Computational analysis of target hub gene repression regulated by multiple and cooperative miRNAs

Abstract

MicroRNA (miRNA) target hubs are genes that can be simultaneously targeted by a comparatively large number of miRNAs, a class of non-coding RNAs that mediate post-transcriptional gene repression. Although the details of target hub regulation remain poorly understood, recent experiments suggest that pairs of miRNAs can cooperate if their binding sites reside in close proximity. To test this and other hypotheses, we established a novel approach to investigate mechanisms of collective miRNA repression. The approach presented here combines miRNA target prediction and transcription factor prediction with data from the literature and databases to generate a regulatory map for a chosen target hub. We then show how a kinetic model can be derived from the regulatory map. To validate our approach, we present a case study for p21, one of the first experimentally proved miRNA target hubs. Our analysis indicates that distinctive expression patterns for miRNAs, some of which interact cooperatively, fine-tune the features of transient and long-term regulation of target genes. With respect to p21, our model successfully predicts its protein levels for nine different cellular functions. In addition, we find that high abundance of miRNAs, in combination with cooperativity, can enhance noise buffering for the transcription of target hubs.

Figure 1.

Illustration of miRNA target hub regulatory network. Once an miRNA target hub gene is transcribed, its mRNAs are post-transcriptionally regulated by many miRNAs whose expression can be activated or inhibited by numerous TFs. In animals, the complementarity between an miRNA and the target hub mRNA usually results in reduced protein expression through translational repression or mRNA deadenylation followed by degradation. Target hubs are highly interconnected via protein–protein interactions, so that its regulatory network may contain interlocked network motifs, such as feedback and feedforward loops.

Figure 2.

Schematic representation of the approach. Biomedical knowledge and quantitative data are integrated using mathematical modelling to describe the regulatory role of miRNAs in signalling pathways and gene regulatory networks. It considers data retrieval, iterative cycles of model construction and calibration and computational simulations. The mathematical model obtained is suitable to test hypotheses and simulate complex biological scenarios associated with cellular function-related variability.

Figure 3.

The regulatory map of p21. Visualization of the annotated SBML file integrating information about the regulation of miRNA target hub p21. miRNAs are classified according to their mechanisms of p21 regulation (deadenylation denoted by blue box and translation repression denoted by green box). TFs are divided into three groups: TFs regulating only p21 expression (red box), TFs regulating only miRNAs (yellow box) and those that regulate the expression of p21 and some of the miRNAs (light blue box). The latter accounts for putative feedforward loops. p21 protein interaction partners are clustered into groups (grey boxes). The small boxes, highlighted in green, account for putative feedback loops formed by p21, given miRNAs and the interacting partner (e.g. p21→AKT1→miR-132⊣p21). Highlighted in blue are TFs that down-regulate expression of the considered gene targets. The purple box accounts for cellular functions assigned to the considered TFs. A high resolution version of the regulatory map is available in Supplementary Data.

Figure 4.

(A) Matrix of cooperating and non-cooperating p21 targeting miRNAs. The intersections of pairs of miRNAs denote their potential for cooperation. Grey cells indicate a non-interacting pair and blue cells denote potentially cooperating pairs based on binding site proximity range (13–35 nt) defined by Saetrom et al. (7). Red cells, however, indicate that cooperation cannot be established due to binding sites with extensive overlap or miRNA pairs that share the same binding site. The figures inside specify the fraction of binding sites interacting with the respective partner (the figures in brackets represent the miRNA on the x-axis). (B) Binding sites of regulatory miRNAs in the p21 3′-UTR.

Figure 5.

Regulation of p21 expression by miR-572 and miR-93. The experimental data of p21 protein expression in response to genotoxic stress (Exp) are compared with the model predictions (Sim) in four biological scenarios: (i) both miRNAs are normally expressed (NTC), (ii) miR-572 is overexpressed (miR-572), (iii) miR-93 is overexpressed (miR-93) and (iv) both miRNAs are overexpressed (miR-572 + 93). n.u.: normalized unit.

Figure 6.

The three regulatory mechanisms proposed, particularized for the miRNA pair miR-93 and miR-572. For each depicted target regulation mechanism, the p21 expression level is computed for different combinations of the initial concentrations of miR-93 and miR-572. The black symbols stand for four different miRNA initial concentration scenarios: both miRNAs are normally expressed ([1, 1], filled inverted triangle), miR-572 is overexpressed ([1, 10], filled diamond), miR-93 is overexpressed ([10, 1], filled circle) and both miRNAs are overexpressed ([10, 10], filled triangle). The legend bar represents the p21 expression ranging from basal levels (1) to full repression (0). n.u.: normalized unit.

Figure 7.

(A) p21 expression levels at different strengths of cooperativity (K) between miR-93 and miR-572. The dashed vertical lines represent thresholds of K, for which the miRNA-mediated p21 repression displays different behaviours. Different symbols correspond to different miRNA initial concentration scenarios: both miRNAs are normally expressed ([1, 1], filled inverted triangle), miR-572 is overexpressed ([1, 10], filled diamonds), miR-93 is overexpressed ([10, 1], filled circles) and both miRNAs are overexpressed ([10, 10], filled triangles). The small plot zooms in on the first scenario for illustrating the sigmoid shape of the expression levels of p21 at different K. (B) The role of RBPs in multiple miRNA-mediated target repression. Our analysis suggests the activity of RBPs in combination with the repression by multiple miRNAs here described can induce a tunable-like target repression. n.u.: normalized unit.

Figure 8.

The dynamics of p21 to upstream signals. (A) The sketch of the stimulus signal computed (left) and the p21 response (right). The parameters μ and τ account for the amplitude and duration of the stimulus signal, respectively. (B) The predicted p21 response peak to different stimulus signals for different miRNA abundance scenarios. Each plot refers to a set of stimulus signals with the same amplitude range but different durations (µ ∈ [10⁻² 10¹] n.u., τ = [1, 10, 24] h). Five miRNA abundance scenarios were considered and they are as follows: (i) off (solid red line): no expression of any miRNA, (ii) on (black dash-dot line): normal expression of all miRNAs, (iii) on + C (blue circles): normal expression with miRNA cooperativity, (iv) on(x10) (green dotted line): overexpression of all miRNAs and (v) on(x10) + C (magenta dashed line): overexpression with miRNA cooperativity. The grey dashed lines represent the threshold, which is equal to 10% of the basal p21 expression level. (C) The time-series plots of the p21 response when the p21 is stimulated by the low (µ = 0.1 n.u.) or high (µ = 1 n.u.) amplitude and long-lasting (τ = 10 hr) stimulus signals. n.u.: normalized unit.

Figure 9.

Cooperative miRNA regulation of p21 expression in different cellular functions. The statuses (on/off) of miRNAs for specific cellular functions are derived from the GO terms of their TFs. The horizontal bars represent the predicted p21 steady-state levels for different cellular functions when the miRNA cooperativity is considered (C) or not (no C). The equations and parameter values can be found in Supplementary Materials Section 3.3. n.u.: normalized unit.

Figure 10.

Examples of function-specific regulation of p21. The TFs (rectangles) regulate p21 directly and indirectly through miRNAs (parallelograms): inflammatory response (A) and cell cycle (B).

Figure 11.

Target hub regulation by cooperative miRNA repression and feedforward loops. The regulatory loops detected in our analysis and cooperative miRNA repression can synergize in the regulation of the target. This synergy can generate further fine-tuning in the response. As an example, we here display a module which combines an incoherent feedforward loop with transcriptional delay in miR₁ expression (TF₁→TgHub, TF₁→miR₁, miR₁⊣TgHub) and cooperative target repression ([miR₁ AND miR₂]⊣TgHub, left). When miR₂ expression is triggered by a different transcription factor TF₂, it can modulate the transient peak and the long-term target expression levels. Our simulations indicate that, beyond a threshold in TF₂-mediated expression of miR₂, a step-like transcriptional activation of TF₁ induces a transient peak in target hub expression (middle). The intensity of that transient peak is modulated by the levels of TF₂, which promotes the expression of the second cooperative miRNA, miR₂ (right). n.u.: normalized unit.