What is the direct effect of a mutation?

Motivation: Mutations play fundamental roles in evolution by introducing diversity into genomes. Missense mutations in structural genes may become either selectively advantageous or disadvantageous to the organism by affecting protein stability and/or interfering with interactions between partners. Thus, the ability to predict the impact of mutations on protein stability and interactions is of significant value, particularly in understanding the effects of Mendelian and somatic mutations on the progression of disease. Here, we propose a novel approach to the study of missense mutations, called mCSM, which relies on graph-based signatures. These encode distance patterns between atoms and are used to represent the protein residue environment and to train predictive models. To understand the roles of mutations in disease, we have evaluated their impacts not only on protein stability but also on protein–protein and protein–nucleic acid interactions.

Results: We show that mCSM performs as well as or better than other methods that are used widely. The mCSM signatures were successfully used in different tasks demonstrating that the impact of a mutation can be correlated with the atomic-distance patterns surrounding an amino acid residue. We showed that mCSM can predict stability changes of a wide range of mutations occurring in the tumour suppressor protein p53, demonstrating the applicability of the proposed method in a challenging disease scenario.

Availability and implementation: A web server is available at http://structure.bioc.cam.ac.uk/mcsm.

Contact:[email protected]; [email protected]

Supplementary information:Supplementary data are available at Bioinformatics online.

1 INTRODUCTION

1.1 Background

Mutations play fundamental roles in evolution by introducing diversity into genomes, most often through single nucleotide polymorphisms (SNPs). Non-synonymous single nucleotide substitutions (nsSNPs) are of particular interest, as they can disrupt function by interfering with protein stability and/or interactions with partners. Such mutations can be selectively advantageous in evolution or they may cause a change in stability often leading to malfunction and resulting in disease. Thus, predicting the impacts of mutations in proteins is of major importance to understanding function, not only of molecules and cells but also of the whole organism.

Mutagenesis studies that experimentally determine free energy differences between wild-type and mutant proteins (Fersht, 1987) produce accurate results but are usually costly and time-consuming. However, the advent of databases with experimental thermodynamic parameters for both wild-type and mutant proteins such as ProTherm and ProNIT (protein-nucleic acid) (Kumar et al., 2006) and more recently the SKEMPI (Moal and Fernandez-Recio, 2012), which describes protein–protein complexes, has been helpful to the study of mutations on a larger scale. These provide an experimental basis for novel in silico paradigms, models and algorithms to study more extensively missense mutations and their impacts on protein stability and function.

The several different approaches used to study the impacts of mutations on protein structure and function can be broadly classified into those that seek to understand the effects of mutations from the amino acid sequence of a protein alone, and those that exploit the extensive structural information now available for many proteins. The first group includes well-established and widely used sequence-based methods such as SIFT (Ng and Henikoff, 2003) and PolyPhen (Adzhubei et al., 2010). Here, we focus on the second approach that takes advantage of the protein structural information that has been accumulated on the impact of mutations within the 3D space of a natively folded protein.

Structure-based approaches, which may be categorized as machine learning methods and potential energy functions, typically attempt to predict either the direction of change in protein stability on mutation (as a classification task) or the actual free energy value (⁠

The second set of methods is based on potential energy functions. Environment-specific substitution tables, which describe the propensities of residues to mutate in a certain protein-structural environment during evolutionary time, have been used to derive a statistical potential energy function used by the method SDM (Topham et al., 1997; Worth et al., 2011). In the PoPMuSiC method (Dehouck et al., 2009), the estimated stability change on mutation is expressed as a linear combination of 26 different energy functions, whose parameters were trained using an artificial neural network. Empirical energy functions have also been used in a method that performed Monte Carlo optimization (Bordner and Abagyan, 2004), which has also been used to study the role of conformational sampling as a way to assess the impact of single-point mutations in protein structures (Kellogg et al., 2011).

Although there have been attempts to predict the affinities of particular protein–protein complexes (Moal et al., 2011; Yan et al., 2013), there has been much less attention to the challenge of predicting the impact of mutations on affinity in large sets of protein–protein and protein–DNA complexes. A significant exception has been the report of Guerois et al. (2002) on predicting the effects of mutations in a set of 82 protein–protein complexes. Another important study refers to the identification of binding energy hot spots in protein complexes by predicting the impact of mutations to alanine in >200 mutations (Kortemme and Baker, 2002). More recently, a method derived from PoPMuSiC, called BeAtMuSiC (Dehouck et al., 2013), has been developed to predict the impacts of mutations on protein–protein affinities for a set of 81 proteins.

An alternative approach to study mutations is to represent residue environments as graphs where nodes are the atoms and the edges are the physicochemical interactions established among them. For instance, the method Bongo (Cheng et al., 2008) attempts to predict structural effects of nsSNPs by evaluating graph theoretic metrics and identifying key residues using a vertex cover algorithm. From these graphs, distance patterns can also be extracted and summarized in a structural signature, which may then be used as evidence to train predictive models. Da Silveira et al. (2009) first reported the use of inter-residue distance patterns or signatures to define protein contacts, demonstrating that they are conserved across protein folds. The Cutoff Scanning Matrix (CSM) is a protein structural signature (Pires et al., 2011) successfully used in large-scale protein function prediction and structural classification tasks. Pires et al. (2013) extended the inter-residue signature to an atomic level (aCSM) and successfully applied it in large-scale receptor-based protein ligand prediction.

Here, we use the concept of graph-based structural signatures to study and predict the impact of single-point mutations on protein stability and protein–protein and protein–nucleic acid affinity. The approach, called mutation Cutoff Scanning Matrix (henceforth called mCSM), encodes distance patterns between atoms to represent protein residue environments.

1.2 Method outline

The mCSM signatures are calculated in two steps:

• For a given mutation site, we define the wild-type residue environment by the atoms within a distance r from its geometric centre. A pairwise distance calculation between the atoms of the environment generates an atom distance matrix, which accounts for a wide spectrum of distances, from short to long range. From this matrix, distance patterns are then extracted and summarized as a feature vector;

• To account for the atom changes induced by the mutation, we introduce a ‘pharmacophore count’ vector. Each one of the 20 amino acid residues is represented by a different vector, where each position denotes the frequency of a certain pharmacophore in that residue. The difference vector between the wild-type and mutant pharmacophore vectors is then appended to the signature.

Thus, each mutation is represented as a signature vector that is used to train and test predictive machine learning methods in regression and classification tasks. In Section 2, we describe in detail how the signatures are calculated and the data sets used in this study.

1.3 Summary of results

We show that the mCSM signatures can be used successfully to tackle different tasks related to the prediction of the impacts of mutations in proteins. We have conducted a series of comparative experiments that indicate that mCSM performs as well as or better than several other widely used methods. mCSM is able to predict not only the direction of the change in stability of proteins and affinity of protein–protein and protein–DNA complexes but also the actual numerical experimental value, with correlation coefficients up to 0.824 for a large data set of mutations. We have also applied our methodology to predict changes of stability resulting from mutations occurring in the tumour suppressor protein p53. mCSM outperforms other methods, demonstrating its applicability to understanding mutations that lead to disease.

2 MATERIALS AND METHODS

2.1 mCSM: graph-based signatures

Here, we extend the concept of the inter-atomic distance patterns to a residue environment called mCSM. mCSM signatures can be divided into three major components:

• Graph-based atom distance patterns: The major components of the signatures are distance patterns in the vicinity of the wild-type residue encoded as a cumulative distribution. The wild-type residue environment, which here is defined as the set of atoms within a distance r from its geometric centre, can be modelled as a contact graph, where the atoms are the nodes and the edges are defined by a cutoff distance. In this way, the signatures encode distance patterns by varying the edge-defining cutoff distance, used in computing the number of edges of the resulting atomic contact graph. A cumulative distribution is obtained from an atom distance matrix, which is a pairwise distance calculation between atoms of the environment. Here, as in a previous study (Pires et al., 2013), we use three types of atom classification to segment the cumulative distribution: one class (no distinction between atoms), a binary classification (atoms labelled as polar or hydrophobic) and using the Pmapper pharmacophoric classification, which classifies atoms into eight possible categories: hydrophobic, positive, negative, hydrogen acceptor, hydrogen donor, aromatic, sulphur and neutral. mCSM considers only the residue environment in the wild-type protein. In this way, our method is applicable even when no mutant structures are available. Furthermore, it does not require the generation of homology models.

• Pharmacophore changes: To take into account the changes in atom types due to the mutation, a ‘pharmacophore count’ vector is introduced. Wild-type and mutant residues are represented as pharmacophore frequency vectors as shown in the lower part of Figure 1a and calculated as follows. Let L be a set of pharmacophore types and f a labelling function that assigns a pharmacophore

  
  to a given atom x:
  
  ⁠. The frequency of each type of pharmacophore in a residue is then summarized in a vector p. The difference pchange between pharmacophore count for mutant (pmt) and wild-type (pwt) residue is calculated (⁠
  
  ⁠) and appended to the signature. The atom pharmacophores are characteristics described by PMapper and belong to eight possible classes: hydrophobic, positive, negative, hydrogen acceptor, hydrogen donor, aromatic, sulphur and neutral.

• Experimental conditions: The experimental conditions, in which the thermodynamic data, such as pH and temperature are collected, are also appended to the signatures when available. Relative solvent accessibility of the residue is also included.

with a standard error of ⁠. For this data set, comparing with the AUTOMUTE method (Masso and Vaisman, 2008), mCSM presented a performance equivalent or even better for both regression and classification tasks as showed in Supplementary Tables S3 and Supplementary Data.

with ⁠, whereas for the ProNIT data set the results were with ⁠. In classification tasks, for both data sets the predictive models trained with the mCSM signatures were able to achieve accuracies of >82% and Area Under ROC Curves (AUCs) of 0.826 and 0.853, respectively, as described in Supplementary Table S1.

Supplementary Table S4 shows the mCSM performance in comparison with the program BeAtMuSiC. mCSM achieves a correlation of with ⁠, in comparison with with achieved by the BeAtMuSiC method. mCSM also achieves a correlation of with in a blind test as described in Supplementary Section S4.

We also evaluate our approach by using it to identify disease-related mutations, comparing it with well-established sequence-based methods. Supplementary Table S2 summarizes the obtained results. mCSM achieves a comparable level of accuracy, whereas presenting much better Matthews Correlation Coefficient (MCC) and AUC values.

In addition, we performed experiments on low-redundancy data sets where all mutations in a protein (or position) are either in the test or training set exclusively, as described in Supplementary Section S4. As shown in Supplementary Tables S8 and Supplementary Data, at the protein level the performance tended to be slightly inferior. The distribution of mutations per protein in the data sets is unequal, meaning that information about hundreds of mutations may be available for a single protein, which may be a significant source of bias when defining the folds in cross-validation.

3.3 Case study: predicting stability changes for p53 mutants

Cancer is a complex disease that arises from a combination of genetic and epigenetic changes accumulated over many years. Although there is large variability in the genes implicated in tumorigenesis, >50% of human cancers carry loss of function mutations in the transcription factor p53 (Beroud and Soussi, 2003; Olivier et al. 2002). In response to DNA damage, p53 transactivates a range of genes to induce cell cycle arrest, DNA repair, senescence and apoptosis, depending on the extent and types of DNA damage (Sionov et al., 1999; Vousden et al., 2002).

p53 is composed of three main domains: an N-terminal transactivation domain (amino acid residues 1–45), a DNA binding domain (residues 102–292) and a C-terminal oligomerization domain (residues 319–359) (Sionov et al., 1999; Vousden et al., 2002). Unlike most tumour suppressors that are inactivated by deletion or truncation mutations, mutations in p53 most often result in a protein with a single nucleotide substitution. The majority of these mutations (95%) are located in the DNA binding domain; they either directly interfere with residues involved in DNA binding or disrupt the wild-type conformation and stability of p53 (Olivier et al., 2002). Both of these classes of mutations prevent the transcriptional activation of p53 target genes in a dominant-negative fashion (Vousden et al., 2002).

A new approach in cancer therapy is to find drugs that can rescue the activity of mutant p53. This has primarily focused on trying to enhance the stability p53, potentially reducing the effect of destabilizing mutations and restoring wild-type activity. Several approaches have identified stabilizing molecules that also show a stimulatory effect on p53 DNA binding. These have included antibodies (Hupp et al., 1995), peptides (Selivanova et al., 1999) and small molecules identified via structure-guided design (Boeckler et al., 2008) and screening approaches (Bykov et al., 2002). The most advanced of these is the small molecule PRIMA-1MET (APR-246) that has successfully completed Phase I/II clinical trials, where it was observed that it could induce p53-dependent biological effects in tumour cells in vivo (Lehmann et al., 2012).

We have used mCSM to predict the effect of mutations on the stability of p53. We used the published crystal structure of p53 (PDB ID: 2OCJ), and predicted the change in stability of >42 single mutations within the DNA binding domain of p53 whose thermodynamic effects have previously been experimentally characterized. None of these mutations was present in the training set. In addition to mCSM, we also used SDM and PoPMuSiC to predict the stability changes of these mutations. These predictions were compared directly with the experimentally determined thermodynamic effects (Supplementary Table S6).

mCSM predicted stability changes correlated strongly with the experimentally observed thermodynamic effects (⁠⁠), as shown in Supplementary Table S5. In addition, mCSM was a much better predictor of stability changes in p53 than either SDM (⁠⁠) or PoPMuSiC (⁠⁠), consistent with our larger analysis. In Supplementary Figure S1, we can see that compared with the experimental observations, mCSM predictions did not have a large variation. Significantly, the interquartile range and 95% confidence interval from mCSM predictions were tighter than either SDM or PoPMuSiC.

In general there was good agreement between the algorithms in predicting the direction of change when compared with the experimental data. One interesting deviation, however, was the mutation of arginine 282 to tryptophan, clinically a commonly observed p53 mutation. Arginine 282, shown in Figure 1a and Supplementary Figure S2, is involved in a network of interactions underpinning the loop-sheet-helix major groove DNA binding motif. Mutation to tryptophan results in large structural perturbations, resulting in p53 being largely unfolded, and hence inactive, under physiological conditions (Bullock et al., 2000). mCSM was able to predict accurately the effect of this mutation on the overall stability of p53. Interestingly, both SDM and PoPMuSiC predicted that the R282W mutation would be stabilizing, with SDM predicting that it would actually be stabilizing (Supplementary Table S1). In the case of SDM, the version used in the comparison considers only side chain H-bonds to main chain residues (Worth et al., 2011), which tend to be less critical; however in this case the arginine side chain makes three hdydrogen bonds to other buried or partially buried side chains. This highlights the power of mCSM using the local environment to predict the effects of mutations.

In summary, we have shown that mCSM can predict the effects of mutations on the stability of p53, and can identify disease-associated destabilizing mutations. mCSM provides a reliable way to quickly assess the impact of mutations within the p53 gene, and hence the likelihood of success of stabilizing molecules, which is important within a clinical setting.

4 DISCUSSION AND CONCLUSIONS

We present a new approach, mCSM, for studying the impact of missense mutations in proteins. mCSM, which relies on graph-based signatures, was successfully applied and evaluated in different predictive tasks and was shown to outperform earlier methods. We have successfully applied this methodology to predict stability changes of mutations occurring in p53, demonstrating the applicability of mCSM in a challenging disease scenario.

The results achieved by mCSM support the idea that the impact of a mutation can be correlated with the atomic distance patterns surrounding an amino acid residue. The distant patterns describe the nature of the environment of the residue in the wild-type protein. They contrast with amino acid substitution patterns as used in SDM, which define the environment as a function of the residues immediately surrounding the residue. Thus, a solvent inaccessible residue in SDM will have the same defined environment whether it is on the core of the protein or close to the surface. On the other hand, the distance matrix of mCSM will differ according to the depth of the residue and the curvature of the protein surface that lies within a large radius, in this work, up to 10 Å. This is reminiscent of the focus on ‘depth’ championed by Chakravarty and Varadarajan (1999). The mCSM distance matrix will also be sensitive to the nature of the electrostatic environment, which appears to be less well described in PoPMuSiC. These advantages are evidenced in the feature selection analysis (Supplementary Fig. S3), which shows that long-range distances are the most discriminative attributes of the signatures, usually polar–polar and hydrophobic–hydrophobic atom frequencies for distances beyond 6 Å for the SKEMPI data set. A latent semantic analysis (Supplementary Fig. S4) shows that the majority of the variability of the signatures can be explained with atom frequencies for long-range distances. In this way, mCSM perceives residue environment density and depth implicitly, without relying on direct calculations or thresholds. A similar analysis was done for the other data sets (Supplementary Figs. S5 and Supplementary Data).

The mCSM approach shows that a good description of the effect of mutations on the wild-type structure can be achieved using the pharmacophore count. This contrasts with use of the immediate environment of the residue to define a pharmacophore, which is de-emphasized in mCSM. Although in mCSM the pharmacophore is not described in terms of a defined structure of the mutant protein, there may still be further gains in the quality of the prediction if this could be incorporated into mCSM. A further difference from many previous methods is the omission of an assessment of the effect of the mutation on the unfolded state. This is explicitly described in perturbation methods and in programs like SDM. It appears this is either less important to the estimation of the change of stability that arises from the majority of mutations or the unfolded state is insufficiently well described in methods such as SDM.

mCSM does not depend on the observation of mutations that have occurred in evolution of proteins. These have generally been selected over long evolutionary times for minor positive or negative changes on stability that were selectively advantageous to the organism. Mutations that occur in cancer, such as those in p53 described here, are often deleterious to the function of the protein and may not be sampled in a statistically significant manner in evolution when information on residue structural environment is required in the calculation. Therefore, they may not be properly accounted for in SDM environment-specific substitution tables. This is almost certainly the case for the mutation of arginine 282 to tryptophan in p53, which has significant effects on the stability of the protein. This underlines the importance of understanding theoretically the effects of various parameters that influence stability a major focus of both mCSM and PoPMuSiC.

In future works we intend to apply our methodology for predicting affinity changes to protein-ligand complexes, to understand the affects of mutations that occur in drug resistance in cancer and other diseases. We believe our machine learning approach is complementary to those based on potential energy functions like SDM and PoPMuSiC. This way, we intend to combine them into a hybrid method.

ACKNOWLEDGEMENT

The authors thank Harry Jubb and Bernardo Ochoa for helpful discussions and feedback.

Funding: Brazilian agency Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)—Brazil (to D.E.V.P). Victoria Fellowship from the Victorian Government and the Leslie (Les) J. Fleming Churchill Fellowship from the The Winston Churchill Memorial Trust (to D.B.A.). University of Cambridge and The Wellcome Trust for facilities and support (to T.L.B.).

Conflict of Interest: none declared.

REFERENCES

, et al. 

A method and server for predicting damaging missense mutations

, et al. 

Protein–DNA interactions: structural, thermodynamic and clustering patterns of conserved residues in DNA-binding proteins

, et al. 

Effects of common cancer mutations on stability and DNA binding of full-length p53 compared with isolated core domains

,  . 

The umd-p53 database: new mutations and analysis tools

, et al. 

Targeted rescue of a destabilized mutant of p53 by an in silico screened drug

,  . 

Large-scale prediction of protein geometry and stability changes for arbitrary single point mutations

, et al. 

Quantitative analysis of residual folding and DNA binding in mutant p53 core domain: definition of mutant states for rescue in cancer therapy

, et al. 

Restoration of the tumor suppressor function to mutant p53 by a low-molecular-weight compound

,  . 

Improving the prediction of disease-related variants using protein three-dimensional structure

, et al. 

I-mutant2. 0: predicting stability changes upon mutation from the protein sequence or structure

, et al. 

Predicting protein stability changes from sequences using support vector machines

,  . 

Residue depth: a novel parameter for the analysis of protein structure and stability

, et al. 

Prediction of protein stability changes for single-site mutations using support vector machines

, et al. 

Prediction by graph theoretic measures of structural effects in proteins arising from non-synonymous single nucleotide polymorphisms

, et al. 

Protein cutoff scanning: a comparative analysis of cutoff dependent and cutoff free methods for prospecting contacts in proteins

, et al. 

Fast and accurate predictions of protein stability changes upon mutations using statistical potentials and neural networks: PoPMuSiC-2.0

, et al. 

BeAtMuSiC: prediction of changes in protein-protein binding affinity on mutations

. 

Dissection of the structure and activity of the tyrosyl-trna synthetase by site-directed mutagenesis

, et al. 

Predicting changes in the stability of proteins and protein complexes: a study of more than 1000 mutations

, et al. 

Small peptides activate the latent sequence-specific DNA binding function of p53

, et al. 

Structural basis for understanding oncogenic p53 mutations and designing rescue drugs

Proc. Natl Acad. Sci. USA

, et al. 

Role of conformational sampling in computing mutation-induced changes in protein structure and stability

,  . 

A simple physical model for binding energy hot spots in protein–protein complexes

Proc. Natl Acad. Sci. USA

, et al. 

Protherm and pronit: thermodynamic databases for proteins and protein–nucleic acid interactions

, et al. 

Targeting p53 in vivo: A first-in-human study with p53-targeting compound apr-246 in refractory hematologic malignancies and prostate cancer

,  . 

Accurate prediction of stability changes in protein mutants by combining machine learning with structure based computational mutagenesis

,  . 

SKEMPI: a structural kinetic and energetic database of mutant protein interactions and its use in empirical models

, et al. 

Protein–protein binding affinity prediction on a diverse set of structures

, et al. 

SCOP: a structural classification of proteins database for the investigation of sequences and structures

,  . 

Sift: predicting amino acid changes that affect protein function

, et al. 

Semirational design of active tumor suppressor p53 DNA binding domain with enhanced stability

Proc. Natl Acad. Sci. USA

, et al. 

Mechanism of rescue of common p53 cancer mutations by second-site suppressor mutations

, et al. 

The iarc tp53 database: new online mutation analysis and recommendations to users

, et al. 

Cutoff Scanning Matrix (CSM): structural classification and function prediction by protein inter-residue distance patterns

, et al. 

aCSM: noise-free graph-based signatures to large-scale receptor-based ligand prediction

, et al. 

Reactivation of mutant p53 through interaction of a c-terminal peptide with the core domain

,  . 

The cellular response to p53: the decision between life and death

, et al. 

Predicting changes in protein thermostability brought about by single-or multi-site mutations

, et al. 

Prediction of the stability of protein mutants based on structural environment-dependent amino acid substitution and propensity tables

,  . 

Live or let die: the cell’s response to p53

, et al. 

SDM – a server for predicting effects of mutations on protein stability and malfunction

, et al. 

Specificity and affinity quantification of protein-protein interactions

© The Author 2013. Published by Oxford University Press.

This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.

What is direct effect of mutation?

What is the direct cause of mutations in organisms?

What are the 3 outcomes of mutation?

What is the effect of most mutations?