Association of serum biomarkers of dietary purine intake with glycaemic control and risk of preterm birth: two prospective cohort studies among pregnant women

Summary

Background

To examine serum purine metabolites as biomarkers of maternal dietary purine intake and their associations with glycaemic control and preterm birth during pregnancy.

Methods

We included 1480 pregnant Chinese women with gestational diabetes mellitus (GDM) from the Westlake Precision Birth Cohort (WeBirth) at mid-pregnancy, divided into discovery (n = 1230) and internal validation (n = 250) subcohorts. Linear regression was used to assess associations between serum metabolites, dietary purine intake, and continuous glucose monitoring (CGM) metrics. A purine score was developed and evaluated for associations with glycaemic control and preterm birth, with replication in 936 pregnant women from the Tongji-Huaxi-Shuangliu Birth Cohort (THSBC).

Findings

Serum purine level was associated with dietary purine intake and with the mean of daily difference (β = 0.096 [95% CI, 0.038, 0.15]). The association between higher serum purine and preterm birth risk in the WeBirth cohort (OR = 1.29 [95% CI, 1.06, 1.52]) was replicated in the THSBC cohort. Early-pregnancy serum purine also predicted GDM risk (OR = 1.38 [95% CI, 1.13, 1.63]) in the THSBC. A purine score comprising purine, adenine, 6-O-methylguanine, and uric acid showed consistent associations with glycaemic variability and preterm birth across both cohorts.

Interpretation

Reflecting dietary purine intake, serum purine level and purine score are associated with greater glycaemic variability and higher preterm birth risk. Monitoring dietary purine intake may improve glycaemic control in GDM and prevent pregnancy complications.

Funding

National Key R&D Program of China; National Natural Science Foundation of China; Zhejiang Provincial Key Laboratory Construction Project; Sichuan Provincial Natural Science Foundation.

Research in context.

Evidence before this study

Gestational diabetes mellitus (GDM) is a common pregnancy complication associated with adverse maternal and neonatal outcomes. These long-term consequences highlight the critical importance of effective glycaemic control during pregnancy. Modern dietary patterns characterised by increasing consumption of red meat, seafood, and sugar-sweetened beverages have led to increased dietary purine consumption, which is closely related to metabolic dysfunction. However, the role of purine metabolism in pregnancy, particularly among women with GDM, remains poorly characterised. Besides, the majority of investigations relying on self-reported dietary assessments are limited by recall bias. Moreover, the commonly used purine biomarker serum uric acid, an end product of purine metabolism, is not able to reflect the full scope of purine metabolism.

Added value of this study

This study assesses the association of serum purine metabolites, dietary purine intake and glycaemic traits in a GDM cohort. Using objective serum metabolite measurements related to dietary purine intake, we identified specific purine-related metabolites associated with glycaemic variability and preterm birth in women with GDM. We also developed a composite purine score including purine, adenine, 6-O-methylguanine and uric acid that outperforms single biomarkers like uric acid and shows consistent associations across two independent cohorts.

Implications of all the available evidence

This study establishes serum purine metabolites and a composite purine score as objective indicators of dietary purine exposure. The purine score captures metabolic dysregulation linked to glycaemic variability and may inform personalised interventions to improve pregnancy outcomes.

Introduction

Gestational diabetes mellitus (GDM), characterised as hyperglycaemia that develops during pregnancy,¹ is the most common pregnancy complication and poses threats to both mothers and offspring.² Achieving optimal glycaemic control is therefore critical for improving maternal health and pregnancy outcomes, highlighting the need to identify modifiable factors that influence metabolic regulation during pregnancy. Among such factors, dietary patterns have emerged as key determinants of maternal metabolic health.

Over the past decade, modern lifestyles and dietary patterns (e.g. alcohol, red meat, and sugar-sweetened beverage intake) have led to a gradual increase in dietary purine intake.³ This shift has also contributed to the rising global prevalence of hyperuricemia and gout,⁴ which are closely linked to comorbidities such as diabetes.⁵ Hyperuricemia is also associated with an increased risk of GDM⁶ and other complications such as pregnancy-induced hypertension, low birthweight, and preterm birth.⁷ However, uric acid is the final product of purine metabolism and is affected by both production and excretion, so it may not fully reflect dietary purine exposure or its metabolic effects. The role of maternal dietary purine intake itself in glycaemic control and preterm birth, especially in women with GDM, remains unclear.

Previous studies have largely relied on self-reported intake of purine-rich foods or focused exclusively on serum uric acid.⁸ Other purine-related metabolites, such as adenosine and hypoxanthine, have also been implicated in insulin resistance, endothelial dysfunction and increased diabetes risk.^9,10 So far, no study has systematically profiled circulating purine metabolites during pregnancy and examined their association with continuous glucose monitoring (CGM)–derived glycaemic traits. Unlike traditional metrics such as fasting glucose or the oral glucose tolerance test (OGTT), CGM captures real-time glucose variability, an independent risk factor for diabetic complications.¹¹ Leveraging CGM-based measures may therefore provide a more comprehensive, physiologically relevant, and clinically actionable assessment of maternal glycaemic control.

Therefore, in the present study, we aimed to investigate the associations of maternal purine exposure during pregnancy with glycaemic control and preterm birth by integrating dietary assessment, serum purine metabolites, and CGM data among pregnant women with GDM.

Methods

Study population

This study was based on the Westlake Precision Birth Cohort (WeBirth). WeBirth is an ongoing prospective birth cohort that enrolled 2001 pregnant Chinese women aged 18 years or older who were diagnosed with GDM between 24 and 28 weeks of gestation at the Hangzhou Women's Hospital, China.¹² GDM was diagnosed using the criteria of fasting blood glucose ≥5.1 mmol/L, and/or 1-h blood glucose ≥10.0 mmol/L, and/or 2-h blood glucose ≥8.5 mmol/L.¹³ At the enrolment visit, we collected maternal demographics, lifestyle, dietary habits, and simultaneously instructed the participants to wear a CGM device (Freestyle Libre Pro, recording interstitial glucose concentrations every 15 min) for 14 consecutive days, during which fasting blood samples were collected. In the present study, a total of 1480 participants from the WeBirth cohort with complete data on dietary habits, serum purine metabolites and CGM measurements were included (Supplementary Fig. S1). This cohort was further divided into two subcohorts based on the timing of enrolment relative to the COVID-19 pandemic (Jan 2020), during which recruitment was interrupted for four months, resulting in two temporally distinct populations. These were designated as a discovery subcohort (post-COVID, n = 1230) and an internal validation subcohort (pre-COVID, n = 250), with the discovery subcohort shown to be representative of the original cohort (Supplementary Table S1).

External validation was conducted in a case-cohort within the Tongji-Huaxi-Shuangliu Birth Cohort (THSBC), a prospective birth cohort of 7281 pregnant women aged 18–40 years initiating prenatal care at ≤15 weeks of gestation in Chengdu, China. To efficiently investigate metabolic biomarkers for adverse pregnancy outcomes, we selected a case-cohort restricted to singleton pregnancies for serum metabolomic profiling. This case-cohort comprised 1059 participants, including 514 women with incident cases of preterm birth (n = 240), low birthweight (n = 137) or macrosomia (n = 216), and 545 randomly selected participants who did not experience the above three adverse pregnancy outcomes.¹⁴ In the present study, we included 936 participants with available early- and mid-pregnancy circulating purine metabolite profiles and OGTT measures (Supplementary Fig. S1).

Dietary purine intake measurements

Baseline purine intake was assessed using a food frequency questionnaire (FFQ), which captured habitual dietary intake over the month preceding enrolment. Purine intake was estimated based on the China Food Composition Table (Sixth edition).¹⁵ Dietary purine content was further categorised into guanine, adenine, hypoxanthine, xanthine, and total purine (mg per 100 g of food; Supplementary Table S2). Cumulative purine intake was calculated by multiplying the purine content of each food item by its reported consumption frequency and portion size.¹⁶

Measurement of serum purine metabolites

Serum samples were collected after at least 8 h of fasting in both the WeBirth and THSBC cohorts, then stored at −80 °C until analysis. Serum metabolomic profiling was performed using an LC-ESI-MS/MS system (UPLC, ExionLC AD, MS, QTRAP® System).¹⁴

In the WeBirth cohort, we extracted 17 purine and related metabolites from the metabolomics panel, measured using serum samples collected at the enrolment visit (24–28 gestational weeks), including purine, guanosine, 2′-O-methylguanosine, 1-methylguanosine, adenosine, N6-methyladenosine, 7-methylguanine, 6-O-methylguanine, adenine, inosine, inosine diphosphate, 1-methylinosine, hypoxanthine, xanthine, 7-methylxanthine, uric acid, and allantoin. All were identified with Level 1 or Level 2 confidence, with a coefficient of variation (CV) < 30%, and were detected or expected in human biofluids according to the Human Metabolome Database (HMDB, version 5.0).

In the THSBC cohort, blood samples were collected at approximately 6–16 weeks (defined as early-pregnancy) and 16–28 weeks (mid-pregnancy) of gestation. Serum purine metabolites were extracted for both time points, while 7-methylguanine, 6-O-methylguanine, inosine diphosphate and allantoin were not detected. To ensure analytical consistency, identical preprocessing procedures were applied across both cohorts. Given the extremely low proportion of missing values, all of which were below the detection limit, missing values were imputed using the global minimum.¹⁷ Batch effects were corrected with a support vector regression (SVR) using the metaX R package (version 1.5.2).¹⁸

Continuous glucose monitoring and glycaemic metrics calculation

Valid CGM data were defined as valid recordings of ≥72 h, with the exclusion of the first and last incomplete days, and days with extreme hypoglycaemia (>99th percentile time spent below 3.9 mmol/L).¹² As a result, we included 21,437 valid person-day readings for further analysis.

A comprehensive panel of CGM-derived metrics was analysed to capture key aspects of glycaemic control and variability. Metric definitions and calculation methods were adopted from previous approaches, and computations were performed using the CGMTSA R package, as detailed in Supplementary Table S3. Processed glycaemic control traits included mean blood glucose (MBG), area under the curve (AUC), glucose management indicator (GMI), estimated haemoglobin A1c (eA1c), time in range of 3.5 mmol/L to 7.8 mmol/L (TIR),¹⁹ time in range above 7.8 mmol/L (TAR), time in range below 3.5 mmol/L (TBR), high blood glucose index (HBGI), and low blood glucose index (LBGI); while glycaemic variability traits included mean of daily differences (MODD), CV, mean amplitude of glucose excursions (MAGE), standard deviation of blood glucose (SDBG), and J-index.20, 21, 22, 23, 24

All CGM metrics were calculated separately for the all-day, daytime (6:00–24:00), and nighttime (00:01–6:00) periods, based on the median sleep-wake patterns of the WeBirth cohort,²⁵ which were assessed using the Axivity AX3 accelerometer (Axivity Ltd.). Daily CGM metrics were first calculated and then averaged across all valid days to account for intraindividual variability.

Assessment of covariates, blood traits and birth outcomes

In the WeBirth cohort, baseline characteristics, including education level, household income, smoking status, alcohol consumption and glycaemic-control medications (insulin or metformin) were collected via structured questionnaires. Baseline energy intake, as well as protein, fat, and fibre intake were evaluated using dietary data collected via FFQ. Physical activity was assessed using the total metabolic equivalent of task (tMET) from the Chinese version of the Pregnancy Physical Activity Questionnaire (PPAQ-C).

Biochemically measured blood traits such as HbA1c, fasting blood glucose (FBG) and total triglycerides (TG) were measured at baseline using capillary electrophoresis, hexokinase and GPO-PAP methods during hospital visits. Gestational age was determined based on the last menstrual period and confirmed by ultrasound. Foetal sex and pregnancy plurality were obtained from prenatal care and delivery medical records. Preterm birth, defined as delivery before 37 weeks of gestation,²⁶ was identified and categorised into spontaneous and medically indicated subtypes based on clinical medical records.

For participants in the THSBC cohort, maternal age, pre-pregnancy BMI, parity, and gestational age at baseline were also obtained through questionnaires, while OGTT results at mid-pregnancy, GDM diagnosis, preterm birth, as well as foetal sex were extracted from clinical medical records.

Statistical analysis

All statistical analyses were performed using R software (version 4.4.3). Prior to analysis, dietary purine intake was categorised into quartiles (Q),²⁷ percentage-based CGM metrics (TAR, TIR, TBR) underwent arcsine square root transformation, and purine metabolites were log-transformed. Following these transformations, purine metabolites, CGM metrics and biochemically measured blood traits were standardised to Z-scores to enable comparison of effect size. Consistent statistical significance criteria were applied across all analyses: in the WeBirth discovery subcohort, statistical significance was defined as false discovery rate (FDR) < 0.05 after Benjamini–Hochberg correction. In the two validation cohorts (the internal validation subcohort and THSBC cohort), P < 0.05 together with a consistent direction of effect relative to the discovery cohort was considered indicative of replication.28, 29, 30

We first examined the associations between dietary purine intake and circulating purine metabolites (log-transformed and Z-scored) using a linear regression model, with model assumptions checked. Models were adjusted for pre-pregnancy BMI, gestational age at baseline, maternal age, and parity. These variables were included as potential confounders given their known biological plausibility and impact on both dietary patterns and systemic metabolism.31, 32, 33 In the sensitivity analyses, we additionally adjusted for education level, household income, smoking status, and alcohol consumption in the linear models.^16,32,33

Primary analyses: association of serum biomarkers of dietary purine intake with glycaemic control

We evaluated the associations between each serum purine metabolite (per SD) and CGM-derived metric (per SD) using multivariable linear regression adjusting for pre-pregnancy BMI, gestational age at baseline, maternal age, and parity (Model 1). In sensitivity analyses, Model 2 was additionally adjusted for physical activity and total energy intake; Model 3 further included education, income, smoking, and alcohol status; and Model 4 additionally adjusted for dietary macronutrients, including protein, fat, and fibre intake. To ensure the robustness of our findings, we also conducted a sensitivity analysis by excluding participants who reported using glycaemic-control medications (insulin or metformin) to evaluate whether these treatments influenced the observed associations. To identify the key CGM metric most strongly associated with purine metabolites, we performed permutational multivariate analysis of variance (PERMANOVA, Euclidean distance, n = 999 permutations) using the vegan R package, with the same covariates adjusted.

Based on these associations, we constructed a purine score for the CGM metric most strongly associated with purine metabolites. We included purine metabolites that showed significant associations with this CGM metric (defined as FDR <0.05 in the WeBirth discovery subcohort). The purine score for each individual was calculated as the sum of the selected metabolites, weighted by the direction of their association observed in the WeBirth discovery subcohort. The equation is defined as follows:

$${Purinescore}_i=\sum _{j=1}^n\left(d_j\times z_{ij}\right)$$

where i represents the individual subject; j represents the specific purine metabolite included in the panel; n represents the total number of selected purine metabolites; d_j is the directional weight assigned to metabolite j, where d_j = 1 if the metabolite was positively correlated with the outcome in the WeBirth discovery subcohort, and d_j = −1 if it was negatively correlated; z_ij denotes the Z-scored level of metabolite j for individual i. This score was subsequently applied in the WeBirth internal validation subcohort and the THSBC cohort.

We examined the associations of serum purine (per SD) and the purine score (per SD) with CGM metrics and six biochemically measured blood traits, including HbA1c, FBG, total cholesterol (TC), TG, low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) using multivariable linear regression with sequential adjustment for covariates (Models 1–4, as described above). In addition, sensitivity analyses were performed by restricting the study population to participants not receiving glycaemic-control medications. In the independent THSBC cohort, we evaluated the association of serum purine level and the purine score (in early and mid-pregnancy) with fasting, 1-h, and 2-h OGTT glucose levels using multivariable linear regression, and we examined the associations of serum purine and purine score with GDM risk using logistic regression. Both models were adjusted for maternal age, parity, pre-pregnancy BMI, gestational age at sampling, and the specific birth outcome case status of the subcohort (macrosomia, low birthweight, or preterm birth).

Secondary analyses: association of serum biomarkers of dietary purine intake with preterm birth and gestational duration

Due to the limited number of preterm birth cases within each individual subcohort, we pooled participants from both the discovery and internal validation subcohorts of the WeBirth study for the preterm birth analyses. We used logistic regression to examine the associations of serum purine level (per SD) and the purine score (per SD) with preterm birth, and linear regression to assess their associations with gestational duration (weeks) in the WeBirth cohort, adjusting for pre-pregnancy BMI, gestational age at baseline, maternal age, parity, and foetal sex. Foetal sex was adjusted based on evidence from previous studies showing a higher preterm birth risk in male foetuses.³⁴ We replicated these analyses in the THSBC cohort using mid-pregnancy serum purine and purine score (per SD) with same covariates adjusted. To assess robustness, we conducted sensitivity analyses by excluding: (1) medically-induced preterm birth cases; (2) multiple pregnancies to account for the confounding effect of plurality, with the same covariate adjustments applied. We further examined whether the association between serum purine (per SD) and the purine score (per SD) with preterm birth was mediated by the related CGM metric in the WeBirth cohort using the mediation R package, with same covariates adjusted.

Ethics

Westlake Precision Birth Cohort study was approved by the Ethics Committee of Westlake University (20190701ZJS0007). The Tongji-Huaxi-Shuangliu Birth Cohort was approved by the Ethics Committee of the Tongji Medical College, Huazhong University of Science and Technology (2017) (S225)-1. All participants provided written informed consent in accordance with the principles of the Declaration of Helsinki.

Role of the funders

The funders had no role in the study design, data collection, data analyses, interpretation, or writing of the report.

Results

Population characteristics

Baseline anthropometrics, clinical characteristics, dietary purine intake, and CGM metrics of the study populations are summarised in Table 1 and Supplementary Table S4. In the WeBirth discovery subcohort, a total of 1230 pregnant Chinese women were enrolled at a mean (SD) gestational age of 25.8 (1.7) weeks, with a mean age of 31.3 (3.8) years. Among them, 69.3% were primiparous, and 72 experienced preterm birth. In the internal validation subcohort, 250 women were enrolled at 26.4 (2.1) weeks of gestation, with a mean age of 31.2 (3.6) years. Of these, 66.0% were primiparous, and 14 had preterm births. In the THSBC cohort, 936 pregnant women were enrolled at a mean age of 26.4 (3.7) years, with a mean gestational age of 10.2 (2.0) weeks, of whom 99.6% were primiparous. There were 72 cases of GDM, and 181 cases of preterm birth in the THSBC cohort.

Table 1.

Characteristics of the study populations.

	WeBirth discovery (n = 1230)	Internal validation (n = 250)	THSBC (n = 936)
Age, years	31.33 (3.75)	31.18 (3.63)	26.42 (3.66)
Pre-pregnancy BMI, kg/m2	22.07 (3.26)	22.74 (4.92)	21.05 (3.01)
Parity
≤1	852 (69.3%)	165 (66.0%)	932 (99.6%)
≥2	378 (30.7%)	85 (34.0%)	4 (0.4%)
Gestational weeks (baseline)	25.81 (1.71)	26.40 (2.11)	10.17 (2.02)
GDM	1230	250	72
Preterm birth	72 (5.9%)	14 (5.6%)	181
CGM metrics, all-day
MBG, mmol/L	4.73 (0.51)	4.81 (0.57)
AUC, mmol/L	112.45 (12.18)	114.42 (13.42)
GMI, %	5.29 (0.22)	5.33 (0.24)
eA1C, %	4.59 (0.32)	4.64 (0.36)
TAR, %	0.01 [0.00, 0.02]	0.01 [0.00, 0.03]
TIR, %	0.91 [0.83, 0.96]	0.91 [0.82, 0.96]
TBR, %	0.06 [0.02, 0.15]	0.06 [0.01, 0.15]
HBGI, mmol/L	0.63 [0.40, 0.95]	0.67 [0.40, 1.05]
LBGI, mmol/L	5.64 (2.74)	5.48 (2.73)
MODD, mmol/L	0.24 (0.11)	0.24 (0.10)
CV, %	0.21 (0.04)	0.21 (0.04)
MAGE, mmol/L	2.29 (0.59)	2.42 (0.65)
SDBG, mmol/L	0.97 (0.21)	1.02 (0.22)
J-Index, mmol/L2	10.65 (2.41)	11.18 (2.73)

Normally distributed continuous variables were presented as mean (SD), continuous variables with non-normal distribution were presented as median [interquartile range], and categorical variables were presented as number (%).

Abbreviations: WeBirth, Westlake Precision Birth Cohort; THSBC, Tongji-Huaxi-Shuangliu Birth Cohort; BMI, body mass index; GDM, gestational diabetes mellitus; CGM, continuous glucose monitoring; MBG, mean blood glucose; AUC, total glucose area under the curve; GMI, glucose management indicator; eA1C, estimated haemoglobin A1c; TAR, time above range (>7.8 mmol/L); TIR, time in range (3.5–7.8 mmol/L); TBR, time below range (<3.5 mmol/L); HBGI, high blood glucose index; LBGI, low blood glucose index; MODD, mean of daily differences; CV, coefficient of variation; MAGE, mean amplitude of glycaemic excursions; SDBG, standard deviation of blood glucose.

Serum purine, a biomarker of purine intake, is associated with glycaemic control during pregnancy

In the WeBirth discovery subcohort, dietary total purine intake was positively associated with serum purine level (Q4 vs. Q1, β = 0.22 [95% CI, 0.068, 0.38], Fig. 1A). Although not statistically significant in the internal validation subcohort, this association remained directionally consistent (Fig. 1A). Sensitivity analysis yielded similar results (Supplementary Fig. S2). We also observed positive associations of dietary guanine, adenine, hypoxanthine, and xanthine intake with serum purine level (Q4 vs. Q1, Fig. 1A), and observed an inverse association of dietary purine intake with 6-O-methylguanine (Q4 vs. Q1, Supplementary Table S5).

Figure 1.

Serum purine, a biomarker of purine intake, is associated with glycaemic control during pregnancy. A: Associations of dietary purine intake (Q4 vs. Q1) with serum purine level. Differences in serum purine level (log-transformed and Z-scored) associated with the highest (Q4) vs. lowest quartile (Q1) of dietary intake in both WeBirth discovery (red) and internal validation (blue) subcohorts. Data are presented as beta coefficients with 95% confidence intervals (CIs). Asterisks (∗) indicate that the presented P-trend and FDR values are specific to the WeBirth discovery subcohort. Models were adjusted for pre-pregnancy BMI, gestational age at baseline, maternal age, and parity, with the linear regression model assumption validated. B: Associations of serum purine (log-transformed and Z-scored) with all-day, daytime, nighttime MODD (Z-scored; left panel) and biochemically measured blood traits (Z-scored; right panel) in both WeBirth discovery (red) and internal validation (blue) subcohorts. Data are presented as beta coefficients with 95% CIs, with the same covariates adjusted. C: Association of serum purine (log-transformed and Z-scored) at early-pregnancy (pink) and mid-pregnancy (green) with OGTT glucose level (Z-scored; left panel) and GDM risk (right panel) in the THSBC cohort. Early-pregnancy (6–16 gestational weeks) serum purine was positively associated with fasting, 1-h OGTT glucose levels and the risk of GDM. Multivariable linear regression was used for fasting, 1-h, and 2-h OGTT glucose levels, and logistic regression was used for GDM risk, adjusted for same covariates. Abbreviations: FDR: false discovery rate; MODD: mean of daily differences; HbA1c, Haemoglobin A1c; FBG, fasting blood glucose; TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; HDL-C, high-density lipoprotein cholesterol; OGTT, oral glucose tolerance test; GDM, gestational diabetes mellitus.

Serum purine demonstrated consistent significant associations with all-day MODD in the discovery (β = 0.096 [95% CI, 0.038, 0.15]) and the internal validation subcohort (β = 0.14 [95% CI, 0.0090, 0.27]), as well as daytime and nighttime MODD (Fig. 1B). To further assess its broader metabolic relevance, we tested serum purine level against six biochemically measured blood traits relevant to glucose metabolism or diabetes.²⁴ In the WeBirth discovery subcohort, we also observed positive associations between serum purine level with HbA1c (β = 0.16 [95% CI, 0.11, 0.21]) and TG (β = 0.12 [95% CI, 0.064, 0.19]), all of which were replicated in the internal validation subcohort (HbA1c: β = 0.23 [95% CI, 0.10, 0.37]; TG: β = 0.17 [95% CI, 0.019, 0.30], Fig. 1B). In addition, in the THSBC, higher serum purine at early-pregnancy was associated with higher fasting OGTT (β = 0.093 [95% CI, 0.030, 0.16]), 1-hr OGTT (β = 0.095 [95% CI, 0.032, 0.16]) glucose level, and risk of GDM (OR = 1.38 [95% CI, 1.13, 1.63], Fig. 1C).

Association of purine-related metabolites with glycaemic control during pregnancy

In addition to serum total purine, we identified 105 significant associations between 17 purine metabolites (including serum purine) and 52 CGM metrics across all time periods in the discovery subcohort, and 42 of these associations were replicated in the internal validation subcohort (Fig. 2A and Supplementary Fig. S3). Notably, measures such as MBG, AUC, GMI, and eA1C showed similar positive associations with purine metabolites, while metrics such as TBR and LBGI showed inverse associations that were not replicated in the validation subcohort (Fig. 2A and Supplementary Fig. S3). Similar results were obtained after additional adjustment for energy intake and physical activity (Supplementary Fig. S4) or further adjustments for education level, household income, smoking status, and alcohol consumption (Supplementary Fig. S5), or additional adjustment for dietary patterns associated with high purine diets using protein, fat, and fibre intake (Supplementary Fig. S6). Furthermore, the associations remained stable when restricting the analysis to participants not using glycaemic-control medications (Supplementary Fig. S7).

Fig. 2.

Association of purine-related metabolites with glycaemic control during pregnancy. A: Association between individual serum purine metabolites (log-transformed and Z-scored) and the purine score (Z-scored) with all-day CGM-derived glycaemic metrics (Z-scored) in the WeBirth discovery subcohort. Data are presented as beta coefficients indicated by colour scale, with linear regression models adjusted for pre-pregnancy BMI, gestational age at baseline, maternal age, and parity. FDR was controlled using the Benjamini–Hochberg method. B: Associations of the purine score (Z-scored) with biochemically measured blood traits (Z-scored) in both WeBirth discovery (red) and internal validation (blue) subcohorts, with the same covariates adjusted. C: Associations of the purine score (Z-scored) at early- (pink) and mid-pregnancy (green) with OGTT glucose levels (Z-scored; left panel) and GDM risk (right panel) in the THSBC cohort. Early-pregnancy purine score was positively associated with fasting, 1-h, and 2-h OGTT glucose levels and the risk of GDM. Mid-pregnancy purine score was positively associated with 2-h OGTT glucose levels and the risk of GDM. Multivariable linear regression was used for fasting, 1-h, and 2-h OGTT glucose levels, and logistic regression for GDM risk, adjusted for corresponding covariates. Abbreviations: FDR: false discovery rate; MBG, mean blood glucose; AUC, total glucose area under the curve; GMI, glucose management indicator; eA1C, estimated haemoglobin A1c; TAR, time above range (>7.8 mmol/L); TIR, time in range (3.5–7.8 mmol/L); TBR, time below range (<3.5 mmol/L); HBGI, high blood glucose index; LBGI, low blood glucose index; MODD, mean of daily differences; CV, coefficient of variation; MAGE, mean amplitude of glycaemic excursions; SDBG, standard deviation of blood glucose; HbA1c, Haemoglobin A1c; FBG, fasting blood glucose; TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; TG, triglycerides; HDL-C, high-density lipoprotein cholesterol; OGTT, oral glucose tolerance test; GDM, gestational diabetes mellitus.

Among all CGM metrics, MODD was the CGM-derived biomarker most strongly associated with the 17 circulating purine metabolites, showing the highest explained variance of 0.47%–0.50% across all periods in the discovery subcohort (Supplementary Fig. S8A). Hence, we developed a composite purine score based on four metabolites significantly associated with all-day MODD in the WeBirth discovery subcohort, including adenine, purine, 6-O-methylguanine, and uric acid (Supplementary Fig. S8B).

The purine score showed consistent positive associations with all-day MODD (β = 0.057 [95% CI, 0.035, 0.078]; β = 0.081 [95% CI, 0.034, 0.13]), daytime MODD (β = 0.055 [95% CI, 0.033, 0.076]; β = 0.081 [95% CI, 0.035, 0.13]) and nighttime MODD (β = 0.052 [95% CI, 0.030, 0.073]; β = 0.083 [95% CI, 0.037, 0.13]) in both discovery and internal validation subcohorts, respectively (Fig. 2A). The purine score was also positively associated with other glycaemic control-related CGM metrics in the WeBirth discovery subcohort, including MBG, AUC, GMI, and eA1C (Fig. 2A). Taking AUC as an example, higher purine score was associated with all-day AUC in both the discovery subcohort (β = 0.040 [95% CI, 0.019, 0.061]) and internal validation subcohort (β = 0.073 [95% CI, 0.027, 0.12]). Sensitivity analyses yielded similar results across both subcohorts, including further adjustments for lifestyle and dietary factors (Supplementary Fig. S9–S11) and the exclusion of individuals on glucose-lowering medications (Supplementary Fig. S12).

In the WeBirth discovery subcohort, higher purine score was also associated with higher HbA1c (β = 0.059 [95% CI, 0.039, 0.79]) and TG (β = 0.063 [95% CI, 0.041, 0.086]), which were replicated in the internal validation subcohort (HbA1c: β = 0.059 [95% CI, 0.039, 0.79]; TG: β = 0.063 [95% CI, 0.041, 0.86], Fig. 2B).

In the THSBC cohort, higher mid-pregnancy purine score was associated with higher 2-h OGTT glucose levels (β = 0.029 [95% CI, 0.014, 0.058], Fig. 2C). Notably, a higher early-pregnancy purine score was significantly associated with greater risk of developing GDM (OR = 1.14 [95% CI, 1.03, 1.25], Fig. 2C).

Glycaemic variability mediates the association between maternal purine biomarkers and preterm birth

Higher serum purine (per SD) was significantly associated with shorter gestational duration (week) (β = −0.14 [95% CI, −0.22, −0.054]) and higher risk of preterm birth (OR = 1.29 [95% CI, 1.06, 1.52]) in WeBirth cohort (Fig. 3A). Similar association was also observed for purine score (β = −0.073 [95% CI, −0.10, −0.044]), OR = 1.15 [95% CI, 1.07, 1.23], Fig. 3A). In the THSBC, these associations were replicated using the mid-pregnancy serum purine (β = −0.16 [95% CI, −0.30, −0.016], OR = 1.22 [95% CI, 1.04, 1.39]) and the purine score (β = −0.077 [95% CI, −0.14, −0.012], OR = 1.09 [95% CI, 1.01, 1.17], Fig. 3A). Consistent results were observed when the analysis was limited to spontaneous preterm births or singleton pregnancies (Supplementary Fig. S13).

Fig. 3.

Glycaemic variability mediates the association between serum purine and preterm birth. A: Higher serum purine (log-transformed and Z-scored) and purine score (Z-scored) were significantly associated with higher risk of preterm birth (odds ratios) and shorter gestational duration (beta coefficients; weeks) in the WeBirth (red) and the THSBC (green) cohorts. Data are presented as effect estimates with 95% Cis, with models adjusted for maternal age, prepregnancy BMI, gestational age at baseline, parity, and foetal sex. B: Mediation analysis of all-day MODD (Z-scored) on the associations of serum purine (log-transformed and Z-scored) and the purine score (Z-scored) with preterm birth (left panels) and gestational duration (right panels) in the WeBirth cohort. Numbers on paths indicate beta coefficients or odds ratios with P-values. PropMedi indicates the proportion of the mediated effect by MODD. P_medi indicates the P-value of the indirect effect. All models were adjusted for pre-pregnancy BMI, gestational age at baseline, maternal age, parity, and foetal sex. Abbreviations: MODD, mean of daily differences.

Results of the mediation analysis suggested that MODD mediated the relationship between serum purine with both gestational duration across all time periods (mediation proportion: 9.52%–14.22%) and with preterm birth (mediation proportion: 11.91%–16.28%, Fig. 3B and Supplementary Fig. S14). As for the purine score, we also observed significant mediation of MODD across all periods with gestational duration (mediation proportion: 7.78%–10.36%) and preterm birth (mediation proportion: 9.05%–11.43%, Fig. 3B and Supplementary Fig. S14).

Discussion

In the present study, we demonstrated that dietary purine intake was positively associated with serum purine level. Besides, multiple purine-related metabolites showed consistent and reproducible associations with CGM-derived glycaemic traits. Based on these findings, we developed a purine score composed of four key metabolites, which showed strong associations with glycaemic traits across different cohorts. Notably, higher serum purine as well as the purine score were both associated with a higher risk of preterm birth among participants with or without GDM, with glycaemic variability (MODD) serving as key mediators.

We observed that several circulating purine metabolites including guanosine, 6-O-methylguanine, adenine, inosine, uric acid and total purine, were positively associated with higher risk of GDM and poorer glycaemic profiles. These results align with prior epidemiological evidence linking purine-rich food intake and hyperuricemia with type 2 diabetes and insulin resistance.³⁵ A meta-analysis across 11 cohort studies (n = 42,834) reported that elevated serum uric acid level precedes the development of type 2 diabetes.^36,37 Mechanistically, purine metabolism is involved in energy production, nucleotide synthesis, and signal transduction.³⁸ Perturbations in this pathway may influence hepatic glucose production, peripheral glucose utilisation, mitochondrial oxidation and reduced glucose uptake in adipocytes and muscles,39, 40, 41, 42, 43 providing biological plausibility for the observed associations between elevated purine metabolites and impaired glycaemic control.

Previous epidemiological studies on dietary purine intake have often relied on self-reported consumption of purine-rich foods, yet accurate quantification remains challenging due to substantial variability in food composition and processing.⁴⁴ In contrast, serum purine metabolites offer an integrated and objective measure of systemic purine exposure, reflecting not only dietary intake but also endogenous production, and nucleic acid turnover.⁴⁵ In our study, dietary purine intake was positively correlated with some serum metabolites (e.g., purine), but showed no association or even inverse trends with others (e.g., 6-O-methylguanine). This observation may arise from pregnancy-induced changes, possibly caused by increased renal perfusion and glomerular filtration,⁴⁶ as well as increased cell turnover, oxidative stress, and placental purine metabolism contributed endogenous purine production. This complexity underscores the advantage of serum metabolites, which capture systemic purine exposure beyond diet, suggesting that a multi-metabolite score may offer greater clinical insight into metabolic regulation during pregnancy.

Therefore, beyond using serum total purine level as a biomarker of dietary purine intake, we also developed a composite purine score integrating four key circulating metabolites. Among them, purine and adenine can be acquired through dietary sources or synthesised endogenously⁴⁷; 6-O-methylguanine, a DNA methylation byproduct, is involved in DNA repair and apoptosis⁴⁸; and uric acid is the final oxidation product of purine catabolism.⁴⁹ The purine score captures an integrated signature of both exogenous exposure of dietary purine intake and intrinsic metabolic activity. As a result, both serum purine and the purine score demonstrated strong and reproducible associations with glycaemic traits across multiple cohorts, predicting glucose variability and control.

Moreover, previous metabolomics studies on GDM have predominantly focused on amino acids, carbohydrates, and lipids, while purine metabolism remains largely unexplored.⁵⁰ One study investigated urinary purine metabolites in GDM using a relatively small cohort of 61 participants.⁵¹ Our findings highlight serum purine level and the metabolomic-derived purine score as key indicators of glycaemic control during pregnancy. Their strong and reproducible associations with glucose variability and control suggest that they may serve as valuable tools for guiding personalised glucose management, dietary modification, and lifestyle interventions in women with or at risk of GDM during early pregnancy.

Across CGM-derived outcomes, purine metabolites and the purine score showed consistent associations with multiple indicators of glycaemic control. Similar effect sizes were observed for MBG, AUC, GMI and eA1C, which is biologically expected given that these metrics represent closely related measures of overall glycaemic exposure.²⁴ CGM traits are hierarchically structured and cluster into a limited number of physiological domains, with mean glycaemic control forming a dominant and highly correlated cluster.²⁴ Therefore, the concordant associations across these traits likely reflect a shared underlying mechanism influencing baseline glucose regulation. Besides, inverse associations with TBR and LBGI are also biologically coherent, as reductions in hypoglycaemia-related metrics typically accompany a shift toward sustained hyperglycaemia, particularly among women with GDM.

Notably, MODD emerged as the CGM metric most strongly associated with circulating purine metabolites. MODD captures inter-day glycaemic variability, reflecting instability in glucose regulation across consecutive days.^21,24 Mechanistically, enhanced purine catabolism can increase reactive oxygen species (ROS) and promote inflammatory responses.^52,53 Such oxidative and inflammatory stress can impair insulin signalling and endothelial function, reducing the resilience of glucose regulatory systems and rendering maternal glycaemia more susceptible to day-to-day fluctuations and glycaemic variability.^54,55

Regarding preterm birth, previous metabolomic studies yielded heterogeneous and often non-overlapping biomarker candidates, likely due to variations in biological sample types, gestational ages at sampling, population characteristics, or small sample sizes.⁵⁶ In the context of purine metabolism, one case–control study reported altered purine-related metabolites in spontaneous preterm placentas (n = 102).⁵⁷ In our study, higher serum purine and the purine score were both associated with shorter gestational duration and higher risk of preterm birth. Importantly, these associations were mediated by MODD, suggesting that purine-induced metabolic stress may contribute to preterm birth through destabilisation of glycaemic regulation. Emerging experimental evidence shows that acute glucose fluctuations, as captured by MODD, induce disproportionate oxidative stress and endothelial dysfunction, which may impair placental perfusion and activate pro-inflammatory pathways involved in parturition, thereby increasing susceptibility to preterm birth.^58,59

To strengthen the robustness of our findings, we validated results from the WeBirth cohort in an independent THSBC cohort. Differences between the two cohorts, including maternal age distribution, parity and GDM prevalence, likely contributed to variations in effect sizes and the stronger associations observed with early-pregnancy metabolites in THSBC.60, 61, 62 Nonetheless, successful replication of the key associations across these phenotypically distinct cohorts underscores the robustness of our findings and supports their potential generalisability across diverse pregnancy risk profiles.

A major strength of our study is the comprehensive identification of multiple purine metabolites associated with CGM-derived glycaemic traits in women with GDM, representing, to our knowledge, the largest investigation of its kind. By integrating self-reported dietary data with serum metabolomics, we captured both exogenous and endogenous purine exposure. Validation across independent cohorts further enhances the generalisability of our findings. Collectively, serum purine level and the metabolomic-derived purine score may serve as early, integrative biomarkers for identifying women at higher risk of adverse glycaemic profiles and preterm birth, enabling timely metabolic or dietary interventions during pregnancy.

This study also has several limitations. First, in the THSBC cohort, validation of the purine score was constrained by the availability of only three metabolites, which may have limited its robustness. Second, the observational design precludes causal inference, and residual confounding and reverse causality cannot be excluded, as glycaemic status may influence related behaviours such as dietary intake. This limitation is particularly relevant to the mediation analysis, where purine metabolites and CGM metrics were measured concurrently in the second trimester, preventing a clear temporal sequence and limiting causal interpretation of the mediation pathway. Additionally, data on certain clinical confounders, such as maternal renal function, were unavailable. Future studies incorporating these clinical records are warranted to validate our findings. Finally, our findings require validation in ethnically and geographically diverse populations to improve generalisability.

Conclusion

Our study demonstrates that serum purine and related metabolites, as objective biomarkers of dietary purine intake, are positively associated with maternal glycaemic variability and risk of preterm birth among pregnant participants. Both serum purine level and the purine score provide comprehensive and biologically relevant information for metabolic dysfunction and adverse pregnancy outcomes, independent of GDM status, and have the potential to serve as dynamic tools for identifying at-risk individuals and guiding targeted dietary interventions. Our findings suggest that interventions targeting purine pathways through dietary modification, lifestyle intervention, or pharmacologic modulation may hold promise for improving maternal glycaemic control and reducing the risk of adverse birth outcomes.

Contributors

The authors gratefully acknowledge all participants for their invaluable contributions to data collection, management, visualisation, and interpretation. J.-S.Z. and X.-F.P. contributed to the study conception and design. M.-Q.S., L.S, X.-H.W, J.Y., J.W., W.H. were responsible for data collection. Y.D., Y.L., K.Z., C.X., J.C., W.G., Y.F., T.W., F.L. supported data curation, visualisation, and interpretation. S.H. conducted the statistical analyses and interpreted the results. S.H., Z.M., and J.-S.Z. drafted the initial manuscript. J.-S.Z. supervised the overall project. J.-S.Z., X.-F.P., A.P., and H.S. critically reviewed, edited, and approved the final version of the manuscript prior to submission. S.H. and J.-S.Z. have accessed and verified the underlying data. All authors contributed to the discussion and critical revision of the manuscript, and approved the final draft.

Data sharing statement

Analysis code used in this study is available via https://github.com/IrisHue/Purine-and-CGM. Other raw genotype and phenotype data of the present research are available from the corresponding authors (X.-F.P. or J.-S.Z.) upon reasonable request.

Declaration of interests

The authors declare no competing interests.