Biological Relationships between miRNAs used for Colorectal Cancer Screening

MicroRNAs have been proposed to function as biomarkers in several disorders including cancer. Particularly, those present in plasma samples have been validated for colon cancer screening purposes. However, mechanisms underlying miRNA regulating CRC development have not been fully investigated yet. The aim of this study was to perform a biological contextualization of a particular 6 miRNAs signature (Hsa-miR-15b-5p, Hsa-miR-18a-5p, Hsa-miR-29a-3p, Hsa-miR335-5p, Hsa-miR-19a-3p and Hsa-miR-19b-3p) already described as CRC screening tool in terms of its downstream targets and associated pathways, to understand how this set of miRNAs could be regulating colon cancer development. Our results demonstrated that Hsa-miR-15b-5p, Hsa-miR-18a-5p and HsamiR-29a-3p are the best contributors to the diagnostic value of the signature by regulating epithelial to mesenchymal transformation though a gained balance of inhibiting tumoral suppressors genes such as ETS2, TP53, TLE3, ACVR2A.


Introduction
MicroRNAs (miRNAs) are a class of noncoding RNAs of about 22 nucleotides that mediate posttranscriptional gene expression mainly by either repressing the translation or inducing the degradation of their mRNA targets, to which they bind by complementarity [1].
Several authors have demonstrated that miRNAs are present in human circulation, plasma and serum, and profiles are altered in cancer [2,3] being then suggested as a powerful new class of blood-based biomarkers.
Previously we published a cohort of 196 individuals where patients with CRC and AA have significantly different patterns of miRNA expression than healthy individuals [5]. A set of miRNAs (Hsa-miR-15b-5p, Hsa-miR-29a-3p, Hsa-miR-335-5p and Hsa-miR-18a-5p, Hsa-miR-19a-3p, Hsa-miR-19b-3p) demonstrated to have discriminative capacity when found in plasma samples. Then, in a case-control study of 300 subjects, a robust predictive model able to distinguish healthy individuals from those with advanced neoplasm (CRC or AA) was validated (data to be published).
Nevertheless, mechanisms underlying miRNA regulating CRC development have not been fully investigated yet. These studies are needed in order to construct robust panels that would be clinically validated. miRNA targets can be identified by prediction algorithms based on biophysical models, but further experimental strategies are essential to ensure that they are genuine [11,12].
The aim of this study was to identify key proteins and pathways involved in how this particular 6 miRNA signature (Hsa-miR-15b-5p, Hsa-miR-18A-5p, Hsa-miR-29a-3p, Hsa-miR335-5p, Hsa-miR-19a-3p and Hsa-miR-19b-3p) already described as CRC screening tool could be regulating colon cancer development. A biological contextualization of the 6 miRNAs within colorectal carcinogenesis through a network-based approach was performed. By using systems biology based methods, these miRNAs were ranked based on the mechanistic relationship of their protein targets with CRC.

Materials and Methods miRNAs targets research
The lists of gene targets from the 6 miRNAs signature were retrieved from miRTarBase [13,14]. Only experimentally validated targets were selected, in order to reduce the noise. Then, these genes were translated to proteins by mapping them to unique UniProt identifiers [15].

Molecular characterization of colon cancer pathogenesis
Colon cancer was characterized at molecular level via manual review of the literature. This characterization was performed in two steps: first identifying the main pathophysiological processes involved in CRC development and progression (referred to as physio pathological processes or motives), which were then further characterized at the protein level to provide a final list of 160 "effector" proteins (herein "effectors"). Hence, colon cancer 'effectors' are those proteins that, according to the existing literature, have previously been reported to play a critical role in this condition (databases: PubMed, ScienceDirect and Scopus).

Generation of a protein map around colon cancer effectors
"Effector" proteins were used to focus the analyses on the biological condition of interest in the human biological network. The direct interactions (physical interactions or functional relationships) between the miRNAs targets and the effectors of colon cancer, as well as the interactions among the protein targets, were assessed. Different publically available databases were consulted for the human protein network generation (e.g., Reactome, MINT, BioGrid) [16]. Cytoscape 3.5.1. software was used to perform the representation of all the reported interactions.

Mechanistic evaluation of the miRNA target genes/proteins in the context of colon cancer: Artificial neural networks analysis
The possible molecular relationship between the targets of miRNAs and colon cancer was evaluated by means of Artificial Neural Networks (ANNs), following TPMS technology protocols [17]. This approach implies the generation of mathematical models of the biological process through the use of artificial intelligence techniques. Then mathematical models were solved by ANNs, which are supervised algorithms that identify relations between the different nodes in the network. The system attempts to find the shortest distance between miRNA targets and the effectors of the disease or pathways, thus ranking the list of miRNA targets ordered by their association with the selected disease or pathway. ANNs analysis yields a predictive value (p-value) that defines the probability of the existence of a relationship between each target and the diseases. Aiming to facilitate the understanding of the results, the obtained scores were divided in three categories: High (p<0,05); medium (p<0.25); low (p>0.25). At this point, these individual target scores were aggregated into a cumulative value for each miRNA to allow their prioritization on a rational basis.
Relationships between the miRNAs targets and colon cancer were assessed both for individual proteins and for combinations of two proteins, to identify combinations of targets with a synergistic effect on colon cancer. The synergy criterion was applied according to the approach described by Berenbaum [18], in which a significant synergic effect is considered when the overall effect is > 20% of the sum of the individual effect of the two proteins. Additionally, the relationship between the miRNA targets and the different colon cancer motives (individually) was also calculated.
When broad miRNA targets were investigated, Hsa-miR-335-5p is the miRNA presenting a higher proportion of unique targets compared to the rest of miRNAs. However, it should be noticed that Hsa-miR-335-5p is the one presenting the highest number of targets (2405 protein targets, 2 times more targets than the next miRNA, Hsa-miR-15b-5p with 1058). Hsa-miR-19a-3p and Hsa-miR-19b-3p are the miRNAs that present the lowest proportion of unique targets (2.8% and 18.4% respectively). In fact, their target profile overlaps substantially, probably because they belong to the miR-17-92 cluster located on chromosome 13q31.3 and could have potential redundant functions [19]. By contrast, Hsa-miR-18a-5p and Hsa-miR29a-3p had low proportion of targets, more than half of them being unique. Attending to the intensity of the reported miRNA target interaction (MTIs), Hsa-miR-29a-3p has a great proportion of MTIs identified through strong experimental techniques (72 strong and 112 weak interactions) ( Table 1). Taking into consideration variability across miRNAs, it has to be noted that the proportion of shared targets over the total when considering duplicates (36.6%) is clearly shorter than those unique to a single miRNA (63.4%). When we examined these frequencies on all distinct protein miRNA targets, we observed that most targets are related to only one miRNA (79.9%), whereas the rest (20.1%) are silenced by two or more miRNAs.

miRNA targets in CRC pathogenesis
We then characterized CRC at molecular level by literature search ( Table 2) and, based on these data, by using artificial neural networks we generated mathematical models of the disease (see Methods section). Through these models, we established a ranking of the identified unique or a combination of two miRNA targets according to their relationship to CRC and its pathophysiological processes (motives).
Our data demonstrated that the 6 miRNAs evaluated presented several proteins targets closely related to CRC showing different scores (  CRC contributors include both-oncogenes and tumoral suppressor proteins. In fact, a gained balance is obtained when looking to proteins acting as tumoral suppressors such as PDCD1, TET1, ETS2, TP53, TLE3, ACVR2A, TLE3. By contract, these 3 specific miRNAs could be targeting proteins that induce tumoral growth such as HIF1A, IGF1, GSK3b, VEGFA. These results show that altered expression of miRNAs is involved in deregulation of developmental programs and tumorigenesis. In CRC context, miRNA overexpression leads to an imbalance in tumoral suppressor genes and oncogenes haemostasis.

miRNA targets by pathophysiological pathways (motive analysis)
An analysis of the relationship of each miRNA protein target with CRC pathophysiological pathways was performed. According to these results, tissue invasion and metastasis-Epithelial to Mesenchymal transition (EMT) is assigned as the most relevant process (Table 5). In fact, EMT is the process that exhibits a higher number of miRNA targets with high association to effectors within CRC pathways (Supplementary Table 2). It has to be noted that process like immune system suppression/resistance and evading apoptosis possesses low levels of associations or weak functional relationship. Table 5 shows that EMT is on average better represented within miRNA as opposed to evading apoptosis or inflammation. These data suggest that evaluated miRNAs may be related to CRC mainly through EMT. Results also demonstrate that Hsa-18a-5p and Hsa-miR29a-3p are the two miRNAs more associated with CRC. On the contrary, values for Hsa-miR335-5p demonstrated its low importance probably due to the high number of targets reported for this miRNA.

Data integration and final score for targets
Targets combinations were also evaluated in order to identify combinations of proteins with effect on CRC. We found that when this effect was evaluated, results are similar than when evaluating individual targets: Hsa-miR-18a-5p and Hsa-miR29a-3p present total proteins showing low scores on average. However, there are some miRNAs that diverge from these values such as Hsa-miR-18a-5p and Hsa-miR-29a-3p.
The miRNAs presenting the highest proportion of targets highly related to CRC are Hsa-miR-18a-5p and Hsa-miR-29a-3p. Following them, we found Hsa-miR-19a-3p and Hsa-miR-19b-3p and Hsa-miR-15b-5p being Hsa-miR-335-5p the miRNA presenting the lowest proportion of targets highly related with CRC development. These data showed us that Hsa-miR-18a-5p and Hsa-miR-29a-3p are the miRNAs that binds with higher affinity to certain targets implicated in CRC development. Among the 6 miRNAs, Hsa-miR-29a-3p stands out as the candidate that displays better properties (high proportion of highly related) (more high and less low) as regulator of targets affecting colorectal carcinogenesis.
Looking at the 20 miRNA targets which display the best scores due to their relationship with CRC pathogenesis (Supplementary Table 1) we found proteins most usually described as oncogenes but also some tumor suppressor genes.

Tissue invasion and metastasis and Epithelial to Mesenchymal Transition (EMT) 71
Sustained angiogenesis 27 Immune system suppression/resistance 25

Cell growth and proliferation 22
Inflammation 17 Proteins related to epigenetic instability 17 Evading apoptosis 9 Note: The last column contains the arithmetical ratio -low/(high+medium).  Hsa-miR-29a-3p was previously described as contributor for its diagnostic potential for high-risk adenomas [10]. In this study, miRNA29a together with miRNA92a in serum samples correlated with adenoma size and adenoma total number within the colorectum. Yang et al. [21] in a recent study have demonstrated that Hsa-miR-15b-5p are significantly associated with progression free survival (PFS) and/ or overall survival (OS) of colorectal cancer patients independent of tumor stage and age at diagnosis.
Although these publications showed individual miRNAs as potential biomarkers for early diagnosis of AA and CRC, there is a need for more careful evaluation of targets and biological function that could support further clinical studies. A comprehensive literature research evidenced in our study indicated a functional role of miRNAs in colorectal carcinogenesis giving priority to targets and effectors driving CRC through EMT [23].
The more general targets the miRNA has, the less they seem to be associated to CRC development: while hsa-miR-29a-3p just covers 184 targets overall and presents a L/HM rate of 1,6, hsa-miR-335-5p covers 1887 targets (mapping for more high and medium targets, in absolute terms) and has the highest ratio [3,6]. This phenomenon could be caused by the fact that the targets of the miRNAs with less reported targets are better known proteins, which are better represented in the models, whereas some of the targets of hsa-miR-335-5p are less known proteins and are worse represented in the models. In fact, most of the MTIs reported for hsa-miR-335-5p (2299 out of 2405) have been identified in the same study [12], and a great proportion of its targets are reported through weak techniques [24].
Hsa-15b-5p has been associated to cell survival by affecting TP53 pathway through Bcl2 interaction in patients with inflammatory bowel diseases and associated cancer [25]. Also, this miRNA was found overexpressed in exosomes of colorectal cancer patients presenting shorter time to relapse for liver metastasis [26].
As commented previously, EMT has been extensively characterized in CRC while evading apoptosis, inflammation or epigenetic instability either has lower role in CRC or they are less studied. Attending to the two genes that seem to function as best targets as well as CRC effector with high scoring in EMT are ETS2 and ACVR2A. Both of them could be acting as promoters of carcinogenesis when inhibited by miRNAs. ETS2 gene encodes a transcription factor which regulates the highest proportion of high-associated target pairs whereas Hsa-miR335-5p presents the highest proportion of low-associated target pairs. Importantly, Hsa-15b-5p presents the highest proportion of medium-associated target pairs but the low-associated targets pairs (even less than Hsa-miR-18a-5p and Hsa-miR-29a-3p) (Figure 1 and data not shown).
Finally, aiming to elaborate a final ranking that could enable effective prioritization of the miRNAs based on their targets, the results of previous analysis taking into account only high and medium relationship with CRC were integrated into a single ranking and a composite score was assigned accordingly. To that purpose, each parameter for each miRNA was assigned a value in a normalized score between 0 (worst) and 1 (best) and the average of single and combined was indicated as global score (Table 6). According to this final ranking, Hsa-miR-18a-5p, Hsa-miR29a-3p and Hsa-15b-5p are the miRNAs presenting highest scores of sensitivity in CRC targeting.

Discussion
In the present study, we performed a biological contextualization of a 6 miRNAs signature (Hsa-miR-15b-5p, Hsa-miR-29a-3p, Hsa-miR-335-5p and Hsa-miR-18a-5p, Hsa-miR-19a-3p, Hsa-miR-19b-3p) previously described as biomarkers, within CRC screening setting. We elucidate which of these 6 miRNAs conferers the highest contribution to its diagnostic potential based on targets and effectors taking part in colorectal cancer development. We also provided evidence of functional cooperation between miRNA targets and EMT related effectors in CRC.
Our data highligthed that although 6 miRNAs included in the signature target proteins implicated in CRC development, Hsa-miR-18a-5p, Hsa-miR29a-3p and Hsa-15b-5p are likely the most important contributors to the diagnostic value of the signature by regulating EMT process.
The signature evaluated in this study was originally identified on miRNAs overexpressed in AA and CRC tissues and further validated in plasma samples [5]. However, generally speaking-only a few miRNAs abundantly expressed in cancer cells are detectable in blood [20] and are those one with potential for early diagnosis purposes when minimally invasive methods are considered.

Proteins related to epigenetic instability
Hsa-mi-15b-5p 0. genes involved in development and apoptosis. Its contribution may be specific to intestinal tumors driven by the Wnt pathway. Particularly, it has been described that ETS2 has an epithelial cell autonomous, tumor repressive activity for colon tumors [27]. ACVR2A encodes activin A type II receptor which belong to the transforming growth factor-beta (TGF-beta) superfamily of structurally related signalling proteins. In hypermutated tumours, ACVR2A is frequently mutated [28]. In fact, TGFBR2 and ACVR2A are commonly mutated in microsatellite unstable (MSI) colon cancers, which is the second most common genomic subtype [29].  Table 6: Final ranking of miRNA biomarker candidates considering single targets and combinations.
Surprisingly, LAMC2, IGF1 we also found as targets although have been traditionally described as overexpressed during cancer proliferation. IGF1 seems to be activated before carcinogenesis, and may promote the growth and malignant transformation of adenomatous polyps [30] and expression of laminin gamma 2 (LAMC2), matrix metallopeptidase 7 (MMP7) and EMT-related molecules were elevated in the colon tumors front [31]. Different interferences between other targets and effectors, as well as mutated, inactivated or inactivated genes could explain their implication here.

Conclusion
In summary, this study provides evidence for the potential targets of miRNAs and its usefulness of this signature for minimally invasive screening purposes. Hsa-miR-18a-5p, Hsa-miR29a-3p and Hsa-15b-5p are key regulators of CRC development mediating EMT, as proposed by different authors separately in previous studies.
Being a hot topic nowadays, miRNAs are a novel and specific type of biomarkers derived from tumor cells and its concentration is associated with invasive ability of the tumor cells and tumor microenvironment. Several publications hypothesized that certain miRNAs can be transferred to other cells and tissues free in plasma or by means of exosomes and build a suitable microenvironment for tumour growth and cell metastasis. Further studies are required in order to explore in detail biological function of miRNA and interactions in vivo.