MELLODDY: Cross-pharma Federated Learning at Unprecedented Scale Unlocks Benefits in QSAR without Compromising Proprietary Information

Federated multipartner machine learning has been touted as an appealing and efficient method to increase the effective training data volume and thereby the predictivity of models, particularly when the generation of training data is resource-intensive. In the landmark MELLODDY project, indeed, each of ten pharmaceutical companies realized aggregated improvements on its own classification or regression models through federated learning. To this end, they leveraged a novel implementation extending multitask learning across partners, on a platform audited for privacy and security. The experiments involved an unprecedented cross-pharma data set of 2.6+ billion confidential experimental activity data points, documenting 21+ million physical small molecules and 40+ thousand assays in on-target and secondary pharmacodynamics and pharmacokinetics. Appropriate complementary metrics were developed to evaluate the predictive performance in the federated setting. In addition to predictive performance increases in labeled space, the results point toward an extended applicability domain in federated learning. Increases in collective training data volume, including by means of auxiliary data resulting from single concentration high-throughput and imaging assays, continued to boost predictive performance, albeit with a saturating return. Markedly higher improvements were observed for the pharmacokinetics and safety panel assay-based task subsets.

Chemistry-Based Modeling on Phenotype-Based Drug-Induced Liver Injury Annotation: From Public to Proprietary Data

Drug-induced liver injury (DILI) is an important safety concern and a major reason to remove a drug from the market. Advancements in recent machine learning methods have led to a wide range of in silico models for DILI predictive methods based on molecule chemical structures (fingerprints). Existing publicly available DILI data sets used for model building are based on the interpretation of drug labels or patient case reports, resulting in a typical binary clinical DILI annotation. We developed a novel phenotype-based annotation to process hepatotoxicity information extracted from repeated dose in vivo preclinical toxicology studies using INHAND annotation to provide a more informative and reliable data set for machine learning algorithms. This work resulted in a data set of 430 unique compounds covering diverse liver pathology findings which were utilized to develop multiple DILI prediction models trained on the publicly available data (TG-GATEs) using the compound's fingerprint. We demonstrate that the TG-GATEs compounds DILI labels can be predicted well and how the differences between TG-GATEs and the external test compounds (Johnson & Johnson) impact the model generalization performance.

Leveraging Cell Painting Images to Expand the Applicability Domain and Actively Improve Deep Learning Quantitative Structure-Activity Relationship Models

The search for chemical hit material is a lengthy and increasingly expensive drug discovery process. To improve it, ligand-based quantitative structure-activity relationship models have been broadly applied to optimize primary and secondary compound properties. Although these models can be deployed as early as the stage of molecule design, they have a limited applicability domain─if the structures of interest differ substantially from the chemical space on which the model was trained, a reliable prediction will not be possible. Image-informed ligand-based models partly solve this shortcoming by focusing on the phenotype of a cell caused by small molecules, rather than on their structure. While this enables chemical diversity expansion, it limits the application to compounds physically available and imaged. Here, we employ an active learning approach to capitalize on both of these methods' strengths and boost the model performance of a mitochondrial toxicity assay (Glu/Gal). Specifically, we used a phenotypic Cell Painting screen to build a chemistry-independent model and adopted the results as the main factor in selecting compounds for experimental testing. With the additional Glu/Gal annotation for selected compounds we were able to dramatically improve the chemistry-informed ligand-based model with respect to the increased recognition of compounds from a 10% broader chemical space.

Atom-to-atom Mapping: A Benchmarking Study of Popular Mapping Algorithms and Consensus Strategies

In this paper, we compare the most popular Atom-to-Atom Mapping (AAM) tools: ChemAxon,^[1] Indigo,^[2] RDTool,^[3] NameRXN (NextMove),^[4] and RXNMapper^[5] which implement different AAM algorithms. An open-source RDTool program was optimized, and its modified version ("new RDTool") was considered together with several consensus mapping strategies. The Condensed Graph of Reaction approach was used to calculate chemical distances and develop the "AAM fixer" algorithm for an automatized correction of erroneous mapping. The benchmarking calculations were performed on a Golden dataset containing 1851 manually mapped and curated reactions. The best performing RXNMapper program together with the AMM Fixer was applied to map the USPTO database. The Golden dataset, mapped USPTO and optimized RDTool are available in the GitHub repository https://github.com/Laboratoire-de-Chemoinformatique.

Reaction Data Curation I: Chemical Structures and Transformations Standardization

The quality of experimental data for chemical reactions is a critical consideration for any reaction-driven study. However, the curation of reaction data has not been extensively discussed in the literature so far. Here, we suggest a 4 steps protocol that includes the curation of individual structures (reactants and products), chemical transformations, reaction conditions and endpoints. Its implementation in Python3 using CGRTools toolkit has been used to clean three popular reaction databases Reaxys, USPTO and Pistachio. The curated USPTO database is available in the GitHub repository (Laboratoire-de-Chemoinformatique/Reaction_Data_Cleaning).

Image-based profiling for drug discovery: due for a machine-learning upgrade?

Image-based profiling is a maturing strategy by which the rich information present in biological images is reduced to a multidimensional profile, a collection of extracted image-based features. These profiles can be mined for relevant patterns, revealing unexpected biological activity that is useful for many steps in the drug discovery process. Such applications include identifying disease-associated screenable phenotypes, understanding disease mechanisms and predicting a drug's activity, toxicity or mechanism of action. Several of these applications have been recently validated and have moved into production mode within academia and the pharmaceutical industry. Some of these have yielded disappointing results in practice but are now of renewed interest due to improved machine-learning strategies that better leverage image-based information. Although challenges remain, novel computational technologies such as deep learning and single-cell methods that better capture the biological information in images hold promise for accelerating drug discovery.

Tales of 1,008 small molecules: phenomic profiling through live-cell imaging in a panel of reporter cell lines

Phenomic profiles are high-dimensional sets of readouts that can comprehensively capture the biological impact of chemical and genetic perturbations in cellular assay systems. Phenomic profiling of compound libraries can be used for compound target identification or mechanism of action (MoA) prediction and other applications in drug discovery. To devise an economical set of phenomic profiling assays, we assembled a library of 1,008 approved drugs and well-characterized tool compounds manually annotated to 218 unique MoAs, and we profiled each compound at four concentrations in live-cell, high-content imaging screens against a panel of 15 reporter cell lines, which expressed a diverse set of fluorescent organelle and pathway markers in three distinct cell lineages. For 41 of 83 testable MoAs, phenomic profiles accurately ranked the reference compounds (AUC-ROC ≥ 0.9). MoAs could be better resolved by screening compounds at multiple concentrations than by including replicates at a single concentration. Screening additional cell lineages and fluorescent markers increased the number of distinguishable MoAs but this effect quickly plateaued. There remains a substantial number of MoAs that were hard to distinguish from others under the current study's conditions. We discuss ways to close this gap, which will inform the design of future phenomic profiling efforts.

Industry-scale application and evaluation of deep learning for drug target prediction

Artificial intelligence (AI) is undergoing a revolution thanks to the breakthroughs of machine learning algorithms in computer vision, speech recognition, natural language processing and generative modelling. Recent works on publicly available pharmaceutical data showed that AI methods are highly promising for Drug Target prediction. However, the quality of public data might be different than that of industry data due to different labs reporting measurements, different measurement techniques, fewer samples and less diverse and specialized assays. As part of a European funded project (ExCAPE), that brought together expertise from pharmaceutical industry, machine learning, and high-performance computing, we investigated how well machine learning models obtained from public data can be transferred to internal pharmaceutical industry data. Our results show that machine learning models trained on public data can indeed maintain their predictive power to a large degree when applied to industry data. Moreover, we observed that deep learning derived machine learning models outperformed comparable models, which were trained by other machine learning algorithms, when applied to internal pharmaceutical company datasets. To our knowledge, this is the first large-scale study evaluating the potential of machine learning and especially deep learning directly at the level of industry-scale settings and moreover investigating the transferability of publicly learned target prediction models towards industrial bioactivity prediction pipelines.

High-Parameter Mass Cytometry Evaluation of Relapsed/Refractory Multiple Myeloma Patients Treated with Daratumumab Demonstrates Immune Modulation as a Novel Mechanism of Action

Daratumumab is a CD38‐targeted human monoclonal antibody with direct anti‐myeloma cell mechanisms of action. Flow cytometry in relapsed and/or refractory multiple myeloma (RRMM) patients treated with daratumumab revealed cytotoxic T‐cell expansion and reduction of immune‐suppressive populations, suggesting immune modulation as an additional mechanism of action. Here, we performed an in‐depth analysis of the effects of daratumumab on immune‐cell subpopulations using high‐dimensional mass cytometry. Whole‐blood and bone‐marrow baseline and on‐treatment samples from RRMM patients who participated in daratumumab monotherapy studies (SIRIUS and GEN501) were evaluated with high‐throughput immunophenotyping. In daratumumab‐treated patients, the intensity of CD38 marker expression decreased on many immune cells in SIRIUS whole‐blood samples. Natural killer (NK) cells were depleted with daratumumab, with remaining NK cells showing increased CD69 and CD127, decreased CD45RA, and trends for increased CD25, CD27, and CD137 and decreased granzyme B. Immune‐suppressive population depletion paralleled previous findings, and a newly observed reduction in CD38⁺ basophils was seen in patients who received monotherapy. After 2 months of daratumumab, the T‐cell population in whole‐blood samples from responders shifted to a CD8 prevalence with higher granzyme B positivity (P = 0.017), suggesting increased killing capacity and supporting monotherapy‐induced CD8⁺ T‐cell activation. High‐throughput cytometry immune profiling confirms and builds upon previous flow cytometry data, including comparable CD38 marker intensity on plasma cells, NK cells, monocytes, and B/T cells. Interestingly, a shift toward cytolytic granzyme B⁺ T cells was also observed and supports adaptive responses in patients that may contribute to depth of response. © 2018 The Authors. Cytometry Part A published by Wiley Periodicals, Inc. on behalf of International Society for Advancement of Cytometry.

Large-scale comparison of machine learning methods for drug target prediction on ChEMBL

Deep learning is currently the most successful machine learning technique in a wide range of application areas and has recently been applied successfully in drug discovery research to predict potential drug targets and to screen for active molecules. However, due to (1) the lack of large-scale studies, (2) the compound series bias that is characteristic of drug discovery datasets and (3) the hyperparameter selection bias that comes with the high number of potential deep learning architectures, it remains unclear whether deep learning can indeed outperform existing computational methods in drug discovery tasks. We therefore assessed the performance of several deep learning methods on a large-scale drug discovery dataset and compared the results with those of other machine learning and target prediction methods. To avoid potential biases from hyperparameter selection or compound series, we used a nested cluster-cross-validation strategy. We found (1) that deep learning methods significantly outperform all competing methods and (2) that the predictive performance of deep learning is in many cases comparable to that of tests performed in wet labs (i.e., in vitro assays).

Repurposing High-Throughput Image Assays Enables Biological Activity Prediction for Drug Discovery

In both academia and the pharmaceutical industry, large-scale assays for drug discovery are expensive and often impractical, particularly for the increasingly important physiologically relevant model systems that require primary cells, organoids, whole organisms, or expensive or rare reagents. We hypothesized that data from a single high-throughput imaging assay can be repurposed to predict the biological activity of compounds in other assays, even those targeting alternate pathways or biological processes. Indeed, quantitative information extracted from a three-channel microscopy-based screen for glucocorticoid receptor translocation was able to predict assay-specific biological activity in two ongoing drug discovery projects. In these projects, repurposing increased hit rates by 50- to 250-fold over that of the initial project assays while increasing the chemical structure diversity of the hits. Our results suggest that data from high-content screens are a rich source of information that can be used to predict and replace customized biological assays.

Identification of cis-regulatory mutations generating de novo edges in personalized cancer gene regulatory networks

The identification of functional non-coding mutations is a key challenge in the field of genomics. Here we introduce μ-cisTarget to filter, annotate and prioritize cis-regulatory mutations based on their putative effect on the underlying "personal" gene regulatory network. We validated μ-cisTarget by re-analyzing the TAL1 and LMO1 enhancer mutations in T-ALL, and the TERT promoter mutation in melanoma. Next, we re-sequenced the full genomes of ten cancer cell lines and used matched transcriptome data and motif discovery to identify master regulators with de novo binding sites that result in the up-regulation of nearby oncogenic drivers. μ-cisTarget is available from http://mucistarget.aertslab.org .

ExCAPE-DB: an integrated large scale dataset facilitating Big Data analysis in chemogenomics

Chemogenomics data generally refers to the activity data of chemical compounds on an array of protein targets and represents an important source of information for building in silico target prediction models. The increasing volume of chemogenomics data offers exciting opportunities to build models based on Big Data. Preparing a high quality data set is a vital step in realizing this goal and this work aims to compile such a comprehensive chemogenomics dataset. This dataset comprises over 70 million SAR data points from publicly available databases (PubChem and ChEMBL) including structure, target information and activity annotations. Our aspiration is to create a useful chemogenomics resource reflecting industry-scale data not only for building predictive models of in silico polypharmacology and off-target effects but also for the validation of cheminformatics approaches in general.

Virology analyses of HCV isolates from genotype 1-infected patients treated with simeprevir plus peginterferon/ribavirin in Phase IIb/III studies

BACKGROUND & AIMS

Simeprevir is an oral hepatitis C virus (HCV) NS3/4A protease inhibitor approved for treatment of chronic HCV infection. Baseline NS3 polymorphisms in all patients and emerging mutations in patients who failed to achieve sustained virologic response (SVR) with simeprevir plus peginterferon/ribavirin (PR) in Phase IIb/III studies are described.

METHODS

Baseline sequencing data were available for 2007 genotype 1 (GT1)-infected patients. Post-baseline data were available for 197/245 simeprevir-treated patients who did not achieve SVR. In vitro simeprevir susceptibility was assessed in a transient replicon assay as site-directed mutants or in chimeric replicons with patient-derived NS3 protease sequences.

RESULTS

Baseline NS3 polymorphisms at positions associated with reduced in vitro susceptibility to simeprevir (43, 80, 122, 155, 156, and/or 168; EC50 fold change >2.0) were uncommon (1.3% [26/2007]), with the exception of Q80K, which confers ∼10-fold reduction in simeprevir activity in vitro (13.7% [274/2007]; GT1a 29.5% [269/911], GT1b 0.5% [5/1096]). Baseline Q80K had minor effect on initial response to simeprevir/PR, but resulted in lower SVR rates. Overall, 91.4% of simeprevir-treated patients [180/197] without SVR had emerging mutations at NS3 positions 80, 122, 155, and/or 168 at failure (mainly R155K in GT1a with and without Q80K, and D168V in GT1b), conferring high-level resistance in vitro (EC50 fold change >50). Emerging mutations were no longer detectable by population sequencing at study end in 50% [90/180] of patients (median follow-up 28.4weeks).

CONCLUSIONS

Simeprevir treatment failure was usually associated with emerging high-level resistance mutations, which became undetectable over time in half of the patients.

Analysis of genotype 2 and 3 hepatitis C virus variants in patients treated with telaprevir demonstrates a consistent resistance profile across genotypes

Study C209 evaluated the activity of telaprevir in treatment-naïve patients with genotypes 2 or 3 (G2, G3) hepatitis C virus (HCV) infection. Telaprevir monotherapy showed potent activity against HCV G2, but limited activity against G3. This analysis was performed to characterize HCV viral variants emerging during telaprevir-based treatment of G2/G3 HCV-infected patients. Patients were randomized to receive 2 weeks of treatment with telaprevir (telaprevir monotherapy), telaprevir plus peginterferon alfa-2a and ribavirin (triple therapy), or placebo plus peginterferon alfa-2a and ribavirin (control), followed by 22-24 weeks of peginterferon/ribavirin alone. Viral breakthrough was defined as an increase >1 log10 in HCV RNA from nadir, or HCV RNA >100 IU/mL in patients previously reaching <25 IU/mL. Twenty-three patients (47%) had G2 and 26 (53%) had G3 HCV. Viral breakthrough occurred during the initial 2-week treatment phase in six G2 patients (66.7%; subtypes 2, 2a and 2b) and three G3 patients (37.5%; all subtype 3a), all in the telaprevir monotherapy arm. Four breakthrough patients (three G2, one G3) subsequently achieved sustained virologic response (SVR). In all patients with breakthrough and available sequence data, mutations associated with reduced susceptibility to telaprevir in genotype 1 (G1) HCV were observed. No novel G2/G3-specific mutations were associated with telaprevir resistance. The telaprevir resistance profile appeared consistent across HCV genotypes 1, 2 and 3. Although viral breakthrough with resistance occurred in patients receiving telaprevir monotherapy, half of these patients achieved an SVR upon addition of peginterferon/ribavirin highlighting the importance of combination therapy.

High-resolution profiling of the LEDGF/p75 chromatin interaction in the ENCODE region

Lens epithelium-derived growth factor/p75 (LEDGF/p75) is a transcriptional coactivator involved in stress response, autoimmune disease, cancer and HIV replication. A fusion between the nuclear pore protein NUP98 and LEDGF/p75 has been found in human acute and chronic myeloid leukemia and association of LEDGF/p75 with mixed-lineage leukemia (MLL)/menin is critical for leukemic transformation. During lentiviral replication, LEDGF/p75 tethers the pre-integration complex to the host chromatin resulting in a bias of integration into active transcription units (TUs). The consensus function of LEDGF/p75 is tethering of cargos to chromatin. In this regard, we determined the LEDGF/p75 chromatin binding profile. To this purpose, we used DamID technology and focused on the highly annotated ENCODE (Encyclopedia of DNA Elements) regions. LEDGF/p75 primarily binds downstream of the transcription start site of active TUs in agreement with the enrichment of HIV-1 integration sites at these locations. We show that LEDGF/p75 binding is not restricted to stress response elements in the genome, and correlation analysis with more than 200 genomic features revealed an association with active chromatin markers, such as H3 and H4 acetylation, H3K4 monomethylation and RNA polymerase II binding. Interestingly, some associations did not correlate with HIV-1 integration indicating that not all LEDGF/p75 complexes on the chromosome are amenable to HIV-1 integration.

Docking motif-guided mapping of the interactome of protein phosphatase-1

The ubiquitous protein Ser/Thr phosphatase-1 (PP1) interacts with dozens of regulatory proteins that are structurally unrelated. However, most of them share a short, degenerate "RVxF"-type docking motif. Using a broad in silico screening based on a stringent definition of the RVxF motif, in combination with a multistep biochemical validation procedure, we have identified 78 novel mammalian PP1 interactors. A global analysis of the validated RVxF-based PP1 interactome not only provided insights into the conserved features of the RVxF motif but also led to the discovery of additional common PP1 binding elements, described as the "SILK" and "MyPhoNE" motifs. In addition to the doubling of the known mammalian PP1 interactome, our data contribute to the design of PP1 interaction networks. Notably, an interaction network linking PP1 interactors discloses a pleiotropic role of PP1 in cell polarity.

The transcriptional repressor NIPP1 is an essential player in EZH2-mediated gene silencing

EZH2 is a Polycomb group (PcG) protein that promotes the late-stage development of cancer by silencing a specific set of genes, at least in part through trimethylation of associated histone H3 on Lys 27 (H3K27). Nuclear inhibitor of protein phosphatase-1 (NIPP1) is a ubiquitously expressed transcriptional repressor that has binding sites for the EZH2 interactor EED. Here, we examine the contribution of NIPP1 to EZH2-mediated gene silencing. Studies on NIPP1-deficient cells disclose a widespread and essential role of NIPP1 in the trimethylation of H3K27 by EZH2, not only in the onset of this trimethylation during embryonic development, but also in the maintenance of this repressive mark in proliferating cells. Consistent with this notion, EZH2 and NIPP1 silence a common set of genes, as revealed by gene-expression profiling, and NIPP1 is associated with established Polycomb target genes and with genomic regions that are enriched in Polycomb targets. Furthermore, most NIPP1 target genes are trimethylated on H3K27 and the knockdown of either NIPP1 or EZH2 is often associated with a loss of this modification. Our data reveal that NIPP1 is required for the global trimethylation of H3K27 and is implicated in gene silencing by EZH2.

An Essential Oligomannosidic Glycan Chain in the Catalytic Domain of Autotaxin, a Secreted Lysophospholipase-D

Autotaxin/NPP2, a secreted lysophospholipase-D, promotes cell proliferation, survival, and motility by generating the signaling molecule lysophosphatidic acid. Here we show that ectonucleotide pyrophosphatase/phosphodiesterase 2 (NPP2) is N-glycosylated on Asn-53, Asn-410, and Asn-524. Mutagenesis and deglycosylation experiments revealed that only the glycosylation of Asn-524 is essential for the expression of the catalytic and motility-stimulating activities of NPP2. The N-glycan on Asn-524 was identified as Man8/9GlcNAc2, which is rarely present on mature eukaryotic glycoproteins. Additional studies show that this Asn-524-linked glycan is not accessible to alpha-1,2-mannosidase, suggesting that its non-reducing termini are buried inside the folded protein. Consistent with a structural role for the Asn-524-linked glycan, only the mutation of Asn-524 augmented the sensitivity of NPP2 to proteolysis and increased its mobility during Blue Native PAGE. Asn-524 is phylogenetically conserved and maps to the catalytic domain of NPP2, but a structural model of this domain suggests that Asn-524 is remote from the catalytic site. Our study defines an essential role for the Asn-524-linked glycan chain of NPP2.

The translation initiation factor eIF2beta is an interactor of protein phosphatase-1

It is reasonably well understood how the initiation of translation is controlled by reversible phosphorylation of the eukaryotic translation initiation factors eIF2alpha, eIF2Bepsilon and eIF4E. Other initiation factors, including eIF2beta, are also established phosphoproteins but the physiological impact of their phosphorylation is not known. Using a sequence homology search we found that the central region of eIF2beta contains a putative PP1-(protein phosphatase-1) binding RVxF-motif. The predicted eIF2beta-PP1 interaction was confirmed by PP1 binding and co-immunoprecipitation assays on cell lysates as well as with the purified components. Site-directed mutagenesis showed that eIF2beta contains, in addition to an RVxF-motif, at least one other PP1-binding site in its C-terminal half. eIF2beta functioned as an inhibitor for the dephosphorylation of glycogen phosphorylase and Ser51 of eIF2alpha by PP1, but did not affect the dephosphorylation of Ser464 of eIF2Bepsilon by this phosphatase. Strikingly, eIF2beta emerged as an activator of its own dephosphorylation (Ser2, Ser67, Ser218) by associated PP1, since the substrate quality of eIF2beta was decreased by the mere mutation of its RVxF-motif. These results make eIF2beta an attractive candidate substrate for associated PP1 in vivo. The overexpression of wild-type eIF2beta or eIF2beta with a mutated RVxF-motif did not differentially affect the rate of translation, indicating that the binding of PP1 is not rate-limiting for translation under basal conditions.

Approaches to defining the ancestral eukaryotic protein complexome

In this paper, we integrate and summarize the currently available information on the ancestral eukaryotic protein complexome, which is defined as the set of protein complexes that extant eukaryotes inherited from their last common ancestor. From the literature, we compiled lists of complexes with three or more distinct protein components from well-studied eukaryotic model organisms. Combinatorial complexes of membrane-associated signalling proteins and specific transcription factors were disregarded. A stringent but sensitive novel orthology detection algorithm, complemented with manual sequence similarity searches and with published data on whole genome or segmental and tandem gene duplications, enabled us to map the vast majority of these complexes to a virtual primitive eukaryote termed Eukaryotic Virtual Ancestor (EVA). EVA is intended to resemble the last common eukaryotic ancestor and to emulate the biological common denominator of the major extent eukaryotic lineages at the molecular level. The dataset was then used for the functional and domain annotation of the ancestral eukaryotic complexome. Furthermore, we illustrate its usefulness for inferring complexes of poorly studied eukaryotes and for the recognition of highly divergent orthologs. We also discuss the evolution of the circa 1,400 complex-associated ancestral proteins. As about 90% of these proteins have been conserved in all thirteen studied free-living eukaryotes, the evolutionary reduction and loss of complexes seems minimal. Moreover, the available data suggest that, in general, the acquisition of stable complexes of novel design occurs too slowly to be a major contributor to evolutionary innovation. Finally, given the stability of the ancestral eukarotic complexome we propose its use in the formulation of the mathematical systems that aim to simulate biological processes. Our data suggest that these simplified formulations can apply to most free-living model eukaryotes.

A tighter RVxF motif makes a finer Sift

Most partners of protein phosphatase 1 rely on an instance of the so-called RVxF motif for interaction with the enzyme. In this issue of Chemistry & Biology, a stringent definition of the motif targeting high-affinity instances enabled Meiselbach and colleagues to recognize novel binding partners with high specificity .

Substrate specificity and activity regulation of protein kinase MELK

Maternal embryonic leucine zipper kinase (MELK) is a protein Ser/Thr kinase that has been implicated in stem cell renewal, cell cycle progression, and pre-mRNA splicing, but its substrates and regulation are not yet known. We show here that MELK has a rather broad substrate specificity and does not appear to require a specific sequence surrounding its (auto)phosphorylation sites. We have mapped no less than 16 autophosphorylation sites including serines, threonines, and a tyrosine residue and show that the phosphorylation of Thr167 and Ser171 is required for the activation of MELK. The expression of MELK activity also requires reducing agents such as dithiothreitol or reduced glutathione. Furthermore, we show that MELK is a Ca2+-binding protein and is inhibited by physiological Ca2+ concentrations. The smallest MELK fragment that was still catalytically active comprises the N-terminal catalytic domain and the flanking ubiquitin-associated domain. A C-terminal fragment of MELK functions as an autoinhibitory domain. Our data show that the activity of MELK is regulated in a complex manner and offer new perspectives for the further elucidation of its biological function.

Determinants of the nucleolar targeting of protein phosphatase-1

The ubiquitously expressed protein Ser/Thr phosphatase-1 isoforms PP1alpha, PP1beta and PP1gamma1 are dynamically targeted to distinct, but overlapping cellular compartments by associated proteins. Within the nucleus of HeLa cells, EGFP-tagged PP1gamma1 and PP1beta were predominantly targeted to the nucleoli, while PP1alpha showed a more diffuse distribution. Using PP1 chimaeras and point mutants we show here that a single N-terminal residue, i.e., Gln20 for PP1alpha, Arg19 for PP1beta and Arg20 for PP1gamma1 accounts for their distinct subnuclear distribution. Our data also suggest that the N-terminus of PP1beta and PP1gamma1 harbours an interaction site for one or more nucleolar interactors.

A novel glycogen-targeting subunit of protein phosphatase 1 that is regulated by insulin and shows differential tissue distribution in humans and rodents

Stimulation of glycogen-targeted protein phosphatase 1 (PP1) activity by insulin contributes to the dephosphorylation and activation of hepatic glycogen synthase (GS) leading to an increase in glycogen synthesis. The glycogen-targeting subunits of PP1, GL and R5/PTG, are downregulated in the livers of diabetic rodents and restored by insulin treatment. We show here that the mammalian gene PPP1R3E encodes a novel glycogen-targeting subunit of PP1 that is expressed in rodent liver. The phosphatase activity associated with R3E is slightly higher than that associated with R5/PTG and it is downregulated in streptozotocin-induced diabetes by 60-70% and restored by insulin treatment. Surprisingly, although mRNA for R3E is most highly expressed in rat liver and heart muscle, with only low levels in skeletal muscle, R3E mRNA is most abundant in human skeletal muscle and heart tissues with barely detectable levels in human liver. This species-specific difference in R3E mRNA expression has similarities to the high level of expression of GL mRNA in human but not rodent skeletal muscle. The observations imply that the mechanisms by which insulin regulates glycogen synthesis in liver and skeletal muscle are different in rodents and humans.

Geminate structures of African cassava mosaic virus

Two types of geminate structures were purified from African cassava mosaic geminivirus (ACMV)-infected Nicotiana benthamiana plants and analyzed by electron cryomicroscopy and image reconstruction. After cesium sulfate density gradient centrifugation, they were separated into lighter top (T) and heavier bottom (B) components. T particles comigrated with host proteins, whereas B particles were concentrated in a cesium density typical for complete virions. Both particles were composed of two incomplete icosahedra of 11 capsomers each, but T particles were slightly larger (diameter, 22.5 nm) and less dense in the interior than B particles (diameter, 21.5 nm). T particles were frequently associated with small globules of approximately 14 nm diameter of unknown origin. The overall structure of ACMV, a begomovirus transmitted by whiteflies, was similar to that of Maize streak virus (MSV), a mastrevirus transmitted by leafhoppers, although the vertices of the icosahedra were less pronounced. Models of ACMV coat proteins based on Satellite tobacco necrosis virus support the exposure of parts of the molecule essential for transmission specificity by whiteflies and provide possible structural explanations for the smaller protrusion of the ACMV capsid relative to MSV. The differences of ACMV and MSV virion shapes are discussed with reference to their different animal vectors.

Fast Fitting of Atomic Structures to Low-resolution Electron Density Maps by Surface Overlap Maximization

The complexities of X-ray crystallography and NMR spectroscopy for large protein complexes, and the comparative ease of approaches such as electron microscopy mean that low-resolution structures are often available long before their atomic resolution equivalents. To help bridge this gap in knowledge, we present 3SOM: an approach for finding the best fit of atomic resolution structures into lower-resolution density maps through surface overlap maximization. High-resolution templates (i.e. partial structures or models for multi-subunit complexes) and targets (lower-resolution maps) are initially represented as iso-surfaces. The latter are used first in a fast search for transformations that superimpose a significant portion of the target surface onto the template surface, which is quantified as surface overlap. The vast search space is reduced by considering key vectors that capture local surface information. The set of transformations with the highest surface overlap scores are then re-ranked by using more sophisticated scores including cross-correlation. We give a number of examples to illustrate the efficiency of the method and its restrictions. For targets for which partial complexes are available, the speed and performance of the method make it an attractive complement to existing methods, as many different hypotheses can be tested quickly on a single processor.

Structure-Based Assembly of Protein Complexes in Yeast

Images of entire cells are preceding atomic structures of the separate molecular machines that they contain. The resulting gap in knowledge can be partly bridged by protein-protein interactions, bioinformatics, and electron microscopy. Here we use interactions of known three-dimensional structure to model a large set of yeast complexes, which we also screen by electron microscopy. For 54 of 102 complexes, we obtain at least partial models of interacting subunits. For 29, including the exosome, the chaperonin containing TCP-1, a 3'-messenger RNA degradation complex, and RNA polymerase II, the process suggests atomic details not easily seen by homology, involving the combination of two or more known structures. We also consider interactions between complexes (cross-talk) and use these to construct a structure-based network of molecular machines in the cell.

Functional diversity of protein phosphatase-1, a cellular economizer and reset button

The protein serine/threonine phosphatase protein phosphatase-1 (PP1) is a ubiquitous eukaryotic enzyme that regulates a variety of cellular processes through the dephosphorylation of dozens of substrates. This multifunctionality of PP1 relies on its association with a host of function-specific targetting and substrate-specifying proteins. In this review we discuss how PP1 affects the biochemistry and physiology of eukaryotic cells. The picture of PP1 that emerges from this analysis is that of a "green" enzyme that promotes the rational use of energy, the recycling of protein factors, and a reversal of the cell to a basal and/or energy-conserving state. Thus PP1 promotes a shift to the more energy-efficient fuels when nutrients are abundant and stimulates the storage of energy in the form of glycogen. PP1 also enables the relaxation of actomyosin fibers, the return to basal patterns of protein synthesis, and the recycling of transcription and splicing factors. In addition, PP1 plays a key role in the recovery from stress but promotes apoptosis when cells are damaged beyond repair. Furthermore, PP1 downregulates ion pumps and transporters in various tissues and ion channels that are involved in the excitation of neurons. Finally, PP1 promotes the exit from mitosis and maintains cells in the G1 or G2 phases of the cell cycle.

The relationship between sequence and interaction divergence in proteins

There is currently a gap in knowledge between complexes of known three-dimensional structure and those known from other experimental methods such as affinity purifications or the two-hybrid system. This gap can sometimes be bridged by methods that extrapolate interaction information from one complex structure to homologues of the interacting proteins. To do this, it is important to know if and when proteins of the same type (e.g. family, superfamily or fold) interact in the same way. Here, we study interactions of known structure to address this question. We found all instances within the structural classification of proteins database of the same domain pairs interacting in different complexes, and then compared them with a simple measure (interaction RMSD). When plotted against sequence similarity we find that close homologues (30-40% or higher sequence identity) almost invariably interact the same way. Conversely, similarity only in fold (i.e. without additional evidence for a common ancestor) is only rarely associated with a similarity in interaction. The results suggest that there is a twilight zone of sequence similarity where it is not possible to say whether or not domains will interact similarly. We also discuss the rare instances of fold similarities interacting the same way, and those where obviously homologous proteins interact differently.

Degeneracy and function of the ubiquitous RVXF motif that mediates binding to protein phosphatase-1

Most interactors of protein phosphatase-1 (PP1) contain a variant of a so-called "RVXF" sequence that binds to a hydrophobic groove of the catalytic subunit. A combination of sequence alignments and site-directed mutagenesis has enabled us to further define the consensus sequence for this degenerate motif as [RK]-X(0-1)-[VI]-[P]-[FW], where X denotes any residue and [P] any residue except Pro. Naturally occurring RVXF sequences differ in their affinity for PP1, and we show by swapping experiments that this binding affinity is an important determinant of the inhibitory potency of the regulators NIPP1 and inhibitor-1. Also, inhibition by NIPP1-(143-224) was retained when the RVXF motif (plus the preceding Ser) was swapped for either of two unrelated PP1-binding sequences from human inhibitor-2, i.e. KGILK or RKLHY. Conversely, the KGILK motif of inhibitor-2 could be functionally replaced by the RVXF motif of NIPP1. Our data provide additional evidence for the view that the RVXF and KGILK motifs function as anchors for PP1 and thereby promote the interaction of secondary binding sites that determine the activity and substrate specificity of the enzyme.

Functional characterization of the non-catalytic ectodomains of the nucleotide pyrophosphatase/phosphodiesterase NPP1

The ubiquitous nucleotide pyrophosphatases/phosphodiesterases NPP1-3 consist of a short intracellular N-terminal domain, a single transmembrane domain and a large extracellular part, comprising two somatomedin-B-like domains, a catalytic domain and a poorly defined C-terminal domain. We show here that the C-terminal domain of NPP1-3 is structurally related to a family of DNA/RNA non-specific endonucleases. However, none of the residues that are essential for catalysis by the endonucleases are conserved in NPP1-NPP3, suggesting that the nuclease-like domain of NPP1-3 does not represent a second catalytic domain. Truncation analysis revealed that the nuclease-like domain of NPP1 is required for protein stability, for the targeting of NPP1 to the plasma membrane and for the expression of catalytic activity. We also demonstrate that 16 conserved cysteines in the somatomedin-B-like domains of NPP1, in concert with two flanking cysteines, mediate the dimerization of NPP1. The K173Q polymorphism of NPP1, which maps to the second somatomedin-B-like domain and has been associated with the aetiology of insulin resistance, did not affect the dimerization or catalytic activity of NPP1, and did not endow NPP1 with an affinity for the insulin receptor. Our data suggest that the non-catalytic ectodomains contribute to the subunit structure, stability and function of NPP1-3.

Binding of the concave surface of the Sds22 superhelix to the alpha 4/alpha 5/alpha 6-triangle of protein phosphatase-1

Functional studies of the protein phosphatase-1 (PP1) regulator Sds22 suggest that it is indirectly and/or directly involved in one of the most ancient functions of PP1, i.e. reversing phosphorylation by the Aurora-related protein kinases. We predict that the conserved portion of Sds22 folds into a curved superhelix and demonstrate that mutation to alanine of any of eight residues (Asp(148), Phe(170), Glu(192), Phe(214), Asp(280), Glu(300), Trp(302), or Tyr(327)) at the concave surface of this superhelix thwarts the interaction with PP1. Furthermore, we show that all mammalian isoforms of PP1 have the potential to bind Sds22. Interaction studies with truncated versions of PP1 and with chimeric proteins comprising fragments of PP1 and the yeast PP1-like protein phosphatase Ppz1 suggest that the site(s) required for the binding of Sds22 reside between residues 43 and 173 of PP1gamma(1). Within this region, a major interaction site was mapped to a triangular region delineated by the alpha4-, alpha5-, and alpha6-helices. Our data also show that well known regulatory binding sites of PP1, such as the RVXF-binding channel, the beta12/beta13-loop, and the acidic groove, are not essential for the interaction with Sds22.

Regulator-driven functional diversification of protein phosphatase-1 in eukaryotic evolution

We have used the (nearly) completed eukaryotic genome sequences to trace the evolution of thirteen families of established vertebrate regulators of type-1 protein phosphatases (PP1). Two of these families are present in all lineages of the eukaryotic crown and therefore qualify as candidate primordial regulators that determined the surface of PP1. The set of regulators of PP1 has continued to expand ever since, often in response to functional innovations in different eukaryotic lineages. In particular, the development of metazoan multicellularity was accompanied by an explosive increase in the number of regulators of PP1. The further increase in the functional diversity of PP1 in the vertebrate lineage was mainly achieved by the duplication of genes for regulatory subunits and by the conversion of already existing proteins into regulators of PP1. Unexpectedly, our analysis has also enabled us to classify nine poorly characterized proteins as likely regulators of PP1.

Nucleotide pyrophosphatases/phosphodiesterases on the move

Nucleotide pyrophosphatases/phosphodiesterases (NPPs) release nucleoside 5'-monophosphates from nucleotides and their derivatives. They exist both as membrane proteins, with an extracellular active site, and as soluble proteins in body fluids. The only well-characterized NPPs are the mammalian ecto-enzymes NPP1 (PC-1), NPP2 (autotaxin) and NPP3 (B10; gp130(RB13-6)). These are modular proteins consisting of a short N-terminal intracellular domain, a single transmembrane domain, two somatomedin-B-like domains, a catalytic domain, and a C-terminal nuclease-like domain. The catalytic domain of NPPs is conserved from prokaryotes to mammals and shows remarkable structural and catalytic similarities with the catalytic domain of other phospho-/sulfo-coordinating enzymes such as alkaline phosphatases. Hydrolysis of pyrophosphate/phosphodiester bonds by NPPs occurs via a nucleotidylated threonine. NPPs are also known to auto(de)phosphorylate this active-site threonine, a process accounted for by an intrinsic phosphatase activity, with the phosphorylated enzyme representing the catalytic intermediate of the phosphatase reaction. NPP1-3 have been implicated in various processes, including bone mineralization, signaling by insulin and by nucleotides, and the differentiation and motility of cells. While it has been established that most of these biological effects of NPPs require a functional catalytic site, their physiological substrates remain to be identified.

Structural and catalytic similarities between nucleotide pyrophosphatases/phosphodiesterases and alkaline phosphatases

Nucleotide pyrophosphatases/phosphodiesterases (NPPs) generate nucleoside 5'-monophosphates from a variety of nucleotides and their derivatives. Here we show by data base analysis that these enzymes are conserved from eubacteria to higher eukaryotes. We also provide evidence for the existence of two additional members of the mammalian family of ecto-NPPs. Homology searches and alignment-assisted mutagenesis revealed that the catalytic core of NPPs assumes a fold similar to that of a superfamily of phospho-/sulfo-coordinating metalloenzymes comprising alkaline phosphatases, phosphoglycerate mutases, and arysulfatases. Mutation of mouse NPP1 in some of its predicted metal-coordinating residues (D358N or H362Q) or in the catalytic site threonine (T238S) resulted in an enzyme that could still form the nucleotidylated catalytic intermediate but was hampered in the second step of catalysis. We also obtained data indicating that the ability of some mammalian NPPs to auto(de)phosphorylate is due to an intrinsic phosphatase activity, whereby the enzyme phosphorylated on Thr-238 represents the covalent intermediate of the phosphatase reaction. The results of site-directed mutagenesis suggested that the nucleotide pyrophosphatase/phosphodiesterase and the phosphatase activities of NPPs are mediated by a single catalytic site.

Syntenin-syndecan binding requires syndecan-synteny and the co-operation of both PDZ domains of syntenin

Syntenin is an adaptor-like molecule that binds to the cytoplasmic domains of all four vertebrate syndecans. Syntenin-syndecan binding involves the C-terminal part of syntenin that contains a tandem of PDZ domains. Here we provide evidence that each PDZ domain of syntenin can interact with a syndecan. Isolated or combined mutations of the carboxylate binding lysines in the inter-betaAbetaB loops and of the alphaB1 residues in either one or both the PDZ domains of syntenin all reduce syntenin-syndecan binding in yeast two-hybrid, blot-overlay, and surface plasmon resonance assays. PDZ2 mutations have more pronounced effects on binding than PDZ1 mutations, but complete abrogation of syntenin-syndecan binding requires the combination of both the lysine and the alphaB1 mutations in both the PDZ domains of syntenin. Isothermal calorimetric titration of syntenin with syndecan peptide reveals the presence of two binding sites in syntenin. Yet, unlike a tandem of two PDZ2 domains and a reconstituted PDZ1+PDZ2 tandem, a tandem of two PDZ1 domains and isolated PDZ1 or PDZ2 domains do not interact with syndecan bait. We conclude to a co-operative binding mode whereby neither of these two PDZ domains is sufficient by itself but where PDZ2 functions as a "major" or "high affinity" syndecan binding domain, and PDZ1 functions as an "accessory" or "low affinity" syndecan binding domain. The paired, but not the isolated PDZ domains of syntenin bind also strongly to the immobilized cytoplasmic domains of neurexin and B-class ephrins. By inference, these data suggest a model whereby recruitment of syntenin to membrane surfaces requires two compatible types of bait that are in "synteny" (occurring together in location) and engages both PDZ domains of syntenin. The synteny of compatible bait may result from the assemblies and co-assemblies of syndecans and other similarly suited partners in larger supramolecular complexes. In general, an intramolecular combination of PDZ domains that are weak, taken individually, would appear to be designed to detect rather than drive the formation of specific molecular assemblies.

Glypican-6, a new member of the glypican family of cell surface heparan sulfate proteoglycans

The glypicans compose a family of glycosylphosphatidylinositol-anchored heparan sulfate proteoglycans. Mutations in dally, a gene encoding a Drosophila glypican, and in GPC3, the gene for human glypican-3, implicate glypicans in the control of cell growth and division. So far, five members of the glypican family have been identified in vertebrates. By sequencing expressed sequence tag clones and products of rapid amplifications of cDNA ends, we identified a sixth member of the glypican family. The glypican-6 mRNA encodes a protein of 555 amino acids that is most homologous to glypican-4 (identity of 63%). Expression of this protein in Namalwa cells shows a core protein of approximately 60 kDa that is substituted with heparan sulfate only. GPC6, the gene encoding human glypican-6, contains nine exons. Like GPC5, the gene encoding glypican-5, GPC6 maps to chromosome 13q32. Clustering of the GPC5/GPC6 genes on chromosome 13q32 is strongly reminiscent of the clustering of the GPC3/GPC4 genes on chromosome Xq26 and suggests GPCs arose from a series of gene and genome duplications. Based on similarities in sequence and gene organization, glypican-1, glypican-2, glypican-4, and glypican-6 appear to define a subfamily of glypicans, differing from the subfamily comprising so far glypican-3 and glypican-5. Northern blottings indicate that glypican-6 mRNA is widespread, with prominent expressions in human fetal kidney and adult ovary. In situ hybridization studies localize glypican-6 to mesenchymal tissues in the developing mouse embryo. High expressions occur in smooth muscle cells lining the aorta and other major blood vessels and in mesenchymal cells of the intestine, kidney, lung, tooth, and gonad. Growth factor signaling in these tissues might in part be regulated by the presence of glypican-6 on the cell surface.

A capping domain for LRR protein interaction modules

Leucine-rich repeats (LRR) are protein interaction modules which are present in a large number of proteins with diverse functions. We describe here a novel motif (16-19 residues) downstream of the last, incomplete, LRR in a subfamily of LRR proteins. In the U2A' spliceosomal protein, this motif is folded into a cap that shields the hydrophobic core of the LRRs from the solvent. Modelling of the LRR-cap in the imidazoline-1 candidate receptor, using the known structure of U2A' as template, showed a conservation of the basic structural features.

Molecular polymorphism of the syndecans. Identification of a hypo-glycanated murine syndecan-1 splice variant

We have identified a cDNA that encodes a variant form of murine syndecan-1. The variant cDNA lacks the sequence corresponding to the first 132 nucleotides of the third exon of the syndecan-1 gene. The corresponding message is rare. The alternative splice respects the reading frame and deletes 44 amino acids from the protein, joining the S45GS47GT sequence to a variant immediate downstream context. This sequence context initiates with alanine instead of glycine as residue 50, reducing the number of SGXG sequence motifs in the protein from two to one. Expression of this variant syndecan-1 in Madin-Darby canine kidney or MOLT-4 cells yielded a recombinant proteoglycan with a reduced number and clustering of the heparan sulfate chains. Both the conversions of Ala50 and of Lys53 into glycine enhanced the heparan sulfate substitution of the variant protein. These findings support the concept that serine-glycine dipeptide signals for glycosaminoglycan/heparan sulfate synthesis depend on sequence context (Zhang, L., David, G., and Esko, J. D. (1995) J. Biol. Chem. 270, 27127-27135) and imply that alternative splicing mechanisms may in part control the molecular polymorphism of syndecan-1 and, therefore, the efficiency and versatility of this protein in its co-receptor functions.

Structure and splice products of the human gene encoding sds22, a putative mitotic regulator of protein phosphatase-1

sds22 is a regulatory subunit of protein phosphatase-1 that is required for the completion of mitosis in yeast. It consists largely of 11 tandem leucine-rich repeats of 22 residues that are expected to mediate interactions with other polypeptides, including protein phosphatase-1. In this paper, we report on the structure of the human gene encoding sds22, designated PPP1R7. This gene (33 kb) comprises 11 exons, but these do not coincide with the sequences encoding the leucine-rich repeats. Up to six splice variants can be generated by exon skipping and alternative polyadenylation, as revealed by expressed sequence tag database analysis, RT-PCR and Northern blot analysis. The sds22 transcripts are expected to encode four different polypeptides. sds22alpha1 corresponds to the variant cloned previously from human brain [Renouf et al. (1995) FEBS Lett. 375, 75-78]. Sds22beta1 is truncated within the ninth repeat and has a short and different C-terminus. Both variants also exist without the sequence corresponding to exon 2, and these are termed sds22alpha2 and sds22beta2. The 5'-flanking region of PPP1R7 contains two NF-Y-binding CCAAT boxes near the transcription start site and potential binding sites for the transcription factors c-Myb, Ik-2 and NF-1, which are conserved in the mouse gene.

Measurement of resting energy expenditure in patients with chronic obstructive pulmonary disease in a clinical setting

There is a growing tendency to estimate energy requirements by means of the assessment of resting energy expenditure (REE) by indirect calorimetry. In this study a computerized open-circuit ventilated hood system is described that was constructed for assessing REE in a clinical setting. Measurement error of the device, tested by ethanol combustion was +2% for VO2 and VCO2 and less than 1% for respiratory quotient. To assess the within-patient variability of REE measurements performed in a daily clinical routine, we studied the following aspects of the measurements in several groups of patients with chronic obstructive pulmonary disease: (1) reproducibility, (2) the influence of routine physical activities before the measurement, (3) measurement duration, and (4) difference between measurements using a ventilated hood or a mouthpiece. Reproducibility of measurements with a 2-month interval in 12 weight-stable patients was good (1415 +/- 128 and 1398 +/- 138 kcal/day). Variations due to limited activities and different measurement durations (between 10 and 30 minutes) were not significant. Variations between measurements with a mouthpiece and ventilated hood were larger in patients than in healthy control subjects, but for both groups no systematic difference was established. REE can be assessed reliably by short-term measurements with a ventilated hood in stable chronic obstructive pulmonary disease patients on an outpatient basis, provided a short rest is taken before the measurement.