Transcriptional regulatory networks underlying gene expression changes in Huntington's disease

• DOI: 10.15252/msb.20167435
• Publication Date: 2018-03-26T11:30:08Z
• Authors: Seth A Ament; Jocelynn R Pearl; Jeffrey P Cantle; Robert M Bragg; Peter J Skene; Sydney R Coffey; Dani E Bergey; Vanessa C Wheeler; Marcy E MacDonald; Nitin S Baliga; Jim Rosinski; Leroy E Hood; Jeffrey B Carroll; Nathan D Price
• Keywords: Proteomics; 0301 basic medicine; 570; Medicine (General); QH301-705.5; 610; Article; SMAD3; Network Biology; Mice; 03 medical and health sciences; R5-920; Animals; Humans; Molecular Biology of Disease; Gene Regulatory Networks; Protein Interaction Maps; Smad3 Protein; Biology (General); transcription factor; Gene Expression Profiling; Huntington's disease; Articles; Corpus Striatum; Disease Models, Animal; Huntington Disease; Gene Expression Regulation; Genome-Scale & Integrative Biology; transcriptional regulatory networks; Transcription Factors

Article26 March 2018Open Access Transparent process Transcriptional regulatory networks underlying gene expression changes in Huntington's disease Seth A Ament Institute for Systems Biology, Seattle, WA, USA Genome Sciences and Department of Psychiatry, University Maryland School Medicine, Baltimore, MD, Search more papers by this author Jocelynn R Pearl Molecular & Cellular Biology Graduate Program, Washington, Altius Biomedical Sciences, Jeffrey P Cantle Behavioral Neuroscience Psychology, Western Washington University, Bellingham, Robert M Bragg Peter J Skene Basic Division, Fred Hutchinson Cancer Research Center, Sydney Coffey Dani E Bergey Vanessa C Wheeler Neurogenetics Unit, Center Human Genetic Research, Neurology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, Marcy MacDonald Nitin S Baliga orcid.org/0000-0001-9157-5974 Jim Rosinski CHDI Management, Foundation, Princeton, NJ, Leroy Hood B Carroll orcid.org/0000-0003-1711-8868 Nathan D Price Corresponding Author [email protected] orcid.org/0000-0002-4157-0267 Information Ament1,2,‡, Pearl1,3,4,‡, Cantle5, Bragg5, Skene6, Coffey5, Bergey1, Wheeler7, MacDonald7, Baliga1, Rosinski8, Hood1, Carroll5 *,1 1Institute 2Institute 3Molecular 4Altius 5Behavioral 6Basic 7Molecular 8CHDI ‡These authors contributed equally to work *Corresponding author. Tel: [phone]; E-mail: (2018)14:e7435https://doi.org/10.15252/msb.20167435 PDFDownload PDF article text main figures. Peer ReviewDownload a summary the editorial decision including letters, reviewer comments responses feedback. ToolsAdd favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InMendeleyWechatReddit Figures Info Abstract occur presymptomatically throughout (HD), motivating study transcriptional (TRNs) HD. We reconstructed genome-scale model target genes 718 transcription factors (TFs) mouse striatum integrating genomic binding sites with transcriptome profiling striatal tissue from HD models. identified 48 differentially expressed TF-target modules associated age- CAG repeat length-dependent Htt knock-in replicated many these associations independent transcriptomic proteomic datasets. Thirteen predicted were also human disease. experimentally validated specific prediction that SMAD3 regulates HD-related using chromatin immunoprecipitation deep sequencing (ChIP-seq) striatum. found occupancy confirmed our model's are downregulated early Synopsis This models network controlling striatum, predicts central role 13 whose patterns change as result expansion is built TF data. The Experimental validation was produced Introduction Massive accompany diseases, yet we still know relatively little about how mediate changes. Comprehensive characterization disease-related can help clarify potential mechanisms prioritize targets novel therapeutics. variety approaches have been developed reconstruct interactions between TFs their genes, focused on reconstructing physical locations factor (Gerstein et al, 2012; Neph 2012), well computational algorithms utilizing co-expression infer relationships (Friedman 2000; Bonneau 2006; Margolin Huynh-Thu 2010; Marbach Reiss 2015). These yielded insights into regulation range biological systems, accurate, mammalian TRNs remain elusive. Several lines evidence point (HD). fatal neurodegenerative caused dominant inheritance polyglutamine (polyQ)-coding expanded trinucleotide (CAG) HTT (MacDonald 1993). Widespread detected post-mortem brain cases versus controls (Hodges 2006), among earliest detectable phenotypes (Luthi-Carter Seredenina Luthi-Carter, preprint: 2016; Langfelder 2017). particularly prominent most profoundly impacted region (Vonsattel 1985; Tabrizi 2013). Replicable patients include downregulation related synaptic function medium spiny neurons accompanied upregulation neuroinflammation (Seredenina Labadorf Some may be directly functions huntingtin (HTT) protein. Both wild-type mutant (mHTT) protein shown associate DNA, mHTT interacts histone-modifying enzymes states (Benn 2008; Thomas Seong 2010). Wild-type has regulate activity some (Zuccato 2007). Also, high concentrations nuclear aggregates sequester co-factor proteins interfere finding, though it unknown whether occurs at physiological (Wheeler Shirasaki Li 2016). Roles several characterized 2003; Arlotta Tang Dickey 2015), but lack global downstream pathological processes they regulate. availability large transcriptomics datasets now making possible begin comprehensive analysis disease, al (2016) generated RNA-seq 144 mice heterozygous an allelic series mutations differing lengths, 64 littermate controls. They used identify co-expressed altered However, analyses did not attempt any responsible Here, investigated roles core drive HD, biology approach. machine learning strategy combining publicly available DNase-seq data overrepresented least five fifteen conditions defined mouse's age length, differential multiple Based coordinated transcripts Smad3 its hypothesized regulator Using (ChIP-seq), demonstrate repeat-dependent tissue. In conclusion, results TRN ChIP-seq studies reveal new drivers complex Results information (TFBSs) (Fig 1A). 871 genome digital footprinting. footprints 23 tissues (Yue 2014) Wellington (Piper Footprints short regions reduced accessibility DNase-I enzyme one Our goal single TFBS could make useful predictions even which available. 3,242,454 footprints. Genomic often indicative DNA-binding scanned 2,547 sequence motifs TRANSFAC (Matys JASPAR (Mathelier 2014), UniProbe (Hume high-throughput SELEX (Jolma 2013) predict (TFBSs), compared TFBSs start sites. considered if had site within 5-kb upstream TSS, previously maximize footprinting (Plaisier Figure 1. Reconstruction (TRN) Schematic reconstruction tissue-specific co-expression. Training (black) test set (blue) accuracy model. Genes ordered x-axis according training (r2, actual expression). dotted black line indicates cut off number explained > 50% variation Distribution regulators per gene. TF. Enrichments supported DNase tissues. Download figure PowerPoint To assess model, experiments ENCODE ChEA (Lachmann Appendix Fig S1). For 50 52 TFs, there significant overlap sets (FDR < 1%). median 78% recall 22% precision. That is, majority true-positive made false-positive predictions. Low precision expected since typically occupy only subset given Nonetheless, low need additional filtering steps relevant context. sought active evaluating profiles 208 (Langfelder general idea likely strong TF-gene co-expression, allow us direct interactions. step removes expression: Of predictions, retained fit regression each based combined or ±5 kb gene's site. LASSO regularization select together approach extends previous methods (Tibshirani, 1996; Friedman Chandrasekaran 2011; Haury 2012) introducing TFBS-based constraints. preliminary work, elastic net (α = 0.2, 0.4, 0.6, 0.8, 1.0) penalties evaluated performance fivefold cross-validation (see Materials Methods). selected highest correlation sets. predictive comparing observed levels 13,009 1B). Prediction nearly identical accurately (r 0.94; generally (Appendix S2). removed poorly before moving set. final contains regulated via 176,518 (Dataset EV1). 14 regulating 147 1C D). Fifteen 1,000 S3). Importantly, striatum-specific enriched whole brains 8-week-old C57BL/6 male (P 1.4e-82) fetal 2.1e-88), supporting reflect 1E). "TF-target modules" TFs. modules, 135 functional category Gene Ontology (Ashburner FDR 5%, adjusting 4,624 GO terms). 337 0.01) specifically major neuronal non-neuronal cell type (Doyle Dougherty Zhang known cell-type-specific activities both (e.g., Npas1-3) glia-specific Olig1, Olig2) S4). suggest types. next dataset reconstruction, exon 1 allele knocked endogenous locus 1999; Menalled 2016), remaining C57BL/6J Six distinct alleles length used. humans, shortest alleles—HttQ20—is non-pathogenic, alleles—HttQ80, HttQ92, HttQ111, HttQ140, HttQ175—are progressively earlier onset phenotypes. 2016) four female genotype three time points: 2-month-old, 6-month-old, 10-month-old mice. undergo subtle allele-dependent behavior, all ages profiled precede death (Carty 2015; Rothe Alexandrov differences HttQ20/+ pathogenic point, total 15 comparisons. extent increased fashion, extensive DEGs condition 2). 8,985 showed (DEGs; conditions, 5,132 stringent false discovery rate 1%. robust replicable other overt pathology. 2. Robust 2-, 6-, miceCounts (allele Q20; edgeR log ratio test; nominal P-value 0.01). 209 (three × models; Fisher's exact test, 1e-6; Dataset EV2). Repeating permuted indicated enrichments level significance never occurred than (i.e., zero instances 718,000 tests across permutations networks). therefore conditions. Notably, 44 S5). refer Replication replicate testing enrichment First, conducted meta-analysis microarray (Kuhn 2007; Becanovic Fossale Giles 2012). Targets 46 (meta-analysis 0.01; 3A B) control primary significantly greater chance (Fisher's test: 5.7e-32). preserved 3. Venn diagram showing dataset, −log10(P-values) strength Next, asked abundant level. studied quantitative proteomics 6-month-old dataset. 22 B). CAG-expanded 5.7e-20). Third, might complicated fact samples almost universally late-stage whereas focus much points. addition, heavily degraded dead astrogliosis 1985). reasons, closely pathogenesis masked multitude secondary overcome issues ability detect models, performed two either restrictive (the regulators), broader (all dataset). map 41 types (Neph 36 30 2006). As 5 S6). 616 one-to-one orthology ≥ 10 tested caudate nucleus (part dorsal striatum) Durrenberger Predicted statistically (odds 1.79; 0.05; when (28 shared modules; odds 3.6, 5.0e-5; S6D) upregulated (26 1.8, 0.02; S6E). programs stages molecular progression (assayed models) late support few Fourth, analyzed cortex, hippocampus, cerebellum, liver mice, 168 tissue, equivalent Htt-allele-dependent EV3). ove