The human protein coevolution network

Coevolution maintains interactions between phenotypic traits through the process of reciprocal natural selection. Detecting molecular coevolution can expose functional interactions between molecules in the cell, generating insights into biological processes, pathways, and the networks of interactions important for cellular function. Prediction of interaction partners from different protein families exploits the property that interacting proteins can follow similar patterns and relative rates of evolution. Current methods for detecting coevolution based on the similarity of phylogenetic trees or evolutionary distance matrices have, however, been limited by requiring coevolution over the entire evolutionary history considered and are inaccurate in the presence of paralogous copies. We present a novel method for determining coevolving protein partners by finding the largest common submatrix in a given pair of distance matrices, with the size of the largest common submatrix measuring the strength of coevolution. This approach permits us to consider matrices of different size and scale, to find lineage-specific coevolution, and to predict multiple interaction partners. We used MatrixMatchMaker to predict protein–protein interactions in the human genome. We show that proteins that are known to interact physically are more strongly coevolving than proteins that simply belong to the same biochemical pathway. The human coevolution network is highly connected, suggesting many more protein–protein interactions than are currently known from high-throughput and other experimental evidence. These most strongly coevolving proteins suggest interactions that have been maintained over long periods of evolutionary time, and that are thus likely to be of fundamental importance to cellular function.Cellular proteins rarely work in isolation, but are more usually involved in pathways and interaction networks. Dysfunctional networks are commonly implicated in disease, so an elucidation of protein–protein interactions can greatly contribute to our understanding of pathological states, and much more broadly, of molecular biology. Systems biology relies on accurate representations of interaction networks, but these are often hard to describe. Interactions can be conditional or contextual, and may not always be captured in a given study, regardless of its attention to quality. Complementary approaches based on experimental data as well as on sequence and evolutionary analyses are required in order to describe a system with a sufficient degree of detail so that it may effectively be understood.Large protein–protein interaction networks are commonly obtained from high-throughput experiments. Such large-scale efforts have described the interaction maps for organisms such as Helicobacter pylori (Rain et al. 2001), Escherichia coli (Butland et al. 2005), Saccharomyces cerevisiae (Ito et al. 2000; Schwikowski et al. 2000; Uetz et al. 2000; Gavin et al. 2002; Ho et al. 2002), Caenorhabditis elegans (Walhout et al. 2000; Simonis et al. 2009), Drosophila melanogaster (Giot et al. 2003), and Homo sapiens (Rual et al. 2005; Stelzl et al. 2005). These studies have been extremely valuable toward our understanding of protein interaction networks, but they suffer from inherent experimental biases in terms of the types of interaction that can be detected. The first S. cerevisiae network, for example, was estimated to have a false-negative rate of 90% and a false-positive rate of 50% (von Mering et al. 2002; Sprinzak et al. 2003). Nevertheless, accuracy can be increased by combining data sets (Bader and Hogue 2002; von Mering et al. 2002; Yu et al. 2008), by repeated screening (Venkatesan et al. 2009) and confidence evaluation (Yu et al. 2008; Braun et al. 2009), and by comparing data sets from different species (Kalaev et al. 2008). Several underlying principles of protein interaction networks have thus been discovered. Their scale-free nature (Barabási and Oltvai 2004) suggests that there are a small number of very highly connected proteins (hubs), which form highly connected modules of proteins sharing well-defined functions (Snel et al. 2002; Rives and Galitski 2003; Spirin and Mirny 2003; Wuchty and Almaas 2005). These hub proteins tend to be found in more species, to have higher sequence conservation and thus to be slow to evolve, and are essential for survival (Fraser et al. 2002, 2003; Jordan et al. 2003; Wuchty et al. 2006). Although many protein interactions appear to be lineage-specific (Mika and Rost 2006; Cusick et al. 2009), evolutionarily conserved proteins maintain their interaction partners (Wuchty et al. 2006; Yellaboina et al. 2008).Many examples of interacting protein families have been shown to coevolve, with interacting members of the different families displaying similar phylogenetic trees (Moyle et al. 1994; van Kesteren et al. 1996; Goh et al. 2000). This observed correlation between the pattern and rate of evolution of members of functionally related protein families is thought to be the result of the functional complementarity of gene duplication and divergence events (Fryxell 1996). Kim et al. (2004) analyzed protein family pairs and found significant correlation in the evolutionary patterns for 78% of the 454 that were reliably known to interact. Hakes et al. (2007) showed that the correlation of evolutionary rates between interacting proteins in yeast cannot simply be explained by the covariation between functional residues in isolation. However, others have been able to predict interacting domains from coevolving residues between domains or proteins (Jothi et al. 2006; Kann et al. 2007, 2009; Yeang and Haussler 2007).Great efforts have been made in using sequence information alone to predict interacting proteins and domains (for review, see Shoemaker and Panchenko 2007; Jothi and Przytycka 2008), all based to some degree on coevolutionary arguments. One method has been to consider the conservation of gene order, whereby an interaction influences the relative placement of these genes in the genome, in some cases fusing the genes altogether. Another has been to consider the phylogenetic profile of genes, whereby the presence or absence of a gene influences the presence or absence of other genes.Coevolution, as more directly measured by the similarity of evolutionary trees, is often used to predict interacting proteins and domains (Ramani and Marcotte 2003; Valencia and Pazos 2003; Pazos et al. 2008). The technique does not normally compare trees directly, but rather through their implied distance matrices (although see Jothi et al. 2005 for a tree-based approach). Correlated values between two matrices reflect similar evolutionary trees, which can indicate coevolution, although as with amino acid covariation, the difficulty arises in separating the background signal of functional coevolution from the species phylogeny (Pazos et al. 2005; Sato et al. 2005). Recently, Juan et al. (2008) extended the mirror tree approach for pairwise observations toward a complete cellular “coevolutionary context.” This was achieved by considering not only the individual correlations between pairs of proteins in the E. coli proteome, but also the secondary correlations with the entire set of proteins (a technique previously described by Sato et al. 2006). By using this coevolutionary context information, they predicted the interactome of E. coli with a degree of accuracy and coverage comparable with that of the high-throughput experimental techniques.The mirror tree approach for detecting coevolution between proteins identifies protein partners by maximizing the correlation between the distance matrices describing a pair of phylogenies. Our own earlier program Codep (Tillier et al. 2006) finds the mutual information between all of the columns from two alignments to determine pairs of proteins that maximize this proxy for the coevolutionary signal. Tree-based and coevolutionary methods can detect interactions between genes and have worked particularly well for proteins from the smaller, compact bacterial genomes, for which many orthologous sequences are available for comparison, and where unique copies of protein-coding genes in genomes can be assessed for coevolution with other such unique genes. In situations in which there has been gene duplication, binding specificities of paralogous proteins may diverge after duplication, and there may be concerted duplications of the interaction partners. Particular members of a gene family within a genome may then have a particular affinity to select members of another protein family. However, the combinatorial problem raised by the presence of paralogs and the potential for multiple interacting partners make predicting protein interactions using coevolution in large genomes with many paralogs difficult. Another complication arises owing to gene loss, either from deletion or through high sequence divergence, which can make identifying orthologous copies and accurately measuring divergence also difficult. Both gene duplication and gene loss will give rise to protein families of different sizes, which remains outside the scope of these bioinformatics methods.Here we describe a novel approach for measuring the coevolution of protein families that address many of the drawbacks of previous methods. Our approach, that we call MatrixMatchMaker (MMM), finds the largest common submatrix that is compatible between the evolutionary distance matrices of two protein families. It can measure the coevolution between families of different size, including both paralogs and orthologs, and predict multiple coevolving partners. The method allows us to detect coevolution present in only some lineages, thus suggesting lineage-specific interactions. Here, we use MMM to analyze the human protein interaction network, to show that a large fraction of known protein interactions, as well as proteins belonging to the same biochemical pathway, do indeed coevolve.MMM, applied to the analysis of protein families, proposes several novel interactions. Although only a subset of coevolving proteins would be expected to interact, and vice versa, the predictions afforded by the use of the present analysis provide a highly enriched set of hypotheses that can more efficiently direct laboratory validation. The human protein coevolution network is large, but can be focused to cover a more restrictive scope. Where proteins are involved in a tight coevolutionary relationship implying a strict interdependency, we may find some of the more fundamental interactions, core to the system in which these proteins are engaged.     

The phospholipase A2 inhibitor methyl indoxam suppresses diet-induced obesity and glucose intolerance in mice

Background and purpose:

Previous results have shown that mice lacking in the group 1B phospholipase A₂ (Pla2g1b) are resistant to obesity and diabetes induced by feeding a diabetogenic high-fat/high-carbohydrate diet. This study examined the potential of using the Pla2g1b inhibitor methyl indoxam as therapy to suppress diet-induced obesity and diabetes.

Experimental approach:

Male C57BL/6 mice were fed the diabetogenic diet with or without methyl indoxam supplementation. Body weight gain, fasting plasma glucose levels, glucose tolerance and postprandial lysophospholipid absorption were compared.

Key results:

Wild-type C57BL/6 mice fed the diabetogenic diet without Pla2g1b inhibitor showed 31 and 69% body weight gain after 4 and 10 weeks respectively. These animals also showed elevated plasma glucose levels and were glucose intolerant. In contrast, C57BL/6 mice fed the diabetogenic diet with 90 mg·kg⁻¹ of methyl indoxam gained only 5% body weight after 10 weeks. These animals were also euglycaemic and displayed normal glucose excursion rates in glucose tolerance test. Methyl indoxam suppression of diet-induced body weight gain and glucose intolerance was correlated with the inhibition of Pla2g1b-mediated postprandial lysophospholipid absorption.

Conclusions and implications:

These results show that oral supplementation of a diabetogenic diet with the Pla2g1b inhibitor methyl indoxam effectively suppresses diet-induced obesity and diabetes in mice. This suggests that Pla2g1b inhibition may be a potentially effective oral therapeutic option for treatment of obesity and diabetes.     

Correlates of suicidal ideation among HIV-positive persons

OBJECTIVES: The present investigation sought to determine the extent to which demographic characteristics, illness-related burdens, alcohol and other substance use, and psychosocial factors are independently associated with suicidal ideation in HIV-positive individuals. DESIGN: HIV-positive individuals in four US cities (San Francisco, Los Angeles, Milwaukee, and New York City) were screened between July 2000 and January 2002 for recruitment into a randomized behavioral prevention trial. Utilizing data from this screening visit, rates and correlates of suicidal ideation were examined in a diverse sample of 2909 HIV-positive individuals. METHODS: Using binary logistic regression study sites, demographic characteristics, illness-related burdens, alcohol and substance use, and psychosocial factors were entered as predictors of suicidal ideation. This cross-sectional model thus examined the independent effects of each factor. RESULTS: Approximately one-fifth (19%) of participants reported thoughts of suicide in the past week. We observed that participants who were not heterosexual, rated HIV-related symptoms and medication side effects as more severe, reported regular marijuana use, and described elevated affective symptoms of depression were those who were more likely to report suicidal ideation. Conversely, participants who identified as Hispanic/Latino, individuals in a primary romantic relationship, and those who reported greater self-efficacy for coping were less likely to report suicidal ideation. CONCLUSION: Suicidal ideation among HIV-positive individuals is relatively common and is associated with multiple factors. These independent correlates may assist with identifying HIV-positive individuals who are at increased risk of suicidal ideation so that they may be assessed regularly and referred for psychological treatment when appropriate.     

Insulin-like growth factor binding protein 7 exhibits tumor suppressive and vessel stabilization properties in U87MG and T98G glioblastoma cell lines

Insulin-like growth factor binding protein 7 (IGFBP7) is downregulated in several solid cancers. IGFBP7 has been proposed to act as a tumor suppressor gene through mechanisms involving senescence and apoptotic pathways. The tumor suppressor effect of IGFBP7 in glioblastoma multiforme (GBM) was examined in this study using two human GBM cell lines, U87MG and T98G. Exogenously applied IGFBP7 (20 and 100 nM) significantly reduced U87MG (~70 and ~75%, respectively) and T98G (~37 and ~50%, respectively) cell growth in soft agar. IGFBP7 stimulated senescence-associated β-galactosidase in both U87MG and T98G cells without stimulating apoptosis (annexin V and propidium iodide staining, expression of SMARCB1 or BNIP3L and caspase cleavage) or affecting phosphorylation of p44/42 MAPK. The inhibitory effect of IGFBP7 on U87MG cell growth was further assessed in vivo using U87MG cells grafted on the chick chorioallantoic membrane. In this model, U87MG cells formed solid and highly vascularized tumors that were reduced in size (~40%) when treated with 500 nM IGFBP7 compared with control tumors. Vessels in IGFBP7-treated tumors were clustered, unevenly distributed and associated with higher number of α-SMA positive cells compared with those in untreated tumors. IGFBP7 induced both aortic smooth muscle cell (AoSMC) chemoattraction and endothelial cell (EC) transdifferentiation into a SM-like cell phenotype. U87MG conditioned media-induced IGFBP7 expression in ECs was significantly inhibited by the cross-talk/interaction with SMCs. This study indicates that IGFBP7 suppresses U87MG tumor cell growth, induces cell senescence and participates in tumor vessel stabilization by promoting SMC/pericyte recruitment and differentiation.     

Autologous bone marrow transplantation in acute myelogenous leukemia: in vitro treatment with myeloid cell-specific monoclonal antibodies

Second or third chemotherapy-induced remissions in acute myelogenous leukemia (AML) are limited by early relapse of the leukemia. We developed monoclonal antibodies (MoAbs) that are cytotoxic to myeloid leukemia cells to treat bone marrow from these patients ex vivo for autologous transplantation. In this pilot study, bone marrow was harvested from ten patients with AML in remission, treated with one or two complement-fixing MoAbs, PM-81 and AML-2-23, which react with myeloid differentiation antigens, incubated with rabbit complement, and cryopreserved. These MoAbs were chosen because they have broad reactivity with AML cells but not with pluripotent progenitor cells. At the time of transplant, 6 patients were in second complete remission, 1 each was in third complete or partial remission, and 2 were in early first relapse. The patients were treated with cyclophosphamide (60 mg/kg a day for 2 days) and total body irradiation (200 cGy twice a day for 3 days) and given infusions of MoAb-treated bone marrow. Full bone marrow reconstitution was observed in eight patients; two patients did not recover platelets. Seven of the ten patients are surviving and disease-free at 21.0, 15.0, 13.0, 10.0, 6.0, 3.0, and 2.0 months posttransplant. Treating bone marrow with MoAbs to myeloid differentiation antigens does not interfere with pluripotential stem cell engraftment. Longer follow-up and a controlled study are necessary to prove that the apparent efficacy of this therapeutic approach in some patients is attributable to MoAb-mediated killing of leukemia cells.     

Marrow transplantation with or without donor buffy coat cells for 65 transfused aplastic anemia patients

Sixty-five multiply transfused patients with severe aplastic anemia were given cyclophosphamide followed by grafts anemia were given cyclophosphamide followed by grafts from HLA-identical siblings. The effect of the administration of viable donor buffy coat cells following the marrow inoculum was evaluated with regard to graft rejection and survival. Results in 43 patients so treated are presented along with those in 22 concurrent patients given marrow alone. Most patients given buffy coat had positive in vitro tests of sensitization indicating a high risk for graft rejection, while all but one of the patients given marrow alone had negative tests. Thirty of the 43 (70%) patients given marrow and buffy coat are alive between 10 and 61 mo (median 36) after grafting; 4 died after graft rejection and 6 with acute or chronic graft-versus-host disease (GVHD). Eleven of the 22 (50%) patients given marrow alone are alive between 29 and 65 mo (median 52); 7 died after graft rejection and 3 with GVHD. The addition of buffy coat cell infusions to the marrow inoculum reduced the risk of rejection and increased survival in the currently reported transfused patients when compared to patients grafted before 1976. However, there was an increased risk of chronic GVHD. Recipients of marrow from female donors survived slightly better (73%) than recipients of male marrow (58%).     

Effect of hepatic dysfunction on oral cyclosporin pharmacokinetics in marrow transplant patients

The effect of hepatic dysfunction, defined as abnormal serum bilirubin level, on oral cyclosporin (CSP) pharmacokinetics was examined in 28 marrow transplant patients who received CSP for prophylaxis of graft-v- host disease. Serum CSP concentrations were measured by radioimmunoassay. Forty-one concentration-time courses were studied, divided among patients with no (less than 1.2 mg/dL), mild (1.2 to 2.0 mg/dL), and moderate (2.0 to 5.0 mg/dL) hepatic dysfunction. CSP elimination, as determined by elimination rate constant and clearance, was delayed in patients with moderate hepatic dysfunction compared to those with no hepatic dysfunction (P less than .05). The volume of distribution, lag time for absorption, maximum serum concentration, and time at which the maximum concentration was achieved was not affected by hepatic function. These data indicate that patients with moderate hepatic dysfunction have delayed CSP or CSP metabolite elimination and may be at higher risk for developing CSP-related toxicity.     

Marrow transplantation for chronic myeloid leukemia: a randomized study comparing cyclophosphamide and total body irradiation with busulfan and cyclophosphamide

A prospective randomized study was conducted comparing two conditioning regimens for the treatment of patients with chronic myeloid leukemia in chronic phase by marrow transplantation from HLA identical siblings. Sixty-nine patients received 60 mg/kg of cyclophosphamide on each of 2 successive days followed by 6 fractions of total body irradiation each of 2.0 Gy (CY-TBI), and 73 patients received 16 mg/kg of busulfan delivered over 4 days followed by 60 mg/kg CY on each of 2 successive days (BU-CY). There was no significant difference between the CY-TBI and the BU-CY groups in the 3-year probabilities of survival (0.80 for both), relapse (0.13 for both), or event-free survival (CY-TBI, 0.68; BU-CY, 0.71) or in speed of engraftment or incidence of venocclusive disease of the liver. The 4-year probabilities of survival and event- free survival for patients transplanted within 1 year of diagnosis were 0.86 and 0.72, respectively, for each group. Significantly more patients in the CY-TBI group experienced major creatinine elevations. There was significantly more acute graft-versus-host disease in the CY- TBI group. Fever days, positive blood cultures, hospitalizations, and inpatient hospital days were significantly more common in the CY-TBI group than in the BU-CY group. In conclusion, the BU-CY regimen was better tolerated than, and associated with survival and relapse probabilities that compare favorably with, the CY-TBI regimen.     

Isothermal microcalorimetry: an analytical technique for assessing the dynamic chemical stability of UHMWPE

In this work, isothermal microcalorimetry (IMC) was utilized to measure the exothermic heat flow from specimens of ultra-high-molecular-weight polyethylene (UHMWPE), that had been sterilized by various standard methods, under simulated shelf storage (air at 25 degrees C, 30% r.h.) and simulated implantation (phosphate buffered saline, PBS, at 37 degrees C) conditions. Gamma-radiation sterilized UHMWPE yielded initial heat flow rates approximately 7-10 times higher in simulated shelf storage and 2-3 times higher in simulated implantation (even after 1 month in PBS) than specimens which were unsterilized or sterilized using either ethylene oxide gas (ETO) or gas plasma (GP). These results show that gamma sterilization of UHMWPE produces many more unstable bonds in the polymer than is the case when ETO or GP is used, and that the net exothermic physico-chemical change proceeds steadily in a diffusion-limited manner in air or saline. In addition, gamma sterilization in nitrogen rather than in air did not prevent the creation of unstable bonds, but did defer physico-chemical change until the UHMWPE was exposed to oxygen. These results demonstrate the usefulness of IMC as a viable method for studying the stability of polymeric implant materials.     

Discrimination between myocardial infarct and ventricular tachycardia patients using magnetocardiographic trajectory plots and iso-integral maps.

Magnetocardiograms were recorded from 30 normal (N) subjects, 15 myocardial infarct (MI) patients, and 15 ventricular tachycardia (VT) patients. Discrimination between the groups was affected by iso-integral magnetic field mapping (MFM) and trajectory plotting of MFM extrema. Iso-integral MFM for the QRST, QRS, and ST-T intervals was created for each test group member. A polarity score, based on the number of extrema features present, was assigned to each iso-integral MFM. Differences in group mean integral of QRST map polarity scores were significant (p less than 0.05) between MI and N, between VT and N (p less than 0.005), and between MI and VT (p less than 0.05) subjects. integral of ST-T map polarity scores were significantly (p less than 0.0001) different between VT and N and between MI and VT (p less than 0.001) subjects. Discrimination between MI and VT patients, based on polarity score difference, was 56% accurate using integral of QRS maps and 73% accurate using integral of ST-T maps. For each subject, time-normalized MFM was used to construct trajectory plots of the maxima and minima in the QRS and ST-T intervals. Discrimination between MI and VT patients was based upon intergroup differences in fragmented trajectory plots. When the number of discrete trajectories and/or the total number (F) of trajectory points at which discrete trajectories coexist were considered, QRSmin trajectory plots were significantly (p less than 0.05) different for VT and N, but not for MI and N subjects. The significant (p less than 0.05) difference between MI and VT trajectory plots enabled 76% accuracy for MI and VT identification. ST-Tmax trajectory plots show significantly (p less than 0.0001) higher F values for VT patients facilitating accurate (87%) discrimination between MI and VT patients. These results suggest that the abnormalities of repolarization processes, displayed by MFM as multipolar integral of ST-T maps and/or as fragmented trajectory plots of ST-T extrema, may be useful indicators of the arrhythmia substrate/processes that characterize VT and vulnerable MI patients.