Confounding Factors Impacting microRNA Expression in Human Saliva: Methodological and Biological Considerations

Abstract

There is growing interest in saliva microRNAs (miRNAs) as non-invasive biomarkers for human disease. Such an approach requires understanding how differences in experimental design affect miRNA expression. Variations in technical methodologies, coupled with inter-individual variability may reduce study reproducibility and generalizability. Another barrier facing salivary miRNA biomarker research is a lack of recognized "control miRNAs". In one of the largest studies of human salivary miRNA to date (922 healthy individuals), we utilized 1225 saliva samples to quantify variability in miRNA expression resulting from aligner selection (Bowtie1 vs. Bowtie2), saliva collection method (expectorated vs. swabbed), RNA stabilizer (presence vs. absence), and individual biological factors (sex, age, body mass index, exercise, caloric intake). Differential expression analyses revealed that absence of RNA stabilizer introduced the greatest variability, followed by differences in methods of collection and aligner. Biological factors generally affected a smaller number of miRNAs. We also reported coefficients of variations for 643 miRNAs consistently present in saliva, highlighting several salivary miRNAs to serve as reference genes. Thus, the results of this analysis can be used by researchers to optimize parameters of salivary miRNA measurement, exclude miRNAs confounded by numerous biologic factors, and identify appropriate miRNA controls.

Keywords: biomarker; housekeeping; miRNA; salivary; stabilizer.

Figure 1

Differences in RNA stabilizer introduce greater miRNA expression variation than differences in method of collection. PCA plot of samples collected with expectorant and RNA stabilizer (red), swab and no RNA stabilizer (green), and swab with RNA stabilizer (purple).

Figure 2

Salivary miRNA expression is affected by differences in three technical factors. PLS-DA plot of 291 samples that were aligned with both Bowtie1 and Bowtie2 for within subject comparison (A). For PLS-DA plots visualizing differences in methods of collection (B) and use of RNA stabilizer (C), 734 samples were used.

Figure 3

There is minimal sex-specific miRNA expression variation in saliva. PLS-DA-plot of the most consistently present miRNA in saliva, grouped by female (F, red) or Male (M, green) sex.

Figure 4

Differences in prandial status are responsible for a moderate amount of saliva miRNA variation. PLS-DA scores plot between groups fasting (red) and post-prandial (green) samples.

Figure 5

Differences in exercise state confer minimal miRNA variation. PLS-DA plot of pre- (green) and post-exercise (red) samples.

Figure 6

Top salivary miRNAs that correlate with participant BMI and age. Spearman rank correlations between top saliva miRNAs and BMI (A) and age (B).

Figure 7

Intra-participant miRNA expression variability is greater in infants compared to young adults. PLS-DA revealed 3 distinct groups between time points when all samples are combined (A). When stratified by age, high miRNA expression overlap was shown in young adult samples (B), whereas distinct variation is shown between infant time points (C).

Figure 8

Top “control miRNA” candidates and ranked confounders for consideration. The top three control miRNA candidates identified by coefficient of variation (CoV) and abundance (center). Technical and biologic variables that may impact miRNA analyses are rank-ordered based upon the number of miRNAs differentially expressed between conditions (except age and BMI, where rank is based on the number of miRNAs with an |R| ≥ 0.5). Zero miRNA correlated with BMI with an |R| greater than or equal to 0.5. All miRNAs listed outside of the variable circles are the top significantly different miRNA between conditions.