Functional Importance andSelective Constraints in Human

non-coding DNA

Shamil Sunyaev

Is it all evolutionary junkyard?… and probably is an evolutionary junkyard

Most of the Genome is Non-coding

acgtcttcccttaggatc

gcatcttcccttaggcgc

However, many genomic regions

are highly conserved!

Definition:

Conservation \Con`ser*va"tion\, n. [L. conservatio:cf. F. conservation.] The preservation of a genetic

sequence over time due to natural selection.

Natural selection

Phenotype

Effect on molecular

function

Conservation as a

functional genomics assay

Practical aspect of conservation

• Prediction of new non-coding functionalelements

• Search for allelic variation underlying humanphenotypes (even if we do not know function)

Theoretical aspect of conservation

• Estimation of genomic rate of deleteriousmutations (U > 1 already implies synergistic epitasis;

estimates including non-coding sequence are much larger).

• Genetics of common phenotypes (Why do they exist

in spite of purifying selection?).

Individual human genome is a targetfor deleterious mutations

Common disease / Common variant

Trade off (antagonistic pleiotropy)

Balancing selection

Recent positive selection

Reverse in direction of selection

Examples

APOE Alzheimer’s disease

AGT Hypertension

CYP3A Hypertension

Multiple mostly rare variants

Many deleterious alleles in mutation-

selection balance

Examples

Plasma level of HDL-C

Plasma level of LDL-C

Colorectal adenomas

• How many nucleotides in the human genome areselectively constrained?

• How are selectively constrained nucleotidesdistributed along the genome?

• How strong is selective pressure in non-codingregions?

• Do comparative and functional genomics dataagree?

Questions:

Selection coefficient

Wild type New mutation

N1= 4 N2= 3

N1

N2= 1 – s

Selection coefficient

Selection coefficient - is a measure of reproductive disadvantage(or advantage) associated with a mutation.

0 . 0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

0 . 6

0 . 7

0 . 8

0 . 9

1 . 0

Divergence (K)

Every new mutation eventually will be either fixed or lost

s – selection coefficient

Ne - effective population size

For humans estimated to be ~ 10 000

Selection coefficient, s

K/K0

CompleteconservationNeutral

behavior

10-6 10-5 10-4 10-3

For mice estimated to be ~ 85 000

• Mutation rate

• Phylogenetic distance (mostly unknown)

• Selection (negative or positive)

Divergence (conservation) is not informative

about strength of selection.

Divergence depends on…

0 . 0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

0 . 6

0 . 7

0 . 8

0 . 9

1 . 0

Nucleotide diversity ( )

Database SNP density is proportional to

Selection coefficient, s10-6 10-5 10-4 10-3

• Mutation rate

• Coalescent variance

• Background selection

• Hitchhikings

• Selection (negative or positive)

Nucleotide diversity depends on…

0 . 0 0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 8 0 . 9 1 . 0

0 . 0 0

0 . 0 1

0 . 0 2

0 . 0 3

0 . 0 4

0 . 0 5

0 . 0 6

0 . 0 7

0 . 0 8

0 . 0 9

0 . 1 0

0 . 1 1

0 . 1 2

Allele frequency spectrum

The higher selection coefficient s (stronger selection) –the higher proportion of low frequency alleles

F0.05

– the proportion of new alleles with frequency below 5%

New allele frequency

Proportion of new alleles with particular frequency

s = 0.0003

s = 0.0001

s = 0

• Demographic history

• Biased gene conversion

• Background selection (if associated with inefficientnegative selection)

• Hitchhikings

• Selection (negative or positive)

Allele frequency spectrum dependson…

Relationships between K, , F0.05

and s

• Conservation (very low fixation rate of new mutations) doesnot necessary mean high selection coefficients

• Conservation is not enough to estimate the level of selectivepressure

(s) / (s=0)

Kh(s) / K

h(s=0)

F0.05

(s)

Km

(s) / Km

(s=0)

Genome analysis

Km K

h

conserved protein coding

conserved intronic

conserved intergenic region>90% in 50 nt window

F0.05

mouserat

chimpanzee human

Divergence in the mouse lineage (Km) inconserved genomic regions

0

0.2

0.4

0.6

0.8

1

Neutral Conserved

protein-

coding

Conserved

intronic

Conserved

intergenic

Divergence in the human lineage (Kh) inconserved genomic regions

0

0.2

0.4

0.6

0.8

1

Neutral Conserved

protein-

coding

Conserved

intronic

Conserved

intergenic

Nucleotide diversity ( ) in

conserved genomic regions

0

0.2

0.4

0.6

0.8

1

Neutral Conserved

protein-

coding

Conserved

intronic

Conserved

intergenic

Selective pressure in conserved genomicregions: theory

Fraction of rare new alleles (F0.05

) inconserved genomic regions

0%

5%

10%

15%

20%

25%

4-fold degenerate sites

in protein coding

(neutral standard)

Nondegenerate sites in

protein coding

Conserved intronic

Selective constraints in conservedgenomic regions

• A genome-wide relaxation of selective constraint inthe primate lineage

• This relaxation most likely resulted from a smaller

human effective population size

• Relaxation is much more profound in conserved

non-coding regions than in protein-coding regions

• Mutations at a large proportion of sites in conserved

non-coding regions are associated with very small

fitness effect

Distribution of selective coefficients

Protein coding Introns Intergenic

Evolution of non-coding regions

What can explain this staggering enrichmentin sites at the borderline of neutrality inconserved non-coding regions?

? ? ??

?

Evolution of non-coding regions

What can explain this staggering enrichmentin sites at the borderline of neutrality inconserved non-coding regions?

? ? ??

?

...synergistic epistasis!

Model of non-coding regions evolution

• Non-coding region “is trying” to maintain anoverall similarity to some optimal sequence

• All nucleotides are of equal importance

• Mutations on average tend to “move away”

region from the optimal sequence

• With each new deleterious mutation the

remaining nucleotides become more

important

SPECULATIONS!

Non-coding regions evolution

New deleterious mutations are not fixedin population any more

X

Smallselection

coefficient (s)

High selectioncoefficient (s)

NOFIXATION

NEUTRALFIXATIONRATE

Sequence evolution in the framework of birth-death processes

The recursive formula for an equilibrium distribution

Transition probabilities from state to state

fitness dependence on number of deleterious mutations

Sequence evolution in the framework of birth-death processes

Model of non-coding regions evolution

This model predicts mutations of mostly smalleffect even in very conserved non-coding regions

Conserved non-coding region

Conserved coding

All nucleotides have small selection coefficient

Some nucleotides have very high selection coefficient

• How many nucleotides in the human genome areselectively constrained?

• How are selectively constrained nucleotides distributedalong the genome?

• How strong is selective pressure in non-coding regions?

• Do comparative and functional genomics data agree?

Questions:

Chimp

Dog

Mouse

Rat

4GCBs

Humans

Model

• All non-4GCBs are neutral (this is the mostconservative assumption)

• 4GCBs are a mixture of neutral and functionalsites

• All functional 4GCBs are associated with thesame selection coefficient (this is the mostconservative assumption)

( )( )

( )( )

( )( )

+

=1

0

1

0

221

11

12

11

%1

dxxxx

dxxxmm

xmxx

F

mm

mm

neutral

( )

=

=1

1

12

3%1

m

i

neutral

i

F

Fraction of rare neutral alleles

( )( ) ( )

neutralfunctional

neutralfunctional

mixturenn

nnF

+

+=

%1%1%1

( )( )( )

( )( )( )

( )( )+=

1

0

221

2

12

12

11

11

1%1 dxxx

mmxmx

exx

en

mm

sN

xsN

functionale

e

n functional =e 2Nes 1 x( ) 1( )

x 1 x( ) e 2Nes 1( )1 xm 1 x( )

m( ) dx0

1

Mixture of neutral and functionalsites

Direct simulation

1001001100111101010010010111010100001111001100011100010111001

How many functional sites are

needed to produce observed allele

frequency shift?

Polymorphism to divergence ratio

= 4Neμ2Nes+ e

2Ne s 1

Nes 1 e 2Ne s( )+ 4Neμ

D = 2Neμt1 e 2Nes

1 e 4Nes+ tμ

R =

2Nes+ e 2Nes 1

Nes 1 e 2Nes( )+

2Ne

1 e 2Nes

1 e4Nes+

Selective constraints in non-codingregions of the genome

• Selectively constrained bases are diffusely distributed alongthe genome rather than condensed to highly conservedregions

• At least ~20% of 4GCBs are electively constrained (2% of

the genome sequence)

• Probably additional constrained positions in non-alignable

regions

• How many nucleotides in the human genome areselectively constrained?

• How are selectively constrained nucleotides distributedalong the genome?

• How strong is selective pressure in non-coding regions?

• Do comparative and functional genomics data agree?

Questions:

Regions selected for the ENCODE

project have 22 mammalian species

sequenced

… and a lot of functional genomics data

AGC ACC AGC

AGC ACC

AGC

Human Chimp Baboon

Mutation rates are modeled as

asymmetric and context

specific.

The model incorporates

insertions and deletions

Instantaneous rate matrix of transitions Q

P(t) = eQt

• Ignores mutation rate heterogeneity along the genome

• Assumes uniformity between species

• Computes Bayesian estimate of evolutionary rate at the

site

•Computes p-value via simulations

SCONE (Sequence CONservationEvaluation)

SCONE vs. ENCODE SNPs

Conservation of functional features

Clustering of conserved positions

Conservation of functional feature

• Conservation of most of ENCODE functional elements is dueto a small number of positions.

• Most of conserved positions are outside of long conserved

elements

• Conserved positions tend to cluster along the sequence

Acknowledgments

The lab: Saurabh Asthana,Gregory Kryukov, Steffen Schmidt

University of Washington: John Stamatoyannopoulos, William Noble