Optimizing machine-learning models for mutagenicity prediction through better feature selection.

Assessing a compound's mutagenicity using machine learning is an important activity in the drug discovery and development process. Traditional methods of mutagenicity detection, such as Ames test, are expensive and time and labor intensive. In this context, in silico methods that predict a compound mutagenicity with high accuracy are important. Recently, machine-learning (ML) models are increasingly being proposed to improve the accuracy of mutagenicity prediction. While these models are used in practice, there is further scope to improve the accuracy of these models. We hypothesize that choosing the right features to train the model can further lead to better accuracy. We systematically consider and evaluate a combination of novel structural and molecular features which have the maximal impact on the accuracy of models. We rigorously evaluate these features against multiple classification models (from classical ML models to deep neural network models). The performance of the models was assessed using 5- and 10-fold cross-validation and we show that our approach using the molecule structure, molecular properties, and structural alerts as feature sets successfully outperform the state-of-the-art methods for mutagenicity prediction for the Hansen et al. benchmark dataset with an area under the receiver operating characteristic curve of 0.93. More importantly, our framework shows how combining features could benefit model accuracy improvements.

Optimization of secretion and surface localization of heterologous OVA protein in mycobacteria by using LipY as a carrier.

BACKGROUND: Mycobacterium bovis Bacille Calmette-Guérin (BCG) is not only used as a vaccine against tuberculosis but also protects against leprosy and is used as part of bladder cancer treatment to induce a protective immune response. However, protection by BCG vaccination is not optimal. To improve vaccine efficacy, recombinant BCG expressing heterologous antigens has been put forward to elicit antigen-specific cellular and humoral responses. Cell surface localized or secreted antigens induce better immune responses than their cytosolic counterparts. Optimizing secretion of heterologous proteins or protein fragments holds therefore unexplored potential for improving the efficacy of recombinant BCG vaccine candidates. Secretion of heterologous antigens requires crossing the mycobacterial inner and outer membrane. Mycobacteria have specialized ESX or type VII secretion systems that enable translocation of proteins across both membranes. Probing this secretion system could therefore be a valid approach to surface localize heterologous antigens. RESULTS: We show that ESX-5 substrate LipY, a lipase, can be used as a carrier for heterologous secretion of an ovalbumin fragment (OVA). LipY contains a PE domain and a lipase domain, separated by a linker region. This linker domain is processed upon secretion. Fusion of the PE and linker domains of LipY to OVA enabled ESX-5-dependent secretion of the fusion construct LipY-OVA in M. marinum, albeit with low efficiency. Subsequent random mutagenesis of LipY-OVA and screening for increased secretion resulted in mutants with improved heterologous secretion. Detailed analysis identified two mutations in OVA that improved secretion, i.e. an L280P mutation and a protein-extending frameshift mutation. Finally, deletion of the linker domain of LipY enhanced secretion of LipY-OVA, although this mutation also reduced surface association. Further analysis in wild type LipY showed that the linker domain is required for surface association. CONCLUSION: We show that the ESX-5 system can be used for heterologous secretion. Furthermore, minor mutations in the substrate can enhance secretion. Especially the C-terminal region seems to be important for this. The linker domain of LipY is involved in surface association. These findings show that non-biased screening approaches aid in optimization of heterologous secretion, which can contribute to heterologous vaccine development.

Mutation-based selection and analysis of Komagataeibacter hansenii HDM1-3 for improvement in bacterial cellulose production.

AIMS: A low yield of bacterial cellulose (BC) always results from an excessive accumulation of organic acids. Screening and the selection of bacterial mutants with a low accumulation of organic acids is an efficient approach for improving BC production. METHODS AND RESULTS: In combination with the proton suicide method (medium containing NaBr-NaBrO3 ), diethyl sulphate chemical mutagenesis coupled with 60 Co-γ irradiation treatment were performed for the screening and selection of desired mutant lines with a high yield of BC. Two high-yield strains, Br-3 and Co-5, as well as a low-yield strain, Br-12, were obtained. Amplified fragment length polymorphism (AFLP) was applied to explore the differences between the mutant lines and the wild type. For the Br-12 line, three specific fragments were verified, corresponding to TonB-dependent transport (TBDT), exopolysaccharides output protein (PePr) and an unknown gene. For Co-5, two specific fragments were matched, acsD and UDP-galactose-4-epimerase. In addition, metabolic analysis for the mutant lines indicated that BC production may be limited by excessive accumulation of organic acids in the fermentation. The limitation would be resolved by the cross-talk of genes involved in BC biosynthesis. CONCLUSIONS: Reduced organic acid by-products from glucose in bypasses were found to be responsible for the high-yield BC synthesis in Komagataeibacter hansenii mutant strains. SIGNIFICANCE AND IMPACT OF THE STUDY: The metabolic process was varied by mutagenesis-induced gene disruption of the metabolic products. A new idea was provided for the targeted screening and characterization of mutants in the future.

Isolation and characterization of two cellulose morphology mutants of Gluconacetobacter hansenii ATCC23769 producing cellulose with lower crystallinity.

Gluconacetobacter hansenii, a Gram-negative bacterium, produces and secrets highly crystalline cellulose into growth medium, and has long been used as a model system for studying cellulose synthesis in higher plants. Cellulose synthesis involves the formation of ß-1,4 glucan chains via the polymerization of glucose units by a multi-enzyme cellulose synthase complex (CSC). These glucan chains assemble into ordered structures including crystalline microfibrils. AcsA is the catalytic subunit of the cellulose synthase enzymes in the CSC, and AcsC is required for the secretion of cellulose. However, little is known about other proteins required for the assembly of crystalline cellulose. To address this question, we visually examined cellulose pellicles formed in growth media of 763 individual colonies of G. hansenii generated via Tn5 transposon insertion mutagenesis, and identified 85 that produced cellulose with altered morphologies. X-ray diffraction analysis of these 85 mutants identified two that produced cellulose with significantly lower crystallinity than wild type. The gene disrupted in one of these two mutants encoded a lysine decarboxylase and that in the other encoded an alanine racemase. Solid-state NMR analysis revealed that cellulose produced by these two mutants contained increased amounts of non-crystalline cellulose and monosaccharides associated with non-cellulosic polysaccharides as compared to the wild type. Monosaccharide analysis detected higher percentages of galactose and mannose in cellulose produced by both mutants. Field emission scanning electron microscopy showed that cellulose produced by the mutants was unevenly distributed, with some regions appearing to contain deposition of non-cellulosic polysaccharides; however, the width of the ribbon was comparable to that of normal cellulose. As both lysine decarboxylase and alanine racemase are required for the integrity of peptidoglycan, we propose a model for the role of peptidoglycan in the assembly of crystalline cellulose.

Experimental validation of candidate schizophrenia gene ZNF804A as target for hsa-miR-137.

MicroRNAs (miRNAs) are small non-coding RNAs that mainly function as negative regulators of gene expression (Lai, 2002) and have been shown to be involved in schizophrenia etiology through genetic and expression studies (Burmistrova et al., 2007; Hansen et al., 2007a; Perkins et al., 2007; Beveridge et al., 2010; Kim et al., 2010). In a mega analysis of genome-wide association study (GWAS) of schizophrenia (SZ) and bipolar disorders (BP), a polymorphism (rs1625579) located in the primary transcript of a miRNA gene, hsa-miR-137, was reported to be strongly associated with SZ. Four SZ loci (CACNA1C, TCF4, CSMD1, C10orf26) achieving genome-wide significance in the same study were predicted and later experimentally validated (Kwon et al., 2011) as hsa-miR-137 targets. Here, using in silico, cellular and luciferase based approaches we also provide evidence that another well replicated candidate schizophrenia gene, ZNF804A, is also target for hsa-miR-137.

Creation of salt-insensitive 3'(2'),5'-bisphosphate nucleotidase by modeling and mutagenesis approach.

3'(2'),5'-Bisphosphate nucleotidase, (EC 3.1.3.7) (BPntase) is a ubiquitous enzyme. Recently, these enzymes have drawn considerable attention as in vivo targets of salt toxicity as well as therapeutic targets of lithium that is used for the treatment of manic-depressive disorders. They belong to the Mg2+-dependent Li+-sensitive phosphomonoesterase super-family and are highly sensitive to lithium and sodium ions. However, the molecular mechanism of inhibition of this group of enzymes by monovalent cations has not been completely understood. Previously we have identified a BPntase (Dhal2p) from a highly halotolerant yeast Debaryomyces hansenii. Molecular characterization revealed a number of unique features in Dhal2p, indicating this is an extraordinary member of the family. In this study, we have carried out the structure-function analysis of Dhal2p through the combination of molecular modeling and in vitro mutagenesis approach. We have not only provided the explanation for the role played by the functionally important elements that are conserved among the members of this family but also identified important, novel structural elements in this enzyme. Our study for the first time unraveled the role of a flap as well as a loop region in the functioning of this enzyme. Most importantly, mutations in the loop region resulted in the creation of a BPntase that was insensitive to salt.

Genes required for intrinsic multidrug resistance in Mycobacterium avium.

Genes required for intrinsic multidrug resistance by Mycobacterium avium were identified by screening a library of transposon insertion mutants for the inability to grow in the presence of ciprofloxacin, clarithromycin, and penicillin at subinhibitory concentrations. Two genes, pks12 and Maa2520, were disrupted in multiple drug-susceptible mutants. The pks12 gene (Maa1979), which may be cotranscribed with a downstream gene (Maa1980), is widely conserved in the actinomycetes. Its ortholog in Mycobacterium tuberculosis is a polyketide synthase required for the synthesis of dimycocerosyl phthiocerol, a major cell wall lipid. Mutants of M. avium with insertions into pks12 exhibited altered colony morphology and were drug susceptible, but they grew as well as the wild type did in vitro and intracellularly within THP-1 cells. A pks12 mutant of M. tuberculosis was moderately more susceptible to clarithromycin than was its parent strain; however, susceptibility to ciprofloxacin and penicillin was not altered. M. avium complex (MAC) and M. tuberculosis appear to have different genetic mechanisms for resisting the effects of these antibiotics, with pks12 playing a relatively more significant role in MAC. The second genetic locus identified in this study, Maa2520, is a conserved hypothetical gene with orthologs in M. tuberculosis and Mycobacterium leprae. It is immediately upstream of Maa2521, which may code for an exported protein. Mutants with insertions at this locus were susceptible to multiple antibiotics and slow growing in vitro and were unable to survive intracellularly within THP-1 cells. Like pks12 mutants, they exhibited increased Congo red binding, an indirect indication of cell wall modifications. Maa2520 and pks12 are the first genes to be linked by mutation to intrinsic drug resistance in MAC.

Genes required for mycobacterial growth defined by high density mutagenesis.

Despite over a century of research, tuberculosis remains a leading cause of infectious death worldwide. Faced with increasing rates of drug resistance, the identification of genes that are required for the growth of this organism should provide new targets for the design of antimycobacterial agents. Here, we describe the use of transposon site hybridization (TraSH) to comprehensively identify the genes required by the causative agent, Mycobacterium tuberculosis, for optimal growth. These genes include those that can be assigned to essential pathways as well as many of unknown function. The genes important for the growth of M. tuberculosis are largely conserved in the degenerate genome of the leprosy bacillus, Mycobacterium leprae, indicating that non-essential functions have been selectively lost since this bacterium diverged from other mycobacteria. In contrast, a surprisingly high proportion of these genes lack identifiable orthologues in other bacteria, suggesting that the minimal gene set required for survival varies greatly between organisms with different evolutionary histories.

Genes required for mycobacterial growth defined by high density mutagenesis

Despite over a century of research, tuberculosis remains a leading cause of infectious death worldwide. Faced with increasing rates of drug resistance, the identification of genes that are required for the growth of this organism should provide new targets for the design of antimycobacterial agents. Here, we describe the use of transposon site hybridization (TraSH) to comprehensively identify the genes required by the causative agent, Mycobacterium tuberculosis, for optimal growth. These genes include those that can be assigned to essential pathways as well as many of unknown function. The genes important for the growth of M. tuberculosis are largely conserved in the degenerate genome of the leprosy bacillus, Mycobacterium leprae, indicating that non-essential functions have been selectively lost since this bacterium diverged from other mycobacteria. In contrast, a surprisingly high proportion of these genes lack identifiable orthologues in other bacteria, suggesting that the minimal gene set required for survival varies greatly between organisms with different evolutionary histories.

Molecular and functional analyses of the gene (eshA) encoding the 52-kilodalton protein of Streptomyces coelicolor A3(2) required for antibiotic production.

Analysis of proteins recovered in the S100 precipitate fraction of Streptomyces griseus after ultracentrifugation led to the identification of a 52-kDa protein which is produced during the late growth phase. The gene (eshA) which codes for this protein was cloned from S. griseus, and then its homologue was cloned from Streptomyces coelicolor A3(2). The protein was deduced to be 471 amino acids in length. The protein EshA is characterized by a central region that shows homology to the eukaryotic-type cyclic nucleotide-binding domains. Significant homology was also found to MMPI in Mycobacterium leprae, a major antigenic protein to humans. The eshA gene mapped near the chromosome end and was not essential for viability, as demonstrated by gene disruption experiments, but its disruption resulted in the abolishment of an antibiotic (actinorhodin but not undecylprodigiosin) production. Aerial mycelium was produced as abundantly as by the parent strain. Expression analysis of the EshA protein by Western blotting revealed that EshA is present only in late-growth-phase cells. The eshA gene was transcribed just preceding intracellular accumulation of the EshA protein, as determined by S1 nuclease protection, indicating that EshA expression is regulated at the transcription level. The expression of EshA was unaffected by introduction of the relA mutation, which blocks ppGpp synthesis.

Assessment of the in vitro and in vivo genotoxicity of Thalomid (thalidomide).

Thalomid is the FDA-approved commercial formulation of thalidomide currently used in the US to treat erythema nodosum leprosum, a complication of leprosy. The genotoxicity of Thalomid thalidomide was assessed in the Ames reverse mutation, AS52/XPRT mammalian cell forward gene mutation, and mouse bone marrow micronucleus assays. The Ames and AS52 assays were performed with and without S9. In the Ames, Salmonella typhimurium strains TA1535, 1537, 98, 100, and 102 and Escherichia coli strain WP2 uvrA were used. Assays were performed by using plate incorporation and liquid pre-incubation systems at thalidomide doses of 50-10,000 microg/plate. In the AS52 assay, Chinese hamster ovary cells were plated with fortified Ham's F12 medium and incubated overnight. The medium was then incubated with 1-1000 microg/ml thalidomide. After a series of aspirations, washings, reconstitutions, and incubations, mutant AS52 cells were fixed and stained. Colonies were then counted and the relative survival frequencies compared to negative controls. In the mouse micronucleus assay, Crl:CD-1 albino mice were dosed with 500, 2,500, and 5,000 mg/kg thalidomide and sacrificed over 72 h. Femurs were flushed with fetal bovine serum and the suspensions centrifuged. The supernatant was aspirated and the cell pellet resuspended and stained. Polychromatic erythrocytes were scored for micronucleated polychromatic and normochromatic erythrocytes. Thalidomide did not increase revertant frequencies in all bacterial strains. It also did not produce any significant increase in the average mutant frequencies of AS52 cells and mouse micronucleated polychromatic erythrocytes. We conclude that Celgene's Thalomid thalidomide is non-genotoxic.

Identification of tandem genes involved in lipooligosaccharide expression by Haemophilus ducreyi.

A transposon insertion mutant of Haemophilus ducreyi 35000 possessing a truncated lipooligosaccharide (LOS) failed to bind the LOS-specific monoclonal antibody 3E6 (M. K. Stevens, L. D. Cope, J. D. Radolf, and E. J. Hansen, Infect. Immun. 63:2976-2982, 1995). This transposon was found to have inserted into the first of two tandem genes and also caused a deletion of chromosomal DNA upstream of this gene. These two genes, designated lbgA and lbgB, encoded predicted proteins with molecular masses of 25,788 and 40,236 Da which showed homology with proteins which function in lipopolysaccharide biosynthetic in other gram-negative bacteria. The tandem arrangement of the lbgA and lbgB genes was found to be conserved among H. ducreyi strains. Isogenic LOS mutants, constructed by the insertion of a cat cartridge into either the lbgA or the lbgB gene, expressed an LOS phenotype indistinguishable from that of the original transposon-derived LOS mutant. The wild-type LOS phenotype could be restored by complementation with the appropriate wild-type allele. These two LOS mutants proved to be as virulent as the wild-type parent strain in an animal model. A double mutant with a deletion of the lbgA and lbgB genes yielded equivocal results when its virulence was tested in an animal model.

A fully modular vector system for the optimization of gene expression in Escherichia coli.

The new expression vector system CYTEXP is designed to facilitate the optimization of both transcription and translation in Escherichia coli, while at the same time allowing the exchange of its major components using unique restriction sites. In vitro mutagenesis can be performed in situ using single-stranded DNA generated from the bacteriophage f1 ORI sequence. The basic vector pCYTEXP1 bears a synthetic copy of the intercistronic sequence that enhances the translation of the E. coli atpE gene. Reading frames can be inserted directly downstream of this sequence. The bacteriophage lambda promoters, the atpE sequence, the bacteriophage fd transcriptional terminator, the f1 ORI, and the amp antibiotic resistance gene are all borne on exchangeable "modules." Thus, both the efficiency and the conditions of expression of cloned genes can be readily optimized.

Insertional mutagenesis and illegitimate recombination in mycobacteria.

Mycobacteria, particularly Mycobacterium tuberculosis, Mycobacterium leprae, and Mycobacterium avium, are major pathogens of man. Although insertional mutagenesis has been an invaluable genetic tool for analyzing the mechanisms of microbial pathogenesis, it has not yet been possible to apply it to the mycobacteria. To overcome intrinsic difficulties in directly manipulating the genetics of slow-growing mycobacteria, including M. tuberculosis and bacille Calmette-Guérin (BCG) vaccine strains, we developed a system for random shuttle mutagenesis. A genomic library of Mycobacterium smegmatis was subjected to transposon mutagenesis with Tn5 seq1, a derivative of Tn5, in Escherichia coli and these transposon-containing recombinant plasmids were reintroduced into mycobacterial chromosomes by homologous recombination. This system has allowed us to isolate several random auxotrophic mutants of M. smegmatis. To extend this strategy to M. tuberculosis and BCG, targeted mutagenesis was performed using a cloned BCG methionine gene that was subjected to Tn5 seq1 mutagenesis in E. coli and reintroduced into the mycobacteria. Surprisingly for prokaryotes, both BCG and M. tuberculosis were found to incorporate linear DNA fragments into illegitimate sites throughout the mycobacterial genomes at a frequency of 10(-5) to 10(-4) relative to the number of transformants obtained with autonomously replicating vectors. Thus the efficient illegitimate recombination of linear DNA fragments provides the basis for an insertional mutagenesis system for M. tuberculosis and BCG.