Discovering and validating between-subject variations in plasma lipids in healthy subjects

Abstract

Lipid levels are commonly used in clinical settings as disease biomarkers, and the advent of mass spectrometry-based (MS) lipidomics heralds the possibility of identifying additional lipids that can inform disease predispositions. However, the degree of natural variation for many lipids remains poorly understood, thus confounding downstream investigations on whether a specific intervention is driving observed lipid fluctuations. Here, we performed targeted mass spectrometry with multiple reaction monitoring across a comprehensive spectrum of 192 plasma lipids on eight subjects across three time-points separated by six hours and two standardized meals. A validation study to confirm the initial discoveries was performed in a further set of nine subjects, subject to the identical study design. Technical variation of the MS was assessed using duplicate measurements in the validation study, while biological variation was measured for lipid species with coefficients of variation <20%. We observed that eight lipid species from the phosphatidylethanolamine and phosphatidylcholine lipid classes were discovered and validated to vary consistently across the three time-points, where the within-subject variance can be up to 1.3-fold higher than between-subject variance. These findings highlight the importance of understanding the range of biological variation in plasma lipids as a precursor to their use in clinical biochemistry.

Figure 1. General outline of the pilot and validation studies.

(a) Study population distribution. Two study cohorts were used for the pilot and validation studies. The validation study was performed on a different set of subjects (duplicate measurements) to confirm the pilot study findings and to determine technical variability. Biological variation was assessed in both studies (data in text). Ethnic coding: C = Chinese, I = Indian, M = Malay. Gender coding M = Male and F = Female. (b) Schematic representation of study design and blood sampling. All samples were taken within the same day for each study across the three time-points (t1 to t3).

Figure 2. Extent of technical variation.

Coefficient of variation (CoV) is shown across 63 (positive ionization, left panel) and 65 (negative ionization, right panel) lipid species measured via mass spectrometry (MS). CoV is represented by mean and 95% confidence interval (2 technical replicates of 26 samples) of each lipid species in the validation batch. Lipid species are annotated by their lipid class followed by their parent ion mass measured by tandem MS; e.g., PC (834) = phosphatidylcholine species measured at 834 precursor ion m/z.

Figure 3. Global profile showing –log10 p-values of 128 measured lipid species across three timepoints t1 to t3 (unadjusted).

Horizontal line indicates Bonferroni correction of –log10 (p-value = 0.01). Eight lipid species are significantly altered across t1 to t3 after Bonferroni correction across both discovery pilot and validation studies. The identities of these eight species are represented in the table above before and after Bonferroni adjustment across both independent and combined studies.

Figure 4. Representative lipid species significantly altered across the three time-points (t1 to t3).

Panels (a–h) show representative lipid species significantly altered across t1 to t3, in contrast to lipid species with no significant differences across the time-points shown on panels (i–j). Technical Coefficient of variation (CoV) for all these species are less than 15% (Fig. 2). Y axis indicate normalized lipid intensity (μg/ml) across all 17 subjects (discovery pilot and validation studies combined). t1 = baseline readings; t2 = post-breakfast readings; t3 = post-lunch readings. Meals were standardized for each subject.

Figure 5. Correlation analysis showing p-values and correlation coefficients across the eight lipid species significantly altered across timepoints t1 to t3.

Correlation p-values are shown numerically (bottom left) and correlation coefficients are represented by colour and shape (top right). Darker blue indicates stronger positive correlation and darker red indicates stronger negative correlation. Thinner ellipse indicates stronger correlation, and rounder oval indicates weaker correlation. (a) All timepoints (t1 to t3) (b) t1 (c) t2 (d) t3.