Evolutionary and genomic approaches to find gene regulatory sequences

Penn State University, Center for Comparative Genomics and Bioinformatics: Webb Miller, Francesca Chiaromonte, Anton Nekrutenko, Kateryna Makova, Stephan Schuster, Ross Hardison

University of California at Santa Cruz: David Haussler, Jim Kent

Children’s Hospital of Philadelphia: Mitch WeissNimbleGen: Roland Green

University of Nebraska, Lincoln February 14. 2007

Major goals of comparative genomics

• Identify all DNA sequences in a genome that are functional– Selection to preserve function– Adaptive selection

• Determine the biological role of each functional sequence

• Elucidate the evolutionary history of each type of sequence

• Provide bioinformatic tools so that anyone can easily incorporate insights from comparative genomics into their research

Known types of gene regulatory regions

G.A. Maston, S.K. Evans, M.R. Green (2006) Ann. Rev. Genomics & Human Genetics 7:29-59.

Regulatory regions tend to be clusters of transcription factor

binding sites

Sequence-specific

SV40 promoters and enhancer

Properties of known regulatory regions

• Binding sites for transcription factors, many with sequence specificity

• Clusters of binding sites• Conventional promoters encompass major start sites for transcription

• Conserved over evolutionary time???

Structures involved in transcription are probably more

complex

Peter R. Cook, Oxford University, http://users.path.ox.ac.uk/~pcook/images/Images.html

Middle image: Green: active transcription (Br-UTP label) Red: all nucleic acids HeLa cellSides: EM spreads of transcripts

Domain opening is associated with movement to non-heterochromatic regions

Schubeler, Francastel, Cimbora, Reik, Martin, Groudine (2000) Genes & Dev. 14: 940-950

Other possible activities for sequences involved in gene

regulation• Opening or closing a chromosomal domain• Move a gene to or away from a transcription factory

• Control how long a gene is in a transcription factory– Long association

• High level expression• Really long gene

– Short association• Lower level expression• Rapid regulation

• Are these conserved over evolutionary time?

3 modes of evolution

Sequence matches at longer phylogenetic distances could reflect purifying selectionSequence differences at closer phylogenetic distances could reflect adaptive evolution.

Conservation vs. Constraint

• Conserved sequences are those that align between two species thought to be descended from a common ancestor

• Constrained sequences show evidence in their alignments of negative (purifying) selection– E.g. change at a rate significantly slower than “neutral” DNA

Ideal cases for interpretation

Neutral DNASimilarity

Human vs mouse

Position along chromosome

DNA segments with a function common to divergent species.

DNA segments in which change is beneficial to at least one of the two species.

Negative selection(purifying)

P (not neutral)Neutral DNA

Similarity

Positive selection(adaptive)

Neutral DNA

Human vs rhesus

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

Finding all gene regulatory regions is a challenge for comparative

genomics

• Known regulatory regions for the HBB complex• 23 total• 19 conserved (align) between human and mouse• Many others show no significant difference in a measure of constraint (phastCons) from the bulk or neutral DNA

Two extremes of

constraint in TRRs

ENCODE projects

• ENCODE (ENCyclopedia Of DNA Elements): consortium aiming to find function for all human DNA sequences– Phase I focused on 1% of human DNA– 30 Mb, 44 regions

• About 10 regions had known genes of interest (CFTR, HOX)

• Others were chosen to get a sampling of regions varying in gene density and alignability with mouse

• Major areas– Genes and transcripts– Transcriptional regulation– Chromatin structure– Multiple sequence alignment– Variation in human populations

Biochemical assays for protein-binding sites in DNA

Purified protein& Naked DNA

Chromatin Immunoprecipitation:DNA sites occupied by a protein inside cells.

ChIP-on-chip to examine many sites

Putative transcriptional regulatory regions = pTRRs

• Antibodies vs 10 sequence-specific factors: – Sp1, Sp3, E2F1, E2F4, cMyc, STAT1, cJun, CEBPe, PU1, RA Receptor A

– High resolution ChIP-chip platforms: Affymetrix and NimbleGen

– Data from several different labs in ENCODE consortium

• High likelihood hits for ChIP-chip– 5% false discovery rate

• Supported by chromatin modification data– Modified histones in chromatin: H4Ac, H3Ac, H3K4me, H3K4me2, H3K4me3, etc.

– DNase hypersensitive sites (DHSs) or nucleosome depleted sites

• Result: set of 1369 pTRRs

A small fraction of cis-regulatory modules are conserved from human to

chicken

310

450

91

173

Millions ofyears

• About 4% of pTRRs, 4% of DNase HSs, 4-7% of promoters active in multiple cell lines

• Tend to regulate genes whose products control transcription and development

David King

Most pTRRs are conserved in eutherian mammals

310

450

91

173

Millions ofyears

Within aligned noncoding DNA of eutherians, need to distinguish constrained DNA (purifying selection) from neutral DNA.

Percentage of class that align no further than:

Primates: 3%

Eutherians: 71%

Marsupials: 21%

Tetrapods: 4%

Vertebrates: 1%

pTRRs DNase HSs Promoters

11%

70%

14%

4%

1%

1-13%

63%

16-28%

4-7%

2-4%

Measures of conservation and constraint capture only a subset of

pTRRs

Fraction overlappingan MCS

phastCons (background rate corrected)

Composite alignability (background rate corrected)

Stringent constraint Allows a range of constraint

Aligns, but no inference about purifying selection

Different measures perform better on specific functional regions

Sensitivity

1-Specificity

Examples of clade-specific pTRRs

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

Regulatory potential (RP) to distinguish functional classes

Good performance of ESPERR for gene regulatory regions (RP)

-

James TaylorFrancesca Chiaromonte

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

Conservation of predicted binding sites for

transcription factorsBinding site for GATA-1

Genes Co-expressed in Late Erythroid Maturation

G1E-ER cells: proerythroblast line lacking the transcription factor GATA-1. Can rescue by expressing an estrogen-responsive form of GATA-1Rylski et al., Mol Cell Biol. 2003

Predicted cis-Regulatory Modules (preCRMs) Around Erythroid Genes

B:Yong Cheng, Ross, Yuepin Zhou, David KingF:Ying Zhang, Joel Martin, Christine Dorman, Hao Wang

preCRMs with conserved consensus GATA-1 BS tend to be active on transfected

plasmids

preCRMs with conserved consensus GATA-1 BS tend to be active after integration into a chromosome

Examples of validated preCRMs

Correlation of Enhancer Activity with RP Score

Validation status for 99 tested fragments

preCRMs with High RP and Conserved Consensus GATA-1 Tend To Be

Validated

Compare the outputs

C C N C M C C C W

Consensus for EKLF binding site:

All validated preCRMs

All nonvalidated preCRMs

Same parameters

CCNCMCCCWCCNCMCCCW

CACC box helps distinguish validated from nonvalidated preCRMs

Ying Zhang

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

preCRMs with conserved consensus GATA-1 binding sites are usually occupied by

that protein: ChIP assay

Design of ChIP-chip for occupancy by GATA-1

1. Non-overlapping tiling array with 50bp probe and 100bp resolution (NimbleGen)

2. Cover range Mouse chr7:57225996-123812258 (~70Mbp)3. Antibody against the ER portion of

GATA-1-ER protein in rescued G1E-ER4 cells

50 50

100

Yong Cheng, with Mitch Weiss & Lou Dore (CHoP), Roland Green (NimbleGen)

Signals in known occupied sites in Hbb LCR

1) Cluster of high signals2) “hill” shape of the signals

HS1 HS2 HS3

Peak Finding Programs

• TAMALPAISMark Bieda from Peggy Farmham’s lab Focus more on the cluster of the signals4 thresholds based on number of consecutive probes with signals in the 98th or 95th percentiles

• MPEAKBing Ren’s labFocus more one the “hill” shape of the signal4 thresholds, for a series of probes with at least one that is 3, 2.5, 2 or 1 standard deviations above the mean

ChIP-chip hits for GATA-1 occupancy

Mpeak TAMALPAIS

275 hits in both 276 hits in both216 6059

321 total ChIP-chip hits

Technical replicates of ChIP-chip with antibody against GATA1-ER

ChIP-chip hits validate at a high rate

Validation determined by quantitative PCR.19 of the 321 hits were tested.13 (~70%) were validated.

9 regions were “hits” in only one of the two technical replicates.None were validated.

Validation rate is similar at different thresholds

ChIP DNA

Association of WGATAR and conservation with ChIP-chip

Hits

1. 249 out of the 321 (78%) have WGATAR motifs, binding site for GATA-1

2. Of the GATA-1 binding motifs in those 249 hits, 112 (45%) are conserved between mouse and at least one non-rodent species.

Expected and unexpected ChIP-chip hits

Distribution of ChIP-chip hits on 70Mb of mouse chr7

Yong Cheng, Yuepin Zhou and Christine Dorman

Almost half the GATA-1 ChIP-chip hits increase expression of a

transgene, K562 cells

0

1

2

3

4

GHP181GHP10GHP7GHP182GHP309

GHP1GHP186GHP205

GHP4GHP314GHP172GHP167GHP74GHP193GHP27GHP9

GHP170GHP18GHP16GHP243GHP15GHP28GHP17GHP31GHP11GHP198GHP169GHP14GHP173GHP29GHP199GHP12GHP3GHP24GHP164GHP13GHP30GHP19GHP26GHP161GHP191GHP197GHP183GHP184GHP6GHP23GHP206GHP194GHP202

GHP0GHP200

GHP8GHP185GHP118GHP20GHP204GHN534GHN006GHN133GHN037GHN322

YC3

GHN213

Fold change over parent

GATA-1 occupied sites by ChIP-chip No GATA-1

15 6 6

24 validated out of 56 fragments with ChIP-chip hits tested 43%

Conserved and nonconserved ChIP-chip hits can be active

as enhancers

Conserved, active

Conserved, not active Not conserved, active

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

Polymorphism as a transient phase of evolution

Slide from Dr. Hiroshi Akashi

Test of neutrality using polymorphism and divergence data

Test for recent selection in human noncoding DNA

• McDonald-Kreitman test• Use ancestral repeats as neutral model (MKAR test)• Count polymorphisms in human using dbSNP126• Count divergence of human from

– Chimpanzee (great Ape, diverged from human lineage 6 Myr ago)

– Rhesus macaque (Old World Monkey, diverged from human lineage 23 Myr ago)

• Tiled windows, most analysis on 10kb windows• Compute p-value for neutrality by chi-square test• Ratio of polymorphism to divergence ratios gives

indication of direction of inferred selection

Heather Lawson, Anthropology, PSU

pTRR apparently under positive selection

A promoter distal to the beta-like globin genes has a signal for recent

purifying selection

Selection on a primate-specific promoter

The distal promoter is close to the locus control region for beta-globin

genes

Messages about evolutionary approaches to predicting regulatory

regions• Regulatory regions are conserved, but not all to the same phylogenetic distance.

• Incorporation of pattern and composition information along with with conservation can lead to effective discrimination of functional classes (regulatory potential).

• Regulatory potential in combination with conservation of a GATA-1 binding motif is an effective predictor of enhancer activity.

• In vivo occupancy by GATA-1 suggests other activities in addition to enhancers.

• Comparison of polymorphism and divergence from closely related species can reveal regulatory regions that are under recent selection.

Many thanks …

B:Yong Cheng, Ross, Yuepin Zhou, David KingF:Ying Zhang, Joel Martin, Christine Dorman, Hao Wang

PSU Database crew: Belinda Giardine, Cathy Riemer, Yi Zhang, Anton Nekrutenko

Alignments, chains, nets, browsers, ideas, …Webb Miller, Jim Kent, David Haussler

RP scores and other bioinformatic input:Francesca Chiaromonte, James Taylor, Shan Yang, Diana Kolbe, Laura Elnitski

Funding from NIDDK, NHGRI, Huck Institutes of Life Sciences at PSU

Computing Regulatory Potential (RP)

Alignment seq1 G T A C C T A C T A C G C A seq2 G T G T C G - - A G C C C A seq3 A T G T C A - - A A T G T ACollapsed alphabet 1 2 1 3 4 5 7 7 6 8 3 6 3 9

• A 3-way alignment has 124 types of columns. Collapse these to a smaller alphabet with characters s (for example, 1-9).

•Train two order t Markov models for the probability that t alignment columns are followed by a particular column in training sets:

–positive (alignments in known regulatory regions)–negative (alignments in ancestral repeats, a model for neutral DNA)–E.g. Frequency that 3 4 is followed by 5:

0.001 in regulatory regions0.0001 in ancestral repeats•RP of any 3-way alignment is the sum of the log likelihood ratios of

finding the strings of alignment characters in known regulatory regions vs. ancestral repeats.

€

RP = logpREG (sa | sa−1...sa−t )

pAR (sa | sa−1...sa−t )

⎛

⎝ ⎜

⎞

⎠ ⎟

a in segment

∑

Stage 1: Reduced representations

G

T

gap

ESPERR: Evolutionary Sequence and Pattern Extraction using Reduced Representations

Stage 2: Improve encoding

Train models for classification

Note that many different columns are reduced to single “encoding” (a number in the figure). E.g. Four different columns are each called “3”.

6 6 2 may occur frequently in positive training set and rarely in the negative training set, and thus contribute to discrimination.If the positive training set is known regulatory regions, this would contribute to a positive RP.

Categories of Tested DNA Segments

Example that suggests turnover

GATA-1 BSs

All validated preCRMs All nonvalidated preCRMs

Background:

Mouse chr 19 (42.8% C+G) - NCBI Build 30

CLOVER (Zlab)

EKLF PWM(Dr. Perkins)

ELPH (UMaryland)

Hexamer Counting

Motif P(mm_chr19.m)EKLF 0.0008

Motif P(mm_chr19.m)

none none

Output for validated preCRMs

Output for nonvalidated preCRMs

validated non-validated6-mer TTATYT GGCAGR7-mer CCWCAGM RGRCAGR8-mer CASCCWGC CAGGGAWR9-mer CCWGGCWGM CWGRGAWRA

counts validated nonvalidatedNCACCC 60 32CACCCW 56 27expected validated nonvalidatedNCACCC 16.31 5.81CACCCW 11.74 4.36

Additional methods find CACC box as distinctive for validation

Using Galaxy to find predicted CRMs