Polycystic ovary syndrome (PCOS) is a complex and heterogeneous endocrine disorder with well established metabolic abnormalities. The prevalence of PCOS is approximately 5% to 10% in reproductive-age women [1]. Hyperandrogenism, oligomenorrhea, chronic anovulation, and hyperinsulinemia are common clinical manifestations of PCOS [2]. Women with this syndrome have an increased risk of developing type 2 diabetes mellitus (DM2) and cardiovascular disease (CVD) [1, 3, 4]. The pathogenesis of PCOS remains a mystery, although it is considered as a polygenic trait that might result from the interaction of susceptible genomic variants and environmental factors [5, 6].

At present, alterations in several metabolic pathways have been implicated in the pathophysiology of PCOS, including abnormalities in steroid hormone regulation and insulin signaling pathway [7–9]. Moreover, there is increasing focus on the complications associated with metabolic disturbances among women with PCOS. Obesity, insulin resistance, dyslipidemia and inflammation have been recognized as risk factors for developing DM2 and CVD in PCOS [10, 11]. This highlights the need to understand the metabolic dysfunction in PCOS for prevention of long-term complications through appropriate screening, diagnosis and intervention.

In addition, the diagnosis of PCOS has also been a controversial issue [12]. Following the 2003 Rotterdam criteria [13], PCOS can be divided into four different phenotypes: (A) hyperandrogenism, chronic anovulation and polycystic ovaries (HA+AO+PCO); (B) chronic anovulation and polycystic ovaries but no clinical or biochemical hyperandrogenism (AO+PCO); (C) hyperandrogenism and chronic anovulation but normal ovaries (HA+AO); (D) hyperandrogenism and polycystic ovaries but ovulatory cycles (HA+PCO). However, it is not clear whether routine characteristics of metabolic abnormality markers could be found in all these phenotypes.

The systemic changes that occur in PCOS reflect not only changes in ovarian function, but also changes in whole-body metabolism. Metabolomics enables the characterization of endogenous small molecules that serve as direct profiles of biochemical activity and correlate with the phenotype [14]. As compared with genomics, transcriptomics and proteomics, metabolomics provides the most predictive biomarkers of this disease. In order to investigate potential metabolic profiles for different phenotypes of PCOS, as well as for the early prognosis of complications, we conducted a plasma metabolomic analysis to investigate the metabolic changes underlying different phenotypes of PCOS using both ¹H nuclear magnetic resonance (NMR) and gas chromatography/time-of-flight mass spectrometry (GC/TOF-MS); these demonstrated to be complementary analytical technologies [15].

The etiology of PCOS is complex, and the present results demonstrated the changes of metabolite profiles in the different PCOS phenotypes, which reflected the metabolic heterogeneity of the PCOS population and offered potential to study the underlying causes. Our findings clearly show that PCOS is associated with aberrations in carbohydrate metabolism. The significant elevation of lactate and glucogenic amino acids and the reduction of glucose in PCOS plasma implied elevated glycolysis in muscle and decreased gluconeogenesis in liver during PCOS pathogenesis. The strongly positive correlation of lactate level to insulin resistance further suggested insulin stimulated glucose uptake and consumption in the muscle of these PCOS patients (Additional files 1 and 4).

In terms of lipid metabolism, subjects with PCOS had higher triglycerides, LDL and VLDL levels and a lower HDL level (Tables 1 and 2), which is consistent with previous reports [20, 21] and manifested lipid disorders and dyslipidemia development. Moreover, plasma metabolic profiles in our results indicated the dramatically increased levels of three long-chain fatty acids (palmic acid, stearic acid, linoleic acid) in PCOS samples compared with the controls, irrespective of obesity or insulin resistance (Table 3 and Additional file 1). Previous reports have suggested the levels of linoleic acid in the follicular fluid significantly decreased during follicle size increase in cattle, and linoleic acid supplementation could inhibit bovine cumulus expansion, leading to reduce oocyte maturation and developmental potential [22]. The increase of linoleic acid levels in PCOS plasma may be accompanied by a similar change in the follicular fluid, thus resulting in blocked oocyte maturation and ovulation in PCOS. Additionally, linoleic acid displayed potent proinflammatory activities [23], so the higher level of linoleic acid might be not only linked with increased lipolysis and ovarian dysfunction, but also to the chronic low-grade inflammation in PCOS patients.

Additionally, our study showed different amino acid profiles in PCOS phenotypes for the first time, and the distinct patterns of free amino acids in PCOS and control subjects in the current study provided us important biochemical information and metabolic signatures that enabled the diagnosis of PCOS. More recently, some prospective studies have reported potential amino acid biomarkers for IR and diabetes. Newgard et al. [24] reported that circulating concentrations of branched-chain amino acids (Val, Leu, Ile) contributed to development of obesity-associated insulin resistance. Wang et al. [25] identified 5 branched-chain and aromatic amino acids (Val, Leu, Ile, Phe, Tyr) from 61 metabolites profiled as the markers of insulin resistance and predictors of the future development of DM2. Further, Wurtz et al. [26] reported the alterations in branched-chain and aromatic amino acid metabolism precede hyperglycemia in the general population. In our study, glycine was a novel amino acid we found which was closely related to IR except for valine and leucine, (Additional file 2). Elevated level of valine and reduced level of glycine were also observed in the women with PCOS without insulin resistance as compared with controls, and these changes were further aggravated when the patients had impaired insulin sensitivity (Additional file 4), which underlined valine and glycine were associated with other metabolic disturbances except for IR in the development of PCOS. Additionally, glycine has been shown to improve the proinflammatory profile and upregulate adiponectin gene expression in vitro [27]. Adiponectin levels seem to be lower in women with PCOS compared with non-PCOS controls after controlling for BMI-related effects [28]. Consequently, the reduced level of glycine might downregulate the expression of adiponectin and lead to the inflammation in women with PCOS independently of obesity. Thus, glycine could also be useful as a modulator of the inflammatory state observed in PCOS.

Leucine has been shown to rescue insulin signaling via activation of the mTOR pathway, and increasing dietary leucine intake can improve insulin sensitivity and restore many metabolic abnormalities [29, 30]. Leucine uptake gradually increases during follicle development, whereas this increasing rate decreases in preovulatory follicles [31]. Considering there was no significant difference of leucine level between women with PCOS with normal insulin sensitivity and control subjects in our results, we speculated that alterations of leucine plasma level in PCOS patients with IR might be entirely due to the impairment of insulin signaling. In terms of aromatic amino acids, the obvious changes of AAA levels and BCAA/AAA ratio were independent of insulin resistance and obesity, which were inconsistent with other findings in DM2 [25, 26]. Although IR is a common manifestation of PCOS and women with PCOS have an increased risk of developing DM2, the pathogenesis of PCOS and DM2 were entirely different, which was indicated by these distinct amino acid profiles in women with the two diseases.

Some previous reports have shown the clinical and endocrine disorders in different PCOS phenotypes [32, 33]. In this study, the classic phenotype of PCOS was associated with more adverse biochemical and metabolic changes than other phenotypes when compared with controls. Alterations of LH level, LH/FSH ratio, AAA levels and BCAA/AAA ratio were much more severe in classic PCOS than in other PCOS phenotypes (Tables 1 and 3).

Moreover, ovulatory PCOS phenotype had different changes of metabolic profile than the anovulatory PCOS phenotypes. The total concentration of endogenous amino acids was suppressed in ovulatory women with PCOS compared with the control group (Table 3), which demonstrated increased protein synthesis in those patients as Carmina et al. had reported that lean muscle mass was increased in women with PCOS [34]. However, the total level of endogenous amino acids was elevated in PCOS patients accompanied with the clinical feature of polycystic ovary and anovulation, indicating elevated protein degradation during ovarian dysfunction. We further noted ovulatory dysfunction of PCOS patients with raised production of the following amino acids: serine, threonine, phenylalanine, tyrosine and ornithine were significantly elevated only in the anovulatory PCOS subgroups, which implied that enhancements of these five amino acids might be directly related to ovulatory dysfunction by their increased ovarian uptake in PCOS patients. The main pathway to de novo biosynthesis of serine starts with the glycolytic intermediate 3-phosphoglycerate, so the increase of serine in anovulatory PCOS patients probably arises from increased glycolysis. Specially, the levels of serine and threonine were obviously reduced in ovulatory PCOS subtype, which might be due to the significantly negative correlations of serine and threonine to obesity and insulin resistance (Additional files 3 and 4). The concentrations of these two amino acids were indeed inhibited in PCOS patients with obesity or insulin resistance as compared with PCOS controls, respectively (Additional files 1 and 2). All of these findings together confirmed the metabolic heterogeneity of PCOS due to various clinical features. In addition, the elevated aromatic amino acids levels and decreased BCAA/AAA ratio in anovulatory patients were strongly related to the severity of the PCOS phenotypes (Table 3). In relation to this point, the significantly decreased BCAA/AAA ratio may be considered as a crucial marker of the development of PCOS.

Furthermore, androgen excess was the most common phenotype of PCOS and was somehow associated with insulin resistance. However, we observed different roles of these two phenotypes on metabolic components of PCOS. Androgen excess was closely related to the lipid metabolic disorder, the levels of three long-chain fatty acids (for example, palmic acid, stearic acid, linoleic acid) were all significantly reduced in PCOS patients with hyperandrogenism compared with women with PCOS without clinical or biochemical hyperandrogenism (Additional file 5), whereas no significant difference of these fatty acids levels were observed between PCOS cases with IR and without IR. Additionally, lactate, leucine and glycine were closely related to insulin resistance of PCOS, but the levels of these metabolic components were not influenced by androgen excess of PCOS (Additional files 4 and 5).

Otherwise, our results of metabolic signature in PCOS are partially inconsistent with previous reports. Escobar-Morreale et al. found PCOS was associated with decreased alanine concentrations [35], but GC/TOF-MS analysis in our data revealed obviously elevated levels of alanine in PCOS plasma. Alanine is transferred to the circulation mainly by skeletal muscle. There are two main pathways of alanine production: directly from protein degradation, and via the transamination of pyruvate by alanine aminotransferase (ALT). Women with PCOS have been implicated to have higher levels of ALT in the serum [36], which could accelerate the transamination of pyruvate to alanine. Additionally, increased expression of pyruvate dehydrogenase kinase 4 (PDK4) mRNA in PCOS patients [37] can enhance the peripheral concentration of this enzyme and subsequently promote the conversion of pyruvate to lactate, supporting the higher lactate concentration and glycolytic rate in our results. In addition, all control subjects have normal weight and insulin sensitivity, and we need samples from control women with obesity or insulin resistance for comparison to further analyze the effect of obesity and insulin resistance on the metabolic changes in PCOS. Another limitation is that we did not perform the subsequent replication using more samples. We are now enlarging our sample size to confirm these findings and trying to make a possible diagnostic model using the combined values of differentiate metabolites.

PCOS is a persisting challenge for clinical and basic research to elucidate its origins and distinguish primary pathological changes from secondary environmental disruptions. This study has important significance not only in identifying novel metabolic biomarkers that predict PCOS and its long-term complications, but also in better understanding of the metabolic pathways directly affected during PCOS occurrence, which may provide a rational basis for the development of novel therapeutic targets. Further correlation-based network analysis may help us to generate a diagnostic index of PCOS based on the plasma concentrations of amino acids and other metabolites detected in this study.

JQ, LF and RL participated in the study concept and design, analysis and interpretation of data, and critical revision of the manuscript. YZ did the statistical analyses and wrote the manuscript. LF, L-NW, YY, YW and PL helped to collect the samples. N-NL and C-MZ completed the clinical information. B-BT and XZ participated in sample preparation. All authors are guarantors of the work. All authors participated in the revision and final approval of the manuscript, and had full access to the data of the study.