Amino acid conservation is strongly associated with functional and structural significance, and thus can be used to identify functionally or structurally important features, such as active sites and protein-protein interaction domains. However,

1
Inferring Property Selection Pressure from
Positional Residue ConservationRose Hoberman,
Yong Lu, Judith Klein-Seetharamanand Roni
Rosenfeld Carnegie Mellon University,
University of Pittsburgh Medical School,
Pittsburgh, PA
Summary of Results for GPCRA
Evaluating Gaussian Goodness-of-Fit Assumption
We believe that conserved properties will most
often have a unimodal distributions, therefore we
use a Gaussian to model property
conservation. Method
Introduction

• Amino acid conservation is strongly associated
  with functional and structural significance, and
  thus can be used to identify functionally or
  structurally important features, such as active
  sites and protein-protein interaction domains.
  However, identifying positional conservation is
  only a first step. Different positions have
  different functional and structural roles, and
  thus the amino acids in each position will be
  subject to different constraints. A more
  difficult task is to understand the specific
  selective pressures that have influenced the
  distribution of residues at each position.

• Fit a maximum-likelihood Gaussian to amino acid
  frequencies in property space
• From Gaussian calculate expected amino acid
  frequencies
• Calculate goodness-of-fit to quantify
  difference between expected and observed amino
  acid frequencies
• Identifies unimodal distributions
• Penalizes missing amino acids (holes)

Objectives

• Develop a method for systematically identifying
  conservation of specific physical and chemical
  properties at each site in a multiple sequence
  alignment.
• Design a test to determine whether the apparent
  conservation of a particular property in a
  position is statistically significant.

• Statistical Significance
• The probability of a property appearing conserved
  purely by chance is high when the entropy is low
  (when sequence conservation is high). Therefore
  we use a Monte-Carlo method to estimate p-values
  for each significant property. For each
  position, property pair
• Generate a large set of random (shuffled)
  property scales
• Calculate goodness-of-fit of each random scale
  to a Gaussian
• Distribution of these random values allows
  us to estimate the probability of getting a
  value more extreme than the observed
  value merely by chance.
• We use a Bonferroni correction to correct for
  testing of multiple properties. For the chance of
  a type I error to be no more than 5 per
  position, then a Bonferroni correction for
  multiple testing (of 240 properties) yields a
  signi?cance threshold of a 0.05/240 0.0002.

Summary of results for GPCRA, where significant
property conservation at a position is indicated
by a dot, and the color of the dot indicates the
type of property predicted to be conserved.
Entropy and location of transmembrane helices are
also indicated. Reference positions are taken
from the Rhodopsin PDB structure.
Validation
Method
We compared our predictions for the GPCRA family
to current knowledge about the structure,
function, and dynamics of GPCR. We analyzed only
those positions with significant property
conservation and low sequence conservation
(entropy of 3.5 - 4). Our method identifies many
instances of known property conservation
Previous Work
Existing methods use amino acid similarity
matrices, set-theoretic methods, or ML
optimization of only a small handful of
properties 1,2 and generally dont provide a
test of statistical significance.

• Charge conserved at 134
• part of D/E R Y motif of importance to binding
  and activation of G-protein
• Size conserved at 54, 80, 87, 123, 132, 153, 299
• helix faces one or two other helices
• Dynamic properties conserved
• especially in third cytoplasmic loop
• in Rhodopsin this is the most flexible
  interhelical loop
• in TM at 215 and 269
• close proximity to retinal ligand the
  ligand-binding pocket is the most rigid part of
  crystal structure

Mapping Amino Acid Distributions into Property
Space
Results
Downloaded about 250 physical, chemical, and
structural properties of amino acids from the
online database PDBase 3. Each property scale
assigns a numerical value to each of the 20 amino
acids.
1 Hydrophobicity
2 Volume
3 Net charge
4 Transfer free energy
...
240 Average flexibility

• We applied our method to the
• G-Protein Receptor Family A.
• GPCRA is a family of transmembrane proteins
  containing 7 transmembrane helices
• Respond to a variety of ligands bind to
  G-proteins
• Diverse in sequence, but believed to share
  similar structure (only known structure is for
  Rhodopsin)
• Used PFAM alignment of GPCRA proteins, but
  analyzed only the 253 positions aligned to
  nongapped positions in Rhodopsin.
• Used significance threshold of a 0.025/240
  0.0001

Conclusions
A C D E F G H I K L M N P Q R S T V W Y 0.66 2.87 0.10 0.87 3.15 1.64 2.17 1.67 0.09 2.77 0.67 0.87 1.52 0.00 0.85 0.07 0.07 1.87 3.77 2.67

• Proposed a method for systematically identifying
  conservation of specific physical-chemical
  properties at each site in a multiple sequence
  alignment.
• Our test identifies which properties are
  significantly conserved in each position in a
  protein family.
• We show that our automated method correctly
  identifies many instances of known property
  conservation in Rhodopsin.
• We make additional predictions regarding
  previously unknown sites of property
  conservation, providing guidance for future
  site-directed mutagenesis studies by experimental
  biologists.

FAMLRIVM LAMLRIVC IAMLRIVM P-EL-IVP GAELRIVA PGEI
RIVS L-EVYIVA L-EVRIVM I-MLKIVP WAELRIVP HAELYIVS
YAILYIVP WAML-IVA
For each column in an alignment we calculate the
frequency of each amino acid at that site. By
mapping the amino acids to their respective
property values we can then generate a histogram
of showing the distribution of property values at
this site. For example, the first column in the
alignment at left is shown below plotted against
two different property scales.
Sequence vs. Property Conservation
Future Work

• Simple extension to multivariate Gaussian to
  identify conservation of more than one property.
• Incorporate method into a full phylogenetic model
  to model the effect of codon distances as well as
  reduce bias from unequal availability of
  subfamilies and non-independence of sequences.

We expect conserved properties to have lower
variance (indicated by red arrows above), so at
this site hydrophobicity (left) appears more
conserved than the radius of side chain (right).
This graph compares locations of significant
property conservation (yellow lines) with
sequence conservation (indicated by entropy,
black lines) in the GPCR family A. In GPCR the
TM helices (red bars) are much more highly
conserved than the loops (blue bars), and the
entropy and property conservation both illustrate
this trend. However, property conservation can
also be found in many positions with very low
sequence conservation (high entropy).

• References
• 1 Valdar, W. 2002. Scoring residue
  conservation. Proteins, 4822741.
• 2 Koshi, J, Mindell, D, and Goldstein, R.
  1997. Beyond mutation matrices
  Physical-chemistry based evolutionary models.
  Genome Inform Ser Workshop, Genome Inform,
  78089.
• 3 PDbase
• www.scsb.utmb.edu/comp_biol.html/venkat/prop.html
  

However, low variance doesnt always indicate a
conserved property. In the graph on the left, for
example, the missing cystein makes it is unlikely
that this is the sole property that is important
at this site.
For Additional Information Please Contact
roseh_at_cs.cmu.edu
2

• 3.3 Gaussian Goodness-of-Fit
• Fit a maximum-likelihood Gaussian to amino acid
  frequencies in property space
• From (discretized) Gaussian calculate expected AA
  frequencies
• Calculate goodness-of-fit to learned
  Gaussian
• Identifies unimodal distributions
• Penalizes missing amino acids (holes)
• Use Monte-Carlo method to calculate statistical
  significance (p-value)
• Otherwise will have high false discovery rate
  when entropy is low

• Estimating statistical significance
• What is the probability of a property being
  conserved purely by chance?
• Generate a large set of random (shuffled)
  property scales
• Calculate goodness-of-fit to a Gaussian
• p-value estimated from distribution of the
  statistic under random null-hypothesis

However, the Bonferroni is an overly conservative
adjustment, especially since there are signi?cant
correlations between many of the properties we
are testing.