Genome Comparisons and Gene Regulation

1
Genome Comparisons and Gene Regulation

• Penn State University, Center for Comparative
  Genomics and Bioinformatics Webb Miller,
  Francesca Chiaromonte, Anton Nekrutenko, Ross
  Hardison James Taylor, David King, Hao Wang
• University of California at Santa Cruz David
  Haussler, Jim Kent
• National Human Genome Research Institute Laura
  Elnitski
• Childrens Hospital of Philadelphia Mitch Weiss
• Lawrence Livermore National Laboratory Ivan
  Ovcharenko

CSH Nov. 6, 2005
2
DNA sequences of mammalian genomes

• Human 2.9 billion bp, finished
• High quality, comprehensive sequence, very few
  gaps
• Mouse, rat, dog, oppossum, chicken, frog etc. etc
  etc.
• About 40 of the human genome aligns with mouse
• This is conserved, but not all is under
  selection.
• About 5-6 of the human genome is under purifying
  selection since the rodent-primate divergence
• About 1.5 codes for protein
• The 4.5 of the human genome that is under
  selection but does not code for protein should
  have
• Regulatory sequences
• Non-protein coding genes
• Other important sequences

3
Silent and repressed chromatin
4
Transcription initiation and pausing
Repressors bind to negative control elements
General transcription initiation factors, GTIFs
Assemble on promoter
5
Basal and activated transcription
Activators bind to enhancers
6
Contact for activation
Enhancer
IID
PolII
Promoter
Coactivators
Coactivators and/or activators sometimes recruit
enzymes that modify chromatin structure to
facilitate transcription. Histone
acetylation Nucleosome remodeling
7
Promoter for RNA Polymerase II
DPE
Regulate efficiency at which minimal promoter is
used
Minimal promoter binding of GTIFs and RNA Pol II
Bad news for prediction TATA box is moderately
well-defined, but in large datasets of mammalian
promoters, only about 11 have TATA boxes ! Inr
(YANWYY) and DPE are not well-defined sequences.
Good news for prediction of promoters About 70
are in CpG islands Almost all encompass the 5
end of genes
8
Enhancers Specific DNA sequences that cause an
increase in transcription

• Can act in a variety of positions
• 5 to gene (similar to an upstream activation
  sequence)
• Internal to a gene (e.g. in an intron)
• 3 to a gene
• Can act at a considerable distance from the gene
• Current studies implicate enhancers as far as
  200kb to 500kb away from genes.
• Other genes can be between an enhancer and its
  target gene.
• Contain a set of binding sites for
  transcriptional activators.
• Sequence-specific binding sites
• Short roughly 6-8bp

9
Interferon beta Enhancer-Promoter
10
Many regulatory DNA sequences in SV40 control
region
Sequence-specific
11
Domainopening is associated with movement to
non-hetero-chromatic regions
12
Expected properties of regulatory elements

• Conserved between species
• Examine interspecies alignments
• Enhancers and promoters clusters of binding
  sites for transcription factors
• Use TRANSFAC, TESS, MOTIF (GenomeNet), etc to
  find matches to binding sites for transcription
  factors
• Binding sites conserved between species
• Servers to find conserved matches to factor
  binding sites
• Comparative genomics at Lawrence Livermore
  http//www.dcode.org/
• zPicture and rVista
• Mulan and multiTF
• ECR browser
• Consite http//mordor.cgb.ki.se/cgi-bin/CONSITE/co
  nsite
• The database GALA records conserved (and
  nonconserved) matches to factor binding sites
  (http//www.bx.psu.edu/)
• Can be almost anywhere
• 5 or 3 to gene
• Within introns
• Close or far away

13
Conservation score S in different types of regions
Red Ancestral repeats (mostly neutral) Blue
First class in label Green Second class in label
Waterston et al., Nature
14
Use measures of alignment texture to discriminate
functional classes of DNA

• Mouse Cons track (L-scores) and phastCons are
  measures of alignment quality.
• Match gt Mismatch gt Gap
• Alternatively, can analyze the patterns within
  alignments (texture) to try to distinguish among
  functional classes
• Regulatory regions vs bulk DNA
• Patterns are short strings of matches,
  mismatches, gaps
• Find frequencies for each string using training
  sets
• 93 known regulatory regions
• 200 ancestral repeats (neutral)
• Regulatory potential genome-wide
• Elnitski et al. (2003) Genome Research 13 64-72.

15
What types of regulatory sequences may we hope to
find?

• Sequence signature specific binding sites
• Promoters
• Enhancers
• Repressor binding sites
• But these signatures are short and occur
  frequently in any long sequence
• Sequence signature unknown, maybe none
• Compact, silent chromatin
• Insulators, boundaries
• Release from pausing
• Movement from inactive to active compartments

16
Coverage of human by alignments with other
vertebrates ranges from 1 to 91
Human
5.4
91
Millions of years
92
173
220
310
360
450
17
Neutral DNA cleared out over 200Myr
Chick
Frog
Fish
Platypus
Opossum
Mouse, Rat
Cow
Dog
Chimp
Most human DNA is not alignable to species
separated by more than 200 yr. Divergence dates
from Kumar and Hedges (Nature 1998) and Hedges
(Nature Rev Genet 2002)
18
Distinctive divergence rates for different types
of functional DNA sequences
19
Large divergence in cis-regulatory modules from
opossum to platypus
20
Marsupial genome adds substantially to the
conserved fraction of regulatory regions
21
The distal Major regulatory element of the human
HBA gene complex is conserved in opossum but not
beyond
22
cis-Regulatory modules conserved from human to
fish

• About 20 of CRMs
• Tend to regulate genes whose products control
  transcription and development
• Recent reports
• Sandelin, A. et al. (2004). BMC Genomics 5 99.
• Woolfe, A. et al. (2005). PLoS Biol 3 e7
• Plessy, C., Dickmeis, T., Chalme,l F., Strahle,
  U. (2005) Trends Genet. 21 207-10.

Millions of years
91
173
310
450
23
cis-Regulatory modules conserved from human to
chicken

• About 40 of CRMs
• Noncoding sequences conserved from human to
  chicken tend to clusters in gene-poor regions
• Conservation jungles
• Hillier et al. (2004) Nature
• Stable gene deserts are conserved from human to
  chicken
• Ovcharenko et al., (2005) Genome Res. 15
  137-145.
• Conserved noncoding sequences in stable gene
  deserts tend to be long-range enhancers
• Nobrega, M.A., Ovcharenko, I., Afzal, V., Rubin,
  E.M. (2003) Science 302 413.

Millions of years
91
173
310
450
24
cis-Regulatory modules conserved in eutherian
mammals (and marsupials?)

• About 80-90 of CRMs
• Within aligned noncoding DNA of eutherians, need
  to distinguish constrained DNA (purifying
  selection) from neutral DNA.

Millions of years
91
173
310
450
25
Score multi-species alignments for features
associated with function

• Multiple alignment scores
• Binomial, parsimony (Margulies et al., 2003)
• PhastCons
• Siepel and Haussler, 2003 Siepel et al. 2005
• Phylogenetic Hidden Markov Model
• Posterior probability that a site is among the
  10 most highly conserved sites
• Allows for variation in rates and autocorrelation
  in rates
• Factor binding sites conserved in human, mouse
  and rat
• Tffind (from M. Weirauch, Schwartz et al., 2003)
• Score alignments by frequency of matches to
  patterns distinctive for CRMs
• Regulatory potential (Elnitski et al., 2003
  Kolbe et al., 2004)

26
Binding sites conserved between species

• tffind Identify high-quality matches to a weight
  matrix in one sequence (e.g. human) that also
  aligns with other sequences (e.g. mouse and rat)
• Look for matches to weight matrix in 2nd and 3rd
  sequences, in the part of the alignment that
  aligns to match to weight matrix in first species
• GALA records these matches

Program does not find this, but some studies show
that it can happen.
Matt Weirach
27
Conserved transcription factor binding sites

• Track on UCSC Genome Browser (human)
• GALA (www.bx.psu.edu)
• rVista
• Can export alignments from zPicture and Mulan
• ECR browser
• All at dcode.org
• ConSite

28
Use measures of alignment texture to discriminate
functional classes of DNA

• Compute the probability of matching a pattern
  characteristic of regulatory regions
• Analyze alignments as short strings of matches,
  mismatches, gaps
• Find probabilities for each string using as
  training sets
• 93 known regulatory regions
• 200 ancestral repeats (neutral)
• Construct Markov models that give good separation
  of regulatory regions from neutral DNA
• Regulatory potential of all 100 bp windows in the
  genome

29
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 A
Collapsed 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 regions
• 0.0001 in ancestral repeats

30
RP and phastCons in HBB locus control region
LCR
HBB
HBD
HBG2
HBG1
HBE
- Both RP and phastCons are high in exons - RP
peaks in many cis-regulatory modules - phastCons
peaks in more regions
http//genome.ucsc.edu/
31
More species and better models improve
discriminatory power of RP scores
ROC curves for different RP scores, tested on a
set of known regulatory regions from the HBB gene
complex
32
RP and phastCons can discriminate most known
functional elements from neutral DNA
33
Leveraging genome evolution to discover function

• Overall goals and core concepts
• All-vs-all whole-genome comparisons
• Comparison of no two species is ideal for finding
  all functional sequences
• Alignment scores
• Aid in finding functional elements
• Discriminate between functional classes
• Example of experimental tests of the
  bioinformatic predictions

34
Genes co-expressed in late erythroid maturation

• G1E-ER cells proerythroblast line from mice
  lacking the transcription factor GATA-1.
• Can restore the activity of GATA-1 by expressing
  an estrogen-responsive form of GATA-1
• Allows cells to mature further to erythroblasts
• Use microarray analysis of each to find genes
  that increase or decrease expression upon
  induction.
• Walsh et al., (2004) BLOOD Image from k-means
  cluster, GEO

35
Predicting cis-regulatory modules (preCRMs)
Identify a genomic region with a regulated gene.
Find all intervals whose RP score exceeds an
empirical threshold.
Subtract exons
Find all matches to GATA-1 binding sites that are
conserved (cGATA-1_BS)
Intervals with RP scores above the threshold and
with a cGATA-1_BS within 50bp are preCRMs.
36
Predicted cis-regulatory modules (preCRMs) around
erythroid genes

-
37
Test predicted cis-regulatory modules (preCRMs)

• Enhancement in transient transfections of
  erythroid cells
• Activation and induction of reporter genes after
  site-directed, stable integration in erythroid
  cells
• Chromatin immunoprecipitation (ChIP) for GATA-1

38
Validation of preCRM in Alas2
39
Negative controls do not enhance transient
expression
Negative controls are segments of mouse DNA that
align with rat and human but have low RP scores
and do not have a match to a GATA-1 binding site.
They have almost no effect on the level of
expression of the reporter gene in erythroid
cells.
40
7 of 24 Zfpm1 preCRMs enhance transient expression
41
9 of 24 Zfpm1 preCRMs enhance after stable
integration at RL5
42
All preCRMs in Gata2 are functional in at least
one assay
ChIP data are from publications from E.
Bresnicks lab.
43
About half of the preCRMs are validated as
functional
Assay Number Number tested
positive validated Transient 62 21 34
transfections Site-directed 62 21 34
integrants Either expression assay 62 33 53
GATA-1 ChIPs 17 11 65
44
Positive correlation between enhancer activity
and regulatory potential
45
Developmental regulation of the HBB gene complex
transcription, in erythroid cells
46
High throughput DNase I hypersensitive sites find
known regulatory regions
R
47
Long transcripts run through OR genes into globin
genes
48
Conclusions

• Particular types of functional DNA sequences are
  conserved over distinctive evolutionary
  distances.
• Multispecies alignments can be used to predict
  whether a sequence is functional (signature of
  purifying selection).
• Alignments can be used to predict certain
  functional regions, including some cis-regulatory
  elements.
• The predictions of cis-regulatory elements for
  erythroid genes are validated at a good rate.
• Databases such as the UCSC Table Browser, GALA
  and Galaxy provide access to these data.
• Expect improvements at all steps.

49
Many thanks
PSU Database crew Belinda Giardine, Cathy
Riemer, Yi Zhang, Anton Nekrutenko
Wet Lab Yuepin Zhou, Hao Wang, Ying Zhang, Yong
Cheng, David King
RP scores and other bioinformatic
input Francesca Chiaromonte, James Taylor, Shan
Yang, Diana Kolbe, Laura Elnitski
Alignments, chains, nets, browsers, ideas, Webb
Miller, Jim Kent, David Haussler
Funding from NIDDK, NHGRI, Huck Institutes of
Life Sciences at PSU