Lower Activation in Frontal Cortex and Posterior Cingulate Cortex Observed during Sex Determination Test in Early-Stage Dementia of the Alzheimer Type.

Face-labeling refers to the ability to classify faces into social categories. This plays a critical role in human interaction as it serves to define concepts of socially acceptable interpersonal behavior. The purpose of the current study was to characterize, what, if any, impairments in face-labeling are detectable in participants with early-stage clinically diagnosed dementia of the Alzheimer type (CDDAT) through the use of the sex determination test (SDT). In the current study, four (1 female, 3 males) CDDAT and nine (4 females, 5 males) age-matched neurotypicals (NT) completed the SDT using chimeric faces while undergoing BOLD fMRI. It was expected that CDDAT participants would have poor verbal fluency, which would correspond to poor performance on the SDT. This could be explained by decreased activation and connectivity patterns within the fusiform face area (FFA) and anterior cingulate cortex (ACC). DTI was also performed to test the association of pathological deterioration of connectivity in the uncinate fasciculus (UF) and verbally-mediated performance. CDDAT showed lower verbal fluency test (VFT) performance, but VFT was not significantly correlated to SDT and no significant difference was seen between CDDAT and NT for SDT performance as half of the CDDAT performed substantially worse than NT while the other half performed similarly. BOLD fMRI of SDT displayed differences in the left superior frontal gyrus and posterior cingulate cortex (PCC), but not the FFA or ACC. Furthermore, although DTI showed deterioration of the right inferior and superior longitudinal fasciculi, as well as the PCC, it did not demonstrate significant deterioration of UF tracts. Taken together, early-stage CDDAT may represent a common emerging point for the loss of face labeling ability.

White Matter Deterioration May Foreshadow Impairment of Emotional Valence Determination in Early-Stage Dementia of the Alzheimer Type.

In Alzheimer Disease (AD), non-verbal skills often remain intact for far longer than verbally mediated processes. Four (1 female, 3 males) participants with early-stage Clinically Diagnosed Dementia of the Alzheimer Type (CDDAT) and eight neurotypicals (NTs; 4 females, 4 males) completed the emotional valence determination test (EVDT) while undergoing BOLD functional magnetic resonance imaging (fMRI). We expected CDDAT participants to perform just as well as NTs on the EVDT, and to display increased activity within the bilateral amygdala and right anterior cingulate cortex (r-ACC). We hypothesized that such activity would reflect an increased reliance on these structures to compensate for on-going neuronal loss in frontoparietal regions due to the disease. We used diffusion tensor imaging (DTI) to determine if white matter (WM) damage had occurred in frontoparietal regions as well. CDDAT participants had similar behavioral performance and no differences were observed in brain activity or connectivity patterns within the amygdalae or r-ACC. Decreased fractional anisotropy (FA) values were noted, however, for the bilateral superior longitudinal fasciculi and posterior cingulate cortex (PCC). We interpret these findings to suggest that emotional valence determination and non-verbal skill sets are largely intact at this stage of the disease, but signs foreshadowing future decline were revealed by possible WM deterioration. Understanding how non-verbal skill sets are altered, while remaining largely intact, offers new insights into how non-verbal communication may be more successfully implemented in the care of AD patients and highlights the potential role of DTI as a presymptomatic biomarker.

Sequence amplification via cell passaging creates spurious signals of positive adaptation in influenza virus H3N2 hemagglutinin.

Clinical influenza A virus isolates are frequently not sequenced directly. Instead, a majority of these isolates (~70% in 2015) are first subjected to passaging for amplification, most commonly in non-human cell culture. Here, we find that this passaging leaves distinct signals of adaptation, which can confound evolutionary analyses of the viral sequences. We find distinct patterns of adaptation to Madin-Darby (MDCK) and monkey cell culture absent from unpassaged hemagglutinin sequences. These patterns also dominate pooled datasets not separated by passaging type, and they increase in proportion to the number of passages performed. By contrast, MDCK-SIAT1 passaged sequences seem mostly (but not entirely) free of passaging adaptations. Contrary to previous studies, we find that using only internal branches of influenza virus phylogenetic trees is insufficient to correct for passaging artifacts. These artifacts can only be safely avoided by excluding passaged sequences entirely from subsequent analysis. We conclude that future influenza virus evolutionary analyses should appropriately control for potentially confounding effects of passaging adaptations.

Functional Sites Induce Long-Range Evolutionary Constraints in Enzymes.

Functional residues in proteins tend to be highly conserved over evolutionary time. However, to what extent functional sites impose evolutionary constraints on nearby or even more distant residues is not known. Here, we report pervasive conservation gradients toward catalytic residues in a dataset of 524 distinct enzymes: evolutionary conservation decreases approximately linearly with increasing distance to the nearest catalytic residue in the protein structure. This trend encompasses, on average, 80% of the residues in any enzyme, and it is independent of known structural constraints on protein evolution such as residue packing or solvent accessibility. Further, the trend exists in both monomeric and multimeric enzymes and irrespective of enzyme size and/or location of the active site in the enzyme structure. By contrast, sites in protein-protein interfaces, unlike catalytic residues, are only weakly conserved and induce only minor rate gradients. In aggregate, these observations show that functional sites, and in particular catalytic residues, induce long-range evolutionary constraints in enzymes.

The utility of protein structure as a predictor of site-wise dN/dS varies widely among HIV-1 proteins.

Protein structure acts as a general constraint on the evolution of viral proteins. One widely recognized structural constraint explaining evolutionary variation among sites is the relative solvent accessibility (RSA) of residues in the folded protein. In influenza virus, the distance from functional sites has been found to explain an additional portion of the evolutionary variation in the external antigenic proteins. However, to what extent RSA and distance from a reference site in the protein can be used more generally to explain protein adaptation in other viruses and in the different proteins of any given virus remains an open question. To address this question, we have carried out an analysis of the distribution and structural predictors of site-wise dN/dS in HIV-1. Our results indicate that the distribution of dN/dS in HIV follows a smooth gamma distribution, with no special enrichment or depletion of sites with dN/dS at or above one. The variation in dN/dS can be partially explained by RSA and distance from a reference site in the protein, but these structural constraints do not act uniformly among the different HIV-1 proteins. Structural constraints are highly predictive in just one of the three enzymes and one of three structural proteins in HIV-1. For these two proteins, the protease enzyme and the gp120 structural protein, structure explains between 30 and 40% of the variation in dN/dS. Finally, for the gp120 protein of the receptor-binding complex, we also find that glycosylation sites explain just 2% of the variation in dN/dS and do not explain gp120 evolution independently of either RSA or distance from the apical surface.

Computational and Functional Analysis of the Virus-Receptor Interface Reveals Host Range Trade-Offs in New World Arenaviruses.

UNLABELLED: Animal viruses frequently cause zoonotic disease in humans. As these viruses are highly diverse, evaluating the threat that they pose remains a major challenge, and efficient approaches are needed to rapidly predict virus-host compatibility. Here, we develop a combined computational and experimental approach to assess the compatibility of New World arenaviruses, endemic in rodents, with the host TfR1 entry receptors of different potential new host species. Using signatures of positive selection, we identify a small motif on rodent TfR1 that conveys species specificity to the entry of viruses into cells. However, we show that mutations in this region affect the entry of each arenavirus differently. For example, a human single nucleotide polymorphism (SNP) in this region, L212V, makes human TfR1 a weaker receptor for one arenavirus, Machupo virus, but a stronger receptor for two other arenaviruses, Junin and Sabia viruses. Collectively, these findings set the stage for potential evolutionary trade-offs, where natural selection for resistance to one virus may make humans or rodents susceptible to other arenavirus species. Given the complexity of this host-virus interplay, we propose a computational method to predict these interactions, based on homology modeling and computational docking of the virus-receptor protein-protein interaction. We demonstrate the utility of this model for Machupo virus, for which a suitable cocrystal structural template exists. Our model effectively predicts whether the TfR1 receptors of different species will be functional receptors for Machupo virus entry. Approaches such at this could provide a first step toward computationally predicting the "host jumping" potential of a virus into a new host species. IMPORTANCE: We demonstrate how evolutionary trade-offs may exist in the dynamic evolutionary interplay between viruses and their hosts, where natural selection for resistance to one virus could make humans or rodents susceptible to other virus species. We present an algorithm that predicts which species have cell surface receptors that make them susceptible to Machupo virus, based on computational docking of protein structures. Few molecular models exist for predicting the risk of spillover of a particular animal virus into humans or new animal populations. Our results suggest that a combination of evolutionary analysis, structural modeling, and experimental verification may provide an efficient approach for screening and assessing the potential spillover risks of viruses circulating in animal populations.

Evolution. Systematic humanization of yeast genes reveals conserved functions and genetic modularity.

To determine whether genes retain ancestral functions over a billion years of evolution and to identify principles of deep evolutionary divergence, we replaced 414 essential yeast genes with their human orthologs, assaying for complementation of lethal growth defects upon loss of the yeast genes. Nearly half (47%) of the yeast genes could be successfully humanized. Sequence similarity and expression only partly predicted replaceability. Instead, replaceability depended strongly on gene modules: Genes in the same process tended to be similarly replaceable (e.g., sterol biosynthesis) or not (e.g., DNA replication initiation). Simulations confirmed that selection for specific function can maintain replaceability despite extensive sequence divergence. Critical ancestral functions of many essential genes are thus retained in a pathway-specific manner, resilient to drift in sequences, splicing, and protein interfaces.

Geometric Constraints Dominate the Antigenic Evolution of Influenza H3N2 Hemagglutinin.

We have carried out a comprehensive analysis of the determinants of human influenza A H3 hemagglutinin evolution. We consider three distinct predictors of evolutionary variation at individual sites: solvent accessibility (as a proxy for protein fold stability and/or conservation), Immune Epitope Database (IEDB) epitope sites (as a proxy for host immune bias), and proximity to the receptor-binding region (as a proxy for one of the functions of hemagglutinin-to bind sialic acid). Individually, these quantities explain approximately 15% of the variation in site-wise dN/dS. In combination, solvent accessibility and proximity explain 32% of the variation in dN/dS; incorporating IEDB epitope sites into the model adds only an additional 2 percentage points. Thus, while solvent accessibility and proximity perform largely as independent predictors of evolutionary variation, they each overlap with the epitope-sites predictor. Furthermore, we find that the historical H3 epitope sites, which date back to the 1980s and 1990s, only partially overlap with the experimental sites from the IEDB, and display similar overlap in predictive power when combined with solvent accessibility and proximity. We also find that sites with dN/dS > 1, i.e., the sites most likely driving seasonal immune escape, are not correctly predicted by either historical or IEDB epitope sites, but only by proximity to the receptor-binding region. In summary, a simple geometric model of HA evolution outperforms a model based on epitope sites. These results suggest that either the available epitope sites do not accurately represent the true influenza antigenic sites or that host immune bias may be less important for influenza evolution than commonly thought.

Time dependence of evolutionary metrics during the 2009 pandemic influenza virus outbreak.

With the expansion of DNA sequencing technology, quantifying evolution in emerging viral outbreaks has become an important tool for scientists and public health officials. Although it is known that the degree of sequence divergence significantly affects the calculation of evolutionary metrics in viral outbreaks, the extent and duration of this effect during an actual outbreak remains unclear. We have analyzed how limited divergence time during an early viral outbreak affects the accuracy of molecular evolutionary metrics. Using sequence data from the first 25 months of the 2009 pandemic H1N1 (pH1N1) outbreak, we calculated each of three different standard evolutionary metrics-molecular clock rate (i.e., evolutionary rate), whole gene dN/dS, and site-wise dN/dS-for hemagglutinin and neuraminidase, using increasingly longer time windows, from 1 month to 25 months. For the molecular clock rate, we found that at least three to four months of temporal divergence from the start of sampling was required to make precise estimates that also agreed with long-term values. For whole gene dN/dS, we found that at least two months of data were required to generate precise estimates, but six to nine months were required for estimates to approach their long term values. For site-wise dN/dS estimates, we found that at least six months of sampling divergence was required before the majority of sites had at least one mutation and were thus evolutionarily informative. Furthermore, eight months of sampling divergence was required before the site-wise estimates appropriately reflected the distribution of values expected from known protein-structure-based evolutionary pressure in influenza. In summary, we found that evolutionary metrics calculated from gene sequence data in early outbreaks should be expected to deviate from their long-term estimates for at least several months after the initial emergence and sequencing of the virus.

Predicting growth conditions from internal metabolic fluxes in an in-silico model of E. coli.

A widely studied problem in systems biology is to predict bacterial phenotype from growth conditions, using mechanistic models such as flux balance analysis (FBA). However, the inverse prediction of growth conditions from phenotype is rarely considered. Here we develop a computational framework to carry out this inverse prediction on a computational model of bacterial metabolism. We use FBA to calculate bacterial phenotypes from growth conditions in E. coli, and then we assess how accurately we can predict the original growth conditions from the phenotypes. Prediction is carried out via regularized multinomial regression. Our analysis provides several important physiological and statistical insights. First, we show that by analyzing metabolic end products we can consistently predict growth conditions. Second, prediction is reliable even in the presence of small amounts of impurities. Third, flux through a relatively small number of reactions per growth source (∼10) is sufficient for accurate prediction. Fourth, combining the predictions from two separate models, one trained only on carbon sources and one only on nitrogen sources, performs better than models trained to perform joint prediction. Finally, that separate predictions perform better than a more sophisticated joint prediction scheme suggests that carbon and nitrogen utilization pathways, despite jointly affecting cellular growth, may be fairly decoupled in terms of their dependence on specific assortments of molecular precursors.

Identifying structural variation in haploid microbial genomes from short-read resequencing data using breseq.

BACKGROUND: Mutations that alter chromosomal structure play critical roles in evolution and disease, including in the origin of new lifestyles and pathogenic traits in microbes. Large-scale rearrangements in genomes are often mediated by recombination events involving new or existing copies of mobile genetic elements, recently duplicated genes, or other repetitive sequences. Most current software programs for predicting structural variation from short-read DNA resequencing data are intended primarily for use on human genomes. They typically disregard information in reads mapping to repeat sequences, and significant post-processing and manual examination of their output is often required to rule out false-positive predictions and precisely describe mutational events. RESULTS: We have implemented an algorithm for identifying structural variation from DNA resequencing data as part of the breseq computational pipeline for predicting mutations in haploid microbial genomes. Our method evaluates the support for new sequence junctions present in a clonal sample from split-read alignments to a reference genome, including matches to repeat sequences. Then, it uses a statistical model of read coverage evenness to accept or reject these predictions. Finally, breseq combines predictions of new junctions and deleted chromosomal regions to output biologically relevant descriptions of mutations and their effects on genes. We demonstrate the performance of breseq on simulated Escherichia coli genomes with deletions generating unique breakpoint sequences, new insertions of mobile genetic elements, and deletions mediated by mobile elements. Then, we reanalyze data from an E. coli K-12 mutation accumulation evolution experiment in which structural variation was not previously identified. Transposon insertions and large-scale chromosomal changes detected by breseq account for ~25% of spontaneous mutations in this strain. In all cases, we find that breseq is able to reliably predict structural variation with modest read-depth coverage of the reference genome (>40-fold). CONCLUSIONS: Using breseq to predict structural variation should be useful for studies of microbial epidemiology, experimental evolution, synthetic biology, and genetics when a reference genome for a closely related strain is available. In these cases, breseq can discover mutations that may be responsible for important or unintended changes in genomes that might otherwise go undetected.

Predicting evolutionary site variability from structure in viral proteins: buriedness, packing, flexibility, and design.

Several recent works have shown that protein structure can predict site-specific evolutionary sequence variation. In particular, sites that are buried and/or have many contacts with other sites in a structure have been shown to evolve more slowly, on average, than surface sites with few contacts. Here, we present a comprehensive study of the extent to which numerous structural properties can predict sequence variation. The quantities we considered include buriedness (as measured by relative solvent accessibility), packing density (as measured by contact number), structural flexibility (as measured by B factors, root-mean-square fluctuations, and variation in dihedral angles), and variability in designed structures. We obtained structural flexibility measures both from molecular dynamics simulations performed on nine non-homologous viral protein structures and from variation in homologous variants of those proteins, where they were available. We obtained measures of variability in designed structures from flexible-backbone design in the Rosetta software. We found that most of the structural properties correlate with site variation in the majority of structures, though the correlations are generally weak (correlation coefficients of 0.1-0.4). Moreover, we found that buriedness and packing density were better predictors of evolutionary variation than structural flexibility. Finally, variability in designed structures was a weaker predictor of evolutionary variability than buriedness or packing density, but it was comparable in its predictive power to the best structural flexibility measures. We conclude that simple measures of buriedness and packing density are better predictors of evolutionary variation than the more complicated predictors obtained from dynamic simulations, ensembles of homologous structures, or computational protein design.

Analyzing machupo virus-receptor binding by molecular dynamics simulations.

In many biological applications, we would like to be able to computationally predict mutational effects on affinity in protein-protein interactions. However, many commonly used methods to predict these effects perform poorly in important test cases. In particular, the effects of multiple mutations, non alanine substitutions, and flexible loops are difficult to predict with available tools and protocols. We present here an existing method applied in a novel way to a new test case; we interrogate affinity differences resulting from mutations in a host-virus protein-protein interface. We use steered molecular dynamics (SMD) to computationally pull the machupo virus (MACV) spike glycoprotein (GP1) away from the human transferrin receptor (hTfR1). We then approximate affinity using the maximum applied force of separation and the area under the force-versus-distance curve. We find, even without the rigor and planning required for free energy calculations, that these quantities can provide novel biophysical insight into the GP1/hTfR1 interaction. First, with no prior knowledge of the system we can differentiate among wild type and mutant complexes. Moreover, we show that this simple SMD scheme correlates well with relative free energy differences computed via free energy perturbation. Second, although the static co-crystal structure shows two large hydrogen-bonding networks in the GP1/hTfR1 interface, our simulations indicate that one of them may not be important for tight binding. Third, one viral site known to be critical for infection may mark an important evolutionary suppressor site for infection-resistant hTfR1 mutants. Finally, our approach provides a framework to compare the effects of multiple mutations, individually and jointly, on protein-protein interactions.

Alternate splicing of dysferlin C2A confers Ca²âº-dependent and Ca²âº-independent binding for membrane repair.

Dysferlin plays a critical role in the Ca²âº-dependent repair of microlesions that occur in the muscle sarcolemma. Of the seven C2 domains in dysferlin, only C2A is reported to bind both Ca²âº and phospholipid, thus acting as a key sensor in membrane repair. Dysferlin C2A exists as two isoforms, the "canonical" C2A and C2A variant 1 (C2Av1). Interestingly, these isoforms have markedly different responses to Ca²âº and phospholipid. Structural and thermodynamic analyses are consistent with the canonical C2A domain as a Ca²âº-dependent, phospholipid-binding domain, whereas C2Av1 would likely be Ca²âº-independent under physiological conditions. Additionally, both isoforms display remarkably low free energies of stability, indicative of a highly flexible structure. The inverted ligand preference and flexibility for both C2A isoforms suggest the capability for both constitutive and Ca²âº-regulated effector interactions, an activity that would be essential in its role as a mediator of membrane repair.

Maximum allowed solvent accessibilites of residues in proteins.

The relative solvent accessibility (RSA) of a residue in a protein measures the extent of burial or exposure of that residue in the 3D structure. RSA is frequently used to describe a protein's biophysical or evolutionary properties. To calculate RSA, a residue's solvent accessibility (ASA) needs to be normalized by a suitable reference value for the given amino acid; several normalization scales have previously been proposed. However, these scales do not provide tight upper bounds on ASA values frequently observed in empirical crystal structures. Instead, they underestimate the largest allowed ASA values, by up to 20%. As a result, many empirical crystal structures contain residues that seem to have RSA values in excess of one. Here, we derive a new normalization scale that does provide a tight upper bound on observed ASA values. We pursue two complementary strategies, one based on extensive analysis of empirical structures and one based on systematic enumeration of biophysically allowed tripeptides. Both approaches yield congruent results that consistently exceed published values. We conclude that previously published ASA normalization values were too small, primarily because the conformations that maximize ASA had not been correctly identified. As an application of our results, we show that empirically derived hydrophobicity scales are sensitive to accurate RSA calculation, and we derive new hydrophobicity scales that show increased correlation with experimentally measured scales.

PeptideBuilder: A simple Python library to generate model peptides.

We present a simple Python library to construct models of polypeptides from scratch. The intended use case is the generation of peptide models with pre-specified backbone angles. For example, using our library, one can generate a model of a set of amino acids in a specific conformation using just a few lines of python code. We do not provide any tools for energy minimization or rotamer packing, since powerful tools are available for these purposes. Instead, we provide a simple Python interface that enables one to add residues to a peptide chain in any desired conformation. Bond angles and bond lengths can be manipulated if so desired, and reasonable values are used by default.

Cross-species comparison of site-specific evolutionary-rate variation in influenza haemagglutinin.

We investigate the causes of site-specific evolutionary-rate variation in influenza haemagglutinin (HA) between human and avian influenza, for subtypes H1, H3, and H5. By calculating the evolutionary-rate ratio, ω = dN/dS as a function of a residue's solvent accessibility in the three-dimensional protein structure, we show that solvent accessibility has a significant but relatively modest effect on site-specific rate variation. By comparing rates within HA subtypes among host species, we derive an upper limit to the amount of variation that can be explained by structural constraints of any kind. Protein structure explains only 20-40% of the variation in ω. Finally, by comparing ω at sites near the sialic-acid-binding region to ω at other sites, we show that ω near the sialic-acid-binding region is significantly elevated in both human and avian influenza, with the exception of avian H5. We conclude that protein structure, HA subtype, and host biology all impose distinct selection pressures on sites in influenza HA.

Computational tests of a thermal cycling strategy to isolate more complex functional nucleic acid motifs from random sequence pools by in vitro selection.

The dual information-function nature of nucleic acids has been exploited in the laboratory to isolate novel receptors and catalysts from random DNA and RNA sequences by cycles of in vitro selection and amplification. This strategy is particularly effective because, unlike polypeptides with random amino acid sequences, nucleic acids with random base sequences are often capable of stably folding into defined three-dimensional structures. However, the pervasive base-pairing potential of nucleic acids is also known to lead to kinetic traps in their folding landscapes. That is, the same DNA or RNA sequence can often adopt alternative base-paired structures that are local energy minima, and these folds may interconvert very slowly. We have used simulations with nucleic acid folding algorithms to evaluate the effect of misfolding on in vitro selection experiments. We demonstrate that kinetic traps can prevent the recovery of novel families of complex functional motifs by two mechanisms. First, misfolding can lead to the stochastic loss of unique sequences in the first round of selection. Second, frequent misfolding can reduce the average activity of multiple copies of a sequence to such an extent that it will be outcompeted after multiple rounds of selection. In these simulations, adding thermal cycling to sample multiple folds of one sequence during a selection for a self-modifying catalytic activity can improve the recovery of rare examples of more complex structures. Although newly isolated sequences may fold poorly, they can represent footholds in sequence space that can be improved to reliably fold after a few mutations. Thus, it is plausible that thermal cycling by day-night cycles or other mechanisms on the primordial earth may have been important for the evolution of the first RNA catalysts, and a fold sampling strategy might be used to search for more effective nucleic acid catalysts in the laboratory today.

Integrating sequence variation and protein structure to identify sites under selection.

We present a novel method to identify sites under selection in protein-coding genes. Our method combines the traditional Goldman-Yang model of coding-sequence evolution with the information obtained from the 3D structure of the evolving protein, specifically the relative solvent accessibility (RSA) of individual residues. We develop a random-effects likelihood sites model in which rate classes are RSA dependent. The RSA dependence is modeled with linear functions. We demonstrate that our RSA-dependent model provides a significantly better fit to molecular sequence data than does a traditional, RSA-independent model. We further show that our model provides a natural, RSA-dependent neutral baseline for the evolutionary rate ratio ω = dN/dS Sites that deviate from this neutral baseline likely experience selection pressure for function. We apply our method to the influenza proteins hemagglutinin and neuraminidase. For hemagglutinin, our method recovers positively selected sites near the sialic acid-binding site and negatively selected sites that may be important for trimerization. For neuraminidase, our method recovers the oseltamivir resistance site and otherwise suggests that few sites deviate from the neutral baseline. Our method is broadly applicable to any protein sequences for which structural data are available or can be obtained via homology modeling or threading.

Modeling coding-sequence evolution within the context of residue solvent accessibility.

BACKGROUND: Protein structure mediates site-specific patterns of sequence divergence. In particular, residues in the core of a protein (solvent-inaccessible residues) tend to be more evolutionarily conserved than residues on the surface (solvent-accessible residues). RESULTS: Here, we present a model of sequence evolution that explicitly accounts for the relative solvent accessibility of each residue in a protein. Our model is a variant of the Goldman-Yang 1994 (GY94) model in which all model parameters can be functions of the relative solvent accessibility (RSA) of a residue. We apply this model to a data set comprised of nearly 600 yeast genes, and find that an evolutionary-rate ratio ω that varies linearly with RSA provides a better model fit than an RSA-independent ω or an ω that is estimated separately in individual RSA bins. We further show that the branch length t and the transition-transverion ratio κ also vary with RSA. The RSA-dependent GY94 model performs better than an RSA-dependent Muse-Gaut 1994 (MG94) model in which the synonymous and non-synonymous rates individually are linear functions of RSA. Finally, protein core size affects the slope of the linear relationship between ω and RSA, and gene expression level affects both the intercept and the slope. CONCLUSIONS: Structure-aware models of sequence evolution provide a significantly better fit than traditional models that neglect structure. The linear relationship between ω and RSA implies that genes are better characterized by their ω slope and intercept than by just their mean ω.