The relationship of human tissue microRNAs with those from body fluids

It is known that many microRNAs (miRNAs) stably exist in various body fluids, however, the relationship of body fluids miRNAs (BF-miRNAs) with those from tissues (T-miRNAs) remains largely unclear but is important for understanding the potential of BF-miRNAs to be biomarkers of specific diseases. Here by analyzing miRNA expression data from 40 human healthy tissues and those from human body fluids, including plasma, serum, urine, bile, and feces, we revealed a positive correlation between BF-miRNAs and T-miRNAs. Moreover, plasma and serum have the most communication with pericardium, adipose, liver, and spleen. Urinary miRNAs show the highest correlation with kidney miRNAs. For fecal miRNAs, gastrointestinal tract (colon, ileum, jejunum, small intestine, stomach, proximal colon, duodenum, and distal colon) miRNAs show the strongest correlation. Moreover, miRNA set enrichment analysis revealed that highly expressed fecal miRNAs are mostly associated with gastric and colon cancers etc. Additionally, bile miRNAs from suspected cholangiocarcinoma patients show a positive correlation with the cholangiocarcinoma tumor tissue. Interestingly, the relationship of BF-miRNAs and T-miRNAs shows significant sex differences. Serum miRNAs showed higher correlation with T-miRNAs in males, whereas plasma miRNAs and urine miRNAs showed higher correlations with T-miRNAs in females. These findings together indicate a potential role of BF-miRNAs as biomarkers to monitor corresponding specific human diseases.


Materials and methods
miRnA expression datasets. We obtained the miRNA expression dataset of 40 healthy human tissues from Liang et al. 's study 3 . The miRNA expression dataset of human body fluids which includes plasma, serum, urine and bile were obtained from Sinivasan et al. 's study 11 and the miRNA expression dataset of feces were obtained from Liu et al.'s study 10 . Besides, we downloaded miRNA expression dataset of cholangiocarcinoma (CHOL) from the TCGA database 12 . Data analysis. For all expression data, we deleted the NA values and averaged the expression values of the redundant IDs. Then we mapped the expression profile of tissue miRNAs and each body fluid miRNAs according to the miRNA name. Spearman's correlation analysis was performed using R, a free statistical platform (https:// cran.r-project.org/). To compare the relationship of BF-miRNAs and T-miRNAs between male and female, we separated the samples into two groups using gender information of samples. miRNA set enrichment analysis was performed using the TAM tool v2.0 13,14 .
The global relationship between BF-miRNAs and T-miRNAs has been dissected. It is necessary to know the functional roles of the highly expressed miRNAs in the body fluids. We have uncovered the enriched functions, clusters, families, and TFs of the highly expressed miRNAs in feces using the TAM tool. Here, using the same procedure, we firstly selected highly expressed miRNAs in plasma (expression level >10000), serum (expression level >10000) and urine (expression level >100). We found those miRNAs are highly overlapped in 3 body fluids. For those highly expressed miRNAs in three types of body fluids, serum miRNAs are all present in plasma, 74% www.nature.com/scientificreports www.nature.com/scientificreports/ (14/19) of serum miRNAs are in urine, and 75% (18/24) of plasma miRNAs are in the urine. Therefore, we performed enrichment analysis for those miRNAs which was highly expressed in all three body fluids. As a result, the top terms of significantly enriched function include aging, cell death, adipocyte differentiation, innate immunity, and tumor suppressor etc (Fig. 2C). The terms were listed in Table S3. Also, those miRNAs mainly enrich in the let-7 family, mir-26 family, mir-25 family and are significantly regulated by the TFs, for instance, MYC, TRIM32, LIN28B and so on (Fig. 2D). Besides, we observed vascular hypertrophy and lymphoma are the most significant related diseases.
BF-miRNAs showed biased correlations with miRNAs in specific tissues. We further studied the relationship of miRNA expression between the four body fluids and the 40 healthy tissues. Plasma and serum show highly similar tendency of correlation with the 40 tissues (Fig. 3A,B). Pericardium, adipose, liver and spleen are the top tissues of strongest correlation with both plasma and serum. It is well known that pericardium, liver, and spleen are indeed organs with a profusion of blood supplies [15][16][17] . For adipose, a previous study has proposed the adipose cells can release large amounts of miRNAs into the plasma 18 , suggesting that plasma and serum miRNA biomarkers could do better on diseases related with heart, liver, spleen, and adipose than on other diseases. Interestingly, urinary miRNAs have the highest correlation with kidney-borne miRNAs (rho = 0.55, P = 6.74E-22, Spearman correlation; Fig. 3C), which hints the prospect of urinary miRNA signature in kidney-related diseases. In addition, several studies have reported urinary miRNAs as biomarkers for the diagnosis of kidney disease 19,20 . As mentioned above, feces should have direct contact with gastrointestinal tissues. As a result, we found that the miRNAs from colon, ileum, jejunum, small intestine, stomach, proximal colon, www.nature.com/scientificreports www.nature.com/scientificreports/ duodenum, and distal colon which all belong to gastrointestinal tract, show the strongest correlations with fecal miRNAs (Fig. 3D). This result means that fecal miRNAs could mostly communicate with gastrointestinal tract and thus could monitor the physiological and even pathophysiological status of gastrointestinal tract. In short, the miRNA expression status of not only the above four body fluids but also all body fluids might be the same as their most frequently communicated tissues. Taking account of the characteristics of miRNA expression, we also applied other proper correlation methods, including Kendall coefficient of concordance and distance correlation. The results showed high similarity with that of spearman correlation. For example, liver and pericardium are still the tissues of more correlation for serum and plasma. For urinary and fecal miRNAs, the most correlation tissues are also similar (Table S3), which further suggested the stable existence of contacts between body fluids and specific tissues.

Bile miRNAs show positive correlation with cholangiocarcinoma tumor tissue.
Circulating miRNAs only partially originate from the secretion of tumor tissue and may also come from sources such as tissue damage and micrometastatic cells 21 . We have found a significantly positive correlation between the miRNA expression of tissues and body fluids in normal samples, however, the overall relationship of miRNA expression between pathological tissues and body fluids is not known. Therefore, we analyzed the bile miRNAs from www.nature.com/scientificreports www.nature.com/scientificreports/ suspected CHOL patients and the miRNAs from CHOL tumor samples in TCGA database. As shown in Fig. 4, there is a significantly positive correlation between bile-borne miRNAs and the miRNAs from tumor tissue (rho = 0.52, P = 1.25E-80, Spearman correlation), suggesting that T-miRNAs have similar expression state with BF-miRNAs in the abnormal condition. The results hint the body fluid miRNAs might be the ideal biomarkers for diagnosis, development, and prognosis of the diseases.

Comparison of the relationship of BF-miRNAs and T-miRNAs between male and female.
We previously have reported sex-biased miRNAs in healthy human tissue and peripheral blood cells and revealed  www.nature.com/scientificreports www.nature.com/scientificreports/ some patterns that female-biased miRNAs are more conserved and are significantly associated with metabolism process and cell cycle process 22 . Therefore, we studied whether there is sex difference between the relationship of BF-miRNAs and T-miRNAs. We separated the miRNA expression profiles of body fluids by the gender information of the samples and compared the correlation of BF-miRNAs and T-miRNAs between males and females. Consequently, It was indeed found that the serum miRNAs had a significantly higher correlation with T-miRNAs of 40 tissues in males (paired t-test P-value = 1.90E-20, Fig. 5A), whereas miRNAs from female samples had significantly higher correlation with T-miRNAs in plasma (paired t-test P = 3.25E-18, Fig. 5B) and urine (paired t-test P-value = 4.32E-12, Fig. 5C). In addition, we also compared the correlation of T-miRNAs with plasma and serum miRNAs. We found the correlations of serum with tissues are significant higher than that of plasma with tissues (median: 0.43 vs. 0.40, P-value=2.73E-23, Fig. 5D), which suggested serum miRNAs could perform better to reflect the state of tissues.

Discussions
In conclusion, we revealed that miRNAs of body fluids tend to be positively correlated with various human tissues globally. For plasma and serum miRNAs, they show the strongest correlations with pericardium, adipose, liver, and spleen miRNAs. Urinary miRNAs have the highest correlation with kidney miRNAs. And fecal miRNAs show the strongest correlations with miRNAs from colon, ileum, jejunum, small intestine, stomach, proximal colon, duodenum, and distal colon. In addition, we uncovered that feces highly expressed miRNAs are mostly enriched in functions regulation of stem cell, EMT, response to cytokine, inflammation, adipogenesis, aging, and tumor suppressor, and diseases of digestive system cancer (colon cancer, stomach cancer etc). Similarly, we found the most expressed miRNAs in all other 3 body fluids. Those miRNAs are enriched in the functions of aging, cell death, adipocyte differentiation, innate immunity, and tumor suppressor. Furthermore, bile miRNAs have a significantly positive correlation with miRNAs in tumor tissue under disease status. Intriguingly, in different types of body fluids, the relationship of BF-miRNAs and T-miRNAs shows different sex-preferences. Specifically, male BF-miRNAs are significantly higher correlated with T-miRNAs in serum, whereas the opposite pattern appears in plasma and urine. This finding suggests that for blood-borne miRNA based disease diagnosis, it could be better to take serum samples for male patients and plasma for female patients. Together, these findings suggest frequent communication between body fluids and corresponding tissues and potential miRNA signature in body fluids, which are not random but have regular patterns. And thus body fluid miRNAs could be a non-invasive biomarker for facilitating the diagnosis and prognosis of various human diseases. Nevertheless, it should be noted that the answers to a number of important questions are still unknown. For example, where the body fluid miRNAs are from and what their functions are. Therefore, more analysis is needed when more data becomes available in the future.

Data availability
The raw data used in our study are from the publications 3,10,11 .