In this study, we incorporated miRNAs into a traditional GRN to investigate the correlations between PPIs and regulatory motifs formed by miRNAs, TFs, and target proteins/genes. The regulatory motifs were classified into four types: single-regulation, co-regulation, crosstalk, and independent. Traditionally, random sampling methods are usually applied to evaluate the significance of PPI numbers among a group of proteins, but this is very time-consuming. In addition, random sampling is not suitable for analyzing complicated regulatory networks, because the whole process should be redesigned for different motif members. In order to improve the efficiency of the evaluation process without loss of general applicability, we calculated the significance of PPI enrichment for different motifs based on the Bernoulli distribution; in other words, we regarded PPI gain and lost as a Bernoulli process. This allowed the whole evaluation process to be kept under constant time (O(1)).
Among the four types of motifs, the strong correlation between single-regulation and PINs has been well-discussed [21
], and a correlation with the co-regulation type has also been reported [37
]. Single-regulation motifs analyzed here showed consistent conclusions with previous studies. Our investigation into co-regulation motifs has further provided complementary analysis and given insights that have not been addressed in any previous studies. More importantly, we proposed that the third type of motif -- the crosstalk motif -- could be another prominent pattern in GRNs. Crosstalk motifs were defined as the private target gene sets of two corresponding regulators, TFs and/or miRNAs, which shared at least two targets. In human PINs, crosstalk motifs were significantly enriched in PPI contents and network properties. To summarize the analysis of network properties, crosstalk motifs displayed several features: 1) high degree, 2) high closeness, 3) high density, 4) high clique level, and 5) short characteristic path length. In PINs, proteins with a high degree are usually called "hub proteins", those with high closeness centrality are usually called "central proteins", and those with high density, short characteristic path length, and high clique level are usually called "modular proteins". Therefore, the regulators which participate in crosstalk motifs tend to regulate hub proteins, which are usually more essential than non-hub proteins [39
], and modular proteins, which usually form important protein complexes or modules in human PINs [42
]. Additionally, we investigated the enriched functions of the crosstalk motifs. For all three types of regulator pairs, the majority of enriched crosstalk functions are associated with positive/negative regulation of cellular metabolic processes. Notably, miRNA-miRNA crosstalk motifs are not only associated with regulation-related functions, but also response to insulin stimulus. This is consistent with previous findings that miRNAs preferentially regulate downstream components, such as TFs, in signaling networks [19
]. Moreover, we demonstrated the functional features within the crosstalk motifs with the highest PPI z
-score and proposed a potential cancer-related motif, TP53-miR-200bc/429/548a. Consequently, this crosstalk motif might play an important role in living cells through regulating those essential or pivot proteins in human PINs.
Since our analysis relies on limited data sources from online databases to construct human PINs and GRNs, we carried out further examinations to test the robustness of our conclusions. With respect to miRNA regulation, all current online databases which provide predicted human miRNA targets still have room for improvement both in approach and performance [32
]. Accordingly, we repeated our analysis with another database, miRBase [36
], and were able to reach a consistent conclusion (Figure S8-S13, Table S1 and S2 in Additional file 1
). Considering the incomplete and noisy human PPI data, we performed the same analysis with combined PPI data from HPRD and BioGRID [31
] databases and also obtained consistent conclusions (Figure S14-S22, Table S3-S5 in Additional file 1
). Therefore, these re-analyses provide further evidence to support the robustness of our conclusions. With ongoing efforts to improve the completeness of PPI data and GRNs, we will be able to further investigate and confirm the correlations between PPIs and regulatory motifs in the future.
In summary, we proposed a computational approach to investigate the significance of regulatory motifs formed by TFs/miRNAs and their corresponding targets in human PINs. With this approach, we screened four types of regulatory motifs, single-regulation, co-regulation, crosstalk, and independent, from human GRNs and investigated their correlations with PPIs. Among the four types of motifs, the crosstalk motif emerged as a potentially significant motif with important roles in PINs, which has not been previously reported. We suggested that this motif might play an important role in living cells because of its strong correlations with PPIs and significant network properties in human PINs.