MicroRNA-mediated feedback and feedforward loops are recurrent network motifs in mammals

Abstract

MicroRNAs (miRNAs) are regulatory molecules that participate in diverse biological processes in animals and plants. While thousands of mammalian genes are potentially targeted by miRNAs, the functions of miRNAs in the context of gene networks are not well understood. Specifically, it is unknown whether miRNA-containing networks have recurrent circuit motifs, as has been observed in regulatory networks of bacteria and yeast. Here we develop a computational method that utilizes gene expression data to show that two classes of circuits-corresponding to positive and negative transcriptional coregulation of a miRNA and its targets-are prevalent in the human and mouse genomes. Additionally, we find that neuronal-enriched miRNAs tend to be coexpressed with their target genes, suggesting that these miRNAs could be involved in neuronal homeostasis. Our results strongly suggest that coordinated transcriptional and miRNA-mediated regulation is a recurrent motif to enhance the robustness of gene regulation in mammalian genomes.

Figure 1. Two Classes of miRNA-containing Circuits

U denotes an upstream factor or a process that regulates the transcription rate of a miRNA (m) and one of its targets (T). In Type I (II) circuits, the transcription rate of m and T are positively (negatively) co-regulated by U. Both Type I and II circuits are topologically characterized by miRNA-mediated feedback or feedforward loops acting on T. Note that in the case of the positive (Type II) and negative (Type I) feedback circuits, U regulates the transcription rate of m indirectly through T.

Figure 2. Targeting Bias Analysis of Human Embedded miRNAs

(A) The expression profiles of human embedded miRNA host genes in the Novartis atlas. Four prominent clusters were identified by hierarchical clustering. (B) Targeting bias enrichment scores of miRNAs from major expression clusters. The color scale reflects the degree of deviation from the 5% significance level (ΔZ), while black denotes insignificant enrichment/depletion (i.e. P>0.05).

Figure 3. Computing the CE Score Profile of a miRNA given an Expression Data Set

Step 1: sort all genes on the microarray based on their expression correlation to the miRNA host gene. Step 2: for each sliding-window subset, the conservation rate (CR) of the miRNA seed-matches is computed by counting the number of occurrences of conserved and non-conserved matches (gray box). Step 3: The CR is used to compute the CE score based on the background CE score distribution obtained from randomly drawn gene sets (σ denotes the standard deviation of the background CR distribution). The CE score is the number of standard deviations (σ) away from the expected conservation rate ( $\overline{CR}$) of gene sets drawn at random. Step 4: the centers of each sliding window subset (e.g. the top-10 percentile set centers at 95) are plotted against the corresponding CE scores. We summarize the data by showing the scores of the top-10, middle-10, and bottom-10 percentile sets as a heatmap. The yellow scale reflects the amount of deviation (ΔCE) from the 5% significance level (i.e. CE>1.65). Insignificant scores (CE <= 1.65) are shown in black.

Figure 4. CE Analysis of the Novartis Expression Atlas for Human and Mouse

(A) ΔCE scores of the top-10, middle-10, and bottom-10 percentile sets of human embedded miRNAs we analyzed. The scores of miRNAs in Group II (immune/cancer expression signature) and IV (brain-enriched) are shown. Since some host genes have multiple microarray probes, the probe IDs are also shown. (B-D) Representative CE profiles of biased miRNAs along with their expression patterns. Expression conditions: im=immune/cancer; br=brain; org= organs; ts=testis-related. (B) CE score profile of miR-198. Expression patterns of genes in the top-10 percentile set are also shown. (C) miR-153 is brain-enriched and its Type I CE score profile is typical of miRNAs in Group IV. (D) miR-15b and miR-16 (embedded in the same host gene) exhibit both Type I and II biases. (E) ΔCE scores of the top-10, middle-10, and bottom-10 percentile sets of mouse embedded miRNAs we analyzed. Brain-enriched miRNAs only exhibit significant CE scores in their top-10 percentile sets.

Figure 5. CE Analysis of the Mouse Motor Neuron Development (MDEV) and Homogeneous Neuronal Cell (NCELL) Data Sets

(A) ΔCE scores of the top/middle/bottom percentile sets of mouse embedded miRNAs in the MDEV data set. (B-C) CE score profiles of miR-103 and miR-342, representative of miRNAs showing Type I and II bias, respectively. (C) ΔCE scores of the top-10, middle-10, and bottom-10 percentile sets of embedded miRNAs in the NCELL data set.

Figure 6. Circuits with miRNA-mediated Positive and Negative Feedback Loops

(A) The NRSF/miR-29 network. A miRNA can be embedded in both Type I and II circuits. In addition to potentially targeting NRSF, miR-29 putatively targets a significant number of neuronal genes (T) that are co-repressed by NRSF. (B) Illustrates how a miRNA can be embedded simultaneously in both Type I and II circuits in relation to a single target. U1 and U2 regulate the miRNA and its target via different cis regulatory modules. Depending on whether U1 or U2 is active, the expression of the miRNA will positively or negatively correlate with that of the target. (C) A hypothetical toggle-switch network with two transcription factors (T₁ and T₂) and two miRNAs (m₁ and m₂) with both Types I (T₁-m₁/T₂-m₂) and II (T₁-m₁-T₂ and T₂-m₂-T₁) circuits. Such networks are typically characterized by two stable states where only one of T₁ or T₂ is active. Each miRNA is involved in a positive and a negative feedback loop. The former functions in conjunction with other positive feedbacks to reinforce the active state (T₁ or T₂ active) and allow transient signals to turn the circuit on or off. The latter could buffer fluctuations and prevent random switching events. MiRNAs may be especially effective at providing feedback with short delays. (D) The c-Myc/E2F1/miR-17-20 network in human. c-Myc and E2F1 can activate each other's transcription and both can activate the transcription of the miR-17 miRNA cluster. Two Type I circuits are present: the miR-17-mediated negative feedback to E2F1, and the c-Myc-activated feedforward loop to E2F1. These negative loops mediated by miRNAs could prevent random activation of c-Myc/E2F1 due to fluctuations in their expression.