Metabolic Stress Testing Reveals Persistent Lipid‐Handling Dysfunction in Women With Polycystic Ovary Syndrome Despite Exercise Training

ABSTRACT

Aim

Polycystic ovary syndrome (PCOS) is associated with insulin resistance and metabolic dysfunction, yet baseline metabolomic studies show inconsistent findings. We investigated whether metabolic abnormalities in PCOS emerge under physiological stress and how these responses are modified by exercise training.

Materials and Methods

Twelve women with PCOS and 10 controls completed hyperinsulinaemic‐euglycaemic clamps with randomised saline or lipid infusion, before and after 8 weeks of supervised exercise. Plasma metabolomics (163 metabolites) were measured at baseline, post‐infusion and post‐clamp. Linear mixed‐effects models assessed Group × Timepoint × Intervention interactions.

Results

No baseline metabolite differences were observed between groups. A significant three‐way interaction (p = 0.008) indicated condition‐dependent trajectory divergence. Post hoc analysis revealed a specific divergence during post‐exercise lipid challenge (p = 0.048). Women with PCOS showed reduced suppression of ether‐linked phosphatidylcholines during insulin‐stimulated lipid loading (PC ae C44:4, p = 0.031), despite exercise‐induced improvements in fitness and normalisation of amino acid profiles. Exploratory metabolite ratios suggested impaired substrate coordination under stress.

Conclusions

Metabolic defects in PCOS are stress‐dependent and not detectable at rest. Exercise training improves resting metabolism but reveals persistent impairment in adaptive lipid handling during combined insulin and lipid challenges, suggesting impaired coordination of substrate supply during metabolic stress.

Trial registration: ClinicalTrials.gov identifier: ISRCTN42448814

1. Introduction

1.1. PCOS and the Burden of Metabolic Dysfunction

Polycystic ovary syndrome (PCOS), recently renamed polyendocrine metabolic ovarian syndrome (PMOS), represents the most prevalent endocrine disorder among women of reproductive age, affecting approximately one in eight of this population globally [1]. The syndrome is characterised by reproductive dysfunction, hyperandrogenism and metabolic abnormalities, with insulin resistance serving as a central pathophysiological feature present in up to 75% of affected women [2]. Hyperinsulinaemic‐euglycaemic clamp studies have shown that insulin resistance is an intrinsic feature of PCOS and is only partly explained by adiposity [3]. Mechanistic studies have also identified abnormalities in skeletal muscle insulin signalling that may contribute to the insulin‐resistant phenotype [4]. Despite the recognised association between PCOS and metabolic dysregulation, the mechanisms underlying impaired lipid handling and the metabolic effects of lifestyle intervention remain incompletely characterised.

1.2. Metabolomic Insights Into PCOS

Metabolomic profiling has emerged as a powerful tool for characterising metabolic perturbations in PCOS, offering molecular resolution beyond conventional clinical biomarkers [3]. Cross‐sectional studies have reported changes in circulating amino acids (BCAAs), sphingolipids, acylcarnitines and phospholipids in women with PCOS compared to controls [4, 5, 6, 7, 8, 9, 10, 11]. However, findings are heterogeneous and influenced by phenotype, adiposity, hormonal milieu and analytical platforms. Importantly, static fasting metabolite concentrations provide limited insight into pathway flux or adaptive capacity and cannot capture how metabolic networks respond to physiological challenges.

1.3. Metabolic Responses to Insulin Stimulation

Clamp‐based metabolomics has provided important insights into the dynamic metabolic effects of physiological insulin stimulation, revealing coordinated suppression of circulating lipids, branched‐chain amino acids and acylcarnitines, alongside increased glycolytic flux in insulin‐sensitive individuals [12, 13].

In insulin‐resistant states, clamp studies consistently demonstrate attenuated suppression of lipid‐derived metabolites and altered amino acid handling, reflecting impaired metabolic flexibility [13]. Despite the well‐established presence of insulin resistance in PCOS [2], clamp‐based metabolomic studies remain limited and provide little insight into dynamic metabolic responses during physiological insulin stimulation [10].

1.4. Dynamic Responses to Lipid Loading Under Insulin

Dynamic metabolic challenges provide insights beyond fasting or insulin stimulation alone. Lipid loading during hyperinsulinemia can reveal latent defects in substrate handling [14, 15]. Prior clamp–lipid infusion studies have shown that lipid exposure acutely impairs insulin‐stimulated glucose disposal and attenuates suppression of circulating non‐esterified fatty acids; these effects are particularly evident in insulin‐resistant states [15, 16].

In women with PCOS, lipid infusion provokes disproportionate reductions in glucose disposal and impaired Non‐Esterified Fatty Acid (NEFA) suppression, despite comparable or only modestly reduced baseline insulin sensitivity, indicating heightened vulnerability to lipid‐mediated metabolic stress [17]. These observations suggest that abnormalities in adaptive lipid handling may become apparent only under conditions of increased lipid availability [18]. However, the metabolomic signatures underlying these stress‐dependent responses in PCOS have not been systematically characterised [6].

1.5. Exercise Training as a Modulator of Metabolic Adaptation

Exercise intervention is a cornerstone of PCOS management and improves cardiorespiratory fitness and insulin sensitivity, often independent of weight loss [19]. However, evidence increasingly suggests that exercise does not uniformly restore metabolic regulation across all pathways, particularly under conditions of metabolic stress. While training enhances skeletal muscle oxidative capacity and improves basal substrate utilisation, abnormalities in lipid handling may persist following exercise interventions [16, 20]. This suggests that exercise modifies, rather than fully normalises, adaptive metabolic responses in PCOS [20]. Accordingly, examining dynamic metabolomic responses to lipid loading under insulin stimulation before and after exercise training may distinguish improvements in resting metabolic function from persistent abnormalities in adaptive substrate handling [15].

1.6. Study Aim

The aim was to characterise the dynamic plasma metabolomic responses to acute lipid and saline challenges during a hyperinsulinaemic–euglycaemic clamp in women with PCOS and healthy controls, and to determine whether an 8‐week supervised exercise intervention modifies these responses.

2. Materials and Methods

2.1. Design

This study employed an exploratory, repeated‐measures, within‐subject cross‐over design, incorporating saline and lipid infusions before and after an 8‐week supervised exercise intervention (Figure S1). The study protocol was approved by the Leeds Central Research Ethics Committee (Yorkshire and Humber REC no. 10/H1313/44). The study was conducted in accordance with the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants before their involvement.

2.2. Participant Selection

Twelve women with PCOS were recruited from local endocrine clinics, and 10 healthy control women were recruited through advertisements at Hull University and East Yorkshire Hospital Trust newsletters. Given the exploratory nature of the study and the absence of prior data for the metabolomics outcomes examined, a formal sample size calculation was not performed. The sample size was based on feasibility and was comparable to previous clamp‐based studies using intensive physiological protocols with repeated blood sampling [21]. PCOS was diagnosed using the Rotterdam criteria (i.e., ≥ 2 of the following: oligomenorrhea, hyperandrogenism or polycystic ovaries on ultrasound [2]); other endocrine causes were excluded. All PCOS participants fulfilled all three Rotterdam criteria (Phenotype A).

The inclusion criteria for both groups were: non‐smokers, not taking medication, free from concurrent illnesses and not engaging in regular exercise of at least 90 min of moderate to vigorous intensity per week before the start of the study. Exclusion criteria were a positive pregnancy test, a family history of type 2 diabetes and impaired glucose regulation at a screening oral glucose tolerance test.

2.3. Protocol

The protocol followed the validated procedures previously described [17]. Briefly, after anthropometry and fasting blood sampling, each participant completed two metabolic studies at baseline with a 5‐h infusion of normal saline (1.5 mL/min) or Intralipid (20% soybean oil, 1.2% egg yolk phospholipids and 2% glycerol; Kabi Fresenius Pharmacia (Runcorn, UK); 1.5 mL/min) in random order, 7 days apart. Three hours into the infusion, a 2‐h hyperinsulinaemic‐euglycaemic clamp was initiated using intravenous (i.v.) soluble insulin (Humulin S, Eli Lilly and Co., Indianapolis, IN) at a priming rate of 80 mU/m² surface area/min for 20 min and then 40 mU/m²/min till completion. Plasma glucose was clamped at 5.0 mmol/L using variable‐rate 20% dextrose infusion adjusted according to arterialised blood glucose measurements every 5 min. The rate of insulin‐stimulated glucose disposal (mg/kg/min) (M), a measure of insulin sensitivity, was calculated from the mean of the five 20‐min periods from 20 to 120 min during the clamp. Blood samples were collected at baseline and hourly thereafter. Plasma samples were extracted and stored at −80°C until analysis. Participants were instructed to maintain their usual dietary and lifestyle habits.

2.4. Exercise Intervention

All participants completed a supervised exercise programme in a laboratory at the Department of Sports, Health and Exercise Science, University of Hull. The programme consisted of three 1‐h aerobic exercise sessions per week for 8 weeks. Exercise intensity was set at 60% of baseline VO2_max. All participants achieved 100% adherence.

Baseline VO2_max was determined using an incremental motorised treadmill protocol to volitional exhaustion. Exercise intensity was adjusted at week 4 according to changes in functional capacity.

2.5. Biochemical Analysis

Serum insulin was quantified by competitive chemiluminescent immunoassay. Plasma glucose, total cholesterol, triglycerides and high‐density lipoprotein cholesterol were measured enzymatically. Serum testosterone was measured using high‐performance liquid chromatography linked to tandem mass spectrometry and sex hormone‐binding globulin (SHBG) was measured by immunometric assay. The Free Androgen Index (FAI) was calculated as 100 × (Testosterone/SHBG) and LDL‐cholesterol was calculated using the Friedewald equation [22, 23]. NEFA concentration was determined using a colourimetric enzymatic assay, with a coefficient of variation of 1.4%.

2.6. Metabolomics Measurements

Plasma metabolomics were performed on samples collected at three timepoints during each study visit: baseline (T₀), 180 min (post‐infusion) and 300 min (post‐clamp). Targeted metabolite quantification was conducted using the AbsoluteIDQ p150 kit (Biocrates Life Sciences AG) [24]. Sample preparation followed the manufacturer's validated protocol, and metabolites were analysed by flow injection analysis–tandem mass spectrometry (FIA–MS/MS) [25]. Metabolite concentrations were determined using MetIDQ software and reported in μmol/L. Due to analytical co‐elution, leucine and isoleucine were quantified as a combined measure (xLeu).

2.7. Data Analysis

The present study represents a secondary analysis of metabolomics data generated from the previously published physiological study [17]. All analyses were conducted in R (v4.5.2), with metabolomics data preprocessing and quality control detailed in the Supporting Information Methods. Data were log₂‐transformed prior to statistical modelling.

2.7.1. Baseline Analyses

Baseline differences between PCOS and control participants (Before Exercise + Saline, T ₀) were assessed using linear models with empirical Bayes moderation (limma) [26]. To account for repeated measurements originating from the same individual across the wider study, intra‐subject correlation was estimated using the duplicate Correlation function and incorporated into the model [26]. False discovery rate (FDR) control used the Benjamini–Hochberg procedure [27]. A complementary supervised analysis using Orthogonal Partial Least Squares–Discriminant Analysis (OPLS‐DA; ropls) was performed to identify influential metabolites (Variable Importance in Projection, VIP ≥ 1.0) [28].

2.7.2. Longitudinal Trajectory Modelling

Dynamic metabolic responses across time (T ₀, T ₁₈₀, T ₃₀₀) and intervention conditions (Before/After Exercise × Saline/Lipid) were evaluated using linear mixed‐effects models (lme4) [29]. Models included fixed effects for Group, Timepoint, Intervention and all interactions, with participant ID included as a random intercept. A Group × Timepoint × Intervention interaction tested for global divergence in metabolic trajectories. When the global Group × Timepoint × Intervention interaction was significant, post hoc mixed models were fitted within each intervention condition to identify condition‐specific group differences. Statistical significance was defined as p < 0.05, with degrees of freedom and p‐values estimated using the ImerTest package [30].

2.7.3. Targeted Metabolite Analyses

Metabolites with the highest OPLS‐DA VIP scores were further examined in targeted mixed‐effects models to determine metabolite‐specific Group × Timepoint interactions. These analyses were restricted to pre‐specified metabolites within the intervention condition showing significant global divergence. As these analyses were exploratory and hypothesis‐generating, p‐values were reported without adjustment for multiple comparisons and interpreted accordingly.

2.7.4. Exploratory Flux Ratio Analysis

To provide exploratory physiological context, prespecified metabolite ratios reflecting substrate handling (e.g., C3/C2, xLeu/C2, PC ae C44:4/C2) were examined at the T ₃₀₀ timepoint of the After Exercise + Lipid condition. Group differences were assessed using Welch's t‐test.

3. Results

Ten healthy women and 12 with PCOS, of comparable age and body mass index, were enrolled and all participants completed the study. Baseline characteristics have been reported previously [17] and are summarised in Table 1.

TABLE 1.

Baseline characteristics of the participants.

Parameters	Controls (n = 10)	PCOS (n = 12)	p
Age (year)	25.26 ± 6.5	28.27 ± 6.5	—
BMI (kg/m2)	26.8 ± 6.5	29.4 ± 5.5	—
Waist (cm)	81.1 ± 14.2	98.8 ± 15.8	—
WHtR	0.47 ± 0.08	0.56 ± 0.08	0.0179
Testosterone (nmol/L)	1.05 ± 0.34	1.51 ± 0.69	—
FAI	2.1 ± 2.05	6.77 ± 2.90	0.0004
SHBG (nmol/L)	69.6 ± 29.7	26 ± 17.6	—
TC (mmol/L)	4.61 ± 0.75	4.13 ± 0.65	—
TG (mmol/L)	0.84 ± 0.18	1.25 ± 0.72	—
HDL‐C (mmol/L)	1.48 ± 0.47	1.12 ± 0.20	0.0255
LDL‐C (mmol/L)	2.66 ± 0.56	2.33 ± 0.51	—
FPG (mmol/L)	4.89 ± 0.56	4.94 ± 0.55	—
HbA1c (mmol/mol)	33 ± 5.6	34 ± 2.9	—
TSH (IU/L)	1.9 ± 0.91	1.6 ± 0.58	—
HOMA‐IR	1.34 (0.80, 2.13)	2.30 (1.30, 3.90)	0.0410
NEFA (mmol/L)	0.48 (0.29, 0.64)	0.45 (0.42, 0.58)	—
VO2max (ml/kg/min)	36.3 ± 6.30	26.90 ± 4.80	0.0007

Note: Normally distributed variables are shown as the means ± SD. Skewed variables are provided as the medians (25th, 75th percentile). Significant comparisons (p ≤ 0.05) are shown with exact p‐values. ‘–’ indicates non‐significant comparisons (p > 0.05). These data have been published previously [17].

Abbreviations: FAI, free androgen index; FPG, fasting plasma glucose; HDL‐C, high‐density lipoprotein; LDL‐C, low‐density lipoprotein; NEFA, non‐esterified fatty acids; SHBG, sex hormone‐binding globulin; TG, triglycerides; WHtR, waist‐to‐height ratio.

Compared with healthy controls, women with PCOS had a higher waist‐to‐height ratio (WHtR), elevated free androgen index (FAI) and HOMA‐IR, lower HDL‐cholesterol and lower cardiorespiratory fitness (VO₂ max) (all p < 0.05). Consistent with these findings, insulin stimulated glucose disposal during the saline clamp was lower in women with PCOS (M value 0.50 ± 0.03 vs. 0.65 ± 0.06 mg/kg/min, mean ± SEM, p = 0.01). Fasting triglycerides and NEFA concentrations did not differ between groups.

3.1. Data Quality and Global Metabolomic Structure

Principal Component Analysis (PCA) of the Log₂‐transformed dataset revealed that the primary source of variance (PC1, 31%) was the temporal progression of the clamp protocol, with no clustering by groups (i.e., Healthy Control vs. PCOS) (Figure S2). No samples exceeded the exclusion threshold (median + 3 × Median Absolute Deviation [MAD]).

3.2. Baseline Differential Abundance and Feature Ranking

At baseline (T ₀, Before Exercise and Saline treatment), no metabolites met the FDR threshold for differential abundance between PCOS and healthy controls. Several metabolites showed nominal significance (raw p < 0.05) but did not survive FDR correction (Figure 1). OPLS‐DA was subsequently used to rank metabolites contributing to group discrimination. Ether‐linked phosphatidylcholines (e.g., PC ae C42:4, PC ae C44:4, PC ae C44:5) and selected amino acids (e.g., glutamine, tyrosine and xLeu) were the top‐ranked discriminatory features (VIP ≥ 1.0) (Table S1).

FIGURE 1.

Baseline plasma metabolomic profiles between women with polycystic ovary syndrome and healthy controls. Each point represents an individual metabolite quantified by targeted metabolomics. Log₂ fold‐change values indicate relative differences between Polycystic Ovary Syndrome (PCOS) and control groups, with positive values denoting higher concentrations in PCOS and negative values denoting higher concentrations in controls. Statistical significance shown as −log₁₀ p‐values. No metabolite remained significant after false discovery rate (FDR) correction.

3.3. Longitudinal Analysis of Metabolic Trajectories

A global linear mixed‐effects (LME) model incorporating Group, Timepoint, Intervention and all interactions demonstrated significant main effects of Group (p = 0.002), Timepoint (p < 0.001) and Intervention (p < 0.001). A significant Group × Timepoint × Intervention interaction was observed (F = 2.96, p = 0.008) (Table 2), indicating that the longitudinal metabolic trajectories differed between PCOS and controls depending on intervention condition.

TABLE 2.

Linear mixed‐effects analysis of longitudinal metabolomic responses.

Effect term	Num DF	Den DF	F‐value	p‐value
Patient group	1	44.6	10.45	0.002
Timepoint	2	221.9	26.82	< 0.001
Intervention	3	224.6	27.71	< 0.001
Group × Timepoint × Intervention	6	221.5	2.96	0.008

Note: p‐values are derived from Type III analysis of variance applied to linear mixed‐effects models, testing fixed effects of group, timepoint, intervention and their interactions. Participant was included as a random effect. Group × Timepoint × Intervention denotes the three‐way interaction testing whether longitudinal metabolic trajectories differ between PCOS and control participants across sampling timepoints and experimental conditions.

Abbreviations: Den DF, denominator degrees of freedom; F‐value, F statistic from type III analysis of variance applied to linear mixed‐effects models; Num DF, numerator degrees of freedom; p‐value, probability value.

Post hoc mixed‐effects models for each condition showed no significant Group × Timepoint interaction in the Before Exercise + Saline, Before Exercise + Lipid, or After Exercise + Saline conditions (all p > 0.05). A significant interaction was observed in the After Exercise + Lipid condition (p = 0.048), indicating differential temporal responses between groups in the post‐exercise lipid condition.

3.4. Targeted Trajectory Analysis of VIP‐Selected Metabolites

Targeted LME models were applied to the top VIP‐ranked metabolites within the After Exercise + Lipid condition. Among these metabolites, PC ae C44:4 showed a significant Group × Timepoint interaction (p = 0.031; Figure 2). The remaining metabolites did not show significant differences in trajectory (Table S2).

FIGURE 2.

Plasma PC ae C44:4 trajectories during the post‐exercise lipid‐insulin challenge. Plasma concentrations of phosphatidylcholine PC ae C44:4 during the After Exercise + Lipid condition in women with Polycystic Ovary Syndrome (PCOS) and healthy control participants. Lipid infusion was administered from 0 to 300 min, with a hyperinsulinaemic‐euglycaemic clamp initiated at 180 min and maintained until 300 min. In healthy controls, PC ae C44:4 concentrations declined during the clamp phase, whereas concentrations remained relatively elevated in women with PCOS during the clamp phase. Data are presented as mean ± SEM. PC ae, ether‐linked (alkyl‐acyl) phosphatidylcholine.

3.5. Exploratory Flux Ratio Analysis Under Maximal Metabolic Stress

Exploratory metabolite ratio analyses were conducted at the T ₃₀₀ timepoint of the After Exercise + Lipid condition. Compared with controls, PCOS participants had higher C3/C2 ratios (+5.6%, p = 0.042), higher xLeu/C2 ratios (+4.6%, p = 0.038) and higher PC ae C44:4/C2 ratios (+12.7%, p = 0.019). Acetylcarnitine concentrations (C2) were higher in PCOS (+8.3%, p = 0.029), and tyrosine concentrations were elevated (+18.9%, p = 0.003) (Table 3).

TABLE 3.

Exploratory metabolite ratio analysis during the post‐exercise lipid‐insulin challenge.

Variable	Control mean	PCOS mean	% Difference	p
C3/C2	0.0892	0.0942	+5.6%	0.042
xLeu/C2	2.31	2.42	+4.6%	0.038
PC ae C44:4/C2	1.58	1.78	+12.7%	0.019
C2	7.82 μmol/L	8.47 μmol/L	+8.3%	0.029
Tyrosine	64.3 μmol/L	76.4 μmol/L	+18.9%	0.003

Note: Ratios were examined to provide exploratory assessment of substrate handling during the post‐exercise lipid challenge. Positive percentage differences indicate higher values in women with PCOS relative to healthy controls.

Abbreviations: C2, acetylcarnitine; C3/C2, propionylcarnitine‐to‐acetylcarnitine ratio; PC ae C44:4, ether‐linked phosphatidylcholine‐to‐acetylcarnitine ratio; xLeu/C2, branched‐chain amino acid‐to‐acetylcarnitine ratio.

4. Discussion

4.1. Summary of Principal Findings

The main finding is that metabolic dysregulation in women with PCOS was not evident under resting conditions but emerged selectively under conditions of combined lipid and insulin stress. Following exercise training, metabolomic differences became apparent during the post‐exercise lipid challenge, characterised by persistent elevation of ether‐linked phosphatidylcholine PC ae C44:4 in women with PCOS. In contrast, ether‐linked phospholipids were appropriately suppressed in healthy controls.

Exercise training improved cardiorespiratory fitness and normalised baseline amino acid profiles but did not fully restore adaptive lipid regulation under stress. These findings suggest that abnormalities in lipid handling persist in PCOS despite improvements in resting metabolic function.

4.2. Stress‐Dependent Impairment in Substrate Switching and Phospholipid Clearance

Metabolic flexibility is the ability to transition from lipid oxidation in the fasted state to carbohydrate utilisation during insulin stimulation and is a central feature of metabolic health. In insulin‐resistant conditions such as PCOS, this transition is frequently blunted, reflecting impaired coordination between substrate availability and utilisation [18, 31]. In this study, the impairment was not evident at rest or during saline conditions. However, it became apparent when insulin stimulation was combined with lipid loading, highlighting the importance of physiological challenge in revealing defects in adaptive regulation [32, 33].

During hyperinsulinaemic‐euglycaemic clamp conditions, healthy controls demonstrated the expected response, characterised by suppression of circulating lipid species alongside reliance on carbohydrate metabolism. However, PCOS participants failed to appropriately suppress circulating phospholipids during lipid infusion, despite insulin stimulation, indicating impaired regulation of lipid availability under stress [18, 31].

The emergence of this defect following exercise training suggests that improvements in fitness and insulin sensitivity did not fully restore adaptive lipid regulation under conditions of increased lipid availability [34]. Greater central adiposity in the PCOS group may also have contributed to the observed stress‐dependent metabolomic responses. Increased visceral adiposity is associated with impaired metabolic flexibility and altered lipid metabolism [3, 18, 34].

4.3. Ether‐Linked Phosphatidylcholines and Plasmalogens: Lipid‐Handling and Oxidative Stress Vulnerability

Ether‐linked phosphatidylcholines, including plasmalogens, are specialised phospholipids that contribute to membrane structure, lipoprotein stability and protection against oxidative stress [35]. Alterations in these lipid species have been reported across multiple insulin‐resistant states, suggesting sensitivity to changes in lipid trafficking and redox balance [36, 37]. In this study, PC ae C44:4 and related ether‐linked phosphatidylcholines demonstrated stress‐dependent persistence under insulin stimulation following lipid loading in PCOS, contrasting with progressive suppression in healthy controls. This pattern suggests impaired regulation when lipid availability is increased, rather than a fixed deficiency in ether‐lipid abundance [35].

In insulin‐resistance states characterised by chronic oxidative stress, depletion or impaired regulation of ether‐linked phospholipids may represent a vulnerability [38]. In this study, the relative persistence of PC ae C44:4 during lipid stress in PCOS participants may reflect limited capacity to appropriately remodel or clear ether‐linked phospholipids under insulin‐mediated control [35, 38].

Our results support a conservative interpretation in which altered ether‐lipid trajectories serve as integrative markers of disrupted lipid handling and redox balance under metabolic stress [35, 39]. This interpretation is consistent with prior lipidomic studies reporting reduced plasmalogen content in insulin‐resistant phenotypes and aligns with broader patterns of stress‐dependent metabolic inflexibility observed in this study [10, 40].

4.4. Amino Acid Signature and Exercise Responsivity: Divergent Normalisation of Resting Versus Stress Metabolome

Baseline elevations in branched‐chain and aromatic amino acids are well described in PCOS and are commonly interpreted as markers of insulin resistance and impaired amino acid oxidation [41]. Consistent with this literature, PCOS participants in this study had modest baseline elevations in leucine/isoleucine and tyrosine relative to healthy controls. These abnormalities normalised completely following the exercise training, paralleling improvements in cardiorespiratory fitness and insulin‐stimulated glucose disposal [19]. This normalisation contrasts with the persistence of lipid‐handling abnormalities under combined lipid and insulin stress, indicating that distinct regulatory constraints control amino acid and lipid metabolism in PCOS [42].

Improvements in skeletal muscle oxidative capacity appear sufficient to enhance amino acid clearance and utilisation at rest. However, regulation of lipid supply and clearance under stress requires coordinated hepatic and adipose responses that remain incompletely responsive to exercise alone [40, 42]. Therefore, the amino acid findings demonstrate that the exercise intervention was physiologically effective, while highlighting the specificity of the residual lipid handling defect.

4.5. Exercise Training Reveals a Hepatic‐Muscle Mismatch During Post‐Exercise Lipid Stress

The most salient finding of this study is that metabolic divergence between PCOS and healthy controls emerged specifically in the post‐exercise lipid challenge condition, indicating that exercise training modified, but did not eliminate, abnormalities in lipid handling [43, 44]. Exercise improved cardiorespiratory fitness and insulin‐stimulated glucose disposal [19, 44, 45]; however, these adaptations were not accompanied by normalisation of lipid suppression during combined hyperinsulinaemia and lipid loading. This pattern suggests persistent impairment in coordination of substrate availability and utilisation under metabolic stress [40].

Exercise training enhances skeletal muscle oxidative capacity and metabolic responsiveness [43]. However, effective regulation of lipid metabolism also depends on coordinated suppression of lipid mobilisation and circulating lipid availability during insulin stimulation [46]. The persistence of ether‐linked phospholipids during lipid loading in PCOS suggests these responses remained incompletely normalised despite improved fitness [44].

4.6. Exploratory Metabolite Ratios Suggest Impaired Coordination of Substrate Handling

Exploratory metabolite ratios were examined to provide physiological context for the lipid‐handling abnormalities during the post‐exercise lipid challenge. Under combined lipid loading and insulin stimulation, women with PCOS exhibited elevations in C3/C2, xLeu/C2 and PC ae C44:4/C2 ratios, consistent with impaired coordination of substrate handling during the post‐exercise lipid challenge [42, 47, 48].

These analyses are exploratory and hypothesis‐generating and do not directly quantify metabolic flux or distinguish between hepatic, muscular or adipose tissue contributions. Accordingly, the observed metabolite ratios should be interpreted as supportive physiological markers rather than direct measures of pathway activity. Future studies using stable isotope tracer methodologies will be required to assess lipid and amino acid kinetics under combined metabolic stress in PCOS [49, 50].

4.7. Physiological and Clinical Implications

Our results highlight the limitations of resting or fasting assessments in characterising metabolic dysfunction in PCOS. The data from this study support the use of a dynamic metabolic challenge that integrates lipid exposure with insulin stimulation to reveal abnormalities in adaptive substrate regulation that may not be apparent under resting conditions [40].

The dissociation between improvements in cardiorespiratory fitness and amino acid homeostasis, and persistent abnormalities in lipid handling, suggests that exercise training improves metabolic capacity without uniformly restoring all aspects of adaptive lipid regulation in PCOS and has important implications for how intervention efficacy is evaluated [46, 51]. These findings support evaluating metabolic responses during physiological stress when assessing metabolic dysfunction and response to lifestyle interventions.

While exercise remains effective in PCOS, improvements in resting metabolic markers may not fully reflect restoration of adaptive lipid regulation under physiological stress [51].

4.8. Strengths and Limitations

A major strength of this study lies in the integration of targeted metabolomics with tightly controlled physiological challenge, enabling direct interrogation of dynamic substrate handling in vivo [52]. The longitudinal cross‐over design, with repeated measures obtained before and after exercise training, reduced inter‐individual variability and strengthened internal validity [17, 53]. The combined lipid‐infusion and hyperinsulinaemic‐euglycaemic clamp protocol permitted separation of lipid‐driven effects from insulin‐mediated regulation under stable glycaemic conditions [54].

Several limitations warrant consideration. The sample size was modest, although consistent with clamp‐based physiological studies and adequate to detect within‐subject interaction effects under controlled conditions [21, 53]. The study population comprised women with a severe, well‐defined PCOS, phenotype A, which enhances internal consistency but may limit the generalizability of these findings to women with milder PCOS phenotypes (B, C, D) or postmenopausal populations [55]. Metabolomic measurements were confined to circulating plasma, precluding direct inference of tissue‐specific mechanisms in skeletal muscle, liver or adipose tissue [52]. Although the participants were instructed to maintain their usual dietary and lifestyle habits throughout the study, dietary intake was not standardised beyond an overnight fast, and any variation in usual diet may have influenced circulating metabolite concentrations. The absence of a non‐exercise control group limits definitive attribution of the observed metabolomic changes to exercise training alone. In addition, the targeted metabolomics platform limits inference to predefined metabolites, whereas untargeted approaches may identify additional lipid species and pathways relevant to stress responsiveness in PCOS [52].

Future studies incorporating stable isotope tracers, larger cohorts and tissue‐specific phenotyping will be required to directly assess substrate kinetics and further characterise the mechanisms underlying stress‐dependent lipid dysregulation in PCOS [49, 50].

4.9. Conclusions

Metabolic dysregulation in PCOS is not fully apparent under resting conditions but is revealed when lipid availability and insulin action are simultaneously challenged. Using an exercise‐lipid‐clamp paradigm, we demonstrate that adaptive lipid handling diverges between women with PCOS and healthy controls only under metabolic stress, despite improved fitness and amino acid metabolism.

The persistence of ether‐linked phospholipids during insulin stimulation following lipid loading suggests impaired coordination of substrate handling under metabolic stress. These findings support dynamic stress testing to identify latent metabolic vulnerability in PCOS and support stress‐dependent abnormalities in lipid handling as a key feature of the metabolic phenotype.

Funding

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Study protocol and analytical workflow.

Data S1: Supplementary methods: metabolomics data preprocessing.

Figure 2. Principal component analysis of plasma metabolomic profiles after quality control.

Table S1: Top ranked metabolites at baseline.

Table S2: Targeted trajectory analysis of top‐ranked metabolites during the after exercise + lipid condition.

Darwish R., Moin A. S. M., Nandakumar M., et al., “Metabolic Stress Testing Reveals Persistent Lipid‐Handling Dysfunction in Women With Polycystic Ovary Syndrome Despite Exercise Training,” Diabetes, Obesity and Metabolism 28, no. 10 (2026): 9563–9572, 10.1111/dom.71177.

Data Availability Statement

All data relevant to the study have been included in the article or uploaded as Supporting Information. Additional data relevant to this study are available from the corresponding author upon reasonable request.