Elaine Holmes

Background and aims: Diabetic retinopathy (DR) is a major complication of diabetes mellitus and a leading cause of blindness globally. Early detection and intervention underpin optimal DR management and remain challenging. The underlying biological mechanisms are also poorly understood. This study examined comprehensive plasma lipid profiles using a targeted approach to detect potential DR biomarkers.

Materials and methods: We utilised data and samples from 762 adult participants with type 2 diabetes (mean age 64.5 years, 54.3% males, median diabetes duration 7.0 years) from the community-based longitudinal Fremantle Diabetes Study Phase II. All had DR status (none, mild non-proliferative diabetic retinopathy (NPDR), moderate NPDR, or severe NPDR or worse) assessed by colour fundus photography at baseline and at the Year 4 or 6 review. Ultra-performance liquid chromatography-tandem mass spectrometry was used for plasma lipid profiling using baseline samples. Multiple logistic regression was used to identify baseline associates of i) any new or worsening DR and ii) any incident DR. The likelihood ratio test (LRT) was used to evaluate the incremental contribution of potential new lipid biomarkers. The net reclassification improvement (NRI) was also calculated.

Results: Any new/worsening DR was observed in 121 participants (16%) and 35 of 495 without DR at baseline (7.0%) developed DR during follow-up. We detected five lipids independently associated with one or both DR outcomes, specifically cholesterol ester (20:4), fatty acid (20:3), lysophosphatidylglycerol (18:0), lysophosphatidylglycerol (18:1) and phosphatidylcholine (18:0_20:3). For any new/worsening DR, the inclusion of lipid parameters (cholesterol ester (20:4), fatty acid (20:3) and lysophosphatidylglycerol (18:1)) in addition to conventional risk factors (systolic hypertension, HbA1c, blood glucose-lowering treatment intensity and urinary albumin:creatinine) added significantly to the conventional model (LRT, P=0.00002). The NRI gain for at a moderate cut-off of 10% was 5.7% (SE 2.8%; P=0.043). For incident DR, inclusion of lipid parameters (fatty acid (20:3), lysophosphatidylglycerol (18:0) and phosphatidylcholine (18:0_20:3)) with HbA1c as the only conventional risk factor improved model performance (LRT, P=0.00001). The NRI gain at a moderate cut-off of 5% risk of incident DR was 28.3% (P=0.002).

Conclusion: These data demonstrate that disturbances in lipid metabolism are associated with DR progression in type 2 diabetes. The present five lipid biomarkers have not been identified as determinants of incident DR in limited previous longitudinal studies but have the potential to improve DR risk prediction and provide novel insights into the mechanistic pathways underlying the development and progression of DR.

Understanding the distribution and variation in NMR-based inflammatory markers is crucial in the evaluation of their clinical utility in disease prognosis and diagnosis. We applied high resolution 1H NMR spectroscopy of blood plasma and serum to measure the acute phase reactive glycoprotein signals (GlycA and GlycB) and the subregions of the lipoprotein based Supramolecular Phospholipid Composite signals (SPC1, SPC2 and SPC3) in a large multi-cohort population study. A total of 5702 samples were measured to determine the signal variations in a range of chronic and acute inflammatory conditions. We found that while the GlycA and GlycB were increased in inflammation, the SPC regions behaved independently of Glyc signals, with SPC2 and SPC3 being reduced in chronic inflammation in comparison to healthy controls (p-value SPC2=2.9x10-10, p-value SPC3=2.2x10-3) and SPC1 (p-value=0.29) being unchanged. SPC1 was decreased in acute inflammation indicating a link to the immune response (p-value=2.5x10-11). These findings confirm the independent biological relevance of all 3 SPC subregions and contraindicate the use of aggregate SPC values as general inflammatory markers.

Nuclear magnetic resonance (NMR) spectroscopy is increasingly employed in research to quantify lipoprotein subfractions, offering potential utility in clinical diagnostics, particularly for cardiovascular risk assessment. However, the independent validation of proprietary NMR-based lipoprotein profiling methods is crucial for verifying clinical accuracy and reliability. This study presents a posthoc evaluation of concordance between the NMR-based B.I.LISA method and standard enzymatic assays for total cholesterol (TC), triglycerides (TGs), and high-density lipoprotein cholesterol (HDL-C), measured in 620 plasma samples from the OMNI-Heart study, focusing on their performance in evaluating the dietary intervention outcomes. Despite involving independently acquired data not designed for an intermethod validation, the comparison showed a high correlation between methods (R = 0.85–0.92), with median deviations of −4, −5, and −15% for HDL-C, TC, and TGs, respectively. The larger TG deviations are attributed to known issues arising from heterogeneity in high-TG samples, although intervention outcomes remained unaffected. Albumin was identified as a potential interfering factor affecting the TC and HDL-C measurements. HDL-C could also be affected by lipoprotein degradation, contributing to divergence in comparisons of marginal intervention outcomes. Extreme discrepancies were observed in atypical hypercholesterolemia samples. These findings highlight the reliability of the NMR approach despite revealing minor but significant deviations that warrant further research.

Interplay between the maternal diet and gut microbiome may impact fetal immune development and allergic disease risk. This study investigated associations between maternal prenatal urinary metabolites and infant food allergy and then extended to potentially relevant dietary and microbial precursors.
We investigated 599 mother-infant dyads from an Australian population-derived prebirth cohort. Maternal dietary data and fecal and urine samples were collected in the third trimester. NMR was used to measure prenatal urinary metabolites. Infant food allergy status was determined at 1 year by skin prick allergy testing and food challenge. Regression techniques were used to investigate associations and adjust for pre-specific confounding factors.
Higher concentration of hippuric acid in maternal urine, an end-product of dietary polyphenol metabolism, was associated with a lower risk of infant food allergy (odds ratio (OR) 0.62 (95% CI 0.42, 0.93)). Consistent with this, dietary proanthocyanidins, a polyphenol, were positively associated with both higher urinary hippuric acid concentration (0.11 log units, CI 0.01, 0.22) and lower risk of infant food allergy (OR 0.58 (CI 0.36, 0.96)). Maternal carriage of the gut commensal Prevotella copri, previously associated with protection against infant allergic disease, was associated with 21% higher urinary hippuric acid concentrations (CI 4%, 40%, corresponding to 0.19 log units CI 0.04, 0.34); however there was no evidence of mediation.
Further studies are required to confirm whether higher dietary intake of proanthocyanidins during pregnancy is associated with protection against allergic disease in the infant via gut microbiome production of hippuric precursors and other immune-active metabolites.

Clinical chemistry retains its position as a cornerstone of toxicological assessment, yet inter-laboratory variability in baseline values remains a challenge for the integration and interpretation of multisite datasets. This study leveraged a publicly available clinical chemistry database to assess the impact of inter-laboratory variability in response to hydrazine-induced steatosis. Seventeen clinical chemistry and physico-chemical parameters were evaluated in response to a single dose of hydrazine (at 30 mg/kg or 90 mg/kg) administered to Sprague-Dawley rats (n = 83) across five different pharmaceutical companies and compared with sham-dosed control animals. Hydrazine exposure produced a distinct and consistent biochemical signature at 48 h post-dose across the combined sample set from all laboratory sites, characterised by increased serum bilirubin and BUN and decreased serum protein concentrations, alongside atypical reductions in ALT and AST due to transaminase inhibition. Despite sizable inter-laboratory differences in response when considering single assays, multivariate analysis of the complete dataset was able to extract a core pathological response signature. Early changes at 24 h post-dose in AST, ALT, total protein, and calcium demonstrated strong predictive value for 48-h toxicity profiles (AUROC 0.98), underscoring the translational potential of early biomarkers. This study highlights both the robustness and contextual limitations of clinical chemistry data in toxicological studies. It underscores the importance of matched-control designs and multivariate approaches for multisite studies and advocates for the integration of early predictive modelling to optimise study design and align with the principles of the Replace, Reduce, and Refine initiative.

As part of a strategy for accommodating missing data in large heterogeneous datasets, two Random Forest-based (RF) imputation methods, missForest and MICE were evaluated along with several strategies to help navigate the inherently incomplete structure of the dataset. Background: A total of 3817 complete cases of clinical chemistry variables from a large-scale, multi-site preclinical longitudinal pathology study were used as an evaluation dataset. Three types of ‘missingness’ in various proportions were artificially introduced to compare imputation performance for different strategies including variable inclusion and stratification. Results: MissForest was found to outperform MICE, being robust and capable of automatic variable selection. Stratification had minimal effect on missForest but severely deteriorated the performance of MICE. Conclusion: In general, storing and sharing datasets prior to any correction is a good practise, so that imputation can be performed on merged data if necessary.

Microbiome engineering has emerged as a promising strategy to drive biotechnological developments across diverse fields. Microbiome-based fertilizers could significantly contribute to the gradual replacement of synthetic chemical fertilizers, potentially leading to substantial environmental and economic impacts. This study employed microbiome engineering to develop a self-assembled nitrogen-fixing microbial community utilizing carbon compounds from animal waste. This was achieved by enriching soil samples in bioreactors supplied with nitrogen via air pumping and fed with volatile fatty acids (VFAs) as the only carbon source. VFAs are the most common by-products of anaerobic waste fermentation. Results show a self-assembled community, dominated by Sinirhodobacter spp. (44.4%), Aureimonas spp. (17.7%), and Taibaiella spp. (12.4%), capable of fixing 2.7 times more nitrogen than the initial microbiome. During cultivation, inorganic nitrogen forms were detected in the supernatant at concentrations of up to 12.7 mg·L −1 . Once the self-assembled community was inoculated in tomato plants, Pseudomonas spp. and Exiguobacterium spp. became the most abundant and significantly enhanced tomato plant growth in both hydroponic and soil-based systems. Plant height and yield were comparable to those achieved with conventional synthetic nitrogen fertilizers. This study shows the potential of this methodology for developing effective biofertilizers while promoting a circular economy strategy that transforms waste into high-value bioproducts. This approach, combined with the simplicity of the bioreactor system, offers a viable and sustainable solution for developing countries with limited technological resources, and materializes the One Health vision while simultaneously protecting the health of people, crops, and animals.

Background
Nuclear magnetic resonance (NMR) spectroscopy enables the characterisation of lipoprotein sub-particles, providing a more detailed lipid profile than the conventional lipid measurements, with potential clinical relevance, particularly in cardiovascular disease (CVD), which remains the leading cause of mortality worldwide. Nonetheless, for clinical implementation, it is essential to first determine the normal variation of lipoprotein parameters by age and sex.

Methods
This cross-sectional study analysed a large dataset of 31,275 serum or plasma samples from five different countries using the B.I.LISA™ NMR-based platform, quantifying 112 lipoprotein parameters, including subclass size and concentration. Lipoprotein parameters from specific cohorts were fitted to a Quantile Generalised Additive Model (QGAM) to calculate the different percentiles as a function of age and sex.

Findings
A sub-cohort of individuals belonging to non-oriented cohorts (27,470 individuals) showed that lipoprotein parameters exhibit distinct sex- and age-dependent patterns, with inflection points observed around 44 and 60 years in women and around 60 years in men, aligning with known ageing acceleration models. The sub-cohort of 3021 individuals showing cardiometabolic risk factors was used to evaluate the effect of obesity, hypertension and diabetes in the lipoprotein distribution. Finally, we analysed the lipoprotein parameters that align with SCORE2 (a well-known CVD risk predictor) in an age- and sex-dependent manner. Many NMR-derived parameters effectively distinguish between low and high/very high CVD risk profiles, with very low-density (VLDL)-associated parameters demonstrating the highest sensitivity across a broad age range.

Interpretation
Our findings provide reference values for NMR-derived lipoprotein parameters by age and sex, enabling their accurate interpretation in the context of cardiovascular disease risk stratification.

Funding
The specific funding of this article is provided in the acknowledgements section.

Pooled quality control (PQC) samples are the gold standard for data quality monitoring in metabolic phenotyping studies. Typically composed of equal parts from all study samples, PQCs can be challenging to generate in large cohorts or when sample volumes are low. As an alternative, externally sourced matrix-matched surrogate QCs (sQC) have been proposed. This study evaluates the performance of sQCs against PQCs for assessing analytical variation, data pre-processing, and downstream data analysis in a targeted lipidomics workflow.
Plasma samples (n = 701) from the Microbiome Understanding in Maternity Study, along with PQC (n = 80) and sQC (n = 80) samples, were analyzed using a lipidomics assay targeting 1162 lipids. QC samples were injected throughout acquisition, and data pre-processing was performed using each strategy. For simplicity, a subset (n = 381) of the study samples was used to assess differences in downstream statistical analyses.
Both QC approaches demonstrated high analytical repeatability. While PQC and sQC compositions differed, use of PQCs retained less than 4 % more lipid species during pre-processing. Univariate analysis identified more statistically significant lipids with PQC-based pre-processing, but multivariate model performance was similar between datasets.
This study provides a comprehensive comparison of QC strategies and emphasizes the importance of careful QC workflow selection. While PQCs offer advantages, sQCs serve as a suitable alternative for quality assessment and pre-processing. Their commercial availability also supports use as intra- and inter-laboratory long-term references, aiding data harmonization across studies and laboratories.
[Display omitted]
•Comparison of two quality control workflows; pooled study and surrogate QC samples.•In-depth assessment of lipid composition, precision, and filtering.•OPLS-DA model predictive power maintained with both QC pre-processing strategies.•Surrogate QC samples are a robust alternative to a pooled QC in targeted lipidomics.