metabolic clearance rate (MCR) over the last 30 min of each phase were calculated using the Steele non-steady-state equation, accounting for “spiked” glucose in the 20% dextrose infusion (3). Muscle insulin sensitivity was calculated by glucose MCR during the 80 mU·m²·min⁻¹ insulin phase. The insulin concentration required to suppress 50% of the basal glycerol (glycerol IC₅₀) or glucose R_a (glucose IC₅₀) was calculated for each individual (5, 11). Adipose and hepatic tissue insulin sensitivity were measured by glycerol IC₅₀ and glucose IC₅₀, respectively.

Indirect calorimetry.

Indirect calorimetry was performed during steady state at the end of each clamp phase (Carefusion, San Diego, CA).

Body composition.

Body composition by dual X-ray absobitometry (DEXA) was performed to determine percent total body fat (26).

Postexercise muscle mitochondrial function measurement.

A custom-built, 12-cm phosphate coil (Clinical MR Solutions, Brookfield, WI) was used to collect phosphate spectra of the calf at rest, during isometric plantar flexion exercise for 90 s at 70% of the maximum volitional contraction, and during 5 min of recovery postexercise on a 3 T GE (Milwaukee, WI) research magnet, as previously described (8, 9, 11). An MRI-compatible, custom-built plantar flexion device was used as the exercise device, and the force was monitored during exercise. The spectra were analyzed using the jMRUI software package with time domain fitting and using AMARES with curve fitting of the time constant data, as previously described (7–9, 27, 43). The outcomes of interest were the calculated adenosine diphosphate (ADP) time constant, phosphocreatine (PCr) time constant, calculated rate of oxidative phosphorylation, and Q_max (8, 9, 11).

Laboratory analysis.

Laboratory analyses, other than the metabolomics, were performed at the Children’s Hospital Colorado Laboratory and the University of Colorado CTRC Laboratory. Plasma glycerol (R-Biopharm, Marshall, MI), FFAs (Wako Chemicals, Richmond, VA), triglycerides, and total cholesterol (Hitachi 917 autoanalyzer; Boehringer Mannheim Diagnostics, Indianapolis, IN) were analyzed using enzymatic assay (11). Low-density lipoprotein cholesterol was determined using the Friedewald equation (17). To analyze HbA_1c, Diabetes Control and Complications Trials-calibrated high-performance ion exchange liquid chromatography was used (Bio-Rad Laboratories, Hercules, CA). Serum adiponectin, leptin, and insulin were analyzed using a radioimmunoassay (EMD Millipore, Billerica, MA). C-reactive protein was analyzed using an immunoturbidimetric assay (Beckman Coulter, Brea, CA). C-peptide and estradiol were analyzed using the chemiluminescent immunoassay (DiaSorin, Stillwater, MN, and Beckman Coulter). Alanine aminotransferase was determined via VITROS 5600 (Ortho Clinical Diagnostics, Rochester, NY). Sex hormone-binding globulin was measured using an electrochemiluminescence immunoassay (Esoterix, Calbassas Hills, CA). Total testosterone samples were analyzed using liquid chromatography-tandem mass spectrometry (LC-MS; Esoterix, Calbassas Hills, CA). The free androgen index (FAI) was calculated using total testosterone and sex hormone-binding globulin (SHBG) values.

Metabolomics.

Untargeted metabolomics analysis was performed at the Michigan Regional Comprehensive Metabolomics Resource Core. Samples were drawn in EDTA tubes and frozen immediately at −80.0°C. LC-MS was used for chromatographic separation and mass chromatography was used for mass detection. This analysis yielded both identified and unidentified compounds. Identification was made comparing masses and retention times from samples to an in-house library. Quality of analysis, assessed by visual inspection of the chromatographic traces and relative quantification of internal standards, showed that the analysis methods were both stable and reproducible across all samples.

Statistics.

Descriptive statistics for demographic characteristics, fasting laboratories, and measurements of glucose and mitochondrial metabolism are presented as means ± SD. Univariate analysis with an unpaired t-test was performed to compare participant characteristics and clinical laboratories between controls and PCOS. Raw data was log transformed and mean-centered. All variables were compared between the two groups, using the Benjamini and Hochberg’s P value correction methods to correct for multiple comparisons. False discovery rate (FDR)-adjusted P values (q values) are reported. Univariate statistical analysis with unpaired t-tests was performed to compare metabolites between controls and PCOS, and a P value of <0.05 was considered significant. Spearman’s correlation coefficients were used to examine the relationship between AA, FFA, and their breakdown products with FAI, tissue-specific insulin sensitivity, and mitochondrial measures. All statistical analyses, except for the metabolomics data, were performed using SigmaStat software version 13.1 (Systat Software, San Jose, CA). MetaboAnalyst software (http://www.metaboanalyst.ca/) was used for statistical analysis and interpretation of the metabolomics data (55, 56).

RESULTS

Participants included 21 obese girls: 15 with PCOS [age 14.5 yr (SD 1.6)] and six without PCOS [controls; age 13.2 yr (SD 1.2)]. Demographic characteristics, physical attributes, and laboratory measures are listed in Table 1. BMI percentile, waist circumference, waist/hip ratio, percent body fat, physical activity, and diet were similar between the two groups. There were no statistical differences in race and ethnicities between the groups. Androgens were significantly higher, estradiol and SHBG significantly lower, and markers of the metabolic syndrome significantly worse in girls with PCOS. Fasting glucose and fasting insulin values were similar for both groups in addition to the fasting C-peptide and HbA_1c (Table 1). As expected, muscle insulin sensitivity assessed with glucose MCR (dl·kg⁻¹·min⁻¹) during the highest dose of insulin was significantly lower in girls with PCOS compared with controls (P = 0.006). Also, muscle postexercise oxidative phosphorylation measured by ³¹P-MRS tended to be lower in girls with PCOS (P = 0.07). However, in this sample, we did not find any significant differences in glucose IC₅₀ (P = 0.16), reflecting hepatic insulin sensitivity, or in glycerol IC₅₀ (P = 0.77), reflecting adipose tissue insulin sensitivity.

Table 1.

Participant descriptors

	Control	PCOS	P Values
No.	6	15	NA
Age, yr	13.2 (1.2)	14.5 (1.6)	0.07
Ethnicity (C/H/B/O), %	(17/33/50/0)	(40/47/7/7)	NS
Menarche age, yr	10.7 (1.2)	11.4 (1.3)	0.29
%BMI	98.0 (1.1)	98.5 (1.0)	0.28
Waist circumference, cm	92 (10)	103 (10)	0.04
Waist/hip ratio	0.80 (0.34)	0.88 (0.02)	0.56
Percent body fat from DEXA	43 (4)	43 (1)	0.75
Daily mets from 3DPAR	58.8 (6.7)	56.2 (8.4)	0.50
Accelerometer: %sedentary time	0.68 (0.09)	0.70 (0.09)	0.68
Accelerometer: %lifestyle + light time	0.24 (0.14)	0.24 (0.10)	0.88
Daily kcal from FFQ	1770 (250)	1893 (150)	0.19
Daily carbohydrates from FFQ, g	249 (36)	265 (21)	0.20
Daily protein from FFQ, g	68 (10)	73 (6)	0.22
Daily fat from FFQ, g	61 (9)	65 (5)	0.19
Fasting laboratory measures
Total testosterone, ng/dl	23.8 (9.9)	46.2 (14.6)‡	0.003
SHBG, nmol/l	25.4 (11.3)	16.6 (7.6)*	0.05
Free androgen index	3.4 (0.7)	12.1 (8.3)*	0.02
Estradiol, pg/ml	93.3 (53.5)	42.5 (12.2)‡	0.002
Leptin, ng/ml	38.1 (20.4)	30.9 (10.3)	0.29
Adiponectin, ng/ml	8.3 (3.3)	5.9 (1.8)*	0.05
Cholesterol, mg/dl	130 (19)	160 (31)*	0.04
Triglycerides, mg/dl	91 (34)	135 (42)*	0.04
LDL, mg/dl	79 (15)	113 (31)*	0.02
ALT, U/l	31 (3)	41 (11)	0.05
hs-CRP, mg/dl	0.9 (0.8)	4.5 (3.8)*	0.03
Glucose metabolism
HbA1c, %	5.2 (0.3)	5.3 (0.3)	0.50
Fasting glucose, mg/dl	85 (6)	83 (5)	0.59
Fasting insulin, µIU/ml	17 (6)	23 (7)	0.11
C-peptide, ng/ml	2.8 (1.2)	3.6 (0.7)	0.08
HOMA-IR	4.1 (1.7)	5.5 (2.1)	0.16
Glucose MCR, ml·kg−1·min−1	8.4 (2.3)	5.6 (1.6)‡	0.006
IC50 glucose, IU/ml	56 (21)	77 (32)	0.16
IC50 glycerol, IU/ml	122 (116)	143 (151)	0.77
Muscle mitochondrial metabolism
PCr time constant, s	29.9 (3.9)	34.6 (8.4)	0.22
ADP time constant, s	20.1 (4.5)	25.2 (7.3)	0.13
Qmax, mmol/s	1.6 (2.7)	0.4 (0.1)	0.11
Oxidative phosphorylation, mmol·l−1·s−1	0.24 (0.18)	0.13 (0.09)	0.08

Data are presented as means (SD). ALT, alanine aminotransferase; ADP, adenosine diphosphate; B, Black; C, Caucasian; DEXA, dual X-ray absorptiometry; 3DPAR, 3-day physical activity recall; H, Hispanic; HbA_1c, hemoglobin A_1c; HOMA-IR, homeostatic model assessment of insulin resistance; hs-CRP, high-sensitivity C-reactive protein; IC₅₀, insulin concentration at which 50% suppression is achieved; LDL, low-density lipoprotein; MCR, metabolic clearance rate; NA, not available; NS, not significant; O, other; PCr, phosphocreatine; PCOS, polycystic ovarian syndrome; Q_max, maximal mitochondrial capacity; SHBG, sex hormone-binding globulin.

^*P < 0.05 and

^‡P < 0.01, significant differences.

Initial untargeted analysis of metabolomics identified 5,015 compounds, with 249 known metabolites. From fasting samples, levels of 125 metabolites were significantly different between the two groups, whereas 116 metabolites differed in response to hyperinsulinemia. Analysis of the 249 known metabolites at fasting showed only an increase in dehydroascorbic acid in girls with PCOS (q < 0.05). Under the hyperinsulinemic condition, bilirubin, bilirubin II, phytanic acid, uric acid, homovanillic acid, C4 AC, and spermine were the only known metabolites that were different between girls with PCOS and those without (q < 0.05). The overall metabolomic signature was also significantly different between the fasting and hyperinsulinemic conditions. Girls with PCOS had 68 known metabolites with greater than twofold changes between the fasting and insulin-stimulated conditions, whereas girls without PCOS had 82 known metabolites with a greater then twofold change with the clamp.

In examining AAs specifically (Fig. 1), glutamine and histidine were significantly higher in adolescents with PCOS in the fasting state (P = 0.04 and P = 0.02, respectively) and in the hyperinsulinemic state (P = 0.03 and P = 0.009, respectively). Fasting arginine, lysine, phenylalanine, and tyrosine were also significantly higher in girls with PCOS compared with controls (P = 0.003, P = 0.05, P = 0.007, and P = 0.006, respectively). However, no significant differences were found in response to hyperinsulinemia. Asparagine, methionine, tryptophan, and proline were similar between the PCOS and control groups in both conditions. In terms of BCAA, girls with PCOS had significantly higher levels of valine in response to hyperinsulinemia (P = 0.0002), whereas no significant difference was found in the fasting state. Isoleucine and leucine were similar between both groups fasting and during hyperinsulinemia.

Fig. 1.

Amino acids fasting and in response to hyperinsulinemia. Data presented are individual data [■, control (CON); ○, polycystic ovarian syndrome (PCOS)], median, and interquartile ranges, comparing obese girls with PCOS vs. without PCOS in both the fasting and hyperinsulinemic state. *P < 0.05, **P < 0.01, significant differences.

Analyzing AC specifically (Fig. 2), fasting short-chain AC C3 and C4, medium-chain AC C6 and C8, and long-chain AC C18:2 were all lower in girls with PCOS compared with girls without PCOS (P = 0.004, P = 0.004, P = 0.002, P = 0.04, and P = 0.02, respectively). Under hyperinsulinemic conditions, short-chain AC C4 and long-chain AC C20 were also significantly lower in girls with PCOS (P = 0.009 and P = 0.008, respectively).

Fig. 2.

Acylcarnitines fasting and in response to hyperinsulinemia. Data presented are individual data [■, control (CON); ○, polycystic ovarian syndrome (PCOS)], median, and interquartile ranges, comparing obese girls with PCOS vs. without PCOS in both the fasting and hyperinsulinemic state. *P < 0.05, **P < 0.01, significant differences.

Total fasting FFAs were not different between groups [632 (463–727) vs. 647 (434–708), P > 0.99]. However, fasting saturated FFAs (Fig. 3) C18:0 (stearic acid) and C20:0 (arachidic acid) were lower in obese girls with PCOS (P = 0.008 and P = 0.03, respectively). Fasting unsaturated FFA C18:2 (linoleic acid) was significantly elevated, whereas FFA C20:5 (cis-11-eicosenoic acid) and C22:1w9 (erucic acid) were significantly lower in girls with PCOS (P = 0.008, P = 0.011, and P = 0.012, respectively). In response to hyperinsulinemia, total FFAs tended to be higher in girls with PCOS [56 (34–97) vs. 38 (14–53), P = 0.09]. FFA C14:0 (myristic acid) was significantly elevated in girls with PCOS (P = 0.009) and had a lower percent suppression under hyperinsulinemia [55 (49–62%) vs. 63.5% (59.5–70%), P = 0.05]. Whereas the levels of C16:0 and C18:0 were the same at the end of the clamp in both groups, the percentage of suppression from the fasting to the hyperinsulinemic condition was lower in girls with PCOS (C16:0 84 vs. 90%, P = 0.04, and C18:0 51 vs. 62.5%, P = 0.02). Unsaturated FFAs C16:1 (P = 0.05) and C18:1 (P = 0.003) were also significantly higher in girls with PCOS in the hyperinsulinemic state. No significant differences were observed in the other saturated or unsaturated FFAs at the end of the clamp. Fasting branch-chained FFA phytanic acid was significantly lower in girls with PCOS in comparison with controls (P = 0.0006), whereas no significant difference was noted during the hyperinsulinemic state.

Fig. 3.

Free fatty acid fasting and in response to hyperinsulinemia. Data presented are individual data [■, control (CON); ○, polycystic ovarian syndrome (PCOS)], median, and interquartile ranges, comparing obese girls with PCOS vs. without PCOS in both the fasting and hyperinsulinemic state. *P < 0.05, **P < 0.01, significant differences. DHA, [³H]docosahexaenoic acid.

Spearman’s correlations between FAI and fasting AA, AC, and FFA were performed within the entire sample population. Significant associations are shown in Fig. 4. FAI related only to AC metabolites and none of the AAs or FFAs. FAI negatively correlated to short-chain AC C3 (r = −0.57, P = 0.003) and C4 (r = −0.70, P < 0.0001), medium-chain C6 (r =−0.54, P = 0.01), and long-chain AC C18:1 (r = −0.56, P = 0.008) and C18:2 (r = −0.59, P = 0.005). Fasting medium-chain AC C10 (r = −0.41, P = 0.06) and C14 (r = −0.42, P = 0.06) had negative correlations with FAI that were not statistically significant.

Fig. 4.

Correlation between acylcarnitine metabolites and free androgen index. Data are presented showing the results of Spearman correlations between free androgen index and acylcarnitine metabolites. ■, Controls; ○, polycystic ovarian syndrome. r = Correlation coefficient. AUC, area under the curve.

Spearman’s correlations between tissue-specific IR and metabolites of interest under hyperinsulinemic conditions are shown in Table 2. Adipose tissue IR was negatively associated with glutamine (r = −0.52, P = 0.02) and positively associated with long-chain AC C18:1 (r = 0.48, P = 0.03) and C20:0 (r = 0.51, P = 0.02). Hepatic IR strongly correlated with higher isoleucine (r = 0.62, P = 0.003), long-chain AC C16:0 (r = 0.56; P = 0.009) and C18:1 (r = 0.48, P = 0.03), and long-chain FFA C16:1 (r = 0.55; P = 0.01) and C18:2 (r = 0.45; P = 0.04). Peripheral IR was associated with higher end-clamp medium and long-chain AC C6, C8, C10, C14:0, C16:0, and C18:1 but with lower short-chain AC C4 (r = −0.48, P = 0.03). Conversely, higher valine, the direct precursor of C4, was associated with peripheral IR (r = 0.55; P = 0.01). AA phenylalanine (r = 0.54, P = 0.01) and FFA C16:1 (r = 0.77, P < 0.0001), C18:2 (r = 0.58, P = 0.007), and cis-11-eicosenoic acid (r = 0.44, P = 0.05) were also associated with peripheral IR.

Table 2.

Correlation between amino acids, free fatty acids, acylcarnitine metabolites, and tissue specific insulin resistance

	Adipose IR IC50 Glycerol	Hepatic IR IC50 Glucose	Peripheral IR (1/MCR, dl·kg−1·min−1)
Amino acids
Glutamine	r ≥ 0.5, P = 0.049–0.01*
Isoleucine		r ≥ 0.5, P = 0.01–0.001*
Phenylalanine			r ≥ 0.5, P = 0.049–0.01†
Valine			r ≥ 0.5, P = 0.049–0.01†
Acylcarnitines
C4			r ≥ 0.5, P = 0.049–0.01*
C6			○
C8			r ≥ 0.5, P = 0.049–0.01†
C10			r ≥ 0.5, P = 0.049–0.01†
C14:0			r ≥ 0.5, P = 0.049–0.01†
C16:0		r ≥ 0.5, P = 0.01–0.001*	r ≥ 0.5, P = 0.049–0.01†
C18:1	r < 0.5, P = 0.049–0.01†	r < 0.5, P = 0.049–0.01†	r ≥ 0.5, P = 0.049–0.01†
C20:0	r ≥ 0.5, P = 0.049–0.01†
Free fatty acids
C16:1			r ≥ 0.5, P < 0.001†
C18:2			r ≥ 0.5, P = 0.01–0.001†
Cis-11-eicosenoic acid			r ≥ 0.5†

Data are presented showing the results of Spearmen correlations between makers of tissue-specific insulin sensitivity, and amino acids, free fatty acids, and their breakdown metabolites. IC₅₀, insulin concentration at which 50% suppression is achieved; IR, insulin resistance; MCR, metabolic clearance rate.

^*Significant negative associations;

^†significant positive associations.

In terms of muscle mitochondrial function, fasting tyrosine (r = −0.56, P = 0.009), arginine (r = −0.55, P = 0.009), and tryptophan (r = −0.45, P = 0.040) were negatively associated with ³¹P-MRS-assessed oxidative phosphorylation. C16:1 (r = 0.48, P = 0.03), C20:5 (r = 0.52, P = 0.02), and C20 (r = 0.56; P = 0.01) positively correlated with oxidative phosphorylation.

The rate of whole body fat oxidation under hyperinsulinemia related strongly to valine (r = 0.72, P < 0.001) and to long-chain C18:0 (r = 0.52, P = 0.02), C16:1 (r = 0.52, P = 0.02), and C16:0 (r = 0.48, P = 0.03).

DISCUSSION

Obese girls with PCOS defined by NIH criteria are at a higher risk for metabolic disease, including significant IR and future development of T2DM (5, 29). We have previously shown that this patient cohort, relative to BMI similar controls, has significant skeletal muscle and hepatic and adipose insulin resistance in addition to postexercise skeletal muscle mitochondrial dysfunction (5, 10). Using metabolomics, we now describe here differences in several plasma AA, FFA, and AC metabolite levels in obese girls with PCOS compared with age, pubertal stage, diet, physical activity, and BMI-similar controls both during fasting and following a hyperinsulinemic clamp. To our knowledge, we are the first to evaluate AA, FFA, and AC levels in obese adolescents with PCOS. During fasting, several essential and nonessential AAs were higher in girls with PCOS, with no differences in BCAA between groups but decreased short-chain AC. During hyperinsulinemia, similar to findings in other populations with IR, such as obesity and T2DM, valine (a BCAA) and some long-chain FFA (C14:0, C16:1, and C18:1) were elevated and related to hepatic and/or peripheral IR. Furthermore, insulin-stimulated levels of several short-, medium-, and long-chain acylcarnitines were associated with peripheral IR but were similar between groups. However, insulin-stimulated AC C4 was reduced in obese girls with PCOS and lower levels related to IR and higher androgens. This specific signature of fasting and hyperinsulinemic metabolomic profiles may help elucidate the unique metabolism that contributes to long-term metabolic disease in girls with PCOS.

BCAA and breakdown products.

We demonstrated that whereas there were no significant differences between groups in BCAA during fasting, valine was significantly higher in obese girls with PCOS during hyperinsulinemia. The few studies of metabolomic profiles in PCOS have found different results concerning BCAA from each other but were limited by small sample sizes (4) of enrolled participants, with variable criteria to define PCOS, including the Rotterdam criteria, which does not require elevated androgens and did not control for dietary intake or activity, which can significantly affect measures of BCAA. Two studies showed higher fasting BCAA (only valine or all BCAA, including leucine, isoleucine, and valine) (4, 57) in obese women with PCOS, and another found the opposite, with lower levels of fasting isoleucine and valine in women with PCOS in comparison with BMI-matched controls (49). Thus, it is not clear whether BCAA are systematically affected by PCOS status, but in our sample of obese, physically inactive adolescents with PCOS, we did not find any difference in fasting BCAA concentrations.

However, the associations between fasting BCAA and IR measured by oral glucose tolerance test or fasting indices has been shown consistently in obese and lean youth and adults (21, 32, 38, 40, 48). Furthermore, prospectively, elevated fasting concentrations of BCAA and aromatic AAs predict future IR in young adults (54). We confirmed that insulin-stimulated valine strongly relates to skeletal muscle IR assessed with gold standard hyperinsulinemic euglycemic clamp in youth, but there was no association between peripheral IR and leucine or isoleucine. However, there was a strong correlation between isoleucine and hepatic IR. To our knowledge, no other studies have looked at tissue-specific IR or associations with different BCAA in PCOS.

Short-chain AC were generally lower in obese girls with PCOS compared with controls. AC C3 was significantly lower in PCOS during fasting and tended to be lower in the insulin-stimulated state. Whereas the BCAA valine was higher in girls with PCOS during insulin stimulation, C4 AC, its direct mitochondrial metabolite, was significantly lower in girls with PCOS during fasting and hyperinsulinemia. This signature differs from results observed in various metabolomics studies in adults with obesity, metabolic syndrome, T2DM, and IR, where BCAA and short-chain AC are both elevated (38). The elevation in both BCAA and short-chain AC is thought to result from lower catabolism in hepatic and adipose tissue, increased production by the gut microbiome (39), and substrate mitochondrial overload. Similar to our results, fasting AC were lower in obese youth with T2DM; however, these youth had lower BCAA, suggesting increased complete catabolism (35). Our results thus reveal a signature specific to adolescent PCOS, and this is supported by the strong negative associations between FAI and short-, medium-, and long-chain AC. In support of our findings, a study investigating the effects of pioglitazone treatment on AC in obese women with PCOS also showed lower levels of fasting AC compared with healthy BMI-matched controls (50). Furthermore, treatment with pioglitazone for 16 wk led to increased insulin sensitivity in addition to an increase in the levels of AC, especially AC C3, which is a derivative of BCAA isoleucine and valine catabolism. Alternatively, valine can also be catabolized into either β-aminoisobutyric acid (BAIBA) or 3- hydroxyisobutryate (3-HIB) (39), and as it was not possible to measure those metabolites in our sample, we cannot exclude the possibility that valine was converted more into these metabolites instead of AC C4. Perhaps further study, including tracer methodologies, is needed to better understand valine metabolism in women with PCOS, especially as it relates to insulin sensitivity.

Non-BCAA amino acids.

Several fasting AA levels other than BCAA were higher in girls with PCOS, including nonessential (arginine, glutamine) and essential amino acids (histidine, lysine, phenylalanine) as well as tyrosine. Elevations of many of these AA have been associated with IR in the literature (15, 20, 36). In our sample, only insulin-stimulated phenylalanine was related to peripheral IR. Interestingly, a treatment consisting of an insulin sensitizer, pioglitazone, plus metformin in women with PCOS led to a reduction in lysine and phenylalanine (23). Altogether, it is possible that muscle IR also manifests as lower stimulated protein synthesis, and the elevation of EAA may reflect net protein breakdown, and this potential mechanism needs to be further explored in adolescents with PCOS.

FFAs.

Lipidomic studies in PCOS are conflicting, confounded by variable degrees of IR and obesity between groups. However, most studies argue toward an increase in fasting saturated FFA in women with PCOS in comparison with controls (22, 25, 31). In our sample, whereas total fasting FFAs were similar between groups, obese girls with PCOS conversely had lower saturated FFAs species (C18:0 and C20:0), unsaturated FFAs (C20:1n9 and C221w9), and branched-chain fatty acid phytanic acid. Also, medium-chain saturated FFAs (C12:0 and C14:0) tended to be lower. The discrepancies with the literature of lower levels of some FFA species in girls with PCOS may relate to the study-provided standardized diet for 3 days before sample collection, which affects the acute but not chronic impact of diet on the FFA profile. Additionally, women with PCOS demonstrated lower fasting saturated FFA (C12:0-C20:0), suggesting a unique fasting FFA profile in PCOS (33). However, the lower fasting C18:0 with a trend toward higher C18:1 and a higher ratio C18:1/C18:0 in girls with PCOS may be explained by upregulation of stearoyl-CoA desaturase enzyme-1 (SCD-1). This enzyme is responsible for desaturation of C18:0 to C18:1 or C16:0 to C16:1. It is also upregulated in liver and adipose tissue of obese IR rodents and is thought to be important in the development of hepatic IR (19, 46). Our findings may imply that liver SCD-1 activity is increased in PCOS independently of obesity. We have previously shown that C16:1 n7 in adolescents with PCOS was higher than in controls (5), and the fact that we do not see any difference in this sample may represent a power issue. On the other hand, we also found that fasting C18:2 was higher in girls with PCOS, which was associated with decreased C18:2 acylcarnitines. The increase in C18:2 has also been found consistently in other studies comparing obese women with PCOS to obese controls (4, 14). As suggested by the decrease in C18:2 acylcarnitines, a different endogenous metabolism of C18:2 in girls with PCOS is likely, but a difference in chronic diet intake between groups cannot be excluded.

Similar to what we found with short-chain AC, fasting medium-chain AC, reflecting incomplete β-oxidation of longer-chain fatty acids through mitochondria, were lower in girls with PCOS. Again, this is in opposition to data from adult populations with IR but in line with what has been shown in adolescents with T2DM (35). The decreased fasting AC, together with some lower fasting long-chain saturated and unsaturated FFA, may suggest increased fasting β-oxidation of some species of saturated FFA as well as increased desaturation of stearic acid (C18:0) to oleic acid (C18:1) in girls with PCOS.

We observed that C14:0, C16:1, and C18:1 were higher under hyperinsulinemia in girls with PCOS, and they were correlated with IR. Furthermore, unsaturated C18:2 FFA was also strongly associated with IR. Although medium- and long-chain AC were not different between groups, they were associated with IR, reflecting incomplete mitochondrial β-oxidation of long-chain lipids or overload of mitochondrial products as contributors to IR.

Surprisingly, the levels of plasma AC and saturated FFA were not associated with postexercise mitochondrial function measured by muscle ³¹P-MRS. However, plasma levels are a mixed reflection of processes throughout the body, including not only muscle tissue but hepatic metabolism as well. There is evidence that increased hepatic mitochondrial oxidation and TCA cycle activity occurs in the early stages of hepatic steatosis, obesity, and metabolic syndrome (47). Obese adolescents with PCOS in the AIRS cohort have a 50% prevalence of hepatic steatosis (5). Altogether, these data may suggest increased fasting hepatic FFA turnover in obese girls with PCOS.

Effect of androgens.

The significant differences in the metabolomic profile between girls with and without PCOS seen in the current study may be related to elevated androgens, as this is the defining feature of PCOS. BCAA and their AC breakdown products are elevated in adolescent males compared with females with a similar BMI, suggesting that testosterone contributes to differences in BCAA (37). In adults, men also had higher fasting medium- and long-chain AC compared with women (41). However, in our sample, AC, including short, medium, and long chain, related in the opposite way to FAI. This suggests that androgens may influence AC levels, fasting FFA β-oxidation, and mitochondrial handling of BCAA in a different way in obese female adolescents with PCOS. This is consistent with evidence demonstrating increased whole body lipid oxidation in hypogonadic men treated with testosterone (42). Whereas increases in the proportion of the mitochondrial fraction of the muscle cell related to increases in serum testosterone, there was no apparent change in the number or activity of mitochondrial enzymes, genes involved in biogenesis, or lipid metabolism (24, 42). Thus the exact role of excess or supplemental testosterone in mitochondrial function per se remains unclear. Of note, girls with PCOS have testosterone concentrations between 40 and 125 ng/l in comparison with testosterone concentrations of 400–700 ng/dl in men. This difference in degree of testosterone concentrations may explain a different relationship between testosterone and muscle protein turnover in girls with PCOS.

Liver, muscle, and adipose tissue contribution.

The literature in metabolic disease has focused mainly on the contribution of the different tissues to the BCAA plasma pool (39). In general, it is has been suggested that elevated serum BCAA concentrations may reflect increased production by gut microbiome in combination with reduced utilization by adipose tissue and liver, but not in skeletal muscle, in people with obesity and insulin resistance. However, in our sample of obese adolescents we found a positive association between muscle IR and elevated valine levels and between isoleucine levels and hepatic IR arguing toward differential contribution of tissue-specific metabolism of the different BCAAs. Further complex isotope tracer studies are needed to better model specific amino acid and lipid metabolism in hepatic and muscle tissues. Understanding these tissue-specific alterations in metabolism is critical for the development of new therapeutics to treat IR, in particular in PCOS and/or adolescence, where physiology seems different from what have been shown in obese adults.

Our study has several strengths and weaknesses. The participants included in this study were extensively phenotyped, and this is the first time that multiple-stage hyperinsulinemic euglycemic clamp data, fat oxidation with indirect calorimetry, and muscle mitochondrial metabolism with ³¹P-MRS have been combined with metabolomics analyses. Furthermore, our participants were provided a controlled diet and restricted from exercise, both of which can affect metabolomic profiles. Our participants were tightly classified in terms of PCOS status and the groups matched for BMI. Additionally, our samples were processed at one of the six NIH metabolomics cores and had not undergone multiple freeze/thaw cycles. However, due to the intensive protocol, we do have a limited sample size and a trend for racial and ethnic differences between the groups. Furthermore, we performed an untargeted analysis, whereas targeted analysis for AA, FFA, and AC would allow for comparisons of true concentrations, not just relative levels between the groups. Whereas we have similar stored samples in normal-weight youth with PCOS, we did not have the funding to include them in this analysis, and future directions could include similar analysis to distinguish the role of obesity in PCOS.

In conclusion, obese girls with PCOS have a metabolomic phenotype similar to other populations with significant IR, namely fasting elevation in EAA, lower insulin-stimulated elevations in valine, and unsaturated long-chain FFA (C16:1, C18:1), all of which relate to the degree of peripheral IR. However, girls with PCOS uniquely have decreased fasting short-, medium-, and long-chain Ac C18:2 as well as insulin-stimulated valine breakdown products (AC C4). Testosterone concentrations relate to several AC levels, pointing toward a link between testosterone with whole body mitochondrial handling of the various substrates. Further work is needed to determine how these unique patterns of metabolism reflect the increased risk for metabolic diseases in these girls and whether a treatment approach unique to PCOS can be developed.

GRANTS

K. J. Nadeau received National Center for Research Resources (NCRR) Grant K23 RR020038-01, NIH/NCRR Colorado Clinical Translation and Science Institute (CTSI) Co-Pilot Grant TL1 RR025778, National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Grant 1R56-DK-088971-01, JDRF Grant 5-2008-291, and American Diabetes Association Grant 7-11-CD-08. M. Cree-Green received American Heart Association Grant 13CRP-14120015, a Thrasher Pediatric Research Foundation Mentored Pilot grant, NIH/NCRR Colorado CTSI Co-Pilot Grant TL1 RR025778, a Pediatric Endocrinology Society Fellowship, NIDDK T32-DK-063687, Building Interdisciplinary Research Careers in Women’s Health Grant K12HD057022, NIDDK Grant K23-DK-107871, Doris Duke Foundation Grant 2015212, MRC2, and Michigan Regional Comprehensive Metabolomics Pilot Grant DK-097153. This research was also supported by NIH/National Center for Advancing Translational Sciences Colorado Clinical and Translational Science Awards Program Grant No. UL1 TR001082.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

M.C.-G. conceived and designed research; M.C.-G. and Y.G.-R. performed experiments; M.C.-G., A.-M.C., H.R., L.P., and K.J.N. analyzed data; M.C.-G., A.-M.C., H.R., B.C.B., L.P., and K.J.N. interpreted results of experiments; M.C.-G., A.-M.C., and H.R. prepared figures; M.C.-G., A.-M.C., and H.R. drafted manuscript; M.C.-G., A.-M.C., Y.G.-R., B.C.B., L.P., and K.J.N. edited and revised manuscript; M.C.-G., A.-M.C., H.R., and K.J.N. approved final version of manuscript.