In collaboration with Mikkelsen Lab Dissection of regulatory motifs in 2,000 human enhancers using a massively parallel reporter assay Pouya Kheradpour Kellis Lab In collaboration with Mikkelsen Lab

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Identifying conserved motif instances movement mutations missing short branches CTCF_8mer chr1 32054304 32054323 + 63199 2.233603 Overlaps with CTCF experimental regions BLS = 2.23sps (78%) Allows for: Mutations permitted by motif degeneracy Misalignment/movement of motifs within window (up to hundreds of nucleotides) Missing motif in dense species tree CTCF

Conserved instances more likely to be bound by corresponding factor Conservation of motif match significantly increases enrichment in ChIP regions for factor Enrichment in bound regions also bound in orthologous mouse region significantly higher

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Combinations of chromatin marks define Chromatin States Ernst, Kheradpour, et al. Nature 2011

Coordinated activity reveals activators/repressors Oct4 predicted activator of embryonic stem cells Gfi1 predicted repressor K562/GM12878 cells Ernst, Kheradpour, et al. Nature 2011

Coordinated activity reveals activators/repressors

HNF1 and HNF4 are predicted activator of HepG2 enhancers Model: Disruption of the motif site would abolish enhancer state

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Systematically looking at HepG2 and K562: all motifs with at least 1 Systematically looking at HepG2 and K562: all motifs with at least 1.3-fold enrichment/depletion Zfp161_3 Nrf-2_3 HNF4_6 GATA_14 Gfi1_1 HNF1_1 Foxa_2

Motifs predicted to be functional in HepG2 and K562 cells selected

Model for cell line specific motifs

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Massively parallel reporter assay (MPRA) Multiplexed enhancer assay 10,000s of elements using unique tags Normalize mRNA counts by Plasmid counts Tarjei Mikkelsen Melnikov, Murugan, Zhang, et al.

Experimental design Manipulations (*expected to disrupt enhancer): Scramble of bases* Least change in score Complete removal of motif* Greatest increase in score Greatest decrease in score* Random change (x2) Activators Repressors HepG2 HNF1, HNF4, FOXA ZFP161 K562 GATA, NRF2 GFI1 Same cell type 160 18 + scramble + other manip (x7) 15 Opposite cell type No. tested instances (x2 for ignoring/high conservation)

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Ex. activator: conserved HNF4 motif match Found in enhancer state of target cell line (HepG2) Coincident with “Dip” in H3K27ac chromatin signal, suggestive of nucleosome exclusion

Ex. activator: conserved HNF4 motif match Motif match disruptions reduce expression to background Non-disruptive modifications maintain expression Random changes depend on effect to motif match WT expression specific to HepG2

All conserved HNF4 in HepG2 results Most are tested sequences not highly expressed Maybe not sufficient context? Those that are almost always reduce in expression when scrambled Max-increase and min-change in motif match score do not seem to reduce when scrambled

Scrambling has the same effect for all activators Conserved instances consistently have higher expression than instances ignoring conservation

Additional modifications for 15 conserved instances per factor The “disruptive” modifications (scrambling, removal, max 1-bp decrease) all reduce expression The “neutral” modification does not The max 1-bp increase significantly increases expression when looked in aggregate

Opposite cell line enrichments: Surprise functional factor Conserved motif instances (18 per factor) found in the opposite cell type (K562 for HNF1, HNF4, FOXA; HepG2 for GATA, NRF2) Do not generally show reduction on scrambling Exception is NRF2 – shows signature of activity despite lack of expression or enrichment

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Estimating the number of functional enhancers tested 71% of conserved instances for activators drop expr in their target cell line when scrambled (n≈800) We expect non-functional enhancers to increase in expression upon scrambling of motif instances with 50/50 probability If we assume (conservatively) that only ‘fake’ enhancers will go up on scrambling, we estimate at most 2*(100 – 71) = 58% ‘fake’ enhancers, and at least 42% real enhancers

Expression of conserved Context matters: identical motif matches considerably varied expression Expression of conserved NRF2 motif instances Motif match score explains very little of variability in expression (r=0.15) Even for these cell-type specific enhancers and conserved motif instances, context plays an important role

Features of functional enhancers Instances ignoring conservation for all 5 activators in matched cell line H3K27ac dip score (suggestive of nucleosome exclusion) Motif conservation level (BLS) No. different factors with matching motifs in 145-bp seq From large database of known motifs Strength of motif match

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Model of GFI1 activity in HepG2 / K562 enhancers

Ex. GFI1 match in HepG2 enhancer Disruption leads to increase in K562 expression

Disruption of repressor motifs leads to aberrant expression Much more subtle effect than for activators Not surprising, we don’t expect every enhancer in one cell type to be “primed” for the other Significant effect seen for GFI1, but not for ZFP161

Dissecting motifs in 2000+ enhancers Predict activator and repressor motifs in specific cell lines Motif instances using comparative genomics Activator/repressor prediction using chromatin state dynamics Motif disruption experiments in 2000+ human enhancers Selecting 5 activators and 2 repressors in two cell lines (x160) Massively parallel reporter assay: 200,000+ expr measurements Experimental results: Activators: Disrupting, enhancing, and neutral motif changes Context matters: sequence features of functional WT enhancers Repressor disruption: aberrant enhancer activity in other cells

Contributions Hundreds of experimentally validated enhancers with necessary motif instance. Resource for data mining enhancer sequence elements General principles of activator/repressor motifs: Motif matters: Scrambling, removing, or disrupting predicted activator motifs abolishes enhancer activity PWM matters: Silent/positive changes maintain activity Context matters: (1) Conservation, (2) nucleosome exclusion, (3) binding of other TFs, (4) motif strength Repressors matter: help activators maintain specificity by repressing enhancer activity in aberrant cell types General methodology for enhancer motif dissection

Acknowledgements Jason Ernst Alexandre Melnikov Peter Rogov Li Wang Xiaolan Zhang Jessica Alston Brad Bernstein Tarjei S. Mikkelsen Manolis Kellis