PMC

Functional analysis of motif modules. (A) Global view of specific transcription-factor motifs that are enriched in gene sets (by hypergeometric P-value). Different types of gene sets are distinguished by color, and their intensity corresponds to the significance of the motif module enrichment. Unsupervised clustering was applied to the resulting matrix. Several clusters are shown in detail (indicated by arrows), with the figure location in parentheses. (B) Detailed image of motif module clusters with function-specific enrichment. Novel conserved motif modules that we experimentally validated are highlighted in red. Notations following some transcription-factor names (ex. Q6_01) are identifiers for variants of the motifs according to TRANSFAC. Stand-alone sequences or sequences in parentheses following a transcription-factor name represent consensus binding motifs from ref. () (key for combination of nucleotides: Y = C or T; R = A or G; W = A or T; S = C or G; K = T or G; M = C or A; N = unknown). Individual motifs of motif combinations are separated by a colon. (C) Detailed image of tissue/organelle-specific enrichments. (D) Detailed image of novel enrichments.

Uncharacterized motifs regulate cell cycle progression. (A) Clinical annotation enrichment of the array signature of E2F_Q6_01 and the four uncharacterized motif modules. (B) Significance of enrichment of the indicated motif module target genes with all other motif targets is shown. (C) Shown is the average expression of the indicated motif module genes found in a set of periodically expressed cell cycle genes in HeLa cells (). The stage of cell cycle that the pattern of expression most resembles is indicated on the left. Orange bars signify S phase; arrows signify mitoses. (D) Decoy oligos corresponding to the uncharacterized motif sequences inhibit target gene expression in HeLa cells. Shown is the average log₂ gene expression (± SE) of genes predicted to have the indicated motif (orange) or those that lack the motif (gray). (*) P < 0.05 compared with genes without motif, Student’s t-test. (E,F) Decoy oligos inhibit cell cycle progression in HeLa cells. (E) Example of BrdU immunofluorescence staining and DAPI counterstain after introduction of scrambled oligos or decoy oligos for the ACTWSNACTNY motif. (F) Average percentage (± SE) of BrdU-positive cells after transfection of the indicated scrambled or decoy oligos. The cell cycle phase in which the cells arrested (as determined by FACS) is indicated on the left. (*) P < 0.01 compared with the respective scrambled oligo, Student’s t-test.

Overview of our approach and comparison to other methods. (A) Flowchart of our motif functional characterization pipeline. (B) Comparison of the number of significant enrichments between gene sets from Gene Ontology (GO) () and motif modules defined using our approach, GO and an alternative approach based on best motif occurrences, or GO and our approach applied to permuted promoters. For each particular FDR threshold f, the comparison shows the number of GO-motif target set pairs with FDR < f. (C) Comparison of the ability of our method (Y-axis) and best motif occurrences method (X-axis) to predict TP53 binding to 542 sequences experimentally measured to be strongly bound to TP53 by genome-wide chromatin immunoprecipitation followed by sequencing of paired end ditags (PET) (). For each experimentally identified PET region, shown is the predicted binding P-value of each method, determined by comparing the likelihood score of the original region to that obtained in 10,000 regions randomly selected from the human genome. PET regions are separated into two types by their length (green and blue). Of the 542 PET regions, 397 have a fivefold lower P-value in our method (gray shaded area above diagonal), compared with nine with a fivefold lower P-value in the score-based method.

Global view of cis-regulation in cancer using motif modules. (A) Clinical annotations organized by their associated activated and deactivated motif modules. We used the “module map” method () on a compendium of 1975 arrays spanning 22 different tumor types. We found arrays in which the genes of each motif module were coordinately induced or repressed (P < 0.05). We then tested the enrichment of these coregulated arrays of each motif module in clinical annotations of the arrays () (P < 0.05, corrected for multiple hypothesis testing using FDR), and applied unsupervised hierarchical clustering to group together clinical annotations that show enrichments for the same motif modules. Intensity of each enrichment corresponds to the percentage of microarray samples that have motif module target genes significantly induced (up) or repressed (down). The colors of the branch arms represent specific groups of clinical annotations (the specific tissues listed correspond to the tissue origin of the cancer). (*) Groups of metastatic/high-grade cancers. Locations of clusters analyzed in detail are shown at the top and left of the figure. (B) Clinical annotation enrichment of the array signature of E2F and NFY-containing motif modules (E2F-NFY). (C) Motif modules enriched in the leukemia/lymphoma/breast/liver (Lk/Lym/Bst/Lvr) cluster. (D) Shown are PAX4 motif module genes noted in C that are significantly induced or repressed in the indicated grade/stage of breast/lung cancer, respectively. PAX4 target gene induction is enriched in the higher grade/stage tumors (P < 0.05, χ² test). (E) Motif modules enriched in metastatic/high-grade tumors (*Met) of different histologic origins.

Extensive coupling between mechanisms of transcriptional and post-transcriptional control. (A) Shown is the number of gene sets of predicted microRNA targets (; ) that are significantly enriched for High or Low CpG promoter genes () (P < 0.05). The P-value of enrichment of the microRNAs analyzed in C is shown. (B) Concomitant change in genes expression in High vs. Low CpG promoter genes. Average expression (± SE) is shown. (*) P = 0.001, Student’s t-test. (C) Gene repression by microRNA overexpression (). (Left) Coordinate repression of predicted microRNA target genes. (Right) Coordinate repression of high CpG promoter genes. Each column is a microRNA overexpression experiment; each row is a gene set of microRNA targets or CpG promoters. The degree of coordinate gene regulation over expectation by chance alone is quantified by the color scale. The enrichment of miR-206 targets in miR-1 overexpression is due to the 98% overlap of their predicted targets by PicTar (). (D) A common modularity between motif modules and microRNA targets. We compared singleton motif modules with fewer than 1000 member genes against predicted microRNA targets (PicTar). (Left) Percentage of motif modules whose targets are enriched in the targets of k different microRNAs, for k = 1, 2, 3, 4, >5. (Right) Percentage of microRNA target modules whose targets are enriched in the targets of k different motif modules, for k = 1, 2, 3, 4, >5. (E) CEBPA_01 targets significantly overlap with miR-20a targets (P = 6 × 10⁻¹¹), and FOXO1_01 targets significantly overlap with miR-9 targets (P = 5 × 10⁻¹⁰).