Comprehensive Analysis of Individual Variation in the Urinary Proteome Revealed Significant Gender Differences

Abstract

Disease biomarkers are the measurable changes associated with a pathophysiological process. Without homeostatic control, urine accumulates systematic changes in the body. Thus, urine is an attractive biological material for the discovery of disease biomarkers. One of the major bottlenecks in urinary biomarker discovery is that the concentration and composition of urinary proteins are influenced by many physiological factors. To elucidate the individual variation and related factors influencing the urinary proteome, we comprehensively analyzed the urine samples from healthy adult donors (aged 20-69 years). Co-expression network analysis revealed protein clusters representing the metabolic status, gender-related differences and age-related differences in urinary proteins. In particular, we demonstrated that gender is a crucial factor contributing to individual variation. Proteins that were increased in the male urine samples include prostate-secreted proteins and TIMP1, a protein whose abundance alters under various cancers and renal diseases; however, the proteins that were increased in the female urine samples have known functions in the immune system. Nine gender-related proteins were validated on 85 independent samples by multiple reaction monitoring. Five of these proteins were further used to build a model that could accurately distinguish male and female urine samples with an area under curve value of 0.94. Based on the above results, we strongly suggest that future biomarker investigations should consider gender as a crucial factor in experimental design and data analysis. Finally, reference intervals of each urinary protein were estimated, providing a baseline for the discovery of abnormalities.

Keywords: Biofluids*; Label-free quantification; Mass Spectrometry; Protein Identification*; Urine analysis.

Graphical abstract

Fig. 1.

The flowchart of the urinary proteome analysis of individuals.

Fig. 2.

Protein identification and CV distributions. A, Number of protein groups identified in each sample. Darker color: number of proteins identified by MS/MS spectral match; Lighter color: number of proteins identified by match between runs. The first and second letters of each label on the x axis denote the age and gender group of the sample, respectively. B, Technical and biological CVs. The x axis denotes the median log10 protein abundance among the QC runs.

Fig. 3.

Hierarchical clustering of the 49 individual urinary proteomes. Samples were categorized into ten groups based on donor age and gender, as illustrated by the colored bars beside both rows and columns of the heatmap. Based on the clustering tree, the 48 samples were classified into 10 clusters, and 1 was an outlier. The clusters are labeled by their dominant sample groups on the corresponding branches of the dendrogram with the same color code.

Fig. 4.

Protein co-expression networks. A, Characteristics of the 9 protein co-expression subnetworks. #Vertices. Number of vertices (i.e. proteins) in the subnetwork; #Edges, the median edge number of all vertices; Correlation, the median absolute Spearman correlation coefficient of all vertex pairs, regardless of whether they are linked by edges; CV, the median biological CV of all proteins in that subnetwork; Hub proteins, proteins with a hub score > 0.9. B, Expression profiles of the protein co-expression subnetworks. For the convenience of visualization, the high dimension abundance matrix in each subnetwork was summarized into a one-dimensional array (an eigengene in co-expression network analysis[45]) by principal component analysis. In detail, the abundance values for each protein were log2 transformed and scaled to a mean of 0 and standard deviation of 1. Then, the eigengene for each subnetwork was calculated as its first principle component, which explained 49.65% to 75.59% of the total protein variance. Thus, the major information of protein abundance distribution in a subnetwork could be represented by the distribution of its eigengene. C, The metabolism subnetwork. Proteins involved in different metabolic pathways are displayed in different colors. All of the proteins were positively correlated with each other.

Fig. 5.

Gender-related differential protein analysis. A, Volcano plot for protein abundance changes between the male and female urine samples. B, Comparison of the subcellular localization distributions of the proteins that were increased in the different gender groups. C, Functional analysis by IPA. The significance of the enrichment was tested by Fisher's exact test. The x axis represents the -log₁₀ p value of this test. The color of the bar represents the Z-score, which was calculated to predict the effect of protein changes. A positive/negative Z-score indicates that proteins with observed changes in abundance have a potential activating/inhibitory effect on a function. D, Pathway analysis with REACTOME. The significance of the enrichment for each pathway is shown as the -log₁₀ p value adjusted for multiple hypothesis tests. E, Enrichment of related diseases analyzed by IPA.

Fig. 6.

Urinary gender model. A, MRM validation results for nine gender-related proteins. B, Distribution of the “gender score” among the training (2D LC-MS/MS) set. C, Performance of the urinary gender model on the test (MRM) set. ROC curves represent the urinary gender model as well as five predictors using individual proteins in this model.