[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1449 1449–1454
BIOINFORMATICS REVIEW
Vol. 27 no. 11 2011, pages 1449–1454
doi:10.1093/bioinformatics/btr175
Structural bioinformatics Advance Access publication April 14, 2011
Proteins without 3D structure: deﬁnition, detection and beyond
Ferenc Orosz∗ and Judit Ovádi∗
Institute of Enzymology, Biological Research Center, Hungarian Academy of Sciences, Karolina út 29, Budapest,
H-1113 Hungary
Associate Editor: Jonathan Wren
ABSTRACT
Motivation: Predictions, and experiments to a lesser extent,
following the decoding of the human genome showed that a
signiﬁcant fraction of gene products do not have well-deﬁned 3D
structures. While the presence of structured domains traditionally
suggested function, it was not clear what the absence of structure
implied. These and many other ﬁndings initiated the extensive
theoretical and experimental research into these types of proteins,
commonly known as intrinsically disordered proteins (IDPs). Crucial
to understanding IDPs is the evaluation of structural predictors
based on different principles and trained on various datasets,
which is currently the subject of active research. The view is
emerging that structural disorder can be considered as a separate
structural category and not simply as absence of secondary
and/or tertiary structure. IDPs perform essential functions and their
improper functioning is responsible for human diseases such as
neurodegenerative disorders.
Contact: orosz@enzim.hu; ovadi@enzim.hu
Supplementary information: Supplementary data are available at
Bioinformatics online.
Received on February 3, 2011; revised on March 25, 2011; accepted
on March 30, 2011
1 DEFINITION AND CLASSIFICATION
There is currently a lack of consensus regarding the deﬁnition,
terminology or nomenclature of proteins without well-deﬁned 3D
structure in their native (functional) state. These proteins lack a
stable equilibrium conformation but exist as dynamic ensembles
within which atom positions exhibit extreme temporal ﬂuctuations
without speciﬁc equilibrium values (Uversky and Dunker, 2010).
The intrinsically disordered proteins (IDPs), the most frequently
used term, just as others do not always differentiate whether the
whole or (a) signiﬁcant segment(s) of the sequences of these
proteins are without deﬁned structures. The Database of Protein
Disorder (DisProt) deﬁnes IDP as ‘a protein that contains at least
one experimentally determined disordered region’ (Sickmeier et al.,
2007). There are proteins disordered in full length while others
contain both ordered and disordered parts termed as intrinsically
disordered regions (IDRs; Obradovic et al., 2005). These proteins
are often named also as IDPs; however, the name ‘proteins with
IDR(s)’ appears to be more correct.
A number of terms have been used to indicate the disordered
characteristics of these proteins which are as follows: natively
∗To whom correspondence should be addressed.
denatured (Schweers et al., 1994), natively unfolded (Weinreb et al.,
1996), intrinsically unfolded (Baskakov et al., 1999), intrinsically
unstructured (Wright and Dyson, 1999), intrinsically disordered
(Dunker et al., 2000), exceptionally ﬂexible (Ahmed et al., 2007),
natively unstructured (Schlessinger et al., 2007a), naturally ﬂexible
(Uversky et al., 2009), etc. The most often used synonyms are
intrinsically unstructured/disordered proteins (IUPs/IDPs), although
the pioneers of this ﬁeld (Dunker et al. 2008, Uversky et al., 2009)
consider IUPs as a subset of IDPs without hydrophobic core and
signiﬁcant amounts of stable secondary structure.
The central dogma of the new protein structure–function
paradigm, the protein trinity/quartet, consisting of the folded
(ordered) state, the molten globule and the random coil (Dunker
et al., 2001), plus the pre-molten globule as the fourth unique
thermodynamic state (Uversky, 2002), is that any of these states
may be the native state, which is relevant to the biological function
of a protein. Accordingly, the three types of intrinsic disorder can
be classiﬁed as the native coil, native pre-molten globule (both
considered as intrinsically unstructured) and native molten globule
(Uversky et al., 2009). Native coils, characterized by extended
disorder, arise from chains having repulsion arising from a net
charge, and these proteins and regions resemble the more classical
idealized random coil. Pre-molten globules have no well-deﬁned
tertiary structure, may contain regions with transient and small
amount of secondary structure and 3-fold larger hydrodynamic
volume as expected for the folded state. Molten globules (collapsed
disorder) possess secondary structure and folding pattern similar to
the folded state (Uversky, 2002), with loosened, i.e. molten, tertiary
interactions and exhibit an increase in hydrodynamic volume of no
more than 50% (Uversky, 2002; Wright and Dyson, 1999). This
classiﬁcation, which is slightly artiﬁcial, is useful from conceptual
and practical point of view, e.g. for the development of speciﬁc
disorder predictors. However, we emphasize that according to our
concept, in agreement with recently published ones (Rauscher and
Pomés, 2010; Xue et al., 2009), there is a continuum of these states
with various degrees of compactness due to the different amounts
and distribution of secondary and tertiary structure. Nevertheless, it
is important to remember that this view does not inﬂuence the fact
that each of these states, not only the folded one, can occur as the
native state.
Structural disorder can be considered as a separate structural
category, and not merely as a lack of secondary and/or tertiary
structure (Tompa and Kalmar, 2010). The length distribution of
disorder in the human proteome is scale free (follows a power law),
with many short regions and also a signiﬁcant incidence of very
long disordered regions. This is in sharp contrast with the length
distribution of conventional secondary structural elements, which
© The Author 2011. Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com 1449
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1450 1449–1454
F.Orosz and J.Ovádi
shows a length limit near 50 residues, however, is highly reminiscent
of the distribution of tertiary structural units (domains) in proteins.
This behavior was correlated with the direct functional involvement
of disorder (Section 5), which place structural disorder as a unique
‘structural’ level between secondary and tertiary structures (Tompa
and Kalmar, 2010). In line with this, the sequences of intrinsically
disordered domains differ signiﬁcantly from random sequences;
the deviation is comparable with that found between ordered and
disordered domains (Teraguchi et al., 2010) suggesting a ‘structure–
function’ relationship for disordered regions.
2 OCCURRENCE
Bioinformatics predictions (Section 4) suggest that intrinsic
structural disorder is a widespread phenomenon, especially in
eukaryotes, where conservative estimations suggest that 5–15% of
proteins are disordered in full sequence (IDPs), and about 35–50% of
proteins have at least one IDR (more than 30 residues) (Ward et al.,
2004). It was interpreted that more disorder is needed for signaling
and coordination among the various organelles of eukaryotes being
more complex than prokaryotes (Dunker and Obradovic, 2001).
Indeed, in mammals, 75% of signaling proteins are predicted to
contain long disordered regions (Dunker et al., 2008). In general,
the more complex the organism is the more frequent occurrence
of disorder can be found. [An interesting exception is that some
protist parasites have the highest prevalence of disorder (Feng et al.,
2006)]. The various genome wide in silico studies based on Gene
Ontology annotations and Swiss-Prot functional keywords suggest
that the biological processes involving IDPs are as follows (Tompa,
2009; Tompa et al., 2006; Ward et al., 2004; Xie et al., 2007): (i)
transcription and its regulation; (ii) signal transduction and cell-cycle
regulation; (iii) functioning of nucleic acid containing organelle; (iv)
mRNA processing and splicing; and (v) cytoskeleton organization.
These results from genome-wide predictions of intrinsic disorder
and the results from other bioinformatics studies drew attention
to these proteins. However, prediction of disorder foreruns its
experimental identiﬁcation and only relatively few experimentally
characterized examples are known. There is experimental evidence
for the structural disorder of about 1100 regions within 500 proteins,
which is collected in the DisProt database (Sickmeier et al.,
2007).
3 EXPERIMENTAL IDENTIFICATION
Disordered proteins can be identiﬁed and characterized by using
wide arsenal of the experimental methodologies (Daughdrill et al.,
2005; Eliezer, 2009; Mittag and Forman-Kay, 2007; ReceveurBréchot
et al., 2006; Uversky and Longhi, 2010). Nevertheless,
the absence of a well-deﬁned structure in disordered proteins
complicates their investigation, since the determination of unique
high-resolution structure of IDPs is frequently not attainable.
Instead, the goal usually is to obtain experimental constraints on
the ensemble of states, including the detection of residual secondary
structure, transient long-range contacts and regions of restricted or
enhanced mobility (Eliezer, 2007). There is neither time nor space
to cover these important experimental results. Here, we give only
a brief description of the techniques which are most successfully
used for identiﬁcation of IDPs and/or IDRs. Most of them provide
information for the global structures and do not identify the speciﬁc
disordered region(s) within the molecule. For a more detailed
description of speciﬁc methods, please see Supplementary Material
File 1.
4 PREDICTION
Since the ﬁrst predictors of protein disorder were published (Li
et al., 1999; Romero et al., 1997), almost 60 predictors have been
developed so far. The properties and the advantages/disadvantages
of these predictors are summarized in several papers (Dosztányi
and Tompa, 2008; Dosztányi et al., 2010; Feng et al., 2006, Ferron
et al., 2006; He et al., 2009; Tompa, 2009; Uversky and Dunker,
2010). A very comprehensive recent review has been published by
He et al. (2009), which contains detailed descriptions of the most
often used predictors and references to the publicly available ones.
The majority of the programs are accessible via public servers; links
to many of them can be found in the Disordered Protein Database
(http://www.DisProt.org) (Sickmeier et al., 2007). Here, we shortly
introduce merely some representative and frequently used disorder
predictors (cf. also Supplementary Material File 2).
The predictors are based on different principles and can be
classiﬁed into three main categories (Csizmók and Tompa, 2009;
Tompa, 2009): (i) propensity-based predictors; (ii) machine learning
algorithms; and (iii) algorithms based on interresidue contacts.These
categories are not absolute since some of the methods use more than
one of these features. Moreover, combined meta-servers also exist.
4.1 Propensity-based predictors
IDPs are signiﬁcantly depleted in so-called order-promoting
residues, including bulky hydrophobic (Ile, Leu and Val) and
aromatic amino acid residues (Trp, Tyr and Phe), which would
normally form the hydrophobic core of a globular protein, as well
as Cys and Asn. On the other hand, there are so-called disorderpromoting
amino acids, namely,Ala,Arg, Gly, Gln, Ser, Pro, Glu and
Lys, which are substantially overrepresented in IDPs (Dunker et al.,
2001; Romero et al., 2001). This speciﬁc amino acid composition
is usually indicative for disorder. This propensity of IDPs was
used to develop sophisticated prediction methods as well (Section
4.2).
Several methods are based on simple amino acid propensity
scales. Their advantage is that they are easy to calculate and to
interpret; however, they are limited to a single property. For example,
due to their amino acid composition, low overall mean hydropathy
and high mean net charge represent a unique structural feature of
IDPs/IDRs (Uversky et al., 2000). The mean hydropathy is deﬁned
as the sum of the hydropathies of all residues divided by the number
of residues in the polypeptide. The mean net charge is the net
charge, at pH 7.0, divided by the total number of residues. A plot of
mean net charge versus mean hydropathy (the CH plot or Uversky
plot) separates ordered and disordered proteins into distinct regions
(Uversky, 2002). By calculating the distribution of these features
for a pre-deﬁned sequence window, Prilusky et al. (2005) used this
idea to design a per-residue disorder predictor, FoldIndex. Another
predictor, the GlobPlot algorithm (Linding et al., 2003a), uses the
relative propensity of amino acid residues to be in an ordered or
disordered state applying an amino acid scale based on the difference
in the probability for a given amino acid to be in regular secondary
1450
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1451 1449–1454
Proteins without 3D structure
structure or to be in random coil. The basic algorithm behind
GlobPlot is simple and very fast, representing a sum function.
4.2 Machine learning algorithms
The prediction of protein disorder can be viewed as a classic binary
classiﬁcation problem and can be addressed by standard machine
learning techniques as artiﬁcial neural networks (NNs) and support
vector machines (SVMs). The majority of the methods developed
belong to these categories. They are trained on datasets of disorder
and order and evaluate intrinsic disorder on a per-residue basis. Their
underlying assumption is that sequence features calculated from a
local sequence window can be directly mapped into the property of
order or disorder.
The Dunker’s lab was the pioneer of predictors (Romero
et al., 1997); then their PONDR® family of algorithms has been
continuously developed and improved (Li et al., 1999; Obradovic
et al., 2005; Peng et al., 2005, 2006; Romero et al., 1997, 2001).
They typically use NNs, although in a few cases SVMs are also
included. PONDR® algorithms based on the fact that the amino
acid composition in a window of N amino acids for ordered proteins
are distinguishable from the composition for disordered proteins.
Beside merely amino acid composition itself, they use as inputs some
attributes derived from composition as well. These various types
of attributes are weighted and combined in a non-linear manner.
The training datasets are different in the various family members
(missing residues of X-ray structures; variously characterized long
disordered regions; DisProt). Accordingly, there are predictors for
short and long (>30 amino acid) regions, and for N- and Cterminal
and internal ones. For their detailed descriptions, see the
authors’ recent review (He et al., 2009) and the group’s home page
(http://www.pondr.com).
DisEMBL designed by Linding et al. (2003b) consists of three
separate NN predictors, to predict three kinds of disordered
structures in proteins, which represent residues within ‘loops/coils’,
‘hot loops (loops with high B-factors, i.e. with high degree of
mobility)’ or those that are missing from the PDB X-ray structures.
Thus, it performs better on short disordered regions.
Another NN algorithm, RONN, developed by Yang et al. (2005)
is based on ‘functional alignments’. The main idea is that if two
proteins have similar biological functions, in this case the similar
tendencies to be ordered/disordered, then their sequences are also
similar. In the training process, the similarity of sequences is
evaluated by sequence alignment techniques using a mutation matrix
to score the similarity. These scores of sequence alignments are then
used for training.
The most often used SVM method is DISOPRED2 (Ward et al.,
2004) where the input data are generated by sequence alignment
using PSI-BLAST, and which is trained on a database of amino acids
missing from PDB structures. Thus, the prediction is better on short
disordered regions in the context of globally ordered proteins. The
fact that the database contain much more ordered than disordered
residues (176 550 versus 4590) is balanced by formulating the SVM
to place greater cost of misclassiﬁcation for points from the minority
(disordered) class than from the majority (ordered) class. This is the
reason for the low false positivity of DISOPRED2. Compared with
other disorder predictors, the main difference is that DISOPRED2 is
directly trained on the whole sequence rather than measures of amino
acid composition, sequence complexity or biophysical properties.
Prediction accuracy of this method depends on the number of
homologs used for sequence alignment.
Additional methods which apply a second level of prediction using
the output of the ﬁrst level prediction as an input are the POODLE
algorithms. They employ SVMs with radial basis kernels for
training; the input is constructed from physico-chemical properties
using PSI-BLAST proﬁles. POODLE-S (Shimizu et al., 2007) and
POODLE-L (Hirose et al., 2007), which aim to predict disordered
segments shorter and longer than 40 residues, respectively, calculate
the input vector by using physico-chemical features and a reduced
amino acid set of position-speciﬁc scoring matrices or from
hydropathy, average contact density propensity, mean net charge,
sequence complexity, amino acid compositions relative to the
composition of disordered and ordered training sets and secondary
structure preferences.
4.3 Prediction methods based on interresidue contacts
Limitations due to the biased and insufﬁcient databases can
be overcome by the methods based on structural and energetic
considerations, which do not rely on experimental data on protein
disorder. The prominent representatives of these methods are
FoldUnfold (Galzitskaya et al., 2006; Garbuzynskiy et al., 2004),
IUPred (Dosztányi et al., 2005a, b), and Ucon (Schlessinger et al.,
2007a). The main idea of these methods is that the disorder
of proteins is originated from the lack or low level of the
interresidue contacts which cannot compensate the large decrease
in conformational entropy during folding (Tompa, 2009). The
importance of interresidue contacts, especially that of the heavily
interacting residue clusters (stabilization centers) is essential in the
maintenance of the folded protein structure (Dosztányi et al., 1997).
Intuitively, it can be thought that the lack of them favors protein
disorder, as it was found indeed in several cases (Orosz et al., 2004).
FoldUnfold based on the statistical analysis of residue contact
numbers. The summation of the interresidue contact numbers of
the amino acids of a protein is indicative for its folded/unfolded
character. Two residues are considered in contact if any pair of
their heavy atoms is within 8.0 Å to each other. To express the
average contact number of residues within a given distance in a
protein structure, the mean packing density of residues is calculated.
It was demonstrated that regions with low-expected packing density
correspond to the disordered segments.
Ucon combines a former predictor, PROFcon, for long-range
protein-speciﬁc internal contacts (Punta and Rost, 2005) with a
generic pairwise potential to predict unstructured regions longer
than 30 amino acids. It combines information from alignments, from
predictions of secondary structure and solvent accessibility, from the
region between two residues and from the average properties of the
entire protein.
The core of IUPred is a method that renders the direct estimation
of the interaction energies using exclusively the protein sequence
possible. In this approach, the estimated energy for each residue
depends on the amino acid type and on the amino acid composition of
the sequential neighborhood. Generally, residues with less favorable
predicted energies are more likely to be disordered. The parameters
of this method are derived exclusively from a globular protein
dataset without the use of speciﬁc datasets of disordered proteins.
As globular protein datasets are considerably larger than that
of disordered proteins, this stabilizes the method substantially if
1451
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1452 1449–1454
F.Orosz and J.Ovádi
compared with methods where a large number of parameters are
trained on a limited and sometimes biased disordered protein dataset.
IUPred performs comparatively well for predicting long disordered
segments, and has a good sensitivity, i.e. does not miss a signiﬁcant
number of disordered residues.
However, a problem arises by the presence of conserved cysteines
and/or of metal-binding motifs which can cause uncertain local
predictions of disorder within these regions using the methods based
on interresidue contacts. These predictors may display features
typifying disorder while the protein region gains structure upon
disulﬁde formation or binding to metal ions (Ferron et al., 2006).
This problem can in part be handled using methods predicting metalbinding
sites and disulﬁde bridges of proteins from their sequence
(Lippi et al., 2008).
4.4 Metapredictors
Generally, it is a good idea not to rely on one single algorithm when
predicting disorder. Instead, as these algorithms all capture different
aspects of the structural properties of proteins, they can complement
each other to give a more complete picture. Recently, a new direction
in the development of disorder predictors based on the creation
of metapredictors has attracted attention. These metapredictors
(metaPrDOS, MeDor, MD and PONDR-FIT) combine the outputs of
several individual predictors (Ishida and Kinoshita, 2008; Lieutaud
et al., 2008; Schlessinger et al., 2009; Xue et al., 2010a). They can
be applied either at the residue level or at the whole sequence level.
The individual predictors constituting metapredictors are based on
different philosophies, the strength and weakness of which can be
balanced by their combination. MetaPrDOS (Ishida and Kinoshita,
2008) uses SVM to integrate residue-level predictions from several
algorithms and was trained on a group of PDB-extracted proteins
that all have regions of missing electron density in their crystal
structures, and the sequence identities among these proteins are
<20%. Meta-Disorder predictor (MD) (Schlessinger et al., 2009)
uses NN and the training datasets were proteins from PDB and
DisProt. The metapredictors improved the prediction accuracies
which were several percentage points higher (max. 10 %) on various
datasets in comparison with the values estimated for the individual
predictors.
4.5 Future directions
The performance of disorder predictors has been compared in the
CASP (Critical Assessment of Structure Prediction) experiments
(Bordoli et al., 2007; Jin and Dunbrack, 2005; Melamud and
Moult, 2003; Noivirt-Brik et al., 2009). However, these comparisons
are considered to be rather biased since ‘the performance of the
methods depends on both the type of disorder and evaluation criteria’
(Tompa, 2009) as discussed in details in several other papers as
well (Dosztányi et al., 2010; Schlessinger et al., 2007b). Moreover,
the predictors focus on different type (‘ﬂavors’) of disorder, thus
predictors trained on disorder of one type of protein often achieve
poor accuracy on disorder of proteins of a different type, as
recognized already at the advent of IDP research (Vucetic et al.,
2003).
Currently, the per-residue prediction accuracies of these methods
have risen to about 80% (Dunker et al., 2008). The limitation to
further improvement comes from inaccuracy in the ordered and
disordered protein data (Dosztányi et al., 2010; He et al., 2009). The
performance of disorder prediction methods critically depends on the
dataset used for testing and the type of disorder (e.g. extended or
collapsed) studied (Dosztányi et al., 2010; Schlessinger et al., 2007b;
Vucetic et al., 2003). Datasets of experimentally veriﬁed ordered and
disordered regions contain many mis-classiﬁed segments; moreover,
the latter ones are not sufﬁciently large for prediction of very
high level of accuracy. Various datasets of disordered protein
sequences exhibit variations in their sequential bias. Differences can
be observed depending also on the experimental method used for
identiﬁcation of the disordered regions (Dosztányi et al., 2010), on
their length, and on the location in the sequence (N- and C-terminal,
middle regions) (Li et al., 1999). Although these differences are
smaller compared with the differences observed between ordered
and disordered proteins, they should be taken into account during
the development of prediction methods. Thus, it was suggested that
predictors that go beyond the binary classiﬁcation of proteins as
ordered or disordered are necessary (Dosztányi et al., 2010; He et al.,
2009).
5 FUNCTION
Disordered proteins can be separated into two main functional
classes, based on their in vivo activities: entropic chains and IDPs
involved in molecular recognition (Tompa 2002, 2005). There
are IDPs which do not have a folded or ordered state under
any known conditions, while others are capable of folding under
certain circumstances, i.e. upon binding to a partner, termed as
‘non-folders’ and ‘folders’, respectively (Rauscher and Pomés,
2010). Entropic chains are necessarily ‘non-folders’, since their
functions rely on their high-conformational entropy: their functions
are derived by populating many accessible conformations without
well-deﬁned folded structure (Tompa, 2009). On the contrary,
IDPs involved in molecular recognition due to their interacting
potencies are generally ‘folders’ that become (partly) ordered upon
binding to their targets. The binding can be permanent (scavengers,
effectors and assemblers) or transient (display sites and chaperones)
(Tompa and Kovacs, 2010). Scavengers and assemblers usually
bind to multiple partners. Scavengers store and neutralize small
molecules, while assemblers support the assembly of multi-protein
complexes. Effectors regulate the activity of partner proteins.
Display sites expose sites for post-translational modiﬁcations such
as phosphorylation or limited proteolysis, whereas chaperones bind
to the partner molecule to facilitate its correct folding preventing its
aggregation or proteolysis.
5.1 Binding to partners
The folding of IDPs during molecular recognition is analogous to
protein folding of globular proteins, since both processes involve
a thermodynamically stable folded state and an unfolded state of
higher conformational entropy (Verkhivker et al., 2003, 2005).
Since many IDPs and IDRs fold upon binding to their targets
(Wright and Dyson, 2009), a challenging question is whether
folding occurs before binding or binding occurs before folding?
The two extremes are induced folding and conformational selection
(Wright and Dyson, 2009). In the case of the ﬁrst mechanism, the
protein associates with its binding partner in a disordered state and
subsequently folds in association with the target protein. In the
conformational selection mechanism, the target protein ‘selects’ a
1452
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1453 1449–1454
Proteins without 3D structure
conformation closely approximating that of the bound form from
the ensemble of conformations populated by the IDP when free
in solution. This question is analogous to and generalization of
the induced ﬁt—ﬂuctuation ﬁt (conformational selection) duality of
molecular recognition (Vértessy and Orosz, 2011). In real systems,
one or another or both mechanism(s) can be favored (Wright and
Dyson, 2009).
In general, the binding of IDPs differs from that of ordered
proteins since they often bind their partner via short recognition
elements [MoRFs (molecular recognition f eatures)] (Fuxreiter et al.,
2007; Mohan et al., 2006; Oldﬁeld et al., 2005) in a structurally
adaptive process termed disorder-to-order transition (Verkhivker
et al., 2003). Structural disorder may confer signiﬁcant functional
advantages for IDPs, such as rapid binding to the partner molecule,
the combination of high speciﬁcity with weak and reversible
interaction and the ability to carry out more than one function
either via multiple interaction sites or through regions speciﬁc to
distinct partners (Tompa, 2002, 2005; Tompa et al., 2005). They
can fold into different structures on binding to different target
proteins (‘one-to-many binding mode’), with different functional
outcomes (functional promiscuity or moonlighting) (Tompa et al.,
2005). In contrast to this, the ‘many-to-one binding mode’ was also
suggested, when many different IDPs may bind alternatively to one
site on a single ordered partner, by which different IDPs fold into
similar conformations (Uversky et al., 2009). It was suggested that
these binding features of IDPs may explain their ‘organizing role’
in protein–protein interaction networks as so-called ‘hub’ proteins
(Dunker et al., 2008).
Some binding sites can be found as linear motifs (LMs; Puntervoll
et al., 2003), short segments involved in the molecular recognition
of proteins. It was shown that there is a connection between LMs
and molecular recognition elements of IDPs. LMs are embedded
in locally unstructured/highly ﬂexible regions, while their amino
acid composition exhibits a mixture characteristic of folded and
disordered proteins (Fuxreiter et al., 2007).
As we discussed above, there are many algorithms for predicting
IDPs; however, the methods for predicting regions undergoing
disorder-to-order transition upon protein binding is rather limited.
An algorithm proposed by the Dunker’s and Uversky’s labs,
α-MoRF-PredII, combines two bioinformatic tools, sequence
alignment and disorder prediction, to ﬁnd possible binding partners
in protein databases and identify the interaction sites. The method
is based on the identiﬁcation of the above-mentioned MoRFs,
which are short segments expected to have a high propensity
for folding upon binding and that are located within regions of
disorder (Cheng et al., 2007; Mohan et al., 2006; Oldﬁeld et al.,
2005). Very recently, the authors hypothesized that not only MoRFs
with similar sequences can be aligned but also those of with
reversed sequential order (‘retro-MoRFs’). Applying this theory,
they developed a software package named PONDR-RIBS, which
aligns protein segments, predicts disorder and interaction regions
(Xue et al., 2010b). However, experimental veriﬁcation of this new
method is needed.
A recent method, ANCHOR, based on the principles behind the
IUPred algorithm (Section 4.3), has been developed for this aim
(Mészáros et al., 2009). The essential feature of these binding
segments is that they exist in a disordered state in isolation, but
they can favorably interact with a globular protein and adopt a rigid
conformation upon binding. Based on this model, the combination
of the high disordered tendency of the sequential environment, the
unfavorable intrachain interaction energies and high energetic gain
by interacting with a globular protein partner indicates the presence
of a disordered binding region (Dosztányi et al., 2010).
For a discussion of the connection of intrinsic disorder
and alternative splicing and diseases, respectively, please see
Supplementary Material File 1.
Funding: Hungarian National Scientiﬁc Research Fund
OTKA (T-067963 to J.O.); the European Commission (DCIALA/19.09.01/10/21526/245-297/ALFA
111(2010)29 to J.O.); the
European Concerted Research Action (COST Action TD0905 to
J.O.).
Conﬂict of Interest: none declared.
REFERENCES
Ahmed,M.A. et al. (2007) The BG21 isoform of Golli myelin basic protein is
intrinsically disordered with a highly ﬂexible amino-terminal domain. Biochemistry,
46, 9700–9712.
Baskakov,I.V. et al. (1999) Trimethylamine N-oxide-induced cooperative folding of
an intrinsically unfolded transcription-activating fragment of human glucocorticoid
receptor. J. Biol. Chem., 274, 10693–10696.
Bordoli,L. et al. (2007) Assessment of disorder predictions in CASP7. Proteins, 69
(Suppl. 8), 129–136.
Cheng,Y. et al. (2007) Mining α-helix-forming molecular recognition features α-MoRFs
with cross species sequence alignments. Biochemistry, 46, 13468–13477.
Csizmók,V. and Tompa,P. (2009) Structural disorder and its connection with misfolding
diseases. In Ovádi,J. and Orosz,F. (eds) Protein Folding and Misfolding:
Neurodegenerative Diseases. Springer, pp. 1–19.
Daughdrill,G.W. et al. (2005) Natively disordered proteins. In Buchner,J. and
Kiefhaber,T. (eds) Protein Folding Handbook. Wiley-VCH, Verlag GmbH & Co.
KGaA, Weinheim, Germany, pp. 271–353.
Dosztányi,Z. and Tompa,P. (2008) Prediction of protein disorder. Methods Mol. Biol.,
426, 103–115.
Dosztányi,Z. et al. (1997) Stabilization centers in proteins: identiﬁcation,
characterization and predictions. J. Mol. Biol., 272, 597–612.
Dosztányi,Z. et al. (2005a) The pairwise energy content estimated from amino acid
composition discriminates between folded and intrinsically unstructured proteins.
J. Mol. Biol., 347, 827–839.
Dosztányi,Z. et al. (2005b) IUPred: web server for the prediction of intrinsically
unstructured regions of proteins based on estimated energy content. Bioinformatics,
21, 3433–3434.
Dosztányi,Z. et al. (2010) Bioinformatical approaches to characterize intrinsically
disordered/unstructured proteins. Brief. Bioinform., 11, 225–243.
Dunker,A.K. and Obradovic,Z. (2001) The protein trinity – linking function and
disorder. Nat. Biotechnol., 19, 805–806.
Dunker,A.K. et al. (2000) Intrinsic protein disorder in complete genomes. Genome
Inform. Ser. Workshop Genome Inform., 11, 161–171.
Dunker,A.K. et al. (2001) Intrinsically disordered protein. J. Mol. Graph. Model., 19,
26–59.
Dunker,A.K. et al. (2008) The unfoldomics decade: an update on intrinsically disordered
proteins. BMC Genomics, 9 (Suppl. 2):S1.
Eliezer,D. (2007) Characterizing residual structure in disordered protein States using
nuclear magnetic resonance. Methods Mol. Biol., 350, 49–67.
Eliezer,D. (2009) Biophysical characterization of intrinsically disordered proteins. Curr.
Opin. Struct. Biol., 19, 23–30.
Feng,Z.P. et al. (2006) Abundance of intrinsically unstructured proteins in P. falciparum
and other apicomplexan parasite proteomes. Mol. Biochem. Parasitol., 150,
256–267.
Ferron,F. et al. (2006) A practical overview of protein disorder prediction methods.
Proteins, 65, 1–14.
Fuxreiter,M. et al. (2007) Local structural disorder imparts plasticity on linear motifs.
Bioinformatics, 23, 950–956.
Galzitskaya,O.V. et al. (2006) FoldUnfold: web server for the prediction of disordered
regions in protein chain. Bioinformatics, 22, 2948–2949.
Garbuzynskiy,S.O. et al. (2004) To be folded or to be unfolded. Protein Sci., 13,
2871–2877.
1453
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom
[15:05 12/5/2011 Bioinformatics-btr175.tex] Page: 1454 1449–1454
F.Orosz and J.Ovádi
He,B. et al. (2009) Predicting intrinsic disorder in proteins: an overview. Cell Res., 19,
929–949.
Hirose,S. et al. (2007) POODLE-L: a two level SVM prediction system for reliably
predicting long disordered regions. Bioinformatics, 23, 2046–2053.
Ishida,T. and Kinoshita,K. (2008) Prediction of disordered regions in proteins based on
the meta approach. Bioinformatics, 24, 1344–1348.
Jin,Y. and Dunbrack,R.L. Jr (2005) Assessment of disorder predictions in CASP6.
Proteins, 61 (Suppl. 7), 167–175.
Li,X. et al. (1999) Predicting protein disorder for N-, C-, and internal regions. Genome
Inform. Ser. Workshop Genome Inform., 10, 30–40.
Lieutaud,P. et al. (2008) MeDor: a metaserver for predicting protein disorder. BMC
Genomics, 9, S25.
Linding,R. et al. (2003a) GlobPlot: exploring protein sequences for globularity and
disorder. Nucleic Acids Res., 31, 3701–3708.
Linding,R. et al. (2003b) Protein disorder prediction: implications for structural
proteomics. Structure, 11, 1453–1459.
Lippi,M. et al. (2008) MetalDetector: a web server for predicting metal-binding sites
and disulﬁde bridges in proteins from sequence. Bioinformatics, 24, 2094–2095.
Melamud,E. and Moult,J. (2003) Evaluation of disorder predictions in CASP5. Proteins,
53 (Suppl. 6), 561–565.
Mittag,T. and Forman-Kay,J.D. (2007) Atomic-level characterization of disordered
protein ensembles. Curr. Opin. Struct. Biol., 17, 3–14.
Mohan,A. et al. (2006) Analysis of molecular recognition features (MoRFs). J. Mol.
Biol., 362, 1043–1059.
Mészáros,B. et al. (2009) Prediction of protein binding regions in disordered proteins.
PLoS Comput Biol., 5, e1000376.
Noivirt-Brik,O. et al. (2009) Assessment of disorder predictions in CASP8. Proteins,
77 (Suppl. 9), 210–216.
Obradovic,Z. et al. (2005) Exploiting heterogeneous sequence properties improves
prediction of protein disorder. Proteins, 61 (Suppl. 7), 176–182.
Oldﬁeld,C.J. et al. (2005) Coupled folding and binding with alpha-helix-forming
molecular recognition elements. Biochemistry, 44, 12454–12470.
Orosz,F. et al. (2004) TPPP/p25: from unfolded protein to misfolding disease: prediction
and experiments. Biol. Cell, 96, 701–711.
Peng,K. et al. (2005) Optimizing long intrinsic disorder predictors with protein
evolutionary information. J. Bioinform. Comput. Biol., 3, 35–60.
Peng,K. et al. (2006) Length-dependent prediction of protein intrinsic disorder. BMC
Bioinformatics, 7, 208.
Prilusky,J, et al. (2005) FoldIndex: a simple tool to predict whether a given protein
sequence is intrinsically unfolded. Bioinformatics, 21, 3435–3438.
Punta,M. and Rost,B. (2005) PROFcon: novel prediction of long-range contacts.
Bioinformatics, 21, 2960–2968.
Puntervoll,P. et al. (2003) ELM server: a new resource for investigating short functional
sites in modular eukaryotic proteins, Nucleic Acids Res., 31, 3625–3630.
Rauscher,S. and Pomés,R. (2010) Molecular simulations of protein disorder. Biochem.
Cell Biol., 88, 269–290.
Receveur-Bréchot,V. et al. (2006) Assessing protein disorder and induced folding.
Proteins, 62, 24–45.
Romero,P. et al. (1997) Identifying disordered regions in proteins from amino acid
sequence. Proc. IEEE Int. Conf. Neural Netw., 1, 90–95.
Romero,P. et al. (2001) Sequence complexity of disordered protein. Proteins, 42, 38–48.
Schlessinger,A. et al. (2007a) Natively unstructured regions in proteins identiﬁed from
contact predictions. Bioinformatics, 23, 2376–2384.
Schlessinger,A. et al. (2007b) Natively unstructured loops differ from other loops. PLoS
Comput. Biol., 3, e140.
Schlessinger,A. et al. (2009) Improved disorder prediction by combination of orthogonal
approaches. PLoS One, 4, e4433.
Schweers,O. et al. (1994) Structural studies of tau protein and Alzheimer paired helical
ﬁlaments show no evidence for beta-structure. J. Biol. Chem., 269, 24290–24297.
Sickmeier, M. et al. (2007) DisProt: the database of disordered proteins. Nucleic Acids
Res., 35, D786–D793.
Shimizu,K. et al. (2007) POODLE-S: web application for predicting protein disorder
by using physicochemical features and reduced amino acid set of a position-speciﬁc
scoring matrix. Bioinformatics, 23, 2337–2338.
Teraguchi,S. et al. (2010) Intrinsically disordered domains deviate signiﬁcantly from
random sequences in mammalian proteins. BMC Bioinformatics, 11 (Suppl 7), S7.
Tompa,P. (2002) Intrinsically unstructured proteins. Trends Biochem. Sci., 27, 527–533.
Tompa,P. (2005). The interplay between structure and function in intrinsically
unstructured proteins. FEBS Lett., 579, 3346–3354.
Tompa,P. (2009) Structure and Function of Intrinsically Disordered Proteins. CRC
Press/Taylor and Francis Group, Boca Raton, FL.
Tompa,P. and Kalmar,L. (2010) Power law distribution deﬁnes structural disorder as a
structural element directly linked with function. J. Mol. Biol., 403, 346–350.
Tompa,P. and Kovacs,D. (2010) Intrinsically disordered chaperones in plants and
animals. Biochem. Cell Biol., 88, 167–174.
Tompa,P. et al. (2005) Structural disorder throws new light on moonlighting. Trends
Biochem. Sci., 30, 484–489.
Tompa,P. et al. (2006) Prevalent structural disorder in E.coli and S.cerevisiae proteomes.
J. Proteome Res., 5, 1996–2000.
Uversky,V.N. (2002) Natively unfolded proteins: a point where biology waits for
physics. Protein Sci., 11, 739–756.
Uversky,V.N. and Dunker,A.K. (2010) Understanding protein non-folding. Biochim.
Biophys. Acta, 1804, 1231–1264.
Uversky,V.N. and Longhi,S. (eds) (2010) Instrumental analysis of intrinsically
disordered proteins: assessing structure and conformation. In Uversky,V.N. (series
ed.) The Wiley Series in Protein and Peptide Science. John Wiley & Sons, Inc,
Hoboken, NJ, USA.
Uversky,V.N. et al. (2000) Why are “natively unfolded” proteins unstructured under
physiologic conditions? Proteins, 41, 415–427.
Uversky,V.N. et al. (2009) Unfoldomics of human diseases: linking protein intrinsic
disorder with diseases. BMC Genomics, 10 (Suppl. 1), S7.
Verkhivker,G.M. et al. (2003) Simulating disorder-order transitions in molecular
recognition of unstructured proteins: where folding meets binding. Proc. Natl Acad.
Sci. USA, 100, 5148–5153.
Verkhivker,G.M. (2005) Protein conformational transitions coupled to binding in
molecular recognition of unstructured proteins. Proteins, 58, 706–716.
Vértessy,B.G. and Orosz,F. (2011) From “ﬂuctuation ﬁt” to “conformational selection”:
evolution, rediscovery, and integration of a concept. Bioessays, 33, 30–34.
Vucetic,S. et al. (2003) Flavors of protein disoder. Proteins, 52, 573–584.
Ward,J.J. et al. (2004) Prediction and functional analysis of native disorder in proteins
from the three kingdoms of life. J. Mol. Biol., 337, 635–645.
Weinreb,P.H. et al. (1996) NACP, a protein implicated in Alzheimer’s disease and
learning, is natively unfolded. Biochemistry, 35, 13709–13715.
Wright,P.E. and Dyson,H.J. (1999) Intrinsically unstructured proteins: re-assessing the
protein structure-function paradigm. J. Mol. Biol., 293, 321–331.
Wright, P.E. and Dyson, H.J. (2009) Linking folding and binding. Curr. Opin. Struct.
Biol., 19, 31–38.
Yang,Z.R. et al. (2005) RONN: the bio-basis function neural network technique applied
to the detection of natively disordered regions in proteins. Bioinformatics, 21,
3369–3376.
Xie,H. et al. (2007) Functional anthology of intrinsic disorder. 1. Biological processes
and functions of proteins with long disordered regions. J. Proteome Res., 6,
1882–1898.
Xue,B. et al. (2009) CDF it all: consensus prediction of intrinsically disordered
proteins based on various cumulative distribution functions. FEBS Lett., 583,
1469–1474.
Xue,B. et al. (2010a) PONDR-FIT: A meta-predictor of intrinsically disordered amino
acids. Biochim. Biophys. Acta, 1804, 996–1010.
Xue,B. et al. (2010b) Retro-MoRFs: identifying protein binding sites by normal
and reverse alignment and intrinsic disorder prediction. Int. J. Mol. Sci., 11,
3725–3747.
1454
atMasarykUniversityonSeptember18,2011bioinformatics.oxfordjournals.orgDownloadedfrom