Datasets:

chemNLP

/

ncbi-text

like 0

RNA plays a surprising and previously unsuspected role in many biological processes, such as post-transcriptional regulation, conformational switches, expansion of the genetic code (such as selenocysteine insertion), ribosomal frameshift, metabolite-binding and chemical modification of specific nucleotides in the ribosome. Apart from its catalytic role as a ribonucleic enzyme () (), RNA can regulate genes in several ways. For example, by hybridizing to a portion of messenger RNA, small ∼ 22 nt RNA molecules perform post-transcriptional gene regulation by RNA interference (RNAi), a process so important that for its discovery the 2006 Nobel Prize in Physiology or Medicine was awarded to A. Z. Fire and C. C. Mello. In addition, by very different means, RNA can perform transcriptional and translational gene regulation by allostery, where a portion of the 5′ untranslated region (5′ UTR) of mRNA, known as a (,), undergoes a conformational change upon binding a specific ligand such as adenine, guanine or lysine. As the field of matures, many sophisticated computational tools, e.g. RNA structure prediction, alignment and gene finding, have been developed—see (,) for recent overviews. Recently developed programs that are of most relevance here include the program (,), that computes a low energy ensemble of structures by sampling from the partition function (), and an earlier program () that computes all suboptimal structures within a user-specified number of kcal/mol of the minimum free energy (MFE). In addition, the program () provides a useful description of RNA branching structure by computing the Boltzmann probability of various shapes and also the MFE structure for various shapes. Here, an RNA shape is an equivalence class of secondary structures, describing the overall branching; for instance the shape of a typical cloverleaf tRNA would be [ [ ] [ ] [ ] ]. In this article, we describe the web server , which computes the Boltzmann probability and MFE structures which differ by δ base pairs from a given initial structure. Unlike most of the tools just described, which focus on the MFE structure or a low energy ensemble, yields information concerning the secondary structure folding landscape. Potential applications of include the design of RNA aptamers (see () for a suggestion how RNA might be designed to inhibit the function of the viral enzymes such as HIV-1 reverse transcriptase and hepatitis C NS3 protease), detection of conformational switches, understanding the role played by biologically active structural intermediates and improvement in secondary structure prediction. sup #text Due to the time and space constraints of the algorithm, RNA sequences may be of length up to 300 nucleotides. Sequences of length up to 60 are processed interactively and output is displayed in the user's browser window. For sequences of length 61–300, the computation is done off-line and the results are returned to the user by email; for this, the email address is required. The user can either paste an input sequence (with optional secondary structure), or upload a file of the same. The full input consists of up to four lines, illustrated by the following example.The temperature is set to a default value of 37 C; however the user can enter any integer temperature between 0 and 100. The only required input is an RNA sequence of length at most 300 nucleotides; the FASTA comment, initial secondary structure and upper bound Δ are optional inputs. If no secondary structure is given, then the initial structure is taken to be the MFE structure, as computed by . If the optional input Δ is missing, then Δ is defined to equal the length of the input sequence ; otherwise Δ is the minimum of the input value and . Here, the summation is made over all secondary structures of which are δ-neighbors of . The full partition function Z = ∑ Z is computed by McCaskill's algorithm () if Δ ≥ . In addition to computing probability , computes the number of δ-neighbors of , the MFE over all δ-neighbors of and the MFE secondary structure. Tables of the values and , as well as their graphs, are made available as downloadable files. The five-column text file output, consisting of δ, , , MFE and the MFE structure, is depicted in . can be used to generate alternative low energy structures, which differ markedly from the MFE structure, or from any initially given structure. shows the output for a short 3 ′-UTR sequence of an mRNA with NCBI accession number MUSGBPS. The input structure in this example is the MFE structure (as predicted by ). The output indicates two ranges of δ that show higher probabilities than the rest, 0–9 and 20–24. The MFE structures at distance δ between 0 and 9 from the MFE structure all have very similar folds and the probability of finding the RNA in a structure at δ between 0 and 9 is 0.63. The probability of finding a structure at δ 20–24 is also relatively high, 0.35, and the MFE structures in this range are similar to each other but completely different from the MFE structure. Thus the two highly probable δ ranges represent two possible alternative folds of the RNA. Analyzing the same sequence with gives similar results. finds three types of structures (three clusters), with probabilities 0.65, 0.22 and 0.13, respectively. One cluster contains the MFE structure corresponding to the folds at δ values from 0 to 9, another cluster has a centroid structure resembling the structures at δ between 20 and 24, and the third cluster has a centroid structure similar to the MFE structure. on the other hand is less successful for this example since the alternative folds as predicted by have the same shape [ ], even though the folds are very different. displays the MFE structure and the MFE structure of the 101 nt SAM riboswitch with EMBL accession number AP004597.1/118941-119041, with sequence taken from Rfam (). The MFE structure over all 30-neighbors, the MFE structure, is clearly much closer to the real structure than the global MFE structure. displays the Boltzmann probability density, showing a peak for the value δ = 30. inline-formula xref italic sup #text

Comparative analyses of multiple bacterial genomes have revealed that some bacterial species possess an extremely plastic genome (,). Horizontal gene transfer events have led to the integration of foreign DNA segments into species-specific syntenic backbones, often within tRNA and tmRNA gene sites (,). This ‘optional’ genomic repertoire, termed ‘mobilome’ (mobile genome) (,), which can vary considerably between members of the same bacterial species, includes episomal plasmids, transposons, integrons, prophages and a growing list of pathogenicity islands (PAIs) or genomic islands (GIs) (,). Many approaches for detecting mobile genetic elements in sequenced bacterial genomes have been developed recently. These include methods based on anomalous codon usage, G+C content, dinucleotide bias, and amino acid usage patterns (), identification of archetypal GI-specific features () and comparative genomics (,); for excellent reviews see (,,). The main barrier to high-throughput prospecting of the mobilome has been a paucity of bacterial genome sequence information, and so it has become a major challenge to develop rapid and cost-effective approaches to discover strain-specific DNA that is dispersed amongst hundreds of members of bacterial species of principal interest to man (). Recently we have developed a high-throughput strategy, dubbed MobilomeFINDER, for experimental and discovery of bacterial GIs (). This approach combines the newly proposed ‘MAmP’, ‘tRNAcc’ and ‘tRIP’ comparative genomics-based approaches with an experimental island capture step facilitated by island probing () and/or a yeast-based homologous recombination system (). MAmP (Microarray-Assisted mobilome Prospecting) is underpinned by comparative genomic hybridization (CGH), ArrayOme and pulsed-field gel electrophoresis (PFGE) genome sizing (A) and is used to screen large numbers of isolates to identify strains that are particularly rich in mobilome DNA sequences to which the species meta-array would have been ‘blind’. tRNAcc (tRNA gene contents and contexts analysis), complemented by an PCR approach (B), is used to identify putative GIs in closely related complete and near-complete genomes. Finally, the tRIP (tRNA site interrogation for pathogenicity islands, prophages and other GIs) (C) strategy permits high-throughput experimental identification and characterization of new GIs through PCR-based profiling of MAmP-selected or otherwise chosen test strains, followed by large-scale targeted capture and full-length sequencing of GIs. We have now incorporated and improved the previously reported ArrayOme and tRNAcc standalone tools into a user-friendly MobilomeFINDER web-server as a public resource: . xref table-wrap #text MobilomeFINDER runs on a Linux platform and has integrated and improved the ArrayOme () and tRNAcc standalone packages () that we reported previously. Specific enhancements include: (i) a newly developed tRIP tool, (ii) an intuitive DNAnalyser tool that generates additional outputs comprising a circular genome map with the locations/sizes of GIs marked and a plot of the negative cumulative GC profile of the genome, (iii) a schematic ArrayOme output, (iv) hyperlinks to visualize DNA fragment details using NCBI Sequence Viewer, (v) an ExtractFlank tool that automatically generates ClustalW multiple sequence alignment files and (vi) ‘example’ and ‘format’ prompts, ready access to tutorial files and enhanced warnings re file formatting errors aid file construction and entry. In addition, the following freely available components were employed: Mauve 1.2.2 (); NCBI Blast 2.2.9 (); ClustalW (); Primaclade (); Electronic PCR (); CGview (); gnuplot () and Bioperl (). Each run is assigned a job-id and the output files are kept on the server for 7 days allowing the user to inspect the results at any given time. The server web site includes a step-by-step tutorial for general users as well as detailed technical documentation and the open source codes of tRNAcc and ArrayOme for software developers. In addition, users can download the standalone versions of tRNAcc and ArrayOme to run locally. Because sequence alignment algorithms, such as the multigenome comparison tool Mauve () and the pairwise alignment tool BLAST (), are computationally intensive, it may not be possible to return results to users immediately when the input is large. With the current hardware configuration using two Dual-Core Intel Xeon 2.8GHz processors and 8GB RAM, the MAUVE-facilitated tool IdentifyIsland takes about 1 h to discover islands by comparative analysis of the contents and contexts of ∼80 tRNA sites across three closely related bacterial genomes. Three tools, LocateHotspots and GenomeSubtractor that use BLASTN and IdentifyIsland that uses MAUVE, display a URL for subsequent retrieval of results if the job cannot be completed promptly. Alternatively, if users supply their e-mail address, results will be emailed automatically upon completion of the job. MobilomeFINDER and the related experimental methodologies are applicable to a wide range of bacterial species. To date the web-server has been used to perform comparative bacterial genomic analyses for several species including nine genomes, four genomes, two genomes (), two genomes and two genomes; the resulting data are shown at . We have used the MobilomeFINDER web-server to characterize the GI contents of blood culture-derived isolates obtained from patients with no laboratory evidence of concurrent urinary tract infections. CGH analyses using the metagenome microarray () (), together with PFGE-based genome sizing has been used to identify mobilome-rich strains by MAmP (A). In addition, PCR-based tRNA site interrogation (tRIP) (C) coupled with chromosome walking and sequencing has been used to investigate sixteen tRNA loci in ten selected isolates. Approximately half of the 85 GIs identified were related to UPEC strain CFT073 islands, with an equal number resembling elements in and EAEC, EHEC, EPEC pathotypes of . Based on a limited preview data, at least seven GIs contained sequences novel to , with six possessing stretches of sequence without any counterparts in the entire DNA database (K. Rajakumar, unpublished data). In addition to the 95 GIs we identified within sequenced genomes in our recent study (), we have also discovered by tRIP analysis a large tRNA gene-associated GI that contains a likely DNA modifying gene cluster (,) within the unfinished genome of enterotoxigenic (ETEC) B7A (RefSeq accession no. NZ_AAJT00000000). xref email ext-link #text italic xref #text

Protein–protein interactions are involved in most biological processes. Identifying their associated networks comprehensively is the key to understanding cellular mechanisms (). Some systematic identification of protein–protein interactions have been constructed by high-throughput experimental methods, such as large-scale two-hybrid system () and affinity purifications (). A basic problem with most large-scale experimental methods is the high false-positive rate (). Many computational methods have been developed to predict protein–protein interactions by using gene expression profiles (), domain–domain interactions (), phylogenetic profiles (), known 3D complexes (,) and interologs (,). These large-scale methods are often unable to respond how a protein interacts with another one. To identify interacting domains from three-dimensional (3D) structural complexes is able to study domain–domain interactions. A known 3D structure of interacting proteins provides interacting domains and atomic details for thousands of direct physical interactions. In addition, it is usually possible to build an interaction model of two proteins by comparative modeling if a known complex structure comprising homologs of these two sequences is available (, , , ). For a pair sequences, these methods often search a 3D-complex library to find homologous templates and score how well the query protein pair fit the known template structures by using a scoring matrix. In this way, they should evaluate all possible protein pairs (18 000 000) in one species if it has 6000 proteins. Our previous study proposed ‘3D-domain interologs’ which is similar to ‘interologs’ (). The 3D-domain interologs is defined as ‘Domain (in chain A) interacts with domain (in chain B) in a known 3D complex, their inferring protein pair A′ (containing domain ) and B′ (containing domain ) in the same species would be likely to interact with each other if both protein pairs are homologous.’ Based on this concept, we are able to search protein databases to predict protein–protein interactions for many species by using a 3D-dimer complex (). Here, we report the development of an automatic server, 3D-partner, for interacting partners and binding models prediction by using 3D-domain interologs through structure complexes and a knowledge-based scoring function which is the key novelty in this article. The 3D-partner utilizes IMPALA and PSI-BLAST to identify homologous structures (templates) and interacting partners of a query protein sequence from a 3D-dimer template library and protein sequence databases [i.e. SwissProt ()], respectively. These homologous structures and interacting partners were evaluated by a scoring function which considered steric and special-bond matrices (i.e. hydrogen bonds, electrostatic interactions and disulfide bonds) but also the template consensus scores (couple-conserved residue score and template similarity). After interacting partners were identified, the 3D-partner provides 3D interacting domains and contact residues for visualizing molecular details of any protein pairs between the query and interacting partners. The 3D-partner server was tested on 275 mutated residues selected from the Alanine Scanning Energetics database (ASEdb) () to predict the binding affinities. The correlation between experimental energies and predicted energies is 0.91. In addition, the average precision of this server for interacting partner prediction is 0.72 by using a non-redundant set. presents the details of the 3D-partner server for inferring interacting partners and binding models of a query sequence through structure complexes and a new scoring function by the following steps. First, the server uses IMPALA to search template candidates of a query () from 3D-complex profile library (1894 heterodimers). IMPALA, widely used for local sequence alignments, searches the query sequence against each of the template profiles, which constitute a database of PSI-BLAST-generated position-specific score matrices (PSSMs). A template is considered as a candidate if the -value is <0.05 and the aligned contact residue ratio () between the and candidate is >0.5. The aligned procedure of IMPALA is a sequence (Q) to profile (template) alignment. Second, our scoring function is applied to calculate the interacting score and -value for each candidate, which is selected as a homologous template ( in ) of if its -value > 3.0, according to the aligned contact pairs on the template. After homologous templates are identified, the 3D-partner identified interacting partner candidates of the query. For each homologous template (), this server applies PSI-BLAST to scan the interacting-partner sequence profile ( in ) of against each of protein sequences in the SwissProt version 51.3 (containing 250 296 protein sequences). The sequence profile, built by using the same procedure for template sequence profiles, is the initial PSSM of PSI-BLAST and the number of iteration is set to one. Therefore, this search procedure can be considered as a profile-to-sequence alignment. The sequences whose -value < 0.05 and  > 0.5 are selected as homologous sequences of . Finally, for each homologous sequence, our scoring function is applied to calculate the interacting score and to evaluate the -value between the query and the homologous sequence according to the aligned contact pairs on the hit template. A homologous sequence is considered as an interacting partner of the query if the -value >3.0. The server reports interacting partners of the query ordered by -values which represent the statistical significances of hit interacting partners. fig #text italic #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R o n l i n e .

Phylogenetic and evolutionary analyses of sequences are among the most often used methodologies in laboratories working in functional, comparative and structural genomics (). Since 1980, when the first version of the PHYLogeny Inference Package (PHYLIP) () was introduced by Felsenstein, a high number of programs for phylogenetic inference have been developed. Currently, PHYLIP(), PAUP*(), MEGA(), PhyML(), PAML() and MrBayes() are well-known programs that are used by thousands of users around the world. Other more specific programs, designed to test evolutionary hypotheses for model selection, tree topology, molecular clock or adaptation, are less popular among common users, but they are, nevertheless, of great interest for users familiar with evolutionary enquiries. Currently, the most comprehensive list of phylogenetic resources can be found at the University of Washington in Seattle (), which listed 292 phylogeny packages and 38 web servers, by 2003. Web servers for phylogenetic and evolutionary analyses provide a direct means for addressing several evolutionary questions, ranging from the computation of a multiple alignment and a neighbor-joining tree using ClustalW program () (), to the more sophisticate analysis of molecular adaptation for detection of positively selected sites in DNA sequences (using methods as those available in the HYPHY package () ()). Many such servers run a single tool or program whereas others bring together many of the most popular programs of phylogenetic reconstruction (e.g. see ). Despite this diversity there is, so far, no single integrated web server that provides a common framework to run the most frequent analyses on DNA and protein sequences from a phylogenetic and evolutionary perspective. Non-expert users are then often overwhelmed by the variety of servers, formats and options available and by the difficulty of concatenating analyses performed on different servers. The main objective of Phylemon is to fulfil this need by providing users with the possibility of finding all the necessary applications in a single integrated web framework that guides them throughout the whole evolutionary analysis. Phylemon is a web server that integrates a selected suite of more than 20 different tools from the most popular stand-alone programs of phylogenetic and evolutionary analysis (A). Three features characterize all tools integrated in Phylemon: tools have available examples in order to familiarize users with the correct input data and expected results, input formats (preferentially FASTA or PHYLIP) are automatically transformed in order to move among alternative tools and all the input and output result files can be saved in default or user-defined projects (folders). Phylemon can be accessed by anonymous login or by registered users. The only difference between these choices is that registered users, from whom only an e-mail is required, can store project results and use them at a later time for further analysis (G). Phylemon runs distance-based methods, maximum parsimony analyses and statistical methods of phylogenetic reconstruction. Distances and parsimony methods for DNA or protein sequence data are provided by the most often used algorithms of the PHYLIP package () v3.65: DnaDist, ProtDist, DnaPars and ProtPars, respectively. Pairwise distance matrices can be represented in a phylogenetic tree using the neighbor-joining (NJ) algorithm (Neighbor) or applying a least square (LS) method or a minimum evolution (ME) criterium (Fitch program). In order to obtain trees with statistical support on internal nodes, a re-sampling method (i.e. bootstrap option included in the Seqboot algorithm) and the corresponding trees summarizing algorithm (i.e. majority rule tree using Consense program) of PHYLIP are included. Basic maximum likelihood (ML) analyses of DNA and protein sequence data are provided with the DnaML and ProML algorithms of the PHYLIP package in Phylemon. When a more sophisticated ML analysis is required users can run PhyML version aLRT (,) or TREE-PUZZLE v5.2 (). Major differences between these ML programs are: PhyML is faster than any other ML algorithm of phylogenetic reconstruction, TREE-PUZZLE uses a quartet-puzzling method instead the more classical heuristic searches for tree searching, TREE-PUZZLE reports reliability values while the PhyML method reports Felsenstein's bootstrap values and aLRT-related statistics branch support (), TREE-PUZZLE can quantify the amount of the phylogenetic signal contained in a data set (the probability of the data producing a tree-like phylogenetic representation) through the likelihood-mapping method () and TREE-PUZZLE computes ML pairwise distances that can easily be represented in an NJ/LS/ME tree. Finally, Phylemon runs Bayesian phylogenetic analysis using MrBayes v3.1.12 (). MrBayes runs in Phylemon with the same characteristics that users have in Windows or Linux interfaces. Users can define all the parameters of MrBayes in a file to upload to the server or, alternatively, edit the parameters in the specific text box when entering the data. A valuable list of examples showing alternative data and parameters is available in the server. In addition, when users edit sequence data or parameters, the MrBayes command list is available on the fly. At the end of the run, Phylemon asks for the sump and sumt parameters in order to define the burnin limit (D). Once the analysis is finished, 10 outfiles are listed, 8 corresponding to the usual files produced by MrBayes and 2 corresponding to: all information printed to the standard output the program runs (outfile.out) and the last topologies retained (with branch lengths and posterior probabilities on the nodes—tree.nw), the last of which can be visualized in a tree-viewer program available in the utilities section of Phylemon. #text xref #text Molecular adaptation is an exciting topic in molecular evolution and phylogenetic studies (). Three alternative programs are included in Phylemon that allow the inference of molecular adaptation events, these are YN00 and CodeML from PAML v3.15 () and the sitewise likelihood-ratio (SLR) () method. YN00 program implements pairwise computations of ω (dN/dS) from synonymous and non-synonymous substitutions rates as defined in different counting methods (), such as NG (), LWL (), Li (), PB () and YN00 () (F). CodeML uses numerical optimization algorithms to maximize the log-likelihood values under a specific model of evolution. CodeML requires that users provide option parameters in a control file in which all variables of the ML models are listed. Although likely straightforward for advanced users, the configuration and compatibility of the different options are not evident for novel users. Therefore, we have developed a web interface with more than 20 examples covering the branch, site and branch-site ML models. Branch models allow searching for positive selection acting on a particular lineage in a phylogeny (), whereas site models detect adaptive evolution on codon positions in the alignment (,), and branch-site models detect positive selection affecting only a few sites along a few lineages (,) (see (,) for its application on the human genome). The SLR method uses a site by site approach to test for neutrality but, in contrast to similar methods such as SG (), SLR does so by using the entire alignment to determine quantities common to all sites, such as evolutionary distances. At the end of the test, a necessary correction for multiple testing is completed. Readers can see () for a comparison on estimations of SLR and CodeML site models. Substitution rates between DNA or protein sequences, whether grouped or not in phylogenetically defined lineages, can be statistically compared in Phylemon. This is done by using relative rates test () as computed in RRTree vs 1.1.11 (,) program. RRTree computes relative rates tests among user-defined lineages. When a lineage includes many species, RRTree computes relative rates tests with a weighted scheme for species based on the tree topology provided by the user (,). RRTree computes differences in rates for coding DNA sequences using different parameters: the number of synonymous substitutions and synonymous transitions per synonymous site (Ks and As, respectively), the number of non-synonymous substitutions and non-synonymous transversions per non-synonymous site (Ka and Ba, respectively) and, finally, the number of synonymous transversions per 4-fold degenerate site (B4). Kimura two parameters (K2P) () and Jukes and Cantor (JC) models are available for non-coding DNA sequences. For protein sequences, RRTree computes a modification of JC model (). #text xref #text The Phylemon web server integrates two different programs for the alignment of multiple sequences: ClustalW v1.83 () and MUSCLE v3.52 (). Furthermore, Phylemon provides additional pre- and post-analysis utilities. These include file format conversion, gene concatenation, tree visualization and the computation of distances between trees. Conversion between sequence formats can be made by means of the ReadSeq program (GNU/Linux program). Users can transform alternative files to FASTA or PHYLIP format and run with confidence any of the Phylemon tools. The concatenation of individual multiple alignments (PHYLIP format) with equal or different number of species, generally employed in phylogenomic studies, can be made using a facility specifically developed by us for such a purpose. Rooted and unrooted newick tree formats can be visualized in rectangular, radial and circular diagrams using ETE program (Environment for Tree Exploration, developed by JHC, (C)). Finally topological distances between trees are computed by the TreeDist program from PHYLIP. This program measures symmetric differences or branch score distances between two or more trees [see () for an application of its use]. Molecular evolution and phylogenetics embrace a wide range of scientific enquiries. Following the development of the field in the last 20 years, researchers have developed tools ranging from the most complete packages to the more specific programs. Although some of these are available online in separate, dedicated web servers, many of the programs available in Phylemon cannot be found on any public web server. This is the case for the frequently used MrBayes, Tree-Puzzle, CodeML in full, SLR and RRTree programs. Altogether, Phylemon addresses an important, yet unanswered, necessity of users working with evolutionary and phylogenetic analysis of molecular sequences; namely, the need for a public web server providing a core set of format compatible tools truly integrated in an independent platform.

The biggest challenge in human genetics currently is to identify the genes whose alleles confer susceptibility to disease. It is believed that there will be many loci that increase the risk for each common disease (). Since each causative gene may make only a very modest contribution to disease risk, identification of particular susceptibility variants becomes quite difficult. While genetic association studies have been used in gene mapping, their efficiency has been limited because they have typically assessed only one or a few genes at a time. The development of new SNP genotyping technologies, which can handle from dozens to hundreds of thousands of SNPs, and large numbers of samples, promises to accelerate gene mapping. Current platforms include the Illumina BeadArray and BeadChip systems (,), the Affymetrix GeneChip Mapping Arrays () and Applied Biosystem's TaqMan SNP Genotyping Assays (). The newest technologies, while powerful, bring with them substantial costs, as they can involve as many as hundreds of millions of genotypes. For this reason, researchers have been trying to devise ways to maximize efficiency of resource utilization given a set of SNP-based gene-mapping goals. It has been found that the pattern of linkage disequilibrium (LD) varies across the human genome and that there are discrete regions of high LD in the genome, called haplotype blocks (). Most variation in populations can be characterized by a small number of common haplotypes. By selecting SNPs that uniquely identify or ‘tag’ these haplotypes, the number of markers and, hence, the cost of genotyping can be significantly reduced. The approach became more powerful with the availability of genetic data from the International HapMap Project (), which contains genotype data for ∼4 million SNPs from each of four populations: Yoruba from Ibadan, Nigeria (YRI), Japanese from Tokyo (JPT), Chinese from Beijing (CHB) and United States residents with European ancestry (CEU). Even with the increased efficiency introduced by tagSNPs, investigators are typically in the position of having to make strategic decisions about which set of tagSNPs to study. One strategy is to focus on those within genes, as these have the greatest likelihood of being functionally relevant or being in LD with those that are functional (). Recently, a similar question was explored using empirical data from the HapMap-ENCODE project; tagSNPs chosen to capture common variation in exonic as well as evolutionarily conserved regions yielded genotype savings compared with a tagging approach that captured all common variation across the region (). While the extent to which functionally important elements in the genome reside strictly within and near genes is not known, a gene-centric genotyping strategy may be a reasonable approach to searching for disease susceptibility alleles in the setting of limited resources. The choice of SNPs for genetic association testing, thus, is a crucial step that will directly affect both the cost and the outcome of studies. Since the number of SNPs can range into the thousands, manual selection can be extremely time-consuming. There are some useful internet-based tools available for selection and prioritization of SNPs for genotyping. These include SNPper () (), TAMAL () (), SNPSelector () (), SNPHunter () (), PupasView () () and tagger () (). These programs have a variety of strengths as well as limitations. Among the gaps: most of them do not allow for automated selection of gene-based SNPs in a region, and none examines SNP coverage on genome-wide microarray SNP genotyping platforms. We have developed a web server named to provide selection of tagSNPs in a chromosomal region, and to fill in some of the gaps in existing SNP selection tools. One useful feature of is the option to input the coordinates of a chromosomal region and have the program select SNPs, in an automated fashion, only from the genes within that region. Other useful features include automated selection of coding non-synonymous SNPs, SNP filtering based on inter-SNP distances, and reporting of whether SNPs have available assays or are present on whole genome chips. There are several situations where we believe this tool will be particularly useful, including: (i) planning an LD-mapping study of a region , where one has decided for any of a number of reasons to focus on genes and (ii) one is planning to obtain, or has obtained, data from a genome-wide association chip, and one wants to ‘fill in’ a particular region either because the chip scan produced a positive result or because of other information (e.g. a linkage peak or interest in a particular gene pathway), and one wants to find additional tagSNPs as well as coding non-synonymous SNPs in genes in the region. utilizes Apache as its web server, and CGI (Common Gateway Interface) scripts are used to handle dataflow and validation to and from a dynamic HTML interface that utilizes cascading style sheet objects and integrated JavaScript. The data extraction and manipulation portion of the program is written in PERL (practical extraction and report language) modules and features two other programs embedded in the main code—, a freely available Java-based utility, and , a freely available Linux command-line application. is available at the URL . It is located on a cluster of processors running Linux OS at the Johns Hopkins McKusick-Nathans Institute for Genetic Medicine. All databases are locally downloaded and placed in the storage space of the Linux cluster. Files are not copied to a fileserver during user data uploads, but instead data is extracted dynamically from these files using CGI file handles, and thus information uploaded by users will not be retained on a file server. The basic function of is to generate a list of tagSNPs in a given chromosomal region, or for the genes in that region, or for any specified list of genes. For genomic position (whole region) searches, genotype data for SNPs lying in the region are extracted from the HapMap database (). For genomic position (genes only) and gene name-based queries, gene coordinates are first extracted from the Entrez gene database and then SNP genotype data for those positions are extracted from HapMap. If the genomic position entered is not for NCBI build 35 (May 2004), it is first converted to that by the program. Also, genomic position is adjusted according to the length of flanking sequence used. The resulting SNP list is passed to the program, which generates tagSNPs based on the tagger algorithm () using the user-specified , minimum minor allele frequency and include/exclude tags specifications (if any). There are categories of options that the user can select to obtain the best results from (see below). Input options like include/exclude tags and coding non-synonymous SNPs are used before tagSNP selection and they affect the list of SNPs that is used by for tagSNP selection. After selection of tagSNPs, the results can be further filtered by options such as removing SNPs lying too close (by a user-defined distance criterion). The user sees a results web page with two types of output. One is the core output consisting of a summary statistics table, a file displaying tagSNPs selected, and another file displaying pairwise LD tests as well as LD bins. The other type of output consists of additional information, including tables for genotype and allele frequencies, for the occurrence of SNPs in whole-genome chips and assays, and for the cost of genotyping. There is also a link to a graphical display of tagSNPs in the . We have attempted to create a simplified user interface for so that it can be employed without the need for sophisticated computational skills. The input screen is divided into three main sections: input method, search conditions and additional options. The user may enter either genomic positions or gene names in the search window of the input method section. Multiple gene names can be entered by either typing in the corresponding window or uploading a file with a list of gene names. For the genomic position-based searches, users can further specify whether they wish to consider the whole region or just the genes within the specified region for tagSNP selection. The user is then required, in the search conditions section, to enter the desired , minor allele frequency and HapMap population. To access the basic functionality of the server, the user need not consider the section containing additional options. However, depending on the study design, these options can enable more judicious and efficient selection of tagSNPs. For example, the user may want to include or exclude certain SNPs (based on availability of PCR primers or on past performance of genotyping assays). The user has the option to include flanking sequence around genes, reject SNPs that are too close to each other (because they are less likely to work with certain genotyping platforms), and force include coding non-synonymous SNPs, which can be identified and included automatically by through a search of the whole-genome coding SNP database. There are other result-related options that display various kinds of information for the chosen tagSNPs. These include the cost of genotyping using some popular methods, allele and genotype frequencies of tagSNPs in four HapMap populations, and occurrence of tagSNPs in available whole-genome chips and assays. For genomic position-based queries, the user also has the option to graphically visualize the tagSNPs in relation to genes, transcripts, conserved regions and other genomic features in that region using the (). The results are generated in the form of a zipped archive containing multiple files (for multiple genes), as well as text files corresponding to an individual gene or region. A file containing a list of tagSNPs chosen and another file with details regarding LD tests and bins is generated for each gene/region. A summary table is also generated that displays the number of SNPs in the HapMap database for the gene/region queried and the eventual number of tagSNPs selected by for individual genes as well as the whole region. If include/exclude existing tags and/or coding non-synonymous SNPs were implemented in a search, an additional result table would be generated that lists the included or excluded SNPs, which of them were used in the tagSNP search (only those with genotype data in HapMap database can be used for tagging), and their type (user-specified include/exclude tags versus coding non-synonymous SNPs). Other results are also generated based on the additional options used for search (). There are three levels of help available to users: (a) QuickHelp, which can be accessed by clicking on the [?] symbol next to each option, and which briefly explains the purpose of that option; (b) frequently asked questions, which provides more detail and (c) direct contact with the authors, available by emailing us at . We validated the core functionality of for various genes and genomic regions by comparing results from to those derived from a manual tagSNP selection using HapMap and the tagger algorithm in Haploview. Since many options and features are unique to this tool, they could not be compared to existing automated resources for tagSNP selection. For those cases, we manually performed steps of analyses for some of the options (for example, gene-based searches in a genomic region including coding non-synonymous SNPs), and compared the results with those generated in an automated manner by . The results generated by were always in agreement with those generated by the manual procedures. We extensively used to select tagSNPs for a 6 Mb region on chromosome 17 that produced evidence for linkage to major depressive disorder in our Genetics of Recurrent Early Onset Depression (GenRED) collaborative project (). Our aim was to select SNPs for an initial LD mapping association study of this region using the Illumina BeadStation custom genotyping platform. The region contained a total of ∼8000 HapMapII SNPs. Using criteria of  ⩾ 0.8 and MAF ⩾ 0.1, there were 1526 tagSNPs selected from across the full region and an additional 438 coding non-synonymous SNPs. Our project budget allowed us to study approximately 800 SNPs from the region in this initial experiment, so that excellent tagSNP coverage could be achieved if we focused on genes and their associated regulatory regions. We searched the region with using the genomic position, genes-only input method, force including coding non-synonymous and some previously genotyped SNPs, and rejecting SNPs that were closer than 60 bp. We performed various searches for different , MAF values and lengths of flanking region around genes. shows the number of tagSNPs selected by using different combinations of parameters. We elected to genotype the 809 SNPs that resulted from tagging with  = 0.8, MAF ≥ 0.1 and a 5 kb flanking region on either side of each gene. offers many useful features (see for a comparison with other available programs): In the last few years, millions of new SNPs have been identified, and SNP genotyping technologies have developed rapidly. Investigators need to determine how to select SNPs for study in a chromosomal region in a manner that is efficient while still preserving power. There is a need for new tools, which can perform these functions in an automated manner. provides all of the basic SNP selection functions present in existing tools, while adding additional features. #text

Accurate detection of sequence similarity between distantly related proteins is essential for many fields, including protein structure prediction, protein engineering, and comparative genomics. The performance of an automatic method for sequence comparison can be characterized by sensitivity, selectivity and accuracy of produced sequence alignments. All these parameters can be significantly improved by comparing multiple sequence alignments (MSAs) rather than individual sequences. The improvement comes from evolutionary information about residue preferences at sequence positions in the family represented by the MSA. This information can be extracted from MSAs in two numerical forms: ‘traditional’ position-specific profiles and hidden Markov models (HMMs). The well-known and popular methods for profile-sequence or HMM-sequence comparison include PSI-BLAST (,), HMMER (), SAM-T (,) and others. A newer generation of methods involves the comparison of two profiles () or two HMMs (,), with several corresponding web servers available (). These methods further improve the quality of homology detection and alignment construction (,). There is a number of publicly available web servers aimed at protein structure prediction that use these and a variety of other techniques [for example, ()]. COMPASS () is an established method for profile-based comparison of MSAs. COMPASS derives numerical profiles from given MSAs, constructs optimal local profile-profile alignments, and analytically estimates E-values for the detected similarities. As previously shown by us () and independently verified by others (,), COMPASS is a sensitive and selective tool for detection of remote sequence similarity that offers accurate local alignments. In many cases, COMPASS provides accurate homology detection and structure prediction that would be difficult or impossible to produce by PSI-BLAST (,). As a standalone package, COMPASS has been used by different research groups (). Until now, COMPASS was only available for download and local installation. Here, we present a new web server featuring the recently improved version of COMPASS. To compare two MSAs, COMPASS performs four steps: (i) processing input MSAs and generating numerical profiles; (ii) calculating scores between individual positions of the compared profiles; (iii) finding optimal local alignment of the two profiles; and (iv) assessing statistical significance of the optimal alignment score (). Methodologically, COMPASS is a generalization to profile-profile comparison of the PSI-BLAST approach to profile-sequence comparison. Numerical profiles represent effective counts and frequencies of 21 symbols (20 residue types and gaps) at each position of the input MSAs. To search with a query MSA against a database of MSAs, the database profiles are pre-computed in advance. Scores for the similarity between individual profile positions are calculated using our original formula () and then rescaled so that their distribution is similar to a standard distribution with well-known properties (such as BLOSUM62 substitution scores). Rescaled positional scores are used to find the optimal local alignment using the Smith–Waterman algorithm. The statistical significance of the optimal alignment score is estimated using a simple formula for E-value (the expected number of hits in a random database with a score equal to or greater than the observed score). The parameters of this formula are based on our extensive simulations of random profile comparisons (). As the final result of the search, a list of the most significant hits for the submitted query is displayed, followed by the optimal profile-profile alignments. According to our results () and independent evaluations (,), COMPASS performance has been demonstrated to be among the top methods for profile comparison, by both the quality of homology detection and the accuracy of local alignment construction. The presented web server features a newer version of COMPASS, with several major modifications to improve performance. In order to gain more confidence in detected similarities and to find the best search conditions for a specific query, tuning the parameters controlling the generation of profiles and the construction of profile-profile alignments is advisable. The user can modify several such parameters. First, the input MSA (or sequence) can be used as a query for PSI-BLAST search, in order to produce a more diverse MSA of this family. The user can adjust the maximal number of iterations, as well as the requirements for a detected homolog to be included in the alignment: maximal E-value, minimal coverage of the query and minimal sequence identity to the query. Second, ‘Gap fraction threshold’ allows the user to control the maximal content of gaps in the MSA columns included in the COMPASS profile. If a column contains too many gaps, it is disregarded in the process of profile comparison, and shown in the final output as lower-case letters for residues and dots for gaps. The default value of this parameter is 0.5. In the construction of profile-profile alignments, ‘Gap penalties’ are score penalties for opening and extending a new gap. ‘Effective length of the database’ is the parameter used in the calculation of E-values for the profile-profile alignments. For a given optimal alignment score, there is roughly a linear dependence of E-value on the assumed database length. ‘Matrix’ is a substitution matrix of the user's choice, BLOSUM62 by default. As described above, the choice of the matrix affects the rescaling of scores between individual profile positions that are used in the construction of the profile-profile alignment. Changing the scale of the positional scores would (i) make gap insertion more or less likely, affecting the resulting alignments, and (ii) change the optimal alignment scores and E-values. Among the output formatting options, many are similar to those of PSI-BLAST. ‘Expect’ and ‘significance threshold’ are, respectively, the E-value cutoffs for the hit to be included in the output and to be considered significant. The hits outside the significance threshold are shown as potentially not meaningful. The user can also limit the total number of hits to display (‘Display up to’). Some output options are specific to profile-profile comparison. For example, the displayed profile-profile alignments can include different numbers of top sequences from the input MSAs (‘Top sequences to show’), as well as consensus sequences (‘Show consensus sequences’). Brief help sections are provided for every adjustable parameter, as well as a link to more detailed documentation (A).

Due to the importance of electrostatic interactions in biomolecular systems, a variety of computational methods have been developed for evaluating electrostatic forces and energies [see () and references therein]. Typical computational electrostatics methods for biomolecular systems can be loosely grouped into two categories: ‘explicit solvent’ methods, which treat solvent molecules in full molecular detail, and ‘implicit solvent’ methods, which include solvent–solute interactions in averaged or continuum fashion. Implicit solvent methods are, by definition, limited in detail and therefore lack the atomic-scale accuracy of their explicit solvent counterparts. However, implicit solvent methods have gained increasing popularity, in part due to their elimination of the extensive sampling of solvent configurations required with explicit models (,). The basic ingredients of an implicit solvent electrostatics calculation are environmental parameters such as temperature, solvent dielectric and ionic strength; biomolecular atomic coordinates; and parameters for atomic charges and radii. While the environmental parameters are relatively straightforward to specify, the remaining two ingredients can often be difficult to supply. In particular, most biomolecular structures in the Protein Data Bank (PDB) () do not contain hydrogen atoms, and many are also missing a fraction of the heavy atom coordinates. The addition of hydrogens and the reconstruction of these missing coordinates is not a trivial process; electrostatic properties obtained from the ‘repaired’ structures can often be very sensitive to the manner in which missing atoms are added and protonation states are assigned (,). Furthermore, inconsistent atomic nomenclature and other force field idiosyncrasies can often make the assignment of atomic charges and radii a cumbersome task. An additional obstacle to the use of PDB structures in electrostatics calculations and other biomolecular computational tasks is the accurate assignment of parameters to ‘non-standard’ residues and ligands. Previously (), we introduced the freely available PDB2PQR service (), which was designed to facilitate the setup and execution of continuum electrostatics calculations from PDB data, particularly by non-experts. The original PDB2PQR server automated many of the common tasks of preparing structures for continuum electrostatics calculations, including adding a limited number of missing heavy atoms to biomolecular structures, estimating titration states and protonating biomolecules in a manner consistent with favorable hydrogen bonding, assigning charge and radius parameters from a variety of force fields, and finally generating ‘PQR’ output (a PDB-like format with the occupancy and temperature factor columns replaced with charge ‘Q’ and radius ‘R’, respectively) compatible with several popular computational biology electrostatics [APBS () and MEAD ()], docking [AutoDock ()], simulation [AMBER ()] and visualization [VMD (), PyMOL () and PMV ()] packages. Since its inception, we have continued to expand the capabilities of the PDB2PQR server to address the challenges associated with ligand parameterization in PDB files and to include several new features. The PDB2PQR web service is driven by a modular, Python-based collection of routines, which provides considerable flexibility to the software and permits non-interactive, high-throughput usage. The service is available via a number of web mirrors listed at . The source code is also available for download from this link, and due to the portability of Python, PDB2PQR can be executed on a wide range of platforms. outlines the typical workflow of a PDB2PQR job and summarizes the features described in more detail below. The procedures for reconstruction of missing atoms, hydrogen optimization and APBS input generation were described previously () and are essentially unchanged in the current version of the software. Since their initial development, these atom reconstruction options have been greatly improved through a number of bug fixes and code optimization, robust support for separate biomolecular chains, and improved chain termini optimization. The following sections describe modified and new elements of the PDB2PQR pipeline. sub #text

Analyses of atomic interactions in tertiary structures of proteins contribute richly to our understanding of sequence–structure relationships, structural basis of protein stability and protein evolution. Studies on interactions between sidechains are commonly used in designing methods and identifying strategies for remote homology detection, protein fold recognition, protein structural comparisons and comparative protein modelling (). For example, remote homology detection between proteins can rely on conservation of structural motifs involving interacting sidechains (). Such studies also serve as guidelines in designing site-directed mutagenesis experiments () and in the understanding of the basis for residue conservation in homologous proteins (). Interactions between subunits of multimeric proteins and interactions between interacting protein modules are also areas of intense study (). Analyses on nature of sidechain–sidechain interactions across interacting interfaces between protein modules have enlightened us on evolutionary conservation of protein–protein interactions and in distinguishing transient and permanent complexes (). Different kinds of interactions have been noted in the stabilization of tertiary structures, quaternary structures and assemblies of proteins. Roles and importance of interactions between apolar residues and hydrogen bonds are very well known (). Importance of interactions such as aromatic–aromatic, aromatic–sulphur, cation–π and ionic interactions in the structure and function of proteins is also well realized (). Observations and analyses of less common features such as exposed cluster of hydrophobic residues, partially buried salt bridges and interactions of buried charged residues are also of specific interest () Here we report the development of a web-based service, PIC (Protein Interactions Calculator), to aid recognition and analyses of various kinds of interactions in tertiary structures of proteins and structures of protein–protein complexes. We have also integrated solvent accessibility calculations in PIC to aid recognition of interacting motifs that are exposed or buried. Further, the residue depth calculations are also made possible in PIC so that interactions deep inside the protein structure or near the surface can be recognized. Advantage of using residue depth parameter is that it can distinguish residues with no solvent accessible surface area in terms of how deep they are from the protein surface. PIC server accepts atomic coordinate set of a protein structure in the standard Protein Data Bank (PDB) format. The user is prompted with selecting one or more of the following interaction types: Interaction between apolar residues, disulphide bridges, hydrogen bond between main chain atoms, hydrogen bond between main chain and sidechain atoms, hydrogen bond between two sidechain atoms, interaction between oppositely charged amino acids (ionic interactions), aromatic–aromatic interactions, aromatic–sulphur interactions and cation–π interactions. The input coordinate set is accepted, under each section of the page, for recognition of interactions within a polypeptide chain. If an ensemble of NMR-derived structures is input then the first model in the file is taken as a representative and is used by the PIC server. The output corresponds to the list of residues involved in interaction type of interest. An option is provided, using RasMol () interface and Jmol interface, for enabling visualization of structure in the graphics with interactions highlighted. It is possible to get the results by e-mail. It is also possible to download the output files of the original programs. A separate panel is available for identification of various types of interactions between polypeptide chains when a multi-chain PDB file is subjected to the analysis. All the said interactions could be explored for their occurrence across the inter-polypeptide chain interface. Thus this panel facilitates recognition of interactions between different subunits in a multimeric protein structures or between proteins in a protein–protein complex structure. show ionic interactions between oppositely charged sidechains across the interface, formed between cyclin-dependent protein kinase and bound cyclin (), recognized using PIC server. Solvent accessibility calculations could be used to identify different kinds of interactions between buried or between solvent exposed residues. Solvent accessibility calculations are performed using NACCESS program (Hubbard, S.J. and Thornton, J.M., 1993, NACCESS Computer Program, Department of Biochemistry and Molecular Biology, University College London.). The exposed and buried residues are identified by >7% and ⩽7% residue accessibility, respectively. Under this facility list of all the interaction types are displayed prompting the user to select list of interaction types of interest. For example, a user may prefer to identify interactions between apolar residues that are exposed. shows interactions between solvent exposed apolar residues, in crambin (), recognized using PIC server. Depth of an atom in a protein is defined as the distance from the nearest atom in the surface of the protein structure. Mean depths of atoms of a residue defines the residue depth (,). Analogous to the panel on solvent accessibility, panel on residue depth enables the users to identify specific types of interactions near the protein surface or deep inside the core of the structure. Based on the analysis of residue depth parameter by Chakravarty and Varadarajan () we consider those residues with depths ⩽5 Å as close to the protein surface and others as deep inside. Using this part of the PIC server it is possible to identify interactions between, say, aromatic residues near the protein structural surface. As calculation of residue depths takes a few minutes for most protein structures, results involving depth calculation are sent by e-mail to the user if a valid e-mail address is provided. Various types of interactions are recognized from the atomic coordinates using the standard criteria that are published. We used mainly the criteria suggested by NCI server () to identify non-canonical interactions in proteins. The aromatic–aromatic, aromatic–sulphur and cation–π interactions are recognized between appropriate sidechains using the criteria proposed by Burley and Petsko (), Reid . () and Satyapriya and Vishveshwara (), respectively. Disulphide bonds are recognized using the distance criteria employed originally in the MODIP program (). Hydrogen bonds are recognized using HBOND routine developed by Overington . () and described in Mizuguchi . (). The hydrogen bonds are categorized as main chain–main chain, main chain– sidechain and sidechain–sidechain. Only standard hydrogen bonds are recognized in PIC as NCI server () is available for identification of interactions such as C–H … O. Interactions between hydrophobic sidechains are identified using a distance cut-off of 5 Å between apolar groups in the apolar sidechains. Though various interactions are recognized using the standard criteria, user has an option of changing the distance cut-off in recognizing any of the types of interactions. Analysing the interactions that stabilize tertiary and quaternary structures and protein–protein complexes is a common situation in structural biology. For example, a structure just solved using X-ray analysis or nuclear magnetic resonance may be subjected to such an analysis. In protein engineering and design experiments, a good understanding of the structural roles of various residues is essential before taking decisions on residues to mutate by site-directed mutagenesis and the replacing residue. Use of combinations of features available in PIC such as various kinds of interactions and solvent exposure or buried nature or depths of residues is also expected to aid recognition of common and uncommon structural features in a given protein structure. Analysis of such interactions in homologous protein structures () enables recognition of evolutionary constraints critical for the retention of fold of the protein family. The PIC server is available at:

Bilipid membranes divide eukaryotic cells into various types of organelles containing characteristic proteins and performing specialized functions. Thus, subcellular localization information gives an important clue to a protein's function. Although localization signals in mRNA appear to play some role (), the main determinant of a protein's localization residues in the protein's amino acid sequence. (We recommend wikipedia.org/wiki/Protein_targeting for a brief overview and Alberts . () for a textbook description.) Numerous experiments to determine protein localization have been performed to date. These can broadly be classified as: small-scale experiments—the results of which continue to accumulate in public databases, such as UniProt () and Gene Ontology (); and large-scale experiments using epitope () or green fluorescent protein (GFP) () tagging, or by separation of organelles by centrifugation combined with protein identification by mass spectrometry (,). Although they provide invaluable information, the coverage of experimental data is only high for model organisms, particularly yeast. Moreover, the agreement amongst large-scale experimental data is only 75–80% (). Thus, computational prediction of localization from amino acid remains an important topic. Numerous computational methods are available [reviewed in (,)]. Some (including WoLF PSORT) have recently been benchmarked by Sprenger . (), who found the computational methods to be useful for sites, such as the nucleus, for which many training examples can be easily obtained from UniProt (which is the source of most or all of the training data for most prediction methods—including WoLF PSORT). The different methods they benchmarked were found to have different strengths. Here, we describe the public server for our WoLF PSORT method. italic #text xref italic #text italic disp-quote xref #text ext-link #text W o L F P S O R T n o t o n l y p r o v i d e s s u b c e l l u l a r l o c a l i z a t i o n p r e d i c t i o n w i t h c o m p e t i t i v e a c c u r a c y , b u t a l s o p r o v i d e s d e t a i l e d i n f o r m a t i o n r e l e v a n t t o p r o t e i n l o c a l i z a t i o n t o h e l p u s e r s t o f o r m t h e i r o w n h y p o t h e s e s .

Terminal restriction fragment length polymorphism (T-RFLP) analysis is a microbial fingerprinting technique capable of discriminating microbial communities quickly and relatively inexpensively (). T-RFLP is increasingly used in high-throughput studies of microbial communities in combination with or even in lieu of clone library analysis (,). Briefly, the method involves PCR amplification of a gene of interest (often 16S rRNA genes) with fluorescent dye-labeled primers, followed by multiple single restriction digests done in parallel. The resulting fragments are then separated by capillary electrophoresis with an internal size standard to determine the lengths of the terminal (fluorescently labeled) fragments. Each distinct terminal restriction fragment is considered an operational taxonomic unit (OTU), thus the choice of restriction enzymes can impact the number of OTUs observed in each sample and the calculation of diversity statistics. When analyzing uncharacterized and very diverse bacterial communities, sufficient community discrimination can often be accomplished with multiple randomly-chosen tetrameric restriction enzymes (). However, a brief review of the literature indicates that there is still no standard in even this simplified case. We examined 26 papers (,) that were published between 1997 and 2007 and used T-RFLP. Of those papers, 38% used universal bacterial primers combined with a single restriction enzyme, but the choice of enzyme was not consistent. MspI was used most frequently (four studies), followed by TaqI (two studies), and one study each used AluI, CfoI, HhaI and HaeIII. Overall, only three of the 26 papers included a rationalization of enzyme selection (,,). An alternate approach to T-RFLP can be taken if the microbial community has been characterized (by clone library analysis or by prediction from previous studies) or if a particular taxonomic group is being targeted with specific primers. In this case, a more reasoned choice of restriction enzymes can be conducted. In particular, specific species or microbial taxa of interest to the researcher—particularly closely related taxa that may share some restriction sites—can often be differentiated if the proper restriction enzymes are selected. There are, however, few resources available to narrow down the selection process. Over 600 Type II restriction enzymes are commercially available, accounting for 262 distinct specificities (). Existing computer programs for assisting in the choice of restriction enzymes include TAP-TRFLP (), MiCA Enzyme Resolving Power Analysis () and TRF-CUT (). These programs perform restriction digestions of a predefined sequence database or user-provided sequences, but these results must still be manually examined to determine which enzymes are best suited to discriminate that set of sequences. CLEAVER (), a stand alone program, provides the above features as well as the ability to assign sequences to taxonomic groups at multiple levels and to search for enzymes that cut one group but not another group. However, it is limited to comparing only two groups at once. Restriction Endonuclease Picker (REPK) addresses this gap by finding enzymes that are able to discriminate an unlimited number of user-designated sequence groups on the basis of their terminal restriction fragment lengths. If no single enzyme can discriminate all groups, REPK reports sets of four restriction enzymes that together are able to differentiate the groups of interest. An important component of REPK is this ability to specify the taxonomic rank of sequences to be differentiated, which is particularly useful in the case where a diverse microbial community has been characterized by clone library analysis or there is an existing database of several subgroups of sequences that amplify with the same specific primers. xref fig #text W e f o u n d t h a t r e s e a r c h e r s o f t e n f a i l e d t o r e p o r t t h e i r r a t i o n a l e i n c h o o s i n g a p a r t i c u l a r s e t o f r e s t r i c t i o n e n z y m e s f o r T - R F L P a n a l y s i s , y e t t h i s c h o i c e i s c r u c i a l f o r r e s o l v i n g t h e m i c r o b i a l c o m m u n i t y a n d i n t e r p r e t i n g t h e r e s u l t s . W e p r o v i d e R E P K i n t h e h o p e t h a t i t w i l l a l l o w m i c r o b i a l e c o l o g i s t s t o m a x i m i z e t h e i r a b i l i t y t o d i s c r i m i n a t e t e r m i n a l r e s t r i c t i o n f r a g m e n t s o b t a i n e d d u r i n g T - R F L P a n d t h e r e b y t a k e g r e a t e r a d v a n t a g e o f t h i s p o w e r f u l c o m m u n i t y f i n g e r p r i n t i n g t e c h n i q u e .

Protein interfaces are drawing much attention of the structural bioinformatics community as well as the rest of the biological world. Many articles have been published classifying complexes according to function, and analyzing the properties that characterize them. Several prediction engines have been developed in order to analyze interfaces, and predict their location for the various interaction types (). The ongoing discussions rising from the literature show large divergence concerning basic aspects. This appears already at the level of how interfaces are defined (change in accessible surface area or various cutoff distances either between heavy atoms or Cα atoms). Further it goes through the definition of successful prediction, which is measured at the level of proteins, amino acids or predefined surface patches (). Finally, it concerns basic issues such as the role of evolutionary conservation in binding that is still controversial (,,). We contributed to this effort, through the development of ProMate (), a protein-binding sites prediction server. ProMate uses various structural features measured on unbound proteins to identify potential binding sites. Thirteen different properties were examined in ProMate, and using a reduced brute-force optimization, a subset of nine of them was selected to be counted in the final prediction. At this point, having examined many different alternatives, the field of binding-site prediction has matured to be able to converge to common guiding definitions that are considered most suitable, and are needed in order to focus on the goal itself. A community-wide effort is required for this to be accomplished. Moreover, as research evolves, new insights can improve the prediction. In some cases, the original motivation might not be directly related to the prediction of protein-binding sites rather provide an independent measure related to proteins. In others, the information regarding binding sites location can amplify the significance of a new result. In any case, the value of having a simple tool for testing the relevance of new features to proteins’ binding potential is evident. To this aim, we leveraged ProMate into a generic protein-binding sites analysis web tool, ProMateus. Here, we present this tool and its utilization for three types of results that can be drawn from this type of analysis: improving ProMate's prediction accuracy, extending ProMate for the prediction of protein DNA binding sites, using a relevant training dataset, and for the comparison of a newly suggested definition of secondary structure compositions of proteins based on interaction networks to traditional secondary prediction methods for their binding site occupancy. However, the real goal of ProMateus is to promote a new idea of open research with ProMateus providing an open web tool that facilitates the examination of features that are relevant for binding sites prediction. In the first phase, ProMateus uses a simple filtering scheme. A histogram is produced, presenting the distribution of the feature values at interfaces versus the rest of the surface. For categorical features, a bootstrap procedure is used to evaluate the 70% confidence intervals of each category. The histogram of continuous scores is assigned a -value using the Kolmogorov–Smirnov test. As an alternative, the log file also presents the -value evaluated from the Pearson's correlation coefficient. If the suggested feature passes this filter namely, the curves are significantly different; ProMateus uses a logistic regression (LR) optimization in 5-fold cross-validation of the new feature together with ProMate's original properties. To simplify the model, the weights assigned by the optimization are limited to the range, thus no complicated dependencies between features are allowed. Due to the equivalence of ProMate's scoring scheme to LR (explained in the Methods section) this would generally be classified as an embedded scheme. However, note that the LR procedure differs from ProMate by the fact that it acts on the space of the surface dots (see Supplementary Data for detailed description of methods), and not yet at the level of the proteins. The results of re-estimating the features used in ProMate shows some disagreement with its original feature selection. One score is the probability distribution of the different atom types at interfaces. Second is the preference of interfaces to be populated by longer ‘loops’, i.e. unstructured flexible regions of the protein. Also, the sequence distance that is the distance between residues along the peptide chain that tends for the longer distances at interfaces was found significant. Finally, the number of bound water molecules proved to be higher at interfaces already at the unbound structure. All these features are described in detail elsewhere (). Using these four scores with a weight of 1, the prediction improves from 36 correct predictions out of 51 predictions produced, to 38 out of 55. Thus, an increased coverage was achieved. In recent years, several other features were claimed to be significant for interface recognition; among those are improved evaluation of evolutionary conservation (such as WHISCY () and conSurf ()), the distance of each atom from the center of mass in enzymes (), and high-frequency vibrating residues (). All these features were tested by ProMateus. Specifically, WHISCY was run through its web server, limited to the calculation of the evolutionary score alone, to avoid overlap with the AA propensities which already exist in ProMate. ConSurf was run through its web server, with all the default values except homologs that are collected from UniProt. The distance of each atom from the center of mass was calculated both at the level of atoms, and at the level of amino acids. High-frequency vibrating residues were extracted from the iGNM database () taking the residue mean-square fluctuations driven by the joint contribution of the highest 10 modes. All these features failed before the last phase, namely, though they contain relevant information to interfaces, they do not improve the interface prediction within the suggested model. This conclusion might be inaccurate for the feature of the distance of the atom from the center of mass since the authors claim it should only apply to enzymes, which are a small fraction of the database used. The result of ProMate together with each of these features is presented in . A predicted interface patch is extracted from the full range of predicted interface probabilities. The bounds for this are optional parameters in ProMateus, allowing the user to experience with a full range of bounds, determining the sensitivity and specificity of the prediction. Throughout our analysis, a prediction is defined successful if it is reliable, namely, if at least 50% of the predicted interface patch is truly so. The number of successful predictions achieved over the proteins is presented in . We found a hard core of 20 proteins that were predicted by all the different combinations, 13 proteins that were not predicted by any of the combinations and 24 proteins that were predicted correctly by only some combination. Excluding the distance from the proteins' center of mass, one can see that the scores are overlapping, and the differences are within the noise. Moreover, one clearly sees a difference in the success rate versus coverage. The re-optimized ProMate has the highest coverage (0.96) but a success rate of only 0.69 (resulting in 38 correct predictions), while ProMate + WHISCY has a coverage of only 0.77 but with a success rate of 0.75 resulting in 33 correct predictions. Which of the two choices is better depends whether one needs high coverage or a higher success rate. The strongest property that characterizes DNA-binding proteins is their positive electrostatic potential used to attract the negatively charged DNA. Studies also incorporated sequential properties such as evolutionary conservation and the frequency of favored residues like lysine and arginine, as well as structural properties of surface curvature and accessible surface area, together with the helix-turn-helix motif that is abundant in DNA-binding proteins (but is not limited to them) (). To demonstrate the great advantage in the simplicity of ProMateus, we utilized the available features used in ProMate and applied them to a database of protein-DNA interfaces. Due to its importance, a simple electrostatic score was added. This potential was constructed by simulating a negative charge at the center of every circle on the protein surface using the program () (see Methods section). Since the available unbound database was limited to six proteins, we replaced it by a database of simple models constructed by the fast calculation option of ModWeb, the web server running MODELLER (). ModWeb assigned each protein from the bound database a template from the PDB, based on sequence similarity. Then, a structure was constructed based on this template and a force-field optimization. As the number of water molecules and temperature factor distribution are biased in case of the bound structures, and unavailable for the models, both were excluded. The role of secondary structures at interfaces has been discussed previously, raising many contradicting conclusions. Jones . () found α-helices to be favored at interfaces. Gutteridge . () used it in a neural network aiming to predict interfaces and found it had a low weight in the prediction. In a recent article, Hoskins . () showed that β-strands that participate in protein–protein interactions exhibit characteristics similar to internal strands rather than regular edge strands, and used this property for interface prediction. In the analysis of ProMate we found interfaces to be richer in β-sheets and poorer in α-helices, in cases where both these structures appear at the same protein. Thus, there is still an uncertainty about the significance that should be associated with the secondary structure in this context. The classification to secondary structure is an example for a characteristic that could be misleading. Ascribing an amino acid to one of the classes is strongly dependent on its sequence neighbors, and indeed the most popular secondary structure prediction algorithms are based on sequential relations, e.g. Hidden Markov Models. Therefore, we re-analyzed this property more carefully, at the protein level, using hierarchical bootstrapping. The resulting picture is somewhat different (). In addition, the sampling at the level of the protein instead of the amino acid widens the confidence intervals, to an uncertainty level. As a counterexample, one can consider the amino acid distribution that does not differ between the two ways of analysis (data not shown). In comparison to the traditional definition of secondary structures discussed above, we examine a new definition of secondary structures suggested recently (). While the common definition used in PROMOTIF () is based on predefined angels between consecutive amino acids, the definition suggested by Raveh . is based on spatial features extracted by clustering the proteins contact map defined by the backbone and hydrogen bonds. Comparing the distributions based on the two definitions shows that the latter is superior from aspects of the protein function. Class number 4, which is associated with a subset of the loops, shows a significant preference for non-binding surfaces, while no class shows a significant difference with the traditional definition. Thus, the new definition is more relevant from aspects of the protein function. The internet revolution dictates a communal way of research. The simplified communication in the ‘global village’ increases the creation of new knowledge and its utilization around the world. This is in fact one of the driving forces of bioinformatic research. Many servers that provide various scientific services have been established, and are used as daily scientific tools. The vision lying at the base of ProMateus suggests taking this community-research approach one step further. The industrial community has already acknowledged the advantages of the open source model of system development, in which portions of source-codes are freely distributed by individuals and companies from around the world. In addition to a fast development rate, such projects are considered superior in contribution to world standards, in improved project modularity and even from financial aspects. Inspired by this, ProMateus is an initiative demonstrating the open research approach. The potential of such tools to advance specific areas of research is tremendous, and suggests a way of worldwide research communication that up until now was only available though important contest projects, such as CASP and CAPRI. The open-research approach differs from the open source by one intrinsic complication that should be acknowledged. When testing many features over the same limited data, the probability of overfitting, i.e. that a random feature would be found significant by chance increases. However, this should not prevent the development of such projects. Employing careful filtering (and cross validation) and updating the available databases should restrain these effects. Taken to the extreme, consider all the structural bioinformatics labs working on one database—the PDB. Applying a careful analysis would enable to gain from a worldwide research effort. Yet, one has to keep in mind the theoretical limitations of such a system, and consider the results carefully. In the context of binding-site prediction, new features gaining from the expertize of different labs can be easily checked and incorporated into the exiting framework. Another aspect is the creation of standards. One of the hardest problems in the field is the disagreement over the basic definitions, such as those of the binding site itself, and of a successful prediction. Having future similar tools, the inherent natural selection of the web will enable the objective comparison of the various tools and definitions and the convergence to the most promising ones. The goal of enhancing the understanding of protein recognition is by itself interesting, and important for applications that are not directly related to binding-site prediction. The comparison of the alternative definitions of secondary structures demonstrates this. This article is the first known to us that analyzes DNA-binding site of protein models. Due to lack of unbound structures, studies in the field usually focus on analyzing the bound structures. Some of them use a small unbound dataset as supporting evidence. Homology models are a neglected class, though these occupy most of the currently available structure space. The goal of predicting the location of the binding site on modeled structures might be more complicated than unbound structures but is certainly worth pursuing. The results in this article showed that the electrostatic potential, considered most important for DNA-binding site identification, significantly loses its strength to a more general chemical character representation when only a model is available. A comparison to a model built on an unbound template is expected to yield further insights. An important issue that is raised in this article is the alternative ways of analyzing the same property. The analysis at the protein versus AA level exemplifies a problem that most probably rises in many similar studies. Two different sampling models are suggested, and can sometime lead to different conclusions. There is no strict answer regarding which level of analysis is superior. The analysis at the protein level reduces the effect of intra-protein dependencies, but at the same time loses confidence due to smaller sample size. On the other hand, taking the mean over all the proteins removes the bias in favor of larger proteins, but will ascribe higher weight to outlier proteins. When analyzing a new property, one should carefully examine both options, and choose the one which seems more appropriate for the specific case. To conclude, the main objective of this article is to export the idea of ‘open-research’ into a simple, user-friendly web-based server. Using this idea, the bioinformatics research can leverage above small independent projects and evolve towards fewer centralized worldwide cooperations that integrate different modules contributed by labs around the world. We believe that due to its nature, biological research that often requires high expertize in a small-scale area of research would gain significantly from this approach. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

During the past years there has been a wide interest in studies of specificity-determining residues within protein subfamilies (). Consequently, an increasing number of methods and web applications has become available that offer functional analyses of subtype specificity from multiple alignments (). Previously we evaluated advantages and limitations of several of the state-of-the-art methods and introduced a new method called Sequence Harmony () (). accurately detects positions within an alignment that are responsible for functional differences between two protein subfamilies. To further facilitate the use of for a broad audience, we have implemented a comprehensive web server. The web server offers a fast, one-step analysis with a number of options, yielding results that can be interpreted easily. In this article we will guide the user through all the steps of the web-application by means of a biologically relevant example. We will look for subtype specific sites for the two subfamilies of the alternative oxidases (AOX) protein family of plant alternative oxidases. The subtype specific sites found are the best candidates to explain the functional differences. Other relevant applications of this method include pathway specificity, ligand specificity, host-specific viral infection, viral disease progression differences and viral drug resistance. The SH method, as described previously (), is currently implemented as an awk program. The main steps are as follows. Sequences are read from the alignment and separated into two user-specified groups. For each group separately and combined entropies () are calculated. values are calculated as , with = − ∑ ( + ) log ( + ), i.e. using the sum of the normalized frequencies of groups A and B. values range from zero for completely non-overlapping residue compositions, to one for identical compositions. Next, sites are selected that have a score below a cutoff. Stretches of neighbouring selected sites are identified and the size of each of these stretches is assigned to the sites as the `Rank'. Finally, selected sites are sorted on i) increasing , ii) decreasing Rank and iii) increasing entropy. This sorted list of selected sites is the primary result of the algorithm. Currently alignments of up to about 5 million residues (including gaps) can be processed with runtimes on the order of tens of seconds, excluding the generation of the high-quality output PyMol image. The separate chains of the optional protein structure (PDB file) are aligned with the input alignment using the `profile' option of Muscle (). Subsequently, the chains with the highest average SH score with respect to the input alignment are selected for graphical display as an image in an interactive Jmol applet. The web server contains basic and advanced options. Sanity of the input options is checked, e.g. whether an input alignment is uploaded or pasted, and not both, whether it is proper FASTA format. Problems are reported with an error message and offending input fields are highlighted. An example screenshot of the main page and input form is shown in . The main input is a multiple sequence alignment of the protein family and a subdivision into two groups. Sequence data in FASTA format is case-insensitive and may be split over multiple lines; gaps should be represented with a dash (`–'); and lines beginning with a semi-colon (`;') are ignored as comment. From the alignment the program automatically generates a list of sequence IDs. The user defines the two input groups by selecting the identifier of the first sequence of the second group. By default, a cutoff of 0.2 for the score is used. Advanced features allow more control over the analysis and output, but the default settings usually suffice for a basic analysis. The cutoff can be adjusted between zero (allowing no compositional overlap) and one (allowing full overlap). A higher cutoff will lead to the selection of a larger number of sites. A reference sequence can be selected from the alignment and a starting position can be set to provide a reference numbering in the output tables. Additionally, a PDB identifier can be specified by its four-letter code, to visualize the selected sites in the protein structure. Alternatively, a PDB file can be uploaded. The PDB file is automatically aligned with the multiple alignment, or manually when the reference position is set. The output is a sorted table and optional graphical representations of the selected sites as an image (generated by PyMol, ) and as an interactive Jmol applet (). The Jmol applet includes buttons to set the SH cutoff for displaying selected sites, and to show or hide non-aligned chains in the PDB file. An example screenshot of the main output page is shown in . An additional, non-sorted, table is available that lists all sites in the alignment. The tables are coloured from red (predicted subtype specific sites) to blue (sites not predicted). The cyanide-insensitive plant alternative oxidase (AOX) is a mitochondrial non-haeme quinol oxidase (). An established function of AOX is thermogenesis in the spadices of lilies, but little is known about AOX function in other tissues (). AOX is encoded in two discrete gene subfamilies (). AOX1 is found in monocot and dicot plants and is induced by stress stimuli (,). AOX2 is usually constitutive and has at present not been found in monocot plants (). We retrieved 31 AOX sequences from the NCBI non-redundant database (); 24 of the AOX1 and 7 of the AOX2 subfamily. Sequences were aligned using PSI-Praline (,) using secondary structure prediction information from PSIPRED (). The alignment is available as Supplementary Data. Lacking an experimental AOX structure, we constructed a homology model using of the catalytic iron-binding N-terminal domain. Swiss-Model () was used to align the AOX1 sequence with the template structure 1AFR (), following the description of Andersson and Nordlund (). The PDB file is provided as Supplementary Data. The multiple alignment obtained was uploaded in the appropriate box of the web server and the first AOX2 sequence (AOX2_Arab_th) was selected as the first sequence of the second group (). The default cutoff of 0.2 was used, meaning that partial overlap in residue composition is allowed. We chose AOX1 from as reference sequence so that the subtype specific sites obtained will be numbered accordingly (). The PDB file of the model structure was uploaded to the web server. The results page of the SH server () shows at the top a summary of the input parameters, and optionally the graphical representations of the protein structure. Below that is the table containing all sites with a score (SH) below the cutoff value (default 0.2). These are sorted first on increasing score, then on decreasing rank (Rnk, number of neighbouring sites below the cutoff) and finally on increasing total Entropy (AB) of the alignment position, as described previously (). Position Ali and Ref show the sequence position in the alignment and reference sequence, respectively. The `Consensus' columns give all residue types present in groups A and B, respectively, in order of decreasing frequency and in lowercase when the frequency is less than half of the highest. In addition, a link is provided to the raw table, that holds all information of the complete alignment without selection or ranking. The raw table is available as Supplementary Data. For the AOX family we found nine sites with a value of zero (bright red in ). An additional eight sites have SH values between zero and 0.2 (from light red to white). In the `Rnk' column, a value of 2 signifies a stretch of two consecutive sites with SH values below the cutoff (alignment positions 244–245 and 121–122). Interestingly, out of nine sites with zero harmony, four have non-zero entropy in at least one group, i.e. they are not totally conserved within both groups (alignment positions 136, 245, 269, 354). This is also reflected in the `Consensus' columns, which show multiple residue types occurring. The graphical representations of the protein structure show the predicted sites in our homology model of the C-terminal domain (positions 182–352), see and also . This domain contains two sites of zero harmony and six of low harmony. Four low-harmony sites on the upper-left (Phe224, Val227, Ala228 and Asn284) are close to the Glu and His iron binding residues () (), and could conceivably play a role in ligand binding or activation mechanisms. Three other low-harmony sites, Arg199, Tyr249 and Gly252 are located around a putative membrane-binding region (), and could conceivably play a role in modulating the protein-membrane interactions or substrate access. The functional prediction of from the multiple alignment seems to be consistent with the structural prediction using homology modelling, although a different structural model could lead to different conclusions. Our SH web server identifies putative subtype specific sites based on an alignment and selection of two groups as input only. In addition, sites can be easily linked to structural information using a structure from the PDB directly, or, as shown in our example, using a homology model of the protein. Using this structural information, selected subtype specific sites can be grouped into spatial clusters of sites that are likely to share functional relationships. The combination of the SH algorithm and protein structural information has yielded a useful tool to interpret multiple sequence alignments, and to guide subsequent selection of interesting sites for experimental investigation. #text

Tumor markers are widely used to detect cancer and to monitor cancer progression. They can be grouped into markers that are identified in cancer cells and markers that are secreted into body fluids. To perform early stage cancer diagnosis, the second group of markers is more appropriate. A promising method that allows minimal invasive tumor diagnosis based on markers is mass spectroscopy. Matrix-Assisted Laser Desorption and Ionization (MALDI) mass spectroscopy evaluated by ‘peak probability contrasts’ revealed an accuracy of around 70% for ovarian cancer (). Similar approaches for pancreatic cancer performed slightly better with 88% sensitivity and 75% specificity (). Tumor antigens in blood sera represent an alternative approach for minimally invasive cancer diagnosis. A popular example is the prostate specific antigen (PSA) that is widely used in the diagnosis of prostate cancer (). Since PSA is also present in the blood sera of 33% of unaffected people, PSA as a single tumor marker shows a lack of specificity. Likewise, other single antigen markers including CA-19.9 (pancreatic cancer) and CA-15.3 (breast cancer) show severe limitations (). Recent studies strongly indicate that antigen marker sets significantly improve the specificity and sensitivity of cancer diagnosis compared to single antigen markers (). Our Minimally Invasive Multiple Marker (MIMM) approach for meningioma () e.g. is based on 57 meningioma-associated antigens. Meningiomas are frequently occurring, generally benign intracranial tumors that are grouped by the World Health Organization (WHO) in three grades, grade I (common type), grade II (atypical) and grade III (anaplastic) meningioma. On a data set of 183 seroreactivity profiles from 83 meningioma and 90 normal sera, MIMM reached a specificity of 96.2% [95% confidence interval (CI) = (96.0–96.5%)], sensitivity of 84.5% (95% CI = 84.3–84.8%), and accuracy of 90.3% (95% CI = 90.1–90.4%). The area under the receiver operator curve (AUC-value) was 0.957 (95% CI = 0.956–0.957%). We developed a web-based application, called ‘Seroreactivity Profile Classification Service’ (SePaCS) that gives experimental groups easy access to several supervised statistical learning approaches for classifying seroreactivity profiles. The results of SePaCS are summarized in an easy interpretable table that contains for each seroreactivity profile and each classification method the predicted class label. Our tool also provides a detailed result file containing for example, graphical representation of computed results. We demonstrate the capabilities and the ease-of-use of our web-based application on the example of meningioma. We tested a variety of supervised learning methods on a meningioma data set. The approaches that yielded the best results were ‘Naïve Bayes (NB) Classifiers’ (NB), ‘Support Vector Machines’ (SVM) with radial basis kernel functions (), ‘Linear Discriminant Analysis’ (LDA), and ‘Diagonal Discriminant Analysis’ (DLDA). Since further evaluations on glioma autoantibody profiles confirmed the results, these statistical learning approaches were included in SePaCS. If the WHO grades of cancer sera in the data sets are also provided by the user, SePaCS additionally offers the possibility to predict the WHO grade of these sera using a modified ‘NB Classifier’ (NBC). All statistical computations are performed using the R language (). The mutual information (MI) is a well-known measure in information theory introduced by Shannon (). The MI of an antigen and the disease state (for example, cancer and control) represents a measure of the information content that provides for the classification task, i.e. the disease state. Given two random variables X and Y, the MI I(X,Y) is a measure of the reduction in uncertainty about X due to the knowledge of Y. In our case, X and Y are binary random variables. The two possible states of the random variable X are ‘normal’ (X = 0) or ‘diseased’ (X = 1). In our application, the binary random variable Y represents the occurence of the antigen s, i.e. Y can take the states ‘s not detected’ (Y = 0) or ‘s detected’ (Y = 1). Thus, we can consider I(X,Y) as the reduction in uncertainty about the disease state due to the occurrence of antigen . The higher the value of the MI of antigen s, the more ‘valuable’ is for the classification task. Variable selection is a widely used machine learning technique to increase the performance of classification methods by focusing on a subset of relevant features. Basically, two methods for feature subset selection exist, filter and wrapper approaches (). Filter approaches perform the subset selection as a pre-processing step independent of the classification algorithm (). One disadvantage of filter approaches is that two correlated features may be both included in the selected subset. In contrast, wrapper approaches conduct the search for an appropriate feature subset using the classification algorithm as part of the function for evaluating variable subsets (), avoiding the problem of correlated features. Thus, in general wrappers should be preferred over filter approaches although they are computationally much more demanding. We compared different wrapper and filter approaches that revealed comparable effectiveness. The subset selection method that showed the best performance is based on the so called ‘MI’ (). By computing the MI of an antigen and the class label (0 for normal and 1 for cancer sera), it allows for measuring the diagnostic information that the antigen provides. We use a greedy algorithm that adds in each step the antigen that provides the highest MI. Using 10-fold cross validation we determine the subset that shows the lowest error rate for classification. SePaCS offers two different modes of operation. In the first mode, usage of own training data, the user can upload two antibody profile sets, a training and a test set. In the second mode, no training data set has to be provided. Instead, classification methods that are already trained on our data sets can be applied to the uploaded antibody profiles. Currently, SePaCS provides trained classifiers for meningioma. Similar models for other cancer entities will be available soon, starting with predictors for gliomas (manuscript in preparation). Additionally, we plan to provide classification methods trained with autoantibody profiles of prostate cancer patients, nephroblastoma patients, and patients with lung cancer. In both operating modes the results of SePaCS are summarized in tabular form, i.e. for each classification method and each test sample the result table contains a ‘1’ if a tumor is predicted and a ‘0’ otherwise. The web-interface of SePaCS is implemented in Perl and consists of three modules: parameter specification, data upload, and data processing and output. The required parameters can be specified using the web-interface. On this interface, the user has to select the operating mode and has to choose at least one of the offered classification methods. At present, the user can choose from the following statistical learning methods: four different versions of a NB Approach, SVM with a radial basis function kernel (), LDA and DLDA. In the second step the data is uploaded. If the user intends to train the selected statistical learning algorithms with his own training set (operating mode one), he can optionally assign names to the antigens. The antigen names have to be separated by a semicolon. If no names are given, the antigens are numbered. Afterwards, the user has to upload the training data that should be imported as a matrix of size   (  ), where each of the rows represents one serum. The first column denotes the class label, i.e. a ‘0’ for each normal serum and a ‘1’ for each patient's serum. Each of the following columns represents an antigen. The matrix entry  [  ] contains the information whether antigen has been detected in serum ( [  ]  ) or not ([  ]  ). The entries of the data matrix are delimited by white spaces. The described format allows for an easy data-upload by ‘copy and paste’ from spreadsheets. An example for a training data matrix M is provided in the supplemental material. In both operating modes, the antigen profiles to be classified are uploaded next. If sera are to be diagnosed, this data matrix is expected to be of dimension   , i.e. the matrix has one row per serum and one column per antigen. These sera can also be named. If sera names are given, they have to be separated by a semicolon, if no names are given, the test sera are numbered. Additionally, SePaCS offers the option to upload a data file if a user intends to use own seroreactivity profiles. For details on the data file, we refer to the SePaCS tutorial, where an example data file can be downloaded. The statistical analysis starts with an evaluation of the training data, including among others the mean antigen reactivity in cancer and control sera, the MI of single antigens, and the estimation of the classification methods’ performance using standard 10-fold cross validation. Considering the MI profile, users can easily detect the most ‘valuable’ antigens that are especially suited to perform an accurate classification. The cross-validation error rates enables researchers to assess the classification results obtained for the test set. If a data set shows a low cross-validation error, the predictions of the test data are likely to be correct. In contrast, if a high cross-validation error rate is reached (maybe due to noisy data), the classification results of the test data may be incorrect. Thus, the first part of the statistical analysis facilitates the interpretation of the data set. Thereafter, the supervised statistical learning methods are applied to the antibody profiles and the classification results are provided on a web-page starting with a summarizing table. This table shows the output for each classification method and each antibody profile. Positive predictions (tumor) are colored red and negative (normal) predictions are colored green. An example of a table is shown in . The web-page additionally contains links to supporting plots, e.g. MI profiles. Besides this summary page, details of the analysis are provided as PDF report. This report can be either downloaded or accessed online via a unique job ID. For example, the report generated with the meningioma data set described in ‘Results’ can be accessed by using the job ID 000001. The PDF report is divided into up to five sections, depending on the chosen parameters. In the first section, a summary of the analyzed data set is presented, including images of the data matrices as well as basic statistics of the training data set. For example, the mean antigen reactivity of healthy and diseased sera and a balloon plot of the antigen distribution are shown. An example of such a balloon plot for the meningioma data set is given in . The second section contains the classification results for each serum and each classification method. For some of the statistical learning methods, as the NB approaches, additional graphical output is provided. Here, the quotient of the probabilities that a serum is a normal serum and that the serum is a cancer serum is plotted. An example of such a plot is shown in . Test and training set are divided by the vertical blue line, and all sera above the horizontal green line are classified as cancer sera. If a subset selection based on MI has been performed, the next section presents the MI of all antigens. This section also contains the performance of the classification methods that have been evaluated on the training data using 10-fold cross validation as function of the subset size. Additionally, the classification results computed with the shrunken subsets are provided. In the last section, all available classification results are summarized in tabular form. The functionality of SePaCS is demonstrated exemplarily on a meningioma data set including a total of 183 sera (90 meningioma patients and 93 controls). These sera have been screened for reactivity against 57 antigens that are known to be meningioma associated (). For example, we divided the data set such that classification was performed for 10 randomly selected sera based on a training data set of 173 sera. The pdf report for this example can be reviewed on , using the job ID 000001 and is also available in the supplementary material. The analysis of the training set showed that the mean antigen reactivity in meningioma sera (20%) is significantly higher than in control sera (11%). The antigen distribution shown in reveals that in meningioma sera up to 53 of the 57 considered antigens have been detected, whereas in normal sera only up to 41 antigens have been found. Out of the 57 antigens 8 antigens react only with meningioma sera, but not with control sera as detailed in . In , the MI of all antigens is provided. The diagnostic value of each antigen can be directly compared to the value of all other antigens. The antigen with the highest MI (0.2) is NKTR that occurs in 31 of 87 meningioma sera (37%), but not in a single control serum. NKTR represents the last antigen in the first column of the balloon plot in . The antigen with the second highest MI (0.14), NIT2, is detected in 3 of 85 control sera and in 32 of 87 meningioma sera. The two antigens providing the highest MI are highlighted in . shows the cross validation error rates of the training set for the subset selection together with the respective subset sizes. This information enables the user to judge the predictive power of the different classification methods regarding his data set. In our example, best performance is obtained with the NB Classifiers, whereas LDA and DLDA show significantly decreased performance. As shown in , the first five sera that stem from control persons are classified by all prediction methods as non-meningioma sera. With the exception of LDA (without subset selection) that misclassifies one meningioma serum, all meningioma sera have been correctly predicted. We also tested the server with seroreactivity patterns of 95 glioma sera versus 82 control sera. The results were of comparable quality as the results computed by using meningioma antibody profiles. SePaCS grants non-experts in the field of statistical learning easy access to a comprehensive analysis framework for classifying seroreactivity profiles. Our tool offers the possibility to analyze seroreactivity profiles not only from different tumor entities, but from a wide variety of other human diseases that trigger a complex immune response, e.g. autoimmune diseases. Although, SePaCS was designed to analyze autoantibody profiles, it can be used for any kind of binary data. The easy usage of our statistical framework and its diagnostic value have been demonstrated with a meningioma data set. A second test with glioma seroreactivity profiles showed a similar performance. Currently, we are preparing analyzes of seroreactivity profiles for other tumor types.

xref #text FRalanyzer (Fold Recognition alignment analyzer) is a web tool, aimed at visualizing and annotating sequence–structure alignments. Functionally important residues are displayed and highlighted if they are conserved in the alignment. Functional residues in a query sequence of unknown function are identified based on the conservation of functionally significant residues in sequence–structure and structure–structure alignments. Thus, the input to FRalanyzer is an alignment, obtained independently (i.e. not a part of FRalanyzer). To facilitate the human interface and to avoid the need of cutting and pasting of alignments, FRalanyzer can directly retrieve the alignments from other tools. For FR sequence–structure alignments, a user can enter the job ID of the (independently generated) FR results from the Bioinfo Meta server () or the FR Inub server (). FRalanyzer displays a copy of the FR server results with an additional column labeled as ‘Fralanyzer’ which consists of the PDB codes of the templates identified by the FR server. A sequence–structure alignment is defined by selecting one of the PDB codes. For structure–structure alignments, FRalanyzer accepts the results of the VAST server, which are displayed so that the user selects one of the templates as before. FRalanyzer displays the selected alignment along with a number of options. A sample output is shown in . We submitted the conserved hypothetical protein EF3048 to Bioinfo, which returned the Bioinfo ID ‘49416’. We used this ID as input to FRalanyzer (see for the initial output window as well as our online documentation files), and selected the first template (PDB 1w1a). shows the main FRalanyzer output. Along with the alignment, the sequence identity % and the secondary structure identity of the aligned query and template are shown (14 and 70%, respectively). Identical residues and identical secondary structure states in the query and template sequences are highlighted. The secondary structure of the query and template are obtained from the FR server and/or PDBSum (). Further automatic annotation of the alignment can be obtained from functional features of the template from the PDBSum and the SwissProt databases (). PDBSum functional features include ConSurf residue conservation, active site residues, Prosite motifs, contacts to DNA, ligands, metals and H-bonds to ligands. Users can select any of the available features or all. Matched Prosite motifs are highlighted in the target sequence in bold red fonts. There are two options to obtain PDBSum features: ‘To Bring’ (access the PDBSum site) and ‘Local Copy’ (access local copies instead, if they exist). In the present example, the PDBSum features, active site residues, contacts to ligand, metal and Hbonds to ligand are retrieved and the respective checkboxes are checked (). Each feature is displayed in a separate row with 1-letter amino acid of the template (if present in the template) or ‘-’s (if present in a structural homolog). In addition, the functionally important positions that are conserved in the alignment (aligned to identical residues in the query) are further highlighted with vertical bars. For example, see the template active site residues (D, G, H and H) highlighted in . Other possible annotations include: catalytic CSA, Prosite patterns, ligand type or residue conservation. For instance, the ligand types in the present example are GOL and NDG. The residue conservation of the template is displayed as colored bars according to the Consurf () coloring scheme, where the most variable positions are colored turquoise, intermediately conserved positions are colored white and the most conserved positions are colored maroon. The user can access further functional information for the template or the query from SwissProt. If selected, a BLAST search is executed to identify the SwissProt sequences that are most similar (if any). Features are extracted from the SwissProt FT lines from entries having sequence identity >90%. e d i c t i n g f u n c t i o n a l l y i m p o r t a n t r e s i d u e s o f a p r o t e i n o f u n k n o w n f u n c t i o n i s a u s e f u l s t e p i n c h a r a c t e r i z i n g i t . A n e f f e c t i v e w a y t o i s t o i d e n t i f y c o n s e r v e d f u n c t i o n a l k e y r e s i d u e s i n a n a l i g n m e n t w i t h a t e m p l a t e o f k n o w n 3 D - s t r u c t u r e . I f a n e x p e r i m e n t a l s t r u c t u r e e x i s t s f o r t h e q u e r y p r o t e i n , t h e a l i g n m e n t c a n b e o b t a i n e d f r o m s t r u c t u r a l a l i g n m e n t s . O t h e r w i s e , i t c a n b e o b t a i n e d b y e . g . s e q u e n c e s e a r c h e s o r F R . R e l e v a n t f u n c t i o n a l f e a t u r e s o f t h e q u e r y c a n b e p r e d i c t e d b a s e d o n t h e f u n c t i o n a l f e a t u r e s o f t h e t e m p l a t e . F R a l a n y z e r w a s d e v e l o p e d t o h e l p i n t h i s p r o c e s s . F R a l a n y z e r a u t o m a t i c a l l y e x t r a c t s i m p o r t a n t f u n c t i o n a l r e s i d u e s f r o m a n n o t a t e d d a t a b a s e s a n d d i s p l a y s t h e m a l o n g w i t h t h e s e q u e n c e – s t r u c t u r e a l i g n m e n t , h i g h l i g h t i n g t h e c o n s e r v e d p o s i t i o n s . T h i s h e l p s t h e u s e r t o q u i c k l y l o c a t e f u n c t i o n a l l y i m p o r t a n t r e s i d u e s i n t h e q u e r y , w h i c h i n t u r n c a n h e l p i n i t s f u n c t i o n a l c h a r a c t e r i z a t i o n . F R a l a n y z e r c a n a l s o b e u s e f u l i n t h e v i s u a l i z a t i o n o f s t r u c t u r a l a l i g n m e n t s t o k n o w n t e m p l a t e s o f n e w l y d e t e r m i n e d p r o t e i n s o f u n k n o w n f u n c t i o n . F u t u r e v e r s i o n s m a y i n c l u d e l i n k s a n d a u t o m a t i c a c c e s s e s t o o t h e r a n n o t a t e d d a t a b a s e s a n d s e r v e r s , a s w e l l a s g r a p h i c a l e n h a n c e m e n t s , f a s t e r r e s p o n s e t i m e s a n d d i s t i n c t i o n b e t w e e n d i f f e r e n t t y p e s o f l i g a n d s .

Allergy, including food allergy, is a major and increasing ailment (). The disease is strictly associated with atopy, i.e. a genetic predisposition to develop allergic immune reactions to otherwise innocuous components, generally proteins. Several forms of this disorder are described and a major one is designated IgE-mediated allergy, also known as hypersensitivity type I (). This disease involves reactions to a variety of aerial proteins typically occurring in tree, grass and weed pollen as well as proteins present in a wide range of foods. Animal dander and insect venoms can also cause disease reactions (). The establishment of allergy consists of two separate phases: sensitation and triggering, i.e. education of the immune system and the actual reaction(s), respectively. The former part involves maturation of naïve T- and B-cells into immunocompetent effector cells, as dictated by a series of complex cellular interactions (,). The type-2 helper T-lymphocyte (T2) has a key function in this process, since it preferentially promotes class switch to IgE-expression. Moreover, a variety of regulatory T-cell subsets play an essential function in the orchestration of an immunological educational procedure (,). IgE immunoglobulins can readily bind to high-affinity receptors on tissue mast cells or basophilic granulocytes. The triggering phase is commenced by renewed contact with the antigen, involving binding to cell-anchored IgE molecules and an accordingly elicited release of inflammatory substances, causing anyone or several among a range of symptoms (). Asthma, rhinitis, rhinoconjunctivitis, eczema, contact dermatitis, angioedema and abdominal pain are common allergic reactions, but anaphylactic shock—entailed to impaired respiratory and circulatory function—can also follow. A sensitized individual may also respond similarly to substances that share certain structural features with the molecule that elicited the initial immune reaction (). This phenomenon, designated cross-reactivity, is tightly connected to the epitopes, i.e. parts of an allergenic protein that are recognized by immunoglobulins—particularly IgE—or receptors present on T-lymphocytes. Broadly defined, such cross-reactivity can engage either IgE- or T-cell epitopes, but that involving IgE-binding (generally referred to as B-cell cross-reactivity) is much better understood (). IgE epitopes can occur either as uninterrupted segments of amino acid residues (continuous epitopes) or distributed as patches on the protein (discontinuous epitopes), the latter sort being brought into juxtaposition in a native (folded) protein configuration. Some common examples of IgE-type cross-reactivity are the pollen-fruit and the latex-fruit syndromes, both categories being associated with promiscuous IgE recognition due to protein structural similarity across species (,,). This phenomenon typically, but not necessarily, occurs between protein allergens from phylogenetically related species (,,). Moreover, a relatively high degree of identity at the amino acid sequence level is commonly seen between IgE cross-reactive proteins (). Nonetheless, high levels of homology without conservation of allergenicity and low degree of sequence similarity with conservation of the offending property are also reported (,). The complex mechanisms involved in allergy have prompted for several inherently different methods to safely conclude on potential protein allergenicity. Major schemes suggest a tiered set of tests involving amino acid sequence comparison (simple bioinformatics) as well as several and assays (,). Notably, bioinformatics-type inspection represents a key prescription for allergenicity testing in the subsequently adopted guideline on safety assessment of genetically modified foods and that of the European Food Safety Authority (EFSA) (,). The bioinformatics testing scheme, being an early computational design, is built to recognize both general homology-type similarity (to known allergens) and B-cell epitopes; T-cell counterparts may, though, be outside the remit of this allergenicity assessment (). Intricate relationships between amino acid sequence similarity of query proteins to known allergens and their type-I hypersensitivity potential have, however, spurred further development within this field of risk assessment. Dedicated comparison approaches in conjunction with statistical learning algorithms have enabled increasing overall performance of computational assessment methodology (). Over the last decade, a variety of Internet-based bioinformatics testing tools for protein potential allergenicity have been developed (). Recently, we reported an method for evaluation of potential protein allergenicity, designated ‘Detection based on Filtered Length-adjusted Allergen Peptides’ (DFLAP) (). This system is founded on a novel principle, which involves two main features: First, a flexible peptide-selection procedure from allergens, as accomplished by comparing allergens with presumed non-allergens, and secondly, specific education of an support vector machine (SVM) to which a query amino acid sequence can be presented. In this article, we present EVALLER, a web server-based on a DFLAP core algorithm and an intuitive interface that allows for expedient interrogation of query amino acid sequences, occurring in FASTA format. Subsequent to processing by the DFLAP machine EVALLER returns an assessment of its potential allergenicity, viewed as a rather comprehensive textual and graphical output. EVALLER should be feasible for scanning purposes and as a key part of an integrated allergenicity assessment procedure. The EVALLER server is built as a Perl front end on top of a DFLAP original core, which is implemented in MATLAB. Briefly, the DFLAP algorithm was allowed to construct a set of ‘Filtered Length-adjusted Allergen Peptides’ (FLAPs), derived by extraction from an allergen database through a process involving comparison with a data set (largely compiled from the human proteome) holding proteins with low probability of being allergenic. Selection criteria for both data sets are described elsewhere (). Amino acid sequences of both allergenic and presumably non-allergenic proteins were subsequently compared with the accordingly extracted FLAPs. Based on the resulting similarity score values for each protein, an SVM was ultimately educated to decide whether a query protein is sufficiently akin to any FLAP to be assigned as an allergen. The allergen data set (762 amino acid sequences), the non-allergen set for FLAP extraction, (52 081 amino acid sequences) and that of presumable non-allergens for DFLAP training (1524 amino acid sequences) were identical to those reported in our earlier study (). This also applies to DFLAP parameter settings in EVALLER (minimal peptide length = 22, FLAP threshold = 48, number of FLAP matches = 4 and the cost parameter C = 100 in SVM). For details on these parameters and on the algorithm, see Soeria-Atmadja . (). In its present configuration, EVALLER permits updates of both data and system. Nonetheless, efforts are underway to confer enhanced technical simplicity to this function. A decision statistic, DFLAP score, being an output parameter by SVM algorithm of DFLAP, assigns a query protein as either presumably an allergen or presumably not an allergen depending on whether the aforementioned number is positive or negative. A confidence measure accompanies each risk assessment in order to inform the user on EVALLER's uncertainty regarding assignments. This measure is derived from decision statistics of a test set (not used in the design procedure) holding both allergens and presumed non-allergens. While we needed to hold some data outside the design procedure a DFLAP system was trained using two sets encompassing 500 allergens and 1000 presumed non-allergens, respectively, employed in our earlier published report (), of which the latter data set was created by random sampling from Swiss-Prot. To achieve minimal incorporation of spurious entries into the non-allergen set, amino acid sequences shorter than 50 amino acid residues or with high sequence similarity to any of the allergens were dismissed. Compliance with the last criterion was accomplished by alignment, allowing a maximum Smith–Waterman score of 100 to be considered as a presumed non-allergen. An identical procedure was applied to compiling a test set of 1000 presumed non-allergens. These test examples are, however, easily assigned as presumable non-allergens by DFLAP. Therefore, additional examples of this category, which are more difficult to assess, were also included (104 vertebrate tropomyosins, 39 animal profilins and 14 mammalian parvalbumins). Of the total 762 allergens the remaining 262 allergens, set apart from the design procedure, were used to estimate decision statistics typical for allergens (all sequence data sets are publicly available on the EVALLER server, under ‘Supplementary Data’). Thus, decision statistics for 1157 (1000+157) presumed non-allergens and 262 allergens were estimated, using the accordingly designed DFLAP system. Uncertainty scores (US) are differently defined for allergen assignments (i.e. the decision statistic DFLAP score >0), relative to those of presumed non-allergens (i.e. the decision statistic DFLAP score <0). Both definitions are specified as the uncertainty, given that the DFLAP score served as a decision threshold instead of zero. In the former (allergen) case, the uncertainty score reflects the probability of false alarms (1-specificity), whereas the probability of overlooking an allergen (1-sensitivity) applies to the latter (non-allergen) situation: FN(DFLAP score) and TP(DFLAP score) are numbers of misclassified and correctly assigned allergens, respectively, when test statistic DFLAP score is used as a decision threshold. Analogously, FP(DFLAP score) and TN(DFLAP score) are the numbers of incorrect/correct assignments of presumed non-allergens, when DFLAP score applies as a decision threshold. EVALLER was developed to enable bioinformatics assessment of protein potential allergenicity by virtue of the corresponding amino acid sequences. In its present design, protein sequences occurring in FASTA format are accepted. The interrogation procedure, involving submission of a query amino acid sequence (one at a time), allows for user-defined options regarding presentation of results; a range of top-scoring matches and resizable views of the test sequence and matching peptides (). Detector design and classification of the test sequence are, however, not available to the user since parameters involved in these procedures have already been optimized (). As outlined in ‘Material and Methods’ section, the EVALLER decision statistic for allergenic potential has been split into two separate categories: ‘presumably not an allergen’ and ‘presumably an allergen’. An uncertainty score, indicating the confidence level of EVALLER, is also presented. It is based on the decision statistics from interrogation of a reduced DFLAP, i.e. 262 allergens and 1157 presumed non-allergens (see ‘Material and Methods’ section). An uncertainty score for a negative (presumably a non-allergen) decision statistic, DFLAP score, represents the probability of neglecting an allergen, assuming that the decision threshold had been set at the DFLAP score instead of zero. Analogously, an uncertainty score based on a positive (presumably an allergen) DFLAP score indicates the probability of false alarm, on the assumption that the DFLAP score is identical to the decision threshold. Uncertainty score numbers decrease with increased distance of the DFLAP score to the decision threshold, which is zero (). Notably, EVALLER is much more confident in its assignment to the ‘presumably an allergen’ category than that of non-allergens. The uncertainty scores are estimated for a reduced DFLAP (educated with 500 allergens and 1000 presumed non-allergens) rather than for the final DFLAP that EVALLER is based upon (762 allergens and 1524 presumed non-allergens), and may therefore be overly conservative. For clarity, the textual assignment output has been color-coded as determined by category (). Green color indicates low allergenic potential whereas red corresponds to a presumable allergen, as judged by EVALLER. A 2D graphic representation of the scanned sequence, alongside with a color-coded printout (according to the aforementioned assignment scheme) of best matching FLAPs, are presented to the user by virtue of a modified version of the Perl open source package EBioForms (unpublished). The display is dynamic and can be zoomed for detailed views of specific parts of the sequence (). Apart from presumed allergen/presumed non-allergen assignment, being integrated in the aforementioned view, EVALLER also provides information on FLAP (see ‘Materials and Methods’ section) to which similarity is identified (). Moreover, a link to the UniProt of the cognate allergen accompanies each FLAP. The server and related information is available at the following Internet address: . The EVALLER core algorithm has been extensively validated for sensitivity and false-alarm rates, using test sets of amino acid sequences (). Nonetheless, several potential allergens of the polcalcin family of proteins, being members of a widely known protein family involved in pollen–pollen cross-sensitization (), were not included in those evaluations and may therefore qualify as feasible for a simple test of EVALLER performance. Polcalcins hold two EF-hand calcium-binding domains and typically show high intra-family sequence similarity (). Presumable non-allergenic homologues to the polcalcin family include the calmodulins and calmodulin-like proteins, as well as other related calcium-binding proteins. An assembly of these proteins, derived from both plant and animal kingdoms, were mined from UniProt and interrogated for potential allergenicity by EVALLER as well as by two additional web servers—Allermatch and APPEL—dedicated to assessment of potential allergenicity (,). The former of these already reported methods is designed to apply bioinformatics testing principles outlined by the , i.e. either >35% identity in a segment of 80 amino acid residues (criterion 1) alone or in conjunction with an exact match in stepwise searches for continuous identical sequence segments (criterion 2), whereas the latter implementation derives an assessment from sequence-derived structural and physicochemical properties in combination with an SVM (,,). The EVALLER and APPEL servers assigned all calmodulins or calmodulin-like proteins as presumably non-allergenic, whereas a traditional alignment approach (35% similarity over 80 amino acid segments) gives preference to resemblance of input proteins to known allergens (). Although EVALLER and Allermatch differ in assignment on (presumably non-allergen) query proteins, both methods identified peptides (or entire amino acid sequences) in allergen polcalcins as best matches. In two cases, however, EVALLER reported differently. The top-scoring FLAP hits for calcium-binding protein-5 (bovine and human) are derived from a troponin-like protein occurring in the fish parasite . This protein, however, also contains EF-hand Ca binding motifs, typical of polcalcins, which confers reason to its identification by the EVALLER algorithm (). Furthermore, to probe for sensitivity of the aforementioned web servers, three polcalcins—neither of them incorporated in the EVALLER system—were submitted for interrogation. Two of them occur in tobacco and one (Syr v 3) in lilac (). The Allermatch database already holds these potential allergens, thereby easily finding perfect matches. Both EVALLER and APPEL, though, assigned Syr v 3 as presumably allergenic, a conclusion supported by a recent report (). The two tobacco polcalcins were also recognized as potential allergens. For the time being, though, these proteins are devoid of documentation on allergenicity in the scientific literature. Although scored as allergens by EVALLER and the two additional bioinformatics testing tools referred to above, further interrogation is needed to arrive at a scientifically justified conclusion on the allergenicity of the three polcalcin proteins. For this purpose, immunoassay analysis, involving reactivity measurement of IgE antibodies in the sera of patients with clinically validated allergic responses to other polcalcins should apply (). Other tests, such as resistance to pepsin digestion under appropriate conditions, may further improve assessment accuracy (,). Most of the hitherto accumulated literature on bioinformatics protein allergenicity assessment can be assorted into any of the following several categories: straightforward alignment (e.g. the two-part procedure), alignment-based feature-extraction combined with statistical learning, methods based on comparison to either allergen-derived motifs or reported IgE epitopes and, lastly, similarity search of entire proteins as represented by a variety of coding protocols (,,,). Moreover, by bringing a special kind of segment-reduction procedure into action on allergens, the alignment/statistical learning model has been considerably enhanced (,). Hence, EVALLER—built on a DFLAP core—exploits on a tripartite procedure, including a highly specialized filtration of peptides from allergens involving usage of the human proteome, to create material, which permits efficient education of an SVM algorithm. Each output, in response to query submission, is accompanied by an uncertainty numeral, which allows the user to appraise assignment accuracy. Additionally, parts of the query proteins deemed most akin to cognate segments in any or several among the local repository of allergens, are depicted and made available to users of EVALLER in several ways, which altogether support risk assessment. p p l e m e n t a r y d a t a a r e a v a i l a b l e a t N A R O n l i n e .

The superimposition of a set of related protein structures is a straightforward methodology for identifying structurally conserved residues that cannot be detected using sequence alignments (,). Structurally conserved regions are usually of functional significance and can also improve our understanding of protein evolution. Despite the fact that a variety of multiple structure alignment algorithms have been reported over the years [(); for a comprehensive review see ()], methods for automatic identification of variable and conserved regions within a multiple structure alignment of proteins have not been fully exploited. The approach of identifying structurally conserved residues in a set of protein structures has been already used () to improve the sensitivity and/or specificity of poorly performing PROSITE patterns (). Clearly, the construction of structure and sequence motifs is an important area, which would greatly benefit from the possibility of easily detecting residues structurally conserved in diverse protein structures. The 3dLOGO server makes accessible to web users a generalized and automated tool for the detection of 3D conserved residues in a multiple structure alignment (MStA), for the interactive improvement of an MStA, for obtaining a pictorial view of a 3D consensus and for deriving a sequence motif from a structural consensus. Starting from a set of protein structures and a number of specified residues on at least one of them (i.e. a substructure), 3dLOGO identifies the substructures common to the set of the input structures and performs a 3D alignment onto the substructures’ residues. A tabular view is presented highlighting, if present, other similar and well-superimposed amino acids. The output set of conserved residues can be used for several purposes: to recognize a putative functional region, to build a 3D pattern for structure database searches, to identify event(s) of convergent evolution, to derive a customized sequence pattern in the form of a regular expression, which is usable for sequence database searches. Interesting results can be obtained with a careful selection of the input residues: such as a set of amino acids known to specifically bind the same ligand in different protein structures [e.g. some of the p-loop residues in proteins that bind ATP or GTP ()] or also residues belonging to a PROSITE pattern true positive match in diverse protein structures (). Users must supply ( ≥ 2) protein chains, which can be specified using PDB codes () or whose coordinates can be uploaded. For reasons of speed, the maximum number of input structures is 25. For each given structure, a single chain ID must be provided. The user must also supply the number (ID) of at least three amino acids of at least one input protein chain. This minimal substructure (the ‘object’ substructure) cannot comprise more than ten residues and is needed in order to localize the protein area to be studied. For instance, the minimal substructure can be a region known to bind a ligand (e.g. the solvent exposed residues of an SH3 domain or a set of residues binding ATP or GTP), to have a specific function (e.g. the catalytic triad of a serine protease), to match a known functional motif (e.g. PROSITE) or it can be any other protein area of interest. The server uses the Query3d application () for the detection of local similarities, in order to identify a structural match of the ‘object’ substructure (query) in the other − 1 input structures. A match is valid if (a) it is made up of at least three residues, (b) pairs of matching residues are physico-chemically similar and display an averaged rmsd <2 Å. If no local similarities are found in one of the − 1 input structures, this last is skipped and the subsequent steps are carried out only for the remaining − 2 structures. The ‘residue similarity options’ in the input page, makes it possible to set the degree of physico-chemical similarity under which 3D conserved positions are identified, e.g. by setting ‘identical’, only positions made up of identical residues will be shown. All structures are superimposed using two points for each given residue, namely the C-alpha and a pseudo-atom, calculated as the average of the coordinates of the residue side-chain atoms. In the case of input structures, the first structure is assumed to be the (or ‘target’) structure for the structural superimposition(s). The remaining − 1 structures are aligned pairwise to the reference structure. After superimposing the input structures on the given residues, all the nearby conserved residues are identified. Two amino acids belonging to different structures are assumed to be conserved if they are spatially well superimposed and display physico-chemical similarities, according to a substitution matrix [see () and the documentation web page for more details]. The user can make a selection from the input page if only identical, very similar, similar or all the spatially well-superimposed residues have to be displayed. Identification of the conserved residues is carried out using the 3D profiles method (). The multi-alignment output () is displayed in a table containing a column for each input structure and a row for each conserved position. User-supplied residues are highlighted by green dots, whereas yellow marks pinpoint newly identified well-superimposed residues. In the score column, the similarity measure derived from the substitution matrix is shown. Its value is maximum if the row's residue types are identical. The ‘rmsd’ column reports the average rmsd between the residues belonging to a row of the table with respect to the reference structure. Only positions whose residues display an average rmsd <2.5 Å are shown in the table. The structural alignment can be visualized with the Astex Viewer™ applet () or downloaded in text format (). Furthermore, the user can visualize the variable or conserved positions of the structural multiple alignment through a specifically designed Java application, , which is a 3D implementation of a sequence logo (). A sequence logo is a graphical representation of an amino acid or nucleic acid multiple sequence alignment (MSA) which provides an instant way of visualizing variable and conserved positions of the MSA (). In the sequence logo, the logo is constructed by calculating the information content of each position of the aligned sequences, and then displaying the characters representing the amino acids stacked on top of each other. The height of each letter is made proportional to its frequency. The height of each stack is then adjusted according to the information content at that position, as detailed in (). Similarly, 3dProLogo is a 3D graphical representation of the residues in a multiple structural alignment. The total information for each position in the structural alignment is calculated as in () and used to derive the height of each residue type's letter present in the 3D position. The resulting representation (3D logo) can be seen in , where the conserved positions in P-loop containing proteins 3D multiple alignment (see the last section) are displayed. The strong conservation of the residues involved in the nucleotide binding can be detected immediately as can be their arrangement in space. 3D logos can be rotated according to the users need. In order to build a sequence consensus, the user is allowed to manually select the positions in the MStA that better fit with his/her requirements. When the user is satisfied with the proposed structural alignment, a sequence (pattern) can be generated from the matched residues. By using all the identified conserved residues, the sequence pattern is constructed as a regular expression for searching sequence databases. The format of the regular expression provided is the one required for direct submission to the ScanProsite () server. The search for structurally conserved residues can be repeated by superimposing the structures on a new choice of conserved residues, including the newly detected ones. The whole process can be iterated until the identification of an ‘extended’ structure or sequence ‘consensus’. The user can obtain a new alignment by starting from a manual selection of the conserved positions listed in the structural alignment output table and using these as a new input of the procedure. The P-loop is a phosphate-binding loop commonly found in ATP and GTP-binding proteins (,). The ‘consensus’ sequence of this loop is [AG]XXXXGK[TS] (). A typical example of a P-loop-containing GTP-binding protein is the small GTPase c-H-ras p21, belonging to the Ras family (SwissProt: P01112). Guanylate kinase (SwissProt: Q8UGD7) is an ATP-binding protein belonging to the guanylate kinases family. The two proteins display 11% sequence identity, and the SSM () structure alignment of the two proteins finds only a few highly similar matching residues (concentrated around the P-loop), and an overall lack of structural similarity. The P-loop in the c-H-ras p21 structure (PDB 5P21) matches residues 10–18, whereas in the guanylate kinase structure (PDB: 1gky) the P-loop is found at residues 8-16 (). We considered 5p21 and 1gky plus a set of four different P-loop-containing proteins (PDB IDs: 1grn,1gua,1tad,1ega) extracted from () to perform a 3dLOGO query. By using these six structures as 3dLOGO input, all the residues belonging to the P-loop were found, as well as several new well-superimposed residues which are close to the P-loop in space, but very distant along the sequence (). In particular we retrieved the ras N116 (correctly superimposed to the kinase N168), which belongs to the well-known G protein conserved motif NKXD, whose residues are involved in the interaction with the nucleotide. More examples are described in detail and graphically displayed in the 3dLOGO examples Web page (). e 3 d L O G O w e b s e r v e r i s d e s i g n e d f o r t h e l o c a l s t u d y o f p r o t e i n s t r u c t u r e s . I t r e p r e s e n t s a n e w u s e f u l t o o l f o r t h e a n a l y s i s o f c o n s e r v e d s t r u c t u r a l r e g i o n s a n d f o r t h e d e r i v a t i o n o f m o r e s p e c i f i c s t r u c t u r e a n d s e q u e n c e p a t t e r n s . T h e 3 d L O G O t o o l w a s a l s o i m p l e m e n t e d t o p r o d u c e a n o v e l 3 D r e p r e s e n t a t i o n o f t h e c o n s e r v a t i o n o f r e s i d u e s i n a s e t o f p r o t e i n s t r u c t u r e s . R e p r e s e n t a t i v e e x a m p l e s a r e r e p o r t e d i n t h e R e s u l t s a n d i n t h e 3 d L O G O W e b p a g e s .

Intrinsically disordered proteins or protein regions exhibit unstable and changeable three-dimensional structures under physiological conditions (). Although lacking fixed structures, protein disorder has been identified to carry out important functions in many biological processes (,). In addition, it is observed that the absence of a rigid structure allows disordered binding regions to interact with several different targets (,). These regions, sometimes called ‘molecular recognition elements’, usually undergo a disorder-to-order transition when binding to their targets (,). In this regard, predicting protein disorder and investigating its potential of induced folding is a necessary preliminary to understanding protein structure and function (). The proposed web server iPDA aims at providing an integrated environment for detecting disordered regions and exploring their functional roles. In our recent work DisPSSMP (), it is demonstrated that the accuracy of protein disorder prediction can be greatly improved if the disorder propensity of amino acids is considered when generating the condensed position-specific scoring matrix (PSSM) features. For iPDA, we implement a two-stage classifier of Radial Basis Function Networks (RBFN) to further enhance the predicting power of DisPSSMP. As unbalanced datasets, a large amount of ordered residues over disordered residues, are employed when training the new classifier DisPSSMP2, an alternative decision function is recently adopted and the cutting threshold is dynamically determined by the proportion of predicted secondary structure in the query protein. iPDA takes an amino acid sequence as the input and reports the prediction of disordered residues with graphical plots, along with various sequence characteristics which are believed to be important when investigating the so-called induced folding behavior (). The provided information includes sequence conservation from multiple sequence alignment (ClustalW) (), concurrent sequence conservation from pattern mining (WildSpan) (,), secondary structure prediction (Jnet and PSIPRED) (,), low-complexity regions (CARD) () and hydrophobic clusters. Romero . stated that low-complexity regions are usually located in the long disordered regions (), where the sequence complexity is measured by Shannon's entropy. In addition, Ferron . mentioned in their recent study that hydrophobic clusters and secondary structures can provide distinct clues for investigating induced folding (). Meanwhile, we observe that sequence conservation is essential for disordered regions to maintain their functionality. Therefore, iPDA further provides a pattern mining utility to detect motifs in the specified disordered and/or ordered regions in order to predict potential intra- and inter-molecular interactions. fig #text In this section, we first evaluate how DisPSSMP2 performs in comparison with DisPSSMP and other existing packages for disorder prediction. After that, several interesting examples are provided to illustrate how iPDA helps users to explore the functional roles of the detected disorder regions. Many measures have been introduced to evaluate the performance of protein disorder predictors (,,,). Since and listed in are seriously affected by the relative frequency of the target classes (), two more appropriate measures are included in to reveal the properties of different packages. The first one, , is widely used in many bioinformatics problems (,). The other evaluation measure, named , was recommended by CASP (,) and Yang . () for this problem. To evaluate the performance of DisPSSMP2, we use a benchmark proposed by Yang . (), comprising 239 proteins. When preparing the training data of DisPSSMP2, the redundancy between the training and testing data has been avoided using the same criterion adopted in Yang's paper (). This benchmark also helps to judge whether a predictor tends to over-predict or under-predict disorder (,,,). As listed in , DisPSSMP2 has a better performance than DisPSSMP no matter which evaluation measure is used. The improvement of DisPSSMP2 is mainly from its including more ordered residues as training samples and the two-stage architecture employed. In addition, the performance of the existing packages for predicting protein disorder is ranked by its in . It should be aware that these packages were trained with different databases and some of them have different definitions for protein disorder from ours. Although many methods achieve specificity in excess of 90%, they usually result in low sensitivity. Since iPDA expects to discover potential disorder-to-order transitions, it is expected that employed predictor should deliver a high sensitivity rate of disordered regions without an explicit drop on specificity. Next we provide some examples discussed in () to illustrate how iPDA facilitates the study of protein disorder and induced folding. The first example used is a DNA-binding protein GCN4. According to the prediction shown in A, this protein might be largely unstructured. Meanwhile, it is observed that the region 225–281 is provided with large helical components. WildSpan also indicates high concurrent conservation in this area. B shows that one pattern found by WildSpan identifies the important residues with respect to the DNA-binding region. Similar discoveries are observed on the proteins NFATC1 and RXR discussed in (). In many cases, we observed that the regions undergoing disorder-to-order transitions when binding DNA usually possess both high disorder and secondary structure propensity, and additionally at least one pattern is found within this region to indicate potential intra- and/or inter-molecular interactions. This observation can be again justified by another protein, SecA, which undergoes locally disorder-to-order transition upon ADP binding in high temperature (). The partial result of analyzing SecA is shown in A. It shows that the range of 500–600 exhibits both disorder and concurrent conservation property. In this region, WildSpan detects 26 residues, and 10 of them are predicted as disorder. We highlight these 10 residues as red sticks in B to examine their positions with respect to the molecule ADP. Another example of disordered regions containing functional motifs is the protein p53. The iPDA result is shown in . In the disordered N-terminal domain (NTD) of p53, a short motif ‘FxxLW’, called the MDM2 functional motif, is discussed by Dawson's . (). The key residues are detected by ClustalW, as well as the second run of WildSpan ( = 3 and = 1). Those residues were not found in the first run of WildSpan, because the DNA-binding domain of p53 is more conserved than the MDM2 binding domain. If only the first disordered region (1–117) predicted by DisPSSMP2 are selected, the motif will be detected, as shown in A and B. In C, an available PDB structure shows the interaction of this polypeptide with the protein MDM2. r o v i d e s c o m p r e h e n s i v e i n f o r m a t i o n f o r a n n o t a t i n g t h e d i s o r d e r e d r e g i o n s o f a q u e r y s e q u e n c e . T h e i n t e g r a t e d r e s o u r c e r e c o g n i z e s i n t r i n s i c a l l y u n s t r u c t u r e d p r o t e i n s a n d h e l p s t o t e l l w h e t h e r a d i s o r d e r e d p r o t e i n o r p r o t e i n f r a g m e n t i s w i t h t e n d e n c y t o w a r d b e i n g f o l d e d u p o n b i n d i n g o t h e r m o l e c u l e s . A c c o r d i n g t o t h e e x p e r i m e n t s c o n d u c t e d i n t h i s s t u d y , t h e d i s o r d e r p r e d i c t o r D i s P S S M P 2 a c h i e v e s a h i g h e r s e n s i t i v i t y r a t e t h a n o t h e r e x i s t i n g p a c k a g e s p e r f o r m i n g t h e s i m i l a r t a s k w i t h o u t s a c r i f i c i n g t h e s p e c i f i c i t y r a t e . B e s i d e s , i P D A e m p l o y s s e q u e n t i a l p a t t e r n m i n i n g t o i d e n t i f y c o n c u r r e n t c o n s e r v a t i o n i t e r a t i v e l y , f r o m h i g h l y c o n s e r v e d r e g i o n s t o l i g h t l y c o n s e r v e d r e g i o n s o n e a t a t i m e . I t i s o b s e r v e d i n m a n y c a s e s t h a t t h e d i s o r d e r e d r e g i o n s u n d e r g o i n g d i s o r d e r - t o - o r d e r t r a n s i t i o n s u p o n b i n d i n g u s u a l l y e x h i b i t h i g h c o n c u r r e n t c o n s e r v a t i o n a n d c l e a r s e c o n d a r y s t r u c t u r e p r o p e n s i t y . T h i s a s s o c i a t i o n d e s e r v e s f u r t h e r s t u d i e s i n t h e n e a r f u t u r e . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The growth of protein structures in the Protein Data Bank (PDB) and expansion of our understanding of protein physical properties are constantly enhancing our ability to predict structural and functional features. Many prediction methods are now automated and accurate. We have contributed three such methods: WESA for predicting the solvent accessibility of amino acids from a protein sequence (), cons-PPISP for predicting interface residues from the structure of a protein which binds a second protein (,), and DISPLAR for predicting interface residues from the structure of a protein which binds DNA (). Both WESA and DISPLAR were found to have higher prediction accuracy than competing methods (,). Cons-PPISP was shown to be able to complement experimental techniques such as NMR chemical shift perturbation in mapping protein–protein interfaces (). The methods have found uses in protein structure prediction () and in docking of protein complexes (). For wide access to them, we have now developed web servers for these methods. Here we describe the functionality of the web servers and illustrate their utility by a structural model, built from predictions of the web servers, for a multicomponent protein–DNA complex. The web servers are components of PIPE, the protein interface/interior prediction engine, located at . Together, they serve as a pipeline from protein sequences to tertiary structures, then onto quaternary structures of binary complexes, and finally onto superstructures of functioning, multicomponent complexes. The three prediction methods have similar designs. The input in each case consists of data for a list of residues. In WESA, the list is comprised of a central residue and equal numbers of its ‘sequential’ neighbors on the left and on the right. In both cons-PPISP and DISPLAR, the list is comprised of a central residue and a number of its ‘spatial’ neighbors. The input data in WESA are sequential profiles, as given by the position-specific scoring matrix produced by PSI-blast (). In cons-PPISP and DISPLAR, solvent accessibilities (defined as percentages of exposed surface areas of residues) are also included in the input. All the three prediction methods used structures collected from the PDB for training. The dataset of WESA was comprised of 2148 proteins chains with sequence identity <25%. For cons-PPISP, the dataset was comprised of 1256 protein chains that form either heterodimers (involving 458 chains) or homodimers (accounting for the remaining 798 chains of the dataset). The sequence identities among the 1256 chains were <30%. For DISPLAR, the dataset was comprised of 264 protein multimers that form complexes with DNA in their PDB entries. The identities among multimeric entries were <50%. The three protein lists can be downloaded at . To illustrate why the prediction methods work, in we show the distributions of the 20 types of amino acids in interior and surface regions of folded proteins, and in interface and non-interface portions of protein surfaces that bind proteins or DNA. The results were calculated on the respective datasets. It is clear that the distributions exhibit distinctive patterns. Expectedly non-polar amino acids show preference for the interior whereas polar and charged residues for the surface. Non-polar amino acids similarly (albeit less prominently) prefer interfaces of protein–protein complexes. On the other hand, protein–DNA interfaces appear to be more dictated by electrostatic complementarity (instead of hydrophobicity), with positively charged arginine and lysine significantly enriched while negatively charged aspartate and glutamate significantly depleted in the interfaces. Another feature, captured by sequence profiles, that distinguishes interior from surface and interface from non-interface is sequence conservation. For structural and functional reasons, interior and interface positions are expected to be more conserved. Such a trend is indeed shown through a comparison in conservation scores between interior and surface positions and between interface and non-interface positions (Supplementary Figure 1). In addition to sequence profiles, cons-PPISP and DISPLAR also use solvent accessibilities as part of the input. In protein–protein complexes, interface positions consistently have higher solvent accessibilities than non-interface positions. On the other hand, in protein–DNA complexes, positively charged arginine and lysine have higher solvent accessibilities in interface positions than in non-interface positions, but negatively charged aspartate and glutamate show the opposite tendency (Supplementary Figure 2). Both cons-PPISP and DISPLAR are based on training neural networks. WESA is a metamethod, based on a weighted ensemble of five separate methods, one of which is neural network training. Further details on the implementations of the methods can be found in the original papers (). WESA has consistently shown a two-state (interior/surface) prediction accuracy ∼80%. Cons-PPISP predictions cover >50% of actual protein–protein interface residues and have >70% accuracy. DISPLAR predictions cover >60% of actual protein–DNA interface residues and have >80% accuracy. The direct web link for the WESA web server is . Once there, the user is asked to provide the sequence, in FASTA format, of the protein on which solvent accessibility is predicted. In addition, the user is asked to type in a unique identifier, at the user's choice, for referencing the particular WESA submission, and an e-mail address, for receiving the prediction. A link () also provides a sample output (obtained from submitting the sequence of PDB entry 1who), with explanations for the columns of numbers for each residue. In short, for each residue in the sequence, the results predicted by five separate methods and the WESA metamethod are given as 1 for exposed or 0 for buried, along with the prediction confidence (ranging from 0.00 for no confidence at all to 1.00 for full confidence). The threshold for an exposed residue is set at 20% solvent exposure. Two figures displaying the actual and WESA-predicted buried residues of 1who are found at the web server. The direct web link for the cons-PPISP web server is . The protein structure, on which an interface prediction is to be made, must be provided in PDB format, either by uploading or by pasting. In addition, the user must specify the chain(s) in the structure to be used for prediction. Here there are three common possibilities. The PDB file does not have chain ID; “_”is to be entered. The PDB file has a single chain ID (say “A”), or, it has multiple chains but only a single chain is to be used for prediction; “A” must be entered. Multiple chains (e.g. A, B and C) in a PDB file are to be treated as a single structure and used together for interface prediction; the user must enter “A,B,C”. A sample output can be found at . The direct web link for the DISPLAR web server is . Input and output formats are very similar to cons-PPISP predictions. For convenience of using the web servers, we provide scripts for running predictions in a batch mode. The scripts allow the user to submit multiple jobs by a unix command from his/her terminal, as if the prediction programs are installed on the local computer. In reality the scripts upload the sequences or PDB files to the PIPE web servers. The predictions are run at the servers, and results are sent back to the local computer and saved in files specified by the scripts. We now present interface residues predicted by the cons-PPISP and DISPLAR web servers to illustrate their utility. First, results are given for three proteins that competitively bind proteins and DNA and have unbound structures in the PDB. Next, predictions for a protein that simultaneously binds DNA and another protein are given, and the results are used to build structures for the protein–protein binary complex and the protein–DNA ternary complex. Full lists of predicted interface residues for the proteins are given in Supplementary Table S1. A displays the cons-PPISP predictions for the ribonuclease barnase, in reference to the structure of the complex with its inhibitor, barstar (PDB entry 1brs). The predictions were made on the unbound structure of barnase, 1a2p. Twenty predicted residues line up the actual interface. DISPLAR also correctly predicted the same portion of the barnase surface for binding DNA. As shown in B, the 27 predicted interface residues are concentrated around the binding site for a tetradeoxynucleotide, d(CGAC) (as defined in 1brn). Six residues (I55-S57, R59, Y97 and Y103) are common in the two sets of predictions. It is of interest to note that cons-PPISP predicted two clusters on the unbound structure of barstar (1bta), one is in the actual interface with barnase (shown in A) and the other defines an unknown binding site. DISPLAR did not make any positive predictions on 1bta, which is not expected to bind DNA. We applied cons-PPISP to the tumor repressor p53 core domain (2fej). In A, the predictions are displayed on the structure of p53 in complex with either 53BP1 or 53BP2 (1gzh or 1ycs). The five predicted residues (N239-M243) lie in the interfaces with the p53-binding proteins. A second cluster of 22 residues (listed in Supplementary Table 1) was also predicted; many of these residues are found in the dimer–dimer interface of a tetramer of the p53 core domain bound to DNA (). Some of these residues are also implicated in the binding with the E6/E6-AP complex (). When DISPLAR was applied on 2fej, 19 residues were predicted. These also lie in the interface with DNA (as found in 1tsr) (B). The protein- and DNA-binding sites on p53 partly overlap. Cons-PPISP predicted residues are close to some of the DISPLAR predicted residues, but no residue was predicted by both methods. The two-component transcriptional activator PhoB regulates its DNA-binding activity through transiently forming a domain–domain complex, thereby blocking the DNA-binding site. We used cons-PPISP to predict interface residues on the unbound structures of both the effector and receiver domains (1qqi and 1b00). The structure for full-length PhoB is not available, but we found the structure (1ys6) for a close homolog, PrrA, in the PDB. The root-mean-square deviations (RMSD) of 1qqi and 1b00 from the corresponding domains in 1ys6 are 2.4 and 2.1 Å, with sequence identities over aligned positions at 35 and 41%, respectively. When 1qqi and 1b00 are aligned onto 1ys6, the 23 and 21 predicted residues on the two respective domains indeed mostly line the interface as modeled on 1ys6 (A). DISPLAR predicted 25 residues on 1qqi, which are located in the actual interface between the effector domain and DNA (as found in 1gxp) (B). The binding sites for the receiver domain and DNA on the effector domain (1qqi) overlap, and eight residues (R68, T69, D71, H73 and V93-T96) were predicted by both cons-PPISP and DISPLAR. No DNA-contacting residues were predicted by DISPLAR on the receiver domain (1b00), which is not known to binding DNA. Core binding factors (CBF) form a heterodimer between the α and β subunits, which in turn forms a ternary complex with DNA. Using the unbound structures of CBFα and CBFβ (1eaq and 1ilf, respectively), we predicted interface residues between these two proteins by cons-PPISP. Examined on the dimer structure of CBFα and CBFβ (1e50), the 20 and 23 predicted residues on the two subunits indeed line the actual interface (A). Previously we have used cons-PPISP predictions to drive the docking of unbound structures (). Here we use the predictions to score docked structures, obtained by running ZDOCK 2.3 () with 15° sampling. The best 200 ZDOCK structures (out of a total of 2000) were re-ranked according to the number of cons-PPISP predicted residues among the interfacial residues (defined as having 10-Å contacts across the interface). A structure with an RMSD of 2.2 Å was ranked second according to cons-PPISP predictions (improved from ZDOCK's ranking of 49th). This docked structure is shown in B. The dimer structure of CBFα and CBFβ, 1e50, was used to predict DNA-contacting residues by DISPLAR. Nineteen residues, all on CBFα, were predicted. As shown in A, CBFα is indeed the subunit that contacts DNA in the ternary complex (1h9d), and the predicted CBFα residues line the DNA-binding site. We docked the dimer structure with a B-DNA decamer built in InsightII (Accelrys Software Inc., San Diego, CA, USA) from the sequence in 1h9d. Parameters for DNA nucleotides required for running ZDOCK were taken from Fanelli and Ferrari (). A docked structure for the ternary complex, ranked 113th in ZDOCK but improved to 11th by DISPLAR predictions, was found to have an RMSD of just 1.2 Å from 1h9d and is shown in B. The continuous growth of the sequence database () and the PDB () will further improve the accuracy of the three prediction methods. We plan to periodically upload the NCBI the non-redundant (nr) onto PIPE (). In addition, we plan to expand the datasets for the three predictors by including new entries from the PDB; re-training will be done. The high accuracy of WESA predictions suggests that solvent accessibility is now a matured field. Future methodological developments will thus focus on protein–protein and protein–DNA interface predictions. A strategy that contributed to the success of WESA is the combination of complementary methods. We plan to combine cons-PPISP and DISPLAR with approaches based on phylogenetic tree (,), surface patch characteristics (), secondary structure (), empirical scoring function (), support vector machine () and Bayesian network (,). Ultimately proteins need to be studied within their functioning units, which often are multicomponent protein complexes. The PIPE web servers will contribute to better understanding of these complexes. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Reliable and accurate predictions of protein structure are important for many biologists. For many years it was believed that manual experts significantly outperformed all automatic methods. However since consensus-based approaches () were introduced it has been found that at the most a handful experts in the world can outperform the ‘community’ of web-servers. It has also been shown consistently in CASP that consensus methods are superior compared to individual methods in predicting the structure of a protein sequence (). Pcons has been among the top performing automated predictors since CASP5 and was the best method for assessing model quality in CASP7 (). Here, we introduce the meta server () which provides improved automated tools for protein structure prediction and analysis using consensus. The whole process is fully automated and a potential user only submits the protein sequence. This makes it easy to acquire structural information without any prior knowledge of remote homology detection, model building and model quality assessment. Pcons has previously been available as a downloadable program as well as through several other meta servers (genesilico.pl and bioinfo.pl). meta server provides significant improvements over these servers. It has an improved web interface and prediction accuracy, the local accuracy for each residue is also provided and for easy targets an accurate 3D model is build within minutes of submission. The Meta Server () essentially implements all the steps necessary to produce a high quality model of a protein sequence: An overview of the method is shown in . In the first step domains are assigned using Pfam () and a quick database search against known protein structures (PDB90) is performed using BLAST () and RPS-BLAST (). This also establishes the difficulty of the submitted sequence. If a significant hit is found using RPS-BLAST, an all-atom model is produced using, Pfrag, a novel rapid homology modeling program based on segment matching and assembly. If the sequence identity is above 50% this model will be quite close to the native structure, comparable to low-resolution X-ray and NMR structures (,). The whole process from sequence to all-atom model takes ∼30 s, making it one of the fastest comparative modeling servers available. RPS-BLAST is also used to parse the sequence into structural domains by analyzing the significance and span of the best RPS-BLAST alignment. If the hit is (i) significant and (ii) the alignment contains more than 30 unaligned residues, the unaligned residues are parsed out and resubmitted to the servers as a separate submission. In many cases, these domains agree well with the domains obtained using Pfam. It is only if no significant hits are found using RPS-BLAST, that the sequence is submitted to publicly available more advanced fold-recognition servers (). The user has the possibility to force the submission of sequences that has clear RPS-BLAST hits. However, we strongly discourage overuse of this possibility in order to not overload the external servers with trivial queries. The alignments from the initial BLAST, RPS-BLAST as well as the alignments from the fold-recognition servers are collected as they finish and all-atom models are built using Pfrag. When the model building is finished, the quality of the models is assessed using Pcons (,,). Pcons benefits from the use of as many individual servers as possible. Thus, it is important to not put too much weight on a consensus analysis that is only based on the results from a few servers. In parallel to the consensus analysis, the model quality is also assessed purely based on structural features using ProQ (). Both Pcons and ProQ give an overall quality to each model as well as a local quality score to each individual residue (). In CASP7, Pcons was one of the best method for assessing the overall quality of protein models and the best method for assessing the local quality of residues (). In summary, the major advances over other web servers are: The server takes a protein sequence in one-letter amino acid format as input. The user has the possibility to name the sequence and to give their e-mail address. Both the name and e-mail address can be used to filter the results in the job queue (). Results for a specific job are provided through the web interface by clicking on the job id listed in the job queue table (). This page is updated continuously as more predictions are finished. If an e-mail is provided the top 10 ranked model coordinates are e-mailed after 46 h. The 46 h time limit is set to allow for as many fold-recognition servers as possible to finish and provide the basis for the consensus analysis. However, if a significant hit indeed is found using the locally run RPS-BLAST, an accurate model should be ready within minutes of submission. In addition to the web interface, the meta server will also be made available as a web service using the Web Service Description Language (WSDL) (). The idea behind web services is to allow applications to communicate with each other in a standardized way. WSDL is used to conceptually describe the operations available at the service, and expresses the data formats using XML Schema definitions. Communication between web services and clients is done using the SOAP language (Simple Object Application Protocol) (). For this will mean that a user who has access to a web service client, such as Taverna (), will be able to make submissions to the meta server and also build in these submissions into more complex analysis workflows. An additional novel feature is the representation of the different alignments (), which enables a quick overview of the alignment quality and facilitates comparisons of many alternative alignments. The alignment is represented as a line that is color-coded according to the secondary structure. For the template structure STRIDE () is used to assign secondary structure based on the coordinates, for the target sequence PSIPRED () is used to predict secondary structure and assign it to each residue. Both the target and the template sequence are represented as full-length sequences, making it possible to see which parts of the target and template that are covered; and if the alignment spans only a part of the whole template structure. Here, the user also has the possibility to submit unaligned regions that did not fulfill the criteria for automatic domain resubmission (see above). #text A key component for any successful protein structure protocol is the ability to assign quality scores to the created models. scores models using the best methods currently available. For each model three global quality scores are provided, one based on consensus (Pcons), one based solely on structure (ProQ) and one using a combination of the two (Pmodeller). All are presented in the job summary page. The reason for providing more than one score is that they contain complementary information. The Pcons score, for instance, is only meaningful if a sufficient number of models are available. If this is not the case, a structural evaluation using ProQ might be more suitable and for other cases the ProQ score might be a useful aid in the process of choosing the best model. From a user perspective it is important to know when to trust a certain score. Based on results from the quality assessment category in CASP7 () the Pcons score correlates well with the correct quality of the models as measured by LGscore () ( = 0.96). Moreover a Pcons score above 1.1 separates correct from incorrect models almost perfectly (only 2.5% false predictions). The ProQ and Pmodeller scores are the predicted LGscore and score values above 1.5 correspond to -values better than 10. In addition to the global quality scores, each amino acid in the models is given an estimate of the CA–CA error as measured by the local S-score ( = 1/(1 + error/5)). The S-score varies between 0 and 1 corresponding to high and low error, respectively, e.g. if the S-score is larger than 0.5 the error is predicted to be <2.24 Å. The advantage with this type of score is that it focusses on the regions that have low error and gives the same score value for regions that are wrong. As for the global scores the local quality is predicted using either consensus (Pcons) or structural features (ProQres). In terms of performance, Pcons is superior to ProQres (). In fact, no non-consensus-based approach is nearly as good as consensus-based approaches (). However, ProQres still provide some additional value as a complement when there is no clear consensus or as additional augmentation when the consensus is weak. The local quality predictions are accessible by clicking either on the Pcons score or on the ProQ score in the job summary page (). The local quality scores predicted by Pcons are also added to the B-factor column of all models for easy visualization in any coordinate viewing program (). #text

Genome sequencing projects have lead to a surge in the number of protein sequences that lack experimental functional data. At the same time the rise of structural genomics initiatives has meant that the structural databases are starting to fill up with unannotated structures. Experimental approaches for function characterization are expensive and difficult to automate and this has meant that researchers have turned increasingly to computational methods to try to close the gap between the number of new unannotated sequences and the number of sequences with known function. The standard way to overcome this deficit is the homology-based transfer of functional annotation. The transfer of general functional information [in the form of GO terms (), etc.] typically requires little more than a BLAST () homology search, as long as the protein does not have more than one domain and as long as the query sequence meets a certain threshold of similarity to the annotated protein template. The main constraint for this classical approach is that transfer of function based solely on the percentage of identity of two sequences is not always 100% reliable (). This constraint has meant that it has been necessary to develop techniques that are capable of assigning function in a more sophisticated manner. In many cases the most interesting functional information, such as catalytic residues and those residues bound to ligands, is to be found at the residue level. Here transference of function is considerably more laborious since binding sites are disperse, and alignments must be generated and checked by hand before the interesting residues can be mapped onto the query sequence. Homology-based transfer of functional information at the residue level has been explored in numerous works. The Catalytic Site Atlas () is built from annotations extracted from literature and these annotations are extended to the whole PDB trough PSI-BLAST transference. The same authors used this to explore the evolution of catalytic sites in homologous families (). Automated functional information transfer is also carried out by databases such as Swissprot-Trembl (). However, while there are web servers that use homology to predict functional features such as GO terms (), and web servers that predict probable binding sites based on clefts or cavities () we do not know of an available server that is capable of mapping functional residues onto a target with the aim of highlighting potential functionally important residues. Here we present , an expert system that merges the time consuming tasks of alignment and mapping into a single server with a simple input. It combines the FireDB () database, a large inventory of structure-based functionally important residues, and SQUARE (,), a method for the assessment of the local reliability in sequence alignments, to predict likely residues of functional importance in query sequences. This simple tool also includes measures of reliability for the predictions, a set of methods for the visualization of the results on the corresponding sequences and structures and a multiple alignment option for easy comparison. The FireDB database is a databank containing a comprehensive and detailed repository of known functionally important residues. It integrates biologically relevant data filtered from the close atomic contacts in Protein Data Bank () crystal structures and reliably annotated catalytic residues from the Catalytic Site Atlas. Residues in close contact with ligands are defined as those residues with atoms that are closer than 1.0Å plus the sum of Van der Waals radius of the atoms involved. Redundancy in the PDB was addressed when designing the database. PDB sequences are clustered with cd-hit () at 97% sequence identity and a consensus sequence is built for each cluster. The consensus sequences form the basis of FireDB and of . All functional information is associated to the consensus sequences—equivalent binding sites from proteins within the same cluster are conveniently mapped onto the consensus sequence. The functional residues used by are derived from all the proteins in each cluster, but associated to just a single sequence. As of 8 January 2007, FireDB contained a total of 16 843 clusters, of which 9021 had associated functional information. evaluates the probability that a residue is involved in ligand binding with a version of SQUARE, a method that was developed to predict regions of reliably aligned residues in pairwise sequence alignments. For SQUARE to evaluate the reliability of an alignment one of the two sequences in the alignment must be associated with a PSI-BLAST-generated profile. In the case of , PSI-BLAST profiles are pre-generated for all the FireDB consensus sequences. This allows SQUARE to evaluate all the pairwise alignments in —the sequence and structural alignments generated as part of are all between the query sequence and the stored FireDB consensus sequences. SQUARE assigns conservation-based reliability scores by extracting values for each aligned residue from the PSI-BLAST profiles generated from the FireDB consensus sequences. The alignment scores are smoothed with a triangular five-residue window. From the SQUARE reliability scores it is possible to discern which residues are aligned reliably, and which of the binding and functional residues are likely to have some level of functional conservation. While SQUARE was developed to evaluate the reliability of pairwise alignments, it has been shown that the method is even more effective at predicting the conservation of residues in binding sites (). A stand-alone version of SQUARE for pairwise alignment reliability is available at . Accepted input forms are sequences in fasta format, or structures described by their PDB codes or coordinates in PDB format. Target sequences are subjected to standard PSI-BLAST searches. Profiles are generated with an nrdb90 database from the EBI () and the final search is made against the FireDB consensus sequence database. Functional residues mapped onto the resulting alignments and the reliability of each position in the alignment is evaluated with SQUARE. The residue scores from SQUARE represent the probability that a given target residue is aligned to the evolutionary equivalent residue in the consensus sequence. It has been shown that evolutionary conserved binding site residues are almost always involved in ligand binding in the target protein (). If the user input is a structure (with PDB code or uploaded structure) can generate structural alignments between the query and the templates selected by PSI-BLAST with the structural alignment program LGA (). In this case SQUARE evaluates the reliability of each position in the structural alignment. There are three output types generated by the server: The evaluation of active site conservation by SQUARE is sensitive to alignment quality, a poor alignment may mean SQUARE does not tag a binding residue as conserved. On occasions, automatic alignments may not be the most appropriate, so user-defined pairwise alignment inputs are also possible. Users may upload their own pairwise alignment with the constraint that the template they use must be in the FireDB database. FireDB is updated regularly from the PDB database, so the FireDB template database contains all the structures present in the PDB at the time of the most recent FireDB update. In this case will produce an output in pairwise mode. #text One complication with the functional data in the PDB is that bound ligands may or may not be biologically relevant. This problem was addressed in the FireDB database; ligands classified as solvent by mmCif () are ignored by FireDB and biological relevance can be assessed by the co-occurrence of sites in homologous proteins. Conserved sites in two or more homologues imply an evolutionary pressure in residue conservation and suggest biological relevance. adds additional capacity since it makes it easier to find several homologues with same binding site conserved in aligned positions. We developed the server as one of the tools for our evaluation of the function prediction section of the 7th edition of Critical Assessment in Structure Prediction (CASP) (). was essential to determine context and conservation in other structures and whether the ligands bound to the target structures could be considered as biologically relevant or not. Only the biologically relevant ligand-binding sites formed part of the CASP evaluation. CASP Target T0312 is a hypothetical DNA-binding protein from . It has a homo-trimeric form in the coordinates file (PDB: 2H6L) and the three monomers bind both zinc and acetate in the same cleft. Acetate is a common solvent and it is not supposed to be relevant. We ran with the T0312 sequence, the pairwise output returned by PSI-BLAST showed two hits (a). The first alignment in a shows 2H6L itself and shows the zinc binding at Tyr 61 and at three conserved histidines, the second alignment is with template 2hx0 that was deposited in the PDB after the CASP7 prediction deadline. The third alignment shows the best template available to predictors, 1xv2 (), a distant template that was found by PSI-BLAST with a high e-value and a poor alignment. As it turns out 1xv2 also binds zinc, but only one of the three binding histidines (His 104) is reliably conserved in the PSI-BLAST alignment with the target sequence. A second histidine (His 91) is conserved but less reliable, while the third histidine and the tyrosine are not conserved. In fact two of the histidines are misaligned in the PSI-BLAST alignment, something that becomes clear in the MUSCLE alignment (b). The histidines involved in binding in 1xv2 are also aligned in LGA structural alignment (b) and here it can be seen that all the three histidines are also reliably aligned. Moreover the Jmol window shows that the side-chain orientations of the histidines are conserved (c). The target clearly contains a biologically relevant zinc-binding site even though the other residues involved in zinc binding (Tyr 61 from T0312 and Glu 45 from 1xv2) were not conserved. In addition it is clear that even though this target would be regarded as ‘difficult’, the binding residues could have been predicted by before the structure was solved if the server was provided with a good alignment. The LGA structural alignment was available only because the structure has recently been released, but it would have been possible to use the tool to build the correct alignment and predict the binding residues from even without access to the structure. Future releases of will include improvements to the server interface and the development of new features. We plan to make a version available at the INB web services (), and the central web services in Canada as part of the services offered by the NoE Embrace. We plan to exploit the predictive abilities of and to make results available in the context of large annotation efforts, in particular in the BIOSAPIENS and GENEFUN projects. At the same time we are working on a version of that will validate the biological relevance of all bound ligands found in PDB entries. In addition to updating the server we are planning to test with functionally interesting residues such as post-translational modifications, mutations or even residues linked to diseases in OMIM (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Protein phosphorylation, which is an important reversible mechanism in post-translational modifications, is involved in many essential cellular processes including cellular regulation, cellular signal pathways, metabolism, growth, differentiation and membrane transport (). Phosphorylation of substrate sites at serine, threonine and tyrosine residues of eukaryotic proteins is performed by members of the protein kinase family. Additionally, phosphorylation on histidine plays an important role in signal transduction in prokaryotes known as two-component histidine kinase (). It is estimated that one-third of proteins are phosphorylated and around half of kinome are disease- or cancer-related by chromosomal mapping (). Experimental identifications of kinase-specific phosphorylation sites on substrates and are the foundation of understanding the mechanisms of phosphorylation dynamics and important for the biomedical drug design (). However, these experiments are often time-consuming, labor-intensive and expensive. Therefore, prediction of phosphorylation sites with high predictive performance could be a promising strategy to conduct preliminary analyses and could heavily reduce the number of potential targets that need further or confirmation. With the recent exponential increase in protein phosphorylation sites identified by mass spectrometry (MS), many researches are undertaken to identify the kinase-specific phosphorylation sites. Our previous work, KinasePhos 1.0, incorporated profile hidden Markov model (HMM) for identifying kinase-specific phosphorylation sites, whose overall predictive accuracy is ∼87% (,). NetPhos () developed neural networks to predict phosphorylation sites on serine, threonine and tyrosine residues; however, it cannot provide information on the kinases involved and NetPhosK () applied an artificial neural network algorithm to predict 17 PK groups-specific phosphorylation sites. DISPHOS () took advantage of the position-specific amino acid frequencies and disorder information to improve the discrimination between phosphorylation sites and non-phosphorylation sites. Scansite 2.0 () identified short protein sequence motifs that are recognized by modular signaling domains, phosphorylated by protein serine/threonine, tyrosine kinases or mediate specific interactions with protein or phospholipid ligands. PredPh2ospho () predicts phosphorylation sites limited to four protein major kinase families, such as CDK, CK2, PKA and PKC, and four protein kinase groups (AGC, CAMK, CMGC and TK) with predictive accuracy 83–95 and 76–91%, respectively. GPS (,), is a group-based phosphorylation site predicting and scoring platform which clustered the 216 unique protein kinases in 71 groups. PPSP () developed an approach based on Bayesian decision theory for predicting the potential phosphorylation sites accurately for around 70 protein kinase groups. This work proposes a kinase-specific phosphorylation site prediction server which incorporates support vector machines (SVM) with two features, i.e. protein sequence profiles surrounding the modified sites and coupling patterns surrounding the modified sites. The coupling pattern of proteins, which is first used for analyzing the protein thermostability (). In this work, we incorporate the protein coupling pattern as a feature for training computer models for identifying phosphorylation sites. After evaluating the computational models by -fold cross-validation and Jackknife cross-validation, the overall predictive accuracy of KinasePhos 2.0 is ∼91%, which is better than the previous version and the other tools previously developed. The details of the proposed method and predictive performance are described below. depicts the system flow of the proposed method. The experimentally validated phosphorylation sites are extracted from Phospho.ELM (release 6.0) () and Swiss-Prot (release 50) (), containing 13 612 phosphorylation sites within 3674 proteins and 6832 sites within 3148 proteins, respectively. After removing the redundant sites between Phospho.ELM and Swiss-Prot, the number of serine (S), threonine (T), tyrosine (Y) and histidine (H) substrate are 11 888, 2433, 2179 and 43, respectively, as given in . Since the flanking sequences (position −4 ∼ +4) of the phosphorylation sites (position 0) are graphically visualized as sequence logos (), the conservation of amino acids in the phosphorylation sites can be observed. The 9-mer sequences (−4 ∼ +4) of kinase-specific phosphorylation sites are extracted and constructed as training sets. Table S1 (See Supplementary Data) summarizes the statistics of 60 kinase-specific phosphorylation sites in the data set constructed. To avoid the overestimation of the predictive performance, the redundant training sequences should be discarded. After the construction of non-redundant training set of kinase-specific phosphorylation sites, two features, i.e. sequence of surrounding catalytic sites and coupling pattern of surrounding catalytic sites, are extracted. As to sequence surrounding catalytic sites, 9-mer sequences (−4 ∼ +4) of kinase-specific phosphorylation sites are encoded in three ways: BLOSUM62 profile encoding (the corresponding row number of amino acids in BLOSUM62 matrix), reduced alphabet (sparse encoding with fewer letters) () and 20-dimensional vector (each amino acid is mapped to a 20-dimensional vector), as given in Table S2. It was found that amino acids have a great variety of properties such as mass, polarity, hydrophobicity, so many groupings are possible (). With the hydrophobicity (), for instance, the 20 amino acids are reduced into three classes, such as polar (R,K,E,D,Q,N), neutral (G,A,S,T,P,H,Y) and hydrophobic (C,V,L,I,M,F,W). The coupling pattern of surrounding catalytic sites is extracted from the flanking sequences of kinase-specific phosphorylation sites. Let [] denote the coupling pattern of amino acids and that are separated by amino acids. Since the protein sequence is directional, the sign of is determined by the relative positions of and . For example, as shown in , a coupling pattern [R3Q] occurs in the training set, another coupling pattern [Q-3R] also occurs. Herein, we would not consider the coupling pattern with minus symbol. The coupling strength between and of the pattern [] is given by If  ≥ 1, then and are positively correlated with respect to the distance , and they are negatively correlated if  < 1. The differences of coupling strength between the training set of phosphorylation sites and the background set, which is extracted from all 9-mer sequences centering at residue serine, threonine, tyrosine and histidine in Swiss-Prot protein sequences, are computed and used to determine the number of coupling patterns trained by SVM. The higher differences of mean that the coupling pattern [] is the most important feature for separating the training set from the background set; therefore, the values of differences of the coupling strength between training set and background set should be tuned for determining the number of coupling patterns used to train a SVM model. Each coupling pattern is a dimension of features used in SVM. For instance, when set up the cutoff value of the differences of between training set and background set to 1.5, there are about 400 coupling patterns which is higher than the cutoff; thus, the number of dimensions trained by SVM is about 400, which is equal to the number of selected coupling patterns. This work incorporates support vector machine (SVM) with the protein sequences and profiles of coupling pattern for training the predictive models for kinase-specific phosphorylation site prediction. A public SVM library, namely LIBSVM (), is applied for training the predictive models. The SVM kernel function of radial basis function (RBF) is selected. In general, the experimental kinase-specific phosphorylation sites are defined as the positive set, while all other residues (S, T, Y or H) in the phosphorylated proteins are regarded as the negative set. -fold cross-validation is used to evaluate the predictive performance of the models trained from the large data sets including PKA, PKC and MAPK, and Jackknife cross-validation is applied for models trained from the data size smaller than 30. We balance the positive set and negative set and the sizes of positive set and negative set are equal during the cross-validation processes. The cross-validation is performed for 30 times. The following measures of predictive performance of the trained models are defined: Precision (Prec) = TP/(TP + FP), Sensitivity (Sn) = TP/(TP + FN), Specificity (Sp) = TN/(TN + FP) and Accuracy (Acc) = (TP + TN)/(TP + FP + TN + FN), where TP, TN, FP and FN are true positive, true negative, false positive and false negative predictions, respectively. For finding the best predictive performance of SVM models in each kinase-specific group, the SVM models trained with various features such as coupling pattern (CP), sequence and the combination of coupling pattern and sequence are evaluated based on cross-validation. As shown in , the average precision (Prec), sensitivity (Sn), specificity (Sp) and accuracy (Acc) of the SVM models trained with various features are calculated for phosphoserine, phosphothreonine, phosphotyrosine and phosphohistidine. Two methods are used to extract the coupling patterns, i.e. ‘CP difference’ and ‘CP ratio’. ‘CP difference’ indicates the coupling strength of training set subtracted the coupling strength of background set, and ‘CP ratio’ indicates the coupling strength of training set divided the coupling strength of background set. As to the feature of sequence profile, there are various coding methods used for encoding amino acids surrounding the phosphorylation sites, such as reduced alphabet (3-classes, 7-classes and 8-classes), BLOSUM62 profile encoding and 20-dimensional vector. Because the average predictive performance of the kinase-specific phosphorylation sites with small training set may be overestimated, the SVM models of kinase-specific group whose data size is smaller than 20 training sequences are not considered. gives the average predictive accuracies of models trained with coupling patterns (CP difference or CP ratio) of phosphoserine, phosphothreonine, phosphotyrosine and phosphohistidine are 86, 93, 88 and 93%, respectively. The overall predictive performance of SVM models trained with the features of coupling patterns, whose accuracy is close to 90%, is performing better than the SVM models trained only with sequence profiles (Seq). Since the features of coupling patterns (CP ratio) and sequences (7-classes) with best predictive performance are combined, the average predictive accuracy of SVM models trained with the combined features of phosphoserine is 89%, which is slightly better than the SVM models trained only with coupling patterns. However, the average predictive performance of the SVM models trained with the combined features of phosphothreonine, phosphotyrosine and phosphohistidine is close to the SVM models trained only with coupling patterns. The overall predictive accuracy of SVM models trained with the combined features of coupling patterns and sequences is close to 91%. In addition, the method of KinasePhos 1.0 is evaluated based on the data set constructed in this work. The average predictive accuracies of phosphoserine, phosphothreonine, phosphotyrosine and phosphohistidine are 84, 88, 84 and 83%, respectively. Since the SVM models trained with various features, the most accurate model of each kinase-specific phosphorylation sites are selected and used to implement a prediction server. As shown in Table S3, the trained features, SVM Cost value, SVM Gamma value, precisions, sensitivity, specificity and accuracy of the selected models are presented for 37 kinase-specific groups with at least 20 experimentally verified phosphorylation sites. In the column of trained features, the value in the parentheses behind the coupling pattern (CP) is the value of difference or quotient of coupling strength between the training set against the background set. The average predictive accuracies of phosphoserine, phosphothreonine, phosphotyrosine and phosphohistidine are 90, 93, 88 and 93%, respectively. fig #text The models trained with various features, including sequence profiles and coupling patterns, were evaluated by 5-fold and Jackknife cross-validation, the predictive performance of the models trained with coupling patterns are better than the models trained with sequence profiles. In general, the previous works of phosphorylation site prediction focused on residues serine, threonine and tyrosine; like our previous work (KinasePhos 1.0). Herein, KinasePhos 2.0 first considers phosphohistidine from Phospho.ELM and Swiss-Prot, which contain one and 42 phosphorylated histidine, respectively. Moreover, the proposed web server is compared with several previously developed phosphorylation prediction tools, such as DISPHOS, PredPhospho, GPS, PPSP and KinasePhos 1.0. As given in , the number of kinases, sensitivity and specificity of prediction and the overall predictive performance of these tools are compared. GPS, PPSP, PredPhospho, KinasePhos 1.0 and the proposed methods all support the identification of kinase-specific phosphorylation sites. Although only the kinase groups containing at least 20 experimental phosphorylation sites were selected to evaluate the average predictive performance, the web server of KinasePhos 2.0 provided the predictive models of 60 kinase-specific groups with at least 10 experimental phosphorylation sites. Because the average predictive performance of serine, threonine and tyrosine of GPS and PPSP cannot be obtained, the predictive performance of three representative kinases such as PKA, PKC and CK2 are compared. As given in , the predictive performances of three representative kinases in KinasePhos 2.0 are comparable with PredPhospho, GPS, PPSP and KinasePhos 1.0. In particular, KinasePhos 2.0 provides the predictive model for phosphohistidine, whose predictive accuracy is 93%. The overall predictive accuracy of the kinase-specific groups with at least 20 phosphorylation sites of the proposed method is 91%. However, as given in Table S4, the overall predictive accuracy of the kinase groups which are smaller than 20 experimental phosphorylation sites is 94%. The protein structural properties, such as accessible surface area (ASA) and secondary structure, can be considered in the future to improve the predictive performance of the models. For instance, ASA may be used for reducing the number of false-positive predictions of phosphorylation sites which locate in buried regions. However, the number of experimental phosphorylation sites located in the protein regions with known structure from PDB () is few for each kinase-specific group. Although ASA and secondary structure can be predicted by several published tools such as RVP-net () and PSIPRED (), respectively, the predictive performance of phosphorylation sites may be affected by the predictive structural properties. #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Functional genomics research often generates lists of genes with observed common properties, such as coordinated expression. For many studies, a key challenge is the generation of relevant and testable hypotheses about the regulatory networks and pathways that underlie observed co-expression. Our strategy for elucidating regulatory mechanisms identifies over-represented sequence motifs that are present in the upstream regulatory regions of genes. The motifs may represent transcription factor binding sites (TFBSs) that have a role in regulating expression. oPOSSUM () and oPOSSUM2 () were developed to identify over-represented, predicted TFBSs and combinations of predicted TFBSs, respectively, in sets of human and mouse genes. The user inputs a list of related genes, selects the TFBS profile set to be included in the analysis, and the algorithm determines which, if any, predicted TFBSs occur in the promoters of the set of input genes more often than would be expected by chance. Both analytic approaches rely on a database of aligned, orthologous human and mouse sequences, and the delineation of conserved regions within which TFBS predictions are analyzed. While the approach does not explicitly address uncharacterized transcription factors (TFs), the effective coverage is broadened by the fact that members within certain structural families of TFs can exhibit similarities in binding specificity. While intra-class similarity is not always the case, as exemplified by the zinc-finger family of TFs (), the observation holds true for many TF families (,). Here we describe the new release of the oPOSSUM system, which integrates the two previously developed applications, and has been expanded to accommodate new species (yeast and worms). It also includes new methods for orthology assignment, transcription start site (TSS) determination and sequence alignment. italic #text italic xref #text In addition to enhancements to the human/mouse oPOSSUM database, we introduce new species databases for studies of over-represented TFBSs in yeast and worms. While the SSA over-representation analysis remains the same for all species, differences in gene structure require that the construction of the underlying databases be particular to each species. For the metazoan species, we search for matches to TFBS profiles contained in the JASPAR CORE and JASPAR PhyloFACTS database collections (,). Additionally, we include a set of profiles compiled for TFs from literature review for Worm SSA (Table S2). Binding sites are predicted for the sequences using the TFBS suite of Perl modules for regulatory sequence analysis (). A predicted binding site for a given TF model is reported if the site occurs in the promoters of both orthologs above a threshold PSSM score of 70% and at equivalent positions in the alignment. Overlapping sites for the same TF are filtered such that only the highest scoring motif is kept. The genomic location, profile score, motif orientation and local sequence conservation level of each TFBS match in orthologous genes are stored in the respective species databases. For , we compiled a collection of yeast-specific TFBS motifs from both the Yeast Regulatory Sequence Analysis (YRSA) system () and the literature (Table S3), and record the genomic location, profile score and motif orientation for each prediction. Based on the observation that members of the same structural family of TFs often bind to similar sequences, plant and insect matrices are available for inclusion in the analysis. The MADS family of TFs is an excellent example of conservation of binding domains between plants and vertebrates (,), and there are numerous examples of conservation of binding domains across vertebrates, flies and worms. Thus, in cases where a profile for the TF of interest is not available in the database, oPOSSUM can still provide insights into the underlying regulation by suggesting a particular TF family that may be involved. Each oPOSSUM component was validated on sets of reference genes. The results of all validations are available as Supplementary Data (Tables S4–S13). In the interest of space, selected examples are described for each system. The four oPOSSUM systems, Human SSA, Human CSA, Worm SSA and Yeast SSA, have been integrated into a use-friendly website at: . We recommend that users of the system begin with the SSA to quickly identify TFBSs that may be relevant to their input data sets. For sets of human and mouse genes, this can be followed with the CSA, which takes longer to process, but which can provide insights into TFBSs that may be acting in concert to regulate the set of genes. Upon submission, oPOSSUM SSA generates a summary of the input parameters, and produces a single table that ranks the over-represented TFBSs by descending Z-score. The table may be sorted by TF name, TF class, supergroup, information content (IC), Z-score and Fisher score (A). Pop-up windows linked to each TFBS foreground count display the genes in which the putative site is located, the promoter region(s) for each gene, as well as the TFBS's co-ordinates and score (B). TFBSs that occur in overlapping promoter regions are marked by an asterisk and highlighted in yellow. The TF names are linked to the JASPAR database for easy access to information regarding the binding site profiles. The output for oPOSSUM CSA is similar, providing (i) a ranked list of over-represented TFBS class combinations, and (ii) a list of the most significant TFBS combinations (found in the set of expanded top-ranked class combinations). Based on the underlying assumption of the statistics employed that DNA sequences are randomly generated, there is little reason to accept the calculated scores as accurate reflections of significance. Instead, as suggested in the original published description of the oPOSSUM algorithm, we recommend that the scores are best used as rankings rather than significance measures. For this reason, a multiple testing correction is not applied as it does not alter the relative ranks. Empirically, we determined that TFBS profiles with Z-scores ⩾10 and Fisher scores ⩽0.01 facilitate the identification of relevant TFBSs for our sets of reference genes (). However, these are relatively stringent thresholds, and we encourage users to examine the scores of top-ranked TFBS profiles before applying any cutoffs. e o P O S S U M s y s t e m i s u n d e r c o n t i n u e d d e v e l o p m e n t . E f f o r t s a r e u n d e r w a y t o a l l o w u s e r s t o s u b m i t c u s t o m T F p r o f i l e s t o b e i n c l u d e d i n t h e a n a l y s i s . A n i m p r o v e d s e a r c h m e t h o d f o r n u c l e a r h o r m o n e r e c e p t o r s , w h i c h t y p i c a l l y c o n t a i n t w o h a l f s i t e s s e p a r a t e d b y a v a r i a b l e l e n g t h s p a c e r , h a s b e e n d e v e l o p e d a n d w i l l b e i n c l u d e d i n a f u t u r e r e l e a s e . W e w i l l c o n t i n u e t o a d d T F B S p r o f i l e s a s t h e y b e c o m e a v a i l a b l e , w i t h a n e m p h a s i s o n e x p a n d i n g t h e r e p e r t o i r e o f w o r m T F B S p r o f i l e s . W e b e l i e v e t h e o P O S S U M w e b s e r v e r i s a n d w i l l c o n t i n u e t o b e a u s e f u l r e s o u r c e f o r i n f e r e n c e o f m e c h a n i s m s o f c o - r e g u l a t i o n b a s e d o n o b s e r v e d c o - e x p r e s s i o n . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The sequential deletion method and other biological and biochemical experiments are generally used to locate the functional domain of a protein. For example, a previous study () used the sequential deletion method accompanied by manual PCR primers design to generate the N-terminal truncated mutants of different lengths (), which in turn were used in further biological experiments to decipher the functional domain of the 5S RNA-protein complex (5S rRNP). The 5S rRNP is believed to be formed by a co-translation event leading to the binding of the 5S rRNA to the nascent ribosomal protein L5. The formation of 5S rRNP complex facilitates the nuclear entry of the protein L5. Lin . () used an in vitro translation system to investigate how and when 5S rRNA triggers the formation of the eukaryotic 5S rRNP. The L5 and truncated L5 mutant mRNAs were prepared on a large scale for their investigation and a great amount of time was needed to manually modify the in-frame pattern of ATG start codon for conventional PCR and truncated mutant translation experiments. In order to save time and money needed in the traditional sequential deletion method, this web-based application system NTMG is proposed to automatically do the multiplex PCR assays design in order to generate the various N-terminal truncated mutants. Given a protein cDNA sequence, the NTMG will first find those ATG-like codons that are suitable to act as the starting positions of truncated mutants. Then, the NTMG will design the forward primers for all possible truncated mutants. Finally, with all these primers, the NTMG will choose those primers that can be divided into the least number of groups such that each group constitutes a multiplex PCR assay. In this section, we describe the input to the NTMG, the methodology of the NTMG and the output of the NTMG. Since the primer design and the multiplex PCR primer design are two important parts of the NTMG, some factors concerning the primer design such as the primer length and the melting temperature are input as parameters. There are also some factors that may affect the multiplex PCR amplification with multiple primers in the same tube (,). These factors include the cross-dimerization, the melting temperature, the products co-existence and others. All these factors are also input as parameters. Each input parameter has a default value but users may change that value. We first introduce the input parameters in the following subsection. A web-based application system called the NTMG is provided for researches who need to apply the sequential deletion method to locate the functional domain of a protein. After input the cDNA sequence, the NTMG automatically generates groups of primers. Each group of primers can be put in a tube and all tubes can be accommodated in a single batch of the multiplex PCR amplification under the same condition. Thus, time and money can be saved. We conducted a wet laboratory experiment on the multiplex PCR assay design proposed by the NTMG on input HL5 cDNA. The NTMG found 48 forward primers and one reverse primer and it divided them into 8 groups. In the wet PCR experiment, 44 PCR products had been found and the success rate is 91.7% (see Supplementary Data). Hence, the NTMG is of practical use to researchers who need to apply the sequential deletion method. As a future work, we plan to develop the more general multiplex PCR assay design. Given a set of PCR experiment requirements, we plan to develop a system that can automatically find the primers and try to divide the primers into as few groups as possible such that the primers in each group can be put in a tube and all tubes can be accommodated in a single batch of the multiplex PCR experiment. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Proteins involved in a majority of cellular processes usually perform their function by binding to some target proteins and forming protein–protein complexes. Interactions between two or more proteins often occur over short contiguous stretches of amino acids within one protein. For example, recognition of substrate proteins by various protein kinases during cell signaling events is governed primarily by specific interactions between the kinase and a contiguous peptide stretch containing the phosphorylation site. Several receptors have peptide fragments as ligands e.g. the major histocompatibility complex (MHC) (). Thus understanding molecular details of interactions between proteins and short peptide motifs is essential for dissecting underlying mechanism of several major cellular processes. Among the various proteins which interact specifically with short peptide motifs, protein kinases and MHCs represent two major protein families whose substrate specificities have been extensively studied by various experimental approaches (). Although a number of computational tools such as NetPhosK (), KinasePhos (), GPS (), Scansite (), SYFPEITHI (), ProPred (), etc. are available for predicting the putative substrate peptides for protein kinases and MHC proteins, these methods are mostly based on available experimental binding data for a given class of protein kinase or MHC. These tools predict substrate peptides based on identification of the conserved motifs in a set of known peptide substrates and do not use information from the three dimensional structure of the protein–peptide complex. Hence, these sequence-based prediction tools do not give information about key residues in kinases and MHCs which control substrate specificity. Information about specificity determining residues (SDR) can help in design of novel peptide ligands. Correct identification of SDRs of a given protein kinase or MHC can help in prediction of substrates for those protein kinases or MHCs for which no peptide-binding data is available, as demonstrated successfully in structure-based substrate prediction methods like PREDIKIN () and PREDEP (). These studies have demonstrated that structural analysis of interactions in protein–peptide complexes can lead to novel insight into the mode of substrate recognition. Therefore, molecular modeling of peptide–MHC and peptide–kinase interactions have been carried out by several groups using docking () or MD simulation approach (). However, the compute intensive nature of these calculations has limited such studies to few protein–peptide complexes. Since knowledge-based methods are less compute intensive, and have better prediction accuracy, development of suitable knowledge-based tools for modeling protein–peptide complexes would permit quick structural analysis of MHCs and protein kinases with their substrate peptides. A knowledge-based approach has been used recently for developing kinDOCK (), a powerful tool for modeling of ATP analogs into the active site pocket of protein kinases. However, no such user-friendly tool is presently available for knowledge-based modeling of peptides in the binding pockets of MHCs or protein kinases. Therefore, we have developed MODPROPEP, a web server for structural modeling of peptides of any desired sequences in the active site pockets of kinases/MHCs having known crystal structures or homology models of kinases/MHCs. In this manuscript, we give a brief description of the development of MODPROPEP, various assumptions made in the knowledge-based modeling protocol, various features of MODPROPREP and few examples of its use. The available crystal structures of MHC and protein kinases were downloaded from PDB website at (). The structures were divided into two groups, i.e. structures in complex with substrate peptide ligand and structures without the bound peptide ligand. These crystal structures were manually examined and chain/residue numbering was appropriately edited if necessary. All the crystal structures were categorized into three major classes, i.e. class I MHC, class II MHC and protein kinases. Each of these three classes was further grouped into various functional families of protein kinases or MHC alleles. Detailed analysis of these crystal structures indicated, that all the protein kinases shared a conserved structural fold despite their sequence divergence. For example, crystal structures of IR and PHK, which share a sequence identity of only 40% can be superposed with a C RMSD of 1.6 Å. Similar conservation of structures was also observed both for class I and class II MHC structures which share a higher degree of sequence identity within themselves. BLAST alignment of large number of protein kinases and MHC proteins available in sequence databases with these crystal structures indicated that, homology models can be obtained for most of these sequences with reasonable accuracy. Comparison of the bound peptide structures indicated that in all these three classes of proteins, the substrate peptides bind at a structurally homologous site on the conserved fold and the bound peptides maintain a more or less similar extended conformation. This suggested the possibility that bound peptides from peptide–protein complexes can be transformed to the protein structures lacking the bound peptide based on optimum superposition of the protein structures. It may be noted that similar assumption has been used successfully in structural modeling studies of protein–ligand complexes involving protein kinases (), MHCs (,) and other enzyme families (,). There are several examples where more than one crystal structure of an allele is found with bound peptides of different length. It is generally assumed that, three residues on each side of the phosphorylation site make significant contact with the protein kinase and are responsible for the specificity of a kinase (). Therefore, bound peptides having more than seven amino acids were truncated to three amino acids on either side of the phosphorylation site. All these structures were stored in the template library of MODPROPEP. The current template library of MODPROPEP has protein–peptide complex crystal structures for 16 alleles of class I, 12 alleles of class II MHC proteins and six different protein kinase families. shows a flowchart depicting various tasks which can be performed using MODPROPEP. For these MHC alleles and protein kinase families, substrate peptide of any desired sequence can be modeled. Modeling of peptide in the binding pocket of MHC or protein kinase is carried out by using the same backbone conformation as in the template complex and the side-chain conformations are generated by the program SCWRL (), which uses a backbone-dependent rotamer library approach. The template library of MODPROPEP has structures for many MHC alleles or kinases families without the bound peptide substrate. For modeling of peptide substrates in complex with any of these MHC alleles or kinase families, peptide conformations are transformed from the available crystal structures of the protein–peptide complexes after optimum superposition of the proteins. If no crystal structures are available for a given protein kinase or MHC protein, the program can model its structure in complex with peptides of desired sequence using the crystal structure of the closest homologous protein–peptide complexes. Sequences of various MHC alleles have been obtained from the IMGT/HLA database () and stored locally so that the user can select from the list of alleles the protein to be modeled. The crystal structure having maximum sequence similarity is used as a template for modeling the structure of query allele. All sequence alignments are carried out using a local version of the program BLAST. The SCWRL program is used for mutating the residues as per the BLAST alignment and generate the desired homology model. Since only protein–peptide complexes are used to generate the homology models, the backbone of the bound peptide is appropriately mutated by SCWRL to model the substrate of desired sequence. Thus, MODPROPREP provides options for modeling peptide of any desired sequence in complex with any MHC protein or protein kinase. In order to analyze the interactions between the peptide and the protein, the residues of the MHC or the kinase, which are in contact with different side chains of the modeled peptide, are identified using a distance-based cut off. Based on these contact residues, putative binding pockets are defined for each of the residues in the peptide. MODPROPEP provides a user-friendly Jmol java applet interface () for visualizing the modeled complexes and analyzing the binding pockets in detail. Apart from structural modeling of the peptide of a given sequence in the substrate-binding pocket of MHC protein or protein kinase, MODPROPEP is also capable of scanning an antigenic protein for potential MHC-binding peptides. Similarly putative substrate proteins for various protein kinases can be scanned for potential phosphorylation sites. Scanning of input sequence is done by breaking the protein sequence into all possible overlapping peptides of a given length. This length is usually the length of the bound peptide present in the template protein–peptide complex, i.e. 9 or 10 mer for class I MHC and longer peptides for class II MHC. However, for protein kinases only heptameric peptides containing Ser/Thr/Tyr as central residue are chosen. For each of these peptides, instead of building all atom side-chain conformations, as a first step, contacting residue pairs between peptide and the protein are identified based on C–C distances. The binding score of these peptides with the MHC or kinase is evaluated using residue-based statistical energy function by Miyazawa and Jernigan (MJ) (). It may be noted that a similar scoring scheme has been used earlier for identifying MHC-binding peptides using a threading approach (). Apart from MJ statistical potential, the program also has options for ranking peptide-binding affinities using residue-based statistical energy function by Betancourt and Thirumalai (BT) () or other user-defined residue-based schemes. The peptides are sorted according to their binding score and the user can select some or all of these peptides for detailed side-chain modeling by SCWRL depending on their preliminary scores. The program requires user to select a MHC allele or protein kinase from the pull-down menu. The program automatically shows the peptide length options available for modeling for that MHC allele or protein kinase. Program takes the user to available crystal structure templates for the selected protein and peptide length. From here the user can decide a task, which is either modeling of peptides or scanning a protein sequence for favorable binders. The user is prompted to enter the sequence of peptides as one letter code of amino acids. The program models the peptides in complex with the selected protein that are available for download as files in PDB format. If no ligand bound structure is available for the selected protein, the peptide is modeled by transferring the ligand peptide coordinates from a homologous protein–peptide complex. shows an example where a peptide has been modeled in complex with the kinase GSK3-beta by transferring the coordinates from CDK2. In order to test the accuracy of this ligand transformation approach, we modeled a peptide in complex with PKB by transforming the bound peptide from PKA. The tutorial section of MODPROPEP shows the superposition of the modeled and the experimentally determined bound peptide in the active site of PKB. As can be seen, backbone of both the peptides superpose quite well with an RMSD of 1.3 Å. MODPROPEP provides a user-friendly interface to analyze each modeled peptide in detail for contact with the protein. Inter-residue contacts can be calculated either based on the distance between C atoms or based on the distance between any two atoms in a pair of residues. A list of neighboring residues in the protein is displayed for each residue in the peptide. These amino acids on the protein define the binding subsite for each of the peptide residues. Residue pairs having steric clashes are highlighted in yellow. The program also provides interface for analyzing detailed atomic contacts between each pair of residue. Additionally, MODPROPEP uses Jmol applet for the rapid visualization of these subsites in the proteins. Mouse click on a peptide residue shows that residue and the neighboring residues in the protein in Jmol applet on right-hand side. Clicked peptide residue is depicted in ball and stick, while the neighboring residues are shown in CPK. The protein backbone is shown in ribbon while the peptide backbone is shown in the sticks. As mentioned earlier, the current version of MODPROPEP permits scoring various bound peptides using residue-based statistical scoring matrices given by MJ and BT. Both these scoring matrices have been used in the literature for evaluating binding energy of protein–peptide complexes. It has been reported that, while MJ potential gives better results for binding of peptides involving hydrophobic interfaces, BT potential is more appropriate for binding of peptides involving polar contacts. Here, we discuss a typical example of ranking the site of phosphorylation on the beta-adducin protein (accession no: P35612) by protein kinase A (). Out of a total of 118 S/T containing heptamers, RTPSFLK containing the experimentally identified phosphorylation site S713, is ranked 8 by MJ potential, while scoring by BT matrix gives it a rank of 3. Modeling of this peptide in complex with PKA shows R710 is stabilized by contacts with E127 and E170. The screenshots for this example are available at the tutorial page of MODPROPEP. Prediction of phosphorylation site in myosin III by PKA (), and PS1 by GSK3-beta () indicates that, the true phosphorylation sites identified in recent experiments are ranked as high-scoring peptides by MODPROPEP using BT matrix. shows the ranking of a recently identified class I MHC allele HLA-A*0201 ligand by MODPROPEP (). As can be seen, out of a total of 625 nonameric peptides present in the antigen CABL1_HUMAN (accession no: Q8TDN4), VALEFALHL has a rank of 13 and 26 by MJ and BT potentials, respectively. Analysis of inter-molecular contacts indicates that, this peptide is stabilized by interactions involving K66, A150, V152, Y159 and W167. We have also tested the predictive ability of MODPROPEP using all the known substrates of PKA cataloged in phospho.ELM (). Our results indicate that, in 76% of cases the true phosphorylation site can be ranked within top 30% using BT matrix. Similar benchmarking on 90 class I MHC–peptide complexes shows that, MODPROPEP can rank the true binder within top 30% in 61% of cases. MODPROPEP has been implemented using Perl, CGI scripts, java scripts, Jmol applet and apache web server. BLAST program downloaded from NCBI website is used for local alignments. SCWRL3 is used for the side-chain modeling. Various structural superpositions have been carried out using the program ProFit (). s a w e b s e r v e r f o r k n o w l e d g e - b a s e d m o d e l i n g o f p e p t i d e l i g a n d s i n t h e a c t i v e s i t e o f v a r i o u s M H C s a n d p r o t e i n k i n a s e s . T h e s o f t w a r e u s e s a v a i l a b l e c r y s t a l s t r u c t u r e s a s t e m p l a t e s a n d u s e s t h e p r o g r a m S C W R L t o m u t a t e t h e s e q u e n c e o f t h e p r o t e i n a s w e l l a s t h e p e p t i d e t o m o d e l a n y p e p t i d e – M H C o r p e p t i d e – k i n a s e c o m p l e x . I t p r o v i d e s a n u m b e r o f u s e r - f r i e n d l y i n t e r f a c e s f o r v i s u a l i z a t i o n a n d a n a l y s i s o f b i n d i n g p o c k e t s i n t h e s e p r o t e i n – p e p t i d e c o m p l e x e s . T h i s s o f t w a r e h a s b e e n d e v e l o p e d b a s e d o n t h e a s s u m p t i o n t h a t M H C s a n d p r o t e i n k i n a s e s h a v e c o n s e r v e d s t r u c t u r a l f o l d a n d t h e l i g a n d p e p t i d e s b i n d e s s e n t i a l l y a t t h e s a m e s i t e . A m a j o r a d v a n t a g e o f M O D P R O P E P o v e r o t h e r s t r u c t u r a l m o d e l i n g p r o g r a m s i s t h a t , i t c a n b e u s e d t o q u i c k l y m o d e l a l a r g e n u m b e r o f p e p t i d e s i n t h e b i n d i n g p o c k e t s o f M H C s a n d p r o t e i n k i n a s e s . T h u s M O D P R O P E P w i l l c o m p l e m e n t v a r i o u s a v a i l a b l e s e q u e n c e - b a s e d p r o g r a m s f o r p r e d i c t i n g p e p t i d e l i g a n d s f o r M H C s a n d p r o t e i n k i n a s e s . U s i n g t h i s s o f t w a r e t h e u s e r c a n i d e n t i f y a m i n o a c i d s o n t h e M H C o r k i n a s e s , w h i c h a r e c r u c i a l f o r s e l e c t i o n o f a p e p t i d e l i g a n d . S u c h i n f o r m a t i o n i s i m p o r t a n t f o r d e s i g n o f n o v e l p e p t i d e l i g a n d s o r a s s i g n i n g s p e c i f i c i t i e s t o n e w a l l e l e s o f M H C s o r n o v e l f a m i l i e s o f k i n a s e s . T h i s s o f t w a r e a l s o h a s a n o p t i o n f o r s e a r c h i n g t h e M H C - b i n d i n g p e p t i d e s i n t h e s e q u e n c e o f a n a n t i g e n o r p h o s p h o r y l a t i o n s i t e s o n t h e s u b s t r a t e p r o t e i n o f a p r o t e i n k i n a s e u s i n g s t r u c t u r e - b a s e d a p p r o a c h . P r e s e n t l y , t h e b i n d i n g e n e r g y i s b e i n g a c c e s s e d u s i n g r e s i d u e - b a s e d s t a t i s t i c a l p o t e n t i a l . T h i s s c o r i n g f u n c t i o n i s a p p r o p r i a t e f o r q u i c k p r e l i m i n a r y r a n k i n g o f p u t a t i v e p e p t i d e l i g a n d s . H i g h - r a n k i n g p e p t i d e s n e e d t o b e m o d e l e d a n d d e t a i l e d i n t e r a c t i o n s w i t h t h e p r o t e i n s s h o u l d b e a n a l y z e d f o r p r e d i c t i o n o f a c t u a l b i n d e r s .

Recent transcriptome studies have revealed that a large number of transcripts in mammals and other organisms do not encode proteins but function as noncoding RNAs (ncRNAs) instead. experiments have demonstrated important biological roles of noncoding RNAs, including regulation of transcription and translation, RNA modification and epigenetic modification of chromatin structure (). There is immense interest within the biological community to identify and study new noncoding RNAs. As millions of transcripts are generated by large-scale cDNA and EST sequencing projects every year, there is a need for automatic methods to accurately and quickly distinguish protein-coding RNAs from noncoding RNAs. Since to date no web server and few standalone tools have been designed for this purpose, researchers sometimes used tools developed for other purposes such as cDNA annotation and functionally domain identification (). However these methods showed varied performance on different datasets (,). Recently a new algorithm and standalone software named CONC was published that classifies transcripts as ‘coding’ or ‘noncoding’ using machine learning methods (). CONC showed improved performance over previous tools such as ESTScan (). However, CONC is slow for large datasets and does not have a web-server interface, limiting its usefulness. It works well with high-quality transcripts but may suffer from errors such as frameshifts which are common in ESTs and even occur occasionally in full-length cDNAs (). Furthermore, CONC only outputs the ‘coding’/‘noncoding’ classification but does not provide an explanation or related information. New tools are desired that are more accurate, run faster, and have a more user-friendly web-based interface. To assess a transcript's coding potential, we extract six features from the transcript's nucleotide sequence. A true protein-coding transcript is more likely to have a long and high-quality Open Reading Frame (ORF) compared with a non-coding transcript. Thus, our first three features assess the extent and quality of the ORF in a transcript. We use the software () to identify the longest reading frame in the three forward frames. Known for its error tolerance, can identify most correct ORFs even when the input transcripts contain sequencing errors such as point mutations, indels and truncations (,). We extract the LOG-ODDS SCORE and the COVERAGE OF THE PREDICTED ORF as the first two features by parsing the raw output with Perl scripts (available for download from the web site). The LOG-ODDS SCORE is an indicator of the quality of a predicted ORF and the higher the score, the higher the quality. A large COVERAGE OF THE PREDICTED ORF is also an indicator of good ORF quality (). We add a third binary feature, the INTEGRITY OF THE PREDICTED ORF, that indicates whether an ORF begins with a start codon and ends with an in-frame stop codon. The large and rapidly growing protein sequence databases provide a wealth of information for the identification of protein-coding transcript. We derive another three features from parsing the output of BLASTX () search (using the transcript as query, -value cutoff 1-10) against UniProt Reference Clusters (UniRef90) which was developed as a nonredundant protein database with a 90% sequence identity threshold (). First, a true protein-coding transcript is likely to have more hits with known proteins than a non-coding transcript does. Thus we extract the NUMBER OF HITS as a feature. Second, for a true protein-coding transcript the hits are also likely to have higher quality; i.e. the HSPs (High-scoring Segment Pairs) overall tend to have lower -value. Thus we define feature HIT SCORE as follows: The higher the HIT SCORE, the better the overall quality of the hits and the more likely the transcript is protein-coding. Thirdly, for a true protein-coding transcript most of the hits are likely to reside within one frame, whereas for a true non-coding transcript, even if it matches certain known protein sequence segments by chance, these chance hits are likely to scatter in any of the three frames. Thus, we define feature FRAME SCORE to measure the distribution of the HSPs among three reading frames: The higher the FRAME SCORE, the more concentrated the hits are and the more likely the transcript is protein-coding. We incorporate these six features into a support vector machine (SVM) machine learning classifier (). Mapping the input features onto a high-dimensional feature space via a proper kernel function, SVM constructs a classification hyper-plane (maximum margin hyper-plane) to separate the transformed data (). Known for its high accuracy and good performance, SVM is a widely used classification tool in bioinformatics analysis such as microarray-based cancer classification (,), prediction of protein function (,) and prediction of subcellular localization (,). We employed the LIBSVM package () to train a SVM model using the standard radial basis function kernel (RBF kernel). The C and gamma parameters were determined by grid-search in the training dataset. We trained the SVM model using the same training data set as CONC used (), containing 5610 protein-coding cDNAs and 2670 noncoding RNAs. We evaluated our method, named Coding Potential Calculator (CPC), by 10-fold cross-validation on the training data sets. The accuracy was 95.77%. For further evaluation we tested CPC on three large datasets including two non-coding RNA datasets from the Rfam 7.0 database () and RNAdb databank (), respectively, and a protein-coding RNA dataset from the EMBL nucleotide databank based on cross-links to the UniProt/SwissProt protein knowledgebase (,). We recorded the accuracy and computation time of CPC in , and compared it with CONC (version 1.01 downloaded from the authors’ website and installed locally). Both CPC and CONC were run in a Linux box with Intel Xeon 3.0G CPU and 4G RAM. Overall, CPC showed better accuracy on all three datasets with an order-of-magnitude faster speed (). For more stringent evaluations we removed those sequences in the three test datasets that were similar to one or more sequences in the training set (BLASTN E-value cutoff 1-2) and tested CPC on the remaining sequences. We also tested CPC on new entries in the latest UniRef90 release (version 10.1) which were not included in the previous release used to train CPC (version 9.4). In both cases the accuracy of CPC remained high (see section ‘More Stringent Evaluation’ and Table S1 in Supplementary Data). We then compared CPC with other prediction algorithms following the same evaluation strategy proposed by Frith . (). The results showed that CPC had the highest consistency with expert curation and performed well for the six challenging cases hand-picked by Frith . () (see section ‘Comparison with other protein-prediction algorithms following Frith .’ and Table S2 in Supplementary Data). CPC was also able to accurately predict 92% of the 2,849 short peptides with less than 100 amino acids (see section ‘Performance on Short Peptides’ in Supplementary Data). We developed a user-friendly web interface for CPC (). The CPC web server accepts a set of nucleotide FASTA sequences as input (allowing symbols ‘A’, ‘T’, ‘G’, ‘C’, and ‘U’). The sequences can be pasted directly into the input box or uploaded from a local sequence file. By default, the CPC server runs in ‘interactive mode’ in that results will be shown in the browser once the computation is finished. For a large set of sequences the user can input an email address to run his/her job in ‘batch mode’. The server will send a notice to the user's mailbox upon completion of the job. A unique ‘Task ID’ (TID) is assigned to each job by the web server. Users can use TID to track the job progress and retrieve the results which are saved on the server for 1 week. CPC summarizes the main output in a table (a). Each row corresponds to one input sequence. The columns show the sequence ID, the coding/noncoding classification, the SVM score (the ‘distance’ to the SVM classification hyper-plane in the features space), and a ‘Details’ link (as described later). In general, the farther away the score is from zero, the more reliable the prediction is. As a rule of thumb from our experience, the transcripts with score between −1 and 1 are marked as ‘weak noncoding’ or ‘weak coding’. Results in the summary table can be sorted interactively by sequence id, coding/noncoding classification, and SVM score; they can also be filtered by coding/noncoding classification, and SVM score. The current version of CPC cannot accurately discriminate transcripts falling entirely within UTR regions from true non-coding transcripts, because neither of them produces amino acid sequences. To handle this limitation, CPC provides the users the option to search database of known UTR sequences, UTRdb (), using BLAST (see section ‘Recognizing Potential UTR regions’ and Figure S1 in Supplementary Data). To ‘explain’ why a transcript is classified as coding or noncoding, CPC server provides detailed supporting evidence and other related sequence features of the input transcript in an Evidence page (b). The Evidence page shows the six features of the transcript, color coded for better visualization. It shows graphically the putative ORF identified by and the BLASTX hits. Mousing over, users can view details of each ORF and BLASTX hits. The Evidence page also provides options for querying the input transcript against well-annotated database, such as the functional domain database Pfam (), SMART () and SuperFamily (), UTRdb () and ncRNA database RNAdb (). The Evidence page aims to facilitate the user's detailed investigation of the transcript. We developed the CPC web server on a Java platform using JSP to render the dynamic HTML pages and Apache/Tomcat as the J2EE container. The web site is in compliance with W3C XHTML 1.0 Strict specification and works in both the Microsoft Internet Explorer and Mozilla Firefox browsers. A standalone version of the software is freely available for download on the web site, distributed under GNU GPL. A parallel version with simple distributed computing support is available upon request. t h t h e r a p i d l y i n c r e a s i n g a m o u n t o f d a t a g e n e r a t e d b y l a r g e - s c a l e t r a n s c r i p t o m e s e q u e n c i n g a n d i n t e n s i f y i n g a t t e n t i o n o n t h e s t u d y o f n o n c o d i n g R N A s , m e t h o d s t h a t c a n d i s c r i m i n a t e n o n c o d i n g R N A s f r o m p r o t e i n - c o d i n g o n e s w i t h h i g h r e l i a b i l i t y a n d f a s t s p e e d a r e i m p o r t a n t . I n t e g r a t i n g m u l t i p l e s e q u e n c e f e a t u r e s w i t h b i o l o g i c a l s i g n i f i c a n c e , C P C i s s h o w n t o h a v e g o o d a c c u r a c y i n b o t h c r o s s - v a l i d a t i o n a n d s e v e r a l t e s t d a t a s e t s . I t a l s o r u n s a n o r d e r - o f - m a g n i t u d e f a s t e r t h a n t h e p r e v i o u s s t a t e - o f - t h e - a r t t o o l , a n d t h u s i s m o r e s u i t a b l e f o r h i g h - t h r o u g h p u t a n a l y s i s . C P C u s e s f a r f e w e r f e a t u r e s t h a n C O N C d o e s ( 6 v e r s u s 1 8 0 ) b u t a c h i e v e d c o m p a r a b l e , e v e n b e t t e r , p e r f o r m a n c e i n t h e e v a l u a t i o n . T h e r e s u l t s d e m o n s t r a t e d t h a t t h e s e q u e n c e f e a t u r e s u s e d b y C P C h a v e p o w e r f u l d i s c r i m i n a t i n g p o w e r a n d m a y r e f l e c t t h e i n t r i n s i c p r o p e r t i e s o f c o d i n g t r a n s c r i p t . U s i n g f e w e r , s e q u e n c e - b a s e d f e a t u r e s a l s o s i g n i f i c a n t l y r e d u c e d c o m p u t i n g c o s t , t h u s r e m o v i n g a h u r d l e f o r a w e b s e r v e r t o b e d e v e l o p e d . A d d i t i o n a l i n f o r m a t i o n s u c h a s p o t e n t i a l f u n c t i o n a l d o m a i n s a n d s i m i l a r i t y t o k n o w n U T R r e g i o n s o r n c R N A i s u s e f u l t o u s e r s . T h i s a n d o t h e r s u p p l e m e n t a r y i n f o r m a t i o n i s a v a i l a b l e i n t h e E v i d e n c e p a g e s o f C P C , m a k i n g t h e r e s u l t s o f C P C m o r e e a s i l y i n t e r p r e t a b l e a n d b i o l o g y - m e a n i n g f u l . p p l e m e n t a r y d a t a a r e a v a i l a b l e a t N A R O n l i n e .

Increasing structural genomics projects have led to the exponential growth of the number of available protein structures. A few of these structures are annotated as hypothetical proteins as biochemical information is not available for them. Experimental functional characterization of proteins is a labor expensive and time consuming process. A computational tool is therefore useful to predict the functional site in a protein. The importance of such a tool is strengthened by the automation required for structural genomics projects. A large number of theoretical tools exist that attempt to predict functions of proteins on the basis of sequence or structural homology of the query protein with well-characterized proteins. However, proteins having sequence or structural similarity might not always perform similar biological functions (,). Proteins possessing different folds are also known to perform similar functions such as subtilisin-like proteases and trypsin-like proteases (). This discrepancy has led to the development of structure-based approaches wherein function is predicted on the basis of similarity of the spatial arrangement of functionally significant residues. Structure-based approaches () typically attempt to identify residues that might be non-contiguous in the primary sequence but are structurally analogous with a known structural template. Such approaches are guided by the fact that proteins perform similar function by maintaining the physicochemical environment of their functionally significant residues. This fact can be exploited to generate structural templates from active site geometries of known enzymes, and then comparing the newly determined structures with these templates. Methods that create structural templates from C atoms of the active site residues and their spatial neighbors have a drawback of lacking specificity and thereby giving rise to a large number of false positives. The other methods that use all the side-chain atoms of the key residues are too constrained and often overlook the small variations that might occur in the side chain placements. In our method, we use C atoms of the key residues along with the corresponding C atoms to form a template. These templates possess optimum specificity and flexibility to identify active site residues in query structures. The current method is highly specific for each functional class of proteins included here. The method is not affected by the small conformational variations and ambiguities in the placement of side-chains in the query structure. In addition to this the algorithm employed does not require any similarity to overall sequence or fold of known proteins. In this article we describe a web-server that executes this structure-based approach for predicting function. The method of Iengar and Ramakrishnan () has been modified and implemented in the current server. Structural templates are generated for the active site residues of different protease classes, glycolytic pathway enzymes and metal-binding sites. A training set was formulated from a set of known proteins of each functional family. The structural templates consist of active site residues’ identity and the geometrical parameters derived from their spatial environment. The geometrical parameters considered are the distances between the C and C atoms of the active site residues and the angle between the C plane and the C plane. The C and C planes are defined by the C or C atoms, respectively of the residues comprising the active site (a). Structural templates derived for proteases also considered the primary sequence order. Geometrical parameters for all the structural motifs are calculated and stored for the prediction (). MEROPS database (,) was used to form training sets for different protease classes. MEROPS classifies proteases into 47 different clans on the basis of their evolutionary origin. Structural templates could be generated for six of these clans. The clan identifiers along with the active site residue pattern are shown (). Templates could not be generated for other clans either due to non-availability of sufficient representative structures or due to involvement of less than three residues in the catalytic activity. The algorithm employed here performs well for all the structures solved by X-ray crystallography, NMR spectroscopy and theoretical structure prediction tools. In the case of NMR structures only first three models are considered for the prediction of functional site residues. However, the server is especially useful for structures modeled by threading, because inaccurate side chain placement in the threading-based models does not affect the accuracy of active-site residue prediction. A two-step procedure is used to identify active site residues for every query structure. In the first step coordinates of residues that can form the active site are extracted from the query structure file. In the second step spatial arrangement of the probable active site residues is determined in terms of geometrical parameters. These parameters are compared with the stored geometries of different functional classes. The user is required to provide a single query structure file in a PDB format. The user submitted structure files are accepted through an HTML form generated using CGI-Perl script. In order to search for a RCSB file, it has to be downloaded on a local machine and then submitted to PAR-3D. PAR-3D stores structural motifs for different protease classes, glycolytic pathway enzymes and metal-binding sites. The user can specify if they wish to search against one family of motifs or all the PAR-3D motifs. The uploaded file is first tested and verified for the PDB format. The structure data obtained are then processed by a set of PERL scripts that search for the stored structural templates. Output is provided in a tabular format describing the list of predicted active site residues. A sample output produced using yeast YDR533c structure (PDB ID: 1QVV) is shown in b. The first column lists the chain identifier. The second column provides residue name and the third column lists the residue number as defined in the uploaded file. The structural motifs stored here for the comparison are specific for different protease classes, glycolytic pathway enzymes and metal-binding sites. Therefore, output also provides information about the functional class of the predicted site in the query structure. g o r i t h m i m p l e m e n t e d h e r e p r e d i c t s t h e f u n c t i o n a l c l a s s o f t h e q u e r y s t r u c t u r e o n t h e b a s i s o f t h e s p a t i a l a r r a n g e m e n t a n d t h e r e s i d u e i d e n t i t y o f t h e p r e d i c t e d c a t a l y t i c r e s i d u e s . H o w e v e r , i t i s k n o w n t h a t s e v e r a l f u n c t i o n a l c l a s s e s b e l o n g i n g t o t h e s u p e r f a m i l y h y d r o l a s e s , s u c h a s a c e t y l - c h o l i n e e s t e r a s e s , a c e t o n i t r i l e s , l i p a s e s , s e r i n e c a r b o x y p e p t i d a s e s , a l l p o s s e s s s i m i l a r c a t a l y t i c t r i a d s a s s e r i n e c a r b o x y p e p t i d a s e s d u e t o s i m i l a r c a t a l y t i c m e c h a n i s m . T h e r e f o r e , a q u e r y s t r u c t u r e f r o m a n y o f t h e s e h y d r o l a s e s w i l l b e p r e d i c t e d a s s e r i n e c a r b o x y p e p t i d a s e s . The server (PAR-3D) has been tested extensively and response time has been observed to be 1–2 min. The local running time of the same was 0.3 CPU seconds on a silicon graphics workstation with R10000 processor. Server has been used to scan the complete PDB database. The statistics of PAR-3D performance upon scanning the complete PDB can be accessed at . The PAR-3D web server is freely available at . It can be accessed through a browser using any operating system. e b s e r v e r i d e n t i f i e s a c t i v e s i t e r e s i d u e s i n t h e q u e r y s t r u c t u r e u s i n g s t r u c t u r a l m o t i f s . T h e a l g o r i t h m u s e d f o r t h e s e r v e r h a s b e e n u s e d t o s c a n t h e e n t i r e P D B . P r e s e n t l y t h e s e r v e r s e a r c h e s f o r s t r u c t u r a l m o t i f s d e r i v e d f r o m p r o t e a s e s , g l y c o l y t i c p a t h w a y e n z y m e s a n d m e t a l - b i n d i n g s i t e s . W e a r e c u r r e n t l y w o r k i n g t o g e n e r a t e s t r u c t u r a l m o t i f s f o r o t h e r f u n c t i o n a l l y i m p o r t a n t s i t e s i n p r o t e i n s . W e a r e a l s o w o r k i n g t o i n c l u d e a f e a t u r e , w h i c h w i l l h e l p t o e n g i n e e r n e w c a t a l y t i c s i t e s i n e x i s t i n g p r o t e i n s . W i t h t h e a v a i l a b i l i t y o f s t r u c t u r a l m o t i f f o r s e v e r a l f u n c t i o n a l c l a s s e s o f p r o t e i n s , t h i s t o o l w i l l b e b e n e f i c i a l f o r s t r u c t u r a l g e n o m i c s p r o j e c t s .

Genomic structures corresponding to CRISPRs were observed first in 1987 in () and were subsequently reported in other organisms under different names [TREP (), SRSR (,), DRVs (), LCTR (), SPIDR ()] until the CRISPR acronym was proposed by Jansen . (). The direct repeat sequences carry in general a low level of palindromic symmetry; they are remarkably well conserved within a species (up to 248 exact copies in ). However, one of the flanking DRs is frequently truncated or diverged (see Supplementary Data). The DR size varies from 24 to 47 bp whereas the spacer sequence is generally within the range of 0.6–2.5× the DR size. The originality of spacers is that they apparently derive from conjugative plasmids or bacteriophages (,). A prokaryotic genome may harbour up to 16 CRISPR clusters with the same or a different DR. In a genome, a single CRISPR is generally associated with a family of genes called for CRISPR-associated (,), encoding proteins showing functional similarity with components of the eukaryotic RNA interference (RNAi) systems (). In addition, it was demonstrated in two archaea, () and (), that the CRISPR locus is transcribed into small RNAs (smRNA) probably from one of the flanking regions, the leader, acting as a promoter. These observations and the viral origin of spacers have led to the hypothesis that the CRISPR-associated system (CASS) is a prokaryotic defence mechanism against genetic aggressions (,,). Within species, CRISPRs may be present in a subset of strains, where they sometimes show polymorphism. The DR and the order of the spacers are well conserved, but the number of motifs (DR + spacer) differs from strain to strain. To better understand the mechanisms underlying the CRISPRs’ evolutionary scenario, three evolution rules were proposed by Pourcel . () and confirmed by Lillestol . (): (i) polarized acquisition of spacers near the leader sequence; (ii) random loss of motifs and (iii) shared ancestry when spacers are identical. CRISPRs’ analyses started in 1995 () but no specific stand-alone CRISPR software tool was created. Several software were used by different authors to identify these particular repeats but usually a manual discard of background was necessary, and generally some CRISPR clusters were missed or neglected, especially the shortest one (less than three motifs). This is the case, for example, of Tandem Repeat Finder () when considering a motif (DR + spacer) as a degenerate repeat (,), or Locating Uniform poly-Nucleotide Areas (LUNA), a program for finding degenerate repeats in microbial genomes on a desktop computer. The repeats can be filtered using several parameters including length, distance and level of conservation. LUNA was used especially for finding CRISPRs in archaea (,). Another program, Patscan () a pattern-matching tool that searches sequences fitting the introduced pattern, was applied to identify CRISPRs containing at least three () or four exact direct repeats (). PYGRAM () is a visualization program browsing all the repeats in the submitted genomic sequence and showing perfectly conserved palindromic repeats as pyramids. The PYGRAM program is mostly efficient in visually displaying large CRISPRs (CRISPRs with as many as seven motifs are considered as being very short in this work) since they will be recognized as a concentration of horizontal bars referring to a group of co-occurring repeats that differ by only a few nucleotides. Finally, Haft . () used REPfind (), a part of the REPuter package () and BLASTN to identify smaller repeat clusters. These programs are the most used tools in CRISPR detection, although none of them is especially conceived for this purpose. They require further manual manipulations to eliminate background data (tandem repeats for example) and importantly, do not define accurately the DR consensus (due to errors on the boundaries). Recently, two CRISPR- dedicated software tools were proposed, CRT () and PILER-CR (). Both of them run fast and perform well in finding CRISPRs. However, CRT results in a considerable background since tandem repeats are considered as putative CRISPRs and in addition, the same CRISPR is sometimes detected more than once with different consensus DRs. PILER-CR has also some drawbacks since it often misidentifies the DR boundaries and omits the truncated DR. In addition, there is no user-friendly dedicated web site. A specialized program to automatically identify CRISPRs seems to be mandatory for their optimum, rapid exploration and in-depth analysis, in order to increase the efficiency of CRISPRs investigations. CRISPRFinder is a web service offering fundamental tools for CRISPR detection, including the shortest ones, allowing an accurate definition of the DR consensus boundaries and extraction of the related spacers. It offers also additional tools to analyze the CRISPR loci: (i) obtain the CRISPR and the flanking sequences according to flexible size; (ii) make a blast of selected spacers or flanking sequences against the Genbank database and (iii) check if the DR is found elsewhere in prokaryotic sequenced genomes. CRISPRFinder core routines were developed in Perl under Debian Linux. The input of the web tool is a genomic query sequence of length up to 67 Mb in ‘FASTA’ format. Possible locations of CRISPRs (consisting of at least one motif) are detected by finding maximal repeats. A maximal repeat () is a repeat that cannot be extended in either direction without incurring a mismatch. The total number of maximal repeats in a sequence of size is linear (less than ) which is interesting since the computation may be done in linear time using a suffix-tree-based algorithm. A CRISPR pattern of two DRs and a spacer may be considered as a maximal repeat where the repeated sequences are separated by a sequence of approximately the same length. The operation of the program can be divided into four main steps summarized in : (Step 1) browsing the maximal repeats of length 23–55 bp interspaced by sequences of 25–60 bp, (Step 2) selecting the DR consensus according to a defined score taking into account the number of occurrences of the candidate DR in the whole genome and privileging internal mismatches between the DRs rather than mismatches in the first or the last nucleotides, (Step 3) defining candidate CRISPRs after checking if they fit CRISPR definition, (Step 4) eliminating residual tandem repeats. In the first step, maximal repeats are found by the software Vmatch (), the upgrade of REPuter (). Vmatch is based on a comprehensive implementation of enhanced suffix arrays () which provides the power of suffix trees with lower space requirements. A one nucleotide mismatch is allowed permitting minimal CRISPRs with a single nucleotide mutation between DRs to be found. Hereafter, the obtained maximal repeats are grouped to define regions of possible CRISPRs with a display of consensus DR candidates related to each cluster. The second step is aimed at retrieving the DR consensus of each cluster. The difficulty resides especially in the identification of boundaries, which is very important to extract the correct spacers and compare DRs. In fact, the consensus DR is selected as the maximal repeat which occurs the most in the whole underlying genome sequence with respect to the forward and the reverse complement directions (since two CRISPRs having the same DR consensus may be in opposite directions). Thus, ambiguity in the choice of a DR will be eliminated in the case of presence of similar DRs in other CRISPRs of the related genomic sequence. However, if occurrence numbers are equal, more than a single DR consensus candidate are kept and later compared. Given a candidate consensus DR, the pattern search program fuzznuc of the EMBOSS package () is applied to get DRs’ positions in the related cluster. As the first or the last DR in a CRISPR may be diverged/truncated, a mismatch of one-third of the DR length is allowed between the flanking DRs and the candidate consensus DR, whereas smaller nucleotide differences are allowed between the other DRs to take into account possible single mutations. In case of multiple DR candidates, a score is computed and the best one (minimum) is picked. This score favours candidates which are encountered more frequently, rather than consensus DR showing less internal mismatches. Once the DR consensus is determined, the corresponding spacers (Step 3) are extracted according to the DR boundaries determined previously. The spacer length is not allowed to be shorter than 0.6 or longer than 2.5 times the DR length. These sizes are in the range of CRISPRs described in the literature. The last step consists in discarding false CRISPRs. Therefore, tandem repeats are eliminated by comparing the consensus DR with the spacer if there is only one spacer, or by comparing spacers between each other. The comparison is done with the CLUSTALW program () and the percentage of identity between spacers is not allowed to exceed 60%. Finally, candidates having at least three motifs and at least two exactly identical DRs are considered as confirmed CRISPRs. The remaining candidates are considered as questionable. These should be critically investigated by, for example, checking for intraspecies size variation of the locus. e q u e r y s e q u e n c e m u s t b e i n ‘ F A S T A ’ f o r m a t . N s c h a r a c t e r s a r e a c c e p t e d , I U B / G C G l e t t e r s ( M R W S Y K V H D B X ) w i l l b e c o n v e r t e d t o N s a n d c o n s i d e r e d a s m i s m a t c h e s b u t a n y o t h e r c h a r a c t e r s w i l l b e d e l e t e d . O n e c a n e i t h e r p a s t e t h e g e n o m i c s e q u e n c e i n t o t h e i n p u t f i e l d o r u p l o a d i t f r o m a f i l e o n t h e l o c a l m a c h i n e . M u l t i s e q u e n c e f i l e s a r e a l s o a l l o w e d b y t h e p r o g r a m a n d w i l l b e t r e a t e d i n d e p e n d e n t l y . U s e r s m a y u s e t h e d e f a u l t v e r s i o n o r c l i c k o n t h e ‘ a d v a n c e d v e r s i o n ’ l i n k t o s e t a n d m o d i f y a l l t h e p r o g r a m p a r a m e t e r s , w h i c h m a y b e e s p e c i a l l y u s e f u l f o r f i x i n g t h e D R s i z e . italic fig #text CRISPRFinder is a program that allows the identification of structures with the principal characteristics of CRISPRs, the smaller being composed of a truncated or diverged DR, a spacer and a complete DR. In their analysis, Godde . () using Patscan had chosen to retain only CRISPRs with at least three exact repeats (eliminating CRISPRs constituted of a first truncated repeat plus two exact repeats) thus ignoring most CRISPRs containing less than three spacers. Similarly in the work by Durand et .(), the PYGRAM program is mostly efficient in visually displaying large CRISPRs. Such stringent criteria were appropriate in order to avoid ambiguities in early investigations which were essentially describing these new structures. However, it is now important, in order to better understand the evolution and spreading of CRISPRs, to provide tools which will not eliminate the smallest CRISPRs. This is what we chose to achieve with CRISPFinder. The major drawback is that when looking for the shortest structures, such as those with a unique spacer, it is clear that the background of spurious candidates can be very high. The output of Patscan and CRT also contains a large quantity of noised data that needs a manual treatment. CRISPRFinder is accessible on the web and submission is very simple. We provide several samples on the website as demonstrators. Upon submission of the complete genome of VF5 (sample1), five confirmed and five possible CRISPRs are displayed in the following pages. On the contrary, while using the webservice for Patscan (), it is necessary to first define a pattern (which is not straightforward) and it is not possible to seek for CRISPRs in a single genomic sequence but rather in an entire predefined database. In addition, Patscan requires a Sun machine for local implementation. Similarly, PYGRAM only runs on linux systems and its installation requires advanced skills. CRT requires either to install JRE (Java Runtime Environment) or compile the source files, and PILER-CR needs to be compiled before use. A comparison between layouts of available online programs (REPuter, Patscan, TRF) and of CRISPRFinder is provided in the Supplementary Data. To check that CRISPRFinder was efficient in recovering all the CRISPRs from a genome, we compared the results to other available studies on CRISPRs (,). The data were generally in good agreement, the differences being always in the DR boundaries’ identification (more accurate with CRISPRFinder) or in the number of motifs found, as the truncated DR is sometimes neglected or short clusters are not detected with other programs. Interestingly, some strains were claimed to be devoid of CRISPRs by Godde and colleagues but proved to have short CRISPRs with CRISPRFinder, such as in different sp. ( Ss046, 2a str. 301, 2a str), or even long CRISPRs such as in . The latter example is shown on the CRISPRfinder website (sample2), and as can be seen by using the BLAST spacer function, six spacers out of thirty six at two different CRISPR loci, correspond to a bacteriophage sequence (bacteriophages F116, B3, D3112, DMS3 and phi CTX). The tools developed here will assist in future CRISPRs’ analysis. Furthermore, the possibility to identify CRISPRs containing one or two motifs may help understand how new CRISPRs are created. The very small candidates will need to be typed across different isolates within the same species or very closely related species to search for variations. For instance, as shown with the sample file provided on the website (YP1 Yersinia), five strains possess at the same CRISPR locus two to eight spacers, some being unique and others shared by two or more strains (). This strain-dependent polymorphism is especially interesting for epidemiological and phylogenetic studies (,). A tool to easily create a dictionary of spacers from different strains is proposed in a CRISPR-dedicated web database (). The CRISPRFinder web server is an interface to extract with precision and to further analyse CRISPRs from genomic sequences. Four main advantages may be cited: (i) short CRISPR-like structures are detected, they are labelled questionable but may be of great interest if later confirmed; (ii) DRs are accurately defined to single base pair resolution; (iii) summary files may be uploaded (CRISPR properties summary and spacers file in Fasta format) and (iv) flanking sequences or spacers can be easily extracted and blasted against different databases. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

A variety of experimental genome annotation technologies provide counts, probabilities or intensity values for chromosomal positions spread over a complete or partial genome. Such technologies include cDNA and tag-sequencing protocols to map the 5′ and 3′ ends of mRNA (,), ChIP–chip analysis to reveal transcription factor binding sites and epigenetic markers, and high-density SNP profiling platforms for various kinds of genotype–phenotype association studies. A recurrent problem in the processing of such data is the identification of individual clusters (alternatively called peaks or islands) by some kind of signal detection and noise-filtering algorithm. The analysis of genome annotation data challenges any clustering software in many respects. Genome coordinates with available quantitative measurements are often not evenly distributed over the analyzed chromosomal range. Furthermore, the horizontal axis is discrete in nature and the size and shape of the target objects are in most cases largely unknown in advance. Moreover, the position-specific readout from the experimental protocol may be a function of several overlaid biological and technical processes For instance, in the case of promoter mapping, the number of cDNA 5′ ends at a given position is thought to reflect primarily transcription initiation events, but technical artifacts or premature termination of cDNA synthesis may also contribute to the signal. Peak-recognition algorithms exist in many variants (,) and their application is described for instance for the analysis of MS data and chromatographic profiles or time series (). However descriptions of applications of these methods to genome annotation data are sparse. Existing methods frequently lack flexibility and at the example of the package mclust in R (), implementation of additional constraints is difficult. These methods are typically based on more or less explicit physical assumptions about the signal-generating process and peak shapes, and therefore cannot be directly ported to new applications. Standard clustering algorithms are furthermore prone to be perturbed by atypical distributions. Often, there is no experimental gold standard available for evaluation of the results. In such cases, the ultimate reference remains human intuition applied to representative examples. We initially developed the program MADAP for the inference of promoters from mRNA 5′ end profiles obtained from the mapping of full-length cDNAs to the genome sequence (). More recently, we discovered the usefulness of MADAP for the interpretation of ChIP–chip data. Given the potential of an even more general use, the web server presented here is primarily intended to enable a rapid evaluation of the suitability of MADAP for new kinds of data. Finding appropriate parameter settings likely represents the largest obstacle in applying MADAP to a new genome annotation problem. Please note that previously established parameters might not be suited for a novel data set with distinct characteristics such as background noise. The server may also be used for small-scale productive applications. However, for large-scale genome annotations task, we recommend to use a local installation of the program which can be freely downloaded from our FTP site. The MADAP web server takes as input an uploaded file containing a set of tabulator-separated numbers representing for instance positions on a chromosome. The numbers describing the data points occur at a frequency corresponding to the strength (or intensity) of a measure at this given position, for instance the number of times a 5′ end of a full-length transcript () is observed at this position. Alternatively, data can be supplied in a file in gff format (). In this case, a frequency score of each feature has to be indicated in column 6 (‘score’) of the gff file. The function of MADAP is to determine the most likely model describing the input data set. A mixture of normal (Gaussian) distributions with centers, standard deviations and relative frequencies are used to model the data points, an approach also known as a mixture modeling. Although the shape of the distributions of the clusters hidden in the input data is frequently not known and not necessarily resembling a normal distribution, we observe that the algorithm of MADAP as further described later copes remarkably well with most kinds of distributions. Using normal distributions has the advantage that the probability of observing a unit event (e.g. one cDNA 5′ end) at a specific position can easily be computed given the center position and standard deviation of a set of normal distributions. The number, the location of the centers and the relative frequency of the normal distributions are initially deduced from the data. Using a standard Expectation Maximization (EM) approach () MADAP optimizes the center positions of zero to many clusters and their standard deviation. A known drawback in this approach is that isolated points distant from the cluster centers (leverage points) can have an undesirable high impact on the selection of the model that best describes the clusters of points. In order to control for disturbing isolated points, we add to the mixture model an additional non-Gaussian, uniform ‘background’ component, thereby reducing the negative influence on the model selection. MADAP thus differs from the standard algorithm () in the addition of a background distribution and in the possibility to specify a set of additional constraints explained in the following summary of the optimization steps of the model. In a first model initialization step, the program generates several initial models for each possible number of clusters. Minimal and maximal numbers of clusters can be defined by user-specified parameters; the initial numbers of clusters are additionally limited by the number of distinct data points in the input data. The center positions of the clusters are initially attributed to data points with the highest frequency. Data points within a neighboring ‘integration range’ are included into each initial cluster. A second parameter for background control subtracts a user-defined constant from all positions for model initialization. The subsequent steps are calculated again with full data. In next steps, each of the initial models is iteratively evolved using the EM algorithm. The initial centers of clusters are optimized until stabilization of the data likelihood or discarded if a maximal number of iterations is reached. In a third step, models resulting from the EM step are required to comply with user-defined model constraints. Such constraints include a minimal number of data points attributed to a cluster and a minimal distance between peaks of neighboring clusters. Noncompliant clusters in models are removed and the EM step described earlier is repeated with a model with a reduced number of clusters. If there is no cluster left, the model is rejected. Models satisfying all constraints are recorded. Upon the optimization of all initial models, the final model is chosen as the one with the highest data likelihood. Two variants for likelihood computation are offered: the usual likelihood under a mixture model (), and a likelihood that is calculated after attributing each data point to the cluster with the highest density at that position (see explanatory document on web server). Due to limits in computational resources, the web server version of MADAP features a few restrictions, notably in the number of initial clusters. Users are advised to either split up their data sets to ranges putatively containing less than 50 clusters, or to install MADAP on their local computers. For the analysis of larger data sets on our infrastructure, please contact the authors. Further descriptions of the algorithm and its parameters can be found on the site of the MADAP web server. In the following we are going to describe with two examples how MADAP can be used for definition of promoters from full-length cDNA data or from ChIP–chip data. Note that this is a partly exploratory (unsupervised) data analysis problem. Given a spectrum of transcription start sites over a genomic range, there is no objective way to answer the question of how many promoters they represent. The consensus answer is likely to come from an interacting learning process with new methods and data. We therefore equipped the MADAP algorithm with a variety of parameters that enable the user to guide the partitioning process in a desired direction. Default parameters on the web server correspond to optimized values for the transcription start site (TSS) task. In particular, we try to relate the parameters chosen in this application to assumptions about the biological signal, in the following presented in the estimated order of importance. Aiming for high robustness and precision of TSS mapping, we required at least 10 counts (here cDNA 5′ ends) per clusters (parameter = 10). We assumed that in average ∼70% of all full-length transcripts initiate within 20 bp of its ‘main’ TSS, thus being best described by a fixed standard deviation of the initial Gaussian components equal to 20 ( = 20). Alternative promoters were defined as neighboring TSS having a minimal distance of 50 bp ( = 50). Parameters ( = 1) and ( = 16) specify the range for the number of clusters in the initial models. Computation time increases significantly with increasing ranges, because the more models have to be tested. Parameter ( = 0) defines the background subtraction for model initialization and parameter ( = 0.02) represents an estimate of the proportion of data points that belong to a random point background distribution. Optimal values for these parameters strongly depend not only on the kind of application, but also on the actual data set with, for instance specific noise levels. Parameter ( = 5) defines an integration range within which data points are initially attributed to a cluster center. Parameters ( = 6) and ( = 11) specify extended reporting ranges, for which the number and the fraction of points is reported in the text output, without influence on the clustering. This led to the following parameter settings (default on the web server, resulting output shown in Supplementary Figure 1): Due to the fact that exact characteristics of the putative clusters in the input data are not known, some parameter settings may appear arbitrary. A second demo file provided on the web server is derived from a ChIP–chip experiment using the Nimblegen platform with an antibody against a component of the preinitiation complex (). Data describing a ∼250 kb segment of chromosome 12 was extracted from the GEO database, and the locations of the probes were remapped onto the current human genome assembly (NCBI 36). The intensity of the hybridization signal on the chip was transformed into integers by a simple exponentiation to the power of 10, setting an arbitrary maximum of 200 (the numerical values provided by GEO represent log-intensities). The input file provided in gff format contains 336 genome coordinates associated with a total of 3357 digitized intensity units. This demo gff file was submitted to MADAP with the following parameters: Changes in parameters , , , and relate to the larger cluster width expected for ChIP signals. The background probability ( = 0.002) was adjusted to a lower noise level. shows the output of the MADAP web server on this second demo file aligned with the corresponding genome annotations visualized by the ENSEMBL genome viewer (). On the MADAP web server, the results are presented graphically on top of histograms of the input data. In an overview plot, the distribution of determined clusters is displayed as numbers plotted at the approximate location of the corresponding cluster. A detailed view of each cluster allows the determination of the center and the estimated standard deviation. In addition to the visual representation of the location of the inferred clusters, the output of the MADAP server includes parseable output files in text format. The main results of MADAP are reported in a file ‘output’ containing a recapitulation of the parameters used and a description of the clusters found under the models described above. Supplementary files are intended to help to track the iteration behavior of MADAP, including the files ‘components’ with an outline of iteration steps, and a ‘summary’ file resuming properties of optimized models. Eventual problems encountered during execution of the program are reported in an ‘error’ file. We present here a web server for the clustering program MADAP, which was initially developed for the determination of TSS (). Although MADAP uses internally normal distributions, it was designed to model non-contiguous distributions of any shape and proved to be remarkable robust in this aspect. Others have exploited cDNA full-length sequencing data or 5′ tags (5′ SAGE, CAGE) for promoter mapping and have presumably developed alternative solutions to the TSS clustering problem. However, to our knowledge none of these methods has been explicitly described or made public via web servers. MADAP is in principle versatile enough to interpret data from any source in terms of a finite number of clusters characterized by center positions, volume and extension. In the scope of primary genome annotation data, we envisage an extended usage of MADAP in the analysis of 3′ ends of transcripts and the inference of polyA signals, and on data derived from ChIP–chip, or from tiling arrays. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The function of RNA molecules often depends on both the primary sequence and the secondary structure. RNAs are involved in translation (tRNA, rRNA), splicing (snRNA), processing of other RNAs (snoRNA, RNAseP) and regulatory processes (miRNA, siRNA) (). Furthermore, parts of mRNAs can adopt structures that regulate their own translation (SECIS (,), IRE ()). Since prediction and experimental determination of 3D RNA structures remain difficult, much work focuses on problems associated with its secondary structure, which is the set of base pairs. The problem of predicting the secondary structure of an RNA is called the ‘RNA folding problem’. Existing computational approaches are based on a thermodynamic model that gives a free energy value for each secondary structure (). The structure with the lowest free energy [called the ‘minimum free energy (mfe) structure’] is expected to be the most stable one. Here, we consider the ‘inverse RNA folding problem satisfying sequence constraints’, which is the design of RNA sequences that fold into a desired structure and fulfill some given constraints on the primary sequence. These constraints can restrict certain positions to fixed nucleotides or to a fixed set of nucleotides. The INFO-RNA web server is applicable to the design of RNA elements that include conserved nucleotides, which are essential for binding of proteins. #text The INFO-RNA web server allows biologists to design RNA sequences, which fold into a given structure, in an automatic manner. The procedure is fast, as most applications are completed within seconds. As shown in (), INFO-RNA (not considering sequence constraints) proceeds better and faster than other existing tools. Artificial as well as biological test sets were analyzed. The biological test sets divide into computationally predicted structures for known RNA sequences and structures from the biological literature. INFO-RNA turned out to be the algorithm having the highest succession rates as well as the lowest computation times for all test sets. Additional stability tests showed that the designed sequences are more stable than the biological ones. The novel extension of INFO-RNA including sequence constraints allows the design of -acting mRNA elements such as the ‘iron responsive element’ () and the ‘polyadenylation inhibition element’ (). Both elements have conserved sequence positions in loops. The IRE is essential for the expression of proteins that are involved in the iron metabolism (). It consists of a stem-loop structure, and the first five nucleotides in the hairpin loop as well as the bulged nucleotides were found to be essential for binding of iron-regulatory proteins. The PIE contains two binding sites for U1A proteins (). It consists of a stem structure with two asymmetric internal loops that serve as U1A-binding sites (). Using the INFO-RNA web server, we designed artificial IREs and PIEs having a much higher folding probability compared to natural elements. While designed sequences for the IRE having a single C bulge fold into the target structure with an average probability of 88%, natural sequences do so only with an average probability of 15%. Regarding IREs having an interior loop with left size 3 and right size 1, the results are similar. Furthermore, the average probability of the designed PIE sequences folding into the target structure is more than 20 times higher than the probability of the natural PIE sequences (Supplementary Figure 1). Besides, all IREs designed by the INFO-RNA web server adopt the wanted structure as its mfe structure whereas only a small fraction of the natural ones does (Supplementary Figure 2). Furthermore, we demonstrated the usability of the INFO-RNA web server by designing artificial microRNA (miRNA) precursors that are as stable as possible. To this end, artificial miRNA sequences published in () were used. Applying the INFO-RNA web server, we designed precursors of these artificial miRNAs as well as of the natural miRNA. All of the designed sequences have a free energy that is at least twice as low as the free energy of the natural precursor sequences. On average, their probability of folding into the target miRNA precursor structure is five times as high as the folding probability of the natural precursor sequences. For more details see Supplementary Table 1. Other potential application areas are the design of ribozymes and riboswitches (), which may be used in research and medicine, and the design of non-coding RNAs, which are involved in a large variety of processes, e.g. gene regulation, chromosome replication and RNA modification (). We have shown that the INFO-RNA web server is a very fast and successful tool to design RNA sequences, which fold into a given structure and fulfill some sequence constraints. The core of the algorithm was introduced in (). There, we already showed that INFO-RNA (not considering sequence constraints) proceeds better and faster than other existing tools. Here, we have demonstrated that the INFO-RNA web server, which can handle additional constraints on the primary sequence, also performs well and fast. Most of the sequences designed by the INFO-RNA web server are highly stable and have very low free energy. This might result from the high GC content that most of the sequences show since G–C base pairs are energetically most favorable. It is not clear whether such highly stable structures are always of advantage or how the high GC content may influence the kinetics of the folding process. To reduce the GC content, the user can constrain some positions to A and/or U. In the future, it is desirable to extend the algorithm to allow the user to specify the GC content. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The most intuitive decomposition of the binding free energy involves four terms (): van der Waals (vdW) interactions, electrostatic, hydrophobicity and configurational entropy. The relative contribution of the changes between the bound and free states of these four terms is not the same. For stability (), the main contributions appear to be electrostatic and desolvation interactions. For refined docked conformations, vdW interactions are expected to balance between the bound and unbound state, as they seemingly do in protein folding (). This is good news, since it is not yet possible to readily estimate solute–solvent interactions. It should be noted, however, that solute–solute vdW has been shown to be an important consideration for complex refinement (). Configurational entropy loss upon binding, including rotational and translational degrees of freedom, is always important, rough estimates based on crystal complexes varying between 5 and 15 kcal/mol (,). For the most part, this entropy depends on the flexibility of the unbound or free state with respect to the bound, with smaller corrections depending on the docking geometry. Since there is no robust estimate of entropy for a given protein, empirical free energy estimates, like ‘FastContact’, are always subject to an entropic correction. Hence, the server is most useful for discrimination between protein–protein docked complexes, and, more generally, for identifying energetically important contacts at the interface. ‘FastContact’, originally published in (,) rapidly estimates the electrostatic and desolvation component of the free energy based on a classic distance dependent dielectric 4 () and an empirical contact potential for the desolvation contribution () developed using a database of crystal (no complexes) structures from the PDB. Because of the pairwise nature of the empirical interactions, ‘FastContact’ is also able to report the contribution of individual residues and pairs of residues to the free energy. The latter should prove useful for site-directed mutagenesis studies since rankings of these interactions consistently identify the hot spots in the interface. The code behind the server was written in Fortran 77 and the server itself was written in PHP. ‘FastContact’ performs a fast computational estimate of the binding free energy between two proteins based on atomic pairwise interactions: The first two values (i–ii) can be used to calculate the overall free energy of the protein–protein interactions, assuming solute and/or solvent vdW cancellation between the bound and free proteins, and a correction factor for the configurational entropy loss. The application uses the definition of the atomic composition of each amino acid consistent with CHARMm19 parameters. shows a snapshot of the input page. The user uploads two Protein Data Bank (PDB) format files (), one ‘receptor’ and one ‘ligand’, along with their email address. The web server currently makes no distinction between chains; it simply reads in each line in the PDB file starting with an ‘ATOM’ field. The maximum number of residues is limited to 1500. The email address is where the output/results will be sent (as a file attachment). Hydrogen bonds and missing atoms are built and optimized on the uploaded structures using the molecular software CHARMm. #text T h e d e f a u l t s e t t i n g f o r H y d r o g e n b o n d o p t i m i z a t i o n a n d r e m o v a l o f m i n i m a l o v e r l a p s p r e s c r i b e s a s h o r t 3 × 2 0 A B N R m i n i m i z a t i o n s t e p s w i t h f i x e d b a c k b o n e u s i n g t h e p r o g r a m C H A R M m a n d t h e P A R A M 1 9 r e s i d u e t o p o l o g y f i l e ( R T F ) . H o w e v e r , t h e u s e r i s f r e e t o c h a n g e t h i s s e t t i n g t o a f u l l a t o m m i n i m i z a t i o n . T h i s s e t t i n g w i l l w o r k f o r s i n g l e c h a i n s o n l y a n d n o g a p s . d e f a u l t , t h e e n d t e r m i n a l r e s i d u e s w i l l b e p a t c h e d b y C H A R M m . I n c a s e t h e e n d t e r m i n a l s a r e m i s s i n g f r o m t h e s t r u c t u r e , t h e u s e r h a s t h e o p t i o n o f t u r n i n g t h e p a t c h i n g f e a t u r e o f f . The main results file (‘output.txt’) returns two components of a free energy function, electrostatic energy and desolvation free energy, and evaluates the solute vdW energy using CHARMm. The latter is sometimes useful to compare between different models (), but here it is given only as a reference since it is not used in the analysis of contacts. Often vdW energies larger than about −500 kcal/mol suggest structural overlaps. Although ‘FastContact’ smoothes the potentials to tolerate some limited overlaps, these are, in general, detrimental to the quality of the computational estimates. shows the summary energy output and part of the contact analysis, for the barnase–barstar complex 1BRS. We should caution that, when submitting co-crystallized receptor and ligand structures, the automated minimization implemented in the server leads to an over optimization of the electrostatic contacts of ∼10–20% (). The reason is because the direct electrostatic term used in the server does not have an angular dependence for Hydrogen bonds. Hence, these interactions tend to ‘double-dip’. This effect is compensated when scoring unbound models that always have some built in frustration due to the less optimal backbone and side chain conformations. p ext-link xref #text ext-link #text

Searching for pertinent literature is an essential part of every scientist's life. There are many stages in the scientific process in which intimate knowledge of the appropriate literature is critical: (i) familiarization of a new area by a young scientist or a scientist whose research is taking on a new direction, (ii) monitoring the literature as the research progresses to capitalize on recent developments, measure ones competitiveness and avoid duplication of effort (), (iii) development of reference lists during manuscript or grant application writing and (iv) compiling suggested reviewers when called upon to do so as part of a manuscript submission to a journal. For mature scientists, the reasons for interaction with the literature expand: (i) development of very broad knowledge when writing, for example, a review article, (ii) mastery of new areas in the role of student mentor or examiner and (iii) acquiring focused knowledge when called upon as a manuscript or grant application reviewer. For other scientific professionals, the literature is a resource for identifying colleagues: (i) identification of experts for advisory or steering committees, (ii) selection of reviewers for grants or proposals by government or private agencies, (iii) identification of experts for legal proceedings and testimony, (iv) finding starting points into the literature for novice or lay individuals by librarians and (v) identification of manuscript reviewers by journal editors. The primary portal for the biomedical literature is PubMed (,). This web-based tool searches the Medline database using keywords and Boolean operators. The selection of appropriate keywords by the user requires some knowledge to choose wisely, and this often requires numerous iterations to sample the literature with hopes of finding the most relevant literature. Once the results of a query are presented to the user, the lists can be sorted by date, author or journal. Recent research has focused upon improving the quality and navigation of output (). There is sufficient information contained within the Medline database to overcome these limitations given a tool with appropriate query entry and result presentation methods. Scientists or professionals either generate in the course of manuscript or grant writing or are presented with concentrated information in the form of an abstract or other document. Given this, the keyword selection and optimization process can be bypassed if natural language free text, such as an abstract, can be submitted directly to a literature search engine. To do this, we have developed eTBLAST, which uses a hybrid scheme to extract and weight keywords contained within the submitted query to identify a subset of literature in Medline, and then performs a sentence alignment to compute a final quantitative score as a measure of similarity and, presumably, relevancy. This tool then outputs a list, similar to PubMed, but ranked instead by this similarity score. At this point, scientists can interact with the most relevant Medline literature much as they have done traditionally via date, author or journal sorting methods in PubMed. This similarity-ranked output can be further processed to compile lists and present output views which add value for the specific uses just outlined; identifying the most frequent and prominent authors as experts/reviewers, identifying the most frequent journals as targets for submission and inspection of the publication rate over time as a measure of novelty and topic popularity. It should be noted that eTBLAST and PubMed both find similar abstracts, but by different methods and PubMed's Related Links is limited to only finding similarity among the records currently in Medline, not arbitrary text, as is used by eTBLAST. There also are numerous other Medline keyword-based search tools (CiteXplore, HubMed and GoPubMed, for example) (), including some of which have results post processors with some similar functionality (author and journal finding). Summarized herein are a set of parsers for the code, eTBLAST (,), that can take an abstract or any text as input to identify lists of ‘experts’, target journals and publication trends. The server requires a text specimen that can be input via copy/paste, or by uploading a text-only file. Additionally an email input option is available to allow users to receive a URL pointing to the results. Results are stored for at least 1 month. The analysis is currently performed on a 20 CPU Linux cluster. The eTBLAST webserver has been up since 2003 and typical searches (of abstracts containing 100–200 words) against Medline, which currently has >16 million records, usually takes from 1 to 3 min and is roughly proportional to the query length. Although Medline is expanding by about 500 000 records per year, eTBLAST performance is continuously being improved through code optimizations and expansion of the number of CPUs in the cluster. There is also a backup 20 CPU Linux cluster which mirrors the primary cluster to guarantee high availability. eTBLAST [see () for a detailed description of methods and performance statistics] returns a list of PubMed IDs (PMID) ordered by statistical similarity to the input text. Briefly, using a two-step process, eTBLAST computes a quantitative score. In step one a weighted keyword set extracted from the query is used to quickly search a database of indexed keywords in Medline, gathering the top 400 most similar records. In step two, a novel sentence alignment algorithm is used to refine the rank order of those similar records and compute a -score. Each of the utilities presented herein performs a similar set of tasks on these results: (i) results are parsed to extract relevant articles (with similarity z-score > 3), (ii) authors or journals which are overrepresented are calculated and (iii) the results are returned to the user (A). On January 17, 2007 at 17:40 the abstract from () was submitted to eTBLAST via the web browser at . Results were returned after 120 s. The query text contained 149 words, of which 58 were ‘stop words’. A collage of some of the output web pages is presented in , discussed above, to illustrate the output user interface. The primary methods in which users interact with the results of Medline searches can be improved and expanded to enable quick and efficient suggestions for optimizing the manuscript writing and publication process, including review. Quantitative similarity scores computed for a text query, such as the abstract for a manuscript submitted to a journal for publication, against the primary biomedical bibliographic database, Medline, can be used to generate a ranked list of similar documents from which summary information about the authors, journals, similar work and dates can be of high utility. Scientists submitting to or editors of the more than 5000 journals represented in Medline can use this free web-based utility to speed the process of selecting or confirming appropriate journal selection, estimate a given articles novelty based on the relative similarity of its abstract and to select potential reviewers (experts), typically requested by journals at manuscript submission time. Several caveats and potential enhancements to the system should be noted. First, as with any search system, similar articles sharing keywords but belonging to different fields may appear as relevant. Secondly, journal targets or experts are calculated based on frequencies of journals and authors in the eTBLAST results; these suggestions do not account for the publication volume of each journal. Finally, journal impact factors may be indicators of expertise level and are not considered. These enhancements may improve performance and are being evaluated as potential upgrades to the system. . None declared.

The ever increasing number of fully sequenced genomes results in a rapidly expanding knowledge of the metabolic capabilities of a wide variety of organisms. This information is collected and made accessible through biochemical databases such as KEGG (), Brenda () or BioCyc (). Whereas our knowledge on the wiring of the metabolic networks is far advanced, precise data on the kinetic properties of the catalyzing enzymes is still sparse. However, even without the ability to formulate kinetic models, the topology alone can be used for a wide variety of structural analyses, from which informative properties such as principle biosynthetic capabilities or feasible flux distributions can be derived. Established structural approaches include the concept of elementary flux modes (,), the closely related concept of extreme fluxes (), flux balance analysis () as well as graph theoretical approaches (,). We have recently introduced the concept of network expansion for the structural analysis of large-scale metabolic networks (). The algorithm allows to calculate for a given network and predefined external resources (the seed compounds) those chemical compounds which the network is in principle able to produce. This set of products is called the scope of the seed. Because scopes characterize the synthesizing capacities of metabolic networks, this concept is well suited for relating structural to functional properties of the networks. With this method we explored the hierarchical structuring of metabolic networks, where we focused on a complete network comprising enzymatic reactions originating from a wide variety of organisms (). We also compared metabolic capabilities of organism-specific networks () and developed a model of metabolic evolution (). Furthermore, we analysed the changes of metabolic capacities in response to environmental perturbations (). These results demonstrate the general usefulness and wide applicability of the concept of network expansion. With the development of the here-described web server application ‘MetaPath Online’, we provide a public access to this method enabling scientists to investigate specific metabolic hypothesis on particular networks. Such hypotheses include the question whether certain metabolites can be produced by a particular organism and, if so, what may be a possible synthesis route. Moreover, with the inclusion of user specified sets, it is possible to analyse the metabolic performance of mutants in which one or several reactions are removed or added. The size of a scope strongly depends on the specific choice of seed compounds as well as on the investigated metabolic network. Besides the scope, which is the final result of the expansion algorithm, the analysis of the expansion process itself is of interest. The expansion curve shows in many cases characteristic features which depend on structural properties of the network (). In metabolic networks, many reactions require the presence of particular metabolites, so-called cofactors, which typically participate in a large number of reactions and are responsible for specific functions. ATP acts as a cofactor by transferring one phosphate group to glucose. This pattern occurs in a large number of reactions and can be considered as a main function of ATP. Due to its importance, a cell in a typical physiological state ensures that ATP is always present in a sufficient amount. When asking for products which can be synthesized from glucose in such a cell, it is therefore reasonable to assume that a reaction such as () can proceed. In a similar way, we implemented other variants accounting for the presence of the important cofactors NADH/NAD and NADPH/NADP which mediate redox reactions, as well as CoA which is a carrier of acyl groups. We have developed a web-based tool which provides public access to the algorithms described earlier. The user may select for his or her analysis a wide range of metabolic networks, including organism-specific metabolic networks from a list of over 400 species as well as a reference network comprising over 5000 reported reactions or define a customized network by uploading a list of reactions. The user may choose to include information on the reversibility of reactions or to consider all reactions as reversible. The latter choice may be reasonable because for some reactions the information on reversibility is arguable, in particular since the direction of a reaction may depend on environmental conditions or the specific cell type. In its current state, the web server offers three types of analyses. It is possible to calculate the scopes of arbitrary sets of seed compounds, visualize the course of the expansion process and identify shortest synthesis pathways between chemical compounds. These three applications are described in more detail later in this chapter. The web server allows to analyse predefined metabolic networks which have been extracted from the KEGG database. In future, we plan to integrate other sources of biochemical information, by extracting networks from other databases such as BioCyc. The extension to a wider data source will increase the reliability of the results. However, metabolic databases are generally error-prone, and therefore curation of input data is an important prerequisite. In particular, the expansion process critically depends on correct stoichiometries because otherwise results would indicate that chemical species can be produced from nothing. Therefore, we applied very strict criteria for the acceptance of a reaction. On the other hand, due to this strictness, the algorithm might miss target metabolites which are in fact producible. To resolve this conflict, we plan to refine our curation procedure to allow more reactions, for example the inclusion of metabolites with chains of chemical groups of arbitrary length. The possibility to analyse user-defined networks by uploading a list of KEGG reactions considerably extends the applicability of the tool because it lifts the restriction to predefined networks. In order to further widen the usability of our application, we intend to establish a function allowing the user to upload his or her own networks in a portable format, which may be the result of his or her own research activities. This makes the user independent from the reactions specified in the KEGG database allowing for the consideration of a wider spectrum of reactions. In this work we presented ‘MetaPath Online’, a web server application for the structural and functional analysis of large-scale metabolic networks. It is based on the method of network expansion and can be used to calculate synthesizing capacities of over 400 species-specific networks as well as the reference network comprising all KEGG reactions. With the described implementation of our web server, we provide for a wide audience an easy to use interface for the network expansion method. MetaPath Online is freely available for use at . p p l e m e n t a r y D a t a a r e a v a i l a b l e t h r o u g h t h e w e b s i t e .

It is well known that RNA secondary structure prediction via energy minimization is limited in accuracy due to efficiency constraints imposed on the models and missing or imprecise thermodynamic data (). Nonetheless, simulations of RNA folding pathways () can provide interesting insights into RNA folding dynamics and structural rearrangements. Since complete RNA secondary structure landscapes have become tractable (), the limitations of predictions may be overcome by sampling and browsing candidate structures sorted by free energy or other properties. To facilitate this new approach of interrogation of an RNA secondary structure landscape a dedicated visualization tool was developed. RNA Movies () uses as input a script—essentially a list of secondary structures—that it then draws at an adjustable rate using the NAVIEW RNA secondary structure layout algorithm (), while smoothing the transitions between the nucleotide positions in subsequent structures linearly. The main aim was to create an intuitively usable web application. Therefore, the user interface matches the layout of a typical media player with play, pause, forward, backward and rewind buttons below the display area. Structures in the central viewing area can be scaled, moved and rotated by mouse controls. Multiple configuration options, such as animation speed and nucleotide colouring options are provided in a menu system. To provide dynamic extensibility of the application the menu paths are determined by an XML framework at the program start-up. This enables the adaptation of the menu structure and its associated actions to match the requirements of a given application domain. Currently, RNA Movies supports three input formats. The RNM-format makes use of the Vienna (dot-bracket) format for secondary structures. Pseudoknots may be modelled using additional brace characters {}, [], and <>. Furthermore, RNA Movies now supports the DCSE format () and the RNAStructML format (). The DCSE format has been extended to include pseudoknots and entangled helices. RNAStructML is an XML-based and XML Schema language defined format for all kinds of RNA secondary structure information including pseudoknots and entangled helices. The well-known NAVIEW algorithm () was chosen because of its many favourable properties. The layout is nearly free of overlaps, consistent and intuitively interpretable. Consistency is a very important property with regard to animation, as common parts of consecutive structures remain fixed while the rest of the structure moves about changing its shape. Therefore, a linear interpolation of coordinates associated with every base suffices as transition between different structures and works very well in practice. Moreover, co-transcriptional folding can be visualized by consecutively longer secondary structures. A great benefit of RNA Movies is the easy integration of the program into application workflows on other web servers. shows the integration of the RNA Movies applet in a typical web browser/web server environment. For simplicity the source code was reduced to a minimum, using Perl as server side language and HTML and Java as client side languages. Note the use of JavaScript in the HTML page to set input parameters of the Java applet. RNA Movies is available online as applet. Moreover, it can also be run as fully fledged application, either locally or from a web page using the Java WebStart technology. Currently, two tools on the Bielefeld Bioinformatics Server use the applet as visualization front-end. shows a screen shot of the third RNA secondary structure of the calculated by RNAshapes () using the shape folding mode with the most abstract shape class. PknotsRG is a fast algorithm for pseudoknot computation (). Since the online version supports suboptimal folding of RNA secondary structures with pseudoknots, RNA Movies is the preferred visualization front-end. shows a pknotsRG folding of the . i n t e n d t o a u g m e n t R N A M o v i e s w i t h n e w l a y o u t a l g o r i t h m s f o r b e t t e r v i s u a l i z a t i o n o f p s e u d o k n o t s t r u c t u r e s , a s c u r r e n t l y a v a i l a b l e a l g o r i t h m s d o n o t f u l f i l t h e p r o p e r t y o f l a y o u t c o n s i s t e n c y t o a d e g r e e n e e d e d f o r t h i s a p p l i c a t i o n d o m a i n . C o m m a n d l i n e f u n c t i o n a l i t y i s d e s i r a b l e f o r b a t c h p r o c e s s i n g a n d w e b s e r v i c e s a p p l i c a t i o n s t o a l l o w i n t e g r a t i o n i n t o e x i s t i n g w o r k f l o w e n g i n e s . F u r t h e r m o r e , t h e o u t p u t o f m o v i e f i l e f o r m a t s w o u l d e n a b l e e n d u s e r s t o e a s i l y p u b l i s h f i l e s g e n e r a t e d w i t h R N A M o v i e s .

The availability of a structural model of a protein is one of the keys for understanding biological processes at a molecular level. The recent advances in experimental technology have led to the emergence of large-scale structure determination pipelines aimed at the rapid characterization of protein structures. The resulting amount of experimental structural information is enormous. The application of computational methods for the prediction of unknown structures adds another plethora of structural models. The latest NAR web server issue, e.g. lists about 50 tools in the category ‘3D Structure Prediction’. The assessment of the accuracy and reliability of experimental and theoretical models of protein structures is a necessary task that needs to be addressed regularly and in particular, it is essential for maintaining integrity, consistency and reliability of public structure repositories. ProSA is a tool widely used to check 3D models of protein structures for potential errors. Its range of application includes error recognition in experimentally determined structures (4–6), theoretical models (7–10) and protein engineering. Here we present a web-based version of ProSA, ProSA-web, that encompasses the basic functionality of stand-alone ProSA and extends it with new features that facilitate interpretation of the results obtained. The overall quality score calculated by ProSA for a specific input structure is displayed in a plot that shows the scores of all experimentally determined protein chains currently available in the Protein Data Bank (PDB). This feature relates the score of a specific model to the scores computed from all experimental structures deposited in PDB. Problematic parts of a model are identified by a plot of local quality scores and the same scores are mapped on a display of the 3D structure using color codes. A particular intention of the ProSA-web application is to encourage structure depositors to validate their structures before they are submitted to PDB and to use the tool in early stages of structure determination and refinement. The service requires only C atoms so that low-resolution structures and approximate models obtained early in the structure determination process can be evaluated and compared against high-resolution structures. The ProSA-web service returns results instantaneously, i.e. the response time is in the order of seconds, even for large molecules. ProSA-web requires the atomic coordinates of the model to be evaluated. Users can supply coordinates either by uploading a file in PDB format or by entering the four-letter code of a protein structure available from PDB. A chain identifier and an NMR model number may be used to specify a particular model. A list with possible values of these parameters is presented to the user if the entered chain identifier or model number is invalid. If no chain identifier or model number is supplied by the user, the first chain of the first model found in the PDB file is used for analysis. The computational engine used for the calculation of scores and plots is standard ProSA which uses knowledge-based potentials of mean force to evaluate model accuracy. All calculations are carried out with C potentials, hence ProSA-web can also be applied to low-resolution structures or other cases where the C trace is available only (a set of C potentials is included in the stand-alone version of ProSA, see Supplementary Data 1). After parsing the coordinates, the energy of the structure is evaluated using a distance-based pair potential and a potential that captures the solvent exposure of protein residues. From these energies, two characteristics of the input structure are derived and displayed on the web page: its -score and a plot of its residue energies. The -score indicates overall model quality and measures the deviation of the total energy of the structure with respect to an energy distribution derived from random conformations. -scores outside a range characteristic for native proteins indicate erroneous structures. In order to facilitate interpretation of the -score of the specified protein, its particular value is displayed in a plot that contains the -scores of all experimentally determined protein chains in current PDB (an example is shown in A). Groups of structures from different sources (X-ray, NMR) are distinguished by different colors. This plot can be used to check whether the -score of the protein in question is within the range of scores typically found for proteins of similar size belonging to one of these groups. The energy plot shows the local model quality by plotting energies as a function of amino acid sequence position (see B and D for example). In general, positive values correspond to problematic or erroneous parts of a model. A plot of single residue energies usually contains large fluctuations and is of limited value for model evaluation. Hence the plot is smoothed by calculating the average energy over each 40-residue fragment , which is then assigned to the ‘central’ residue of the fragment at position  + 19. In order to further narrow down those regions in the model that contribute to a bad overall score, ProSA-web visualizes the 3D structure of the protein using the molecule viewer Jmol (). Residues with unusually high energies stand out by color from the rest of the structure (C and E). The interactive facilities provided by Jmol, like distance measurements, etc. are available for exploring these regions in more detail. In what follows, we provide a typical example for the application of ProSA-web in the validation of protein structures. We analyze two structures determined by X-ray analysis and deposited in PDB. The first is the structure of MsbA from , a homolog of the multi-drug resistance ATP-binding cassette (ABC) transporters (PDB code 1JSQ, release date 12 September 2001) determined to a resolution of 4.5 Å. The structure consists of an N-terminal transmembrane domain and a soluble nucleotide-binding domain. Doubts regarding the quality of 1JSQ were raised after the X-ray structure of a close homolog became available which turned out to be surprisingly different. This second structure, multi-drug ABC transporter Sav1866 from (PDB code 2HYD, release date 5 September 2006) was determined to a resolution of 3.0 Å. Based on the newly determined structure, it was realized that the published structure of the MsbA model is incorrect and as a consequence the related publication had to be retracted. A–C shows the results of ProSA-web obtained for 1JSQ (chain A). The -score of this model is −0.60, a value far too high for a typical native structure. This can clearly be seen when the score is compared to the scores of other experimentally determined protein structures of the size of 1JSQ (A). Furthermore, large parts of the energy plot show highly positive energy values, especially the N-terminal half of the sequence which contains part of the membrane spanning domain (B). In the C trace of the model, residues with high energies are shown in grades of red (C), and it is evident from these figures that the N-terminal transmembrane domain as well as the C-terminal globular domain contain regions of offending energies. A also shows the location of the -score for 2HYD (chain A). The value, −8.29, is in the range of native conformations. Overall the residue energies are largely negative with the exception of some peaks in the N-terminal part (D). These peaks are supposed to correspond to membrane spanning regions of the protein. In the C trace, these regions show up as clusters of residues colored in red (E, lower left). The C-terminal domain shows a high number of residues colored in blue and an energy distribution that is entirely below the zero base line, consistent with the parameters of a typical protein (D and E). The protein structure community is, to some extent, aware of the fact that the RCSB protein data base contains erroneous structures. But it is quite difficult to spot these errors. Grossly misfolded structures are sometimes revealed after the results of subsequent independent structure determinations become available. Errors in regular PDB files generally remain unknown to the structural community until the corresponding revisions are made available. Hence, diagnostic tools that reveal unusual structures and problematic parts of a structure in a manner that is independent of the experimental data and the specific method employed are essential in many areas of protein structure research. ProSA is a diagnostic tool that is based on the statistical analysis of all available protein structures. The potentials of mean force compiled from the data base provide a statistical average over the known structures. Structures of soluble globular proteins whose -scores deviate strongly from the data base average are unusual and frequently such structures turn out to be erroneous. For proteins containing membrane spanning regions, the significance of deviations from the average over the data base is less clear. Here, we provide an example of a published structure (1JSQ) that is known to be incorrect as is revealed by subsequent independent X-ray analysis of a related protein yielding a completely different conformation. The ProSA-web result obtained for 1JSQ shows extreme deviations when compared to all the structures in PDB (A). In contrast, the score obtained for the related 2HYD structure is close to the data base average. The result demonstrates that also for membrane proteins large deviations from normality may indicate an erroneous structure. #text

Interest in the topological properties of biological systems was greatly accelerated with the discovery of knots in single-stranded DNA in 1976 (). Subsequently, knots in DNA were investigated extensively () and even created artificially in polymeric materials (), but it took another 20 years before the first systematic studies of protein knots appeared (). Topology is particularly relevant for proteins because the 3D structure of a protein directly determines its functionality. Recently, we performed a comprehensive analysis of the Protein Data Bank () and demonstrated that knotted structures tend to persist across species and kingdoms. However, when a knot appears or vanishes in the course of evolution, the function of the protein is also altered accordingly (). We uncovered some knotted proteins that have significant biomedical importance, such as the Parkinson's disease-associated ubiquitin hydrolase UCH-L1 () or its structural homolog UCH-L3 (,), which contain the most complicated knots found in proteins so far. Other challenges include understanding the folding and unfolding of knotted proteins. The underlying mechanisms are not yet well understood and are the subject of active research (,). Surprisingly, most discovered knots were not reported at the time the structure was solved, since finding knots in protein structures by naked eye is virtually impossible. Moreover, widely used protein structure verification tools like WHATIF (), VERIFY3D () and PROCHECK () do not have the capability to detect knots. We hope that with our contribution, the discovery of knots in newly solved protein structures becomes part of the standard routine, similar to identification of secondary structure elements or classification of protein's architecture. To address this challenge, we developed a web server that allows a user to check a new or a known protein structure for knots by entering its PDB id or uploading a coordinate file. Mathematically, knots are only well defined in closed (circular) loops (). However, both the N- and C-termini of open proteins are typically located close to the surface of the protein and can be connected unambiguously: We reduce the protein to its backbone and draw two lines outward starting at the termini in the direction of the connection line between the center of mass of the backbone and the respective ends. The two lines are joined by a big loop, and the structure is topologically classified by the computation of its Alexander polynomial (,). To determine an estimate for the size of the knotted core, we successively delete amino acids from the N-terminus until the protein becomes unknotted (). The procedure is repeated at the C-terminus starting with the last N-terminal deletion structure that contained the original knot. For each deletion, the outward-pointing line through the new termini is parallel to the respective lines computed for the full structure. Unfortunately, the size of a knot is not always precisely determined by this procedure, so reported sizes should only be regarded as approximate. To speed up calculations, the KMT reduction scheme is used (,,,). This algorithm successively deletes amino acids that are not essential to the topological structure of the protein. It is also employed to create a reduced representation of the knot (). In the course of our investigations () we came up with a number of stringent criteria that a structure should satisfy to be classified as knotted: presents a typical output of the server—the summary page reporting a knot. If a knot is found, the server reports the type of the knot (e.g. 3- the trefoil knot, 4- the ‘figure eight’ knot, 5, etc.), its location in the protein structure, and a simplified representation of the knot (A). At this point, a user may choose to download results of the calculation as a collection of Rasmol/Jmol scripts or to proceed to the second page that has Jmol visualization of the knot on our server. The second page (B) has a two-window GUI to examine, rotate and further analyze the structure of the knot. The left window visualizes the protein structure with the knot using a Jmol Java applet. The knotted part is colored in rainbow colors to facilitate following the chain and visualizing the knot. The right window presents a simplified representation of the knot obtained by the reduction algorithm, making it easier to see that the protein structure is indeed knotted. The structures in both windows can be rotated, magnified and further analyzed using the tools of Jmol applet. Two buttons below (i) hide or show the rest of the protein structure in the left window, thus allowing a user to focus on the knot or to examine it in the context of the structure; and (ii) to spin structures in both windows simultaneously. An expert user familiar with Rasmol/Jmol commands can further analyze the structure using the command-line interface by entering individual commands or a whole script into a field below the windows. The bacterial tRNA(mG37)methyltransferase (TrmD) is an enzyme that transfers methyl group from -adenosyl--methionine (AdoMet) to a G nucleotide in the anti-codon region of certain bacterial tRNA species. The methylation of anti-codon nucleotides is essential for reducing the error rate in anti-codon binding to the complementary codon on mRNA during translation. The crystal structure of the enzyme from has recently been solved and is known to have a trefoil knot in the AdoMet-binding pocket. The specific configuration of the pocket allows AdoMet to adopt an unusual strongly bent conformation with its methyl group protruding from the pocket and accessible for transfer reaction. The Knots output for a PDB entry 1uam, TrmD protein, is shown in . A trefoil knot has been correctly identified for residues 86–130A in the 1uam structure. Clicking on ‘Jmol visualization’ link leads to the second page showing a protein ribbon diagram (B, left), and the simplified representation of the knot. The knot can be easily seen in the protein structure by eye if the surrounding structure is hidden from view using the button provided. The reduced representation of the knot (B, right panel) is generated by the KMT reduction algorithm. The first and the last segments in this representation are not part of the protein but represent the connection lines to ‘infinity’, which are required to circularize the structure and calculate the Alexander polynomial (see the Section ‘How knots are determined’). More examples of knot in protein structure and their analysis can be found in our recent publication (). lists all novel protein knots that were discovered with our software in 2006. A complete list with all knotted proteins is available online. t h i s a r t i c l e , w e p r e s e n t e d o u r k n o t d e t e c t i o n s e r v e r a n d a n i l l u s t r a t i o n o f i t s u s e . T h e s e r v e r i s e a s y t o u s e , a c c u r a t e a n d f a s t . I n f u t u r e , w e p l a n t o a d d a u t o m a t i c m o d e l i n g o f u n r e s o l v e d p a r t s i n t h e s t r u c t u r e s b y u s i n g h o m o l o g y .

Expressed sequence tags (EST) represent short, unedited, randomly selected single-pass sequence reads derived from cDNA libraries, providing a low-cost alternative (also called ‘poor’ man's genome) to whole genome sequencing (,) and specifically relevant to the transcriptome of an organism at various stages of development or under different experimental conditions. The analysis of EST data can enable gene discovery, complement genome annotation, aid gene structure identification, establish the viability of alternative transcripts, guide single nucleotide polymorphism (SNP) characterization and facilitate proteomic exploration (). ESTs are highly error prone and require several computational methods for pre-processing, clustering, assembly and annotation to yield biological information. Furthermore, it is extremely important to be able to store, organize and annotate ESTs using a comprehensive analysis pipeline due to their ‘high-throughput’ nature. We recently compared () available web resources (), individual tools and pipelines pertaining to EST analysis. We also evaluated currently available methods for each step of analysis, including EST clustering, assembly, consensus generation and tools for DNA and protein annotation, employing benchmark EST datasets. A detailed investigation of different EST analysis platforms () revealed that they all terminate prior to functional annotations, such as gene ontologies, motif/pattern analysis and pathway mapping. Some platforms terminate at the assembly level, providing contigs and singletons as an output (). Other platforms solely run nucleotide-based programs with limited annotation at the protein level (,,,). Therefore, we developed ESTExplorer, a complete EST analysis suite which employs programs for both nucleotide- and protein-based annotation. Moreover, we have carefully selected the most appropriate combination of programs for each stage of EST analysis, based on their ability to accurately reproduce partial gene sequences from ESTs and annotate them as correctly as possible (). ESTExplorer comprises a suite of programs with a customizable web interface to manage and analyse EST data. Optionally, EST assembly datasets generated elsewhere, e.g. EGAssembler (), can be further functionally annotated at the ESTExplorer website. Users have the option of selecting specific analysis phases (detailed below). Besides pre-processing and assembly from EST sequences, ESTExplorer annotates input sequences extensively, using gene ontologies (GO), domain analysis and pathway mapping. ESTExplorer has been used extensively for the analysis and annotation of large EST datasets from parasitic nematodes generated in our laboratories, and to identify key nematode molecules as potential targets for anti-parasite intervention. ESTExplorer has been also used for the analysis of differential transcription between adult male and female by oligonucleotide microarrays (unpublished data). The ESTExplorer workflow can be divided into three phases (shown in Supplementary Figure 1). Phase I is dedicated to EST sequence pre-processing and assembly, Phase II carries out DNA- level annotation and Phase III provides for protein-level annotation. ESTExplorer can accept nucleotide sequence input of two types (A; arrows in Supplementary Figure 1). ESTs in FASTA format can be submitted to Phase I for EST pre-processing and assembly, followed by analyses in Phases II and III. Alternatively, ESTs assembled using another program or pipeline into contigs and singletons, may be submitted directly for functional annotation (Phases II and III). Phase I comprises three programs run sequentially, to convert input EST sequences into high quality ESTs. SeqClean accepts ESTs in FASTA format and performs vector removal (using NCBI's UniVec database), PolyA removal, trimming of low quality segments at the 5′ and 3′ ends and cleaning of low complexity regions (using the DUST module). Additionally, all short ESTs (<100 bp) are eliminated as uninformative. The output from SeqClean is processed by RepeatMasker () to mask repeats. Species-specific repeat masking is done using Repeat Masker which in turn employs Cross_Match and up-to-date repeat libraries for different species from RepBase. For a novel species, the nearest organism listed in ESTExplorer, using NCBI Taxonomy, may be selected. CAP3 () then accepts repeat-masked high quality EST sequences and performs clustering and assembly into contigs (containing multiple ESTs) and singletons, based on an overlap percent identity threshold cutoff of 80. The user can modify this, with the recommendation to provide a value >65. Output files from each program are provided. Phase II carries out annotation at the nucleotide level, of assembled EST contigs and singletons from Phase I or directly uploaded by the user, using the BLASTX () program and NCBI's non-redundant protein database, followed by the assignment of functionality Gene Ontologies () using BLAST2GO (). BLAST2GO extracts GO terms for each BLAST hit obtained by mapping to extant annotation associations, using a default cutoff of E-03, which the user can modify. Additionally, BLAST2GO provides a data file which can be used to reconstruct GO relationships and perform statistical analysis on gene function information. ESTExplorer, in turn, retrieves gene ontologies from BLAST2GO and links each GO identifier to its ontology tree, displayed by the AmiGO Browser. Protein-based annotation is effected in Phase III. At the outset, ESTScan () accepts contigs and singletons from CAP3 and provides conceptual translations, using the genetic code from the nearest organism, in a two-step process. In the first step, coding regions or open reading frames (ORFs) are detected and extracted, while correcting for frame shift errors. In the second step, these ORFs are translated into putative peptides. ESTExplorer currently implements the genetic codes ( files generated from mRNA sequences) for the ten organisms: human, mouse, rat, rice, zebrafish, chicken, fly, dog, thale cress ( and roundworm () provided by the authors of ESTScan. For a novel species, the nearest organism listed in ESTExplorer, using NCBI Taxonomy, may be selected. The peptide sequences from ESTScan are simultaneously passed on to InterProScan () and KOBAS () for processing. InterProScan matches protein sequences against InterPro, an integrated resource for protein families, domains and functional sites from member databases such as PROSITE, PRINTS, Pfam, ProDom and SMART. ESTExplorer runs InterProScan in the backend and provides an html output that users can download and analyse, with details of domain/motif architecture for each sequence. KOBAS (KEGG orthology-based annotation system) maps protein sequences to pathways based on KEGG (). KOBAS uses controlled vocabularies (KO) to annotate a set of sequences and assigns pathways to individual proteins, using a two-step process. In the first step, it takes a set of sequences and assigns KEGG orthology terms based on a BLASTP similarity search against KEGG GENES or direct cross-sequence identifier mapping. In the second step, KO is used for respective pathway identification. ESTExplorer provides an html output for the mapped pathways through which the user can directly access the pathways at the KEGG website. Proteins that are mapped from the processed EST dataset are highlighted and coloured differently for easy identification. Once an EST or contig dataset has been submitted to ESTExplorer, a status page is accessible (B), for monitoring the progress of the analysis, at the program level. As each selected program is completed, the status page is updated and the output from that program becomes available immediately. ESTExplorer provides an integrated workflow approach to EST analysis, by combining assembly with traditional and well-established resources, such as BLAST2GO and InterPro. While some components are available separately as web servers, ESTExplorer has extended functionality over these as well as added additional features, interfaced seamlessly together. Phase I of ESTExplorer roughly maps to the functionality of EGAssembler. However, there is no functional annotation after assembly into contigs from EGAssembler. Additionally, we have also provided the ability to use quality values during the assembly process. Phase II involves DNA-level orthologue mapping, directly from the Phase I output. When there are several contigs and singletons after Phase I, the user does not have to submit each one to NCBI to run BLAST. Additionally, we can process each of the contigs and singletons from the Phase I for protein-level annotation, Phase III where the complete InterproScan, GO mapping and KEGG pathway mapping are carried out. Recently, Pavy and coworkers () have used GO and Pfam matches for annotating their ESTs at a functional level. ESTExplorer provides these along with the additional advantage of KEGG and the complete InterProScan currently comprising 12 modules in addition to Pfam, for protein and domain analysis (details available from our website). The outcome for each run is summarized, with links to output files from each selected program. An email with the URL of the results will be sent to the user after the completion of the entire run. Users can either download output files from the download page for each step or as a single zipped file for each phase of the analysis (C). The results are stored for one week, after the completion of the run. Some programs are run by default, whereas others are optional. In Phase I, SeqClean and CAP3 are run by default while RepeatMasker is optional. All of the programs in Phase II and III, excepting ESTScan, are optional. We update the backend databases (non-redundant protein and UniVec databases from NCBI, Repeat Database from RepBase, Gene Ontologies, InterProScan and KEGG) every month using automated scripts. A detailed tutorial and FAQ () are available for running sample EST datasets and understanding the different analysis programs. It is usually difficult to collate the analysis results at the final output stage when a large dataset is analysed using a workflow containing several phases and multiple programs. To address this issue, ESTExplorer tracks each assembled sequence (contig/singleton) which has been functionally annotated (more details are available from the example section). p l o r e r h a s b e e n d e v e l o p e d u s i n g o p e n s o u r c e t e c h n o l o g i e s ; Z o p e ( V 2 . 8 . 1 ) , P y t h o n ( V 2 . 4 . 3 ) a n d M y S Q L ( V 4 . 1 . 1 0 a ) , f o r E S T d a t a m a n a g e m e n t a n d a n a l y s i s . E S T E x p l o r e r r u n s o n a 1 6 - n o d e L i n u x c l u s t e r ( 1 . 3   G H z , I t a n i u m 2 R e v , 5 P r o c e s s o r s , 1 6   G B R A M ) r u n n i n g o n R e d H a t E n t e r p r i s e L i n u x A S R e l e a s e 3 . T h e w o r k f l o w a r c h i t e c t u r e h a s b e e n d e s i g n e d b a s e d o n a ‘ d i s t r i b u t e d c o n t r o l a p p r o a c h ’ . T h e u s e r r e q u e s t f r o m t h e c e n t r a l Z O P E c o n t r o l l e r i s d i v e r t e d t o o n e o f t h e d a t a - p r o c e s s i n g m a c h i n e s a f t e r a p p r o p r i a t e l o a d b a l a n c i n g . B r o w s e r a n d p l a t f o r m i n d e p e n d e n t j a v a s c r i p t s h a v e b e e n u s e d f o r d a t a v a l i d a t i o n , i n o r d e r t o e n h a n c e t h e f l e x i b i l i t y o f q u e r y a n d o u t p u t p a g e s . T h e s e r v e r r e f r e s h e s t h e i n t e r m e d i a t e r e s u l t p a g e e v e r y 3 0 s a n d u p d a t e s t h e u s e r w i t h t h e s t a t u s o f p r o c e s s i n g i n t h e i n d i v i d u a l p r o g r a m s i n t h e p i p e l i n e . A f i n a l o u t p u t p a g e p r o v i d e s t h e u s e r w i t h d e t a i l e d o u t p u t f i l e s f o r v i e w i n g a n d f o r d o w n l o a d i n g t h e r e s u l t s . O u t p u t f i l e s a r e s t o r e d o n t h e s e r v e r f o r s e v e n d a y s . From dbEST (), we provide a small dataset of 372 ESTs (Input Option 1 in Supplementary Figure 1) for the plant and the complete analysis results from ESTExplorer. Additionally, assembled sequences (contigs/singletons) from these ESTs have been provided as an example for Input Option 2 (Supplementary Figure 1). Detailed sequence-wise annotation summaries are provided to facilitate rapid functional analysis of EST datasets (). The detailed summary of the analysis of contig 9 shows the contributing ESTs, protein domains, gene ontologies and mapped pathway (shown in Supplementary Figure 2). One of our research projects involves gene discovery from parasitic nematodes. ESTExplorer has allowed the rapid and accurate analysis of ESTs by providing robust annotation at the gene and protein levels, matching evidence from multiple sources. Using ESTExplorer to analyse 873 ESTs from a parasitic nematode () yielded 133 contigs and 314 singletons, compared with 128 contigs and 388 singletons reported by Cottee . (). Overall, 29 entries were annotated with gene ontology data, 44 sequences had protein domain information and 246 sequences were mapped to KEGG pathways. This rapid and comprehensive analysis together with additional analyses of specific molecules enabled the identification of novel genes and molecules predicted, based on comparisons with extensive data in WormBase (), to be involved in biological pathways critical for development, reproduction and survival. With ESTExplorer, the analysis was systematic and additional information on domain and pathway mapping made it easier to validate functional annotation with low scoring hits. This dataset is provided as the second example dataset (Input Data 1) on the server (). A moderate dataset of 10 651 ESTs for , downloaded from dbEST (), is also available, as ESTs in FASTA format (Input Option 1) and assembled ESTs (Input Option 2). Additionally, we have also applied ESTExplorer for the analysis of a number of EST datasets ranging from 717 ESTs from a related parasitic nematode () to 21 967 ESTs from for subsequent analysis of differential transcription between adult male and female worms by oligonucleotide microarrays. We used two types of data that were annotated using ESTExplorer: the first comprised unprocessed 21 967 ESTs and the second contained 1885 contigs. By annotating both the ESTs as well as these contigs, we have been able to get better representation of biologically relevant genes for oligonucleotide design and subsequent microarray analysis (unpublished data). ESTExplorer has been used extensively for the annotation of transcript and protein sequence data for the and fungal genomes, a collaborative effort of our group (N.D. and S.R.) with DOE Joint Genome Institute (JGI), USA. #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Traditional transmembrane topology predictors often predict signal peptides as transmembrane segments, and vice versa signal peptide predictors often predict N-terminal transmembrane segments as signal peptides. This fact is often overlooked when testing prediction methods, and is the main cause for very different test results. A frequent advice how to circumvent the problem of these cross-predictions is to remove predicted signal peptides before predicting transmembrane proteins (), or to remove proteins with transmembrane segments when predicting signal peptides (). However, as the number of errors due to cross predictions is roughly the same for the two kinds of predictors (), the gain will be as high as the loss by such approaches. To resolve the ambiguities we have, in a previous study, designed a hidden Markov model, Phobius, containing submodels for both signal peptides and transmembrane segments (see ). We obtain better discrimination by forcing the predictor to chose between the two types of features. A benchmark () showed that false classifications of signal peptides were reduced from TMHMM's () 26 to 4% and false classifications of transmembrane helices were reduced from SignalP 2.0's () 19 to 8%. An advantage is that the method even increased the high accuracy of TMHMM in predicting pure transmembrane topologies from 44.5 to 53.9% correctly predicted topologies. Since this benchmark, a new version SignalP 3.0 () has been published. Its false positive rate on transmembrane proteins is however as high as before. On the same set of transmembrane proteins without signal peptides used in the previous benchmark, SignalP 3.0 produces false predictions on 21% (52 of 247) of the test sequences. Here, we present an overlap analysis between signal peptides predictions and transmembrane segment predictions done by conventional predictors on five proteomes. We also give a description of the Phobius web interface. table-wrap #text The Phobius web server provides an easy and accurate mean to predict signal peptides and transmembrane topology from an amino acid sequence. The sequences should be submitted in fasta format, preferably uploaded as a file. The predictions are given either in ‘short’—single line text output or ‘long’—UniProt feature table styled output (see ). All predictions made by the Phobius server can optionally be accompanied by a posterior label (location) probability plot. The posterior label probability is the probability for a location (cytoplasm, non-cytoplasm, membrane or signal peptide) of a residue given the whole sequence (see ). Note that the posterior probability plot is not a prediction in itself. The pattern of the plot might even deviate from the prediction, which would be a sign of uncertainty in the prediction. In ‘normal prediction’ mode as well as in the ‘constrained prediction’ mode described below, sequences are decoded with the 1-best algorithm (). The accuracy of the predictions can be greatly improved if we can include information about the location of a part of the sequence in a constrained prediction (). Typically we could have experimental data at hand from reporter fusions (), antibody experiments, or have knowledge of the location due to functional requirements of a site (). The Phobius web server provides a service to let the user specify such constraints for a prediction. The user may specify that a residue resides in a cytoplasmic loop, non-cytoplasmic loop or a transmembrane segment. One can also specify that the N-terminal part of the sequence is a signal peptide. Here we maximize ( | ) ( | ). This is implemented by a modification in the forward–backward () calculations; we multiply the forward probability for a state with the ( | ) in the constrained sequence positions. As the membrane, signal peptide or a cytoplasmic loop states are uniquely identified by one single label in the Phobius model (), we set ( | ) to 1 for the label corresponding to the constraint and 0 for all other labels in the constrained position. Non-cytoplasmic loops, on the other hand, can have two different labels. Here we assign 0.5 probability to each of the two constrained labels, and 0 to all other labels. Since homologous sequences are likely to share both transmembrane topology and absence or presence of signal peptides, we can gain extra support for a prediction by examining the query sequence's homologs. This is the supporting idea for PolyPhobius, whose algorithm is described in a separate paper (). Here the server BLASTs the query sequence against UniProt. Hits with an E-value lower than 1E–5 covering more than 75% of the sequence length are used as support for the prediction. The full-length sequences are then realigned using a multiple sequence alignment program, and weighted with the Henikoff and Henikoff weighting scheme (). When we measured the performance of the approach, we found a significant increase in accuracy for transmembrane topology prediction accuracy (from 67.8 to 74.7% correct topologies) and as well as improvement in signal peptide prediction accuracy (increase in Matthews correlation from 0.901 to 0.921) as compared to Phobius without homology–enrichment (). The user can also submit his own alignment in Fasta format. In this case, the transmembrane topology and presence of signal peptide of the first sequence will be predicted taking the other sequences in the alignment into account. #text ext-link #text

Analysis of gene expression profiles across various tissue types is essential for understanding gene functions. Among various available expression data sources, expressed sequence tags (ESTs) have been valuable for rapid expression profiling. Based on the premise that EST clone frequency is proportional to the corresponding gene's expression level (), we and others have developed algorithms and tools to perform expression analysis based on EST data (). Meanwhile, EST data continue to accumulate at a rapid pace, and there are a growing number of databases that organize general or species-specific EST information, including EST data for sea bass, wheat, chicken, pig and tomato (). Despite the surge of recent progress in other species, the number of public EST entries for human and mouse still far exceed those for any other species, based on the January 2007 summary from dbEST (). Since the reliability of EST-based expression analysis is dependent on the size of EST libraries, the human and mouse data remain an attractive source for expression analysis, and the tools built for analyzing these data will likely benefit expression analysis for other species. We previously developed the GEPIS server that utilizes EST abundance information to calculate gene expression levels in a panel of normal and cancerous human tissues for a given input DNA sequence (). We showed that such EST-based (or ‘digitally’ derived) expression units exhibit a linear correlation with TaqMan-determined expression levels. Since its release, the GEPIS server has provided expression results for over 30 000 requests by researchers from >60 countries. Despite its usefulness, GEPIS suffers from several limitations. The method relied on the BLAST program to assign EST sequences to a given input mRNA sequence. However, BLAST often erroneously links ESTs to input sequences due to high-percentage regional matches. As a result, a given EST could be matched to multiple genes, thus leading to miscalculated expression data. In addition, there were insufficient data for performing reliable analysis for mouse genes. The design of the system also did not allow easy development of new functionality commonly requested by users, such as URL linking to the expression results, text searching and display of detailed results. We have now developed a new web server, named GeneHub-GEPIS, which performs digital expression analysis based on an integrated database for human and mouse genes. One distinguishing characteristic of this tool is that ESTs are mapped to pre-defined gene structures along the genome. The GeneHub component of the application is designed to define gene boundaries based on mRNA transcript sequences from major databases and to establish extensive cross references for commonly used gene identifiers. Based on the precise genomic locations of ESTs, as determined by the GMAP algorithm (), we link ESTs to genes for subsequent expression analysis. The new design offers several major advantages. First, this genome-based approach increases the accuracy of EST mapping to genes, thus enhancing the overall reliability of the EST-based expression values. Second, the new gene-centric design makes the system more extensible, so that we could easily add a new collection of genes, such as microRNAs, to the system. Third, the integrated gene database accepts text-based searches and diverse input values. In addition to DNA sequences, the input values can be identifiers from common gene databases or commercial microarray platforms. Fourth, the orthologous relationships stored in the GeneHub database allow easy navigation between mouse and human genes. Finally, the program provides basic information about input genes and allows direct linking to expression results from any web site. Meanwhile, we retained the useful features from the previous GEPIS application, such as the capability to draw a regional expression atlas for a given genomic region. Here, we present GeneHub-GEPIS as a new and useful tool for performing gene expression analysis across many normal and cancer tissues for both mouse and human genes. The genomic structures of protein-coding genes were first defined using transcripts from several reliable sources. The collection of such high-quality transcripts, which we also call the core gene set, contains mRNA sequences from RefSeq, the Known gene set of Ensembl genes, Proteome and FANTOM (mouse only) (see Supplementary Table 1 for details). Each of the core gene set sequences was mapped to their respective genome (human NCBI Build 36 and mouse NCBI Build 35) using GMAP (), and only the genomic match with the highest percent identity and percent coverage was chosen. GMAP has been shown to provide very accurate mapping and alignment results for both mRNA transcripts and ESTs (), but occasionally matches with lower matching percentage can be found due to low-quality sequences. As a conservative precaution, we removed transcripts (and ESTs for a later step) with <90% coverage of the entire transcript or with <90% identity as measured by GMAP. For instance, about 0.4% RefSeq sequences were filtered out during this step. For any two transcripts to be clustered into one gene, we required that their exon sequences overlap, be in the same orientation and share at least one exact exon boundary or splice site. The requirement for shared exon boundary was used to limit the inclusion of antisense transcripts, since the orientation of these transcripts could be occasionally mis-annotated (,). Each group (or cluster) of transcripts was considered as a ‘GeneHub gene’. Using this approach, we defined 31 999 non-redundant genes for the human and 34 794 for mouse. Overall, these GeneHub genes can be considered as a superset of known genes from the above data sources. After the initial collection of human and mouse genes were obtained, additional sequences were mapped to the GeneHub genes. Transcripts from GenBank and the Ensembl Novel collection were mapped to the genomes with GMAP () and then compared with those GeneHub genes derived above. For a transcript to be linked to a GeneHub gene, at least one of the splicing junctions was required to match perfectly with those of the GeneHub gene. We next tried to assign microarray-related probe sequences (see Supplementary Table 1 for the full list) from commonly used commercial array platforms to GeneHub genes. For Affymetrix expression arrays, we used the target sequences obtained from Affymetrix to link to known genes. For Agilent oligo-based expression microarrays, we directly used the 60-mer oligo-nucleotide sequences for gene linking. Using GMAP (), we determined whether the array probe sequences overlap with the exon sequences collected above. Next, for sequences that did not overlap with any exons, we examined whether they were located in the vicinity of any GeneHub genes. We assigned a probe sequence to the closest gene in the same orientation if the probe sequence was located within 5 kb to the 3′ end or 2 kb to the 5′ end of the gene. The ortholog linkings between human and mouse GeneHub genes were based on the hmlg_ftp.txt file from HomoloGene () Release 50.1. We used the orthologous Entrez Gene pairs of human and mouse if they were established by reciprocal best match between three or more organisms, or reciprocal best match, or sequence similarity with match identity >70%. We were able to link 15 868 human GeneHub genes to their mouse counterpart. EST data were downloaded from NCBI () and processed following a previous protocol to retain only non-normalized, non-subtracted usable cancer and normal libraries (). From the 11 July 2006 release of dbEST, we collected 4 175 880 human ESTs from all usable libraries, including 1 912 573 ESTs in 1 995 normal libraries and 2 263 307 ESTs in 3 812 cancer libraries. This represents a 35% increase in total usable EST counts from 2 years ago. For mouse, we collected 1 879 315 total ESTs, including 1 461 621 from 285 normal libraries and 417 694 from 37 cancer libraries. See Supplementary Table 2 for total number of ESTs in each category and tissue type. We first mapped EST sequences to their genomic locations using GMAP (), followed by a secondary filtering step as described earlier (>90% identity and >90% coverage). To avoid ambiguity, we discarded the ESTs (1.3% of total) that were mapped to multiple genomic loci with identical percent identity and coverage. We also eliminated those ESTs (5.5% of total) with near identical matches (with <2% difference in percent identity or coverage) to multiple genomic loci. The genomic coordinates were compared with the gene structure information (intron/exon boundary) of GeneHub genes. An EST was considered to be the product of a gene if the two entities were mapped to the same locus and share at least one exon (with minimum overlap of 30 bp). Multiple EST reads from the same clone were reduced to a single read if clone information was available. The digital expression unit (DEU) for a given gene in each tissue category is defined as the number of matching EST clones from a normalized library size of 1 million. The DEU values were calculated iteratively for each tissue type to profile expression levels across all tissues. The -test was applied to determine whether DEUs in two samples were statistically different using a method described previously (). Comparisons were made between normal and cancer samples from the same tissue, and across different types of tissues. To improve the efficiency of the GeneHub-GEPIS program, the expression result for each of GeneHub genes was pre-computed for fast access. For ∼1000 randomly selected human genes, we compared how ESTs were mapped to genes by BLAST (≥98% identity over >60 bp region) or by the above method. For 74% of tested genes, BLAST alone would identify at least one incorrect EST, as judged by its genomic location. By design, the new method disallows this type of erroneous mapping. It is worth noting that the median DEU level for human genes in normal tissues is 50.0 based on our new method, as compared to a median level of 70.0 derived from BLAST-based EST-mapping approach (). The average number of ESTs mapped to a gene also reduced from 128.7 to 107.5 (or a 16% reduction). The significant reduction of ESTs mapped to genes reflects the high accuracy of EST mapping by GMAP () and the rejection of ESTs with promiscuous matches. Manual review of EST mapping results for randomly sampled genes also confirmed the much improved mapping quality. As a confirmation step, we compared our results with a set of EST-gene mapping data independently generated by the UCSC Genome Browser team, and we found a concordance of >97%. The UCSC data are based on BLAT () and a series of filtering steps, and are used for the EST alignment track in the UCSC genome browser. MicroRNA expression analysis is based on the observation that miRNA precursor sequences can be found among ESTs (,), and that pre-miRNA expression levels correlate with mature miRNA expression levels (). Given this relationship, we could use EST data to approximate miRNA expression levels in various tissues. We first collected genomic locations of the miRNA stem-loop sequences from version 9.0 of miRBase (), and then obtained all EST sequences that had any overlap with the miRNA stem-loops for expression analysis as described earlier. The Regional GEPIS Atlas, which depicts the expression level of all genes in selected tissues for a given genomic window, was created in the similar fashion as described previously () but with the exception that we stored the genomic coordinates for all genes in a MySQL database instead of in plain text files. The web front-end was written in HTML, javascript, CSS and Perl CGI. The Perl template module (HTML::Template) was used to achieve consistent look across different web pages. The Ajax technology was used to make the application more interactive so that text searches could be performed without leaving the input web page. We used the Prototype Javascript library () to implement AJAX calls. This library supports AJAX interactions and provides utility functions for accessing page components and DOM manipulations. A MySQL database was used for data storage and retrieval (Supplementary Figure 1). For text-based searching, the query string can be a record identifier (e.g. accession) or gene name. The program queries DBXREF, GENE and GENE_SYNONYMS tables in sequential order to find a best match from the selected target species for the given query, regardless of the species of input record. The text search is case-insensitive and a begin-search is automatically performed if no exact match is found. All of the source code is available upon request. fig #text xref #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

xref ext-link #text italic #text is an approach to the functional characterization of unknown proteins based on a cascade of similarity searches. It is well known that protein databases do not completely overlap and differ in their annotation quality (). This task takes into account the significant differences among databases (Supplementary Table 1) to improve the quality of the protein characterization. It selects the order in which the databases have to be searched and combines the annotation found depending on the results. Protsweep classifies proteins into the following categories: , , and proteins. The query protein starts the BLAST () cascade against Swissprot () first (). We do take into account three parameters to classify the BLAST hits: (i) percentage of identity, (ii) ‘’ and (iii) ‘’. The two last parameters are related to the length of the total alignment, being ‘’ the percentage of the query sequence length covered in the alignment with the database hit and ‘’ the percentage of the hit (subject) sequence length covered by the alignment (). Depending on the classification of the BLAST hits according to these parameters and the hit protein annotation, three different approaches will be followed. If the hit has 100% ‘’ and ‘’ and more than 98% identity, it is considered an identical protein and the Swissprot ID will be searched in Ensembl (). If it is successful, all information from both databases will be combined (Supplementary Table 2) and stored in the XML output. If the ID cannot be found in Ensembl then a BLAST search is performed with the query protein against Ensembl. The best Ensembl hit is selected and compared against the Swissprot hit using the Smith–Waterman algorithm implemented in Water (EMBOSS) (). If the identity between sequences is greater than 98%, then the information from both sources and the BLAST alignment will be added to the final output, if the identity is less, only Swissprot annotation and the alignments will be added to the XML. If the ‘’ and ‘’ is between 80% and 98% and the identity is between 85% and 98%, the hit is classified as and follows the same strategy with Ensembl as already described (). In case, the identity is between 20% and 85% and ‘’ and ‘’ are greater than 85%, then the BLAST cascade continues with SpTrembl and RefseqProt. In the case that no or hits can be found in any of the databases, the best similar hit among the three databases is selected and classified as or (). Depending on the classification, the task displays different kinds of information. If the protein is characterized, information concerning the coding gene, about the splicing variants and orthologous genes is also provided. Depending on the degree of homology, protein function, transcript of origin, genomic localization, and GO annotation or partial similarities will also be shown. Proteins annotated as ‘hypothetical’ are further analysed. Hypothetical proteins will only be presented in the result when no other information about identical or homologous proteins can be found in any of the databases (Supplementary Figure 1). The web output of ProtSweep (Supplementary Figure 1) is divided in five sections: (i) General Information, (ii) Identified Protein and Transcripts, (iii) Features and Functions, (iv) Genomic Localisation and (v) Homology to Other Organisms/Genes. The information provided in each of these sections is provided in and Supplementary Table 2. The user has immediate access to all complete application outputs and database entries via hyperlinks. At the bottom of the HTML output there is a link to the explanatory legend as well as to the XML output containing all the generated information. identifies the domain architecture within a protein sequence and therefore aids in finding correct functional assignments for uncharacterized protein sequences (). It employs different database search methods to scan a number of protein/domain family databases. Among these models, in increasing complexity, are: PRODOM (), automatically generated protein family consensus sequences, PROSITE () regular-expression patterns, BLOCKS (), ungapped position-specific scoring matrices of sequence segments, PRINTS () sequence motifs, PROSITE profiles (), gapped position-specific scoring matrices and Hidden Markov Models like PFAM (), SMART (), TIGRFAMS () and SCOP (). Each database covers a slightly different, but overlapping set of protein families/domains. Each model has its own diagnostic strengths and weaknesses and for each of these protein/domain family databases used we have established different thresholds. For example, in the case of the database PFAM-A, we compare the input sequence against the Hidden Markov model profile of each PFAM protein family. In principle, it is possible to decide the significance of a match upon its -value. However, there are a few complications such as that there is no analytical results available for accurately determining -values for gapped alignments, especially profile HMM alignments. We use as threshold the trusted cut-off value (TC) existing for each PFAM family. This value is the lowest score for sequences included in the family (e.g. in the full alignment). Therefore, we consider a hit very significant if scores better than the trusted cut-off and at the same time has a significant -value. In the case of SCOP, individual protein families are described by several HMMs. We use the SCOP filtering mechanism to look for consistency in the HMMScan output, and filtering out inconsistent hits. In the case of SMART we use only the -value. For each of the protein/domain databases used, we have established different thresholds and rules. Afterwards DomainSweep takes all true positive hits of all individual database searches for further data interpretation. Domain hits are listed as ‘significant’: All other true positive hits are listed as ‘putative’ (). It is clear that any automatically produced sequence analysis implies a reasonable compromise between sensitivity and selectivity, and that no ideal recognition threshold exists that would allow for perfect separation of true and false similarities. Our thresholds tend to be rather conservative and stringent and thus the possibility of extending false positives is very limited. The output in the web consists of two groups of graphs, those corresponding to the significant and putative hits, and one table output containing all reported protein domains (Supplementary Figure 2). The graphical outputs display for each ‘significant’ or ‘putative’ hit a cartoon of the sequence with the domain corresponding to the match, the hit ID, description, begin, end and Gene Ontology (GO) annotation. The user has immediate access to all complete application outputs and database entries (via hyperlinks) by clicking on the corresponding part of the picture. At the bottom of each graph there is a link to the task explanatory legend. The table output contains all hits, IDs, descriptions and links to the original output. The XML output containing all the generated information is available via hyperlink at the bottom of the task output. identifies the structural domains in the protein and therefore aids in finding structural elements. It reports on predictions for alpha-helix, beta-strand, coiled-coil and helix-turn-helix motifs, transmembrane regions, signal sequences, hydrophobicity, antigenicity, protease cleavage sites and more. When predicting the secondary structure of a protein, it is useful to exploit the features of several available prediction algorithms rather than to rely on a single program. Unfortunately, combining prediction methods on a large scale is complicated by the fact that prediction programs have very different input requirements and output formats. Some of them perform much better when they have a multiple sequence alignment covering different degrees of similarity as input instead of a single sequence. We have developed MSFGenerator, a program, which creates a multiple sequence alignment for a single protein sequence according to user, defined rules (Supplementary Data MSF). It performs a BLAST search against a non-redundant protein database following different strategies that will generate different kind of alignments (Supplementary Data MSF, ). The output of MSFGenerator is an alignment in MSF format (multiple sequence file). The generated MSF will be used as input for four different structure prediction programs: PsiPred (), Jnet (), Prof (), and DSC (). Each derives its prediction using a different heuristic. PsiPred is a two-stage neural network that bases its prediction on position specific scoring matrices, Jnet is a neural network method that works by utilizing an alignment as input, alongside Psiblast () and HMM profiles. Prof is a classifier that combines linear discriminations and neural networks. DSC is based on decomposing secondary structure prediction into basic concepts and then uses simple and linear statistical methods to combine them. Since DSC is known to perform worse than the other prediction methods employed in 2Dsweep, the usage of DSC is optional. As a second concept, 2DSweep searches for DSSP (Definition of Secondary Structure of the Protein, () annotation for the input protein. 2DSweep runs a Blast against the PDB database. For all local alignments found it extracts secondary structure elements (if any) from the structure definition of the DSSP database. If there is more than one element covering the same sequence region, 2DSweep uses a simple majority vote to determine the structure at each position. The result of this procedure is shown together with the prediction of the different secondary structure prediction tools. Additionally, 2DSweep shows several other common measures of secondary structure. First, the distribution of small, charged and hydrophilic amino acids are shown and probable antigenic regions are indicated. Furthermore, the task searches for transmembrane helices and intervening loop regions using four different methods: TmHmm (), DAS (), TMap () and TmPred. In eukaryotic protein sequences, it additionally searches for signal peptides. Finally, information is given about molecular weight, isoelectric point, the distribution of protease cleavage-sites, and the possible sub-cellular localization of the protein. The web output of 2DSweep (Supplementary Figure 3) is divided in five sections: (i) General Information, (ii) Secondary structure, (iii) Features and (iv) Cleavage sites. The information provided in each of these sections is shown in and Supplementary Table 3. The complete results can be viewed by clicking on the corresponding part of the picture. At the bottom of each graph there is a link to the corresponding explanatory legend. As in the other tasks the XML output containing all the generated information is available via a hyperlink at the bottom of the task output. These servers have been implemented using the W3H task framework (), which allows the execution of compound jobs using work and data flow descriptions in a heterogeneous bioinformatics environment using meta-data information. The system regulates the dataflow by specifying dependency rules between the used applications in the meta-data, which allows the design of high complexity bioinformatics tasks, and stores the results of the different applications together with the new results computed during the process. The final output of the task is an XML file which contains all relevant information generated. The XML information is transformed by means of W2H's () post-processing mechanism into an HTML page for the task report using the Extensible Style-sheet Language Transformations (XSLT; for facilitating a final visual inspection of the results. Furthermore, the XML output can be also required and used for further analysis (i.e. direct integration in user's databases, additional pipeline analysis). All public databases used by these servers are installed under the Sequence Retrieval System (SRS) at the DKFZ (). The DKFZ SRS server contains more than 500 databases that are automatically updated whenever new releases become available; this means that the webservers will be using the very last version of each database. The use of this integrated approach provides great flexibility and extensibility of the process. Therefore, as new and improved algorithms and methodologies are developed, they are incorporated into the protein analysis process without having to redesign the entire task. It is also possible to incorporate specific sets of databases as they become available, and to implement arbitrary configuration parameters. The development of the three pipelines presented here, has been user-driven from the beginning. Their functionalities are continually being updated and extended in response to requests and suggestions emerging from our core users like LIFEDB (,), where these servers are actively used in their protein analysis and annotation. We are currently developing checks especially through the application of filtering strategies and algorithms that will take into account the relationships between domain structure and homology searches. At the moment we are starting to develop a filtering system for the homology searches results taking into account the different quality of annotation in different protein databases with the idea to assign confidence levels and cross-checking results between tasks. We are additionally working on the implementation of directed text mining using the keywords of the proteins description. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

It has always been a major goal of biological research to understand how the behavior of biological systems is governed by the properties of and interactions between their parts. Today's experimental high-throughput technologies allow us to simultaneously determine the state of thousands of system components at a cellular or molecular level. More and more powerful bioinformatics methods are applied to high-throughput data in order to unravel the interaction networks underlying a system's apparent behavior. Examining those networks in aberrant systems, e.g. malignant tissues, may lead to new insights into molecular etiology and thus suggest possible targets for the development of new therapeutic drugs. A wide range of technologies is available for obtaining data on the expression level of genes. Most frequently used are methods like SAGE or microarrays, which measure transcript abundances as an estimate for gene expression height. In principle, both methods allow for two different experimental set-ups (,): (i) Comparison of expression levels between different tissues or the same tissue under different conditions (treated/untreated) or dignities (normal/malignant). In this case, all genes which are over-expressed in one sample with regard to the other one are considered to be co-regulated and are generally believed to be also functionally related (‘guilty by association’). (ii) Time series experiments typically monitor the temporal change in expression in just one tissue type after a stimulus has been applied. Genes for which changes in expression height are observed at an earlier time point are suspected to have a causal influence on the expression of genes displaying altered expression levels at a later stage. Various methods exist to derive possible gene interaction networks from such data. In either case, the detection of significant expression differences may be improved by taking into account other sources of information, e.g. on the participation of the examined genes in common biochemical pathways (). However, is has turned out not to be sufficient to rely on expression levels alone in order to generate useful hypotheses about the gene interaction networks eventually responsible for the realization of a certain phenotype. Rather, additional biological knowledge seems indispensable again for narrowing down the vast number of possible interactions that can be deduced from observed expression heights to a realistic and comprehensible amount. A resource frequently used for augmenting expression analysis methods is GeneOntology (), which may provide additional hints on whether genes found to be co-expressed or inducing each other's expression do also share a common functional context. However, even for knowledge-based methods, additional (manual) expert evaluation is still indispensable in order to clarify which of the generated hypotheses on causal relationships seem plausible and should be subject to further experimental examination. The majority of the available gene expression analysis methods takes primarily into account those genes which have been observed to be expressed differentially between the different conditions or time points; genes which are not found to be differentially expressed are considered to be of lesser interest. These approaches, however, neglect the fact that a gene doesn't necessarily have to be differentially expressed in order to exert effects specific for, e.g. a certain tissue or disease state. Rather, its gene product may just be subject to functional modulation by other, differentially expressed genes, rendering it functional under the one condition (where the modulator is over-expressed or, in case the modulator acts as an inhibitor, under-expressed) and inactive under the other (modulator under- or over-expressed, respectively). Thus by being dependent in its activity on a differentially expressed gene, such a non-differentially expressed gene acts as an effector specific for a certain expression profile. Consequently, if (i) a gene (or gene product) is known to be modulated in its activity by a gene at least under some circumstances and (ii) is found to be differentially expressed between the examined conditions and , we generate the hypothesis that 's activity will also differ between conditions and according to 's abundance and hence may be called a (putative) expression-profile-specific effector. In order to identify such additional effectors, one has to identify among non-differentially expressed candidates those genes (or gene products) which are already known to be subject to modulation by differentially expressed ones. We present a method (and provide an implementation thereof, called DEEP—Differential Expression Effector Prediction) to identify such molecules by applying already existing biological expert knowledge about biomolecular interaction networks, as provided by resources like the TRANSPATH database on signal transduction (), to user-supplied, newly generated gene expression data. DEEP thus also demonstrates how the vast amount of information on potential interactions contained in databases like TRANSPATH can be utilized by, e.g. filtering for data sets of actual relevance in a certain expression background. DEEP has initially been designed for the analysis of user-supplied SAGE data in the form of two sets (one for each condition and ) of absolute tag counts , mapped to their corresponding genes . Alternatively, DEEP can access all human SAGE libraries as provided by CGAP on their SAGE Genie server (), pooled according to CGAP's organ/tissue and dignity (normal/cancer/tumor associated) classification. If such a library set is selected, or more than two SAGE gene lists have been supplied for any of the two conditions, the corresponding list set is being merged into a ‘meta library’ by summing up the tag counts for each gene found in at least one of these lists. In any case, values for -fold over-expression (default:  = 2) are calculated using the method described in (). Eventually, all values are re-mapped to significance values , with  = 1 −  in case of over-expression in tissue/condition and  =  − 1 otherwise. Only those genes with || values exceeding a user-defined threshold are processed further. Apart from direct support for SAGE data, DEEP is, in principle, capable of processing all kinds of expression data sets that consist of a list of genes with corresponding significance values −1 ≤  ≤ 1. Such values can, for instance, easily be derived from microarray data, rendering DEEP also applicable to this widely used experimental method for assessing differential gene expression. TRANSPATH features a molecule classification schema with various levels of abstraction (); the mapping process described above, for instance, yields a list of so-called ‘basic molecules’, which represent species-specific sequence variants. For easing further processing, the identified basic molecules are further mapped to TRANSPATH ‘ortholog’ entries , which subsume all species-specific variants. All finally get their corresponding differentially expressed gene's assigned to them. The differentially expressed ortholog molecules identified in the previous step are used as starting points for reconstructing a signal transduction network, represented as a graph with molecules as nodes and signal transduction events (also taken from TRANSPATH) as directed edges. Reconstruction is performed by depth-first searching the network implicitly described by TRANSPATH to a user-defined depth, i.e. by calling the procedure depicted in the pseudocode outline () for each starting point with parameters (, 0). In a subsequent step, the reconstructed network graph gets traversed, again starting from each node which represents a differentially expressed gene (or its corresponding TRANSPATH ortholog entries, respectively) , by calling the procedure (, , 0) from the pseudocode outline for each ). During this traversal, each encountered—and not necessarily differentially expressed!—successor node ‘inherits’ the current starting node's value. More precisely, each node keeps track of a weighted average of the values of all starting nodes it has been reached from, with 1/ as weighting factor and being the number of steps that were required to reach , starting from the node representing . If an gets visited several times—be it from the same node via different paths, be it from different starting nodes—each visit will contribute to the node's as described. Consequently, each molecule node's value is based on the significance by which one or more genes were considered differentially expressed in a certain experiment, yielding a measure for the degree to which its activity may be influenced in a tissue- or dignity-specific manner. Finally, the graph is visualized, representing each molecule node's (initial or calculated) value mapped to a color spectrum ranging from green ( = 1, i.e. specific for tissue/condition A) over yellow ( = 0, i.e. not tissue/condition-specific) to red ( = −1, i.e. specific for tissue/condition B). Thus a node's, or sub-network's, predominant color immediately gives a visual clue on which molecules—differentially expressed or not—may play pivotal roles in the tissues or conditions under examination. Different types of interaction imply different semantics of the propagated values. For instance, if an interaction between genes/molecules and is known to be of type ‘ inhibits ’, with being over-expressed under condition and not being expressed differentially, can be assumed to be more active under condition than under condition . This fact can be accounted for by inverting the propagated value for inhibitory interactions. DEEP allows the user to choose which interaction types are to be treated this way and re-calculates the node coloring accordingly. In addition to reconstructing the signal transduction network in the above-described manner, DEEP also partitions the network into so-called signal percolation clusters. These clusters represent self-contained units of information flow, each starting from one differentially expressed gene and containing all its putative effector molecules found in the network. Percolation clusters hence represent the sub-networks, which may be subject to causal influence by a differentially expressed gene. Consequently, if another starting node is encountered during the calculation of the starting node 's percolation cluster, 's cluster gets merged into the cluster currently under construction since it is well imaginable that 's activity may be modulated by . DEEP has been implemented in the Java programming language and consists of three parts: the DEEP core server, a JSP-based web interface and a downloadable Java client. This architecture prevents users from having to care for maintenance tasks, like updating the utilized databases. Just like the browser-based solution, the Java client communicates with the core server via firewall-friendly HTTP. DEEP can be accessed in two ways. On the one hand, a web interface provides all basic functionality, while, on the other hand, a Java client is available which runs on the user's local computer. Both routes of access are available from the DEEP homepage (). Since the local Java client allows for much more interactivity when inspecting the calculation results (e.g. zooming and panning), its use is strongly encouraged. It can be launched via a Java Web Start link, rendering its installation very straightforward; the only requirement is the presence of a moderately recent Java runtime environment (Java 5.0 or newer). Hypotheses generated by tools like DEEP need to be verified either experimentally or by checking against previously published experiments. To this end, we present two literature-based case studies of DEEP usage. Another straightforward way of extending the system would be to utilize databases other than TRANSPATH, which contain data on biomolecular interaction networks. The next data resource projected to be thus included into the DEEP server is KEGG's LIGAND () section, providing information about the enzymes and metabolites involved in metabolic pathways. In a metabolic network, two enzyme genes will be considered to interact if one enzyme's product can serve as the other enzyme's substrate, or vice versa. The Reactome knowledgebase of biological pathways () is another candidate for inclusion. Finally, the inclusion of molecules upstream to differentially expressed genes into the reconstructed network might also be worth implementing. A propagation of -values ‘backwards’ through (directed) interaction edges will lead to the identification of molecules preceding differentially expressed gene products in, e.g., signaling cascades. The molecules thus identified can be considered as only being enabled to trigger certain functions in a tissue- or condition-specific manner by differentially expressed genes. We have shown that by combining gene expression data with biological knowledge about biomolecular interaction networks, additional genes (or gene products) can be identified which may play distinct roles in different tissues or under different conditions, though not being differentially expressed themselves. While most gene expression analysis methods focus only on those genes found to be expressed differentially and hence will not present additional effectors in their result lists, our method extends the range of genes that may be crucial for the processes under examination, thus shedding new light on our understanding of the molecular basis of physiological processes as well as their pathological aberration, culminating in the prediction of new plausible targets for rational drug design.

T cell development in the thymus produces peripheral T cells with a useful repertoire of T cell antigen receptors (TCRs) and is coordinated by extracellular signals from antigen receptors, cytokines and stromal cells. These stimuli link via tyrosine kinases to a diverse network of GTPases and serine kinases. The importance of serine kinases for thymocyte development is revealed by studies of phosphoinositide-dependent kinase l (PDK1), a kinase that phosphorylates a key ‘T' loop site within the catalytic domain of AGC serine kinases (). PDK1 phosphorylates and activates multiple AGC kinases, including phosphatidyl inositol-3 kinase (PI3K)-controlled serine kinases such as Akt or protein kinase B (PKB), the 70-kDa ribosomal S6 kinase-1 (S6K1) and the 90-kDa ribosomal S6 kinase (RSK). Deletion of PDK1 in T cell progenitors causes a block in T cell development (). PDK1-deficient thymocytes fail to differentiate past the pre-T cell stage and do not make peripheral T cells. The TCR has variable α/β subunits that recognise peptide/MHC complexes and the selection of cells that have successfully rearranged their TCRβ gene locus is an essential step in T cell development (; ). This occurs in T cell precursors, which do not express the major histocompatibility complex (MHC) receptors CD4 and CD8 (double-negative (DN) thymocytes). T cell progenitors initiate rearrangements of the TCRβ locus and if successful produce a TCRβ chain that permits surface expression of the pre-TCR complex comprising TCRβ, pre-Tα and CD3 antigens. The pre-TCR, in combination with Notch, then induces proliferation and differentiation into CD4/CD8 double-positive (DP) thymocytes (; ; ; ; ; ; ). The process is referred to as β-selection, because only cells that have successfully rearranged their TCRβ locus undergo proliferative expansion and differentiate to DPs. The pre-TCR activates PDK1-regulated kinases (; ) and these are clearly essential for thymocyte development as PDK1-null pre-T cells do not proliferate or differentiate yet express TCRβ subunits and have a functioning pre-TCR (). Loss of PDK1 causes a size reduction of β-selected pre-T cells indicative of a metabolic defect (). In this context, Notch signals support metabolism throughout β-selection () and the failed development of PDK1-null cells could thus reflect defective trophic responses to Notch. A number of facts are consistent with this hypothesis: deletion of floxed Notch1 alleles via excision blocks thymocyte development at the same stage as PDK1 deletion (); Notch-ligand interactions in pre-T cells activate the PDK1 substrate PKB (); expression of a constitutively active PKB mutant can partially substitute for Notch and maintain thymocyte metabolism during β-selection (); and PKB serine kinases are required for the transition of DN thymocytes to the DP stage, partly by enhancing the proliferation and survival of cells undergoing β-selection (). A key question then is whether the impact of PDK1 loss on thymocyte development stems only from its key role in regulating PKB and/or reflects the unresponsiveness of cells to Notch-induced trophic signals. To address these issues, the present study compares the development of wild–type (WT) and PDK1-null T cell progenitors in an model that uses OP9 stromal cells expressing the Notch ligand delta-like 1 (OP9-DL1 cells) to drive thymocyte differentiation (; ). To determine the contribution of the PDK1/PKB pathway to thymocyte development, we studied the differentiation of thymocytes whose WT PDK1 allele were substituted with a PDK1 L155E mutant, that permits phosphorylation of PKB, but not other substrates such as S6K1, PKC, SGK or RSK (, ). The substitution of leucine (L) 155 in PDK1 with glutamate (E) disrupts the integrity of an important PDK1 domain termed the PIF-binding pocket. This domain is not required for PKB phosphorylation, but is necessary for PDK1 to interact with carboxy-terminal hydrophobic motifs in substrates such as S6K1 and RSK (, ; , ). The PDK1 L155E mutant can thus support normal activation of PKB, but not S6K1 and RSK activity (). The value of PDK1 L155E in dissecting the contribution of different PDK1 substrates has been demonstrated (; ). It can substitute for WT PDK1 in insulin responses in skeletal muscle demonstrating that PKB is the relevant target for PDK1 in these cells (). However, PDK1 L155E does not support normal murine embryo development, indicating that PDK1 activation of PKB is not sufficient for all PDK1 functions (). The present results show that PDK1-null pre-T cells cannot respond to Notch-induced trophic signals, because Notch signals via PDK1 to induce and sustain expression of key nutrient receptors. In the absence of PDK1, pre-T cells are blocked at the DN stage of thymocyte differentiation. Expression of PDK1 L155E, which supports activation of PKB is able to replace WT PDK1 and restore nutrient receptor expression and pre-T cell differentiation, but does not restore normal thymus cellularity. These results identify an important role for the PDK1/PKB pathway during thymocyte differentiation, but show that the importance of PDK1 in the thymus cannot be ascribed solely to its role upstream of PKB. T cell development is thus equally dependent on PDK1 substrates that interact with PDK1 via its PIF domain. To assess whether PDK1 is required for Notch-induced thymocyte growth, differentiation and proliferation, we compared the responses of WT versus PDK1-null pre-T cells in an system using OP9 stromal cells expressing the OP9-DL1. The OP9-DL1 system allows an assessment of Notch responsiveness in pre-T cells (; ). PDK1-null pre-T cells were obtained from PDK1 mice (hereafter referred to as T-PDK1 mice), which were generated by crossing mice with floxed PDK1 exons 3 and 4 on both alleles (PDK1) () with mice expressing recombinase under the control of the proximal p56 proximal promoter (), which induces expression in DN T cell progenitors in the thymus (; ). DN thymocytes can be subdivided on the basis of differential surface expression of CD44 and CD25: the first T cell progenitors are CD44/CD25 (DN1) followed sequentially by CD44/CD25 (DN2), CD44/CD25 (DN3) and CD44/CD25 (DN4) populations. In T-PDK1, thymocytes are blocked in development at the DN4 stage (). WT DNs cultured on OP9-DL1 cells increase in cell size, whereas cells cultured on OP9 cells in the absence of DL-1 do not (). DNs cultured on OP9-DL1 cells also proliferate (70- to 80-fold in 6 days) and differentiate to DP thymocytes that co-express CD4 and CD8 (). In contrast, DN thymocytes cultured on OP9 cells in the absence of Notch ligand do not proliferate, although they differentiate to DPs (). T-PDK1 DN thymocytes fail to increase in cell size in response to OP9-DL1 () and do not proliferate or differentiate to DPs (). The small size of T-PDK1 DN thymocytes cultured on OP9-DL1 cells ( versus ) parallels what is seen in PDK1 deleted pre-T cells (), which are small compared to controls (). PDK1-null pre-T cells thus do not grow, differentiate or proliferate in response to Notch. We examined whether this defect could be attributed to defective expression of Notch on T-PDK1 DNs. A DL4-IgG fusion protein was used to monitor Notch1 expression (). shows that WT and PDK1-null pre-T cells express Notch1. There is a two-fold reduction in Notch1 expression in PDK1-null cells, but does not explain failed development of T-PDK1 DNs as Notch1 haploinsufficient thymocytes develop normally (F Radtke, personal communication) and within WT DN4s, two-fold differences in Notch1 levels had no impact on cell size (). Moreover, PDK1 DN4s expressing identical levels of surface Notch1 as WT DN4s (Region A in ) are still smaller (). T lymphocyte cell size is dependent on the regulated expression of amino-acid transporters that include CD98 (42F) as a key component (; ; ; ; ). The upregulation of CD71, the transferrin receptor, mediates iron uptake and is also critical for cell growth (; ). Therefore, we considered whether defective growth response of PDK1-null pre-T cells might result from defective expression of nutrient transporters during thymus development. In initial experiments, we explored whether thymocyte differentiation is accompanied by changes in expression of CD71 and CD98. shows that DN3 thymocytes express low levels of CD98 and CD71, whereas surface expression of these nutrient transporters is high in DN4s. Progression through β-selection is thus accompanied by upregulation of CD71 and CD98. We then asked whether Notch controls CD71 and CD98 expression. shows that DN3 thymocytes cultured on OP9-DL1 cells upregulate expression of CD71 and CD98, but this does not happen when they are cultured on OP9 cells in the absence of Notch signalling. DN3s cultured on OP9-DL1 cells also increase in cell size (). DN4 thymocytes have high levels of CD71 and CD98 expression (); they strikingly lose expression of these receptors if cultured on OP9 cells, but maintain CD71 and CD98 expression if cultured on OP9-DL1 cells (). Hence, sustained Notch signalling is required to maintain CD98 and CD71 surface expression on DN4 thymocytes. The cell size of DN4 thymocytes is also sustained by Notch: DN4s cultured on OP9 cells rapidly decrease in size, but maintain cell size when cultured on OP9-DL1 cells (). Notch1 thus regulates nutrient receptor expression on pre-T cells, but a key question is whether pre-TCR signalling is essential for T cells to become competent to respond to Notch1 in the context of CD98 and CD71 expression. To address this issue, we cultured DN3 cells from Recombinase gene 2-null mice (Rag2) that lack a pre-TCR, on OP9-DL1 cells. shows that Rag2 DN3s fail to upregulate CD98 and CD71 when cultured on OP9-DL1 cells. Hence, Notch is not sufficient to regulate CD71 and CD98 expression in the absence of the pre-TCR, rather Notch is required to sustain expression of CD98 and CD71 on β-selected pre-T cells. Next, we examined CD71 and CD98 expression in thymocytes from T-PDK1 mice. In T-PDK1 mice, there is partial deletion of PDK1 in DN3s, complete PDK1 loss in the DN4s and a failure of these cells to develop beyond the DN4 stage (). shows that CD98 and CD71 expression levels are defective in PDK1 DN4 thymocytes. DN4 thymocytes from T-PDK1 mice were uniformly low for CD98 expression and had a high frequency of CD71 low cells compared to WT controls. The loss of CD71 and CD98 expression is not reflective of a global signalling defect in T-PDK1 DN4s, because the upregulation of a number of other surface receptors occurs normally as PDK1-null DN3s transit to DN4s. T-PDK1 DN4s can thus upregulate CD2 and CD5 expression normally; a marker of a functioning pre-TCR (). The reduced expression of CD98 and CD71 on PDK1-null DN4s reveals that the expression of these key nutrient receptors is mediated by PDK1. Is reduced expression of nutrient receptors sufficient to explain the defective growth responses of pre-T cells? Experiments to limit nutrient availability revealed that the ability of pre-T cells to increase cell size and develop in response to Notch was dependent on the availability of exogenous amino acids (data not shown). We also examined whether differences in levels of expression of nutrient receptors on the surface of pre-T cells had an impact on trophic responses in pre-T cells. The data () show that WT DN4s expressing low levels of CD71 or CD98 are small cells compared to cells expressing high levels of CD71. Hence, expression of CD98 and CD71 is rate limiting for pre-T cell growth. There are multiple PDK1 substrates in T cells, including PKB, S6K1 and RSK. Expression of constitutively active PKB mutants can substitute for cytokines and induce expression of CD71 and CD98 in haematopoietic cells (). PDK1 is necessary for PKB activation and hence the impact of PDK1 loss on thymocyte development could reflect loss of PKB activity in these cells. In this context, simultaneous loss of three PKB isoforms blocks thymocyte development at the DN/DP stage, similar to the effects of deleting PDK1 (; ). However, it is not known if PKB is the only effector of PDK1 in the thymus and it is important to address the involvement of additional PDK1 substrates in T cell development. One way to explore the contribution of PKB to PDK1 function is to use ectopically expressed constitutively active PKB mutants, a strategy used to determine much of what we know about PKB in T cells (; ; ; ; ; ). However, PKB mutants may not necessarily mimic the actions of endogenous PKB. One alternative and powerful strategy to explore the relative contributions of different PDK1 substrates to a biological response is to analyse mice with ‘knock-in' mutations of PDK1 alleles (, ; ; ). One such PDK1 mutant, containing a L155E mutation, cannot support activation of S6K and RSK but allows activation of PKB (; ). Mice that are homozygous for the PDK1 L155E mutation do not survive beyond developmental stage E12. However, it is possible to generate thymocytes that selectively express PDK1 L155E (T-PDK1) by backcrossing mice that express a single PDK1 L155E allele and a single PDK1 floxed allele (PDK1) with mice expressing recombinase under the control of the p56 proximal promoter (). The presence of the single PDK1 floxed allele in all tissues allows normal mouse development. However, as progenitors enter the thymus and express recombinase, the WT allele is deleted thereby generating pre-T cells that express a single PDK1 L155E allele (PDK1 , referred to as T-PDK). A comparison of the ability of a single PDK L155E allele versus a single WT allele to support T cell development can then be assessed relative to the impact of complete PDK1 deletion. The crucial control for the analysis of pre-T cells that express the single L155E mutant allele are pre-T cells that express a single WT PDK1 allele (PDK1 , referred to as T-PDK1). The impact of PDK1 haplo-insufficiency on thymocyte development has been described and of relevance to the present work is the fact that a single WT PDK1 allele restores DN4 cell size and cell cycle progression and allows thymocyte differentiation from DNs to DPs (). Experiments with PDK1 L155E explore whether kinases that interact with PDK1 via its PIF domain are required for thymocyte development. Initially, we compared CD71 and CD98 expression in DN4 thymocytes from T-PDK1 and T-PDK1 mice. shows that expression of PDK1 L155E restores CD71 and CD98 expression to a level comparable to that seen in DN4s expressing WT PDK1. PDK1 L155E could also restore thymocyte differentiation in PDK1 null pre-T cells. The distribution of the different DN subpopulations in T-PDK1 and T-PDK1 mice was comparable (). shows that unlike T-PDK1 thymi, which have virtually no DPs and single positives (SPs), T-PDK1 thymi contain DP and SP subsets at relatively normal ratios. Mature WT SP thymocytes express high surface levels of the TCRβ and there was a normal frequency of these cells in T-PDK thymi (). PDK1 L155E can thus substitute for PDK1 loss and is sufficient for normal CD71 and CD98 expression and for pre-T cell differentiation into DPs and SPs. A genomic PCR analysis verified that -mediated deletion of the flΔneo allele had occurred in T-PDK DPs (). There was also functional loss of PDK1 as judged by failed phosphorylation of the ribosomal S6 subunit in T-PDK1DN4 thymocytes. S6 phosphorylation is mediated by S6K1, which requires phosphorylation at its ‘T loop' site by PDK1. A crucial step that enables this phosphorylation event is the docking of PDK1 via its PIF domain at a hydrophobic-motif in S6K. The L155E mutant of PDK1 is thus defective in activating S6K1 (). The activity of S6K1 can be monitored by quantification of phosphorylation of its substrate, the ribosomal S6 subunit, by intracellular staining with specific phosphorylated S6 antisera. Basal levels of S6 phosphorylation are high in WT DN4s, but absent following PDK1 deletion (). shows that expression of a single WT PDK1 allele can support activation of S6K1 in pre-T cells, but the PDK1 L155E allele cannot. These data were confirmed by western blot analysis (). Western blot data also confirm that PDK1 L155E can support PKB activity as determined by the normal phosphorylation of GSK3α on its PKB substrate sequence Serine 21 in both WT and PDK1 L155E pre-T cells (). In contrast, GSK3α Serine 21 phosphorylation is absent in PDK1-null pre-T cells. PDK1 L155E substitutes for WT PDK1 in the context of CD98 and CD71 expression and DN to DP differentiation. However, DN to DP transition is normally accompanied by a large proliferative expansion of β-selected pre-T cells and there was no such proliferative expansion in T-PDK1 mice (). T-PDK1 mice thus have very low numbers of DPs and SPs and peripheral α/β T cells compared to control mice (). PDK1 L155E thus supports DN to DP differentiation, but does not restore thymus cellularity. In accordance with the failed proliferative expansion of T-PDK1 thymocytes, there was a decrease in the frequency of cells in S/G2 phases of the cell cycle in T-PDK1 DN4s (). Moreover, the normal increase in cell size that accompanies the DN3 to DN4 transition is lost in T-PDK1 mice (). Hence, expression of a single WT PDK1 allele can maintain normal growth and proliferation of pre-T cells, whereas PDK1 L155E cannot. Thus, PDK1 activation of PKB is sufficient for pre-T cell differentiation, but PDK1 substrates that interact with PDK1 via its PIF domain are required for optimal pre-T cell growth and proliferative expansion. The PDK1 PIF domain is required for the phosphorylation of multiple kinases, such as RSK and S6K1 (). We examined the role of RSK in pre-T cell development using a recently described specific RSK inhibitor BI-D1870 (). This inhibitor inhibits all four RSK isoforms, but not other related AGC kinases (). shows that RSK inhibition does not prevent the DN to DP differentiation of WT pre-T cells, but does inhibit Notch-induced proliferative expansion of these pre-T cell subpopulations (). Inhibition of RSK with BI-D1870 thus allows cells to differentiate, but does not allow them to proliferate normally. We also considered the possible contribution of S6K1 to thymus development. S6K1 must be phosphorylated by PDK1 at its T loop site to become activated, but also has an additional requirement for PDK1 function as its activation is dependent on PKB, which regulates S6K1 via modulation of TSC-1/2 function and mTOR (mammalian target of rapamycin) (; ; ). In this respect, in mature T cells, S6K1 couples PKB and mTOR to the control of T cell cycle progression (). Previous studies have shown that inhibition of mTOR with rapamycin reduces thymus cellularity (; ). show that rapamycin blocks Notch-dependent proliferative expansion of pre-T cells , but not DN to DP differentiation. As rapamycin immediately inactivates S6K1, these results are consistent with a role for S6K1 in thymus development. However, important caveats are that rapamycin can disrupt sustained activation of PKB (). Moreover, mTOR is needed for PKB activation of S6K1, but can also regulate S6K1-independent pathways (). Accordingly, the ability of rapamycin to inhibit T cell proliferation may be explained by a requirement for mTOR signalling to mediate PKB function, rather than reflecting loss of S6K1 activity. In this context, in a pro-B cell line, PKB could induce expression of CD98 and CD71 via a mTOR-dependent pathway (). To explore the mTOR requirement for nutrient receptor expression in pre-T cells, we examined the impact of rapamycin on CD98 and CD71 expression in the OP9 DL-1 system. The data () show that the ability of Notch signals to sustain expression of CD98 and CD71 is partially sensitive to rapamycin inhibition. Hence, the expression of nutrient receptors in pre-T cells is regulated by PDK1- and mTOR-sensitive pathways. The differentiation and proliferation of pre-T cells in the thymus is dependent on sustained Notch receptor/ligand interactions (; ; ). The present work shows that Notch-mediated trophic and proliferative responses are lost following deletion of PDK1. As T cell progenitors progress through β-selection, there is induced cell surface expression of CD98, an essential subunit of mammalian -type amino-acid transporters and CD71, the transferrin receptor. The expression of these nutrient receptors is induced by the pre-TCR in a Notch- and PDK1-dependent manner. Moreover, sustained Notch signalling is required to maintain CD71 and CD98 expression as pre-T cells progress through β-selection. Strikingly, PDK1 deletion prevents expression of these key nutrient receptors and prevents Notch-induced growth responses. The observations that PDK1 regulates the expression of key amino-acid and iron transporters during thymocyte development and is required for Notch trophic responses offer an explanation as to why the loss of PDK1 has such a major impact on T cell development. The loss of nutrient receptor regulation in PDK1-null T cells will thus render the cells metabolically insufficient to support the demands made by the massive proliferative expansion occurring as thymocytes make the DN to DP transition. To investigate PDK1 substrates that control expression of CD98 and CD71, we looked at the expression of these nutrient receptors in pre-T cells expressing a PDK1 L155E allele that can activate PKB, but not AGC kinases that bind to PDK1 via its PIF domain. PDK1 L155E restores expression of CD71 and CD98 and also supports both the differentiation of pre-T cells to DP and mature SP cells. PKB thus has key functions regulating the expression of amino-acid transporters and transferrin receptors during thymocyte development. However, PDK1 L155E did not restore normal thymus cellularity. Hence, PKB activation is sufficient for thymocyte differentiation, but not sufficient for optimal proliferation. Accordingly, kinases that interact with PDK1 via its PIF domain are necessary for cell growth and proliferation of pre-T cells in the thymus. The role of PDK1 during thymocyte development is thus multifaceted and not simply a reflection of its role as an upstream activator of PKB. Serine/threonine kinases that interact with PDK1 via its PIF domain include RSK and S6K1. The present results identify RSK as an important signalling molecule for pre-T cell development, as inhibition of this kinase inhibited proliferative expansion of pre-T cells, but did not prevent their differentiation. The importance of the PDK1 PIF domain could thus be explained by its role in RSK regulation. However, there is still the question of the involvement of S6K1. It is difficult to assess a thymus autonomous role for S6Ks in mice doubly deficient for S6Ks, as these show perinatal lethality (). Nevertheless, we tested the role of another S6K regulator mTOR in thymocyte growth and proliferation by examining the impact of rapamycin on Notch-induced thymocyte development. These experiments revealed that the inhibition of mTOR with rapamycin prevents optimal nutrient receptor expression and growth and proliferation of DN thymocytes, but permits their differentiation. Accordingly, there is a dual requirement for PDK1 and mTOR for cell growth and proliferation during thymocyte development. In summary, the present study shows that PDK1 is important during thymocyte development, because it regulates expression of key nutrient receptors on the surface of pre-T cells and mediates Notch-induced cell growth and proliferative responses. PDK1 was first identified as a key activator of PKB, but was also found to mediate activation of other AGC kinases such as RSK and S6K1 (, ; , ). A PDK1 L155E mutant, with a disrupted PIF pocket can support normal activation of PKB and restore nutrient receptor expression and DP and SP thymocyte differentiation in PDK1-null cells. However, PDK1 L155E cannot support normal thymocyte proliferation. The importance of PDK1 for β-selection thus reflects the role of multiple AGC serine kinases in this process rather than just reflecting PKB regulation. Mice (5- to 7-week-old) (backcrossed for at least seven generations to C57BL6) were maintained in SPF conditions under Home Office project license PPL60/3116. T-PDK1 mice (PDK1 or PDK1 were generated as described previously (; ; ). T-PDK1 mice have partial deletion of PDK1 in DN3 thymocytes and complete ablation of PDK1 in DN4 pre-T cells (). Control mice used for analyses of T-PDK1 mice were age-matched WT littermates or PDK1 where indicated. Mice containing a knock-in mutation of PDK1, wherein leucine (L) at residue 155 was changed to glutamate (E) (PDK1) () were crossed with PDK1 mice to generate mice expressing a single PDK1 L155E allele and a single PDK1 floxed allele (PDK1). These were then backcrossed with mice expressing recombinase under the control of the p56 proximal promoter in T cell precursors to generate PDK1 mice (referred to as T-PDK1). Control mice PDK1 (referred to as T-PDK) were generated by crossing mice expressing a single PDK1 floxed allele (PDK1) with mice expressing recombinase under the control of the p56 proximal promoter. Cre-mediated deletion of the floxed PDK1 allele from DP sorted thymocytes in T-PDK1 mice was confirmed by PCR using primers p99 (5′-ATC CCA AGT TAC TGA GTT GTG TTG GAA G) and p100 (5′-TGT GGA CAA ACA GCA ATG AAC ATA CAC GC). A PCR product of 200 bp was generated for WT and PIF alleles and a 250 bp band was generated when the flΔneo allele was present. A 200 bp product generated by PCR analysis using primers p80 (5′-CTA TGC TGT GTT ACT TCT TGG AGC ACA G) and p100 was indicative of deletion of exons 3 and 4 of PDK1. Antibodies conjugated to fluorescein isothiocyanate, phycoerythrin, allophycocyanin and biotin were obtained from either Pharmingen (San Diego, CA, USA) or eBioscience (San Diego, CA, USA). TriColour-conjugated antibodies were obtained from Caltag (Burlingame, CA, USA). Cells were stained for surface expression of the following markers using the antibodies in parentheses: CD4 (RM4-5), CD8 (53-6.7), CD25 (7D4), CD44 (IM7), CD71 (C2), CD98 (RL388), Thy1.2 (53-2.1), TCR β (H57-597), B220 (RA3-6B2) and TCR γ/δ (GL3). Cells were stained with saturating concentrations of antibody in accordance with the manufacturer's instructions. Data were acquired on either a FACS Calibur (Becton Dickinson, Franklin Lakes, NJ, USA) or an LSR1 flow cytometer (Becton Dickinson) using CellQuest software and were analysed using either CellQuest (Becton Dickinson) or FlowJo (Treestar, San Carlos, CA, USA) software. Viable cells were gated according to their forward scatter and side scatter profiles. CD4 and CD8 DN subsets were gated by lineage exclusion (lineage) of all CD4, CD8 DP and SP cells and TCR γ/δ. DN3s and DN4s were further defined as CD25CD44 and CD25CD44 thymocytes, respectively. Mature SP thymocytes were defined as Thy-1, TCRβ and positive for CD4 or CD8 expression. A DL4 IgG fusion protein was used to monitor Notch1 expression as described previously (). Cellular DNA content was measured by DAPI staining of saponin permeabilised cells. Phospho-S6 levels in thymocytes was assessed as described previously (). OP9 bone marrow stromal cells expressing OP9-DL1 () and control OP9 cells were a gift from Juan Carlos Zúñiga-Pflücker (Toronto, Canada). OP9 cells were maintained in αMEM supplemented with 50 μM 2-mercaptoethanol, 100 U/ml penicillin, 1 mg/ml streptomycin and 20% heat-inactivated FBS. Sorted thymus DN3 (CD25CD44) and DN4 (CD25CD44) subsets were co-cultured on OP9 and OP9-DL1 monolayers for times indicated in Figure legends. DN3 and DN4 thymocytes were purified by first depleting thymic populations of CD4 and CD8 cells using an AutoMACs magnetic cell sorter (Miltenyi Biotech, Auburn, CA, USA) before sorting to a purity greater than 95%, using a FACS VantageSE cell sorter (Becton Dickinson). On day of harvest thymocytes were filtered through 50 μm filters to remove OP9 cells before developmental progression of T lineage cells was assessed. Sorted DN3 and DN4 thymocytes were lysed on ice in NP-40 lysis buffer (50 mM Hepes (pH 7.4), 75 mM NaCl, 1% Nonidet P-40, 10 mM sodium fluoride, 10 mM iodoacetimide, 1 mM EDTA, 40 mM β-glycerophosphate, protease inhibitors, 1 mM phenylmethylsulfonyl fluoride, 100 μM sodium orthovanadate). Lysates were centrifuged at 1600 for 15 min at 4°C. Protein samples were separated by sodium dodecyl sulphate 4–12% polyacrylamide gel electrophoreisis, transferred to nitrocellulose membrane and detected by western blot analysis using standard techniques. Blots were probed with antibodies that recognise phosphorylated GSK3 α/β on Ser21/9 (Cell Signaling Technologies, Danvers, MA, USA), total GSK3 α/β (Upstate, Hampshire, UK), phosphorylated S6 ribosomal protein on Ser235/236 (Cell Signaling Technologies) and total S6 ribosomal protein (Cell Signaling Technologies).

Endocytosis induced by ligand−receptor interaction has been directly linked to signal transduction mediated by Rab5 and its effector APPL1 (daptor protein containing H domain TB domain and eucine zipper motif; ; ). The small GTPase Rab5 is a generally acknowledged prominent regulator of vesicle trafficking enroute from the plasma membrane to early endosomes (), whereas APPL1 (also called DIP13α) is identified with signaling pathways of adiponectin, insulin, EGF, follicle stimulating hormone receptor, neurotrophin receptor (TrkA), oxidative stress, and DCC-mediated apoptosis (; ; ; ; ; ). Within this milieu, APPL1 specifically binds to the GTP-bound, active form of Rab5. In response to extracellular stimuli, Rab5 hydrolyzes its bound GTP, releasing APPL1 from an endocytic structure, and allowing APPL1 to further interact with components of nucleosome remodeling and histone deacetylase complexes. The interaction with Rab5 is essential for APPL1 localization to the endosomes and is indispensable for the functional cycle of APPL1 (). Human APPL1, a multidomain protein 709 amino-acid (aa) residues in length contains an amino (N)-terminal BAR (in1/mphiphysin/VS167) domain and a PH (leckstrin omology) domain followed by a carboxy (C)-terminal PTB (hosphoyrosine inding) domain (; ; ). The Rab5-binding site is located in the N-terminal BAR-PH region (), while the C-terminal region is found to interact with a host of other proteins, including the adiponectin receptor (), Akt2/PKBβ kinase (), tumor suppressor DCC (), TrkA, and TrkA interacting protein GIPC1 (). Based on aa sequence analysis, BAR domains have been identified in many proteins involved in intracellular trafficking, but sequence homology is low in general among known BAR domains (; ). The BAR domain typically contains three long kinked α-helices (α1, α2, and α3) that form a well-packed, crescent-shaped, symmetrical, six-helix bundle, side-by-side antiparallel homodimer; a structure proposed to exert its function as a convex membrane-curvature sensor or stabilizer. The concave surface of the BAR dimer is proposed to bind preferentially to a negatively charged, curved membrane largely through electrostatic interactions. Furthermore, some BAR domains have been found to bind to small GTPases, a class of intracellular molecular switches (; ); thus, their membrane association is directly linked to regulation of signal transduction and trafficking. However, currently available structural information suggests that bindings of the BAR domain to GTPases and to membrane lipids are incompatible, because both interactions appear to compete for the same concave surface region of the BAR dimer (). The BAR domain of APPL1 is required for Rab5 binding and membrane recruitment (), although the mechanisms remain to be elucidated. The PH domain is approximately 100-residue long, and has been identified in over 100 different eukaryotic proteins such as kinases, isoforms of phospholipase C (PLC), GTPases, and their regulators; most of which participate in cell signaling and cytoskeletal regulation (). Despite their minimal sequence homology, the three-dimensional (3D) structures of PH domains are remarkably conserved. They possess a common core consisting of seven β-strands and a C-terminal α-helix (). Some PH domains specifically bind to phosphatidylinositol phosphates, suggesting that one possible function of this family is to anchor the host proteins to membranes. PH domains are also suggested to bind to the Gβγ complex of the heterotrimeric G protein, protein kinase C, and small GTPases. Nevertheless, none of these functions is absolutely conserved. For instance, the PH domain of APPL1 alone is insufficient for binding to the membrane (). The PH domain immediately follows the C-terminus of the BAR domain; such a BAR-PH motif is essential for Rab5 binding. The same motif has also been found in a homolog Rab5 effector APPL2, centaurin-β family members, GRAF2, and oligophrenin (), but the 3D structure organization of BAR-PH motif and its functional implication remained elusive until now. In order to address the functional roles of the BAR-PH motif in APPL1 and related proteins, we have carried out structure-function studies on human APPL1 and determined the crystal structures of the Rab5-binding region of APPL1 as well as the BAR domain alone. The results show that two BAR-PH molecules form an integrated, symmetric homodimer, and the PH domain has extensive intermolecular interactions with the BAR domain. The BAR dimer of APPL1 has a stronger curvature than other reported BAR structures. Further mutagenesis analyses allowed us to identify the binding sites on both APPL1 and Rab5. In sharp contrast to the presumed conflict between concurrent membrane association and GTPase binding by the BAR dimer (), the novel binding mode of the BAR-PH dimer should permit simultaneous interactions with both. Recombinant proteins of human APPL1 N-terminal fragments including the BAR (residues 5−265) and BAR-PH domains (residues 5−385) were expressed in , then purified using His tag affinity chromatography. The samples were crystallized after removing the tag with thrombin, which generated a four-residue (Gly–Ser–His–Met) peptide N-terminal to the native Asp5 residue. The BAR domain crystal diffracted up to 1.8-Å resolution on a beamline at the Argonne Advanced Photon Source (APS) synchrotron facility. The crystal belongs to P222 space group. Phases of the structural factors were determined using the Se-Met-based single-wavelength anomalous dispersion (SAD) method (). There is one APPL1 BAR molecule per asymmetric unit, with ∼41% solvent content. Regions of the N-terminus (up to Thr12), Leu75−Asp79, and C-terminus (i.e., Pro260−Asp265) were missing from the final refined model because of lack of interpretable electron density. The BAR-PH crystal diffracted to 2.05-Å resolution at the synchrotron facility. The crystal also belongs to P222 space group. There is one APPL1 BAR-PH molecule per asymmetric unit, with ∼45% solvent content. Phases of this crystal form were calculated using a combination of molecular replacement and SAD methods, and further improved with density modification. Regions of N-terminal non-native tripeptide (i.e., Gly–Ser–His), Gly76−Asp78, Asn288−Ser295, and C-terminus (i.e., Ser380−Glu385) lacked interpretable electron density and were omitted from the final refined model. Data collection and refinement statistics are summarized in . From the two crystal forms of APPL1 peptides, we obtained two crystallographically independent BAR domain models. In both cases, the APPL1 BAR domain has three long helices, namely α1, α2, and α3. In addition, the APPL1 BAR domain contains an extra nine-turn α-helix, α4 (; ). The two models could be superimposed onto each other with a moderate, 1.3-Å, Cα-atom root mean square deviation (r.m.s.d.) if flexible terminal and loop regions (i.e., residues 5−18, 75−79, 151−153, and 255−265) were omitted. Thus, the overall structure of the BAR domain remains the same either alone or in the context of BAR-PH motif. In each of the two crystal forms, two BAR molecules form a tightly packed dimer, which assumes a crescent-like shape, a hallmark of the BAR dimer structure (). In the BAR dimer, the helix α1 forms an antiparallel helix bundle with its symmetry counterpart, giving shape to the concave surface of the crescent-like dimer. Helix α4 packs against α3 of the symmetry mate on the convex side of the dimer, and its C-terminus points to the tip of the crescent. Over 4400 Å solvent accessible surface (SAS) from each protomer is buried in the dimer interface. The addition of each α4 helix to the canonical BAR motif results in approximate 1900 Å buried SAS on the two protomers, corresponding to over 40% of the total buried SAS. Although the overall folding of APPL1 BAR domain is similar to previously reported BAR domain 3D structures (i.e., arfaptin2, PDB file 1I4T; amphiphysin, 1URU; and endophilin, 1ZWW), those structures are in general more similar to each other than to the APPL1 BAR domain. For instance, using 150 Cα atoms of the common helical regions, the r.m.s.d. values between the dimer of APPL1 BAR and 1I4T, 1URU, and 1ZWW were 3.7, 3.8, and 4.4 Å, respectively, while those among 1I4T, 1URU, and 1ZWW range between 2.4 and 2.6 Å. In addition, the APPL1 α1 and α2 helices lack extensive patches of positively charged aa residues on the concave surface (); such patches are thought to be essential for some BAR containing proteins to induce tubule formation (). The curvature of the concave face of the BAR dimer is thought to play an important role in membrane bending and/or curvature sensing (). We implemented a computing algorithm to calculate the curvature radius () and found that the APPL1 BAR dimer has an about 55 Å (; ), significantly smaller than the values of other BAR dimers (). Thus, the APPL1 BAR dimer has the strongest curvature among known BAR dimer structures. The APPL1 PH domain encompasses residues Asn276−Leu379 and has a typical PH folding (). The core structure of PH domain consists of a pair of nearly orthogonal β-sheets of four and three antiparallel β-strands (β1–β2–β3–β4 and β5–β6–β7; ). The C-terminal α-helix, α, packs against both β-sheets and contributes to the core of the domain. In the PH domain, connecting loops are named after the preceding β-strands (e.g., the loop between β1 and β2 is called L1, etc). The canonical ligand-binding site is composed of β1, L1, β2, L3, and L6 (exemplified in the crystal structure of PLC-δ1, PDB file 1MAI) and, roughly speaking, is confined to a triangular area with L1, L3, and L6 as the three vertices. Some positively charged or polar residues that have been previously identified as critical for lipid binding in this ligand-binding triangle are not conserved in APPL1 (), consistent with the fact that APPL1 alone lacks membrane binding ability. In our crystal structure of the APPL1 BAR-PH dimer, the two PH domains are located at the opposite ends of the crescent-shaped dimer, and each has fairly extensive contact with the BAR domain of its symmetry mate (). The addition of the PH domain expands the BAR dimer in the longest dimension from 140 to 170 Å, but hardly changes the height of the dimer (i.e., the dimension along the two-fold axis direction) and its curvature. Using its β1, β2, L3, and L7 regions, the PH domain contacts the BAR domain of its symmetry mate in two places ( and ; ). First, the motif of DSPxxR (where x stands for any aa residue) at the N-terminal of BAR domain contacts β1, β2, and L3 regions of PH domain. For instance, the hydroxyl group of Tyr283 in β1 forms a 2.6-Å hydrogen bond with the backbone carbonyl oxygen of Asp15 (). Second, the conserved DxxDRRYCF motif in the loop L7 of the PH domain is directly in contact with the loop connecting α2 and α3 in the BAR domain (). The buried SAS from each BAR-PH molecule in the dimer is about 6600 Å. Thus, in the presence of PH domain, the buried SAS is 50% larger than that of the dimer formed by BAR domain alone (4400 Å). The canonical ‘ligand'-binding triangle of the PH domain is oriented about 60° from the concave side of the BAR dimmer, so that both the PH triangle and BAR concave surface could be brought within the vicinity of a curved membrane simultaneously. Nevertheless, as discussed earlier, key residues for lipid interaction are not conserved in APPL1. The C-terminus of PH is exposed to solvent in the dimer, consistent with the fact that it connects to the C-terminal region including the PTB domain. Because of the interaction between DSPxxR motif and PH domain, the rest N-terminus peptide (residues 5−12) clearly became ordered in the BAR-PH crystal structure, in comparison to the BAR domain-alone crystal structure, where residues N-terminal to Leu13 were invisible in the electron density map. The fixed N-terminal peptide in the BAR-PH structure has an extended backbone conformation between Met4 (remnant from the His-tag cleavage) and Pro8, followed by a one-turn 3 helix (). This region has several important intramolecular contacts mainly with helices α1 and α3. For instance, the Leu7 side chain inserts between the aromatic rings of Phe26 in α1 and Phe182 in α3. Meanwhile, the side chain of Asn186 forms two hydrogen bonds with the backbone amide and carbonyl groups of Leu7, respectively. All these hydrophobic and hydrogen bond interactions appear conserved among BAR-PH containing proteins (). Furthermore, the N-terminus is surrounded by a number of regions from the dimer partner, including the helix α4 and flexible loop connecting α1 and α2 (where residues 76−78 were mobile in the crystal structure). For instance, Lys6' (where the prime stands for the dimer partner) forms a salt bridge with Asp243 in α4 between the protomers. In addition, Ile9' forms hydrophobic interactions with Met247, Ile251, and Leu254 in α4 (). To investigate roles of the BAR−PH interaction in solution, a double point mutation, S16E/P17E, at the BAR−PH interface was made. These residues are located in the region N-terminal to the BAR domain and form close contacts with β2 and L3 of the PH domain (). The recombinant protein of S16E/P17E double mutant in the context of BAR-PH was expressed predominantly in the insoluble fraction of cell lysate; however, the same mutations behaved normal in the BAR-only construct (data not shown). Moreover, expression of the APPL1 PH domain alone in did not produce soluble recombinant protein. The data suggest that the dimer interaction between PH and BAR domains is critical for the solubility and stability of the APPL1 PH domain. Consistent with this, our analytical ultracentrifugation (AUC) data showed that BAR-PH protein has a higher dimerization affinity in solution (=0.34 nM) than BAR domain alone (=0.13 μM; ). To study APPL1−Rab5 interaction in solution, we performed glutathione -transferase (GST)-mediated pull-down assays. The APPL1 BAR-PH domain (residues 5−385) and a longer fragment with a 40-residue extension downstream of the PH domain, APPL1 (5−419), were each effectively pulled down by GTP-bound GST−Rab5 fusion protein (). The APPL1 protein was pulled down by either WT Rab5 preloaded with non-hydrolysable GTP analog (GppNHp) or Rab5-Q79L defective in GTP hydrolysis (with or without preloaded GTP analog), but could not be effectively pulled down by either the WT Rab5 preloaded with GDP or Rab5-S34N defective in GTP binding ( and data not shown). In contrast to the BAR-PH domain, we confirmed that APPL1 BAR domain alone (residues 5−265) cannot directly interact with Rab5 (data not shown) (). Furthermore, using different Rab5 truncation variants, we demonstrated that the N-terminus (residues 1−15) and C-terminus (residues 185−215) of Rab5 are dispensable for interaction with APPL1 (). Binding affinity between full-length Rab5-Q79L and APPL1 (5−419) was quantitatively determined in a surface plasmon resonance (SPR) experiment. Rab5 was coupled to the SPR biosensor chip in random orientations, and APPL1 (5−419) was applied as the analyte at concentrations of 0.15−12 μM (). The dissociation constant, , for the Rab5−APPL1 interaction measured from this experiment was 0.9 (±0.7) μM, with and of 1.3 (±0.6) × 10 M s, and 1.2 (±0.4) × 10 s, respectively. This value is typical for an interaction between a Rab and its effectors (). To identify Rab5-binding site(s) in APPL1, GST−Rab5-Q79L (full length) was used to pull down APPL1 variants having surface point mutations. The WT APPL1 (5−419) fragment was used as the parental construct for the mutagenesis, because this fragment is easily distinguishable from GST−Rab5 by size on SDS–PAGE gels without the need for Western blot analysis. A total of 31 point mutations were made at 27 distinct, solvent-exposed positions (; ), based on the structural information of BAR-PH motif. Most of these point mutations were located in the PH domain or near the BAR−PH interface, which are the surface regions most conserved between APPL1 and APPL2 (). Substitution mutants were designed to maximize potential mutational effects on Rab5 binding (e.g., by flipping charges or switching between hydrophobic and hydrophilic residues) without disrupting the overall structure. In addition, the flexible L1 loop (residues 289−294) was truncated and replaced with one Gly residue. All of these APPL1 variants, as well as the WT construct, were expressed in , with comparable yields from the soluble fractions (data not shown), in contrast to the mutations at BAR−PH interface mentioned earlier. This suggested that the thirty or so surface mutations had little effect on the stability of BAR-PH dimer. Among them, seven mutants, including V25D, N308D, M310K, A318D, G319R, L321D, and D324A, either abolished or significantly reduced (i.e., retaining <30%) Rab5 binding compared with the WT APPL1 (). In the 3D structure, most of these residues cluster in an elongated surface area formed by β3, L3, and β4 of the PH domain, defining a major Rab5-binding site (; ). In addition, the effect of the V25D mutant suggests that the BAR domain also contributes to Rab binding either directly or indirectly. On the other hand, L1 loop seems not to be required for Rab5 binding; significance of the apparent, positive effect of the truncation mutant () remains to be studied. We further extended these binding studies and confirmed the above Rab5-binding site in the cell, by monitoring Rab5-mediated APPL1 recruitment to early endosomes in the cell via confocal microscopy. In this case, the RFP (DsRed-monomer)−Rab5-Q79L fusion protein was expressed in PC12 cells, targeted to the early endosomes, and recruited effectively the coexpressed GFP (green fluorescence protein)−APPL1 to these early endosomes (). Importantly, APPL1 (5−385), that is, the BAR-PH domain, was sufficient to target to Rab5-Q79L containing early endosomes (). In contrast, one of the Rab5-binding defective mutants (A318D) failed to target the early endosomes and exhibited a diffused pattern throughout the cytoplasm in the cell (). Interaction between small GTPase and BAR domain has been exemplified in a complex crystal structure of Rac and arfaptin2 before (). Based on the following observations, however, we excluded the possibility of a Rac−arfaptin2-like binding mode for the Rab5−APPL1 interaction. First, the linear dimension of Rab5 is less than 50 Å, which is significantly smaller than the distance (∼60 Å) between the putative Rab5-binding site in the PH domain and the central region of the BAR dimmer, where Rac binds with arfaptin2. Second, the isolated APPL1 BAR domain did not bind to Rab5 in our pull-down assay. Third, we mutated APPL1 Asn52, which is at the position equivalent to Rac-binding site in arfaptin2, to either a smaller (Ala) or larger (Arg) side-chain residue, and the mutations showed no effect on the binding to Rab5. Rab5 subfamily contains several members, including Rab5, Rab21, and Rab22. Among them, Rab5 and Rab22 share a higher overall sequence identity with each other than with Rab21 (). This difference was used to explain the ability of Rab5 and Rab22, but not Rab21, to share some common effectors such as EEA1 and rabenosyn5 (; ). Therefore, we tested APPL1 binding specificity towards other members in the Rab5 subfamily, using GST−Rab21 (full length) and GST−Rab22 (2−192) to pull down APPL1 (5−419). Interestingly, APPL1 would bind to Rab21 in a GTP-dependent manner (), indicating that APPL1 is an effector for both Rab5 and Rab21. On the other hand, we were unable to detect any binding between APPL1 and Rab22 in the pull-down assay (). We could not rule out possible interaction between them because our recombinant Rab22 might not have folded correctly in based on the following observations: the expression level of Rab22 was 10- to 20-fold lower than Rab5 and Rab21, and the GTP loading rate of Rab22 was lower too (data not shown). Therefore, we focused our study on Rab5 and Rab21 for their interactions with APPL1. We demonstrated that Rab21 and Rab5 have similar but not identical binding profiles towards APPL1 variants (), which may be explained by their sequence divergence. This differential binding to Rab5 and Rab21 by APPL1 may allow analysis of the functional roles of each Rab−APPL1 interaction, for example, by specifically abolishing one interaction while retaining others. Next we investigated the APPL1-binding regions in Rab5. Since residues responsible for APPL1 binding are likely to be located in the switch I, switch II, and interswitch regions, whose conformations change between different nucleotide binding states, these regions became the main objects of our investigation. In addition to relevant Rab5 mutations that we made in previous studies, several point mutations in the Rab5 switch I region (i.e., residues 40−53) were tested for APPL1 binding. We found that point mutations in the 42−48 region significantly reduced APPL1 binding, while L38R, Q49A, E50A, and I53N showed little or no detectable effect (). Consistent with our previous structural studies on Rab5–rabaptin5 interaction (), all mutations within the 38−50 region did not interfere with Rab5−rabaptin5 binding (). Furthermore, with the knowledge of crystal structures of Rab5–rabaptin5 and Rab22–rabenosyn5 complexes, it is clear that both Rab5 effectors rabaptin5 and rabenosyn5 bind to the so called invariant hydrophobic triad of Rab5 (i.e., Phe57, Trp74, and Tyr89) (). Mutation of any of these residues usually strongly inhibits the Rab-effector binding (; ). Interestingly, in our mutagenesis analysis, the APPL1-binding was affected by W74R and Y89R, but not by F57R point mutation in Rab5 (). Taken together, our results indicate that APPL1 binds to Rab5 regions including the 40−48 loop and switch II, ∼30 Å across. In addition, we showed that the two effectors, APPL1 and rapaptin5, could compete for Rab5-binding (data not shown), confirming that the binding sites of APPL1 and rabaptin5 on the Rab5 surface overlap with each other. To further define the Rab5−APPL1 binding mode, we performed extensive pull-down analyses between variants of Rab5 and APPL1, looking for reversal mutants that could rescue the lost binding ability of others. We identified one such pair; APPL1-N308D abolished the binding to Rab5, while Rab5-L38R had no effect on APPL1 binding. However, Rab5-L38R was found to bind with APPL1-N308D, but not with the other tested APPL1 variants of similar hydrophobic-to-charged mutations, including V25D, A318D, and L321D (). This result suggests that Rab5-L38R restores binding for APPL1-N308D through complementary, electrostatic, yet specific interactions. It further implies that the position 308 in the β3 strand of APPL1 PH domain is in the vicinity of position 38 in the α1 helix of Rab5 in their complex. Since both APPL1 and APPL2 bind to Rab5, their Rab-binding sites are likely located in a surface region that is conserved between the two APPL proteins. There are no deletion/insertion differences in the BAR-PH region between them (), and an inspection of the APPL1 BAR-PH dimer surface indicates that the most conserved surface region is located on the PH domain surface and the BAR-PH junction (). Furthermore, neither PH domain () nor BAR domain alone (data not shown) can directly bind Rab5, suggesting that the dimer interface between PH and BAR domains plays a critical role in Rab5 binding directly or indirectly. This binding mode between Rab5 and APPL1 is apparently distinct from that between Rac and arfaptin, which only requires BAR dimerization (). To investigate further the structural basis of APPL1 and Rab5 interaction, we have performed extensive mutagenesis analyses. A BAR dimer breaking mutant (F210D/F211D) and the BAR−PH interface mutant (S16E/P17E) are both insoluble when expressed in (data not shown), supporting the notion that the functional form of APPL1 BAR-PH domain is a dimer. Importantly a series of surface point mutants are soluble, allowing us to analyze the binding properties between these APPL1 mutants and Rab5 (). The results indicate that Rab5 specifically binds to the PH domain of APPL1 in the context of BAR-PH dimer, and this binding may marginally extend to the neighboring BAR domain. Our structure-functional analyses are consistent with existing biochemical data. For example, a previously reported triple mutation of APPL1 within the PH domain, K280E/Y283C/G319R, disrupts Rab5 binding (). This effect can be fully explained based on the importance of the BAR−PH and PH−Rab5 interfaces. Combined results from our mutagenesis pull-down experiments (, and ), crystal structures of the BAR-PH domain of APPL1 (), and structures of GTPase domain of human Rab5 in different nucleotide binding modes (, ) clearly explain the requirement of GTP-bound Rab5 for APPL1 binding. Based on available information, we have modeled the interaction between the two proteins. With the assumption that both proteins remain rigid bodies, our complex model satisfies constraints imposed by the mutagenesis pull-down results (). Over 1200 Å SAS area combined from both the APPL1 dimer and Rab5 would be buried in their interface. In this putative Rab5−APPL1 binding mode, APPL1 interface includes L2, β3, L3, and β4 regions. Note that the L3-β4 region showed weak electron density in the crystal structure, indicating its higher mobility and possible adaptability in forming a complex with Rab5. On the Rab5 side, two regions that harbor binding-defective mutations are involved in the complex formation: the loop 42−48 and switch II (). Furthermore, the reversal mutation pair, Rab5-L38R and APPL1-N308D () would directly interact with each other inside the interface of our complex model. The bound Rab5 molecules would extend the concave surface of the APPL1 dimer, with both the N- and C-termini of Rab5 exposed to solvent. Considering that there are about 30 residues C-terminal to our Rab5 model, which are necessary for membrane association but excluded from the crystallography study, our model would allow Rab5 molecules to anchor to the membrane through the added C-terminal tails and to interact with APPL1 at both ends of the BAR-PH dimer (). The Rab5 C-terminal tail is likely flexible, supporting that recruitment of APPL1 to the endocytic vesicle may not require its direct contact with the membrane. In the complex, the Rab5 molecule does not block the C-terminus of the PH domain, allowing peptide extension of the APPL1 molecule from the BAR-PH domain. In contrast to the α-helix dominant Rab-binding motifs of all other effectors of known 3D structures (; ; ; ; ), the Rab5-binding motif of APPL1 is mainly composed of two β-strands, β3 and β4, and their connecting loop L3. Although the exact binding position on the Rab protein and orientation of the effector helices may differ among available complex structures, all these Rab-binding domains interact with the invariant hydrophobic triad. However, we have identified a Rab5 mutation in the hydrophobic triad, F57R, that does not interfere with APPL1 binding, but abolishes the binding to another Rab5 effector, rabaptin5 (; ). In contrast, several point mutations in the switch I region of Rab5 affect the binding of APPL1 but not rabaptin5. The 42−48 region in Rab5 has not been previously reported to be involved in effector binding. GTPase binding has emerged as a major function of PH domains in addition to lipid binding (). For example, PH domains in some guanine nucleotide-exchange factors (GEF) have been shown to bind directly to their cognate small GTPases (, ; ), and our data now show direct interaction between the APPL1 PH domain and Rab5. So far, only two crystal structures of small GTPase−PH domain complexes are available. One is Ran−RanBD1 (PDB file 1RRP). The interactions between the Ran GTPase domain and RanBD1 PH core domain is fairly minor, occurring between the switch I region of the GTPase (equivalent to the 40's in Rab5) and strand β2 of the PH domain. This interaction alone is unlikely to be sufficient to form a stable complex. Indeed, Ran has a long C-terminal peptide beside the GTPase domain, while the PH domain of RanBD1 has an extra N-terminal peptide. These two terminal peptides wrap around the partner proteins forming the major interaction between Ran and RanBD1. Such an interaction seems not to be required for Rab5 and APPL1, because the GTPase domain of Rab5 and BAR-PH domain of APPL1 are sufficient to mediate their interaction. The second published small GTPase−PH complex is that of Ral−Exo84 (PDB file 1ZC3). In this complex, the PH domain of Exo84 uses L1, β5, and L6 to interact with the interswitch and switch II regions of Ral forming an intermolecular β-sheet extension mediated by the PH β5 strand and GTPase β2 strand (). Our mutagenesis analysis points to a different surface region (β3, L3, and β4) of the PH domain for Rab5 binding. Therefore, the Rab5−APPL1 interaction represents a new GTPase−PH binding mode. Both APPL1 and APPL2 are identified as Rab5 effectors, and their overall aa sequences are highly homologous. In particular, residues on the APPL1 BAR dimer interface, BAR−PH interface, and the presence of the α4 helix seem well conserved in APPL2 (). Therefore, APPL2 BAR-PH domain most likely forms a homodimer very similar to that of APPL1. Furthermore, these conserved structural features may also extend to other BAR-PH containing proteins (; ). For instance, no helix breaking aa sequence appears in the middle of their predicted α4 regions. Based on the APPL1 BAR-PH crystal structure, we find that, in general, the PH domain is more conserved than the BAR domain, and most of the highly conserved positions are located closer to the BAR−PH interface rather than the central region of the symmetric dimer. For example, the two major contact regions between PH and BAR domains (i.e., DSPxxR and DxxDRRYCF) are conserved at the aa sequence level among BAR-PH containing proteins. In addition, correlated mutations are present between these proteins at the BAR−PH interface. Thus, we propose that all BAR-PH containing proteins share similar 3D structures in the corresponding regions and that the BAR-PH motif may function as a general structural unit to interact with membrane-bound proteins and other molecular moieties. In some BAR containing proteins, it is proposed that there exists an amphipathic helix N-terminal to the α1 helix of BAR domain, and they are called an N-BAR motif (; ). It is suggested that this extra N-terminal region facilitates membrane binding (or bending). A similar N-BAR structure was predicted for APPL1 and APPL2 (), but our current APPL1 crystal structure does not show such a structural motif. Instead, the N-terminal region assumes an extended conformation and packs in the groove formed between helices α1 and α3 on the convex side of the crescent-shaped dimer (). Since the N-terminal regions of the other BAR-PH containing proteins () share similar sequences, we suggest that none of these proteins contains an N-BAR motif in their 3D structure. Lacking both the amphipathic helix N-terminal to the BAR domain and the lipid-binding motif in the PH domain () may explain the Rab5-dependent membrane association of APPL1. In contrast, the PH domains of centaurin-β1/2 contain the key, basic residues for phosphoinositide binding (; ). If their PH domains are oriented similarly to that in the APPL1 dimer, the canonical (i.e., L1–L3–L6), ligand-binding triangle in their PH domains likely contributes to direct membrane association of these proteins. While it is suggested that Rab5−APPL1 interaction mediates a signal transduction pathway between the plasma membrane and the nucleus, the mechanism by which Rab5 binding stimulates APPL1 translocation to the nucleus remains elusive. The current BAR-PH structure may help to clarify the mechanism. Interestingly, the sequence of ‘PKKKENE' was identified in the BAR domain of APPL2 as a potential nuclear localization signal (). The corresponding region in APPL1 is the solvent exposed loop connecting α2 and α3 at the tip of the dimer () and has a fairly conserved sequence (). In addition, our preliminary data suggest that there is no detectable binding between the BAR-PH domain and the C-terminal region of APPL1 (data not shown), which makes it unlikely that Rab5 may regulate APPL1 through interference with the intramolecular interaction of the latter. It seems more probable that the Rab5−APPL1 complex recruits downstream effectors to propagate the signal transduction process. Unlike other Rab effectors, APPL1/2 proteins function in the signaling pathway from the so-called signaling endosome to nucleus. Our data show that APPL1 interacts with the Rab5 protein using a novel binding mode; it remains to be proven whether such a binding mode is essential for APPL1 function. Whereas it has been shown that APPL1 does not bind other Rab proteins miscellaneously (), we demonstrate that APPL1 is also an effector of Rab21, indicating that APPL1 adopts a binding mode shared by both Rab5 and Rab21. It raises the possibility that, besides Rab5, other members of this Rab subfamily may also be involved in the APPL1 signaling pathway. Constructs of human APPL1 (GenBank ID: NP_036228) (5−265) (i.e., the BAR domain) and APPL1 (5−385) (i.e., the BAR-PH domain) were inserted into the vectors pET28a and pET15b (Novagen), respectively, between I and HI restriction sites. The N-terminal few residues in the native sequence are hydrophobic and were deleted in an attempt to improve the solubility. Point mutations were introduced into the pET15b-APPL1 (5−419) parental construct using QuickChange site-directed mutagenesis kit (Stratagene). His-tagged proteins of APPL1 (5−265) and APPL1 (5−385) were expressed as soluble recombinant proteins in BL21 Star (DE3) strain of (Invitrogen), and cells were harvested after induction with 0.1 mM isopropyl-β--thiogalactopyranoside (IPTG) for 8 h at 25°C. The cells were lysed with lysozyme, and the lysate supernatant was purified with His-Select affinity beads (Sigma). In both cases, the His tag was removed with thrombin. After further purification with Resource-Q anion-exchange chromatography (GE Healthcare), both protein samples were concentrated to ∼30 mg ml in (20 mM Tris–HCl (pH 8.0) and 0.1% (v/v) β-mercaptoethanol (βME)) and stored at −85°C until needed. APPL1 (5−419) mutants were expressed similarly. Se-Met-substituted proteins were expressed in B834 (DE3) pLysS cells (Novagen) in minimal media supplemented with 40 mg l Se-Met (Sigma) and purified using the same procedure as the native protein. Recombinant proteins of human Rab5a variants (GenBank ID: NM_004162), human Rab21 (BC021901), and human Rab22a (NM_020673) fused with an N-terminal GST were expressed in BL21 and purified with GST-affinity chromatography. The sample was concentrated to ∼20 mg ml and stored in 1 × phosphate-buffered saline (PBS) with 0.1% (v/v) βME at −80°C. Recombinant protein of human rabaptin5 (551–862) (GenBank ID: CAA62580) was expressed and purified as described previously (); two additional point mutations, C719S and C734S, were introduced to reduce aggregation. Crystals of APPL1 (5–265) were grown at 20°C with the hanging drop vapor diffusion method. The Se-Met incorporated protein sample diluted to 10–20 mg ml was mixed 1:1 (v/v) with the reservoir solution of 0.1 M magnesium formate and 0.1% (v/v) βME. Crystals were transferred to a cryo-protectant solution of (88% saturated LiSO, 14 mM magnesium formate, 20 mM Tris–HCl (pH 8.0), and 0.1% (v/v) βME) by gradually changing the drop solution in 20 min, followed by cooling in liquid nitrogen. A data set was collected at selenium edge at sector 22 BM of the Argonne APS facility. In the Rab5−APPL1 pull-down experiment, 30 μg GST−Rab fusion protein (52 kDa) was incubated with 60 μl of 30% slurry of GSH–Sepharose 4B (GE Healthcare) at 22°C for 30 min. Nucleotide loading reaction was performed on the GSH beads in an exchange buffer of (1 × PBS, 2 mM DTT, 1 mM MgCl, 4 mM EDTA, and 400 μM GppNHp or GDP) at 22°C for 30 min. Increasing the magnesium ion concentration to 20 mM terminated the loading reaction. Soluble fractions of cell lysate containing all His-tagged APPL1 (5–419) variants were analyzed by SDS–PAGE to confirm their comparable expression level and solubility. The GSH resin carrying nucleotide-loaded GST−Rab fusion protein was incubated with ∼50 μl cell lysate (∼200 μg APPL1 variant, 50 kDa) at 22°C for 30 min, then washed three times with 200 μl of (1 × PBS, 2 mM DTT, and 4 mM MgCl) and resuspended in 20 μl of 2 × reducing SDS sample buffer. The sample was subjected to SDS–PAGE analysis, visualized with Coomassie blue stain. The same samples were analyzed with chemiluminescence Western blot (GE Healthcare) and His-tag antibody then detected on films which were semi-quantified using the computer software ImageJ () including its default calibration. The relative band intensity of each mutant versus WT from multiple experiments is shown in . Coordinates and the structural factors of the APPL1 crystal structures have been deposited to PDB under codes 2Q12 (BAR domain structure) and 2Q13 (BAR-PH domain structure).

Many signaling systems use heterotrimeric (αβγ) G proteins to relay signals from heptahelical receptors to downstream effectors. To accomplish signal transduction, G proteins act as conformational sensors of a guanine nucleotide, which is bound to the α subunit. G proteins that are charged with guanosine diphosphate (GDP) are in the inactive state, where the α and the βγ subunits are associated with each other. Receptor activation accelerates the exchange of bound GDP for free GTP () followed by the dissociation of active Gα-GTP from βγ subunits. Hydrolysis of the bound GTP by a GTPase reaction brings the Gα subunit back to the inactive state (), which is characterized by tightly bound GDP and a reassociation with the βγ complex. To ensure specificity, high effective concentrations, and speed of interaction, the G protein signaling components are usually attached to the membrane domain as peripheral membrane proteins. Membrane attachment of heterotrimeric G proteins has been extensively investigated, and the effect of lipid modification on membrane localization has been addressed by several studies (; ; ; , ; ). All G protein α subunits (with the exception of transducin) are palmitoylated, and some are additionally modified by myristoylation. The α subunits of Gq/G, including the eye–specific Gqα, as well as Gsα, Gα, and Gα are modified only by palmitoylation. The corresponding βγ subunits undergo isoprenylation of a cysteine residue at the so-called CAAX box of the γ subunit (for review see ; ; ). Plasma membrane attachment of the α subunits Gsα and Gqα is dependent on coexpression with the βγ subunits (, ). Furthermore, the βγ subunits, having only one membrane attachment signal on the γ subunit, are poorly targeted to the plasma membrane and require coexpression of the α subunit for efficient plasma membrane attachment (; ; ). Altogether, these studies led to a model of two membrane attachment signals that are needed for plasma membrane attachments and localization of heterotrimeric G protein subunits (; ). It should be noted, however, that most of these studies have been performed by using various culture cells that were transfected with vectors yielding overexpressed proteins (usually the α and βγ subunits of the heterotrimeric G protein). This procedure is bound to cause distortion of the original stoichiometry of α and βγ subunits, which is difficult to control under these conditions. The extensively studied visual system combined with the large repertoire of visual mutants offer a unique opportunity to study in vivo the various roles of the βγ dimer, its cellular localization, and the functional consequences of altering α/βγ stoichiometry. The visual system is a specialized system that is composed of highly polarized and compartmentalized cells that sequester the phototransduction machinery in a specific signaling compartment called the rhabdomere (; ). This signaling compartment is functionally equivalent to the vertebrate rod photoreceptor outer segment, which also sequesters the phototransduction machinery in a specific cell compartment. Phototransduction in is initiated upon the activation of rhodopsin by light and proceeds through a photoreceptor-specific Gq protein (Gq; ), which, in turn, activates the phospholipase C enzyme effector (). Upon activation, the eye-specific Gqα subunit (Gqα) dissociates from the eye-specific βγ dimer (Gβγ) and translocates, at least in part, from the membrane to the cytosol (; ). In this study, we show (by using a series of eye-specific β hypomorph mutants) that the βγ dimer has a crucial role in both membrane attachment and rhabdomeral targeting of the α subunit that can account for the decreased light sensitivity previously observed in these mutants (). On the other hand, by using the almost null mutant for the eye-specific Gqα subunit α , we found that the βγ dimer is dependent on the α subunit for membrane attachment but not for targeting to the rhabdomere, suggesting a role for the βγ dimer in targeting the heterotrimer to the photoreceptor signaling compartment (the rhabdomere). An analysis of the protein levels of Gqα and Gβ subunits revealed a surprising twofold excess of the Gβ subunit over the Gqα subunit. Mutants that eliminated this excess showed a dramatic increase in spontaneous activity of the phototransduction cascade. Conversely, double mutations that also reduced the level of Gqα and, thereby, restored the excess of Gβ over Gqα completely reversed this phenotype. Together, these results provide a significant insight into the strategy used by the photoreceptor cell in vivo to avoid spontaneous activity at the G protein level. The α subunit of the heterotrimeric G protein and the tightly associated complex of βγ subunits undergo dissociation and reassociation during activation of the phototransduction cascade. Therefore, it is expected that these subunits would influence one another's level, localization, and function. Previous studies that addressed these questions used tagged subunits and heterologous expression in tissue culture cells. Qualitatively, it is now generally accepted that plasma membrane attachment of the α subunit requires coexpression of the βγ subunit complex (; , ), and, reciprocally, plasma membrane attachment of the βγ subunit complex requires coexpression of the α subunit (; ). Although a great deal has been learned from these previous studies, little is known about the localization of G protein subunits in their natural environment and how the stoichiometry of these subunits affects the level, localization, and function of G protein subunits under physiological conditions. To test the effect of various subunits on the level of one another, we have used the eye–specific Gβ subunit mutants (β) that were described by and the eye-specific Gqα subunit mutant (α ) that was described by . The hypomorph β mutants β , β , and the heterozygote of the most severe mutant, β /+, express the Gβ subunit protein at levels of 4, 13, and 50% of wild-type flies, respectively (). Despite the progressive decrease in the Gβ subunit level in these mutants, the level of the α subunit was undiminished and is the same level as in wild-type flies ( A). Similarly, in the α mutant, which expresses negligible levels of the α subunit, the level of the Gβ subunit was the same as in wild-type flies ( C). Although the levels of the Gβ subunit that we found in the β and β mutants were higher than those previously reported (), the progressive decrease of the Gβ subunit protein among these mutants was similar ( B). The eye-specific Gγ subunit, which forms an extremely tight complex with the Gβ subunit, completely disappeared in the severe β mutant but, like Gβ, was undiminished in the α mutant ( D). Therefore, we can conclude that the β mutants are, in fact, βγ mutants and that the effects observed in β mutants can be ascribed to a decrease in the level of the βγ subunit dimer without effecting the level of the α subunit. To understand how Gβ affects the localization of Gqα, we extended our analysis to membrane attachment and targeting of the α subunit in β mutants. As shown in , the low levels of βγ subunits in β mutants cause a progressive decrease in the fraction of the α subunit that is attached to the membrane. Quantitatively, the decrease in membrane attachment of the α subunit is proportional to the percent decrease in the level of the β subunit. The molecules that participate in phototransduction, including the eye-specific DGq subunits, are confined to a specific signaling compartment (the rhabdomere). Thus, we investigated how the decreased levels of βγ subunits affect the targeting of the α subunit to the signaling compartment. Using immunogold EM with antibodies against the eye-specific α subunit, we counted the gold particles in 20 cross sections of equal size from wild-type and mutant rhabdomeres. This analysis revealed that the quantity of the Gqα subunit in the rhabdomere of different mutants corresponds with the level of the α subunit that is membrane attached () and indicates that the βγ subunit complex controls both membrane attachment and rhabdomeral targeting of the α subunit. One of the major advantages of for the study of phototransduction in vivo is the ability to examine the electrophysiological response in detail and characterize the phenotype that results from a decrease in a specific phototransduction component, which is caused by mutation. Two physiological phenotypes were observed for β mutants (). The first phenotype was a dramatic loss of light sensitivity (reaching a decrease by two orders of magnitude in the β mutant), and the second phenotype was a slow termination of the light response. To address the possibility that the reduced sensitivity to light in β mutants arises from a reduction in membrane-bound Gqα, we reexamined the sensitivity to light in four mutants with reduced levels of Gβ. and show a correlation between the level of membrane-bound Gqα () and the sensitivity of the response to light () in which low levels of membrane-bound Gqα correspond to low light sensitivity. The latter was accompanied by a modified waveform of the light-induced current (, inset). The fact that heterozygous β /+ showed only a minor reduction in the sensitivity to light is consistent with previous results showing that 50% of Gqα is sufficient to maintain normal sensitivity to light (). Together, these results indicate that the loss of light sensitivity is caused by the effect of β mutants on membrane attachment and targeting of the Gqα subunit to the signaling compartment (the rhabdomere). Clearly, when rhodopsin and Gqα are present in different cellular compartments, the Gqα subunit cannot transfer signals from rhodopsin to the phospholipase C enzyme. To examine the effect of Gqα on the localization of Gβ, we measured the distribution of Gβ between the membrane and cytosol in wild-type and Gαq mutant flies. In contrast to the light-dependent translocation of Gqα from the membrane to the cytosol (), Gβ was about equally distributed between the membrane and the cytosol under both light and dark conditions (). A longer period of illumination for up to 4 h did not alter the Gβ distribution (not depicted). These results suggest that the βγ complex remains partly bound to the membrane even when the α subunit is translocated to the cytosol. Indeed, it has been shown that although rhodopsin–Gα interactions are reduced upon activation, rhodopsin–Gβγ interactions remain undiminished (). Moreover, electrostatic calculations showed that upon dissociation from the Gα subunit, the β subunit of transducin exposed a prominent patch of basic residues that enhanced the membrane affinity of the βγ dimer by about an order of magnitude (). However, it has also been shown in the rat visual system that Gβγ subunits translocate from the outer to the inner rod segment in response to light, albeit at a slower rate than the translocation of the α subunit (). The effect of Gqα on the membrane attachment of Gβ was further studied using the Gαq mutant. In this mutant, which has a negligible level of Gqα, the Gβ subunit is localized mainly to the cytosol (>80%; ), suggesting that a newly synthesized Gβγ complex is dependent on the α subunit for membrane attachment. Failure of the βγ complex to bind by itself to the plasma membrane was previously observed in transfected cells (; ; ) and in Gα RNA interference of embryos (). However, immunogold EM using specific antibodies against Gβ revealed that the βγ complex is targeted to the rhabdomere even in the near absence of the α subunit ( C) but apparently remains soluble within this compartment. This result indicates that the βγ complex is targeted to the rhabdomere independently of Gqα but depends on the α subunit for tight membrane attachment. The presence of soluble Gβ in the rhabdomere can be a result of interactions with protein partners like phosducin () and regulators of G protein signaling proteins (). Although homologues of these proteins are present in the genome, their cellular localization in photoreceptors are currently unknown. The cellular localization of the Gq heterotrimer may be determined by the βγ complex. This finding is consistent with a previous report that ectopic targeting of the βγ complex to the mitochondria leads to mitochondrial localization of the Gzα subunit (). The presence of 80% of Gqα in a membrane-bound form in wild-type dark-adapted flies ( B, left), whereas only 50% of Gβ is membrane bound ( B, left), raised the question of the stoichiometry of these two components. To determine the levels of the subunits in vivo, we performed immunoblot analysis with a mixture of Gqα- and Gβ-specific antibodies at a concentration five times that required for their saturation. Furthermore, two different anti-Gβ antibodies that were raised against two different sequences of the Gβ protein gave similar results (see SDS-PAGE and immunoblotting). In wild-type flies, the amount of Gβ was ∼2.5 times higher than the amount of Gqα (). To verify the excess of Gβ over Gqα subunits in wild-type flies, which was determined by Western blotting, we calibrated the immunoblot with the use of purified recombinant Gqα and Gβ proteins. We determined the concentrations of the recombinant proteins spectrophotometrically by using calculated extinction coefficients of Gqα = 42,350 cm M and Gβ = 60,000 cm M at 280 nm. This quantitative analysis of two samples of wild-type fly head homogenate again revealed an excess of Gβ over Gqα of ∼2.5 times ( C). Most of the excess Gβ was present in the cytosol, whereas the membrane-bound fraction contained both the α and β subunits in about equal amounts (). Therefore, in rhabdomere membranes, all of the Gα molecules, which are in close proximity to rhodopsin, may be associated with the Gβ subunit. This finding also indicates that in the photoreceptor cells, there is a soluble pool of free Gβ subunit in the rhabdomere. The localization of soluble Gβ in the signaling organelle, the rhabdomere (), could be functionally important. The unexpected excess of Gβ over Gqα was almost completely abolished in the β heterozygous mutant (β /+). The ratio between Gβ and Gqα in this mutant was ∼1:1. The decrease in Gβ levels of this mutant did not change the ratio between membrane-bound Gqα and Gβ, which remained ∼1:1. In the soluble fraction, however, we found a large decrease of excess Gβ. Although the ratio between soluble Gβe and Gqα in wild-type flies was ∼7:1, the ratio in the β /+ mutant was reduced to ∼2.5:1 (). A new and striking phenotype of β mutants was revealed in this study. Whole cell patch-clamp recording of dark-adapted mutant photoreceptor cells showed spontaneous, unitary, inward currents that were similar in shape to the single photon responses known as quantum bumps (; ). The frequency of these spontaneous responses was different for the various β mutants. For the β mutant, only a low frequency of spontaneous bumps was observed, which was not much different from the frequency of spontaneous bumps observed in wild-type flies. A higher frequency of spontaneous bumps was clearly noted for the β mutant, whereas the most dramatic increase in the frequency of spontaneous bumps was observed for the β heterozygous mutant (β /+). The high frequency of spontaneous bumps in the heterozygous β mutant is surprising because this mutant has normal sensitivity to light in contrast to the β homozygote, which is the most severe mutant but has an almost normal frequency of spontaneous bumps ( and ). This complex behavior can be explained by the decreased levels of Gqα observed in the signaling compartment of these mutants (). Indeed, when the bump frequency was normalized to the number of rhabdomeral Gqα, a similar bump frequency per rhabdomeral Gqα was observed for all of the β mutants, whereas the wild-type bump frequency remained much lower ( C). To find out whether the high frequency of spontaneous activity is caused by activation of the G protein and not by the spontaneous activation of rhodopsin, we generated a heterozygous β /+ mutant with highly decreased levels of rhodopsin. To reduce the rhodopsin level in β /+ flies, we reduced the chromophore level by raising the flies on a carotenoid-deficient medium () for three generations (β /+ Vit A−). The metarhodopsin potential (M potential) is a linear electrical manifestation of the level of rhodopsin in fly photoreceptors (; ). A shows the amplitude of the M potential in β /+ flies raised on standard medium (top) compared with β /+ flies raised on carotenoid-deficient medium (bottom). The virtually complete elimination of M potential after carotenoid deprivation clearly shows that the rhodopsin level was largely reduced in these flies. This conclusion was further supported by measuring the sensitivity to light after carotenoid deprivation, which resulted in a reduction of ∼300-fold in sensitivity to light without a change in the distribution of Gqα in carotenoid-deprived flies (not depicted). shows that the high rate of spontaneous bumps, which is characteristic of β /+ flies, was not significantly changed by reduced levels of rhodopsin. This indicates that the high frequency of spontaneous bumps in the β /+ mutant does not arise from the spontaneous activation of rhodopsin in the dark. The excess of Gβ over Gqα that was observed in wild-type flies is almost abolished in the β /+ heterozygous mutant (); this finding raised the possibility that the excess in wild-type flies prevents the spontaneous activity of Gqα observed in the β /+ heterozygous mutant. To further test this hypothesis, we crossed the β mutant with the α mutant to generate a double mutant containing one copy of the α gene and one copy of the β gene (α /+β /+). The double mutant had about half the level of both Gqα and Gβ as wild-type flies ( A), restoring the excess Gβ over Gqα that was observed in wild-type flies (). This mutant showed almost normal sensitivity to light () and no spontaneous activity in the dark (). This result strongly suggests that the excess of Gβ over Gqα, rather than the absolute amount of the Gβ subunit, prevents the spontaneous activation of Gqα in photoreceptor cells. When Gβ mutants were first isolated (), it was reported that these mutations caused a dramatic decrease in the sensitivity to light, which was ascribed to participation of the β subunit in G protein–rhodopsin coupling. Our finding that the decrease in Gβ in β mutants is accompanied by a proportional decrease in Gqα in the rhabdomeral compartment does not support the previously claimed catalytic effect of the β subunit on light sensitivity (). Rather, we conclude that the decrease in light sensitivity of these mutants is caused by the presence of rhodopsin and the major fraction of the G protein α subunit in two different cellular compartments. Clearly, when these two components are present in different cellular locations, the photo-excited rhodopsin is unable to catalyze the exchange of GDP that is bound to the Gqα for free GTP, and the transduction process is prematurely terminated. The mechanism that underlies the decreased sensitivity to light in β mutants, therefore, is a structural change in the localization of the Gqα subunit. We also examined how a decrease in the α subunit of the α mutant influences membrane attachment and targeting of the βγ subunits to the rhabdomere. In this case, the βγ dimer is soluble and not membrane attached but is still targeted to the rhabdomere. The presence of βγ in the rhabdomeral cytosol may be physiologically important for preventing spontaneous activity because the βγ subunits are in close proximity to the membrane-bound signaling molecules. We have previously shown that the eye-specific Gqα subunit translocates from rhabdomeral membranes to the cytosol in response to illumination (). Gqα behaves like many other Gα subunits, which demonstrate activity-dependent translocation from the membrane to the cytosol (for review see ; ; ). The eye–specific βγ dimer behaves differently from the Gqα subunit, as it does not show any significant change in its distribution even after prolonged illumination (). A possible reason for this result might be an interaction between the γ subunit of the βγ dimer and the photoactivated rhodopsin. Such an interaction has been reported for the transducin γ subunit and the active form of vertebrate rhodopsin (). Both vertebrate and invertebrate photoreceptor cells contain high concentrations of rhodopsin, and even a weak interaction could be significant as a result of mass action. It should be noted, however, that studies in rat retina detected light-dependent movement of both the α and βγ subunits from the rod outer to inner segment, although the βγ subunits moved more slowly than the α subunit, suggesting that it might be caused by an interaction of the βγ complex with phosducin (, ). The different behavior of Gβγ subunits in vertebrate and might be caused by the difference in stability of the active rhodopsin in these two systems. Whereas vertebrate rhodopsin undergoes bleaching and inactivation, the activated rhodopsin of is stable for hours (). The Gβγ subunits are known to bind to Gα-GDP switch regions, thereby stabilizing the binding of GDP and suppressing spontaneous receptor-independent activation (; ; ; ). To find out whether this interaction is relevant to the eye–specific Gq heterotrimer, whose three-dimensional structure has not been determined, we have constructed a homology model of the DGq heterotrimer (αβγ) based on the crystal structure of transducin (unpublished data). It appears from the model that the Gβγ subunit complex directly contacts the switch I and switch II space regions of Gqα as was previously reported for other G proteins. Therefore, it is conceivable that the Gβγ complex prevents the spontaneous activation of Gqα by binding to Gqα switch regions. It should be pointed out, however, that the physiological consequences of this effect in vivo have not been described previously. Furthermore, the mechanism that suppresses the spontaneous activity of G proteins under physiological conditions is unknown. In this study, we report () the observation of a high frequency of spontaneous activity in the heterozygous β /+ mutant in which the level of the β subunit was decreased to 50% of its level in wild-type flies. Surprisingly, a further decrease in the level of Gβ in the β and β mutants did not increase the frequency of spontaneous activity but rather decreased the frequency. This is easily seen in the most severe mutant (β ), which has only 4% of Gβ, as the spontaneous activity is not much different from the low frequency in wild-type flies. This is probably the reason why the increase in spontaneous activity was not detected in the initial characterization of β mutants (). These results indicate that the relationship between the level of Gβ and the spontaneous activity is not straightforward. We suggest that the observed spontaneous activity of β mutant photoreceptor cells is regulated by two opposite effects of the βγ dimer. On the one hand, the decrease in βγ levels leaves some Gα-GDP unassociated with βγ, and this free Gα-GDP undergoes spontaneous exchange of the bound GDP for free GTP, leading to spontaneous activity. On the other hand, the decreased level of βγ leads to a proportional decrease of Gα in the signaling compartment, resulting in a diminished ability to activate the phototransduction process. To test this notion, we normalized the observed rate of spontaneous activity to the number of Gqα subunits in the rhabdomeres of Gβ mutants that lack excess Gβ over Gqα. We found ( C) similar frequencies of normalized spontaneous activity for all of the β mutants, which is consistent with a role of excess βγ over Gqα in suppressing spontaneous activity. One of the unexpected and novel findings of this study is the presence of the eye–specific Gβ subunit in ∼2.5-fold excess over the Gqα subunit. Because the levels of α and β subunit proteins are maintained independently of one another, unequal levels of these subunits are mechanistically possible. Our calibration curves using purified recombinant Gβ and Gqα proteins ( C) verified the excess of Gβ over Gqα subunits, which was determined by immunoblot analysis with a mixture of Gqα and Gβ antibodies ( A). Furthermore, we have shown that as long as the two antibodies are maintained at saturating concentrations and determinations are performed in the same gel, levels of the α and β subunits are obtained that nicely fit the expected results from gene dosage effects ( A). Furthermore, according to the “two-signal model” for membrane attachment of peripheral membrane proteins, one expects to find equal amounts of membrane-bound Gqα and Gβ subunits. In accord with this notion, although we found about twofold excess of total Gβ over Gqα, an analysis of these subunits in the membrane-bound fraction gave a ratio of 1:1. In the heterozygous β /+ mutant, in which there is a reduction of 50% in the level of the β subunit, yielding a β/α ratio of ∼1, we found a dramatic increase in the spontaneous activity of photoreceptor cells ( and ). The critical role of the excess of Gβ over Gqα was revealed in the α /+β /+ double heterozygous mutant, in which the rate of spontaneous activity was dramatically reduced by restoring the excess of Gβ over Gqα. This indicates that the excess of Gβ, rather than its absolute amount, is important to maintain a low frequency of spontaneous activity. Furthermore, this mutant rules out the possibility that the spontaneous activity we observed was caused by side effects of the β mutation. Altogether, this is the first demonstration of the strategy of excess βγ over the α subunit in vivo for the suppression of spontaneous activity at the G protein level. Two possible mechanisms can explain how the excess of Gβ over Gqα prevents spontaneous activity. One mechanism could be through participation of the soluble pool of rhabdomeral Gβ in accelerating the hydrolysis of Gqα-GTP if spontaneous exchange occurs. This mechanism is currently under investigation. The second mechanism could be through the stabilization of Gqα-GDP, thus preventing the exchange of bound GDP for free GTP. In an insightful, theoretical paper dealing with the spontaneous activity of G proteins by using thermodynamic model simulations, it was found that the concentration of β equal to that of α is barely sufficient to suppress spontaneous activity, whereas a twofold excess of βγ over the α subunit produces a large decrease in spontaneous activity (). Altogether, our in vivo studies point to the importance of βγ subunits as principle modulators of spontaneous activity and to the relevance of this strategy in vivo. of the following strains were used: wild-type, Oregon-R (obtained from W.L. Pak, Purdue University, West Lafayette, IN); , a severe hypomorph for Gqα (obtained from C.S. Zuker, University of California, San Diego, San Diego, CA; ); β , a severe hypomorph mutant of eye-specific Gβ; and β , a less severe hypomorph mutant of eye-specific Gβ (obtained from C.S. Zuker; ). Assay for the light-dependent localization of Gβ was performed as described previously (). In short, dark-adapted flies were subjected to illumination with activating blue light (18-W white light lamp with a 1-mm–thick wide band filter [Schott BG 28; Bes Optics] 12 cm away from the flies) for various durations at 22°C. Termination was performed by moving the flies to 4°C in the dark and promptly separating the fly heads. 10 flies were used for each time point. Heads were separated from 10 flies that were dark adapted overnight (except in ) and homogenized in 1 ml isotonic homogenization buffer (20 mM Tris, pH 7.5, 120 mM KCl, 0.1 mM MgCl, 0.1 mM PMSF, and 5 mM β-mercaptoethanol). Homogenate was either directly precipitated with 5% TCA or subjected to fractionation. Membranes and cytosol fractions were separated by centrifugation (15,800 for 15 min at 4°C). The pellet was washed and centrifuged again, and the supernatants were combined. Ultracentrifugation at 150,000 for 30 min did not change the distribution of α and β subunits between the fractions. The proteins were precipitated by 5% TCA, ran on SDS-PAGE, and subjected to quantification as described in SDS-PAGE and immunoblotting. cDNA clones of Gqα and Gβ genes were obtained from the Medical Research Council UK gene service. Gqα cDNA was amplified and cloned into pQE-80 vector (QIAGEN) that contained an NH-terminal 6× His tag and was expressed in Rosetta bacterial cells (Novagen). The recombinant (His)-Gqα protein was then purified on a Ni-Sepharose column (GE Healthcare) and eluted with a 20–250-nm imidazole gradient using fast protein liquid chromatography Akta explorer (GE Healthcare). Gβ cDNA was amplified and cloned into pHis-parallel 1 (pET22) vector (obtained from P. Sheffield, University of Virginia, Charlottesville, VA) that contained an NH-terminal 6× His tag and was expressed in HMS174 bacterial cells (Novagen). Purified recombinant (His)-Gβ was extracted from inclusion bodies by applying 6 M guanidine HCl on a bacterial membrane extract that had been washed three times with 1% Triton X-100. Both proteins were ∼95% pure as determined by SDS-PAGE and Coomassie blue staining. Recombinant protein concentrations were determined spectrophotometrically by using a calculated molar extinction coefficient of 42,350 for Gqα and 60,000 for Gβ at 280 nm. Equal protein amounts that were determined by Bradford assay were loaded on the specified gel. For detection of the α or β subunits of DGq, a 10% SDS-PAGE was used. To detect both subunits (α and β) on the same gel, proteins were separated on a gradient 7.5–15% SDS-PAGE. For the detection of Gγ, a 20% SDS-PAGE with 4 M urea was used. The urea was needed for separation of the γ subunit from the β subunit. Subsequent to SDS-PAGE separation, proteins were subjected to Western blot analysis using the specified antibodies. Two different anti-Gβ polyclonal antibodies were made in rabbit as described previously (). One antibody was made against a peptide from the COOH terminus of the protein (residues 333–346), and the other was made against a peptide from the NH terminus (residues 3–13). For Gqα detection, we used anti-Gqα polyclonal antibodies that were previously made by us (), and for Gγ detection, rabbit polyclonal antibodies that were directed against the Calliphora Gγ protein were used (obtained from A. Huber, University of Karlsruhe, Karlsruhe, Germany; ). To determine the ratio between Gqα and Gβ subunits, we performed Western blot analysis using a mixture of anti-Gqα and anti-Gβ each at a 1:1,000 dilution, which is five times higher than their saturating concentration. To rule out the possibility that the Gβ excess we observed is caused by the antibodies, we repeated these experiments with the two different Gβ antibodies and obtained the same results. To further ensure that the Gβ excess over Gqα was not a result of the antibody concentrations, we repeated this procedure with a higher concentration of anti-Gqα or with a twofold dilution (1:2,000) of anti-Gβ. In all of these cases, an excess of Gβ over Gqα was observed. Immunogold EM was performed as described previously (). All sections were made from flies that were dark adapted overnight. Sections were incubated with either Gqα antibodies (dilution of 1:80) or Gβ affinity-purified antibodies (dilution of 1:20). Gβ antibodies were affinity purified by using Affi-Gel 10 gel (Bio-Rad Laboratories) according to the manufacturer's instructions for anhydrous coupling followed by elution with glycine-HCl, pH 2.5. The secondary antibody used was goat anti–rabbit conjugated to 18 nm of gold particles. ERG recordings were performed on intact flies as described previously (). Orange light (OG-590 Schott edge filter; Bes Optics) from a Xenon high pressure lamp (operating at 50 W; model LPS 220; Photon Technology International) was delivered to the compound eye by an optic fiber and was attenuated by natural density filters. The maximal luminous intensity at the eye surface was 12.5 mW/cm. M potential recordings were performed as described previously (). In brief, an adapting light of maximal intensity 20-s blue light (Schott BG-28) from the Xenon high pressure lamp was delivered 1 min before each white test stimulus (70 jouls of photographic flash light). Dissociated ommatidia were prepared from newly eclosed white-eyed adult flies (<1 h after eclosion; ) that were maintained in a 12-h dark/12-h light cycle and kept in the dark 24 h before the experiment. Whole cell patch-clamp recordings were performed as previously described (). Signals were amplified with a patch-clamp amplifier (Axopatch 200B; Axon Instruments, Inc.), sampled at 2,000 Hz, and filtered below 1,000 Hz. The bath solution contained 120 mM NaCl, 5 mM KCl, 10 mM -Tris buffer, pH 7.15, 4 mM MgSO, and 1.5 mM CaCl. The pipette solution contained 120 mM K gluconate, 2 mM MgSO, 10 mM -Tris buffer, pH 7.15, 4 mM MgATP, 0.4 mM NaGTP, and 1 mM NAD. Transillumination of the halogen light source (100 W) was used as previously described (). The orange stimulating light (Schott OG-590) was applied via a condenser lens (Carl Zeiss MicroImaging, Inc.) and was attenuated by neutral density filters.

In many cases, amplification is indispensable for the analysis of nucleic acids. Currently, nucleic acid amplification methods include but are not limited to polymerase chain reaction (PCR) (), strand-displacement amplification (SDA), nucleic acid sequence-based amplification (NASBA), rolling-circle amplification (RCA) and the Q replicase reaction. Among these methods, PCR has been the most popular due to its simplicity; however Peltier effect or metal-block-based PCR system are characterized by high thermal mass, large reaction volume and thus slow heating/cooling rates. The PCR speed can be improved by increasing the heat transfer rate or decreasing the thermal mass. With the advent of micro-electro-mechanical-systems (MEMS) technology, the development of miniaturized PCR chips becomes possible (,). The miniaturization of PCR devices offers several advantages such as short assay time, low reagent consumption and rapid heating/cooling rates, as well as great potential of integrating multiple processing modules to reduce size and power consumption. The number of publications on PCR chips has grown rapidly recently, and the articles are spread over a large number of journals. The development of PCR microchips has been discussed in recent reviews (). In this article, we will review the latest advances and future trends based on literature published since January 2005. In addition, we will also discuss some practical issues related to the development of PCR chips. As a supplement to this review, the reader may wish to refer to several reviews of general microfluidic technologies (). The organization of this article is as follows. First, several important topics on the microfluidic PCR chips will be presented. Those topics, which are crucial in the development of PCR chips, include chip substrates and surface treatments, PCR chip architecture, on-chip PCR reaction volume and reaction speed and approaches to eliminating cross-contamination. Then, the temperature and fluidic controls and measurements in PCR chips are discussed, which include thermal insulation, evaporation and gas-bubble formation and measures to counteract these phenomena, semi-invasive or noninvasive temperature and fluidic measurements and numerical simulation of temperature and fluid fields in PCR chips. Finally, product detection methods used in PCR chips, e.g. off-line and on-line detection, are covered, followed by integration of functional components in PCR chips, biological samples used in PCR chips and potential applications of PCR chips, as well as practical issues related to the development of PCR chips. Most PCR microchambers or microchannels are fabricated from silicon () or glass () substrate. Polymers, such as polydimethylsiloxane (PDMS) (), polycarbonate (PC) () and polymethylmethacrylate (PMMA) () have increasingly been utilized as alternative substrates. New substrates, such as SU-8 (), cyclic olefin copolymer (COC) (), Gene Frame® (), perfluoroalkoxy-modified polytetrafluoroethylene (PFA) (,), LiNbO () and 317 stainless steel (), have also been used in PCR microfluidic devices. Each substrate has different properties and therefore different advantages and disadvantages. The superior thermal conductivity of silicon makes rapid PCR cycling possible. Silicon fabrication processes are well developed, and thus precise and complex chip structures can be achieved (). However, silicon can be problematic: bare silicon inhibits PCR; its high thermal conductivity requires thermal insulation and therefore results in structural complexity (,,,,); its opacity limits optical detection; and its electrical conductivity makes it difficult to combine micro PCR with micro capillary electrophoresis (CE) (,,,,) on a single silicon chip. Transparent glass is suitable for optical detection. The electro-osmotic-flow (EOF) property of glass allows the integration of PCR and CE on a monolithic chip (,,,,). However, the PCR chips made from silicon or glass cannot be disposed due to the high cost of fabrication. The use of polymers as substrates may overcome these disadvantages. PDMS, an inexpensive elastomeric polymer, has emerged as a promising substrate. It exhibits high flexibility, better optical transparency, lower cost of fabrication and better biocompatibility than silicon. The flexibility of the polymers gives rise to the highly integrated PCR chips incorporating PDMS micropumps and/or microvalves (,,,,,,,,,,). PDMS adsorbs less PCR sample than other materials and thus can be used to coat glass surfaces to avoid the possible inhibitory effect of glass on PCR (,,). However, the permeability of PDMS can result in diffusional loss of biological sample, and implantation of polyethylene (PE) vapor barrier in the PDMS has been proposed to prevent this fluid loss (). Air bubbles can be formed on PDMS surface during sample loading due to its hydrophobicity (). Other chip substrates include PC, which has a high glass transition temperature () (∼150°C) and can withstand the sustained high temperatures associated with PCR or ligation detection reaction (LDR) (,). Biocompatible PMMA is another suitable candidate. It displays less autofluorescence than PC () and is suitable for conventional optical detection. Inexpensive and versatile PMMA PCR chips can be fabricated by CO laser ablation within minutes (,). In addition, the nonspecific adsorption between PMMA and DNA/protein is minimal. However, its low (∼105°C) () may hinder widespread use. The PCR efficiency is often limited by interactions between the chip surface and the biomolecules in the PCR solution, primarily due to the increase of the surface-to-volume ratio in a micro-scale environment. In general, the hydrophilic PCR solution is not easily introduced into the hydrophobic chip (). Therefore, a proper surface treatment is required to ensure the success of on-chip PCR. The treatment processes can be classified as static treatment and dynamic treatment. The static treatment involves the pre-coating of chip surface, during fabrication of PCR chip or immediately before use, with one of the following substances: SiO (,,,,), bovine serum albumin (BSA) (,,,,,,), polyethylene glycol (PEG) (,,) or silanizing agents (for example 3-glycidoxypropyl trimethoxysilane (), dichlorodimethylsilane (,), Sigmacoat® (,) or trimethylchlorosilane (). The SiO pre-coating is a reproducible and inexpensive standard MEMS process and can be accomplished in a batch fashion. BSA and other agents become popular due to their simplicity but the reproducibility is argued. It is worth noting that diacrylated PEG (DAPEG) can be grafted into PDMS polymer for polyelectrolyte multilayers (PEM) deposition (), and that the surface characteristic of PDMS polymer shifts from hydrophobic to hydrophilic after oxygen plasma treatment (,). The oxygen plasma activation on the surface of PDMS can be performed at the same time with the bonding processes, and thus it is a standard MEMS process with high reproducibility and low costs. Silanization is also a commonly used process to prevent on-chip adsorption of biomolecules, but it is time-consuming and labor intensive. Moreover, its reproducibility is also problematic. Recently, Legendre . developed a rapid (<5 min) and simple (one-step) Sigmacoat® silanization method on a dry glass surface without extensive rinsing and cleaning steps often associated with other chip coating procedures (). This dynamic treatment occurs during the actual operation of PCR chips because the reagents are in the PCR solution. Other widely used dynamic reagents are BSA (,,,,,,,,,,,,,,,,,) and polyvinylpyrrolidone (PVP) (,,). Glycerol (,), formamide (,), polyethylene glycol (PEG) 8000 (), and Tween 20 (), have also been studied as dynamic coating agents. It is known that BSA can improve on-chip PCR efficiency because it can compete with DNA polymerase for active adsorption sites on the inner surface. However, it is not yet clear that how components such as (NH)SO, MgCl and/or KCl in the PCR mixture affect the BSA coating. In addition, when the fluorescence detection technique is applied, possible interactions between BSA and fluorescent probe/dye may be problematic. In this format of PCR chip, the PCR solution is kept stationary and the temperature of the reaction chamber is cycled between different temperatures. The first PCR chip developed by Northrup . in 1993 was based on such a stationary chamber (). Since then, this format has been widely replicated and improved (). During the past two years, the miniaturization of such PCR devices has been of growing interest. In general, they can be classified as single chamber stationary PCR chip (A) (,,,,,,,,,,,,,,,,,,,,,) and multi chamber stationary PCR chip (B) (,,,,,,,,,,,,). The former can perform very well in terms of fluidic and thermal controls and offer beneficial properties such as reduced thermal and fluidic crosstalk between chambers. However, they cannot realize high-throughput and cannot readily be used for special purposes such as single cell gene expression analysis. To circumvent these issues, great efforts have been made to develop multi-chamber stationary PCR. This type of PCR chip allows generation of standard curves, use of multiple primer sets and optimization of microfluidic PCR. However, special care must be taken to achieve the thermal optimization of chamber array in order to obtain homogeneous temperature fields between chambers. In addition, precise handling and processing of sample microfluid on such PCR chips still face challenges (,). Chamber-based PCR chips are constructed as closed systems where the reaction chamber(s) are actually microfabricated on the chip. Their design does not differ significantly from the model presented by Northrup . (). Recently, Guttenberg . proposed a completely different approach to handle small-scale sample using a hydrophobic/oleophobic surface to provide virtual fluid confinement (). The concept of virtual reaction chamber (VRC) (C) was applied for the first time on the PCR chip in 2005, in which each PCR sample droplet was covered with a drop of mineral oil. Neuzil . further optimized this approach by placing a disposable microscope glass cover slip on top of a silicon chip to eliminate cross-contamination between samples (,). The continuous-flow PCR concept of using a capillary passing through different temperature baths was introduced by Nakano . in 1994 (). In 1998, Kopp . reported a continuous-flow PCR chip using a serpentine channel passing through three thermostable copper blocks (A) (). Since then, especially during the last two years, such PCR chips have undergone substantial improvements (,,,,,,,,,,). However, continuous-flow PCR poses several problems during thermal cycling: (i) Gas bubbles are easily generated in the microchannels, which adversely affect PCR amplification. (ii) Pressure-driven flow easily produces a hyperbolic flow profile that may lead to progressive sample dispersion (), and often requires an external bulky syringe pump (,,,,,,,,,,,,), which adversely affect the development of compact, portable and integrated continuous-flow PCR chips. (iii) The rate at which the PCR solution travels between different temperature zones is difficult to regulate, and thus most of the continuous-flow PCR systems lack the flexibility to regionally control fluid flow velocity so as to meet different PCR requirements. (iv) Highly integrated continuous-flow PCR chips are rarely reported due to high fabrication cost and difficulties in controlling the continuous liquid flow. In order to overcome these problems, some alternative approaches have been utilized. For example, by introducing a highly viscous fluorinated oil cap immediately before the introduction of the PCR sample, Nakayama . have overcome the generation of air bubbles in the continuous-flow microchannel (). Gui and Ren have proposed the possibility of performing a novel EOF-driven continuous-flow PCR on a serpentine channel chip on the basis of numeric simulation (). Similar work was also reported by Chen . (). Li . have reported a continuous-flow PCR chip with a serpentine channel of varying width for ‘regional velocity control’ (). Most recently, a hybrid chip incorporating the continuous-flow PCR, LDR and hybridization assays has been proposed by Hashimoto . to perform analysis of single base mutations in genomic DNA (). This is a significant work because it opens the door to the integration of the continuous-flow PCR with pre-PCR or post-PCR analytical step. Several new approaches have been developed to partially or fully circumvent these problems. The roles of spiral-channel-based continuous-flow PCR devices (B) have been recently emphasized by several groups since this approach can effectively avoid the possible formation of the DNA double strands (,,,,,,,). But it is still hard to perform parallel PCR amplifications using such devices. An oscillatory-flow-based approach (C) (,,) not only combines the cycling flexibility of the stationary chamber PCR with quick temperature transitions associated with the continuous-flow PCR, but also provides the possibility of performing high-throughput PCR amplifications in a parallel format (). Microfluidic digital PCR (,,,) represents another example of the power of microfluidic PCR chips. This technology can provide precise control of sample volume and high-throughput analysis of serial (,,,) or parallel (,) format. In the serial format, the PCR solutions can flow continuously through a reaction channel path as a ‘droplet train’. However, such systems easily suffer from cross-contamination between samples and sample dispersion, and consequently appropriate two-phase flow systems are often required (,). In spite of this shortcoming, it is still possible to implement the totally automated, contamination-free, reusable and robust microfluidic digital systems for high-throughput PCR (). Recently, Quake and colleagues reported another type of microfluidic digital PCR, where digital PCR was performed in parallel microarray format (,). The array chips used are commercially available, thus allowing single usage to effectively prevent carryover contamination. Importantly, the micropumps and microvalves not only can be conveniently used to distribute PCR fluid into a number of isolated reaction chambers for high-throughput PCR, but can also effectively seal the individual reaction chambers so as to realize a cross-contamination free microfluidic digital PCR system. In addition, for these microfluidic digital PCR systems, quantification relies only on binary, positive/negative calls for each subreaction within the partitioned analyte. This affords an absolute readout of DNA copy number with single-molecule resolution (), allowing for transcription factor profiling in individual hematopoietic progenitors () and multigene analysis of individual environmental bacteria (). Of course, such microfluidic digital PCR can also be used for other potential applications such as detection of base substitution mutations, chromosomal translocations, alternatively spliced products. It is also useful for allelic discrimination and detection of allelic imbalance (). One of the motivations for the development of on-chip PCR is to process a sample of small volume. Based on the reaction volumes of the conventional PCR, we define small volume as ≤3 μl. Within the last two years many small-volume PCRs have been performed (see ). Quake's group performed 72 parallel reverse transcription (RT)-PCRs with a volume as low as 0.45 nl (). An array of 3072 real-time, 33-nl RT-PCRs was recently reported by Morrison . (). Matsubara . performed PCR in 40-nl reaction chambers (). Neuzil . reported a VRC-based PCR system that could amplify a PCR sample of 100 nl covered by 1.0 μl of mineral oil (). Similarly, Guttenberg . amplified a PCR sample of 200 nl covered by 5 μl of mineral oil to form a VRC (). Landers's group developed a glass/PDMS hybrid PCR-CE chip with on-chip pressure injection using elastomeric valves and the product of 278 bp could be amplified within a PCR chamber of 280 nl (). Recently, this group also reported a solid-phase extraction (SPE)-PCR-CE integrated glass/PDMS hybrid chip with a 550-nl reaction chamber (). However, large-volume (>3 μl) on-chip PCRs are still in use (see ). For example, Lee's group developed several PCR chip devices with different functions and volumes of 10 μl or more (,,,,,). A PCR chip device proposed by Shen . had a reaction volume of 25 μl (). When working with samples containing a very low concentration of the target (e.g. in diagnostics), the large-volume PCR has several obvious advantages over small-volume PCR: (i) Evaporative loss of sample during thermal cycling on PCR amplification may be negligible. (ii) The large-volume PCR allows the routine detection techniques such as gel electrophoresis to be used for the analysis of PCR products. (iii) The large-volume PCR sampling is accomplished using a conventional manual sampling gun, making large-volume sample handling during reaction set up relatively easy. However, the large PCR reaction volume can be disadvantageous for low-molecule PCR amplification, especially for single-molecule or single-cell analysis. On the contrary, the submicroliter, nanoliter or picoliter PCR systems are robust in performing analysis with single molecule or cell sensitivity (,,,,,,,). However, as the sample volume is decreased, evaporation of sample solution and introduction of a small amount of solution into the reaction chamber/channel can be the major drawbacks for PCR analysis. In addition, as PCR volumes are decreased, amplification is increasingly prone to biochemical surface absorption problems at the chamber/channel walls due to the increasing surface-to-volume ratio. Small-volume PCR should not only reduce the costs but also allow rapid thermal cycling. However, the use of small-volume PCR does not necessarily reduce amplification time if the heating method does not fully use the advantages of small-volume reaction system (,,,). On the contrary, the large-volume PCR can result in the reduction of amplification time when a better mode of heating is utilized (,,,,,). As yet, there is no definitive correlation between PCR time and PCR volume. PCR speed ultimately depends on the thermal mass of the entire PCR chip system. Although changeable in a wider range, the sample volume of continuous-flow PCR is usually <10 μl (see ). Pursuing high-speed PCR is one of the major motivations in the development of on-chip PCR. The acquisition of high-speeds for PCR depends on reduction of the thermal mass of the entire PCR system. The basic methods used to decrease the system's thermal mass are to choose a better heating method and to adopt a desirable chip architecture. For chamber-based stationary PCR chips, thermal cycling can be performed either with contact heating methods or noncontact heating methods. The former is defined as having heaters fabricated within the chip or in thermal contact with the outside of the chip, where the thermal mass of the microchip is in contact with the heating element. The latter uses a heating method that is not in physical contact with the PCR chamber (). Among the contact heating methods, the MEMS-based film heating elements have smaller thermal mass, faster thermal response and higher heating rates (e.g. >10°C/s) (,,,,,,,,,,,,,,) (see ). To date, the fastest heating rate (175°C/s) and cooling rate (125°C/s) have been obtained using this method (). However, since these heating elements are usually fabricated in a complicated process, the costs are considerably higher than other designs. To reduce costs, the temperature control chip can be physically separated from the PCR reaction chip and reused after initial temperature calibration (,,,,). To develop a cost-effective heating element, interest was focused on the commercially available flexible thin film heaters with a heating rate of 6.5°C/s or higher (,,,,). Alternatively, contact heating is also often realized by a Peltier device (,,,,,). Although its high thermal mass weakens the thermal response of the entire PCR chip, ∼5°C/s heating rates can be acquired and thus the system performances are better than most of Peltier effect or metal block-based PCR machines. A disadvantage associated with the contact heating is that a certain amount of thermal mass is added in the PCR chip assembly, which inevitably hinders fast thermal transitions. Moreover, when PCR and analytical function (e.g. CE) are integrated on a single chip, it is very difficult to confine the contact heating to the PCR chip itself and not analysis part of the chip. In order to overcome these issues, interest in noncontact heating continues to grow (). Recently, Landers's group successfully realized the integration of noncontact infrared (IR)-mediated PCR with CE separation () or with SPE and CE separation () on a single glass chip. Hu . proposed a new method to control PCR thermal cycling using an alternating-electric-current induced buffer Joule heating effect without an external heater component (). Although this approach obtained a low heating rate of 3°C/s and cooling rate of 2°C/s, there is still much room to improve the thermal response of Joule heating if the total thermal mass of the chip can be reduced. As described above, the heat inertia of a continuous-flow PCR system is considered as only the sample thermal mass, and the temperature transition time depends only on the sample flow rate and its time to reach thermal equilibrium. Consequently, the speed of continuous-flow PCR is limited only by the synthesis rate of the DNA polymerase. For example, in the 30-cycle continuous-flow PCR chip proposed by Kim ., only 8–30 min were required to produce a detectable amount of 430-bp PCR products (). Münchow . reported that 372-bp PCR products could be achieved within 5 min by over 40 thermal cycles in the oscillatory-flow PCR chip (). The fastest PCR reported to date was obtained on the continuous-flow PCR chip developed by Soper's group — amplification of a 500 bp λ-DNA fragment in 1.7 min and a 997-bp fragment in 3.2 min, respectively (). italic list #text Thermal interaction or ‘crosstalk’ has emerged as an important issue as chip device size decreases. Most DNA-based assays [e.g. PCR, restriction endonuclease reaction, temperature gradient gel electrophoresis (TGGE) and PCR-sequence-specific oligonucleotide polymorphism (SSOP)] are highly temperature sensitive and require precise temperature control. When integrating these analytical components on a single chip, thermal crosstalk will deteriorate chip device's performance and the thermal insulation is often required. The thermal isolation should also be used to eliminate or weaken the thermal crosstalk between chambers within the PCR array chip because PCR conditions vary slightly from one gene target to another although DNA hybridization occurs at the same temperature for many different gene targets. In addition, the thermal isolation between the temperature zones and the substrate is also usually considered to prevent heat loss to the surroundings. For these reasons, there have been increasing attempts to develop thermal isolation solutions on PCR chips. Shih . recently proposed a novel technology using parylene-cross-linking structure to achieve air gap thermal isolation for on-chip continuous-flow PCR (). This technology provides excellent thermal isolation efficiency. Its simplicity of integration with other analytical components also makes applications of micro total analysis systems (μTAS) feasible. Burn's group reported in detail two cost-effective thermal isolation techniques: the thermal conduit technique based on a selective conduction mechanism (,), and the silicon back-dicing technique based on a selective insulation mechanism (). They are inexpensive alternatives to the silicon back etching technique. In addition, most existing thermal isolation techniques also adopt thin substrate structures (e.g. cantilever beam (,) or deep trenches (,,,)) to thermally insulate the PCR region. These structures can provide excellent thermal isolation due to their high thermal resistance, but they usually have the low mechanical stability and require complicated microfabrication processes. Recently, Zou . proposed a simple conductive polymer flip-chip bonding technique to accomplish the thermal isolation of multi-chip array to allow the PCR chambers to be thermally and independently controlled (). Sample evaporation is often problematic because PCR volumes are usually very small. Especially as denaturation temperatures approach 100°C the sample evaporation is so rapid that the sample would dry up quickly under standard atmospheric pressure. To circumvent such evaporation, a number of measures can be taken, but all have their respective advantages and disadvantages. A mineral oil cover layer is frequently used as a vapor barrier to prevent evaporation (,,,,,,,). The mineral oil is a suitable liquid cover because it has a boiling point far above 100°C and a density slightly below 1.0 g/cm. However, its applicability is questionable for highly integrated PCR systems. Another approach is using a solid cover or valve to resist the internal pressure generated during PCR (,,,,,,,,). It is known that the evaporation rate decreases with the increase of the gas pressure around a liquid. Recently, Cheng . extended this concept to the oscillating-flow PCR chip (). In their approach, a single opening serves for both sample loading and syringe pump port. When the sample plug is pumped to high-temperature zones, the internal pressure increases by six times and thus the sample evaporation is greatly reduced. Noteworthy is that within the continuous-flow PCR chip, the relative sample evaporation loss can be decreased because of the decrease in the free surface area of the liquid. In addition, sample evaporation is also affected by the PCR chip substrates. For example, the water diffusion/vapor loss property of PDMS could lead to the sample evaporation loss and thus special approaches such as vapor barrier (,) should be considered to reduce the sample loss at elevated temperatures. An important drawback of the PCR chip microsystems is the generation of air bubbles. which not only cause large temperature difference in the sample but also expel the sample from the PCR chamber. The formation of air bubbles has two prerequisites: the liquid must be superheated and there must be nucleation site(s). Several methods have been used to avoid the prerequisites and inhibit the bubble generation: (i) The structural design of PCR chamber. A diamond-shaped or rhomboidal chamber is superior to a circular chamber in preventing bubble formation (,). Recently, Gong . reported that the deeper the PCR chamber, the more difficult it is for the PCR solution to flow into the chamber without trapping bubbles. However, the size of the chamber or the shape and size of the inlet and outlet have little or no influence on the bubble formation (). (ii) The surface treatment of the PCR chamber. In general, the wetting properties of the PCR chamber and its inlet/outlet have an obvious effect on the bubble formation. When the chamber surface is highly hydrophilic, the PCR sample can flow into the chamber smoothly and rapidly without bubble formation (,,,). (iii) The sealing pressurization of the PCR chamber. Under pressurization and high-temperature, the gas solubility will increase and the dissolved gases and microbubbles in the PCR sample cannot grow up in volume, thus preventing the air bubble formation (,,,,). (iv) Degasification of the PCR sample. This process can eliminate non-condensable gases in the PCR sample before loading and consequently decrease the risk of bubble formation (). (v) The addition of high boiling-point biocompatible reagents to the PCR sample. When a solvent with a boiling point above 100°C (e.g. glycol, glycerol or poly(ethylene glycol)) is included in the PCR sample, the boiling point of the resulting sample is increased, and thus preventing bubble formation at high temperatures (). Since the PCR is temperature sensitive, it is essential to choose a proper temperature measurement technique to determine or at least to estimate the temperature within the PCR chip. In general, the measurement techniques can be classified into three categories based on the nature of the contact mode between the temperature measuring element and the PCR chip (solid) or PCR solution (liquid): (i) Invasive. The measuring element is in direct contact with the PCR chip or solution. (ii) Semi-invasive. The PCR solution is treated in some manner to enable remote observation, e.g. inclusion of dye whose color changes with temperature. (iii) Noninvasive. The PCR solution is observed remotely, e.g. IR thermography. Due to their low cost and convenience, invasive temperature measurements are still most widely used (). However, there are several obvious challenges associated with these measuring techniques: First, regardless of the nature of the temperature sensing element, it will add some thermal mass to the PCR chip, ultimately decreasing the chamber PCR thermal cycling rates. Second, they can obtain temperature data only at a few discrete points or lines and thus do not reflect the overall temperature field of the PCR chip or solution. Third, the temperature sensor within the PCR solution will inhibit the PCR and increase the risk of sample cross contamination. Fourth, the use of invasive temperature measurement involves a disturbance, which manifests itself as a difference between the temperature being measured and that which would exist in the absence of the temperature-measuring element. To address these issues, attention has been given to the semi-invasive or noninvasive temperature (and fluidic) measurement techniques, which are usually accomplished by the optical techniques. In these techniques, temperature-sensitive materials such as thermochromic liquid crystals (TLCs) or fluorescence dye indicators are included in the PCR solution, variations in optical properties can be observed remotely. These techniques are categorized as semi-invasive since they involve a modification of the component in the solution and therefore could cause disturbances to the temperature field. Most of the noninvasive techniques measure temperature using the electromagnetic spectrum. For example, IR devices are sensitive to the spectrum in the infrared region. Optical techniques such as absorption and emission spectroscopy are sensitive in the visible region. It is essential to know the flow state of the micro fluids in the PCR chip since results depend on the fluid distribution and variation in the reaction system. Less attention has been devoted to the fluid mechanics of micro PCR devices than to other aspects. One notable advance has been to use semi-invasive fluid measurement techniques to measure flow fields inside a continuous-flow PCR chip. Curtin . used the micro particle image velocimetry (μ-PIV) and pressure measurements to study the effect of PCR on biofluid viscosity in continuous-flow PCR chips (). It is reported that for low molecular weight substances, the biofluid viscosity will not increase after PCR. Li . used μ-PIV to measure the velocity fields inside the continuous-flow PCR microchannel at various downstream locations (). Using μ-PIV in the PCR chips represents a novel application of existing measurement techniques. For this technique, the size choice of the fluorescence seed particle is essential, as a particle that is too large or too small will cause undesirable effects on flow field. Due to the small dimensions of the PCR chips, the direct temperature/fluid measurement remains a challenge. Moreover, direct measurement cannot provide any static and/or dynamic information on temperature/fluid before micromachining of PCR chips. In order to address this issue, numerical simulation is thought to be a very effective approach and can provide the opportunity to evaluate flow, thermal and even chemical processes during PCR (). Within the past two years, this approach has been widely used on the chamber-based stationary PCR chips (,,,,,,,,,). Meanwhile, a great attention has been focused on the numerical simulation of the continuous-flow PCR chips (,,,,,,). For example, Gui and Ren developed a 3D model to simulate the electrical potential field, flow field and temperature field in an electro-osmosis-based continuous-flow PCR chip (). Tsai and Sue numerically simulated the flow field at the continuous-flow channel turning corners, as well as the performance of designed heaters (). Wang . described a 2D numerical model to study the effects of chip geometries, materials, heater temperatures, flow rates and boundary conditions on the thermal performance of the continuous-flow PCR chip (). Commercial software often used in the numerical simulation include ANSYS (,,,,,,), CFD-RC (), CFD-ACE (+) (,,), FLUENT (,), FLOTHERM (), CoventorWare (,,,) and COSMOS (). ConventorWare (IMAG Inc.) comprises a simulation module devoted to the microfluidics, and thus can simplify the development of the microfluidic mathematical model and accelerate the study of the microfluidics theory. FLOTHERM (Flometrics Inc.) is a thermal analysis software tool developed for the electronics industry and it has been widely used for the thermal modeling of microfluidic systems. All existing DNA detection methods can be used for off-line detection of on-chip PCR products. The use of intercalators in combination with gel electrophoresis is the most widely used method for the detection of post-PCR products (,,,,,,,,,,,,,,,,,). In this combination, the DNA molecules are effectively labeled with an intercalating dye and subsequently separated according to their sizes. Upon binding to double-stranded DNA, the intercalator molecules exhibit significant enhancement in their fluorescence quantum efficiencies. Ethidium bromide (EtBr) and SYBR Green I are the most popular intercalator dyes, but other intercalating dyes, such as GoldView™, have also been investigated (). The use of intercalator to detect DNA molecules has two main advantages: real-time detection (,,,,) and versatility. However, indiscriminate binding is also a major disadvantage: both specific and nonspecific PCR products can produce the same type of signal and it is difficult to differentiate between them. In addition, this technique is time-consuming and labor intensive. CE is another widely used off-line detection technique (). The use of CE for DNA separation detection has several obvious advantages including low operating costs, high separation efficiency, small sample volume, short analysis time, versatility and simplicity. Incorporation of CE on a chip fully utilizes these advantages. CE chips were first demonstrated in 1990 by Manz . () and have now become commercially available. One notable example is the Agilent 2100 Bioanalyzer (Agilent Technologies). Recently, this commercial CE chip has been used for off-line detection of PCR products (,,,,,). Other microchip electrophoresis systems include the Hitachi SV 12-channel electrophoresis microchip (Hitachi Electronics Co.) (), LabChip® 90 Automated Electrophoresis System (Caliper Lifesciences) () and CE chips with laser-induced fluorescence (LIF) detection (,,,). The off-line CE chip platform can offer walkaway and unattended analysis of DNA molecules, eliminate the time-consuming and messy slab gel process, generate much more reproducible and high quality data, and allow high-throughput laboratory analysis; however, the manual sample loading may increase the risk of cross contamination and the total analysis time. In order to circumvent the drawbacks of off-line DNA detection methods, great efforts have been made to develop on-line DNA detection methods. Some of the on-line detection methods described below may soon be routinely implemented for on-chip quantitative and/or qualitative PCR detection. One of the main challenges in miniaturization of PCR is the integration of functional components to perform several operations without the need of external macro apparatus or manual operation. With the advance of the PCR chips, more and more on-chip PCR systems have been developed. Lien reported a device which could accommodate the following operations: virus loading, virus capture, virus purification, cell lysis, plus RT and stationary PCR (). In addition, micropumps, microvalves, micromixers, microcoils, film-resistive heaters and film temperature sensors were integrated on the chip. Another system developed by Hashimoto . possessed following capabilities: continuous-flow PCR, LDR and hybridization (). Other integrated PCR chips have also been investigated (see ). The stationary chamber PCR seems to be preferred for integrated systems when compared to the continuous-flow PCR system. This might be due to an increased simplicity in sample handling and structure designing for the former system. The most commonly integrated physical elements are film-resistive heaters, film-temperature sensors, micropumps and microvalves (see ). Although the partially integrated PCR chips have been successfully developed, the PCR-based ‘complete’ lab-on-a-chip still requires further development. The bottlenecks blocking the realization of a truly and highly integrated PCR chip include on-chip pre-PCR sample preparation and on-chip PCR product detection. Since the source of raw template samples is varied and the sample preparation methods are diverse, the miniaturization of conventional sample preparation and functionalities on a chip remains a challenge (). As for the on-chip on-line DNA detection, the bulky optical detection systems such as charge-coupled device (CCD) and LIF are difficult to miniaturize onto a single chip. However, along with the development of MEMS technology, the optical-MEMS apparatus such as optical fiber, LED and PMT can be applied onto the PCR chip to realize a portable DNA analysis device (,). Starting DNA/cDNA samples used in PCR chips include bacterial genomic DNA (for example, () (,,,,,,), () (), () (), cyanobacterial genomic (), () (,), () (), () (), () (), () genomic (,,) DNA), viral genomic DNA (for example, human papilloma virus (HPV) (), hepatitis C virus (HCV) (,), hepatitis B virus (), human immunodeficiency virus (HIV)-1 (), Dengue II virus (), () () and severe acute respiratory syndrome (SARS) () DNA), λ phage DNA (), yeast genomic DNA (), human genomic DNA (,,,,) and others (,,,,,). For RNA analysis, cDNA are reversed transcription from the corresponding RNAs. They may include cDNA from sorted CD19+ malignant B cells (), cDNA from brain tissue from transgenic GFAP–GFP mice (), cDNA from human breast cancer cell lines T47D (), cDNA from human heart and human liver samples () and influenza viral cDNA (). In order for a PCR microfluidic chip to have value in clinical diagnostics or genetic profiling analysis, it must be capable of accepting the crude biological samples (other than purified DNAs) as an analytical target, such as cells, total RNA, virus, whole blood, urine, sperm, or nasal aspirate (see ). As seen from a wide range of biological samples being on-chip amplified and analyzed, the PCR chips could be used for broad applications including molecular diagnostics of diseases (,,,,), gene expression analysis (,,,,), forensics, environmental testing, food safety testing and biothreat sensing. For example, the PCR/LDR/hybridization chip reported by Hashimoto . has been used to detect low-abundant DNA mutations in gene fragments (K-) that carry point mutations with high diagnostic value for colorectal cancers (). The lowest mutant:wide-type ratio that could be detected by this chip was up to 1:80, and the total assay time was ∼50 min, including 18.7 min for PCR, 8.1 min for LDR, 5 min for hybridization, 10 min for washing and 2.6 min for fluorescence scanning (). Since the sample preparation processes have been integrated on a single chip, along with the robust product detection techniques, the total analysis process can be completed within less than 25 min (less than 10 min for DNA extraction, 11 min for PCR and less than 3 min for injection, CE separation and detection) (). In addition, due to high integration and miniaturization of PCR chips, the biological sample and costly reagent consumption, as well as the possible contamination resulting from manual processes can be decreased. In principle, the PCR chips can be applied in any field where minute amounts of nucleic acid sample needs to be rapidly amplified and subsequently analyzed. s p i t e g r e a t p r o g r e s s e s i n m a n y a s p e c t s o f P C R c h i p s t h a t h a v e m a d e t h e m a c e n t r a l p a r t o f μ T A S , P C R c h i p s a l s o f a c e s o m e p r a c t i c a l i s s u e s . F i r s t , d u e t o t h e c o m p l i c a t e d M E M S p r o c e s s , m a n y P C R c h i p s a r e r e l a t i v e l y e x p e n s i v e a n d t h u s n o t d i s p o s a b l e . T h e r e f o r e , c r o s s c o n t a m i n a t i o n b e t w e e n s a m p l e s i s d i f f i c u l t t o a v o i d . S e c o n d , f u l l y i n t e g r a t e d P C R c h i p s a r e d i f f i c u l t t o m a k e . M a n y P C R c h i p s c a n o n l y p e r f o r m t h e D N A / R N A a m p l i f i c a t i o n . T h e d e v e l o p m e n t o f t h e h i g h l y i n t e g r a t e d P C R c h i p s i s l i m i t e d b y c o m p l i c a t e d d e s i g n a n d f a b r i c a t i n g p r o c e s s . T h i r d , t h e p r o d u c t d e t e c t i o n m e t h o d s h a v e n o t a d v a n c e d a s r a p i d l y a s o t h e r a s p e c t s o f c h i p d e v e l o p m e n t . M o s t o f t h e P C R c h i p s s t i l l u t i l i z e t h e c o n v e n t i o n a l g e l e l e c t r o p h o r e s i s t e c h n i q u e s t o d e t e c t t h e p r o d u c t s . W i t h t h e d e c r e a s e i n P C R v o l u m e , d e t e c t i n g t h e p r o d u c t s b y t h i s t e c h n i q u e b e c o m e s a c h a l l e n g e . F o u r t h , i n t e l l e c t u a l p r o p e r t y i s s u e s , w h i c h m a y l i m i t t h e a b i l i t y t o c o m b i n e d i f f e r e n t t e c h n o l o g i e s i n a s i n g l e s y s t e m h a v e y e t t o b e f u l l y e x p l o r e d . F i n a l l y , w h e n P C R c h i p s b e c o m e w i d e l y u s e d c l i n i c a l l y , e t h i c a l i s s u e s s u c h a s g e n e t i c a l l y - a l l y - b a s e d e m p l o y m e n t d i s c r i m i n a t i o n w i l l b e c o m e m o r e u r g e n t a n d p r o t e c t i o n l e g i s l a t i o n w i l l n e e d t o b e c o n s i d e r e d . r s u r v e y o f r e c e n t l i t e r a t u r e s o n m i n i a t u r i z e d P C R c h i p s c o n c l u d e s t h a t s u c h P C R c h i p s h a v e b e e n w e l l d e v e l o p e d a n d h a v e f o u n d i m p o r t a n t a p p l i c a t i o n s i n m i n i a t u r i z a t i o n t e c h n i q u e s . T h e a u t o n o m o u s o r q u a s i - a u t o n o m o u s h i g h - s p e e d i m p l e m e n t a t i o n o f n u c l e i c a c i d a m p l i f i c a t i o n a n d a n a l y s i s i n t h e c a s e o f a s m a l l - v o l u m e b i o l o g i c a l s a m p l e w i l l p r o v i d e c o n t i n u e d i m p e t u s f o r t h e d e v e l o p m e n t a n d i m p r o v e m e n t o f m i n i a t u r i z e d P C R c h i p s . A l t h o u g h t h e P C R c h i p p r o v i d e s m a n y a d v a n t a g e s o v e r t h e c o n v e n t i o n a l P C R d e v i c e , t h e m i n i a t u r i z a t i o n a l s o r a i s e s s o m e c h a l l e n g i n g i s s u e s . T h u s , m a n y p r o b l e m - s o l v i n g s t r a t e g i e s a r e r e q u i r e d t o t a c k l e c h a l l e n g e s s u c h a s a d s o r p t i o n o f t h e r e a g e n t s t o t h e c h i p s u r f a c e , b e i n g p r o n e t o e v a p o r a t i o n o f t h e s a m p l e s o l u t i o n a n d f o r m a t i o n o f g a s b u b b l e s , t h e r e q u i r e m e n t o f p r e c i s e t e m p e r a t u r e c o n t r o l a n d t h e p a t e n t d i s p u t e . D e s p i t e t h e s e o b s t a c l e s , t h e p o t e n t i a l o f m i n i a t u r i z e d P C R , a s a f u t u r e n u c l e i c a c i d a m p l i f i c a t i o n a n d a n a l y s i s t o o l , i s s t i l l a t t r a c t i v e . A s t h e d e v e l o p m e n t o f P C R c h i p s c o n t i n u e s , n e w l y d e s i g n e d m i c r o s t r u c t u r e s s h o u l d m a k e a f u l l u s e o f t h e b e t t e r M E M S p r o c e d u r e s t o m e e t t h e r e q u i r e m e n t s o f e a c h b i o a s s a y p r o c e d u r e t o e n a b l e t h e m t o b e w i d e l y c u s t o m i z e d t o a c c o m p l i s h s p e c i f i c b i o a s s a y s . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

According to rules for double helix formation proposed by Watson and Crick in 1953, antiparallel DNA strands are held together by nucleobase pairs that obey two rules of complementarity: (large purines pair with small pyrimidines) and (hydrogen bond donors from one nucleobase pair with hydrogen bond acceptors from the other) (,). The former permits the structure that underlies faithful replication. The latter helps achieve the specificity that gives rise to the simple rules for base pairing (‘A pairs with T, and G pairs with C’) that underlie genetics and molecular biology. Some time ago, it was noticed that the DNA alphabet need not be limited to this architecture. For example, many groups, including those of Rappoport (), Kool (), Hirao . (,), Minakawa . (), Romesberg () and Schultz (), have shown that the Watson–Crick structural design can be drastically altered, removing hydrogen bonding, introducing steric determinants of specificity, or changing substantially the size of the nucleobases. To date, however, the most useful ‘expanded genetic alphabets’ have come from more subtle modifications of the Watson–Crick architecture. One class of these involves simply rearranging hydrogen bond donor and acceptor groups within a pair while retaining the overall Watson–Crick geometry () (,). By doing this, 12 nucleobases forming six base pairs joined by mutually exclusive hydrogen bonding patterns are readily available within that geometry. shows the standard and non-standard hydrogen bonding patterns obtained by this rearrangement, together with a nomenclature to designate them (). These non-standard nucleotides and the pairs that they form have had particular value as ‘orthogonal binders’, recognition elements that bind with DNA-like specificity, but without interference by natural DNA. This orthogonality substantially lowers noise in a range of nucleic acid-targeted assays. For example, non-standard nucleotides that implement the pyAAD:puDDA hydrogen bonding pattern () are used in the ‘branched DNA’ diagnostic assay developed at Chiron and Bayer. Having now FDA approval, this diagnostic helps manage the care of some 400 000 patients annually infected with the HIV, hepatitis B and hepatitis C viruses (). The binding properties of these artificially expanded genetic information systems (AEGIS) could have still greater value, however, if their components could be incorporated as parts of architectures to detect, quantitate and sequence nucleic acids. For example, expanded genetic alphabets could support architectures for highly multiplexed amplification of DNA and RNA, the movement of PCR-amplified nucleic acids to specific spots on microarrays (‘binning’) and low-cost, high capacity re-sequencing of the genomes of individual patients. For these architectures to be practical, however, the components of an expanded genetic alphabet must interact with standard DNA polymerases with sufficient efficiency that they can be copied, and their copies copied. To expand the potential application of expanded genetic alphabets in dynamic assays in molecular-, systems- and synthetic-biology, we returned to the structure of DNA polymerases. Studies in many laboratories suggested that polymerases might ‘scan’ the minor groove of a growing DNA duplex searching for the electron density that is presented by nitrogen-3 of adenine and guanine, and by the 2-position exocyclic carbonyl group of cytosine and thymine (). This is a structural feature shared by the four standard nucleobases, making it potentially a convenient ‘handle’ for a polymerase, even though it does not seem to be a mandatory specificity determinant. As can be seen by inspecting , many of the non-standard nucleotides that are pyrimidine analogs do not have a 2-position exocyclic carbonyl group. Therefore, they do not present electron density to the minor groove at this position, as do thymine and cytosine. However, one non-standard pair ‘does’ present electron density at this position by ‘both’ components; this is the pair implementing the pyDDA:puAAD hydrogen bonding pattern (). For this reason, we focused on the pyDDA:puAAD hydrogen bonding pattern in our most recent work developing dynamic DNA sequencing, detection and quantitation architectures. A heterocycle to implement the pyDDA hydrogen-bonding pattern proved to be difficult to find, however. For example, implementation of the pyDDA hydrogen-bonding pattern was first attempted on a simple pyridine heterocycle; this failed to give a species that was stable to oxidation (). The same pattern implemented on a pyrazine gave a nucleoside analog that was prone to specific acid-catalyzed epimerization (). The same pattern implemented on a pyrimidine heterocycle gave rise to tautomeric ambiguity (). Implementing the pyDDA hydrogen-bonding pattern on a nitropyridine heterocycle solved these problems, however. We recently reported that 6-amino-5-nitro-3-(1′-β--2′-deoxyribofuranosyl)-2(1H)-pyridone (trivially designated d) could implement the pyDDA hydrogen bonding pattern (). The nitro group rendered the otherwise electron-rich heterocycle stable against both oxidation and epimerization under standard conditions. When paired with the corresponding puAAD nucleotide, duplexes were formed with stabilities that, in many cases, were higher than those observed in comparable strands incorporating the d:d nucleobase pair (). We therefore developed chemistry to efficiently prepare d, together with its nucleoside complement, 2-amino-8-(1′-β- -2′-deoxyribofuranosyl)-imidazo[1,2-a]-1,3,5-triazin-4(8H)-one (implementing puAAD, trivially designated as d) (). These syntheses made d and d efficiently available as both their triphosphates and their protected phosphoramidites suitable for solid phase DNA synthesis. It also yielded their alpha-thiotriphosphates. We report here studies of the interaction between DNA polymerases and the d:d pair. Following a survey of polymerases, we found that both dZTP and dPTP are accepted by DNA polymerases representative of both Families A and B. We also showed that the d:d pair can participate in PCR amplification using DNA polymerase with >94.4% retention per round; using Vent (exo-) and Deep Vent (exo-) polymerases, 97.5% retention per round is measured. A study of the pH-dependence of mismatches suggests that the principal route for the loss of the d:d pair is via a transition to a d:d pair through a mismatch between d and protonated d (at low pH), or a mismatch between d and deprotonated d (at high pH). Here, the canonical Watson–Crick model for the nucleobase pair, which includes both size and hydrogen bonding complementarity, is adequate to explain these behaviors. Further, this level of fidelity is sufficient to allow the d:d pair to participate as a dynamic component of many architectures for multiplexed detection and sequencing of DNA. Oligonucleotides (), except those containing d and d (Z-Temp and P-Temp), were synthesized by Integrated DNA Technologies (Coralville, IA). All oligodeoxynucleotides were purified by PAGE (10–20%). Z-Temp (containing d at position 26) and P-Temp (containing d at position 26) were synthesized in-house on an Expedite-8900 DNA synthesizer employing standard β–cyanoethylphosphoramidite chemistry using the d and d protected phosphoramidites reported recently (). Other reagents were purchased from Glen Research (1 μmol scale, CPG 1000 column). The triphosphates and α-thiotriphosphates of d and d were prepared as described by Eckstein . (). The S- and R-diastereoisomers of dPTPαS were separated by preparative rp-HPLC (Nova-Pak® HR C18 Column (7.8 × 300 mm). Natural deoxynucleoside triphosphates were purchased from Promega (Madison, WI). Klenow Fragment (exo), , Vent®, Deep Vent®, 9°N, Phusion (high-fidelity DNA polymerase) and DyNAzyme™ EXT DNA polymerases were purchased from New England Biolabs (Beverly, MA). and DNA polymerases were purchased from Promega (Madison, WI). (exo), native and cloned DNA polymerases were purchased from Stratagene (La Jolla, CA). Exonuclease III was purchased from Promega (Madison, WI). They were generally used in the buffers provided by the supplier of the polymerase. Recognizing that the pH of Tris buffers, which are routinely recommended by manufacturers (for example, Thermopol buffer used for , Vent®, Deep Vent® and 9°N DNA polymerases is 20 mM Tris–HCl, pH 8.5 measured at 25°C, 10 mM KCl, 10 mM (NH)SO, 2 mM MgSO, 0.1% Triton X-100) is known to strongly vary with temperature (see ‘Results’), the pH of buffers was measured at the elevated temperatures (using a temperature calibrated Accumet® AB15 pH Meter, Fisher Scientific) used for the extension reactions and PCRs. As expected, these pHs were ca. 1.4 units below those measured in the same buffer at room temperature. Four Family A polymerases ( and , all exo) and 10 Family B polymerases [Vent and Deep Vent (both exo and exo)], (exo, native and cloned), 9°N, (exo) and Phusion (exo) were initially screened for their ability to incorporate dZTP opposite template d, and dPTP opposite template d (data not shown). In general, the incorporation of dZTP opposite template d appeared to be more facile than the incorporation of dPTP opposite template d. These experiments identified ‘polymerases of interest’ from both families, in particular, , Vent (both exo and exo), Deep Vent (both exo and exo), and 9°N. Based on our experience with other screens that sought to incorporate nucleoside variants that do ‘not’ present electron density in the minor groove, (,) finding this number of polymerases able to incorporate the d:d pair was surprising. We then sought to determine how well a d:d pair survives in duplex DNA after multiple rounds of PCR. To do this, we wished to apply a strategy that combines the incorporation of alpha-thiotriphosphates and Exo III digestion to estimate the amount of d and d in an oligonucleotide (Yang ., in press). This strategy is based on the fact that phosphorothioate linkages, incorporated into an oligonucleotide by a polymerase from the corresponding -α-thiotriphosphate, can resist hydrolysis by Exo III. To the extent that d or d is present in a template, therefore, primer extension on that template with dPTPαS or dZTPαS, respectively, will generate products containing phosphorothioate linkages at the positions where dPTPαS or dZTPαS is incorporated. Exo III digestion of these products, in turn, will give bands in a gel at positions where the dPTPαS or dZTPαS were incorporated. If that nucleotide is d, this band implies the presence of d surviving in the PCR product. If that nucleotide is d, this band implies the presence of d surviving in the PCR product. Some stereochemical features of the analysis are relevant to the interpretation of these experiments. First, a phosphorothioate linkage having an R configuration is believed to be highly resistant to Exo III cleavage. Conversely, the S configuration is believed to be sensitive to Exo III digestion, with the extent of degradation depending on the amount of Exo III used and time of the digestion. Given the reasonable assumption that polymerases invert configuration at the alpha phosphorus of the triphosphate being incorporated, S-α-thiotriphosphates should deliver the cleavage-resistant R-phosphorothioate linkage to the product oligonucleotide, while R-α-thiotriphosphates should deliver the cleavage-prone S-phosphorothioate linkage. Therefore, the analysis is expected to work best if pure S-α-thiotriphosphates are used. Unfortunately for the application of this strategy here, while the diastereomers of dPTPαS could be resolved using HPLC resolution, the diastereomers of dZTPαS could not. Also unfortunately, both and 9°N DNA polymerases were found to accept ‘both’ S and R isomers of dPTPαS (and therefore presumably dZTPαS) to some extent (Yang ., in press). This means that the use of dZTPαS in a primer extension will give some Exo III sensitive product (a phosphorothioate linkage with S configuration), which implies an underestimation of the amount of d remaining in the PCR product. The stereochemical details also are relevant to the design of an internal standard to control for these factors. To quantitate total product, a phosphorothioate linkage is incorporated into the primer by chemical synthesis. Chemical synthesis delivers phosphorothioate linkages as a ≈50:50 mixture of R and S diastereomers. Thus, only ca. 50% of the oligonucleotide will be highly resistant to degradation. This must be considered when interpreting a reference band arising from a chemically introduced phosphorothioate linkage. To implement the phosphorothioate–exonuclease combination analysis, we first examined the ‘polymerases of interest’ for their ability to incorporate the alpha-thiotriphosphates of both d and d. Both and 9°N were shown to allow primers to be fully extended past templates containing d and d with dZTPαS and dPTPαS, respectively, after incubation for 3 min by (, left). The full-length product (FLP) was then treated with Exo III (100 U) and the digestion products were resolved by gel. As shown in (right), the 9°N polymerase generates more phosphorothioate-containing product than , as determined by a greater intensity of the band at 26-mer (relative to the intensity of the 17-mer reference band). Indeed, the faintness of bands at position 26 when was used suggested that mismatched standard nucleotides opposite template d and template d rather than accept dZTPαS or dPTPαS. Less prominent features in are low molecular weight products that presumably arise following digestion past the 16–17 phosphorothioate linker that has the Exo III-sensitive stereochemistry. This is expected, as this phosphorothioate linker, introduced by chemical synthesis, is expected to be a mixture of the S and R diastereomers. These were noticeable in these experiments, where 100 U of Exo III were used; they were greatly reduced when the amount of Exo III was reduced to 20 U ( and , below). Last, extra pausing bands (larger than the 26-mers) are evident in the 9°N lanes using dZTPαS. These are positioned in a way that suggests that 9°N incorporates small amounts of dZTPαS opposite d in the template. Inspection of also suggested that dZTPαS was incorporated more efficiently than dPTPαS by both polymerases, but in particular by 9°N. This is indicated by the greater intensity of the band at position 26 relative to the intensity of the band at position 17 in the appropriately compared lanes (, right). Therefore, we decided that the incorporation of dZTPαS by 9°N is a more reliable analytical tool to detect d in a template coming from multiple rounds of PCR, than the incorporation of dPTPαS is to detect d in the PCR product. Various unknowns make the reference (the band at position 17) especially important. First, the rate at which Exo III digests DNA need not be independent of local sequence or secondary structure. For example, Linxweiler and He . (,) reported that Exo III digests through nucleotides in the order C > A,T > G. Second, dZTPαS, as presented to this assay, is a mixture of unresolvable diastereoisomers. If the polymerase accept both isomers (as indicated by a study of and 9°N for dPTPαS), both the R (Exo III resistant) and S (Exo III sensitive) phosphorothioate linkages will result. While we suspect that polymerases prefer the S isomer of dZTPαS over the R isomer, we do not know by how much. Hence, the reference band at position 17 is needed to normalize for this unknown as well. Other Family A (, all exo) and Family B polymerases (Vent, Deep Vent, and , all exo) were also examined using this assay. We found also Vent (exo) and Deep Vent (exo) were comparable to 9°N in their ability to incorporate dZTPαS opposite d in the template (data not shown). shows the strategy used to apply the phosphorothioate–exonuclease combination to determine the amount of d remaining in oligonucleotides after a specified number of PCR cycles at different pHs. The starting material was a synthetic template (P-Temp) that contained a single d at position 26 (the length of the extended primer after d is incorporated opposite d). Forward and reverse primers (P-RS and Z-RS, respectively) were designed as usual (). PCR cycles were then done with P-Template at concentrations that generated a theoretical number of cycles ranging from 3.3 to 16.6, using dZTP, dPTP and the four natural dNTPs. At this point, with all primer consumed, the duplex PCR products were separated from the polymerase and excess triphosphates. Then, dZTPαS was added to incorporate phosphorothioate linkages opposite any d remaining in the PCR products using 9°N or Vent (exo), dPTP, dNTPs, radiolabeled forward primer containing a synthetic phosphorothioate joining nucleotides 16 and 17 (P-RS-S16, 50% of the amount of PCR product) and unlabelled reverse primer. Seven additional cycles of ‘analytical primer extension’ were used to convert all of the radiolabeled primer to FLP suitable for Exo III digestion. The presence of reverse primer ensured that sufficient d-containing template was present to lead to full conversion of radiolabeled primer (P-RS-S16). The 2-fold excess of PCR product over P-RS-S16 primer also served this purpose. These products were then isolated and subjected to Exo III digestion. To determine whether the pH influenced the amount of d surviving in the PCR products after multiple rounds of PCR, the strategy shown in was applied at pH 7.5, 7.8, 8.0 and 8.5 for 16.6 theoretical rounds of PCR. The buffers were Tris–HCl (20 mM), as provided by the manufacturer for these polymerases. These must be regarded as ‘nominal pH's’, as they were measured at room temperature, and the pH of Tris-HCl buffers is well known to change as a function of temperature (,). shows the result of these experiments where polymerase was used to generate the PCR products, and Vent (exo) was used to incorporate the phosphorothioate linkage. The five lanes on the left show that the fully extended products (obtained by primer extension of the PCR amplicons obtained at various pHs) co-migrated with the product from the positive control. These products were then subjected to Exo III digestion, and the intensity of the bands indicative of a phosphorothioate linkage joining nucleotides 25 and 26 (and therefore indicative of the incorporation of alpha-thiotriphosphate of dZTP, and therefore indicative of the survival of d in the PCR products) relative to the bands arising from the phosphorothioate joining nucleotides 16 and 17 (introduced into the primer by chemical synthesis) showed a clear pH dependence, with an optimum at nominal pH between 7.8 and 8.0 (actual pH during elongation is ∼1.4 pH units lower). This is graphed in . All other polymerases of interest were then examined using a similar strategy. shows that and Deep Vent (exo) convert all of the PCR primers into the expected PCR product at pH 8.0. 9°N and Vent (exo) also effected this conversion (data not shown). The amounts of PCR products were, for the most part, the same when d and d were used as in the positive control, which incorporated only standard nucleotides in the templates and standard triphosphates; appeared to perform better than Deep Vent in this regard. These PCR products were then used as templates for the ‘analytical primer extension’ step () to introduce a phosphorothioate linkage from dZTPαS using, in this case, 9°N instead of Vent (exo) ( left). These products were then digested with Exo III (, right). (right) has three prominent features. First, with the amplicon derived from PCR using , the ratio of intensities of the bands at positions 26 and 17 (including band at position 18) was quite high, but decreased with the number of PCR cycles used to generate the amplicons. This provides direct evidence for the gradual loss of the d:d pair over multiple rounds of PCR (additional loss during the ‘analytical primer extension’ steps was normalized through the reference). A similar loss was not obvious with Vent (exo) and Deep Vent (exo). Here, the relative intensity of the bands at positions 26 and 17 (including the band at position 24) does not decrease with increasing PCR cycles as much as seen with (). This implies that both of these polymerases retain the d:d pair better than does . We then fit the data to the equation = (0.5 + fidelity/2) () where is the fraction of the original d remaining in the PCR product, and is the number of theoretical rounds of PCR. This formula correctly reflects the fact that after rounds of PCR only half of the PCR product has survived rounds of PCR; a half of this remainder has survived (−1) rounds, a half of the next remainder has survived (−2) rounds, and so on. The estimated values for fidelity per round for were 94.4%, 97.5% for Vent (exo) and 97.5% for Deep Vent (exo). Interestingly, both Vent (exo) and Deep Vent (exo) generated PCR products that gave an Exo III degradation pause band at position 28 as well as at position 26. This suggests that the amplicons contain a small amount of a phosphorothioate linkage joining nucleotides 27 and 28. This, in turn, implies the incorporation of d at position 28. This position should complement a d, however, not d. Thus, the presence of this band suggests that during the PCR, d must have been replaced by d at this position. This infidelity is not general, however, as other dG:dC pairs are not replaced; this phenomenon was not further explored. Further, for Vent (exo) and Deep Vent (exo), it appears as if the amount of d inferred at position 28 in the PCR product ‘increases’ as the number of PCR cycles increases. This implies that during PCR with these polymerases, a d:d nucleotide pair is gradually converted to a d:d nucleotide pair. Thus, the higher level of retention of the d:d pair by Vent (exo) and Deep Vent (exo) at the site where retention is desired is paralleled by a higher level of misincorporation of d and/or d opposite d and/or d, respectively. The third feature of the gel in (right) is the apparent resistance of all products to initial digestion by Exo III, giving pause bands at the length expected for a 51-mer (approximately; the bands are also doubled). These bands may arise because 9°N added a terminal N + 1 thiotriphosphate in an untemplated extension reaction, a process seen with certain polymerases (,). The fact that these bands disappear with Vent (exo), which is believed to be less prone to non-templated addition (), is consistent with this analysis. The resistance of the product to initial Exo III degradation also suggests, however, the misincorporation of dZTPαS opposite a d at position 49 or 51, possibly in the ‘analytical primer extension’ by 9°N (, right). Such misincorporation might be expected to occur most easily at the end of a template, where the nucleobase pairing is perhaps the least specific. To date, just three examples have reported where six different nucleotides have been introduced into a PCR. The extra base pair in the first example was joined by the pyDAD:puADA hydrogen bonding pattern implemented on 2,4-diaminopyrimidine and xanthine heterocycles (). This required a double mutant of HIV reverse transcriptase to achieve. As HIV reverse transcriptase is unstable to heating, it was necessary to add new enzyme after each cycle of thermal denaturation. Therefore, only five rounds of PCR were reported. The reported fidelity per round (99%) was quite high. The pyAAD:puDDA hydrogen bonding pattern implemented on 5-methylcytidine and guanosine was used in the next two examples of six-nucleotide PCR (,). The success of one of these PCRs was mitigated slightly by fact that guanine exists in two tautomeric forms (), a keto tautomer (presenting the puDDA hydrogen bonding pattern) that is complementary to 5-methylcytosine (as desired), and the enol tautomer (presenting the puDAD hydrogen bonding pattern) that is complementary to thymidine (creating the possibility of an G:T mismatch) (). Consistent with the hydrogen bonding ambiguity of guanine, significant loss of the nonstandard base pair was observed after multiple PCR cycles (). Johnson . () reported a 96% retention per round, but obtained this number by assuming that the slope of a plot of the amount of non-standard nucleotide remaining directly indicates fidelity, an assumption that overlooks the fact that much of the PCR product present in a mixture after theoretical rounds of cycling is derived from fewer than cycles. Applying a correct formula to the same data gives a fidelity of ca. 93%, a fidelity per round that is comparable to the amount of minor tautomer of guanosine present at equilibrium (ca. 10%). Sismour . () managed the tautomerism of guanine by exploiting the fact that adenine forms only two of the three canonical hydrogen bonds. Sismour . () replaced thymidine by 2-thiothymidine in a six letter PCR that included 5-methylcytidine and guanosine. This strategy assumed that the thiol group makes an unfavorable sulfur-proton interaction with the 2-hydroxyl group of the undesired (enol) tautomer of guanosine (), thereby destabilizing the guanosine:thymidine mismatch. Because adenine does not present a hydrogen bonding opportunity to the C = S unit, the 2-thioT:A match is not destabilized similarly, creating enhanced specificity. Sismour . reported substantially higher fidelity (98% per round) with 2-thiothymidine than with thymidine (a per round fidelity of 93%), permitting over 30 cycles of PCR. The principal feature (and possible disadvantage) of this strategy is that it produces PCR products that are rich in sulfur. The 6-amino-5-nitro-2(1H)-pyridone heterocycle (d, implementing the pyDDA hydrogen bonding pattern), paired with the 2-amino-imidazo[1,2-a]-1,3,5-triazin-4(8H)-one heterocycle (d, implementing the puAAD hydrogen bonding pattern), appears to successfully support a six-nucleotide PCR in a way that shares certain advantages, and avoids certain disadvantages, of these other examples. Most noticeable among the advantages is the fact that many native polymerases accept this non-standard pair; this extends even to some exo polymerases (data not shown). While it is clear that presenting electron density to the minor groove of a non-standard pair is not an absolute requirement for a non-standard nucleobase to be accepted by all polymerases, experience with many over the past decade makes it noteworthy how easily the d:d pair is accepted. This suggests that minor groove scanning, as discussed by Joyce, Steitz . (), is a feature of many polymerases that contributes to nucleoside recognition, even if it is not an absolute requirement. It is difficult to compare directly the reported fidelities of the different literature six-nucleotide PCR experiments, as different methods were used to estimate fidelity for different expanded genetic alphabets, and these methods were often specific to the non-standard nucleobase pair incorporated. In the first example (), where a 99% fidelity per cycle was reported, a polymerase was available that stopped extension when it encountered the pyDAD non-standard nucleotide. This made estimation of the amount of non-standard nucleotide remaining direct. Analysis of the loss of pyAAD after multiple rounds of PCR likewise relied on a specific chemical feature of this non-standard nucleotide, its sensitivity to acid (,). Using comparable formulas, both Johnson . () and Sismour . () arrived at comparable retention per rounds (93%) for the cytidine-guanosine pair using standard dA, dG, dC and T. As acid sensitivity was also used to measure fidelity in a system that exploited thiothymidine to suppress infidelity arising from tautomerism, these 93% values can be directly compared with the 98% fidelity observed with thiothymidine replacing T. While the phosphorothioate-exonuclease analysis tool is in principle general for all non-standard nucleotides, it has limitations that are apparent in this work. Polymerases must be found that incorporate the alpha-thiotriphosphate of the non-standard nucleotide, as well as the non-standard nucleotide itself. The results must be quantitated against a reference to manage a relatively large number of unknown parameters. It is best when the thiotriphosphate that is incorporated is also the thiotriphosphate that is easily resolved into its diastereomers. For these reasons, the 94.4%, 97.5% and 97.5% fidelities reported here for , Vent (exo) and Deep Vent (exo), respectively, are best used in comparison with each other. In particular, the analysis is adequate to support comparison across a series of closely related experiments, such as those used to determine the pH-dependence of fidelity. These pH-dependency experiments suggest that the acid-base properties of the components of the d:d pair contribute most to infidelity. The canonical Watson–Crick model (which assigns an important role to inter-strand hydrogen bonding) is able to explain the results of experiments to detect infidelity. Specifically, a route for the conversion of a d:d pair to a d:d pair may involve deprotonated d (presenting a pyDAA hydrogen bonding pattern) complementing d (presenting a puADD hydrogen bonding pattern) (). Conversely, protonated d (presenting a pyDDA hydrogen bonding pattern) can complement d (presenting a puAAD hydrogen bonding pattern). Given this model, one expects fidelity to be lowest at the extremes of pH, and highest at a pH between the pK of protonated cytidine (ca. 4.5) () and the pK of d (ca. 7.8). This model, together with the experiments reported here, suggest different conditions to achieve different goals. To achieve the highest retention of the d:d, a nominal pH of 7.8–8.0 and Vent/Deep Vent (both exo) are best. This retention comes, however, with the risk of converting d:d pairs to d:d pairs. Conversely, for optimal fidelity overall, appears to be better. A PCR with a fidelity of 94.4% with (which shows little evidence of any ability to convert d:d pairs to d:d pairs) is certainly sufficient to allow non-standard nucleotides to participate as dynamic components of multiplexed DNA and RNA sequencing, detection, quantitation and characterization tools. These are widely sought in architectures for molecular biology, systems biology and synthetic biology. Several of these are presently being developed at the Foundation for Applied Molecular Evolution and the Westheimer Institute. Other polymerases, and polymerase variants, are now being explored.

Mutations that influence pre-mRNA splicing represent a substantial proportion of gene alterations leading to Mendelian disorders (). cDNA-based mutation studies of disease genes that have a large number of introns showed that splicing mutations accounted for about half of mutated alleles (,). In contrast, estimates derived from DNA-based mutation screening designed to scan coding regions and flanking intronic sequences have generally been lower (,). As a significant fraction of mutated alleles in both recessive and dominant conditions has not been identified, and the availability of RNA samples from affected individuals and their families is often problematic, the overall contribution of intronic alterations acting at the level of pre-mRNA splicing could be substantial. In addition to single-gene disorders, DNA variants that influence splicing may modify the risk of developing complex diseases and their phenotypic manifestations, but the overall role of this variability in the pathogenesis of such conditions is only beginning to be explored (). The most common consequence of splicing mutations is skipping of one or more exons, followed by the activation of aberrant 5′ (donor) splice sites (5′ss), 3′ (acceptor) splice sites (3′ss) and full intron retention (,,). Mutation-induced aberrant splice sites found in disease genes often involve disruption of the consensus sequence of the authentic sites, while activating a cryptic splice site nearby. However, aberrant splice sites can also be generated by mutations that create splice-site consensus sequences. As described earlier (), we refer to these aberrant splice sites as cryptic and , respectively, even though the distinction between cryptic and sites may occasionally be vague, because disruption of the authentic site can also create a new splice site consensus. Cryptic 3′ss are preferentially located in exons whereas 3′ss usually reside in introns, which has been attributed to splicing signal sequences upstream of the 3′ss that are required for selection of acceptor sites, including the polypyrimidine tract (PPT) and the branch point sequence (BPS) (). In contrast to cryptic 3′ss, cryptic 5′ss have a similar frequency distribution in exons and introns and their number decreases with increasing distance from the authentic 5′ss (). The human 5′ss consensus sequence is MAG|GURAGU (M is A or C; R is purine), spanning from position −3 to position +6 relative to the exon–intron junction. This sequence is critical but often insufficient for accurate 5′ss recognition, and may require auxiliary sequences in both introns and exons. These sequences can repress or activate splicing and are referred to as splicing silencers or enhancers, respectively (). The complementarity of the 5′ss consensus to the 5′ end of U1 small nuclear RNA (snRNA) exerts a dominant effect on 5′ss selection, but auxiliary sequences may exhibit a more prominent role in selection of competing 5′ss with lower base-pairing complementarity (,). In addition, the intrinsic structural properties of the RNA molecule may hinder 5′ss availability for basal splicing factors, thus controlling splicing efficiency (). Moreover, 5′ss selection can also be influenced by the presence of sequence motifs specific for RNA-binding proteins () and by the rate at which the pre-mRNA is transcribed (). A variety of methods have been used to computationally predict the 5′ss strength and recognition, including nucleotide frequency matrices (,), machine-learning approaches and neural networks (NNs) (,) and methods employing putative base-pairing interactions of 5′ss with U1 snRNP () and interdependence between adjacent or more distant positions of the splicing consensus sequences (). Exon-prediction algorithms that take into account protein-coding information may perform better than those that rely only on signals present in the splice sites (). However, it is unknown which models best predict the localization of cryptic or 5′ss that were activated . In the present study, we compiled nucleotide sequences of cryptic and 5′ss that have been reported previously in human disease genes since the first description of disease-causing aberrant splice sites (). This resource is being made available to the public through an online retrieval and submission tool termed DBASS5 (ataase of berrant ′ plice ites). In addition, we provide a detailed characterization of the underlying mutation pattern, a comparison of the nucleotide composition of aberrant and corresponding authentic 5′ss, and we evaluate the performance of computational tools that predict their utilization. Aberrant 5′ss were identified by searching home pages of peer-reviewed journals and PubMed (). They were included in the database and selected for further analysis if (i) they resulted from disease-causing or -predisposing mutations or variants in human genes; (ii) aberrant RNA products spliced to new 5′ss were verified by nucleotide sequencing; and (iii) their sequences or reliable identifiers were published in peer-reviewed journals between 1981 and January 2007. We also included 22 cases of aberrant 5′ss that were confirmed by minigene assays with wild-type and mutated reporter constructs transfected into mammalian cells, but from which patients’ RNA samples were not available. These criteria were similar to those used for a recently published analysis of aberrant 3′ss (). Aberrant 5′ss were manually validated by mapping the information in the literature to sequences in the Human Genome Project databases. Nucleotide sequences of authentic, mutated and aberrant 5′ and 3′ss are available online in the Database of Aberrant Splice Sites , which consists of the recently described DBASS3 () and the newly developed DBASS5. Validated sequences of aberrant and corresponding authentic 5′ss were used as input files for seven publicly available splice-site prediction algorithms. The Shapiro and Senapathy (S&S) matrix is based on nucleotide frequencies of 5′ss and assumes independence between individual positions of the 9-nt consensus (,). The S&S matrix scores were computed using an online tool available at . To take into account known dependencies between adjacent and non-adjacent positions of the 5′ss consensus, the compiled sequences were analysed using the first-order Markov model (MM) and the maximum entropy (ME) model (). The former method considers dependencies between adjacent positions, whereas the latter model approximates short-sequence motif distributions with the ME distribution and may include dependencies between non-adjacent as well as adjacent positions. The maximum dependence decomposition model (MDD) is a decision-tree approach that accentuates the strongest dependencies in the early branches of the tree (). The MM, ME, MDD and weight-matrix (WMM) scores, which extract single nucleotide probabilities for each position from a training set (), were computed using online tools at . The HBond algorithm, which analyses individual hydrogen-bonding patterns to the U1 snRNA 5′ end irrespective of nucleotide frequencies and assumes that the threshold values for U1 snRNP binding are influenced by specific SR proteins () was computed using a web application available at . The NN algorithm is a machine-learning approach that recognizes sequence patterns once it is trained with DNA sequences encompassing authentic splice sites (). We used the NN splice site predictor NNSPLICE (v. 0.9) at . The free energy (Δ) of predicted 5′ss/U1 base-pairing was computed using OligoArrayAux (), which is available at . Finally, the number of H bonds (#H) between 5′ss and U1 was computed using a web tool at . To compare the strength of aberrant or authentic 5′ss with a large number of human 5′ss, we used the sequences of 8415 5′ss reported previously (). The non-parametric Wilcoxon–Mann–Whitney rank test (Stat-200, v. 2.01, Biosoft Ltd., UK) was employed to test the significance of score differences between authentic and aberrant 5′ss in each category. DBASS5 (ataase of berrant ′ plice ites) is an online retrieval and submission tool for mutation-induced aberrant 5′ss available at , complementing a recently described sister database of aberrant 3′ss, termed DBASS3 (). The web application was created using the Microsoft ASP and ASP.Net server technology (), and Microsoft SQL Server database software (). In addition to aberrant 5′ss induced by disease-associated germline and somatic mutations, DBASS5 contains naturally occurring DNA variants that were shown to modify both the relative expression of RNA products spliced to alternative 5′ss and the disease predisposition. Polymorphisms that control exon skipping levels or full intron retention events have not been included in DBASS5. An exhaustive search for previously published aberrant 5′ss identified 305 unique aberrant 5′ss in 166 genes (). They were generated by a total of 26 deletions/duplications, 3 insertions, 2 complex alterations and 315 point mutations (). These alterations were described in a total of 264 publications. The number of reported cryptic 5′ss was almost three times higher than the number of 5′ss (). Cryptic 5′ss were usually activated by single-nucleotide substitutions of guanosine (G) residues, which were ∼3-times more common than mutations of the remaining nucleotides (177 versus 57, < 10, ). Conversely, substituting adenosines accounted for almost every other point mutation. Among single-nucleotide substitutions leading to 5′ss, cytosine was the most frequently mutated nucleotide (32/81, 40%). In contrast, no 5′ss have thus far been reported to be created by a point mutation introducing cytosine (). The overall distribution of unique point mutations within the 9-nt consensus sequence was highly non-random both for cryptic and 5′ss (). For cryptic 5′ss, point mutations were most common at the highly conserved position +1 relative to the natural intron/exon junctions (39.4%). Interestingly, the second most frequently mutated position was the fifth intron nucleotide (21.6%), followed by positions +2 (14.7%) and –1 (14.3%). Point mutations at positions +3, +4, +6 and −2 each accounted for <3% of all the single-nucleotide substitutions. In contrast to cryptic 5′ss, the most frequent point mutations resulting in 5′ss were at the highly conserved first (28%) and second (34%) intron nucleotides (). Single-nucleotide substitutions at position +5 were found only in 5/81 (6%) unique 5′ss as opposed to 50/234 unique cryptic 5′ss (χ = 8.6, = 0.003). The ratio of point mutations in cryptic over 5′ss in the authentic and new 9-nt consensus, respectively, was highest for position +5 (10.0), followed by position −1 (5.5) and +1 (4.1), with an average ratio for all positions of 2.9. The overall proportion of point mutations in patients with aberrant 5′ss that created the 5′GT consensus was ∼55% (). Newly created 5′GT dinucleotides were utilized by the spliceosome in 100% of the observed cases. In contrast, although mutations generating 3′AG dinucleotides found in individuals with aberrant 3′ss are also present in about half of the cases, only ∼95% are used , owing to the presence of ‘AG exclusion zones’ downstream of the BPS (). and show the breakdown of point mutations by nucleotide and by highly conserved positions of the 5′ss consensus. Transitions (R-to-R or Y-to-Y, Y is pyrimidine), which account for 62.5% of point mutations in human disease genes (), were found in 58.7% of cases (). Comparison of mutations in highly conserved positions of the 5′ss consensus with those expected based on previously published mononucleotide mutation rates corrected for a number of confounding effects () suggested that the biased distribution is unlikely to be fully explained by differential mutability (; = 0.002 and 4.3 × 10 for position +1 and +2, respectively). However, comparison with the published dinucleotides rates that take into account nearest-neighbour effects no longer showed a significant -value for position +1, consistent with a severe block of splicing following mutations to any nucleotide (). Nevertheless, the distribution of point mutations at position +2 was still unlikely to be fully explained by differential mutabilities (, = 0.035), raising the possibility that the observed under-representation of +2C/A among cryptic 5′ss may be attributed to higher residual levels of accurately spliced pre-mRNAs with 5′GC or 5′GA dinucleotides. This would be consistent with a previously observed +2T>+2C>+2A>+2G hierarchy in splicing efficiency (,) and with efficient recognition of the 0.56% of mammalian introns that have 5′GC-3′AG splice sites (). Finally, the distribution of point mutations at positions +1 and +2 and that for all splicing mutations in the Human Gene Mutation Database (HGMD, ) were not significantly different ( = 0.77 for position +1 and = 0.15 for position +2, ). This suggests that the mutation spectrum of the 5′GT dinucleotide is similar for aberrant 5′ss and exon skipping events, which represent the bulk of HGMD entries. Interestingly, all point mutations at position +5 of authentic 5′ss that activated cryptic 5′ss were substitutions of G, and not any other nucleotide (), raising the possibility that 5′ss with +5G are more susceptible to aberrant splice-site activation than 5′ss with +5H (non-G). However, assuming ∼78% occupancy of this nucleotide in human 5′ss () and a G/C substitution rate of ∼70% derived from the HGMD data (), the expected number of +5H substitutions among authentic sites whose mutation induces cryptic 5′ss activation would only be ∼4 in our dataset and not significantly different from zero ( = 0.1, Fisher's exact test). A prominent influence of differential mutability rates on the mutation spectrum was also supported by the observed predominance of +5G>A transitions over transversions (). In addition, the distribution of point mutations activating cryptic 5′ss was significantly different from that resulting in sites ( < 0.0001, ), with the latter showing peaks in the most conserved positions +1 and +2 and exclusive +5A>G transitions relative to new 5′ss that were located both in exons () or introns (). However, unique point mutations in the 9-nt consensus logged in the HGMD () and in our sample () had significantly different distributions (χ = 27.7, = 0.0005), with position +5 clearly overrepresented in our dataset (∼22% versus ∼12%). This suggests that the mutation spectra underlying cryptic 5′ss activation and exon skipping events are distinct. A search for literature reports of point mutations that produce either aberrant 5′ss activation or exon skipping in the same 5′ss consensus revealed several discordant cases. For example, the substitution IVS46+5G>A resulted in cryptic 5′ss activation 33-nt downstream of the authentic exon–intron junction (), whereas the IVS46+1G>A mutation caused exon 46 skipping (). Similarly, the mutation IVS7+1G>A activated a cryptic 5′ss 75-nt downstream of the authentic exon–intron boundary (), whereas mutation IVS7+2T>G in the same 5′ss led exclusively to exon 7 skipping (). In the latter case, IVS7+2T>G creates a putative splicing silencer containing the AGGG motif, which may prevent activation of the downstream cryptic 5′ss, whereas IVS7+1G>A results in no consecutive Gs in the 5′ss consensus. The presence of IVS+5H in authentic 5′ss, which is not predicted to base-pair with U1 or U6 snRNAs, was proposed to be compensated by having G at the last exon position (–1G) (). The –1G can base-pair to U1 snRNA () and is almost completely conserved (97.5%) in IVS+5H 5′ss (). The +5/−1 association was confirmed with a large sample of homologous human-mouse 5′ss (). In our dataset, only 18/35 (51%) of unique authentic 5′ss that were repressed by mutations of IVS+5G in favour of a cryptic 5′ss had -1G. This percentage is significantly (χ = 10.9, = 0.001) lower than for a large set of authentic 5′ss (5142/6716, ∼77%). In addition to position –1, adenosine –2 was less frequent in our sample (31%; 11/24) as compared with 57% in average 5′ss (3830/6716, = 0.002), while the number of uracils at position +6 was higher (25/35; 71% versus 3415/6716; 51%; χ = 5.1, = 0.02). These results are consistent with previously described +5 dependencies (,) and suggest that authentic 5′ss that are susceptible to IVS+5 mutations are less likely to make sufficient contacts between positions –1/–2 and their interacting factors, but may exhibit stronger putative base-pairing interactions between U1/U6 snRNAs and intron position +6. shows the relative representation of each nucleotide in the consensus sequence of aberrant 5′ss (upper panels) and the corresponding authentic sites (lower panels). The consensus sequence of cryptic 5′ss had lower proportions of conserved residues than for authentic 5′ss at each position, except for the invariant position +1 (A). This difference was much reduced in 5′ss, in which conserved nucleotides at positions +3 through +6 had even higher frequencies than those in their authentic counterparts (B). Sequence alignments of cryptic and sites generated in exons and introns are shown in Supplementary together with their authentic counterparts. Apart from Δ and #H between 5′ss and U1 snRNA, we used seven different algorithms that predict utilization of 5′ss in multiple sequences and are publicly available (A,B). Cryptic 5′ss had significantly lower scores with each algorithm, lower #H and higher Δ than their authentic, wild-type counterparts. Cryptic 5′ss were most effectively discriminated from the authentic sites by the ME model, followed by MDD and MM algorithms. -values obtained for the HBond and NN scores were higher, even when we disregarded cryptic 5′ss with non-canonical 5′ss dinucleotides to obtain the scores and replaced them with group means. All these models clearly outperformed the matrix-based prediction scores—S&S and WMM. The #H and Δ values gave the poorest, albeit still significant discrimination. The weakness of cryptic 5′ss was well illustrated by a shift of the #H peak frequency from seven in the authentic counterparts to six in the cryptic 5′ss (C). In contrast to cryptic 5′ss, 5′ss were not distinguished from their authentic counterparts by any of the tested algorithms (B). Although the number of 5′ss was smaller than cryptic 5′ss (), random selections of the same number of cryptic 5′ss and their comparison with authentic sites gave consistently significant discrimination with several algorithms (data not shown), indicating that computational prediction of 5′ss is poor. However, newly created 5′ss activating pseudoexons had higher ME scores than the remaining 5′ss (8.66 ± 3.00 versus 6.07 ± 4.83, = 0.0002) or the remaining intronic 5′ss (, = 0.002). The corresponding 3′ss of these pseudoexons were slightly stronger than intronic 3′ss (ME scores 6.79 ± 3.39 versus 5.24 ± 4.50, = 0.04) ascertained previously (), but were not significantly different from exonic 3′ss or their authentic counterparts. Thus, activation of cryptic exons through 5′ss use requires their high strength and may be facilitated by intrinsically stronger decoy 3′ss across the newly formed exon. We then tested each computational method separately for aberrant 5′ss in exons and introns (). Although cryptic 5′ss in exons were best discriminated by the ME scores, the lowest -values for cryptic 5′ss in introns were achieved by the NN model. To test whether this could be explained by having to disregard 5′GC splice sites for the NN method in both datasets and replace them by group means, we recalculated the NN and ME scores after removing 5′GC splice sites, but we obtained a similar result ( = 1.2 × 10 versus 1.0 × 10, respectively). Authentic counterparts of intronic 5′ss were intrinsically weak and therefore less likely to challenge newly created competitors. However, this was not evident for exonic sites, strongly suggesting that their activation is more reliant on splicing regulatory sequences in exons rather than on the intrinsic strength of the 5′ss consensus (). The overall performance of ME, MDD, MM, HBond and NN models for the whole set of aberrant 5′ss was very similar, with minimal differences in -values. Finally, mutated authentic 5′ss were on average weaker than cryptic 5′ss, confirming an earlier observation (). Again, the lowest -values of the non-parametric test were observed for the ME model ( and data not shown). Thus, as shown for 3′ss (), the ME algorithm discriminated best both wild-type and mutated authentic 5′ss from cryptic 5′ss (), thus providing a method of choice for computational prediction of aberrant splice sites. Next, we carried out pair-wise comparisons of cryptic and 5′ss with their authentic counterparts. For each computational algorithm, we determined the proportion of aberrant 5′ss that showed equal or higher scores than their respective wild-type authentic sites (D). This proportion was on average significantly higher for 5′ss than for cryptic 5′ss and roughly reflected the ability of each method to discriminate between aberrant 5′ss and their authentic counterparts. The percentage of exonic cryptic 5′ss with equal or higher scores than their authentic counterparts was lowest for the ME algorithm (10.5%). For intronic cryptic 5′ss, the same proportion was lowest for the NN method (14.4%, D). Using the best-performing algorithms, ∼12.3% of cryptic 5′ss were computationally stronger than their wild-type authentic counterparts, yet they were used only if the wild-type 5′ss consensus was inactivated or weakened by mutation. This underscores the importance of factors that repress utilization of decoy splice sites that are present in excess over natural sites in the genome. Importantly, the authentic counterparts of cryptic 5′ss were significantly weaker than a large collection of 8415 natural 5′ss (), with the ME scores of 7.75 ± 2.50 and 8.37 ± 2.08, respectively ( = 2 × 10; Wilcoxon–Mann–Whitney rank test). The distribution of #H in the authentic counterparts of cryptic 5′ss and natural 5′ss was also significantly different ( = 0.02, χ = 13.5, 5 df for 4–9 #H), with a maximum difference at 8 #H (C). The relative weakness of the authentic counterparts of cryptic 5′ss is consistent with the notion that mutations in less conserved positions of stronger 5′ss produce, on average, higher amounts of natural transcripts and less severe phenotypes than identical alterations in intrinsically weaker 5′ss. The predicted strengths of authentic sites that were mutated at position +5 were significantly lower than the average authentic 5′ss (ME scores 7.55 ± 1.81 versus 8.37 ± 2.08, = 0.0002) and also somewhat lower than the authentic counterparts of all unique cryptic 5′ss in our dataset (7.55 ± 1.81 versus 7.75 ± 2.52), despite all having +5G and a higher than average relative frequency of +6T (). Guanine at position +5 was proposed not to be obligatory for 5′ss selection if the two preceding positions are purines (); nevertheless 24/35 (69%) of unique authentic 5′ss with point mutations of +5G had only purines at positions +3 and +4 and only a single authentic counterpart had pyrimidines at both positions. shows a comparison of the ME scores of cryptic and 5′ss by mutated position in the authentic and new 5′ss consensus, respectively. Cryptic 5′ss had similar ME values irrespective of the location of the point mutation ( > 0.05, F-test). In contrast, sites created by mutations in a subset of intronic positions of the new 5′ss consensus tended to be stronger, with statistically significant differences between and cryptic 5′ss for the highly conserved positions +1 and +2. In addition, we compared the ME scores of aberrant 5′ss with their respective counterparts (). The authentic counterparts of cryptic 5′ss induced by point mutations at positions +2 and +5 of natural sites were weaker than the authentic counterparts of cryptic 5′ss activated by substitutions at position +1 (7.24 ± 1.93 for position +2 and 7.22 ± 2.06 for position +5 versus 8.34 ± 2.21 for position +1, < 0.01 for both comparisons). This is consistent with the notion that mutations at less conserved positions of authentic 5′ss are less likely to completely inactivate the 5′ss and result in recognizable phenotypes than mutations at position +1. The authentic counterparts of sites induced by mutations at position +1 were significantly weaker than the authentic counterparts of cryptic 5′ss induced by mutations at the same position (6.33 ± 3.39 versus 8.34 ± 2.21). The number of mutations for the remaining positions of the 5′ss consensus was too small for meaningful comparisons. The average intrinsic strength of aberrant and authentic 5′ss in each category is schematically summarized as the mean ME score in . Taken together, cryptic 5′ss generated were best predicted by models that accommodate nucleotide dependencies in the 5′ss, particularly by the ME algorithm, which takes into account non-adjacent positions (A). Discrimination of exonic cryptic 5′ss from their authentic counterparts was more efficient than that for intronic cryptic 5′ss, because the former category of aberrant 5′ss was weaker than the latter ( = 0.02), for which the NN model gave the best performance (). Computational discrimination of 5′ss and their authentic counterparts was poor (B) as 5′ss were, on average, stronger than cryptic 5′ss (), particularly when generated by point mutations in highly conserved intronic positions of the new 5′ss consensus (). The intrinsic strength of exonic 5′ss could not be distinguished from their authentic sites at all, pointing to the importance of exonic regulatory sequences in their selection. Finally, the authentic counterparts of aberrant (both cryptic and ) 5′ss were weaker than a large collection of human 5′ss, highlighting the practical importance of ranking splice sites in human disease genes using efficient computational tools. We propose that their systematic categorization may facilitate identification of intronic mutations or polymorphisms that affect pre-mRNA splicing, improve the interpretation of unknown alterations and, ultimately, increase the cost-effectiveness of mutation screening. The DBASS5 () provides access to the database of aberrant 5′ss through the search option (Supplemental A). DBASS5 can be searched by phenotype, gene, mutation, location of aberrant 5′ss and their distance from authentic 5′ss. If more than one database entry is found, the user can manually choose the details page (Supplemental B), which shows nucleotide sequences flanking the authentic and aberrant 5′ss, the estimated strength of both authentic and aberrant 5′ss and literature references with a PubMed hyperlink. DBASS5 visitors can register to obtain regular updates by email and can submit published data through a submission tool. Potential applications of DBASS5 include optimization of splice-site prediction algorithms, leading to improved prediction of aberrant 5′ss, identification of genes and gene segments frequently involved in aberrant splice-site activation, detection of splicing mutations in a gene or phenotype of interest and selection of models for studying basic mechanisms of 5′ss utilization. This report presents the first comprehensive and publicly available database of aberrant splice sites in human disease genes. Together with a recently described database of aberrant 3′ss (), this combined resource now contains over 600 unique mutations that create or activate a total of 562 aberrant splice sites. The overall number of reported aberrant 5′ss was higher than aberrant 3′ss, consistent with sequence limitations imposed by additional signal sequences upstream of 3′ss (BPS and PPT) that are important for recognition of splice acceptor sites. The relative ratio of non-repetitive aberrant 5′ss ( = 305) and 3′ss ( = 257) [() and I.V., (unpublished data)], was smaller than that reported for unique splicing mutations in the HGMD that were arbitrarily selected to reside in 5 exonic and 15 intronic nucleotides adjacent to natural splice sites (), i.e, 1.2 versus 1.5, respectively. The lower ratio might reflect a reporting bias towards mutations closer to authentic splice sites for exon skipping events. Mutations located upstream of intronic splicing signals that are required for 3′ss selection could not be detected in many published mutation reports, because these regions were amplified only for a subset of introns or were not scanned at all. In addition, the lower ratio could be due to an under-representation of mutations leading to splice sites in the HGMD as compared to our dataset. Also, the availability of suitable decoy splice sites near mutated sites is likely to determine if the outcome of a splicing mutation is exon skipping or aberrant splice-site activation (). The higher number of cryptic than 5′ss () can, probably to a large extent, be explained by a detection bias of DNA-based mutation screening, a method used to identify most aberrant 5′ss in this dataset, towards coding regions and flanking intronic sequences. As explained above, classification of aberrant 5′ss as cryptic and 5′ss may occasionally be vague, but DBASS5 contains only two ambiguous examples (,). Both cases were induced by G-to-T substitutions in 5′ss that had G at position −1, creating a new 5′GT 1-nt upstream of the authentic 5′'ss. Both cases were classified as cryptic 5′ss in our analysis. The rarity of such cases confirms the validity of the previously proposed () categorization of aberrant 5′ss. The most frequent point mutations that activated aberrant 5′ss were purine transitions, accounting for 45.7% cases (11.1% A>G and 34.6% G>A mutations; ). This figure seems to be somewhat lower ( = 0.08) than the ∼54% (113/211) observed for aberrant 3′ss (), probably due to a higher prevalence of transitions in the 3′YAG than those in the 5′ss consensus. Cryptic 5′ss resulted from point mutations in each nucleotide of the 9-nt consensus except for position –3, consistent with this position being the least conserved. However, position –3 has previously been implicated in pathological exon skipping in well-documented cases (), suggesting that –3 substitutions in weak 5′ss are also likely to result in aberrant 5′ss, although these cases must be rare and aberrant splicing and putative phenotypic manifestations could be subtle. As for positions adjacent to the 9-nt 5′ss consensus, each of the reported single-base substitutions at intron positions +7 and +8 created new 5′GT dinucleotides that were used (,). Despite position +7 exhibiting a predominance of purines after several rounds of functional 5′ss selection experiments (), point mutations downstream of the 5′ss consensus resulting in activation of cryptic 5′ss have thus far not been reported. Among upstream substitutions in DBASS5, we found only a single case of an exonic 5′ss generated by a C>T transition 11-nt upstream of an authentic 5′ss (), consistent with a disruption of exonic splicing regulatory sequences. Are point mutations at any position of the 5′ss consensus sequence particularly prone to aberrant 5′ss activation? As observed for mutations in the HGMD (), position +1 led the frequency table in our dataset, with 49.8% and 39.4% mutations observed in the two studies, respectively (). However, the overall distribution of mutations within the 5′ss consensus was significantly different between the two. In particular, the proportion of mutations at position +5 was almost twice as high among cryptic 5′ss than in the HGMD [ and ref. ()]. For unique point mutations leading to cryptic 5′ss activation, this position was in the second place and position +2 in the third, whereas this order was opposite for unique mutations in the HGMD (50 and 34 mutations versus 347 and 456, respectively; = 0.004). G at position +5 is nearly invariant in (). In contrast, +5G is present in only ∼88% of introns () and ∼78% of human introns (), indicating that relief from the absolute requirement for G was an ancient evolutionary event. Comparison of exonized and non-exonized intronic Alu repeats revealed a higher number of +5Gs in exonized sequences (). Mutations at position +5 have resulted in frequent activation of cryptic 5′ss both in yeasts () and humans (, all references available at: ). Our study is the first to provide statistical evidence that this position is important for distinct aberrant splicing outcomes. DBASS5 gives many examples of natural 5′ss in which different point mutations resulted in the same cryptic 5′ss. Similarly, there are numerous cases in the literature of identical exon skipping events caused by different point mutations in the same 5′ss. The identification of several exceptions in humans using the DBASS5 and HGMD data () is consistent with an earlier observation in , namely that cryptic 5′ss activation by +5G>A mutation was not replicated for another 5′ss point mutation in the same intron (). These rare examples may provide important insights into the requirements for activation of aberrant 5′ss, as opposed to exon skipping events. In addition to the local sequence context, the frequent occurrence of +5G>A substitutions underlying aberrant 5′ss activation ( and ) can be explained by a more severe splicing outcome of these transitions. More dramatic splicing defects for +5G>A transitions than +5G transversions were found in (). In contrast, each IVS1 + 5G>H mutation in the human proinsulin gene promoted activation of a competing decoy 5′ss 26 nt downstream of the authentic 5′ss to the same extent, irrespective of the substituting nucleotide [(); J.K. and I.V., unpublished data], consistent with a position effect. What interaction(s) at position +5 is crucial for aberrant 5′ss activation? Authentic 5′ss in which mutation at position +5 generated cryptic 5′ss had a high proportion of +5G+6T (). Interestingly, the +5G+6T dinucleotide signifies the most frequent location for alternative 5′ss across several species, and this preference was suggested to result from U1 binding rather than U6 binding (). However, compensatory mutations in U1 snRNA that restore base-pairing with the mutated intron frequently fail to suppress aberrant splicing, suggesting that position +5 is engaged in additional interactions (,). Interaction of U6 snRNA with the 5′ss () at intron position +5 () was partially suppressed by U6 mutations predicted to increase base-pairing (,). Although the 5′ ss has very limited complementarity to the U6 ACAGAG motif, this interaction seems to be important for accurate 5′ss selection also in mammals (), albeit not in all systems (). In addition, cryptic splice sites have been induced by co-expression of splicing reporters with mutated snRNAs, including U1 (), U5 (,) and U6 (,), both in yeast (,,,) and mammals (,,). As U1 snRNP binds sequences that are not used as 5′ss and is present in excess over U6, sequential occupancy by both snRNPs may be absolutely essential for accurate 5′ss utilization (,). This would be consistent with rate-limiting U6 snRNA interactions with the pre-mRNA observed for U1-independent splicing () and loose requirements for U1 binding in numerous introns (,,,). Mutations of IVS+5G that resulted in cryptic 5′ss occurred in relatively weak authentic 5′ss and they are likely to further reduce U1 binding so that it may no longer be sufficient for accurate 5′ss recognition. U6/U4.U5 snRNP would then bind to nearby pseudo-sites that are, on average, intrinsically stronger than the mutated authentic 5′ss [ and ref. ()]. The strength of predicted base-pairing interactions at positions +5 and +6 in the authentic counterparts () may hamper the transfer of these 5′ss from U1 to U6. In fact, strong 5′ss were reported to be inhibitory in , potentially delaying the release of U1 and productive interactions with U6 (), although extended complementarity between U1 snRNA and a human immunodeficiency virus 1 donor site did not inhibit splicing (). Mutations that destabilized a yeast 5′ss/U6 duplex improved the second step of splicing and hyperstabilization of the 5′ss/U6 interaction had the opposite effect, suggesting that changing the stability of these interactions alters the equilibrium between the first and second step conformations (). Suppression of 5′ss mutations by U6 in a hybrid reporter was more efficient when U1 could pair nearby than when pairing was restored further away (). In addition, position +5 may directly interact with U6 residues that base pair to the BPS recognition region of U2 () as well as with other splicing factors, such as PRP8 (,). Finally, sequential recognition of position +5 is likely to require contacts with exon-bound factors that may substitute for U1 interactions (,), and that may be essential for spliceosome assembly at authentic 5′ss and contribute to the observed high number of cryptic sites as compared to 5′ss ( and ). In summary, we have shown that cryptic 5′ss in human disease genes are best predicted by computational methods that accommodate nucleotide dependencies and not by methods employing only nucleotide frequency matrices. Discrimination of intronic cryptic 5′ss from their authentic counterparts was less effective than for exonic cryptic 5′ss, as the former were intrinsically stronger than the latter. Computational prediction of exonic 5′ss was poor, suggesting that their activation critically depends on exonic splicing enhancers or silencers, rather than on the strength of the 5′ss consensus, and that improved algorithms for their prediction will need to accommodate auxiliary splicing sequences. The authentic counterparts of both and cryptic 5′ss were weaker than the average human 5′ss, highlighting the practical importance of ranking splice sites in disease genes to improve detection of splicing mutations. The mutation spectra of cryptic and 5′ss were distinct and differed also from that underlying exon skipping events, implicating point mutations at position +5 in frequent activation of cryptic 5′ss. Finally, the development of an online database of aberrant 5′ss will facilitate detection of introns and exons frequently involved in aberrant splicing, identification of auxiliary sequences that control selection of aberrant splice sites, fine-tuning of splice-site prediction algorithms, identification of splicing mutations, as well as studies of the basic mechanisms of splice-site selection. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

S-adenosyl--methionine (SAM)-dependent methyltransferases (MTases) transfer the methyl group from SAM to carbon, oxygen or nitrogen atoms of the target. Substrates for SAM-dependent MTases include DNA, RNA, proteins, polysaccharides, lipids or small molecules, implying their key importance in most biological processes (,) Methylation of DNA has crucial roles in DNA damage repair, regulation of expression and embryonic development (). In prokaryotes, DNA MTases are part of restriction–modification systems, which protect the cells from viral invasion (). Protein methylation is a post-translational process that typically occurs on arginine or lysine residues and is found in prokaryotic and eukaryotic signal transduction pathways, and a role in intracellular signaling has been identified (). Recently, numerous RNA MTases have been discovered and their function studied (). RNA plays a central role in the flow of biological information. Ribosomal RNA undergoes modifications by a number of enzymes during the maturation of ribosome. To date, over 100 different types of nucleotide modifications have been identified, out of which about one-third are present in rRNA (). Of all species, rRNA modifications are the best characterized in (). There are three basic types of post-transcriptional modifications found in rRNA, namely base methylation, ribose methylation and pseudouridylation. Base methylation is the simplest and most common type of modification found in rRNA of prokaryotes. It occurs at the final stages of ribosome maturation and the rRNA sequences in which it occurs are highly conserved. Ribose methylation, occurring at the 2′ hydroxyl position on the sugar backbone, is more common in eukaryotes, and is less frequent in bacteria (). rRNA MTases play a crucial role in the assembly, maturation and regulation of the protein synthetic cellular machinery (). In spite of the wealth of literature on rRNA methylation, the structural information currently available for RNA MTases is insufficient to elucidate their mechanism of action. The structure of an rRNA MTase in complex with the substrate RNA is available only for the ternary complex of 23S rRNA mU MTase RlmD (previously called RumA) with the ribosomal substrate and SAH (S-adenosyl--homocysteine) (). RNA MTases are thus relatively less well-understood compared to DNA MTases, whose structure–function relationships are well established (). As a continuation of our efforts to understand the structure and function of rRNA modifying enzymes, we have undertaken structure determination and functional analysis of RsmC that specifically methylates the N2 atom of G1207 in 16S rRNA. The modified residue mG1207 occurs in a region of the rRNA that is involved in the recognition of peptide chain termination codons. , transversion mutants of G1207, namely C1207 and U1207, were shown to have dominant lethal phenotypes (). Here we report the crystal structure of RsmC from refined at 2.1 Å resolution. RsmC is the first structurally characterized MTase, which exhibits the phenomenon reported earlier for many enzymes, including those involved in RNA modification: presence of duplicated, mutually homologous domains, which preserved the ancestral 3D fold, but accumulated divergent mutations in different regions, leading to the complementary loss of conserved motifs and selective retention of different aspects of function present in the ancestral non-duplicated enzyme. Thus, we combined computational and experimental analyses to identify the key amino acids involved in different functions and to assign the roles to the two domains of RsmC. The gene, cloned into pCA24N vector with a non-cleavable N-terminal His6 tag and corresponding strain with the knocked-out gene, were obtained as a gift from the ASKA recloned library [NBRP (NIG, Japan): ]. Site-directed mutagenesis of the rsmC gene was performed by a PCR-based technique according to the QuikChange site-directed mutagenesis strategy (Stratagene) following the manufacturer's instructions. NTD- and CTD-RsmC variants were constructed by recloning single domains into the pET28 vector by removing single domains in the PCR reaction. The mutant genes were sequenced and found to contain only the desired mutations. For the native protein, plasmid DNA carrying gene was transformed into BL21 (DE3) and grown in 1 l of LB media at 37°C till the OD reached 0.5–0.6. Induction of the culture was then carried out with 100 µM IPTG after cooling it down to room temperature. The cells were continuously grown overnight at 25°C in a shaking flask at 180 rpm. The next day, cells were harvested by centrifugation (9000  for 30 min, 4°C) and pelleted. The cell pellet was first washed with pre-binding buffer (10 mM Na-Hepes pH 7.9, 0.17 M NaCl), and resuspended in 20 ml of binding buffer (20 mM Na–Hepes pH 7.9, 0.5 M NaCl), 5 mM imidazole (pH 7.0), 5%(v/v) glycerol, 10 mM BME, 0.5% (v/v) Triton-X-100 and 1 tablet of Complete™ EDTA-free protease inhibitor cocktail (Roche diagnostics). The buffer conditions were slight modifications to the ones mentioned in an earlier work describing the purification, cloning and characterization of RsmC (). For the selenomethionine (SelMet) substituted RsmC, the cells were grown in Le-Master medium (), using the DL41 strain of (methionine auxotroph). The purification of RsmC was carried out at room temperature. Both native and SelMet RsmC were purified using the same two-step protocol: DEAE sepharose (Amersham biosciences) column followed by Ni-NTA beads (Qiagen) purification. After binding the protein to the Ni-NTA resin for 30–40 min, the beads were washed with binding buffer (without Triton-X100). The protein was then eluted with 20 mM Na-Hepes pH 7.9, 0.5 M NaCl, 5 mM BME, 0.5 M imidazole, 5%(v/v) glycerol. Furthermore, RsmC was passed through a Superdex-200 gel filtration column using an AKTA-FPLC UPC-900 system (Amersham Biosciences). The gel filtration buffer was the same as the final protein storage buffer: 20 mM HEPES pH 7.9, 0.5 M NaCl, 5 mM BME, 10 mM MgCl and 5%(v/v) glycerol. The protein eluted as a monomer (∼40 kDa). The peak fractions were pooled together and concentrated to 4.5 mg/ml by ultra filtration, using a Centriprep centrifugal filter device from Millipore, with a molecular weight cut-off of 10 kDa. Inclusion bodies were collected from the cell extract by centrifugation at 20 000 rpm and resuspended in buffer B (10 mM Tris, 50 mM NaCl, 10 mM imidazole) supplemented with 6 M urea. Dissolved pellet was then centrifuged, followed by addition of buffer B and Ni-NTA resin was equilibrated with buffer B. After 1 h incubation, Ni-NTA resin with the bound protein was washed three times with buffer B. The deletion mutant protein RsmC-CTD was eluted with elution buffer (10 mM Tris, 50 mM NaCl, 10 mM imidazole) supplemented with 6 M urea. Refolding of the purified RsmC-CTD was achieved by sequential dialysis with reducing urea concentrations from 6 M to 4 M, 2 M, 1 M, 0 M against RF buffer (100 mM Tris pH 8.8 400 mM -arginine, 10% glycerol, 0.5% TritonX-100, 1 mM EDTA,1 mM DTT). The dialysis buffer was exchanged every 24 hours. The composition of the dialysis buffer (suitable for the subsequent ITC analyses) was 20 mM Na-Hepes pH 7.0, 300 mM NaCl, 5%(v/v) glycerol, 10 mM MgCl, 10 mM BME. 30S ribosomal subunits were isolated as described previously (). Quantitation of subunits was determined by absorbance at 260 nm (1 A unit is equivalent to 34.5 pmol of 30S ribosomes). methylation reactions were carried out using 2 µg pure RsmC protein or its variants, 6 µM [methyl-14C]-SAM (52.8 mCi/mmol, NEN) and 3 µM 30S RNA ribosome subunit isolated from the _knockout (K.O.) strain in the total volume 60 µl of the buffer (50 mM PIPES [piperazine-’-(2-ethanesulfonic acid)]-Na (pH 7.0), 4 mM MgCl). After 60 min incubation at 37°C, methylation was stopped by heating the reaction mixture to 70°C for 10 min. The RNA was precipitated with 10% TCA onto Whatman GF/C filter disks. The disks were washed twice with 5% TCA, once with 5 ml ethanol and air-dried. The filter-bound radioactivity was determined by liquid scintillation counting. The native and SelMet substituted RsmC was further analyzed for the incorporation of selenium on a Voyager-STR MALDI-TOF mass spectrometer (Applied Biosystems) by comparing the experimentally measured molecular weight of the native protein with that of the SelMet protein. DLS measurements were performed at room temperature on a DynaPro (Protein Solutions) DLS instrument. The homogeneity of the protein samples was monitored during the various stages of concentration in order to avoid aggregation. The percentage of polydispersity was below 16% and the SOS error was less than 10 for all protein samples at various concentrations. SAM was procured from MP biomedicals. For the titration experiments, the protein (both native and variants), was extensively dialyzed against a 500-fold excess volume of the buffer containing 20 mM Na-Hepes pH 7.0, 5%(v/v) glycerol, 10 mM MgCl, 0.3 M NaCl, 10 mM BME, for ∼14 h. SAM solutions were prepared by weight, in the same dialysis buffer. The ITC experiments were carried out using VP-ITC calorimeter (Microcal, LLC) at 20°C using 0.02–0.06 mM of the protein in the sample cell and 1–2 mM of SAM in the injector. All samples were thoroughly degassed and then centrifuged to get rid of precipitates. Injection volumes of 4–5 µl per injection were used for the different experiments and for every experiment, the heat of dilution for each ligand was measured and subtracted from the calorimetric titration experimental runs for the protein. Consecutive injections were separated by at least 4 min to allow the peak to return to the baseline. The ITC data was analyzed using a single site fitting model using Origin 7.0 (OriginLab Corp.) software. RsmC was crystallized using the hanging drop vapor diffusion method. Initial crystals were obtained from a Jena Biosciences (Jena, Germany) screen and further optimized. The best crystals were obtained when a volume of 1 µl of reservoir solution containing 25% (w/v) PEG MME 5000, 0.1 M Tris–HCl pH 8.5, 0.2 M ammonium sulfate was mixed with 1 µl of protein (Hanging drop). Diffraction quality crystals formed in 3 days, with the smallest dimension measuring ∼0.14 mm. RsmC crystals belonged to the space group C2 with one molecule in the asymmetric unit. The cell parameters were = 123.94, = 51.50, = 73.33, ß = 121.52. The Matthew's co-efficient was 2.49 Å/Da and the solvent content 50.7% (). The crystals were directly taken from the drop, and flash cooled in a N cold stream at 100°K. The native crystals diffracted up to 2.5 Å resolution using an -axis 1V++ image plate detector mounted on a RU-H3RHB rotating anode generator (Rigaku Corp., Tokyo, Japan). Synchrotron data were collected at beam lines X12C and X29, NSLS, Brookhaven National Laboratory for the SelMet protein. A complete SAD dataset was collected () using Quantum 4-CCD detector (Area Detector Systems Corp., Poway, CA, USA) to 2.1 Å resolution. Data were processed and scaled using the program HKL2000 (). Of the expected seven selenium sites in the asymmetric unit, five were located by the program SOLVE (). The N-terminal, as well as the C-terminal methionine, was disordered. The initial phases were further improved by density modification using Sharp (v 3.0.15) () that improved the overall figure of merit (FOM) to 0.73. The ARP/wARP () built ∼65% of the molecule. The remaining parts of the model were built manually using the program O (). Further cycles of model building alternating with refinement using the program CNS () resulted in the final model, with an -factor of 0.21 ( = 0.26) to 2.1 Å resolution with no sigma cutoff used during refinement. The final model comprises 334 residues (Ala3-Met336) and 231 water molecules. The N-terminal His-tag and the linker residues were not visible in the electron density map. PROCHECK () analysis shows no residues in the disallowed regions of the Ramachandran plot. A simulated annealing oc omit map of the putative SAM-binding site of RsmC is shown (c). Sequence searches were carried out with PSI-BLAST (), and multiple sequence alignment was constructed with MUSCLE (). Sequence conservation was calculated from the sequence alignment and mapped onto the protein structure using ConSurf (). Structure manipulations and modeling was carried out with SwissPDBViewer and PyMol. Structure database searches and superpositions were done with DALI (). Coordinates and structure factors have been deposited with RCSB Protein Data Bank with code 2PJD. The structure of RsmC from was solved by the single-wavelength anomalous dispersion (SAD) () method from synchrotron data using SelMet-labeled protein and was refined to a final -factor of 0.21 ( = 0.26%) at 2.1 Å resolution. The asymmetric unit contains one RsmC molecule comprising 334 residues from Ala3 to Met336 and a total of 232 water molecules. Neither the N-terminal His-tag nor the C-terminal residues Thr337-Gly343 had interpretable density and were not modeled. The RsmC molecules eluted as a monomer from the gel filtration column. This was consistent with observations in the dynamic light scattering experiments as well as the analysis of intermolecular contacts in the crystal (data not shown). Analysis of the Ramachandran plot using the program PROCHECK () showed 88.6% of all residues within the most favored regions and no residues in the disallowed regions. The crystallographic statistics are given in . The structure of the full-length RsmC with overall dimensions of ∼35 × 40 × 60 Å reveals the presence of two homologous domains of a mixed α/β fold, characteristic for SAM-dependent MTases ( ribbon diagram). The existence of intramolecular homology in RsmC has been earlier predicted by bioinformatics analysis (). The N-terminal domain (NTD) consists of seven β-strands and five α-helices and the C-terminal domain (CTD) has nine β-strands and six α-helices. The NPPF (N269-F271) tetrapeptide motif, which is conserved in mG MTases () is located in a loop between β4 and α5 of the CTD (). This motif is absent from the NTD. The DALI search () shows that there is no structure in the PDB with global similarity to the entire RsmC. However, the isolated NTD and CTD show expected similarity to SAM-dependant MTases from the RFM superfamily () as well as to each other. In particular, the NTD shows higher similarity to the CTD than to any other structure: RMSD 2.4 Å for 135 Cα atoms, DALI -score of 13.1. As predicted by bioinformatics analyses (), among other proteins of known structure, MJ0882, a putative MTase from (PDB code 1dus) is the closest homolog of both NTD and CTD: it superimposes onto the NTD with 2.5 Å RMSD over 138 Cα atoms, DALI -score of 13.1 and onto the CTD with RMSD 2.0 Å over 173 Cα atoms, DALI -score of 23.3. Other MTases from the large RFM superfamily show significant, but lower structural similarity (data not shown). Although the structures of the NTD and CTD of RsmC are highly similar to each other, the structure-based sequence alignment of the two domains indicates that there is only 12% amino acid identity between them (). Clearly noticeable is the preservation of a non-polar character of the residues forming the ß-sheet core of both domains and the lack of conservation of residues at the surface. These features suggest that both domains of RsmC originated by intragenic tandem duplication from a primitive single-domain ancestor similar to MJ0882, and that they accumulated divergent mutations that made them dissimilar on the surface, while preserving the structural scaffold. It is important to note that the NTD appears to have accumulated more sequence and structural changes than the CTD with respect to MJ0882: while the CTD exhibits 22.7% amino acid sequence identity to MJ0882, the NTD shows 11.4% identity both to the CTD and to MJ0882 (see also the aforementioned DALI Z-scores, 23.3 versus 13.1). Although the sequence analysis of RsmC had been reported (), thus far no high-resolution structure was available to provide a 3D framework for sequence-function considerations. Both domains of RsmC are members of the RFM superfamily of MTases, which is characterized by the presence of a series of motifs conserved at the structural level, and typically also at the sequence level (). Motifs I, II and III form a SAM-binding pocket, while motifs X and IV usually form the ‘floor’ and the ‘roof’ of the catalytic site and may be important for the methyl group donor SAM and substrate binding, positioning them in optimal orientation for the methyl group transfer to occur. Motif VI often participates in the formation of the active site from the substrate side, motifs V and VII are typically important for the structural stability and motif VIII can participate in substrate binding. On the sequence level, motif I is strongly conserved among nearly all members of the RFM superfamily and typically assumes the pattern similar to (D/E)XGXGXG. Motif IV typically contains the key substrate-binding and/or catalytic residues and assumes very different sequence patterns in MTase families acting on different molecules. In MTases acting on exocyclic amino groups of nucleic acid bases (those yielding mA, mC and mG modifications), the typical pattern of conservation is (N/D/S)PP(Y/F/W/H) (). To identify the potential functionally important sites in both domains of RsmC, we calculated the sequence conservation in the RsmC family and mapped it onto the protein surface. This analysis reveals two conserved patches: the larger one lining up a deep pocket in the CTD formed by motifs: X, I, II, III, IV and VI, and the smaller one on the exposed protuberance of the NTD formed by motifs VII and VIII. Importantly, the conservation is asymmetric across the domains—neither the NTD pocket nor the CTD protuberance shows any significant conservation (A). On the other hand, mapping of the electrostatic potential on the surface of RsmC reveals that the protein is almost uniformly negatively charged with the exception of a small positive patch on the conserved NTD protuberance (B). We carried out analogous analyses for a comparative model of RlmG (YgjO), a MTase closely related to RsmC and also exhibiting two domains, but specific for mG modification at the G1835 another position of 23S rRNA (). The distribution of conservation in the RlmG family is similar to that in the RsmC family, with high conservation in the CTD pocket and on the NTD protuberance ( and Supplementary Figure 1). Interestingly, while the CTD pocket is conserved between RsmC and RlmG, the NTD protuberance is not, i.e. motifs VII and VIII in both families exhibit different conserved amino acids (). RlmG is also negatively charged, with positive patches on both NTD and CTD protuberances (Supplementary Figure 2). Conservation of the pocket with motifs I and IV suggest that the CTD of RsmC and RlmG is important for binding of the SAM cofactor and the catalysis of the methyl transfer reaction. On the other hand, a positively charged protuberance that shows differential conservation in MTase families of different specificity is likely to be important for the recognition and binding of their different rRNA substrates. This prediction is further supported by bioinformatics methods for prediction of RNA-binding sites RNABindR () and BindN () that identify region 130–145 (encompassing motif VIII in the NTD) as a likely RNA-binding site (data not shown). To characterize the function of each domain of RsmC and to confirm the predicted role of individual residues, we designed and constructed two deletion mutants corresponding to the isolated NTD and CTD (amino acids 1–158 and 159–336, respectively), and a series of point mutants of conserved residues in the full-length RsmC that mapped to the predicted SAM-binding site, guanosine-binding/catalytic site and the RNA-binding site. For the potential RNA-binding site we constructed three double mutants in the neighboring positively charged residues (C). The NTD as well as the point mutants expressed and purified easily using procedure optimized for the wild-type protein, while the isolated CTD turned out to be very difficult to purify in these conditions and only the purification and refolding from inclusion bodies enabled us to obtain sufficient amounts of the deletion mutant protein for further experiments. It is known that a maltose-binding protein (MBP) can act as a ‘passive chaperone’ to improve the solubility and promote the proper folding of their fusion partners (). Thus, we constructed two variants of the RsmC CTD, fused to the MBP either in the N- or C-terminus of the isolated domain (i.e. MBP–CTD or CTD–MBP). We found that the MBP–CTD fusion protein purifies well, similar to the wild-type RsmC (NTD–CTD), while the CTD–MBP fusion protein purifies poorly, similar to the isolated CTD (data not shown). In the MBP–CTD fusion, the MBP domain physically replaces the NTD of the wt RsmC and has the opportunity to fold before the CTD, as it leaves the ribosome earlier. On the other hand, in the CTD–MBP fusion CTD leaves the ribosome first, and it is likely that it starts to fold before it has a chance to interact with the MBP domain. Our results suggest that the RsmC CTD has lost the ability to fold on its own and requires a pre-folded ‘intramolecular chaperone’ localized at its N-terminus, be it the NTD or another well-folded domain such as MBP. In order to characterize the function of individual residues, we carried out the functional, biochemical and biophysical characterization of the point mutants. The biochemical assay involving the methylation of ribosomes isolated from the rsmC▵ strain (see ‘Materials and methods’ section for details) revealed that all mutant proteins exhibit reduced activities compared to the wild-type RsmC (). In particular, alanine substitution of residues predicted to be important for SAM binding showed the most severe loss of activity (D202A in motif I to 4% and D227A in motif II to 13%). The alanine substitution of the Asn residue in the predicted catalytic NPPF motif IV (N268A) has reduced the activity to 20% of the wild-type. On the other hand, substitutions of individual residues in the predicted RNA-binding site had relatively mild effects on the RsmC activity—their activity was typically reduced only to 30–50% of the wt enzyme (). Double mutants exhibited further reduction of activity, e.g. K86S/K88S to 16%. These results are very similar to those obtained in the course of mutagenesis of the rRNA:mA methyltransferase ErmC’ (), where it was also impossible to obtain a mutant that would be completely inactive even with multiple substitutions in the predicted RNA-binding site. The interactions between RsmC (and its mutant variants) and the methyl group donor SAM were studied by the Isothermal Titration Calorimetry. The thermodynamics of binding is given in . The mutants D202A and D227A in the potential SAM-binding site in the CTD showed complete inability to bind the cofactor (), while the N268A mutant in the predicted catalytic motif NPPF that coordinates interactions between SAM and the target guanosine showed almost 5-fold reduction in the SAM-binding affinity (Supplementary and ). On the other hand, mutants in the predicted RNA-binding site in the NTD could still bind SAM with wild-type-like affinities () indicating that their reduced activity is not due to the compromised cofactor binding. Domain duplication and functional specialization is a common evolutionary process. The duplication of a gene encoding a primitive multifunctional protein yields two independent proteins or one protein with two similar domains, which may experience relaxation of functional constraints and increased rate of mutations [review: ()]. A number of primitive homooligomeric enzymes have been reported to possess heterooligomeric counterparts with specialized subunits, the best known example being probably the proteasome [review: ()]. Among enzymes involved in RNA metabolism, the most frequent specialization in enzymes composed of two or more homologous domains concerns substrate-binding, catalysis or structural stability, accompanied by the degeneration of ancestral activities. Examples include heterodimeric tRNA deaminases () and heterotetrameric tRNA:mA58 MTases (,). Similar mechanisms have been postulated for other MTases, including the protein-modifying enzyme PRMT7 comprising two domains in the single polypeptide () and eukaryotic DNA MTases Dnmt3a/Dnmt3b/Dnmt3L, where the ‘degenerated’ Dnmt3L is a regulatory subunit in the heterodimeric complex with Dnmt3a or Dnmt3b (,). However, thus far no structural information existed to analyze this phenomenon in detail. The structure of RsmC provides the first atomic-level picture of an RNA-modification enzyme as well as of an MTase which comprises two domains apparently derived from a common ancestor, which underwent differential functional specialization. According to ITC measurements, RsmC binds only one SAM molecule, and mutational analyses clearly demonstrate that conserved residues in the CTD are responsible for SAM binding. The direct involvement of the NTD in rRNA binding remains to be established, nonetheless mutational analyses of residues in the conserved charged patch on the NTD surface, predicted to be involved in RNA binding by the RNABindR and BindN methods, give strong support for this prediction. Substitutions of these residues significantly affected the MTase activity, while they had no effect on the SAM-binding ability of the enzyme. Despite the conservation of the structural ‘MTase-like’ scaffold, two RFM domains of RsmC exhibit complementary pattern of sequence loss or conservation in motifs implicated in substrate-binding (NTD) versus cofactor-binding and catalysis (CTD). Not surprisingly, the isolated domains are unable to carry out the methylation reaction. Moreover, even when the two isolated domains of RsmC are mixed together, they fail to form a catalytically active complex, suggesting that cooperation between the domains requires physical linkage or begins already at the stage of protein synthesis. Indeed, we found that the CTD requires a well-folded N-terminal partner to fold correctly. It is also possible that the peptide linker between the NTD and the CTD plays a role in coordinating binding and catalysis. This specialization of complementary functions and resulting mutual dependence of domains (concerning both protein stability and enzymatic activity) are likely to be common to other ‘pseudodimeric’ MTases with partially degenerated motifs, such as the protein-arginine MTase PRMT7 and in RNA modification enzymes composed of several homologous domains. Recently, Dontsova and coworkers characterized experimentally three rRNA:mG MTases: RlmL that modifies G2445 in 23S rRNA (), RlmG that modifies G1835 in 23S rRNA () and RsmD that modifies G66 in 16S rRNA (). They have demonstrated that RsmD is encoded by the YhhF open reading frame (ORF), and that the YgjO ORF encodes not the RsmD enzyme as previously believed (,), but RlmG. They have also determined the structure of YhhF/RsmD, which revealed a single catalytic domain (). Based on these findings, Dontsova and coworkers proposed a hypothesis that rRNA:mG MTases can be divided into two categories based on the domain structure and substrate specificity: MTases composed of multiple domains would recognize protein-free ribosomal RNA and most probably, unfolded early assembly intermediates , while MTases comprising only the catalytic domain would recognize only late assembly intermediates resembling the completed 30S particle and not the free RNA (). They predicted that RlmG and RlmL (whose structures remain unknown) are composed of multiple domains, and that RsmC closely resembles RsmD in that it is composed only of a single domain (). On the other hand, our results clearly show that RsmC is composed of two domains and is closely related to RlmG (YgjO) rather than RsmD (YhhF). Besides, RsmD has been shown to require the presence of proteins S7 and S19 with the 16S rRNA to be recognized by the enzyme (). Thus, it appears that the relationship between structure and substrate specificity in rRNA:mG MTases is more complex and cannot be inferred simply from the number of domains in different proteins. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R o n l i n e .

Polycyclic aromatic hydrocarbons (PAHs) are environmental pro-carcinogens that are produced during combustion of organic materials (). Benzo[]pyrene (BP), one of the most extensively studied PAHs, is found in a wide range of substances ingested or inhaled by humans, such as automobile exhaust, tobacco smoke and food (). Several BP metabolic activation pathways have been delineated (). The predominant diol epoxide pathway yields, among others, the major reactive (+)--benzo[]pyrene diol epoxide (BPDE) (). The (+)--BPDE enantiomer is highly mutagenic and tumorigenic in mammals (). This metabolite can react chemically with DNA, forming several bulky BP-base adducts, among which the 10(+)---[BP]--dG ([BP]G*) adduct () is dominant and is the focus of this study (). This adduct can lead to mutations when replicated and (). Although such bulky adducts mainly block replicative DNA polymerases, error-prone bypass can occur if the blocked replicative polymerase is replaced by one or more bypass polymerases (). Nevertheless, replicative polymerases are also able to bypass such adducts, though rarely, with potentially mutagenic consequences (,). Bacillus fragment (BF) is a Pol I family replicative DNA polymerase large fragment from a thermophilic strain of (). Our focus on this particular replicative polymerase is motivated by the availability of high-resolution BF crystal structures of a binary complex (PDB ID: 1L3S) with enzyme and primer/template DNA, and a ternary complex (PDB ID: 1LV5) which contains, in addition, an incoming Watson-Crick paired 2′-deoxynucleotide triphosphate (dNTP) (). BF adopts an open conformation in the binary complex. Placement of the correct dNTP partner opposite the template base causes the fingers domain to close tightly around the active site via an induced-fit mechanism () to form a reaction-ready ternary complex (, ). Three important sites have been delineated at and near the active site of BF, namely the pre-insertion, insertion and post-insertion sites. The pre-insertion site, in the fingers domain, is composed of two helices, O and O1, and the loop connecting them. In the open BF, the pre-insertion site is occupied by the base next-to-be replicated, while the insertion site is blocked by a critical polymerase amino acid residue, Tyr-714. When the polymerase closes, the base which was previously in the pre-insertion site moves to the insertion site, where it pairs with the incoming dNTP and the insertion site is no longer blocked by Tyr-714. The pre-insertion site is blocked by the loop connecting the O and O1 helices. After the incoming dNTP is incorporated into the 3′-end of the primer, the polymerase opens again, the DNA translocates, and the new base pair moves to the post-insertion site. The insertion site is now again blocked by Tyr-714, the pre-insertion site is occupied by the next-to-be-replicated template base and the replication cycle continues (). Thus, the pre-insertion, insertion and post-insertion sites are vital to the functioning of this type of polymerase. An incompletely resolved open BF crystal structure with the [BP]G* adduct as the base next-to-be-replicated and a well-resolved open BF crystal structure with [BP]G* at the post-insertion site paired with cytosine (PDB ID: 1XC9) () provide partial insights into the inhibition of a replicative polymerase by a bulky adduct. These two structures represent the stages before and after nucleotide insertion opposite the adduct, respectively. In the incompletely resolved structure, the adducted guanine, which would occupy the pre-insertion site if not modified by the BP moiety, is disordered, is not located in the pre-insertion site, and is not poised for the next step of nucleotide incorporation. In the well-resolved structure, the modified dG* adopts a normal conformation with the BP rings on the DNA minor groove side. Normally on this side there are extensive interactions between DNA and the BF polymerase, constituting a scanning track that is most important in ensuring polymerase fidelity and processivity (). Placement of the BP in this position disrupts the track. In addition, the presence of the adduct distorts the primer end so that the 3′-OH is no longer in an appropriate position for attacking the α-phosphate of the incoming dNTP during nucleotide incorporation. The position of the BP is such that it would interfere with dNTP binding and polymerase closing. In the present work we investigated, using molecular modeling and dynamics simulations, how the representative replicative polymerase, BF, treats the [BP]G* adduct through an entire replication cycle. The availability of the series of crystal structures with well-characterized sites employed in the replicative cycle presents an excellent opportunity to elucidate molecular details of lesion processing in this model system. We investigated the effect of the [BP]G* adduct on the crystallographically well-determined BF structures, considering for the first time both open binary and closed ternary complexes, as well as the pre-insertion, insertion and post-insertion sites. This work builds on earlier studies involving the [BP]G* adduct in closed ternary complexes of the Pol I family DNA polymerase from phage T7 (). Our primer extension and single dNTP insertion studies provide the experimental foundation for the modeling studies. The experimental data clearly show predominant polymerase blockage with a small amount of lesion bypass and preferential insertion of purines opposite the adduct. Our modeling studies provide molecular views of the various intermediates in the multi-step process of replication past the bulky [BP]G* adduct, enabling us to gain a better understanding of the structural properties leading to predominant polymerase blockage and to rationalize the rare mutagenic lesion bypass. Blockage is engendered by substantial distortions of the active site in the glycosidic bond conformation of the dG*, while some bypass appears feasible when this bond is in the less distorting conformation; the conformation also supports the preferential misincorporation of purines opposite the lesion. [γ-P] ATP (3000 ci/mmol) was purchased from Perkin-Elmer Life Sciences, Inc. (Boston, MA). The dNTPs, dATP, dCTP, dGTP and dTTP were purchased from New England Biolabs, Inc. (Beverly, MA). BF was prepared as described previously (). The 11-mer oligonucleotides with a central triple base motif CACACGGACAC were synthesized by standard phosphoramidite methods on a Biosearch Cyclone automated DNA synthesizer (Milligen-Biosearch Corp., San Rafael, CA) and were purified and desalted by standard HPLC protocols. The site-specifically modified oligodeoxynucleotides were ligated to a 13-mer and a 19-mer to form 43-mer template oligonucleotide strands, 5′-GAC TAC GTA CTG TCA CAC G*GA CAC GCT ATC TGG CCA GAT CCG C-3′. The ligated oligonucleotides were further purified and desalted by three successive ethanol precipitations. The quality of the samples was verified by gel electrophoresis; fluorescence and UV absorption spectroscopy were employed to further verify that the BP residue was intact. The 43-mer templates were annealed with a 5′-end P-labeled 22-mer primer. The BP modified guanine residue was positioned at template position 25 counted from the 3′ end of the template strand (a). The time course of primer extension assays of the adducted templates were carried out at 37°C. The standard 30 μl reactions contained 50 mM Tris-HCl (pH 8.0), 5 mM MgCl, 1 mM DTT, 50 μg/ml BSA, 4% glycerol, 4.5 nM primer:template (1:1.5 ratio) complexes, 200 μM dNTPs and 20 nM of BF. The experiments were stopped after pre-selected time intervals (3, 6, 9, 12, 15 and 30 min) by the addition of a 5 μl stop solution (20 mM EDTA in 95% formamide, 0.05% bromophenol blue, and 0.05% xylene cyanol) to the reaction mixture. The samples were then heated at 90°C for 5 min, chilled on ice and then applied to a 15–20% denaturing polyacrylamide gel containing 7 M urea. The replication products were visualized by autoradiography and quantitatively analyzed by a Storm 840 phosphorimager and the Storm ImageQuant software (Amersham). The 43-mer templates were annealed with a 5′-end P-labeled 24-mer primer. The one-step insertion assays of each dNTP (2 mM) opposite the modified guanine were performed utilizing 15 nM primer/template complexes and 2 nM BF concentrations at 37°C for 30 min. The products were visualized and analyzed in the same manner as the running-start primer extension assays. The crystal structures of the open binary complex (PDB ID: 1L3S) and the closed ternary complex (PDB ID: 1LV5) of BF () were used as the starting structures for molecular modeling with coordinates obtained from the Protein Data Bank (). All molecular modeling was carried out using Insight 2000.1 (Accelrys, Inc., a subsidiary of Pharmacopeia, Inc.). The original crystal structures of the ternary and binary complexes were remodeled (full details provided in Supplementary Data). The remodeled complexes were then used to build the initial models for the molecular dynamics (MD) simulations. The [BP]G* was modeled into (i) the pre-insertion site of open BF; (ii) the insertion site of closed BF; (iii) the post-insertion site of open BF; and (iv) the post-insertion site of closed BF, representing four important steps as the adduct threads through the polymerase (a). The DNA sequences were remodeled to match that used experimentally. According to the different positions of [BP]G* in BF at the four steps, four corresponding structures were built (b). We then made a grid search, at 5° intervals in combination of χ, α′, and β′, for structures with minimal close contacts, within the ranges: χ() = 230 ± 50°, χ() = 40 ± 40°, α′ = 0–360°, and β′ = 270 ± 60° [near the low energy domain of β′, which governs the orientation of this stereoisomer ()]. Thus we investigated both and conformations of the glycosidic torsion angle χ () of the BP modified dG* at each of the four steps. Starting from Step 2, we placed the normal partner C opposite the -[BP]G* and all four partners opposite the -[BP]G*. For the case, one [BP]G* conformation with fewest collisions was found for each of Steps 1, 2 and 4; two conformations were found for Step 3, one similar to the available crystal structure and the other different from it. We also simulated the available crystal structure. Therefore, three initial models were investigated for the -[BP]G* at Step 3. For the case, one [BP]G* conformation was found for each of the four steps. The torsion angles in the initial models for all four steps are given in Table S1. For each step, an unmodified control, containing normal partner C, where appropriate, was also constructed for comparison. All initial models are shown in Figure S2. Full details are given in Supplementary Data. MD simulations were carried out using the SANDER module of AMBER 7.0 () with the Cornell . force field () and the parm99.dat parameter set (). Following equilibration, 2 ns of production MD were carried out. Full details are given in Supplementary Data. Trajectories were collected for each modified system and its corresponding unmodified control, and were analyzed using the PTRAJ and CARNAL modules of the AMBER package () to obtain ensemble average values for properties of interest. The RMSDs (root-mean-square deviations) of the whole structure of each system, calculated relative to the first production frame, are shown in Figure S3. The RMSDs of the active site, composed of all the residues within 5 Å of the nascent base pair relative to the first frame, are also provided (Figure S3). We found that all systems are stable after 800 ps. The following analyses are based on this range. All structural figures were prepared using PyMOL (). Running-start primer-extension experiments were carried out with the primer terminus several nucleotides away from the lesion site (a). The 22-mer primer strand is quickly extended by two nucleotides up to the nucleotide opposite the template base flanking the adduct [BP]G* on the 3′ side. Insertion of the nucleotide opposite [BP]G* is slowed but is almost complete (90%) after a 30 min reaction time (b). In contrast, further extension of the primer strand beyond the [BP]G* adduct on the template strand is significantly inhibited. Nevertheless, a small amount of full extension (∼1%) to a 43-mer is also observed after an incubation time of 30 min. It is evident from b that the incorporation of a dNTP opposite the lesion is significantly more effective than primer extension beyond the lesion site. In order to determine which 2′-deoxyribonucleotides are incorporated opposite [BP]G* in the current sequence context, standing-start single nucleotide insertion assays were carried out using a 24-mer primer strand with its 3′-terminal base paired with the template base flanking [BP]G* on the 3′-side. The experimental conditions were adjusted in order to limit the incorporation of nucleotides to <10%. At a [DNA]/[Enzyme] ratio of 7.5 employed in these experiments, it is evident that purines rather than pyrimidines are preferentially inserted opposite [BP]G* (c, and Supplementary Data). The steady-state and Michaelis-Menten parameters, determined for dATP and dGTP insertion opposite [BP]G* (Figure S4), are summarized in Table S3. The values of the insertion efficiencies = / are ∼7000 and ∼20 000 times smaller, respectively, than the value of insertion of the normal dCTP opposite an unmodified G template base in the same sequence context (Table S3). In order to investigate the structural basis of these observations, especially the relatively facile insertion of dNTP opposite the lesion, and the strong inhibition of primer extension beyond the lesion, we constructed initial models with - and -[BP]G* at (i) the pre-insertion site of open BF; (ii) the insertion site of closed BF; (iii) the post-insertion site of open BF; and (iv) the post-insertion site of closed BF. These are four consecutive steps as [BP]G* advances through the active site of BF (a). For studies of mismatched dNTP insertion, we considered only the -[BP]G* case, based on our modeling results that -[BP]G* would be totally blocking irrespective of the incoming dNTP. The models were subjected to 2 ns of MD following equilibration, and achieved stability after 800 ps (Figure S3). The structures, after MD, are shown in , S5 and S6, and analyses of the trajectory ensembles are provided in Tables S4–S11 and Figures S7–S11. The structural properties relevant to polymerase function were analyzed in each case in the presence of the [BP]G* lesion and compared to the corresponding unmodified control. We chose structural and geometric features of the active site region of the binary polymerase-DNA and ternary polymerase-DNA-dNTP complexes deemed important for faithful and efficient catalysis (,,). Our hypothesis is that distortions to the correct alignment, as observed with unmodified templates, impede or slow polymerase activity. The structural characteristics of the active site we examined are: (i) the width of the pre-insertion site (Figure S7 and Table S4) at Steps 1 and 3 (the open binary complex): the pre-insertion site is open and occupied by the next-to-be-replicated base in the binary complex, but is blocked by a polypeptide loop in the closed ternary complex (); (ii) the number and occupancies of the hydrogen bonds in the [BP]G*-containing base pair (Figure S8) and its neighboring base pairs; hydrogen bond occupancies are the percent of time during the stable region of the MD trajectory (0.8–2.0 ns) that the hydrogen bond is present according to the following criteria: 3.3 Å between heavy atoms (donor and acceptor) and a hydrogen bonding donor-hydrogen-acceptor angle of 135°; specifically, we monitored the hydrogen bonds at the +1 base pair at Step 1; 0 and +1 base pairs at Step 2; +1 and +2 base pairs at Step 3; and 0, +1 and +2 base pairs at Step 4 (b and Table S5); (iii) the stacking interaction between Tyr-714 and the templating base at the post-insertion site () at Steps 1 and 3 (the binary complex): Tyr-714 blocks the insertion site and stacks with the template base at the post-insertion site in the binary complex, but this stacking interaction is not present in the ternary complex (); (iv) base stacking interactions in the nascent base pair and the DNA duplex region around the [BP]G (); (v) minor groove interactions at the post-insertion site (+1 base pair) (b and Table S6), where the newly formed base pair first meets the minor groove scanning track; at this site, the N3/O of the template base forms hydrogen bonds with Gln-797, and the O/N3 of the primer base forms hydrogen bonds with Arg-615 (); (vi) the width of the post-insertion site, measured by the distance between the C1’ atoms of the template and primer strands at this site (Table S7); (vii) hydrogen bonds between amino acid residues and incoming dNTP; these stabilize nucleotide binding at Steps 2 and 4 (the ternary complex) (Table S8); (viii) the active site pocket geometry (defined in a and b) at Steps 2 and 4 (the ternary complex) ( and Table S9); (ix) blockage by BP of the nucleotide binding pocket at Steps 1 and 3 (the binary complex) (); (x) frequency of sampling a near reaction-ready distance, in the range of 3.1–3.5 Å, between Pα of the dNTP and the O3′ of the primer end (,) (Figure S9 and Table S10); (xi) the angles formed by the O3′, Pα of the dNTP and the O connecting Pα and Pβ, ideally 180° for in-line attack and chemical reaction () (Figure S9 and Table S10); (xii) distance between the two Mg ions (Table S11): normal values are ∼3.5 Å (); (xiii) octahedral coordination of the Mg ions (Table S11). Properties (x)–(xiii) are for Steps 2 and 4 (the ternary complex). In the present study, we have carried out a computational study, based on key experimental observations, to investigate how the bulky [BP]G* adduct is processed during DNA replication by the polymerase BF. Primer extension and single nucleotide insertion assays reveal that the [BP]G* lesion allows slow incorporation of purine dNTPs when these are paired with the modified guanine on the template strand, but that further extension is strongly inhibited. To account for these observations, we have constructed two sets of initial models with BP-modified dG* in the normal and the less prevalent conformations. The conformation has been observed experimentally by solution NMR methods in the case of an unpaired guanine residue of the (+)--[BP]G* adduct positioned at a template-primer junction (,), revealing that the conformation is feasible in such structures. We have investigated four consecutive steps in DNA replication as [BP]G* transits through the BF active site region, with polymerase open and closed, in binary and ternary complexes (a). MD simulations were carried out and important structural properties were analyzed in detail and compared to the unmodified control. The MD results structurally interpret our experimental data and are consistent with the available crystal structure (). We hypothesize that increased distortions at the polymerase active site are a hallmark of reduced polymerase fidelity and efficiency. The distortions for each property were scored, and a composite score was derived for each system (). A numerically more negative score indicates a more distorted structure. The properties evaluated and the scoring criteria employed were selected based on our current best understanding of proper active site geometry. A recent crystal structure reveals a well-organized active site (), whose features closely resemble those in our unmodified models, suggesting that our criteria for evaluating distortions are reasonable. However, the scoring function is an evolving one which is being improved as our knowledge base advances. At this stage, the composite scores reflect the degrees of distortion, but the relationship between the scores and the biological outcome is very likely non-linear. #text S u p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Aminoacyl-tRNA synthetases (aaRSs) play a central role in the assembly of amino acids into polypeptide chains. They catalyze the specific esterification of a given amino acid to its corresponding tRNA through a two-step reaction (). In the first step, the specific amino acid and ATP substrates are recognized and then are converted into a reactive aminoacyl-adenylate (aa-AMP) intermediate in the presence of magnesium ions. In the next step, the amino acid moiety from the aa-AMP is transferred to the 3′-CCA terminus of the cognate tRNA. This enzymatic function is crucial for the fidelity of protein synthesis, in which the genetic code is translated to the amino acid sequence. The primary sequence analyses as well as the tertiary structure determinations allowed the partition of the 20 aaRSs into two exclusive classes, I and II, each consisting of 10 enzymes (). Each class I enzyme has a Rossmann-fold domain as the catalytic domain. In addition, two consensus motifs, HIGH and KMSKS, are conserved among the class I enzymes. The class I enzymes are further divided into three sub-classes: Ia, Ib and Ic (,). The class Ic enzymes are aaRSs for tyrosine and tryptophan and are unusual in that they act as dimers, while the other class I (Ia and Ib) enzymes act as monomers. Tyrosyl-tRNA synthetase (TyrRS) is the first aaRS to have its crystal structure solved (). The specific aminoacylation of tRNA by TysRS relies on the identity determinants (the anticodon bases, the C1-G72 base pair, and the discriminator base A73) in the cognate tRNA (). To date, a number of crystal structures of the TyrRSs have been solved: they are from four bacteria, [BstTyrRS, ()], [SauTyrRS, ()], [TthTyrRS, ()] and [EcoTyrRS, ()]; from eukarya [HsaTyrRS, ()]; and from four archaea, [MjaTyrRS, (,)], and [AfuTyrRS, PhoTyrRS, and ApeTyrRS, respectively, ()]. Because the archaeal/eukaryotic TyrRSs–tRNAs pairs do not cross-react with their bacterial counterparts (), the recognition modes of the identity determinants by the archaeal and eukaryotic TyrRSs were expected to be similar to each other but different from that by the bacterial TyrRSs. Such orthogonality is used for the incorporation of unnatural amino acids into proteins with engineered pairs of TyrRSs and tRNAs (). In such situations, structural information on archaeal, eukaryotic and bacterial TyrRSs complexed with their cognate tRNAs has long been awaited. In a half-decade, crystal structure analyses of bacterial () () and archaeal () () TyrRSs complexed with their cognate tRNAs have been reported. Although previous experiments showed that TyrRS could bind only one tRNA per dimer () in solution, the crystal structure analyses (,) have shown that two tRNAs are bound to each dimer in a symmetrical fashion in the crystal. A plausible explanation for this discrepancy (asymmetry in solution vs symmetry in crystal) has been described by Yaremchuk . (). A structural comparison revealed the structural basis for orthogonal specificities of archaeal and bacterial TyrRSs (). On the other hand, no structures are available for eukaryotic TyrRSs complexed with their cognate tRNAs. To understand the molecular basis for the recognition of their cognate tRNAs by eukaryotic TyrRSs, we initiated the structure analysis of TyrRS from (SceTyrRS), the model organism for lower eukaryotes. Here we present the crystal structure at 2.4-Å resolution of the ternary complex of SceTyrRS complexed with a Tyr-AMP analog and the native tRNA(GΨA). The present structure of SceTyrRS complexed with the cognate tRNA and the previously reported structures of bacterial and archaeal TyrRSs (TthTyrRS and MjaTyrRS, respectively) complexed with their cognate tRNAs provide a full set of the recognition modes of the identity determinants of tRNAs by TyrRSs from three kingdoms. Chemicals were purchased from Wako Pure Chemical Co. (Tokyo, Japan). The purification of native modified tRNA was performed in a way similar to the method as described (). A Tyr-AMP analog (-(adenosine-5′--yl) -(-tyrosyl)phosphoramidate (Tyr-AMPN), (A) was prepared as described (). We expressed and purified a C-terminally truncated SceTyrRS (hereafter simply SceTyrRS, residues 1–364), which has full TyrRS activity, for the present crystal structure analysis. SceTyrRS was expressed and purified in a way similar to the method described previously for the full-length SceTyrRS (residues 1–394) (). Crystals of the ternary complex of SceTyrRS were obtained by the hanging-drop vapor diffusion method, as described elsewhere (). Briefly, a droplet was prepared by mixing an equal volume of a protein solution containing . 0.2 mM SceTyrRS, 5 mM Tyr-AMPN (A), . 0.2 mM tRNA, 40 mM KCl in 20 mM Tris buffer at pH 7.5 and a reservoir solution containing 25% (v/v) polyethyleneglycol 400 (PEG400) and 100 mM CaCl in 100 mM Tris buffer at pH 7.5. The crystals belong to tetragonal space group 422 with cell dimensions of = = 63.85 Å and = 330.3 Å (under the cryogenic conditions described below). Assuming one SceTyrRS subunit and one tRNA molecule per asymmetric unit, we obtained a V value of 2.55 Å/Da, corresponding to a solvent content of 52%. Since the crystallization conditions of SceTyrRS contained 25% (v/v) PEG400 in reservoir solutions, X-ray data collections could be performed under cryogenic conditions without further addition of a cryo-protectant. Crystals were mounted in nylon loops and flash-cooled in a cold nitrogen gas stream at 100 K just before data collection. Crystals of the ternary complex of full-length TyrRS were obtained in a similar condition as described above and had similar morphology and cell dimensions to those of the truncated TyrRS. However, they diffracted quite poor (. 10-Å resolution). Initially, a native dataset was collected and several attempts were made to solve the structure of SceTyrRS by the molecular replacement techniques. The structures of several TyrRSs complexed with or without cognate tRNA and deposited in the Protein Data Bank, having . 10–20% sequence identity with SceTyrRS, were used as search models. Secondly, attempts were made to find good heavy-atom derivatives for phasing by the isomorphous replacement techniques. Since both of these attempts failed, we prepared a Se-Met substituted SceTyrRS using LeMaster medium () and B834(DE3) cells for phasing by the multiwavelength anomalous diffraction (MAD) method. The MAD data collection was performed at beamline 38B1, SPring-8. XAFS measurements were carried out around the selenium K absorption edge using an Se-Met SceTyrRS crystal in a cold nitrogen gas stream at 100 K. Subsequently, four datasets were collected from a new single crystal of Se-Met SceTyrRS on and around the selenium K absorption edge at 100 K using an ADSC Quantum-4R CCD detector. All datasets were integrated using the program package DPS (). Scaling and processing were performed using the CCP4 program suite (). Thereafter, a high-resolution dataset was collected from a single crystal of native SceTyrRS at beamline 40B2, SPring-8 (λ = 1.00 Å) in a similar way. The data collection statistics are summarized in . Initial phase calculation was carried out at 3.0-Å resolution using the program SHARP/autoSHARP (). Of the nine selenium sites, seven were found. Interpretation of the electron density maps and model-building procedures were carried out on a Linux PC with the aid of the program X-fit as implemented in the program XtalView version 4.0 (). The obtained model was refined at 2.4-Å resolution with the programs CNS () and REFMAC (). After each refinement calculation, the obtained model was corrected with difference Fourier maps using XtalView. Water molecules were picked by the water-add routine in XtralView. The stereochemistry of the model was verified using the program PROCHECK in the CCP4 program suite. The present model includes residues 8 to 356 of SceTyrRS, one tRNA molecule, one Tyr-AMPN molecule, one magnesium ion and 57 water molecules per asymmetric unit. Residues 224–233 of SceTyrRS, the base moieties 16, 20, 31, 40, 41 and 46 of tRNA, and the whole nucleotides 17, 20a, 20b, 32 and 33 of tRNA, were disordered. The current -factor is 0.245 ( = 0.289) for the resolution range of 40.0–2.4 Å. The root-mean-square-distances (RMSDs) from ideal values are 0.006 Å for bond lengths and 1.124 for bond angles. The refinement statistics are summarized in . The atomic coordinates have been deposited in the Protein Data Bank with the entry code 2DLC. Figures were produced using both the DINO () and POV-Ray () programs (B and D) or both the Ribbons () and POV-Ray programs (C, , , , and ). The crystallization trials conducted to date have not successfully obtained crystals of ligand-free SceTyrRS. Fortunately, however, crystals of the ternary (SceTyrRS/Tyr-AMP analog/tRNA) complex of SceTyrRS were successfully obtained. For this study, a Tyr-AMP analog having an -acyl phosphoramidate linkage where the oxygen atom of the mixed anhydride bond (-C-O-P-) of Tyr-AMP was replaced by an amino group (-C-NH-P-) (-(adenosine-5′--yl) -(-tyrosyl)phosphoramidate [(hereafter Tyr-AMPN), A], was used for crystallization. Initial phase calculation was performed by the MAD method using the Se-Met-substituted SceTyrRS at 3.0-Å resolution (B). Further model building and structure refinement were performed using the native SceTyrRS, and we refined resulting model to an -factor of 0.245 ( of 0.289) at 2.4-Å resolution. The data collection and refinement statistics are summarized in . The overall structure of the ternary complex of SecTyrRS is shown in C. As observed in the cases of TthTyrRS () and MjaTyrRS (), SecTyrRS forms a homo dimer and two tRNAs are bound to each dimer in a symmetrical fashion. The asymmetric unit contains one SceTyrRS subunit and one tRNA molecule (one-half of a 2:2 complex). The molecular two-fold axis coincides with a [1 1 0] crystallographic two-fold axis. In addition, a Tyr-AMP analog, Tyr-AMPN, is bound at the active site of SceTyrRS (B). A perfectly symmetrical SceTyrRS/tRNA/Tyr-AMPN complex in crystal presented here is contrary to the functional asymmetry of TyrRSs in solution () that the enzyme exhibit half-of-the-sites’ reactivity with respect to the binding of tyrosine or Tyr-AMP and one tRNA molecule is bound per dimer of TyrRS. As for the explanations for this discrepancy, we completely agree with the notion pointed out by Yaremchuk . () that perfectly (or nearly) symmetrical structure observed in the crystal structures of TyrRSs is due to the fact that (i) there is ample time for substrate binding to both active sites in a crystallization experiment and there is quite likely preferential crystallization of a symmetrical form and (ii) in the case of tRNA binding, the same arguments hold. In addition, we postulate that considerably higher concentrations (as compared with the previous functional studies for TyrRSs in solution) of SceTyrRS (. 0.2 mM), tRNA (. 0.2 mM) and Tyr-AMPN (5 mM) and a large excess of Tyr-AMPN in the crystallization solution favor the formation of symmetrical dimer in the crystal. The subunit of SceTyrRS consists of two domains. One of these, the catalytic domain, provides the groups necessary for converting the substrates Tyr and ATP into reactive intermediate Tyr-AMP (the first step of the aaRS reaction) and for transferring the amino acid moiety from the Tyr-AMP to the 3′-CCA terminus of the cognate tRNA (the second step of the aaRS reaction). The other domain is responsible for the recognition of the anticodon bases of the cognate tRNA. The two domains are unequal in size; the catalytic domain is somewhat larger and comprises 232 residues, whereas the anticodon-binding domain comprises 117 residues. The catalytic domain comprises residues 8 to 239. The structural core of this domain is an α/β structure (or Rossmann fold) comprised of a six-stranded parallel β-sheet and 10 surrounding α-helices (C, light colors). The Tyr-AMPN molecule is bound in the central region of the carboxyl end of the parallel β-sheet in the center of the domain. The anticodon-binding domain comprises residues 240 to 356. The basic element of the secondary structure in this domain consists of six α-helices and a two-stranded anti-parallel β-hairpin (C, dark colors). A loop region between the two domains (residues 224–233), including the KMSKS signature motif, which is one of the two consensus motifs conserved among the class I aaRSs, is disordered. The tRNA molecule forms an L-shaped structure. The acceptor stem and anticodon loop of the tRNA interact with different subunits of the dimeric TyrRS molecule (C). The structural discontinuity in the anticodon-loop of tRNA is due to the disordered nucleotides C32–U33. However, the anticodon triplet of tRNA (GΨA) was well ordered (D). The catalytic domain of one subunit (yellow) recognizes the acceptor stem of a tRNA (blue), while the anticodon-binding domain of the other subunit (green) recognizes the anticodon bases of the same tRNA (blue). The overall structure of the ternary complex of SceTyrRS is similar to that of MjaTyrRS (A and B), which is expected from the amino acid sequence similarity (). In the TthTyrRS structure, on the other hand, a characteristic long variable arm of bacterial tRNA is recognized by an additional C-terminal domain of TthTyrRS (C). Amino-acid sequence alignment of the C-terminal domain of bacterial TyrRSs suggested that the conserved sequences of the C-terminal domains determined a conserved secondary structure (). The structure of the C-terminal domain of BstTyrRS was determined using NMR (), and was found to have a very similar structure to that of TthTyrRS (). Recent advances in genome sequencing revealed that bacterial tyrosyl-tRNA synthetases occur in two large subfamilies; TyrRS and TyrRZ that possess about 25% amino-acid sequence identity (21 and references therein). More detailed functional and structural analyses of the TyrRZ–tRNA complex are necessary to shed more light on the evolutionary divergence of the enzyme–tRNA interactions of the TyrRS and TyrRZ subfamilies in the bacterial domain. As observed in the structures of the ternary complexes of TthTyrRS and MjaTyrRS, SceTyrRS has a class II mode of tRNA recognition, i.e. it interacts with tRNA from the variable loop and acceptor stem major groove side. This is in strong contrast to canonical class I enzymes, which approach cognate tRNA from the acceptor stem minor groove side. In the present structure analysis, the anticodon triplet of tRNA (GΨA) was well ordered (D). The first anticodon, G34, is flipped out and base-specifically recognized by SceTyrRS (A). The guanine ring moiety of G34 shows a stacking interaction with Phe296. The opposite face of the base has hydrophobic contact with Pro320. The N1 and O6 atoms of G34 have base-specific interactions with Asp321 through bifurcated hydrogen bonds. It is reported that mutation of G34 in yeast tRNA impairs aminoacylation by SceTyrRS (). The second and third anticodon bases, Ψ35 and A36, have fewer base-specific interactions with the enzyme; only N3 of Ψ35 hydrogen bonds with the main chain carbonyl group of Cys255 (A). They are accommodated in a hydrophobic patch composed of Phe254, Pro257, Pro319 and Pro320 (D). This observation is also consistent with the results of the functional analysis of SceTyrRS by Fechter . (). It is of note here that the tyrosylation activity is compatible with an ‘a+1 shift’ [ ()] of the anticodon in the 3′-direction (G35-U36-A37) but is strongly inhibited in the opposite 5′-direction (G33-U34-A35) (). This result is explained by the present structure analysis, which shows that the electron density of the anticodon stem region was rather poor: the base moieties 31, 40 and 41 and the whole nucleotides 32 and 33 of tRNA were disordered. In addition, our recent biochemical analysis () also suggests the flexibility of the anticodon stem region of Sce-tRNA. This is consistent with the low G-C content at the anticodon stem region of Sce-tRNA as compared with that of Mja–tRNA. The flexibility in the 5′-portion of the anticodon loop enables the shift of the anticodon in the 3′-direction, which can be regarded as an insertion of one nucleotide before G34. The anticodon recognition mode of archaeal MjaTyrRS (B) is quite similar to that of eukaryotic SceTyrRS (A), except that the hydrophobic interaction between Pro320 and G34 in SceTyrRS is replaced by the stacking interaction between His283 and G34 in MjaTyrRS. The anticodon recognition mode of prokaryotic TthTyrRS (C) is markedly different from those of eukaryotic and archaeal enzymes. In the TthTyrRS complex, the guanine base of G34, base-specifically recognized by Asp259, is stacked with the third anticodon base, A36. The second anticodon base, Ψ35, is oppositely flipped out and base-specifically recognized by Asp423. The specific aminoacylation of tRNA by TysRS relies on the identity determinants (the anticodon bases, the C1-G72 base pair, and the discriminator base A73) in the cognate tRNA (). A previous observation showed that the strongest determinants are base pair C1-G72 and discriminator base A73 for SceTyrRS, while the three anticodon bases (G34, Ψ35 and A36) contribute to lesser extents (). A similar observation was reported for MjaTyrRS (). Before the present crystal structure analysis, the recognition modes of the identity determinants by the archaeal and eukaryotic TyrRSs were expected to be similar to each other but different from that by the bacterial TyrRSs. Interestingly, however, the tRNA recognition modes of SceTyrRS have both similarities and differences compared with those in MjaTyrRS: the recognition of the C1-G72 base pair by SceTyrRS is similar to that by MjaTyrRS, whereas the recognition of the A73 by SceTyrRS is different from that by MjaTyrRS but similar to that by TthTyrRS (). The expected feature is that the recognition mode of the identity base pair C1-G72 by SceTyrRS (A) is similar to that by MjaTyrRS (B). The recognition mode of the C1 base by the side chain of Arg193 in SceTyrRS is equivalent to that by the side chain of Arg174 in MjaTyrRS. The G72 base is recognized by a slightly different manner between SceTyrRS and MjaTyrRS. The 3′-terminal strand of the acceptor stem of SceTyrRS shows a helical conformation. The G72 base in SceTyrRS is recognized by a base-specific hydrogen bond with the side chain of Arg151 and a stacking interaction with the discriminator base A73, which is fixed by a hydrogen bond with Arg193. On the other hand, the 3′-terminal strand of the acceptor stem of MjaTyrRS shows a rather extended conformation. The G72 base is far from the side chain of Arg132, corresponding to Arg151 in SceTyrRS, and is recognized by the side chain of Lys175. The side chain of Arg132 is involved in the recognition of the C1 base, rather than the G72 base, via a water-mediated hydrogen bond (B). Although the G72 base recognition mode of SceTyrRS and that of MjaTyrRS have some differences, the archaeal/eukaryotic TyrRSs recognize the identity base pair C1-G72 by conserved residues () and we assume that the recognition mode of the C1-G72 base pair by the archaeal/eukaryotic TyrRSs are essentially conserved. The unexpected feature is that the recognition mode of A73 by SceTyrRS (A) is similar to that by TthTyrRS (C) but is different from that by MjaTyrRS (B). In the case of SceTyrRS (A), the N3 of A73 is recognized by Arg193 via a single hydrogen bond, and the discriminator base A73 is stacked with the G72 base. An equivalent recognition mode is observed for the TthTyrRS complex (C): the N3 of A73 is recognized by Arg198 via a single hydrogen bond, and the discriminator base A73 is stacked with the C72 base. In the case of MjaTyrRS (B), the discriminator base A73 is unstacked with the G72 base and out of the helical continuity of the acceptor stem. The N1 and N6 atoms of A73 are base-specifically recognized by the main-chain amino and carbonyl groups, respectively, of Val195. Since Arg193 in SceTyrRS (Arg198 in TthTyrRS) is also conserved in MjaTyrRS (Arg174), the observation that A73 is bound in a different manner in SceTyrRS and MjaTyrRS may reflect different modes of binding, rather than species-specific difference. However, it should be noted here that the Arg residues are not conserved in BstTyrRS (Trp196) and EcoTyrRS (Trp201) (). The different recognition pattern of A73 would be observed in BstTyrRS and EcoTyrRS. In the present crystal structure of SceTyrRS, the 3′-CCA terminus of tRNA is well ordered by a triplex stacking interactions of the C74, C75 and A76 bases, but was flipped-out from the active center. Manual model adjustment of the 3′-CCA terminus allows the 2′-OH of the terminal ribose to be correctly positioned for aminoacylation (D). This model is consistent with the fact that TyrRSs preferentially aminoacylate the 2′-OH of A76 in accordance with other class I enzymes, although it can also aminoacylate the 3′-OH (). Thus we assume that the present structure of the acceptor region of Sce-tRNA is not an artifact at least up to the discriminator base A73. As for the binding mode of the 3′-CCA terminus of tRNA, the CCA terminus of MjaTyrRS () and that of TthTyrRS () were disordered, while that of SceTyrRS was flipped out from the active site. Because neither of the available complex structures (SceTyrRS–tRNA-Tyr–AMPN, MjaTyrRS–tRNA-tyrosine and TthTyrRS–tRNA-tyrosinol–ATP) contains true reactive intermediate, Tyr-AMP, the non-productive binding of the 3′-CCA terminus may occur. The present study successfully revealed the recognition modes of Tyrosyl–AMP, the reactive aminoacyl-adenylate (aa-AMP) intermediate, by SceTyrRS, as a result of our use of a Tyr-AMP analog, Tyr-AMPN (A). The Tyr-AMP recognition mode is well conserved among the archaeal, bacterial and eukaryotic TyrRSs as described below. The tyrosine moiety is accommodated in a deep pocket of the enzyme (A). The hydroxyl group of the tyrosine moiety makes hydrogen bonds with the side chains of Tyr43 and Asp177. The main-chain amino group of the tyrosine moiety is specifically recognized by three hydrogen bonds: they are from the side chains of Tyr170, Gln174 and Gln192. The carbonyl group of the tyrosine moiety is recognized by the side chain of Gln192. In archaeal and bacterial TyrRSs (B and C), the recognition mode of the tyrosine moiety of Tyr-AMP is apparently similar to that of SceTyrRS. Tyr170, Gln174, Asp177 and Gln192 in SceTyrRS are well conserved () and play the same roles in the binding of the tyrosine moiety of Tyr-AMP (or its analog) among TyrRSs. In the case of BstTyrRS/TyrAMP complex (), however, the side chain of Gln195 (Gln192 in SceTyrRS) does not form a hydrogen bond with the amino group of Tyr-AMP. Instead, the side chain of Asp78 is involved in the hydrogen bond with the amino group of Tyr-AMP together with the side chains of Tyr169 and Gln173 (Tyr170 and Gln174 in SceTyrRS). As for the recognition mode of the AMP moiety of Tyr-AMPN, the adenine ring is base-specifically recognized by the main-chain atoms of Val219 in SceTyrRS [N6(Ade)—O(Val229) and N1(Ade)—N(Val229), A]. A similar recognition mode is also found for the main chain atoms of Leu224 in TthTyrRS (C). The oxygen atom of the phosphate group of Tyr-AMPN is fixed to the protein surface [Ala47(O), Tyr56(OH), and Tyr101(OH)], via a magnesium ion (A). The conserved signature motifs of class I aaRSs, HIGH and KMSKS (blue and red, respectively, in ), are involved in the catalysis of tyrosine activation with ATP. It is reported that the HIGH motif is involved in binding with the γ-phosphate group of ATP in EcoTyrRS (). In the present structure analysis, the HIGH motif's important role in tyrosine activation is not observed because of the absence of β- and γ-phosphate groups in Tyr-AMPN (A). The KMSKS motif shows conformational changes in tyrosine activation (): initially, the KMSKS motif adopts the open form and then, upon binding of the adenosyl moiety of ATP, shifts to the semi-open form before finally assuming the ATP-bound closed form. In that study, Kobayashi .() assumed that after the amino acid activation, the KMSKS motif adopts the semi-open form to accept the 3′-CCA terminus of tRNA for the aminoacyl transfer reaction. In the present structure of SceTyrRS, residues 224–233, including the KMSAS sequence (), are disordered. This flexibility of the loop containing the KMSKS motif would allow the Tyr-AMP to be fully exposed and the 3′-CCA terminus of tRNA to access the aminoacyl transfer center. Unfortunately, however, the 3′-CCA terminus of tRNA is ordered but flipped out from the active center in the present crystal structure (D). In the case of TthTyrRS (C), the loop containing the KMSKS motif interacts with ATP and is structurally well ordered. In the aminoacylation of tRNAs, each amino acid is matched with a tRNA that contains the anticodon that corresponds to that amino acid. Although the anticodons within tRNAs are conserved for a given amino acid throughout evolution, the aaRS from one species does not aminoacylate its cognate tRNA from another species in some cases. Typical example is TyrRSs that exhibit species-specific tRNA recognition (,). The origin of species-specific tRNA recognition is the presence of a G1-C72 base pair in bacterial tRNA and a C1-G72 pair in archaeal/eukaryotic tRNA (). Because the archaeal/eukaryotic TyrRSs-tRNAs pairs do not cross-react with their bacterial counterparts, the recognition modes of the identity determinants by the archaeal and eukaryotic TyrRSs were expected to be similar to each other but different from that by the bacterial TyrRSs. Interestingly, however, a structural comparison between the present crystal structure of the ternary complex of SceTyrRS with the available crystal structures of the ternary complexes of TthTyrRS () and MjaTyrRS () revealed (i) an unexpected similarity in the recognition mode of the discriminator base A73 between SceTyrRS and TthTyrRS (A and C) and (ii) some differences in the recognition mode of the G72-A73 bases between SceTyrRS and MjaTyrRS (A and B). These features indicate that the interaction mode between TyrRS and the cognate tRNA appears to have evolved separately for the three kingdoms of life, i.e. TyrRSs/tRNAs pairs have diverged after the kingdoms separated (). The present crystal structure analysis of the eukaryotic SceTyrRS and structural comparisons strongly support the notion pointed out by Wakasugi . () that the lack of cross-reactivity between archaeal/eukaryotic and bacterial TyrRS-tRNA pairs most probably lies in the different sequence of the last base pair of the acceptor stem (C1-G72 vs G1-C72) of tRNA. On the other hand, the recognition modes of Tyr-AMP are conserved among the TyrRSs from all three kingdoms (). In the class I aaRSs, the amino acid binding pocket lies at the bottom of an active site cleft in the Rossmann-fold domain. In general, the class I aaRSs use a lock-and-key mechanism to recognize the side chains of their amino acid substrates, although some exceptions exist. Detailed sequence alignment of TyrRSs and TrpRSs as well as the crystal structure analyses of human TyrRS (HsaTyrRS) and human TrpRS (HsaTrpRS) by Yang . () provided a unique example for amino acid discrimination by TyrRS and TrpRS. An environment for recognition of the hydroxyl group of Tyr side chain is provided by the universal presence of aspartate (Asp173 in HsaTyrRS) and the presence of either tyrosine (Tyr39 in HsaTyrRS) or lysine (Lys41 in TthTyrRS) ( and ). Interestingly, TrpRS uses the structurally equivalent residues (either Pro287 or Tyr159, respectively, in HsaTrpRS, but not both) to hydrogen bond to the indole nitrogen of tryptophan. In the case of yeast system [Asp177 and Tyr43 in SceTyrRS (A) and Thr233 and Tyr106 in SceTrpRS(GI:51013347)], the same arguments appear to hold, although crystal structure analysis of SceTrpRS has not yet reported. The present crystal structure analysis of SceTyrRS and structural comparisons of the amino acid binding site of TyrRSs () are consistent with the structural and phylogenetic studies of TyrRS and TrpRS by Yang . (). The archaeal/eukaryotic TyrRSs-tRNAs pairs do not cross-react with their bacterial counterparts (). Such orthogonality is used for the incorporation of unnatural amino acids into proteins with engineered pairs of TyrRSs and tRNAs (,,). We have been trying to utilize the yeast TyrRS/tRNA pair as a candidate for the carrier of unnatural amino acid in the translation system () or (). We previously showed that the substitution of tyrosine at position 43 to glycine (Y43G mutation) in SceTyrRS led to a drastic change in amino acid specificity. The Y43G mutant was found to be able to utilize several 3-substituted -tyrosine analogs, rather than -tyrosine, as substrates for aminoacylation (). Used together with yeast amber suppressor tRNA, the Y43G mutant should serve as an effective tool for site-specific incorporation of 3-substituted tyrosine analogs into proteins in an appropriate translation system (). The present crystal structure analysis can explain the structural basis for recognition of 3-substituted tyrosine analogs by the Y43G mutant SceTyrRS. Since the side chain of Tyr43 is directly involved in binding with the tyrosine moiety of Tyr-AMPN (A), substitution of the tyrosine residue to a smaller residue creates a space to accommodate an extra functional group at position 3 of the substrate. Similar replacements are also reported for EcoTyrRS (,). The proteins containing unnatural amino acids will be used as molecular switches for signaling pathways, as photocrosslinkers, fluorescently labeled probes, or heavy-atom derivatives for phasing in X-ray structure determination. It is reported that human TyrRS is secreted during apoptosis in cell culture and is cleaved with an extracellular elastase, and the two released fragments (the N-terminal ‘mini-TyrRS’ and the EMAP II-like C-terminal domain) are active cytokines (). The mini-TyrRS has an ELR motif in the Rossmann-fold domain. This motif is responsible for IL-8-like cytokine activity and is conserved among segmented animals, whereas it is absent in yeast and lower eukaryote. In the case of yeast, SceTyrRS is inactive as a cytokine and has a NYR motif instead of an ELR motif. Interestingly, Liu . () reported that substitution of the tripeptide NYR to ELR in SceTyrRS resulted in mutant TyrRS with cytokine activity. This result suggests that it is the E and L that are strong candidates for a direct involvement in cytokine receptor binding. However, the Arg side chain appears to be also important for the cytokine activity of TyrRS, because an Arg93-to-Gln mutation in human mini-TyrRS abolished cytokine activity (). A structural comparison between the ELR motif of human mini-TyrRS and the NYR motif of SceTyrRS () reveals that the overall structures around the motifs are similar to each other. Since the side chains of the ELR motif of human mini-TyrRS and the NYR motif of SceTyrRS are exposed, it is quite reasonable that substitution of the tripeptide NYR to ELR in SceTyrRS resulted in mutant TyrRS with cytokine activity.

A serious limitation of the use of many types of synthetic oligonucleotides (ON) and their analogues as therapeutic antisense agents has been their poor cellular delivery (,). Many types of vector have been designed to aid ON delivery both for cell culture and . Amongst such strategies, conjugation to cell penetrating peptides (CPP) has received much recent attention (). In the case of negatively charged antisense ON, the potential of conjugated CPPs for delivery has not been realized, since there are very few publications that have shown significant biological activity (,). Indeed, a recent study with a well-controlled assay dealing with inhibition of -activation of the HIV-1 LTR showed some significant cell internalization of a number of CPP-ONs, but a complete lack of biological activity (). In addition, only very modest biological activity was found for similar CPPs conjugated to synthetic short interference RNA (siRNA) targeted to a P38 MAP kinase mRNA (). A particularly useful HeLa cell assay for assessing the activity of CPP-ONs conjugates in a comparative manner is that established by Kole and colleagues () involving splice correction of an aberrant -globin intron by 16-mer synthetic oligonucleotides (705 site) and subsequent up-regulation of firefly luciferase. This assay is straightforward to carry out and has a very high dynamic range, such that even very low activity levels can be seen as a positive luminescence read-out. CPPs conjugated to ONs that are not negatively charged, such as peptide nucleic acids (PNA) or phosphoramidate morpholino oligonucleotides (PMO) have shown significant promise in splicing correction assays and other steric block applications, for which PNA is particularly suited. For many PNA conjugates, biological activity in this and other splice alteration assays has been observed when the PNA is attached to cationic, amphipathic or other CPP peptides, but concentrations of conjugates in the 5–10 µM range almost invariably have been needed for incubation with cultured cells to see significant splice alteration activity (). Recent studies by our laboratories () and by other groups (,) have demonstrated that a major barrier for nuclear delivery, required for splicing correction, is the release from endocytic vesicular compartments. This was not surprising since, for polycationic CPPs such as Tat, Penetratin, R or K, the vast majority of the material is internalized by an active mechanism of endocytosis, which involves electrostatic interactions with cellular heparan sulphates, and has little access to the nuclear compartment (). Endosomolytic agents, such as chloroquine, calcium ions or high sucrose concentration (,), are necessary to obtain a significant splice correction activity (,), but the use of such agents is difficult to envisage. One possible solution is to complement the CPP with a membrane-destabilizing agent (e.g. viral fusogenic peptide or membrane-destabilizing peptide), such as has been proposed by Dowdy to improve CPP-mediated protein transfection (), or to screen for a new peptide additive that might improve the biological activity of the CPP conjugate. In addition to the increased complexity of such a delivery system and to its cost, we have not been able to find to date a peptide or lipopeptide that showed substantially enhanced steric-block biological activity for a PNA ON conjugated to the Tat peptide (). Likewise the co-incubation of 5 μM HA2–Penetratin fusion peptide with various CPP–PNA constructions had only a moderate effect on splice correction (). We, therefore, concluded that a better approach is to modify existing CPPs in order to search for peptides that may have enhanced intrinsic endosomolytic activity. Two vector strategies have been adopted, both taking into account the key roles played by Arg side chains in CPP uptake. We recently showed that (R-Ahx-R)–PMO705 conjugate had significant splicing correction activity in the luciferase up-regulation model at 1 µM concentration in the absence of an endosomolytic agent (). Similarly we showed that a (R-Ahx-R)–PNA705 conjugate also had significant splice correction activity at 1 µM concentration (). In parallel studies, we found substantial activity in an HIV-1 -activation inhibition assay (also requiring nuclear delivery) when a derivative of the known CPP Penetratin, in which six Arg residues were added to the N-terminus of the CPP, R-Penetratin (R6Pen), was disulfide-conjugated to a PNA complementary to the -activation responsive element RNA (). We show now that this Arg-modified CPP when conjugated to a PNA targeted to the luciferase splice correction site shows by far the highest up-regulation of luciferase at both protein and RNA levels at 1 µM concentration compared to all previous CPPs studied. We also begin to characterize some aspects of the structure–function relationship and show that, for example, a W→L mutant that was reported to substantially reduce the cell penetration of Penetratin peptide () does not reduce the splice correction activity of the R6Pen–PNA conjugate. These results show that R6Pen might be a very good lead CPP towards further development of a suitable PNA–peptide conjugate candidate for studies. N-terminal nitropyridyl (Npys) cysteine-containing PNA oligonucleotides with additional lysine residues were synthesized on an Apex 396 Synthesizer by the Fmoc/Bhoc method as previously described (,) to give the general structure NH-Cys(NPys)-Lys-PNA-(Lys)-amide. PNA705 antisense is CCTCTTACCTCAGTTACA and PNA705 scrambled sequence is CCTGTTATACCACTTACA. Note that we have found recently that higher overall synthesis yields are obtained when the final deprotections are carried out in the absence of phenol scavenger. In some cases, N-terminal Cys-containing PNA was obtained from Panagene () and activated with dipyridyldisulfide (Pys2) as follows. To the PNA (500 nmol) was added 150 µl Pys2 (6.75 µmol, 13.5 eq.) in DMF (10 mg ml), 15 µl 2 M triethylammonium acetate solution (pH 7) and 135 µl water. After standing for 1 h the solution was loaded on to a Sephadex NAP-10 column and eluted with 0.1% TFA solution, collecting the excluded volume. This solution was used directly in conjugation after quantification by measurement of the absorbance at 260 nm. Npys and Pys activated PNA could be used interchangeably in the conjugation reactions to form disulfide linkages. Stably Linked K-PNA705 [NH-(Lys)-CCTCTTACCTCAGTTACA-Lys-amide] and Tat-PNA705 [NH-Gly-Arg-Lys-Lys-Arg-Arg-Gln-Arg-Arg-Arg-Pro-(O-linker)-CCTCTTACCTCAGTTACA-amide] peptide–PNA conjugates were synthesized by continuous PNA/peptide synthesis as previously described (,). An O-linker was added with an Fmoc-AEEA spacer (Applied Biosystems). The α-N-bromoacetyl-Lys-PNA-(Lys)-amide (both 705 and scrambled 705) were obtained from Panagene (Korea). MALDI-TOF mass spectrometry was carried out on a Voyager DE Pro BioSpectrometry workstation with a matrix of α-cyano-4-hydroxycinnamic acid, 10 mg ml in acetonitrile/3% aqueous trifluoroacetic acid (1:1, v/v). The accuracy of the mass measurement in linear mode is regarded by the manufacturer as ± 0.05%, but since internal calibration was not used, the determined values varied in a few cases from the calculated by ± 0.1%. underline italic xref #text I n a t y p i c a l c o n j u g a t i o n r e a c t i o n , 5 0   n m o l b r o m o a c e t y l P N A w a s d i s s o l v e d i n 4 5   µ l f o r m a m i d e a n d 1 0   µ l B i s T r i s – H B r b u f f e r ( p H 7 . 5 ) a n d 1 5 . 6   µ l C - t e r m i n a l - C y s c o n t a i n i n g p e p t i d e ( 8   m M , 1 2 5   n m o l . 2 . 5 e q . ) w a s a d d e d . T h e s o l u t i o n w a s h e a t e d a t 4 0 ° C f o r 2   h a n d t h e p r o d u c t w a s p u r i f i e d b y r e v e r s e d p h a s e H P L C a t 4 5 ° C u s i n g w a t e r b a t h h e a t i n g a n d a n a l y s e d b y M A L D I - T O F m a s s s p e c t r o m e t r y ( S u p p l e m e n t a r y T a b l e S 1 ) . #text This was carried out similarly to that described previously (). The conjugates () were incubated for 4 h in 1 ml OptiMEM medium with exponentially growing HeLa pLuc705 cells (1.75 × 10 cells/well seeded and cultivated overnight in 24-well plates). The conjugates were then diluted with 0.5 ml complete medium (DMEM plus 10% fetal bovine serum) and incubation continued for 20 h. Cells were washed twice with ice-cold PBS and lysed with Reporter Lysis Buffer (Promega, Madison, WI, USA). Luciferase activity was quantified with a Berthold Centro LB 960 luminometer (Berthold Technologies, Bad Wildbad, Germany) using the Luciferase Assay System substrate (Promega, Madison, WI, USA). Cellular protein concentrations were measured with the BCA™Protein Assay Kit (Pierce, Rockford, IL, USA) and read using an ELISA plate reader (Dynatech MR 5000, Dynatech Labs, Chantilly, VA, USA) at 550 nm. Levels of luciferase expression are shown as relative light units (RLUs) per microgram protein. All experiments were performed in triplicate. Each data point was averaged over the three replicates. To analyse the cell permeabilization of CPP–PNA conjugates, exponentially growing HeLa pLuc705 cells (3 × 10 cells seeded and grown overnight in 30 mm plates) were incubated for 4 h with the CPP–PNA705 conjugates at different concentrations. The cells were then washed twice with PBS, detached by incubating with trypsin for 5 min at 37°C (0.5 mg ml)/EDTA.4Na (0.35 mM), and washed by centrifugation (5 min, 900 × ) in ice-cold PBS containing 5% FCS. The cell pellet was resuspended in ice-cold PBS containing 0.5% FCS and 0.05 µg/ml propidium iodide (PI) (Molecular Probes, Eugene, OR, USA). Fluorescence analysis was performed with a BD FacsCanto flow cytometer (BD Biosciences, San Jose, CA, USA). A minimum of 20 000 events per sample were analysed. HeLa pLuc705 cells were plated at 30 000 cells/well in a 24-well plate 24 h before treatment. After overnight incubation, the cells were washed with PBS and incubated in 1 ml OptiMEM containing 1 µM of the indicated conjugates (naked PNA705, Pen-s-s–PNA705, R6Pen-s-s–PNA705, R6Pen-s-s–PNA705scr or R6Pen(W–L)-s-s–PNA705) for 4 h and the conjugates were then diluted with 0.5 ml of DMEM containing 10% FCS and allowed to grow for 20 h. Cells were then washed twice with PBS. Total RNA was extracted from the cells using the High pure RNA isolation Kit (Roche Applied Science). The extracted RNA was examined by RT-PCR (MJ Research PTC200 Peltier Thermal cycler) with forward primer 5′TTG ATA TGT GGA TTT CGA GTC GTC3′ and reverse primer 5′TGT CAA TCA GAG TGC TTT TGG CG3′. The products were analysed on a 2% agarose gel (A). For dose-dependence experiments (B), cells were treated as described above with increasing concentrations of R6Pen-s-s–PNA705 or R6Pen–PNA705 conjugates. After carrying out the luciferase assay and BCA™ Protein Assay, the remaining cell lysates (about 270 µl) were transferred into 2 ml microfuge tubes and total RNA was extracted with 1 ml TRI Reagent (Sigma). Minor changes to the manufacturer's protocol were made to accommodate the presence of Reporter Lysis Buffer. Thus, 0.3 ml of chloroform was used for extraction and the amount of iso-propanol for RNA precipitation was increased to give a 1:1 mixture with the aqueous phase. The RT-PCR was carried out as described above and agarose gels were scanned using Gene Tools Analysis Software (SynGene, Cambridge, UK). shows a comparison of the splice correction activities at 1 µM concentration of unconjugated PNA705, K8-PNA705 and Tat-PNA705, the activity of each of which is known to be chloroquine-dependent (,,,), together with R6Pen–PNA705 and (R-Ahx-R)4–PNA705 () in the absence of an endosomolytic agent. In all cases, PNAs were conjugated to the carrier peptides through stable amide or thioacetyl linkages ( for construct details). R6Pen conjugate, and to a lesser extent (R-Ahx-R)4 conjugate, gave rise to a strong up-regulation of luciferase under conditions where K8 and Tat peptide conjugates were essentially inactive. Note that the scale of light units is shown in relative light units per microgram protein, demonstrating the very high level of activity seen for R6Pen–PNA705. The low level of activity for Tat-PNA705 agrees with results recently reported by two other laboratories, where similarly low splice correction was seen also for Penetratin, R and Transportan at 1 µM concentration (,) and only at 5–10 µM concentrations did some conjugates (notably Transportan) show significant splice correction activity. Thus, R6Pen appears substantially more effective as a CPP and leads to much stronger splice correction activity compared to our previously used (R-Ahx-R)4–PNA705. The splice correction activity of the R6Pen conjugate is sequence-specific, since no splice correction activity is seen when this CPP vector is conjugated to a scrambled version of PNA705. Note that luciferase activity levels vary somewhat between experiments as pointed out by Bendifallah . (). Normalization of the data to the basal luciferase expression in untreated cells, as proposed by these authors, gives rise to much less apparent variation between experiments (see Supplementary Data, and ), but we have chosen here to show un-normalized values just to demonstrate the high activity levels. To characterize further the properties of the R6Pen–PNA705 conjugate, we monitored the dose-dependence of splice correction, as measured by luciferase up-regulation, at concentrations between 0.1 and 2.5 µM (). R6Pen–PNA705 allows an efficient dose-dependent splice correction activity in the absence of chloroquine (, white bars) under conditions where no toxicity was seen, as judged by measurement of PI uptake by flow cytometry (Supplementary Data, ). The proportion of permeabilized cells remained <3% as compared to the untreated controls in cells incubated with the various CPP–PNA conjugates at 1 μM (e.g. at the concentration allowing almost complete splicing correction). The addition of chloroquine improved the splice correction activity, which demonstrates that some of the conjugate still remains entrapped in endosomal compartments in keeping with an endocytotic mechanism of cell uptake. However, the incremental improvement in splice correction activity afforded by chloroquine addition was somewhat smaller at the higher concentrations (approximately 2- to 3-fold, , grey bars), than those we obtained previously with K8 and Tat conjugates of PNA or PMO, where a 10-fold increase or more was often observed (,). We next investigated the importance of the stability of the linkage between the delivery peptide and the PNA cargo. It has been suggested by others that if a disulfide-linked conjugate is able to escape from the endocytic compartments and reaches the cytosol, the disulfide bridge might be reduced, thus allowing free PNA to be released (). A new conjugate R6Pen-s-s–PNA705 () was therefore constructed with a linker containing a disulfide bridge, similar to that which we have previously used in studies of HIV-1 Tat-dependent -activation inhibition (). This conjugate was tested in the splice-correction assay in parallel with the stably linked R6Pen–PNA705 and indeed showed a slightly (but reproducibly) higher activity (). However, the relatively small difference demonstrates that the nature of the linkage is not a principal factor governing splice correction activity. Nevertheless, we decided to use the more active disulfide-bridged conjugates for further studies on the structure–function relationship. In order to determine the effect of the N-terminal Arg stretch on splice correction activity, we constructed a series of R(z)Pen-s-s–PNA705 conjugates with {0, 3, 6 and 9}. These R(z)Pen-s-s–PNA705 conjugates were tested at 0.5 and 1 µM in the splice correction assay in the absence of chloroquine (). Pen-s-s–PNA705 at 1 µM displays only a very weak activity, consistent with previous results of others (,). The activity level is strongly enhanced by the addition of an Arg tail by factors of 16, 43 and 28 for = 3, 6 and 9, respectively. Thus, at 1 µM concentration, the optimum activity is obtained for R. No significant differences were seen in cell toxicity for any of the conjugates at this concentration as judged by flow cytometry and PI uptake (Supplementary Data, ). Previous studies () have shown that the substitution of the tryptophan residues that occurs naturally in the homeodomain helix 3 sequence by a leucine residue decreased the cell internalization of Penetratin peptide. Surprisingly, the R6Pen(W–L)-s-s–PNA705 conjugate displayed a slightly higher splicing correction activity than the unmodified R6Pen–PNA705 (). This indicates that the Penetratin part of the R6Pen–PNA conjugate has a completely different effect in enhancement of membrane permeabilization when it is located within the PNA conjugate context as compared to the Penetratin peptide alone. In most studies using the HeLa-pLuc705 model, splice-correction is monitored by the quantification of luciferase luminescence activity (,,,). However, this assay gives only a relative appreciation of splice correction activity between different conjugates. In contrast, use of RT-PCR allows the evaluation of the completeness of splice correction by comparison of the amounts of uncorrected and corrected mRNA, as has been used with this splice correction assay for cationic lipid-based transfection methods (,). We, therefore, carried out RT-PCR on RNA samples extracted from HeLa-pLuc705 cells incubated with various conjugates (A). As expected, no RT-PCR products corresponding to the correctly spliced mRNA were detected in cells treated with 1 µM of free PNA705, Pen-s-s–PNA705, or scrambled control R6Pen-s-s–PNA705sc, as seen in lanes 1, 2 and 3, respectively. In contrast, a very high proportion of correctly spliced mRNA was found in cells treated with 1 µM R6Pen-s-s–PNA705 (lane 4) or with R6Pen(W–L)-s-s–PNA705 (lane 6). The dose-dependences of splice correction for R6Pen-s-s–PNA706 and stably linked R6Pen–PNA705 were assessed by the RT-PCR assay (B). The ECs of splice correction at the RNA level were estimated as 0.7 ± 0.3 µM and 1.0 ± 0.3 µM, respectively. ECs were also estimated from the amounts of conjugate required to raise the luciferase luminescence levels to 50% of the observed plateaux values (data not shown). These values were found to be 0.9 ± 0.2 µM and 1.0 ± 0.2 µM, respectively. The nuclear delivery of steric-block ON analogues conjugated with most CPPs for splice correction or exon skipping has been hampered by endosome trapping, unless an endosome disturbing drug or peptide is added, or high CPP–PNA conjugate concentrations are used. Bearing in mind the key role played by cationic amino acids for CPP uptake, we have appended varying numbers of arginine residues to the N-terminal end of Penetratin, a CPP which by itself does not impart on the PNA a significant amount of splice correction ability. R6Pen turned out to be the most active. The level of activity obtained for splice correcting conjugated PNA is higher than for all other CPPs tested to date, including the recently described (R-Ahx-R)4 vector (,). Remarkably, R6Pen–PNA705 conjugates are highly active at 1 µM concentrations in the absence of any endosomolytic agents. Quantification of luciferase expression, as carried out here and also in most published work to date, is a sensitive and convenient assay, which allows one to compare several conjugates quickly in terms of efficiency or specificity, and is thus the method of choice for structure–activity relationships studies. However, such data are expressed in relative light units and do not allow direct determination of the extent to which aberrant splicing has been corrected. RT-PCR products from the aberrantly and correctly spliced luciferase pre-mRNA can be separated easily by agarose gel electrophoresis, thus allowing evaluation of the extent of splice correction under various conditions. RT-PCR data closely parallel luciferase luminescence measurements and indicate that the R6Pen-ss–PNA705 and the W→L variant allow sequence-specific splicing correction at 1 μM concentration to a high level (about 60–70%), whilst PNA705 alone or Pen-s-s–PNA705 are totally inactive. The levels of activity we have obtained (EC of 0.7–1.0 µM) now start to approach those obtained with the same assay by cationic lipid transfection using leashed PNA or other modified ON types (,). The achievement of a fair proportion of correction at low conjugate concentration is a key issue in the development of steric block ONs as potential therapeutics. By use of PI as an index of membrane permeabilization, we have indeed verified that R6Pen did not perturb membrane integrity of HeLa cells at the active dosage. Previous studies from our group have established that high (>5 μM) concentrations of CPP–ON as R or K–ON led to significant increase of PI uptake thus precluding further developments (). We have no explanation at this stage for the dramatically increased splice correction activity of R6Pen as compared to Pen or as compared to several Arg-rich CPPs. It is worth emphasizing in this respect that the W→L mutation in the Penetratin moiety, which is known to inhibit Penetratin peptide uptake (), does not affect splice correction by R6Pen–PNA705 and instead gave rise to a slightly higher activity ( and ), thus inferring different mechanisms by which this CPP operates. Along the same lines, chloroquine has a significantly lower effect on splice correction by R6Pen–PNA705 () as compared to Tat-PNA705 () or K8-PNA705 (), in keeping with its improved intrinsic endosomal escape. We are also able to rule out significant effects of the Lys residues on the PNA part on splice correction activity. Indeed we have found recently that R6Pen disulfide linked to a PNA 18-mer containing just one Lys residue on each end behaved identically to the corresponding conjugate containing four Lys residues (data not shown). Further mechanistic studies are in progress, but it should be noted that we have deliberately avoided on these conjugates the use of fluorescent labels, which are commonly used to track cellular uptake by confocal microscopy. Such labels alter the hydrophobicity of the conjugate at a particular region. This may alter the ability of the PNA-peptide to be released from endosomal compartments. Concerns about this have emerged recently in the case of our parallel studies on inhibition of HIV-1 Tat-dependent -activation (). We have been unable so far to construct a conjugate that contains a fluorescein label on the PNA part of a R-Penetratin–PNA conjugate targeted to TAR without losing all intra-nuclear inhibition activity in the absence of chloroquine in our HeLa cell assay (Turner, J.J., Arzumanov, A.A., Ivanova, G.D. and Gait, M.J., unpublished results). Further, there does not appear to be a strong correlation of the amount of fluorescent oligonucleotide reagent seen to be taken up by cells and their biological activity (,,,), as has also become apparent in the design of lipid-based reagents for delivery of siRNA (). Thus, more sophisticated ways of tracking locations of nucleic acids-based reagents and determining the precise compartments where activity takes place will be needed before such types of experiment become fully meaningful. Whether CPP delivery peptides and their cargoes should be conjugated through stable or unstable linkers has often been debated, but few direct comparisons have been provided. In our case, a disulfide-linked conjugate was slightly (but reproducibly) more active than a stably conjugated PNA. Thus, we are now in the process of carrying out further more detailed structure–function analyses using such disulfide linkers to try to understand how the various parts of the R6-Penetratin peptide contribute to obtaining intra-nuclear splice correction activity. The disulfide linker strategy may also be less susceptible to problems arising from steric interference by the conjugated delivery vehicle, or from potential non-specific binding of the vector to non-targeted entities. However, use of PNA–peptide conjugates may require a more stable linkage and our work shows that a thioacetyl linker is also compatible with high-level splice correction activity. The fact that strong splicing correction (as judged by the RT-PCR analysis) can be achieved at much lower (1 μM) concentration of the correcting ON than has previously proved possible opens up promising perspectives for applications. We hope that further optimization of the peptide–PNA construct will lead to a construct suitable for studies, and eventually for instance towards the treatment of disease-associated splicing defects [cancer, thalassemia, etc. ()] or in exon-skipping strategies, as are now being considered for the treatment of Duchenne muscular dystrophy (,). p p l e m e n t a r y d a t a a r e a v a i l a b l e a t N A R O n l i n e .

A conserved structural blueprint for building a ribosome might be anticipated given the common function of ribosomes in protein synthesis. Numerous studies have revealed the evolutionarily conserved secondary structure of ribosomal RNA (rRNA) across species [reviewed by ()]. Among ribosomal protein constituents, several families of highly conserved proteins are recognized between evolutionarily divergent groups [e.g. ()]. Primary rRNA-binding proteins are within the group of structurally conserved proteins that recognize highly conserved features of rRNA (,). Other ribosomal proteins are less well-conserved in structure between lineages. Variability in ribosomal protein composition contributes to diversity in ribosome composition observed in many organisms. Thus, within a given species evolutionarily conserved and structurally divergent ribosomal components may comprise the translation apparatus. An important ribosomal protein family is defined by protein EL23. Numerous structural equivalents have been recognized in many organisms, identified as ‘L23’ in most prokaryotes, ‘L25’ in yeasts and as ‘L23a’ in most eukaryotes, including insect lineages discussed in this paper. Ribosomal protein L23(a)/L25 is an essential protein, binding to precursor large subunit rRNA early in ribosome assembly (). The multifaceted roles of L23 become increasingly apparent; the protein functions in early assembly events as well as in early and late stages of rRNA processing in yeast (,,). After the completion of ribosome maturation, L23 later assumes a co- and post-translational role as the nascent polypeptide emerges from the ribosome. L23 interacts with several protein folding and targeting apparatus components, including the chaperone trigger factor (TF) in , signal recognition particle, Sec61 translocation channel, and the nascent polypeptide-associated complex (NAC) (). Yeast L25 is the most widely studied eukaryotic member of the L23(a)/L25 protein family. Three distinct functional domains, essential for rRNA maturation, have been identified within the yeast protein: an N-terminal region which harbors the nuclear localization signal (NLS) (), a central domain required for rRNA binding (,), and a C-terminal region which is required for 60S large subunit assembly (). Although L25 is reportedly involved in early and late rRNA processing steps, a specific function for L25 during rRNA processing has yet to be defined. L23(a)/L25 binds within Domain III of 23S–28S rRNA, near the peptidyl transferase center of the ribosome (,,). Within the large ribosomal subunit, L23 assumes a prominent position on the subunit surface at the exit tunnel for the nascent polypeptide, based on structural studies of the ribosome (,), residing next to proteins L29, L19 and L39e (,). The interior surface of the exit channel consists primarily of rRNA domains. The proximity of L23 to the nascent polypeptide and the membrane translocon, and its documented interactions with several protein folding and translocation components (), favor the proposal that L23 also plays a key role in targeting inner membrane proteins (). Numerous L23(a)/L25 homologues have been identified, each with the signature rRNA-binding domain and the nearly invariant RNA-binding motif KKAYVRL, found in the C-terminal portion of the protein. Several studies have confirmed the ability of L23 family members to interact with 25S–28S rRNA-binding sites from different organisms, demonstrating that a core of conserved interactions must exist between L23(a)/L25 proteins and the rRNA-binding site [e.g. (,)]. In two cases, L23a (L23aA) () and rat L23a () have both been confirmed as functional homologues of yeast L25 through their abilities to rescue yeast strains grown under nutritional conditions in which the endogenous L25 gene was not expressed. In each case, the plant and mammalian homologues are structurally similar to yeast L25 (). The L23 member (called L23a) has been identified as such due to structural similarity in the C-terminal rRNA-binding domain with known L23(a)/L25 family members (). Although all known eukaryotic members of this protein family have an N-terminal extension of variable length, harboring the NLS (,,), the proposed L23a homologue (and several other L23a homologues discussed in this paper) is even more structurally divergent in the N-terminal region. Fruit fly L23a carries an extra domain of approximately 135 amino acids (aa) with similarity to histone H1 (), extending its overall size to 277 aa (33; accession NP_523886) compared to the yeast L25 protein of 142 aa (34; accession P04456; and ). While it has been presumed that fly L23a is the functional equivalent of other L23(a)/L25 proteins based on C-terminal half homology, no previous studies have confirmed this. In order to test the function of L23a and to gain insight into the interchangeability of L23 ribosomal components, we examined whether L23a could function in yeast. We have created a yeast strain that is completely dependent on fly L23a for survival, providing the first report of an L23a protein containing a novel domain as a functional member of the L23(a)/L25 ribosomal protein family. Within the yeast strain, pulse-chase experiments demonstrate that the most significant difference in rRNA maturation kinetics is a delay in a late step(s) that converts 27S rRNA into mature 25S rRNA, likely contributing to the slow growth phenotype of the strain. A search of available databases shows that the extended L23a insertion is not unique to L23a. Rather, multiple insect species have varying sizes of insertions. Given the functional homology of L23a to yeast L25, demonstrated in this study, we suspect the other insect genes are functional as well. This would suggest a remarkable tolerance of L23a to novel insertions within this essential ribosomal component. L23a amino acid sequences were aligned pairwise with L23a using the database and BLASTP 2.2.13 software (). Following alignment of the more highly conserved C-terminal domain of L23a proteins using default SEG filtering in the BLASTp subprogram, unaligned amino acids in the N-terminal domains of L23a proteins were aligned pairwise by removing default SEG filtering in the BLASTp subprogram, allowing for the greatest degree of similarity between N-terminal domains. Multiple amino acid sequence alignments were performed with Clustal W software (). The yeast strains and oligonucleotides used in this study are detailed in and , respectively. For most analyses strains were grown in synthetic complete (SC) medium or rich medium (YP) containing 2% dextrose as a carbon source, as described (). The pL23a-FLAG plasmid was derived from BIT700 (), a YCplac33-based plasmid () containing yeast fused to a FLAG-(His6) epitope (FH), whose expression is controlled by the promoter and terminator regions. The gene and 5′ regulatory sequences were excised from BIT700 as a BamHI-BamHI fragment and replaced with a BamHI-BamHI fragment containing the L23a cDNA fused to the FLAG epitope. The BamHI L23a-FLAG fragment was generated by PCR using an L23a cDNA as a substrate, oligonucleotides OCR1 (forward) and OCR2 (reverse) as primers () and HotStarTaq polymerase (Qiagen). OCR2 encodes the FLAG sequence fused in frame to the L23a sequence just 5′ of the stop codon. Next, the promoter and 5′ regulatory sequences were inserted into the vector's SacI and SmaI sites as a SacI-blunt ended fragment. This promoter fragment, generated by yeast colony PCR on YCR9 () using OCR3 (forward) and OCR4 (reverse) as primers () and HotStarTaq polymerase (Qiagen), contains the 420 bp sequence immediately 5′ to the translational start codon, and includes both RPG boxes previously reported as required sequence elements for L25 gene transcription (). A DNA fragment containing sequence disrupted by the gene was obtained by yeast colony PCR on YCR10 using OCR3 (forward) and OCR5 (reverse) as primers () and HotStarTaq polymerase (Qiagen). The resulting fragment is comprised of 420 bp of 5′ regulatory sequence, the coding region disrupted by , followed by 367 bp of 3′ regulatory sequence. The fragment was introduced into YCR11 () through transformation and colonies were selected on synthetic complete media lacking leucine (SC-LEU). The colonies were screened for disruption of the locus through colony PCR using OCR3 and OCR5 as primers, resulting in strain YCR16 (). Overnight cultures of strains YCR10, YCR11 and YCR16 () grown in SC-URA media were diluted and added to 100 ml SC-URA in sidearm flasks to give an OD between 0.07 and 0.16. OD readings were measured spectrophotometrically at approximately half hour intervals for a period of 10 h in order to determine doubling times for each strain during logarithmic growth. Crude cellular lysates from strains YCR9, 10, 11, 13 and 16 () were prepared as previously described (). Crude lysates were fractionated by 15% SDS–PAGE. Duplicate gels were run: one gel was stained with Coomassie Blue and the other was used for immunoblotting. Arrays of protein markers, including Biorad Precision markers were used. Proteins were transferred by electroblotting onto Schleicher and Schuell Optitran 0.2 μm nitrocellulose membranes. For Western blot analysis, filters were blocked in 1XPBS, 10% non-fat dry milk, 0.3% Tween 20® (Sigma), and then incubated with anti-FLAG M2 mouse monoclonal antibody (Sigma) at a concentration of 10 μg/ml. FLAG-tagged proteins were detected using goat anti-mouse IgG whole molecule alkaline phosphatase conjugate affinity-isolated antibody (Sigma) at a 1:30,000 dilution. Yeast L25 was detected using rabbit anti-yeast L25 (a generous gift from A. Faber and H.A. Raué, Vrije Universiteit, Amsterdam) at a dilution of 1:10,000 and secondary antibody goat anti-rabbit IgG alkaline phosphatase conjugate (Sigma) at a dilution of 1:30,000. Blots were developed using BCIP/NBT (Sigma) as a substrate. Ribosomes were purified from crude extracts derived from YCR10 (containing plasmid-encoded FLAG-tagged L25 and a disruption of chromosomal L25 by LEU2) and YCR16 (containing plasmid-encoded FLAG-tagged L23a with endogenous L25 disrupted by LEU2) strains using an anti-FLAG M2 affinity gel resin (Sigma) according to the established procedure of Inada . (). Crude extracts from strain YCR9 were used as a negative control for affinity purification as this strain lacks any FLAG-tagged proteins. Carboxyl terminal FLAG bacterial alkaline phosphatase (Sigma) at a concentration of 5 mg/ml was used as a positive control to verify the affinity purification protocol. FLAG peptide (Sigma) at a concentration of 100 μg/ml was used to elute bound protein complexes from the anti-FLAG M2 affinity gel resin. Ribosome-associated proteins from affinity-purified ribosomes were fractionated by 15% SDS–PAGE and visualized by silver staining (Biorad Silver Staining Plus). RNA was isolated from mid-log phase yeast YCR cells using a mechanical disruption method detailed in the Qiagen RNAeasy protocol (Qiagen, Valencia, CA). Column-purified RNA was subjected to DNase treatment using RQ1 RNase-free DNase (Promega, Madison, WI) at 37°C for 30 min. Following phenol/chloroform/isoamyl (25:24:1, v/v/v) extraction and ethanol precipitation, RNAs were either resuspended in RNase-free water for immediate use in RT-PCR or were stored as pellets at −70°C for later use. RNA was extracted from adult flies using a guanidine hydrochloride procedure (). Ethanol-precipitated fly RNA was stored at −70°C. RT-PCR (50 μl) was performed using the Qiagen One-Step RT-PCR kit. Reactions were incubated sequentially for 30 cycles at three specific temperatures in order to achieve template denaturation (94°C), primer annealing (55°C) and primer extension (72°C). Forward OVW1 and reverse OVW2 primers () were used to generate an L23a-specific H1 domain PCR product of 460 bp. RT-PCR products were analyzed on 2% agarose gels. YCR9 () cells were grown at 30°C in 10 ml of SD-ura media, supplemented with uracil (final concentration 200 μg/ml), to an OD of ∼0.4–0.5. YCR16 () cells were grown at 30°C in 10 ml of SD-ura media to an OD of ∼0.4–0.5. Cells were pelleted and then resuspended in 1.5 ml of the appropriate media. To each set of cultures, 200 μCi of [5,6 H]-uracil (GE Healthcare) was added. Following a 5-min labeling period, 4 ml of SD-ura media supplemented with an excess of uracil (final concentration 20 mg/ml) were added and 0.5 ml samples collected at various time points after the addition of media containing cold uracil. RNA was isolated using a LiCl extraction method () with volumes adjusted to accommodate the smaller culture volumes used in labeling. Approximately equal counts per minute (∼40,000 cpm) per sample (based on liquid scintillation counting) were loaded onto a 1.2% formaldehyde–agarose gel to fractionate newly synthesized RNAs. H-labeled RNAs were transferred onto 0.2 μm Optitran nitrocellulose membrane, sprayed with EnHance (PerkinElmer), and exposed to Kodak BioMax XAR film. Total RNA was extracted from log phase cells, fractionated on a 1.2% formaldehyde–agarose gel and blotted onto a 0.2 μm Optitran nitrocellulose membrane. Filters were hybridized sequentially with P-labeled oligonucleotide probes complementary to yeast 25S, 18S or PGK1 () used as a control for loading, according to hybridization conditions outlined in Sambrook . (). RNAs were visualized by phosphorimaging. We are generally interested in L23a structural variation and what effect(s), if any, this variation may have on L23a function in a subset of lineages where the protein binding site within 28S rRNA is specifically cleaved (within the D7a expansion segment), producing a ‘hidden break’ or ‘gap’ within the rRNA [e.g. ()]. A large number of L23a protein sequences are now available in databases; however, only a few such sequences are available from organisms where 28S rRNA ‘gap’ processing has been documented. The proposed L23a ribosomal protein features the C-terminal rRNA-binding signature KKAYVRL () and an N-terminal extension with homology to the C-terminal region of histone H1 (). Proposed L23a proteins from other organisms where gap processing has been documented [e.g. : (46; accession AAV34835); sp.: (47; accession AAP06228); s: (48; accession AAY41436)] or where gap processing is likely to occur based on taxonomic relatedness between organisms [e.g. (accession XP_393135), (accession AAY41435) and (accession XP_316083; EAA11004)] also show C-terminal amino acid conservation and larger N-terminal extensions than are characteristic for most eukaryotic L23a proteins (; ; Supporting ). Within the N-terminal region (; Supporting ), amino acid sequence divergence is more pronounced; yet, some similarity is noted particularly in repeated stretches of basic amino acids within insect lineages but not in , suggesting a common evolutionary origin for the extra protein domain among the insect lineages (). Interestingly, preliminary analyses suggest a coincident increase in the size of the L23a N-terminal domain in some lineages and the structural complexity of the 28S rRNA D7 expansion segment within the L23a-binding site, suggesting co-evolution of L23a proteins and the 28S D7a expansion segment as well (Ware and Mendelson, in preparation). To detect and analyze L23a protein expression in yeast, we designed a construct that expresses L23a with a FLAG epitope tag fused to its C-terminus (pL23a-FLAG). This particular tagging scheme was chosen because it has been shown that the addition of a FLAG tag to the C-terminus of yeast Rpl25p does not inhibit its function (). We used this L23a-FLAG construct to transform a yeast strain (YCR9) wild-type for genomic , to generate strain YCR11 (). Next, we disrupted the genomic locus in strain YCR11 through homologous recombination with a DNA fragment containing the gene and flanking L25 gene sequences to generate strain YCR16 (). Colony PCR analysis using OCR3 and OCR5 primers (), directed against the locus confirmed that strain YCR16 carries a disruption of genomic , as does strain YCR10 (; see ). Since L25 is an essential protein in , one prediction about YCR16 is that it should require the L23a-FLAG plasmid which contains a marker for growth. We tested this prediction by plating strains YCR10, YCR11 and YCR16 onto media containing 5-fluoroorotic acid [5-FOA; ()], which is converted to the toxic product, fluorodeoxyuridine when yeast cells express . Under these plating conditions, growth was inhibited in strains YCR10 and YCR16, indicating that neither strain could lose the plasmid and remain viable (). In order to rule out the possibility that sequence from the locus had replaced L23a sequence on pL23a-FLAG through homologous recombination, we rescued the pL23a-FLAG plasmid from YCR16 and analyzed it by digestion with several restriction enzymes. Comparison of restriction fragment sizes with the original vector showed no alteration in the plasmid (data not shown). Sequencing of the L23aFLAG insert within pL23a-FLAG showed three nucleotide changes that alter the deduced amino acid sequence at three positions relative to the GenBank L23a sequence (accession number NP_523886). All three amino acid changes occur at positions where amino acid variation is present within L23a sequences (Supporting ). Although this L23a variant may have arisen due to errors induced in RT-PCR by Taq DNA polymerase, it is likely that the variant is a naturally occurring form. From numerous independent isolations of cloned PCR products derived from RT-PCR experiments using different RNA preparations, we have only been able to recover the canonical version of L23a (accession number NP_523886) or the variant described here containing all three amino acid changes (accession number DQ450529). With either explanation, it is clear that the variant encodes a functional protein. Confirmation that pL23a-FLAG was expressed within strain YCR16 was obtained by Western blot analysis using an anti-FLAG M2 mouse monoclonal primary antibody followed by detection using goat anti-mouse IgG whole molecule alkaline phosphatase conjugate. Initial analyses clearly showed the presence of FLAG-tagged L23a protein within strain YCR16 (a and b). FLAG-tagged yeast L25 protein was also readily detected within strain YCR10 (a and b). However, no FLAG-tagged L23a protein was detected within strain YCR11 containing the pL23a-FLAG plasmid and an intact chromosomal copy of L25 (a and b). Our inability to detect L23a-FLAG protein within strain YCR11 was confirmed in several instances (see a and b, for example). Interestingly, yeast FLAG-tagged L25 protein was detected in strain YCR13 that carries a plasmid encoding L25-FLAG along with an intact chromosomal copy of L25 (b). Protein expression levels for L25 and L25-FLAG in strain YCR13 are not identical. Within strain YCR13, L25 expression appears to be less than the level that accumulates in strain YCR11; however, in a separate isolate of this strain (carrying a chromosomal copy of L25 and a plasmid-encoded FLAG-tagged L25; called strain YCR14 not described in this study), no differences in L25 and L25-FLAG protein levels were seen, indicating expression variation between strains when both L25 and L25-FLAG genes are present (data not shown). Several possibilities might account for the lack of fly L23a-FLAG protein within strain YCR11. There was no rationale for proposing a negative interaction between chromosomal and plasmid genes since simultaneous expression of both genes was readily detectable within strain YCR13. By restriction analysis of rescued plasmid pL23a-FLAG, we ruled out any gross changes in sequence that might affect plasmid gene expression, although small differences would not be detectable with this method. By RT-PCR analysis, we determined if the plasmid was transcribed within strain YCR11. Using L23a H1 domain-specific primers, L23a-specific RT-PCR products were detected within YCR11, but not in YCR10 where only FLAG-tagged L25 is expressed (). The YCR11 product is the same size as the product generated using adult RNA as a template for amplification. A noticeable reduction in the amount of PCR product for strain YCR11 compared to strain YCR16 (shown in this particular RT-PCR experiment) was not consistently observed in all experiments, likely reflecting experimental variation. Equivalent amounts of PCR product for YCR16 and another strain that is genotypically equivalent to YCR11 (not discussed here, called YCR12; note that L23a-FLAG protein also fails to accumulate in this strain—data not shown) were observed in other RT-PCR experiments (data not shown). Since the analysis was not designed as a quantitative PCR, we cannot confirm if there is reduced L23a plasmid transcription within strain YCR11 compared to strain YCR16. The data implicate a post-transcriptional mechanism to account for the lack of fly L23a-FLAG protein within strain YCR11. As an essential ribosomal protein L23a-FLAG protein should be a component of the ribosome population within strain YCR16. Inada . () have previously shown that ribosomes and associated proteins can be purified using a one step affinity purification method where an anti-FLAG M2 antibody resin is used to capture FLAG-tagged L25 protein and associated components. Using an identical strategy, we captured FLAG-tagged L25- and FLAG-tagged L23a-associated proteins from strains YCR10 and YCR16, respectively (). Compared to the crude protein extract input for each strain, a subset of proteins was bound and eluted from the resin with a high concentration of FLAG peptide. In the absence of any FLAG-tagged protein in strain YCR9, no proteins were specifically bound to the anti-FLAG resin. Notably the affinity-purified protein patterns from strains YCR10 and YCR16 are nearly identical, indicating that the same subset of proteins is associated with tagged L25 and tagged L23a. Although the identity of the L23a-/L25-FLAG-associated proteins was not confirmed in this analysis, the data are consistent with the conclusion that fly L23a-FLAG protein is a component of ribosomes within strain YCR16 just as FLAG-tagged L25 is a component of ribosomes within strain YCR10 [equivalent to strain YIT613 (); see ]. For further phenotypic analysis of YCR strains 10, 11 and 16, we compared strain growth characteristics at different temperatures. At 30°C strain YCR16 grew significantly slower than strains YCR10 and YCR11 (). The growth of all strains was retarded at 23°C with strain YCR16 growing much slower relative to strains YCR10 and YCR11 (). Doubling times for each strain at 30°C, calculated based on logarithmic growth of strains in liquid culture, were 130 min for YCR10, 140 min for YCR11 and 260 min for YCR16. It is unclear if the difference between the growth rates for strains YCR10 and YCR11 is significant or not (). No obvious morphological differences were observed between strains YCR10 and YCR11 (data not shown). Yeast cells, dependent on fly L23a for survival and growth were significantly delayed in growth relative to cells that were dependent on endogenous L25, requiring twice the amount of time for doubling. Previous studies in which yeast L25 was genetically depleted or mutated have shown that L25 is required but is not sufficient for the removal of ITS2 and is necessary for efficient cleavage at the early sites A, A and A [(); see a]. The slow growth pattern of strain YCR16 suggested that rRNA maturation kinetics might be affected. A pulse-chase analysis of pre-rRNA processing was performed to determine if any step(s) in rRNA maturation was affected by fly L23a substitution for yeast L25 in the maturation pathway. b shows that while there is no measurable delay in the appearance of 27S and 20S rRNA precursors from processing of 35S rRNA, there is an apparent delay in the conversion of 27S rRNA into mature 25S rRNA within strain YCR16. There is no evidence for a delay in 35S rRNA synthesis as this precursor is present at the beginning of the chase period in both strains YCR9 and YCR16 (b). Although the products of 35S rRNA processing are present at the beginning of the chase period in both strains, processing of 35S rRNA at early sites A, A, or A (a) may be slightly less efficient in strain YCR16 since a greater amount of 35S rRNA persists in the chase period compared to amounts in strain YCR9 (b). A delay is clearly evident at a later processing step (C; a) to form mature 25S rRNA within strain YCR16 compared to strain YCR9 (b). At 5 min into the chase period, 25S rRNA is clearly evident within the YCR9 strain (b). It is not however until 10–15 min into the chase period that 25S rRNA becomes apparent within YCR16. A delay in 27S rRNA processing was consistently seen in replicates of this experiment. No significant differences in the kinetics of 20S rRNA processing into18S rRNA were detected between strains. While it is unknown if the 27S rRNA processing delay is accompanied by any changes in rRNA synthesis and/or turnover in the YCR16 strain, it is plausible that the slower processing kinetics alone may be sufficient to explain the strain's slower growth rate. It is possible that a decrease in 25S rRNA maturation kinetics would affect the steady state levels of 25S rRNA in strain YCR16. Northern blot analysis of RNAs isolated from cultures grown overnight shows that the steady state levels of 35S, 25S and 18S rRNAs are diminished compared to levels in strain YCR9 (c), even though equal amounts of RNA were loaded onto gels (confirmed by using as a loading control). Although the rate of rRNA synthesis within strain YCR16 may be diminished, increased rRNA degradation may be a contributing factor, affecting not only the steady state levels of 25S rRNA, but the levels of 35S and 18S rRNAs as well. In light of the reduced amount of rRNA and yet nearly equivalent amount of L23a-FLAG compared to L25-FLAG detected by Western blot in strains YCR16 and YCR10, respectively (a), the level of L23a-FLAG appears to be in excess of the amount of rRNA substrate in strain YCR16. Rapid degradation to eliminate excess yeast L25 protein was previously reported when the L25 gene dosage was increased nearly 50-fold (). Within strain YCR16, the level of L23a-FLAG protein may be below a threshold required for rapid degradation. Protein turnover may also be affected by slower kinetics of ribosome maturation observed in this strain. The ability of several heterologous L23a proteins to replace the function of yeast L25 has been documented previously [e.g. (,)]; however, in no case has the disparity in L23a protein structure been as remarkable as is the case for L23a compared to L25. Translation function within strain YCR16 is sufficient to maintain cell viability although there are clearly phenotypic deficiencies highlighted by the slow growth pattern in the strain at 23°C and at 30°C. Several factors may account for the growth deficiency in strain YCR16, including inefficient utilization of the fly L23a NLS by the yeast transport machinery affecting the rate of L23a-FLAG protein nuclear transport or the presence of the small FLAG tag on L23a interfering with interactions required for ribosome synthesis or for post-ribosome maturation steps. Although neither of these possibilities can be completely excluded, the N-terminal extension itself may impede the kinetics of ribosome assembly or post-ribosome maturation processes at crucial steps, ultimately affecting the rate of growth. Earlier studies in which yeast L25 was either mutated () or genetically depleted () revealed early and late rRNA processing defects. Our studies suggest that 35S rRNA processing (A–A) to produce pre-27S rRNAs is not severely compromised, but that the efficiency of late steps (including C) to produce mature 25S rRNA is affected in strain YCR16. This may affect the ribosome pools available for translation (as suggested by the diminished steady state levels of mature rRNAs), even if translation and post-translation processes proceed normally. Although the plasmid-encoded L23a-FLAG gene is transcribed within strain YCR11, no L23a-FLAG protein accumulates. In principle, the difference in L25 gene dosage between strains YCR11 and YCR16 () might elevate protein levels beyond what is required for pre-rRNA binding and large ribosomal subunit assembly, resulting in degradation of excess protein in a mechanism previously described in other studies [e.g. ()]. If so, we would then expect both L25 and L23a-FLAG proteins to be represented in the YCR11 ribosome population, with excess protein degraded. Yet, lack of any detectable accumulation of L23a-FLAG suggests that an alternative explanation must account for the strain's protein accumulation pattern. A reasonable hypothesis is that yeast L25 competes more efficiently for nuclear import and/or for incorporation into yeast 60S ribosomal subunits than fly L23a-FLAG in this strain, ultimately excluding L23a-FLAG protein from being assembled and leading to its rapid turnover. On the other hand, the impact of protein competition is not a factor within strain YCR16 because the fly L23a-FLAG gene provides the sole source of this essential protein. Although the rate of L23a-FLAG protein turnover may be affected due to a change in ribosome maturation kinetics overall, it is clear that fly L23a-FLAG protein accumulates, ribosome assembly proceeds and growth is supported, albeit at a slower rate. Our results differ from those of Jeeninga . () in which rat L23a competed efficiently with yeast L25 for incorporation into yeast ribosomes, even in the presence of endogenous L25. Rat L23a shows 62% sequence identity with yeast L25 (), is comparable in size to yeast L25, and lacks the extended N-terminal domain. This general structural similarity may be sufficient to minimize binding affinity differences that would affect ribosome assembly. How does the yeast ribosome accommodate the extra domain of fly L23a-FLAG? To propose a hypothesis for positioning the fly L23a-FLAG extra domain within a chimeric yeast ribosome, we have considered several L23 and protein chaperone interactions as well as the spatial organization of other ribosomal proteins at the exit tunnel. Along with proteins L19, L22, L24, L29 and L31e, L23 surrounds the base of the exit channel () and interacts with the protein folding apparatus (see Supporting for the position of L23a on the 50S subunit). High-resolution structural studies of the 50S subunit from () and () have revealed L23 structural differences between organisms. All L23 protein family members have a conserved globular domain, but unlike eubacterial L23, archaea and eukaryotic members lack an internal loop that extends into the exit tunnel interior cavity [reviewed by ()]. The other end of L23 is positioned on the solvent side, close to the opening of the tunnel (). Forming a hydrophobic cradle to nestle the emerging peptide just outside the exit tunnel (), TF binds to the ribosome on a composite surface formed from two separate exposed regions in the L23 globular domain, a single exposed region of L29, and regions of Domain III of 23S rRNA (). One exposed region in L23 includes glutamate-14 (Glu-18 in L23) and the other region includes C-terminal amino acid positions 92–94 (Supporting ). Glutamate-18 is located in the center of the binding surface, with 23S rRNA interactions on one side of the binding surface and interactions with L29 and the C-terminal portion of L23 on the other side (). Although TF is only found in eubacteria and chloroplasts, yeast NAC reportedly binds to and yeast ribosomes, suggesting that even in the absence of a direct test of which eukaryotic L23a residues are involved in binding, it is likely that NAC interacts through an identical binding surface that includes yeast L25 (). Based on amino acid alignment of L23a with and L23 (Supporting ), critical residues required for TF (and NAC) binding are found at fly L23a amino acid positions 196–200 (20; also Supporting ), within the conserved C-terminus. Thus, within the chimeric ribosomes of strain YCR16, binding surface residues for NAC would include fly L23a positions 196–200 and additional residues at the C-terminal end. Taken together, we speculate that the N-terminal extra domain of fly L23a forms an exposed ‘appendage’ on the chimeric ribosome, distal to the C-terminal portion of L23a itself and projecting away from the peptide tunnel and ribosomal proteins L29 and L19. Other ribosome interactions may stabilize the position of the proposed appendage. This configuration is favored over a proposal in which the extra domain assumes an internal position within the tunnel where it could sterically occlude the tunnel in a fashion similar to some macrolide antibiotics (,) and is consistent with structural models in which an internal L23a loop is absent from eukaryotic L23a (). Whether or not the ‘surface appendage’ proposal can be extrapolated to fly ribosomes remains to be determined. Interactions between L23a and poly-ADP ribose polymerase (PARP), mediated through the novel H1-like domain, have already been identified (), but not fully explored. PARP plays an important role in many cellular processes, including DNA repair, transcription and apoptosis (,). The possibility that L23a may function along with PARP in non-ribosomal pathways is intriguing. In fact, a recent finding that fly L22 co-purifies with histone H1 and is involved in chromatin interactions required for transcriptional repression broadens the concept of ribosomal protein function in general (). Both fly L22 and L23a carry the histone H1-like N-terminal extra domain (). A possible role for the L22 N-terminal extension along with PARP in mediating these novel interactions with chromatin was not investigated (). It is unknown if similar chromatin-L23a interactions occur. Other L23a extra domain interactions, yet to be discovered, may contribute further to the multifunctional capacity of the L23a ribosomal protein family. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Genomic DNA in mammalian cells is commonly methylated at position 5 of cytosine residues in CpG sequences. This epigenetic modification plays an important role in the regulation of gene expression and chromatin structure and is essential for normal development, cell differentiation, X chromosome inactivation and genomic imprinting (,). Cell-type-specific methylation patterns are established in early developmental stages by the action of Dnmt3a and 3b and are then maintained through subsequent cell generations primarily by the action of Dnmt1 (,). In somatic cells, Dnmt1 is the predominant DNA methyltransferase in terms of abundance, contribution to global methyltransferase activity and to genomic methylation levels (,). Reduction of Dnmt1 levels leads to hypomethylation, genomic instability and cancer (). Aberrant genomic methylation is often associated with human disease and tumorigenesis (). Due to its strong preference for hemimethylated substrate DNA () and its accumulation at replication sites during S-phase (,) Dnmt1 is thought to act on hemimethylated CpG sites generated during DNA replication. The association of Dnmt1 with replication factories was proposed as an efficient mechanism for coupling maintenance of genomic methylation patterns to DNA replication (). Later, direct interaction with the replication processivity factor PCNA was shown and a small PCNA-binding domain (PBD) was mapped to the N-terminal region of Dnmt1 (,). This domain was also shown to be necessary for recruitment of Dnmt1 to DNA damage sites, suggesting that it is responsible for coupling DNA repair and restoration of methylation patterns (). Thus, direct interaction with PCNA would ensure that methylation patterns are faithfully preserved in different situations involving DNA synthesis. However, DNA replication is highly processive taking about 0.035 s per nucleotide (), while steady-state kinetic analysis of purified recombinant Dnmt1 revealed rather low turn-over rates of about 70–450 s per methyl group transfer (). Although DNA methylation by Dnmt1 may be faster , it is not likely to come close to the 3–4 orders of magnitude faster DNA replication. In fact, cytosine methylation is a highly complex reaction involving recognition of hemimethylated CpG sites, binding of S-adenosylmethionine, flipping of the cytidine base out of the double helix, formation of a covalent bond between the enzyme and the cytidine, transfer of the methyl group and release of the covalent bond by β-elimination (). These considerations leave open the question of how DNA replication and methylation are kinetically and mechanistically coordinated. In addition to Dnmt1 several other factors directly and indirectly involved in DNA replication, such as DNA Ligase I, Fen1, CAF-1 and Cyclin A have been shown to redistribute to replication foci during S-phase (). Many of these factors have been found to interact directly with PCNA, which forms a homotrimeric ring around the DNA helix and serves as a platform for tethering them to the replication machinery (). Even taking into account the trivalent nature of the PCNA ring, the sheer number of its potential binding partners during replication makes it clear that they cannot all possibly bind at the same time in a constitutive manner. The functional relevance of the interaction between Dnmt1 and PCNA and its contribution to the maintenance of epigenetic information after DNA replication, however, remains unclear. We have addressed this question by comparing the kinetics and activity of GFP-tagged wild-type Dnmt1 and PCNA-binding-deficient mutants in live cell assays and Dnmt1 deficient embryonic stem (ES) cells. Our data show that the interaction of Dnmt1 with PCNA is highly transient, increases the efficiency of postreplicative methylation by 2-fold, but is not required for restoring CpG methylation in Dnmt1 deficient ES cells. The expression constructs RFP-PCNA, GFP-Ligase, GFP-Dnmt1, GFP-Dnmt1, GFP-Dnmt1 as well as the PBD-GFP construct were described earlier (,,,). The GFP-Dnmt1 and GFP-Dnmt1 expression constructs were derived from the GFP-Dnmt1 construct by overlap extension PCR mutagenesis (,) using the outer forward primer 5′-CAG ATC TCG AGC TCA AGC TTC-3′, the inner reverse primer 5′-GTG TCA AAG CTC TGA TAG ACC AGC-3′, the inner forward primers 5′-GAACCACCAGGGAGACCACCATC-3′ for Q162E and 5′-CACGGCTCACTCCACGAAGG-3′ for F169S and the outer reverse primer 5′-CTGGAATGACCGAGACGCAGTCG-3′. The final PCR fragments containing the mutations were digested with BglII and HindIII and exchanged with the corresponding fragment in the GFP-DNMT1 construct. Mutations were confirmed by DNA sequencing and molecular size of fusion proteins was tested by expression in HEK 293T cells and western blot analysis. For stable transfections we inserted the cassette containing the wt Dnmt1 cDNA fused to GFP from the GFP-Dnmt1 construct into the pCAG-IRESblast vector (). Human embryonic kidney (HEK) 293T cells and mouse C2C12 myoblasts were cultured in DMEM supplemented with 10% and 20% fetal calf serum, respectively, and 50 µg/ml gentamycine. HEK 293T cells were transfected with polyethylenimine (Sigma) (). For live cell observations C2C12 myoblasts were grown to 30-40% confluence on Lab-Tek chamber slides (Nunc) or µ-slides (Ibidi) and transfected with TransFectin transfection reagent (Bio-Rad) according to the manufacturer's instructions. Cells were then incubated overnight before performing live cell analysis. Nuclear localization of GFP-Dnmt1 was identical to endogenous Dnmt1 as determined by immunostaining with an affinity purified polyclonal antiserum against the N-terminal domain of mouse Dnmt1. GFP-Dnmt1 localization was not affected by additional co-expression of RFP-PCNA (controls not shown). Dnmt1 immunostaining showed that typical expression levels of transfected GFP-Dnmt1 constructs in cells selected for live cell imaging were comparable to those of endogenous Dnmt1 protein (Supplementary Figure 1). For stable expression of GFP-Dnmt1, C2C12 cells were grown in a p100 tissue culture dish and transfected as described earlier. Cells were then cultured with 10 µg/ml blasticidin for at least 20 days before homogeneity and levels of expression were determined by fluorescence microscopy and western blotting (Supplementary Figure 2). Mouse wild type and J1 ES cells (s allele) () were cultured without feeder cells in gelatinized flasks as described (). J1 cells were transfected with Transfectin (BioRad) 3–4 h after seeding and GFP-positive cells were sorted with a FACS Vantage SE cell sorter (Becton–Dickinson). HEK 293T cells were transiently transfected with expression plasmids as described above. After 48 h about 70–90% of the cells expressed the GFP constructs as determined by fluorescence microscopy. Extracts from ∼1 × 10 cells were prepared in 200 µl of lysis buffer (20 mM Tris/HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 2 mM PMSF, 0.5% NP40). After centrifugation supernatants were diluted to 500 µl with lysis buffer without NP40. Extracts were incubated with 1 µg of a GFP-binding protein coupled to sepharose (manuscript in preparation) for 1 h at 4°C with constant mixing. Immunocomplexes were pulled down by centrifugation. The supernatant was removed and 50 µl were collected (referred to as non-bound). The beads were washed twice with 1 ml of dilution buffer containing 300 mM NaCl and resuspended in SDS-PAGE sample buffer. Proteins were eluted by boiling at 95°C and subjected to SDS-PAGE followed by immunoblotting. Antigens were detected with a mouse monoclonal anti-GFP antibody (Roche) and a rat monoclonal anti-PCNA antibody (). Extracts from HEK 293T cells expressing the indicated GFP constructs were prepared and immunoprecipitations were performed as described above. After washing with dilution buffer containing 300 mM NaCl the beads were washed twice with assay buffer (100 mM KCl, 10 mM Tris pH 7.6, 1 mM EDTA, 1 mM DTT) and resuspended in 500 µl of assay buffer. After adding 30 µl of methylation mix {[H]-SAM (S-adenosyl-methionine); 0.1 µCi (Amersham Biosciences), 1.67 pmol/µl hemimethylated ds 35 bp DNA (50 pmol/µl), 160 ng/µl BSA} incubation was carried out for 2.5 h at 37°C. The reactions were spotted onto DE81 cellulose paper filters (Whatman) and the filters were washed 3 times with 0.2 M (NH)HCO, once with ddHO and once with 100% ethanol. After drying, radioactivity was measured by liquid scintillation. Samples without enzyme and with 2 µg of purified human recombinant DNMT1 were used as negative and positive controls, respectively. Genomic DNA was isolated by the phenol–chloroform method () and bisulfite treatment was as described () except that deamination was carried out for 4 h at 55°C. Primer sets and PCR conditions for CpG islands of (region A) and promoters, exon 1 and intracisternal type A particle long terminal repeats (IAP LTRs) were as described (). PCR products were digested with the following enzymes: and promoters and IAP LTRs with HpyCH4IV (New England BioLabs); with Bsh136I and with TaqI (both from Fermentas). Digests were separated by agarose electrophoresis except for IAP LTR fragments, which were separated in 10% acrylamide gels. Digestion fragments were quantified from digital images using ImageJ software (). The results were corrected for PCR bias, which was calculated as described (). Briefly, COBRA assays were performed on genomic DNA from untransfected J1 cells methylated with recombinant SssI methyltransferase (New England BioLabs) and mixed in different proportions with unmethylated DNA from the same cells (Supplementary Figure 6). Bias curves and corrections were calculated using WinCurveFit (Kevin Raner Software). For each amplified sequence digestion with restriction enzymes whose recognition sequence includes cytosine residues in a non-CpG context was used to control for complete bisulfite conversion except for the promoter, where bisulfite sequencing revealed about 99% conversion efficiency in all samples (Supplementary Figure 5). The trapping assay to measure postreplicative methylation efficiency in living cells was previously described (). 5-Aza-2′-deoxycytidine (Sigma) was added at a final concentrations of 30 µM and cells were incubated for the indicated periods before performing FRAP experiments. Microscope settings were as described above except that a smaller ROI (3 µm × 3 µm) was selected and the time interval was set to 208 ms. FRAP data were double normalized as described above. We recently demonstrated a loss of association with replication foci in early and mid S-phase of a GFP-Dnmt1 fusion construct with a deletion of the first 171 amino acids which includes most of the PBD (GFP-Dnmt1) (). To address the function of the PBD in early and mid S-phase more specifically and to exclude potential misfolding of the protein due to the large deletion we have generated GFP-Dnmt1 constructs bearing single point mutations of highly conserved residues within the PIP (PCNA-interacting peptide)-Box () of the PBD (GFP-Dnmt1 and GFP-Dnmt1) (). Either of these mutant constructs were expressed in C2C12 mouse myoblasts together with RFP-PCNA, which served as S-phase marker (). Both Dnmt1 constructs showed a diffuse nuclear distribution in early and mid S-phase cells (B and Supplementary Figure 3), in contrast to the wild-type Dnmt1 construct (GFP-Dnmt1) which was concentrated at replication foci during early and mid S-phase (A). In late S-phase and G2, however, the wild type as well as all the PBD mutant constructs, including GFP-Dnmt1, were similarly concentrated at chromocenters (A and B and Supplementary Figure 1), confirming the additional binding to heterochromatin in late S-phase mediated by the TS domain (). Co-immunoprecipitation experiments confirmed that the Q162E point mutation abolishes the interaction between Dnmt1 and PCNA (C). Similar results were obtained with GFP-Dnmt1 and GFP-Dnmt1 (Supplementary Figure 3). Thus, both Q162E and F169S point mutations prevent accumulation of Dnmt1 at replication foci during early to mid S-phase, while localization at constitutive heterochromatin in late S-phase and G2 is not affected. These results clearly confirm the role of the PBD in mediating the interaction between Dnmt1 and PCNA . To investigate the dynamics of the PBD-mediated interaction at replication sites we measured fluorescence recovery after photobleaching (FRAP) of GFP-Dnmt1 throughout S-phase. GFP-Dnmt1 and RFP-PCNA were co-expressed in C2C12 cells and a small square region of interest (ROI) was bleached (). Consistent with earlier observations (,) RFP-PCNA showed hardly any recovery within the observation period of 100 s, whereas in the same period GFP-Dnmt1 recovered fully, with very similar kinetics in early and mid S-phase, but notably slower in late S-phase. This result shows that the binding of Dnmt1 at replication sites is more dynamic during early and mid S-phase than in late S-phase, when Dnmt 1 is likely slowed down by the additional interaction with chromatin mediated by the TS domain (). To address the kinetic properties of the PBD-mediated binding to replication sites more specifically, we performed quantitative FRAP analysis of GFP-Dnmt1 and GFP-Dnmt1 in G1 and early/mid S-phase. As replication sites are not homogeneously distributed in the nucleus, we chose to bleach half nuclei (half-FRAP) to ensure that the bleached region contains a representative number of potential binding sites (A). In early/mid S-phase nuclei GFP-Dnmt1 recovered with a halftime () of 4.7 ± 0.2 s and reached complete equilibration () in about 56 s. These values place Dnmt1 among the more dynamic factors involved in chromatin transactions previously determined with half-FRAP analyses (). In comparison to GFP-Dnmt1, GFP-Dnmt1 showed a slightly increased mobility ( = 4.4 ± 0.2 s; ∼45 s), which is likely due to the lack of binding to PCNA rings at replication sites. In the absence of active replication sites in G1, GFP-Dnmt1 and Dnmt1 showed nearly identical kinetics, which were remarkably similar to the kinetics of Dnmt1 in early/mid S-phase. These data indicate that PCNA binding has only a minor contribution to Dnmt1 kinetics in S phase. The recovery rates measured for the full-length constructs were considerably slower than the rate of GFP alone ( = 0.8 ± 0.1 s; ∼10 s), which was used to control for unspecific binding events. As is roughly proportional to the cubic root of the molecular mass (,), the ∼8-fold size difference of GFP alone to the full-length construct (27 and 210 kDa, respectively) would only account for about a 2-fold slower recovery. Instead, GFP-Dnmt1 full-length constructs recover more slowly, pointing to one (or more) additional and yet uncharacterized cell cycle independent interaction(s). In order to dissect the PBD-mediated interaction from superimposing effects caused by other potential interactions we assayed the FRAP kinetics of a GFP fusion with the isolated PBD of Dnmt1, i.e. amino acids 159–178 (PBD-GFP). We found that in S–phase the recovery was only about 2 times slower than GFP alone ( = 1.5 ± 0.1 s; ∼16 s), thus confirming the highly transient nature of the PBD interaction with PCNA. In non-S-phase cells, PBD-GFP showed an increase in mobility ( = 1.1 ± 0.1 s; ∼11 s) similar to that observed with the full-length wild-type construct. For direct comparison we further analyzed another PCNA-interacting enzyme, DNA Ligase I fused to GFP (GFP-Ligase), and found a similar mobility shift in S phase ( = 2.4 ± 0.3 s; ∼27 s) compared to non-S-phase cells ( = 2.1 ± 0.3 s; ∼17 s). Thus, with all three PCNA-interacting GFP fusion proteins (GFP-Dnmt1, PBD-GFP and GFP-Ligase), but not with the PCNA-binding mutant, the transient association with the replication machinery (PCNA) caused a slower recovery as compared to G1/non-S-phase (B, inset). Next we investigated the contribution of the highly transient interaction with PCNA to the postreplicative methylation activity of Dnmt1. Earlier it was shown that N-terminal deletions of mouse Dnmt1 comprising the PBD did not alter catalytic activity (,). To test the catalytic activity of the GFP-Dnmt1 constructs used in this study they were expressed in HEK 293T cells, immunopurified and directly assayed for methyltransferase activity . While the catalytically inactive GFP-Dnmt1 mutant () displayed only background activity, GFP-Dnmt1 and GFP-Dnmt1 exhibited enzymatic activity comparable to GFP-Dnmt1 (D), indicating that neither the Q162E point mutation nor deletion of the first 171 amino acids affect the enzymatic activity of Dnmt1. To establish whether the binding to PCNA is required for postreplicative methylation , we tested the activity of GFP-Dnmt1 and GFP-Dnmt1 in living cells using a recently developed trapping assay (). This assay takes advantage of the catalytic mechanism of DNA (cytosine-5) methyltransferases which involves transient formation of a covalent complex with the C6 position of the cytosine residue. When the cytosine analogue 5-aza-2′-deoxycytidine (5-aza-dC) is incorporated into the DNA during replication the covalent complex of Dnmt1 and 5-aza-dC cannot be resolved and Dnmt1 is trapped at the site of action. Time-dependent immobilization, i.e. trapping of GFP-tagged Dnmt1 at replication foci can be visualized and measured by FRAP and reflects enzymatic activity of the fusion protein. C2C12 cells co-transfected with RFP-PCNA and either GFP-Dnmt1 or GFP-Dnmt1 as well as C2C12 cells stably expressing GFP-Dnmt1 were incubated in the presence of 30 µM 5-aza-dC. In early S-phase the focal enrichment of GFP-Dnmt1 at replication sites increased over time reflecting the accumulation of immobilized enzyme and after 40 min GFP-Dnmt1 was completely immobilized (A). Similar kinetics were observed in mid S-phase (data not shown). As shown above GFP-Dnmt1 displayed a diffuse nuclear distribution in early S-phase cells (B). However, with prolonged incubation in the presence of 5-aza-dC an increasing focal accumulation at replication sites was observed. Quantitative FRAP analysis revealed that the immobilization rate of GFP-Dnmt1, which is a direct measure of its enzymatic activity, was only about 2-fold slower than GFP-Dnmt1, resulting in complete trapping after ∼90 min. These results indicate that the PCNA-binding-deficient mutant binds to DNA and is catalytically engaged at hemimethylated sites generated during replication. We then probed the stability of the interaction between Dnmt1 and the replication machinery by trapping Dnmt1 with 5-aza-dC and long-term live imaging. In the case of stable interaction, covalent immobilization of Dnmt1 would be expected to stall the progression of the replication machinery. C2C12 cells co-transfected with the GFP-Dnmt1 and RFP-PCNA constructs were incubated in the presence of 10 µM 5-aza-dC and individual S-phase cells were imaged at consecutive time points for an extended period of time (). Progressive separation of GFP-Dnmt1 and RFP-PCNA foci could be clearly observed over a time period of ∼2 h, indicating that trapping of Dnmt1 did not prevent the progression of replication factories. This result is consistent with the FRAP kinetics of GFP-Dnmt1 demonstrating the transient nature of the interaction between Dnmt1 and PCNA. To investigate the contribution to maintenance of methylation patterns by the PBD-mediated interaction with PCNA we transiently expressed either GFP-Dnmt1 or GFP-Dnmt1 in Dnmt1 deficient mouse ES cells, which are severely hypomethylated in all genomic compartments (,). GFP-positive cells were isolated by FACS sorting 24h and 48 h after transfection and methylation of single-copy sequences and intracisternal type A particle (IAP) interspersed repetitive elements was analyzed by COBRA. An increase in methylation at all tested sites was observed in cells expressing either the wild type or the mutant Dnmt1 constructs already 24 h after transfection ( and Supplementary Figure 4A). Further substantial increase of methylation 48 h after transfection was observed only for the promoter where the methylation level approached that observed in wild-type cells. The result for the promoter were confirmed and extended by bisulfite sequencing (Supplementary Figure 5). It was previously reported that re-expression of wild-type Dnmt1 in ES cells does not lead to restoration of methylation at imprinted genes since passage through the germ line is needed for re-establishment of methylation patterns in these sequences (). We analyzed the promoter of the imprinted gene to control for the specificity of our assay and found that indeed expression of neither GFP-Dnmt1 nor GFP-Dnmt1 resulted in increased methylation of this sequence (). To further evaluate methylation of repetitive sequences we stained transiently transfected cells with an antibody against 5-methylcytidine that detects highly methylated satellite repeats present at mouse chromocenters (pericentromeric heterochromatin). Restoration of high DNA methylation levels at chromocenters was observed in both cells expressing GFP-Dnmt1 and GFP-Dnmt1 (Supplementary Figure 4B). The experimental procedures employed did not allow detection of significant differences in remethylation kinetics between the two constructs. Nevertheless, these results show that the PCNA-binding-deficient mutant is, like wild-type Dnmt1, able to rescue methylation of both single copy and repetitive sequences . Faithful replication of genetic and epigenetic information is crucial to ensure the integrity and identity of proliferating cells. Earlier work has demonstrated that the maintenance methyltransferase Dnmt1 binds to the replication processivity factor PCNA and is thus recruited to replication sites (). The interaction between Dnmt1 and the replication machinery was proposed as a mechanism for coupling maintenance of genomic methylation to DNA replication (). By traveling along with the replication machinery Dnmt1 would be able to restore symmetrical methylation as soon as hemimethylated CpG sites are generated. The estimated kinetics of DNA replication and DNA methylation by Dnmt1 , however, differ by 3–4 orders of magnitude arguing against a stable coupling. Although DNA methylation may be faster it is not likely to come much closer to the DNA replication rate. In other words, it is reasonable to assume that methylating a cytosine takes far longer than to incorporate it. Also, DNA replication is essentially a continuous process, while methylated CpG sites are encountered discontinuously in vertebrate genomes. Stable binding of Dnmt1 to the replication machinery would stall replication at each hemimethylated CpG site. Here we show that the interaction of Dnmt1 with the replication machinery is highly dynamic and that immobilization of Dnmt1 at postreplicative hemimethylated sites does not prevent the progression of DNA replication. The transient nature of the interaction between Dnmt1 and the replication machinery allows rapid release and transfer of Dnmt1 to hemimethylated substrate sites, reconciling the paradox of the relative rates of DNA replication and methylation. According to basic principles of enzyme kinetics the local enrichment of Dnmt1 at replication foci would increase the postreplicative methylation rate. At the same time, transient binding of Dnmt1 enables also other replication factors to interact with PCNA. Similar binding dynamics have been shown for the interaction of PCNA with DNA Ligase I and Fen1 () and may thus represent a common feature of PBD-containing factors. Interestingly, the interaction between Dnmt1 and PCNA is believed to be a major mechanism for the methylation maintenance activity of Dnmt1, but its functional relevance had never been tested experimentally. Here we show two lines of evidence that this interaction is not crucial for the maintenance of methylation patterns in mammalian cells. First, postreplicative methylation rate of wild-type Dnmt1 measured is only 2-fold faster than that of a PCNA-binding-deficient mutant. Second, methylation of both single copy and repetitive sequences in Dnmt1 deficient ES cells was restored by this PCNA-binding-deficient Dnmt1 mutant with efficiency comparable to wild-type Dnmt1. The maintenance of DNA methylation without direct coupling to the replication machinery could in part be explained by the preference of Dnmt1 for hemimethylated sites (). Also, genetic manipulation in the mouse indicate that Dnmt1 is at least 5-fold more abundant than necessary for maintaining normal methylation levels (,). Thus, the combined effect of the affinity for hemimethylated sites, relative abundance and simple diffusion could explain the relatively fast immobilization of PCNA-binding-deficient mutants in the presence of 5-aza-dC. In addition, the ability of Dnmt1 to methylate nucleosomal DNA (), suggests that maintenance of DNA methylation is not necessarily restricted to the short time window before nucleosome assembly. Recent structural data on Ligase I:PCNA and FEN-1:PCNA complexes indicate that for these enzymes PCNA does not simply serves as a loading platform for the replication machinery, but also causes allosteric activation (,). The data presented here cannot rule out a similar mechanism for Dnmt1, but clearly show that interaction with PCNA is not a prerequisite for enzyme activity . The transient nature of this interaction also argues against PCNA as a classic processivity factor for postreplicative DNA methylation. The major role of the PCNA interaction most likely is to increase the local Dnmt1 concentration and thereby enhance methylation efficiency at replication sites. Notably, it is still unclear whether the role and regulation of Dnmt1 is similar in different cell types and species. While Dnmt1 is clearly essential for maintenance of DNA methylation in mouse cells (,), it seemed dispensable in human tumor HCT116 cells (). However, two reports recently showed that DNMT1 is essential for maintenance of DNA methylation also in these human tumor cells (,). It turned out that genomic methylation was maintained by a residual, truncated DNMT1 form lacking the PBD, arguing that PCNA binding is not strictly required in these cells (). Also, the requirement of Dnmt1 for cell viability remains unsettled. In mouse fibroblasts inactivation of the gene caused a continuous loss of genomic methylation leading to apoptotic cell death after several cell division cycles (). Similar results were obtained after depletion of DNMT1 activity by RNA interference in human cells (). Surprisingly, conditional deletion of the locus in the same cells caused immediate apoptotic cell death long before substantial passive loss of genomic methylation could occur () arguing for additional roles of DNMT1. In this regard, the association of Dnmt1 with heterochromatin in G2 phase and occasionally in mitosis in mouse cells () would fit well with the mitotic catastrophe observed upon deletion of the gene in HCT116 cells. In addition, the PBD-mediated association of Dnmt1 with repair sites () may indicate a direct role in the maintenance of genome integrity (). Clearly, more experiments are required to resolve the species and cell-type-specific role and regulation of Dnmt1. In summary, we demonstrate that the association of Dnmt1 with the replication machinery is not strictly required for maintaining global methylation but still enhances methylation efficiency by 2-fold. Whereas the benefit of Dnmt1 to be directly recruited to replication foci seems subtle in short-term cell culture experiments, it may be more relevant in long-lived organisms and in situations where the nuclear concentration of Dnmt1 is limiting. Indeed, Dnmt1 levels vary considerably in different tissues and developmental stages (,). Based on sequence features of the gene a modular structure was proposed to originate from an ancestral DNA methyltransferase that evolved by stepwise acquisition of new domains (). Thus, the improved efficiency of postreplicative methylation achieved by the PBD-mediated transient binding to PCNA likely represents an additional safety mechanism, which was acquired in the course of evolution and contributes to the faithful maintenance of epigenetic information over the entire lifespan of complex organisms. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Antiretroviral resistance is a major threat to successful anti-HIV treatment. This problem is more frequent among individuals that started therapy in the 1990s, who were treated sequentially with single antiviral agents (), allowing sequential development of resistance to each treatment. Therapy with combinations of antiretroviral drugs is more effective, but nevertheless up to 50% of individuals currently in care in the US harbor HIV-resistant viruses (). These resistant strains can also be transmitted—more than 15% of recently infected individuals have acquired viruses that are resistant to at least one of the major antiretroviral classes (). Current treatment guidelines in the United States recommend resistance testing before beginning or changing antiretroviral therapy. In the developing world, where much of the burden of HIV infection is concentrated, combination therapy is increasingly available. A sharp increase in drug resistance is expected as patients become more treatment experienced. Unfortunately, resistance genotyping is generally unavailable in the developing world due to the prohibitive expense. Genotypic and phenotypic methods are commonly used to detect antiviral resistance in clinical specimens. Genotypic methods use bulk sequencing of the protease (PR) and reverse transcriptase (RT) coding regions, which reports the sequence of the predominant circulating HIV variants. Resistance mutations conferring reduced sensitivity to the three most widely used drug classes (nucleoside and non-nucleoside RT inhibitors and PR inhibitors) are well characterized, allowing probable resistance patterns to be inferred from sequence information. Phenotypic methods rely on cloning the RT and PR-coding regions from patient samples into a standard HIV plasmid backbone, allowing generation of viral stocks and functional analysis of viral drug sensitivity in short-term culture (). Several studies have demonstrated that minor drug-resistant HIV populations that are not detectable in the standard assays can impair the response to therapy (,). This problem is particularly apparent in studies of pregnant women that received single doses of nevirapine, a non-nucleoside reverse transcriptase inhibitor (NNRTI), at the time of delivery to prevent vertical transmission of HIV. In these patients, the presence of minor populations with resistance to nevirapine—which were often undetectable by conventional sequencing—compromised the response to subsequent NNRTI therapy (,). A variety of technologies have been devised to allow characterization of minor HIV drug-resistant populations (,,). In one method, microarrays were designed to interrogate positions of drug resistance mutations (). This technique allows analysis of many genomic positions in a single experiment, but the method has not been widely used, in part due to the high cost of each test. Another method involves allele-specific RT-PCR, which allows sensitive detection of single drug resistance mutations (). A third approach takes advantage of the massively parallel polony sequencing method (), and a fourth uses an early version of pyrosequencing to query single nucleotide positions for possible mutations (). However, except for the microarray method, all of the above methods queried individual base pairs at a time (). Given that there are over 60 amino acid positions just in PR and RT that can affect resistance to the three widely used drug classes, these methods are time consuming and difficult to use for comprehensive analysis. An ideal method for investigating drug resistance mutations would yield many complete sequences of HIV genomic regions at risk for mutations for each viral population, and allow analysis of many samples of HIV populations in a single experiment. We have adapted pyrosequencing (), combined with a DNA bar coding system (,), to characterize rare drug-resistant HIV variants in many samples in parallel. In a single experiment, we determined 118 093 sequences from ∼100 bp segments of the PR and RT-coding regions for seven samples of viral populations (). These data identified a variety of minor drug resistance alleles in patient samples of potential clinical significance, and demonstrate the feasibility of using pyrosequencing for efficient HIV genotyping. Multiplex analysis of many bar coded samples in a single sequencing experiment offers the potential to drive down the cost of each genotype determination. For the primer design, representative sequences (341 total) from subtypes A–D and G were downloaded from the Los Alamos HIV Database and Compendia (). Subtype consensus sequences were formed by aligning sequences within each subtype, then ‘Pan-HIV’ primers were designed to anneal to sequences conserved among subtypes (Supplementary Table 1). Patient 1 was treated with tenofovir, FTC, atazanvir and ritonavir; patient 2 is deceased and treatment history is unknown, and patient 3 was treated with retrovir (zidovudine), epivir (lamivudine; 3TC) and viracept (nelfinavir). The conventional clinical analysis of HIV genotypes for the three patients studied was carried out as follows. Plasma viral RNA was extracted, amplified by RT-PCR (Roche Amplicor v 1.0; Roche Diagnostic Systems, Inc., Branchburg, NJ, USA), and sequenced using the Viroseq HIV Genotyping System v 2.0 (Applied Biosystems, Foster City, CA, USA). The sequences were analyzed for the resistance mutations using the HIV Drug Resistance Database at Stanford University () (), then calls were edited to match the slightly different catalog from the International AIDS Society Drug Resistance Mutations in HIV (Fall 2006) (). Pyrosequencing was carried out using the 454 Life Sciences Technology at the University of Florida. The pyrosequencing method, as implemented by 454 Life Sciences, involves the following steps. Genomic DNA samples of interest are amplified using primers that include 5′ extensions providing binding sites for the 454 A and B primers. DNA fragments are then mixed with beads that have bound on their surfaces oligonucleotides complementary to the primers. This step is carried out in dilute solution so that on average a single DNA strand binds to each bead. A dilute mixture of beads is then added to an oil–water emulsion, arranged so that on average each aqueous droplet contains a single bead with a single bound strand. PCR amplification is then carried out in the emulsion. Each DNA strand becomes amplified and then binds by sequence complementarity to the bead, thereby creating beads that are each conjugated to DNA strands of a single homogenous sequence. Beads with bound DNA are then distributed on a picotiter plate at a density of ∼150 000 beads per plate. A primer is then bound to each DNA, and a polymerase used to extend a DNA chain. The four nucleotide triphosphates are sequentially flowed over the plate. An enzyme system is present in the buffer, which directs incorporation of pyrophosphate liberated by nucleotide addition into ATP, which then activates purified luciferase enzyme in the buffer to produce light. A CCD camera records each flash from each well on the plate. Sequential application of the four nucleotides allows DNA sequences of ∼100 bp to be built up ∼150 000 at a time (). The initial sequence reaction, carried out on a single plate, yielded 135 528 sequence reads, of which 118 093 ultimately passed quality control. For the sequences to pass, we required that each have a perfect match to the bar code and primer region, and no more than one N in the determined sequence. A total of 5.1% of the initial sequence reads had bar codes that were not included in the original experiment. The pyrosequencing method is error-prone at homopolymers, and of the incorrect bar codes, 54% created homopolymeric sequences within the bar codes and were excluded on that basis. To suppress bar code ‘crossover’, the different viral samples were separated into separate quadrants of the sequencing plate as follows (numbers from ): samples 1 and 2, quadrant 1; samples 3 and 5, quadrant 2; samples 6 and 7, quadrant 3; sample 4, quadrant 4. Inspection of drug resistance calls for samples within the same quadrant showed no obvious bar code crossover. For scoring drug resistance mutations, we used the Sierra Webservice at the HIV Drug Resistance Database at Stanford University (). To query the Drug Resistance Database, the pyrosequence reads were embedded in ‘dummy’ HIV flanking sequences, and the resistance alleles identified over the sequenced region. To pass quality control at this step, we required that the full sequence be recognized as HIV by the HIV Drug Resistance database, and that all pyrosequence reads cover at least 60% of the genomic window interrogated. Resistance mutations called by the Stanford Database were filtered to match the International AIDS Society definitions. Inspection of the raw counts of drug resistance calls in controls (Supplementary Table 2) showed that different positions showed differing error rates. For this reason, a statistical model was used that took into account the error rate measured at each position. The proportion of drug-resistant mutant calls in the combined HIV LAI DNA and HIV LAI RNA data sets were taken as the background error for each position. We took advantage of the Fisher's exact test to investigate whether drug resistance mutations were significantly enriched in the patient samples compared to controls. To control for multiple comparisons, a Bonferroni correction was applied. The -values were multiplied by the number of comparisons within each individual (62 comparisons corresponding to 62 codons queried for drug resistance). Drug resistance calls with corrected -values <0.05 were judged to be significantly enriched. Statistical analysis was carried out in the environment (). Primers were designed for amplifying the regions of HIV that are known from previous work to be sites of substitutions that result in resistance to PR and RT inhibitors. Because the most useful design would allow amplification of HIV sequences from any of the viral subtypes, we designed ‘Pan-HIV’ primers that would amplify subtypes A, B, C, D and 01_AE, the major subtypes circulating world-wide. Typical read lengths for the pyrosequencing method at the time of this experiment were ∼100 bp. For the sequencing procedure, it is desirable to use fragments that are somewhat longer than this, to allow electrophoretic separation of the PCR products away from short contaminating sequences such as primer dimers. For this reason, we designed 11 overlapping amplicons, as shown in , to allow analysis of all known positions of PR and RT drug resistance mutations [cataloged by the International AIDS Society (Fall 2006 Revision) ()] while allowing purification of fragments in the 200–400 bp size range. A key consideration in analyzing our results was distinguishing authentic drug-resistant mutations from erroneous calls. After isolation of HIV RNA from particles, reverse transcription followed by PCR (RT-PCR) has the potential to introduce mutations. The pyrosequencing procedure, as implemented at 454 Life Sciences (), also requires a PCR amplification step, and the pyrosequencing method is more error prone than the Sanger method. Several controls were therefore included in the experiment to allow estimation of the background error rate (). Plasmid DNA from the HIV isolate LAI was used as template for PCR amplification by the 11 pan-HIV primers and the products analyzed. Because the HIV LAI DNA sequence is known, this serves as a control for error in the RT-PCR and pyrosequencing steps. RNA from HIV LAI viral particles was included as another control. This provided a second measure of error and also allowed us to assess the misincorporation rate due to transcription and reverse transcription of the HIV RNA, which proved to be undetectable using the methods described later. A fourth sample contained a pool of five HIV subtypes (A, B, C, D and 01) (), and was included to illustrate the ability of the pan-HIV primers to amplify sequences from all five, and the downstream bioinformatic methods to distinguish sequence data from each. Since one of our goals was developing methods for testing many samples in single sequencing experiments, we developed a DNA bar coding strategy to allow sample multiplexing (). Each primer consisted of the 454 A and B sequences at the 5′ ends, required for the emulsion PCR, and the HIV-complementary regions at the 3′ end. To distinguish different amplicons in a pool, we added a 4 bp DNA bar code sequence between the 454 A or B sequence and the HIV-complementary region. Because the 454 sequencing method has an increased error rate at positions of adjacent nucleotides of the same type (homopolymers), we avoided using bar codes containing adjacent identical nucleotides. PCR amplification was carried out on the seven HIV nucleic acid templates listed in using the pan-HIV primers. All amplification reactions yielded DNA chains of the expected length when analyzed by agarose gel electrophoresis and ethidium bromide staining. shows a sample amplification of subtype C viral RNA. The PCR amplicons were then pooled and subjected to pyrosequencing using the method implemented by 454 Life Sciences (). Samples were attached to beads and amplified using emulsion PCR, the beads with bound DNA were applied to a picoliter plate, and sequences on beads were determined using the pyrosequencing method. A total of 135 528 sequence reads were returned, of average length 103 bp. After quality control, 118 093 sequences were available for analysis, corresponding to ∼12 million base pairs of DNA sequence. Of the seven bar codes used, the number of sequences returned per bar code after quality control ranged from 28 751 (pooled subtypes) to 7398 (NL4-3 with doped drug resistance allele). More of the PCR products from the subtype pool was used for sequencing because we expected a relatively higher level of diversity from this sample due to the diversity of the templates. Overall, 121 amplification products were pooled and analyzed, and sequences from all were recovered in good yield. We next sought to identify mutations within the HIV sequences expected to confer drug resistance. Sequences passing quality control were analyzed at the HIV Drug Resistance Database at Stanford University () (), and the results tabulated. For our initial analysis, we restricted our attention to the best documented drug resistance mutations as summarized by the International AIDS Society Drug Resistance Mutations in HIV (Fall 2006 Revision) (). Overall, 62 positions of drug resistance were queried, and the frequency of drug resistance calls at each codon tabulated (raw data in Supplementary Table 2). The pool of HIV subtypes (, sample 4) was separated into subtype-specific groups by aligning the sequences against subtype-specific consensus sequences. Sequences were recovered for all five subtypes in good yield. Supplementary Table 3 catalogs the sequences identified and the drug resistance mutations detected in each subtype, and Supplementary Table 4 presents the rate of polymorphism at sites not involved in drug resistance. The proportion of drug resistance calls at each position in the controls (, samples 1 and 2) was assessed and taken as the position-specific background. The frequency of resistance calls in each sample was then tested against the position-specific background using Fisher's exact test. Inspection of the initial collection of -values showed that there were more positions of significant enrichment than expected in the controls (, samples 1–3), which was not surprising because the large number of comparisons carried out (62 for each patient) increased the likelihood of obtaining significant enrichment by chance. To control for the inflation of error due to multiple comparisons, we applied a Bonferroni correction, in which the -value from Fisher's exact test was multiplied by the number of comparisons in each patient. Thirty-six positions achieved significance after correction ( < 0.05), and are shown as black tiles in . Proper performance of this procedure could be verified by analyzing the results for the HIV NL4-3 sample spiked with 5% virions encoding the L10R/M46I/L63P/V82T/I84V resistance substitutions. All five doped drug resistance alleles were found to be significantly enriched after the correction for multiple comparisons, and no additional mutations were called erroneously (). The measured frequencies of the drug resistance alleles, initially added at 5%, were L10: 4.6%, M46: 1.3%, L63: 4.8%, V82: 2.9%, I84: 3.1%. We conclude that the pyrosequencing method together with the above statistical procedure is sensitive and accurate enough to selectively detect drug resistance substitutions present as 5% of the viral population. Drug resistance mutations were identified using pyrosequencing in the three patient samples (, samples 5–7), and the calls compared to those called by the Viroseq HIV Genotyping System (analyzed using the HIV Drug Resistance Database at Stanford University). The pyrosequencing study identified all 15 of the drug-resistant mutations that were called using the Viroseq genotyping pipeline (, black bars without asterisks). The median proportion of mutations called by the Viroseq method was 88% drug resistance calls as measured in the pyrophosphate sequencing data (range: 28–99%). Four additional lower abundance drug resistance mutations were called in the pyrosequencing data that were not called by the Viroseq pipeline (, black bars with white asterisks), one for patient 1, one for patient 2 and two for patient 3 (, black bars with asterisks). The frequency of resistance mutations called by pyrosequencing only ranged from 11.6 to 0.65% of the total. A few positions were represented by many sequences, and appear to be just below the level of detection of the conventional genotyping assay. For example, the RT drug resistance substitution K70R was represented by 40 calls out of 305, and the RT mutation M184V was represented by 46 calls out of 726 (both from patient 3). At the other extreme, significant detection could be achieved by as few as nine drug resistance calls for codons in cases where the background was low—for the M46 position in patient 1, drug resistance for 9 out of 1377 calls achieved significance in the presence of a background of 1 out of 2395 calls ( = 0.0014, Fisher's exact test, two-sided comparison). The potential clinical significance of the minor drug-resistant alleles is discussed later. Eight resistance mutations were also identified in the collection of sequences from different HIV subtypes (, sample 4, data summarized in Supplementary Table 3). The HIV samples studied were isolated from patients worldwide in 1991 (). The subtype B sample was isolated in the United States, subtypes A, C and D in Uganda, and subtype 01_AE in Thailand. None of the patients were on antiretroviral therapy at the time of collection. Eight drug resistance mutations were identified in the PR-coding region, and none in RT. In some cases the resistance mutations were present as substantial fractions of the populations, such as the M36I substitution in subtype D, which comprises more than 78%. Others were less common, comprising ∼20% of the population. It is uncertain how many of these mutations, which appear to be pre-existing polymorphisms, would influence sensitivity to protease inhibitors (,). Taken together, these findings demonstrate the ability of the pyrosequencing method to identify minor drug resistance mutations in individuals infected with diverse HIV subtypes. Here we describe the use of DNA bar coding and pyrosequencing in detecting minor drug resistance mutations in HIV populations. We used pyrosequencing to generate 118 093 sequence reads from the region of seven samples of viral populations or controls. The seven samples were analyzed in a single sequencing experiment, made possible by use of DNA bar coding to distinguish the different samples. As controls we analyzed a DNA plasmid encoding HIV LAI and HIV LAI RNA from viral particles, allowing an empirical estimation of error at each codon at risk for mutation to drug resistance. To test the assay sensitivity, we analyzed a mixture containing HIV NL4-3 wild-type virions mixed with 5% of mutant virions containing five drug resistance mutations. All five mutations were called as present in the mixture without false positives, demonstrating the accuracy of the method. Analysis of viral populations from three patients harboring drug-resistant HIV revealed all the mutations called by the conventional genotyping method (Viroseq), plus four additional less abundant drug resistance mutations comprising from 11.6% to 0.65% of the population. Previously several methods have been reported for sensitive detection of rare drug-resistant mutations, some with impressive sensitivity (,). However, only the combination of DNA bar coding and pyrosequencing offers the opportunity to determine the full sequence of genomic regions with drug-resistant mutations from many samples in a single sequencing experiment. Using this method, sequence polymorphisms that were not specifically targeted for analysis can be identified and analyzed for possible correlations with drug resistance, which is not possible with most of the alternative methods. Here we analyzed seven viral populations in a single picoliter sequencing plate, but there is no reason that the number could not be much larger. In another study, we have successfully sequenced 42 different DNA bar codes in a single plate (unpublished data), indicating the potential. Several of the low abundance drug resistance mutations detected here are of potential clinical significance, since they may confer reduced sensitivity to drugs that otherwise might seem attractive for therapy. For example, patient 3 was found to harbor low level resistance alleles in RT at K70 and M184, which would be likely to impair therapy with several of the NRTIs, though not with all NRTIs. Patient 3 had been treated with NRTIs known to elicit these mutations, probably explaining their presence. Because alternative NRTIs are available for which all of the patient 3 viruses would remain sensitive, knowledge of the minor alleles could have improved the ability to choose effective therapy in this case. Looking forward, it will be important to test more fully the importance of minor HIV drug-resistant populations on antiretroviral responses and the impact of such information on treatment outcomes. Lastly, the methods described here may be useful in implementing drug resistance genotyping in resource-limited settings. Using the DNA bar coding strategy, it should be possible to multiplex large numbers of patient samples in single sequencing runs, thereby driving down costs. Though many logistical obstacles would need to be overcome, the combination of DNA bar coding and pyrosequencing offers a prototype technology for affordable HIV genotyping. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

DNA is a constant target of spontaneous hydrolysis at 37°C (). Two frequent hydrolysis reactions are depurination to produce non-coding abasic (AP) sites, and deamination of cytosine to generate uracil (U) (,). Minor base lesions and single-strand breaks are repaired primarily by base excision repair (BER) in mammalian cells. The first step of BER is removal of the damaged base by a DNA glycosylase, cleaving the -glycosyl bond between the base and deoxyribose (,). This results in forming apurinic or apyrimidic (AP) sites in DNA. As noted earlier, AP sites can also occur from depurination or depyrimidination yielding a common intermediate in BER. Subsequently, AP-endonuclease 1 (APE) incises the damaged strand 5′ to the AP-site generating a 3′-hydroxyl and a 5′ deoxyribose phosphate (dRP) (). Then, short-patch BER proceeds by action of DNA polymerase β (Pol β), filling the single-nucleotide gap and removing the dRP group (). Finally, the nick is sealed by DNA ligase I or III. Generally, replication and transcription are not significantly stalled at these types of base lesions. As a result, they can be mutagenic if not repaired () resulting in genomic instability and such chronic disorders as cancer. BER enzymes must deal with damage generated throughout the genome, and the majority of eukaryotic DNA is packaged into highly condensed structures (). As shown in previous studies using mononucleosomes (,), BER enzymes are significantly suppressed in mononucleosomes. However, recently, considerable flexibility of mononucleosome DNA has been reported (,), particularly at the sites of entry and exit from the nucleosome. Such dynamic properties of nucleosomes may allow certain small proteins to gain access to DNA. For example, it was found that Dnmt3a and Dnmt1 DNA methyltransferases act more efficiently on nucleosome DNA than naked DNA, and do not require disruption of histone octamers (). Furthermore, Thoma and colleagues () have shown that UV photolyase, a light-dependent DNA repair enzyme, recognizes and repairs CPDs in nucleosomes within seconds in intact yeast cells, suggesting this protein takes advantage of nucleosome dynamics to gain access to CPDs in chromatin. Although dynamic properties of nucleosomes may contribute to the accessibility of DNA, they are not sufficient for some bulky proteins or protein complexes to gain access to internal sites. Indeed, several different types of chromatin remodeling factors have been identified that assist DNA accessibility. One class of these factors changes histone-DNA contacts by adding or removing covalent modifications from histone tails. Another class alters chromatin structure in a non-covalent manner using ATP hydrolysis, and this class has been subdivided into different families based on the ATPase subunit (). The effect of ATP-dependent chromatin remodeling on overall DNA repair has been shown using mono- or di-nucleosomes (). Recently, we reported that the SWI/SNF remodeling complex facilitates nucleotide excision repair (NER) of UV damage in yeast cells (). However, the question of whether chromatin remodeling is required prior to damage recognition or during DNA repair has not been addressed. Furthermore, the initiation of DNA repair in highly compact chromatin remains an essential component for the understanding of BER in cells. Most BER enzymes are essential (,), making it more difficult to assess this pathway . However, a complete understanding of DNA repair will remain elusive until the impact of higher order chromatin folding on repair enzymes is understood. Therefore, we are establishing conditions that more closely resemble physiologic templates to study BER in chromatin. In this work, the process of short-patch BER was studied using purified human BER enzymes and highly folded oligonucleosomes containing DNA with uracil at different sites. We measured the efficiency of uracil removal by UDG and APE from G:U mismatch base pairs in oligonucleosomes reconstituted with 12 tandem repeats of a 208 bp segment (208–12) of 5S rDNA. In addition, we examined the effect of yeast chromatin remodeling complexes ISW1 and ISW2 on Pol β DNA synthesis in these oligonucleosome arrays. To our knowledge, this is the first study examining DNA repair at the 30 nm oligonucleosome fiber level of DNA packaging in chromatin and linking chromatin remodeling with BER. Reconstitution of nucleosomal arrays—Histone octamers were prepared from chicken erythrocyte nuclei as described (). Oligonucleosome arrays were reconstituted from histone octamers and purified 208-12 DNA template using salt dialysis methods (,). The degree of template saturation was controlled by varying the ratio of moles histone octamer to moles 208-bp DNA from 0.9 to 1.2. In our system, a ratio of 1.1 provided optimal efficiency of reconstitution. An aliquot of 700 ng of oligonucleosomes (or naked DNA) was digested with EcoRI restriction enzyme in a buffer containing 10 mM Tris, pH 8.0, 125 mM NaCl, 2.5 mM MgCl and 1 mM EDTA for 2 h at 37°C. The reaction was stopped by adding gel loading buffer, containing SDS, and the samples run in 5% polyacrylamide gels in 1× TBE. Sedimentation velocity experiments were performed in a Beckman XL-A analytical ultracentrifuge utilizing scanner optics at 260 nm. The temperature was equilibrated at 20°C under vacuum for at least 1h prior to the run and was controlled during the run to within ±0.1°C. Scans were analyzed by the method of van Holde and Weischet () using UltraScan 7.4 (B. Demeler, San Antonio, TX, USA). The UDG and APE reaction mixture contained 50 mM Hepes, pH 7.5, 2 mM DTT, 0.2 mM EDTA, pH 8.0, 100 μg/ml BSA, 10% glycerol (wt/vol) and 2.5 mM MgCl. A lower Mg concentration was used than in previous studies to prevent oligonucleosomes from aggregating. Reactions were initiated by adding UDG and APE (1 nM or 10 nM final concentration), and incubations were carried out at 37°C for 0 to 1 h. Aliquots were removed at different times and treated with phenol to stop the reaction. Digested DNA was resolved on denaturing alkaline gels, transferred to a Hybond-N+ membrane (GE Healthcare, England), probed with randomly labeled 5S rDNA fragments, and visualized on a PhosphorImager (model 445-P90, Molecular Dynamics). Images were analyzed with IMAGE QUANT software (Molecular Dynamics). To uniformly enhance the signal of each fragment after UDG/APE digestion, the resulting ss-fragments (up to 2.5 kb) were annealed with biotin-labeled primers and purified using streptavidin magnetic beads, as previously described (). Briefly, biotin-attached oligonucleotides, containing a sequence complementary to the 3′-end of 208-12 DNA ([biotin]NNNNNNTTTTTGCATGCCTGCAGGTC), were synthesized and the biotin group was used to separate annealed fragments from the rest. The 6 Ns ensured full-length labeling of the fragment, and the 5 Ts allowed the annealed fragment to be extended with [α-P]dATP. Labeled fragments were separated in a 5.5% polyacrylamide denaturing gels. DNA synthesis with Pol β was performed in a mixture containing 50 mM Hepes, pH 7.5, 2 mM DTT, 0.2 mM EDTA, 100 μg/ml BSA, 10% glycerol (wt/vol), 4 mM ATP, 5 mM MgCl and [α-P]dCTP. Templates were first incubated with UDG and APE (10 nM each) for 10 min. DNA synthesis was then initiated by addition of Pol β and incubations were at 37°C for 0–4 h. Aliquots were removed at different times and treated with phenol to stop the reaction. Samples were then run on a native agarose gels and the gels stained with ethidium bromide. Gels were then blotted onto membrane, visualized on a Phosphorimager and the images analyzed with IMAGE QUANT software. Pol β DNA synthesis reactions were carried out with oligonucleosomes under identical conditions as above, with or without chromatin remodeling complex. Remodeling complexes, yISW1 or yISW2 (generous gift of Dr Tsukiyama) were added at the beginning of repair synthesis. The molar ratio of ISW1 or ISW2 to mononucleosomes was 1. Nucleosome arrays containing G:U mismatches can form higher order structures. The experiments described in this report utilize a DNA fragment composed of 12 tandem repeats of a 208 bp segment of the 5S rRNA gene as a template for reconstitution of oligonucleosomes (A). Each repeat contains sequences that position nucleosomes both translationally and rotationally on the DNA molecule (). To study short-patch BER, uracil was incorporated at cytosine bases in the 208-12 DNA fragment by treatment with sodium bisulfite. This reaction was optimized to yield a majority of the single-stranded (ss) fragments with 1 or 2 uracils. For example, 35% of the fragments used for UDG/APE digestion (1 nM each) contained a single uracil, 23% contained two uracils, and 15% of the fragments contained more than two uracils. DNA-stripped histone octamers from chicken erythrocytes were reconstituted by stepwise salt dialysis (,) onto a non-damaged 208-12 template or a 208-12 template containing G:U mismatches. Initially, varying ratios (r) of histone octamers to 5S rDNA repeats were used for reconstitution to generate a fully loaded oligonucleosome array. To evaluate the degree of nucleosome loading, two different methods were used. First, gel mobility shift assays, following restriction digestion with EcoRI, were used. Since each repeat is flanked by EcoRI sites (A), a mononucleosome or 208-1 free DNA is released after EcoRI digestion. When only about 2–5% of the 5S repeats are released in the free form, the 208-12 rDNA is considered to be fully saturated with nucleosomes (). We obtained fully saturated oligonucleosomes when r = 1.1 (). Importantly, no significant differences in efficiency of oligonucleosome assembly were detected between intact 208-12 oligonucleosomes and those containing uracil (B). This result was expected since the G:U-mismatch has the capacity to form base pairs (). In addition, no additional disruption of the core particles was observed in the reconstituted oligonucleosomes in the BER reaction mixture buffer employed in this study (B). Sedimentation velocity analysis was also carried out to monitor the homogeneity of reconstituted oligonucleosome arrays, as well as the degree of salt-induced higher order structure formation. The majority of nucleosomal arrays both with and without uracil residues sediment as rather homogeneous populations of 28–30S species in low ionic strength buffer (C). The small fraction of material with larger S values was due to super-saturated nucleosome arrays assembled with additional histones. Furthermore, as the BER buffer used in this study contains Mg, the Mg-dependent folding of 208-12 oligonucleosomes was characterized. Hansen and colleagues have shown that saturated 208-12 nucleosome arrays form a maximally folded 55S structure in 1–2 mM MgCl, with an extent of compaction equivalent to the classical higher order 30 nm structures (). In the presence of 2 mM Mg, S values of both intact oligonucleosomes (open symbols) and uracil-containing oligonucleosomes (closed symbols) were shifted to higher values (more compact structures) in a similar manner (C). These results suggest that intramolecular folding of uracil-containing oligonucleosomes in solution is not disrupted in any significant way by G:U mismatches. UDG and APE recognize and act on base lesions in folded oligonucleosomes. Incubation of either uracil-containing naked 208-12 DNA or oligonucleosomes with UDG and APE generates a single strand break that can be visualized on denaturing agarose gels (). When digested with 1 nM each of UDG and APE, shorter bands are generated and the intensity of full-length (undigested) fragments is reduced during increasing digestion times (A). Furthermore, the disappearance of full-length fragments is slower with oligonucleosomes than with naked 208-12 DNA (A, graph), suggesting that the combined activity of these enzymes is reduced when DNA is folded into nucleosome arrays. The calculated initial rates for these digestion curves indicate that nucleosome arrays are digested ∼2–3-fold slower by the combined action of UDG and APE. At higher enzyme concentrations (10 nM each), however, the digestion reaction proceeded to near-completion with folded oligonucleosomes (B). This result indicates that these enzymes can access G:U sites within higher order structures of chromatin and do not require a marked disruption of DNA-histone contacts. High resolution mapping of UDG/APE cleavage sites at uracils can be observed on DNA sequencing gels (C). Overall, incision at damaged sites within linker DNA regions (between ovals) was faster than in core DNA (within ovals). However, a few sites within each of the nucleosome core particles were cleaved considerably faster than others (C, red arrows). The rate of cleavage at these sites may reflect (i) the influence of flanking sequences in each of the 5S rDNA repeats, as observed for other DNA sequences (), and/or (ii) the location of these sites at more accessible locations in the nucleosome cores (e.g. facing away from the histone surface in each nucleosome). In addition, some incised sites are detected within each of the repeats in the absence of UDG/APE digestion (C, dotted arrows on left of 0 min lane), indicating that some fragments contained nicks at specific sites following the sodium bisulfite treatment. Although DNA was treated for short times with high concentrations of bisulfite to limit these nicks, the glycosyl bond at purine residues is still susceptible to hydrolysis under acidic conditions. The resulting AP sites lead to single strand breaks during desulfonation under alkaline conditions. However, because these nicked sites do not change during the UDG/APE digestion, they serve as an intrinsic control for the enzyme-produced sites (C, dotted arrows). Gap-filling by DNA Pol β is inhibited by oligonucleosome formation. To assess activity of Pol β, the single-nucleotide repair patch was labeled with [α-P]dCTP. At a concentration of 1 nM, Pol β efficiently incorporates dCMP into naked DNA and reaches a plateau after 90 min ( inset, open circles). On the other hand, at this concentration, Pol β only incorporates dCMP into ∼20% of the sites in oligonucleosomes ( inset, closed circles). The oligonucleosome template used in this study includes about 60 bp of linker DNA between nucleosome core particles (i.e. ∼30% of the total DNA). Therefore, folded nucleosomes may affect the efficiency of DNA synthesis in the linker DNA region. Indeed, even at a 10-fold higher concentration of Pol β (10 nM), DNA synthesis on oligonucleosomes proceeded to a plateau of 37% in ∼60 min (), indicating that nucleosome core particles are a formidable barrier for the efficient addition of nucleotides by Pol β in oligonucleosomes. Therefore, it is possible that other factor(s) assist this step in intact cells. Chromatin remodeling complex ISWI facilitates DNA synthesis by Pol β in oligonucleosomes. To investigate whether ATP-dependent chromatin remodeling facilitates Pol β DNA synthesis, purified yeast remodeling complexes ISW1 and ISW2 () were tested with our BER-oligonucleosome model system. ISWI complexes induce nucleosome sliding both and (), which makes DNA segments accessible while maintaining the overall packaging of DNA (). The ISWI complex is present in all eukaryotes, with the exception of (), and is relatively abundant. For example, ISWI is expressed during development at levels as high as 100 000 molecules/cell (). Moreover, ISWI induces nucleosome sliding on nicked DNA (), which makes the ISWI remodeling complex a good candidate for assisting the Pol β step of the BER pathway in cells. A quantitative comparison of the synthesis of Pol β on oligonucleosomes in the absence and presence of these complexes was made by setting the maximum DNA synthesis achieved on oligonucleosomes in the absence of these factors to 1. As shown in , both of the yeast ISWI complexes promote efficient dCMP incorporation at cleaved uracil sites in the oligonucleosome templates. Indeed, after 4 h, ISW1 and ISW2 increased incorporation by ∼4-fold and ∼6-fold, respectively. Thus, both remodeling complexes significantly increase the accessibility of nucleosome DNA to Pol β. In facing the steric hurdles to surveying the genome, DNA repair proteins may take advantage of the dynamic properties of chromatin and/or rely on other factors to remodel local regions of DNA damage. It has been shown that nucleosome structure reduces the accessibility of many different types of DNA processing proteins, and chromatin remodellers such as SWI/SNF enable the access by these proteins (,). In the case of DNA repair, the question arises as to whether cells require chromatin remodeling prior to or during the repair process. Here, we show that two different chromatin states may be required for efficient BER by systematically examining BER , using highly folded nucleosome arrays of the 208-12 5S rDNA sequence, that more closely resemble natural substrates in intact cells than mono- (or di-) nucleosomes. Uracil-containing DNA does not inhibit nucleosome formation or compaction. In this study, chemical deamination of cytosine to uracil was carried out to generate 1–2 uracils/ss-fragment (i.e. ∼2–4 G:U mismatches in each array template). A G:U mismatch has the capacity to form a base pair (), which is most likely why these lesions do not affect nucleosome formation or compaction (). This aspect of non-distorting DNA lesions makes their recognition more difficult than that of helix distorting DNA lesions such as cyclobutane pyrimidine dimers caused by UV light, which bend the long axis of DNA ∼ 30° (). Furthermore, our sedimentation velocity results demonstrate that the reconstituted nucleosome arrays containing G:U mismatches form higher-order structures at increased ionic strength. Finally, cytosine deamination occurs more frequently where a segment of ssDNA is temporary exposed during transcription and replication (). Our data supports the notion that such uracil-containing DNA can reassemble into nucleosomes and fold back into more compact structures in intact cells, if not repaired. UDG and APE can access DNA base damage even in highly folded regions of chromatin. Differences occur in the activities of UDG and APE at different sites of oligonucleosomes, depending on the predicted rotational settings of the lesions as well as the flanking sequences (C), in agreement with previous studies on mononucleosomes (,). More importantly, however, these enzymes digest the highly folded arrays to completion in a concentration-dependent manner, indicating that both enzymes are capable of accessing dU-damaged sites within nucleosomes in a higher-order structure of chromatin (). Moreover, the strong inhibition of Pol β activity by oligonucleosomes following digestion with high concentrations of UDG/APE () indicates there is no significant nucleosome disruption by UDG/APE digestion. Whether chromatin remodeling or recognition of DNA lesions would come first during BER in cells is unclear at this time. Phosphorylation of histone variant H2AX following induction of DNA double-strand breaks (DSBs) by ionizing irradiation (,) is known to play an important role in recruiting DSB-recognition and repair proteins (), chromatin remodeling factors () and DNA damage-induced checkpoint proteins (). However, unlike induced DSBs, different types of minor base alterations occur constantly throughout the genome and, to date, there is no evidence showing that a specific histone modification is formed following DNA base damage. Our data suggests that, at least for BER, a variety of base lesions could be recognized directly by a substrate specific glycosylase in chromatin without requiring significant local chromatin remodeling. UDG and APE are relatively small, and bend the long axis of DNA only ∼20° and ∼35°, respectively, upon binding (,). Such bending may be compatible with the DNA flexibility allowed on the nucleosome surface. Furthermore, such ‘direct recognition’ by these enzymes may be critical for the cell to initiate BER in chromatin, especially in heterochromatin. Pol β DNA synthesis requires local chromatin remodeling. Unlike the UDG and APE enzymes, Pol β is strictly inhibited by the folding of DNA into nucleosome arrays (). In addition to our previous observation on mononucleosomes (), our results clearly indicate that the major restriction of BER in oligonucleosome templates is the Pol β step. However, our new results indicate there are two levels of inhibition of Pol β in oligonucleosomes. At a lower concentration of Pol β (1 nM), ∼80% of the total gaps are not filled (, inset) in oligonucleosome arrays, which is equivalent to the portion of potential uracil sites (i.e. dC sites) in 168 bp of nucleosomal DNA within the 208 bp repeat. However, when the concentration of Pol β is increased 10-fold, ∼63% of total gaps remain unfilled (), which is almost equivalent to the fraction of potential uracil sites in the 147 bp nucleosome core DNA (65%). This indicates that the outer 10 bp of DNA at the points of entry and exit to the nucleosome are more accessible in oligonucleosomes than the inner 147 bp of core DNA. Pol β binding requires an ∼90° bend in the DNA molecule (), and such a radical distortion may not be tolerated by nucleosomes. This result may reflect the requirement for additional factor(s) to release the structural constraints of positioned nucleosomes in intact cells. Since BER intermediates such as single-strand breaks are also mutagenic, it is crucial that, once initiated, completion of BER occurs rapidly in cells. Therefore, we explored possible solutions for Pol β to overcome such structural barriers by examining the effects of two yeast chromatin remodeling complexes, ISW1 and ISW2, on Pol β DNA synthesis in the oligonucleosome templates. Both of these complexes significantly facilitated Pol β DNA synthesis on the folded oligonucleosome arrays (). This observation may reflect a need for ATP-dependent chromatin remodeling prior to the Pol β step of BER in chromatin. Different chromatin structural states may be required for initial and latter steps of BER. Based on our results, we propose a mechanistic model for short-patch BER in chromatin (). DNA glycosylases, which are relatively abundant for frequently occurring base damage (), continually ‘survey’ the genome and cleave glycosyl bonds of damaged bases (). Subsequently, APE, which is present at high amounts in cells (350 000–700 000 molecules/cell), will incise at AP sites (,). Both of these enzymes may act directly on chromatin without requiring significant nucleosome disruption, possibly using the dynamic unwrapping of nucleosome DNA to gain access to internal nucleosome sites. In contrast, Pol β exists at lower levels in cells, ∼50 000 molecules (), and may only be able to access strand breaks in nucleosome-free DNA, or linker DNA regions. Access of UDG/APE-induced strand breaks within nucleosome core particles, however, may require nucleosome sliding or disruption prior to gap-filling synthesis. After removal of the dRP group by Pol β, the nick is sealed by DNA ligase. Completion of the repair process would then involve rearrangement of the repaired DNA back to the original chromatin structure (). Finally, Chambon and colleagues reported that thymine DNA glycosylase (TDG) associates with transcriptional coactivators CBP and p300. Thus, acetylation of histone tails by the CBP/p300-TDG complex may promote local chromatin relaxation at these sites. However, relaxation of closed chromatin structure to open form may only be sufficient for Pol β DNA synthesis in linker DNA regions and may not be sufficient for access of Pol β to single-strand breaks in nucleosome core DNA. Indeed, preliminary results do not show a marked promotion of DNA synthesis on oligonucleosomes in the presence of a histone acetyl transferase (HAT1) (Nakanishi and Smerdon, unpublished results). Therefore, complete exposure of nucleosome core DNA, by either nucleosome sliding or disruption may be required to allow Pol β complete access to damaged sites. However, histone acetylation may have important roles in recruitment of additional factors, such as chromatin remodeling complexes, linking these two different chromatin states.

RNA interference (RNAi) is induced by double-stranded RNA (dsRNA) and results in gene silencing through sequence-specific degradation of the target RNA (). RNAi provides plants and animals a defense mechanism against viruses () and retrotransposons (,). The ribonuclease Dicer processes the long dsRNA replication intermediates into small interfering RNAs (siRNAs) of ∼22 nucleotides (nt) (). These siRNAs are incorporated into the RNA-induced silencing complex (RISC) that finds complementary RNA sequences, resulting in cleavage of the target RNA (,). The central catalytic component of RISC is an Argonaute protein, which contains the signature domains PAZ and PIWI responsible for binding the siRNA strand (). Transfection of synthetic siRNAs into cells or intracellular expression of short hairpin RNAs (shRNAs), which are processed into siRNA duplexes by Dicer, are powerful tools to suppress gene expression (). Randomly selected siRNAs against a target show a large variation in their efficacy (). Empirical rules on siRNA duplex features have been reported and improve design of effective siRNAs. The asymmetry rule for siRNA duplex ends requires that the 5′ end of the antisense strand forms a less stable end with its complement than the 5′ end of the sense strand (,). Related to this rule is the described requirement of high A/U content at the 5′ end of the antisense strand and high G/C at the 5′ end of the sense strand (,). In addition, a number of position-specific nucleotides, an unstructured guide-RNA, and an accessible target site have been reported to positively effect siRNA efficiency (,). RNAi can be used as a therapeutic strategy against human pathogenic viruses such as HIV-1 (). HIV-1 replication can be inhibited transiently by transfection of synthetic siRNAs targeting viral RNA sequences or cellular co-factors (). Furthermore, long-term inhibition of HIV-1 replication has been demonstrated in transduced cell lines stably expressing antiviral siRNAs or shRNAs (). However, HIV-1 escape variants with nucleotide substitutions or deletions in the siRNA target sequence do emerge after prolonged culturing (,,). The emergence of RNAi-resistant variants may be blocked by a combination-shRNA therapy, which simultaneously targets multiple conserved viral RNA sequences (,). We demonstrated that HIV-1 can also become resistant against RNAi by placing the target sequence in a stable RNA structure, which prevents binding of the siRNA (). We also suggested that such structure-based target occlusion occurs in the RNA genomes of lentiviral vectors with a shRNA-cassette (). By inserting these cassettes, the target sequence will automatically be present in the vector genome, and self-targeting by the shRNA should reduce the lentiviral production level. However, since the target sequence in the genome is also located in this perfect shRNA hairpin, it is protected against RNAi, ensuring a normal vector titer. Indeed, when the target in the lentiviral genome is unstructured, the titer is significantly reduced by the shRNA (). The inhibitory effect of target RNA structure on RNAi efficiency has been described in several studies (,,). These studies compared the efficiency of different siRNAs on a fixed target, and found a correlation between target availability and RNAi efficiency. Schubert . suggested that the local free energy of base pairing in the target region determines RNAi efficiency (). Ideally, one should test this concept by a mutational analysis of one target instead of comparing different siRNAs with intrinsically different RNAi efficacies. In this scenario, mutations that affect the RNA structure should not affect the target sequence itself, such that the same siRNA inhibitor can be used. In this study, we set out to determine the exact hairpin stability at which RNAi suppression occurs by systematically destabilizing a 21-base pair (bp) hairpin structure that occludes the complete target sequence. We monitored the effects on siRNA binding and RNAi efficiency . The 3′ end of the mRNA target sequence is initially recognized by bases 2–5 of the antisense/guide strand siRNA, therefore named the ‘seed’ sequence (,). Thus, one may expect a more prominent effect of an accessible target 3′ end, which primed us to address positional effects when destabilizing the target hairpin. The results demonstrate a clear correlation between the overall stability of the target hairpin and RNAi efficiency, but positional effects were also apparent. The luciferase plasmids pGL3-wt, pGL3-T1 to pGL3-T7 (B) and pGL3-A to pGL3-G (A) were constructed by annealing of forward (fwd) and reverse (rev) oligonucleotides (Supplementary Data, Table 1) and ligation into the EcoRI and PstI sites of the firefly luciferase expression vector pGL3-Nef (). The pSUPER-shPol vector () encodes an effective shRNA against a conserved 19-nt HIV-1 region (Pol1; ACAGGAGCAGAUGAUACAG) under the control of an H1 polymerase III promoter (). The plasmid pRL-CMV (Promega) expresses Renilla luciferase under control of the CMV promoter. C33A cervix carcinoma cells were grown as a monolayer in Dulbecco's modified Eagle's medium supplemented with 10% FCS, minimal essential medium nonessential amino acids, 100 units/ml penicillin, and 100 units/ml streptomycin at 37°C and 5% CO. C33A cells were grown in 1 ml culture medium in 2 cm wells to 60% confluence and transfected by the calcium phosphate method. The pGL3-variant (100 ng) was mixed with 0.5 ng pRL-CMV, 0.1–100 ng pSUPER-shPol and pBluescriptII (KS) (Stratagene) to have 1 μg of DNA in 15 μl water. The DNA was mixed with 25 μl of 2× HBS and 10 μl of 0.6 M CaCl, incubated at room temperature for 20 min and added to the culture medium. The culture medium was refreshed after 16 h, and cells were lysed after another 24 h. Firefly and Renilla luciferase activities were measured with the Dual-luciferase Reporter Assay System (Promega) as described previously (). C33A cells (2 cm) transfected with 100 ng pGL3-variants and 0 or 10 ng pSUPER-shPol were lysed 2 days after transfection. Total RNA was isolated with TRIZOL® reagent (Invitrogen) according to the manufacturer's protocol. Contaminating genomic DNA was removed by DNase treatment using the TURBO DNA-free™ kit (Ambion). First strand cDNA was synthesized using 1 µg of total RNA, Thermoscript™ reverse transcriptase (Invitrogen), and primers. The gene-specific primers used were EWr6 (5′-GCCCCGACTCTAGACTGCAG-3′) for Firefly luciferase and 3′HC-b-ACTIN (5′-TGTGTTGGCGTACAGGTCTTTG-3′) for actin. The pGL3-variant plasmids were used as template for PCR amplification with primers EWr8 (5′- TCCTTCCCC-3′; T7 RNA-polymerase promoter in italics) and EWr9 (5′- GACTCTAGACTGCAGAAA -3′). The resulting PCR product contains a T7 RNA-polymerase promoter upstream of the hairpin (hairpin nt underlined). DNA products were purified from agarose gel using QiaexII Gel extraction kit (Qiagen). RNA transcripts were produced by transcription with the Megashortscript T7 transcription kit (Ambion), and transcripts were checked for integrity and isolated from an 8% acrylamide gel. RNA concentrations were determined by spectrophotometry. The siRNA-Pol antisense/guide oligonucleotide CUGUAUCAUCUGCUCCU-GU (Eurogentec) was 5′ end labeled with the kinaseMax kit (Ambion) and 1 μl [γ-P] ATP (0.37 MBq/μl, Amersham Biosciences). The target hairpin RNAs were denatured in 30 μl water at 85°C for 3 min followed by snap cooling on ice. After addition of 10 μl 4× MO buffer (final concentration: 125 mM KAc, 2.5 mM MgAc, 25 mM HEPES, pH 7.0), the RNA was renatured at 37°C for 30 min. The transcripts were diluted in 1× MO buffer to a final concentration varying from 0 to 1.0 μM in MO buffer. Unlabeled tRNA (1 μg) was added as competitor to each reaction to minimize aspecific RNA interactions. The 5′-labeled oligonucleotide (1.0 nM) was added and the samples (20 μl) were incubated for 30 min at 37°C. After adding 4 μl non-denaturing loading buffer (50% glycerol with bromophenol blue), the sample was analyzed on a non-denaturing 4% acrylamide gel. Electrophoresis was performed at 150 V at room temperature and the gel was subsequently dried. Quantification of the free and bound oligonucleotide was performed with a Phosphor Imager (Molecular Dynamics). The structure and stability of the target hairpins cloned into the pGL3-variants was predicted with the RNA Mfold program (,) at . The indicated Δ in B and A are derived by importing the hairpin sequences into the program (54 nt total: 5′ CCCC + hairpin sequences + UUU) and did not contain luciferase sequences. The presence of the predicted hairpin structures in the context of the luciferase reporter construct was verified by importing longer sequences (150 nt total; 52 nt + hairpin sequences + 51 nt) into Mfold. We investigated the effect of target RNA structure on RNAi efficiency. As a model system we used a very potent shRNA inhibitor that is directed against the Pol gene of HIV-1 (A; left) and that has been tested extensively against HIV-1 and appropriate reporter genes (). Such a luciferase reporter with the HIV-1 Pol target sequence in the 3′ UTR is shown in A (right). Next, we made the target inaccessible by inclusion in a perfect hairpin of Δ = −36.6 kcal/mol (B; wild type (wt), the target sequence is marked in gray). In fact, this hairpin structure is identical to the shRNA itself. The top 2 bp and the 5-nt loop are standard in the optimized pSUPER system (). We systematically destabilized this target hairpin in mutants T1–T7 by introducing nt substitutions in the descending strand of the stem (encircled in B), thus leaving the target sequence intact. The mutations were chosen such that the predicted thermodynamic stability (Δ) decreases gradually. We first destabilized the hairpin by replacing stable G-C by weak G-U base pairs (mutants T1–T3), followed by more gross destabilizations, e.g. by introducing mismatches (mutants T4–T7). The Δ value was reduced in a step-wise manner to −7.2 kcal/mol for mutant T7. To accurately quantify the RNAi efficiency against these differentially structured targets, we placed them downstream of the luciferase reporter gene (A; right). These constructs were co-transfected into cells with increasing amounts of the shRNA-Pol expression vector and luciferase expression was measured after 48 h (C). Expression of the reporter construct with the target sequence embedded in the wt hairpin was completely resistant against shRNA-Pol. The same expression pattern was observed for the T1 construct, but T2 already showed some susceptibility for RNAi-mediated inhibition with higher amounts of shRNA-Pol, with a maximal inhibition of 34% (66% residual luciferase expression). The next reporter constructs (T3,T4) showed a significant drop in luciferase expression (64% inhibition). Inhibition of the remaining destabilized target hairpins (T5–T7) was very effective, showing more than 80% inhibition. This is similar to the maximal inhibition level that can be obtained with this potent shRNA inhibitor against a reporter with the 19-nt target sequence in an unstructured setting [() and results not shown]. To verify that the reduction of luciferase expression is due to mRNA degradation, we performed a semi-quantitative RT-PCR on cellular RNA (E). Consistent with the luciferase assays, the levels of mRNA are increasingly diminished for transfections with the constructs T2 to T7 and shRNA-Pol. The near absence of PCR product for construct T0, with or without shRNA-Pol, indicates an inefficient RT reaction through a perfect hairpin. There were no PCR products when RNA was used as input for the PCRs (results not shown). We plotted the measured level of luciferase expression against the predicted stability of the target hairpins (D). The results suggest an inverse linear correlation between RNAi-susceptibility and target hairpin stability in the −30/−15 kcal/mol range. The curve shows two plateaus. A reduction in hairpin stability from −36 to −30 kcal/mol does not significantly induce RNAi-mediated inhibition (<20% inhibition), and further destabilization above −15 kcal/mol shows no significant improvement of the already maximal inhibition of ∼86%. To demonstrate that the increased RNAi efficiency on destabilized target hairpins is due to more efficient binding of the siRNA, we performed binding experiments by means of electrophoretic mobility shift assays (EMSA). For this, we used short T7 transcripts with the complete hairpin and a 19-nt RNA oligonucleotide, which corresponds to the antisense/guide strand of the siRNA-Pol (complementary to boxed sequence in B). The siRNA was radioactively labeled and incubated with increasing amounts of target transcript (wt, T1–T7), and subsequently analyzed on a non-denaturing acrylamide gel (A). Binding of siRNA to the target RNA leads to duplex formation that results in a band shift on the gel. Unbound siRNA and the siRNA/target RNA duplex were quantified to calculate the percentage of binding (B). We performed the binding experiment multiple times with 0.2 µM target RNA because efficient binding can be observed, yet most variants stay within the linear range of the binding assay. We plotted these binding percentages against the predicted stability of the target hairpins (C). A general trend can be observed that is the opposite of the graph in D: reduced hairpin stability results in more efficient binding of the siRNA to the target RNA. Thus, a decrease in the stability of the target hairpin increases RNAi efficiency due to more efficient binding of the siRNA. The largest improvement in RNA–RNA interaction and RNAi efficiency is observed for mutant T3 in comparison with T2, indicating that a threshold stability is passed by going from Δ = −27.1 to −21.7 kcal/mol. We globally determined the stability at which hairpin structures become inhibitory to the RNAi machinery. However, not all domains within the 19-nt target sequence may contribute equally to siRNA binding and the RNAi mechanism. For instance, it has previously been suggested that the 3′ end of the target sequence is initially recognized by the siRNA within RISC (). To test this, we made a second set of Luc-target constructs (A, mutants A–G). By introducing clustered mutations in the target hairpin, we destabilized either the 3′ end, the center or the 5′ end of the target sequence. Modest G-U changes were introduced in mutants A (3′), B (center) and C (5′). More gross destabilizing mutations were introduced in mutants D (3′), E (center) and F (5′). However, it is apparent that the two mutations in D have a more modest effect on the Δ value because a realignment of the sequences trigger an alternative folding of the top of the hairpin. We therefore constructed the additional mutant G with three mutations to obtain a hairpin with a destabilized 3′ target end that is comparable in Δ to hairpins E and F. Target hairpins A through G were cloned in the luciferase reporter and co-transfected into cells with increasing amounts of the shRNA-Pol expression vector to quantify the RNAi efficiency (results not shown). The luciferase values obtained with 10 ng shRNA-Pol were plotted against the predicted hairpin stability (B, left) and we zoom in on a smaller Δ range (B; right graph). The target hairpins A, B and C follow the general trend that we described previously (gray dotted trend line). Independent of where the hairpin is destabilized, the introduction of G-U base pairs is a too modest manipulation to trigger RNAi activity. The target hairpins E, F and G have more dramatic changes that reduce the overall hairpin stability to −25/−26 kcal/mol, which should become susceptible to RNAi according to the previous results. However, mutants F (5′) and E (center) remain largely insensitive, but mutant G with a free 3′ end shows increased RNAi sensitivity when compared to the trend line. Even the D mutant with a more modest destabilization of the target 3′ end shows reasonable RNAi activity and clearly drops below the trend line. These results confirm the importance of initial recognition of the 3′ target end, which explains the deviations from the general trend. We performed binding experiments to study the A–G mutants for their ability to bind the siRNA. The radioactively labeled siRNA was incubated with increasing amounts of the target transcripts A–G and analyzed on gel (A). The shifts representing the siRNA/target RNA duplex and the free siRNA bands were quantified to calculate the percentage of binding (B). The percentage of binding with 0.2 µM target RNA was plotted against the predicted stabilities of the target hairpins (C). Remarkably, these binding results differ significantly from the RNAi results. The target hairpins D and G (both 3′), which are efficiently targeted by RNAi in the luciferase assay (B), are inefficient in siRNA binding. In contrast, the target hairpins F (5′) and G (center) showed a slightly increased binding efficiency, although these construct were relatively more RNAi resistant in the luciferase assay. These results may reflect the oversimplification of the binding assay and point to a contribution of the RISC/siRNA complex in the recognition and binding of the target sequence . It has been proposed that RNAi efficiency is influenced by the local RNA structure of the targeted sequence. We investigated this phenomenon in detail by placement of the target sequence in a perfect hairpin structure (Δ = −36.6 kcal/mol), which indeed resisted RNAi. Subsequently we destabilized this tight target structure resulting in a gradual exposure of the target sequence. Destabilization of the hairpin structure has little effect on RNAi activity until a threshold is reached (Δ ≈ −30 kcal/mol). Beyond this threshold we demonstrate an inverse correlation between hairpin stability and RNAi-mediated inhibition. Maximal RNAi efficiency was observed with hairpins of Δ ≥ −15 kcal/mol. binding experiments suggested that the increase of RNAi-mediated inhibition is due to efficient siRNA binding to the destabilized target RNA hairpins. When we introduced position-specific mutations in the target hairpin, we observed RNAi efficiencies that deviate from this trend. Hairpins with an opened 5′ end or central part of the target sequence show less RNAi activity than predicted based on their overall stability. In contrast, hairpins with an opened 3′ end are more susceptible to RNAi than expected. These results are consistent with the current notion that the 3′ region of the target is initially recognized and bound by the RISC/siRNA complex (). This model is supported by structural data on RISC bound to the siRNA strand. The 3′ end of the siRNA is recognized and bound in a pocket by the PAZ domain of the Argonaute protein (). The 5′ end of the siRNA is anchored at the PIWI domain of Argonaute and these 5′ nucleotides are readily accessible for base pairing to complementary 3′ nucleotides of the target RNA (,). The importance of the target 3′ end was also revealed in experiments that selected for viruses that resist RNAi-mediated inhibition. We described a unique HIV-1 escape variant that acquired a mutation outside the 19-nt target, which forces the RNA into an alternative structure that occludes the 3′ end of the target (). Besides the RNAi measurements, we also tested the different RNA targets for their ability to interact with the siRNA . The overall Δ effect of stable target hairpins is confirmed in this simplified setting, demonstrating that RNAi resistance is due to the inability of the siRNA to interact with the base-paired stem of the hairpin. We realize that the siRNA does not act by itself as it is part of RISC, of which the helicase activity may affect local structure in the target RNA (). In fact, we observed an interesting discrepancy between the and results for the 5′/center/3′-destabilized hairpins. We observed that an accessible 3′ target is key for RNAi activity, but this effect was not seen . This result may indicate an important contribution of RISC in the siRNA-target RNA annealing step. Thus, target RNA structure is an important factor when selecting a suitable target sequence, as it can have a negative effect on RNAi efficiency. For instance, it has been shown that the TAR hairpin of the HIV-1 genome is an unsuitable target because of its tight structure (,,). On the other hand, it is obvious that an accessible sequence does not automatically make a good siRNA target (), as the matching siRNA may not meet the criteria of an effective siRNA (). It has been proposed to include a calculation of the amount of hydrogen bonds within the target sequence as a parameter for efficient target sequences (). We provide a Δ threshold at which an hairpin RNA structure becomes inaccessible, and we differentiate between different target positions. When designing antiviral siRNAs one may also consider ways to obstruct viral escape via folding of an alternative target RNA structure (). The local RNA region should be screened for the absence of alternative foldings that occlude the 3′ end of the target and that can be selected by one or two mutations. If not available, the genetic threshold for structure-based escape might prove too high, even for a fast evolving virus like HIV-1. RNA structure-mediated resistance against RNAi is in fact beneficial when expressing highly structured shRNAs or miRNAs in cells. For instance, the incorporation of shRNA cassettes in a lentiviral vector is potentially problematic, because the shRNA will target the viral RNA genome during vector production, thus reducing the titer. Such self-targeting has not been reported (,), we think because the target is not accessible as part of the perfectly base-paired shRNA hairpin. The apparent absence of such self-targeting is particularly important for the development of multi-shRNA lentiviral vectors without titer reduction. However, placing many tight RNA structures in the vector genome may negatively influence the titer by other means. For instance, reverse transcription is very sensitive to excessively stable RNA structure () and RNA polymerase II transcription may pause at sites where the RNA products folds stable hairpin structures (). We did indeed observe that four shRNA cassettes reduce the lentiviral vector titer (ter Brake, unpublished data). Destabilizing the introduced shRNAs may avoid such vector problems, and provide additional benefits for cloning and sequencing of inverted repeat sequences (). In our target model system, we mutated the antisense strand of the shRNA hairpin, leaving the sense target sequence intact. In the case of a true shRNA expression cassette, modifications will be made in the sense (target) strand to leave the guide/antisense siRNA strand unaltered. The obvious advantage will be reduced complementarity between the target and the siRNA inhibitor. The impact of such mutations on self-targeting is likely to depend on the position and type of mismatches that are introduced (,). It is therefore impossible to make general rules for shRNA design and destabilization as each hairpin RNA structure will have its unique characteristics as target and effector in the RNAi mechanism. Here we demonstrate a Δ window for shRNA-Pol destabilization without activating RNAi self-targeting, which may provide a guideline for other shRNAs. Positional effects should be considered, and hairpins may be destabilized to Δ = −25 kcal/mol as long as the target 3′ end remains base-paired. It is too early to define more general guidelines for structured RNA motifs other than the man-made, perfectly base-paired shRNA hairpins, as natural RNA structures differ significantly in their topology and architecture. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Small regulatory RNAs (sRNAs) are expressed in both prokaryotes and eukaryotes, primarily as posttranscriptional regulators. Over the past six years, about 70 sRNAs have been discovered in , and about 20 of them have been assigned a function. Many of these trans-encoded RNAs are involved in metabolic processes [e.g. Spot42, DsrA, RprA, RyhB, SgrS, GadY, reviewed in ()] and at least eight sRNAs regulate the expression of membrane proteins [reviewed in ()]. To date, relatively few systematic searches have been performed in Gram-positive bacteria. Among the recently discovered sRNAs in Gram-positive hosts are RatA from the chromosome (), which came up in a systematic search () together with 12 other sRNAs that proved to be sporulation-controlled, but still await the identification of their targets (). Furthermore, in addition to the well-studied RNAIII from (), 12 novel sRNAs from pathogenicity islands have been detected () as well as three Hfq-binding sRNAs of with still unknown function (), and nine novel sRNAs from within intergenic regions found by -based approaches (). Additionally, more than 100 potential 6S RNA species have been identified by bioinformatics approaches, and many of them were verified experimentally, among them two 6S RNA species in (,). Still, the identification of mRNA targets of the recently discovered sRNAs is a challenging issue, and has been successful only in less than one-third of all cases. One important hallmark of many trans-encoded regulatory RNAs from is their ability to bind the Sm-like abundant RNA chaperone Hfq (). While several sRNAs have been found to require Hfq for their stability, some were shown to need Hfq for efficient complex formation with their target RNA (,). For DsrA//Hfq, the pathway of complex formation has been investigated by biophysical techniques (). However, for sRNAs from Gram-positive bacteria, the putative function of Hfq is still elusive. At least in one case, staphylococcal RNAIII/ interaction, no influence of Hfq has been found (). In contrast to the -encoded sRNAs from accessory genetic elements like plasmids, phages, transposons that have been studied in detail over the past 25 years [reviewed in ()], relatively little is known about structural requirements, binding kinetics and mechanisms or degradation pathways of these new trans-encoded regulatory sRNAs. Although complexes between sRNA and mRNA have been detected in some instances, only in five cases secondary structures of such complexes predicted by Mfold have been confirmed by experimental secondary structure probing. These include MicF/ (), Spot42/ (), RyhB/ (), MicA/ (,) from and RNAIII/ from (). So far, the region of initial contact between a trans-encoded sRNA and a target RNA sharing more than one complementary region has not been narrowed down. The mechanism of action has been proposed in some cases, but not always corroborated by a combination of and experiments. The 205-nt untranslated RNA SR1 from the genome was found in our group by a combination of computer predictions and northern blotting (). Recently, we have shown that SR1 is a antisense RNA that acts by basepairing with its primary target, mRNA, the transcriptional activator of the and arginine catabolic operons (). translation data and translational reporter gene fusions suggested that SR1 might inhibit translation at a post-initiation stage. Hfq was shown to be dispensable for the stability of SR1. Here, we provide a detailed characterization of SR1 and the SR1/ complex with and without Hfq. We determined the region of initial contact between SR1 and . Furthermore, a combination of toeprinting and SR1/ complex probing studies demonstrated that SR1 inhibits translation initiation of mRNA by inducing structural changes between the SD sequence and the first complementary region G. In contrast to many sense/antisense systems, Hfq was shown to be exclusively required for translation of RNA, but not for promoting the SR1/ interaction. The intracellular concentration of SR1 in was calculated to be 30 nM in log phase and 315 nM in stationary phase in complex TY medium. Chemicals used were of the highest purity available. DNA polymerase was purchased from Roche or SphaeroQ, Netherlands, respectively, RNA ligase from New England Biolabs and Thermoscript reverse transcriptase and M-MuLV reverse transcriptase from Invitrogene and Fermentas, respectively. Firepol polymerase was purchased from Solis Biodyne, Estonia. strains DH10B and ER2566(Δ) were used for cloning and for expression of the gene, respectively. strains DB104 () and strains were grown in complex TY medium (). transcription and partial digestions of synthesized, 5′-end-labelled SR1 and RNA species with ribonucleases T1, T2 and V were carried out as described (). For the analysis of SR1/ complexes with T1, T2 and V, either SR1 or were 5′ end-labelled and a 6- to 60-fold excess of the cold complementary RNA was added prior to RNase digestion. Both RNA and SR1 were synthesized from PCR-generated template fragments with primer pairs indicated in Supplementary Data. SR1/ complex formation studies were performed as described previously (). Complex formation in the presence of Hfq was assayed in TMN buffer () using purified Hfq from . For the purification of Hfq, the IMPACT™-CN system from New England Biolabs was used. To prevent the purification of Hfq-heterohexamers, strain ER2566() was transformed with plasmid pTYB11-BsHfq. (All plasmids used in this study are summarized in Table 1). The resulting strain was grown at 37°C till OD = 0.7, induced with 0.25 mM IPTG, and grown at 18°C for further 18 h. The fusion protein was purified by affinity chromatography on a chitin column as described by the manufacturer. On-column cleavage was performed with 20 mM Tris-HCl pH 8.0, 500 mM NaCl and 50 mM DTT for 20 h at room temperature. Millipore microcon columns were used to concentrate the eluted Hfq protein and to exchange the buffer for 50 mM Tris-HCl pH 8.0. The purified protein was stored at 4°C. For the construction of the three translational fusions, chromosomal DNA from DB104 was used as template in three PCR reactions with upstream primer SB979 and the corresponding downstream primers SB980 (pGGA4), SB987 (pGGA6) and SB1065 (pGGA8). All fragments were digested with BamHI and EcoRI and inserted into the BamHI/EcoRI vector pGF-BgaB () encoding the promoterless heat-stable β-galactosidase from . For the construction of plasmid pGGA7 carrying an internal deletion of 11 bp (nt 102 to 112) of , a two-step PCR with outer primers SB979 and SB976 and internal primers SB989 and SB988 was performed on chromosomal DNA as template, the third PCR product obtained with SB979 and SB976 cleaved with BamHI and EcoRI and inserted into the BamHI/EcoRI vector pGF-BgaB. The toeprinting assays were carried out using 30S ribosomal subunits, mRNA and tRNA basically according to (). The 30S ribosomal subunits devoid of initiation factors were prepared from strain MRE600 essentially as described by Spedding (). The 5′-[P]labelled -specific oligonucleotide SB1068 (5′ TAC CGT GGC CTG CGT TAC) complementary to mRNA was used as a primer for cDNA synthesis in the toeprinting reactions. An aliquot of 0.04 pmol of mRNA annealed to primer SB1068 was incubated at 37°C without or with 0.4 pmol of 30S subunits and 8 pmol of uncharged tRNA (Sigma) before supplementing with 1 µl M-MuLV-RT (80 units). cDNA synthesis was performed at 37°C. Reactions were stopped after 10 min by adding formamide loading dye. The samples were separated on a denaturing 8% polyacrylamide gel. For the analysis of the effect of sRNAs on 30S complex formation, mRNA and the corresponding sRNA were incubated for 15 min at 37°C before the addition of 30S ribosomes and initiator tRNA. Toeprint efficiency was determined by PhosphorImaging using the Image-quant software package (PC-BAS 2.0). Preparation of total RNA and northern blotting were carried out as described previously (). So far, only for a few chromosomally encoded regulatory sRNAs, secondary structures have been determined experimentally. Examples include MicF (), OxyS (), RNAIII of (), DsrA (), Spot42 (), RyhB () and MicA (,). Since computer-predicted RNA structures often deviate from experimentally determined ones [e.g. RNAIII of pIP501 () or RNAI/RNAII of pT181 ()], we performed limited digestions with structure-specific ribonucleases to determine the secondary structure of SR1. The wild-type SR1 (205 nt) as well as the 3′ truncated species SR1, the 5′ truncated species SR1 and the 5′ and 3′ truncated species SR1 were 5′-end labelled, gel-purified and treated with RNases T1 (cleaves 3′ of unpaired G residues), T2 (unpaired nucleotides with a slight preference for A residues) and V1 (double-stranded or stacked regions). A shows an analysis of SR1 and the truncated species SR1 whereas B contains the schematic representation of the structure of SR1 derived from the cleavage data. The experimentally determined structure for wild-type SR1 comprises three main stem-loops: SL1 (nt 1 to 112), SL2 (nt 138 to 154) and the terminator stem-loop SL3 (nt 173 to 203) interrupted by two single-stranded regions SSR1 (nt 113 to 137) and SSR2 (nt 155 to 172). It deviates from the structure predicted with Mfold in the 5′ as well as in the 3′ portion: The 5′ part was found to be single-stranded between nt 38 and 51, and the double-stranded stem proved to be much longer than predicted and comprises 20 paired nucleotides (nt) interrupted by three internal loops or bulged-out bases, respectively, compared to only 10 paired nt in the predicted structure. For the 3′ part, two stem-loops and the terminator stem-loop were predicted by Mfold, whereas the structure probing data support in addition to the terminator stem-loop only the second stem-loop SL2 in the centre of a long single-stranded region. Structure probing of the 5′ 132 nt of SR1 (A, right part) showed that this portion of the molecule folded independently and exactly as in the full-length sRNA. The secondary structure for the 3′ 98 nt of SR1 contained exactly the terminator stem-loop as in wild-type SR1 (not shown) and the secondary structure for SR1 comprising nt 109 to 186 revealed the single stem-loop SL2 surrounded by single-stranded regions as expected (not shown). The information on the secondary structures of the truncated derivatives was necessary to assess the data on complex formation between different SR1 species and its target, mRNA. Previously, we have shown that SR1 binds to the 376 3′ nt of mRNA (, C) with an equilibrium dissociation rate constant of 3.21 × 10 M (). Since seven regions of complementarity have been predicted between SR1 and mRNA [() and C], we intended to narrow down the segment of SR1 that is required for the initial contact with its target. To this end, SR1 species of different lengths were generated by transcription with T7 RNA polymerase, 5′ end-labelled, gel-purified and used for binding assays with the RNA. The results are shown in A: 3′ truncated SR1 derivatives SR1 and SR1 comprising only stem-loop SL1 and lacking SL2 and the terminator stem-loop, were not able to form complexes with mRNA even at 400 nM. In contrast, 5′ truncated species SR1 comprising only the single-stranded region, SL2 and the 5′ half of the terminator stem-loop, was as efficient in complex formation as SR1, a species that only lacked the 3′ half of the terminator stem-loop, but otherwise contained the complete wild-type sequences and structures. In accordance with these data, both SR1 lacking SL3 completely and SR1, lacking SL1 and SL3, were significantly impaired in the interaction with their target and only at 400 nM mRNA, a weak complex was observed. From these results we can conclude that for efficient complex formation between SR1 and mRNA, SL1 and the 3′ half of SL3 are not required. Furthermore, the opening of the terminator stem-loop SL3 seems to be essential for an efficient interaction and a sequence located in the 5′ half of SL3 proved to be important for the contact between antisense-RNA and target. To analyse the regions of required for efficient pairing with SR1, five 5′ labelled RNA species (shown schematically in C) were used in complex formation experiments with SR1 (B). As expected, labelled comprising nt 108 to 483 of RNA, but lacking the 5′ part and the SD sequence of formed a complex with unlabelled SR1 with the same as determined previously for the labelled SR1/unlabelled pair. The same efficiency for complex formation was observed for containing region G′ but lacking the SD sequence. By contrast, labelled and comprising the 5′ 136 and 196 nt of mRNA, respectively, including SD sequence and region G′, were significantly impaired in complex formation with unlabelled SR1. The complete mRNA including 5′ end, SD and all complementary regions to SR1 formed a weak complex with SR1 only at 400 nM concentration. These results suggest that the SD sequence of mRNA might be sequestered by intramolecular basepairing and that a factor might be needed to facilitate ribosome binding. The results from the binding assays indicate that SR1 is sufficient for efficient complex formation with mRNA and that without opening of the 5′ half of the terminator stem-loop no efficient complex can form. To investigate the alterations in the secondary structures of SR1 and upon pairing, the secondary structure of the SR1/ complex was determined. To ascertain alterations in the SR1 structure, labelled SR1 was incubated with a 6- to 60-fold excess of unlabelled RNA, the complex was allowed to form for 5 min at 37°C, and, subsequently, partially digested with RNases T1, T2 and V1. In parallel, free SR1 was treated in the same way. A shows the result. As expected, no significant alterations were observed within the 5′ 112 nt of SR1 that contain only region A (nt 15 to 19) complementary to . By contrast, significant alterations in the T1, T2 and V cleavage pattern were observed within the other six complementary regions B, C, D, E, F and G (A, right half). The data are summarized in C: Whereas in region B, only one reduced T1 cut was detected at G, drastic alterations were observed in both regions C and G: In C, all 9 nt complementary to showed reduced T2 cleavages, G and G exhibited reduced T1 cleavage and at U and U, an induction of V1 cleavage was detected indicating that this region became double-stranded upon pairing with . The same was true for region G, where the cleavage pattern at all positions was altered compared to free SR1: nt 175 to 181 showed a decreased T2 cleavage, among them G and G a reduced T1 cleavage, whereas at U and G new V cuts appeared. Fewer changes were found in regions D, E and F, where G (region D), U and A (region E) and G, U and U (region F) were not single-stranded anymore and, instead, U and U (region D), A and A (region E) as well as U and G (region F) showed induced V cleavages, i.e. became double-stranded. To further substantiate these results, secondary structure probing was performed with a complex formed between labelled and a 6- to 60-fold excess of unlabelled SR1. To corroborate our previous hypothesis that SR1 does not inhibit the translation initiation at the SD sequence, both the complex between (5′ 136 nt of including SD sequence and region G′) and the complex between (lacking the 5′ 112 nt of including SD, but comprising all regions complementary to SR1) were probed with RNases T1, T2 and V. The results are shown in B and are summarized in D: In the case of , induced V cuts were visible in regions E and G. Furthermore, between region E and D and in region C, T2 cuts were induced which is expected when one strand of a double-stranded region interacts with SR1, and the other half becomes, consequently, single-stranded. The same holds true for the induced T1 cuts in region B and the induced T2 cut in the region upstream of B. The lower part of B presenting the results of SR1/ interaction clearly shows that the SD sequence itself was not affected upon addition of increasing amounts of unlabelled SR1. Surprisingly, a number of alterations could be observed further downstream from it and upstream of complementary region G′. In particular, prominent V cuts were induced at nt 40, nt 46 to 48, nt 52, nt 56, nt 71 and nt 90, accompanied by induced T2 cuts around nt 56 and 74, 75, 77 and 78 (B left and D). These data suggest that binding of SR1 causes structural changes in the 5′ part of mRNA between the SD sequence and region G′. As published previously (), one out of seven regions of complementarity between SR1 and RNA comprises nt 176 to 181 within the 5′ half of the SR1 terminator stem-loop SL3 (designated G) and nt 113 to 118 of mRNA (designated G′). If these two regions were involved in a first contact between SR1 and RNA, nucleotide exchanges in either SR1 or RNA should impair or abolish complex formation, and compensatory mutations should, at least partially, restore binding. To test this hypothesis, three mutated SR1 species with either a 10 nt exchange (5′AGCAUGCGGC to 5′ UCGUACGCCG) between nt 176 and 185 denoted SR1, a 6 nt exchange (5′AGCAUG to 5′UCGUAC), denoted SR1 or a 2 nt exchange (GC to TT) denoted SR1, were assayed in complex formation with wild-type comprising nt 109 to 196 of mRNA (region G′). The 6 and 10 nt exchanges were designed such that the GC/AU content of the region was not altered compared with the wild-type. As shown in A, no interaction between these three mutated SR1 species and wild-type RNA was observed. By contrast, the exchange of only C to G (SR1) did not impede complex formation, suggesting that either G is most important for the initial contact or that substitution of one nucleotide is not sufficient to cause an effect. Interestingly, when RNA, a derivative of the same length carrying the compensatory mutations to SR1 was used, binding could be restored (A and B) confirming a specific basepairing interaction between SR1 and mRNA. When a longer RNA comprising all seven complementary regions G′ to A′ was analysed, binding was abolished by the above-mentioned mutations too, and partially restored with the compensatory mutation mRNA (not shown). These data indicate that the complementary region G of SR1 (nt 176 to 181) plays an important role for the recognition of mRNA. To investigate the contribution of the other regions of SR1 complementary to RNA to efficient binding with its target, two SR1 species carrying 9 nt exchanges each in either region C (nt 119 to 127)—SR1—or region E and the first 2 nt of region F (comprising nt 146 to 154)—SR1—were analysed for complex formation with RNA carrying the wild-type or mutated regions (C). Complex formation was significantly impaired in both cases: SR1 exhibited about 10-fold and SR1 about 30-fold decreased efficiency to pair with RNA. A combined substitution of regions C, E and 5′ F (SR1) or a combined exchange of regions C, D, E and 5′ F (SR1S) resulted in a complete loss of pairing. D shows a schematic representation of the four mutated SR1 species. To test the importance of region G′ (nt 113 to 118 of -mRNA complementary to nt 176–181 of SR1) for the interaction with SR1 in , the following three translational fusions were constructed: pGGA6 containing nt 1 to 113 but lacking all but one nt of region G, pGGA4 comprising nt 1 to 119, i.e. the entire region G + one additional nt, and, hence, no other complementary region, and pGGA7 identical to pGGA3 (comprising G, F and E, 24) but lacking nt 102 to 112 upstream of G. All fusions were integrated into the locus of the DB104 chromosome, grown till OD ∼ 5 (maximal expression of SR1) and β-galactosidase activities measured. As shown in , β-galactosidase activities measured with pGGA4 and pGGA7 were, in both cases, about 30-fold lower than that of the pGGA6-integration strain lacking any complementary region to SR1. Since pGGA4 yielded the same decrease in β-galactosidase activity compared to a construct lacking any complementarity with SR1 as our previous construct pGGA3 that encompassed regions G, E and F, it can be concluded that region G alone is sufficient to inhibit translation almost completely. The results obtained with pGGA7 and pGGA4 exclude the possibility that the sequences immediately adjacent to region G are involved in the observed decrease of β-galactosidase activity, e.g. by providing a cleavage site for an RNase. To test whether point mutations in region G′ abolish the effect of SR1 on translation, pGGA8 was constructed carrying the same 2 nt exchange as SR1 analysed in the binding assay (A), but lacking any sequences downstream from nt 118 (3′ end of region G′) and integrated into the locus of . The β-galactosidase activity measured with pGGA8 was nearly the same as with pGGA6 (), confirming the result that the 2-nt exchange in region G prevented the interaction between SR1 and . Many small RNAs from need Hfq for either stability or their interaction with their targets (see Introduction section). Previously, we have shown that Hfq is neither required for the stabilization of SR1 nor that of (). However, in the absence of Hfq, but presence of SR1, the expression of the downstream SR1 targets, mRNA and mRNA, was about 3- and 6-fold, respectively, increased. Therefore, we wanted to investigate, whether Hfq is required for the promotion of complex formation with RNA. To investigate whether Hfq binds SR1, different concentrations of purified Hfq were added to labelled wild-type SR1 and two 3′ truncated species SR1 and SR1, and a gel-shift assay was performed. As shown in A, all three SR1 species bound Hfq at concentrations of 3–10 μM. To analyse binding of Hfq to RNA, full-length and truncated species were assayed for Hfq binding: As shown in B, , and (full length) that contain the SD sequence, bound Hfq very efficiently. By contrast, lacking the SD sequence bound Hfq less efficiently than . Since both SR1 and RNA bound Hfq, we analysed whether Hfq is able to promote the complex formation between both RNAs . For this purpose, purified Hfq was added to a final concentration of 10 μM (amount required to bind 50% SR1), to the mixture of 1.0 nM labelled SR1 and different amounts of unlabelled mRNA, incubated for 15 min at 37°C and complexes were separated on 6% native PAA gels. Although a ternary SR1//Hfq complex formed, this complex was not observed at lower concentrations compared to the binary SR1/ complex, and the amount of this complex did not increase with increasing concentrations of unlabelled RNA (C). In contrast, upon higher concentrations of unlabelled RNA (≥100 nM), this RNA, apparently, successfully competed with SR1 for Hfq binding, so that the amount of unbound labelled SR1 increased again (C). In summary, all these data clearly prove that the RNA chaperone Hfq does not facilitate the interaction between SR1 and its target mRNA. To reconcile these observations as well as the lacking effect of Hfq on SR1 stability with the increase of the and mRNA levels in the knockout strain, we tested whether the translation of is affected by Hfq. For this purpose, the translational fusion pGGA6 was integrated into the locus of DB104(Δ), and β-galactosidase activity was measured and compared to that determined in the presence of Hfq in DB104. A 250-fold lower β-galactosidase activity was detected in the absence of Hfq, indicating that this RNA chaperone is required for efficient translation of mRNA (). To substantiate the role of Hfq in promoting translation of mRNA, the secondary structures of mRNA and SR1 were probed with RNases T1 and T2 in the presence and absence of Hfq. As shown in D, one binding site of Hfq on mRNA (5′ AAAUA) is located immediately upstream of the SD sequence. The same assay was used to determine the binding site(s) of Hfq on SR1. Here, one binding site around nt 9–13 in the 5′ part of SR1 and a second in the bulge of stem-loop SL1 (nt 43 to 47) were found (gel not shown). The facts that Hfq gel-shifts with wild-type SR1 and SR1 comprising only the 5′ stem-loop were identical (A), support the absence of Hfq-binding sites on SR1 downstream from nt 104. Although the first complementary region between and SR1 is located 87 nt downstream from the SD sequence, we performed a toeprinting analysis () to examine the effect of SR1 on formation of the translation initiation complex at mRNA. A shows that in the presence of initiator tRNA, 30S ribosomal subunits bind to the translation initiation region and block reverse transcription of a labelled primer, annealed downstream, at the characteristic position +15 (start codon A is +1). This signal provides a measure for the formation of the ternary complex, since it is dependent on both 30S subunits and initiator tRNA. Addition of increasing amounts of SR1 or SR1 prior to addition of 30S subunits and tRNA interfered with ternary complex formation, resulting in a weaker toeprint signal (A and C). Thereby, the inhibitory activity of SR1 was higher than that of SR1 which correlates with its more efficient binding activity to mRNA (). By contrast, both the addition of a noncognate small RNA, SR2 from or RyhB from , failed to decrease the toeprint signal on mRNA (summarized in D) indicating that SR1-dependent inhibition of ribosome binding was specific. To support the specificity of the SR1 inhibitory action on ternary complex formation on mRNA, a control toeprint was performed with SR1 and mRNA (target of RyhB). Since SR1 did not affect ternary complex formation on mRNA (B), whereas RyhB did as expected, it can be excluded that the effect of SR1 on mRNA is simply due to binding to the ribosome. To corroborate the importance of complementary region G for the interaction between SR1 and mRNA, an additional toeprinting assay was carried out with the G-region mutant SR1 compared to SR1 (C). The autoradiogram and the quantification (D) show that this mutant is clearly impaired in blocking the binding of the 30S initiation complex, although it has still some residual activity. This result confirms both the specificity of the SR1/ahrC interaction and substantiates our conclusion from the binding assays ( and ) that G is required for the initial contact between SR1 and mRNA. In summary, these data demonstrate that binding of SR1 to mRNA prevents the formation of translation initiation complexes. To determine the intracellular concentration of SR1 in in logarithmic and stationary growth phase, strain DB104 was grown in complex medium, and samples were withdrawn at OD 2 (log phase) and OD 4.5 (onset of stationary phase). Cell numbers were determined upon plating of appropriate dilutions of the harvested cultures on agar plates. Total RNA was prepared, separated on a denaturing polyacrylamide gel alongside defined amounts of synthesized SR1 and subsequently, subjected to northern blotting (). Losses during RNA preparation were calculated using synthesized SR1 mixed with the same amount of DB104::Δ cells at the beginning of the RNA preparation. A comparison with the same amounts of untreated RNA yielded ∼80% loss. Loading errors were corrected by reprobing with labelled oligonucleotide C767 complementary to 5S rRNA. Using this quantification procedure, the amount of SR1 within one cell was calculated to be ∼20 molecules in log phase and 200–250 molecules in stationary phase, corresponding to an approximate intracellular concentration of 30 and 315 nM, respectively. For all recently discovered trans-encoded sRNAs the targets of which have been identified, only one or two complementary regions were found. In the majority of cases, these regions covered the 5′ part of the target RNA, mostly including the SD sequence, and the mechanism of action was found to be inhibition or activation of translation initiation. Rather unusually, SR1 and mRNA contain seven regions of complementarity that comprise the 3′ half of SR1 and the central and 3′ portion of mRNA (). This prompted us to determine the secondary structures of SR1 and the /SR1 complex and to investigate the structural requirements for efficient /SR1 pairing. B shows that SR1 is composed of one large 5′ stem-loop (SL1) structure with a prominent bulge, a central small stem-loop SL2 and the terminator stem-loop SL3 separated by two single-stranded regions. Six out of seven regions of complementarity to RNA (B to G) are located in the 3′ 100 nt of SR1. Secondary structure probing of labelled SR1 in complex with increasing concentrations of unlabelled and (A and B) revealed structural alterations in six of the seven complementary regions. In SR1, all positions in region C and G as well as a few positions in B, D, E and F were affected (summarized in C). In , alterations in regions C, E, F and G as well as additional alterations between regions D and E were found. Interestingly, structural changes over a stretch of ∼50 nt were also observed upstream of region G (B left), although the SD sequence (nt 21 to 25) and the start codon remained unaffected indicating that binding of SR1 causes structural changes in the 5′ part of -mRNA, too. Whereas for -encoded antisense RNAs from plasmids, phages and transposons, a number of studies have been performed to elucidate binding pathways and to determine structural requirements for the two contacting RNA molecules (), little is known, so far, about the formation of initial contacts between trans-encoded sRNAs and their targets. Here, we show that a solely 78-nt long SR1 species spanning nt 109 to 186 is sufficient for efficient complex formation with mRNA, i.e. the 5′ portion of SR1 is not needed (A). Generally, all SR1 species lacking the 5′ half of SL3 with region G or comprising a complete SL3 were significantly impaired in pairing with RNA. This might indicate that some factor — most likely a protein or an RNase cutting within the loop of SL3 — opens the terminator stem-loop to promote complex formation. Since only full-length SR1 can be observed (northern blots and 3′ RACE, 23), the involvement of an endoribonuclease is highly unlikely. The possibility that the RNA chaperone Hfq that binds upstream of the terminator stem-loop of SR1 is responsible for opening up this structure, can be eliminated, too (see below). Most probably, another, yet unknown RNA-binding protein is needed to open SL3. Two lines of evidence show that the initial contact between SR1 and RNA occurs at complementary region G of SR1: complex formation assays of truncated SR1/ pairs containing mutations and compensatory mutations in region G () and translational - reporter gene fusions with the same point mutations (). Furthermore, complex formation assays with SR1 mutants affected in regions C, D, E/F or a combination thereof and a fusion with regions E′, F′ and G′ revealed a contribution of the other complementary regions to SR1/ pairing. In summary, since, (i) in the absence of region G, no efficient complex could form, (ii) in the presence of wild-type regions A to E, a 2-nt exchange within G inhibits pairing and (iii) in the presence of G, significant simultaneous alterations in regions C, E and F did affect complex formation, we can conclude, that region G is responsible for the initial contact between SR1 and RNA, but the other complementary regions add to efficient antisense/target RNA pairing. Region G′ in unpaired mRNA is double-stranded with a bulged-out G at position 116 (B left). Interestingly, only when this G and the neighboured C were replaced by a C and G (, see ), the interaction with SR was restored indicating that it is crucial for the initial contact. As proposed above, some factor is needed to melt or open up region G in SR1, so that the two regions can interact. Our data suggest that pairing initiates at G, but for subsequent steps and stable complex formation, a contribution of the other complementary regions B to F is needed. This is reminiscent of the binding pathway of the antisense/sense RNA pair CopA/CopT involved in regulation of plasmid R1 replication [reviewed in ()]. Here, binding starts with the interaction of two single-stranded kissing loops and, afterwards, a second region is needed to overcome the torsional stress and to propagate the helix. By contrast, for the antisense/sense RNA pair RNAIII/RNAII of plasmid pIP501, the simultaneous interaction of two complementary loop pairs was found to be required (). In other cases, a single-stranded region and a loop form the first complex [e.g. Sok/hok of plasmid R1 or RNA-OUT/RNA-IN of transposon IS10, reviewed in ()]. For many trans-encoded sRNAs in , the RNA chaperone Hfq has been shown to be required for either stabilization of the sRNA or/and efficient duplex formation with the target RNA (see the Introduction section). Previous experiments have demonstrated that Hfq does not stabilize SR1 (). This report shows that although Hfq binds both SR1 and RNA, it is not able to promote complex formation between SR1 and (). This is in agreement with data obtained for the RNAIII/ interaction in , for which Hfq was found to be dispensable for RNAIII/ complex formation (,). The fact that no requirement for Hfq was observed in the RatA/ system of (), too, suggests that in Gram-positive bacteria Hfq might not be needed for sRNA/target RNA interaction or, alternatively, that another RNA chaperone may fulfil the function of Hfq. One candidate might be HBsu, for which RNA-binding activity was demonstrated (). However, our previous observation that the levels of the secondary targets of SR1, and mRNA, were increased 3- to 6-fold in an knockout strain () raised the question on the role of this chaperone in the SR1/ system. Suprisingly, mRNA proved to be not translated in a knockout strain (). This indicates that Hfq is required for efficient translation of , possibly by opening up some secondary structures that otherwise inhibit binding of the 30S initiation complex. This is supported by the finding of one Hfq-binding site (5′ AAAUA) immediately upstream of the ribosome-binding site (RBS). Interestingly, for mRNA it has been also shown that Hfq is essential for efficient translation (). In contrast to , the binding of Hfq to SR1 does not seem to play a role in this context. The fact that Hfq binds upstream of six out of seven SR1 regions complementary to mRNA supports the failure of Hfq to promote complex SR1/ formation. However, we cannot exclude that Hfq binding might be important for the interaction of SR1 with other, still unidentified target mRNAs. Based on a series of translational - fusions, the dispensability of the SD sequence for pairing with SR1 and translation data with chimeric RNAs, we suggested previously that SR1 might affect translation at a post-initiation stage (). However, the structural alterations found in the mRNA downstream from the SD sequence in the presence of increasing amounts of SR1 prompted us to re-evaluate our previous data using a toeprinting analysis (). Both SR1 and SR1, but not two heterologous RNAs, were able to inhibit binding of the 30S ribosomal subunit and formation of a ternary complex with 30S and tRNA on full-length mRNA. These results — together with the structure probing data — demonstrate that binding of SR1 induces structural changes in a ∼65-nt long stretch of RNA between SD sequence and complementary region G that eventually inhibit formation of the 30S initiation complex. Since the 30S ribosomal subunit covers 54 nt, i.e. 35 (±2) nt upstream and 19 nt downstream from the start codon (), the 5′ part of the SR1-induced structural alterations of mRNA coincides exactly with this region. The analysis of the G region mutant SR1 in the toeprinting assay (C) corroborated that this region is involved in the first contact between SR1 and mRNA and supported the specific basepairing interaction between both RNA molecules. The toeprinting results are not opposed to the previously observed translation inhibition of - fusions (), as this inhibition can be explained by SR1-induced structural changes in the 5′ part of RNA, too. Therefore, we can conclude that the mechanism of action employed by SR1 is inhibition of translation initiation. This is the first case of a small regulatory RNA that binds ∼90 nt downstream from the ribosome-binding site and interferes with translation initiation. In contrast, in the well-studied systems like RyhB/ () or MicA (,), the complementary regions between small RNA and mRNA are located upstream of or overlap the target SD sequence, making an effect on ribosome binding and hence, translation initiation, more plausible. Our results raise the question on the maximal distance between SD sequence and a binding region for a small RNA permitting to affect 30S subunit binding. Furthermore, in many cases the inhibition of translation initiation was accompanied by significantly decreased amounts of the target mRNA(s) [e.g. RyhB/ () or SgrS ()] that was attributed to degradation of the unprotected target RNA by RNase E or of the complex by RNase III (). Surprisingly, levels were found to be independent of the presence or absence of SR1 (). To date, no RNase E has been found in . Although two novel endoribonucleases with homology to RNase E, RNase J1 and J2, were recently discovered (), it is unclear, whether they fulfil the role of the main endoribonucleases as it does RNase E in Gram-negative bacteria. In the few sense/antisense RNA systems, where calculations of the amount of both interacting species were performed (,), an at least 10-fold excess of the inhibitory small RNA over its target was determined. Here, the amount of SR1 in grown in complex medium was found to increase upon entry into stationary phase from 15–20 to 250 molecules per cell. This is much lower than the 4500 molecules measured for OxyS under oxidative stress conditions (), but still in the range of RNAIII of plasmid pIP501 (∼1000 molecules). Since we could not detect mRNA in northern blots under any growth condition, its amount must be significantly lower than 15 molecules/cell ensuring at least a 15-fold excess of SR1. The analysis of the SR1/ mRNA interaction yielded three major issues, which might be important for sRNA/target RNA systems in general: First, whereas the major mechanism of action of trans-encoded sRNAs reported in Gram-negative bacteria is inhibition of translation initiation by direct binding to the RBS or 5′ of it, the SR1/ pair is first case, where translation initiation is prevented by binding of the sRNA ∼90 nt downstream from the RBS. Second, while all sRNA/target RNA pairs studied so far comprise at the most two complementary regions, the SR1/ pair is the first case with seven complementary regions between inhibitor and target RNA, and the major contribution of one region as well as the minor, but measurable contribution of five of the other regions has been demonstrated. Third, whereas in , Hfq was required for either sRNA stabilization or promotion of complex formation with the target RNA, at least complex formation in Gram-positive bacteria does not seem to depend on Hfq. The search for and analysis of other SR1 targets will reveal whether this sRNA exerts its function(s) by the same or alternative mechanisms. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The positive transcription elongation factor, P-TEFb, is an essential cellular transcription factor that controls the transition of RNA polymerase II from abortive to productive elongation (). P-TEFb is composed of one of two isoforms of the cyclin-dependent kinase Cdk9 (,) and a partner cyclin, T1, T2 or K (,). P-TEFb phosphorylates serine 2 residues in the heptapeptide repeat present in the carboxyl terminal domain (CTD) of the largest subunit of RNA polymerase II () and DRB Sensitivity Inducing Factor (DSIF) (,), alleviating the effect of the negative elongation factors DSIF and Negative ELongation Factor (NELF) (). Knockdown of P-TEFb subunits by RNAi in reduces serine 2 phosphorylation in the CTD and the expression of early embryonic genes, and results in an embryonic lethal phenotype (). Inhibition of P-TEFb activity in cells through treatment with the P-TEFb inhibitor flavopiridol, blocks most RNA polymerase II transcription, and extended treatment leads to cell death (). P-TEFb was recently found to be regulated by reversible inhibition via association with the small nuclear RNA 7SK (,) and the RNA-binding proteins HEXIM1 () or HEXIM2 (,). Association of 7SK with a HEXIM1 dimer relieves an autoinhibitory interaction between two regions of HEXIM1, allowing recruitment and inhibition of two P-TEFb molecules (). The large form of P-TEFb contains potentially active P-TEFb molecules, because the activating T-loop phosphorylation of Cdk9 is required for its incorporation into the large form (,). While the exact signals and mechanisms for mobilization of P-TEFb from the large form remain undefined, the association of P-TEFb with 7SK and the HEXIM proteins is highly responsive to intra- and extracellular signaling. For example, the large form of P-TEFb is disrupted by treatment of cells with P-TEFb inhibitors, UV light or actinomycin D (), due to a poorly understood feedback loop between inhibition of transcription elongation and P-TEFb activity. Two physiological models for regulation of P-TEFb activity by 7SK and HEXIM1 have been identified thus far. Cardiac hypertrophy has been shown to be a consequence of activation of P-TEFb in cardiomyocytes through release from the large complex, which results in an increase in RNA polymerase II phosphorylation, gene expression and cell size (). Conversely, HEXIM1 has been identified as a suppressor of breast cancer (,). The human immunodeficiency virus (HIV) replicates its genome by utilizing the host cell RNA polymerase II transcription machinery and is controlled mainly at the level of elongation. Basal transcription from the HIV long terminal repeat (LTR) promoter is extremely inefficient, with most RNA polymerase II initiation events terminating in abortive elongation (). Transcription from the LTR is activated by the viral protein Tat, which facilitates the transition from abortive to productive elongation () by recruiting P-TEFb (,,). Tat associates with the transactivation response element, TAR, which is an RNA stem-loop present in the nascent transcript produced from the HIV LTR, and P-TEFb containing cyclin T1 is recruited through association with both Tat and TAR (). Interestingly, all P-TEFb inhibitors found to date block HIV replication at lower concentrations than those that have a negative effect on cellular transcription or viability (,,). Additionally, over-expression of a kinase dead Cdk9 mutant reduced the expression of endogenous Cdk9 and inhibited HIV replication without affecting cell survival (,), and a partial knockdown of Cdk9 and cyclin T1 did not affect cell viability, yet inhibited Tat transactivation in Magi cells (). While the mechanism of this heightened requirement for P-TEFb activity in Tat transactivation compared to normal cellular transcription is not understood, it is possible that much higher P-TEFb activity levels are required for HIV to overcome the tight repression of the HIV LTR promoter. In support of this, it has been shown that for HIV to replicate in primary blood lymphocytes (PBLs), the cells must be activated, which causes an increase in cyclin T1 expression and overall P-TEFb activity (). Notably, much of the work regarding the role of P-TEFb in Tat transactivation was carried out prior to work on P-TEFb control and the discovery of the large form of P-TEFb, and its role in HIV replication has only begun to be explored. Several lines of evidence led us and others () to hypothesize that the large form of P-TEFb is involved in HIV replication. It has been reported that activation of PBLs, which leads to increased P-TEFb as discussed above, causes an increase in 7SK levels along with the increase in P-TEFb expression, suggesting that activation of PBLs actually results in an increase in the large form of P-TEFb (). The partial knockdown of Cdk9 and cyclin T1 that blocked Tat transactivation in Magi cells also eliminated high salt-activated (large form) P-TEFb (). Furthermore, the majority of P-TEFb within cells is actually sequestered within the large complex. Taken together with the P-TEFb inhibition studies demonstrating that HIV requires much higher levels of P-TEFb activity than cellular promoters, this suggests that the P-TEFb within the large complex is necessary for HIV replication, and furthermore, that HIV has a mechanism to subvert normal cellular control and activate P-TEFb. This activation would provide a larger pool of P-TEFb for use at the HIV LTR (,,). In support of this hypothesis, low levels of actinomycin D, which inhibit transcription and lead to release of P-TEFb from the large form without inhibition of P-TEFb itself (), have been shown to activate the HIV LTR in the absence of Tat (), suggesting that the HIV LTR may be able to use the P-TEFb obtained from the large complex. Our current studies were designed to uncover details of how HIV interfaces with the cellular factors that control P-TEFb. All cells were grown under standard conditions of 37°C, 5% CO. HeLa37 cells, which express CD4 and both CXCR4 and CCR5 (), were grown in DMEM with 10% FBS and 1× penicillin/streptomycin (pen/strep). HeLa S3 cells were grown in DMEM-F12 with 10% FBS. 293T cells were obtained as a kind gift from P. McCray (Univ., Iowa). Cells were maintained in DMEM with 10% fetal calf serum and pen/strep. Cells were trypsinized and split 1:10 every 4 days. Anti-Cdk9 (C-20) was obtained from Santa Cruz (sc-484). Anti-FLAG M2-peroxidase (HRP) was obtained from Sigma (A8592). Anti-cyclin T1 (T-18) used in western blots was obtained from Santa Cruz (sc-8127). Anti-cyclin T1 used in EMSA was generated in sheep recognizing the C-terminal domain (Abcam ab27963). Here, 293T cells were seeded at 5 × 10 cells/well in a six-well tray a day before transfection with 7 µg of p256 proviral DNA or cotransfection with 6.5 µg of p256 proviral DNA and 0.5 µg of VSVG-expressing plasmid using the CaPO procedure () to generate p256 and VSVG pseudotyped p256 viral stocks, respectively. Virus-containing supernatants were collected at 24, 48, 72 and 96 h post-transfection and virus production was measured by titering the cell-free supernatants of p256 on HeLa37 cells and VSVG pseudotyped virions on HeLa S3 cells. HeLa37 cells were infected with dual-tropic HIV p256 at an MOI of 0.1 for glycerol gradient analysis. After 4 days, the cells were lifted from the plate.A total of 4 × 10 cells were re-plated and immunostained for HIV antigens. The remaining cells were harvested and washed with 1× PBS containing 1 µl/ml protease inhibitor cocktail (Sigma). Lysates were prepared to extract all P-TEFb and subjected to glycerol gradient sedimentation at 45 000 r.p.m. for 16 h as described below. pFLAG-CMV2-Tat was created by cloning HIV-1 Tat (amino acids 1–86) in frame with the amino terminal FLAG-tag in pFLAG-CMV2 (Sigma) using EcoRI and BamHI restriction sites. Here, 293T cells were trypsinized and 3 × 10 cells were plated on a 150-mm plate. The next day when cells were ∼80% confluent, the media was refreshed prior to calcium phosphate transfection. Seventy-five micrograms of plasmid DNA was added to 0.4 M CaCl in 1 ml of 1 mM Tris (pH 7.6) and 0.1 mM EDTA. This tube was mixed well and maintained at 37°C until mixed in a dropwise fashion with 1 ml of 2× HBS (50 mM HEPES, 1.5 mM NaHPO and 140 mM NaCl, pH 7.05) that was also held at 37°C. Mixture was vortexed thoroughly and added in a dropwise manner to the plate of cells. Media was refreshed at 12 h post transfection and maintained for an additional 30–36 h. At the termination of the transfection, a small population of the cells were immunostained for FLAG expression in a 48-well tray. Cells were fixed with 3.7% paraformaldehyde for 20 min followed by 3 × 1 ml washes with PBS. Cells were permeabilized with 0.25% Triton X-100 in PBS for 20 min and washed extensively with PBS at the completion of the incubation. Cells were incubated with PBS containing 5% fetal calf serum (FCS) to block non-specific binding. Anti-FLAG monoclonal antibody M2 (Sigma) was diluted 1:1250 in PBS with 5% FCS and incubated with the cells for 2 h at 37°C. Cells were rinsed three times with PBS and peroxidase-conjugated goat anti-mouse antisera (1:1000) was added until the reaction was sufficiently developed. The remaining cells were harvested and washed with 1× PBS, and cell lysates were prepared to extract all P-TEFb and subjected to glycerol gradient sedimentation at 45 000 r.p.m. for 16 h as described below. Cell lysates were prepared in Buffer A (10 mM KCl, 10 mM MgCl, 10 mM HEPES, 1 mM EDTA, 1 mM DTT, 0.1% PMSF and EDTA-free complete protease inhibitor cocktail from Roche) containing 150 mM NaCl and 0.5% NP-40 as previously described (,). The lysates were clarified by centrifugation at 14 000 r.p.m. for 10 min at 4°C in a microfuge. The supernatant was used as input for 4.8 ml, 5–45% glycerol gradients, in Buffer A with 150 mM NaCl, spun at 45 000 r.p.m. for 20 or 16 h in a SW-55Ti rotor in a L7-55 Beckman ultracentrifuge as previously described (,). The fractions were analyzed by western blotting with anti-cyclin T1 or anti-Cdk9 for P-TEFb, and a HRP-conjugated M2 FLAG antibody for FLAG-Tat. Following incubation with the appropriate HRP-conjugated secondary antibody, when necessary, the blots were treated with Super Signal Dura West (Pierce). The western blots were imaged with a cooled CCD camera (UVP) and the P-TEFb signals in the fractions containing the free and large forms of P-TEFb, as indicated in the figures, were quantitated with Lab Works 4.0. Kinase reactions were carried out with recombinant purified P-TEFb [Cdk9/cyclin T1 (1–290)] and DSIF in 30 mM KCl, 20 mM HEPES pH 7.6, 7 mM MgCl, 30 μM ATP, 2.5 µCi of [γ-P]-ATP (Perkin–Elmer) with 1 µg BSA. The reactions were incubated for 20 min at 30°C and stopped by the addition of SDS–PAGE loading buffer. Reactions were resolved by SDS–PAGE on a 7.5% polyacrylamide gel. The dried gel was subjected to autoradiography and quantitation was performed using an InstantImager (Packard). Twelve-microliter reactions were carried out in 25 mM HEPES, pH 7.6, 15% glycerol, 60 mM KCl, 10 mM DTT, 0.01% NP-40, 1 µg BSA with 200 ng yeast tRNA (Invitrogen) with or without 15 μM ZnCl as indicated, and including T7 transcribed, radiolabeled 7SK, recombinant Tat (NIH AIDS reagents program), HEXIM1, and P-TEFb containing cyclin T1 as indicated. Cold RNA oligos for competition studies were chemically synthesized (IDT). 7SK and tRNA were heated for 5 min at 75°C and cooled on ice for another 5 min, prior to addition. Reactions were incubated at room temperature for 20 min and resolved on a 4% polyacrylamide gel in 0.5× Tris/glycine at room temperature for 2 h at 4 W. The dried gel was subjected to autoradiography. Twelve-microliter reactions were carried out in 25 mM HEPES, pH 7.6, 15% glycerol, 60 mM KCl, 15 μM ZnCl, 5 mM DTT and 0.01% NP-40, including 50 ng of chemically synthesized TAR RNA (nucleotides 1–59) (Dharmacon), 640 c.p.m. of T7-transcribed radiolabeled TAR RNA, HEXIM1 and P-TEFb containing cyclin T1 (residues 1–290), as indicated. TAR RNA was heated for 5 min at 75°C and cooled on ice for another 5 min, prior to addition. Reactions were incubated at room temperature for 10 min and resolved on a native 5% polyacrylamide gel in 0.5× Tris/glycine at 4°C for 1.5 hat 4 W. The dried gel was subjected to autoradiography. Because most P-TEFb is found in the large inactive complex, and because several studies suggested that HIV requires higher P-TEFb activity than cellular promoters, we decided to test the hypothesis that HIV is able to release P-TEFb from the large form. First, the effect of HIV on the large form of P-TEFb was investigated through examination of the relative levels of the free and large forms of P-TEFb in HIV infected cells. HeLa37 cells were infected with the dual tropic HIV virus p256 () at an MOI of 0.1. Four days post-infection, the HeLa37 cells were determined to be ∼96% positive for the production of viral proteins by immunohistochemistry. The cells were lysed to extract all of the P-TEFb and the forms of P-TEFb were analyzed by glycerol gradient sedimentation and quantitative western blotting (A). Qualitatively, there is an obvious shift of both cyclin T1 and Cdk9 from the large to the small form following HIV infection. The Cdk9 signals for the small and large forms of P-TEFb from two experiments were averaged and SDs calculated (B) to reveal that 84% ± 1% of P-TEFb was found in the large form in untreated cells, whereas after 4 days of HIV infection, 72% ± 3% of P-TEFb remained in the large form. The change in the relative level of the large form of P-TEFb by HIV infection suggests that HIV is able to release and utilize P-TEFb obtained from the large form. Because the main difference between the activation of normal cellular promoters and the HIV LTR is the requirement for Tat and it is known that Tat is able to interact with P-TEFb, we hypothesized that the shift from large to free form observed during HIV infection was due to the expression of Tat in the HIV infected cells. To investigate this possibility, the effect of Tat on the large and small forms of P-TEFb present in uninfected cells was examined. In two separate experiments, 293T cells were transiently transfected with a construct expressing an N-terminally FLAG-tagged Tat. After 48 h, the cells were lysed and the P-TEFb was analyzed by glycerol gradient sedimentation and fractionation followed by quantitative western blotting. In control untransfected or β-Gal transfected 293T cells, most of the P-TEFb (∼90%) is found in the large, inactive form [A (fractions 11–15) and B (fractions 9–12)]. In cells expressing HIV Tat, there was a significant shift in P-TEFb from the large, inactive form to fractions that normally contain the free form [A (fractions 6–10) and B (fractions 5–8)]. Quantitation of the amounts of Cdk9 indicated that in the first experiment, expression of Tat caused a reduction of the large form from 90 to 56%. In the second experiment, the large form was reduced from untransfected control and β-Gal control (89 and 81%, respectively) to 36%. The transfection efficiency was not the same in the two experiments and, interestingly, the fractional change in the large form of P-TEFb correlated with the fraction of cells observed by immunostaining to be expressing Tat. This suggests that all of the P-TEFb was released from the large form within the transfected cells and that the remaining large form detected on the gradient was derived from untransfected cells. The larger change seen in these experiments compared to the experiment in which cells were infected with HIV is most likely due to the fact that Tat is present in higher amounts after transient transfection with a plasmid expressing Tat from the CMV promoter compared to the expression of Tat during HIV infection. In cells expressing HIV Tat, the glycerol gradient fractions containing the ‘free’ form of P-TEFb appear to be shifted slightly down the gradient compared to untransfected or β-Gal transfected cells, and Tat, with a molecular weight of 15 kDa, appears to co-sediment with the small form of P-TEFb, a complex with a molecular weight of at least 120 kDa (A and B). These observations suggest that Tat is bound to P-TEFb, and, in fact, co-immunoprecipitation of P-TEFb and HIV Tat from these fractions confirmed this interaction (data not shown). This association is consistent with many previous studies demonstrating Tat association with P-TEFb (,,,). Although an interaction with Brd4 and P-TEFb has been recently described () we do not think that expression of Tat in cells leads to the formation of a P-TEFb–Brd4 complex. The P-TEFb in fractions 4–8 of the glycerol gradients does not contain Brd4 because the 150 mM salt conditions used to extract P-TEFb from the nucleus do not extract Brd4. The interaction between Brd4 and P-TEFb can only be observed if nuclei are extracted with high salt (>300 mM) and then dialyzed down to low salt (,). From these studies we conclude that expression of Tat leads to the disruption of the large form and the formation of a Tat–P-TEFb complex. The studies just presented indicate that the expression of Tat leads to the release of P-TEFb from the P-TEFb–HEXIM1–7SK complex and to the formation of a P-TEFb complex. However, due to the ambiguity of the system, it is not clear whether this release is due to a direct effect of Tat on the complex or more a more indirect effect, such as Tat association with P-TEFb complex leading to lower levels of P-TEFb activity and less general transcription that might trigger the cellular mechanism that releases P-TEFb from the large form. Because of this, a defined system was used to investigate the biochemical mechanism of this release. EMSAs were carried out under equilibrium binding conditions with protein in vast excess over P-labeled transcribed full-length 7SK. Two-hundred nanograms of highly structured tRNA (>1000-fold excess over 7SK) was used in each reaction as a non-specific competitor instead of poly (rI):poly (rC) that was used previously (,,,) because we recently found that HEXIM1 binds tightly to dsRNA (). Dimeric HEXIM1 () formed its characteristic specific complex with 7SK (A, H1–7SK). Somewhat surprisingly, as increasing amounts of Tat were added to 7SK a more slowly migrating complex formed, demonstrating that Tat can bind to 7SK forming a Tat–7SK complex. A portion of the 7SK was also shifted into the well and this could be due to association of 7SK with oxidized aggregates of Tat. When HEXIM1 was pre-incubated with 7SK to form the HEXIM1–7SK complex and then increasing amounts of Tat were added before being loaded onto the native gel, the HEXIM1–7SK complex was gradually eliminated and was replaced by a Tat–7SK complex. When the Tat–7SK complex was preformed and then HEXIM1 was added, the competition was even stronger. The amounts of the Tat–7SK complex formed were virtually identical with or without HEXIM1 being present. The fact that larger molar amounts of Tat were required is likely due to the presence of a large fraction of inactive Tat in the preparation used. We conclude that Tat can compete strongly with HEXIM1 for binding to 7SK, and that even when the relatively stable HEXIM1–7SK complex is preformed, Tat can disrupt it. There was no evidence for a significant amount of a Tat–HEXIM1–7SK complex, indicating that Tat and HEXIM1 bind to an identical or overlapping region of 7SK or that Tat and HEXIM1 bind to different mutually exclusive conformations of 7SK. We next examined the effect of Tat on the P-TEFb–HEXIM1–7SK complex. In control reactions, the previously observed HEXIM1–7SK and Tat–7SK complexes were found. P-TEFb, containing Cdk9 and full-length cyclinT1, formed non-discrete complexes with 7SK (B). This is likely due to a weak, perhaps non-specific, interaction of P-TEFb and 7SK. A combination of Tat and increasing amounts of P-TEFb also gave a progression of non-specific shifts. However, when HEXIM1 and increasing amounts of P-TEFb were added together, the HEXIM1 shift gradually disappeared and a very discrete band containing HEXIM1 and P-TEFb formed (B, P–H1–7SK). To investigate the direct effect of Tat on formation of a P-TEFb–HEXIM1–7SK complex, HEXIM1, P-TEFb and increasing amounts of HIV Tat were pre-incubated followed by the addition of 7SK. As Tat was titrated into the reactions, a dose-dependent inhibition of the formation of the P-TEFb–HEXIM1–7SK complex resulted with the reappearance of a band with similar mobility to the HEXIM1–7SK complex. Under these conditions in the presence of P-TEFb, there was no evidence for a Tat–7SK complex suggesting that Tat was able to block formation of the P-TEFb–HEXIM1–7SK complex without blocking HEXIM1 binding. Because Zn has been shown to be required for efficient interaction between P-TEFb and Tat (), EMSAs were performed in the presence of 15 μM ZnCl. HEXIM1 and Tat still formed specific complexes with 7SK while the weak interaction of P-TEFb with 7SK was made even weaker as evidenced by higher migration rate of 7SK compared to the rate in the absence of zinc (A). Now, in the presence of zinc, there was evidence of a P-TEFb–Tat–7SK complex. When roughly equimolar amounts of P-TEFb and HEXIM1 (30 and 10 ng, respectively) were combined with 7SK, a shift of the HEXIM1–7SK complex was again observed, with supershift of the complex by an affinity-purified antibody directed towards the C-terminal domain of cyclin T1 confirming the presence of P-TEFb and the formation of the P-TEFb–HEXIM1–7SK complex. Upon titration of HIV Tat into reactions containing 30 ng of P-TEFb and 10 ng of HEXIM1, there was again a dose-dependent inhibition of the formation of the P-TEFb–HEXIM1–7SK complex. Under these conditions, there was only a very small amount of Tat–7SK complex formed at the highest Tat concentration. When similar reactions containing only 10 ng of P-TEFb were analyzed, more of the Tat–7SK complex was seen at high levels of Tat. In reactions containing 10 ng of HEXIM1 with 30 ng P-TEFb and 30 ng Tat or with 10 ng P-TEFb and 10 ng Tat, most of the P-TEFb–HEXIM1–7SK complex was gone and the HEXIM1–7SK complex was clearly visible. This finding is not easily explained by a simple competition between Tat and HEXIM1 for binding to 7SK, but rather suggests that under these conditions with zinc, a Tat–P-TEFb complex forms which does not allow P-TEFb to be recruited by the HEXIM1–7SK complex. Several shifts are visible at high levels of Tat that likely represent Tat–P-TEFb–7SK complexes (A, Tat–P–7SK). A supershift by anti-cyclin T1 confirmed the presence of P-TEFb in these complexes. It is possible that the two major shifts in this region represent Tat–P-TEFb–7SK complexes with one or two molecules of P-TEFb per complex. To further examine Tat inhibition of P-TEFb–HEXIM1–7SK complex formation and to determine if Tat could disrupt a preformed complex, EMSA was again used (B). Again, Tat was observed to inhibit the formation of the P-TEFb–HEXIM1–7SK complex (B, Inhibition). Additionally, as more P-TEFb was added to a constant amount of HEXIM1 in the absence of Tat (compare the first lane in each set of P-TEFb levels), there was the expected progressive shift of the HEXIM1–7SK complex to the P-TEFb–HEXIM1–7SK complex. As increasing amounts of Tat were included in the reactions, there was a progressive decrease in the P-TEFb–HEXIM1–7SK complex formed and corresponding increases in the HEXIM1–7SK complex and the previously postulated Tat–P-TEFb–7SK complexes. Interestingly, as the amount of P-TEFb was increased in the reactions, there was a progressive shift from the lower Tat–P-TEFb–7SK band to the upper band (compare the last lane of each set of P-TEFb), supporting the hypothesis that the upper band is a Tat–P-TEFb–7SK complex containing two molecules of P-TEFb. Again, only at the highest concentrations of Tat, after all P-TEFb has been bound by Tat, was the Tat–7SK complex observed. To determine if Tat is able to disrupt a preformed large complex, P-TEFb–HEXIM1–7SK complexes were formed before addition of Tat. As more Tat was added to pre-formed P-TEFb–HEXIM1–7SK complexes, there was a disruption of the complex, with formation of HEXIM1–7SK, Tat–P-TEFb–7SK, but not the Tat–7SK complexes (B, Disruption). Disruption of pre-formed P-TEFb–HEXIM1–7SK complexes by Tat appeared less efficient than inhibition of complex formation when the amount of P-TEFb–HEXIM1–7SK remaining at corresponding amounts of Tat is compared between the two experiments. Overall, these data demonstrate that Tat can prevent the formation of a P-TEFb–HEXIM1–7SK complex and can disrupt a preformed P-TEFb–HEXIM1–7SK complex. To gain insight into what sequence Tat recognized in 7SK, the first 100 nt of 7SK were compared to HIV TAR, and a set of oligos were designed and tested for their ability to block Tat binding to 7SK. The target for Tat binding in TAR consists of an AUCUG forming a bulge in the apical region of the TAR stem-loop and this binding of Tat changes the structure of TAR so that a pocket forms between U23 and G26 (). When the sequence of 7SK was analyzed, three AUCUG motifs were found (A). This sequence would be expected to occur randomly once every 4 bp, or about 1 in 1000 nt, and so it may be significant that it is found three times in the first 100 nt of 7SK. A set of oligos were designed that contained AUCUG sequences in the context of sequences that would be unstructured or that could form secondary structure with the AUCUG in loops or in bulges. Other regions of 7SK were also sampled as controls (A). mFold suggested that 7SK () would form a stem-loop with one AUCUG in a bulge and one in the loop and is shown compared to TAR in B. HEXIM1 has been demonstrated to bind to a region of in the 5′ end of 7SK () and to this specific oligo as well as it binds to intact 7SK (). Unlabeled oligos were then used as competitors in EMSAs with labeled full-length 7SK and Tat. No competition was seen for any of the oligos except for 7SK () (C and D). This strongly suggests that Tat associated with the residues from 10 to 48 of 7SK. Because no competition was seen with oligos comprised of 11–34 or 20–36, Tat may require the secondary structure shown in B to be able to interact with 7SK. To further analyze the specificity of the interaction of Tat with 7SK, a new oligo similar to 7SK () was used in a competition EMSA assay. The mutant oligo, 7SK (10–48 M), was identical to 7SK () except that an A was inserted in the first AUCUG site (AUCAUG) and U40 was deleted (A). These two changes eliminated the bulge region to which we thought Tat might bind and increased the length of the stem (B). The mutant oligo serves as a control for the potential binding of Tat to any RNA that contains a stem and loop with a bulge. As was seen above, as increasing amounts of the 7SK () were included in reactions containing Tat and P-7SK, the Tat–7SK shift was decreased with 100 ng of the oligo and completely eliminated with 1000 ng of the oligo (E). In contrast, 7SK (10–48 M) only partially blocked Tat binding to 7SK at the highest level of oligo. We conclude that the changes in the 7SK 10–48 oligo lowered the affinity of Tat. Our results up to this point indicate that Tat binds to 7SK in a specific manner and that the region of 7SK to which Tat binds is from nucleotides 10–48. Given the previous results demonstrating that HEXIM1 and Tat compete for binding to the same small 7SK oligo, and the similarity of this region with the TAR RNA, we explored the possibility that HEXIM1 might bind to TAR. If this is the case, it might also be able to bind and inhibit P-TEFb much the same way that HEXIM1 and 7SK inhibit P-TEFb, thus repressing P-TEFb, and transcription elongation, at the HIV LTR promoter. To begin testing this hypothesis, EMSAs were used to determine if HEXIM1 could bind to TAR and recruit P-TEFb. P-labeled, chemically synthesized TAR RNA was combined with FPLC-purified HEXIM1, and P-TEFb containing Cdk9 and the first 290 amino acids of cyclin T1, as indicated; complexes were resolved on a native gel and both silver-stained and exposed to film to visualize both the RNA and protein shifts (). When HEXIM1 and TAR were combined, a new complex migrating more slowly than either HEXIM1 or TAR RNA was observed, suggesting the formation of a HEXIM1-TAR complex (, H1–TAR). As P-TEFb was titrated onto this complex, there was a further shift of the complex with more mobility than free P-TEFb, accompanied by a change in the color of the silver-stained gel complex, suggesting the addition of a new component and the formation of a P-TEFb–HEXIM1–TAR complex (, H1–P–TAR). More importantly, free P-TEFb, which does not bind to RNA alone, disappeared. While this result supports the hypothesis that HEXIM1 can bind to TAR, with subsequent recruitment and binding of P-TEFb, inhibition of P-TEFb requires that the RNA change the conformation of HEXIM1 to relieve the autoinhibitory properties of the protein. To determine whether HEXIM1–TAR can inhibit P-TEFb, kinase assays were performed using recombinant P-TEFb(1–290), HEXIM1 and TAR RNA as in the EMSAs to monitor γ-P(ATP) incorporation into DSIF, and the complexes resolved by SDS–PAGE (A). Compared to P-TEFb alone and with DRB, an inhibitor of P-TEFb, there was no qualitative effect of HEXIM1 alone on P-TEFb kinase activity. Similarly, TAR alone, or 7SK alone as a control, demonstrated no appreciable effect. However, when HEXIM1 and TAR are combined and titrated in a constant ratio determined by EMSA to give 1:1 HEXIM1–TAR binding, there was a dose-dependent inhibition of P-TEFb, similar to that observed with HEXIM1–7SK. When the assay was quantitated (B), HEXIM1 and the RNAs separately had no significant effect; however, HEXIM1–TAR appeared to inhibit as effectively as equimolar amounts of HEXIM1–7SK. These findings support the hypothesis that HEXIM1 may actually be inhibiting P-TEFb at the HIV LTR. #text

The regulation of CFTR exon 9 splicing has been extensively studied in recent years because of its clear connection with CF disease (). At present, several splicing controlling regions have been characterized near the 3′ and 5′ boundaries of this exon. These include a polymorphic TG(m)T(n) region near the 3′ss, a suboptimal donor site and a Polypyrimidine-rich Controlling Element just downstream of the 5′ss (PCE) (). Moreover, additional controlling regions have been identified inside the exon itself in the form of CERES elements () and as an intronic splicing silencer region (ISS) further away in the IVS9 intron sequence () (A). During the course of these studies several trans-acting elements have also been identified as binding specifically to these regulatory elements: TDP-43 to the (TG)m region near the 3′ss of the exon, which has been recently shown to recruit hnRNP proteins near the 3′ss (,,), TIA-1 to the PCE that promotes exon inclusion (), and unidentified members of the SR protein family to the ISS sequence () (B). The SR protein family () has been predominantly studied in relationship with its involvement in alternative and constitutive splicing control () and indeed may have played a decisive role in the evolution of this process (,). However, it has also been recently shown to participate in a very wide range of functions that include the maintenance of genomic stability (), mRNA export (), mRNA surveillance () and protein translation (,). In splicing regulation, SR proteins are generally considered to bind exonic splicing enhancer (ESE) sequences () and in this way they generally promote exon inclusion in the pre-mRNA molecule that is processed by the spliceosome. This enhancement is achieved in a variety of ways: by antagonizing the effect of negative regulators such as hnRNP proteins (), by directly recruiting basic splicing factors such as U1 and U2snRNPs to the exon acceptor and donor sites (,), and by promoting spliceosome assembly through their RS domains (,). Because of all these functions, SR proteins represent one of the most important factors that promote exon inclusion () and is not surprising that an excess of SR proteins can compensate for complete U1snRNP inactivation and rescue correct splicing (,). In general, most SR proteins share rather common enhancer properties despite they have different sequence binding abilities (), protein domain compositional differences () or nucleo cytoploasm shuttling properties (). However, this is by no means a rule. In fact, some SR protein family members have also been recently identified in connection with splicing repression. For example, a novel SR protein designated SRp38 has been recently demonstrated to posses splicing inhibitory activity in mitotic cells or following heat shock treatment (). In addition, another SR-protein like factor (SR-15) has recently been described to possess general splicing inhibitory activity in the HSV1 virus (). Most importantly, there are many examples of factors that display either enhancer or repression activity in one system can display the opposite behavior in different pre-mRNAs. The SR proteins family is no exception to this observation. Indeed, past research has led to the discovery of a small number of splicing systems in which normally enhancing SR proteins display a inhibitory activity on the splicing process (). As previously mentioned, to this short list of examples we have to add the reports that describe SR proteins as general inhibitors of CFTR exon 9 splicing (,). To this date, however, no clear identification/mapping or functional binding sites for these inhibitory SR proteins has been provided. In this work, we have aimed to cover this gap and investigate the functional reasons that underlie this particular SR inhibitory activity. Plasmid TG11T5 has been previously described by Niksic . (). Plasmid pES was obtained by deleting part of the original IVS9 sequence in TG11T5 and inserting in its place a PstI/KpnI linker just before the NdeI cloning site (A and A). All the other plasmids used in this study were obtained by cloning the sequence of interest in the pES plasmid either in the PstI/KpnI (PK series) or PstI site (P series). This was achieved by annealing two complementary oligonucleotides containing the sequence of interest and ligating according to standard protocols (sequence of the oligonucleotides is available upon request). In order to mutate the cryptic 3′ss sequence in the TG11T5 context (mutant IVS9del3′ss) we used the two following oligos: 5′ctctttttttttctaatttgtagtg3′ sense and 5′cactacaaattagaaaaaaaaagag3′ antisense. To generate the RNA probe of h3′int a corresponding pBluescript II KS plasmid containing this sequence was linearized with NdeI and transcribed with T7 RNA Polymerase (Pharmacia Biotech) in the presence of αP-UTP, according to standard procedures. The UV cross-linking assay was performed by incubating 1 × 10 c.p.m.-labeled RNA probes with 100 μg of total HeLa nuclear extracts (CilBiotech, Mons, Belgium) and 100 μg heparin in a 20-μl final reaction volume containing 20 mM HEPES pH = 7.9, 72 mM KCl, 1.5 mM MgCl, 0.78 mM magnesium acetate, 0.52 mM dithiothreitol, 3.8% glycerol, 0.75 mM ATP and 1 mM GTP for 15 min at 30°C. Samples were transferred to HLA plate (Nunc, InterMed) on ice and irradiated with 0.8 J UV light for 5 min by using a BIO-LINK apparatus (Euroclone). Unbound RNA was digested with 30 μg of RNase A (Sigma) for 30 min at 37°C and incubated for 2 h at 4°C on a rotator wheel with 150 ml of IP buffer (20 mM Tris pH 8.0, 300 mM NaCl, 1 mM EDTA, 0.25% NP-40) and 1 ml of monoclonal antibodies anti-SF2/ASF (mAb 96) (Zymed Laboratories Inc), anti-RS phosphorylated domain (mAb 1H4) (Zymed Laboratories Inc) and an anti-SC35 monoclonal antibody (Sigma). Each mixture was then incubated with 30 ml of Protein A/G-Plus Agarose (Santa Cruz Biotechnologies) at 4°C overnight. Beads were collected by centrifugation, washed four times with 1.5 ml of IP buffer and then loaded onto a SDS-10% PAGE gel. Gels were run at a constant 30 mA for ∼3.5 h, dried under vacuum, and exposed for 4 days with a BioMax Screen (Kodak). Liposome-mediated transfections of 3 × 10 human hepatocarcinoma Hep3B cells were performed using DOTAP Liposomal Transfection Reagent (Alexis Biochemicals) according to manufacturer instructions. After 18 h the transfectiom medium was replaced with fresh medium and 24 h later the cells were washed with PBS and RNA was purified using RNAwiz (Ambion). RT–PCR reactions to specifically amplify the minigene transcripts was performed as previously reported (). In order to quantify the amplified fragments, the PCR reaction was performed in the presence of αdCTP and the samples run on a 5% denaturing polyacrylamide gel. Radioactive intensity was measured using a Cyclone (Packard). Transfection of a siRNA reagent against TDP-43 were performed as previously published (). In order to better define which SR proteins are binding to the ISS region we performed immunoprecipitation analysis using, as substrate, the entire h3′int intronic region (A). C shows an immunoprecipitation analysis with HeLa nuclear extract of this RNA (h3′int) together with two control RNAs from the fibronectin EDA exon, one bearing a well-characterized ESE sequence (hTot) and one where this sequence has been deleted (h▵2e) (). Each RNA was labeled using αP-UTP and incubated with ∼150 μg of Hela nuclear extract before being subjected to UV-crosslinking and digestion with RNAse A. Samples were then run on a 10% SDS–PAGE gel and exposed using BioMax autoradiographic films. Immunoprecipitation was performed using equal amounts of each UV-crosslinked sample and following the addition of specific monoclonal antibodies against SF2/ASF (C, left panel, mAb96), against the phosphorylated RS domain (C, center panel, mAb 1H4) and against the SC35 protein (C, right panel). The mobility of the SR proteins is indicated by arrows. The results show that SF2/ASF and SRp40 are the major SR protein family members binding to the h3′int region. In order to map the binding sites of each SR protein in h3′int we used in immunoprecipitation analysis a set of antisense oligos which targeted the original ISS region (A, AS1–AS4). A shows a schematic diagram of the entire IVS9 sequence inserted in the original TG11T5 plasmid () together with the sequences targeted by the AS1–AS4 antisense oligos that cover the originally mapped ISS region (). Also shown in this figure are the two single-point mutations introduced in h3′int to create a unique PstI cloning site that can be used to remove this entire region from the template minigene plasmid. Based on the results obtained in C, each oligo was then used in immunoprecipitation analysis with mAb 96 and mAb 1H4 together with the h3′int labeled RNA. As shown in B lane 3, oligo AS2 was the most efficient in inhibiting SF2/ASF binding to h3′int whilst oligo AS1 was very efficient in blocking SRp40 binding to this region (C, lane 2). A lesser amount of inhibition for SRp40 binding could also be detected in the presence of the AS4 antisense oligo (C, lane 5). However, as the AS1 and AS4 sequence share considerable similarity (almost 50%, especially in the ‘agaaatt’ central region) the AS4 oligo may have cross-hybridized with the AS1 sequence blocking partially its interaction with SRp40 . This hypothesis is consistent with the observed lack of functional effects of the AS4 sequence alone on CFTR exon 9 inclusion. A shows a schematic representation of the CFTR exon 9 hybrid minigene construct lacking the ISS sequence (pES). In this construct, the IVS9 sequence was shortened by exploiting the creation of a novel PstI site which was directly joined to NdeI through a small linker that also provided a unique KpnI site. The unique PstI/KpnI sites could then be used to insert different combination of the AS1–AS4 sequences in the ISS position (A, lower panel, has a detailed scheme of the inserted fragment position in the pES plasmid). Based on the results of the immunoprecipitation analyses it was then decided to insert in pES the two combinations of AS1 + AS2 (pTB AS1 + AS2PK) and AS3 + AS4 (pTB AS3 + AS4PK). B shows the RT–PCR assays of these plasmids (lane 3 and lane 4, respectively) following transfection in Hep3B cells together with two control minigene constructs: the original TG11T5 (lane 1) and the pES minigene (lane 2). The upper and lower bands correspond to exon 9 inclusion (ex9+) and exclusion (ex9−), respectively. Quantification of these bands from three independent experiments following radioactive RT–PCR (C) demonstrated that the SR-binding AS1 + AS2 region could entirely recover the ISS inhibitory activity displayed by the original TG11T5 minigene construct (compare lanes 1 and 3). On the other hand, the AS3 + AS4 region that had no apparent SR-binding ability could not display any inhibitory activity with respect to the original pES plasmid (compare lanes 2 and 4). A similar pattern could also be observed when each individual sequence was inserted in the pES plasmid (D). A quantitation of the inhibitory activity of each sequence (E) confirmed that only the AS1 and AS2 sequences (D, lanes 2 and 3) could inhibit CFTR exon 9 inclusion whilst AS3 and AS4 did not cause any drop in CFTR exon 9 splicing efficiency (D, lanes 4 and 5). Moreover, the fact that the AS1 and AS4 showed very distinct inhibitory activities supports the conclusions drawn from the immunoprecipitation experiments in C. In consideration of the fact that ISS activity strongly correlates with AS1 + AS2 it was of interest to determine whether different SR protein-binding sequences were also capable of inhibiting CFTR exon 9 inclusion. To this end, it was decided to test the activity of two unrelated SR-binding sequences that have been extensively characterized by our laboratory: the ESE region of the fibronectin EDA exon () and the ISE region (ApoISE) of the Apo AII intron 3 () (A). The advantage of using these sequences is represented by the fact that they all possess different combinations of SR protein-binding abilities (A) and also that in their respective contexts they functionally behave as ESE and ISE elements, respectively. These additional polypurinic sequences were then cloned in the PstI and KpnI sites of the pES plasmid and transfected in Hep3B cells. B and C show that both the EDA ESE sequence (pTB-EDAPK, lane 1), and the ApoISE sequence (pTB ApoISEPK, lane 2) were capable of inhibiting CFTR exon 9 inclusion in a way comparable with that observed with pTB AS1 + AS2PK (lane 3) and higher than the pES plasmid alone (lane 4). For the sequences that can bind SF2/ASF, the functional specificity of SR protein binding on the inhibitory activity was also investigated by overexpressing this particular SR protein in transfected cells (D). As shown in this figure, the increased levels of CFTR exon 9 inhibition for constructs pTB AS1 + AS2PK (D, lanes 3 and 4) and pTB-EDAPK (D, lanes 5 and 6) is comparable to that observed for the TG11T5 plasmid (D, lanes 1 and 2) and greater than the one observed for the pES plasmid alone (D lanes 7 and 8). In addition, SC35 overexpression does not result in increased levels of CFTR exon 9 skipping in the pTB AS1 + AS2PK plasmid (D, lanes 9 and 10). This result is consistent with the immunoprecipitation results shown in . We have also tried overexpression of SRp55 and SRp75 that also induced CFTR exon 9 skipping to different degrees (data not shown), in keeping with the SR protein response profile obtained in the original study by Pagani . (). The binding experiments did not show reproducible direct interaction of these proteins with h3′int, although at times bands compatible with SRp55 and SRp75 were visible (C, middle panel). The lack of correlation between this absence of binding in IP assays and functional experiments may be due to the well-known non-specific effects of SR protein overexpression or to indirect interactions with the ISS sequence. Further studies will be needed to clarify these differences. In any case, taken together, these data further reinforce our conclusion that the AS1 + AS2 sequence contains all the functional properties of the originally mapped ISS element. The results in conclusively show that both the EDA ESE and the Apo AII ISE sequences were capable of acting as ISS elements in the CFTR exon 9 context. Therefore, it was of interest to determine whether the exon 9 AS1 + AS2 ISS sequence and the Apo ISE sequence were capable of acting as ESEs in a heterologous splicing context. To this end, we cloned the AS1 + AS2 and ApoISE sequences in the dsx-XH reporter system (,) (Figure S1). This is a well-known splicing system in which processing of the IVS3 intron from the () gene is dependent on the sequences inserted at the 3′ end of the construct. As shown in Figure S1, both the AS1 + AS2 and ApoISE sequence display higher ESE activity than the control AS3 + AS4 sequence. In order to further test these potential enhancer activities of AS1 + AS2 and ApoISE in the CFTR exon 9 context it was then decided to insert in the pES plasmid a donor sequence that could be used as a viable 5′ss to promote ‘exon’ inclusion. This was achieved by simply cloning the AS1 + AS2, ApoISE, and EDA ESE sequences in the PstI site of the pES plasmid. A shows that this cloning procedure, as opposed to cloning in the PstI/KpnI sites, provided any PstI-inserted sequence with a downstream donor site sequence possessing a score of 0.76 according to the NNSPLICE predictor program (). The results of this cloning procedure on the resulting pTB AS1 + AS2P, pTB ApoISEP and pTB EDAP plasmids are reported in B, lanes 1, 3 and 5, respectively. In all three cases, the cloning in the PstI site alone resulted in the appearance of an extra band which, when sequenced, was shown to consist in a ‘mini-exon’ sequence that exploited the newly inserted 5′ss and a cryptic 3′ss in the h3′int sequence (B, lower panel). Interestingly, the intensity of this extra band in the pTB AS1 + AS2P and pTB ApoISEP is markedly different, with the pTB AS1 + AS2P splicing profile still retaining some of the ex9+/ex9− splicing forms. This is consistent with the different ESE abilities displayed in the dsx-XH plasmid (Figure S1), where the ApoISE sequence is a much more efficient ESE (66%) than the AS1 + AS2 sequence (20%). These results also suggested that the action of the ISS element could be dependent on the presence of the naturally occurring cryptic 3′ss in CFTR IVS9, perhaps through the recruitment of a non-productive spliceosomal complex in this position. However, mutating this 3′ss sequence from ‘ag’ to ‘aa’ in the natural TG11T5 context (mutant IVS9del3′ss) did not significantly affect the efficiency of CFTR exon 9 inclusion (C). This result ruled out the possibility that this acceptor-like sequence alone could be playing a role in wild-type ISS functioning. To better assess the context-dependent effects of these different SR protein-binding sequences in the vicinity of CFTR exon 9 we then improved the definition of this exon by removing TDP-43 through siRNA treatment (Figure S2). As previously demonstrated, TDP-43 is a major inhibitory splicing factor that specifically recognizies the TGm polymorphic locus in IVS8 (,), and its specific removal from the transfected cell can offset many splicing inhibitory effects, including those mediated by the ISS (). It was therefore interesting to test the effect of removing this factor on the splicing patterns of the pTB AS1 + AS2P, pTB ApoISEP and pTB EDAP minigenes. Figure S2 shows that removal of TDP-43 in these different contexts can have very different outcomes depending on the type of enhancer sequence present in the ISS position. Finally, it was interesting to assess whether the inhibitory effects mediated by these different G/A-rich sequences in the ISS position could also be mimicked by a non-polypurinic sequence with well-known splicing enhancer effects. Therefore, an A/C-rich enhancer sequence from the alternatively spliced exon v4 of the CD44 gene was inserted in the ISS position (A) (). This sequence was identified by Stickeler . () as the binding site of YB-1, a member of the family of multifunctional cold shock domain proteins (CSD proteins). As shown in B, cloning this YB-1 binding sequence in the PstI/KpnI sites of the pES plasmid (pTB YB-1PK) can successfully mimick the effect of the original ISS activity in TG11T5 (compare lanes 1 and 3). Analogously, when this sequence is cloned in the PstI site only of the pES plasmid to obtain the pTB YB-1P construct (thus providing a downstream donor site capable of supporting exon recognition) the splicing pattern is totally shifted towards the inclusion of a miniexon containing only the YB-1 sequence (C, compare lanes 1 and 2). CFTR exon 9 splicing is a complex event in which several -acting elements located in both intronic and exonic sequences play an important role (,). In particular, an ISS sequence in IVS9 has been previously shown by Pagani . () to represent a negative element towards its inclusion in the final mRNA molecule. Our work has shown that this region specifically binds two members of the SR protein family, SF2/ASF and SRp40, which are normally associated with binding to enhancer elements that promote exon inclusion. Because of this, the CFTR ISS is one of the relatively few splicing systems in which SR proteins behave as suppressors of splicing. It is therefore interesting to compare its functioning with other systems in which SR proteins have also been identified as negative splicing regulators () (). In the hnRNPA1 gene ISS, a particular SR protein (SRp30c) can recognize the silencer element (CE9) and directly down-regulate exon recognition () (A). Most importantly, CE9 cannot compromise the assembly of U2-dependent complexes on the 3′ss of hnRNP A1 exon 8 () and cannot function as an enhancer element in a heterologous system (). Although the molecular mechanism of CE9 functioning remains unclear, this represents an important functional difference with respect to the CFTR exon 9 ISS, because the AS1 + AS2 sequence can function as an enhancer sequence in a heterologous splicing system. A second type of inhibition has also been found in the case of what is really a ISE element localized downstream of b-tropomyosin gene exon 6A (). In this case, competition occurs between a SR protein enhancer factor (SF2/ASF) and another SR protein that has no enhancer effect in this system (SC35) (B). Also this model does not seem to apply to the CFTR exon 9 ISS working model for a number of reasons. First of all, mapping of the SF2/ASF and SRp40-binding sites on the CFTR ISS sequence has shown that they are physically distinct, making competition between the two highly unlikely (unlike the case of the ISE element of b-tropomyosin exon 6A where the binding sites of SF2/ASF and SC35 actually overlap). In addition, our observation that heterologous SR-binding sequences with rather different splicing specificities (EDA ESE and Apo ISE elements) and even a YB-1 binding sequence can restore CFTR ISS function in the absence of its natural sequence would tend to rule out any specific effects by particular SR proteins. A higher degree of similarity can be found between CFTR exon 9 ISS and the action of the IIIa repressor element (3RE) originally described in Adenovirus () (C). In fact, just like CFTR exon 9 ISS, also the 3RE sequence was observed to function as an enhancer element when inserted in a heterologous splicing system, showing that the mechanisms of action of 3RE is wholly dependent on context. From a functional point of view, the inhibitory mechanism mediated by 3RE was initially thought to reside in the physical inhibition of U2snRNP binding to the IIIa 3′ss, because of its nearness to the IIIa acceptor site (∼30 nt) (). More recently, the inhibitory activity of SF2/ASF on IIIa splicing has been specifically identified as residing in its second RNA binding domain, RBD2, although the exact mechanism still remains to be defined (). It is difficult to determine whether physical hindrance may also represent the mode of action for CFTR ISS as this sequence is localized rather far away from the CFTR exon 9 5′ss (∼80–100 nt). Given this limitation, the potential physical hindrance between U1snRNP binding to the natural donor site and SF2/ASF and SRp40 binding to AS1 + AS2 seems unlikely. Another SR-mediated inhibitory situation is represented by the Negative Regulator of Splicing (NRS) of the Rous Sarcoma Virus (D). In this case, together with SR protein SF2/ASF (), the NRS can also bind a U1snRNP molecule (), a U11 snRNP molecule () and hnRNP H (). Interestingly, also this sequence can function as an enhancer element when inserted in the dsx-HX splicing system (). However, the mechanism of action of the NRS sequence cannot be explained by physical hindrance because it is localized very far away from the 3′ss that is inhibited. Recent research has shown that the exact mechanism through which splicing inhibition occurs may probably reside in the formation of non-productive complexes between the NRS inhibitory splicing complex, the 3′ss of the exon, and the polyadenylation process (,). In addition, it has been suggested that RNA polymerase II can ‘tether’ emerging splice sites in the pre-mRNA (). In this case the NRS might also act as a disturbing presence for the recognition of the natural 5′ss. The CFTR exon 9 ISS is an excellent example of the importance of sequence-context in determining the action of -acting sequences (). In the natural situation, the heterologous 5′ss sequence is absent hence no IVS9 intron sequences may be ‘exonized’. Nonetheless, the presence of the exon enhancer-like complexes formed by SR proteins or YB-1 in the ISS may create a situation in which the U1snRNP molecule approaching CFTR exon 9 would remain ‘undecided’ between binding to the wild-type suboptimal site and waiting for an indication from the SR/YB-1 proteins present in the ISS of a better target immediately downstream (E). This is consistent with the fact that the outcome of TDP 43 removal depends on the strength of the ISS–SR interactions. Weak SR interactions and removal of TDP 43 lead to complete recovery of CFTR exon 9 inclusion while strong SR interaction results in the inclusion of a super-exon 9 sequence by selecting exclusively the new 5′ss downstream of the ISS (see Fig. S2). This data suggests that the ISS may act as a sort of ‘decoy’ system hampering recognition of the exon 9 5′ss. The result of this stalemate would be a net decrease in CFTR exon 9 donor site recognition and, consequently, in a lesser inclusion of CFTR exon 9 in the mature mRNA. In this respect, therefore, a critical issue might be represented by the processing speed of the RNA polymerase II molecule in presenting the ISS sequence after having transcribed the CFTR exon 9 region, and these issues are currently being investigated in our laboratory. Finally, from a pathological point of view, the importance of having mapped exactly the binding sites for the SR trans-acting factors that are responsible for this inhibitory action can provide researchers with a useful target to inhibit their action, for example, by antisense oligonucleotide approaches (,). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

RNA secondary structure is increasingly recognized as a powerful modifier of splicing events (). At the local level, RNA conformations have been shown to regulate the splicing process by affecting the basic identifying features of an exon. Numerous examples of such a regulatory role have been recently reported to occur for the donor site of Tau exon 10 (), SMN2 exon 7 (), the branch/acceptor sites in gene (), the internal splicing enhancer region in the fibronectin gene () or to silencer regions in the gene (). In addition, RNA secondary structures have been shown to involve interactions between very distant regions of the pre-mRNA such as in (,), humans () and (). Recently, complementary intron sequence motifs have also been proposed to mediate the peculiar phenomenon of exon repetition (). Finally, RNA secondary structures are increasingly shown to play a part in other facets of mRNA biology such as in mantaining its stability (), regulating translation () or transport (). Taken together, these examples are consistent with the indications provided by analyses which predict the existence of a vast array of conserved structural features both in selected human protein coding RNA transcripts () and in the human genome in general (). From the latter has emerged the recent small RNAs revolution of functional non-coding RNAs (,). Interestingly, RNA secondary structures have also been proposed to play a role in helping the splicing machinery to distinguish between real exons and pseudoexon sequences (). Pseudoexon sequences are loosely defined as intronic sequences between 50 and 200 nt in length that are flanked by apparently good-to-consensus acceptor and donor-site signals. These sequences, however, are apparently never recognized by the splicing machinery () although this definition may contain many exceptions. In fact, it has recently been proposed that many members of this class may indeed be used to regulate the relative abundance of different pre-mRNA isoforms by selective Nonsense-Mediated Decay of alternatively spliced exons (). In addition, a distinct class of pseudoexon sequences derived from transposable elements known as sequences has been recently established as a major source of ‘real’ coding exonic sequences (). Nevertheless, estimates regarding pseudoexon sequence abundance in a typical pre-mRNA molecule has shown they may outnumber ‘real’ exons by an entire order of magnitude (). It is therefore clear that, no matter how many exceptions to the exclusion rule there may be, avoiding the insertion of these sequences during the normal splicing process would be essential for correct pre-mRNA processing. To achieve this, recent research has proposed that their exclusion may be achieved by a combination of factors (), including an enrichment within their sequence of inhibitory elements (,) or through the indication that pseudoexons flanking regions have a distinct tendency to form double-stranded structures that include the pseudo exon itself (). Regarding human diseases, pseudoexon inclusion events have increasingly been described to occur as a result of a single-point mutation deep within intronic regions. In general, these mutations have the effect of creating either a very good acceptor or donor splice site followed by the selection of ‘opportunistic’ complementary sites. A few exceptions that are worth noting have been described elsewhere where the intronic mutation affects a regulatory element within the pseudoexon itself (). For a general review of this topic see Buratti . () and Supplemetary Table 1 for an updated list of pathological pseudoexon inclusion events. This kind of aberrant insertional events have never had any coding potential, as they all originate from the chance occurrence of a triggering mutation. Therefore, no selection pressure may have been applied during evolution to regulate their inclusion as far as codon usage/regulatory sequences are concerned (,). Interestingly, one characteristic feature of these pseudoexons is their reduced length with respect to the normal length spread of exons which is reported to exist in the human and mouse genomes (Supplementary Figure S1) (). This reduced length, intriguingly similar to the previous estimates of the ‘window’ of naked RNA available for folding after transcription (), suggests that intrinsic structural features of these RNAs could represent a major determinant in their ‘exonization’ process. In this work, we have used representative examples of these pseudoexon inclusion events to test the importance of RNA secondary structures in their splicing regulation. The PY7 plasmid has been described in detail elsewhere (). For our purposes, we have inserted two SmaI and NdeI unique cloning sites at positions 44 and 50, respectively, in its 111-nt long intron. Both the wild-type (wt) and gtaa-deleted (Δ) ATM sequences were amplified from the pATM plasmids used in the original report () and inserted in the SmaI site of PY7 (A). The oligos used for this task were the following: 5′-ttgctcaagctcttaactgcaacagtggt-3′ (s) and 5′-gtcaaacagaaaattcaaatcccag-3′ (as). In order to obtain the ATM 46-48T Δ mutant a two-step PCR method was used with the following primers 5′-tgagggtacgtatgccctagatg-3′ (s) and 5′-catctagggcatacgtaccctca-3′ (as). From this mutant we obtained ATM 46-48T 41A Δ using primers 5′-cactctactgatgaggatacg-3′ (s) and 5′-cgtatcctcatcagtagagtg-3′ (as). Finally, the primers used to insert the 21–23 substitution in both the ATM Δ and ATM 46-48T Δ mutants were the following: 5′-gtgatataccctcactctac-3′ (s) and 5′-gtagagtgagggtatatcac-3′ (as). On the other hand, to obtain the ATM 5′-new Δ mutant we created a unique StuI site by joining together nucleotides 38–40 (agg) with nucleotides 51–53 (cct). The ATM 5′-new Δ mutation was then introduced by ligating in this site the 5′-gtaggtaagtacgaaggc-3′ (s) and 5′-gccttcgtacttacctac-3′ (as) oligos. Donor splice-site scores were calculated according to the NNSplice 0.9 program available at (). The human genomic fragment flanking the CFTR pseudoexon region was cloned into the SmaI site in the modified PY7 vector (CFTR WT) following amplification with oligos: 5′-attggtttttaaaaaaatttttaaattggc-3′ (s) and 5′-ccatattaaatagaaatgagataatttc-3′ (as). In order to obtain the following mutants the CFTR WT construct was subjected to two-step PCR method of mutagenesis using the following primers: 5′-atataagttaggtaactaacaa-3′ (s) and 5′-ttgttagttacctaacttatat-3′ (as) for CFTR WT2, 5′-atataactaaggttagtaacaa-3′ (s) and 5′-ttgttactaaccttagttatat-3′ (as) for CFTR WT3, 5′-gatataacttaggtaagtatcaat-3′ (s) and 5′-attgatacttacctaagttatatc-3′ (as) for CFTR WT4, 5′-tacttgagatgtaagtaaggt-3′ (s) and 5′-accttacttacatctcaagta-3′ (as) for CFTR MUT, 5′-tttattacagcaacaattac-3′ (s) and 5′-gtaattgttgctgtaataaa-3′ (as) for CFTR Del1, 5′-agaatcctatgagatgtaag-3′ (s) and 5′-cttacatctcataggattct-3′ (as) for CFTR Del2. Finally, in order to construct the CFTR Rep mutant a two-step PCR on CFTR MUT mutant plasmid was performed first by using the following oligos: 5′-agaatcctatcatgaagagatgtaag-3′ (s) and 5′-cttacatctcttcatgataggattct-3′ (as) and then the resulting mutant (CFTR Dis) was subjected to the same methodology with the following primers: 5′-tttattacagttcatgcaacaattac-3′ (s) and 5′-gtaattgttgcatgaactgtaataaa-3′ (as) to make the CFTR Rep mutant. Splicing reactions were performed using capped, SP-6 transcribed RNAs. Standard reactions were carried out in a 25 μl volume at 30°C for 2 h. Each reaction contained 15 μl of Nuclear Extract from HeLa cells (CilBiotech, Mons, Belgium, approx. concentration 10 μg/μl), 5 μl of 13% (w/v) polyvinyl alcohol, 1 μl of 80 mM MgCl, 1 μl of 12.5 mM ATP, 1 μl of 0.5 M creatine phosphate and 1.25 μl of 0.4 M Hepes-KOH pH = 7.3 and 2 μl of transcribed pre-mRNA at 200 μg/ml. Therefore, the final concentrations of the various components in a standard processing reaction were as follows: 3.2 mM MgCl, 500 μM ATP, 20 mM creatine phosphate, 2.7% (w/v) PVA, 20 mM Hepes (pH = 7.3), 6 μg/μl of Hela nuclear extract and 16 μg/ml of transcribed pre-mRNA. The processed RNAs were then extracted from the reaction mix using RNAwiz (Ambion, Inc.) and analysed by RT-PCR using a set of primers at the beginning of tropomyosin exons 2 and 3, respectively: 5′-gaatacaagcttgtcgaggaggac-3′ (s) and 5′-agaccggaattcggatcctctagag-3′ (as). In order to insert the various PY7-based sequences in the eukaryotic expression vector pcDNA3 the inserts were amplified using the following oligos T2F 5′-agggtaccagcttgtcgaggaggacatctcag-3′ and T3R 5′-cctctagagtcgatcgacctgcagg-3′ and inserted in the KpnI and XbaI sites of pcDNA3. Liposome-mediated transfections of 3 × 10 human hepatocarcinoma Hep3B cells were performed using DOTAP Liposomal Transfection Reagent (Alexis Biochemicals) according to manufacturer instructions. After 18 h the transfectiom medium was replaced with fresh medium and 24 h later the cells were washed with PBS and RNA was purified using RNAwiz (Ambion). In order to rescue the splicing of CFTR WT and CFTR WT2 we have also expressed a variant U1snRNP molecule (C > G U1) that have been described in a previous work from our lab (). RNA secondary structure determination with the use of limited V1 RNAse (Ambion), T1 RNAse (Ambion) and S1 nuclease (Fermentas) digestion has been described in detail elsewhere (). Briefly, 1 μg aliquots of ATM Δ RNA were digested in 100 μl final volume with 0.002 U of RNAse V1, 0.05 U of RNAse T1 and 19 U of S1 nuclease for 10 min at 30°C. An enzyme-free aliquot was processed together and used as a control. The cleaved RNAs were retrotranscribed according to standard protocols using the following antisense primers labelled with P-end-labelled oligonucleotide primers: 5′-gtcaaacagaaaattcaaatccc-3′ for the pseudoexons and 5′-ccatattaaatagaaatgagataatttc-3′ for CFTR pseudoexons. secondary structure predictions were performed using the mFold program (,). Recently, we have described the inclusion of a 65-nt long pseudoexon between exons 20 and 21 in a patient affected by ataxia-telengiectasia (). The activation event consisted of a 4 nt deletion (GTAA) that occurred within a high-affinity U1snRNP-binding site that surprisingly was not a masked 5′ss but an internal splicing repressor. This region was termed ISPE, for Intron-Splicing Processing Element (,). A shows the nucleotide sequence of the pseudoexon (in bold), together with the ISPE sequence (underlined, the GTAA sequence is in italic) and its intron flanking regions. The major peculiarity of this event is represented by the observation that although the ISPE itself is an excellent 5′ss it is never used by the splicing machinery. The reason for this probably resides in its vicinity to the well-defined 3′ss sequence of the ATM pseudoexon (3′pe) that possesses a score of 0.93 according to the NNSplice program (). In fact, recent experiments using modified U1snRNA molecules binding to consecutive positions along this pseudoexon have suggested that the U1snRNP molecule has to be placed 40 nt away from the 3′ss before pseudoexon inclusion can be observed (). Even more surprisingly, the only donor site (5′pe) used is represented by a AAGaagt non-consensus sequence (,). The GC dinucleotide is a fairly rare donor-site sequence which accounts for only 0.56% of all donor-site junctions in humans (). Moreover, in this specific context this choice is made even more peculiar due to the presence inside the pseudoexon itself of another reasonable potential donor-site sequence (A, the AGGACGA sequence, designated throughout this work as 5′int) in addition to the ISPE itself. In fact, the 5′int site possesses a well above cut-off score of 0.79 by the NNSplice program whilst 5′pe is not even considered as a potential donor site by this prediction program. This difference in predicted strengths in favour of 5′int is also reflected by several other donor-site scoring methods such as MaxEntScan (ME, MDD, MM and WMM scores for 5′int are 7.38/10.38/7.61/6.10, respectively compared to the scores of 3.24/7.72/4.43/4.95 obtained for 5′pe) () or Spliceview () where 5′int has a predicted score of 81 whereas 5′pe is not detected at all. In order to clarify this peculiar mechanism of donor-site usage it was decided to clone part of this sequence in the PY7 splicing system for analysis (A). Two versions of this construct were initially analysed, the first one (ATM WT) containing the wild-type intronic sequence whilst the second one (ATM Δ) carrying the 4-nt (GTAA) deletion that had been previously characterized to cause pseudoexon activation (). The transcribed RNAs from these plasmids were then used in a standard splicing assay and analysed both by polyacrylamide gel electrophoresis and by RT-PCR in a time-course analysis (B, upper and lower panels, respectively). As shown in both B panels the ATM Δ RNA could successfully induce pseudoexon inclusion whilst, as expected, no pseudoexon inclusion could be detected during the processing of the ATM WT RNA. It should also be noted that there is a strong similarity between the results obtained using the radioactively labelled RNAs and the RT-PCR products. Subcloning and sequencing of the RT-PCR reaction in B lower panel, lane 4 confirmed that only the 5′pe sequence was used in the pseudoexon inclusion event and failed to detect any use of the 5′int donor site (data not shown). The data, therefore, wholly confirmed the previous results obtained using minigenes and patient's transcript analyses (,). It was then decided to investigate the eventual activation of the 5′int donor site by introducing two single nucleotide substitutions in position 46 and 48 (ATM 46-48T Δ mutant). As shown in C, this had the effect of improving the predicted NNSplice donor-site score from 0.79 to 0.99. The ATM 46-48T Δ mutant was then spliced and the results showed that this improvement caused the almost complete switching from normal pseudoexon donor-site usage to this internal sequence (D, lane 4). Interestingly, inactivating this improved cryptic 5′gt sequence in the presence of the 46–48 substitutions by mutating it to a 5′at sequence (C, ATM 46-48T 41A Δ mutant) restored splicing to the ATM Δ RNA pattern (D, lane 6). However, D lane 8 shows that a mutation (ATM Δ G41) which strongly improves the predicted donor-site score in the 5′int position from 0.79 to 0.97 (C) did not result in preferential usage of this splice site (compare D lane 4 and lane 8). It was therefore evident that additional factors besides predicted scores were playing a role in defining donor-site usage within the ATM pseudoexon. In consideration that RNA secondary structure represents a potentially very powerful modifier of splice-site usage it was then decided to investigate the potential folding of this sequence. Preliminary analyses obtained using the mFold program (,) suggested that the ATM Δ pseudoexon sequence could fold upon itself to form a very compact, double-stranded structure (B). Its existence was thus tested experimentally by partial RNA digestion using single and double-strand-specific RNAses such as V1 (which cleaves double-stranded RNA and stacked RNA regions), T1 (which cleaves single-stranded guanosines) and S1 nuclease (which cleaves single-stranded RNA without sequence specificity) (A). The overall position of the different cleavages shows that the ATM Δ pseudoexon sequence does indeed display a structure that is consistent with the model proposed in B. In addition, it is interesting to note that two particularly strong RT stops (indicated by arrows in A) seem to enclose the pseudoexon sequence, possibly reflecting the difficulty of the RT to travel across a compact structure. Taken together, these results allow us to propose a working model of splice-site usage in this pseudoexon. In this model, once the repression provided the ISPE sequence is removed the 5′pe sequence is chosen as donor site because the RNA secondary structure hides the internal 5′int sequence. This is consistent with previous data which showed that short artificial hairpins have been long known to be able to inhibit usage of a donor site embedded in their stem (). Moreover, the stem configuration may also provide a competitive advantage by approximating the 3′pe to the 5′pe sequence. The importance of internal 5′ss availability was then tested experimentally by engineering a mutant ATM sequence (ATM 5′-new Δ) that carried a 5′ss sequence designed to remain in an open structure/internal bulge conformation (B). It should be noted that these sequence changes also abolish the original 5′int donor like sequence according to the NNsplice prediction program although not according to the MaxEntScan program (data not shown). The fact that the ATM 5′-new Δ mutant has a more open structure also in the mutated 5′int donor site is suggested by the increase of S1 single-strand cleavages in this region of the ATM 5′-new mutant Δ when compared with the ATM Δ mutant (B and B). The structural probing of this mutant shown in A displays a marked reduction in the number of V1 cleavages (double-strand specific) and an increase in the S1 single-stranded cleavages around the central bulge region carrying the new donor site. Therefore, with respect to the structural analysis shown in A for the ATM Δ mutant, the overall picture of the ATM 5′-new Δ mutant shows a predominantly open configuration (and thus presumably becomes more accessible for splicing factors). In keeping with this, when the activity of this mutant was tested (C) the results demonstrated that this time the new internal 5′ss are preferentially used with respect to the 5′pe sequence (C, compare lanes 3 and 4). Interestingly, direct sequencing of the amplified product has showed that although most fragments (∼75%) derive from the use of the 5′-new donor site whilst the remaining are derived from 5′int donor-site usage (data not shown). This result further highlights the importance of an ‘open’ structure on donor-site sequence availability. In addition, because the efficiency of pseudoexon recognition when transfected in eukaryotic cells using minigene systems was significantly higher than () the behaviour of this mutant was also tested following its introduction in pcDNA3 and transfection in Hep3B cells. Also in this case the ATM 5′-new Δ mutant was shown to splice predominantly to the internal donor sites when transfected in Hep3B cells (D, compare lanes 1 and 2). The functional importance of RNA secondary structure was also tested experimentally by performing a more subtle mutational analysis based on the proposed RNA secondary structure of the ATM pseudoexon in B. According to this model, it should be possible to reduce 5′int activation in the ATM 46-48T Δ mutant by ‘tightening’ the upper stem (A). In order to do this, the thymidines in position 21 and 23 were replaced by cytosine both in the original ATM Δ context as control (mutant ATM 21-23C Δ) and in the ATM 46-48T Δ context (mutant ATM 21-23C 46-48T Δ). The rationale of this change was to replace the natural UG ‘wobble’ base-pairings in these positions with two more canonical CG base pairs. This change, according to secondary structure predictions, would have the effect of raising the free energy of the ATM pseudoexon structure from –20.9 to –26.2 kcal/mol (A). B shows that the introduction of the 21–23C substitutions had no effect in donor-site usage of the ATM Δ context (B, lane 4). However, the same substitutions in the ATM 46-48T Δ context were capable of obtaining the full inhibition of 5′int usage (B, lane 6). Interestingly, and in keeping with the results obtained by the ATM 46-48T 41A Δ mutant (compare B, lane 6 with D, lane 6), the inactivation of the 5′int donor site also partially restored splicing to the 5′pe natural donor sequence. The inhibitory behaviour of the 21-23C substitutions in the 46-48T context with regards to 5′int usage are also clearly visible when the same mutants are cloned in pcDNA3 and transfected in Hep3B cells (C, compare lanes 2 and 3), although a marked reduction in splicing efficiency was also observed in this case. Splice-site mutation 1811 + 1.6kbA > G occurs in intron 11 of the gene and creates a novel 5′ splice site. This mutation results in the inclusion of a cryptic 49 bp exon in the final mature mRNA (). As shown in A, an interesting feature of this pseudoexon is represented by the fact that near to the newly created donor site (5′pe) there is a second, not optimal, but viable donor-site sequence (/guuacu, 5′ss) detected by the NNSplice prediction program. This was an interesting finding, as other likely donor-site sequences had already been identified in the original study (). Indeed, an interesting feature of this CFTR sequence is represented by its containing several downstrean donor-like sequences after the activating mutation (/guaa/guaag/guuacu). However, at the time no indication could be obtained that any of these additional donor-site sequences besides the one created by the A > G mutation were ever used in normal CFTR pre-mRNA splicing (). Using our PY7 system we have confirmed this data. In fact, as shown in B the CFTR WT sequence that lacks the A > G substitution does not give rise to any pseudoexon insertion event originating from the 5′ss natural sequence (lane 2). In addition, when the A > G substitution was inserted in the sequence (mutant CFTR MUT) only one pseudoexon inclusion event could be observed (B, lane 4) that originated from this newly created 5′pe donor site as determined by sequencing of the upper band (data not shown). As in the ATM pseudoexon example, analysis of this sequence (D) suggested that avoidance of 5′ss usage in normal splicing could be ascribed to a particular folding of this RNA. An alternative explanation for the lack of 5′ss usage may also reside in the peculiar arrangement of the CFTR sequence (see earlier) where tandem donor-like sequences (/guaag/guuacu) are closely grouped together, an event that may lead to splicing inhibition probably at the level of the A and B complexes due to sterical hindrance of closely binding U1snRNP molecules, as proposed by Nelson and Green in previous studies (). In this CFTR system, however, this possibility is unlikely as U1snRNP cannot be detected as binding to this position in the normal CFTR WT context, at least as determined by super-shift analysis using a anti-U1A monoclonal antibody (data not shown). Therefore, in order to confirm the structural hypothesis we first performed RNAse digestion analysis on the PY7 MUT transcript. As shown in C, the cleavages were consistent with the secondary structure depicted in D and in particular the inactive 5′ss donor site appears to be embedded in a tightly folded structure, as highlighted by the two strong V1 cleavages that immediately flank the two T1 cuts in correspondence to the GG nucleotides. Although this would seemingly make an energetically very unfavourable, 2 nt loop configuration, it should be noted that in order to cleave the V1 RNAse does not necessarily require a bonded interaction but only a sufficiently close distance between two nucleotides. On the other hand, the newly created 5′pe (D) is localized in the apparently more open region joining two stem-loop structures. In order to better understand the molecular mechanisms that repress 5′ss usage in the CFTR WT transcript it was then decided to test the importance of predicted donor-site strengths on CFTR WT pseudoexon inclusion efficiency. As shown in A, three mutants (WT2, WT3 and WT4) were created in which we substantially improved the predicted donor-site scores (0.98, 1.00 and 1.00, respectively) according to NNSplice without changing the structure and base-composition of the lower stem. The improvement in predicted strengths for these mutants was very evident using also other prediction programs such as MaxEnt scan that focus only on the 9 nt consensus stretch itself. In this case, the predicted 5′ss donor-site strengths according to the ME, MDD, MM and WMM scoring models were 1.25/6.68/2.34/3.92 for WT, 6.43/10.08/5.09/7.20 for WT2, 8.54/13.58/7.21/7.76 for WT3 and 9.66/13.68/9.96/11.04 for WT4. splicing analysis of these mutants (B, compare lanes 2, 4, 6 and 8) clearly showed that only WT3 and WT4 were capable of promoting pseudoexon inclusion from their mutated 5′ss sequences. This difference between the splicing behaviour of the WT2 and WT3 mutants (B, compare lanes 4 and 6) prompted us to focus on the importance of the +5 position. Therefore, we decided to improve the CFTR WT-U1snRNP matching region in this position by co-transfecting a modified U1snRNP molecule (C > G U1) that has been previously used to recover splicing from a +5G > C mutated donor site in exon 3 of the NF-1 gene (see C and D, lanes 1 and 2) (). The results of this experiment showed that CFTR WT could not display pseudoexon inclusion even when cotransfected with the C > G U1 (D, lanes 3 and 4) whilst only a very small amount of pseudoexon inclusion could be observed when the CFTR WT2 mutant was cotransfected with this C > G U1 (D, lane 6). Interestingly, addition of the C > G U1 to the CFTR WT 5′ss donor site gives a complementarity pattern that was identical to the one obtained by the CFTR WT3 mutant with wild-type U1 molecule (C, compare boxes 2 and 3). However, in the first case no pseudoexon inclusion was observed (D, lane 4) whilst the CFTR WT3 mutant could naturally promote inclusion (B, lane 6). This finding suggests that the critical importance of the +5 position might be unrelated to U1snRNP binding and the most likely possibility is represented by the interaction with U6snRNP, that would remain impaired in the CFTR WT+ C > G U1 system whilst it would be unimpeded in the CFTR WT3 mutant. This hypothesis cannot, however, explain the recovery of CFTR WT2 pseudoexon recognition (although at very low levels) in the presence of C > G U1. However, in this case it may well be found that the very extended complementarity between the CFTR WT2 sequence and the C > G U1 (C, lower right schematic diagram) may still promote donor-site usage. It should also be noted that the level of splicing recovery is considerably less than the one obtained for the NF-1 exon 3 donor site in the presence of C > G U1 (D, compare lanes 2 and 6) although there is an even greater complementarity between CFTR WT2 with the C > G U1 sequence (C, compare boxes 1 and 4). Taken together, all these differences suggest that the secondary structure may represent an additional constraint to the use of this splicing system. Indeed, disruption of the major structural features of this RNA region should affect the efficiency of CFTR pseudoexon inclusion. Thus, we began testing this by selectively deleting the supporting lower stem in either of the two strands (mutants CFTR Del1 and Del2, see A for a schematic diagram). In parallel, in order to test the importance of sequence conservation within the pseudoexon the structure was either ‘disrupted’ and then ‘repaired’ by engineering mutants CFTR Dis and CFTR Rep (see C for a schematic diagram). These mutants were then tested either or following transfection in Hep3B cells for ability to recognize the pseudoexon (B and D, respectively). For the first set of constructs, the results demonstrate that pseudoexon inclusion is reduced from 50% inclusion in the CFTR MUT construct to 22% in the Del1 mutant (B, left panel, lane 4). Even more strikingly, pseudoexon inclusion is almost completely abolished in the CFTR Del1 mutant when transfected in Hep3B cells (B, right panel, lane 2). It should also be noted that the Del1 deletion changes only slightly the predicted strength of the 3′splice site of the CFTR pseudoexon (from a score of 0.96 to 0.91 according to the NNSplice program), ruling out the possibility that this deletion may have simply affected the quality of this site. On the other hand, no effect on splicing efficiency with respect to the CFTR MUT could be observed for the Del2 mutant in transfected Hep3B cells (B, right panel, lane 3) whilst a small increase (to 68% inclusion) was observed (B, left panel, lane 6). For the second set of constructs it was observed that the pseudoexon in the CFTR Dis mutant was recognized with reduced effiency both and in Hep3B cells (D, left panel lane 4 and right panel lane 2). In fact, pseudoexon inclusion in the CFTR Dis mutant fell to 30% in the splicing assay (from a 50% starting value of CFTR MUT ) and to 64% inclusion in Hep3B cells (from a starting value of 93% inclusion for CFTR MUT in Hep3B cells). On the other hand, in keeping with the structural predictions, the CFTR Rep mutant fully recovered splicing efficiency with respect to the CFTR MUT construct (D, left panel lane 6 and right panel lane 3). These results were thus in keeping with the proposed regulatory role of RNA structure on CFTR pseudoexon inclusion. Furthermore, they demonstrated that conservation of the primary nucleotide sequences was not required. Nonetheless, on the basis of secondary structure considerations alone (A) we also would have expected the Del1 and Del2 mutants to have similar effects on splicing efficiency. It was therefore of interest to check whether the splicing changes detected could be correlated with unforeseeable differential structural rearrangements introduced by these two mutations. The results of the structural analysis performed on the Del1 mutant (B) showed that this RNA displayed a markedly different RNAse profile with respect to the CFTR MUT profile (A) and CFTR Del2 (C). In particular, with the exception of the 5′ss region, there is very little conservation in cleavage pattern profiles between the CFTR Del1 mutant and the CFTR MUT, CFTR Del2 mutants. As shown in the schematic diagrams below, these results are in keeping with the predicted Del1 structure (E) that includes the 5′pe, 5′ss, and 3′pe in a stem configuration, presumably making them less available to the splicing machinery. On the other hand, the cleavage pattern obtained for the Del2 mutant is very similar to that obtained for the MUT sequence (compare A and C), up to the conservation of several RT stops indicated by arrows. Finally, in support of the structural changes detected in the Del1/Del2 mutants it has to be noted that the cleavages in correspondence to the 5′ss position are mostly conserved in all three structures (A, B and C, 5′ss region), demonstrating that the differences in cleavage efficiencies for the different mutants do not represent an artifactual occurrence owing to different RNAse activities/efficiencies in the three analyses. Pre-mRNA secondary structure is increasingly recognized as a general modifier of splicing (). However, evaluating its influence on the processing of individual exons is often a very difficult task and any research on RNA structures within coding regions is complicated by the need to rule out additional sources of bias. These include conservation of coding potential (,), the vast array of positive and negative cis-acting sequences that are now known to be present in most coding sequences (,,), and that the RNA structure itself can be heavily influenced by RNA–protein interactions (). Indeed, to this date very few examples exist regarding the existence of splicing-regulatory secondary structures residing entirely within exonic coding regions (,). It is difficult to assess if this scant representation of secondary structure is a feature of exons or the fact that they have never been extensively searched for, particularly in view of the laborious task of confirming unreliable computer predictions in naked RNA and the almost impossibility up to now of probing structure . Hopefully, the recent increase in sequencing data from many different organisms coupled with ever more refined folding algorithms and new technologies for gene expression visualization will allow researchers to identify additional likely examples of regulatory structural elements within mRNA molecules (). It is also very possible that the often highly intricate network of regulatory processes embedded in exons/introns (,) and the need for the processed mRNA molecule to be translated, have all contributed to keep the 5′ss and 3′ss of real exons in regions of limited RNA structure. Many of these limitations, of course, do not apply to intronic or non-coding sequences. Indeed, conserved stem-loop regions within introns have been recently shown to play a role in the pre-mRNA splicing processes of human exon 7 (), human exon 2 (), in the highly conserved insect gene () and in the yeast gene (). In addition, the structure-forming ability of introns has been very useful to explain a whole range of splicing phenomena that would otherwise be difficult to explain if the pre-mRNA molecule was considered as existing in a predominantly linear form (,). In this work we have studied the importance of RNA secondary structure in pseudoexon inclusion events that are involved in human disease. These pseudoexons originate from the inclusion of apparently random intronic sequences following the introduction of an activating mutation, in general the creation of a novel acceptor or donor splice site although new observations also include modifications within splicing regulatory elements (see Supplementary Table 1 for a comprehensive list of these events). Our investigation of an event that concerns the deletion of a splicing suppressing element in the gene () or of a splice-site creation event in the gene () has shown that in both cases RNA secondary structure plays a major role on donor-site usage and splicing efficiency. Moreover, our results concerning the reasons that underlie the lack of use of an apparently viable 5′ss splice site in the CFTR genomic sequence has suggested that the RNA secondary structure may play a role of inhibiting post U1snRNP interactions, such as that with U6snRNP, by providing an additional inhibitory influence on the suboptimal +5 position. In this respect, therefore, our observations confirm previous suggestions that RNA secondary structures may have a general inhibitory effect on pseudoexon sequences (). In fact, the presence of potentially fully functional splice-site sequences within the intronic sequences of most genes clearly shows that at least some of these sequences would not really be expected to wait for an activating mutation in order to be partially included in the final mRNA molecule. However, if these splice-site sequences find themselves bracketing sequences containing binding sites for splicing inhibitor molecules or, like in our cases, embedded into unfavourable RNA structures, then an explanation can be offered regarding why additional requirements have to be satisfied in order to obtain their ‘upgrade’ to exon status (for example, their presence in an ‘open’ structural configuration). The difficulty in evaluating these events on the basis of the primary sequence alone can also be appreciated from the performance of 5′ss scoring programs used in this work. In fact, beside the MaxEnt Scan evaluation methods that consider the MAG|GURAGU (M is A or C; R is purine) consensus sequence even methods such as NNsplice that consider in their evaluation sequences beyond the strict consensus fail to accurately predict the exact donor-site usage observed in our experimental systems. Our results will hopefully provide further information to facilitate the fine tuning of these splice-site prediction algorithms. It should be noted, however, that RNA secondary structure may not represent the only determining factor in the ATM or CFTR pseudoexons, as highlighted by some of our experiments (i.e. C) where the structure-affecting mutations also cause a drop in overall splicing efficiency of the pseudoexon. These results raise the possibility that splicing controlling elements may also be affected by these changes. In the future, it will be interesting to study in selected examples the potential interplay between RNA structural features and potential enhancer and silencer splicing controlling regions present in the pseudoexon of the type already demonstrated in the fibronectin EDA exon (). Indeed, the role played by splicing factors in pseudoexon inclusion is a promising issue that remains to be fully investigated in the splicing field. In fact, in at least a minority of cases pseudoexon inclusion has been strongly correlated with creation of splicing regulatory enhancer elements (). A similar situation may be present for nucleotide variations that remove the binding site for splicing inhibitory factors (). Nonetheless, our work suggests that taking into account structural features may represent a useful strategy to understand the functional mechanisms that underlie the activation of pseudoexon sequences in other genes causing human genetic diseases. This may be important also for attempting novel molecular therapy approaches to modify the regulation of pseudoexon inclusion events and for refining computer algorithms for the prediction of the splicing effect of any genomic variation. In fact, from a molecular therapy point of view the mapping of pseudoexon RNA secondary structures may provide useful indications regarding the development of inhibitory molecules such as antisense nucleotide that have already been suggested to possess increased working efficiency when targeted against open structural regions (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The BlaI protein (BLBlaI, 128 amino acids) is a transcriptional repressor of the BlaP β-lactamase. This enzyme is a specific hydrolase of β-lactam antibiotics, induced in response to the presence of this class of antibiotic outside the cell (). BLBlaI is homologous to BlaI (SABlaI, 126 amino acids) and MecI (SAMecI, 123 amino acids) regulators involved in the induction of BlaZ β-lactamase and resistant penicillin-binding protein 2a (SAPBP2a or SAMecA), respectively (). BlaZ β-lactamase and resistant PBP2a are the main factors involved in staphylococcal β-lactam antibiotic resistance. The BlaI/MecI repressors are organized in two domains, an N-terminal domain (NTD) for DNA binding and a C-terminal domain (CTD) for repressor dimerization (). The crystallographic 3D structures of SABlaI and SAMecI dimers in free and in complex with their DNA operators have been determined () and the 3D structure of BLBlaI N-terminal DNA-binding domain has been determined by heteronuclear nuclear magnetic resonance (NMR) spectroscopy (). For the three repressors, the BlaI/MecI-NTD share a common fold composed of three α-helices and three β-strands typical of the winged helix regulator proteins. The SABlaI/SAMecI structures highlight dimers of two independent N-terminal DNA-binding domains and two intertwined C-terminal dimerization domains (A). The BlaI/MecI repressors bind specifically to similar nucleic sequences composed of an imperfect dyad symmetry (24 to 30 bp) containing a central conserved palindrome: 5′-TACANNTGTA-3′ () (nucleotide one-letter code; N for any nucleotide). The study of BLBlaI dimerization and its interaction with its operator (BLOP) has shown that, at a concentration below the dissociation constant of BLBlaI dimer ([BLBlaI] <25 µM), the binding of one BLBlaI monomer to its operator leads to the binding of the second monomer with an infinite cooperativity (). In this way, it has proved impossible to isolate a BLBlaI monomer bound to its DNA operator (B). In addition, for the same repressor, it has been shown that the BLBlaI-NTD obtained by papain proteolysis retains its capacity to bind BLOP. However, its affinity for its DNA target becomes at least 500 to 1000 times lower as determined by DNAseI footprinting experiments (). In and , the genes encoding for β-lactam resistance ( and ) form a divergon with the operons, respectively (,). encodes a penicillin receptor essential for the induction of the gene of resistance. The / operators are located in the intergenic DNA sequence between the gene and the operon. In presence of β-lactam antibiotics, the BlaR1/MecR1 receptor is acylated. The resulting activated receptor launches a cytoplasmic signal which inactivates BlaI/MecI repressor. In , the SABlaI inactivation is achieved by the proteolysis of the peptide bond linking residues 101 and 102, giving rise to SABlaI-NTD dissociated from SABlaI-CTD (). The truncated SABlaI-NTD is a monomer, presents a lower affinity for its DNA operator and is released into the cytoplasm. For BLBlaI repressor, the presence of a coactivator () generated by the activation of the BlaR1 repressor has been postulated. The binding of the coactivator to BLBlaI would then result in a decreased affinity of BLBlaI repressor for its DNA target (). To better understand the mechanisms of the binding of the BlaI/MecI repressors on the DNA, and their inactivation during the induction of BlaP/BlaZ/MecA proteins, we solved the solution structure of the low affinity BLBlaI-NTD/1/2OP complex ([BLBlaI-NTD]/[1/2OP]). The dissociation constants of the SAMecI in complex with the operator OP of (OP) and in complex with the operators of (OP) and (OP) have been determined. NMR spectroscopy allowed us to estimate the dissociation constants of the SAMecI-NTD and BLBlaI-NTD for the semi-operating sequence of the gene OP (1/2OP) and for the semi-operating sequence of the gene (1/2OP). The pET22b was used as vector for the overproduction of the His-tagged SAMecI protein (SAMecI-His6). The SAMecI coding sequence was amplified by PCR from the ATCC 43300 genomic DNA using Taq polymerase (Promega) and the following oligonucleotides as primers: 5′-GAG-CAT-ATG-GAT-AAT-AAA-ACG-TAT-GAA-ATA-TCA-TC-3′ and 5′-CTC-GAG-TTT-ATT-CAA-TAT-ATT-TCT-CAA-TTC-TTC-TA-3′ purchased from EUROGENTEC, S.A., Belgium (). The fragment generated corresponds to the SAMecI coding sequence within the restriction sites for I and I and was cloned into the pCR4 TOPO (Invitrogen) vector to generate the pCR4-. The identity of the sequence was verified before the pCR4- was digested with I and I enzymes and cloned into the pET22b to generate the pCIP451 which contains the MecI coding sequence with a polyhistidine tag at its carboxy-terminal end. SAMecI-His6 samples were overexpressed in strain BL21(DE3). For the production of uniformly N-labeled samples, Luria-Bertani medium was replaced by a M9 minimal medium supplemented with 1.1 g/l NHCl, 2 mM MgSO, 0.1 mM CaCl and 2 g/l glucose. Cells were grown at 37°C to a 600 nm absorbance of 0.6 and 0.5 mM IPTG was added for a 3-h induction period. Cells were then harvested by centrifugation, re-suspended in buffer A (20 mM NaHPO/NaHPO buffer, 500 mM NaCl, pH 7.6) and disrupted by passage through an Inceltech disintegrator. The soluble fraction was separated by centrifugation at 40 000  and loaded onto a NiPDC chelating column (2.6 × 10 cm, Affliland) charged with 50 mM NiSO and equilibrated with buffer A. The SAMecI-His6 protein was eluted by a gradient of buffer B (250 mM imidazole, 500 mM NaCl, pH 8). The protein was dialyzed against buffer A and concentrated. The final yields of labeled and unlabeled purified protein were respectively 9 and 15 mg/l of cell culture. The isotopic labeling of 95% was determined by mass spectrometry. The uniformly C/N-labeled BLBlaI sample was prepared as described in Van Melckebeke . (). The SAMecI-NTD protein was dialyzed into a 75 mM NaHPO/NaHPO buffer, 200 mM KCl, 1 mM EDTA, 1 mM NaN, pH 7.6 and was concentrated to 0.5 mM by ultrafiltration through a 5 kDa cut-off Amicon for further NMR analysis. Unlabeled single-stranded DNA samples of NMR quality were chemically synthesized and purified by EUROGENTEC, S.A., Belgium (). Freeze-dried samples were suspended in the buffers used for interaction studies. Single-stranded DNA were mixed in a 1:1 ratio, subsequently heated to 100°C and slowly cooled down at room temperature, in order to improve intermolecular arrangements. The semi-operator sequence [5′-ATA-AGA-CTA-CAT-3′ and complementary strand 5′-ATG-TAG-TCT-TAT-3′] was designed according to previous papers results () and obtained at a 9 mM final concentration in 75 mM NaHPO/NaHPO buffer, 200 mM KCl, 1 mM EDTA, 1 mM NaN, pH 7.6. The OP half-dyad [5′-AAA-GTA-TTA-CAT-3′ and 5′-ATG-TAA-TAC-TTT-3′] was selected using former interaction results with BLBlaI-NTD and obtained at a 23 mM final concentration. NMR experiments were performed on Varian Inova 600 and Inova 800 spectrometers, both equipped with a triple-resonance (H, C, N) probe and shielded -gradients. Furthermore, the Varian Inova 800 MHz spectrometer is equipped with a cooled probe. The temperature was set to 298 K. Proton chemical shifts were referenced with respect to an external DSS calibration. C and N chemical shifts were accordingly referenced indirectly using the H/X following ratios: 0.251449530 (C) and 0.101329118 (N). All experiments used the pulse sequences provided by the Varian Protein Pack (). Data processing and peak intensity measurements were performed using the NMRPipe program. Peak picking and spectra display were achieved using the NMRView software. Electrophoretic mobility shift assays were carried out using an ALFexpress DNA sequencer as described in the literature (). The CY5-labeled fluorescent double-stranded oligonucleotides used in these experiments are listed in (,). For NMR interaction studies, H-N HSQC spectra were collected along the titration of SAMecI and the BLBlaI truncated repressors with the 12 bp DNA. To limit dilution and favor sensitivity, low volumes of highly concentrated half-operators of and genes were added to each protein sample. N-labeled SAMecI-NTD and BLBlaI-NTD samples concentrations were set to 0.1 mM. Titration experiments led on N-labeled MecI-NTD (respectively N-labeled BLBlaI-NTD) with both unlabeled and semi-operating sequences were performed in a 75 mM (respectively 50 mM) NaHPO/NaHPO buffer with 200 mM KCl, 1 mM EDTA, 1 mM NaN and 10% of DO at a controlled pH of 7.6. Data analysis and Kd calculation were performed with the titration script developed for the NMRView software (). In each case, curves fitting were displayed by the Xmgrace software (). H, C and N assignment of the free BLBlaI-NTD protein has been previously reported and deposited in the BMRB (accession number 5873). Resonance assignment of the unlabeled 12 bp DNA of the OP semi-operating sequence was performed on a 1 mM sample in 50 mM of NaHPO/NaHPO, 200 mM KCl, 1 mM EDTA, 1 mM NaN, pH 7.6 in 90%:10% HO:DO. A TOCSY spectrum with 80 ms mixing time and a NOESY spectrum with 150 ms mixing time were recorded on that sample. H-H NOESY was also collected in 100% DO with 150 ms mixing time. A classical homonuclear DNA assignment strategy was used (). Due to the large number of overlap in H′ and H′/H′′ regions, assignment and chemical-shift mapping were restricted to H′, H′, H′, H and H/H resonances. For the 12 bp 1/2OP assignment in the complex, we reiterated the procedure used for the free DNA but using only a 2D filtered NOESY experiment recorded on the [C-N BLBlaI-NTD]/[1/2OP] sample diluted in 100% HO. The experiment was recorded on the 800 MHz spectrometer with a mixing time of 150 ms. For the [BLBlaI-NTD]/[1/2OP] complex, the H, C and N protein resonances were not re-assigned . Assignment of the protein in complex was obtained by comparison of H-C HSQC and a H-N HSQC of the free and bound forms. Experiments were recorded with 2 mM samples and a [C-N BLBlaI-NTD]/[1/2OP] molar ratio of 1 in 100% DO and in 90% HO. To confirm this assignment, a 3D HC(C)H-TOCSY experiment was also recorded on the complex. The weak dependence of C chemical shifts to long distances variation and then to the complex formation has permitted to verify each corresponding residue assignment. In order to obtain inter-molecular NOE restraints between the two partners of the [BLBlaI-NTD]/[1/2OP] complex, we collected isotopically doubly filtered 2D and 3D C/N NOESY HSQC with mixing times set to 150 ms. A protein chemical-shift mapping was obtained by comparison of H-N HSQC, methyl-selective C HSQC and C HSQC optimized for aromatics recorded on the free and bound forms of the protein. A chemical-shift mapping for the half-operator was obtained by comparison of the H-H NOESY recorded on the free DNA and the 2D filtered NOESY experiment recorded on the 1/1 [C-N BLBlaI-NTD]/[1/2OP] sample. Determination of the structure of molecular complexes from sparse NMR data is a difficult task, requiring the integration of local and long-range molecular plasticity (). In order to allow the maximum available degrees of freedom we have determined the quaternary architecture of the BlaI repressor/DNA operator complex using a approach starting from randomized coordinates. Experimental NOE collected on the N-C BLBlaI-NTD free protein were used to fold the bound polypeptide (). The use of these constraints is based on the observation that few chemical shifts change between the bound and free forms, showing that the fold of the two proteins is essentially the same. The 12 bp operator was constrained using distance restraints extracted from a canonical double-strand DNA. B-DNA standard angles (α, β, δ, γ, ε, ζ, ψ, υ0, υ1, υ2) were also included for each nucleotide to facilitate the DNA helix fitting. Distances restraints extracted from intermolecular NOE were defined at 5 Å (one DNA proton correlated with one protein resonance). A thirdly set of structural data was introduced using chemical-shift mapping performed on the two molecules. Chemical-shift-derived distance restraints were created by combining ‘significant’ chemical-shifts variations identified on the polynucleotide H-H NOESY with shifted residues in N-HSQC and C-methyl-selective-HSQC spectra, and including these constraints in an ambiguous manner (,). Distances inferred from these distance restraints were fixed at 10 Å in the docking simulation. All calculations were performed using the program Discover with the AMBER4 force field (). Simulated annealing was used to explore the conformational space for the structure determination and a restrained molecular dynamics calculation was used to refine each structure (). Detailed initial conditions and physical characteristics of exploratory period have been reported previously (). The binding curves of SAMecI to its operator, OP, and to the two β-lactamase operators, OP and OP, have been determined by band-shift assay. All show sigmoidal-binding curves as previously described for the interaction of BLBlaI repressor with its operators (). So, as for BLBlaI, the binding parameters of SAMecI interaction include two equilibria and only the global dissociation constant Kd = Kd. Kd can be obtained, where Kd and Kd are the dissociation constants of SAMecI dimer and SAMecI dimer-operator complex, respectively (). Titration of N-labeled BLBlaI or SAMecI truncated repressor with progressive amounts of unlabeled DNA half-dyads was performed by monitoring changes in H-N HSQC spectra. Significant chemical-shift changes for correlation peaks in the H-N HSQC spectra were observed when the 1/2OP and the 1/2OP were added to protein samples. [DNA]/[Protein] molar ratios were varied from 0 to 10 for [BLBlaI-NTD]/[1/2OP], [BLBlaI-NTD]/[1/2OP], [SAMecI-NTD]/[1/2OP] and from 0 to 50 for [SAMecI-NTD]/[1/2OP]. The chemical-shift changes observed upon complexes formation reached a plateau over 7 [DNA]/[protein] molar ratios for [BLBlaI-NTD]/[1/2OP], [BLBlaI-NTD]/[1/2OP], [SAMecI-NTD]/[1/2OP] and a plateau over 30 for [SAMecI-NTD]/[1/2OP] (). These results confirm the formation of stable intermolecular interactions with saturating DNA quantities. At any point during the titration, specific single correlation peaks were detected, suggesting that the truncated monomeric repressors/semi-operators complexes are in fast exchange regarding the NMR time scale. Only well-resolved correlation peaks with a chemical-shift cut-off equal or superior to 0.03 ppm were used in the four experiments. The affinity constants were calculated from a non-linear fit of the significant chemical-shift variations versus [DNA]/[protein] ratio using equation given by Morton . (). Titration data were analyzed assuming that the observed chemical-shift perturbation is a weighted average between the two extreme values corresponding to the free (Δδ = 0) and the bound state (Δδ = Δδ) so that: Statistical analysis using Monte-Carlo simulations were used to evaluate the uncertainty of the fitted parameters. Semi-operating sequences of the and genes titration curves of selected individual amino acid residues resulted in averaged binding constants of 190 ± 50 µM (Kd), 170 ± 50 µM (Kd) and 160 ± 60 µM for the [BLBlaI-NTD]/[1/2OP], the [BLBlaI-NTD]/[1/2OP] and the [SAMecI-NTD]/[1/2OP] complexes, respectively (). For [SAMecI-NTD]/[1/2OP], the measured affinity constant reaches 860 ± 80 µM (Kd). Considering the low affinity constant measured previously, we decided to investigate the solution structure of the [BLBlaI-NTD]/[1/2OP] complex using a sparse data approach. Structure prediction using a docking approach remains difficult because of the number and the variety of parameters that should be taken into account. Recent advances in the field () take advantage of structural restraints from experimental interaction data (biochemical and/or biophysical) to determine the relative position of the molecular partners. Such approaches have also been applied to the determination of the quaternary structure of protein/DNA complexes (). Our docking protocol has been performed as follows. To investigate the structure of the complex between the C/N-labeled BLBlaI protein with the unlabeled semi-operating sequence, we collected two filtered NOESY experiments. The 2D isotopically filtered NOESY performed in HO allowed us to observe three NOE. To verify these constraints, we collected an additional 3D NOESY C HSQC experiment in HO, resulting in one additional inter-molecular contact. These four correlations have been assigned ambiguously with a tolerance value of 0.05 ppm for the protein and 0.03 ppm for the DNA. To complement the four NOE distance constraints, a chemical-shift mapping has been carried out for both macromolecular partners as follows. For the protein, we compared H and N (backbone amide), and H and C methyl chemical shifts assigned for the free and bound form of the molecules, measured in H-N and H-C methyl-selective and aromatic-selective HSQC, respectively. Indeed, methyl group and aromatic chemical-shifts perturbations were judged to be good additional probes to precisely localize the interaction site. Concerning the nucleic acids, only the sugar H′, H′/H′′, H′ and the base H/H protons were used for chemical-shift mapping due to the overlap problems observed in the other spectral regions. The chemical-shifts variations resulting from complex formation allowed us to make an inventory of each individual amino acid or nucleotide potentially involved in the interaction. We only considered chemical variations larger than 0.8 ppm for the weighted sum of the H and C, or H and N chemical-shift variations of the protein with respect to the gyro-magnetic ratios, and larger than 0.09 ppm for the proton chemical-shift variation measured for the half operator. In both cases, peaks that disappeared were incorporated in the docking procedure. For the DNA operator, the majority of perturbations concerned nucleotides in the vicinity of the TACA/ATGT motif namely the thymines 5, 7, 8 and 6*, 9*, 11* on the complementary strand, the adenines 6, 9, 7*, 8* and 12* and the guanines 4 and 10* (nucleotides nomenclature is described in legend). For the protein, larger perturbations are observed for the N-terminal part, the H2 and H3 helix and around the wing. The structure calculation started from random coordinates of the entire system, and used 1513 experimental intra-molecular NOE of the free form BLBlaI-NTD protein and simulated distance restraints for a standard B-helix (613 intra-residue distances and 52 inter-residue distances). Classical dihedral angles for DNA subunits were also included, namely, 22 α, 24 β, 23 γ, 23 δ, 22 ε, 22 ζ, 24 χ, 24 υ2, 24 υ1 and 24 υ0. The use of experimental NOE from the free form of the molecule is based on the lack of drastic shift between H-N HSQC of the unbound and the bound protein, indicating that no significant structural rearrangement occurs. NMR dihedral angles were also calculated using TALOS software and incorporated in the calculation. Four intermolecular NOE were associated with 11 and 24 additional ambiguous distance restraints proceeding from C methyl-selective HSQC experiments and from N HSQC spectra, respectively. Ten lowest energy structures from 250 calculations of the [N-C BLBlaI-NTD]/[1/2OP] have been generated using a driven docking (). As expected, the global fold of the NTD repressor domain is not modified by the complex formation. Pairwise RMSD calculated on the NTD backbone atoms between the monomeric complexed BLBlaI structure and free BLBlaI-NTD, SAMecI/OP and SABlaI/OP is 1, 1.6 and 1.5 Å, respectively. The typical structural arrangements of the WHP family protein is conserved without violations in the structure calculation file. The three α-helices H1 (), H2 (), H3 () and the three stranded β-sheets S1 (), S2 (57–62), S3 (65–70) are packed following the sequence H1–S1–H2–H3–W1–S2–S3. The wing motif W1 consists of a short loop (residues 63 and 64). Three-dimensional structures of the free and bound forms of the BLBlaI-NTD are very close except for the more dynamic residues located in the N- and C-terminals extremities of the protein, namely M1 to I4 and Y77 to S82, respectively. Moreover, it should be noticed that restricted conformational modification occurred for the residues of the Wing (G63 and R64) and for few residues surrounding this motif (E62, V65 and F66). In the complex, the B-DNA form is conserved in the final lowest energy structures. The position of the bases is less well-defined at both ends of the molecule than in the central part, possibly due to a lack of conformational restraints. Concaveness observed for the complete operator for SABlaI and SAMecI is not observed for the half operator (). However, alignment of our 12 bp operator with the common base of the full MecI operator shows a very similar conformation (B). The absence of a detectable kink could be due to the reduced length of the half-operator used in the NMR study. To analyze the orientation of the protein relative to the DNA, we have aligned our structure with the two X-ray structures of SABlaI and SAMecI using a superposition of the conserved DNA sequence. Compared to the two crystallographic structures, the relative orientation of the monomeric protein with respect to the DNA helix shows a 30° rotation along the long axis of the DNA and a translation of 3 Å (B). To compare the stability of the different protein/DNA complexes, analysis of the protein/DNA interaction using LIGPLOT software () is presented in . Details of the nominative protein/DNA contacts observed in our structure are presented in supplementary Figure S1. Two helices (H2 and H3), the wing motif and the N-terminal domain of the BLBlaI-NTD have been reported to establish contacts with DNA. In our structure, the H3 helix (P41-K53) is deeply inserted in the DNA major groove, whereas the minor groove is close to the wing motif (G63-R64). For our NMR-based model, residue R64 of the wing motif (G63-R64) binds to the 3′ end of thymines 3* and 2*. Moreover, interaction is occasionally propagated for residues surrounding the wing motif i.e. F66 and H61 which make contacts with A6, T7 and C4*, T3* nucleotides, respectively. In the MecI/BlaI X-ray structures, the wing amino-acid F67 binds the DNA backbone in the opposite side to that observed in the NMR model (A7C and T8C equivalent to position G4 and T5 in our DNA sequence). As a consequence of the global rotation, contacts of the wing with the DNA are also rotated with respect to the ‘crystal wing’ position in SAMecI/OP. The H3 helix residues (P41-K53) contact principally with nucleotides belonging to the TACA/T*GTA* motif. T43, T46 and R50 privilege interactions with nucleotides T9*, G10* and T11*, respectively. Furthermore, K54 (and W39) anchors the position of H3 via interaction with A12* (and G10*). In this low affinity complex, Q45 plays a central role, forming hydrogen bonds network with nucleotides T7 and T8 instead of an interaction with conserved A6-T6* bases as observed for the SAMecI. In the past it has been assumed that in β-lactam resistance regulation system, 749/I BlaI-WT interacts as a preformed dimer with its operator (). The binding constants of the full-length SAMecI and BLBlaI repressors in interaction with palindromic operating sequences have been measured in the range of tens of nanomolar (). Using chemical-shift mapping, we have determined affinity constants for the BLBlaI and SAMecI NTD with the different half DNA operators. For all the complexes, [SAMecI-NTD]/[1/2OP], [SAMecI-NTD]/[1/2OP], [BLBlaI-NTD]/[1/2OP] and [BLBlaI-NTD]/[1/2OP] binding constants are in the same range of hundreds of micromolar, that is, 100 times higher than for the dimeric repressors. The relatively low dimerization constant of 25 μM () observed for BLBlaI has suggested that both monomer and dimer pathways were possible. Now, taking into account our quantitative values of monomer-DNA dissociation constants and concentrations of BLBlaI, estimated to 2 µM (), we can conclude that the monomer pathway contributes significantly to the BlaI repressor binding mechanisms . In various Winged HTH dimeric repressor systems, a similar monomer–dimer equilibrium has been established (,). Two repression systems comparable to BlaI are known: LexA and Rep. The DNA binding constant of LexA-NTD () and Rep-NTD () are respectively about 1 and 20 µM, close to values determined for SAMecI-NTD and BLBlaI-NTD. Moreover, initial assays measuring the dimerization constant of the LexA repressor (,) reported a Kd of about 10–50 µM. The association constant of the full-length Rep repressor has been evaluated to 3–5 nM as well. When Rep monomers bind specific DNA sequences, the CTD is moved away, and stable repressor-operator contacts can be established. If the specificity is not high enough, the tail competes with DNA and prevents DNA/protein non-specific associations. The monomer pathway could provide a method for rapid localization of the binding site involving sliding of the protein along the DNA. Theoretical considerations and experimental evidence of protein sliding along non-specific DNA sequences are now well documented (). This kind of search could combine 1D sliding with 3D diffusion () driven by energetic differences for specific and non-specific DNA-binding. In our system, the localization speed of the correct DNA-binding site depends on the competition of the monomer pathway with the dimer pathway. Although strong cooperative effects in monomer association with DNA do not allow us to experimentally evaluate this process, we note that such a mechanism could play a role in this system. Monomers would be able to reach the adapted DNA motif using such diffusive processes, and this pre-recognition step would allow a simpler contact between the two CTD. This intermediate step might be useful for the protein to establish correct and strong contacts. The presence of low affinity constants for the DNA binding of the monomeric form can be correlated with the adaptability of the repressors for different operators. For example, the dimeric LexA repressor tightly binds the different and DNA sequences. In the case of BlaI, the co-repression mechanism of homologous repressors in has been demonstrated (). The Kd values measured in this work demonstrate that the BLBlaI-NTD protein is able to bind to both cognate and crossed semi-operators in the same affinity range. Thus, it suggests that low affinity complex formation might be an essential intermediate relay in binding variable regulation sequences. Contrary to BLBlaI-NTD with half-operator, we have shown that the affinity for semi-operating sequence is 10 times reduced for the SAMecI-NTD repressor. This parameter might corroborate observations performed on clinical isolates (,). Indeed, it has been shown that the majority of oxacillin-resistant strains contain deleted or mutated genes while sequences remain intact. Thus, in clinical isolates, the exclusive expression of MecI repressor appears to be a drawback in stress condition. Indeed, in this case, bacteria are unable to respond rapidly. The BlaI/BlaR1 system associated with the MecI/MecR1 machinery may enhance antibiotic resistance by improving capability to respond. Affinity of the monomer could reflect the capacity to form some intermediate states required during the induction process and the BlaI-NTD might play a crucial role thanks to its inter-operator adaptability. In the monomer pathway mechanism, a first monomer binds to the DNA, exhibiting the monomer-DNA conformation described in this work. This structure is indeed stabilized by a large number of protein–DNA interactions that are systematically present in the dimer–DNA interface (). Indeed, the comparison between the position of the monomeric and dimeric repressor on the DNA gives a 30° rotation. In this conformational change, the overall position of the protein on the DNA remains the same, that is the helix H3 stays in the main groove, but it shifts from one base (). In this model, the energetically favorable monomer–DNA interface is destabilized upon the binding of another monomer molecule onto the DNA–monomer complex. However, it is easy to interpret the increase of the global affinity by considering the gain in enthalpy and entropy due to the presence of two DNA–monomer interfaces and an additional dimerization interface. We note that some energy is stored in the two monomer–DNA interfaces when a dimer–DNA complex is formed, that can be potentially released if the dimerization domain is destabilized. The values of the monomer–DNA affinities measured in this work (hundreds of micromolar) and the concentrations of repressors measured in the cell (2 µM) () show that monomers are not able to repress the genes. In the presence of β-lactam antibiotics, the induction process modifies the affinity of the protein repressor for its cognate DNA sequence. The lower affinity measured for the monomer to its DNA-binding domain could then be sufficient to explain the derepression of the gene. However, intermediate steps driving the release mechanism are not clearly understood. Biochemical investigations performed on the // β-lactamase regulation system () proposed that proteolysis drives the signal transduction. Indeed, in , the BlaR1 penicillin receptor is a membrane protein containing a zinc metalloprotease motif. Acylation of the sensor-transducer via penicillin binding triggers the signaling mechanism leading to genes transcription. This event has been proposed to be the result of the cleavage of the dimeric BlaI repressor between residues N101 and F102. Two induction models have been proposed for the operator transcription under β-lactams stress conditions (,). The first takes into account the fact that a fourth gene, as in strains, encodes a key protein necessary for the CTD accessibility improvement (,,). The second considers that an inductor produced in the cytoplasm in stress conditions can modulate the quaternary blocked conformation of the regulations elements. In this way, the inductor would be able to act as a proteolysis enhancer like in the TetR system (). To sum up, the lower affinity of the repressor during the induction process could be the result of either a proteolysis of the repressor, leading to monomerization, or a structural change of the dimer that would place the two monomers in an unfavorable conformation, leading to a dimer–DNA dissociation constant almost equal to a monomer–DNA dissociation constant. In both crystal structures of the SAMecI and SABlaI in interaction with the and DNA sequence (,), the cleavage site is not easily accessible, suggesting that conformational changes may be necessary. Furthermore, it has been observed that two forms, open and closed, could co-exist, allowing structural adaptation between and operators with different inter-N-terminal domain distances (,). This structural flexibility may be permitted by the high dynamics observed for the dimerization part of the repressors (). Here, considering the structural differences between the monomeric and dimeric interactions of BlaI with DNA, we propose a model where the dimer placed on DNA plays the role of a tense spring that can be released by a modification of the dimerization domains of the repressors. This tense spring is composed of a translational component similar to the one observed between the open and closed form () and a supplementary rotational contribution of 30°. During this release, the monomer repressor would slide 30° inside the major groove of the DNA until it reaches its equilibrium conformation (as described in this article). C displays the relative position obtained for the two CTDs in case of a complete rigid structure and illustrates an incompatibility between the twist of the two NTDs and conservation of an intact dimerization domain. This twist displacement could in practice modify the accessibility of the cleavage site by involving a modification of the secondary structure arrangement in the core of CTD and could result in breakage of covalent bonds in this region. This structural modification would be responsible for an irreversible separation of the two helices of the dimerization domain, substantiating the possible existence of an inactivation-state DNA-binding domain conformation. In the MexR repressor study (), two conformations have been observed as well. On one hand, the open state exhibits optimal DNA-binding domains interspaces, which privileges the association with the operator. On the other hand, a modification in between the DNA-binding domains avoids repressor stowage on DNA. Furthermore, this sliding mechanism might be an explanation for a potential repressor ability to adjust contact between the closed and open form. Recently published results () underline the fact that mobility in the tertiary arrangement might be a useful requirement for thin regulation processes, even in eukaryotic systems. Indeed, the homeodomain of the transcription factor Pdx1 is able to bind a 15-bp DNA promoter (with a consensus binding site) and adopt two slightly different conformations in the same asymmetric unit. Both complexes differ by a 2.4° rotation. Additional modifications such as DNA curvature and N-terminal local adjustments have been also raised by the author. Similarly, major structural rearrangements controlled by small molecules have been highlighted for other systems as hormone receptors (). For the FadR transcriptional regulator (), interaction with DNA is disrupted by the acyl-CoA binding. Dramatic structural rearrangements happen which leads the DNA recognition helices being separated by 7.2 Å. Another repressor inactivation process has been published concerning bacterial resistance to antibiotic. Indeed, the TetR repressor is released in presence of tetracycline. This molecule associated with Mg plays the role of the inducer. Inducer binding generates structural changes in the CTD. Thus, a pendulum-like motion increases the separation of the attached DNA-binding domains () by 3 Å. These models give strong support to the hypothesis that an inducer or a product participates to the initiation of C-terminal remodeling and propagation to the N-terminal parts (). In presence of the effector, specific and optimized contacts between the N-terminal part and the DNA could be re-established by increasing the interspacing between the TACA/TGTA motifs. The differences in the length of the interspaces between the recognition motifs could limit the DNA-binding domains adaptation quality of SAMecI-WT to operator. o r d i n a t e s f o r t h e m o d e l o f t h e c o m p l e x h a v e b e e n d e p o s i t e d i n t h e p r o t e i n D a t a B a n k ( P D B , I D c o d e 2 P 7 C ) . p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Antisense technology provides a simple and elegant means to regulate gene expression, either for therapeutic purposes or for studying gene function (). With one FDA approved drug and approximately 20 candidates in various stages of clinical trials, there is growing promise for the success of this approach (,). Additionally, the excitement generated by the advent of siRNA has rejuvenated interest in gene silencing technologies in general (). Advances in oligonucleotide (ON) synthesis have now made it relatively simple to modify the chemistry of these molecules. This has enabled the creation of ONs with diverse properties and has thus greatly widened their overall utilization as specific mediators of gene silencing. With the growing progress in functional genomics, there is an increasing need for the routine application of gene silencing tools such as antisense in basic research. Despite the conceptual simplicity, utilization of antisense as a routine tool for studies is greatly impaired by the poor cellular delivery of these molecules. Delivery of ONs in a stable form and relevant dose to the appropriate target site of action remains a considerable challenge to date (,). The reduced cellular entry and rapid degradation of these molecules in the presence of cellular nucleases calls for the utilization of carrier molecules. Various types of carriers, including polymers, lipids and peptides () of diverse chemistries, have been tested for their effectiveness as DNA delivery vectors. However, substantial optimization is generally required to make these carriers work for the particular application at hand. The large variations in vector effectiveness among cell types, as well as the relatively high cytotoxicity of the currently available carriers, continue to fuel the demand for more rationally designed carrier systems (). To improve further the design of DNA carriers, extensive research is being conducted to identify cellular barriers to carrier mediated delivery of ONs (). Efforts are directed towards determining structure–property relationships that relate carrier properties to antisense effectiveness. However, it remains somewhat unclear what factors ultimately dictate the effects observed. Of particular importance is the fact that much of the design principles developed for plasmid DNA carriers are often used interchangeably for antisense ONs or siRNA. Although this may work in certain cases, the small size of ONs (10–22 bases) and variations in their backbone chemistry and structure endow them with unique properties that alter their interactions with carriers in comparison to plasmid DNA, and hence influence the design criteria for ON carriers. There are several examples in literature that point to such differences in terms of ON structure, ON chemistry and sequence composition (). Hence, there is a need for systematic investigations of the interactions of carriers with ONs, both at the molecular level as well as at the cellular level, to acquire a mechanistic understanding of their cellular processing. The goal of the present study is to understand the role of the vector, i.e. the properties of both the carrier and the ON, on the extent and dynamics of the antisense effect. For our study, we chose the cationic polymer, polyethyleneimine (PEI), as the model polymer. Our rationale for doing so is 2-fold. First, PEI is a well characterized carrier molecule (). It has been utilized extensively for the delivery of plasmid DNA due to its high charge density and endosomolytic activity. Several reviews are available detailing the effectiveness and mechanism of action of PEI and its various modifications for the delivery of DNA to a range of cell types (). There are fewer studies focusing specifically on the delivery of ONs using PEI. Second, we wish to determine the optimal PEI molecular weight (MW) for delivery of ONs of various chemistries. Specifically, we utilized five MWs of the branched form of the polymer, in combination with phosphodiester (PO) and phosphorothioate (PS) ON backbone chemistries. While a range of alternative chemistries is now available, PO and PS ONs are still utilized most often for work. We studied the polymer–ON interactions and further evaluated their efficacy in delivering active antisense ONs to cells. We demonstrate that the observed antisense effect is not determined solely by the carrier properties, but by the particular combination of polymer and ON properties. Of particular importance is the strength of interactions between the carrier and the ON, which determines the rate at which the ONs are delivered intracellularly. A 20-mer anti-GFP sequence identified previously (,) as an effective inhibitor of pd1EGFP expression (5′-TTG TGG CCG TTT ACG TCG CC -3′) and a scrambled control (5′- TTG CTT GTA CCG TGC GTG CC -3′) were utilized in the study. The phosphodiester, phosphorothioate and fluorescently tagged (5′ Cy5 end modified) forms of these sequences were obtained from Integrated DNA Technologies (Coralville, IA, USA). Stock solutions were prepared by reconstituting each pellet in water to a final concentration of 100 μM. Branched PEI of molecular weights 1.2 kDa (Cat# 6088), 10 kDa (Cat# 19850) and 70 kDa (Cat# 00618) was purchased from Polysciences, Inc. (Warrington, PA). Additionally, PEI of molecular weight 25 kDa (Cat# 408727) and 600–1000 kDa (Cat# 3880) was purchased from Sigma. Stock solutions at a concentration of 10 residue mM (0.43 mg/ml) were prepared in water and the pH adjusted to 7.0 using HCl. OliGreen, a fluorescent dye that binds strongly to single-stranded DNA, was obtained from Molecular Probes (Eugene, OR, USA). Heparin sodium salt was obtained from Sigma (Cat# H4784). Unless stated otherwise, all cell culture products were obtained from Invitrogen (Carlsbad, CA, USA). PEI/ON complexes were prepared at desired PEI/ON charge ratios by mixing equal volumes of PEI (of varying concentrations) and ONs in PBS. The samples were vortexed briefly, and the solutions were then incubated at room temperature for 10–15 min to ensure complex formation. Experimental evidence (stabilization of fluorescence corresponding to free ON) confirmed that this time was sufficient for complex formation. The PEI/ON charge ratios were calculated on a molar basis. The complexes were prepared at a final ON concentration of 10 μg/ml (approximately 1.64 μM) unless stated otherwise. Complexes between PEI and ONs were prepared at various charge ratios as described above. One hundred microliters of each complex solution was transferred to a 96 well (black-walled, clear-bottom, non-adsorbing) plate (Corning, NY, USA). A total of 100 μl of diluted OliGreen reagent (1:100 in TE buffer at pH 8) was then added to all samples for free ON detection. Fluorescence measurements were made after a 3–5 min incubation using a Cytofluor (Applied Biosystems, CA, USA), at excitation and emission wavelengths of 485 and 520 nm, respectively, and a voltage gain of 55. All measurements were corrected for background fluorescence from a solution containing TE buffer and diluted OliGreen reagent. Complexes were prepared by mixing PEI and ON solutions (final ON concentration of 50 μg/ml, charge ratio of 10:1 in PBS), and immediately analyzed using a Brookhaven Particle Size Analyzer (Holtsville, NY, USA) for 15 min with readings taken at an interval of every 3 min. Measurements were performed in triplicate. PEI/ON complexes (charge ratio 10:1, final ON concentration of 5 μg/ml ∼ 0.8 μM, volume 50 μl) were prepared as described previously and transferred to a 96 well (black walled, clear bottom, nonadsorbing) plate. One hundred microliters of diluted OliGreen reagent were added to each well and mixed manually with a multichannel pipet. Fifty microliters of heparin solution (at various concentrations prepared in TE buffer at pH 8) were then added to the wells, and the plate was maintained at 37°C. Fluorescence measurements were made at the end of 1 h from the time of heparin addition using the Cytofluor plate reader. In each measurement, we corrected for changes in background fluorescence due to heparin addition alone. This change was found to be dependent on the heparin concentration (data not shown), and thus was measured and accounted for in each experiment. Chinese hamster ovary (CHO-K1) cells (ATCC, Manassas, VA, USA) were maintained in F-12K medium (Kaighn's modification of Ham's F-12; ATCC, Manassas, VA) supplemented with 10% fetal bovine serum and penicillin–streptomycin solution. CHO-d1EGFP cells (CHO-K1 cells stably expressing a destabilized green fluorescent protein transgene) were previously produced by transfecting CHO-K1 cells with the 4.9 kb d1EGFP-N1 plasmid (BD Biosciences Clontech, Palo Alto, CA, USA), and maintained under constant selective pressure by G418 (500 μg/ml). All cell lines were cultivated in a humidified atmosphere containing 5% CO at 37°C. CHO or CHO-d1EGFP cells were plated at a density of 1.5 × 10 cells/well in 12 well plates 18 h prior to transfection. Prior to introduction of ONs, cell culture medium in each well was replaced with 800 μl of OptiMEM (reduced serum medium). Further, 200 μl of PEI/ON complex solution prepared in PBS was added to each well so the final ON concentration in each well (1 ml) was 300 nM. To measure the intracellular levels of ONs released from the complexes, cells were treated with complexes prepared with Cy5-ONs. At the end of the incubation period of 4 h, the transfection mixture was replaced with serum containing growth medium and maintained under normal growth conditions (5% CO, humidified atmosphere, 37°C). Medium in the wells was replaced with fresh serum containing medium every 24 h. At various times, cells were assayed for antisense activity [green fluorescent protein (GFP) fluorescence] and/or ON levels (Cy5-ON fluorescence) by flow cytometry. As controls, cells were also exposed to only PEI (in the absence of ONs), only ONs (in the absence of PEI) as well as complexes prepared with scrambled ON sequences to check for sequence specificity. All times indicated are relative to when complexes were first introduced to the cells, which represents = 0. Replicate wells of cells were washed in PBS, detached with trypsin-EDTA, washed with serum-containing medium, pelleted by centrifugation for 5 min at 200 , washed with PBS, resuspended in 500 μl of PBS and maintained on ice before being subjected to flow cytometry analysis. Ten thousand cells were analyzed on a FACSCalibur two-laser, four-color flow cytometer (BD Biosciences) for GFP fluorescence in FL1 (green) channel and Cy5-ON fluorescence in FL4 (far-red) channel. CellQuest software was used to acquire and analyze the results. Viable cells were gated according to their typical forward/side scatter characteristics. The flow cytometer was calibrated with fluorescent beads (CaliBRITE Beads, BD Biosciences) prior to each use to ensure comparable readings over the period of the study. We previously developed a detailed mathematical model describing the cellular events that an antisense ON undergoes in its attempt to reach and block its target in the cell (). The model was based on mass action kinetic equations on the ON and took into account a set of compartments based on cellular location (e.g. cell membrane, endosome, cytoplasm) and molecular state (e.g. free, hybridized or degraded). Here, we present a simplified version that describes the intracellular delivery of ONs and the subsequent antisense response. The model highlights the significance of ON release from polymer/ON complexes to the overall nature of antisense down-regulation. Experimental results (Cy5-ON fluorescence) were fit to Equation () using a non-linear equation solver, and the parameters and were estimated. A combination of these parameters, ( − ), is equal to the initial rate of intracellular uptake/release of ONs (, Equation A4). Antisense ONs released from PEI/ON complexes intracellularly are then capable of binding to the target mRNA to elicit an antisense response. We neglect all other events such as non-target interactions or protein binding. Further, we assume the ON-mRNA hybridization to be in rapid equilibrium, denoted by the equilibrium constant . The total mRNA from the target gene can therefore be present in the unbound or hybridized form. In these equations, the GFP mRNA and protein degradation rates ( and respectively) utilized in the model were 0.069 and 0.69 h, respectively. Given the () profiles separately measured and fit to Equation , only the equilibrium constant remains as an adjustable parameter. Equations () and () were solved simultaneously for the objective of minimizing the error associated with . A more detailed explanation of the methods employed to solve the model and fit its parameters is provided in the . First, we tested the ability of PEI of molecular weights ranging from 1.2 K to 600 K to form complexes with ONs, as a function of the PEI/ON charge ratio. We measured the amounts of free (unbound) DNA in solution using a dye, OliGreen, which fluoresces upon binding to single-stranded DNA (,). As shown in , the fluorescence levels decrease with increasing PEI/ON charge ratios. At higher charge ratios (⩾5 : 1 for PO ONs or ⩾2 : 1 for PS ONs), only residual amounts of free (unbound) ONs are detected in solution, indicating complex formation. All PEI MWs behave similarly in their ability to bind with ONs, except that complex formation is highly inefficient for PEI/PO complexes at the lowest MW PEI (1.2 K). In fact, even at the highest charge ratio (20 : 1), there is no significant binding between PEI (MW 1.2 K) and PO ONs. Therefore, PEI MW 1.2 K was excluded from all further studies. For all PEI MWs, the complexation curve is shifted to lower charge ratios for PS ONs relative to PO ONs, indicating a greater affinity for the PEI-PS ON interaction. In order to characterize further the PEI/ON complexes, particle sizes were estimated in the form of mean hydrodynamic diameter using dynamic light scattering. Complexes prepared at various PEI/ON charge ratios in PBS were subjected to particle size measurements for 15 min at a regular interval of 3 min. For charge ratios below 10 : 1, complexes were found to aggregate, as indicated by the rapid increase in particle size (data not shown). Although the OliGreen binding assay indicated PEI-ON association at these charge ratios, stable, submicron sized complexes were formed only at a charge ratio of 10: 1 or above. At a charge ratio of 10 : 1, only 10 K/PO complexes displayed particle aggregation, with the particle diameter increasing from 200 to 450 nm within 15 min. Irrespective of the PEI MW and ON chemistry, all other particles were stable in the presence of salt and maintained a mean diameter of approximately 200 nm (). All further studies were therefore performed at a charge ratio of 10 : 1. For effective cellular delivery, polycation–ON complexes should be of an appropriate strength to withstand encounters with other macromolecular species while entering the cell and during intracellular trafficking but also to dissociate (unpackage) the ON at some point to allow recognition of the target mRNA. We probed the strength of the PEI/ON interactions by studying the dissociation behavior of these complexes upon exposure to heparin sulfate as a competitive binding agent (). Complexes were prepared at a PEI/ON charge ratio of 10 : 1, after which heparin was added to dissociate the complexes, releasing ONs into solution. OliGreen was used to measure the amount of ONs released, following subtraction due to the minor effect of heparin on OliGreen fluorescence. The dose response of ON release with varying amounts of heparin was determined, using release at the end of 1 h as the measurement. Increasing amounts of ONs were released with higher doses of heparin for PEI/ON complexes of both ON chemistries, with several notable characteristics (). For PEI/PO complexes, the data revealed a threshold heparin concentration above which most of the ONs were released from the complexes. There was no significant effect of the PEI molecular weight on the release of PO ONs from PEI. On the other hand, PEI MW plays a more definitive role on PEI/PS interactions. Higher amounts of PS ONs were released with increasing PEI MW (Single factor analysis of variance (ANOVA) test, = 0.0005, for PEI/PS complexes challenged with heparin at 20 μg/ml), indicating reduced strength of binding between PEI of higher MW and PS ONs. Furthermore, a comparison of the release profiles for the two ON chemistries indicates enhanced strength of binding (lesser release) between PEI and PS ONs as compared to PO ONs. Having studied the physico-chemical properties of PEI/ON complexes, we tested the effectiveness of these polymers in delivering anti-d1EGFP ONs to CHO cells stably expressing the d1EGFP transgene (). Cells were treated with complexes of PEI (various MWs) and ONs (PS and PO backbone chemistries) for 4 h, and subjected to flow cytometry at each of several times over a 72 h time period. Under all conditions, >90% of the cells were gated as live. The fluorescence levels indicated in are normalized to the green fluorescence of time-matched, untreated CHO cells that stably express the d1EGFP transgene. Antisense ONs delivered with PEI produced transient reductions in average GFP levels, with as much as 80% reduction observed 8 h from when cells are treated with PEI/ON complexes under the best conditions. The time scale over which antisense effects were observed varied distinctly with the ON chemistry. As shown in A, PO ONs delivered with PEI produced a rapid and brief response. GFP levels declined steeply, with as much as 60% of the fluorescence lost in the first 4 h, when ONs are delivered with PEI MW 25 K. The maximum down-regulation was observed close to the 8 h time point, shortly after which the GFP levels rise swiftly and return to base level by 48 h from administration. In comparison, PS ONs delivered with PEI exhibited a more gradual antisense response that was sustained for a longer duration (B). The antisense effects were maximal between 8 and 24 h, and only gradually returned to base level over a 72 h time period. These differences in the onset of down-regulation between PO and PS ONs are more distinct at the earlier time points such as 4 h. Whereas ON chemistry determined primarily the dynamics of antisense effects, the molecular weight of PEI used to deliver the ONs influenced strongly the extent of down-regulation observed over the same time scale. PO ONs delivered with intermediate MW PEI (25 K) produced the highest levels of GFP down-regulation, while lower levels of down-regulation were recorded when PO ODNs were delivered with all other MWs (10 K, 70 K and 600 K). Interestingly, for PS ONs, the effect of the carrier MW was more pronounced and quite different from that of PO. Almost no down-regulation was observed when PS ONs were delivered with the PEI of MW 10 or 25 K. In contrast, almost 80% inhibition was obtained when PS ONs were delivered with higher MW PEI (70 and 600 K). A number of controls were utilized to evaluate the contribution of ON labeling dye, free polymer and ON sequence on GFP down-regulation (). When cells were treated with ONs in the absence of PEI, no antisense effects (i.e. decrease in GFP expression) were observed, highlighting the need for a carrier. The carrier itself exhibited minimal non-specific effects. We also exposed cells to complexes of PEI and scrambled anti-GFP sequences to verify the specificity of the anti-GFP sequences. In order to segregate any false effects due to inefficient/incomplete ON delivery, we used the particular PEI MW that provided maximum intracellular ON levels for each of the backbone chemistries, i.e. MW 25 K for PO ONs and MW 70 K for PS ONs. In all cases, the anti-GFP sequences exhibited significantly greater down-regulation than the scrambled ones (Single factor ANOVA test, < 0.01 for complexes of PEI with PO and PS ONs), providing evidence for reasonable sequence specificity of antisense inhibition. In general, utilization of the d1EGFP transgene as the antisense target provides a simple means to capture the antisense down-regulation by measurement of GFP fluorescence (,,). To detect simultaneously the presence of delivered, intracellular ONs, fluorescently (Cy5) end-tagged ONs were utilized in our experiments. Statistically indistinguishable levels of antisense inhibition were observed with dye-labeled vs. unlabeled ONs (Single factor ANOVA test, > 0.05, for complexes of PEI with anti-GFP PO ONs, and for PEI with anti-GFP PS ONs), indicating that the label did not appreciably alter delivery or antisense behavior (). Because Cy5 tagged ONs complexed to PEI do not fluoresce (data not shown), the ON fluorescence measured by flow cytometry corresponds to ONs released from PEI within cells. Absolute fluorescence levels are indicated in and are representative of data from several runs. The levels of intracellular PO ONs increase rapidly to a maximum at around 4 h after exposure (A). PO ONs also disappear quickly from cells with negligible amounts detected after 48 h. Maximum levels of intracellular PO ONs are detected when delivered using the intermediate PEI MW of 25 K. In contrast, the levels of intracellular PS ONs delivered with PEI increase slowly, reaching a maximum between 8 and 24 h after treatment (B). PS ONs are retained within cells for a longer duration, being cleared only after 72 h. It was interesting to note that very little ON was detected intracellularly when PEI of lower MW such as 10 K and 25 K were used as carriers for PS ONs. Higher MW PEI (70 K, 600 K) delivered the highest amounts of intracellular PS ONs. To quantify the relationship implied by and between intracellular ON release and antisense inhibition, we integrated these measurements using a mathematical model based on mass-action kinetics. By fitting the results from the ON uptake experiments to a simple lumped uptake/release process with intracellular and extracellular degradation (, (a) and Equation A3), we simulated the dynamics of intracellular ON levels shown in A and B. A comparison of the model fit to the experimental results in A and B shows that this model captures the dynamics observed in the intracellular release levels. Using a combination of parameters obtained from the model fit, we calculated the initial intracellular ON release rates () for various PEI/ON complexes (). This one parameter captures the observed PEI MW and ON backbone dependences. Maximum ON release rates were obtained with intermediate PEI MW (25 K) for PO ONs, while the higher PEI MW (70K and 600K) release more PS ONs. This parameter also shows quantitatively that the initial uptake and release of PO ONs exceeds that of PS ONs. Using a mass action kinetic description of ON-mRNA hybridization (, Figure A1(b)) and coupling it to a description of translation (, Figure A1(c)), we also modeled the overall antisense dynamics. Estimates of GFP mRNA half-life from literature vary from 3–12 h (), from which we utilized a half-life of 10 h for our calculations. The protein half-life was estimated as 1 h, which is the nominal value for the destabilized version of the enhanced d1EGFP. By incorporating these half-life estimates and the parameters from the ON uptake model, a single value of the ON-mRNA equilibrium constant (0.004 and 0.0084 for PO and PS ONs, respectively) was fit for each ON chemistry across all PEI MWs. As shown in c and d, the GFP down-regulation data (A and B) were captured reasonably well by the model. Although the maximum down-regulation is somewhat under-predicted, the model predicts accurately the trends with respect to PEI MW and the overall time scale of effects for each of the ON chemistries, in particular the delayed onset and sustained activities of PS versus PO backbones ONs. PEI has garnered significant attention in recent times as a building block for creating effective DNA carriers and has been tested both and to target a large number of cell types (,,). Several modifications have been proposed to take advantage of the unique buffering capabilities of this molecule, while reducing the toxicity associated with its use (,). However, most of these reports focus on delivering plasmid DNA for gene therapy applications. There are strikingly fewer systematic studies on the utilization of PEI for single-stranded DNA molecules. Insights obtained from PEI mediated plasmid DNA studies are generally extended for the application of delivering small DNA such as antisense ONs. However, as noted by us and other researchers (,,), these generalizations may not hold, especially given the considerable differences in the size, structure and chemistry between ONs and plasmid DNA. In fact, these small DNA molecules (10–20 bases) are often observed to exhibit weaker electrostatic complexation with polycations due to the small number of charged units per ON molecules. While the linear form of PEI of MW 25 K is often touted as the most effective carrier for delivering plasmid DNA, we found it ineffective for both PO and PS ONs (data not shown). Hence, to specifically identify issues related to PEI mediated ON delivery, we performed a systematic study with a set of branched PEI MWs and ON chemistries. Using the d1EGFP gene as an easily quantifiable antisense target, we screened various combinations of PEI MWs (1.2 K, 10 K, 25 K, 70 K and 600 K) and ON chemistries (PO and PS) for their ability to elicit an effective antisense response. For PO ONs, maximum antisense response was observed with intermediate MW PEI (25 K) as the carrier, while complexes of PS ONs with higher MW PEI such as 70 K and 600 K produced comparable levels of d1EGFP down-regulation. These particular PEI/ON combinations that achieved highest antisense response were also the ones that delivered the most ONs, i.e. maximum intracellular ON levels were recorded for these cases. Indeed, a monotonic relationship is apparent when the maximum antisense inhibition is plotted against the maximum ONs delivered for each combination of PEI MW and ON chemistry (). Conditions under which no antisense inhibition was observed occurred because no ONs were delivered using those particular PEI/ON combinations. A correlation between intracellular levels of short interfering RNA and gene silencing has also been reported (). Why do different combinations of PEI MW and ON chemistry deliver different ON levels? Studies of various PEI MWs for delivery of plasmid DNA do not provide a clear view, as they have produced conflicting trends. In some cases, transfection efficiency was found to increase with PEI MW (), while in others low MW PEI was effective as a gene delivery agent (). The toxicity associated with very high MW PEI offsets its use as DNA carrier (), making the less toxic low MW PEI a more attractive candidate for further improvement despite its lower net charge density. To date, very little mechanistic explanation for the differences in behavior of various MWs has been provided. Previous investigations of the cellular processing of PEI/DNA complexes suggests that complexes are taken up by binding with proteoglycans, such as syndecan, and further trafficked through the endocytic pathway (,). The similar size of all our complexes (200 nm) suggests a low probability of differences in the rate of internalization of the complexes. PEI is believed to enable escape of polyplexes from the endolysosomal pathway by a ‘proton-sponge’ effect, by which complexes and/or DNA are released into the cytoplasm (,). It is not clear where or how the DNA is ultimately released from the complex. Previous studies demonstrate no significant differences in the buffering capacity of PEI of various MWs in the relevant pH range of 5–7 (). Some studies suggest the involvement of cytoplasmic proteins and other charged molecules in the competitive release of DNA from PEI/DNA complexes (). Therefore, we hypothesize that at least part of the MW backbone effects could be due to the kinetics of complex dissociation (unpackaging) and that these could be evaluated using the heparin competition assay. Indeed, the differences in PEI–ON interactions are reflected in the heparin competition assay, where we observe a MW dependence on the amount of ONs released from PEI/PS ON complexes. Specifically, in the presence of heparin, complexes made with higher MW release more PS ONs than complexes made with lower MW PEI. However, when we tested the strength of the PEI–PO ON interactions by competition with heparin, we did not find any significant differences among the various PEI MWs. As such, it appears that the MW influences delivery of PO ONs in a manner not captured by the heparin competition assay, perhaps determining where in the cell (e.g. endosome versus cytoplasm) the DNA gets released. Competition with heparin has been used in previous studies as a measure of the stability of carrier/DNA complexes. Our studies suggest that higher stability may correspond to very tight binding that inhibits the release of DNA from the complexes. There are several such examples in the literature, wherein polymers that bind DNA effectively do not release them intracellularly despite their buffering capabilities (,). Our results provide some insights for such observations and additionally suggest interplay of polymer architecture and ON chemistry in complex stability versus dissociation. Such differences in carrier performance based on ON chemistry have been observed by others (). Similar to Dheur . (), we found PEI 25 K ineffective for PS ONs, but in addition, we find MWs higher than 25 K (such as 70 K and 600 K) to be efficient in delivering PS ONs. Apart from identifying higher PEI MWs as effective carriers for PS ONs, our results more significantly highlight the importance of the strength of the electrostatic interactions between the PEI and ONs, which ultimately dominates the rate and extent of DNA release. Both the PEI architecture and ON chemistry play a role in these interactions. The degree of protonation and the flexibility of the polymer chains are speculated to be significant. For example, fractured dendrimers that have more flexible chains are better at delivering plasmid DNA as compared to intact dendrimers (). In our system, higher MW PEI probably possesses more flexible chains, which are able to interact with the heparin (in our non-cellular assay) and with unspecified species in cells, leading to release of ONs. The role of the ON backbone is somewhat less clear. Both ON chemistries are known to have similar charge densities due to the phosphate groups; however, phosphorothioates are known to be more hydrophobic than phosphodiesters. The PO and PS backbones differ only by a single atom: the non-bridging oxygen is replaced by sulfur in the PS backbone. Compared to oxygen, the sulfur atom has less electronegativity. Studies report that the lower charge density of the sulfur atom increases its polarizability, strengthening the interaction with lower charge density groups in proteins (,). This was reflected in our PEI–ON binding assay, in which PS ONs were found to bind to even the lowest MW of PEI (1.2 K). In contrast, 1.2 K PEI and PO ONs did not bind even at very high charge ratios. Furthermore, PEI–PS ONs displayed higher strength of interactions in the heparin competition assay. The manifestation of these molecular interactions on the cellular processing of the vectors is indeed quite dramatic, in that some complexes are apparently so tightly bound that they are incapable of releasing the ONs. As these interactions can be modulated by appropriate modifications to carrier and ON properties, they present a design opportunity for producing an effective antisense response. We also observed distinctly different antisense dynamics dependent on ON chemistry. The overall antisense dynamics consists of two phases: (i) the initial onset of gene inhibition leading to maximum down-regulation and (ii) the duration for which the antisense effects last. The latter can be easily explained based on the known differences in the resistance of PO and PS ONs to nucleases. The rapid disappearance of PO effects is consistent with the susceptibility of PO ONs to intracellular nucleases. PS ONs are known to be more nuclease resistant and to survive in the cell for longer (), leading to sustained antisense effects, which lasted for up to 72 h in our studies. Based on our previous detailed kinetic modeling (), we would expect, for equivalent delivery characteristics, substantially less silencing of gene expression for PO versus PS ONs based on the higher nuclease degradation rate of the former. The fact that we observed significant, albeit short-lived, silencing with some PO ONs suggests that they are being delivered to the cytoplasm at a faster rate than their PS counterparts. The dynamics of intracellular release and gene silencing support this view. For PO ONs, the onset of gene inhibition was more rapid with maximum inhibition observed at 8 h (A). For PS ONs, the onset was somewhat slower, and antisense activity was greatest between 8 to 24 h from when cells were first exposed to PEI/ON complexes (B). A more rapid intracellular release was also observed with the PO ONs (), although the intracellular release measurements should be considered under the caveat that the measured intracellular fluorescence may include contributions from partially degraded ONs and free fluorescent tag in addition to delivered, fully intact ON. Overall, the results suggest that PEI/PS complexes take longer to release ONs. Our heparin competition assay results highlight these differences in strengths between PEI and the ONs of PO and PS chemistries (). These results suggest the possibility that a mixture of PO and PS chemistries would possibly demonstrate both early and sustained antisense activity (). We are currently testing this hypothesis by using PO backbones with varying degree of PS substitution. Because of the rapidly changing levels of intracellular Cy5-ON fluorescence and d1EGFP fluorescence at early times and the discrete time points at which fluorescence in measured, it can be difficult to discern the differences in dynamics among some of the samples. Our mathematical model of the overall antisense process provides a useful framework for interpreting these measurements. By fitting experimental results to a relatively simple function, we were able to simulate the intracellular ON levels and calculate intracellular ON release rates from various PEI/ON complexes. The release rates account quantitatively for the variation in activity as a function of PEI MW and also demonstrate a markedly increased release rate for PO vs. PS ONs (). To capture the antisense dynamics, we simplified the overall antisense process by incorporating only critical features such as equilibrium ON-mRNA hybridization. The fact that, with the parameters calculated from the intracellular ON function and a single globally optimized parameter (ON-mRNA binding constant), we can fit all of the activity data with fidelity of the trends, is a further indication that intracellular release governs the activity to a considerable extent. Mechanistically this implies that downstream barriers such as trafficking, nucleo-cytoplasmic shuttling, association with proteins, and binding to non-target mRNA do not seem to be as significant in governing the dynamics and MW effects, though they could still be important in a manner that is not dependent on PEI MW. A similar approach has been used to identify rate-limiting steps in siRNA gene silencing and . Overall, the results presented here have important implications for the rational design of polymeric carriers and ON backbones used for antisense applications. We have demonstrated that the final antisense activity observed is determined not solely by either carrier or ON chemistry, but rather by the interplay of both factors. While the extent of down-regulation was determined primarily by the polymer MW, the dynamics were determined chiefly by the ON chemistry. Of particular importance is the strength of interaction between the carrier and the ON, which determines the rate at which the ONs are released intracellularly. From a practical standpoint, our results identify PEI MWs that are effective for delivering ONs of PO and PS chemistries. This approach should be useful in predicting and interpreting results for ONs of other chemistries as well, though it will be interesting to see to what extent the observed trends and correlations hold in other cell types. While PEI MW 25 K is generally considered the golden standard for plasmid DNA delivery, this is not true for the delivery of ONs. More strikingly, we find that the performance of a particular PEI MW is determined by the chemistry of the ON it is used to deliver. For example, PEI 25K was most effective in delivering PO ONs, but no PS ONs could be delivered with the same carrier. Modifications of polymers with targeting molecules such as ligands or PEG could affect the interactions with the ONs and hence influence the overall ON release dynamics. Thus, one should be able to control the onset and duration of antisense activity via biophysically guided selection of ONs and carriers.

The identity of eukaryotic cells is defined by the correct temporal and spatial expression of specific genes. The first step in the selective expression of any gene is the ability to single it out from among all the others genes in the genome. This ability, which lies at the heart of many cellular processes, invariably requires the interactions of proteins with DNA molecules. Protein–DNA interactions proceed through an induced-fit mechanism, similar to the induced-fit mechanism of enzyme action (). Both the DNA and the protein are not passive players, but have active roles, dictated by their structural plasticity. The ability of the DNA to deform upon interaction with proteins (‘deformability’) is determined by the preferred stacking interactions between adjacent base pairs () and gives an indirectly recognized structural signature (). Sequence selectivity, in most sequence-specific DNA-binding proteins, is based on direct hydrogen bonds between amino acid residues and the donor and acceptor groups primarily in the major groove of DNA [‘direct readout’, ()]. In addition, the sequence-dependent features of intrinsic DNA structure and its deformability contribute to specific recognition [‘indirect readout’, ()]. In several specific protein/DNA complexes, DNA conformation is used to the extreme case and there are indications for interaction mainly through indirect readout. The TBP/TATA-box system is such a system (). The formation of the TBP/TATA-box complex is the first step in the assembly of the preinitiation complex on promoters of genes that are transcribed by RNA polymerase II. TBP binds to an eight-base-pair segment of DNA, which thus defines the core TATA box (). The TATA-box consensus sequence is T-A-T-A-W-A-W-R (W = A, T; R = A, G) (,). The crystal structures of various TBP/TATA-box complexes show that the formation of the complex results in a severe bend of the DNA towards the major groove (80–100°), as well as untwisting, which exposes a very wide and shallow minor groove on its convex side where TBP engages (). We have previously studied the indirect readout mechanism of TATA-box recognition by TBP (). Based on this study, we proposed several possible signals for TBP recognizing TATA boxes through indirect readout: The first signal was the helical twist angle in the middle of TATA boxes (base pairs 4 and 5, the only two base pairs with direct hydrogen bonds to the protein). Second signal was the identity of base pairs 7 and 8 and third, recognition of global DNA flexibility (). In TATA boxes containing alternating (T-A) runs (similar to the sequence of the AdE4 TATA box) binding affinity and stability to TBP was recently shown to be significantly dependent on the nature of the sequences flanking the core TATA box (). We suggested that this is a novel form of indirect readout (). The structure of (T-A) runs is polymorphic () and context dependent (). Consequently, this pliable structure can be altered by the DNA structure at the flanking sequences, thereby indirectly influencing the interaction with TBP. The variability observed in TBP binding to E4-like TATA boxes, as a function of changing the flanking sequences, is comparable to that observed when the sequences within the core TATA box itself are changed (). Indirect readout of DNA sequences can sometime manifest itself through non-additivity effects in protein–DNA interactions. By this it is meant that protein-binding affinity cannot be accounted for by successive contributions from individual nucleotide pairs within the target sequence. Such non-additivity has been observed in several systems, notably the Mnt system () and the EGR1 Zn-finger system (). However, in both cases it was concluded that the additive, mononucleotide-based assumption, was good enough for most purposes (). Recently, O’Flanagan . () observed non-additive effects in the TBP system, but in this study a theoretical approach has been used to obtain binding-site data. We study here the binding properties of TBP to all consensus-like TATA boxes. We show that grouping consensus-like TATA boxes by their structural properties reveal differences in the indirect readout of these TATA boxes by TBP, and in TBP-binding mechanism. Statistical analysis indicate that TATA boxes that have a context-independent cooperative structure are best described by a nearest-neighbor non-additive model, whereas TATA boxes that have a flexible context-dependent conformation cannot be described by either an additive model or by a nearest-neighbor non-additive model. The c-terminal domain of yeast TBP (yTBPc) was a kind gift from S. Juo (Yale University). The overexpression and purification of the protein were as described by Kim (). The fraction of yTBPc active for DNA binding was determined as previously described () and found to be 50%. All TATA-box variants in this study were chemically synthesized on an automated DNA synthesizer at the Keck Foundation Resource Laboratory (Yale University) or by Sigma Genosys (Israel), and purified using standard protocols (). TATA-box variants for dissociation kinetics experiments were chemically synthesized as hairpin constructs with 20-base-pair (bp) double-stranded stems and five cytosines in the loop (). TATA-box variants for phasing analysis were chemically synthesized as linear duplexes 21-bp long. They are identical in sequence to the stem of the hairpin variants except for an additional T at the 5′ side, used to create an AvaI site for cloning the fragments as described previously (). These linear duplexes were also used as specific competitors in the dissociation kinetics experiments. Radiolabeled hairpin duplexes (0.4 nM) and yTBPc (27 nM active protein) were incubated for 60 min at 30°C in the binding buffer before adding unlabeled 21-bp linear duplex competitor of the same DNA sequence (1.76 μM, 65-fold excess of the cold competitor over active protein and 4400-fold over labeled DNA targets). We used these experimental conditions to concur with those of our previous study (). The rational for using short hairpin duplexes as DNA targets and short linear duplexes as DNA competitors was previously described (). At the time points indicated in and , samples were removed and immediately frozen in liquid nitrogen (). After the final time point, the samples were thawed and immediately loaded on native gels (10%, acrylamide/bisacrylamide ratio 75:1, 10% glycerol) while the gels were running. The gels were run at 450 V and 30°C, in a running buffer containing 0.5× TG (25 mM Tris.HCl, 190 mM Glycine, pH 8.3) and 5 mM MgAc, until the BPB dye migrated 5.5 cm. yTBPc-induced DNA bending was analyzed by phasing analysis using radiolabeled DNA targets, 569–579 bp long, as previously described (). These DNA probes (0.4 nM) were incubated with 25–200 nM yTBPc for 60 min at 30°C. The relative mobilities of the complexes were analyzed on native gels (6%, acrylamide/bisacrylamide 75:1, 10% glycerol). Gels were run at 450 V and 30°C, in a running buffer containing 0.5× TG (25 mM Tris.HCl, 190 mM Glycine, pH 8.3) and 5 mM MgAc, until the XC dye migrated 12 cm. All gels were dried and quantified using a Fujii Bas-1000 phosphoimager. For the analysis of the kinetic experiments, boxes were defined surrounding each band on the gel. To account for dissociation of the complex during electrophoresis, the band corresponding to the protein/DNA complex was defined as extending from its main band to the free-DNA band (). A similar box in a lane containing the unbound target only defined the background. The fractions of bound DNA at the different time points, (), were calculated from the equation: () = (PSL − bg)/[(PSL − bg) + (PSL − bg)], where PSL is the photostimulated luminescence and bg is the background. ln[()/] was plotted as a function of time () after the addition of the unlabeled competitor. The data was analyzed by a two-phase first-order kinetic equation: ()/ = + , where and are fractions of molecules dissociating with rate constants and , respectively. Half-life of complexes dissociating by and processes were calculated from: = ln2/ and = ln2/. Analysis to determine the local helical parameters in crystal structures of TBP/TATA-box complexes was carried out using the web version of 3-DNA (). However, in structures containing Hoogsteen base pairs this analysis yielded erroneous parameters. These structures were then analyzed using curves (). We have downloaded from the eukaryotic promoter database [EPD, (,)] sequences corresponding to the degenerate consensus sequence YWTAWADN. This consensus sequence corresponds to all TATA-box variants that appear with at least moderate frequency (>10%) in eukaryotic promoters, according to the TATA-box consensus of Bucher (). We have filtered the sequences to exclude unidentified preliminary bulk sequences (as defined in the EPD, i.e. sequences with the ‘OS_bA’ identifier), and those derived from high-throughput studies, which cannot be assigned to a defined homology group. We define a homology group, as in the EPD, as sequence similarity due to common phylogenetic origin. In the EPD, and here, two promoters are considered homologous if they exhibit >50% sequence similarity between −79 and +20. However, as the definition of homologous promoters is based only on similarity of DNA sequence in the promoter region, they can be either orthologs or paralogs. We then deleted from each set multiple sequences coming from the same homology group, as well as TATA boxes in the transcribed region. Thus, the dataset now corresponds to a representative set of not closely related promoters. Similarly, we downloaded sequences corresponding to the more restricted consensus sequences, YWTATADN and YWTAAADN. All datasets were aligned by the program MEME (), using only the one strand given in the EPD. From the YWTAWADN dataset (457 sequences) we constructed mononucleotide position-specific weight matrices whose elements are the log-odds weights: is taken here to be 0.25 for each base. These matrix elements are a maximum probability estimate for the binding energy contribution of each base at each position, assuming that each position contributes independently to the total binding energy (). The base frequencies are corrected here for small sample errors as described by Berg and von Hippel (). To score individual sequences, the weight matrix is multiplied by a matrix () containing only 0's and 1's, corresponding to sequences for which binding data was experimentally determined in this study. The summation of this multiplication yields an individual information score for each sequence (,): To the summation we add a constant (), chosen such that the best binding site scores zero and poorer sites score positively. To test for nearest-neighbor non-additive effects we need to calculate dinucleotide information scores, for which we need to add to mononucleotide information scores the following term (,): Here again a constant is added to each result to make the score of the best binder zero. We have also calculated the Z statistics of tetranucleotide motifs at position 6–9 in TATA boxes. It is calculated by subtracting from the observed number of occurrences of each motif the expected number of occurrences based on the mononucleotide frequency of the component base pairs, and then dividing this value by the expected SD (). In testing the strength of relationships between variables, we calculated Spearman's rank correlation coefficient (denoted by ρ) as a non-parametric measure of correlation (). We have grouped the ten TATA boxes studied here to two groups. The first group contains sequences that resemble the Adeno virus major late promoter (MLP) sequence, TATAAAAG (). All sequences in this group have a central A-A step. The second group contains sequences that resemble the Adeno virus E4 promoter, TATATATA (). All sequences in this group have a central A-T step. Initially, we made this division based on our previous observations () that TATA boxes having a central A-A step have higher twist angle (around 13°) than TATA boxes with a central A-T step (around 3°), even though both angles are untwisted relative to the canonical value, which is 34° for generic B-DNA in solution (). This observation is still valid, but is less distinctive, when we analyze co-crystal structures corresponding to the present ten sequences (). The average twist angle at the central A-A position in TBP/DNA complexes identical to those of group I (pdb codes: 1qne or 1cdw for MLP; 1ngm for A, 1qn4 for T and 1qnb for T) is 11°± 2°, whereas the A-T twist angle in TBP/DNA complexes corresponding to group II sequences (pdb codes: 1qn7 for T; 1tgh for (TA); and 1yth for TA) is 6° ± 2°. Looking at , we observe that sequences belonging to group I all harbor an A-tract, defined as a DNA region consisting of four or more A's in a row (). TA and T have only an A-tract, but it has been shown that AT tracts have similar structural properties to A tracts [ + = ≥ 4, (,)]. A-tracts are known to adopt a dominant unique structure, distinct from that of generic B-DNA (), which is invariant and sequence-context independent (,). A-tract may even confer unique structural properties to sequence adjacent to them (). On the other hand, alternating (T-A) runs are known to be a conformationally flexible DNA element, relative to B-DNA in general and A-tracts in particular (,,). Thus, the sequences of group I have on the whole a more rigid DNA structure than those of group II. This suggestion is supported by looking at variability in roll angle, between same steps in different structures, especially the steps at positions 4/5 and 5/6. Roll angle at the 4/5 position vary between 23.5° ± 0.6°, for group I and 26° ± 3°, for group II. At the 5/6 position the variability is 23° ± 1°, for group I and 22° ± 5°, for group II. Thus, the variability in roll angle at these positions is 5-fold larger for group II sequences than those of group I sequences (compare the SD values of group I to those of group II, i.e. 0.6° to 3° and 1° to 5°). Moreover, the average deviation of the roll angle along any one crystal structure corresponding to those studied here is also larger for sequences of group II than those of group I, 13.2 ± 0.5° for group II, versus 12.1 ± 0.1° for group I. Large roll fluctuations are commonly associated with conformational flexibility. Packer . () have argued that when DNA bending is non-planar, such as in the nucleosome () and in the TBP/TATA-box complex (), the bending motion requires shearing by slide and shift, in addition to utilizing changes in roll, tilt and twist, and that slide and shift are better studied on the tetranucleotide level. We have used the values given by Packer . () for the flexibility of tetranucleotides with respect to slide, to compare group I to II. We calculated the flexibility with respect to slide of the ten sequences studied here by summing the components tetranucleotides in a sliding window along the core 8-bp of each sequence. We then averaged the values from each 5-sequence group. Group I has average flexibility of 71 ± 7 KJ/mol Å, whereas group II has an average flexibility of 35 ± 7 KJ/mol Å. Thus, group I is significantly more rigid than group II, also with respect to slide. We measured the bend angles induced on TATA boxes upon binding of yTBPc () by phasing analysis (,,,). The results () show that when we divide the set to two groups (MLP-like versus E4-like), we can arrange the sequences within each group with the same order of base-pair steps at position 7/8, going from complexes with a large induced bend angle to complexes with smaller bend angles. In both groups sequences with the A-G step harbors the largest TBP-induced bend angle, followed by T-A, A-A, A-T and T-G, spanning the range from 76 (±4)° for TATATAAG to 43 (±4)° for TATATATG (). Thus, the bend angles induced on the TATA box by TBP binding are not only sequence dependent (,), but they depend only on the identity of the dinucleotide at position 7/8. These results are similar to those obtained in the study of Wu . () who measured the solution bend angles for TBP complexes with different AdMLP-related TATA-box variants, using fluorescence resonance energy transfer (FRET). The values of the bend angles in the study of Wu . () ranged from 76° for the complex of TBP with MLP to 30° for the complex of TBP with an A variant (TAAAAAAG). These results contrast with the crystallographic study of Patikoglou . (), who observed a similar structure of the wild-type TBP complexes with eleven naturally occurring variants of the AdMLP, including the T and T sequences. Two different explanations are possible for this discrepancy. First, the difference may be due, at least partly, to the writhed DNA structure observed in the crystalline state, as discussed in (). Variations in electrophoretic mobility, between DNA fragments of the same length, are related to differences in the mean square end-to-end distance of the molecules. Bend angles derived from phasing analysis, or from FRET measurements, are 2D entities, determined from the differences in mobility between the and isomers, or distance differences between the 5′-dye and the 3′-dye. Therefore, the difference between the crystallographic and solution results may thus be partly attributed to the difference in the outcome of projecting a 3D curve onto a 2D plane in the two methods. However, no sequence-dependent differences in writhe have been observed in the crystallographic studies (). Second, it has been shown by Wu . () that the osmolytes used to crystallize TBP/TATA-box complexes increase the bend angle of the DNA in the complex to the bend angle observed in the crystalline structures. The data presented here support this option and points to the last base-pair step, position 7–8, as the origin of the sequence-dependent pattern. We have determined the rate of dissociation of yTBPc from all variants studied here by gel electrophoresis (), as previously described (). However, we could not fit well the data of two of the variants (T and T) to a one-phase dissociation equation, as in our previous study (). Hence, we analyzed these two variants by a two-phase kinetic equation (). Consequently, we have also re-analyzed the remaining data by a two-phase kinetic equation (). This analysis shows () that for all variants there are two macroscopic dissociation processes occurring simultaneously—a fast process (termed ‘A’ in ), which is poorly defined in terms of a rate constant (ranging approximately between 7 and 20 min for different TATA boxes, but with a curve-fitting error of the same magnitude or larger in some cases), and a slow process (termed ‘B’ in ), with a low curve-fitting error. However, whereas in all variants except T and T the molecules mainly undergo the slow process (∼80%), for the T and T variants the picture is different. A significant part of T and T molecules undergo mainly the fast process (∼75%). There are two likely interpretations for the observed behavior. First, the two events could arise from dissociation from non-specific DNA versus dissociation from the TATA box (). According to this hypothesis the T and T variants have very low sequence specificities, and thus dissociation from the TATA-box region of the T and T variants and dissociation from the sequence flanking them has similar (and short) half-life. Second possible explanation is that the kinetics of dissociation could proceeds through a complex mechanism with several intermediates (). Parkhurst . () proposed that these intermediates are different intercalation states of TBP on TATA boxes. To differentiate between these two possibilities, we have studied TATA-box variants with methylated cytosines in their flanking sequences. If the different biphasic behavior observed for the T and T variants are due to dissociation from specific sequences (core TATA box) versus nonspecific sequences (the flanking sequences), i.e. if there is sliding of TBP on the TATA box prior to dissociation (), then we would expect different behavior of methylated and unmethylated targets, because we expect that methyl groups, being a bulky side group on cytosine residues, will inhibit this lateral movement by TBP, or else that the methyl group on the flanking sequences will inhibit binding of TBP to these regions. However, if the biphasic behavior of the T and T variants is due to different dissociation events from the TATA box itself, then we would expect similar dissociation behavior of methylated and unmethylated sequences. To address these issues we have studied two representative TATA-box variants, MLP and T, embedded within sequences with methylated cytosines in the region upstream and downstream to the TATA box (). MLP represents the TATA-box variants that undergo mainly the slow process, whereas T represents the variants that undergo mainly the fast process. shows the results of dissociation kinetic experiments using methylated T and MLP TATA-box variants. Analysis by a two phase kinetic equation () shows that the stability of yTBPc complexes with methylated and unmethylated sequences is similar for the T variants (123 ± 4 and 110 ± 5 min, respectively) as well as for the MLP variants (267 ± 17 and 255 ± 24 min, respectively). The fraction of molecules undergoing the slow process did not change significantly upon methylation (29 ± 2% of the methylated T and 25 ± 3% of the unmethylated T variants; 72 ± 5% for the methylated MLP and 83 ± 3% for the unmethylated MLP sequences). These results indicate no significant difference between the dissociation kinetic behavior of yTBPc complexes with regular TATA boxes versus with TATA boxes containing methylated cytosines in their flanking sequences. Thus, we can conclude that the two processes are dissociation events from complexes having different intercalation states. Powell . () suggested that the overall energy profile for the reaction between either AdMLP or AdE4 with yTBP is similar, but that they are composed of different energetic intermediates. Both binding events are composed of three distinct steps, of which the initial step is the interaction of the similar 5′ TATA part of the TATA box with yTBP, i.e. binding, bending and insertion of the 5′ phenylalanines from the stirrup loops between nucleotides T and A. In the AdE4 target, the next step is ordering of the flexible T-A steps, and intercalation of the second phenylalanine pair between the nucleotides at position 7 and 8 (). The interaction is basically over before the final step, which consists of further structural and energetic adjustments. In the more rigid AdMLP target, the second step is not as facile, and thus the intercalation step is delayed until the final step (). We suggest that this overall view is in accord with the results obtained with our extended set of TATA boxes, which are all either MLP-like or E4-like (). We propose that in group I, the second intercalation event is delayed relative to that in group II. Thus, when the equilibrium TBP/TATA-box mixture is challenged with a large excess of competitor DNA the members of group I that are bound with the least binding stability (T and T), will be found to a larger extent stuck at the second step of the binding reaction. Hence, these sequences will not have the second pair of phenylalanine intercalated at the 3′-side of the TATA box, and as a result a larger percentage of these complexes dissociate fast. Within the group of MLP-like TATA boxes (group I) the values of the dissociation half-life and the global bending induced by TBP binding are correlated to each other (). The TBP/MLP complex is the complex with the longest half-life, and it is also the one with the highest DNA bend induced by the binding of TBP. This trend is followed for the rest of group I sequences (ρ = 1). This is similar to the relationship found by Starr . (). On the other hand, no correlation between global induced TATA-box bending and half-life of the complex with TBP is observed in E4-like TATA boxes (group II, ρ = 0.1). However, in group II other structural correlations emerge. First, there is a correlation (ρ = 0.9) between binding stability and dinucleotide flexibility with respect to slide, when we use parameters taken from the study of Packer . (). Here the most rigid dinucleotide (A-A) forms the most stable complex with TBP, and the most flexible dinucleotide (T-G) forms the weakest complex. Similarly, a strong correlation is observed between the conformational energy of tetranucleotides at position 6–9 and binding stability for group II sequences (ρ = 1), but no such correlation is found for group I sequences (ρ = 0.2), when we use parameters taken from the study of Packer . (). In group II tetranucleotides with the lowest conformational energy (minimized with respect to all six base-pair step parameters for the central dinucleotide) form the most stable complexes with TBP. This correlation holds for group II sequences also when we sum the minimal tetranucleotide conformational energy along the entire sequence of each of our target sites, including the flanking sequences (ρ = 1). Berg and von-Hippel (,) were the first to link between the statistics of binding sites occurrences and their binding free energy. In group I we see such correlations, both at the 8-bp level as well as on the dinucleotide level (). Binding stability is correlated with the number of occurrences of the 8-bp core TATA box in the EPD (ρ = 1), as well as the frequency of occurrence in EPD of dinucleotides at position 7/8 in A-A-containing TATA boxes (ρ = 0.9). No such correlation is observed among group II sequences. Moreover, we have calculated an informational theoretical weight matrix from sequences that conform to the YWTAWADN consensus. The matrix elements are the log-odds ratio per base pair and per position. This degenerate consensus sequence includes all high and moderate probability mononucleotide combinations appearing in the base frequency table of Bucher (). There are 457 such sequences in the EPD, when we take only identified sequences belonging to known homology groups, and we take only one sequence per homology group, i.e. only sequences that are not from closely related promoters (see Material and Methods section for details). The matrix elements are the maximum probability estimate for the binding energy contribution of each base at each position, when we assume that each position contributes independently to the total binding energy (). The sum of the dot product between this matrix and a matrix (containing only 0's and 1's as its elements) corresponding to a sequence studied here gives an informational score for that sequence, which is the calculated total binding energy for that sequence (). If the additivity assumption holds true for the studied binding sites the informational score for these sequences should correlate with their measured binding affinity. When we have indications for non-additivity in protein–DNA interactions, we can correct for nearest-neighbor interactions by calculating the dinucleotide information score for these sequences (). This is done by adding to the term based on independent mononucleotide contributions a term that takes into account doublet correlations (,). We have measured the binding stability of ten TATA boxes to yTBPc, and not the binding affinity, which is the more direct measure of the binding free energy. However, Hoopes . () found a direct correlation between binding affinity and binding stability for yTBP. Based on their experimental results, Hoopes . () concluded that the primary difference among TBP/TATA-box complexes is the dissociation rate, and that the difference in association rate between various yTBPc/TATA-boxes complexes is small. These results are in accord with the study by Grove . (). Grove . () found that increased affinity of yTBP to TATA boxes that are more flexible, because of various sequence mismatches or because of various replacements of T with 5-hydroxymethyluracil, is due almost exclusively to an increase in complex stability rather than in the rate of complex formation. In addition, Starr . () found that the binding kinetics determined for yTBP, paralleled those for yTBPc. In , we present the mononucleotide and dinucleotide information scores for each of the sequences studies here. When we calculate the rank-order correlation coefficient, we observe that at the mononucleotide level there is no link between the measured binding stability and the individual information score of the 10 sequences studied here (ρ = 0.16). When we look at each group of five sequences separately, we find a very weak correlation between the binding stability of group I members and their individual mononucleotide information score (ρ = 0.5), and no such correlation among group II members (ρ = 0.2). If we include a nearest-neighbor doublet correlation term in the individual information score, the overall correlation among all 10 sequences is still very weak (ρ = 0.48). However, among group I members the correlation is now very high (ρ = 0.9), whereas for group II sequences no correlation is found (ρ = 0.2). We suggest that this behavior is due to the different structural properties of the two groups. Group I sequences all have A-tracts in them, a known cooperative-built structural unit that forms in sequences of the form A or AT [ ≥ 4, + = ,]. Thus, the non-additivity in group I is due to the presence of cooperative A-tract motifs in these sequences. Berg and von Hippel () suggested that non-additivity will be observed above the scatter in the calculated and experimental binding energy only if at least half of the binding site is involved in the non-additive effect. In our case, the A-tract motifs are 4-bp long, and thus comprise half of the 8-bp core TATA box. A different long-range cooperative structure exists in group II sequences. Here it is the cooperative structure of the flanking sequences that determines the structure in the core sequences (). Thus, the non-additivity in group II is of different origin than that of group I. In group II sequences we have non-additivity, but it is not influenced by nearest-neighbor interactions within the TATA box, but instead stems from the effects of the flanking sequences on the core TATA box. Thus, no correlation is found between binding stability and individual information score that is based on weight matrices build from probability of occurrences of either mononucleotides or dinucleotides in TATA boxes (see further discussion below). We have calculated the Z statistics for tetranucleotides at position 6–9 from the dataset of sequences of the form YWTAWADN. It measures the deviation of the observed tetranucleotide motif from that expected based on additive mononucleotides. Shorter sequence motifs appear both in the MLP-like sequences as well as in the E4-like group. Hence, calculating the Z statistics of motifs shorter than tetranucleotide is biased by the higher occurrence of MLP-like sequences relative to E4-like sequences in eukaryotic genomes (∼2:1 ratio). In it can be observed that there is a relationship between the Z statistics for the tetranucleotide at position 6–9 and the bend angle induced on TATA boxes by TBP binding in both groups. Larger induced bend angles (63–76°) are linked to tetranucleotides with positive Z-score, whereas those with smaller bend angles (43–53°) have negative Z-scores. This indicates that the latter sequences appear in natural sequences less than their mononucleotide occurrences, i.e. that they are being avoided in natural promoters. This may mean that TATA boxes in which TBP induces smaller bend angles are avoided in natural sequences regardless of the binding stability in complexes with TBP. Since the homology groups, as defined in the EPD and used here, are related to the DNA sequence of the promoter only, and not to its attached gene, it is impossible at this point to deduce whether this property is observed when looking across phylogenetic trees, namely, whether an alternative explanation to these observations is that sequences that incur small bend angle are selected for this property (or a related one) and are equally well conserved but are used less frequently. Specificity is not maximized in evolution. Instead, as Berg and von-Hippel suggested (), evolution minimizes the maximum loss of specificity. Thus, specificity will tend towards a situation where mutational drift have relatively small effects. Hence, if we take a dataset of strong TATA boxes, such as that composed of the sequences conforming to the YWTAWADN consensus, we do get a correlation to binding stability, when we calculate the individual information score for sequences having the context-independent A-tract motif in them, or when we simply take the 8-bp occurrences of these sequences in the EPD, as expected based on the statistical–mechanical selection theory of Berg and von Hippel () (). This is not the case when we look at sequences containing a flexible context-dependent (A-T) motif. There are some indications that E4-like TATA boxes may be more sensitive to base changes within the core TATA box. First, they are more sensitive to base changes at position 7 and 8, as can be observed from the half-life in . Second, the change of T to A has larger effect (greater reduction in binding stability) going from TATATAAG to TAAATAAG, than from TATAAAAG to TAAAAAAG (). However, in addition to the sensitivity to mutation within the TATA box, E4-like TATA boxes are sensitive to base changes in the sequences flanking them. Only in these sequences the role played by the flanking sequences in determining the structure of the core TATA box is dominant (). Thus, E4-like TATA boxes are probably more sensitive to mutational errors than MLP-like TATA boxes, if only for the extended DNA region in which mutations can have an effect on TBP binding This may be the reason why evolution did not select these binding sites to be strong promoters, even though TBP can form very stable complexes with such sites, and in optimal sequence context TBP can form stronger complexes to these sites, than those it forms with the known strong basal promoter MLP. An additional structural rational why evolution did not frequently select E4-like sequences to be strong TBP-binding targets may be that stated in our previous publication (). The pliability of E4-like sequences makes it quite easy to modulate their binding properties using their flanking sequences, whereas for the MLP TATA box these changes are not possible, and it is invariably a strong binding site. Thus, one can extend the specificity of TBP/TATA-box interaction by the use of flanking sequences of certain TATA boxes only. For proteins that recognize their target sites mostly or exclusively by indirect readout, as is the case with TBP, mononucleotide weight-matrix methods do not work well in locating new binding sites. Including nearest-neighbor doublet correlation does significantly improve the correlation to binding free energy. However, as observed here, this is true only for sequences in which non-additivity is local. Sequences, in which non-additivity is of longer range than successive base pairs, are not represented well by probabilistic methods based on frequency of occurrence of base pairs in genomic DNA. For such sequences, we need to use methods that are based on experimental data on binding-site strength, which may soon be available more easily from high-throughput studies ().

The high-throughput functional identification and structural characterization of transcriptional networks are major objectives of post-genomic research (). Predictive methods have an important role to play in this endeavor since the large number of protein/DNA and protein/protein interactions involved in transcriptional regulation precludes their systematic study by X-ray crystallography or NMR. Since transcription factor families are generally specified by highly conserved consensus DNA-binding domains (DBD) as well as common strategies of interaction with target DNA () DBD homology modeling is a particularly relevant approach (see () and references herein). Equally, the prepositioning of a DBD within its DNA-binding site can often be inferred by homology, a step that most docking programs cannot yet address (). However, despite these advantages, the prediction of DBD/DNA complex 3D structures is by no means straightforward, as exemplified by complexes involving the POU DBD. The ‘POU’ (acronym of Pit, Oct, Unc) family of transcription factors is defined on the basis of a common DBD of approximately 160 residues, first identified in the mammalian proteins Pit-1 and Oct-1 and the nematode factor Unc-86 [for a review, see ()]. The POU DBD comprises two distinct, highly conserved sub-domains, termed ‘POUs’ and ‘POUh’, which contain respectively four and three α-helices and are connected by a flexible linker, variable in sequence and length. The crystallographic structure of the complex between the POU domain of the ubiquitous protein Oct-1 and the octamer ATGCAAAT has revealed that POUs interacts with the tetramer ATGC in a similar fashion to the phage repressors, whereas the POUh interaction with the tretramer AAAT resembles that of a homeodomain (). If all the POU domains can bind to the prototypic octamer ATGCAAAT, they also recognize numerous other AT-rich sequences due to the flexibility of the linker joining the two sub-domains (). Remarkably, crystallographic structures of various Pit-1 or Oct-1 POU/DNA complexes have shown that the elements of a DNA target recognized respectively by POUs and POUh neither have to be contiguous nor even to belong to the same DNA strand (). Taken together, these structures have revealed two distinct patterns of POU homodimerization, based on different relative positionings of POUs and POUh, and depending on the type of DNA target. The ‘PORE’ (Palindromic Oct-1 Responsive Elements) DNA motifs induce a POU conformation similar to that found in the initial Oct-1 POU/octamer complex. By contrast, the ‘MORE’ (More palindromic Oct-1 Responsive Element) DNA motifs elicit a POU conformation analogous to that first discovered in Pit-1 POU/DNA complexes. N-Oct-3, the human equivalent of the mouse Brn-2 protein, is widely expressed in the developing central nervous system, and necessary to maintain neural cell differentiation (). It is also implicated in the development of the neural-crest-derived melanocytic lineage and its over-expression in melanocytes leads to tumorigenesis the dysregulation of a number of genes (). The fact that N-Oct-3 can interact with such a variety of targets is due to the structural plasticity of its POU domain. In a previous report (), we have shown that the N-Oct-3 DBD, in addition to forming the classical homodimers in association with PORE and MORE sequences, can also adopt a novel mode of homodimerization when bound to a set of neuronal promoters, including the CRH (corticotropin-releasing hormone) gene promoter. We have demonstrated that this pattern is induced by a structural motif that we have termed ‘NORE’ (N-Oct-3 Responsive Element). In the current study, we have used a combination of hydrodynamic methods, DNA footprinting experiments, molecular modeling and small angle X-ray scattering (SAXS) to address the following questions: (i) How should the N-Oct-3-binding site within the HLA DRα promoter be read structurally and translated into a new POU domain allosteric conformation? (ii) How do transitions between free and bound conformations occur and what are the molecular mechanisms involved? Our results lead us to conclude that there might exist a continuous spectrum of free and ‘pre-bound’ N-Oct-3 POU conformations. In addition, a specific pair of glycine residues in the linker likely acts as a major conformational switch. Twenty-four base-pair oligonucleotides corresponding respectively to the (−127/−104) and (−57/−34) fragments of the rat CRH gene promoter () and the human HLA DRα gene promoter (), and encompassing the N-Oct-3 POU homodimer-binding sites, were prepared and purified as previously described (). The two sequences are as follows: (CRH) GCTCCTGCATAAATAATAGGGCCC - (DRα) AATTGATTTGCATTTTAATGGTCA A 100 bp fragment encompassing the DRα promoter sequence was generated by PCR using the plasmid pSVODRαlacZ (kindly provided by Dr Goding) and two flanking primers. DNAse I footprinting assays were performed as described (). The N-Oct-3 His-tag DBD was purified as before with the exception of the final gel filtration on a Superdex 75 HR 16/60 column instead of the heparin sepharose chromatography (). Protein samples were concentrated and buffer exchanged with 25 mM Tris pH 7.5, 500 mM NaCl, 1% glycerol, 2 mM DTT, by ultrafiltration using Microcon centrifugal filter devices, then stored at –70°C and thawed prior to the experiments. The concentration was calculated from absorption measurements at 280 nm using an estimated molar extinction coefficient of 12 900 M .cm. The dispersity of each protein preparation was assessed by dynamic light scattering (DLS) measurements using a DynaPro molecular sizing instrument. The N-Oct-3 DBD folding was checked by circular dichroism using a Jobin-Yvon Mark VI dichrograph. Analytical size-exclusion chromatography was performed at 5°C on a Superdex 75 16/60 column (Pharmacia) equilibrated with 50 mM Tris pH 7.5, 0.1 M NaCl, 2% glycerol, 2 mM DTT. The column was calibrated using the Pharmacia low molecular weight calibrating kit containing bovine serum albumin (M = 67 kDa, Rs = 35.5 Å), ovalbumin (M = 43 kDa, Rs = 30.5 Å), chymotrypsinogen (M = 25 kDa, Rs = 20.9 Å) and ribonuclease A (M = 13.7 kDa, Rs = 16.4 Å). Hydrodynamic or Stokes radii (Rs) were calculated from the plot of (–log Kav) versus Rs. Sedimentation velocity analysis was performed using a Beckman XL-I analytical ultracentrifuge and an AN-60 TI rotor (Beckman Instruments). Experiments were carried out at 12°C in 50 mM Tris pH 7.5, 0.5 M NaCl, 2% glycerol, 0.3 mM TCPH at protein concentrations of 1 and 2 mg/ml. Samples of 400 µl were loaded into 12-mm path-length double-sector cells and centrifuged at 42 000 r.p.m. Their absorbance was recorded at 280 nm. The solvent density, ρ, and viscosity, η, were measured at 20°C as 1.027 g/ml and η/η = 1.134 using a density-meter DMA 5000 and viscosity-meter AMVn (Anton PAAR). The values at 12°C were determined to be 1.028 g/ml and η = 1.398 cp. The partial specific volume of the protein, , was estimated from the amino acid composition at 0.731 ml/g using the SEDNTERP program (V1.01; developed by Haynes, Laue, and Philo; available at ). Data processing was carried out using the SEDFIT program (). Continuous distributions were obtained considering 200 particles of frictional ratio 1.5 with sedimentation coefficients between 0.1 and 5.0 S, and using a regularization procedure ( ratio 0.7) (). The non-interacting single-component model analysis was used to determine independently the sedimentation coefficient () and molecular mass () from the sedimentation velocity profiles. The two analyses take advantage of a systematic noise evaluation procedure (,). The corrected sedimentation coefficients, , were derived from the experimental ones () using the following equation: The Svedberg equation was used to relate and the hydrodynamic radius as follows: Models were generated using the modules InsightII, Biopolymer, Discover, Docking, Homology and Decipher (version 2005), run on a Silicon Graphics Fuel workstation, following the main outlines as previously described (). Models of the 24 bp DNA fragments from the CRH and DRα gene promoters were built based on respective local homology with the NORE motif () and the MORE motif [PDB accession number: 1E3O ()] after assignment of the POUs and POUh tetrameric binding sites. The four inter base-pair structural parameters (rise, twist, tilt and roll) were inferred from the homologous templates. The N- and C-terminal regions of the N-Oct-3 DBD were modeled in an extended conformation. The two-step docking was performed as before (). An automated conformational search procedure based on torsion driving was applied to the CRH-induced form of the N-Oct-3 DBD. The Gly 98 Φ and Gly 110 ψ dihedral angles were selected as rotors, and systematically modified by 18° increments in the –180° to 180° range. The 441 resulting conformers were first filtered out using an energy threshold (<2.10 kcal/mol), and then divided into structural families. Each cluster was defined by conformations with similar relative orientations of the POUs and POUh sub-domains and overall backbone configurations superimposable within 4–5 Å. The synchrotron radiation X-ray scattering data were collected on the X33 camera (,) of the European Molecular Biology Laboratory (EMBL) at the storage ring DORIS III (Deutsches Elektronen Synchrotron) using a linear gas detector (). The scattering patterns from the free N-Oct-3 DBD and from the 24-bp CRH and DRα promoter fragments, either free or in complex with the DBD, were measured at several solute concentrations between 2.5 and 8 mg/ml and in 50 mM Tris pH 7.5, 0.4 M NaCl, 2% glycerol, 2 mM DTT. The data were collected at 12°C at a sample-detector distance of 2.3 m covering the momentum transfer range 0.15 < < 3.5 nm ( = 4πsin/, where 2 is the scattering angle and = 0.15 nm the X-ray wavelength). The data collected in 15 successive 1-minute frames to check the radiation damage were normalized and processed using the program PRIMUS (). The difference curves after buffer subtraction were extrapolated to infinite dilution following standard procedures (). The maximum particle dimensions were estimated using the orthogonal expansion program ORTOGNOM (). The forward scattering values and the radii of gyration were evaluated using the Guinier approximation () and by using the indirect transform package GNOM (), which also provides the distance distribution functions () of the particles. The molecular masses () of the solutes were evaluated by comparison of the forward scattering with that from a reference solution of bovine serum albumin ( = 66 kDa). The scattering patterns from the predicted models of the free N-Oct-3 DBD, the CRH and DRα DNA fragments, and their respective complexes, were computed using the program CRYSOL (). Given the atomic coordinates, the program fits the experimental scattering curve by adjusting the excluded volume of the particle and the contrast of the hydration layer surrounding the particle in solution to minimize the discrepancy estimated as follows: The N-Oct-3 DNA-binding domain (DBD) purifies as a single species of 20 kDa molecular mass as judged by SDS-PAGE (A). In order to investigate the oligomerization state and hydrodynamic radius of this POU domain, we first carried out dynamic light scattering (DLS) and analytical gel filtration experiments. DLS measurements recorded at 20°C and at a maximal concentration of 4 mg/ml indicated a low polydispersity and a narrow particle size distribution diagram corresponding to a hydrodynamic radius of 29.3 Å (B). The purified N-Oct-3 POU domain eluted from a FPLC-size exclusion chromatography column between the 43 and 25 kDa calibration proteins and the elution volume served to calculate its Stokes radius (C). The resulting Rs value of 27.6 Å was very similar to that calculated by DLS, but significantly higher than those of globular proteins of an equivalent molecular weight. This indicates the presence of either a dimer or an elongated monomer in solution. The N-Oct-3 DBD was then submitted to sedimentation velocity analysis, and the data were processed as described in the Materials and Methods section. A selection of sedimentation profiles performed in the same conditions, along with their best-fits using a single component, are shown in A, the corresponding residuals being displayed in B. Identical sedimentation coefficients were obtained (1.84 S) at the two concentrations used (1 and 2 mg/ml), and the deduced molecular mass (21 kDa) indicates, when compared with the theoretical mass (19.9 kDa), that the N-Oct3 DBD is a monomer. In addition, the analysis of the sedimentation profiles in terms of a continuous distribution of elongated particles showed narrow single peaks at both concentrations (C). This clearly demonstrates the homogeneity of the solution and the lack of any association–dissociation processes, thereby confirming the monomeric status of the free N-Oct-3 DBD. Thus we can conclude that the N-Oct-3 POU homodimers which bind to a variety of DNA targets () do not exist prior to complex formation, but are a consequence of specific interactions with target DNAs. The question then arises as to whether the elongated shape of the free N-Oct-3 DBD indicated by the hydrodynamic data reflects a single conformation or represents the average of a collection of conformers. In addition, we would like to determine the molecular mechanisms responsible for the transitions between the free and DNA-bound conformations. To attempt to answer these questions, we have performed a comparative analysis of two regulatory conformations of N-Oct-3 POU, either induced by the NORE motif of the CRH gene promoter () or by an element of the HLA DRα gene promoter. In the latter case, it was first necessary to characterize the interaction between the N-Oct-3 POU and its DNA target. We have previously shown () that the N-Oct-3 POU domain can adopt three different conformations and corresponding homodimerization patterns in response to the particular distribution of potential POUs and POUh tetrameric binding sites which characterize the respective PORE, MORE and NORE motifs evoked earlier. In the same report, we defined a structural framework suitable for the analysis of any interaction between the N-Oct-3 POU domain and a DNA target. Most importantly, the POUs and POUh tetrameric binding sites for each monomer are non-contiguous and on opposite strands in the MORE mode, whereas they are contiguous and on the same strand in the PORE mode. This results in a different relative positioning of the POUs and POUh sub-domains within each monomer between the two modes. Finally, the NORE motif designates the 14-bp sequence element TNNRTAAATAATRN (N: any nucleotide; R: purine residues) which is common to a set of neuronal promoters, including the CRH gene promoter, and which is capable of eliciting a novel homodimerization mode exclusive to the N-Oct-3 DBD. Both the NORE and PORE motifs elicit a ‘POUh-dominant’ mode of N-Oct-3 DBD homodimerization with a strong anchoring into the DNA minor groove. However, in the case of the NORE mode, the two POUh-binding sites are overlapping, which explains the non-cooperative character of the homodimerization. DNAse I footprinting is a particularly valuable tool to determine which homodimerization mode is elicited by a given DNA regulatory element. Bearing in mind the strong correlation between N-Oct-3 over-expression in melanomas and the up-regulation of HLA-DRα gene expression (,), we used this approach, coupled to molecular modeling, to analyze N-Oct-3 binding to the HLA-DRα gene promoter. Electrophoretic mobility shift assays (EMSA) showed that the N-Oct-3 POU domain binds as a non-cooperative homodimer to the DRα DNA, a 24-bp DNA fragment of the HLA-DRα gene promoter (), with an effective dissociation constant of 5 × 10 M for the first monomer (see A legend) and an apparent dissociation constant ≤ 2.6 × 10 M for the second monomer [see B legend; ()]. DNAse I footprinting of the first N-Oct-3 DBD binding to a promoter fragment encompassing this high-affinity binding site reveals a total protection of both DNA strands (lanes 1 in A and B). We therefore deduce that the relative positioning of the POUs and POUh sub-domains within this first bound monomer must be elicited by a MORE-type motif, the only one with POUs and POUh-binding sites on both strands of the DNA. A MORE motif is characterized by two strong POUs anchoring sites on opposite DNA strands and on either side of the pseudo-dyad axis. The sequence of these binding sites is most often ATG(/A)C, but an ATNN motif is sufficient to establish the highly specific set of interactions with the conserved Gln and Thr residues of the POUs recognition helix. Based on the DNAse I footprint, the A12T13T14T15 tetramer on the upper strand and the overlapping A12BT13BG14BC15B tetramer on the lower strand of the HLA-DRα gene promoter possess the appropriate structural requirements for the two POUs-binding sites in the MORE configuration (C). In line with this, the non-cooperativity of the homodimerization observed by EMSA (B) is consistent with the overlap of the two POUs-binding sites. Furthermore, the mutagenesis of the A12T13T14 triplet is sufficient to abolish the binding of both monomers (data not shown). Following the assignment of the two POUs-binding sites as A12T13T14T15 on the upper strand and A12BT13BG14BC15B on the lower strand, the two corresponding POUh-binding sites can now be predicted as G14BC15BA16BA17B and T14T15T16A17 respectively, based on the known MORE motif organization (,). In this mode, each POUh-binding site overlaps the POUs-binding site of the other monomer on the same strand (see C and its legend). The extent of the DNAse I footprint on the lower strand as a consequence of the first monomer binding designates G14BC15BA16BA17B as the first POUh-binding site (lane 1 in B), and hence the A12T13T14T15 as the first POUs-binding site. This implies that the A12BT13BG14BC15B tetramer on the lower strand is the second POUs-binding site and the T14T15T16A17 tetramer on the upper strand is the second POUh-binding site. It is important to underline that, as for the so-called canonical sequence of the human immunoglobulin heavy chain gene promoters IgG V (,), the prototypic octamer sequence ATGCAAAT on the lower strand is not ‘read’ as a single continuous POU-binding site but, instead, as the second POUs-binding site (ATGC) overlapping the first POUh-binding site (GCAA). As a consequence, the terminal AT is still cleaved by DNAse I since it does not take an active part in the interaction (see green-colored marking in B and C). Now that the POUs and POUh-binding sites have been assigned, the bound structure of the HLA DRα promoter DNA fragment can be built and docked with the corresponding sub-domains. The resulting model is displayed in A and B. It is known that the generic MORE mode can accommodate variable spacings between the two POUs insertion sites. For example the ‘MORE+2’ mode, corresponds to a 2 bp spacing (). Following this nomenclature, the DRα/DBD complex represents a new MORE subtype, which can be designated by ‘MORE-2’. In this mode, the two POUs DNA recognition helices are inserted into overlapping sites in the major groove (see the red-colored star in B). A comparative analysis of the N-Oct-3 POU conformation induced by the DRα DNA (C) with that induced by the CRH DNA (D) taking the position of the POUs as a fixed reference, reveals that the two POUh sub-domain orientations can be superimposed by an ∼180° rotation around the linker taken as a virtual axis. Before dealing with the structural determinants of N-Oct-3 linker flexibility, we first need to recall its distinctive features. Using circular dichroism, we previously observed an increase in the α-helical content of the N-Oct-3 DBD when binding to its DNA targets, in contrast to the Oct-1 DBD (). Since the only significant difference between these two highly conserved DBDs is their respective linker sequences, we engineered chimeric proteins where the N-Oct-3 and the Oct-1 linkers were interchanged. This showed that the replacement of the N-Oct-3 DBD linker by that of Oct-1 abolished the increase in α-helical structure, whereas the replacement of the Oct-1 linker by that of N-Oct-3 resulted in the typical increase in the α-helical content following protein/DNA complex formation. Since a number of reliable secondary structure prediction methods indicated that the heptapeptide motif IDKIAAQ specific to the N-Oct-3 linker could adopt an α-helical structure, we built another set of chimeric proteins where this heptapeptide was removed from the N-Oct-3 linker and embedded within the Oct-1 linker. As the results were similar to those for the entire linker interchange experiments, we concluded that the ability of the N-Oct-3 linker to adopt an α-helical structure when binding to a DNA target could be ascribed to the IDKIAAQ motif (see its location in the DBD sequence in A). We now show that the potential secondary structure of this heptapeptide motif can also be stabilized independently of DNA binding, when free DBD concentrations are greater than 0.7 mg/ml (see Figure S1 and its legend), which are the conditions of the hydrodynamic and SAXS experiments reported here. Note that the link between protein folding and molecular concentration has been revealed in a number of recent works [see for example (,)]. Thus the N-Oct-3 linker has the characteristics of a ‘helical linker’ as defined by George and Heringa based on an extensive compilation of inter-domain linkers (). Interestingly, the helical heptapeptide IDKIAAQ is preceded by the 4-residue motif SPTS (A), shown to form a β-turn in a number of proteins and polypeptides, the structures of which were solved by crystallography or NMR (). A crucial feature of hinge residues is that they have very few packing constraints in their main chain atoms (,). As such, the Gly residues are well suited to promote hinge motion (,). The two Gly residues present in the N-Oct-3 DBD linker (A) could therefore act as major molecular pivots in the conformational transitions. To examine this further, we performed automated conformational searches by systematically sampling the ϕ and ψ dihedral angles of Gly 98 and Gly 110, using the CRH-bound conformation as a starting structure. We found the combination of Gly 98ϕ and Gly 110 ψ dihedral angles to be the most efficient to explore the N-Oct-3 DBD conformational space (see the Materials and Methods section and Figure S3A and B). After filtering using an energy threshold, the resulting conformers could be clustered within a discrete number of conformational families, based on overall R.M.S. values of 4–5 Å and corresponding to different relative orientations of the POUs and POUh sub-domains such as those displayed in B–D. In order to identify potential free forms amongst these structures, we first compared their calculated radius of gyration () to the free N-Oct-3 DBD hydrodynamic radius. To select the most likely candidates, we then combined molecular mechanics with SAXS methodology following the main outlines of a recent study (). Processed X-ray scattering patterns corresponding to the free N-Oct-3 DBD are presented in A and B (data groups 1), alongside those from the free DNA fragments (data groups 2) and from the equimolecular N-Oct-3 DBD/DNA complexes (data groups 3). The structural parameters computed from the experimental data, including the radius of gyration () and maximum particle dimension (), are displayed in . The estimated effective mass () of the free N-Oct-3 DBD agrees within experimental error with the value expected from the sequence (), confirming that the protein is monomeric in solution. The distance distribution functions computed from the experimental data () emphasize the elongated shape of the free form(s), and the similarities between the gyration radii of the free N-Oct-3 DBD and of its complexes with each promoter DNA fragment. Note the good agreement between the free N-Oct-3 DBD gyration and hydrodynamic radii. In all cases, the theoretical scattering patterns of the predicted structures were computed using the program CRYSOL and then compared to the experimental data. The accuracy of the fit was assessed by the discrepancy value χ as explained in the Material and Methods section, where typical values between 0.8 and 1.1 indicate good agreement. Thus, the computed scattering curves corresponding to the models of both the CRH DNA fragment and the N-Oct-3 DBD/CRH complex agree well with the respective experimental curves, with discrepancy values of 1.05 and 1.09, respectively (data groups 2 and 3 in A and ; Figure S2A). The same observations can be made for the models of the DRα DNA fragment and the N-Oct-3 DBD/DRα complex (data groups 2 and 3 in B and respective discrepancy values of 0.82 and 1.09 in ; Figure S2B). Fitting the computed scattering curves of the N-Oct-3 DBD in the predicted CRH- or DRα-bound conformations with the experimental data for the free N-Oct-3 DBD yields slightly higher discrepancy values (see respective χ values of 1.18 and 1.23 in and data groups 1 in A and B). In order to accurately interpret this in terms of similarities versus differences between free and bound conformations, we must first build a referential of free-form models. For this, we systematically computed the theoretical scattering curves of the molecular mechanics-derived structures and fitted them to the free DBD experimental data. According to their values in the 1.06–1.09 range, a number of conformers appear as good candidates to represent free N-Oct-3 DBD conformations. These can be divided into two distinct clusters which are themselves part of larger conformational families, ‘FI’ and ‘FII’, defined by respective overall R.M.S. values of 4.9 Å (B) and 4.4 Å (C). Importantly, the value dispersion observed in both cases, 1.06–1.19 and 1.06–1.27 respectively, is compatible with the conservation of a given overall POU domain conformation within each family. A more detailed analysis indicates that each conformational family contains structural sub-classes characterized by a particular distance between the POUs and POUh recognition helices (‘RH’) within the 18–35 Å range. Interestingly, the conformers with the lowest RH (Figure S3B) tend to be less energetically stable (Figure S3A), but are closer to the respective CRH- and DRα-bound conformations for which RH is comprised within the 15–20 Å range (C and D). Taken together, these results imply that the two populations of putative free forms, F1 and FII, most likely coexist, and also that there could be a structural continuum running from free to less stable ‘pre-bound’ conformations. In line with this, the fitted scattering curve of the CRH-bound modeled structure is very close to that of ‘Cf 183’ (see the respective red- and turquoise-colored curves of data group 1 in A, and the corresponding values of 1.18 and 1.09 in ), Cf 183 being the best FI representative (E). Similarly, the fitted scattering curve of the DRα-bound modeled structure is very close to that of ‘Cf 194’ (see the respective blue- and magenta-colored curves of data group 1 in B, and the corresponding values of 1.23 and 1.08 in ), Cf 194 being the best FII representative (F). By contrast, the fitted scattering curve of ‘Cf 221’ significantly deviates from the free N-Oct-3 DBD experimental data with a value of 1.90 (see the dashed green-colored curve in data group 1 in A and B, and ). Indeed, this conformer (G), with its higher (32 Å) and RH (50 Å) values, cannot represent the free form and belongs to a large conformational family of extended structures, characterized by RH values within the 40–50 Å range (D). Model fitting against experimental SAXS data is a useful means to interpret scattering information in terms of higher-resolution structures (). Fitting of multiple models generated by molecular mechanics or dynamics has also been applied to analyze conformer ensembles in solution, especially in relation to protein unfolding (). Along these lines, a recent report [see () and references therein] has explored how multiple well-defined protein conformations in a sample influence the scattering data. Test cases were established, based on simulation of SAXS data from reconstituted ensembles of protein structures, such as ensembles comprising various weighted proportions of the extended and collapsed states of calmodulin, a protein comprising two globular domains connected by a flexible helical linker. One of the main conclusions of this study is that the ability of modeling to differentiate static structures from dynamic structures depends strongly on the extent of the variability of the ensemble. Hence, an low-resolution model of the free N-Oct-3 DBD can be expected to reflect distinct properties from respective members of the FI and FII conformational families, but probably not from members of the same family. Indeed, a molecular envelope of the N-Oct-3 DBD generated using the GASBOR program () can accommodate the DRα- and the CRH-bound conformations at different sites (see Figure S4 and its legend). As these conformations bear similarities with the respective overall structures of the FI or FII families’ members, this lends support to the likely coexistence of these two conformational families, inasmuch as they are energetically equiprobable (see Figure S3 and its legend). Initially structural studies performed on the POU-type DNA-binding domain showed that individual POUs and POUh sub-domains could be considered as rigid bodies when interacting with DNA (,,). The adaptability of several POU proteins to a variety of DNA targets was then ascribed to the flexibility of the linker joining the POU sub-domains (,). However, despite the critical importance of the linker with regards to the molding of specific regulatory POU conformations to the target DNA, no detailed molecular mechanism for this flexibility has so far been proposed. One of the main reasons for this of course is that neither Oct-1 nor Pit-1 POU linker structures can be resolved in the available crystallographic data derived from POU/DNA complexes. The N-Oct-3 DBD linker has dual structural properties. On the one hand, it contains a helical peptide motif, in common with approximately half of the known inter-domain linkers (), which might constrain the relative orientation of the two POU sub-domains. On the other hand, this linker also functions as a hinge region, as best exemplified in the transition between the CRH- and the DRα-bound conformations. A number of studies dealing with hinge motion () designate the pair of Gly residues present in the linker as potential key-players in the N-Oct-3 DBD conformational transitions. Based on these working hypotheses, we have combined various hydrodynamic and SAXS data with the results of a conformational search through torsion driving. We have shown that the linker flexibility resulting from rotations around this pair of Gly residues is sufficient to generate the transitions between the free and bound conformations, whilst at the same time respecting the local structuring of the linker. We have identified two families of putative free N-Oct-3 POU conformations, which can be interconverted by rotation around a virtual Gly–Gly hinge axis. As specified earlier in the text, the distances between the DNA recognition helices (‘RH’) in these conformers lie within the 18–35 Å range, which favors the concerted DNA-binding activity of the two POU sub-domains. There might exist an equilibrium between these two families of putative free conformers and, for each family, between best free form representatives and less stable ‘pre-bound’ conformers. We propose that NORE- or MORE-2-type DNA motifs select conformers closer to the final CRH-or DRα-bound conformations, respectively. Note that the importance of the Gly residues does not exclude the contribution of other residues to the overall flexibility of the linker, especially in the final adjustements required upon DNA binding. In conclusion, our results indicate that regulatory DNA regions most likely select pre-existing N-Oct-3 DBD conformations, in addition to molding the appropriate DBD structure. More generally, our study emphasizes the necessity not only to employ a structural reading of nucleic regulatory sequences but also to integrate information about protein flexibility when predicting functional structure. Indeed a number of recent studies address the critical issue of the indirect readout of promoter DNA sequences (for example see (), whilst new concepts and methods are emerging to explore protein flexibility and allostery (,). Along these lines, combining an ensemble optimization method with SAXS is a highly promising approach as perfectly illustrated in our recently published study (). p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Understanding genetic regulatory networks is a key to elucidating various biological processes. In eukaryotes, an integrated regulatory network comprises transcription factors (TFs), target genes and their relationships. Although recent studies have also implicated small non-coding RNAs in the regulation of gene expression at the post-transcriptional level (), TFs and their corresponding target genes are still regarded as key components of regulatory networks. Screening new transcription factor binding sites (TFBSs) is the most common approach to identifying transcription factor target genes. Various experimental methods have been applied to search for new TFBSs, e.g. electrophoresis mobility shift assays (), enzyme activity analysis of cellulose D (CELD) fusion protein () and the high-throughput Chromatin Immunoprecipitation (ChIP) chip approach (). However, such experiments are difficult to perform on a large scale because they are costly and time-consuming. For example, there are nearly 2000 TFs in the first complete sequenced model plant species, (), but fewer than 20 of these have been experimentally validated so far. To circumvent the low efficiency of experimental methods, many computational approaches have been proposed to screen for new TFBSs. Moses . () and Wang . () reported TFBSs prediction algorithms based on phylogenetic data and multiple alignments of nucleotide sequences among different species. Anand . () proposed the prediction of TFBSs using an n-gram algorithm by analyzing the results of single base substitution experiments. Holloway . combined gene expression data with genomic sequence data to predict new DNA binding sites (,). Hoglund . () discussed the prospect of employing 3D structural information about protein–DNA complexes to improve the prediction of binding motifs. These methods have yet to address the problems of weak conservation in the upstream regions of genes and/or lack of protein structural data. Known TFBSs are also used to identify transcription factor target genes; Position-Specific Scoring Matrices (PSSMs) are generally created to improve prediction performance (,). However, these approaches overlook the interdependence and variable distances among different bases (), and the genetic contexts of TFBSs in the whole cell are ignored in the computational analysis. As a result, there is a very high frequency of false positive hits, especially when only one PSSM is applied, e.g. more than 30% of genes in the genome could be considered auxin response factor (ARF) targets because they all have the ARF-binding site ‘TGTCTC’ in their promoter regions. Several improvements have been made to reduce false positive predictions. Frith . () introduced genetic context (i.e. a set of functionally related PSSMs) into the model to fine-tune the PSSMs, thus improving the prediction performance. Suckow . () introduced variable gaps between two or more different motifs. Other similar attempts have also been reported, such as including a spacing rule between the TFs (), and limiting the numbers of each contributing TF and combinations of TF positions (). Unfortunately, these improvements must rely on known TFBSs, and this remains rather a sparse resource. In this article, we propose a novel systematic computational approach to predicting transcription factor target genes (TFTGs) directly. Our approach does not necessarily predict new DNA binding sites, which other studies have shown to be difficult. Utilizing known binding sites and gene co-expression data, we modeled the prediction problem as a ‘yes’ or ‘no’ classification task and implemented the classifiers with support vector machines (SVMs). Here the ‘feature generation, feature selection, feature integration’ paradigm was followed to build the SVM classifiers. The promoter sequences of both target and non-target genes within 1000 bp from the transcription start site (TSS) were first profiled by a novel reverse-complementary position-sensitive (RCPS) -gram profiling algorithm, which refers to the position of a known binding site or genes TSS. Then, by applying measurements of the information gain (reduction in entropy), representative RCPS -grams of positive and negative samples (target and non-target genes) were selected to create a vector space in which the promoter sequences were represented. Finally, these vectorial -grams were fed to the SVMs to build prediction models. We used the proposed approach to predict ARF target genes on the basis of published co-expression data () and obtained satisfactory results. Using 10-fold cross validation, the AUC value (area under curve) from our model reaches around 0.73, which is significantly higher than that from a random guess (0.5). We extracted sequences up to 1000 bp upstream from gene TSS from genome sequences () by referring to gene locus data (TAIR6, 01/22/2004 release, ). Auxin response factors (ARF) are capable of activating/suppressing the expression of primary auxin response genes at the transcriptional level by recognizing the specific ‘TGTCTC’ box binding site (,). Of the 7720 genes on the Affymetrix 8K AG Chip, 2787 with the ‘TGTCTC’ motif or its reverse complement ‘GAGACA’ in their upstream sequences were chosen as candidate ARF target genes. Goda . () reported that 637 genes are possibly affected by IAA/BL on the basis of a gene expression study using Affymetrix 8K AG-Chips. Therefore, we selected the 186 genes validated by Godas experiments from the 2787 candidate genes as ARF-target genes and treated the remaining 2601 as ARF-non-target genes (highlighted in ). Their 1000 bp upstream sequences were analyzed by the following steps. An -gram is a subsequence of n letters from a given string (). The -gram profiling is a popular technique for converting natural language strings into histograms, i.e. generating statistics of all the n-grams occurring in a sequence stream. It has been applied to the recognition of splice sites in a genome (). The current -gram algorithm is generally little affected by the order in which different -grams occur; in other words, the -gram profile generated only includes -gram frequencies. However, this is not necessarily valid for our TFTG prediction problem. According to existing models of the regulation of gene expression, the same DNA motif at different positions in the upstream region may exert different regulatory effects via a specific TF. In view of this, we extend the definition of an -gram profile into a position-sensitive -gram (PSNG) profile, formalized as follows. Definition 2 (PSNG Profile): The PSNG profile of an -length sequence = , … … … relative to a -length reference sequence = … , denoted PSNP(), is the enumeration of all possible PSNG, psng(), in the sequence. If the same numbers of -grams are counted on the two flanks of a reference sequence, this number is represented as C, which is set to either ‘ − 1’ or ‘N − − + 1’ at most. An example is given below. and its position-sensitive tri-gram profile is A potential binding site remains active over a range of nucleotides, not just at the point of the promoter region, because the 3D conformations of both protein and DNA are flexible. Thus, the same -grams with different but neighboring positions may also have the same function. In view of this consideration, a parameter , the position-sensitive factor, was introduced to control the position-sensitivity of PSNP profiling. The -grams located in neighboring bp regions are considered to be the same -gram as long as they have identical sequences. The two examples above, the bi-gram and tri-gram profiles, were generated with = 1. Considering the base-pairing property of DNA double strands, a specific TF may bind to either strand of a ds-DNA molecule. We therefore extended the position-sensitive n-gram to a reverse-complementary position-sensitive n-gram, denoted rcpsng(n) ≡ (s/rcs/), e.g. (ATCG/CGAT|3). illustrates the process of building reverse-complementary position-sensitive four-grams from a given sequence. The standard -gram algorithm produces (in principle) 4 possible -grams when DNA sequences are profiled. When our reverse-complementary PSNG profiling algorithm is applied, the maximum possible number of n-grams would be 4 × × 2 if we did not consider -gram repetition. The number of -grams is equivalent to the dimension of vector spaces in SVM algorithm. The demand of computation power will increase exponentially when the dimension of vector spaces increases. In our model building phase, our test indicated the optimal combination should include 4, 5, 6, 7, 8 and 9-grams. Thus, the total numbers of -grams reaches , i.e. SVM needs to search dimension of space. Therefore, it is not practical to train an SVM classifier with such a large volume of reverse-complementary PSNGs. An effective feature selection process, which removes noise and outliers, would improve the prediction performance and reduce the computational cost. In our study, we used the measurement of information gain (IG) to select representative features from both positive and negative samples (target and non-target genes). The idea of IG is based on the evaluation of entropy in fuzzy datasets; it represents the change in entropy after a specific signal (a particular rcpsng -gram) is observed (). Let S be the set of DNA sequences being studied and T be the classes of sequences; in particular, for the target gene/non-target gene binary cases ( = 2, target and non-target) in our study. With a conserved region , the sequence set has two distinct values: are the sequences containing the conserved region , and are the sequences that do not contain . The difference between and is then used to define the information gain by the partitioning of according to : A higher IG value indicates greater information significance, and thus suggests that the corresponding -gram is better able to represent an important feature of the sequence. Here, we chose the top ( = 500, 1000, 1500 and 2000) -grams to represent the features of the DNA sequences, thereby constructing a -dimensional vector space. The upstream DNA sequences were then converted into the -dimensional vector space according to their -gram profiles. is an example showing the conversion of a sequence into vector format in terms of these featured -grams. A large number of classification algorithms have been applied successfully to text classification and information search tasks. In some cases, heuristic learning-based supported SVM perform better than other machine learning methods, because both positive and negative samples are utilized to train the models (,). Given a set of linearly separable vectors = {, , …, }, where denotes the -dimensional vector representing the corresponding training sequence and = total number of input sequences, each belonging to one of the two classes labeled ∈ { − 1, + 1}, (−1: 1: ), SVM seeks a separating hyper-plane, = + , that divides into two parts, each containing vectors that have the same class label only by estimation on an optimal separating hyperplane (OSH) that has the maximal margin in both parts. This is done by minimizing (1/2)|| ||, subject to  (W−X) ⩾ 1. Those vectors closest to the OSH are termed support vectors. The SVM-light package () () was used to construct the SVM classifiers. We also compared the linear kernel with other kernels, such as polynomial and sigmoid kernels of the SVM. In our trial-and-error tests, the linear-kernel SVM yielded satisfactory prediction accuracy with relatively low computational cost. We applied 10-fold cross-validation to estimate the models performance. The entire dataset was randomly divided into 10 groups. Each time we chose a different group of sequences as the test group and the remaining nine as training groups; representative features were simultaneously extracted to create the corresponding vector space from the nine training groups. This procedure was repeated ten times to test all ten groups of sequences. The accuracy under a specific threshold value was evaluated on the basis of the following criteria: Because there were 14 times as many non-targets as target gene samples in the datasets, the high ratio of non-target genes may have generated a high accuracy value, even though the model performance was weak. Therefore, we introduced sensitivity and specificity to evaluate the models, which were computed as follows: The sensitivity is the ratio of target genes correctly predicted in the target gene dataset. The specificity is the ratio of non-target genes correctly predicted in the non-target gene dataset. Obviously, these two measurements generally vary when different SVM thresholds are applied. That is to say, by adjusting the threshold, we can obtain higher sensitivity and lower specificity, or vice versa. We also tested model performance under different settings, e.g. different position sensitive factor , length of -gram , number of -grams , number of representative -grams and SVM kernels. In our experiments, reverse-complementary PSNG profiles were generated with set at 4–9. The numbers of -grams () counted on each flank of the central motif ‘TGTCTC/GAGACA’ was set at 50, 75, 100, 125, 150 or 175. For = 100, = 4, the total number of reverse-complementary PSNG (with = 1) was 27 038 in the upstream regions of both ARF-related and ARF-unrelated genes. We applied IG algorithms to select representative features from the reverse-complementary PSNG generated, as detailed in the Materials and Methods section. lists the top 10 four-grams with highest IG values and their frequencies in ARF-related and ARF-unrelated genes. Our 10-fold cross-validation gave the AUC values of the SVM models, which were constructed from various combinations of and . In our experiments, the best model achieved an AUC value of 0.73, with = 4–9, = 100, = 100 and = 1000. shows the raised ROC curve of the models generated. Since the ROC curve represents the relationship between sensitivity and specificity under different thresholds, weak models or completely random guesses would give a straight line at a −45 degree angle (i.e. AUC = 0.5), whereas our raised ROC curve (AUC = 0.73) indicates that our model has significant predictive power; ARF target and non-target genes can be discriminated. The detailed optimal models and their ranking are available at our online supplementary page, . To test how the number of n-grams, , affects the performance of our models, we generated a set of SVM models with = 50, 75, 100, 125, 150 and 175. a shows an AUC versus curve. The best AUC reaches 0.73 when = 100, suggesting that -grams in the region of 100 bp from the core motif (TGTCTC/GAGACA) may be important in the recognition and binding of TF. We also evaluated the effect of position-sensitive factors () on model performance by varying . b shows an AUC versus -plot. The optimal -value was around 100; the models AUC value declined at higher or lower -values. The optimal model was used to predict ARF target genes in . In summary, of the total 26 751 non-redundant TAIR6 genes, 12 559 were first retrieved by searching for the ‘TGTCTC/GAGACA’ binding motif in their promoter regions and then were sent to our SVM for prediction and ranking. We examined the top 1000 genes in the SVM output and found 172 known ARF target genes listed. We also manually curated the remaining 828 genes and found that 574 have been reported as possible auxin-related genes (the gene list is available at online supplementary page, ). This cross-validation result suggests that the model has potential for predicting new target genes of a transcription factor or its family. Prediction of the target genes of transcription factors often suffers from high false positive rates (). In our study, 2787 of the 7720 genes observed would be identified as possible ARF target genes if we only utilized the known ‘TGTCTC/GAGACA’ binding site for prediction. However, Affymetrix AG-chip co-expression analysis suggested that only 186 were true ARF target genes (,). That is to say, the false positive rate would reach 93% if we only employed the known TFBS ‘TGTCTC/GAGACA’ to screen target genes of ARF. Combinations of known TFBSs are commonly used to improve the prediction performance of TFTG (14–17). We compared our approach with a typical algorithm of this kind, cluster-buster (). In our comparison, 20 transcription factor families were found from the known auxin-related genes, and then 13 PSSMs were created from these families and inputted into the cluster-buster program as matrix files. The cluster-buster program outputted a score for each promoter region in the dataset. Our analysis shows that the AUC value of the cluster-buster algorithm can only reach 0.51, which is obviously lower than our approach (data are available at our online supplementary page). The result of this comparison indicates that, owing to the information from co-expression analysis, our approach performs very well even though the associated binding motifs are unknown. The comparison also shows that the cluster-buster algorithm may be useful for identifying TFTGs when most of the associated binding motifs are known. Since gene co-expression data from microarray experiments are rapidly accumulating in public repositories, our approach holds promise for TFTG prediction. Although -gram algorithms have been applied to the analysis of biological sequences (), relative distance information is generally not considered in the n-gram. In this article, by defining novel reverse-complementary PSNG, we have introduced positional information into the standard -gram algorithm for the first time. Here, a conserved binding site, i.e. ‘TGTCTC’, served as a reference point for relative positional information, further narrowing the scope of searches for potential interacting regions in promoters. The inclusion of positional information reflects the mechanism of interaction between transcription factor proteins and DNA, in which multiple transcription factors usually interact to recognize their corresponding binding sites. For example, ARF TFs possibly interact with members of the bZIP family, which are able to recognize the ‘CCTCG’ motif near ‘TGTCTC’ (,). The 3D structures of TF complexes indicate that their corresponding DNA binding sites are sensitive to relative distance (). It is reasonable to include positional information when we profile -grams. On the other hand, since protein/DNA 3D structures are flexible to some degree, motifs with slightly shifted positions may still have the same binding function. Therefore, positional information that is too loose or too stringent may affect our models performance. Here, we introduced the -factor to represent the sensitivity to position differentiation. Our results (b) suggest that the best prediction results could be achieved when P was set to a small region ( = 100). In the present work, we have employed IG to choose representative -grams between positive and negative training samples. A high IG value represents a strong signal to noise ratio, which indicates that the corresponding n-gram is more valuable for training SVMs. IG determines the relative difference in occurrence of -grams between two groups of sequences. Moreover, compared with another popular measurement, the - test (data not shown), the IG test is more capable of simultaneously filtering out -grams with too low a frequency of occurrence. We considered low frequency -grams as random noise even if their occurrence was significantly different between the two groups of sequences. Comparative genomics and phylogenetic studies suggest that gene co-regulation is highly conserved in the evolution of eukaryotes and prokaryotes (). Our model should be applicable to other species, not just . However, in view of the variation in regions between genes, this claim needs to be further verified by biological experiments. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The DNA-dependent RNA polymerase of bacteria is the sole enzyme capable of producing messenger, transfer and ribosomal RNA by catalyzing the 5′ to 3′ synthesis of phosphodiester bonds between adjacent ribonucleoside triphosphates. In , core DNA-dependent RNA polymerase consists of four different subunits and has the composition αββ′ω. It was discovered that an additional and separable factor, the sigma factor (σ), was required for transcription of certain types of DNA (). Numerous sigma factors have been described in and other prokaryotic organisms () since then. The seven known sigma factors are σ, σ, σ, σ, σ, σ and σ. The holoenzyme complex (Eσ) (,) constituted by core RNA polymerase (E) together with a sigma factor (σ) is required to initiate transcription at specific DNA sequences termed promoters. Through the specificity of its σ subunit, holoenzyme is directed to two conserved DNA hexamers centered ∼10 and 35 bp upstream of the transcription start to initiate transcription. Each sigma factor recognizes and directs RNA polymerase to a different set of promoters. While sigma factors provide the primary interaction with those consensus two-block promoter DNA sequences and play a central role in the regulation of global gene expression, additional transcriptional activators such as FIS (,) and CRP (,) can be utilized to strengthen the promoter–holoenzyme interaction. Flagellar, chemotaxis and motility genes are organized into large complex units. The genes that constitute the flagellar regulon are expressed in a cascade that closely parallels the assembly hierarchy of the flagellar structure (,). The flagellar operons are divided into three gene classes with respect to this transcriptional hierarchy (classes I, II and III or early, middle and late classes, respectively). At the top of the hierarchy is the FlhDC master operon that encodes FlhDC complex as a positive transcriptional activator of σ-dependent transcription from class II promoters which also include the promoter. σ is made, binds to core RNA polymerase, and then transcribes from class III promoters. The synthesis of the flagellar system can be tightly regulated by a master regulator, the transcriptional activator FlhDC complex, as well as by the secondary regulator, the alternative sigma factor, σ. In the previous characterizations of the flagellar regulon, genetic mutagenesis analysis was used as a main approach to study those non-flagellate mutants which included spontaneous, Mu phage-induced, and some transposon-induced flagellar mutants to define genes involved in the formation of a functional flagellar apparatus (). Although several new flagellar genes have been recently identified in (), no additional experiment has been published so far to systematically study the FlhDC and σ regulons in . Due to the high transcriptional and translational level of these two regulators in log-phase, our basic strategy was to delete these two regulators as well as to minimally perturb steady-state cell growth by moderate induction of these genes in defined MOPS minimal medium. We then monitor global RNA transcript abundance change using Affymetrix GeneChip Antisense Genome Arrays. While transcription by σ is known to be modulated by the anti-sigma FlgM (), we will not consider effects of FlgM here. We believe that the genes which are dependent on σ can be identified by the σ deletion or induction experiments presented. In addition, to characterize how the activity of motility is regulated by different carbon sources, we measured the transcriptional level of FlhDC and FliA as well as the level of several well-known genes under the direct control of these two regulators in a range of carbon source conditions by a quantitative RT-PCR analysis. The correlation between the activity of CRP and the motility of cells indicates that activation of CRP plays an important role in triggering ‘foraging’-like behavior () to actively search for better conditions as the quality of the available substrate(s) decreases. On the basis of these first systematic studies of the FlhDC complex and σ regulons in , including the effects of carbon source conditions on motility of cells, we gain insight into the complex networks regulated by these two regulators and how they contribute physiological adaptation to the changes in the external environment. All reagents were purchased from Sigma Chemical Company (St Louis, MO) unless otherwise indicated. 10X MOPS minimal media was prepared as described in Neidhardt (). The media was filter sterilized through a 0.2 μm filter and stored at 4°C. The defined media for log-phase cell growth contained 1 X MOPS minimal media, 0.1% glucose, 0.66 mM KHPO. Because the Genechip probe set is based on the sequenced K-12 strain MG1655 (λ F ilvG rfb50 rph-1, prototroph) (), we chose this bacterial strain for use in our study. In order to disrupt the expression of FlhDC/σ in , we used a simple and highly efficient method (,) to prepared in-frame deletion strains for FliA (σ) and for the master regulator FlhDC. In this procedure (as shown in supplemental material Figure S1), we generated PCR products by using primers with 60- to 70-nt extensions that are homologous to regions adjacent to or gene(s) and a template plasmid carrying kanamycin-antibiotic resistance genes. Using linear DNA to do homologous recombination in requires the phage lambda Red recombinase (λRed system), which is synthesized under the control of an inducible promoter on a low copy number plasmid. Deletion mutants of the gene or operon were isolated as kanamycin antibiotic-resistant colonies after the introduction the respective PCR products into bacteria carrying a λRed expression plasmid. The replication of the temperature-sensitive plasmid pKD46 was inhibited and the loss of this plasmid in mutant strains occurred when clones were grown at 43°C. For controllable induction of individual regulators , we used the P promoter which is controlled by the repressor TetR to construct these overexpression vectors as described previously (). A downstream gene can be induced in the presence of inducer, aTc. All strains used in this study were derivatives of K12 MG1655. strains were grown overnight in MOPS minimal media at 37°C in an air shaker with vigorous aeration (225 rpm). Two milliliters of the overnight culture was used to inoculate 100 ml of fresh MOPS minimal medium. For preparing the total RNA for microarray experiments, 15 ml samples of culture (corresponding to 7.5 × 10 cells) were taken for wild-type and deletion mutant strains when the culture density OD value reached 0.2 and the same amount of culture was taken before and 5 min after induction in FlhDC or σ overexpression strains [based on our previous results from σ time-course experiments (), we choose 5 min after induction in this assay because it is a reasonable time point that provides sufficient time to induce sigma-dependent genes and also reduce potential post-transcription or other indirect effects]. RNA was stabilized immediately by mixing with a double volume of RNAprotect Bacterial Reagent (Qiagen) and incubated at room temperature for 10 min. Cells were centrifuged at 5800  for 20 min and cell pellets were stored at −80°C prior to RNA extraction. Preparation of labeled probes and microarray procedures were performed exactly as described previously () and in Supplemental Material. Quantitative reverse transcription (RT)-PCR primers were designed using Primer Express software (Applied Biosystems) and were synthesized by the UW Biotechnology Center. Two steps of real-time quantitative RT-PCR are performed. Five micrograms of the DNase-treated total RNA was reverse transcribed for first strand cDNA by using Superscript II system (Invitrogen). Reactions were then performed using 1 ng cDNA and 100 nM of each primer in a 50 µl volume with 1XSYBR Green I mixture. Controls lacking AmpliTaq Gold DNA Polymerase or template were used. Reactions were run on an ABI 7700 instrument (Applied Biosystems) using the following cycling parameters: 95°C for 10 min, 40 cycles of denaturation at 94°C for 15 s and extension at 60°C for 1 min. Relative gene expression data analysis was carried out with the standard curve method (). Assays were performed in triplicate. The incorporation of radioactivity in the newly synthesized RNA can be measured by the DE81 filter-binding assay. The DNA fragments used for transcription assays were amplified by PCR with/without the upstream sequence (∼350 bp) of those candidate σ-dependent genes with high confidence identified in our microarray data. The DNA fragment (∼50 nM) was incubated with ∼15 nM purified σ-associated holoenzyme in a buffer containing 50 mM Tris–HCl (pH 7.9), 150 mM K Glu, 150 µM 3 NTP's (CTP, ATP, GTP) and 20 μM UTP plus 1 μl [α-P] UTP (∼10 µCi), 100 μg/ml BSA, 1 mM EDTA, 10 mM MgCl, 1 mM DTT in a total volume of 20 µl. The mixtures were incubated at 37°C and transcription was stopped by addition of 0.5 M EDTA (final 100 mM) 30 min after reactions were initiated. The samples were loaded directly onto DE81 filter discs of 22.5-mm diameter (Whatman). The P-labeled transcripts were bound to DE81 filters by absorbing the total reaction mixture on the filters. Unincorporated nucleoside triphosphates were removed by washing the filters three times with 500 mM NaHPO buffer, pH 7.6 and twice with 95% ethanol. The filter discs were dried, and nucleotide incorporation was quantified by Cerenkov counting. Incorporation was corrected for background radioactivity by measuring the amount of apparent incorporation in the absence of holoenzyme. The DNA fragments (∼350 bp) used for gel mobility shift assays were amplified by PCR from the upstream sequence of operon. The DNA fragment was P-labeled at the 5′ end using T4 polynucleotide kinase. Samples of <40 ng of the labeled DNA fragments were included in 20-μl reaction mixtures containing DNA-binding buffer (10 mM Tris–HCl, pH 7.5, 50 mM KCl, 0.5 mM EDTA, 5% glycerol, 1 mM dithiothreitol), 500 µg/ml bovine serum albumin (BSA) and 25 µg/ml herring sperm DNA. The CRP was added at the following concentrations: 0, 5, 10 and 25 nM, respectively. cAMP was included in all reaction mixtures at a final concentration of 2 mM. Reaction mixtures were incubated for 15 min at 37°C and then were stopped by the addition of 1 μl of loading buffer (0.1% xylene cyanol and 50% glycerol in HO). The samples were loaded on a 4–10% native Tris–glycine Novex Gel (Invitrogen) and dried on a Slab Dryer (BioRad) as described previously (). Biomax MS film (Kodak) was used for autoradiography. The gels were scanned using a PhosphorImager (Molecular Dynamics), and the intensities of the bands were determined using ImageQuant version 5.2 software. The sequenced K-12 strain MG1655 (λ F 50 rph-1, prototroph) (), on which Affymetrix Genechip probe design is based, was chosen for our studies. FlhDC or σ in-frame deletion strains as well as FlhDC or σ overexpression strains were constructed as described in Zhao . () and in Experimental Procedures. RT-PCR was used to examine the expression of and in the respective deletion mutants before microarray analysis. As expected, the FlhDC mutant did not express and , and the σ mutant failed to express (A), confirming inactivation of these genes. Soft tryptone swarm agar plates () were used to evaluate motility or swimming ability of the strains. In the tryptone swarm agar (B), the and deletion mutants were totally non-motile and did not form any swarm rings compared with the wild type. When the or mutant was complemented with a cloned or gene on a low-copy plasmid, the motility was restored as shown in B. The empty vector pACYC184 served as a control and had no influence on the motility of the or mutant (data not shown). RT-PCR (semi-quantitative) was also performed to detect the mRNA level of or genes before and after induction of a plasmid-borne or genes in . Instead of 35 PCR cycles as performed earlier to test gene disruption, the semi-quantitative PCR was performed for 24 cycles. This reduces the chance of saturation of final PCR product. The synthesized DNA was loaded onto an agarose gel and stained with ethidium bromide. A significant increase of RNA level from target genes was observed after a 5-min induction as shown in C, confirming induction of those genes. To characterize the effects of the decreasing or increasing FlhDC protein level on gene expression, global RNA transcript abundance was monitored in the deletion mutant strain and the overexpression strain 5 min after FlhDC induction with cells grown in log-phase (OD = 0.2) in MOPS minimal medium at 37°C. Transcription profiles were obtained as described in ‘Materials and Methods’ section. Expression profiling of transcripts corresponding to the complete set of ORFs in the genome revealed that the response to deletion of FlhDC was quite broad. There are 117 genes (2.7% of the genome) downregulated 2-fold or more in the deletion mutant strain. The wide distribution of FlhDC-dependent genes in genome (as shown in ) indicates that FlhDC might play a larger role in the global gene transcription regulation than just to serve as a master regulator for the flagellar regulon. There are 53 genes in known to be directly involved in flagellar structure and motor function (,). Compared with the transcriptional level of genes in the wild-type strain, DNA microarray results showed the transcriptional level of all these genes are significantly downregulated in the deletion strain (see Supplemental Material, Table S1). Most of these known genes in the flagellar regulon were initially identified through genetic mutagenesis analysis and can be divided into two functional groups: chemotaxis and mobility; surface structures. Comparing the DNA microarray results before and 5 min after FlhDC induction in the overexpression strain, there are no significant changes (no RNA level changed more than 2-fold compared to the uninduced control) of RNA level of genes in the flagellar regulon (data not shown). We are not too surprised with this result. Compared with previous experiments () where initial σ protein level is low and then is induced almost 8-fold after a 5-min induction, the fold induction of FlhDC under the same inducible P promoter control must be limited due to the high initial abundance of this protein. Therefore, we expect the reason for no significant increased transcription of FlhDC-dependent genes is due to the high initial protein level of FlhDC before induction. The low fold increase of FlhDC in this short time period (5 min) is not enough to further increase transcription of FlhDC-dependent genes using σ-associated holoenzyme. In Table S2 (Supplemental Material), we show a group of 12 new candidate genes whose transcriptional level decreased more than 3-fold as well as a non-flagellar gene, , known to be regulated by FlhDC complex in (). Computer prediction of FlhDC-related binding element () shows that the consensus is represented as tNAAcGCc(N)AAATAgcg (C), where lowercase letters indicate less highly conserved sites. This consensus agrees well with the previously reported FlhDC binding consensus that was aligned from several published FlhDC-dependent genes (). Note, the general height of these consensus element displayed on SEQUENCE LOGO is not high. This indicates that FlhDC consensus binding sequence might not be as strict as sigma factor-binding sites. There are 21 known σ-dependent genes in involved in flagellar synthesis and function (,). In , we can see that the transcriptional level of these genes is significantly downregulated in the strain and is slightly increased in σ overexpression strain. These results are consistent with our previous hypothesis that a change of the intracellular level of a given sigma factor will cause a change of the transcriptional level of genes dependent on this sigma factor. Jishage () reported that the intracellular level of σ is maintained at 50% the level of σ during log and stationary phase growth and σ is thought to be the second most abundant sigma factor among seven sigma factors in . Loss of σ in cells will greatly decrease the transcription of σ-dependent genes; especially those genes that can only be transcribed by σ-associated holoenzyme (such as in ). Compared with our previous study where the induction of σ caused high induction of σ-dependent genes, induction of σ in this study did not show a large increase in the transcriptional level of σ-dependent genes. Using a specific monoclonal antibody for σ, we determined that the protein level of σ increased about 2.3-fold after a 5-min induction (A), lower than the 8-fold increase in the previous σ induction experiments. Comparative analysis of the microarray data from the set of genes whose transcription is downregulated in the deletion strain (decrease of σ) and the set of genes with increased transcription at 5 min after σ induction (increase of σ) allow us to assign many additional genes to the σ regulon. In , there are 13 new candidate genes in σ regulon. The transcriptional level changes of three reported σ-dependent genes (,) which were not assigned into traditional flagellar regulon before ( and ) are also listed in . These three genes are non-flagellar genes or have unknown function in the flagellar system. We chose the top 10 genes in for transcription assays because no experiment has been performed so far to test if these genes can be directly transcribed by σ-associated holoenzyme. The upstream sequence of gene, encoding a heat shock protein, was chosen as a negative control for the transcription assay because transcription of this gene is dependent on σ and is not a σ-dependent gene (). transcription assay results (A) show that most of these 10 genes can be directly transcribed by σ holoenzyme. Note both and genes are in one operon. The gene can be co-transcribed by σ-dependent promoter located in the upstream region of the gene. Results from promoter region consensus analysis using the algorithms BioProspector () and HMMER () revealed σ holoenzyme-binding sites in the upstream regulatory sequences of these genes (B). Note that several of these genes have two putative promoters. Based solely on the negative results from our transcription assay, we are not sure if the gene belongs to the σ regulon. While these possible two-block DNA consensus sequences might provide the primary interaction with holoenzyme, additional transcriptional activators such as FIS and CRP might be utilized to strengthen the promoter–holoenzyme interaction which are not available in our transcription assay. Due to the virulence role of pathogenic bacteria flagella system in adhesion, biofilm formation and colonization of host organisms and in secretion of virulence determinants to host (), the homologous counterparts of the newly identified genes presented here might be potentially associated with these functions, especially with that of the pathogenicity island SPI-1 TTSS (type three secretion system) which appears to be most commonly found in pathogenic bacteria. Recently, Liu and coworkers () proposed a new model for carbon source foraging strategy by . By growing in several different carbon sources, they discovered, as carbon substrate quality declines (defined by growth rate), cells systematically increase the number of genes expressed in a hierarchical manner. Concomitantly, cells also increase their motility. They proposed a RNA polymerase (RNAP) reapportioning model to explain the expansion of genes expression. But the mechanism of increasing the motility in a low-quality carbon source remains unknown. The gradual increase of motility activity with decreased carbon quality was unanticipated because motility by means of flagella is very expensive for cellular economy in terms of the number of genes and the energy required for flagellar biosynthesis and functioning (). Using energy-intensive flagella in poor nutrient environments would trigger a high risk of more rapidly exhausting the sole energy supply. In Liu's paper (), a strategy known as risk-prone foraging (,) has been proposed for this behavior; that bacteria take a risk and use the flagellar system to actively search out better conditions. It was observed many years ago that carbon catabolite repression affects flagellar biosynthesis (,). This led us to think that the relief from the carbon catabolite repression might be a key factor involved in the effects on motility activity inversely correlated with different carbon source quality. Relief from the carbon catabolite repression is a complex regulatory circuit that triggers reprogramming of global gene expression patterns to adapt the changes in external environment. This mechanism will activate the cyclic AMP receptor protein (CRP) (), a global transcriptional factor that positively regulates most carbon catabolic pathways. While it is known that carbon catabolite repression affects the flagellar synthesis and the CRP activation might be involved in alleviating this repression (,,,), much less is known regarding the role of CRP in motility regulation under a range of carbon source conditions. To determine the effect of different carbon sources on CRP activity as well as the functional relevance between the active CRP level and the expression of FlhDC operon in a range of conditions with the sequenced strain MG1655, we performed the following assays. CRP protein was purified using the pET expression system (Supplemental Material, Figure S2) for assays. In electrophoretic mobility-shift assays as shown in A, the upstream DNA fragment of operon can be shifted by purified CRP protein. CRP is a dimer of identical subunits. The consensus tandem DNA-binding site for CRP dimer has been identified by analysis as shown in red color which is approximately palindromic and provides two very similar recognition sites, one for each subunit of the dimer. For sequenced strain MG1655, there are 250 bp between the CRP-binding sites and FlhDC translational start site, which has a rare translation start codon with GTG that is different from another strain studied by Soutourina . (). We then measured the transcriptional levels of two well-known CRP positively controlled genes ( and ) (to represent the level of active CRP in the cell) as well as the expression level of operon by the quantitative RT-PCR approaches (B and C). Compared with cells grown in glucose, the transcriptional levels of CRP-dependent genes gradually increased in succinate, alanine and acetate grown cells, with a slight decline in proline compared with the level in acetate. The same change in the transcriptional pattern of the FlhDC operon in cells grown in alternative carbon sources relative to that in glucose-grown cells can be observed as shown in C. In addition, we made a cyaA in-frame deletion strain. cyaA encodes adenylate cyclase that is required to catalyze the formation of cyclic AMP (). The pattern of increasing the transcriptional levels of CRP-dependent genes in low-quality carbon sources, as was observed in the wild-type strain, disappeared in the cyaA deficient strain (as shown in Supplemental Material, Figure S3). This indicates that CRP-cAMP plays an important role in promoter activation in our assays. No significant change of the transcriptional level of CRP was observed (C), suggesting that the induction of CRP-dependent genes might be mainly due to the activation of CRP rather than the increase in CRP expression (). In an intact flagellar system, the mechanism by which cells control flagellar operation involves genes that are under the direct control of alternative sigma factor, σ. In motility control, the cell has a family of transmembrane proteins with receptor functions (,) termed methyl-accepting chemotaxis proteins, or MCPs (). MCPs mediate responses to external environmental stimuli. These receptors bind stimulatory ligands and undergo conformational changes that regulate the activities of a network of signal transduction proteins within the cytoplasm. There are six cytoplasmic signal transduction proteins, the products of the Che genes: and in . The signal transduction pathways then deliver information to flagellar apparatus to mediate cell filament (product of ) as a propeller to do rotation and switching. Genomic organization of these genes in respective operons is presented in D. The apparent coupling of motility activity and operon activation in these experiments prompted us to further measure the σ synthesis rate as well as σ-dependent genes synthesis rates in each culture, since these parameter are known to be correlated with motility control (). We choose the first gene in each operon for the RT-PCR experiments to test the activity of these genes as a function of various carbon sources. Compared with the transcriptional levels of these motility and chemotaxis genes in the fastest growing culture (glucose), a significant increase of the transcriptional level of these genes in slow growing cultures with alternative carbon sources (succinate, alanine, acetate, proline) can be seen in our assays as shown in D. The upregulation of motility genes correlated with increasing motility of cells in poor quality carbon sources, together with the synchronized pattern of increasing CRP activity causing increasing FlhDC transcription, further suggests CRP is a factor (or at least one of multiple factors) to play a positive role in triggering ‘foraging’-like behavior to actively search for better conditions as the quality of the available substrate(s) decreases. Interestingly, the transcriptional level of the costly cell filament encoding gene did not have any significant changes across different carbon sources. Recently, the Kelly Hughes group (,) reported studies on the transcriptional and translational control of the gene in promoter and 5′ untranslated regions in . Our observations here suggest, in a poor nutrient environment, that bacteria may strategically use the precious energy on the basis of fine tuning of flagellar-gene expression in response to environmental challenges, which it possibly does by sequestering a transcriptional repressor or other factor(s) to inhibit expensive gene expression. The biological implications of this finding are discussed later. In this study, we used two different genetic approaches moderately expressing FlhDC or σ from anhydrotetracycline (aTc) inducible and Tet repressor-controlled P promoter in a plasmid-borne or gene; disrupting the expression of FlhDC or σ in or deletion mutant strains, to efficiently and reliably study the regulon members of the two flagellar biosynthesis regulators, FlhDC and σ. Our results demonstrate that there are many more genes than previous known under the control of these two regulators. In our previous studies (), we have demonstrated that our approaches have a significant advantage over those approaches for stimulons and regulons studies in various stress conditions. In this paper, we further discovered that the strength of these two different regulon study approaches depends mainly on the initial concentrations of sigmas or other regulators . Low initial concentration of a regulator can be induced to a high level after a short induction, which in turn increases transcription of its dependent genes to a significant level. Loss by deletion of the high initial concentration regulator might totally shut off the transcription of its dependent genes . Note that, for some important regulators, loss of their functions might cause severe growth problems. Currently, using our lab collection of the monoclonal antibodies for each sigma factor in , we are using the ChIP-chips assay (,) as a complementary approach to pull down the DNA fragment that is bound and crosslinked by a given sigma factor. Combination of two or several different regulon study approaches will give us more confidence in positive results. In addition to extending the repertoire of FlhDC and σ regulon with new candidates, this work demonstrates that the carbon source utility and motility activity are interdependent. This study was motivated by the previous observation that increases its motility in poor nutrition environments. Much less is known about how the activation of CRP is affected as well as how this activation will affect cell motility by different carbon source supplies. Our results shed light on cell motility control in a range of different carbon sources and suggest intriguing hypotheses about its establishment and function during utilizing different energy compounds. The observations reported here have many notable features: in the defined MOPS minimum medium with different carbon sources, CRP is activated in poor quality (defined by growth rate) carbon sources relative to rich quality ones; this activated CRP associates with the operon encoding the master regulator with roles in flagellar regulation and development; the activation of CRP is correlated with the increase of expression of σ as well as the increase of cell motility. Flagellar genes are organized into a transcriptional hierarchy that underlies temporal and spatial control of biogenesis program. A large amount of energy is required to synthesize flagella and a large part of this energy is used for filament synthesis. The filament consists of an assembly of around 20 000 subunits of a single protein, flagellin (), and the amount of flagellin alone composes about 8% of the total cell protein when the flagellar operons are expressed optimally (). Although the motility of cells gradually increase in low-quality carbon sources supplies, we found the transcription of the gene, which encodes costly flagellin, does not increase in those minimum growth conditions. This may be a means of conserving energy. The cost to the cell of flagellar synthesis and flagellar operation is about 2% and about 0.1% of total energy expenditure under normal growth conditions, respectively (). As the growth potential of an environment decreases, the ability to reach a potential food source ahead of siblings is a group competition behavior (). Also the ability to efficiently utilize cellular resources to conserve energy for individual self-protection, such as hibernation, would provide a significant survival advantage. We show that motile bacteria carry out tactical responses to a variety of carbon sources by increasing flagellar operation but restricting costly additional flagellar synthesis. This may provide a paradigm for cost and benefit behaviors in prokaryotes, which result in maximum benefit for survival. The delicate balance between conserving more energy and using the energy-intensive flagella to search out better conditions is modulated by environmental conditions. The inverse correlation between the increasing motility of cell with carbon quality might be in a certain minimum nutrition range. As the growth condition becomes more and more harsh as shown in lowest-quality carbon source (proline), conserving precious energy outweighs the cost of expending energy for potential benefit of locating and utilizing good sources. In this situation, the cell will gradually turn off both of the costly flagellar synthesis (flagellin) and flagellar operation (motility and chemotaxis) and save more energy for a long time survival to passively wait for less adverse (natural) conditions. The downshift of the expression of flagellar genes as well as motility activity in proline might represent these behaviors. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Messenger RNA 3′ end formation is an essential step in the biogenesis of mRNAs in eukaryotic cells. This step consists of a series of events that culminate in the addition of a finite tract of poly(A) to the 3′-end of a processed pre-mRNA. These events are mediated by a complex of multi-subunit factors, the composition of which is generally conserved in eukaryotes (). This complex recognizes specific sequence elements in a pre-mRNA, processes the RNA and adds the poly(A) tract. In animals, the canonical poly(A) signal AAUAAA is recognized by the 160-kDa subunit of cleavage and polyadenylation specificity factor (CPSF; the 160-kDa subunit is termed CPSF160), the U + G-rich downstream sequence element is bound by CstF64, the 64-kDa subunit of cleavage stimulatory factor (CstF) and other auxiliary elements situated 5′ to (or upstream of) the polyadenylation site are recognized by CFI25, the 25-kDa subunit of cleavage factor I (CFI). Two other subunits of the complex, hFip1 and CPSF30, also bind RNA (,), but the roles of these activities in the process remain poorly defined. These RNA–protein interactions, as well as a number of protein–protein interactions between other subunits of the subcomplexes, serve to coalesce the complex and position the RNA for cleavage and polyadenylation. Two different subunits of CPSF, namely CPSF73 and CPSF30, have been suggested as the endonuclease responsible for the cleavage of the mRNA at the polyadenylation site subsequent to polyadenylation (). Subsequent to RNA processing, poly(A) polymerase (or PAP) is brought to the RNA, presumably through its interaction with animal orthologs of Fip1 (), so as to add the polyadenylated tract. This final step in the reaction is modulated by a nuclear poly(A)-binding protein (PabN) that serves to promote processive poly(A) addition and to regulate the final length of poly(A) that is added to the mRNA (,). In the budding yeast , mRNA 3′ end formation is directed by a tripartite polyadenylation signal, the elements of which have been termed as the efficiency element, positioning element and polyadenylation site, respectively. The RNA-binding subunits Hrp1, Rna15 and Yth1 associate with these elements; Hrp1 binding to the efficiency element, Rna15 to the positioning element and Yth1 to the polyadenylation site (). Hrp1 and Rna15 are subunits of the so-called cleavage factor I (CFI, the subscript ‘y’ serving to distinguish this factor from the mammalian CFI) (,) while Yth1 is a subunit of the cleavage and polyadenylation factor (CPF) (). Rna15 and Yth1 are orthologs of CstF64 and CPSF30, respectively, while Hrp1 has no strict counterpart in the animal polyadenylation complex. Presumably as is the case in animals, these RNA–protein interactions that involve different multi-subunit factors serve to promote the assembly of the complete complex on a pre-mRNA, as well as to position the pre-mRNA for processing and subsequent polyadenylation. The nature of the processing endonuclease in yeast has not been explored, but both of the animal proteins (CPSF73 and CPSF30) that have been associated with suggestive nuclease activities have yeast counterparts—Ysh1 is the ortholog of CPSF73, and Yth1 is the counterpart of CPSF30. As in animals, PAP is tethered to the processed RNA through its interaction with Fip1 (). While yeast has no formal counterpart of PabN, two other PabNs, Nab2 and Pab1, contribute to poly(A) length control in this organism (). Plant genomes possess genes with the potential to encode virtually the complete set of polyadenylation factor subunits that are found in animals and yeast (). Moreover, as is the case in animals and yeast, the plant polyadenylation signal consists of several distinct elements—in plants, the so-called Near-Upstream Element, Far-Upstream Element and Cleavage Element, or NUE, FUE and CE, respectively (). The plant polyadenylation complex has not been as extensively studied as its animal and yeast counterparts. However, recent studies have established that several of the plant counterparts of CPSF subunits (specifically, CPSF160, CPSF100 and symplekin) reside in a nuclear complex that also includes the ortholog of the yeast polyadenylation factor subunit Pfs2 (the plant protein has been termed FY) (), and most of the plant subunits may be linked conceptually with PAP (,). Eukaryotic CPSF30 proteins are among the most interesting of the subunits of the polyadenylation complex. As stated in the preceding paragraph, the yeast ortholog (Yth1) associates with the cleavage/polyadenylation site (). The ortholog possesses nucleolytic activity (,), leading some to propose that CPSF30 is a processing endonuclease in the polyadenylation reaction (). In animal cells, CPSF30 has been shown to be the focus of regulatory events in virus infections (,). The animal and yeast proteins share an array of five tandemly arranged CCCH-type zinc finger motifs (typically, Cys-X-Cys-X-Cys-X-His) that have been implicated in RNA binding by these proteins (), in nucleolytic activity () and in interactions with other polyadenylation factor subunits (). Moreover, animal proteins have additional zinc knuckle motifs that are absent from Yth1 (), and the zinc knuckle motif of the bovine protein enhances the RNA-binding activity of the core zinc finger domain (). Among the set of plant genes that encode polyadenylation factor subunits are those that encode orthologs of CPSF30/Yth1. The gene (as well as its counterparts in other plants; unpublished data) encodes two mRNAs (); the smaller of the two (termed herein as AtCPSF30) specifies a . 250-amino acid polypeptide that is analogous to CPSF30/Yth1, whereas the larger is translated to yield a polypeptide in which all but the C-terminal 13 amino acids of AtCPSF30 are fused to a protein domain found in mammalian splicing associated factors [the so-called YT521B motif; ()]. Both polypeptides can be detected in nuclear extracts, and the smaller resides in a nuclear complex with the ortholog of CPSF100 (). AtCPSF30 also interacts with an ortholog of Fip1 (), placing it in a network of interactions that includes PAP (among other polyadenylation factor subunits). Like its other eukaryotic counterparts, AtCPSF30 is an RNA-binding protein; interestingly, the RNA-binding activity of AtCPSF30 is inhibited by calmodulin in a calcium-dependent fashion (), suggestive of a regulatory link between signaling and RNA processing in plants. In our ongoing studies of AtCPSF30, it was noticed that a persistent nuclease activity co-purified with the recombinant protein, an activity that confounded attempts to define RNA-binding sites on substrate molecules. In this report, the properties of this nuclease activity are described. Specifically, we show that AtCPSF30 is itself an endonuclease, that the RNA products of endonucleolytic cleavage possess 3′-hydroxyl groups, and that an Fip1 ortholog inhibits the AtCPSF30-associated nuclease activity. Together with other studies, these results suggest that AtCPSF30 is a processing endonuclease, and that the pre-mRNA cleavage step in plants is mediated by more than one protein. The purification of MBP-AtCPSF30, m4 and m9 proteins () has been described elsewhere (). The zinc finger mutants of AtCPSF30 () were generated by using quick-change site-directed mutagenesis kit (Stratagene) using the pMALC2-AtCPSF30 plasmid () as template as per manufacturer's instructions. The results of mutagenesis were confirmed by DNA sequencing; accordingly, the C-terminus of the first zinc finger was changed from DACGFLHQF to DASTFLYQ, the second zinc finger from QDCVYKHTN to QDSTYKYTN and the third from PDCRYRHAK to PDSTYRYAK. The oligonucleotide primers that were used for the mutagenesis are given in . The cloning of coding sequence corresponding to the N-terminal 137 of the chromosome V-encoded Fip1 protein into bacterial protein expression vectors (Pharmacia) and for making GST and MBP fusion proteins, respectively, was as described (). The coding sequence corresponding to the N-terminal 483 amino acids was cloned into vector (Invitrogen) from the corresponding entry clone () to produce histidine-tagged fusion protein. Histidine-tagged and GST fusion proteins were purified as described by Forbes . () and MBP fusion proteins purified as described in Delaney . (). It should be pointed out that the purification of MBP and GST fusion proteins included a wash of proteins bound to the affinity media with buffers containing 2 M NaCl, a step intended to remove almost all non-specifically bound bacterial proteins from the MBP fusion protein preparations. Control proteins (GST, MBP and histidine-tagged β-glucuronidase) were prepared as described (,). Protein concentrations were estimated by comparing the purified preparations with BSA standards by SDS–PAGE and staining of the gels with Coomassie Brilliant Blue. Labeled RNA was derived from the pea -E9 polyadenylation signal, and contained nucleotides extending from 145 nt upstream to 80 nt downstream from the site noted as ‘+1’ by Mogen . (). Various mutant forms of the polyadenylation signal lacking one or more cleavage sites or sequence elements were also used for making the transcripts. Templates for transcription were prepared by PCR, using the plasmid templates and primers listed in . Uniformly labeled and unlabeled RNAs were prepared by transcription using Ampliscribe kit from Epicentre as per manufacturer's instructions, and purified using MegaClear kits (Ambion). In addition to -E9-derived RNAs, a control RNA from a GFP gene was also used in some experiments. Transcription template for this RNA was produced by PCR using the primers indicated in (‘BS-T3’ and ‘GFP-3′’) and a truncated GFP plasmid as template. This plasmid was produced by subcloning the GFP gene in KY80-GFP (a gift from Dr Randy Dinkins, USDA/ARS, Lexington, KY) into pBluescript vector as an XbaI–SacI fragment. The resulting recombinant plasmid was digested with HincII and ligated to release the first 490 bases of GFP. The nuclease assay reactions contained varying quantities (ranging from 0.12 to 12 pmol) of purified protein, 2 pmol of labeled RNA, 0.6 mM MgCl and RNasin (30 U, Eppendorf) in a volume of 10 μl of Tris–HCl buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl). After incubating the reaction mixture at 30°C for appropriate periods of time, reactions were stopped with phenol–chloroform. Three to five microliters of the aqueous phase was mixed with equal volume of gel loading buffer II (Ambion) and heated at 65°C for 15 min. The samples were subsequently chilled on ice for 2 min before conducting electrophoresis on 8–10% polyacrylamide–urea gels. Following electrophoresis at 7–8 mA constant current, the gels were dried and analyzed by autoradiography. Autoradiographs were analyzed using ImageJ analysis software. For RNA binding, the electrophoretic mobility shift assay described elsewhere () was used. It should be pointed out that the conditions used in these assays are such that appreciable RNA, including partial breakdown products, remains after the 15 min incubation time (A, lanes 8 and 12), and thus that RNA binding can be measured for forms of AtCPSF30 that possess nuclease activity. The procedure is similar to the one followed by Ivanov . (). Unlabeled RNA was blocked at 3′ end using yeast PAP (USB) and cordycepin 5′-triphosphate as per manufacturer's instructions. The RNA so obtained was treated with MBP-AtCPSF30 fusion protein for 15 min and the resulting breakdown products recovered by phenol–chloroform and ethanol precipitation. Subsequently, the RNAs were dissolved in water and end-labeled with T4 RNA ligase (6 U) and [P] cytidine 3′,5′-bisphosphate at 4°C for 20 h. The RNAs were subsequently precipitated with ethanol, re-suspended in gel-loading buffer II (Ambion) and analyzed by 10% denaturing polyacrylamide–urea gel electrophoresis and autoradiography. Autoradiographs were analyzed using ImageJ analysis software. Two hybrid assays of the interactions between an Fip1 ortholog and derivatives of AtCPSF30 were carried out as described (). The different portions of AtCPSF30 were subcloned into pGEM as described (), excised as BglII fragments and cloned into pGAD-C() and pGBD-C() () to yield for activation domain (AD) and binding domain (BD) clones, respectively. AD and BD plasmids were transformed into PJ69-4 () and dual transformants (identified as colonies growing on media lacking leucine and tryptophan, the selective markers for these two plasmids) subsequently tested on media lacking leucine, tryptophan and adenine (the latter being one of the scorable markers for interactions). Positive interactions were those in which all tested colonies (between 4 and 10) grew on the adenine-free media. Negative controls for these tests included transformations with combinations of plasmids that included unmodified pGAD-C() or pGBD-C(). The positive control was that used by Forbes . (), namely the set of plasmids that carried the CstF77 and CstF64 orthologs. Prior work in this laboratory established that AtCPSF30 is an RNA-binding protein (). Attempts to explore the RNA-binding properties of AtCPSF30 by mapping possible sites of binding on labeled RNA () were not successful owing to persistent nuclease activity that was insensitive to commercially available ribonuclease inhibitors (data not shown). Extensive and exhaustive purification, as well as characterizations of equally pure preparations of MBP (the tag used to purify recombinant AtCPSF30), led to the realization that AtCPSF30 itself was the source of this activity. In light of reports ascribing similar activities to the counterpart of CPSF30 (,), the nuclease activity of AtCPSF30 was characterized in some detail. For these studies, an RNA containing a complete suite of polyadenylaton-associated elements was used (A). This RNA is derived from the pea -E9 gene and includes the well-defined FUE)and NUEs (,) from this gene, and was chosen so as to best recapitulate the complex nature of plant polyadenylation signals. Initial studies using uniformly labeled RNA revealed an activity that yielded numerous breakdown intermediates at early times in a typical time-course study, and an eventual loss of all detectable poly- nucleotides or oligonucleotides after extended times (A). Nuclease activity was dependent on enzyme concentration, and was not seen when comparable quantities of purified MBP were used in place of the MBP-AtCPSF30 protein (B). The transient accumulation of RNA breakdown intermediates in the time-course studies suggested that AtCPSF30 might possess endonuclease activity; such intermediates were also observed with RNAs end-labeled at either their 5′ or 3′ ends (data not shown). To further test this, assays were conducted using circular RNAs. As shown in C, such RNAs were also susceptible to the nuclease activity of AtCPSF30. Moreover, the kinetics of breakdown of the circular RNAs were similar to those seen with the linear RNA. These results indicate that AtCPSF30 can act as an endonuclease. One possible function for the endonuclease activity of AtCPSF30 is the processing of the pre-mRNA as a prelude to the polyadenylation. Such a function would require that the nuclease action leaves a 3′-hydroxyl group at the end of the RNA; alternative modes of nuclease action, in contrast, might leave 3′-phosphates or 2′-3′ cyclic phosphates. To examine this, the products of AtCPSF30 endonucleolytic action were end-labeled with RNA ligase + [P] cytidine 3′,5′-bisphosphate ([P]-pCp); such a treatment is expected to label RNAs bearing 3′-hydroxyl groups, but not 3′- or 2′,3′-cyclic phosphates. As shown in (lane 1), the end-labeling treatment of the input (unlabeled) RNA yielded a single discrete species. Pre-treatment of the input RNA with 3′-dATP + PAP (, lane 2) eliminated this labeling, indicating a requirement for a free 3′-hydroxyl for the labeling with [P]-pCp. Subsequent treatment of the 3′-blocked input RNA with AtCPSF30 yielded an array of breakdown products that were readily labeled with ([P]-pCp (, lane 3). This result indicates that the nuclease action of AtCPSF30 yields 3′-hydroxyl groups. It should be noted that the RNAs labeled by [P]-pCp in lane 3 of do not correspond in size to those expected if cleavage was occurring at the three ‘natural’ polyadenylation sites in the substrate RNA (the latter would be expected to yield 5′ cleavage products of 125, 145 and 175 nt) (,). Thus, while purified AtCPSF30 processes the substrate at distinct sites, by itself it does not recapitulate the exact 3′-end profile seen . The distinctive processing pattern seen in , while not identical to the pattern expected based on the handling of this RNA, nonetheless suggests some sequence preference in the action of AtCPSF30. Such preferences may be related to one or more of the polyadenylation-related motifs—FUE, NUE or CE—present in the RNA. To explore this possibility, the action of AtCPSF30 on a battery of other RNA substrates was explored. These RNAs consisted of smaller parts of the E9 polyadenylation signal (A). One of these (‘ΔFUE’ in A) had a deletion in its FUE region. Another (‘PC’) retained the FUE and two of the NUEs, and ended at the middle of the three sites in this 3′-UTR (this site is the predominant site utilized ) (,); this RNA is thus analogous to a pre-cleaved RNA. A third (‘ΔNUE’) had a deletion extending from 60 nt upstream to 40 nt downstream from the middle poly(A) site; this deletion removes all of the NUEs and cleavage sites contained in this signal, but retains the FUE. A fourth (DE) extended from the middle polyadenylation site to 80 nt downstream from this site. As shown in B, all of these RNAs could serve as substrates for AtCPSF30 in the nuclease assay. Moreover, no dramatic differences could be seen in the rates with which these RNAs were degraded. An RNA (labeled ‘GFP’ in ) with no known polyadenylation-related elements was degraded with similar kinetics. The results indicate that AtCPSF30 does not have an obvious preference for polyadenylation-related signals for its nucleolytic activity. AtCPSF30 consists of a number of domains (); these include a central region of three CCCH-type zinc finger motifs that are conserved to some extent in all eukaryotic CPSF30 proteins, an N-terminal region with a highly-conserved, plant-specific region enriched in acidic amino acids and a plant-specific C-terminal region. To study the possible contributions of these various domains to the nuclease activities of AtCPSF30, a number of mutant forms of the enzyme (A) were prepared and assayed. These includes derivatives lacking either the N-terminal or C-terminal plant-specific domains (m4 and m9) () and point mutants in which the last five amino acids of each zinc finger motif (CxxxH) were replaced with the motif STxxY; this replacement eliminates two of the four putative zinc-coordinating amino acid side chains in the motif. As was noted previously (), purified preparations of the full-sized MBP-AtCPSF30 consisted of two polypeptides (B); the smaller of these (noted with an * in B) is a breakdown product whose C-terminus lies near that of the m4 mutant (B, the ‘m4’ lane on the right of the panel). These variants were assayed for nuclease activity. As shown in C, the m9 and ZF1 mutants displayed activities that were similar to that of the wild-type enzyme. The m4 and ZF2 mutants both had activity, but less than that seen with the wild-type enzyme (this is apparent as still-detectable precursor RNA after the 30-min incubation). The ZF3 mutant showed little or no detectable activity under these conditions. These results indicate that several parts of the protein are needed for full activity, but point to the third zinc finger as being most important. Importantly, the effects of the point mutations, and especially the ZF3 mutant, indicate that the nuclease activity under study is a property of the AtCPSF30 polypeptide (as opposed to a co-purifying bacterial contaminant that the stringent washes and ion exchange purification protocols did not separate). One possible explanation for the absence of nuclease activity with the ZF3 mutant is that it no longer binds RNA as does the wild-type protein. This hypothesis was tested using a gel-shift assay and the -E9 RNA depicted in . As shown in , the ZF2 mutant had an RNA-binding activity that was similar to the wild type. The ZF3 mutant was also able to bind RNA, albeit at a somewhat reduced level. The ZF1 mutant displayed no ability to bind RNA. These results indicate that ZF1, which is dispensable for nuclease activity () is primarily responsible for the RNA-binding activity of AtCPSF30. Ablation of ZF3 partially reduces RNA binding but completely eliminates the nuclease activity. Interestingly, alteration of ZF2 had no effect on RNA binding. The third zinc finger of AtCPSF30 corresponds to the motif of Yth1 that is involved in its interaction with Fip1, an interaction that affects the binding of Yth1 to RNA (). Since this is also the motif that is needed for endonuclease activity (), it seemed possible that plant Fip1 orthologs might have effects on the nuclease activity of AtCPSF30. Previously, it was reported that an Fip1 ortholog, encoded by At5g58040, was able to interact with AtCPSF30, and that the N-terminal 137 amino acids of the 1196 amino acid Fip1 ortholog [AtFip1(V)] contained the domain responsible for this interaction (). This interaction was further dissected using a standard yeast two-hybrid assay. For this, the battery of AtCPSF30 mutants illustrated in A, along with others described previously (), were tested for interactions with the N-terminus of AtFip1(V). As summarized in , all of the variants that possessed the third zinc finger retained the ability to interact with AtFip1(V); importantly, alteration of just the third zinc finger eliminated the interaction, indicating that this interaction requires just the third zinc finger motif of AtCPSF30. Accordingly, the effects of the N-terminal portion of AtFip1(V) on the nuclease activity of AtCPSF30 were tested, using three different forms of the N-terminus of the protein. One form consisted of a histidine-tagged segment containing the first 483 amino acids of AtFip1(V); this segment includes the CPSF30-binding region as well as the conserved Fip1 domain that defines this group of proteins (this is illustrated in A). Two other forms consisted of the N-terminal 137 amino acids fused to either GST or MBP. These three forms were purified and added in 2-fold molar excesses to nuclease reactions. The results, shown in the upper panel of B, indicate that all three forms inhibit the nuclease activity of AtCPSF30. Purified histidine-tagged GUS (B, lane 8), MBP (B, lane 9) or GST (B, lane 10) had no effects on the nuclease activity. Moreover, the purified control proteins themselves possessed no noticeable nuclease activity (data not shown). These results indicate that the interaction of AtCPSF30 with the N-terminus of AtFip1(V) inhibits the nuclease activity of this protein. This result suggests that AtFip1(V) regulates the nuclease activity of AtCPSF30. In contrast to this, the GST-Fip137 protein had little apparent effect on RNA binding by AtCPSF30 (C). Thus, RNA binding by AtCPSF30 (C, lane 1) was still apparent in the presence of added GST-Fip137 (C, lane 2) or GST (C, lane 3). A slight reduction in binding in the presence of GST-Fip137 was apparent; this is consistent with the reduced RNA binding of the ZF3 mutant (). However, the extent of diminution of RNA binding by GST-Fip137 was far less than the inhibition of nuclease activity seen with the same preparation. The nature of the events associated with processing of the pre-mRNA remains somewhat unclear. Bai and Tolias (,) reported that the CPSF30 ortholog, the so-called clipper protein, possesses an inherent nuclease activity. Based on this observation, it was suggested that CPSF30 is the processing endonuclease (). Such a suggestion would seem to be consistent with the observation that Yth1 associates with the processing site (). However, others have suggested that CPSF73 is the processing endonuclease. These arguments are based on a number of observations: the AAUAAA-dependent crosslinking of CPSF73 to the cleavage site, the similarity of CPSF73 to Zn-dependent hydrolytic enzymes related to metallo-beta-lactamases, the requirement of amino acids residues in CPSF73 that are required for hydrolytic function in metal-beta-lactamases and the observation that recombinant CPSF73 possesses endonucleolytic activity (,). Based on the results presented in this study, AtCPSF30 would seem to be an excellent candidate for an endonuclease involved in processing pre-mRNAs prior to polyadenylation in . It possesses a distinctive endonuclease activity () and leaves a 3′-terminus that is suitable for subsequent poly(A) tail addition by PAP (). Poly(A) site choice is altered in a mutant deficient in AtCPSF30 (Zhang ., submitted for publication), indicating that normal poly(A) site selection in requires the presence of AtCPSF30. However, if AtCPSF30 is in fact a processing endonuclease, it cannot be the only one active in plants. This follows from the realization that this protein is not essential for plant growth and development (), and that plants deficient in this protein produce bulk poly(A) that is indistinguishable in length and quantity from wild-type plants (Zhang ., submitted for publication). Wickens and Gonzalez () have speculated that the cleavage step in polyadenylation may be performed by any of a multiplicity of processing endonucleases, and has likened the 3′-processing machinery to the ‘package’ of nucleases that comprise the exosome. It may be that such a situation is in fact in force in plants, and that AtCPSF30 and AtCPSF73 are both processing endonucleases. Another possibility merits mention. While the obvious role for an endonuclease in the polyadenylation reaction is the generation of the 3′ end for PAP, alternative functions for such an activity may exist as well. CPSF has been implicated in small RNA 3′ end maturation (,), 3′ end formation of cell cycle-regulated histone mRNAs (,) and cytoplasmic polyadenylation (,). While specific roles for CPSF30 in these processes have not been reported, it is nevertheless possible that the endonuclease activity documented here may reflect roles for AtCPSF30 in these, or other as yet unidentified, activities. The part of AtCPSF30 that is associated with endonucleolytic activity (the third zinc finger) is present in all eukaryotic CPSF30 orthologs (A). The corresponding zinc finger motif in Yth1 is required for RNA binding (). Moreover, with AtCPSF30 and Yth1, this motif is needed for interactions with Fip1 proteins, and interactions with Fip1 inhibit the associated RNA-related activities (nuclease, in the case of AtCPSF30, and RNA binding, in the case of Yth1) of these proteins (). Tacahashi . () have proposed that this interplay between Fip1 and Yth1 may reflect a progression through the polyadenylation reaction, with Fip1 displacing Yth1 from the pre-mRNA after the processing step, thereby bringing PAP to the 3′ end left by processing. The characteristics of AtCPSF30 do not lend themselves easily to an analogous model. In particular, the zinc finger motifs of AtCPSF30 implicated in RNA binding and nucleolytic activity are different, and AtFip1 interacts only with the motif involved in nuclease action. Moreover, AtFip1 inhibits the nuclease activity of AtCPSF30 (B) but has a much more modest effect on RNA binding by AtCPSF30 (C). Therefore, it seems unlikely that AtFip1 would displace AtCPSF30 from a processed pre-mRNA. One plausible alternative to this is a scenario where AtCPSF30 is initially associated with AtFip1, such that the endonuclease activity of AtCPSF30 is repressed (B). The endonuclease would become activated with the dissociation of AtFip1(V), perhaps as a consequence of binding of other subunits to AtFip1(V). Along these lines, it is of interest to recall the PAP isoforms bind to the same 137 amino acid portion of AtFip1(V) that inhibits the nuclease activity of AtCPSF30 (24; B. Addepalli and A. G. Hunt, unpublished data). This remodeling around AtFip1(V) would lead to subsequent processing and polyadenylation. More generally, the mutational analysis of AtCPSF30 presented in this study reveals a protein of interesting and subtle complexity. Two RNA-associated activities of the protein, binding and endonucleolytic activity, are associated with different zinc finger motifs. The third motif and its evolutionarily conserved association with AtFip1 is mentioned above. The first zinc finger motif is the most highly conserved of the motifs when compared with other eukaryotic CPSF30 proteins () and is needed for RNA binding by AtCPSF30. The corresponding motif in the bovine and yeast counterparts is also needed for RNA binding (,,). Interestingly, the corresponding motif in Yth1 is needed for the functioning of this protein in the cleavage step of polyadenylation (); this is consistent with the model proposed in the preceding paragraph, in which positioning prior to cleavage by the nuclease domain is accomplished by the RNA-binding domain of AtCPSF30. In (and likely other plants), CPSF30 is encoded by a complex gene whose transcripts are alternatively processed to yield two mRNAs (). The smaller of these encodes AtCPSF30, whereas the larger specifies a polypeptide that consists of virtually the entire CSPF30 sequence fused to another domain (the so-called YT521B domain) that is found in splicing-associated proteins in mammals () and in proteins that are associated with so-called calcineurin B-like interacting protein kinases [CIPKs; ()]. The function of the larger polypeptide is not known, but the present study suggests that it should possess endonuclease activity. The possible roles that the larger polypeptide might play in gene expression are not clear, but the conceptual linkages between mRNA 3′ end formation and both calcium-mediated signaling and pre-mRNA splicing are intriguing.

mRNA 3′ end maturation is part of a general scheme of pre-mRNA processing comprising 5′-capping and intron-splicing. All these maturation events are essential and tightly coupled and controlled for proper gene expression (,). mRNAs poly(A) tails are produced by cleavage and polyadenylation of the pre-mRNA molecule (). Occurring co-transcriptionally, pre-mRNA 3′-end processing is critical for termination of transcription and mRNA export (). As opposed to the striking divergence of the -acting sequence elements that direct cleavage and polyadenylation, the protein components of the pre-mRNA 3′-end processing complexes are quite well conserved from yeast to mammals. In metazoans, cleavage of the precursor requires the trimeric complex cleavage stimulation factor (CstF) and the cleavage and polyadenylation specificity factor CPSF. Both of them are crucial to identify during a preliminary step the precise sequence elements on the precursor where cleavage, and hence polyadenylation thereafter, would occur (). Additional factors are then recruited, CF I, CF II, and the poly(A) polymerase PAP, to stabilize the initial interaction and trigger the processing. A network of physical interactions between subunits of the 3′-end processing machinery and the transcription apparatus has been partially drawn that could explain to some extent how processing, transcription termination and export can be regulated. Many interactions between the pre-mRNA 3′ end processing factors have been reported for the human, a and yeast systems. CstF is a multimeric complex essential for the reaction to occur. In human and , CstF is formed of CstF-50, CstF-64 and CstF-77 (). The more likely yeast counterparts are respectively, Pfs2p, Rna15p and Rna14p (,). CstF-50 exhibits characteristic WD repeats which are involved in the assembly of multi-protein factors (). CstF-64 bears an RRM-type RNA-binding domain required for the recognition of U/GU-rich elements located downstream of the poly(A) site (). It plays a key role in the choice of the cleavage site and hence, in the efficiency of the reaction (,). CstF-77 is critical for the assembly of the complex, bridging both CstF64 and CstF-50. It is the prototypical Half-a- TPR-containing (HAT) protein as defined by Preker and Keller (). Moreover, CstF-77 is located at the crossroads in the network of interactions with other 3′-end formation factors such as CPSF and CF II. It connects CstF to CPSF-160 and hPcf11 (). Mutations in Rna15p and Rna14p not only impair formation of the mRNA 3′-ends but also prevent RNA polymerase II to terminate properly (,). Export of the imperfect transcripts is affected and, as a consequence, they are subsequently degraded (). A growing number of structural studies have shed light on how the catalytic reactions and the regulation may occur in this complex biological machinery (). The structure of protein interacting domains and protein–RNA complexes have been also reported (,). Many basic questions are still open such as the exact subunit composition of some specific complexes. In this study, we report the crystal structure of the core subunit of the CstF complex at 2.0-Å resolution. CstF-77 is built around 11 HAT repeats that self-assemble to form a tight homodimer. The complex has an overall V-shape with large dimensions. Apart from the conserved dimerization interface, several other phylogenetically conserved areas appear at the surface of the complex that may well represent platforms for the association with other protein partners. Mapping experiments performed with the yeast orthologues of Cstf-77 and CstF-64 allows the identification of the docking domain of Rna15p onto Rna14p. The full-length CstF-77 protein of () was cloned into a modified pET-15b overexpression plasmid allowing the production of an N-terminally His-tagged fusion protein (). Purification was carried out after cell lysis by centrifugation at 4°C for 1 h at 50 000. The supernatant was incubated in batch with an affinity resin (Talon) and the eluate was loaded on a HiQ-Sepharose (Pharmacia). The protein was concentrated to 30 mg/ml in 25 mM pH 7.5 and 100 mM NaCl. Crystallization of the sample was carried out at room temperature using sitting-drop vapour diffusion by mixing 1 volume of protein solution with 1 volume of 10% PEG 2000 MME, 100 mM Tris–HCl pH 8.0 and 70 mM calcium acetate of reservoir solution (Nextal). Crystals were directly cryoprotected in a solution of 25% Methyl-2 Pentane-Diol, 10% PEG 2000% MME, 100 mM Tris–HCl pH 8.0 and 70 mM calcium acetate and flash-frozen in liquid nitrogen for data collection. Data were processed with XDS (). Data collection and phasing statistics are shown in . The crystal structure of full-length CstF-77 (1–493) was solved to 2.55 Å using phases determined from a SAD (single anomalous dispersion) dataset on a crystal grown by macroseeding with selenomethionine-substituted protein. Thirty-six Se-sites were located using SHELXD () and phases were calculated with SHARP (). An initial model was automatically built using Arp/Warp (). This initial model was used as a template for molecular replacement against the best native dataset. The model was improved by manual docking of residues and missing portions of the molecules with Coot (). Model refinement was achieved with REFMAC5 (). The final model was refined to a resolution of 2.0 Å with a working and free R-values of 27.9 and 22.5%, respectively, and good stereochemistry (). Strikingly, the final model contains two monomers arranged into a non-crystallographic homodimer, in which short stretches of residues at the N- and C-terminal ends are missing (). The final model consists of residues 12 to 465 with the exception of three short loops (62–65, 271–280 and 427–429). Chain B is less defined and consists of residues 12 to 454 with the exception of residues (60–68), (92–111), (131–149), (271–280) and (426–429). Surface conservation has been calculated with Consurf server with a sequence alignment including , , , , and sequences (). The Rna14p constructs (1–677), (1–593) and (589–677) were amplified by PCR from yeast genomic DNA and cloned into the NdeI and BamHI site of a modified vector allowing expression of a protein fused to a His-tag at its N-terminus. The full-length Rna15p protein was amplified from the yeast genome and cloned into the NdeI–XhoI a modified vector. Co-expression assays where carried out by co-transformation of Rosetta cells. Cells were grown up to an OD600 of 0.6 and cooled down to 15°C. Overexpression was induced by an overnight incubation with 1 mM IPTG. Cells were harvested by centrifugation and sonicated. A crude extract sample was saved at this point and boiled in Laemmli buffer (T, total extract). After 10 min centrifugation at 4°C, 13 000 r.p.m., the supernatant was incubated for 30 min with His-tag affinity resin and washed three times with 50 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.1% Triton X-100. The resin was boiled in Laemmli buffer and the samples were resolved by SDS–PAGE (B, bound). The proteins were transferred on a blot and analysed with polyclonal antibodies directed to Rna15p and Rna14p. Monoclonal antibodies were used to reveal His-tag fused proteins (Amersham, GE Healthcare). The structure of CstF-77 is entirely α-helical and consists of 23 α-helices arranged in pairs of anti-parallel α-helices forming 11 HAT repeats as described by Preker and Keller (). It can be divided into two domains, an N-terminal domain containing the first 4 HAT repeats (residues 12 to 151), and a middle domain containing HAT repeats 5 to 12 plus the C-terminal α-helices (residues 162 to 465). Residues 466 to 493 are likely to form an independent domain not seen in our electron density map. Helix 8 provides α-helix B of HAT repeat 4 and α-helix A of HAT repeat 5. It links both domains forming a 145° kink (a and ). The dimer has an overall V-shape with dimensions of 140-Å wide and 60-Å thick, each arm of the V measuring ∼80-Å long (b). The 110° angle between both arms is in good agreement with the one measured from electron microscopy pictures obtained with the yeast Rna14p–Rna15p corresponding complex and with the angle measured for the murine CstF-77 complex (,). The two CstF-77 monomers are oriented tail-to-tail and interact extensively through their middle domain to form a tight homodimer burying 4000 Å of the surface area (a and c). The interface between the monomers is provided by the C-terminal α-helix (α-helix 23) of each monomer interacting with HAT-repeat 11 on the one hand. On the other hand, the interaction is built up by HAT-repeats 9 to 11 from one monomer interacting with HAT-repeats 11 to 9 of the opposite monomer and shielded by a well-organized network of water molecules. Superimposition with the murine CstF-77 orthologue HAT-N and HAT-C domains shows limited differences between the two models (). Three extra α-helices defining 1.5 HAT repeat at the N-terminus are observed in the murine CstF-77 HAT-N domain in comparison to that (a and )(). Interestingly, the two last α-helices observed in the murine and the orthologues have similar orientations but structurally equivalent helices belong to opposite monomers (b). In the murine CstF-77, these helices follow the curve formed by the HAT repeats whereas, in the orthologue, the equivalent helices cross the concave surface defined by HAT repeats 8 to 11 to interact with HAT repeats 6 to 8 (). In contrast to murine CstF-77, the prominent pocket observed on the concave surface of the homodimer is likely to be occluded by residues 426 to 430 that could not be placed in our model. Whether this reflects species-specific characteristics has to be tested. Interestingly, homodimerization of the CstF-77 homologue, the Su(f) protein, has been proposed to account for the genetic complementation of lethal alleles of the gene with different domains of the Su(f) protein (). Similarly, homodimerization of human CstF-77, as well as human CstF-50, were detected by analysis in a CstF mapping study (). In yeast, ultracentrifugation analyses of CF IA subunits Rna14p and Rna15p demonstrated a 1:1 stoichiometry of the complex. In addition, these data suggested that the Rna14p–Rna15p heterodimer self-associates via the Rna14p subunit to form a heterotetramer (). Our model of CstF-77 provides the structural basis for the homodimerization of the protein and its conservation through evolution. Taken together, these data strongly support the idea that CstF functions as a complex comprising two copies of each of its subunits. Orthologues of pre-mRNA 3′-end processing factors have been characterized from yeast to human. Functional complementation between and human CstF-77 has been demonstrated, with the exception of the C-terminal domain (). Therefore, the determinants of interaction are likely to be conserved as well. We performed sequence alignment for seven different CstF-77 homologues (). Apart from the dimerization interface, two areas at the surface of the complex cluster a number of conserved residues (a and b). The first area is located on HAT-repeats 9 to 11 and consists of charged residues (a). The tail-to-tail orientation of the two monomers brings into close vicinity the equivalent areas of each molecule leading to a potentially unique extended and conserved surface. The second highly conserved area of the homodimer is located in the N-terminal domain (b). The conserved residues of HAT-repeats 1 to 4 are located on the external portion of the dimer. Tyr53, Val57, Val70 and Phe73 cluster into this region and form a hydrophobic patch. Due to the V-shape of the complex and the location of the various conserved regions, the CstF-77 homodimer exposes four highly conserved areas provided by the neighbouring and equivalent areas of HAT repeats 9–11, and by the two independent N-terminal domains of the complex (). CstF-77 and its orthologues in yeast and are central for CstF complex formation and for the interaction with CPSF. Indeed, Rna14p forms a tight complex with Rna15p and interacts with Pfs2p, Pcf11p and Nab4p/Hrp1p (,,,). The conserved exposed areas of CstF-77 are likely to provide platforms for the interaction with protein partners. On the basis of our crystal structure, we tested this assumption in pull-down experiments with various deletion constructs of Rna14p co-expressed with Rna15p in (a and b). As expected, full-length His-Rna14p could efficiently pull down Rna15p (a, lane 4). However, deletion of the Rna14p C-terminal domain (residues 593 to 677) resulted in the loss of interaction with Rna15p (a, lane 6). Co-precipitation of Rna15p with Rna14p C-terminal domain confirmed that this portion of the molecule is sufficient to establish an interaction between the two polypeptides (a, lane 8). This domain is important for the function of Rna14p since the shortening of the protein by 16 amino acids at its C-terminus (stop codon at residue 633) observed in the yeast mutant, leads to a defect in 3′ end pre-mRNA processing (). Altogether, these data suggest that alteration of the interaction between Rna14p and Rna15p is the molecular basis for the loss-of-function phenotype observed in mutant strain. In metazoans, similar interaction have been described for CstF-77 and CstF-64 (,). Interestingly, in , a single mutation or insertion within the su(f) protein (su(f)) impairs its function (). The C-terminal portion of CstF-77 is highly conserved in metazoans but differs notably from the one in yeast (). In summary, these data suggest that the C-terminal Pro-rich portion of CstF-77 and its homologues carries a similar function, even though it is not strictly conserved in sequence from yeast to human. Further analysis is required in order to determine whether this domain has a similar structure in yeast and metazoans.

Stable RNAs are synthesized in precursor forms that are then processed by endonucleases and exonucleases to mature forms (). Maturation of stable RNA in is known to involve the following endonucleases: RNase M5 catalyzes 5S rRNA maturation (); RNase P cleaves tRNA precursors to generate the mature 5′ end (); Bs-RNase III, the version of RNase III, is involved in rRNA processing (); RNase Z is required for cleavage of CCA-less tRNA precursors (); and RNase J1 has recently been shown to be involved in 16S rRNA processing (). Following endonuclease cleavage, trimming by 3′-to-5′ exoribonucleases is often required to produce the mature RNA species. Deutscher and colleagues have used mutant strains of that are missing one or more of the eight known 3′-to-5′ exonuclease activities to demonstrate a remarkable ability of many exonucleases to participate in maturation of tRNAs (). We have constructed mutant strains of that are deficient in one or more of the four known 3′-to-5′ exoribonucleases in this organism: polynucleotide phosphorylase (PNPase), RNase R, RNase PH and YhaM (). We have shown that RNase PH plays a significant role in CCA-containing tRNA maturation, although both RNase R and YhaM can also perform this function (). PNPase did not appear to be involved in tRNA maturation. small cytoplasmic RNA (scRNA) is the functional homolog of the 4.5S RNA, which is a constituent of the SRP-like complex of that is involved in protein targeting (). In , 4.5S RNA precursor is cleaved endonucleolytically by RNase P (), and final maturation is accomplished by exonuclease RNase T and, to a lesser extent, RNase PH (). scRNA is transcribed as a 354-nt precursor RNA that, based on computer predictions, contains a short 5′-terminal stem-loop, a longer internal stem-loop structure that includes a large 255-nt loop, and a 3′-terminal stem-loop that functions as the transcription terminator (). The mature scRNA is 271 nt, with a 5′ end mapping to the site of Bs-RNase III cleavage [site ‘A’ in ; (,)]. The 3′ end of mature scRNA (asterisk in ) maps 4 nt upstream of a second Bs-RNase III cleavage site (site ‘b’ in ). It has been proposed that scRNA maturation is the result of cleavage by Bs-RNase III at the A and b sites, followed by exonuclease trimming at the 3′ end to remove the last 4 nt (). Our results suggest that 3′ exonucleases can act on precursor scRNA from a site other than the downstream Bs-RNase III cleavage site. exoribonuclease mutant strains were derivatives of the parent strain BG1, which is and which is designated ‘wild type’ in this study. Construction of these strains has been described previously (), as has isolation of the deletion strain (). Construction of various endonuclease mutant strains, using an triple exonuclease mutant as host, is described in the text. RNA was isolated by hot phenol extraction from cultures grown to mid-logarithmic phase in minimal medium containing Spizizen salts with 0.5% glucose, 0.1% casamino acids, 0.001% yeast extract, 50 μg/ml tryptophan and threonine, and 1 mM MgSO, as described (). Northern blot analysis of RNA separated on 6% denaturing polyacrylamide gels or sequencing gels was done as previously described (). For the triple mutant strains containing endonuclease mutations, a rich culture medium was used, containing 1% yeast extract, 2% tryptone, 1% NaCl, 1% glucose and with or without 1 mM IPTG. Cultures were grown until an OD of 0.6, and RNA was isolated by hot phenol extraction, as above. The scRNA riboprobe was synthesized by T7 RNA polymerase (Ambion) in the presence of [α-P]UTP, using as template an isolated PCR fragment containing the scRNA large internal loop sequence. 5′-end-labeled oligonucleotide probes were prepared using T4 polynucleotide kinase (New England Biolabs) and [γ-P]ATP. To control for RNA loading in northern blot analyses of scRNA processing and of Δ mRNA half-life, membranes were stripped and probed for 5S rRNA, as described (). Size markers on sequencing gels were sequencing reactions done on single-stranded M13mp18 DNA. For the northern blot shown in D, the size marker was I-digested plasmid pSE420, as described previously (). Quantitation of radioactivity in bands on northern blots was done with a Storm 860 PhosphorImager instrument (Molecular Dynamics). Δ mRNA half-life was determined by a linear regression analysis of percent RNA remaining versus time. Preparation of cell extracts was essentially as described previously (), except that the dialysis buffer was 20 mM Tris-HCl (pH 8.0), 60 mM KCl, 5% glycerol, 0.1 mM EDTA, 1 mM DTT and 0.2 mM phenylmethylsulfonyl fluoride. Conditions for Bs-RNase III cleavage and analysis of results were as described (). scRNA processing was analyzed first in the single exoribonuclease mutant strains, each of which was deficient for one of the four known 3′-to-5′ exonucleases. We expected that we might observe a deficiency in the exonucleolytic processing that is thought to initiate after Bs-RNase III cleavage at site b (), resulting in an accumulation of the 275-nt RNA. RNA was isolated from the wild-type and mutant strains, and the RNA was separated on a high-resolution denaturing polyacrylamide gel (‘sequencing gel’) alongside a sequencing ladder that served as a size marker. The sequencing gel was electroblotted and the membrane was probed with an scRNA riboprobe that was complementary to the internal 255-nt loop portion. Additional experiments with 5′- and 3′-specific, end-labeled oligonucleotide probes allowed unambiguous identification of the RNA bands detected by the riboprobe. The results of this northern blot experiment are shown in , lanes 1–5. (The relative sizes and extents of all bands detected in this blot are shown schematically in B.) An intense band representing the mature scRNA ran at 271 nt, and a faint band representing the full-length scRNA precursor (354 nt) was detected. The 275-nt RNA was scRNA that had been cleaved at the 5′ and 3′ Bs-RNase III cleavage sites (A and b in ). The average amount of this 275-nt band, in most strains examined, was ∼20–25%, relative to total scRNA (). A 317-nt RNA band was an scRNA precursor that had been cleaved at the Bs-RNase III site A only, as determined by probing with 5′- and 3′-specific oligonucleotide probes (see ). Two very faint bands below the 317-nt band were barely visible on this exposure, but were clearly present on longer exposures. These were 314- and 310-nt RNAs that were not cleaved by Bs-RNase III and that retained the 5′ end of precursor scRNA (as demonstrated by a 5′-specific probe which was complementary to scRNA nucleotides 1–29; see A). Importantly, we did not observe a band that corresponded to cleavage at the Bs-RNase III site b alone, which would be 312 nt. Although the absolute size of bands detected on these sequencing gel northern blots could not be determined to an accuracy of greater than ±2 nt, there was independent evidence that these 314- and 310-nt bands were not the result of Bs-RNase III cleavage, since they were also present in the strain that is deleted for the gene encoding Bs-RNase III (). scRNA processing was analyzed in double exonuclease mutant strains, which were deficient for PNPase and one of the other 3′-to-5′ exoribonucleases (, lanes 6–8). The lack of RNase PH in the double mutant resulted in readily detectable 283- and 279-nt bands, with similar relative amounts to that of the mutant alone (compare , lines 4 and 7). A small amount of these additional bands was also detected in the double mutant (, line 8). Next, processing of scRNA was analyzed in strains deleted for three of the four exoribonucleases, i.e. containing only one of the known 3′-to-5′ exonucleases (). In the strain that had only RNase PH, the processing pattern was identical to wild type (compare lanes 1 and 4, ). This result firmly established RNase PH as the major 3′-to-5′ exonuclease involved in scRNA processing. Strikingly, in the strain that had only PNPase (, lane 2), the 279-nt RNA now became the dominant precursor product, and there was almost as much of this precursor RNA as there was of the fully processed scRNA (, line 9). This result indicated clearly that cleavage at the Bs-RNase III b site was not a primary step in processing of the downstream side of the scRNA stem structure. Rather, 3′ exonuclease activity starting from either the native 3′ end or from a site generated by endonuclease cleavage downstream of the Bs-RNase III b site was required for efficient scRNA processing. PNPase processivity, initiating at a downstream site, was severely inhibited at the 279-nt position. Whatever the reason for the block to PNPase processivity, the other 3′ exonucleases were more efficient than PNPase at digesting through this point (see Discussion section). The strain that contained only YhaM (, lane 5) showed the same level of 283- and 279-nt RNAs as the single and double mutant strains that had significant amounts of these RNAs (, lines 4, 7 and 12). This result demonstrated that, in the absence of RNase PH, YhaM was capable of relatively efficient scRNA processing. (Actually, the two bands that were detected above the 275-nt precursor in the YhaM-only strain ran slightly faster than the 283- and 279-nt RNAs. We have not investigated this further.) The strain that contained only RNase R (, lane 3) had somewhat higher levels of the 283- and 279-nt RNAs (, line 10), suggesting that RNase R was not as efficient as RNase PH or YhaM in processing scRNA precursor. Finally, results from the quadruple mutant strain (, lane 9) showed that this strain had the least amount of mature scRNA and the highest amount of the 283-nt precursor RNA (, line 13). The presence of mature scRNA even in this mutant strain implies the existence of one or more 3′ exonucleases in in addition to the ones tested here. We tested whether the native 3′ end of scRNA was a possible initiation site for 3′ exonuclease processing. The scRNA 3′ end sequence (nts 327–354) is predicted to form a stable stem-loop structure () with a free energy of −14.2 kcal mol, which is typical of a transcription terminator sequence. We wished to determine whether the scRNA terminator structure could function as a barrier to 3′ exonuclease processivity , thus making it unlikely that 3′ exonuclease processing commences at the native 3′ end. For this experiment, the scRNA terminator structure was used to replace the 3′ terminator structure of a known stable mRNA. We have constructed a derivative of Δ mRNA that is quite stable, due to the presence of a strong 3′ transcription terminator structure, which has a predicted free energy of −22 kcal mol, and an inserted 5′-terminal secondary structure [‘Δ + 14/7A’ in ()]. Because this mRNA is protected at both the 5′ and 3′ ends, it has a long half-life of more than 20 min. We reasoned that, if the scRNA 3′ end structure was a poor barrier to 3′ exonuclease processivity, replacing the 3′ terminator structure of the stable Δ mRNA with the 3′ terminator structure of scRNA would result in an unstable RNA, since protection at the 5′ end would be rendered irrelevant by rapid degradation from the 3′ end. Such a chimeric mRNA was constructed, and northern blot analysis was used to assess mRNA half-life (). A small but not significant difference in the half-lives was observed, indicating that the scRNA 3′ end provides a strong barrier to 3′ exonuclease activity as does the Δ 3′ end. Thus, we hypothesized that 3′ exonucleolytic processing of scRNA begins at an endonuclease cleavage site, located in the 3′-proximal portion of scRNA, downstream of the Bs-RNase III b site. The approximate location of this site is labeled site X in . If 3′ processing that initiates at site X was an important step in formation of mature scRNA, then Bs-RNase III cleavage at site b would likely contribute only in a minor way to scRNA processing. To assess quantitatively the relative efficiency of cleavage at the A and b sites, uniformly labeled precursor scRNA was prepared and was incubated with protein extracts prepared from wild-type or mutant strains. A time course of the Bs-RNase III cleavage reaction is shown in . Cleavage at the A site predominated: at later time points when 100% of the molecules were cleaved at the A site, <20% of the molecules were also cleaved at the b site. We therefore hypothesize that, , cleavage at the b site is inefficient, so that rapid accumulation of fully mature scRNA requires processing by 3′ exonucleases of an scRNA precursor that retains sequences downstream of the b site. Although Bs-RNase III is considered an essential enzyme in , we have isolated a rare mutant strain (presumably containing a second-site suppressor) that survived deletion of , the gene encoding Bs-RNase III (). A shows the pattern of scRNA processing in the strain, as detected by a 5′-end specific probe (complementary to nts 1–29). As expected, there was a massive accumulation of full-length pre-scRNA in this strain, as well as readily detectable 314- and 310-nt products. In experiments not shown here, the riboprobe that was used in the experiments shown in and was used to demonstrate that no mature scRNA could be detected in the strain, as we have observed previously (). The 314- and 310-nt RNAs were only faintly visible in the wild-type strain, and they contained the 5′ end of scRNA, since the probe used in this experiment was a 5′-end specific probe. If we assume that the 5′ end of the 314- and 310-nt RNAs is at +1, then their 3′ ends map within a few nucleotides of the 3′ ends of the 283- and 279-nt RNAs seen in the 3′ exonuclease mutants. (The Bs-RNase III A site is at +37.) This is consistent with the suggestion above that 3′ exonuclease processing of scRNA starts downstream of the Bs-RNase III b site, at an endonuclease cleavage site. The blot shown in A was stripped and then probed with a labeled oligonucleotide that was complementary to nts 304–327 (boxed nucleotides in ). The 314- and 310-nt RNAS were not detected by this probe (B), because there was only partial overlap between the 3′ ends of these RNAs and the probe. The 317-nt RNA, which represents scRNA cleaved at the Bs-RNase III site A only, was detected by this probe in the wild-type strain but not in the mutant, as expected. The results in the quadruple mutant strain are discussed below. The blot was stripped again and then probed with a labeled oligonucleotide that was complementary to nts 313–345 (circled nucleotides in ). This probe detected full-length RNA, as well as the 317-nt RNA arising from cleavage at site A, but no other RNAs were detected (C). If endonuclease cleavage occurred at site X, then the downstream fragment generated by such cleavage should be stable enough to detect, as it would have the exonuclease-resistant transcription terminator at its 3′ end. The 3′-terminal probe, complementary to scRNA nts 313–345, was used in a northern blot analysis of RNA isolated from the wild-type strain, with conditions designed to detect very small RNA fragments. The result in D showed that, indeed, a fragment of ∼35 nt could be detected. This indicated clearly the existence of an endonuclease cleavage site at 320 nt, i.e. site X (). Cleavage at site X, together with cleavage at the Bs-RNase III A site, and without cleavage at the Bs-RNase III b site, would give a 283-nt RNA. Such a product was not detectable in the wild-type strain, but was the largest precursor product detected in the exoribonuclease mutant strains that lacked RNase PH, and was also the most abundant precursor detected in the quadruple mutant strain, when the internal riboprobe was used ( and ; ). We reasoned that the upstream product of cleavage at 320 nt would be labile in the wild type, as it would have an unprotected 3′ end. In the exoribonuclease mutants, however, this upstream cleavage product would be more long-lived (and detectable) since there would be limited trimming from its 3′ end. The results with the 3′-proximal oligonucleotide probe (B) confirmed this expectation. This oligonucleotide was expected to hybridize well with the 283-nt RNA but not with shorter RNAs (e.g. 279-, 275- and 271-nt). When total RNA from the quadruple mutant strain was probed with this oligonucleotide, a prominent 283-nt band was observed (B). In fact, upon long exposures, the same band was detectable even in the wild-type strain (not shown). The sum of the two fragments detected in B and D is 35 + 283 = 318, which is approximately the same size as the scRNA that is cleaved at the Bs-RNase III A site only (measured to be 317 nt). Detection of these two products is consistent with endonuclease cleavage at site X. Candidate endonucleases that could be responsible for cleavage at site X included: RNase J1, RNase J2, RNase M5, RNase P and RNase Z. Strains were constructed that could test the involvement of these endonucleases in scRNA processing. The host for these constructions was the triple exonuclease mutant that contained only PNPase (i.e. the strain, see , lane 2). In this strain, the 283-nt band was easily detectable, and, from our analysis so far, we hypothesized that the presence of this band was associated with cleavage at site X. The continued presence of PNPase in this strain was beneficial, since strains lacking PNPase are deficient in competence (), which would make strain construction difficult. The PNPase-containing triple mutant strain was transformed with chromosomal DNA from strains containing gene disruptions of RNase J2 () or RNase M5 (), and from strains with conditionally expressed RNase J1 (), RNA component of RNase P () or RNase Z (). In the three latter cases, the endonuclease is essential, so expression was under control of the IPTG-inducible p promoter. The triple mutant strain transformants with a conditionally expressed endoribonuclease also contained plasmid pMAP65 (), which provided additional copies of the repressor to reduce as much as possible expression in the absence of IPTG. RNA was isolated from the five triple mutant strains carrying endonuclease mutations. Northern blot analysis was performed using a low-resolution polyacrylamide gel to separate the larger scRNA species. The blot was probed with the 3′-proximal oligonucleotide probe, which detects the 283-nt band (see B). The results in showed that an effect on the 283-nt band was observed only in the strain with the RNase J1 mutation, grown in the absence of IPTG. The most straightforward explanation of our results is that cleavage on the downstream side of the scRNA stem structure is accomplished not only by Bs-RNase III but also by RNase J1. In other words, release of scRNA from upstream and downstream precursor sequences can be accomplished either by two Bs-RNase III cleavages or by an upstream Bs-RNase III cleavage and a downstream RNase J1 cleavage. The data suggest that the latter pathway may predominate. Exonucleolytic trimming to the mature form is achieved primarily by RNase PH. The other three known exoribonucleases can perform this processing, albeit less efficiently. Such redundancy is similar to what we have observed in previous work on the processing of tRNA in (), where RNase PH played the major role in tRNA 3′ end maturation, with other exonucleases also able to fill this role in its absence. The data suggest that a significant portion of precursor scRNA is cleaved at site X: In the triple mutant strain that is missing RNase PH and in the quadruple mutant strain, a full 45% of the total scRNA is in the 283- or 279-nt form (, lines 9 and 13). These forms represent scRNAs that have not been cut by Bs-RNase III at site b. The 279-nt RNA is presumably a product of limited digestion from the 3′ end of the 283-nt RNA. The structural representation of scRNA in does not suggest a reason why digestion should be impeded at this point (see below). The 275-nt RNA is presumably the result of Bs-RNase III cleavage at sites A and b. Interestingly, the relative amount of the 275-nt RNA in almost all strains examined () is similar to what was observed in the experiment (). The ability of RNase III to cleave on both sides of a target stem depends on the structure of the stem. In experiments with T7 phage RNA substrates, stems with an internal ‘bubble’ tend to be cut on the downstream side only, or primarily, while substrates with extensively base-paired stems tend to be cut on both sides of the stem (,). Using SP82 phage target RNAs, which have internal loop sequences, we have found that Bs-RNase III cleaves such RNAs at only one site in the downstream sequence of the predicted stem structure (). We are unaware of a report describing a case where a substrate of RNase III is cleaved preferentially at the upstream site. A variant of T7 R1.1 RNA, in which the asymmetric internal loop sequence was rotated, did change the cleavage specificity to recognize primarily the upstream site (). This latter result demonstrated that RNase III, in principle, can cleave efficiently at the upstream site of a substrate with a particular configuration. In the case of scRNA, early evidence from S1 mapping studies indicated that the upstream site was cleaved first (), and cleavage of scRNA with purified Bs-RNase III showed a preference for cleavage at the upstream site (). Our results (), and the lack of a 312-nt fragment that would represent cleavage at site b, support the notion that Bs-RNase III cleaves efficiently at site A and inefficiently at site b. In a previous report on turnover of seven small monocistronic mRNAs (), no decay intermediates could be detected in a strain that contained only PNPase, whereas there was an accumulation of decay intermediates in any strain that was missing PNPase. A detailed analysis of mRNA showed that these intermediates had predicted secondary structure at their 3′ ends, suggesting that PNPase could degrade rapidly past such structures while other 3′ exonucleases could not. Thus it was surprising to find here that, with the exception of the quadruple mutant, the strain containing only PNPase accumulated by far the highest amount of immature scRNA molecules (, line 9). Interestingly, we have found previously an RNA substrate with a particular secondary structure that provides a strong block to PNPase processivity , but the same structure does not block other exonucleases in a strain that lacks PNPase (). It is likely that full-length scRNA forms complex secondary and tertiary structures, with unpredictable consequences on processing. Furthermore, scRNA associates with at least three proteins—Ffh, FtsY and Hbsu (). We note that the strain deleted for Bs-RNase III, in which almost 90% of the scRNA is in an unprocessed form and no mature scRNA is detectable, can still grow well, although it does shows a measurable growth defect (). Thus, it is possible that even unprocessed scRNA interacts with its protein partners to function in protein trafficking, and such interactions could also affect RNA processing. In fact, the presence of bound proteins could explain the striking strong stop to processing at the 3′ end of scRNA (indicated by the asterisk in ). No products smaller than the mature 271-nt scRNA are detectable. Based solely on the 2D predicted structure, it is hard to understand why 3′ exonuclease processivity stops just at this site. Complex RNA structure and/or the presence of bound protein are likely to be factors in determining the ultimate 3′ end of the mature scRNA. A closer inspection of the data in reveals some noteworthy results. The triple mutant strain that contained only RNase PH gave no detectable 283- and 279-nt products, had a lower than normal amount of the 275-nt product, and had the highest amount of fully processed scRNA (, line 11). Compare these results with the data from the double mutant strain that contained RNase PH RNase R (, line 8), where there was less of the fully processed scRNA, more of the 275-nt RNA, and detectable 283- and 279-nt RNAs. Paradoxically, the presence of an additional 3′ exonuclease resulted in a decreased amount of processing from 3′ ends. We speculate that RNase PH trimming is more efficient in a strain where other 3′ exonucleases are not present, since there is less competition for binding to the 3′ ends generated by endonuclease cleavage. Similar observations have been made in , where certain mRNAs and a regulatory RNA decay faster in an RNase II-deficient strain than in the wild type (). Another notable result seen in was the small amount of 275-nt RNA in the triple mutant that contained only PNPase (, line 9; , lane 2). This was significantly less than in all the other strains. We do not have a good explanation for this result. If the 275-nt RNA is generated solely by Bs-RNase III cleavage at sites A and b, one might not expect the amount of this RNA to be greatly affected by reducing the level of exonuclease activity present. However, if the 275-nt RNA could also be generated by cleavage at site X followed by exonuclease processing, one could understand why there is less of this RNA in a triple exonuclease mutant that contained only PNPase. However, if this were the case, then one would expect little 275-nt RNA in the quadruple mutant, and the data showed an amount of scRNA in this mutant that was similar to wild type. The pattern of scRNA in the Bs-RNase III deficient strain was of interest. Besides the full-length scRNA, products of 314 and 310 nt were observed (A), which we propose are products of cleavage at site X, followed by exonuclease trimming that is blocked at nts 314 or 310. (An RNA species that had its 3′ end at the same site [nt 308] as in mature scRNA was not observed in the Bs-RNase III-deficient strain. We conjecture that, in the wild type, exonucleolytic trimming to the mature 3′ end is aided by prior Bs-RNase III cleavage at the A site, which perturbs the double-stranded nature of this part of the molecule.) The 314- and 310-nt products were barely detectable in strains ( and ; A, wild-type and quadruple mutant lanes) but constituted ∼11% of the total scRNA in the strain. If, as we have argued above, cleavage at site X is a significant component in scRNA processing, one might have expected to find a higher abundance of these products in a strain that was cleaved endonucleolytically only at site X. However, one needs to consider the effect of Bs-RNase III cleavage on the susceptibility of site X. In the full-length precursor form, the site X cleavage sequence is in a double-stranded state. If we assume that the endonuclease responsible for site X cleavage is single-strand specific, then little cleavage would occur here. After cleavage by Bs-RNase III at site A, an upstream product of 37 nt with an unprotected 3′ end is formed. It is likely that this first step in scRNA processing is followed by rapid degradation of the 37-nt 5′ fragment, which then leaves site X in a single-stranded form, rendering it susceptible to endonuclease cleavage. [In fact, using the 5′-terminal oligonucleotide probe, we could not detect the 37-nt 5′ fragment of Bs-RNase III cleavage at site A in the wild-type strain. This fragment was detectable, however, in the quadruple mutant strain (data not shown).] In the absence of Bs-RNase III cleavage at site A, site X is likely to be only partially in a single-stranded state and is therefore less susceptible to endonuclease cleavage. [This could also explain why cleavage at site X was not observed in the extract obtained from the strain (). In addition the conditions were optimized for Bs-RNase III cleavage, which are different from the conditions used by others for RNase J1 cleavage]. Initial results showed that RNase J1 was responsible for cleavage at site X (). The difference in the level of the 283-nt band in RNA isolated from the RNase J1 mutant strain grown in the presence or absence of IPTG, in independent experiments, was 3–5-fold. It is likely that the p promoter is somewhat leaky, providing the cells with a low level of RNase J1 even in the absence of IPTG. To eliminate the possibility that the effect on scRNA processing that we see in the RNase J1 mutant is indirect, experiments will be needed to determine whether RNase J1 cleaves scRNA directly. However, such experiments may require reconstitution of scRNA with its bound proteins, as this likely affects the presentation of scRNA to RNase J1 . RNase J1 has now been implicated in the processing of two stable RNAs, 16S rRNA () and scRNA, as well as and leader RNAs (). Much work remains to determine whether RNase J1 is involved more generally in RNA processing and mRNA decay in .

Nonsense mediated decay (NMD) is a surveillance pathway by which cells recognize and limit the expression of mRNAs containing premature stop codons (PTCs) and thus reduce the expression of potentially harmful truncated proteins (). Originally, NMD was thought to represent a control mechanism to limit the expression of faulty transcripts with frameshift or nonsense mutations, which originate from point mutations or from aberrant splicing. The finding of NMD being involved in negative feedback loops regulating normal gene expression foreshadowed a wider role of NMD as a basic post-transcriptional cellular process (). More recently, microarray analyses of yeast (,), () and human cells () have revealed that NMD modulates the levels of a large number of normal transcripts. Furthermore, NMD has been suggested to vary in its efficiency. In , the degradation of the pre-mRNA of CYH2 (an endogenous NMD target) has been reported to vary in different strains (). In humans, the expression of dystrophin and genes carrying identical nonsense mutations has been reported to differ and to modulate disease severity (,). Moreover, tissue-specific differences of NMD efficiency for nonsense-mutated collagen X have been suggested in a patient with Schmid metaphyseal chondrodysplasia (). More recently, intertissue and interindividual variations in NMD efficiency have been proposed in the study of two fetuses diagnosed with Roberts syndrome and carrying a homozygous frameshift mutation in the gene (). These observations led to the hypothesis that variations of NMD efficiency may contribute to the phenotypic variability of hereditary disorders (,). However, it has so far been difficult to quantify NMD efficiency. Here, we have developed an assay system that estimates differences of NMD efficiency based on an internally controlled measurement of the expression of cellular NMD targets. Applying this assay in a HeLa cell model system we demonstrate variable NMD efficiency between strains. Functionally, these differences are shown to be caused by a deficiency of RNPS1, a key protein in at least one of the known NMD pathways (). We thus propose that cell type specific co-factor availability represents a novel principle that controls NMD. HeLa cells were grown in DMEM supplemented with 10% fetal calf serum (FCS) and 1% penicillin/streptomycin at 37°C and 5% CO. HeLa strain A has been used by our laboratory for many years (,). Strain B (ACC 57) was purchased at the German Repository of Cell lines (DSMZ). Strain C was kindly provided by Dr Elisa Izaurralde (EMBL, Heidelberg). For plasmid and siRNA transfections, we used previously described methods (). We isolated RNA according to standard protocols with TRIzol reagent (Invitrogen, CA, USA) and performed northern blot analysis as described previously () using 2–3 µg RNA per lane. Target sequences of siRNAs for luciferase, and were described previously (). For estimations of mRNA half-life, actinomycinD (5 μg/ml) was added to the growth medium 48 h after siRNA treatment and RNA was collected every hour. Transcript abundance was quantified by quantitative RT-PCR as in the other cases. The half-life of the transcript was used to monitor efficient inhibition of transcription. We assessed the integrity of total cytoplasmic RNA from the cultured cells using a Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA, USA). We performed preparation, processing and hybridization of labelled and fragmented cRNA targets to Affymetrix HG_U133A GeneChips™ according to the manufacturer's protocols (Affymetrix Inc., Santa Clara, CA, USA). Oligonucleotide arrays were scanned using a confocal laser scanner (GeneArray™, Hewlett Packard, Palo Alto, CA, USA). Three independent experiments with siRNA or Luciferase siRNA as a negative control were analysed. We used the Affymetrix GeneChip Suite 5.0 software (MAS 5.0) to calculate raw expression values for each of the 22 283 probe sets on the U133A oligonucleotide array. Signal intensities were calculated as average intensity difference (AID) between perfect and mismatch probes. Approximately 8800 probe sets continuously resulting in absent calls were excluded from the analyses. Next, we used GeneSpring 4.2.1 (Silicon Genetics, Redwood City, CA, USA) for scaling, normalization and background correction of all genes and arrays. We performed Student's -test on normalized relative expression ratios to identify significant differentially expressed genes with a minimum factor of difference of >2-fold, within the 95% confidence interval ( < 0.05). Full data sets are available in the Supplementary Data and on the Gene Expression Omnibus (GEO) repository (GSE7009). We synthesized first strand cDNA using MuMLV RNaseH- Reverse Transcriptase (MBI Fermentas) according to the manufacturer's protocol using 4 µg of RNA. We carried out real-time PCR, using the LightCycler system (Roche Diagnostics, Mannheim, Germany), as an independent method to assess differences of gene expression and to validate the microarray expression data. We performed expression analyses of selected genes with single-stranded cDNA and gene-specific primers (primer sequences are available on request). We used the FastStart DNA Master SYBR Green kit (Roche Diagnostics) to quantify the mRNA levels by measuring real-time fluorimetric intensity of SYBR green I incorporation. The working concentrations of gene-specific primer, MgCl, enzyme and SYBR green as well as cycling parameters were optimized according to the LightCycler protocol (LightCycler Operator's Manual, Version 3.5). For the experiments done in exclusively in strain A cells, we used the concentration of () to normalize all other genes tested from identical cDNA samples. For the other experiments also the (), () and () were included as standard controls. The ratio of each analysed cDNA was determined as the mean of 4 or 5 experiments. Melting curves of the PCR products were performed for quality control. The primer sequences of SC35 and GAPDH were described previously (). For : gcagtcatttaccacatgc/tattgtttctgcttcttggat, for : gagtgagactgactgcaagc/tcttattaattcgcaaactgg, for : attgtgctggatgccgaga/acctagcgtggtcactccgta, for : ttgacaacagggttcgtag/ttcttggaggaaacattgtg, for : gaccagtcaacaggggacat/aacacttcgtggggtccttttc and for : gcccatctttacatacaca/acttcaaattat tactggctac. We prepared protein lysates with an isotonic lysis buffer as described previously (). For total extracts, the buffer composition was 50 mM Tris-HCl, pH 7.5, 150 mM NaCl,1 mM EDTA,1% Triton X-100, 0.5% Deoxycholate, 0.1% SDS,1× Complete protease inhibitor (Roche). For the cytoplasmic fraction, the buffer was 50 mM Tris-HCl, pH 7.2, 150 mM NaCl, 0.5% (v/v) NP-40, 0.1% Deoxycholate, 5 mM Vanadyl-Ribosyl-complex, 1 mM Dithiothreitol, 0.5 mM PMSF, 1× Complete protease inhibitor (Roche). We performed immunoblot analysis of protein samples using 10–15 µg of total protein per lane as previously described (). Plasmids for the expression of human β-globin WT and NS39 (), Y14, RNPS1 and RNPS1Δ69-121 () and the loading control () were described previously. We aimed at developing an assay to estimate differences in NMD efficiency based on the expression levels of physiological NMD transcripts. To identify a panel of endogenous direct NMD targets in human cells, HeLa cells were treated with siRNA against the NMD-key factor (,,) or Luciferase as a negative control. UPF1-specific immunoblotting showed that this protein was efficiently depleted to a level of <10% (a). Functionally, the inhibition of NMD was assessed by monitoring the expression of transfected nonsense mutated β-globin mRNA (NS39) (b), and of two known NMD-sensitive splice variants of [, referred to as SC35A and B] (c and d). In UPF1-depleted cells, both the β-globin NS39 reporter and the NMD sensitive isoforms were up-modulated ∼5- and 15-fold, respectively, demonstrating the effective inhibition of NMD. RNA isolated from these cells was analysed on Affymetrix HG_U133A GeneChips™. Of 22 283 probe sets, representing ∼14 500 human genes, 9336 transcripts were expressed at a level of more than two SDs above background and were thus included in the analysis. A total of 265 probe sets (2.8%) representing 227 genes were up-modulated more than 2-fold, while 248 probe sets (2.6%) representing 202 genes were down-modulated more than 2-fold (Supplementary Data, Tables 1 and 2). These data indicate that a substantial number of genes are affected directly or indirectly by UPF1 activity. In order to exclude transcripts that are affected by UPF1 depletion in an NMD-independent, non-post-transcriptional fashion, we analysed mRNA and pre-mRNA levels in a subset of 16 transcripts, chosen because of their strong differential expression in the microarray analyses. In several independent experiments performed on UPF1-depleted HeLa cells that showed efficiently inhibited NMD function (see ), pre-mRNA and mRNA levels for the selected 16 transcripts were quantified by RT-PCR (). The microarray data showing up-regulated mRNA abundance in UPF1-depleted cells could be confirmed by RT-PCR for all 16 transcripts. However, only in the case of TBL2 the abundance of the pre-mRNA remained unchanged while the abundance of the mRNA was up-modulated ∼8-fold. In the case of NAT9, these differences were marginal. In all other 14 RNAs, the abundance of the pre-mRNA and the mRNA did not differ significantly, although in two (, ) the pre-mRNA remained below the threshold of 2-fold up-regulation, whereas the mRNA was up-regulated to a level of >2-fold. These data suggest that most of these mRNAs are likely up-modulated transcriptionally and do not represent NMD targets. By implication, these data also suggest that a substantial fraction, likely most of the almost 230 transcripts that are up-modulated by UPF1 depletion in our microarray data are indirect NMD targets. Transcripts that are targeted by NMD are expected to be stabilized by an inhibition of this pathway. We thus analysed the decay rates of the , , and mRNAs. We also included the transcript, which has previously been suggested to represent an endogenous NMD target by analysis () and is experimentally shown to be up-modulated by UPF1 depletion here (see below). Actinomycin D was added to cells pre-treated with siRNA against UPF1 or Luciferase. The short-lived transcript was used as a positive control to assess the block of transcription (a). Prolonged half-lives in UPF1-depleted cells were detected for , and confirming that UPF1 depletion increases the abundance of these transcripts by reducing degradation (b–d). It is interesting to note that the degradation curve of the transcript appears to be biphasic while those of the and transcripts appear to be monophasic. A biphasic decay curve for NMD substrates has been described previously (,,) and can potentially be attributed to degradation of the nonsense-mutated mRNA during the first round of translation. Those mRNAs that escape degradation at that point are thought to be unaffected thereafter by NMD (). The stability of and did not show any effect on UPF1 depletion (e–f). Because of this and because of the only marginal difference between pre-mRNA and mRNA levels (), we excluded these transcripts from further analysis. The role of NMD in directly modulating the abundance of the , and transcripts was further analysed by depleting UPF2, which interacts with UPF1 in the NMD pathway (,). The efficient depletion of UPF2 to ∼10% was confirmed by immunoblotting (a) and, as a functional control, we assessed the abundance of (A) and (B) isoforms (referred to in the subsequent discussion as SC35). The degree of up-modulation in UPF1-depleted and UPF2-depleted cells was not significantly different for all four analysed transcripts (b). Taken together, these results indicate that , , and are NMD targets that depend on both UPF1 and UPF2. Analysis of the structure of these transcripts using sequence databases show that SC35 (A and B), TBL2 and GADD45B possess a termination codon located more than 55 bases from the last exon–exon junction, while NAT9 contains an upstream open reading frame (uORF) (Supplementary Figure 1). These structural features are typical for cellular NMD targets (,), which may explain the sensitivity of these endogenous mRNAs to cellular NMD activity. The panel of five validated cellular NMD target transcripts (SC35 A+B, TBL2, NAT9 and GADD45B) was used to systematically analyse the NMD efficiency of three different HeLa cell strains (referred to as A, B and C). To avoid a potential bias of quantification against a single housekeeping gene, we selected four different transcripts (HPRT1, CBFB, GAPDH and RPL32) for normalisation purposes (,). This group of control transcripts was selected because they showed <10% variability in all of our microarray experiments (data not shown); they were expressed at different steady-state levels and they belong to different metabolic pathways and are thus unlikely to be co-regulated. The comparison of the degree of up-modulation following UPF1 depletion showed similar results for all transcripts that were used for normalisation (Supplementary Figure 2), which indicated that all of these housekeeping genes can be used as standards. Quantification of the five endogenous NMD targets (SC35 A+B combined, GADD45B, NAT9 and TBL2) against the four standards in these strains gave reproducible results in four independent experiments (a). All the transcripts were ∼2- to 3-fold significantly more abundant in strains B and C in comparison with strain A. Strain C showed a trend towards lower mean expression levels for the NMD targets than strain B, although these differences were not statistically significant. When the data from the individual NMD targets were combined, the same differences existed, which indicates the similar behaviour for all the tested mRNAs and suggests a stronger NMD capacity in cells of strain A (b). To validate our analysis, we estimated NMD efficiency by a direct comparison of the down-modulation of transfected, nonsense mutated β-globin (NS39) reporter in four independent experiments (c and d). The down-modulation of the NS39 reporter differed reproducibly and significantly between strains. In strain A NMD efficiency was ∼2.5-fold stronger than in the strains B and C, while strain C tended to be ∼1.5-fold stronger than B. Thus, the quantification of NMD efficiency by analysis of cellular NMD target transcripts was confirmed by the independent analysis of the NS39 reporter. Subsequently, we aimed at gaining insight into the mechanism of variable NMD efficiency in these HeLa strains. As a starting point, we analysed the abundance of the key NMD proteins UPF1, UPF2 and UPF3b and of the functionally critical exon junction complex components Y14, Magoh and RNPS1 by immunoblotting in both, total and cytoplasmic lysates (a). The abundance of the UPF proteins, Y14 and Magoh did not differ between the three strains. In contrast, RNPS1 is shown to be less abundant in cells of strain B (a). To estimate this difference semi-quantitatively, we compared the abundance of RNPS1 in lysates of cells of strain B relative to dilutions of similar lysates of cells of strain A and C. These results indicate that RNPS1 is ∼50% less abundant in the cytoplasmic fraction in cells of strain B (b, left panel). In total lysates, the abundance of RNPS1 in strain B is ∼30% of that in strains A or C (b, right panel). We next functionally analysed if RNPS1 might be the limiting factor for NMD in these cells and over-expressed functional RNPS1 () in cells that were transfected with β-globin reporter genes. We confirmed that the transfection of pCI-NEO-Flag has no effect on the abundance of endogenous RNPS1 in any strain (a, lower western blot) and that the pCI-NEO-RNPS1 is expressed at similar levels in the all three cell lines (a, upper western blot). Increasing amounts of RNPS1 decreased the steady-state levels of the β-globin NS39 reporter up to 4-fold in cells of strain B but had no effect in cells of strains A and C (a). This effect is specific for RNPS1, because over-expression of RNPS1Δ69-121 (a truncated version of RNPS1 known to be non-functional in NMD ()) does not affect the down-modulation of the NS39 reporter (b). Furthermore, the over-expression of Y14—a critical EJC component for NMD function ()—does not augment NMD efficiency in this strain of HeLa cells (c). Based on the differences of RNPS1 abundance, the reconstitution of NMD efficiency by over-expression of a functional protein but not of a non-functional mutant and finally the lack of an effect of over-expressing another critical NMD protein, we conclude that the abundance of RNPS1 is limiting for NMD efficiency in HeLa strain B. NMD has recently emerged as one of the critical post-transcriptional processes that regulate gene expression by targeting transcripts with truncated reading frames (). While the phenomenon of variable NMD efficiency has been observed by many groups studying NMD (,), we document here that NMD efficiency can be systematically analysed by quantifying cellular NMD targets. Such cellular NMD targets have previously been thought to represent ∼1–10% of the total transcriptome of human cells and yeast (,). However, our simultaneous analysis of pre-mRNA and mRNA abundance and of mRNA stability of selected transcripts ( and ) suggests that only a minority of UPF1-dependent transcripts are up-modulated directly by an inactivation of NMD. This apparently transcriptional effect of UPF1 depletion may be caused by influencing the expression of transcription factors either in an NMD-dependent fashion or in a fashion that is related to the non-NMD functions of UPF1 (,). This would indirectly affect the synthesis and the pre-mRNA abundance of target genes. Alternatively, the UPF1 depletion may stimulate the transcription of the up-modulated genes directly. The five cellular NMD target mRNAs that were analysed (SC35A, SC35B and the identified TBL2, GADD45B and NAT9) here were also shown to be UPF2-sensitive () and to contain structural features (alternative splicing isoforms with premature stop codons and uORFs) (Supplementary Figure 1) that explain their NMD sensitivity. Interestingly, the quantification of this small set of carefully validated cellular NMD targets reflected subtle differences of NMD efficiency in different strains of the same cell line thus demonstrating that NMD efficiency can be measured semi-quantitatively. Such measurements may help to analyse NMD efficiency in more complex systems such as in tissues or even in entire organisms. However, the heterogeneity of the composition of such material will have to be controlled as a likely confounding factor of such measurements. NMD variability has previously been studied systematically only in yeast (). The analysis of the yeast NMD substrate CYH2 pre-mRNA in strain crosses suggested that the variable efficiency of NMD is pleiotropic in this organism. Although we cannot discard a multi-gene effect to also be important in human cells, the findings reported here document that the abundance and functional availability of a single NMD co-factor can be limiting for NMD efficiency. NMD is thought to require the interaction of the exon junction complex (EJC) with the SURF complex that is recruited to the ribosome at the site of translation termination (). The EJC is recruited to the RNA by the spliceosome and is remodelled during nucleo-cytoplasmic export (). Structural analyses have shown that the EJC is anchored to the RNA by a core that consists of the proteins eIF4AIII and MNL51 (BTZ) and the Y14/Magoh heterodimer (,). At the periphery of the complex, a number of other proteins are thought to establish the interaction of the EJC with other protein networks and different cellular functions (,). The protein RNPS1 is one of these peripheral EJC proteins that have previously been shown to activate the NMD pathway following tethering to a NMD competent position of the mRNA (,) and to be an important component of one of two pathways implicated in NMD (). Interestingly, the data reported here now functionally link the reduced abundance of this protein in one of the cell lines to low NMD efficiency, thus for the first time implicating the natural abundance of an EJC protein to the efficiency of NMD. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Degradation of RNA is an important factor in the regulation of gene expression. Impairment of regulation of mRNA stability was implicated in the pathogenesis of cancer, inflammatory diseases and Alzheimer's disease (). Enzymes involved in RNA degradation fall into two major classes: endoribonucleases, which cleave RNAs internally and exoribonucleases, which degrade RNAs from the ends. The phylogenetic distribution of endo- and exoribonucleases () in genomes clearly shows that different species vary considerably with respect to the number and variety of the RNases they harbor. Interestingly, two ribonucleases that are essential in do not have homologous counterparts in : RNase E () and oligoribonuclease (). Belonging to the degradosome, RNase E is widely believed to be the enzyme initiating mRNA decay (). Oligoribonuclease, Orn, is the only known exoribonuclease that is essential in (). The essentiality in is due to its unique ability to degrade RNA oligonucleotides with a length of 5 nt and shorter (), and oligonucleotides of these lengths were shown to accumulate in a conditional -mutant (). We would like to introduce the term ‘nanoRNA’ here to distinguish these extremely short oligonucleotides from the longer microRNAs. We chose the term nano in reference to its roots: Nano originates from the Greek word nanos, which means dwarf. Micro on the other hand descends from the Greek word mikros, which means small. Nano is therefore used in this context simply to articulate ‘smaller than’ micro. A recent study reveals the structural basis for the constraints preventing RNase II as a member of the RNR exoribonuclease family from degrading oligonucleotides shorter than 5 nt (). Another member of this exoribonuclease family, RNase R was shown to processively degrade RNA in a 3′ to 5′ directed manner until a di- or trimer remains which cannot be degraded further by this enzyme (). This size limit is therefore likely to be common at least among the members of this important family of exoribonucleases. This highlights the importance of enzymes that have the ability to degrade nanoRNA and thus bring the degradation of RNA to completion. Absence of an oligoribonuclease in Firmicutes is in contrast to its general presence in Gram-negative prokaryotic genomes as well as in eukaryotic genomes (). This prompted us to question which enzyme could functionally replace oligoribonuclease in these organisms. We had recently discovered an unexpected link between sulfur- and RNA metabolism: oligoribonuclease binds to 3′-phosphoadenosine 5′-phosphate (pAp) and is sensitive to micromolar amounts of the nucleotide (). pAp is generated in sulfur assimilation and was implicated in the molecular mechanism of lithium's action in the treatment of bipolar disorder due to strong inhibition of pAp-phosphatase by lithium (). The interaction between pAp and oligoribonuclease was documented for oligoribonuclease, Orn and its human homolog, Sfn. The purpose of this work was to explore whether the conserved interaction between pAp and oligoribonucleases could be exploited to identify a functional analog of Orn in . Surprisingly, the protein identified by this route, YtqI, points to the existence of an even closer link between sulfur- and RNA-metabolism in this organism: YtqI can degrade both nanoRNA and pAp . The pAp-degrading activity of YtqI is similar in magnitude to that of CysQ, the pAp-phosphatase from . Consistent with its activities, YtqI can replace both Orn and pAp-phosphatase (CysQ) in . strains were grown in LB or MOPS minimal medium () containing 40 μg/ml of amino acids as indicated, K-phosphate at 2 mM, vitamin B1 at 0.0005%, biotin at 0.001% when needed, glycerol at 0.4%, glucose or arabinose as indicated. was grown in minimal medium (). Ampicillin (100 μg/ml), kanamycin (25 μg/ml) or erythromycin (1 μg/ml) was added for plasmid maintenance or to select for chromosomal marker. Anhydrotetracycline (Atc) was added at 250 ng/ml for induction of P. To test growth in the absence of cysteine (), overnight cultures grown in MOPS minimal medium containing all amino acids were washed twice with medium lacking cysteine before inoculation into medium containing cysteine (100 μM) or lacking this amino acid. The plasmid for expression of his-tagged YtqI under control of the arabinose-inducible promoter P (pUM412) was constructed as follows: Primer UM175 and UM176 were used to PCR-amplify from 168 chromosomal DNA. The EcoRI, XhoI digested fragment was used to replace the EcoRI/XhoI fragment of pUM407 coding for Orn leaving the region coding for the C-terminal his-tag and the ribosomal-binding site intact. The conditional mutant (strain UM341) uses the anhydrotetracycline (Atc)-inducible promoter P () together with a Tet-repressor (TetR) to ensure tight control in the absence of Atc. This strain was created by introducing the P promoter in front of together with a cassette coding for TetR and a kanamycin selection marker (Km). Two PCR fragments were amplified: PCR1 amplified P, and the transcription terminator T from pZE21-MCS1 () using primer UM153 and UM156, PCR2 amplified including its constitutive promoter P and terminator T from chromosomal DNA of DH5αZ1 () using primer UM155 and UM154. pZE21-MCS1 and DH5αZ1 were kindly provided by Hermann Bujard. The outside primers UM155 and UM156 and equimolar amounts of PCR fragments 1 and 2 were used to perform overlapping PCR. The obtained PCR fragment was then cloned into pGEMT-Easy (Promega) by TA cloning followed by sequencing using primer UM172 and UM173. A verified clone was used as template for PCR amplification using primers UM155 and UM156. The obtained PCR fragment was transformed into CF10230 to create the mutant by lambda Red-assisted recombination according to the protocol of Yu . (). CF10230 is a nic+ derivative of DY329 () that was kindly supplied by Michael Cashel (Cashel,M., unpublished data). Mutants were verified by confirmation of the 5′ site of integration into the chromosome by means of PCR using primers UM158 and UM159 yielding a 432 bp fragment, as well as the 3′ site of integration using primer UM160 and UM161, yielding a 446 bp fragment. The mutant we used here has been described before (). We will refer to mutant 1 as UM285 from now on. UM285 has a replacement of the complete coding DNA sequence (CDS) for CysQ by . The mutant strain (BSF66) was part of the European/Japanese effort to inactivate the whole gene set of 168 and has an insertion of pMUTIN2MCS after the codon for amino acid 108 (). pAp-agarose-binding experiments were performed as described previously (). Two hundred milliliter cultures of 168 or a protease-deficient mutant, DB430 () were grown in minimal medium containing 1.5 mM MgSO at 37°C to an OD between 1.6 and 1.8. Cells were harvested and washed once with 50 mM NaPO pH 8.0, 300 mM NaCl before freezing. Frozen pellets were resuspended in 2 ml pAp-agarose buffer (50 mM HEPES, pH 7.5, 10 mM CaCl, 50 mM KCl) containing 100 mM NaCl, 0.4 mM phenylmethylsulfonyl fluoride (PMSF) and 130 μg/ml lysozyme. After incubation for 45 min on ice, the cells were opened using a Fastprep apparatus (Bio101). Blocking with agarose beads, incubation with pAp-agarose, elution, PAA electrophoresis and identification of proteins was done exactly as described before (). YtqI was purified from a 200 ml culture of MG1655 carrying pUM412 according to the his-tag purification protocol described previously (). Activity assays determining nanoRNase activity were performed using custom-made RNA oligo 5-mers or 3-mers (5′Cy5-CCCCC3′ or 5′Cy5-CCC3′) as substrates in reactions containing 50 mM HEPES, pH 7.5, 5 mM MnCl, 1.6–3.4 μM substrate. At intervals, 4.5 μl reaction aliquots were taken and stopped by adding to an equal volume of sample buffer (4× TBE, 100 mM DTT, 16% glycerol, 20 mM EDTA and frozen at −20°C. For analysis of the reaction products, 1.5 or 2.5 μl of samples were applied to PAA gel electrophoresis on a 22% SDS-PAA gel containing 2× TBE and run in 2× TBE. Fluorescent RNA oligos were visualized using a Molecular Dynamics STORM 860 in 650-nm long-pass filter mode. Quantification of the data was done by calculating the percent of fluorescence of each band at a given time point relative to the total fluorescence of the same time point. Separating reaction products on 22% SDS-PAA gels, we observed a reverse migration phenomenon. This effect can be accounted for by the fact that cyanine dyes have a lower net negative charge than nucleic acids: thus, removing nucleotides will reduce the charge relative to the mass of the oligonucleotide and cause it to shift up instead of down. Assays determining degradation activity on a longer substrate were done using a custom-made RNA 24-mer (5′CACACACACACACACACACACACA3′) that was 5′-end labeled with [γ-P]ATP. This oligonucleotide was labeled using the MirVana Probe and Marker Kit (Ambion) in a 20 μl reaction containing 100 pmol oligo, 6.7 pmol [γ-P]ATP (20 μCi), 90 pmol ATP and 1 μl T4 Polynucleotide Kinase. Incubation was done for one hour at 37°C. The reaction was stopped by the addition of 2 μl of 10 mM EDTA and incubation at 95°C for 2 min. The reaction mixture was purified from the unincorporated nucleotides using NucAway spin columns (Ambion) according to the instructions of the supplier. Three microliter of the labeled RNA (∼800 000 c.p.m.) were used in a 20 μl reaction containing 5 mM MnCl, 50 mM HEPES pH 7.5 and 3 μg enzyme; incubation was for 30 min at 37°C after which the reaction was stopped by the addition of 20 μl loading buffer and incubation for 3 min at 95°C. An aliquot of 5 μl of the samples were resolved on a 20% PAA, 7 M Urea gel containing 2× TBE that was ran in 2× TBE. Labeling of the decade-marker was done with [γ-P]ATP (125 μCi, 1.7 pmol) as suggested by the manufacturer including purification from unincorporated nucleotides. One-fourth of the total volume of labeled decade-marker was used per gel; this amount corresponded to ∼800 000 c.p.m. A previously characterized nuclease (with DNase and RNase activity) from , YhaM () served as positive control in experiments on RNA 24-mers. This protein was purified employing a C-terminal his-tag. HEPES was replaced by Tris pH 8.0 in the reaction containing YhaM as this enzyme is less active in HEPES. pAp degradation was assayed in 20 μl reactions containing 6 mM pAp, 2 mM MnCl, 50 mM HEPES (pH 7.5) at 37°C. Reactions were started by the addition of 1 μg YtqI. Aliquots of 4.5 μl were taken as indicated and mixed with 0.5 μl 100 mM EDTA before resolving them by polyethyleneimine (PEI) thin-layer chromatography with 0.8 mM LiCl as solvent. Authentic pAp and AMP were used as migration standards. Accumulation of reaction products was estimated after visualization by UV. Expression of his-tagged proteins was monitored by PAA gel electrophoresis followed by staining with Bio-Safe Coomassie stain (BIO-RAD) or by western blot using Anti-His Peroxidase antibodies (Roche) at 1:200 in 1× PBS, 1% skim milk, 0.1% Tween and ECL Plus Western Blotting Detection System (GE Healthcare). For these experiments, different expression levels of the P-controlled genes were achieved by growing cultures in liquid LB in the presence or absence of arabinose (0.2 or 0.02%). A total of 393 completely sequenced bacterial genomes published before 4 January 2007 () were analyzed for the presence of YtqI, Orn and CysQ orthologous proteins. Orthologs were defined by searching for bi-directional best hits (BBH) () based on the following parameters: ⩾40% amino acid similarity and ⩽20% difference in protein length. The phylogenetic tree presented in Figure S1 was constructed from 141 representative species based on 16S rRNA similarity. The conserved interaction of pAp and oligoribonucleases between and human cells encouraged us to ask if we could identify a functional analog of oligoribonuclease in among the pAp-binding proteins from this organism. Extracts of 168 and a protease minus mutant (DB430) () were used in pAp-binding experiments. The protein pattern obtained looked similar for both strains; we therefore present only the data acquired for the wild-type strain. Two major protein bands were visible in the pAp-binding fraction (). Analysis of band A by liquid chromatography tandem mass spectrometry (LC-MS/MS) revealed HisIE (SwissProt, O34912) with an overall score of 2080 and 5 identified peptides covering 22.5% of the total mass of the protein. HisIE was identified previously as pAp-binding protein in (). The second major band gave high scores for two proteins: GuaC (SwissProt O05269), GMP reductase with an overall score of 292 and 4 peptides covering 13% of the total mass, and YtqI (SwissProt O34600) with an overall score of 140 and 4 peptides covering 11.5% of the total mass. YtqI is an unknown protein that belongs to the DHH/DHHA1 family (). This family of proteins consists of enzymes with phosphoesterase activity, including RecJ. YtqI was therefore our best candidate for a potential functional Orn analog. In order to perform complementation experiments, we created a conditional promoter mutant of the essential gene in (). This mutant (strain UM341) uses the anhydrotetracycline (Atc)-inducible promoter P () together with a Tet-repressor (TetR) to ensure tight control in the absence of Atc. A growth defect of this mutant was easily observable in cultures lacking Atc grown in LB liquid medium. While growth of the mutant carrying a plasmid-borne copy of (pUM408) was not affected by the absence of Atc (42 versus 41 min doubling time for cultures minus and plus Atc, respectively), mutants carrying the vector control (pBAD18) had a 1.9-fold longer doubling time when Atc was missing (77 versus 43 min, respectively) and their growth leveled off at an OD of ∼0.5. On plates, a similar effect could be observed in the absence of Atc; transformants of strain UM341 with pBAD18 produced pinpoint-sized colonies that stopped growing, while transformants with the carrying plasmid were significantly larger after overnight exposure and continued to grow (). C-terminally his-tagged YtqI was expressed under control of the arabinose-inducible P promoter (plasmid pUM412) for complementation experiments. As seen in , expression completely rescued the growth defect of the mutant on plates lacking Atc. Expression was induced by the addition of 0.2% arabinose. Expression levels of YtqI and Orn were similar under these conditions as judged from Coomassie-stained gels (data not shown). Complementation could be seen even in the absence of arabinose. As opposed to the expression level of YtqI in the presence of 0.2% arabinose, in the absence of arabinose expression was not visible on a Coomassie-stained protein gel and was below the amount that could be detected by western blotting using Anti-His antibodies (data not shown). We concluded therefore that even low levels of YtqI expression were sufficient for Orn complementation. Purified recombinant YtqI was tested for nanoRNase activity. In the presence of manganese, YtqI was able to degrade nanoRNA 5-mers (). The activity in the presence of other ions tested (magnesium, zinc and calcium) was negligible (data not shown). Comparing YtqI- and Orn-catalyzed degradation of nanoRNA 5-mers, we noticed significant differences: The amount of YtqI required for appreciable activity was two orders of magnitude higher than that necessary for Orn-catalyzed activity. In addition, the pattern of degradation products as well as the kinetics of this reaction looked very different. Here, 3-mers were virtually missing and other intermediates (2-mers and 4-mers) accumulated less than in Orn-catalyzed hydrolysis (). Therefore, we hypothesized that 3-mers might be a preferred substrate for YtqI and as such they might be hydrolyzed so fast that accumulation could not be observed. We tested this hypothesis by comparing degradation of 3-mers and 5-mers ( and ). We used three times more enzyme in the reaction with 5-mers as substrate in order to obtain appreciable conversion into monomers (A) as compared to the reaction on 3-mers (A). Turnover numbers for 3-mers were one order of magnitude higher than for 5-mers (1.5 versus 0.14 pmol/μg/min). In B, we compare the kinetics of the disappearance of different substrates (3-mers or 5-mers) and the appearance of the final reaction product monomers in reactions with equal amounts of YtqI (1.5 μg). These results clearly document that 3-mers were a much better substrate for YtqI than 5-mers. Moreover, it seems that degradation of 3-mers to 2-mer was the fastest step in catalysis as the 2-mers formed here disappeared considerably slower. In order to ask whether YtqI degrades specifically nanoRNA or is active on longer substrates as well, we tested degradation of a RNA 24-mer 5′-end labeled with P. shows that activity of YtqI on this substrate was insignificant. The YtqI-catalyzed turnover of 24-mers into monomers could be roughly estimated from this experiment as 0.01 pmol/μg/min. Binding of YtqI to pAp could point to the following possibilities: (i) activity of YtqI is affected by pAp or (ii) pAp can be a substrate for YtqI. We had reported before that Orn-catalyzed degradation of nanoRNA is highly sensitive to pAp (). Therefore, we decided to test the possibility (i) first. Unlike what we observed with Orn, the addition of small amounts of pAp to the YtqI-catalyzed reaction (10, 20 and 50 μM) did not produce an easily observable effect on degradation of nanoRNA. At 100, 200 or 500 μM pAp the activity of YtqI based on the conversion of 5-mer into monomers in 30 min dropped to 28, 4 and 1%, respectively (data not shown). The effect produced by 500 μM pAp was comparable to the effect seen in the presence of 20 μM pAp in an Orn-catalyzed reaction with 0.07 μg Orn and 3 μM substrate (). The observed effect of pAp on YtqI-catalyzed degradation of nanoRNA did not exclude the possibility of pAp being a substrate for YtqI. We therefore tested the ability of YtqI to degrade pAp . Remarkably, YtqI was able to degrade pAp to AMP (data not shown). YtqI converted 6 nmol of pAp/μg/minute. The pAp-degrading activity of YtqI was similar in magnitude to that of CysQ (33 nmol/μg/min), the pAp-phosphatase from (). Unlike CysQ activity, pAp-degrading activity of YtqI was not affected by either LiCl or CaCl at concentrations of 5 mM (data not shown). To test if this activity of YtqI has physiological relevance, we asked whether the expression of could complement the mutant phenotype, i.e. the growth impairment of CysQ-lacking cells in the absence of cysteine. shows that complementation could indeed be achieved. Transformants of UM285 (Δ) with the vector control formed very small colonies when plated on medium lacking cysteine (A). In liquid medium, growth of the vector control strain was severely affected in the absence of cysteine (B). Transformants of UM285 with a plasmid expressing YtqI (pUM412) or CysQ (pUM404) however formed normal size colonies (A). In liquid medium, UM285 strains transformed with plasmids expressing YtqI or CysQ were not affected in their growth when omitting cysteine (B). A comparison of expression levels of YtqI and CysQ in the presence of 0.02% arabinose showed that CysQ was expressed at a somewhat higher level than YtqI (data not shown). The ability of to complement a mutant in , prompted us to investigate the phenotype of an mutant (BFS66) in . Growth rates of the wild type and BFS66 were compared either in the absence or in the presence of cysteine. Doubling times were similar in the presence of cysteine with 42 and 44 min for wild type and the mutant, respectively, but varied considerably in the absence of cysteine with 43 versus 68 min. This phenotype resembled that of a mutant in . The latter seemed however more pronounced as withdrawal of cysteine affected growth more severely (88 versus 203 min) (B). Analysis of the phylogenetic distribution of Orn and YtqI (Supplementary Figure S1), clearly demonstrated that the majority of bacterial species possess only one of the two proteins. YtqI was present in Firmicutes, Bacteroidetes, Chlorobi and in the delta subdivision of Proteobacteria. Orn, however, was present in beta and gamma-Proteobacteria and in Actinobacteria. This distribution points to some anti-correlation: the presence of one of the genes seemed to exclude the presence of the second one (). The two proteins of different origin might therefore exert the same function. Some Actinobacteria were exceptional in that they had both and . Cyanobacteria and alpha-proteobacteria had neither Orn nor YtqI. shows the distribution of YtqI, Orn and CysQ in 393 completely sequenced genomes. This figure shows that most organisms that had YtqI, did not have CysQ or Orn. Whereas the overlap between organisms carrying both Orn and CysQ was considerable (58% of the species having Orn have also CysQ), only 21% of the species having YtqI had also CysQ, and only 11% of species having YtqI carried also Orn. This distribution supports our hypothesis that YtqI might fulfill the function of two proteins, Orn and CysQ. Interestingly, while Orn was absent in all sequenced archeal genomes (), YtqI was represented in 42% of them. This study was conducted in order to search in the model organism for a functional analog of oligoribonuclease, Orn. Encouraged by the observation that the pAp-oligoribonuclease interaction is conserved between and humans, we identified YtqI as potential functional Orn analog through its binding to pAp. The other proteins interacting with pAp, HisIE and GuaC, are of known function and were not the focus of this study. It is however noteworthy that the interaction between HisIE and pAp was observed previously using extracts (), which points to biological relevance of this interaction. YtqI belongs to the DHH family of phosphoesterases, more specifically to the DHHA1 subfamily (), some members of which are involved in nucleic acid metabolism. It was therefore a good candidate for a functional Orn analog. YtqI can complement a conditional mutant in when expressed at similar levels as Orn. This complementation does not require high amounts of YtqI, as expression levels that are below the detection limit of Anti-His antibodies are sufficient. Recombinant YtqI is able to degrade nanoRNA 5-mers in the presence of manganese. Whereas Orn is essential in , YtqI is not essential in . This points to the existence of at least one more enzyme with the ability to degrade nanoRNA. The pattern of degradation products on the PAA gel as well as the kinetics of their appearance make it clear that Orn and YtqI employ different mechanisms for the degradation of nanoRNA. 5-mers are not a good substrate for YtqI, they might be degraded in a distributive rather than a processive way. Another obvious difference was the absence of 3-mers from the degradation pattern. One possible explanation for this could be a preferred degradation of 3-mers into 2-mers. When used as substrate, 3-mers are degraded much faster than 5-mers, requiring approximately 10 times less enzyme than 5-mers for complete degradation. We therefore concluded that 3-mers are much better substrates than 5-mers. This result could reflect the intriguing possibility that YtqI acts preferentially on 3-mers and cannot efficiently degrade 5-mers. In this case, the fact that YtqI can complement Orn in could suggest that the accumulation of 3-mers and not 5-mers is the main cause of growth deficiency in lacking Orn. According to the literature (,), 90% of RNA degradation in is done hydrolytically, implying a more significant contribution of RNase II and RNase R as compared to PNPase. The relative contribution of RNase II and RNase R is under dispute; RNase II was considered to be the main contributor to mRNA degradation (), but this result was questioned by a genome-wide analysis of mRNA levels in a strain deleted for RNase II (). The end products of degradation catalyzed by RNase II and RNase R differ slightly in size; for RNase II experimental data indicate 3–5-mers () or 4–6-mers (,) as final product and 4-mers according to the structural model (), and RNase R leaves 2–3-mers (,) or 1–2-mers (). The size range of fragments produced by RNase R seems to be more suitable for degradation by YtqI than that of oligonucleotides produced by RNase II. The importance of RNase R is increasingly recognized. RNase R has the ability to degrade stable RNA (,) and contributes to quality control of rRNA (). More recently this enzyme was shown to be involved in the degradation of mRNA substrates with extensive secondary structure (). In addition, RNase R was shown to increase dramatically under different stress conditions (). harbors only one member of the RNR family of exoribonucleases, RNase R, which seems to be equally important for the degradation of highly structured RNA as its counterpart in (). Another requirement for oligonucleotide degradation might also come from systems expressed under the control of cyclic dGMP. Indeed this regulatory molecule is degraded by a phosphodiesterase, which should result in formation of pGpG, a dinucleotide that needs to be further degraded (). We previously demonstrated that Orn and Sfn bind pAp, but cannot degrade it, instead pAp is a strong inhibitor of these enzymes (). YtqI however can degrade pAp ; it also complements a mutant in . Both results clearly indicate that YtqI is a pAp-phosphatase. The phenotype of an mutant in resembles that of an deletion: growth is impaired in the absence of cysteine. Withdrawal of cysteine causes doubling times to increase 1.6-or 2.3-fold in an mutant in or an mutant, respectively. The effect of withdrawal of cysteine seems therefore slightly more moderate in lacking YtqI than in lacking CysQ. In fact, the difference between and in this respect could be somewhat larger, considering the fact that strain MG1655 used in our experiments has a rather leaky mutant phenotype as compared to other strains of (). One possible explanation for this interspecies difference could be bispecificity of protein CysH1. This enzyme has the ability to reduce both PAPS and APS () . In addition, expression of can complement an mutant defective for APS kinase encoded by (). This raises the possibility that APS could be reduced directly in , which would bypass the requirement for PAPS synthesis and thus pAp accumulation could be unnecessary. The direct reduction of APS is commonly used in plants () and was documented for some bacteria including (), () and (). The existence of a second enzyme able to hydrolyze pAp could be an alternative explanation for the only partial growth defect of the mutant in the absence of cysteine. The phylogenetic distribution of YtqI, Orn and CysQ and in particular the anti-correlation is in agreement with the hypothesis that YtqI fulfills the functions of two proteins in , Orn and CysQ. Until now, RNase R was the only exoribonuclease known in the small genomes of species (). Noteworthy is therefore the presence of YtqI homologs among the small set of proteins of unknown function in the genomes of and , where proteins MG371 and MPN140 respectively, are likely to perform the essential function of nanoRNA degradation. Different species seem to have found different solutions to the same cellular problem, the problem being the degradation of nanoRNA or pAp. Yet another solution to the problem of nanoRNA degradation awaits to be discovered, as cyanobacteria and the alpha division of proteobacteria have neither YtqI nor Orn orthologs. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Discovered by Nash and coworkers in 1996, tyrosyl DNA phosphodiesterase I (Tdp1) belongs to the phospholipase D superfamily of phospholipids hydrolyzing enzymes (,). Functionally, Tdp1 is part of the DNA repair complex that resolves the irreversible topoisomerase I (Top1)-DNA cleavage complexes by catalyzing the hydrolysis of 3′-phosphotyrosyl bonds (). In addition to the removal of peptides bound via a 3′ phosphotyrosyl linkage, Tdp1 can catalyze the cleavage of other chemical bonds such as a phosphohistidine bond (). Tdp1 can also remove a 3′-phosphoglycolate or biotin-linked substrate and act at 3′-abasic sites (,). Tdp1 thus participates in the repair of a variety of 3′ adducts/base damages from DNA. An interaction of Tdp1 with DNA ligase III () and XRCC1 (), members of the base excision repair (BER) complex has also been demonstrated. More recently, Tdp1 has also been implicated in the repair of topoisomerase II (Top2)-mediated DNA damage as bacterially expressed yeast Tdp1p processed the 5′ phosphotyrosyl linkage of a peptide derived from yeast Top2 covalently to DNA (). Therefore, Tdp1 may function in multiple DNA repair pathways. Tdp1 is physiologically important since a point mutation in the TDP1 gene causes the neurological disorder called spinocerebellar ataxia with axonal neuropathy (SCAN1) (). SCAN1 cells exhibit hypersensitivity to camptothecin (CPT), a potent Top1 inhibitor (,,). Moreover, overexpression of a human or yeast Tdp1 fusion protein has been shown to alleviate some of the effects of CPT treatment (,). These observations suggest that inhibitors of Tdp1 could act synergistically with CPT in a combined anticancer therapeutic regimen. Additionally, hypersensitiveness to CPT in Tdp1-defective yeast was conditional to deficiencies in the checkpoint (Rad9) and 3′-endonucleases (Mus81/Eme1) pathways (). Thus, in principle, therapeutic selectivity can be achieved by combining Top1 and Tdp1 inhibitors as a significant number of tumors have defective DNA repair and checkpoint pathways (). As Tdp1 inhibitors in association with Top1 inhibitors could confer a selective advantage for cancer chemotherapy, we began searching recently for Tdp1 inhibitors (). Currently, the only reported inhibitors of Tdp1 are vanadate, tungstate, aminoglycoside antibiotics and ribosome inhibitors (,). Our initial aim was to develop a high-throughput assay that would provide a sensitive, reliable and a rapid method to screen chemical libraries for novel Tdp1 inhibitors. Here, we report the development of a sensitive high-throughput electrochemiluminescent (ECL) assay to identify novel inhibitors of Tdp1. Identified by the ECL assay, the dication furamidine (DB75, NSC 305831) was further studied to determine its molecular interactions with recombinant Tdp1 and its DNA substrates. The 1981 compounds of the diversity set were obtained from the Developmental Therapeutics Program (DTP) of the National Cancer Institute (NCI), NIH. Berenil and Pentamidine were purchased from Sigma-Aldrich (St. Louis, MO, USA). High-performance liquid chromatography-purified oligonucleotides and tyrosyl nucleotides were purchased from the Midland Certified Reagent Co. (Midland, TX, USA). Human Tdp1 expressing plasmid pHN1910 (a gift from Dr Howard Nash, Laboratory of Molecular Biology, National Institute of Mental Health, National Institutes of Health) was constructed using vector pET-15b (Novagen, Madison, WI, USA) with full-length human Tdp1 and an additional His-tag sequence of MGSSHHHHHHSSGLVPRGSHMLEDP in its N terminus. The His-tagged human Tdp1 was purified from Novagen BL21 cells using chelating sepharose™ fast flow column (Amersham Biosciences, Piscataway, NJ, USA) according to the company's protocol. The collected fractions were assayed immediately for Tdp1 activity. Fractions that showed Tdp1 activity were pooled and dialyzed in 20% glycerol, 50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10 mM β-mercaptoethanol and 2 mM EDTA. Dialyzed samples were aliquoted and stored at −80°C. Tdp1 concentration was determined using the Bradford protein assay (Bio-Rad Laboratories, Hercules, CA, USA). Tdp1 purity was determined as a single ∼70 kDa band representing over 95% of the detectable proteins stained by Coomassie after SDS–polyacrylamide gel electrophoresis (SDS-PAGE). Our electrochemiluminescent (ECL) assay is based on the BioVeris (BV) ECL technology developed by BioVeris, Inc. (Gaithersburg, MD, USA). ECL is based on the use of ruthenium labels (BV-TAG™), designed to emit light when stimulated. These labels, together with a specific instrumentation (M-SERIES® Analyzer), provided our platform for biological measurements. This technology has been successfully applied to drug discovery, pharmaceutical industry, clinical and industrial detection (,). The present report is the first application of the ECL biotechnology to Tdp1. The 5′-biotinylated 14Y DNA substrate (sequence shown in A) was obtained from Midland Certified Reagent and coupled to an NHS ester BV-Tag (BioVeris Inc.) to generate the ECL substrate BV-14Y. Coupling was achieved by incubating 175 µl of 5′-biotinylated 14Y DNA at 200 µM in phosphate buffered saline (PBS), pH 7.4 with 25 µl of NHS-ester BV-Tag (BioVeris Inc.) at 3 µg/µl in 100% dimethyl sulfoxide (DMSO). After 30 min at room temperature under agitation, the coupling reaction was loaded onto a mini Quick Spin Oligo column (Roche Diagnostics, Indianapolis, IN, USA) pre-equilibrated with 3 volumes of PBS, pH 7.4 containing 0.075% (w/v) sodium azide (Sigma-Aldrich, St. Louis, MO, USA). The recovered fraction was aliquoted and stored at −20°C at 10 µM in PBS. A substrate stock solution for 500 reactions was prepared by mixing 40 µl of 10 µM ECL BV-14Y substrate with 500 µl of streptavidin magnetic beads (Dynabeads M-280, BioVeris Inc.) and incubating for 30 min at 25°C under constant agitation. After magnetic separation and a two-volume wash with the Tdp1 assay buffer (50 mM Tris-HCl, pH 8.0, 80 mM KCl, 2 mM EDTA and 1 mM DTT), the linked ECL substrate was resuspended in 500 µl of assay buffer at a concentration of 800 nM. The linked ECL substrate can be stored at 4°C for up to a month without loss of activity. The linked ECL BV-14Y substrate at a concentration of 0.8 nM was incubated with 1 nM Tdp1 in the absence or presence of the drugs (10 µM) to be tested. Reactions were carried out in 96-well plates at a final volume of 100 µl/well in assay buffer for 60 min at 37°C. The reactions were stopped by adding 1 volume of stop buffer (25 mM MES pH 6.0). Plates were read with an M-SERIES® M8 analyzer (BioVeris Inc.) and the ECL arbitrary units were plotted using the Prism software (GraphPad Software, San Diego, CA, USA). High-performance liquid chromatography-purified oligonucleotides 14Y (see A) () and 14Y-CC (see A) were labeled at their 5′-end with [γ-P]ATP (PerkinElmer Life and Analytical Sciences, Boston, MA, USA) by incubation with 3′-phosphatase-free T4 polynucleotide kinase (Roche Diagnostics, Indianapolis, IN, USA) according to the manufacturer's protocols. Unincorporated nucleotides were removed by Sephadex G-25 spin-column chromatography (mini Quick Spin Oligo columns; Roche Diagnostics). For the production of the oligonucleotide duplexes D14Y, the P-radiolabeled 14Y oligonucleotide was mixed with the complementary oligonucleotide (see A) at equal molar ratios in annealing buffer (10 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 10 mM MgCl), heated to 96°C, and allowed to cool down slowly (over 2 h) to room temperature. Unless indicated otherwise, Tdp1 assays were performed in 20 µl mixtures in Tdp1 assay buffer (see above) plus 40 µg/ml bovine serum albumin. Twenty five nanomolar 5′-P-labeled substrate (14Y or 14Y-CC or D14Y) was reacted with 1 ng of Tdp1 (0.7 nM) in the absence or presence of inhibitor for 20 min at 25°C. Reactions were stopped by the addition of 60 µl of gel loading buffer [96% (v/v) formamide, 10 mM EDTA, 1% (w/v) xylene cyanol and 1% (w/v) bromphenol blue]. Twelve-microliter aliquots were resolved in 20% denaturing polyacrylamide (AccuGel; National Diagnostics, Atlanta, GA, USA) (19:1) gel containing 7 M urea. After drying, gels were exposed overnight to PhosphorImager screens (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, USA). Screens were scanned, and images were obtained with the Molecular Dynamics software. Densitometry analyses were performed using the ImageQuant 5.2 software (GE Healthcare Bio-Sciences Corp., Piscataway, NJ, USA). Tdp1 activity was determined by measuring the fraction of substrate converted into 3′-phosphate DNA product by densitometry analysis of the gel image (). Figures show representative results that were consistently reproduced at least three times. Binding experiments were performed on a Biacore 2000 instrument (Biacore Inc., Piscataway NJ, USA). 5′ biotinylated stem-loop (biotin-GATCTAAAAGACTTTCTCAAGTCTTTTAGATC) and single-stranded oligonucleotides (biotin-GATCTAAAAGACTT) were synthesized by IDT (Coralville, IA, USA). Stem-loop oligonucleotides were annealed by heating to 90°C for 5 min followed by snap cooling on ice for 15 min. Biotinylated oligonucleotides were immobilized to neutravidin-coated sensor chips as described previously (). Approximately 5000 RU's of neutravidin was attached to all flow cells on the sensor chips. Oligonucleotides were reconstituted in buffer consisting of 10 mM Tris, pH 7.5, 300 mM NaCl and 1 mM EDTA. Single-stranded and stem-loop oligonucleotides were injected over flow cell 2 and 4, respectively until approximately 500 RU's of oligonucleotide were captured on the chip surface. Furamidine was diluted into running buffer [10 mM MES, 100 mM NaCl, 1 mM EDTA, 5% DMSO (v/v) pH 6.25] and injected over all flow cells at 20 µl/min at 25°C. Following compound injections, the surface was regenerated with a 10 s 1 M NaCl injection followed by a 10 s running buffer injection. A DMSO calibration curve was included to correct for refractive index mismatches between the running buffer and compound dilution series. Data was analyzed using the Scrubber software version 2 (David Myszka, University of Utah) and the equilibrium binding of furamidine was fit to either a single-site or two-site steady state binding model (C and D). The equation used to fit the single-site steady state binding model was: The equation used to fit the two-site steady state binding model was: To discover inhibitors of Tdp1, we developed a novel ECL high-throughput assay (for details see Materials and Methods section and ). An ECL substrate for Tdp1 was generated after coupling a ruthenium-containing tag (BV-Tag) to the 3′-end of a 14-mer oligonucleotide with a 5′-biotin and a 3′-tyrosyl moiety (A) (,). In the presence of Tdp1, the tyrosyl-bound BV-Tag is hydrolyzed from the ECL substrate (BV-14Y DNA) and washed away, leading to a loss of the electroluminescent signal (B and C). In this assay, a potential Tdp1 inhibitor would be detected as preventing this loss of signal with a level of signal retention reflective of the potency of the putative inhibitor. Thus, in our high-throughput ECL assay, active compounds were identified as restoring to the control levels (without Tdp1) the signal lost in the presence of Tdp1. Using the high-throughput ECL assay, we screened the ‘Diversity Set’ from the Developmental Therapeutics Program (DTP) of the National Cancer Institute against recombinant human Tdp1. The diversity set consists of 1981 compounds representative of the chemical/structural diversity of the available repository of more than 140 000 chemicals (for details, see ). Of the 1981 compounds from the DTP diversity set, 169 positive hits were identified as inhibiting at least 70% of the Tdp1 activity at a fixed concentration of 10 µM (A). Subsequent analyses as HPLC to evaluate chemical purity of the samples reduced the number of potential inhibitors of Tdp1 to 69 compounds. Further testing in gel-based assays led to the identification of 49 compounds as confirmed Tdp1 inhibitors. Thus, ∼8% of the NCI-DTP diversity set (1981 chemicals) were active against Tdp1 at 10 µM drug concentration in the ECL high-throughput assay. This assay is therefore a sensitive and efficient technique for the screening of novel Tdp1 inhibitors. The diamidine NSC 305831 (for structure, see B), more commonly known as furamidine (DB75) (), was among the most potent inhibitors of Tdp1 with complete restoration of the normal Tdp1 signal in the ECL assay (see A). Furamidine was chosen among other positive hits because it belongs to a very interesting group of compounds. Furamidine and closely related heterocyclic amines are currently under clinical investigation as anti-parasitic agents (). DB289, an orally available methamidoxime prodrug of furamidine is currently under phase III clinical trials as a trypanocidal agent against human African trypanosomiasis (,). These compounds have been well studied for their binding to double-stranded DNA. Previous studies have shown that furamidine binds duplex DNA in the DNA minor groove selectively at AT rich sites [(A/T)] (,,). Furamidine can also intercalate between GC base pairs of duplex DNA (,,). Furamidine could therefore interfere with DNA processing enzymes such as Tdp1. Having identified furamidine as a novel Tdp1 inhibitor, we characterized its molecular effect on Tdp1 activity using gel-based assays () (). Since both partially duplex and single-stranded DNA are substrates for Tdp1 (,,), we compared the inhibition of Tdp1 by furamidine using the D14Y and 14Y substrates (sequences as shown in A) (). As demonstrated in (panels B and C) furamidine inhibits the processing of both the single and double-stranded substrates by Tdp1. Since Tdp1 inhibition is slightly more effective in the duplex substrate (B and C), DNA binding/intercalation (,,) may contribute to furamidine's potency. However, inhibition of Tdp1 activity by furamidine is also evident in a single-stranded substrate. This is unlike aclarubicin, a known DNA intercalator that inhibits Tdp1 selectively only with double-stranded DNA (). Thus, additional mechanism of Tdp1 inhibition by furamidine exists besides DNA intercalation or minor groove binding. We next evaluated the ability of furamidine to directly interact with DNA in the absence of Tdp1. Surface plasmon resonance (SPR) analyses were carried out using single-stranded and double-stranded substrates (see Materials and Methods section). As expected, A shows detectable binding of furamidine to duplex oligonucleotide at submicromolar concentrations (,,). Furamidine rapidly reached a steady state binding level with duplex DNA but then disassociated more slowly. The equilibrium binding could only be fit using a two binding-site model with affinities of 0.33 and 19 µM (C and equations described in the Materials and Methods section). This seems reasonable given that the sequence contains an AT rich site [(A/T)] within the duplex oligonucleotide, which corresponds to a high affinity-binding site for a heterocyclic diamidine (,,). The 14 base-pairs duplex probably also support additional compound binding at lower affinity sites. The binding of furamidine to a single-stranded substrate was also examined by SPR. B shows that furamidine both associates with and dissociates from a single-stranded substrate very rapidly in contrast with what was observed for the duplex. This observation most likely reflects the electrostatic interaction between the phosphate backbone and the charged compound. We estimate the of furamidine to be ∼70 µM with single-stranded DNA (D and equations described in the Materials and Methods section). This result is in agreement with the IC obtained for the same single-stranded substrate (C). To elucidate whether the inhibition of Tdp1-mediated cleavage of the single-stranded DNA substrate could be due to a stabilization of a hairpin structure by furamidine, we plugged our sequence in the web application (). The most favorable folding option with three formed base pairs, gave a Δ value of 0.32 kcal/mol at 20°C and a value of 13.6°C, rendering this structure unstable under our reaction conditions (25°C). In addition, the fact that we were able to measure a dissociation rate with the duplex but not with the single-stranded oligonucleotide by SPR leads us to believe that furamidine does not stabilize any hairpin structure. We also evaluated the binding of furamidine to amine-coupled Tdp1 protein (data not shown) and found the interaction to be very weak with of >900 µM. Together, these experiments demonstrate that furamidine does bind DNA with a preference for a duplex substrate. Furamidine has previously been shown to exhibit a strong preference for A/T sequences (,). Since the co-crystal studies showed Tdp1 interaction with the last 4 nt at the 3′-end of the oligonucleotide substrate (,) and those 4 bases were ACTT in our 14Y oligonucleotide, we evaluated the effect of altering the sequence of the oligonucleotide substrate on the inhibition of Tdp1 by furamidine. The terminal thymine dinucleotide (–TT) of the 14Y oligonucleotide was replaced with a cytosine dinucleotide (–CC) to generate the 14Y-CC oligonucleotide (see A). The ability of Tdp1 to process either the 14Y or 14Y-CC substrates was similar (B and C). Kinetic plot analysis shows the processing of either substrate almost completely within 10 min of the reaction time and at the same rate by 1 ng of Tdp1 (C). Upon addition of varying concentrations of furamidine, the processing of both the substrates 14Y and 14Y-CC by Tdp1 was inhibited to the same degree (D and E). Thus, presence of a TT dinucleotide at the 3′ terminus of the DNA is not critical for Tdp1 inhibition by furamidine. Structurally, furamidine can be considered as a bisbenzamidine derivative (A) () and belongs to a family of diamidines (). Of the several diamidines, berenil and pentamidine are clinically active and used against parasitic diseases (,). Though furamidine, pentamidine and berenil share similarities in structure (A), they differ in their central moiety being a furan unit for furamidine, a triazene in berenil and a pentyldioxy chain in pentamidine (represented by the ‘variable portion’ in A) (). To evaluate the contribution of the central furan portion of furamidine for Tdp1 inhibition, we tested the three analogs for their ability to inhibit Tdp1 activity. B shows that pentamidine did not inhibit Tdp1 activity under our assay conditions, and berenil showed some activity, albeit at a high concentration (300 µM). Under the same conditions, furamidine exhibited an inhibition of Tdp1 activity at 30 µM (B) and therefore is the most potent of the three bisbenzamidines examined. The hallmark of reversible inhibitors is that when the inhibitor concentration drops, enzyme activity is restored. Our initial gel assays () were performed at a fixed time (20 min) under conditions where Tdp1 almost fully converted the substrate in the absence of inhibitor (1 ng, pH 8.0). We next evaluated how reaction time and Tdp1 concentration affected furamidine's ability to inhibit Tdp1. As shown in A and B (A, left; and squares in B), 1 ng of Tdp1 converted ∼50% () of the 14Y substrate within ∼1.9 min. Thus, we wished to determine how furamidine affected the kinetics of Tdp1 activity. Tdp1 activity was slowed down as the concentration of furamidine increased (A). Kinetic plots (B) demonstrated that furamidine increased the conversion half-time () of the 14Y substrate from 1.9 min in the absence of drug to 2.7 min in the presence of 30 µM furamidine (diamond in B) and 4.4 min in the presence of 60 µM furamidine (inverted triangle in B). Additionally, increasing Tdp1 concentration was able to overcome Tdp1 inhibition by furamidine (C and D). The 50% inhibition of Tdp1 activity observed by 30 µM furamidine with 0.1 ng of Tdp1 was almost completely reversed by increasing the concentration of Tdp1 to 1 ng (C and diamond in D). Similar effects were seen with 60 µM and 250 µM furamidine (C and D). Thus, free Tdp1 competes with furamidine. Together, these results suggest that furamidine produces reversible and competitive inhibition of Tdp1. Until now, Tdp1 activity has been measured by coupling a 3′-phosphotyrosine to DNA or by isolating small peptide fragments covalently linked to DNA, and resolving the reaction product (3′-phosphate DNA) from substrate in polyacrylamide gel (,,,). In 2002, Cheng . () used chromagenic substrates like the 3′-(4-nitro)phenyl phosphate DNA (). However, spectrophotometric detection of 4-nitrophenol required high concentrations of Tdp1 because of its relatively poor extinction coefficient. More recently, fluorescence-based assays have been designed using oligonucleotide and nucleotide substrates containing 3′-(4-methylumbelliferone)-phosphate (DNA-MUP) (). Tdp1-mediated cleavage releases the 4-methylumbelliferone from the DNA, and the resulting fluorescence can be quantitated (). Our ‘mix and read’ high-throughput electrochemiluminescent (ECL) assay offers two main advantages. First, the ECL tag (BV-TAG) on the DNA substrate undergoes many excitation/emission cycles, which amplify the output signal, and increases sensitivity (C). Since the process is selective for the BV-Tag, the background signal typical of fluorescent methods is eliminated. Second, positive hits are inferred by a gain/restoration of signal as opposed to previous techniques that rely on loss of signal, thereby providing high sensitivity and reliability. More than 70% of the compounds identified by our ECL screen at a single concentration were confirmed as inhibitors of Tdp1 by gel assays. Additionally, the ECL assay is quick, as the high-throughput configuration can process a 96-well plate in ∼10 min. Subsequent to screening the DTP diversity set by the ECL assay (), the antiparasitic furamidine () was identified as a positive hit for Tdp1 inhibition and further characterized by gel-based assays. Our results demonstrate that furamidine inhibits the activity of recombinant human Tdp1 at low micromolar concentrations both with single- and double-stranded DNA substrates (see A and B). Moreover, we provide evidence that furamidine binds single-stranded DNA (see B and D). Our results also show that furamidine exhibits no preference for a TT sequence over a CC at the terminus of a single-stranded oligonucleotide for Tdp1 inhibition (see ). Interestingly, another diamidine, pentamidine was inactive against Tdp1 (see ) despite the fact that furamidine and pentamidine have an overall similar structural curvature, that closely matches the curvature of the DNA minor groove (). We also found that berenil was approximately one order of magnitude less effective than furamidine at inhibiting Tdp1. This suggests that the furan linker in furamidine contributes to furamidine's inhibition of Tdp1 activity. We believe that DNA binding is part of the furamidine's mechanism of action. Further studies such as crystallization will be required to fully elucidate its complete mechanism of action. It is plausible that furamidine could form a ternary complex by binding at the interface created by the DNA substrate and the enzyme (,). To date, vanadate, tungstate, aminoglycoside antibiotics and ribosome inhibitors are the only known inhibitors of Tdp1 (,). Though vanadate and tungstate bound to Tdp1 provided insights into substrate binding and catalytic mechanism of Tdp1 (,), they are general inhibitors of a variety of enzymes involved in phosphoryl transfer. Moreover, vanadate, tungstate, aminoglycoside antibiotics and ribosome inhibitors inhibit Tdp1 activity in biochemical assays only at high (millimolar) concentrations (). The present study demonstrates that furamidine is by far the most potent inhibitor of Tdp1 reported to date. However, determining the effect of furamidine on Tdp1 in cellular systems would be very hard to interpret as furamidine may have additional target(s) because of its DNA-binding activities. Nevertheless, furamidine presents a prototype, to explore the possibility of generating new chemotypes that would specifically target Tdp1. Evaluation and characterization of other novel Tdp1 inhibitors identified by our ECL screen are ongoing. Our present study with furamidine demonstrates that the ECL assay is a useful tool for the high-throughput screening of Tdp1 inhibitors.

Single-stranded oligodeoxynucleotides (ODNs) and in particular certain G-rich ODNs have been widely reported to have effects on cells ranging from the induction of senescence and aging () to inhibition of proliferation (). In some cases, secondary structure formation of ODNs (e.g. G-quadruplexes) facilitates the recognition by cellular protein(s) thus leading to cytotoxicity (). In other cases, anti-proliferative activity of ODNs is related to the ability of the ODN to bind specific cellular proteins independent of secondary structure (,). Known targets of G-rich oligodeoxynucleotides include the human ribosomal protein L7a (), nucleolin (,), elongation factor 1A (,), STAT3 () and growth factors (). DNAzymes are single-stranded DNA molecules that are able to cleave RNA in a site-specific manner (). The molecules consist of a 5′ and 3′ binding arm and a catalytically active core region. The stability and activity of the DNAzymes make them a useful tool for gene silencing, and a number of different therapeutic applications have been proposed (). We recently showed that the cellular effects of some ODNs, originally designed as DNAzymes against the transcription factors and , were not due to the cleavage of the target mRNA (). Indeed, DNAzymes such as Dz13 () and Rs6 () and their catalytically inactive controls were cytotoxic when transfected into several cultured cell lines at concentrations as low as 10–50 nM (). These cytotoxic ODNs all featured G-rich regions at the 5′ extremity suggesting that this sequence element is required for cytotoxicity. Although this cytotoxicity did not appear to be dependent on the formation of G-quadruplexes and other secondary structures (), replacement of any of the three contiguous guanosines of Dz13 with 7-deaza-guanosine abolished cytotoxicity. We therefore postulated that the 5′ extremity sequence element is necessary for cytotoxicity and may be involved in mediating specific interactions with cellular proteins. The aims of the present article were to clarify the sequence elements or ‘motifs’ required for cytotoxicity, to investigate the activity of various cytotoxic ODNs on different cell types, to determine the culture conditions required for cytotoxicity, and finally, to identify proteins that selectively bind to these cytotoxic molecules. HPLC-purified oligodeoxynucleotides were purchased from Sigma and Trilink Biotechnologies. Stock solutions were made up to a concentration of 50 μM in nuclease-free water and stored at −20°C. The oligodeoxynucleotides used are listed in . The 3′ biotinylated and 5′ Oregon-Green labeled congeners of Dz13 had the identical deoxyribonucleotide sequence to DT1549 () in which the 3′-3′T inversion of Dz13 is missing. Human dermal microvascular endothelial (HMEC-1) cells were maintained in MCDB131 medium (Gibco) containing 10% foetal bovine serum (FBS), 2 mM -glutamine, 1 μg/ml hydrocortisone (Sigma) and 10 ng/ml epidermal growth factor (EGF; Sigma) with or without 5 U/ml penicillin–streptomycin. Rat vascular smooth muscle (RSMC), human embryonic kidney (HEK-293), HCT116 human colon cancer, NIH3T3 mouse fibroblast and human retinal pigmented epithelium (ARPE-19) cells were cultured in DMEM F12 medium (Gibco) supplemented with 10% FBS and 2 mM -glutamine with 5 U/ml penicillin–streptomycin. Cells were seeded in 96-well black MicroClear plates (Greiner) (4 × 10 cells/well) or 6-well plates (1.2 × 10 cells/well) in growth medium containing 5% FBS for HMEC-1, HCT116, HEK-293, NIH3T3 and ARPE-19 cells or 10% FBS for RSMC cells. Cells were transfected 24 h after seeding with different concentrations of ODNs in triplicate using FuGene6 (Roche) as previously described (). Cell survival was assessed 48 h post-transfection in the 96-well plate format using the Cell Titer™-Blue cell viability assay (Promega) as previously described (). Briefly, culture medium was replaced with 100 μl OptiMEM and 20 μl of Cell-titer™ blue reagent and incubated for 2 h at 37°C. Fluorescence was measured at 544/590 using FLUOstar OPTIMA (BMG Labtechnologies). For the cell density experiments, ARPE-19, NIH3T3 or HMEC-1 cells were seeded at densities of 4000 and 50 000 cells per well in 96-well plates and transfected with 0–200 nM ODN in triplicate. Transfection efficiencies at the 2 cell densities were determined in duplicate in 60 mm dishes using a 5′-oregon green ODN (DT1565; ) in a single experiment. The 60 mm dishes were seeded with a similar number of cells/cm as for the 96-well plates for both seeding densities. For low density transfection, 2.6 × 10 cells/dish were seeded while for high density transfection 3.25 × 10 cells/dish were seeded. Fluorescence of mock and DT1565-transfected cells was measured 48 h post-transfection using fluorescence-activated cell sorting (FACS). Cells were washed twice with PBS and total proteins extracted using MPER (Pierce) or RIPA (150 mM NaCl, 0.1% w/v sodium dodecyl sulphate, 1% v/v Nonidet P-40, 0.5% w/v sodium deoxycholate, 50 mM Tris-HCl pH 8). Subcellular proteins were extracted using the ProteoExtract kit (Calbiochem). All extractions were performed in the presence of EDTA-free protease inhibitors (Roche). Cellular debris was removed by centrifugation at 10 000 for 20 min at 4°C. Protein pull-down assays were performed using oligodeoxynucleotides as ‘bait’ in order to identify Dz13-binding proteins. Dynabeads-Streptavidin (Dynal) were washed twice in 2× Buffer A (10 mM Tris-HCl, 1 mM EDTA, 2 M NaCl, pH 7.4). Beads (0.5 ml) were resuspended in 1 ml 2× Buffer A to a final concentration of 5 μg/μl beads. For assays, beads were incubated in an equal volume of 2 μM 3′ biotinylated congeners of Dz13 or Dz13scr (DT1309h and DT1310c respectively, ) made up in Buffer B (20 mM HEPES, 100 mM KCl, 0.2 mM EDTA, 0.01% v/v NP-40, 10% v/v glycerol, pH 7.5) and the mixture incubated at room temperature for 10 min with gentle mixing. The beads were washed with 3 × 1 ml Buffer B prior to addition of protein. Protein extract (up to 1 mg) was incubated with the bead–DNA mixture for 10 min at room temperature with shaking. For the majority of protein pull-downs, the beads were washed with 20 × 1 ml Buffer B following incubation with protein extract and proteins were eluted by two washes with 1 μM Dz13 or Dz13scr at room temperature. In one experiment, the elutions were performed with 10 μM Dz13 or Dz13scr. In between each elution, the beads were washed with 3 × 1 ml Buffer B. For protein pull-down assays, cells were transfected with 100 nM biotinylated congeners of Dz13 or Dz13scr (i.e. DT1309h and DT1310c) for 24 h. Beads were washed three times with 1 ml Buffer B and resuspended at 5 μg/μl prior to incubation for 10 min at room temperature with protein extracts prepared from these cells using MPER. Following incubation, beads were washed with 20 × 1 ml Buffer B and resuspended in 20 μl Buffer B. Washing and elution steps were then performed as described above. Fractions generated from pull-down assays were concentrated in a Centricon 10000 Mwt cutoff 0.5 ml centrifugal device (Millipore) at 13 800 for 70 min at 15°C. Proteins (from cell extracts or concentrated pull down fractions) were denatured at 70°C for 10 min in 4× Loading Buffer (Invitrogen), loaded onto NuPAGE 4–12% bis-tris acrylamide gels (Invitrogen) and electrophoresed at 140 V in MOPS-SDS running buffer. Following electrophoresis proteins were either silver-stained as described by Rabilloud . () or transferred to a PVDF membrane at 30 V for 2 h. Membranes were blocked in 3% BSA in TBST (10 mM Tris pH 8, 30 mM NaCl, 0.05% v/v Tween-20) for 1 h at room temperature. Primary antibodies were incubated at concentrations recommended by the manufacturers for 1 h at room temperature or overnight at 4°C in 5% skim milk-TBST. Membranes were washed 3 × 5 min in TBST and incubated for 1 h at room temperature with horseradish peroxidase conjugated antibodies (1:2000 in 5% skim milk-TBST; DakoCytomation). Membranes were washed 3 × 5 min in TBST and visualized by chemiluminescence using ECL (Amersham) and Hypersensitive film (Amersham). The following commercially available antibodies were used: EF1A (CBK-KK1; Upstate Biotechnology), vimentin (V9; Sigma) and STAT3 (F-2; Santa-Cruz Biotechnology). Total protein was digested by incubating 100 μl of sample, 25 μl of 10 mM NHHCO and 1 μg trypsin at 37°C for 16 h. The digested peptides were loaded onto a C18 precolumn (500 µm × 2 mm, Michrom Bioresources) using HO:CHCN (98:2, 0.1% formic acid, buffer A) at 20 µl/min. After a 10 min wash, the pre-column was switched (Switchos) in-line with an analytical column containing C18 RP silica (PEPMAP, 75 µm × 15 cm, LC-Packings) or a fritless C18 column (75 µm × ∼12 cm). Peptides were eluted using a linear gradient of buffer A to HO:CHCN (40:60, 0.1% formic acid-buffer B) at 200 nl/min over 60 min. The column was connected via a fused silica capillary to a low volume tee (Upchurch Scientific) where high voltage (2300 V) was applied and a nano electrospray needle (New Objective) or fritless column outlet was positioned ∼1 cm from the orifice of an API QStar Pulsar i hybrid tandem mass spectrometer (Applied Biosystems). The QStar was operated in an information-dependent acquisition mode. A TOF-MS survey scan was acquired (/ 350–1700, 0.5 s) and the two largest precursors (counts > 10) sequentially selected by Q1 for MS/MS analysis (/ 50–2000, 2.5 s). A processing script generated data suitable for submission to the database search programs. CID spectra were analysed using Mascot MS/MS ion search (Matrix Science). The criteria were: trypsin digestion allowing up to 1 missed cleavage, oxidation of methionine, peptide tolerance of 1.0 Da and MS/MS tolerance of 0.8 Da. A Mascot score >18 indicated identity. Variants of Dz13 and other selected DNAzymes, which reduce proliferation in HMEC-1 cells but do not act through cleavage of the RNA substrate, were designed in order to elucidate the sequence requirements for cytotoxicity. First, a set of oligodeoxynucleotides containing regions of the Dz13 sequence was designed in order to identify any requirement for an active motif. ODNs corresponding to the 5′ sequence of the first 9 bases of Dz13 (DT1530), this 5′sequence plus the 10–23 catalytic core (15 bp) in the reverse orientation (DT1531) and the 5′sequence plus the catalytic core in the correct orientation (DT1532) were tested. All of these were substantially less active than Dz13 (data not shown), suggesting that either length or some other sequence requirement had not been met. Similar results were obtained with the corresponding ODN based on the 5′ Rs6 sequence (DT1533-5, ; Figure S1a). In order to elucidate the requirements of the tail for cytotoxicity, Dz13 analogues of the same length but with modified 3′ tail sequences were tested (). These included both CG dinucleotides in the 3′ tail changed to GC (DT1536), the first 16 bases of Dz13 plus scrambled tail (DT1537), a pool of ODNs all consisting of the first 10 bases of Dz13 followed by a random mix of nucleotides in tail sequence (DT1538) and the first 10 bases of Dz13 followed by a polyA tail (DT1539). The polyA-tailed Dz13 analogue (DT1539) exhibited no cytotoxicity in HMEC-1 cells whereas DT1536 demonstrated intermediate cytotoxicity (A). Both the scrambled tail analogue (designed to remove the tail hairpin structure) and the mixture of random-tailed oligodeoxynucleotides had activity that was indistinguishable from Dz13, indicating minimal sequence requirements for the 3′ tail in these 33-mer oligodeoxynucleotides. To elucidate the positional requirement for a 5′ G-rich element, three pools of ODNs featuring the 5′ motif from Dz13 (first 10 bases) were constructed with either the 5′ motif in the middle of a random sequence pool, at the 3′ end or at the 3′ end in reverse order (DT1600-1602, ). None were significantly cytotoxic, although some activity was noted for the 3′ reverse motif ODN (B). This suggests that the motif may also be recognized in the 3′ to 5′ direction or that part of the activity relates to the rest of the molecule. In any case, this experiment confirmed that the optimal placement for the G-rich sequence of Dz13 is at the 5′ end. Given the lack of activity of the shorter variants (DT1530–DT1532), the next experiment sought to clarify the length requirement for the observed cytotoxicity. This was tested by using ODN pools containing the first 10 bases of Dz13 and 5, 10, 15, 20 or 25 random bases (DT1544-48, ). Cytotoxicity was concentration and length dependent with only the 35-mer ODN pool (DT1548) matching the potency of Dz13 (C), thus confirming the role of the tail length. A similar result was obtained with DT1572-6, which are based on a slightly different 5′ sequence (data not shown). A set of 35-mer oligodeoxynucleotides were designed and synthesized in order to identify the minimal 5′ motif required for activity. This set comprised pools of oligodeoxynucleotides for which the first 10 bases of the Dz13 5′ G-rich sequence (plus a 25 random base tail; DT1548) was progressively reduced (DT1553–DT1554) to the first 4 bases of this sequence plus a 31 base tail (DT1555). Only DT1553 (first 8 + 27 random) and DT1548 (first 10 + 25 random) retained full activity (D). The corresponding experiment with the Rs6 sequence (DT1557–60) produced gradated cytotoxicity profiles with activity increasing with increased retention of the 5’ Rs6 sequence (Figure S1b). As reported earlier (), other published G-rich oligodeoxynucleotides are capable of inducing cytotoxicity under the same conditions in HMEC-1 cells including the 20AG ODN described by Cogoi . (). We examined whether extension of this sequence in a manner analogous to the Dz13 experiments would influence cytotoxic activity. Consistent with the Dz13 results, 16- (DT1591), 24- (DT1592), 28- (DT1593) and 32-mer (DT1594) variants of the 20-mer 20AG yielded length-dependent cytotoxic activities with the longest two molecules having similar profiles (data not shown). Given their potent cytotoxic activity against proliferating cells, Dz13 and the other cytotoxic oligodeoxynucleotides could be of use in several disease states. Dz13 has been reported as being active in several preclinical models of disease including ocular angiogenesis (), vascular intimal proliferation () and cancer (), which are all diseases where inappropriate proliferation is present. To investigate any possible differential activity relative to non-proliferating cells, we examined the activity in two additional cell lines that display contact inhibition, namely ARPE-19 and NIH-3T3. In particular, ARPE-19 cells form differentiated, polarized monolayers () similar to that present in normal retinal epithelium. As shown in A, proliferating ARPE-19 cells were sensitive to Dz13-induced cytotoxicity with an IC between 50 and 100 nM, whereas Dz13scr was inactive. Dz13 cytotoxicity in ARPE-19 cells was abrogated over that concentration range by increasing the cell seeding density to 50 000 cells/well, at which seeding density the cells grew to a uniform dense monolayer. Similar results were obtained for the murine fibroblast 3T3 cells, which are also contact-inhibited and for HMEC-1 cells, which form multilayered sheets at maximal confluence (data not shown). To rule out reduced transfection as the cause for this effect, repeat experiments were scaled up and transfection efficiency was determined using a fluorescent 5′-OregonGreen-488 congener of Dz13 (DT1565; ), which was complexed with Fugene6 and transfected at a final concentration of 100 nM. DT1565 had reduced cytotoxicity compared to Dz13 (data not shown), which reduced the confounding effects of toxicity on analysis, and the transfection efficiency as determined by FACS analysis showed higher transfection at the higher cell density (B). These preliminary experiments indicate selective cytotoxicity of Dz13 to proliferating cells as opposed to those that are contact-inhibited and non-proliferating. In order to identify whether proteins bind the cytotoxic ODNs and if binding correlated with cytotoxic potency, protein pull-downs were performed using 3′ biotinylated congeners of Dz13 and Dz13scr (DT1309h and DT1310c, respectively; ) as ‘bait’. In HMEC-1 cytotoxicity assays, these 3′-biotinylated ODNs had very similar cytotoxic properties to their untagged counterparts (data not shown). For the pull-down experiments, the biotinylated ODN was coupled to magnetic streptavidin beads and a crossover design employed in which beads coupled to the biotinylated ODN of interest were incubated with HMEC-1 protein, washed with the opposite ODN and eluted with the non-biotinylated bait ODN. That is, when the beads were prepared with the biotinylated congener of Dz13scr and loaded with cell protein extracts, they were subsequently washed with a Dz13 solution and proteins eluted with a Dz13scr solution. For the Dz13-coupled beads, a large number of non-specific proteins were washed off with Dz13scr and a number of strong staining bands were obtained by competitive elution with Dz13, the major one being a ∼51 kDa protein (A). Elution with the non-cytotoxic Dz13scr yielded a large number of bands including a predominant band at ∼39 kDa, however unlike Dz13, none of these were specifically eluted by Dz13scr. Silver staining revealed a lack of protein remaining bound to unconjugated beads following extensive washing (data not shown). In order to generate enough material to identify proteins eluting from the beads, the experiment was scaled up 5-fold and 1 mg of protein was loaded onto beads coupled to the biotinylated ODNs. In these experiments, the number of washes was increased to 30 while the elution volume was kept constant. Following concentration of the eluted fractions, the proteins were digested with trypsin and analysed by LC-MS/MS, with liquid chromatography used to separate the reasonably complex mixture of peptides prior to mass spectrometric analysis. A number of proteins eluting with both Dz13 and Dz13scr were identified (). Proteins identified in the Dz13-eluting fraction (from Dz13-coupled beads) included microtubule-associated protein 4 (MAP4), nucleolin, vimentin, elongation factor 1A (eEF1A), plasminogen activator inhibitor 1 (PAI-1) RNA-binding protein, glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and signal recognition particle 14 kDa. Proteins identified in the Dz13scr-eluting fraction (from Dz13-coupled beads) included annexin II, GAPDH, heterogeneous nuclear ribonuclear proteins A1, Lamin A/C isoform 1 and nucleolin. The 51 kDa band was competitively eluted by Dz13 in three independent experiments and eEF1A was identified in these fractions by LC-MS/MS each time. Overall, a total of 10 peptides from this protein were identified that together spanned most of the protein sequence (). Western blotting of pull-down proteins with an antibody directed to eEF1A confirmed that it was the major band identified by silver staining at 51 kDa and that the protein bound to both Dz13 and Dz13scr beads but was only eluted by Dz13 (B). The LC-MS/MS identification of vimentin was also confirmed by western blots in which vimentin was shown to bind to Dz-13 coupled beads as well as to Dz13scr-coupled beads (data not shown). Elution of vimentin from the beads with Dz13 or Dz13scr was not detected by western (data not shown). This may reflect the fact that more protein (5-fold) was loaded onto the beads for LC-MS/MS analysis. Vimentin distribution within the various cell compartments was assessed using cell fractionation and western blotting 24 h post-transfection with Dz13 and Dz13scr. Vimentin was mostly localized to the cytoskeletal and nuclear fractions and this distribution did not appear to be modified 6 or 24 h following transfection with 200 nM Dz13 (data not shown). In order to further define the role of eEF1A in the cytotoxic mechanism of Dz13 and to determine whether cytotoxicity correlates with eEF1A binding, pull-down assays were performed where active and inactive oligodeoxynucleotides were used to elute proteins bound to Dz13. DT1501 (rat Dz13 homologue) and Dz13scr are non-cytotoxic to HMEC-1 cells under the conditions used here (). Cytotoxicity experiments were performed with a GT-rich oligodeoxynucleotide (DT1605; ) previously reported as binding to eEF1A and inducing cytotoxicity in human T-lymphoblasts (). However, this GT ODN was not cytotoxic to HMEC-1 cells at concentrations of up to 200 nM (data not shown). The oligodeoxynucleotides DT1595 (AGGG repeat), Rs6 (DT1556) and NT36 (DT1577; ATM-inducing) all resulted in significant cytotoxicity to HMEC-1 cells (). Elution of eEF1A from Dz13-coupled beads occurred for all the cytotoxic oligodeoxynucleotides tested (). In contrast, elution of eEF1A using non-cytotoxic oligodeoxynucleotides was very poor (). Nevertheless, when Dz13 was subsequently used as an eluting solution, eEF1A was strongly released from beads that had been first eluted with the non-cytotoxic oligodeoxynucleotides Dz13scr and DT1501, and to a lesser extent with DT1605. eEF1A was only weakly released by Dz13 from beads where eEF1A had been strongly eluted using the cytotoxic oligodeoxynucleotides DT1595, DT1577 and DT1556 (). This demonstrates that eEF1A initially captured by the biotinylated Dz13 beads and only weakly released by the non-cytotoxic ODN could still be released with Dz13 solutions. To determine the cellular localization of the eluting eEF1A, the pull-down procedure was performed on the cytoplasmic, nuclear, membrane and cytoskeletal fractions of HMEC-1 cells. Whereas eEF1A from the cytoplasmic, nuclear and membrane/organelle fractions bound to Dz13 beads, the major source of eEF1A eluted by Dz13 was from the cytoplasm (). Pull-downs were also performed with total protein lysates from a number of other cell lines including RSMC, HEK-293 and HCT-116, to which Dz13 is cytotoxic (). Predominant elution by Dz13 of the 51 kDa band was observed in all cases and this band was confirmed as being eEF1A by western blotting (data not shown). The results indicate that eEF1A binding also occurs in other cell lines in which Dz13 is cytotoxic. When examined directly by western blotting, the abundance of eEF1A was not dependent on the cell type and was not affected by treatment with Dz13 for 24 h (data not shown). The amount of eEF1A eluted with Dz13 was concentration dependent in that a concentration of 10 μM Dz13 eluted more eEF1A than a concentration of 1 μM (data not shown). Nevertheless, in all experiments elution was only partial, and residual eEF1A was found associated with beads post elution ( and ). This suggests that the interaction might not be canonical or that eEF1A undergoes a conformational change upon binding that inhibits its release. In addition to binding experiments, cells were transfected with 100 nM biotinylated Dz13 and Dz13scr and pull-downs performed directly on cell extracts. There was binding of eEF1A to both Dz13 and Dz13scr beads (data not shown), confirming the presence of the interaction of both oligodeoxynucleotides when the binding occurs within the cells. Furthermore, despite the high abundance of eEF1A, a possible interaction between Dz13 and eEF1A directly in cells was demonstrated using fluorescence colocalization experiments (Figure S2). Whilst there are a number of reports describing the cytotoxic nature of certain oligodeoxynucleotides, in particular G-rich oligodeoxynucleotides (see Introduction Section), there is little known about the motif or sequence requirements for cytotoxicity. The current study attempted to define further the exact requirements for cytotoxicity and identify proteins involved in binding to these oligodeoxynucleotides, thereby unravelling the mechanisms involved in the eventual cytotoxicity of the molecules. We determined that the 5′ sequence, the core or the tail of Dz13 alone are insufficient to regenerate the cytotoxic activity of Dz13 against HMEC-1 cells, thereby indicating a length and/or sequence requirement. We used the novel strategy of testing pools of oligodeoxynucleotides comprised of random sequences along with the 5′ G-rich elements of Dz13 and Rs6 in various position and length contexts to demonstrate that the G-rich element needs to be present at the 5′ extremity and be followed by a tail component that is preferably composed of mixed nucleotides. A comparison of the active oligodeoxynucleotides, in combination with the 5′-motif reduction experiment, leads us to conclude that the required 5′ sequence is G-rich, composed of 6–9 nt with at least four consecutive purines. The presence of a triple G motif (G-G-G) provides for the greatest potency and the 5′ extremity of the triple G motif needs to be positioned no more than 3 nt from the 5′ end of the ODN. We have previously shown by CD that Dz13 does not assemble into stable secondary structures (). Nevertheless, substitution of any of the guanosines in the triple G motif of Dz13 abrogated its cytotoxic activity (). This suggests that the hydrogen bonding activity of these guanosines does not lead to G-quadruplex formation, but is nevertheless required for cytotoxicity, perhaps by enhancing binding of the oligodeoxynucleotides to intracellular proteins. The tail sequence requirements were intriguing in that most sequences with some degree of ‘complexity’ supported cytotoxic activity, but some individual tail sequences, in particular those free of guanosines, had reduced activity. It is possible that the less complex tail sequences such as the polydA tail are being sequestered by polydA-binding proteins thereby abrogating cytotoxicity. Wu . () have previously demonstrated that tail sequence and complexity encourages the multimeric aggregation of the oligodeoxynucleotides, thereby activating TLR9 and cell uptake. However, we have previously demonstrated a need for transfection and ruled out endosomal TLR engagement in the mechanism of action of Dz13 and Rs6 against HMEC-1 cells (). Therefore, although multimeric assembly remains a possibility, it is unlikely to relate to uptake and TLR engagement. We previously compared the activity of Dz13, a prototypic G-rich oligodeoxynucleotide, with that of other published sequences including NT36, an ATM-inducing oligodeoxynucleotide () and 20AG, an oligodeoxynucleotide originally designed to be a triplex-forming inhibitor of K-ras (). Both of these oligodeoxynucleotides have purine tracts in the 5′ sequence and NT36 is of comparable length to the active ODNs described in the present study. As shown for Dz13, the potency of the shorter 20AG ODN was length dependent. That is, the cytotoxicity of the AG molecule increased as the length of the molecule increased. ODNs have been proposed to exert their cytotoxicity through specific interaction with cellular proteins such as nucleolin () and eEF1A (). Nucleolin was identified as one of the proteins that bound to Dz13 and Dz13scr. The absence of an effect on HMEC-1 proliferation by the nucleolin-binding oligodeoxynucleotide GRO29A provided further evidence for the lack of direct involvement of nucleolin in Dz13-mediated cytotoxicity. Likewise, a representative from a class of GT-rich ODNs, which reportedly binds eEF1A and is cytotoxic to human T-lymphoblasts (), was not cytotoxic to HMEC-1 cells. Furthermore in our experiments, the GT oligodeoxynucleotide did not displace eEF1A from Dz13 capture beads as potently as Dz13 or the other cytotoxic ODNs tested, indicating that it has less binding affinity to the eEF1A found in this cell line. The ODNs described in the present study therefore represent a novel class of potently cytotoxic molecules. The ability of several of the proteins, including eEF1A to bind competitively to the cytotoxic oligodeoxynucleotides and not be eluted by the non-cytotoxic ODNs suggests that the ODNs may competitively bind to proteins within the cells and that this is mechanistically relevant to the cytotoxicity of the molecule. eEF1A is an extremely abundant protein with a multitude of roles including protein synthesis, stress-sensing, apoptosis and cellular proliferation (,). Scaggiante . () recently reported a correlation between eEF1A binding and cytotoxicity of G-rich oligodeoxynucleotides that is independent of secondary structure formation. They also reported that some non-cytotoxic G-rich oligodeoxynucleotides bind eEF1A to a small degree in the absence of competitor (). The fact that both Dz13 and Dz13scr were able to bind eEF1A and in cells indicates that some of the captured eEF1A is also binding via a non-specific mechanism in HMEC-1 cells. However, the selective release of eEF1A from Dz13 capture-beads with those ODNs that are cytotoxic suggests that there is a relationship between eEF1A binding and cytotoxicity. eEF1A has been proposed to act as a signalling molecule in proliferating cells through the binding of other nucleic-acid-binding proteins such as ZPR1 () followed by translocation to the nucleus. Once inside the nucleus eEF1A has the capacity to bind DNA, RNA and RNA polymerase and therefore potentially plays a role in transcriptional regulation. Binding of eEF1A to Dz13 was observed in nuclear, cytoplasmic, membrane/organelle and cytoskeletal fractions; cytoplasmic eEF1A eluted most strongly from Dz13 beads. This could reflect a greater concentration of DNA-binding proteins in the nuclear fraction and consequently a greater number of proteins eluting in general or that nuclear eEF1A has a higher affinity for Dz13. It is possible that the higher affinity cytotoxic oligodeoxynucleotides displace eEF1A from nuclear or cytoplasmic protein complexes such as the nuclear cytotoxicity-related complex (CRC; 7), thereby inhibiting the normal functioning of eEF1A in the cell, resulting in a reduction in cellular proliferation and eventually cell death. Vimentin was also identified in the pull-down assay. Vimentin has been shown to bind to G-rich ODNs, causing the translocation of the vimentin–DNA complex to the nucleus (). However, this was not evident in HMEC-1 cells transfected with Dz13. Collectively the protein pull-down experiments indicate that binding of cytotoxic ODNs to eEF1A and other proteins is potentially a key event in the mechanism of action of the molecules. In conclusion, we have determined that ODNs such as Dz13 and Rs6 which were initially designed as catalytic DNAzymes, belong to a group of cytotoxic G-rich ODNs with novel sequence and length requirements. Although the mechanism is unclear, it is likely to be mediated through binding to eEF1A. These molecules are preferentially active against proliferating cells and therefore constitute part of a new class of potentially useful molecules for the treatment or investigation of diseases characterized by abnormal cell proliferation. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Transcription and pre-mRNA splicing are the two key nuclear steps regulating gene expression in eukaryotes. Both transcription and splicing are highly ordered processes involving intricate protein–protein interactions. Transcription is regulated by protein complex composed of transcription factors, basal transcriptional machinery and a group of co-activators (), while pre-mRNA splicing is carried out by spliceosome that consists of five conserved small nuclear ribonucleoprotein particles (snRNPs) U1, U2, U5, base-paired U4/U6 and a large number of less precisely defined non-snRNP proteins (). Accumulating evidence supports the concept that transcription and splicing are coordinated (). The key factor linking these processes is the C-terminal domain (CTD) of pol II large subunit which plays essential role both in transcription and RNA processing (,). Additionally, CTD associates with general RNA processing machineries, functioning in mRNA capping, pre-mRNA splicing, polyadenylation and alternative splicing (). However, recent evidence supports the idea that CTD is required for efficient co-transcriptional editing of some pre-mRNA, but is not required for efficient splicing of the pre-mRNA (). Thus, the CTD is most likely a coordinator, rather than a simple enhancer of RNA processing. Additional evidence supporting the coupling between transcription and polyadenylation is emerging from recent proteomic analyses of the human spliceosome which revealed that at least 30 spliceosomal proteins are also participating in other gene expression steps besides splicing (). For example, transcription cofactor such as TAT-SF1 (TAT specific factor 1) (), CA150 () and SKIP (Ski-binding protein) () have also been identified in the spliceosome. These results imply that gene expression machineries are extensively coupled via protein–protein interactions. However, although the evidence of CTD recruitment of mRNA processing factors is well documented, the interactions between CTD and other proteins that link transcription and RNA processing have been poorly characterized. Hence, identification of functional protein complexes and the shared protein components between them is an essential step in understanding the physiological mechanisms that couple transcription and RNA processing. p100 protein was first identified as a co-activator of EBNA2 (Epstein–Barr virus nuclear protein 2) (), and subsequently discovered as co-regulator of pim-1 (), and STAT6 transcription factor in IL-4 mediated gene regulation (,). p100 functions also as a co-activator of STAT5 in prolactin (PRL) signaling (), and in mammary epithelial cells p100 protein levels increase in response to PRL during lactation and correlate with induction of β-casein gene expression (). In addition, p100 has been linked to the pathogenesis of autosomal-dominant polycystic kidney disease (ADPKD) (), and identified in the RISC (RNA-induced silencing complex) (). These studies suggest that p100 protein participates in several biological responses and that the protein may play distinct roles in various cellular events. p100 protein consists of four similar domains with homology to the staphylococcal nucleases (SN), followed by a C-terminal Tudor-SN (TSN) domain (). The SN-like domains of p100 have been implicated in protein interactions, and the SN-like domains of p100 protein recruit CBP and RNA pol II to STAT6 and facilitate the formation of STAT6 enhanceosome (). The function of p100 TSN domain has remained elusive, but the Tudor domain is 30% homologous to SMN (survival of motor neuron protein) Tudor domain, which functions in the assembly of snRNP complexes and pre-mRNA splicing process (,; ). In this report, we demonstrate a novel function for p100 protein as an interaction partner for spliceosomal snRNPs. First, we used an pull-down assay to demonstrate that TSN domain of p100 protein associates with a subset of U5 snRNP core proteins, and is capable of precipitating all snRNAs. We next showed that exogenously added p100 protein can kinetically enhance both the spliceosome complex formation and the first step of pre-mRNA splicing. Thus our results suggest that human p100 protein is a novel dual function regulator of both gene transcription and pre-mRNA splicing. HeLa cells and COS-7 cells were cultured as described previously (). The HeLa-p100-Flag stable cell lines were cloned as described previously(). Plasmid encoding AdML pre-mRNA was kindly provided by Dr R. Reed (). The pSG5 vector expression plasmids containing full-length p100 tagged with Flag sequence, GST-p100-TSN and GST-p100-SN were generated as previously described (). GST-SMN-Tudor was generated by cloning PCR products corresponding to amino acids 92–156 of SMN and inserted into pGEXT-4T-1 with EcoRI and NotI. pBluescript plasmids containing different domains of U5-220, which could be used for translation, were constructed by cloning PCR products corresponding to amino acids 1–209 (domain 1), 303–668 (domain 2), 673–907 (domain 3), 914–1170 (domain 4), 1167–1698 (domain 5), 1697–2121 (domain 7), 2040–2332 (domain 7). All PCR products were sequenced. GST pull-down experiments were performed as described previously(). GST fusion proteins were bound on glutathione-Sepharose 4B beads (Amersham Biosciences, Little Chalfont, Buckinghamshire, UK) and incubated with HeLa cell nuclear lysates or -translated S-labeled different domains of U5-220 in binding buffer (12.5 mM HEPES pH 7.4, 0.1 mM EDTA, 0.05% NP-40, 1 mM DTT, 2 mg/ml aprotinin, 0.5% bovine serum albumin). The beads were washed with buffer containing 300 mM NaCl, and then separated by SDS–PAGE and visualized by silver staining, Coomassie blue staining or autoradiography. For mass spectrometric analysis, in order to avoid the contamination of GST fusion protein, the precipitated proteins were dissociated from bead-bound GST-p100-TSN fusion protein by incubating in the elution buffer containing 2 M NaCl, and then concentrated and desalted with Microcon column (cut-off 30 kDa, Millipore Corporation, Bedford, MA, USA) by centrifugation at 13 000 r.p.m. for 30 min. The precipitated proteins were separated by SDS–PAGE and visualized by silver staining or Coomassie blue staining. The bands corresponding to the 220-kDa, 200-kDa and 116-kDa proteins were cut out from Coomassie blue-stained gel and subjected to trypsin digestion as described previously (). The molecular masses of the peptide mixtures were determined by matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry. The molecular masses of the tryptic peptides of the 220 kDa, 200 kDa and 116 kDa proteins were used to search the OWL protein sequence database for candidate proteins using the ProFound program. The nuclear extracts of p100 stable HeLa cells were immunoprecipitated with anti-Prp8 (Santa Cruz Biotechnology, Santa Cruz, CA, USA), anti-Flag M2 agarose (Sigma, St Louis, MO, USA), anti-IgG (Santa Cruz Biotechnology, Santa Cruz, CA, USA), or anti-TMG (trimethyl-guanosine) monoclonal antibody cross-linked to agarose beads (Calbiochem, San Diego, USA). The bound proteins were subjected to SDS––PAGE and blotted with anti-Prp8 antibody, anti-Flag M2 antibody or anti-p100 antibody (rabbit polyclonal sera were raised against TSN domain of p100 protein). COS-7 cells were transfected with pSG5-p100-Flag as previously described (). Transfected COS-7 cells were suspended in Nonidet P-40 lysis buffer (50 mM Tris–HCl, pH 7.6, 300 mM NaCl, 0.1 mM EDTA, 0.5% Nonidet P-40, 20% glycerol, 0.1 mM sodium orthovanadate, 1 mM sodium butyrate). The total cell lysates were incubated with mouse monoclonal anti-Flag M2 agarose at 4°C for 2 h with head-over-tail rotation. After washing with stringent buffer (containing 700 mM NaCl), the purified p100 proteins (P-p100) were eluted from the agarose with Flag peptide (100 μg/ml) (Sigma, St Louis, MO, USA), and concentrated with Microcon column. The purity of P-p100 was confirmed by SDS–PAGE and silver staining. Capped adenovirus splicing substrates that were uniformly labeled with P-UTP were produced by transcription with T7 RNA polymerase (Promega, Madison, WI, USA) using plasmid pAdML linearized with HindIII as template (). The labeled RNAs were purified in denaturing polyacrylamide gel before use. The splicing reactions, containing 40% (v/v) HeLa nuclear extracts, 2 mM MgCl, 10 mM DTT, 20 mM creatine phosphate, 2 mM ATP, were supplemented with various recombinant proteins and pre-incubated at 30°C for 10 min, followed by the addition of 10 000 c.p.m. pre-mRNA. After incubating at 30°C for different time points, the reactions were stopped by addition of 150 µl PK buffer (20 mM Tris, 10 mM EDTA, 300 mM NaCl, 4.5 mg/ml proteinase K, 0.5% SDS and 0.14 mg/ml glycogen), and the incubation was continued for 30 min at the same temperature (). Reactions were extracted with phenol and the RNA was recovered by ethanol precipitation. Subsequently, RNA pellets were resolved in gel loading buffer containing 7.5 M urea, denatured at 95°C for 5 min, chilled on ice before loading on an 8% denatured PAGE with 7 M urea. The gels were visualized by phosphoimager (Fuji FLA-5010). Native gel analysis on 4% acrylamide was performed as described before () and visualized by phosphorimager and autoradiography. Heparin (1 mg/ml final concentration) was added to the splicing reactions prior to loading. The nuclear extracts of HeLa cells that displayed stable expression of p100 were immunoprecipitated with anti-IgG or anti-TMG monoclonal antibody cross-linked to agarose beads. After washing, the beads were incubated with 300 µl PK buffer at 65°C for 60 min. After extraction with phenol/chloroform, RNA was precipitated in ethanol. The bead-bound GST fusion proteins were incubated with HeLa cell nuclear lysates (100 mM or 300 mM NaCl), the precipitated RNA was extracted as described above. The RNA was separated on denatured 6% PAGE gels, transferred to a nylon filter with semi-dry blotter (Owl Scientific, Woburn, MA, USA) in 0.5X Tris–EDTA buffer using a constant 3 mA/cm current for 1.5–2 h, followed by cross-linking with stratalinker (Stratagene, La Jolla, CA, USA). Hybridization condition for snRNA blots has been described previously (). Radiolabeled probes of U1 snRNA, U2 snRNA, U4 snRNA, U5 snRNA and U6 snRNA, were made by transcription of the linearized snRNA plasmids. U7 and 7SK probes were made by PCR and labeled with P-UTP as described before (). Human p100 protein is composed of four repeats of SN-like domains followed by a C-terminal TSN domain (A). The SN consists of two subdomains. The first subdomain belongs to the large oligonucleotide/oligosaccharidebinding (OB)-fold superfamily (), and the second subdomain consists of two independently folded α-helices. The TSN domain is a hybrid SN domain, in which the OB-domain is divided by a domain found in Tudor protein. According to the crystal structure of p100 TSN domain (Shaw,N. ., submitted for publication), the Tudor domain is flanked by two segments of SN, thus we refer it as TSN domain instead, to differentiate it from the bona fide Tudor domains. In order to identify p100 TSN domain interacting proteins, a pull-down assay with GST-p100-TSN fusion protein was carried out. Equal amounts of GST or GST-p100-TSN fusion proteins were bound to glutathione coupled beads and incubated with nuclear extracts of HeLa cells. The precipitated proteins were prepared as described above. Due to the relatively large cut-off (30 kDa) of the Microcon columns and the high salt concentration used in the elution, many of the larger complexes were expected to be dissociated, and the smaller protein components of such complexes were most likely lost during the centrifugation step. As shown in B, several proteins were found to specifically interact with TSN domain of p100 protein. The bands corresponding to 220 kDa, 200 kDa and 116 kDa proteins were recovered and subjected to in-gel trypsin digestion. The molecular masses of the digested peptides were analyzed by MALDI-TOF mass spectrometry. The program ProFound was used to compare the mass maps obtained against theoretical tryptic peptide mass maps in the OWL protein sequence database. The comparison resulted in the identification of a group of U5 snRNP specific proteins: p220 is human U5-220 (splicing factor Prp8 in yeast), p200 is human U5-200 (DEXH box RNA helicase, Brr2 in yeast), p116 is human U5-116 (a putative GTPase homologous to the ribosomal elongation factor EF-2, Snu114 in yeast). To analyze the possible association of p100 protein and U5-220, co-immunoprecipitation experiments were carried out in HeLa cells which stably express Flag-tagged human p100 protein (HeLa-p100). The nuclear extracts of these cells were immunoprecipitated with anti-Flag agarose, or rabbit anti-IgG antibody as a control. The precipitated proteins were separated by SDS–PAGE and blotted with anti-Flag or anti-Prp8 antibodies. As shown in A, anti-Flag antibodies, but not IgG control, precipitated the endogenous U5-220. complex formation was confirmed by a reciprocal experiment in which the nuclear extracts of HeLa-p100 cells were immunoprecipitated with anti-Prp8 antibody, or rabbit anti-IgG antibody as a control, and precipitated proteins were detected by blotting with anti-Prp8 antibody or anti-Flag antibody. As shown in B, U5-220 but not IgG control, precipitated the p100-Flag. These results demonstrate that p100 protein and U5-220 are associated, either directly or indirectly, . We did not have access to antibodies specific to Brr2 or Snu114, and therefore, we were not able to confirm the association of p100 protein and the other two U5 snRNP specific proteins identified in our mass spectrometry analysis. In order to further verify the interaction of p100 protein and U5-220, GST pull-down assay was performed by incubating the -translated different domains of U5-220 with beads-bound GST-p100-TSN fusion protein, or GST as control. As shown in C, domain 2 (aa 303–668) of U5-220 interacted with TSN domain of p100, but not with GST control. Meanwhile, other domains of U5-220 did not associate with either GST alone, or GST-p100-TSN fusion protein. These results demonstrated that p100 protein interacts with U5-220 protein both and . Consistent with the original pulldown and proteomic analysis, the western blotting results showed that the GST-p100-TSN precipitated U5-220, while GST-p100-SN or GST protein alone did not (A). Additionally, northern analysis of the pull-down reactions using spliceosomal snRNA-specific probes and U7-specific probe revealed, similarly to the control SMN pull-down, an efficient pull-down with the GST-p100-TSN fusion protein (C). In contrast, no interaction was observed with the 7SK, which is another nuclear snRNP. This suggests that the common features of spliceosomal U-snRNPs and U7 snRNP, Sm-proteins and trimethylated cap structure may be recognized by p100-TSN domain, because both features are missing from 7SK RNP. U6 snRNP lacks these features as well, but its precipitation is promoted by base-pairing with U4. To further validate the association of p100 protein with snRNPs, the nuclear extracts of HeLa-p100 cells were immunoprecipitated with anti-trimethylguanosine (TMG) cap antibody (), or rabbit anti-IgG antibody as control using the same salt concentration as above (300 mM NaCl). The TMG-cap is present in U1, U2, U4 and U5 snRNAs (), while the U6 snRNA contains a monomethyl cap instead, but is base-paired with the U4 snRNA which contains a TMG-cap. Northern analysis revealed that U1, U2, U4, U5 and U6 snRNAs were all present in the anti-TMG immunoprecipitates, but not in the anti-IgG control sample (E). Additionally, western blot analysis of the anti-TMG-cap immunoprecipitates indicated that the p100 protein was co-immunoprecipitated with anti-TMG-cap antibody, but not with the anti-IgG antibody (D). The efficient co-immunoprecipitation of p100 with TMG antibody indicates that the TMG-cap structure itself is not a target for p100. Together, these results suggest that the p100 protein can interact with both U5-specific proteins and with Sm-proteins. In order to investigate the functional significance of p100-snRNP interaction, we tested the effect of the p100 on splicing using splicing substrate (pre-mRNA) derived from the adenovirus major late transcription unit (AdML). Splicing reactions were either pre-incubated with purified p100 protein or mock-treated, followed by the addition of radiolabeled splicing substrate and subsequent time-course analysis of the splicing reactions. The results in A show that the appearance of the ligated mRNA and exon-lariat intermediate is faster and respond in a dose-dependent manner (lanes 6–8, 10–12) as compared to the reactions containing no exogenously added p100 protein (lanes 2–4). In particular, the first detectable signal from ligated mRNA and exon-lariat intermediate were consistently observed at earlier time point in splicing reactions containing p100. For example, ligated mRNA appeared at 15 min time point with 6-ng/μl p100 protein (lane 10), and at 30 min with 3-ng/μl p100 protein (lane 7), while it was hardly detectable at 30 min in the control (lane 3). Similar trend is seen with ligated exons, which have reached their maximal intensity already at 45 min time point while the control reaction has not (lanes 8 and 12 compare with lane 4). Significantly, there were no major differences in the levels of splicing intermediates and products at the terminal 60-min time point (lanes 5, 9 and 13) or after (data not shown), suggesting that the added p100 does not increase the overall level of splicing, but rather accelerates the kinetics of the splicing reaction. As the second step is not markedly influenced, that is, the amount of excised intron was quantitatively similar in the splicing reactions with or without addition of p100 protein (compare lanes 4, 8 and 12, lanes 5, 9 and 13), the results suggest that either first step, or the assembly steps prior to catalysis were affected. Control reactions with added BSA did not show any stimulatory effect on splicing, thus ruling out possible effects of overall protein concentration (data not shown). To examine the possibility that the assembly steps prior to the catalysis were affected by p100, we analyzed the formation of pre-spliceosomal complex A, fully assembled spliceosomal complex B, and the catalytically active late spliceosomal complex C using native gel electrophoresis with AdML substrate. B shows a time-course analysis of spliceosomal complexes using different amounts of p100 protein. The reactions were either kept on ice (0-min time point), or incubated at 30°C up to 60 min prior to gel analysis. In the control reaction (B, lanes 1–5) the non-specific complex H is first detected at 0 min time point, complex A at 5 min time point and the later B and C complexes after 30 min of incubation. In the presence of p100 protein, we reproducibly observed relatively strong complex A formation in reactions on ice (lanes 1, 6 and 11) in a p100 protein dose-dependent manner. Additionally, in those reactions the complex A band has reached its maximal intensity already at 5 min time point. Furthermore, the kinetic appearance of B and C complexes was also accelerated. The complex B could be detected already at 5 min time point in the reaction with 6-ng/μl p100 protein (lane 12), and at 10 min (lane 8) with 3-ng/μl p100 protein, while at 30 min (lane 4) in the control reaction. Finally, there appears to be a faster turnover of the splicing reaction particularly with the higher concentration of the p100 as the level of complexes detected in the native gel drops significantly after 60 min of incubation (compare lanes 5, 10 and 15). Furthermore, the stimulation of the complex formation appears to be dose-dependent as the complex formation is faster in the reaction containing a large amount of the p100 protein (compare reactions 6–10 and 11–15), which is consistent with the results observed in splicing reactions (A). To validate this point further, we determined the intensities of A and B complex from B using a phosphoimager and normalized each set by setting the highest value of complex A to 1. As shown in C, the addition of p100 protein accelerated the kinetics of complex A and B formation, the transition of complex A to B (at 30 min in control, 10 min with 3 ng/μl and 5 min with 6 ng/μl p100 protein), and also the decay at later time points. To rule out possible stimulatory effects of increased protein concentration, similar native gel analyses were carried out using equivalent amounts of control proteins, such as GST () and BSA, neither of which changed the kinetics of the complex formation (data not shown). To identify the domains of p100 protein responsible for the observed stimulation of splicing kinetics we performed splicing and spliceosome complex analysis using the separated TSN and the SN domains of the p100 protein. As shown in A, addition of the recombinant TSN domain resulted in clear accumulation of final product and lariat intermediate of RNA already after 30 min reaction (lanes 9 and 10, 12 and 13). Particularly the lariat intermediate band is 2 to 3.5-fold (in lanes 9 and 12, respectively) more intense than the corresponding bands in lanes 3 and 6. The GST control (lanes 5–7) did not display any differences compared to the reaction without any additional protein (lanes 2–4). Additionally, the isolated TSN-domain displays similar dose-dependent stimulation of the splicing kinetics as the full-length p100 protein (compare lanes 8–10 and 11–13). At 30 min time point (lanes 3, 6, 9 and 12), the amount of ligated mRNA, especially the exon-lariat intermediates were increased with the addition of TSN protein. Surprisingly, the addition of recombinant SN domain of p100 protein led to a total block of the splicing reaction (lanes 14 and 15). Taken together, the splicing analysis indicates that the TSN domain alone is sufficient to accelerate the kinetics of the splicing reaction. These results were confirmed with the native gel analysis. Consistent with the splicing result, the native gel analysis demonstrated that TSN domain alone has a similar function in splicing as the full-length p100 protein, that is, it accelerated the kinetics of the spliceosomal complex formation (B lanes 11–15, C), while the GST control had no effect on the pattern of spliceosome assembly (B, lanes 6–10). In particular, in reactions containing 6-ng/μl GST-TSN domain the A-complex reached it's maximal intensity after 5 min incubation (B, lane 12; 5C bottom panel) while in control reactions, the maximal intensity appeared after 30 min incubation (B, lanes 4 and 9; C, top and middle panels). In contrast, addition of the recombinant SN domain blocked the formation of all the specific complexes, leading to a formation of an unknown complex with an intermediate mobility between H and A complexes (lanes 16–20). This result is consistent with the observation above (A), which indicated that addition of the SN-domain alone caused a complete inhibition of splicing, but at present we do not know if it has any biological significance. The p100 protein is an evolutionarily conserved protein present in the genomes of various eukaryotes from fungi to vertebrates (for example, in and humans) (,). Human p100 protein is composed of four repeats of SN-like domains and a TSN domain (A). The SN-domains consist of two subdomains of which the first subdomain belongs to the large OB-fold superfamily, and the second subdomain consists of two independently folded-helices (). The TSN domain of p100 is a hybrid of SN-like domain and a domain that is found in multiple copies in the Tudor protein. We have previously shown that the SN domains of human p100 recruit STAT6 and CBP to RNA pol II, resulting in enhanced STAT6-mediated transcriptional activation (,). Here we provide evidence that the TSN domain of p100 protein is associated with U5 snRNP via specific interactions with U5-220 both and . Under high salt (700 mM NaCl) concentrations we detected association with the U5 snRNP specific proteins including U5-220, U5-200 and U5-116, which is consistent with previous reports showing that these proteins form a stable, RNA-free subcomplex (). Additionally, under moderate salt concentrations we observed interactions with U1, U2, U4, U5, U6 and U7 snRNPs, but not with 7SK snRNP. The appearance of U7, but not 7SK in p100-TSN precipitation indicates that p100 protein is likely to interact with snRNP via recognition of TMG-cap or Sm-protein both of which are present in spliceosomal snRNPs and in U7 snRNP. The efficient co-immunoprecipitation of p100 along with the spliceosomal snRNPs using TMG-cap-specific antibody, argues that TMG-cap may not be the recognition structure of p100, and suggests two modes for the interaction of p100 and snRNPs, one is via the recruitment of U5 snRNP, the other one is via the association of Sm-protein. Comparison of the Tudor-domains of p100, splicing factor SMNrp/SPF30 and SMN demonstrates sequence similarity () between the Tudor domains. This is further supported by our recent structural analysis of the isolated TSN-domain of p100 using X-ray crystallography (Shaw,N. ., submitted for publication) demonstrating that the TSN domain has a hydrophobic core, composed of aromatic residues F715, Y721, Y738 and Y741, which is highly similar to the Tudor domain of SMN. The hydrophobic core mediates interactions between SMN and the Sm-proteins, by binding with the dimethylarginine modifications of the Sm-proteins (,,). Consistent with this hypothesis, mutations of the conserved tyrosine residues (Y738A, Y741A) of the aromatic cage (Shaw,N. ., submitted for publication) abolished the interactions between TSN domain and snRNAs. The association of p100 protein and snRNPs resemble the previous reports of SMNrp/SPF30 which interacts with the U4/U6-90 kDa protein present in tri-snRNP. Similarly to our study, all snRNPs are co-immunoprecipitated with SMNrp/SPF30 antibodies, and conversely the SMNrp/SPF30 is present in anti-TMG IP (,). Interestingly, SMNrp/SPF30 has been shown to be essential for the recruitment of the tri-snRNP to the spliceosome (,), indicating that it functions during the same spliceosome assembly step as p100 protein. It is possible that the p100 utilizes similar interaction mechanisms as the SMNrp/SPF30 during the spliceosome assembly, although the specific molecular interactions are most likely different. We investigated the functional significance of the p100: snRNP interaction using an splicing and spliceosome complex-formation assays using a generic splicing substrate (AdML), and found that p100 accelerated the kinetics of the mature spliceosome formation before the first catalysis. Addition of either a purified full-length p100 protein or the plain TSN domain enhanced spliceosome formation, particularly the formation of the complex A and the transition from A to B complex. In contrast, p100 fragment encompassing the SN domains of the p100 severely inhibited the spliceosome assembly, but we do not know if it has any biological significance. Because p100 only accelerated the kinetics of the spliceosome assembly, but did not affect the overall level of splicing, we hypothesize that p100 may function by recruiting snRNPs, particularly the U4/U6.U5 tri-snRNP, onto spliceosome as supported by the specific interactions between snRNPs and p100. An important question related to the function of p100 in spliceosome assembly is that why has p100 passed undetected in the extensive proteomic analyses of spliceosomes () and why it can enhance the splicing of a generic splicing substrate. A possible explanation is that compared to the previously identified factors functioning during spliceosome assembly, such as SMN or SMNrp/SPF30, the p100 is a large modular protein composed of multiple SN-domains in addition to the TSN domain and has thus other functions besides RNA processing. Our previous studies indicated distinct functions for the different domains, and the SN-like domain alone was sufficient to enhance STAT6-mediated transcription activity in response to IL-4 stimulation (), while expression of the TSN alone did not affect the transcriptional response. Here we show that the isolated TSN stimulated the kinetics of spliceosome assembly, while the isolated SN-domain was highly inhibitory in the splicing assay. The stimulation of splicing of a generic splicing substrate may reflect the conditions of the in vitro reaction in which the p100 protein is free in the solution, while it may be associated with specific genes via promoter regions. Taken together, the results from this and previous studies suggest that the human p100 protein is dual function regulator of gene expression, capable of interacting with different protein complexes through distinct domains. Thus, it is possible that p100 protein may cooperate with RNA pol II and selectively coordinate gene transcription and pre-mRNA splicing. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The tumor suppressor protein p53 is a sequence-specific DNA-binding transcription factor that regulates the cell cycle checkpoint pathway in response to DNA damage (). The p53 gene is the most frequent target for genetic alterations in cancer, with mutations occurring in ∼50% of all human tumors (,). The tumor suppressor functions of p53 are directly linked to its ability to control the expression of gene products implicated in cell cycle arrest and apoptosis (,). p53 binds as a tetramer to specific response elements located in the transcriptional control regions of p53 target genes, which initiates the recruitment of other transcriptional co-regulators to assemble a transcriptional complex that initiates RNA synthesis (,). A broad spectrum of p53 downstream target genes have been identified to be controlled by p53 in a positive or negative manner (). The fact that a variety of pathways are mediated by these p53 target genes demonstrates the role of p53 as an integrator of diverse cellular signals. The p53 protein contains two N-terminal activation domains, a DNA-binding domain, and a C-terminal oligomerization domain (). The critical step for p53-mediated transcriptional activation is facilitated by the ability of p53 to simultaneously bind to specific DNA sequences and recruit CBP/p300 and other transcriptional co-regulators to p53-responsive promoters. CBP/p300 recruitment appears to concomitantly bring the general transcription machinery, including TFIIB and TBP, and RNA polymerase II to the target promoters (,). Although a wealth of information exists concerning p53, it is unclear about the actual mechanism by which this critical tumor suppressor protein directly interacts with its target genes and co-regulators to mediate its transcriptional activity. The PIAS proteins (protein inhibitor of activated STAT) were first identified as transcriptional co-regulators of the JAK-STAT pathway (). PIAS1 and PIAS3 can inhibit the activity of STAT1 and STAT3, respectively (). Recent studies imply that the PIAS proteins may play a role in chromatin modulation through sumoylation (,). Sequence analysis indicates that the SUMO E3 ligase RING domain shares significant homology with the Miz domain of PIAS proteins (). Several PIAS proteins, such as PIASxα, xβ, 1 and 3, have been shown to interact with SUMO-1 and Ubc9 to sumoylate a variety of transcriptional factors and other regulatory proteins (). Particularly, it has been shown that the transcriptional activity of p53 can be regulated by PIAS through sumoylation (,). hZimp10 and hZimp7, also named zmiz1 and zmiz2, respectively, are novel PIAS-like proteins that share a ring finger domain, termed Miz (msx-interacting zinc finger), with other PIAS proteins (,). This domain has been shown to be important for PIAS-target protein interactions and post-translational modifications (). A novel gene, termed (), appears to be the ortholog of hZimp7 and 10 (). The protein encoded by genetically interacts with the SWI2/SNF2 and Mediator complexes, implying a potential role for the hZimp proteins in transcription. To further explore their roles in transcription, we performed a yeast two-hybrid screen to seek out potential interacting proteins of hZimp7 and 10. Intriguingly, p53 was identified in the screen. Using different and approaches, we demonstrated that hZimp10 physically interacts with the p53 protein, and through the interaction hZimp10 augments p53-mediated transcription. These data elucidate a link between hZimp10 and p53 and demonstrate that hZimp10 is a transcriptional co-regulator of p53. Yeast two-hybrid experiments were performed as described previously (). The DNA fragments containing truncated hZimp10, hZimp7 or hLZTS2 were fused in frame to the GAL4 DBD in the pGBKT7 vector (CLONTECH Laboratories, Inc., Palo Alto, CA, USA). Truncated p53 (amino acids: 251–383) was fused to the GAL4 TAD in the pVP16 vector (CLONTECH). The constructs were transformed into the modified yeast strain PJ69-4A (). Transformants were selected on Sabouraud Dextrose medium lacking tryptophan, leucine and/or adenine. The specificity of the interaction with p53 was measured by a liquid β-galactosidase (β-gal) assay (). Full-length hZimp10 and hZimp7 cDNA was identified and sub-cloned into pcDNA3-FLAG vector as described previously (,). Subsequently, truncated mutants of hZimp10 or hZimp7 were generated and sub-cloned into the pGBKT7 vector containing a GAL4 DBD for the yeast two-hybrid assay or into pGEX4T3 for making GST fusion proteins. Truncated p53 (amino acids: 251–383) was cloned into the pVP16 vector containing the transcriptional activation domain of VP16. Double-stranded oligonucleotides corresponding to human hZimp7 (Z7-1: 5′-GGACTGCATTATAAGCCTAC-3′, Z7-2: 5′-GGACACCAGGACTACACACC-3′, and Z7-3: 5′-GGTGGAGCAGACAGCTATCA-3′) and hZimp10 (Z10-1: 5′-GGCCTCCATTACATCACAGT-3′, Z10-2: 5′-GGCAGCAGCAGCAGTTCTCA-3′ and Z10-3:5′-GGCACCAACTCCAACGACTA-3′) were cloned into the pBS/U6 vector to generate the short hairpin RNA (shRNA) (). Subsequently, the U6 promoter and the hZimp7 or hZimp10 shRNA sequences were PCR amplified and transferred into the pLentiSuper vector (Invitrogen, Carlsbad, CA, USA). The viral vector was co-transfected with other packaging plasmids into human embryonic kidney 293T cells for the virus production (). The pGL2hmdm-HX-Luc reporter was a kind gift from Dr Moshe Oren (The Weizmann Institute of Science, Israel). The p21-Luc reporter, pCMV-NEO-Bam p53 and the p53-induced promoter/reporter PG13-Luc were generously given by Dr Bert Vogelstein (Johns Hopkins, Baltimore, MD, USA). The hZimp10 adenoviral expression vector was cloned into the pAdTrack shuttle vector (). The plasmids were then cleaved with , and transformed into BJ5183 cells that contain pAdEasy-1 vector. Adenoviral vectors were amplified in DH5α cells, and propagated in HEK293 cells. Viral titers were determined using plaque assays. The HCT116 p53 and HCT116 p53 human colon carcinoma cell lines are gifts from Dr Bert Vogelstein, which were maintained in McCoy's medium with 10% fetal bovine serum (FBS, HyClone Laboratories). The SaOS-2 and U2OS cell lines are gifts of Dr Giannino Del Sal (Laboratorio Nazionale C.I.B. Italy) and were cultured in DMEM with 10% FBS. The human breast cancer cell line, MCF7, and embryonic kidney cell line, HEK293, were grown in 10% FBS-DMEM. Transient transfections were carried out using LipofectAMINE2000 (Invitrogen, Carlsbad, CA, USA). Approximately 1.5 × 10 cells were seeded into a 48-well plate 16 h before transfection. Approximately 200 ng of total plasmid DNA per well were used in transfection. Total cell lysates were collected 8–12 h after transfection and then luciferase and β-gal activities were measured in a Monolight 3010 luminometer (Pharmingen, San Diego, CA, USA). Luciferase activity was normalized by β-gal in the same samples and reported as relative light units (RLU). Individual transfection experiments were done in triplicate and the results are reported as mean RLU (±SD) from representative experiments. Expression and purification of GST fusion proteins were performed as described previously (). Equal amounts of GST-fusion proteins coupled to glutathione sepharose beads were incubated with MCF7 cell lysate at 4°C for 2 h in binding buffer (20 mM Tris-HCl-pH 7.8, 180 mM KCl, 0.5 mM EDTA, 5 mM MgCl, 0.5 mM ZnCl, 10% glycerol, 0.1% NP-40, 0.05% dry non-fat milk, 1 mM DTT, 0.5 mM PMSF). Beads were carefully washed three times with binding buffer and then analyzed by SDS-PAGE followed by western blot analysis using a p53 antibody (DO-1: sc-126 Santa Cruz Biotechnology). The pcDNA3-HA-p53 with or without pcDNA3-FLAG-hZimp10 vector was transfected into HEK293 cells. After 48 h of transfection, cells were incubated with 0.5 mM dithio-bis (succinimidyl propionate) (DSP) in PBS to cross-link for 30 min, washed once with 10 mM Tris-HCl pH 7.6, and incubated with the same buffer for 5 min to neutralize DSP. The whole cell lysates were prepared as described previously (), and incubated with different antibodies at 4°C with gentle rotation overnight. Then equilibrated Protein-A Sepharose beads were added for 1.5 h at 4°C, and then collected. The beads were washed, proteins eluted using 2× sample buffer (125 mM Tris-HCl pH 6.8, 4% SDS, 20% (v/v) glycerol, 0.004% bromphenol blue), and analyzed by western blot. Detection was performed with ECL reagents according to manufacturer's protocol using ECL Hyperfilm (Amersham). MCF7 cells were infected with either the hZimp10 expression adenovirus or a control virus in 5% FBS-DMEM, then UV irradiated at 80 J/m, and cultured in the medium for 9 h. Total RNA was then extracted using RNAWiz RNA isolation reagent (Ambion, TX; Cat#: 9736). The reverse transcription polymerase chain reaction (RT-PCR) was carried out as described previously (). In a 50 µl PCR reaction, 1 μl of cDNA was amplified using 25 cycles of 45 s at 94°C, 30 s at 61°C (for p21) or 50°C (for GAPDH) and 30 s at 72°C. Primers for p21 (5′-ATGTCAGAACCGGCTGGGGAT-3′; 5′-GGAGTGGTAGAAATCTGTCATGC-3′), mdm2 (5′ –ATGTGCAATACCAACATGTCTGTACC-3′; 5′ –TTTGGTCTAACCAGGGTCTCTTGT-3′) and GAPDH (5′-CCATGGAGAAGGCTGGGG-3′; 5′-CAAAGTTGTCATGGATGACC-3′) were used in the PCR reaction. For quantitative PCR, cDNA samples were mixed with SYBR qPCR Super Mix Universal (Invitrogen) and specific primers in the MX 3005P thermocycler (Stratagene). Relative mRNA levels were calculated from the point where each curve crossed the threshold line as reported previously (,). Reactions were done in triplicate and the values were normalized by the GAPDH expression level. MCF7 or HEK293 cells were irradiated at 40 J/m using a UV Stratalinker 1800 (Stratagene) and then incubated at 37°C for 6 h. Subsequently, cells were treated with DSP to cross-link protein complexes and washed with Tris-wash buffer. After the last wash with PBS, cells were treated with formaldehyde and subjected to ChIP analysis as described previously (). Briefly, cells were collected and washed sequentially with cold PBS, Wash Buffer I (0.25% Triton X100, 10 mM EDTA, 0.5 mM EGTA and 10 mM HEPES, pH 6.5), and Wash Buffer II (200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA and 10 mM HEPES, pH 6.5). Cells were then lysed in lysis buffer (1% SDS, 10 mM EDTA, 50 mM Tris pH 8.1 and protease inhibitors). The chromatin was sheared to an average size of 800 bp by sonication and diluted 10-fold in ChIP dilution buffer (2 mM EDTA, 150 mM NaCl, 20 mM Tris-HCl, pH 8.1, 1% Triton X-100), and then subjected to immunoprecipitation with either an anti-Zimp10 chicken antibody or anti-p53 antibody (Santa Cruz) for overnight at 4°C and recovered with goat α IgY-agarose (Aves Labs) or Protein A Sepharose (Amersham). The immunoprecipitates were serially washed with different TSE-based buffers and eluted for PCR analysis. For re-ChIP, the immunocomplexes were eluted with re-ChIP elution buffer (10 mM DTT), and the supernatant was diluted 1:40 in ChIP dilution buffer. Antibodies against the second proteins of interest were added, incubated at 4°C overnight, and collected by incubating with either goat anti-chicken IgYagarose or protein-A beads at 4°C for 2 h. In both ChIP and re-ChIP assays, the immuno-complexes were eluted from the beads through incubation with 10X bead volume of elution buffer (1% SDS, 0.1 M NaHCO). Cross-links were reversed by incubating elution samples at 65°C for 6 h and chromatin fragments were purified with the PCR Purification Kit (Qiagen). ChIP and input DNA were analyzed by PCR using p21 promoter-specific primers, 5′-GTGGCTCTGATTGGCTTTCTG-3′ and 5′-CTGAAAACAGGCAGCCCAAG-3′, respectively (,). P21 PCR parameters were: 95°C for 5 min then 33 cycles of 95°C for 30 s, 60°C for 30 s and 72°C for 20 s. The samples were also amplified with GAPDH primers, 5′-CGGTGCGTGCCCAGTTG-3′ and 5′-GCGACGCAAAAGAAGATG-3′, as controls (). HEK293 cells were seeded overnight and synchronized with 0.5 mM mimosine (Sigma Chemical Co., St. Louis, MO, USA) as described previously (). Cells were released from the mimosine block by washing three times with PBS and incubating in fresh 10% FBS-DMEM at 37°C for 12 h. Cells were fixed with 4% paraformaldehyde and immunostained with either an anti-Zimp10 or anti-p53 antibody followed by incubation with species-specific Alexafluor 488 and 594-conjugated secondary antibodies. Images were analyzed by confocal microscopy with 60× and 40× objectives. Mice heterozygous for a neomycin-disrupted allele of the gene were mated, and embryos were harvested from the females at embryonic day 9.5. Embryos were isolated in cold PBS and digested with 250 μl trypsin (0.05%) as described previously (). Cells were directly plated into 48-well plates to adhere overnight, and used for transient transfection assays. To determine MEF genotype, embryo sacs isolated during the dissection were digested, genomic DNA was extracted, and the wild-type or mutant Zimp10 alleles were determined by PCR with specific primers. Using a bait construct containing the N-terminal region of hZimp7 (amino acids: 1–643), which is also highly conserved in hZimp10, we employed a modified yeast two-hybrid system to identify proteins that potentially interact with hZimp7 and/or hZimp10. Of 3.4 × 10 transformants, 123 grew under selective conditions and showed increased adenine and β-gal production in medium. Rescue of the plasmids and sequencing of the inserts revealed several different cDNAs. Most of them were transcriptional factors and PIAS proteins. Among these clones, a cDNA encoding the tumor suppressor p53, between amino acids 251 and 383, was identified. Since hZimp7 and hZimp10 share significant sequence similarity, we co-transformed the p53 clone with various constructs containing either GAL-DBD alone or the fusion proteins with the N-terminal fragment of hZimp7 (1–643 amino acids), and different truncations of hZimp10 (A–C). The original bait construct, pGBKT7-hZimp7 (1–643aa), showed a specific interaction with pVP16-p53 (251–383aa) (D). Interestingly, the construct containing the fragment of hZimp10 between amino acids 451–753 appeared to interact with the pVP16-p53 as well. The above results provide the first line of evidence to demonstrate an interaction between hZimp proteins and the tumor suppressor protein p53. The interactions between hZimp7 and hZimp10 with p53 were further assessed by GST pull-down experiments. Truncated hZimp7, hZimp10 and the PIAS1 SAP (Scaffold attachment factor, Acinus and PIAS) and MIZ domains were cloned in frame to generate GST fusion proteins. They were expressed upon IPTG induction and equalized on a SDS–PAGE gel with comassie blue staining (A). MCF7 cell lysates were applied to each sample, which was immobilized onto a glutathione-sepharose matrix for the binding assay. The elutions from the above samples were analyzed by SDS-PAGE and western blot with a p53 antibody. PIAS1 and Zimp proteins contain a conserved Miz domain that shares similar sequences with the ring finger domain of MDM2 (). Therefore, we tested whether the Miz domains of PIAS1 and Zimp10 interact with p53. Intriguingly, only a weak interaction was observed between the Miz domains and endogenous p53 proteins (A). In addition to the Miz domain, the SAP (Scaffold attachment factor A/B, Acinus and PIAS) domain has also been shown to be involved in protein–protein interactions (). Thus, we included it in our binding assays, although only slight binding activity above background was observed with this construct. Notably, the GST-fusion construct containing amino acids 451–753 of hZimp10 showed the strongest binding activity with p53. Since this region only covers a small portion of the Miz domain, this result suggests that the Miz domain of hZimp10 may not be required for the interaction with p53. In addition, the GST-fusion protein containing the N-terminal fragment of hZimp7 also showed an interaction with p53, which is consistent with the yeast two-hybrid results. Taken together, the above results show that the regions spanning hZimp10 amino acids 451 to 753 and hZimp7 amino acids 1–643 are mainly responsible for binding to p53. To confirm that hZimp10 interacts with p53 in intact cells, co-immunoprecipitation assays were carried out to detect potential protein complexes. Initially, we co-transfected FLAG-tagged hZimp10 with a p53 expression vector in HEK293 cells (B). Whole cell lysates containing FLAG-hZimp10 and p53 proteins were immunoprecipitated with normal mouse IgG or a anti-p53 antibody. As shown in C, Flag-hZimp10 proteins were only detected in the p53 immunoprecipitate but not in normal IgG immunoprecipitate. These data indicate that p53 can form a protein complex with hZimp10 in intact cells. Next, we further evaluated the interaction between endogenous hZimp10 and p53 proteins in HEK293 cells. With specific antibodies against p53 and hZimp10, we detected the expression of both proteins in HEK293 cells (D). Immunoprecipitation of the whole cell lysates with a homemade hZimp10 antibody () revealed that p53 forms a protein complex with hZimp10, which provides evidence to demonstrate that hZimp10 and p53 interact endogenously. Since hZimp7 and 10 have been shown to act as transcriptional co-activators (,,,), we tested whether these Zimp proteins regulate p53-mediated transcription. In order to avoid potential confounding effects of endogenous p53, we chose p53-null cells for these experiments. A luciferase reporter driven by the MDM2 promoter (MDM2-Luc) was co-transfected with plasmids expressing p53, hZimp7 and hZimp10 into HCT116 p53 colon cancer cells in various combinations (A). An approximate 2-fold increase of p53-mediated transcriptional activity above the baseline was observed when cells were transfected with 0.4 ng of p53 expression vector. The p53 activity was further increased ∼7- or 9-fold in the presence of 40 or 80 ng of hZimp10, respectively, above the baseline. In contrast, co-transfection of hZimp7 showed no significant effect (A), which is consistent with the yeast two-hybrid and GST pull-down assays that showed only a weak interaction between p53 and hZimp7. There was no effect when hZimp7 or hZimp10 were transfected alone with the reporter plasmid, indicating that the Zimp10-mediated induction in reporter activity was indeed through p53 (A). To confirm our findings, we repeated the transient transfection assay in SaOS2 osteosarcoma cells, another p53 negative cell line. A similar augmentation of hZimp10 on p53-mediated transcription was observed on the MDM2 promoter/reporter (B). Moreover, we evaluated the effect of hZimp10 on p21 promoter, another downstream target of p53. A similar enhancement by hZimp10 was observed on p21 promoter in HCT116 p53 and SaOS2 cells (C and D). These results provide evidence to demonstrate that hZimp10 augments p53-mediated transcription. To investigate the effect of hZimp10 in regulating the transcriptional activity of endogenous p53, MCF7 cells, which possess wild-type p53, were infected with either hZimp10 expression adenoviruses or control GFP viruses, and treated with or without UV irradiation. The endogenous transcripts of the p53 target gene, p21, were then measured by semi-quantitative RT-PCR. The level of p21 transcript showed no significant change in cells in the absence of UV treatment (E). However, in cells treated with UV irradiation, the level of p21 transcript was increased ∼1-fold in cells with ectopically expressed hZimp10 in comparison with the cells infected with GFP control viruses. The enhancement of hZimp10 on p53-meidated transcription in MCF7 cells was further evaluated by quantitative PCR assays. As shown in F, the levels of MDM2 and p21 transcripts were higher in cells expressed exogenous hZimp10 proteins than control cells ( < 0.05). These results support our initial observation in the transient transfection experiments, and further demonstrate a functional role for hZimp10 in augmenting p53-mediated transcription. Next, we extended our study to investigate the involvement of endogenous hZimp10 in regulating the transcriptional activity of p53. We first generated three short hairpin RNA (shRNA) constructs for hZimp10 () and tested their knockdown effects on ectopically expressed hZimp10 in CV1 cells. All three hZimp10 shRNA constructs reduced the expression of FLAG-tagged hZimp10 protein (A). However, there was no change in tubulin expression in the same samples. The hZimp10 shRNA construct 2 appeared most effective in the knockdown experiment using overexpressed hZimp10 and also significantly reduced the expression of endogenous hZimp10 protein in HEK293 cells (B). In addition, we evaluated the specificity of shRNA vectors for hZimp7 and hZimp10 in MCF7 cells. As shown in C and D, the shRNA vectors specific for hZimp7 or hZimp10 showed selective knockdown for their respective targets. In HCT116 p53 cells, the hZimp10 shRNA construct selectively reduced hZimp10 enhancement of p53-mediated transcription (E), which is consistent with the western blot results. Moreover, knockdown of endogenous hZimp10 expression by the hZimp10 shRNA vector in MCF7 cells resulted in a 5-fold reduction in p53-mediated transcription on both the p21-Luc and MDM2-Luc reporters (F and G). In contrast, there was no change in p53 transcriptional activity in samples transfected with hZimp7 shRNA constructs. Using quantitative PCR assays, we further assessed the suppressive effect of the hZimp10 shRNA on the expression of endogenous p21 and MDM2 transcripts. The levels of p21 and MDM2 mRNAs were reduced in cells infected with hZimp10 shRNA-2 lentiviruses whereas cells either infected with the control viruses or hZimp7 shRNA viruses showed no effect (H). Taken together, the above data indicate that endogenous hZimp10, but not hZimp7, plays an important role in the augmentation of p53-mediated transcription. To further demonstrate the role of hZimp10 in a more biologically relevant setting, we have recently generated mice in which the gene locus has been disrupted by replacing exons 8–10 with a neomycin resistance cassette. The phenotype of this disruption is embryonic lethality at approximately E10.5 (Beliakoff ., unpublished data). To determine whether endogenous Zimp10 regulates p53-mediated transcription, we generated mouse embryo fibroblasts (MEFs) from E9.5 day embryos and transfected them with the p53-responsive MDM2-Luc and pG13-Luc reporters (). As shown in I, ∼4-fold induction of luciferase activity by ectopically expressed p53 on either MDM2 or pG13 promoters, respectively, was observed in MEFs prepared from wild-type embryos (Zimp10). In contrast, no activity was observed in MEFs where both alleles were disrupted (Zimp10). These data demonstrate a crucial role for endogenous Zimp10 in the regulation of p53-mediated transcription . To examine whether a dynamic interaction between p53 and hZimp10 exists in cells, we detected the cellular distribution of both endogenous p53 and hZimp10 proteins using specific antibodies. HEK293 cells were synchronized by adding 0.5 mM mimosine overnight. Cells were then washed three times with PBS and stimulated with full medium for 12 h (). As shown in A, both p53 and Zimp10 proteins show a strong nuclear distribution in human HEK293 cells, which is consistent with previous reports (,,). Intriguingly, a significant amount of overlay between endogenous p53 and hZimp10 proteins was observed in these cells. Based on these observations, we conclude that hZimp10 can co-localize with p53 in the nucleus, where the proteins may form a ternary transcriptional complex. To demonstrate the direct involvement of hZimp10 to coordinate p53-mediated transcription, chromatin immunoprecipitation (ChIP) assays were performed to detect the occupancy of hZimp10 on p53-regulated promoters. HEK293 cells were grown in DMEM with 5% FBS. Soluble chromatin was prepared after formaldehyde treatment of the cell cultures, and specific antibodies against hZimp10 were used to immunoprecipitate hZimp10-bound genomic DNA fragments. The genomic DNA was analyzed by PCR using specific pairs of primers spanning the p53-binding sites in the p21 promoter (B). Both p53 and hZimp10 recruitment was detected within the region of the p21 promoter that contains a functional p53-binding site () in HEK293 cells after UV treatment. As the average length of the genomic DNA fragments produced in these experiments was ∼800 bp (data not shown), we could not distinguish with certainty whether the occupancy of hZimp10 on the p21 promoter is through an interaction with p53 or through interactions with other DNA-binding proteins. Therefore, we performed re-ChIP assays to assess the relationship between p53 and hZimp10 on the p21 promoter (C). Using the hZimp10-specific antibody, we re-immunoprecipitated the elutions from the immunoprecipitates with a p53 antibody or normal IgG. As shown in D, the presence of hZimp10 on the p21 promoter was selectively detected in the immunoprecipitates using the p53 antibody but not the IgG control. In addition, the samples re-immunoprecipitated with anti-hZimp10 showed a more intense p21 promoter PCR fragment than the ones using normal IgY. These data suggest that the recruitment of hZimp10 onto the p21 promoter is mediated through p53. The p53 tumor suppressor is a DNA sequence-specific transcriptional factor that is mutated in ∼50% of human tumors (). In response to a variety of cellular signals, perhaps the most well studied is its DNA damage function. p53 regulates the transcription of numerous genes involved in different cellular processes, including cell cycle arrest and cell death. Like other transcriptional factors, the transcriptional activity of p53 is largely dependent on its ability to recognize and bind specific DNA sequences and to recruit other necessary transcriptional co-regulators. In recent years, numerous transcription co-regulators have been shown to either directly or indirectly interact with p53 to modulate its transcriptional activity. For instance, physical and functional interactions between p53, p300 and HAT proteins have been well documented (,). The involvement of PRMT1 and CARM1 methyltransferases has also been demonstrated in previous studies (). Importantly, p53 has been shown to facilitate formation of a preinitiation complex via direct interactions with the components of the general transcription complex (). The experiments reported here demonstrate a specific protein–protein interaction between p53 and hZimp10, a novel PIAS-like protein (). The interaction was first identified by a modified yeast two-hybrid screen. Using GST pull-down and immunoprecipitation assays, we then show that p53 binds to hZimp10 both and in intact cells. Moreover, immunofluorescence assays demonstrated that p53 co-localizes with hZimp10 within cell nuclei. Furthermore, analysis of the interaction by ChIP (chromatin immunoprecipitation assay) on the promoter of the p21 gene, a downstream target of p53, showed that hZimp10 is involved in the p53-mediated transcriptional complex. Taken together, these multiple lines of evidence clearly indicate that p53 and hZimp10 can specifically interact in a biologically relevant manner. To search for the biological consequence of the interaction between p53 and hZimp10, we performed a series of experiments to assess the effect of hZimp10 on p53-mediated transcription. As shown in this article, hZimp10 acts as a transcriptional co-activator to augment p53-mediated transcription. We observed that expression of exogenous hZimp10 or knockdown of endogenous hZimp10 affects p53-mediated transcription on both the p21 and Mdm2 promoters. Introducing exogenous hZimp10 into MCF7 cells also augments endogenous p53-mediated transcription by increasing p21 transcript levels. Interestingly, hZimp10 consistently up-regulates p53-mediated transcriptional activity in all cell contexts examined to date. This result is consistent with the observation that hZimp10 harbors a strong intrinsic transactivation domain within its C-terminus (). It appears that through this domain hZimp10 can act as a transcriptional co-activator to augment p53-mediated transcription, which is consistent with previous observations showing that hZimp10 functions as a transcriptional co-activator of the androgen receptor and Smad3 (,). It has been shown that the transcriptional activity of p53 can be regulated by multiple post-translational modifications, including phosphorylation, ubiquitination and acetylation (). In addition, p53 can also be covalently modified by sumoylation, which is mainly regulated through SUMO-1 (Small Ubiquitin-related Modifier 1) (,). Recent studies have shown that PIAS proteins can bind to, sumoylate, and influence the activity of p53 (,). In particular, PIAS1 and PIASxβ act as E3 ligases to enhance sumoylation of p53 and (). Although it has been shown that PIAS proteins negatively regulate the transcriptional activity of p53 through sumoylation, recent data indicated that PIAS1 and PIAS3 may function as activators of p53-dependent gene expression (). Previously, we have shown that Zimp10 co-localizes with the AR and SUMO-1 at replication foci and enhances AR sumoylation. However, the mechanism for hZimp10-mediated enhancement of p53 activity appears to be through a sumoylation-independent pathway because over-expression of hZimp10 and SUMO-1 in HEK293 cells showed no effect on sumoylation of the p53 protein (Supplementary Data). In addition, our results indicate that the hZimp10-mediated enhancement of p53 activity may be at least partially Miz domain-independent because the strongest interaction was observed with a hZimp10 region containing only a portion of the Miz sequence (A). Recently, we have demonstrated that both hZimp7 and hZimp10 enhance the transcriptional activity of several transcriptional factors (,,,). However, the precise mechanism(s) for these Zimp proteins in transcriptional regulation still remains unclear. Our previous data showing that hZimp10 co-localizes with newly synthesized DNA at replication foci throughout S phase suggest that hZimp10 may play an important role in both chromatin assembly and maintenance of chromatin (). Intriguingly, a homolog of human Zimp proteins, termed , has been identified in and was shown to genetically interact with SWI2/SNF2 and the Mediator complexes in complementation studies (). In addition, we have shown previously that the C-terminal proline-rich domains of hZimp7 and 10 possess significant intrinsic transcriptional activity (,), and through these domains, the Zimp proteins can enhance transcription both in trans and in . The finding that hZimp10 augments p53-mediated transcription is consistent with our previous studies showing that hZimp10 functions as a transcriptional co-activator of the androgen receptor and Smad3/Smad4 (,). Interestingly, the C-terminal proline-rich region is not found in other PIAS or PIAS-like proteins, suggesting that the Miz domain family may consist of distinct groups of proteins that contain unique structures and play distinct roles in regulating transcription and other cellular processes. Therefore, it is conceivable that although hZimp10 and other PIAS proteins interact with p53 physically, they may regulate the function of p53 through different mechanisms. Indeed, this is in agreement with our results suggesting that the Miz domain is generally dispensable for the hZimp10–p53 interaction while published reports suggest that the Miz domain is important for PIAS–p53 interactions (). hZimp7 and hZimp10 share significant sequence similarity, particularly within their C-terminal regions (). Both proteins contain an intrinsic transactivation domain and function as transcriptional co-activators (,). These two Zimp proteins show different tissue distribution profiles, which may suggest unique roles for these proteins in regulating different target genes. Our recent data showing that disruption of the Zimp10 gene in mice results in embryonic lethality at approximately E10.5 suggest that hZimp7 and 10 are not functionally redundant. In this study, even though we observed that hZimp7 interacts with p53 in a yeast two-hybrid assay, it showed no interaction between the two intact proteins in immunoprecipitation assays (see Supplementary Data). In addition, we only observed a very weak effect of hZimp7 on p53-mediated transcription. These data suggest that hZimp7 and hZimp10 proteins, although structurally similar, likely play a different role in p53-mediated transcription. In this study, we also assessed the interaction between hZimp10 and p53 in Zimp10 null cells. Using MEFs generated from Zimp10 knockout mice, we demonstrated that the disruption of inhibits p53-mediated transcription. In MEFs with an intact wild-type allele, a clear dose-dependent induction of p53 transcriptional activity was observed in cells transfected with increasing amounts of p53. In contrast, no enhancement was observed in cells where both alleles were disrupted. This perhaps provides the most convincing evidence that Zimp10 can indeed regulate p53 activity in an system. Further study using this system should help to elucidate the biological influence of Zimp10 on p53-mediated tumor repressive effects. In conclusion, this study demonstrates for the first time that hZimp10, a novel PIAS-like protein, augments the transcriptional activity of the p53 tumor suppressor. This interaction provides an additional line of evidence to demonstrate that Zimp10 is involved in transcriptional regulation. Further studies into the molecular mechanisms by which hZimp10 and other PIAS proteins regulate p53-mediated transcription may provide new insight into the biological role of PIAS and PIAS-like proteins in cell growth, apoptosis, differentiation and tumorigenesis. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

MicroRNAs (miRNAs) are noncoding small (∼22 nt) RNAs that regulate the expression of target mRNAs (,). MiRNAs are encoded in the chromosomal DNA and transcribed as longer stem-loop precursors, termed pri-miRNAs (). Upon transcription, pri-miRNA is converted to mature miRNA duplex through sequential processing by the RNaseIII family of endonucleases Drosha/DGCR8 and Dicer (). One strand of the processed duplex is incorporated into a silencing complex and guided to target sequences located at the 3′-terminal untranslated regions (3′-UTRs) of mRNAs by base pairing (), resulting in the cleavage of target mRNAs or repression of their productive translation (). Over the past few years, several hundred miRNAs were identified in animals and plants. It is currently estimated that miRNAs account for ∼1% of the predicted genes in higher eukaryotic genomes (). A growing body of evidence revealed that miRNAs are involved in a variety of biological processes (), such as embryonic development, cell proliferation, cell differentiation, apoptosis and insulin secretion. Moreover, several reports indicate that miRNAs are involved in human tumorigenesis. A systematic search for correlation between the genomic position of miRNAs and the location of cancer-associated regions revealed that over half of the mapped miRNAs in chronic lymphocytic leukemias (CLL) are located at fragile chromosome regions involved in human cancers (). MiRNA expression profiles also indicated that most miRNAs had lower expression levels in tumors compared with normal tissues (). For example, Let-7, targeting the oncogene RAS, is downregulated in lung cancers () and miR-15 and 16, targeting the antiapoptotic factor BCL2, are downregulated in CLL (). Furthermore, a germ-line mutation in the precursor was found to cause its reduced transcription in a patient with familial CLL (). Single-nucleotide polymorphisms (SNPs) are the most frequent variation in the human genome, occurring once every several hundred base pairs throughout the genome. They have been studied extensively for defining the regions of disease candidate genes (). Previously, in order to search for SNP-affected disease susceptibility and outcome, most researchers focused on specific genes, as resources and analytical tools were limited. Moreover, until recently, there was a profound interest in nonsynonymous SNPs because they shift the codons and often change the protein structure and function. However, the majority of SNPs in the genome are not nonsynonymous SNPs that occur in untranslated, intronic or intergenic regions. These SNPs could affect complex diseases through their effect on gene expression quantitatively. Unlike nonsynonymous SNPs, SNPs capable of affecting gene expression may not be easily identified because gene regulatory elements could not be accurately defined in a complex gene regulation process. However, since the thermodynamics of RNA–RNA binding plays an essential role in miRNA interaction with target mRNA, it is expected that sequence variations such as SNPs at miRNA-binding sites may affect the expression of miRNA targets. The rationale for this assumption is based on the principle of miRNA–target interaction. Accumulating evidence revealed that 7 nt at the 5′-terminus of miRNAs from position 2 to position 8, called ‘seed’ region, are essential for their function (). Based on these discoveries, several computational methods have been developed to predict miRNA targets (). Most of these methods have been biologically validated and proved to be very efficient and accurate. For example, up to 90% of the randomly selected miRNA targets predicted by Krek . () have been proved to be true targets (,). The accuracy of these methods has also been proved by gene expression profiling studies (,). These methods have yielded a large number of candidate targets in both plants and animals. The estimated human miRNA targets can account for up to one third of the human genes (,,). Consequently, a SNP located in the miRNA-binding site of a miRNA target (called miRNA-binding SNP in this study) is likely to disrupt miRNA–target interaction, resulting in the deregulation of target gene expression. In this regard, the effect of this type of SNPs on gene expression is predictable. Such SNP-associated deregulation of the expression of an oncogene or tumor suppressor might contribute to tumorigenesis. In this study, we conducted a genome-wide search for SNPs located in miRNA-binding sites of miRNA targets using the dbSNP database and comprehensively defined the display of each SNP in cancers versus normal tissues through mining the dbEST database. We thus identified a number of miRNA-binding SNPs with apparent cancer-associated aberrant allele frequencies, and confirmed through genotyping that some of these SNPs are indeed aberrantly present in tumors. A 3′-UTR dataset and a miRNA target dataset of human genes were obtained from UCSC Genome browser (). The miRNA target dataset, developed by Krek (), contains the human genes, which, at 3′-UTR, have a 7-nt segment (called miRNA-binding ‘seed’ region here) complementary to the ‘seed’ region (the 7 nt from position 2 to position 8 at the 5′-terminus) of human miRNAs. A human SNP dataset (NCBI dbSNP Build 126) was obtained from NCBI databases (). The genomic locations (chromosome number and nucleotide position) on human chromosome of the SNPs, the 3′-UTRs and the miRNA-binding ‘seed’ regions are all indicated in the individual datasets. The SNPs located in the 3′-UTRs and the SNPs located in the miRNA-binding ‘seed’ regions of human genes were identified using the chromosomal location information. Human expressed sequence tag (EST) libraries and EST sequences were obtained from NCBI databases (). The EST libraries were manually curated and cataloged into cancer EST libraries and normal tissue EST libraries. Totally, 2.2 millions of EST sequences were obtained from 3721 cancer EST libraries and 1.9 millions of EST sequences from 2010 normal tissue EST libraries. MiRNA targets predicted by Krek . () were obtained. The SNPs located in the miRNA-binding regions (∼30-nt long), called miRNA-binding SNPs, were identified by blast searching for the dbSNP database using the miRNA target sequences. The EST fragments representing each allele of miRNA-binding SNPs were then identified by blast-searching for the cancer EST libraries and normal tissue EST libraries respectively, using a 30-nt sequence surrounding the SNP site. The sequences for both alleles were used separately for searching. Only the sequences with 100% identity were picked up. For each SNP, the total number of ESTs corresponding to each allele identified from the cancer EST libraries and from normal tissue EST libraries were counted respectively and compared with the number found in the dbSNP database. Fisher's exact test was used to determine the significant difference of the allele frequency of each SNP by comparing the cancer EST libraries with that of the dbSNP database. The allelic distributions of some miRNA-binding SNPs were found in SNP500Cancer database () that contains the allele frequency of the SNPs of cancer-related genes in the four populations (). We also performed Fisher's exact test in order to compare the allele frequencies found in cancer EST libraries and each population in the SNP500Cancer database. About 200 tumor tissue specimens from Caucasian patients with various cancers were obtained from the Cooperative Human Tissue Network (CHTN) in the USA. Genomic DNAs were extracted from freshly frozen specimens using DNeasy Tissue kit following the manufacturer's protocol (Qiagen). As controls, 1000 genomic DNAs of normal subjects were obtained from the British 1958 birth cohort that is based on all persons born in Britain during one week in 1958, and additional 200 genomic DNAs (Caucasian) were from Coriell Institute for Medical Research. Collection and use of the tissue and genomic DNA samples were approved by the National Research Council Canada. For SNP allele identification, ∼300 bp DNA fragments flanking the SNP of interest were amplified by PCR using the genomic DNAs. The PCR products were purified using MinElute 96 UF plates (Qiagen) and subjected to genotyping using one of the following methods. If one of the two alleles of a SNP can be digested by a specific restriction enzyme, a restriction enzyme digestion method was used. Otherwise, DNA sequencing method was used. For restriction enzyme digestion method, the digested PCR products were analyzed by agarose gel electrophoresis, which can distinguish the digested and undigested DNA fragments, for calculation of allele frequencies. Although hundreds of miRNAs have been identified, only a few of them have been functionally characterized, thus the biological functions of miRNAs are largely unknown. If the functions of miRNAs are, as assumed currently, critical for basic biological processes, we expect that the sequence variations, such as SNPs, in miRNA-binding sites of miRNA targets should undergo a purifying selection during evolution. To this end, we downloaded the miRNA targets predicted by Krek . (). Each of the predicted miRNA targets contains at least one 7-nt segment (called ‘seed’ region here) at the 3′-UTRs, which is crucial for the recognition of miRNAs. The SNPs located in the ‘seed’ regions are most likely to affect miRNA–target interaction and thus target expression. We therefore searched for the SNPs located in the ‘seed’ regions of human miRNA targets using the dbSNP database (). For comparison, we also searched for the SNPs that are located in the whole 3′-UTRs of human genes in the same database. We counted the density of the SNPs found in the miRNA-binding ‘seed’ regions (7 nt/seed) versus the whole 3′-UTRs. As shown in , 265 SNPs from 1 400 000 nucleotides, namely 0.182 SNP/kb, were located in the miRNA-binding ‘seed’ regions of 200 000 miRNA targets, whereas 20 588 SNPs from 96 484 523 nucleotides (the total number of nucleotides at the 3′-UTRs of human genes obtained for this study), namely 0.213 SNP/kb, were identified in the whole 3′-UTRs of human genes, indicating that SNPs arise less frequently in the miRNA-binding sites of miRNA targets than in the entire 3′-UTR ( < 0.022). This result suggests that SNPs at the miRNA-binding ‘seed’ regions are negatively selected under evolutional pressure. Since purifying selection (negative selection) eliminates the mutations that have deleterious effects on function, the relative low density of miRNA-binding SNPs at the 3′-UTRs of human genes supports the important role of miRNA–target interaction. Since SNPs are mutations that occur throughout evolution, natural selection should limit the alleles that have deleterious effects on function with time and thus limit the frequency of these harmful alleles. As shown in , we found that more than 70% of the miRNA-binding SNPs have a minor allele frequency less than 0.5% in the dbSNP database. This level is higher than the average level of all SNPs found in the same dbSNP database (∼10 and 25% observed in the HapMap and ENCODE datasets, respectively) and the expected distribution under the standard neutral model (∼50%). This result further confirms the important biological function of miRNAs. To further confirm the essential biological function of miRNAs as elucidated above by the SNP density analysis, we determined the expression variation of different alleles of miRNA-binding SNPs via mining the dbEST database. Because the EST libraries were constructed from cDNAs, the relative frequency of the two alleles for each SNP found in the dbEST database should be similar to that found in the dbSNP database, if the SNP does not affect gene expression. However, if a SNP affects gene expression, the two alleles of this SNP are likely differentially expressed and thus the relative frequency of the two alleles present in the dbEST database might be different from that found in the dbSNP database. Since the 7-nt ‘seed’ region of a miRNA is the most important sequence for miRNA–target interaction and miRNA function, further analysis was conducted by focusing on these regions. For convenience, we termed the allele of a miRNA-binding SNP that has the sequence complementary to the ‘seed’ region of the miRNA as ‘target allele’ and the other with one mismatch as ‘non-target allele’. Among the 930 miRNA-binding SNPs we identified, 297 are located in the miRNA-binding ‘seed’ regions. Half of these 297 SNPs have allele frequency data available in the dbSNP database. We then calculated the relative frequency of the target allele and the non-target allele for each of these 129 SNPs in the dbEST database and in the dbSNP database, respectively. We found that 35 out of the 129 SNPs have a zero frequency of non-target allele in both databases, indicating that the non-target alleles (the minor alleles) of these SNPs are too rare to be detected. Therefore, these 35 were not used for comparison. However, 59 (63%) of the remaining 94 have a higher frequency of non-target alleles in the dbEST database than in the dbSNP database, whereas only 35 (37%) of them have a higher frequency of non-target alleles in the dbSNP database than in the dbEST database (). More importantly, the average frequency ratio of the non-target alleles over target alleles for these SNPs is significantly higher in the dbEST database than in the dbSNP database (0.127 versus 0.086) (), suggesting that the average expression level of the non-target alleles of miRNA-binding SNPs is higher than that of the target alleles. These results indicate that the SNPs/mutations located in the miRNA-binding ‘seed’ regions of miRNA targets can disrupt miRNA–target interaction and lead to up-regulation of miRNA target expression. To identify the miRNA-binding SNPs that may contribute to cancer susceptibility, we carried out a genome-wide scale analysis of the dbSNP database in conjunction with the human dbEST database. To do so, we downloaded the human EST sequences derived from both the cancer EST libraries and the normal tissue EST libraries. In total, 2.2 million EST sequences from 3721 human cancer EST libraries and 1.9 million EST sequences from 2010 human normal tissue EST libraries were obtained. We then searched for the ESTs that correspond to each allele of those SNPs located in the miRNA-binding regions of miRNA targets in the cancer EST libraries and in the normal tissue EST libraries (Table S1). The miRNA targets were obtained from Krek .'s report (). A 30-nt segment for each miRNA-binding site was entered in the original database and used for SNP searching in this study. The allele frequency of each SNP was also obtained from the dbSNP database (Table S1). To determine whether the allele frequency of these SNPs in the cancer EST libraries is similar to or different from that found in the general population, we counted the number of ESTs for each SNP allele found in cancer EST libraries and compared it with that of the general population as found in the dbSNP database. Fisher's exact test was used to determine the statistical significance of the variations observed (). The SNPs above the horizontal line of the figure have a -value less than 0.01 (−log (-value) > 2, ). The analysis of the dbEST database and dbSNP database provided us with some potential candidates for miRNA-binding SNPs with an aberrant allele frequency present in the human cancer EST database and filtered out many SNPs (the majority) that have similar allele distribution in both databases. We next genotyped genomic DNAs derived from human cancer tissues in order to experimentally validate these potential miRNA-binding SNPs. We compared the allele frequency of each SNP found in cancer tissues with that found in the dbSNP database and in the normal subjects. From 65 SNPs tested, we found that 12 have a significantly aberrant SNP allele frequency ( < 0.05) in human cancer tissues as compared with the control present in the dbSNP database ( and Table S2). The aberrant allele frequencies of these SNPs were also found in unaffected tissues, located near these cancer tissues examined, suggesting their germ-line origin. shows the representative sequences of SNP rs16917496 derived from two colon tumor samples. To determine if the number of SNPs (12 out of 65) that have an aberrant allele frequency in cancer samples compared to the dbSNP database is statistically significant, we did genotyping analysis of 16 of the 65 SNPs using normal subjects. We found that none of the 16 SNPs have a significant difference of allele frequency in the normal subjects from that in the dbSNP database ( > 0.1 for all of them, Table S2), whereas 2 of the 16 SNPs have an aberrant allele frequency in cancer population (Table S2, rs1044129 and rs17107469). These results suggest that the analysis method provides us with useful preliminary data for the identification of miRNA-binding SNPs that may affect miRNA target expression, possibly leading to cancer susceptibility. These SNPs with aberrant allele frequencies in cancer can be used as potential tumor markers and drug targets for early cancer detection and prevention. Under the neutral theory of molecular evolution, the majority of DNA variations observed in a population are due to random drift of neutral or nearly neutral mutations. Natural selection, such as purifying selection, may eliminate those mutations that have deleterious effects on function. This will reduce the ratio of those mutations over neutral mutations observed in the present population. In this study, we first found that SNP density is lower in the miRNA-binding motifs (‘seed’ regions) than in the 3′-UTRs, potentially caused by purifying selection. In addition, since SNPs are the result of mutations that occurred one time in human history, natural selection may also lead to rare allele frequencies of the mutations that could not be completely eliminated under rapid population expansion, despite their deleterious effects on function. Indeed, we found that the frequencies of the minor alleles (non-target alleles) of the miRNA-binding SNPs are extremely low (0.079) based on the dbSNP database analysis, which may reflect the deleterious effect of nucleotide substitutions at miRNA-binding sites. Taken together, both the negative selection (lower density) and the rare allele frequencies of miRNA-binding SNPs at the miRNA-binding sites, reflect the importance of miRNAs. Rare SNP alleles may have severe consequences and thus cause various human diseases. By mining the dbEST database and dbSNP database, we identified a number of miRNA-binding SNPs that have aberrant allele frequency in the cancer EST libraries. However, using EST libraries for this analysis has some limitations caused by the quality of the EST sequences, biased sampling and biased ethnic origin. To reduce the effect of these limitations on the accuracy of the results, we performed further analysis. First, we manually searched for these interesting miRNA target genes that have an aberrant allele frequency in the cancer EST libraries compared to that in the dbSNP database, as identified from the screen. We manually detected the quality of each EST sequence to ensure that only the ESTs with high quality are used for our final statistical analysis of these interesting SNPs. In addition, to determine the effect of EST quality on the results, we searched for the EST sequences that contain any two other nucleotides (‘non-SNP’ nucleotides) to replace the defined SNP alleles at the SNP position. We assumed that any EST sequence with the substitution of two other ‘non-SNP’ nucleotides was caused by sequencing errors and that the number of sequencing errors causing change of one SNP allele to the other SNP allele is similar to the number of errors leading to the change of a SNP allele to a ‘non-SNP’ allele (non-existing allele). Using these data, we can calculate the contribution of EST sequence quality to the errors of SNP allele distribution. We found that among 18 845 ESTs located in the two SNP alleles, 24 ESTs were substituted by the ‘non-SNP’ nucleotides, revealing a 0.13% error in our analysis (0.65% for the minor alleles and 0.07% for the major alleles, ). To reduce sampling bias, we counted only one EST for a specific allele of a SNP if more than one EST for the allele were found in the same library. However, many different EST libraries might be constructed with the cDNAs derived from the same donor. EST sequences representing a specific SNP allele found in different libraries derived from the same donor may be counted several times although only one count should be used for statistical analysis. We could not avoid this kind of sampling bias since the identity of donors for the EST library was not available. For the same reason, it is hard to exclude the effect of bias of ethnic origin. Nevertheless, to reduce the limitation caused by the lack of information of the ethnic origin of donors in our statistical analysis, we first compared the allele frequency found in the cancer EST database with that of the dbSNP database from all four populations and also with that from each individual population, if data from different populations was available. In addition, we used the SNP500Cancer database where the SNP allele frequency data from all four populations are available (Table S3). In spite of these limitations, we assessed the efficacy of analysis through experimental studies using human cancer samples. We successfully confirmed by experimental validation 12 out of 65 miRNA-binding SNPs with an aberrant allele frequency in the cancer EST libraries. These results indicate that the aberrant allele frequencies of the miRNA-binding SNPs present in tumors, might be an important factor contributing to tumorigenesis. In addition, these results demonstrated the usefulness of the primary analysis, as it was able to filter out the majority of the miRNA-binding SNPs that showed no difference of allele frequencies between the cancer EST database and the dbSNP database. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Eukaryotic genomic DNA is organized into repeating arrays of nucleosomes that are the structural units of chromatin (). The nucleosome is also known to play a central role in regulating gene transcription from promoters (). For more than 30 years, micrococcal nuclease (MNase), DNase I and restriction endonucleases have been essential tools for chromatin research (). MNase in particular has been the gold standard for characterizing nucleosomal structures in chromatin, because of its relative high specificity toward the internucleosomal linker region. However, this enzyme also digests DNA within nucleosome core particles, attacks single-stranded nucleic acids, possesses exonuclease activity and has significant DNA sequence cleavage preferences (). One of the hallmarks of the terminal stages of programmed cell death or apoptosis is internucleosomal DNA breakdown (). The endonuclease primarily responsible for mediating apoptotic nucleosomal DNA laddering is DNA fragmentation factor-40 (DFF40), also called caspase-activated deoxyribonuclease (CAD) (). DFF is a heterodimer in its inactive form, composed of a 45-kDa inhibitor and chaperone subunit (DFF45), also called inhibitor of CAD (ICAD), and a 40-kDa latent endonuclease subunit (DFF40/CAD) (). This protein complex possesses nuclear localization sequences and resides within the cell nucleus (,). Caspase-3 or -7 cleavage of DFF specifically cuts only DFF45, releasing DFF40 from the complex, which in turn forms enzymatically active homo-oligomers (,,). During the course of our studies on characterizing the enzymatic properties of DFF40 on chromatin substrates, we recognized its potential as an excellent reagent for chromatin research because of its exquisite selectivity for linker region cleavage, its exclusive generation of double-stranded breaks and lack of both exonuclease activity and attack within nucleosome core particles (,,). Indeed, evolutionary pressure to create ‘bite-size’ pieces of chromatin for phagocytotic clearance of apoptotic products appears to have optimized DFF40 to specifically attack the linker regions in polynucleosomal chains. However, DFF's nuclease dependency on caspase-3, which has numerous other cellular protein substrates (), in combination with the protease's high cost, limits DFF's potential for chromatin studies. We therefore decided to engineer DFF nuclease activity to be dependent on TEVP cleavage (DFF-T) because this protease has a seven amino acid recognition cleavage sequence instead of the less stringent four amino acid target sequence of caspase-3 (,). Importantly, caspase-3 expression in yeast is lethal, but expression of the highly sequence-specific TEVP is harmless (). Thus, TEVP expression should not cause any artificial alterations in chromatin structure and gene expression. We demonstrate here the successful expression of recombinant forms of these proteins in either or and that after TEVP cleavage highly active DFF endonuclease is generated. We further demonstrate that DFF-T is an excellent reagent for mapping nucleosome positions and hypersensitive sites in specific genes as revealed by chromatin footprinting of the well-studied gene, whose promoter and upstream region displaces four nucleosomes upon transcriptional induction via chromatin remodeling and binding of Pho4p (). The TEVP cut site (), encoding the amino acid sequence ENLYFQS, was inserted between amino acid positions 117–118 and 224–225, immediately downstream of each of the two caspase-3 cleavage sites in the coding sequence of mouse DFF45. We created all three possible combinations of these insertion mutations, termed IWT, WTI and II. Similarly, we created the corresponding II human DFF45 species, and also substituted the caspase-3 cleavage sites with TEVP cut sites in all combinations in mouse DFF45. Using the two-plasmid expression system for DFF () kindly provided by Gregor Meiss, the individual modified forms of mouse DFF45 along with the corresponding wild-type control were co-expressed with mouse GST-DFF40. The resulting GST-fusion protein-containing complexes were purified on GSH-Sepharose. Human DFF species were cloned into pRSFDuet™ for co-expression in (Novagen) and purification by nickel chromatography. Recombinant caspase-3 was prepared as previously described (), and TEVP was purchased from Invitrogen. One microgram of either wild-type, swap or insertion mutated DFF40/DFF45 heterodimer were incubated with caspase-3 or 0.5 unit of TEVP at 37°C in reaction buffer consisting of 10 mM KCl, 100 mM NaCl, 1.5 mM MgCl, 1 mM EGTA, 1 mM dithiothreitol and 20 mM Tris–Cl, pH 7.5. After 20 min of incubation, samples were separated on 12% polyacrylamide–SDS gels, and then proteins were electrophoretically transferred onto nitrocellulose membranes. Membrane-immoblilized proteins were probed with the following commercial antibodies: rabbit anti-human DFF40 polyclonal antibodies and/or rabbit anti-human DFF45 N-terminus polyclonal antibodies (Pharmingen). The antigen–antibody complexes were visualized using enhanced chemiluminescence (ECL) western blotting detection reagents (Amersham Biosciences). For the endonuclease activity assay, 1 μg of naked plasmid DNA was incubated for 30 min at 37°C with DFF nuclease species pre-incubated with appropriate protease, as indicated in the legends of . Aliquots of the endonuclease reaction were stopped by gel loading buffer containing 0.6% SDS, 50 mM EDTA, 30% glycerol and samples were then separated on 1.5% agarose gels using TAE as the running buffer. After electrophoresis, DNA was stained with ethidium bromide, and gels were scanned with a FluorImager (Molecular Dynamics Inc., Sunnyvale, CA, USA). The TEVP gene's coding region with SV40 nuclear localization signals added to its N- and C-termini, under the control of the promoter, and cloned into a integration vector (), which was kindly provided by Kim Nasmyth and used to stably transform the yeast P2 strain (). DFF40 and wild-type DFF45 or the other three insertion mutants, IWT, WTI and II, were cloned into pESC-His dual yeast expression vector (Novagen) containing two divergently orientated galactose inducible and promoters. The pESC-DFF vectors were introduced into yeast P2 strains containing or lacking integrated TEVP via standard lithium acetate transformation and selection on synthetic complete medium lacking histidine and tryptophan, supplemented with 2% dextrose. To assay the effects of TEVP and DFF expression on cell viability, exponentially growing cultures of individual transformants were serially diluted 10-fold in 10 mM Tris–HCl (pH 8.0), 1 mM EDTA and 5 μl aliquots were spotted onto selective media supplemented with 2% raffinose and 2% galactose. The number of viable cells forming colonies at 30°C was determined. For endonuclease activity assay, yeast genomic DNA was isolated and analyzed. Briefly, 10 ml cultures of yeast cells transformed as indicated in A, were harvested by centrifugation following galactose induction for 6 h. The cells were washed with 1 ml of distilled HO, re-suspended in 200 μl of 2% Triton X-100, 1% SDS, 100 mM NaCl, 10 mM Tris–Cl, pH 8.0, 1 mM EDTA, pH 8.0, vortexed at highest speed for 3 min with 200 μl glass beads and 200 μl phenol/chloroform/isoamyl alcohol, followed by the addition of 200 μl TE buffer and centrifugation for 5 min. The genomic DNA in the aqueous layer was ethanol precipitated and re-suspended and treated with 100 μg/ml RNase for 20 min at 37°C. The DNA was phenol extracted, ethanol precipitated and re-suspended in 50 μl TE buffer. The integrity of 5 μg DNA samples was assessed by agarose gel electrophoresis. We took advantage of the ‘Center for Invention’ resource at UT Southwestern to generate a computer-assisted prediction for the design of synthetic genes and their robotic synthesis from pairs of overlapping 40-mer oligonucleotides (). We have created, through this core, synthetic genes for both mouse and human forms of DFF40 and DFF45. After ligation of oligonucleotides, we amplified the products with high-fidelity PCR in preparation for cloning and sequencing. Any mutations that occurred were cured by additional high-fidelity PCR reactions with appropriate primers and repeated subcloning and sequencing. We have designed these synthetic genes to possess hexa-his tags at the C-terminal ends of DFF40 species, to either lack or possess the TEVP-cut sites as described below (sequences will be made available upon request), as well as the appropriate restriction sites for cloning into pRSFDuet™ expression vectors (Novagen). BL21(DE3) cells transformed with this plasmid were induced with 0.5 mM IPTG for 3 h when the optical density at a wavelength of 600 nm (OD) was 0.5. Cells were lysed by thawing in wash buffer solution (WB; 50 mM NaHPO, 300 mM NaCl containing lysozyme (1 mg/ml), RNase A (10 μg/ml) and DNaseI (5 μg/ml). Lysates were clarified by centrifugation and allowed to bind to nickel affinity resin (Ni-NTA superflow, Qiagen) for 1 h at 4°C with gentle mixing. After washing with WB, the protein was eluted with elution buffer solution (10 mM NaHPO, 100 mM NaCl, pH 8.0, 250 mM imidazole). strain W303 was grown over night at 30°C in yeast nitrogen base phosphate-free medium. Cells were centrifuged, suspended in fresh medium, either phosphate-free or supplemented with 14 mM KHPO and grown for an additional 9 h. Cells were centrifuged and washed with water. One g of wet cell pellet was suspended in 2 ml of yeast lysis buffer (1 M sorbitol, 5 mM β-mecaptoethanol supplemented with 2 mg/ml of Zymolase 100T, ICN) and incubated at room temperature for 3–4 min with occasional agitation in 2-ml round-bottom microcentrifuge tubes. Cells were centrifuged for 15 s, washed in 1 M sorbitol, then suspended in 4 ml of digestion buffer (1 M sorbitol, 50 mM NaCl, 10 mM Tris–Cl pH 7.6, 5 mM MgCl, 1 mM CaCl, 1 mM β-mecaptoethanol, 0.5 mM spermidine, 0.075% NP-40) supplemented with 2 μg/ml of RNaseA and incubated for 15 min at room temperature (DFF is inhibited by RNA) (). Aliquots measuring 400 μl of the resulting suspension were mixed with 100 μl of MNase (Worthington, 1 or 2 U) in digestion buffer. Alternatively, 400 μl aliquots of the resulting suspension were diluted with 400 μl of digestion buffer and mixed with 500 μl solution of TEVP-activated human DFF-T (100 or 200 pmol) in digestion buffer (100 pmol of DFF-T was pre-incubated with 50 U of AcTEV protease for 15 min at room temperature). Digestion mixtures were incubated for 10 min at 33°C, and then reactions were stopped by adding 1/5 vol of stop solution (2% SDS, 100 mM EDTA and 2 mg/ml proteinase K). Mixtures were incubated for 6 h at 50°C and for additional 4 h at 65°C, then phenol/chlorophorm extracted and isopropanol precipitated. DNA was purified by routine RNaseA treatment, phenol/chlorophorm extractions and isopropanol precipitation. DNA concentration was assessed by fluorometry; usually ∼25 μg of DNA was recovered from 0.1 g of wet yeast cell pellet. Ten micrograms of purified DNA was digested with ApaI restriction enzyme (50 U for 15 h at 28°C), electrophorectically separated on a 1.6% agarose gel and transferred onto Zeta-probe (Bio-Rad) nylon membrane using alkali (0.4 M NaOH, 1.5 M NaCl). The membrane was pre-hybridized for 4 h at 64°C with 0.5 M sodium phosphate (pH 7.2), 7% SDS, 1 mM EDTA and then hybridized over night in the same buffer supplemented with 0.1 mg/ml of denatured calf thymus DNA and 10 ng/ml of P-labeled PCR-amplified probe, which was a 200-bp fragment adjacent to the ApaI site generated with the following primers 5′-GTC TTC AGC GTC AAC TTT AG-3′ and 5′-TAA CGA TGA TGG TCC CTT AA-3′ (). In an attempt to convert mouse DFF's nuclease activity to be dependent on TEVP cleavage, we first swapped one or both of the caspase-3 cleavage sites within DFF45 with a TEVP recognition sequence (A). After co-expression of the different recombinant forms of DFF in and purification of the heterodimers, we separately assayed the cleavage pattern of DFF45 mutant forms by either caspase-3 or TEVP (or both) via western blotting, and the activity of DFF40 nuclease by digestion of a plasmid DNA substrate. As shown in B, the various forms of DFF45 were cleaved as expected by the appropriate proteases. However, in spite of the fact that the double swap mutant was effectively cleaved by TEVP (B, lane 12), DFF40 nuclease activity was only weakly generated compared with the wild-type control after cleavage by caspase-3 ± TEVP (C, compare lanes 8 and 12 with 9). Interestingly, the SWT but not the WTS mutant was also fully activated after cleavage by both proteases (C, compare lanes 10 and 11). It therefore appears as if maintenance of the second caspase-3 cleavage site is most important for maintaining DFF45 chaperone function. We conclude that the swapped amino acid sequence at caspase-3 cleavage sites within DFF45 apparently plays a chaperone role in folding DFF40 into a potential nuclease capable of activation. Because the caspase-3 sites within DFF45 are important for its chaperone activity, we created another series of DFF45 mutants in which the TEVP cleavage sequence was inserted immediately downstream of the caspase-3 cleavage sites (D). Surprisingly, even though the caspase-3 recognition sequences DEPD and DAVD in mouse DFF45 were not altered by the adjacent insertions of the TEVP sites, the corresponding caspase-3 sites are no longer cleavable by that enzyme, as demonstrated by the maintenance of intact DFF45 II mutant protein after caspase-3 treatment, as revealed by western blotting (E, lane 8), and by the inability of caspase-3 alone to activate the nuclease (F, lanes 2–4). We conclude that the ability of caspase-3 to cleave at these sites is dependent on the amino acid sequence context and that the additional seven amino acids may affect the spatial properties of the caspase-3 cleavage sites. As shown in F (lane 8), only the heterodimer containing the double-TEVP-site-insertion mutant II could be activated by TEVP treatment alone to a specific activity nearly equal to that of the caspase-3 ± TEVP treated wild-type enzyme (F, lanes 1 and 9). We conclude that a novel modified form of mouse DFF nuclease (DFF-T) has been successfully generated, expressed and purified, whose activation is specifically under the control of TEVP. We intended to develop a highly reproducible system in yeast for the regulated expression and activation of DFF-T for the purposes of footprinting of the positions of nucleosomes and transcription factors on specific genes. Previously, Simpson and Wang () were successful in developing a DNase I expression system in yeast for footprinting experiments by putting the gene into a high copy shuttle expression vector under the control of the promoter. Here we have taken advantage of a system developed by Nasmyth and co-workers (), who engineered the TEVP gene's coding region by adding SV40 nuclear localization signals to its N- and C-termini, which was placed under the control of the promoter, and integrated into the locus. We therefore created a yeast strain with a galactose-inducible TEVP gene. We next inserted various engineered forms of mouse DFF45 and mouse DFF40 genes into the bicistronic, high copy number shuttle expression vector pESC-HIS (Stratagene), which divergently expresses introduced sequences from promoters, and transformed yeast strains that either lacked or possessed the TEVP gene under GAL control with these expression vectors, and optimized expression in this system by titration with different concentrations 3-amino-1,2,4-triazole. As shown by the plating assay in A, expression of TEVP together with various engineered forms of mouse DFF results in cell death only when the DFF45 II mutant and DFF40 are co-expressed together with TEVP. It is significant that this result is in agreement with the activation data for the corresponding recombinant proteins expressed in (F above). Furthermore, yeast DNA undergoes nucleosomal laddering in a galactose-dependent fashion, only in the strain harboring the genes encoding the DFF45 II mutant, DFF40 and TEVP (B, lane 8). We conclude that this form of DFF expression could be useful as a genomics tool to study yeast chromatin structure . It should be noted, however, just as in the case of the DNase I yeast expression system developed by Simpson and Wang (), there is . 6-h time lag after GAL induction before significant DNA breakdown occurs, and it would be most desirable to be able to much more rapidly activate the nuclease to obtain a snapshot of the chromatin structure. This might be achieved by the high-level constitutive expression of codon-optimized DFF-T prior to GAL induction of TEVP. In addition, it may be possible to engineer TEVP for rapid regulated transport from the cytoplasm to the nucleus. A further limitation of DFF, particularly in the yeast system, is that these cells have a very high RNA/DNA ratio and RNA is an inhibitor of the enzyme (). Many codons in DFF cDNAs are rare with respect to their cognate tRNA abundances in , and only a few micrograms of recombinant proteins are routinely obtained upon expression of recombinant forms of DFF in this organism per liter of culture. To provide a more robust source of DFF proteins to the scientific community we have created codon-optimized synthetic genes for wild-type and II mutants of the mouse and human DFF proteins through their robotic synthesis from pairs of overlapping 40-mer oligonucleotides (). After expression of the proteins encoded by these synthetic genes in the pETDuet™ vector (Novagen), more than 20% of the total protein corresponds to DFF bands in Coomassie blue-stained SDS–PAGE gels (A). This level of expression is highly significant because our previous expression systems using the corresponding wild-type cDNA coding sequences gave no visible new bands after induction in such gels of total cell protein (data not shown). Furthermore, DFF nuclease activities produced from these synthetic genes were fully dependent on cleavage by the appropriate protease, and both wild-type and insertion mutant proteins have similar nuclease activities (B). We have demonstrated here that DFF-T could be fully and specifically activated by TEVP both and . To further investigate the utility of DFF-T we mapped nucleosome positions and hypersensitive sites within the promoter region of gene, one of the most thoroughly studied yeast genes with respect to chromatin structure (). Upon induction of gene transcription by phosphate starvation, four nucleosomes are displaced from the upstream promoter region through the binding of Pho4p to UASp1 and UASp2 and the recruitment of chromatin remodeling complexes (, right diagram). To evaluate the effectiveness of DFF-T to detect hypersensitive sites, and both nucleosome positioning and displacement, we made a direct comparison on the same chromatin samples with MNase cutting patterns by indirect end-labeling analysis. Here we utilized human DFF-T produced in high-yield from codon-optimized synthetic genes in an expression system (). This comparison reveals a striking degree of similarity with respect to: (i) the positions of hypersensitive sites HS2 and HS3 on the uninduced gene, which correspond to the accessible UASp1 between nucleosomes −2 and −3, and the 3′ end of the gene, respectively (,); (ii) the positions of nucleosome footprints (, compare lanes 1 and 4) and (iii) in the displacement of nucleosomes −1 to −4 after gene induction (, compare lanes 2 and 5). The partial occupancy by nucleosomes at the −1 position of the activated promoter may be attributed to a steady state of disassembly and reassembly of nucleosomes as previously described (). There is some subtle difference, however, in the hypersensitive sites in the HS1 doublet (, compare lanes 1 and 2 with lanes 4 and 5), which corresponds to the upstream region of the gene. In summary, this comparison allows us to conclude that DFF-T is an excellent reagent to map the chromatin structures associated with specific inactive or active genes. We speculate that this enzyme may detect chromatin structures missed by other conventional enzymes. In this study, we have shown that DFF-T is an excellent reagent for chromatin structure investigations. The enzyme should also prove to be valuable in other types of footprinting experiments. From our previous and current studies, we also know DFF is extremely suitable for fragmenting chromatin to prepare nucleosomes without them possessing internal DNA nicks. We also propose that DFF nuclease is ideally suited to serve as a substitute for sonication to shear chromatin to nucleosome-sized fragments for the chromatin immunoprecipitation (ChIP) technique, because it only cuts between and not within nucleosomes and thus it cannot over-digest the chromatin DNA as other nucleases can. This nuclease should also be ideal for the purpose of generating genomic libraries for gene cloning/subcloning and shotgun sequencing, as it creates nearly exclusively double-stranded blunt DNA ends, and leaves 5′-phosphate and 3′-hydroxyl groups that are ideal for blunt-end cloning by ligation with T4 DNA ligase. Controlled digestion of DNA to yield fragments of various sizes should be easy to achieve with this enzyme. Thus, DFF-T should be a valuable tool for several lines of investigation.

Nucleic acid–protein complexes based on specific interactions have recently been the matter of a great number of contributions in analytical biochemistry (). To study these interactions, the purification of nucleic acid–protein complexes is generally carried over by mixing standard chromatography techniques and specific affinity methodologies. Often, the preliminary chromatographical steps are conventional and aim at removing entire classes of undesirable analytes. For example, Yaneva and Tempst () used a first phosphocellulose fractionation of nuclear extracts to eliminate negatively charged molecular species (saccharides, proteins and nucleic acids). The remaining chromatographic steps are based on the specific or non-specific interaction of proteins with oligonucleotidic target sequences (,). This sequence of chromatographical steps makes the purification of nucleic acids interacting proteins time-consuming and very large amounts of the initial sample might be needed. Affinity chromatography procedures most often involve the coupling of an appropriate specificity determinant molecule (an oligonucleotide bearing a specific target sequence, for example) to a chromatographic support (CNBr-activated or streptavidin-coated agarose beads, for example) in order to craft an affinity chromatography resin. Following washes of the resin, the retained molecules are eluted directly from the chromatographic phase. These methodologies were used to set up so-called one-step purification procedures (,). The authors prepared their affinity chromatography phase with biotinylated oligonucleotides bearing the target sequence that were used to functionalize streptavidin-coated beads. Because the chromatographic support (agarose or polyacrylamide beads, for example) is of huge dimensions with respect to the affinity determinant (the oligonucleotide), it lends itself favourably to non-specific interactions with the analytes in the sample thus leading to high contamination levels upon elution of the analytes of interest by applying either salts or detergents (or both) onto the whole chromatographic phase. The analytes of interest are thus less well purified. In order to limit this adverse effect, we reasoned along with others () that the uncoupling of the affinity determinant (along with potential bound molecular species) from the chromatographic support itself would yield much more useful purifications of analytes present in the initial sample at very low concentrations. A number of systems have been devised in order to allow the uncoupling of the DNA–protein assemblies from the chromatographic support according to biologically compatible mechanisms. Shimkus () chose a disulfide bond-containing linker to couple the oligonucleotidic target sequence to a biotin moiety that was later attached to streptavidin-coated agarose beads. The uncoupling of the nucleic acid–protein assembly from the chromatographic support was triggered by incubating the chromatographic phase with a reducing agent. Bachler () and Hartmuth () devised a competitive elution strategy based on the use of aminoglycosidic antibiotic-coated chromatographic supports onto which an oligonucleotide linked to an aptamer specifically binding to the antibiotic was attached. The uncoupling of the oligonucleotide from the chromatographic support was achieved by adding excess amounts of the antibiotic which competed for the aptamer. Martinez () described a procedure by which the target nucleic acids (a double-strand RNA eicosamer) was linked to a biotin via a photocleavable linker. The chromatographic phase was prepared by attaching the biotin-conjugated target RNA duplex to modified avidin-coated beads. Upon UV irradiation of the chromatographic phase the oligonucleotide is detached from the chromatographic support. Strategies based on the use of reducing agents or antibiotics have some drawbacks such as the addition of molecules in the purification medium that can disturb the subsequent analyses, which prompted us to opt for the system employing the photocleavable linker. In this report, we describe a simplification of this method, specifically aimed at generalizing it to any situation involving the affinity-based purification of DNA-interacting proteins. We simplified experimental conditions that allowed—without loosing any purification efficiency—the use of a single buffer composition throughout the whole purification process. Our bioanalytical conditions were validated by affinity purifying the tetracycline repressor protein (TetR) expressed in quite low amounts in eukaryotic cells () onto an oligonucleotide bearing its cognate sequence (TetO). The purification efficiency was unprecedented with a single contaminant protein being co-purified with the repressor protein. The robustness of our method was challenged with a highly complex DNA–protein system that has the advantage of being well studied: the DNA repair machinery which comprises a large number of proteins that assemble onto DNA damages (double strand breaks, in our case) in supramolecular structures (). We succeeded in purifying these protein assemblies: the purified components were identified as well-known DNA repair proteins. Our work shows that these proteins retained their enzymatic activities throughout the purification, as seen by monitoring DNA ligation products. Further, kinase activities, also monitored in our experiments, were found to be distributed distinctly either on the beads or on the purified DNA–protein complexes. These results showed the major benefits of the uncoupling of the purified DNA–protein assemblies from the beads as far as a detailed and accurate understanding of the biochemical regulatory mechanisms involved in the assembly/disassembly of DNA–protein complexes is concerned. T-Rex HeLa cells (Invitrogen, Cergy Pontoise, France) stably express TetR under the control of the human CMV promoter (). These cells were cultured to subconfluency in minimal essential medium (MEM; Invitrogen) supplemented with 10% fetal bovine serum and non-essential amino acids (Sigma, Lyon, France) with 100 U/ml penicillin and 100 μg/ml streptomycin. Cells were grown at 37°C in a 5% CO atmosphere. Cellular extracts were obtained according to Baron (). Briefly, cells were harvested in the following lysis buffer: 10 mM HEPES pH 7.2, 1.5 mM MgCl, 10 mM KCl, 0.5 mM dithiothreitol and 1 mM phenylmethylsulfonyl fluoride supplemented with the Complete Protease Inhibitor Cocktail (Roche Molecular Biochemicals, Mannheim, Germany) and subjected to freeze/thaw cycles to disrupt their membrane. Cell lysates were cleared by centrifugation at 20 000 g at 4°C for 30 min. The protein concentration in the resulting supernatant was determined according to the Bradford protein assay (Bio-Rad Laboratories, Marnes-la-Coquette, France), bovine serum albumin was used as the standard. The cellular extracts were brought to a final protein concentration of 10 mg/ml with lysis buffer. HeLa nuclear protein extracts were prepared according to Dignam () (Cilbiotech, Mons, Belgium). A photocleavable moiety was coupled to a biotin moiety to form the photocleavable biotin (PCB) linker which was conjugated to the oligonucleotide according to the chemistry described in (,). Briefly, it was covalently linked through its 5′phosphate end to the 1-(2 nitrophenyl)ethyl photo-reactive group that is itself bound to the biotin moiety via a 6-aminocaproic acid linker (A). It was synthesized by Eurogentec (Seraing, Belgium) using commercially available phosphoramidite derivatives. The PCB-TetO was a 58 bp double-strand DNA conjugated to the photocleavable biotin label sequence. The DNA sequence comprised the TetO sequence () and a sequence stretch that was added to the 5′P oligonucleotidic end thus increasing the distance between the TetO target sequence and the chromatographic support so as to diminish the steric hindrance that might hamper the interaction between the proteins of interest and the TetO sequence (). Anti-TetR antibody polyclonal rabbit was from Mobitec (Goettingen, Germany). Anti-DNA-PKcs (clone 18-2), anti-Ku70 (clone N3H10) and anti-Ku80 (clone 111) monoclonal mouse antibodies were from Neomarkers (Fremont, USA) and anti-PARP-1 antibody was from AbD Serotec (Cergy Saint-Christophe, France). Anti-XRCC4 polyclonal rabbit antibody was from Calbiochem (La Jolla, USA). The PCB-TetO was radiolabeled using the T4 polynucleotide kinase (New England Biolabs, Saint-Quentin-en-Yvelines, France) and [γ−P]ATP (GE Healthcare). Binding reactions were performed for 20 min at room temperature with 50 μg of extracts from T-Rex HeLa cells and 25 fmol of 5′P radiolabeled PCB-TetO in a final volume of 10 μl. Binding buffer was 10 mM Tris–HCl pH 7.5, 5% glycerol and 2 mg/ml sheared genomic DNA competitor (fish sperm DNA, Roche), with or without tetracycline (1 μg/ml). Complexes were resolved by non-denaturing electrophoresis on 5% polyacrylamide/0.13% bisacrylamide gels containing 5% glycerol running in buffer 50 mM Tris base, 45 mM boric acid and 0.5 mM EDTA at room temperature and at 15 V/cm (). The oligonucleotide was radiolabeled after the purification (see B) only if a western blot analysis was performed in parallel (see A). Total 1/20 of the purification product was radiolabeled and analyzed by EMSA. The quantification of radioactivity was performed using the Image Quant software (GE Healthcare). In order to purify the TetR protein from T-Rex HeLa cells, the PCB-TetO duplex was immobilized on streptavidin-coated magnetic beads (Roche) according to the protocol provided by the manufacturer. To bind 0.5 pmol of the PCB-TetO oligonucleotide onto the beads, 1.25 pmol of duplex was incubated with 42 μg of magnetic beads followed by two equilibration steps with 84 μl of buffer A containing the lysis buffer supplemented with one volume of 2× buffer B (20 mM Tris–HCl pH 7.5 and 10% glycerol). Incubation of 1 mg protein extract with the bait-coated beads was performed at room temperature for 20 min in buffer A supplemented with 2 mg/ml fish sperm DNA (Roche). Five rounds of washes were performed by alternatively sedimenting the resin with a magnet and resuspending it with buffer A. The resuspended beads (50 μl of buffer A) were irradiated 10 min with a UXM-200HO xenon-mercury lamp (λ > 300 nm, 90 mW/cm at 365 nm, 5% light transmittance at 295 nm) (Lot-Oriel, Palaiseau, France). The beads were pelleted with a magnet; the supernatant contained the released oligonucleotidic bait along with the proteins bound to it. The different purification fractions (10 μl) were analyzed by electrophoretic mobility shift assay (EMSA). In order to purify the proteins involved in DNA-end joining, 40 pmol of the PCB-TetO were immobilized onto the chromatographic support by mixing 100 pmol of duplex with 500 μg of magnetic beads. The chromatographic phase was equilibrated with 1000 μl of buffer C (40 mM HEPES-KOH pH 7.8, 5 mM MgCl, 60 mM KCl, 0.5 mM dithiothreitol, 0.4 mM EDTA, 3.4% glycerol and 0.01% Nonidet P-40 substitute) and was later incubated with 400 μg of HeLa nuclear protein extracts in the absence of DNA competitor at 30°C for 30 min in a final volume of 170 μl. After five cycles of washes with buffer C, the beads were resuspended in 500 μl of buffer C and were irradiated 10 min; the supernatant contained the oligonucleotide and its associated proteins. Protein mixtures (either proteins bound to the DNA or proteins bound to the chromatographic phase) from irradiated or non-irradiated samples were separated on acrylamide denaturing electrophoresis gels. The proteins bound to the beads (irradiated or non-irradiated) were stripped off with a denaturing buffer and the proteins in the supernatants were filtration-concentrated as above. Proteins from TetR or DNA repair complexes purified fractions were mixed with the 2× Tris–glycine SDS sample buffer (Novex, Invitrogen) supplemented with 10% β-mercaptoethanol or Laemmli buffer, respectively and were later separated on 14% acrylamide denaturing Tris–glycine electrophoresis gels (Novex, Invitrogen) or 8% SDS-PAGE gels, respectively. Electrophoresed proteins were then transferred from the gel onto a nitrocellulose membrane (Hybond ECL, GE Healthcare). Protein bands were visualized by incubation of the membrane with Sypro Ruby protein blot stain (Invitrogen; visualization device: Typhoon 9410 fluorescent scanner, GE Healthcare). The same membrane was used for western blotting, revealed using the ECL Plus kit (GE Healthcare) and imaged using the same Typhoon 9410 apparatus. The BenchMark Protein ladder (Invitrogen) or the Perfect Protein marker (Novagen, Wisconsin, USA) was used as molecular weight markers. Home-made 2D electrophoretic gels were prepared and run as described previously (). The concentrated supernatant was mixed with 120 μl of 2D-gel loading buffer (8.75 M urea, 2 M thiourea, 6% CHAPS, 20 mM dithiothreitol). Proteins from the non-irradiated chromatographic phase were eluted with the same buffer. Samples were then loaded onto pH [3–10] 7 cm-long Immobiline DryStrips (GE Healthcare) for the first dimension and on 11% SDS-PAGE gels for the second dimension. The gels were stained with Sypro Ruby protein gel stain (Invitrogen). Spots of interest were digested in-gel with trypsin and the peptidic mixture was analyzed by MALDI-TOF mass spectrometry as previously described (,) except for the reduction and alkylation steps that were performed before the denaturing electrophoresis, during the Immobiline DryStrips equilibration. Sequence editing and mass spectral data simulations were performed using the Free Software package (). Gas-phase fragmentation experiments were performed in the positive-ion mode using a hybrid quadrupole time-of-flight mass spectrometer equipped with a Protana source (nanoESI MS, Q-Star Pulsar i, MDS Sciex-Applied Biosystems). Data acquisition and storage were performed using the Analyst QS software package shipped with the mass spectrometer. Ions were selected based on their [M+2H] / value, in the mass unit mode, using the Analyst QS software package. The protein kinase activities associated with purification fractions were detected by monitoring the phosphorylation level of the XRCC4 protein, a component of the DNA repair protein complexes that assembled onto our DNA damage-mimicking oligonucleotide. The purification of proteins involved in DNA repair was performed as described above, starting from HeLa nuclear extracts depleted in ATP with 8 U· ml hexokinase and 2 mM glucose for 10 min at 30°C. Following incubation of the protein extracts with the chromatographic resin, five rounds of washes were performed and the chromatographic phase was resuspended in 20 μl of buffer C for further kinase assays. Kinase assay conditions were: 1 mM ATP, incubation for 2h at 30°C. When the kinase activities were monitored in presence of the beads (i.e. without irradiation), the chromatographic phase was incubated without or with ATP, and only after were the DNA–protein assemblies released from the beads by irradiation. When the kinase activities were monitored on the released DNA–protein complexes, the irradiation step was performed and the supernatant was collected for kinase assay. In either case, the proteins were migrated on SDS-PAGE gels for further western blot analysis. When indicated, 10 μM wortmannin (Sigma), a kinase inhibitor, was incubated with the sample assayed for kinase activities on ice for 30 min right before the addition of ATP (). The samples were desalted in a microchromatography device prepared according to Rusconi () with small modifications: Poros resin (kind gift from Dr Carole Feltaille, Applied Biosystems, Coutabœuf, France) was packed in a microloader pipette tip (3 μl packed resin bead) and was equilibrated with 500 μl of 1% formic acid. The sample (20 μl) was deposited on top of the resin bead and acidified in place with 2 μl of pure formic acid. Desalting was performed by passing 500 μl of 1% formic acid and the proteins were eluted with 500 μl of 80% acetonitrile–1% formic acid (v/v). The proteins were lyophilized, electrophoresed (14% SDS-PAGE) and later transferred onto a nitrocellulose membrane. Western blotting was performed with anti-XRCC4 antibody. To determine the phosphorylated forms of the XRCC4 protein, 10 or 5 U of calf intestinal phosphatase (New England Biolabs) were incubated at 37°C for 60 min with 30 μg HeLa nuclear extracts or the purification products, respectively. The purification of proteins involved in DNA repair (double-strand DNA breaks) was performed as described in the protein kinase assay paragraph. The proteins associated to the target oligonucleotide were incubated with 1 mM ATP for 2 h at 30°C before or after the UV-irradiation of the chromatographic slurry. The purification products were then incubated with 200 μg ml proteinase K (Ambion, Courtabœuf, France) at 55°C for 60 min. The proteins were removed by phenol-chloroform-isoamyl alcohol treatment and the recovered oligonucleotides were precipitated. The samples were electrophoresed on 8% polyacrylamide/0.42% bisacrylamide gel running in 50 mM Tris base, 45 mM boric acid and 0.5 mM EDTA buffer. Denaturation of the nucleic acids was performed by incubating the gel in 1.5 M NaOH for 30 min; it was then neutralized by soaking it in a 1 M Tris–HCl pH 7.5; 3 M NaCl solution. The Southern blotting was performed in the experimental conditions described in (). The hybridization with the 5′[γ−P] radiolabeled non-photocleavable strand of the PCB-TetO duplex was at 55°C. The positive ligation control was obtained by incubating 25 pmol of the TetO oligonucleotide with 0.4 U of T4 DNA ligase (New England Biolabs) for 120 min at room temperature in the buffer provided by the manufacturer. The schematic in shows that our strategy involves the use of a photocleavable linker molecule between the oligonucleotide that will serve as the bait for the purification and the chromatographic support (magnetic streptavidin-coated beads). The functionalization of the magnetic beads with the oligonucleotidic bait is actually made possible because the linker molecule is itself bound to a biotin moiety. All the chemical components involved in the production of the chromatographic phase are available in the commerce, thus making the method straightforward. We first applied this method to a simple case: the purification of the tetracycline repressor protein (TetR) from crude cellular extracts. The TetR homodimer is known to regulate bacterial genes responsible for tetracycline resistance by binding to the tetracycline operator sequence (TetO). TetR and its derivatives are widely used to modulate the expression of ectopic genes in eukaryotic cells (). T-Rex HeLa cells stably express TetR under the control of the human cytomegalovirus promoter (). Using an anti-TetR antibody and different amounts of the pure TetR protein we estimated by western blot that the TetR protein represents < 0.008% of total proteins in the protein extract from this cell line (data not shown). In a preliminary experiment, the ability of the TetR protein expressed in the T-Rex HeLa cells to bind to the oligonucleotidic cognate sequence was established by incubating—in presence or absence of tetracycline—whole cell extracts with the radiolabeled oligonucleotide. The formation of DNA–protein complexes was monitored by the electrophoretic mobility shift assay (EMSA, A, lanes 1–3). The band shift observed in lane 2 indicates that at least one protein did bind to the oligonucleotide. Addition of tetracycline in the incubation mixture abolished that observed band shift, suggesting that the oligonucleotide had been bound by the tetracycline repressor (lane 3). Indeed, tetracycline inhibits the interaction between the TetR protein and its cognate TetO sequence (). The result from a typical chromatographic experiment is shown in A, lane 4, where the chromatographic phase was first assembled by incubating the biotin-labeled oligonucleotide with the streptavidin-coated magnetic beads. The T-Rex HeLa cellular extracts were then batch-wise incubated with the chromatographic resin. The beads were pelleted and the supernatant (equivalent to a flowthrough) was collected prior to performing five washes of the resin. The chromatographic slurry was UV-irradiated, thus releasing the oligonucleotide along with any protein bound to it. The band shift observed in lane 4 shows that the oligonucleotide had effectively been recognized by a protein. Further, when the purification product was incubated with tetracycline, that band shift was abolished, showing that the oligonucleotide was recognized by the tetracycline repressor (B). The ratio between the free oligonucleotide and the bound oligonucleotide in the product (A, lane 4) was 60% lower than the one obtained in the solution (A, lane 2). This observation might be explained by the fact that the binding of the oligonucleotide onto the beads might sterically hamper the interaction between the TetR protein and the DNA sequence. Inspired by others (,,), we performed a purification experiment in which proteins were first incubated with the oligonucleotide and the formed DNA–protein complexes were attached to the beads only thereafter. In our hands the pre-incubation of the free oligonucleotide with the cellular extracts brought no significant advantage with respect to our first methodology, confirming the hypothesis of steric hindrance. We thus standardized our experiments with the procedure in which the chromatographic resin is prepared prior to its incubation with the cellular extracts. Cellular extracts were incubated with the chromatographic phase and, following thorough washing of the beads, the chromatographic slurry was either UV-irradiated or mixed with denaturing Laemmli sample buffer. In each case, the beads were magnet-pelleted and the obtained fractions—supernatant and pellet—were further resolved by SDS-PAGE. The proteins were transferred onto a membrane (A) that was first stained with the Sypro blot stain (left panel) and then used to probe the presence of the tetracycline repressor protein with a polyclonal antibody (right panel). The left panel shows that the vast majority of the proteins are found attached to the beads, in the pellet fractions, with or without previous UV-irradiation of the chromatographic phase (lanes 2 and 3). The supernatants (lanes 4 and 5) only show a faint band at 40 kDa. The presence of that band in the supernatant fractions—obtained with or without irradiation—indicated that it did not correspond to a molecular species specifically bound to the oligonucleotide but that it had probably leaked from the beads themselves. The western blot shown in the right panel confirmed that the tetracycline repressor protein effectively interacted with the beads (lane 8). Indeed, when no UV-irradiation was performed, the supernatant did not contain the repressor (lane 10), which was found in the pellet fraction instead (lane 8). Upon UV-irradiation of the chromatographic slurry, 75% of the tetracycline repressor protein remained on beads after washes was recovered in the supernatant (lane 9), showing that most of the repressor protein actually interacted with the beads via the oligonucleotide. Because the repressor protein band appeared somewhat thick, the same experiment was performed but was followed by a 2D-gel electrophoresis in order to further resolve potentially overlapping protein variants in that band. Equally significant is the fact that 2D-gel electrophoreses do concentrate proteins in spots that are more suitable for further analysis of their contents by mass spectrometry. The 2D-gel obtained without irradiation (B, left) corresponds to the biological material recovered on the beads in the same way as for lane 3 of A. When the chromatographic slurry was irradiated, the supernatant did contain only two molecular species visible as spots at 40 kDa and 25 kDa (B, right). The lower spot is visible on both gels and is pointed to by an arrow. We suspected this spot to contain the tetracycline repressor protein, which was later confirmed by mass spectrometry. The two detected spots were excised from the gel and subjected to in-gel trypsinolysis. The peptidic mixtures were analyzed by matrix-assisted laser desorption/ionization mass spectrometry. The peptide mass fingerprint results are shown in for the lower spot, indicating that the protein contained therein was the tetracycline repressor protein (30% sequence coverage; 8 matching peptides). The ∼ 40 kDa protein was identified as human β-actin (34% sequence coverage; 13 matching peptides; data not shown). Actin is one of the most abundant proteins in eukaryotic cells, as can be seen on 2D gels where actin makes one of the largest spots at roughly 40 kDa. It is thus possible that this protein leaked from the beads even after the numerous washing steps performed during the purification. The definitive confirmation of the identity of the protein contained in the ∼ 25 kDa spot was brought by nanospray tandem mass spectrometry with gas-phase fragmentation of the [M+2H] parent ion at / 734.34. The obtained sequence is shown in . These results showed that the tetracycline repressor protein could be purified with an unprecedented enrichment factor and that the uncoupling of the affinity determinant (the oligonucleotide) from the chromatographic support (the streptavidin-coated beads) is responsible for that achievement. The specificity of the interaction between the tetracycline receptor protein and its cognate oligonucleotidic sequence is not questionable, as shown by the disruption of that interaction in the presence of tetracycline (B). Equally significant is the fact that the DNA–protein complex is purified in its native form, even after the UV-irradiation. Finally, because the chromatography steps are compatible with 2D-gel electrophoresis, the protein was successfully identified by mass spectrometry. The results obtained with the TetR/TetO system showed the benefits of the photocleavage approach with respect to conventional elution methods. The TetR/TetO system was rather simple as it involved a binary interaction between a protein and its cognate sequence. Therefore, to challenge the robustness of our method, we applied it to the purification of more complex protein assemblies involving both protein–protein and protein–DNA interactions. The experimental system that we chose was the DNA repair machinery that recognizes double-strand breaks on DNA. Two distinct factors motivated this choice: first, on the protein side, the repair machinery is well studied and a number of proteins taking part to the multiprotein repair assemblies are well known and characterized; second, on the nucleic acids side, the DNA double-strand break to be repaired is easily implemented in the test tube because a duplex with free double-strand ends mimicks a double-strand break and as such is able to recruit the DNA repair machinery (,). Two protein complexes are known to be required for DNA double-strand break repair: the DNA-PK holoenzyme and the DNA ligase IV-XRCC4 complex (). DNA-PK comprises three subunits: the DNA-dependent protein kinase catalytic subunit (DNA-PKcs), Ku70 and Ku80. This heterotrimer binds to the DNA ends and recruits the DNA ligase IV-XRCC4 complex that accomplishes the ligation reaction (,). In order to purify these protein assemblies, we coupled to the chromatographic support the PCB-TetO oligonucleotide of which the free blunt ends mimicked double-strand breaks (see sequence in Materials and Methods section). The sequence length of the PCB-TetO oligonucleotide was sufficient to recruit the DNA repair proteins (). The chromatographic phase was incubated with HeLa nuclear extracts. After washes, the proteins were recovered either without irradiation or after irradiation of the chromatographic resin, resolved by SDS-PAGE and stained (A). Without irradiation, proteins from both the beads and the target oligonucleotide were deposited in lane 2, showing a heavy staining pattern. If the DNA–protein complexes were first cleaved off the beads with an irradiation step and the beads removed, the observed staining pattern was much lighter (lane 4), clearly showing that the beads did retain a huge amount of proteins unspecifically bound to them. Interestingly, we performed a purification procedure without DNA on beads and after the irradiation of the chromatographic slurry, the proteins in the supernatant were recovered and stained (lane 4). A very small amount of proteins was released off the chromatographic support, which indicated that the proteins recovered from the purification with DNA were specifically bound to the oligonucleotidic bait (lane 4, compare with lane 3). These results confirmed our initial results with the TetR/TetO binary system, underlining the benefits of the photocleavage step in our method. In lane 3, four major proteins were detected with apparent molecular weights suggesting their identity: Ku70 at 70 kDa, Ku80 at 80 kDa, PARP-1 at 120 kDa () and probably DNA-PKcs > 225 kDa. These hypotheses were confirmed by western blotting the gel with antibodies against each one of these proteins (B and Supplementary ). Because saline conditions were identical throughout the purification process, weak protein–protein and DNA–protein interactions could be preserved, letting us hope that the purified material would have retained its native structure and enzymatic activities. Functional studies were performed on the purified DNA–protein complexes by analyzing enzymatic activities known to operate for such complexes: we first monitored the actual DNA repair via oligonucleotide end-joining and second the phosphorylation of XRCC4 (,). In the following section, these different enzymatic activities operating either on the DNA component or on the proteinaceous components of the DNA–protein complex were systematically compared before and after the irradiation step. In the first experiment, the oligonucleotide ligation was tested either with the DNA–protein assemblies still coupled to the beads (no irradiation) or after their irradiation-mediated release in the supernatant (after removal of the beads). The ligation products were analyzed by Southern blot (). No ligation product could be detected when the DNA–protein assemblies were still coupled to the beads, suggesting that the beads may have reduced the ligation rate too much for its products to be detectable (lane 3). Conversely, the TetO oligonucleotide dimer was detected when the DNA–protein complexes were detached from the beads (lane 4) albeit with a low yield of ligation (∼ 1%), as was expected for blunt DNA ends (). In the next experiments, we analyzed kinase activities that resulted in the phosphorylation of the XRCC4 protein. We focused our attention on XRCC4 because it is a known target of kinases (amongst which DNA-PKcs) (), with the advantage that the non-phosphorylated and the phosphorylated species can easily be resolved on SDS-PAGE gels and later detected by western blot (). To prevent phosphorylation from occurring throughout the purification process, our starting material was prepared by depleting HeLa nuclear extracts of their ATP content. No phosphatase inhibitors were added during this step. A, lane 1 shows the XRCC4 variants in the starting material in the absence of ATP that were detected as a doublet of bands. The heavier variant could be eliminated by incubating the sample with calf intestine phosphatase (CIP), indicating that it was most probably due to phosphorylation (B, lanes 8 and 9). For the next series of experiments, the purification of protein assemblies involved in DNA repair was performed in the same manner as before. However, in order to monitor the molecular species involved in the phosphorylation of XRCC4 and infer their main localization (on the beads or on the oligonucleotide), ATP was added at two different steps: either before or after the irradiation-based release of the DNA–protein complexes off the beads. When ATP was added before the irradiation of the chromatographic phase (i.e. when the DNA–protein complexes were still attached to the beads), a number of XRCC4 variants were found (lane 4) of which the ones migrating in the band labeled with a filled circle are the heaviest. When ATP was added to the DNA–protein complexes alone (i.e. after irradiation, so as to detach and remove them from the beads), XRCC4 variants could also be detected, albeit with lower apparent molecular weights than in the previous experiment (lane 5, filled square). In each of these cases, if no ATP was added, no heavy XRCC4 variants could be detected (lanes 2 and 3). In order to ascertain that the molecular variants detected in these experiments reflected the existence of phosphorylated XRCC4, another experiment was performed in which the DNA–protein complexes—without the beads—were incubated with ATP first (same as lane 5) and subsequently incubated with CIP (or without as a control). The results are shown in B, lanes 10 and 11, which demonstrate that the heavy molecular variants were indeed XRCC4 phosphorylation variants, as the slow-migrating bands (lane 10) were converted into bands migrating at the same rate as the ones in lane 9 (no incubation with ATP). The same results were shown with the DNA–protein assemblies coupled to the beads and incubated with ATP (same as lane 4, data not shown). Overall, these results demonstrated that with or without irradiation, kinases copurifying with the DNA repair complex (either on beads or on the bait oligonucleotide) retained their activity with or without irradiation. Further, one intriguing result obtained in this series of experiments is that the phosphorylation status of XRCC4 varied significantly depending on the presence or absence of the beads during the incubation of the purification products with ATP. To appreciate this, one can compare lanes 4 and 5 (A): if ATP is added when the DNA–protein complexes are still attached to the beads (lane 4), the XRCC4 phosphorylated variants appear conspicuously heavier than those observed in lane 5 (no beads). This result might suggest that kinase activities adsorbed on the beads do phosphorylate XRCC4 in a distinct manner than do the activities interacting with the DNA oligonucleotide. DNA-binding proteins are generally purified by affinity purification using an oligonucleotide as a bait. Salts and/or detergents are generally added to the washing or elution buffers to recover proteins specifically interacting with the target oligonucleotide. A compromise has to be found between a high protein background and the loss of components or activities of interest. In fact, numerous proteins from whole cell extract interact with the chromatographic support that accounts for a huge interaction surface with respect to that of the target oligonucleotide. To circumvent this problem, a number of strategies were devised in the past and were based on the separation of the DNA–protein assemblies from the chromatographic support before elution. The insertion of a disulfide bond between the target oligonucleotide and the chromatographic support (,) (see Introduction Section), was appealing at first sight. However, it might not be widely applicable as reducing conditions are often required for the proper interaction of proteins with their cognate DNA sequence (). In the case of competition-based strategies used to uncouple DNA–protein assemblies from the chromatographic support (,,), one serious problem is that the aptamer usually has non-negligible length (in the range of 20–30 nucleotides) and that, as such, it might recruit proteins impeding it to bind to the chromatographic phase (). In all the strategies above, the uncoupling of the nucleic acid–protein complex is achieved by adding chemicals (either small or large molecules) to the purification medium. This might prove detrimental to downstream analytical steps, like the direct analysis of nucleic acid–protein complexes by mass spectrometry or the isoelectrofocalization of the purified proteins, for example. The near-UV-irradiation strategy used in this work clearly had none of the drawbacks described above. This approach avoids to set physico-chemical conditions optimized to eliminate contaminants from the chromatographic support, since the same weak saline conditions are maintained throughout the purification process. In the case of the purification of DNA ends-binding proteins, we obtained a profile very similar to the ones already published (,). The proscription of stringent conditions (such as high salts and/or presence of detergents) preserved weak interactions (protein–protein and DNA–protein) and allowed to recover DNA–protein assemblies in their native state, as demonstrated by the detection of biologically relevant enzymatic activities copurifying with the DNA–protein complexes. Enzymatic tests performed before or after the UV-irradiation step (i.e. the purified material was respectively in presence or absence of the beads), underlined the fact that photocleaving the DNA–protein assemblies off the beads was highly beneficial for two reasons: first, significant amounts of contaminant enzymatic activities not relating to target oligonucleotide were found to be unspecifically adsorbed onto the beads, leading to spurious phosphorylation of one molecular partner of the purified DNA–protein assemblies; second, monitoring of the ligation activity operating on tethered DNA–protein complexes proved unsuccessful, while it could be achieved if the complexes were first uncoupled from the beads. Taken as a whole, these results demonstrated that the photocleavage step allowed a good compromise between the yield and the purity of the purified proteins. Finally, this method presents the advantage of being fast and easy to carry out as it is a one-step procedure, thus reducing to a minimum the loss of molecular partners from the assemblies of interest. The photocleavable biotin can be incorporated by chemical synthesis in solid phase at the end of the DNA-probe. Short PC-biotin modified primers can be also used to synthesize a longer DNA sequence by PCR (concatemers made of an affinity sequence repetition, for example), or else to hybridize a complementary sequence linked to the nucleic probe of interest (adapter strategy). These different approaches were tested in the case of TetR purification and all gave similar results as the basic duplex PCB-TetO (data not shown). In conclusion, the procedures developed in this work are generalizable to any nucleic acid-binding proteins and permit as well the identification of these proteins as the study of their functions. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The eye is composed of ∼800 repeat units, referred to as ommatidia. Each ommatidium is composed of six elongated peripheral photoreceptor cells (R1–6), which extend across the length of the ommatidium, and two shorter central photoreceptors (R7 and -8; ). Photoreceptors are highly polarized cells composed of two well defined compartments: a cell body and a signaling compartment called the rhabdomere. The rhabdomere contains tightly packed, actin-rich microvilli that harbor the signaling proteins required to generate the photoreceptor potential upon illumination. The transient receptor potential (TRP) protein is a light-sensitive cation channel that provides a major component of the light-induced current (). TRP is also required for anchoring a supramolecular signaling complex that includes the inactivation-no-afterpotential D (INAD) PSD95/DlgA/ZO-1 homology (PDZ) scaffold protein, PLCβ, and the eye-specific PKC (eyePKC) to the plasma membrane (; ; ; ). TRP-like (TRPL) is a second light-activated channel () that, together with TRP, participates in the production of the light-induced current (). Genetic elimination of both TRP and TRPL channels completely eliminates the photoreceptor potential (; ). The photoreceptor potential is only one of several responses to light. Another light-induced response is the translocation of signaling proteins (Gqα and TRPL) and actin between the rhabdomeric membrane and the cell body (; ). The molecular mechanisms underlying translocation of these proteins from the microvilli to the cell body remain largely unknown (). In several tissues, microvillar organization depends on protein members of the ezrin-radixin-moesin (ERM) family, which form a bridge between the actin cytoskeleton and the plasma membrane (for review see ). ERM proteins bind to integral membrane proteins either directly or through PDZ scaffold proteins, such as ezrin binding phosphoprotein 50 (EBP50)/Na+/H+ exchanger regulatory factor and NHE3 kinase A regulatory protein. This binding is a dynamic process, which takes place upon the binding of phosphatidylinositol 4,5-bisphosphate (PIP) and phosphorylation of the ERM protein (; ). Dmoesin, the unique member of the ERM family in (), is required for the specific organization of different actin-rich structures during development. Furthermore, Dmoesin plays an essential structural role in photoreceptor morphogenesis (). Dmoesin mutations disrupt the polarized localization of posterior determinants in oocytes (; ). Mutations that disrupt the dynamics of Dmoesin phosphorylation produce severe defects in actin reorganization and cell shape (; ; ). Although Dmoesin has been shown to accumulate in rhabdomeres (), its physiological function in mature photoreceptors and its relationship to light reception is not known. In this study we used wild-type (WT) and mutant strains to show that Dmoesin only interacts with the TRP and TRPL channels in dark-raised flies. Furthermore, we show that illumination induces dephosphorylation of the conserved COOH-terminal threonine 559 (T559) of Dmoesin, which subsequently dissociates from the channel proteins and moves from the rhabdomeric membrane to the cytosol. Consistent with this conclusion, our results show that mutations that impair phosphorylation of Dmoesin (; ) abolish the movement of Dmoesin upon illumination and result in light-activated degeneration of the photoreceptor cells. We recently showed that actin moves from the rhabdomere to the photoreceptor cell body after the illumination of dark-raised flies (). To investigate a possible mechanism that underlies light-activated actin movement in photoreceptors, we examined the role of Dmoesin, a known regulator of the dynamic reorganization of actin-rich cell structures (). Activation of Dmoesin involves a redistribution of the protein from the cytoplasm (dormant form) to the plasma membrane (active form; ). Thus, modification of Dmoesin intracellular localization might be related to the light-dependent reorganization of actin filaments. To test this possibility, we examined the distribution of Dmoesin between the membrane and cytosolic fractions in head extracts of dark-raised and illuminated . Although Dmoesin was detected in both the soluble and membrane fractions, the majority of Dmoesin was associated with the membrane fraction in dark-raised flies ( A). In contrast, Dmoesin was predominantly in the soluble fraction in illuminated WT flies ( A). Quantification of the Dmoesin membrane/cytosolic ratio further supports the conclusion that illumination induces a substantial movement of Dmoesin from the membrane to the soluble fraction ( B). Upon illumination, nearly 50% of the membrane-associated Dmoesin of the head moved to the cytosol. By using mutants lacking eyes (Fig. S1, available online at ) we found that the Dmoesin protein present in heads, outside the photoreceptors, remained associated with membranes upon illumination and therefore was not affected by light. To further test whether the observed intracellular movement of Dmoesin depends on the activation of the visual signaling cascade, we analyzed Dmoesin distribution in mutants that inactivated phototransduction. The and genes encode the PLCβ isoform and the major light-activated channel TRP, respectively, and both are required for phototransduction. or mutants (). In both mutants, similar amounts of Dmoesin were detected in the membrane and soluble fractions, regardless of the illumination regime. Therefore, the light-induced intracellular movement of Dmoesin is dependent on a functional phototransduction cascade in photoreceptors. To directly visualize the movement of Dmoesin after illumination in the retina, we analyzed the immunocytochemical localization of Dmoesin using anti-Dmoesin antibodies (αDmoesin; ). In dark-raised flies Dmoesin was mainly localized to the base of the rhabdomeres and cortical actin region (). After illumination, Dmoesin level of peripheral R1–6 rhabdomeres was much reduced, with a concomitant increase in nonrhabdomeric regions in WT flies (, A–D) that is consistent with the Western blot analysis of . The failure of Dmoesin to move upon illumination in the central R7 cell can be attributed to the inefficient absorption of visible light by the UV rhodopsin of R7 cells (), which constitutes a negative control (, A–D). (, E–H). Results from our cell fractionations and immunocytochemistry suggest that illumination regulates Dmoesin dissociation from the membrane proteins of the signaling compartment of photoreceptors. Several studies have reported that ERM/EBP50 proteins interact with members of the TRP family of proteins, such as TRPC4 and TRPC5 (; ; ). Therefore, we examined whether Dmoesin interacts with TRP and TRPL, the two major channel proteins of the microvilli in photoreceptors. To test this hypothesis, we analyzed WT and mutant head extracts by immunoprecipitation with monospecific anti-TRP antibodies (αTRP). Protein complexes precipitated with αTRP were fractionated by SDS-PAGE and probed in Western blots with αDmoesin. In dark-raised WT flies, a strong Dmoesin signal was observed, indicating that Dmoesin and TRP formed a protein complex in vivo (). null mutant or in WT extracts immunoprecipitated with nonimmune serum (NIS; ), thus demonstrating the specificity of the Dmoesin–TRP interaction. We then examined the effect of illumination on the Dmoesin–TRP interaction. Interestingly, illumination caused a strong reduction of Dmoesin staining in complexes immunoprecipitated with αTRP, although similar amounts of TRP were immunoprecipitated from dark-raised and illuminated flies (, bottom). Therefore, TRP and Dmoesin interact in vivo only in dark-raised flies, and illumination dissociates the Dmoesin–TRP interaction. Because TRP is known to bind INAD, which is a multi PDZ-domain scaffold protein (; ; ), we examined whether INAD is required for Dmoesin–TRP interaction. Head extracts of dark-raised young (<2-d-old) null mutants were immunoprecipitated with αTRP. mutant flies (), indicating that Dmoesin interacts with TRP independently from INAD. mutant () is consistent with the known slow degradation of TRP in mutants (). As an additional control, αTRP-precipitated complexes were probed with antibodies against another major membrane protein of the microvilli, Chaoptin (). Western blot analysis did not reveal any Chaoptin signal (unpublished data), providing additional evidence of the specificity of the Dmoesin–TRP interaction. To address whether activation of the phototransduction cascade is required for the observed dissociation of the Dmoesin–TRP complex, we examined the coimmunoprecipitation of Dmoesin in a visual transduction mutant. In the PLCβ mutant ( ), a strong Dmoesin signal was observed in protein complexes precipitated with αTRP, regardless of illumination (). Western blot analysis of the same head extracts probed with αINAD revealed that roughly equal amounts of signaling proteins were coprecipitated in each lane (, bottom). eyes, TRP interacts with Dmoesin equally in both dark- and light-raised flies, thus indicating that the dissociation of the Dmoesin–TRP complex depends on activation of the phototransduction cascade. We then addressed whether Dmoesin also interacts in vivo with TRPL, the second light-activated channel of the microvilli. Analysis of protein complexes immunoprecipitated with anti-TRPL antibodies revealed that Dmoesin specifically interacts with TRPL in WT dark-raised flies (). Consistent with this conclusion, Dmoesin is undetectable in similar conditions with extracts from null mutants, or in WT extracts when using NIS (). As observed with TRP, the Dmoesin–TRPL interaction occurs specifically in dark-raised flies, whereas illumination dissociates the Dmoesin–TRPL complex (). Phosphorylation of a conserved threonine residue located in the actin-binding domain of ERM proteins has been shown to regulate both mammalian ERM () and Dmoesin activity and subcellular localization (). Therefore, we examined if the light-sensitive interaction of Dmoesin with the photoreceptor channel depends on Dmoesin phosphorylation. To address the influence of Dmoesin phosphorylation, extracts were immunoprecipitated with an antibody directed against an evolutionarily conserved COOH-terminal peptide of ERM () that specifically recognizes the phosphorylated T559 of Dmoesin (designated hereafter as α-phospho-ERM). Protein complexes precipitated with α-phospho-ERM were then analyzed on Western blots probed with αDmoesin. Although Dmoesin was readily detected in extracts of dark-raised flies, Dmoesin staining was strongly reduced in extracts from illuminated WT flies treated under identical conditions ( A, left). In addition, the TRP channel and the INAD scaffold protein that binds to TRP were also detected only in head extracts of dark-raised flies immunoprecipitated with α-phospho-ERM ( A, right). These results are consistent with our previous findings, which demonstrated that Dmoesin only interacts with TRP in dark-raised flies. To support this notion, extracts were immunoprecipitated with αTRP and analyzed on Western blots probed with α-phospho-ERM ( B). Although α-phospho-ERM staining was detected in the protein complexes of dark-raised head extracts precipitated with αTRP, α-phospho-ERM staining was strongly reduced in the protein complexes of illuminated flies ( B, left). null mutant or in NIS-precipitated WT membranes ( B, middle and right). To directly visualize intracellular movements of Dmoesin in photoreceptors upon illumination, we made use of transgenic lines that allow the expression of functional Dmoesin fused to GFP (). Dmoesin-GFP fusions were expressed under the control of the Rh1-Gal4 driver, which is specific to mature peripheral R1–6 photoreceptors. Upon application of long wavelength excitation light that elicits a strong autofluorescence of the rhabdomeres (without excitation of the GFP), locations and dimensions of the rhabdomeres in each ommatidium are readily observed in the living retina. The typical structure of the ommatidium is visible as seven red circles, representing the R1–6 peripheral rhabdomeres and the smaller R7 rhabdomere, at the center. Live retinae were dissected under dim red light, and the subcellular localization of Dmoesin-GFP was examined with confocal microscopy. A shows a representative image of ommatidia from dark-raised flies expressing Dmoesin-WT-GFP, which localizes to the rhabdomeres and cortical actin of R1–6 photoreceptors. Because of variability in the expression levels of Dmoesin-GFP in the various ommatidia, some photoreceptor cells did not express Dmoesin ( A). In the photoreceptor cells that did express Dmoesin-GFP, the intense fluorescence of the GFP masked the weaker red autofluorescence and the merged images appeared green. When the level of Dmoesin-GFP in the rhabdomeres was reduced ( F) a yellow color appeared in the merged images. The lack of green fluorescence from the central R7 rhabdomere (which does not express Dmoesin-GFP) provides an internal negative control. After illumination a marked redistribution of Dmoesin-GFP is observed, with the green fluorescence moving from the rhabdomeres to the cell body region ( B). Although there are some differences in the detailed localization and movement of the native Dmoesin () and Dmoesin-WT-GFP, the results of confirm our interpretation and clearly indicate that illumination induces a redistribution of the Dmoesin protein from the rhabdomere to the cytoplasm of the cell body. To address the influence of T559 phosphorylation on Dmoesin light-induced movement as observed in vivo, we expressed variant Dmoesin proteins with point mutations. The T559A mutation prevents phosphorylation of Dmoesin to keep it in the dormant state, whereas the T559D mutation is a phosphomimetic mutation that is expected to prevent phosphorylation but to keep the molecule in the “open” (presumably active) state (). When compared with Dmoesin-WT-GFP, distribution of Dmoesin-T559A-GFP in dark-raised flies showed a marked difference, with the major fraction of the GFP fluorescence in the cell body ( C), which is reminiscent of illuminated Dmoesin-WT-GFP flies ( B). In addition, illumination did not induce a significant change in the subcellular distribution of the Dmoesin phosphomutant T559A ( D). Dmoesin-T559D-GFP of dark-raised mutants showed intermediate distribution between rhabdomeres and the cell body. Although GFP fluorescence in the rhabdomere and cortical actin regions was significantly reduced relative to WT retina, it was much higher relative to the Dmoesin-T559A mutant ( E). Quantification of GFP colocalization with the rhabdomere autofluorescence confirmed that the Dmoesin-WT present in rhabdomeres of dark-raised flies was drastically reduced (greater than fivefold) upon illumination ( G). In contrast, most of Dmoesin-T559A fluorescence was not confined to the rhabdomeres in dark-raised retinae, and there was no significant movement after light exposure. Also, similar levels of Dmoesin-T559D were observed in both dark- and light-raised retinae. Thus, these data support the conclusion that phosphorylation of T559 controls the rhabdomeric localization of Dmoesin and that dephosphorylation controls its light-activated movement to the cell body. To support this interpretation using a different approach, we analyzed the impact of T559 mutations on Dmoesin movement through biochemical characterization. The membrane and soluble fractions of head extracts from flies expressing Dmoesin-WT-GFP, Dmoesin-T559A-GFP, and Dmoesin-T559D-GFP were fractionated by SDS-PAGE and analyzed by Western blots using anti-GFP antibodies ( A). Although illumination reduced the Dmoesin level in the membrane fraction and concomitantly increased Dmoesin levels in the cytosol of Dmoesin-WT-GFP, the distribution of phosphorylation-defective Dmoesin mutants was unmodified by light (). As expected, the major fraction of Dmoesin-T559A-GFP was restricted to the soluble fraction, whereas the Dmoesin-T559D-GFP appeared in both the membrane-associated and the cytosol fractions (). Together, these results demonstrate that the phosphorylation/dephosphorylation reactions of T559 regulate light-induced subcellular movement of Dmoesin in photoreceptors. In dark-raised flies, Dmoesin binds to the channels in the rhabdomere and illumination induces both Dmoesin dephosphorylation and relocation to the photoreceptor cytoplasm. Consistent with this interpretation, mutations impairing T559 phosphorylation either prevent () or reduce () association of Dmoesin with the rhabdomere. These mutations totally block intracellular redistribution of the protein in response to light. Estimation of Dmoesin dynamics was obtained from Western blot analysis in which the reduction of Dmoesin levels in the membrane fraction and increase in the soluble fraction were measured after increasing durations of illumination (Fig. S2, available online at ). After 5 min of illumination, a significant fraction of Dmoesin was still detectable in the membranes. After 10, 30, and 60 min of light exposure, the level of Dmoesin further decreased in the membranes and increased in the soluble fraction (Fig. S2). Thus, the time scale of Dmoesin movement from the rhabdomere is similar to the time scale of TRPL translocation (). Together with the redistribution of signaling molecules, we examined if the light-induced movement of Dmoesin was important for the physiology of photoreceptors. We examined the effects of Dmoesin mutations that impair T559 phosphorylation on the retinal structure after extended exposure to light. As prolonged illumination produces toxic effects in photoreceptors expressing GFP (unpublished data), we used transgenic lines carrying the same mutations, with the exception of Dmoesin, which was tagged with the epitope instead of GFP (). Newly eclosed flies expressing WT, T559A, or T559D Dmoesin proteins were subjected to constant light for 2, 4, and 7 d, and the ultrastructure of the ommatidia was analyzed by transmission electron microscopy. Ommatidia of flies expressing Dmoesin-WT- were indistinguishable from WT controls and presented a well organized ommatidium after 7-d illumination (). were abnormal and revealed the initial stages of degeneration () after only 2-d illumination (). Although the ommatidia of the illuminated mutants appeared normal after 2-d illumination, slight degeneration was visible after 4-d illumination and a significant degeneration appeared after 7-d illumination (). Control experiments in which flies that expressed either WT or mutant Dmoesin forms were kept in the dark and did not show any sign of retinal degeneration (unpublished data). These data suggest that the dynamic regulation of Dmoesin phosphorylation is critical for photoreceptor viability upon illumination. The slower degeneration of mutants suggests that the presence of “active” Dmoesin in the rhabdomere is probably necessary, but not sufficient, to prevent degeneration during prolonged illumination and that dynamic phosphorylation/dephosphorylation reactions are required to prevent degeneration. italic #text of the following strains were used: WT Oregon-R ; () and () null mutants for the TRP channel and eye-specific PLCβ, respectively (both obtained from W.L. (obtained from C.S. Zuker, University of California, San Diego, San Diego, CA); and , , (), , , and (). Dmoesin variants encoded in the transgenic lines were expressed using the Gal4/UAS targeted expression () using the driver line (obtained from C. Desplan, New York University, New York, NY). Flies were raised in complete darkness from the first instar larval stage to eclosion. For illumination experiments, live flies were placed in a transparent dish with reflective aluminum foil at the bottom and subjected to illumination with blue light (18 W fluorescent lamp with a wide band filter [1 mm; model BG 28; Schott]) for various durations at 22°C. Illumination with white light of the same intensity produced similar results. After illumination, the flies were moved to 4°C in the dark and the fly heads were promptly dissected. Three to five flies were used for each lane of the Western blots. Fly heads were homogenized in a hypotonic buffer (50 mM Hepes, pH 7, 300 mM NaCl, 3 mM MgCl, 10% vol/vol glycerol, protease inhibitor [Sigma-Aldrich], phosphatase inhibitor [Sigma-Aldrich], and 1 μM caliculin A [Calbiochem]). Membrane and cytosolic fractions were separated by centrifugation at 15,800 for 15 min, at 4°C. The pellet was washed and centrifuged again, and the supernatants were combined. Ultracentrifugation at 150,000 for 30 min did not substantially change the distribution of Dmoesin between the fractions. Protein samples ran on 10% SDS-PAGE and were subjected to Western blotting using anti-Dmoesin or anti-GFP polyclonal antibodies. Quantification of the gels was performed using the BAS-1000 system (Fujifilm Worldwide) with TINA version 2.0 software. Frozen heads (500–2,000) of dark-raised or illuminated flies were homogenized in 200 μl of buffer containing 50 mM Hepes, pH 7.6, 150 mM NaCl, 3 mM MgCl, 10% glycerol, and protease and phosphatase inhibitors (Sigma-Aldrich). The homogenate was centrifuged at 100 for 5 min to remove chitin materials. Membranes were isolated by centrifugation at 20,000 for 30 min at 4°C and resuspended to a final equivalent of 200 μl. Membrane proteins were extracted by incubating the membranes with 1% Triton X-100 and 500 mM NaCl for 1 h and centrifuging at 20,000 for 30 min. For immunoprecipitations, protein A beads were incubated first with the relevant antibody (αTRP, αTRPL, αGFP, αINAD, or αDmoesin) overnight at 4°C. The membrane extracts (of WT, , , and mutants) were incubated with the relevant antibody crosslinked to protein A beads (vol 20–100 μl) in 200 μl of total volume overnight at 4°C. The beads were washed in the Triton X-100 washing buffer (0.1% Triton X-100, 100 mM NaCl, and 50 mM Tris-HCl, pH 8.0). Immunoprecipitated proteins were eluted from protein A–agarose beads with 50 μl of 1× SDS-PAGE loading buffer and subsequently analyzed by Western blot. Immunoblots were incubated with primary antibodies in blocking buffer at the following dilutions: monoclonal mouse αTRP () and mAb83F36 (obtained from S. Benzer, California Institute of Technology, Pasadena, CA), 1:2,000; rabbit αDmoesin (obtained from D. Keihart, Duke University, Durham, NC), 1:2,000; rabbit αINAD, 1:1,000; rabbit α-Phospho-ERM (Cell Signaling Technology), 1:2,000; mouse αTRPL, αRh1, and αChaoptin (Hybridoma Bank), 1:1,000. Immunoreactive bands were visualized by chemiluminescence reaction (obtained from Biological Industries), using HRP-conjugated goat anti–rabbit and anti–mouse as secondary antibodies (Jackson ImmunoResearch Laboratories). Fly heads were separated, bisected longitudinally, and fixed for 12 h in a solution of 1.5% PFA and 2.5% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4. The heads were washed three times in the same phosphate buffer and dehydrated in graded aqueous ethanol concentrations of up to 90%. After fixation, eyes were post-fixed with 1% osmium tetroxide for 4 h, dehydrated in ethanol and propylene oxide, and embedded in Epon. Thin sections were cut and stained with saturated aqueous uranyl acetate and lead citrate. Sections were observed with a Tecnai-12 transmission electron microscope (FEI) and photographed with a Mega-view II charge-coupled device camera (Philips). Live retinae from dark-raised or illuminated transgenic flies were isolated from the cornea and brain and kept in Ringer's solution as described previously (). Optical sections of single ommatidia were visualized using the Fluoview confocal microscope (model 200 IX70; Olympus) using a LUM Plano Fl 60×, 0.9 NA, water objective. Optical sections were recorded from the upper region of the ommatidia, at a depth of 6–10 μm from the tip of the ommatidium. Autofluorescence and GFP fluorescence were recorded sequentially using laser excitation wavelengths of 568 and 488 nm, respectively. Pictures are merged sequential images obtained by 568- followed by 488-nm excitation lights, on separate channels. mutant in background were fixed in 2% PFA in PBS (175 mM NaCl, 8 mM NaHPO, and 1.8 mM NaHPO, pH 7.2) for 1 h at RT, and then washed two times in 0.1 M phosphate buffer (0.1 M NaHPO and 0.1 M NaHPO). This was followed by three washes in 10% sucrose and two washes in 25% sucrose for 15 min each. Eyes were then infiltrated with 50% sucrose overnight at 4°C, cryofixed in melting pentane, and sectioned at 10 μm thickness in a cryostat (Mikrom Laborgeräte GmbH) at −25°C. The cryosections were incubated in 2% PFA in PBS for 10 min, washed two times in PBS, and then blocked in 1% BSA and 0.3% Triton X-100 in PBS (PBS-T) for 2 h at RT. The sections were incubated with α-Dmoesin diluted 1:2,000 or α-phospho-ERM diluted1:50 in PBS-T overnight at 4°C. The sections were subsequently washed three times in PBS and were incubated with a Cy5-coupled secondary goat anti–rabbit antibody (Dianova) and rhodamin-coupled phalloidin (Sigma-Aldrich) in 0.5% fish gelatine and 0.1% ovalbumin in PBS for at least 4 h at RT. The sections were finally washed three times in PBS, mounted in Mowiol 4.88 (Polyscience), and examined with a confocal laser-scanning microscope (LSM-SP). Fig. S1 shows that Dmoesin of eyeless heads does not display light-dependent movement from the membrane to the cytosol. Fig. S2 depicts the kinetics of light-dependent movement of Dmoesin from the rhabdomere to the cell body and time-course analysis of Dmoesin redistribution upon illumination, as observed in Western blot analysis. Online supplemental material is available at .

The acquisition of immune responsiveness is mediated through the assembly, expression, and function of antigen receptors at the cell surface. In T lymphocytes, the antigen T-cell receptor (TCR) is a multiple subunit complex comprising clonotypic, disulfide-linked α and β antigen recognition chains devoid of signaling motifs associated with the invariant chain subunits CD3γ, δ, and ϵ, and TCRζ. These invariant chains assemble in the mature TCR/CD3 complex as noncovalently linked CD3γϵ and CD3δϵ heterodimers and disulfide-linked TCRζ–ζ homodimers and transmit signals inside the cell following TCR ligation-induced tyrosine phosphorylation of immune receptor tyrosine-based activation motifs contained within their extended intracytoplasmic tails. These phosphotyrosine motifs form docking sites for Syk family tyrosine kinases, such as ζ-associated protein of 70 kDa (ZAP-70), which in turn couple the antigen receptor complex to transmembrane adaptor proteins and downstream signaling pathways. The TCRζ subunit differs in its genetic organization, chromosomal localization, and domain structure from the CD3γ, δ, and ϵ invariant chains. Thus, in CD4 and CD8 T cells the 16-kDa TCRζ chain subserves a unique role in TCR/CD3 complex assembly by associating with newly synthesized hexameric αβγϵδϵ complexes resulting in the transport and expression of mature αβγϵδϵζ complexes to the cell surface. Furthermore, metabolic labeling and pulse chase experiments have demonstrated that de novo synthesis of TCRζ is rate-limiting, thereby regulating the amount of TCR/CD3 complex expressed at the cell surface. In NK cells, the TCRζ chain may form homodimers or heterodimers with the FcϵR common γ chain signaling subunit, which in turn may form signaling complexes with several activating receptors, including NK cell protein 46 (NKp46), NKp30, and the low-affinity IgG receptor FcγRIII (CD16). In light of these crucial functions the TCRζ chain could be viewed as a master regulator and sensor of innate as well as adaptive immune responses. It follows from this that aberrations in its expression or function should have profound effects on immune function. The events leading to the down-regulation of TCR/CD3 complex expression on serial ligation by peptide/MHC complexes are well documented. Less well understood are the molecular mechanisms for the sustained down-regulation of TCRζ observed in chronic bacterial, HIV, and mycobacterial infections, autoimmune diseases such as rheumatoid arthritis and systemic lupus erythematosus, and in solid tumor and hematologic malignancies (reviewed by Baniyash). That this phenomenon is associated with so many diverse pathologies has raised the possibility that down-regulation of TCRζ may serve to attenuate T-cell activation at sites of tissue damage, thereby limiting the effects of unbridled T-cell reactivity and pathologic effector responses. Whilst this model is entirely consistent with a central role for antigen in driving autoimmune inflammatory activity, it implies that chronically activated T cells rendered hyporesponsive to TCR engagement through loss of TCRζ expression should play a less prominent role in the established phase of disease. To gain further insight into the immunobiology of T cells expressing low levels of TCRζ (hereafter, TCRζ), we explored the relationship between loss of TCRζ expression and T-cell function. Our results shed light on the nature of persistent effector responses of T cells that become refractory to antigenic restimulation during the evolution of immune responses. They also have implications for cell-based therapies aimed at achieving long-term remission in patients with chronic inflammatory disease. Laboratory staff members, aged 22 to 45 years of age, were enrolled as healthy donors. All patients were recruited from the Hammersmith Hospitals NHS Trust (London, United Kingdom). CMV reactive lymphocytes were obtained from HLA-B*0702 patients with chronic myeloid leukemia selected by CMV seropositivity. Single donor plateletpheresis residues were obtained from the North London Blood Transfusion Service (Colindale, United Kingdom). Patients with severe, active seropositive rheumatoid arthritis (RA) meeting the American College of Rheumatology revised classification criteria for RA, and requiring anti-TNF therapy, were recruited to the study. Prior to anti-TNF therapy, all patients were treated with nonsteroidal anti-inflammatory drugs in addition to methotrexate and sulfasalazine, with or without hydroxychloroquine and low-dose prednisolone. Peripheral blood (PB) and synovial fluid (SF) lymphocytes were obtained from patients with RA, psoriatic arthritis, or reactive arthritis who attended the clinic with a flare of disease requiring therapeutic arthrocentesis. Synovial mononuclear cell suspensions were prepared from RA synovial tissue obtained at joint replacement surgery as described. Written, informed consent was obtained from all participating donors and patients in accordance with the Declaration of Helsinki, and all protocols were approved by the Riverside Research Ethics Committee, Hammersmith Hospitals NHS Trust. PB lymphocytes (PBLs) from heparinized peripheral venous blood (or mononuclear cells from SF) were obtained by Ficoll-Hypaque density gradient centrifugation (specific density 1.077 g/mL; Nycomed Pharma, Oslo, Norway) and used immediately for flow cytometric analysis. Where relevant, additional aliquots were prepared and frozen until required. PB monocytes and T cells were purified by elutriation (JE6; Beckman Coulter, Fullerton, CA) in 1% heat-inactivated fetal calf serum (FCS), as described, and purity assessed by flow cytometry. Single-cell suspensions of synovial membrane mononuclear cells were prepared as described. Cells were cultured in RPMI 1640 supplemented with 10% heat-inactivated FCS, 2 mM -glutamine, 100 U/mL penicillin/streptomycin, 50 μM 2-ME, 1 mM sodium pyruvate, and 25 mM HEPES (complete medium). The following antibodies were purchased from BD Biosciences PharMingen (San Diego, CA): anti–CD8-APC, anti–CD28-FITC, anti–CD45RA-APC, anti–TNFα-FITC, anti–IFNγ-FITC, and anti–IL-10–APC. Anti–CD45RO-FITC was obtained from Serotec (Oxford, United Kingdom), mouse IgG1-FITC and anti–CD3-FITC from Sigma (St Louis, MO), mouse IgG1-PE and anti–TCRζ-PE (TIA2, which recognizes a cytoplasmic domain epitope) from Immunotech (Coulter, Hialeah, FL), and mouse IgG–PerCP, anti–CD3-PerCP, anti–CD4-PerCP, anti–CD8-PerCP, anti–CD16-FITC, and anti–CD56-FITC from Becton Dickinson (San Jose, CA). Anti–TCRζ-FITC (clone 6B10.2, which recognizes a transmembrane epitope of TCRζ) was purchased from Santa Cruz Biotechnology (Santa Cruz, CA). The anti-CD3ϵ antibody used for stimulation (clone OKT3) was obtained from the American Type Culture Collection (Rockville, MD). Human recombinant TNF-α and IL-2 were generous gifts of Prof W. Stec (Centre of Molecular Studies, Lodz, Poland) and Dr P. Lomedico (Hoffman-La Roche, Nutley, NJ), respectively. All other cytokines were purchased from PeproTech EC (London, United Kingdom) or R&D Systems (Wiesbaden, Germany). Cells were stained by standard methods and analyzed using an LSR BD FACScan analyzer and CellQuest software (Becton Dickinson). For analysis of TCRζ and intracellular cytokine expression, cells were fixed with 2% formaldehyde and permeabilized in buffer containing 10 μg/mL saponin, according to established protocols. The efficiency of permeabilization was determined by uptake of trypan blue (always > 99%). For proliferation, unfractionated PBLs were labeled with 5 μM CFSE (Molecular Probes, Eugene, OR) prior to stimulation in complete medium with plate bound OKT3 (2 μg/mL) with or without soluble anti-CD28 (2 μg/mL). Cells were harvested at the indicated times, stained with specific antibodies to TCRζ (TIA2-PE), together with anti-CD3, anti-CD4, or anti-CD8 conjugated to PerCP and analyzed by flow cytometry. CMV-reactive PBLs were stimulated with specific pp65 CMV peptide (amino acid sequence: TPRVTGGGAM, from Proimmune, Oxford, United Kingdom) in the presence of 50 U/mL IL-2 and 10 ng/mL IL-7 for up to 8 days prior to staining with anti-TCRζ (6B10.2-FITC), HLA-B7-tetramer-PE (Proimmune), and anti–CD8-APC and analysis by flow cytometry. Cytokine-activated T cells were generated from elutriated T cells over 8 days using IL-2 (25 ng/mL), TNF-α (25 ng/mL), and IL-6 (100 ng/mL) as described. Cytokine expression was determined by staining fixed and permeabilized PBL with directly conjugated anticytokine antibodies (all from BD Biosciences PharMingen), after stimulation for 6 hours at 37°C with 20 ng/mL phorbol myristate acetate (PMA) and 200 ng/mL ionomycin (both from Sigma). Golgistop (containing monensin, from BD Biosciences PharMingen) was added for the last hour of stimulation. Specificity of staining was confirmed using isotype-specific control antibodies and by ligand-blocking experiments using the corresponding recombinant cytokine. For CD28 stimulation of TCRζ cells, T cells were cocultured with Chinese hamster ovary (CHO) cells transfected with CD80 or CD86 (generously provided by Dr D. Sansom, University of Birmingham, Birmingham, United Kingdom) for 24 hours, and supernatants harvested and assayed for IFN-γ or IL-10 expression by enzyme-linked immunoadsorbent assay (ELISA; BD Biosciences PharMingen). Cell contact-dependent activation of monocytes by TCRζ or TCRζ T cells fixed in PBS containing 0.05% glutaraldehyde and neutralized with an equal volume of buffer containing 0.2 M glycine was performed as described. Stimulation of monocytes with LPS (10 ng/mL) in the absence of T cells was used as a positive control. Culture supernatants were harvested after 24 hours and assayed for TNF-α levels by ELISA (BD PharMingen). Transendothelial migration of PBLs or elutriated T cells 24 hours after plating over monolayers of human umbilical vein endothelial cells (HUVECs) previously stimulated with 10 ng/mL TNF-α for 48 hours was performed essentially as described. Numbers of migrating TCRζ or TCRζ T cells were determined by flow cytometry. Patients with active RA requiring anti-TNF treatment as part of their standard care were identified according to national guidelines. All patients were treated with intravenous infusions of infliximab (3 mg/kg) at week 0, 2, 6, and then 8 weekly thereafter, while continuing existing disease-modifying drug therapy. At baseline and at the indicated times disease activity and response to therapy were determined using the disease activity score (DAS), which includes the 28 joint count (DAS28). At baseline, all patients had DAS28 scores of more than 5.1. Good, moderate, and poor responses were defined according to European League Against Rheumatism (EULAR) DAS response criteria at 14 and 30 weeks. PBLs were obtained at the indicated times after starting therapy and frozen prior to analysis. Differences in phenotype between TCRζ and TCRζ subsets were determined by pair-wise analyses using the Wilcoxon signed rank test or Student test, according to sample distribution. Correlations between changes in TCRζ expression and disease activity were determined by linear regression analysis. The limit for statistical significance was set at .05. To study TCRζ expression at the single-cell level in mixed populations of PBLs from healthy donors we adapted quantitative flow cytometric assays first developed to study deficient ζ-chain expression in T cells from patients with cancer. The TCRζ chain has a short 9 amino acid extracellular domain, so we used mouse monoclonal antibodies (clones TIA-2 or 6B10) that detect cytoplasmic domain epitopes of the ζ chain in T cells after fixing and permeabilization. Both antibodies recognize phosphorylated as well as unphosphorylated forms of the TCRζ chain, implying that any loss of signal is not due to activation-induced phosphorylation of the ζ-chain polypeptide (data not shown). As expected, staining of PBLs from healthy donors defined distinct lymphocyte subsets, including populations of CD3TCRζ cells, CD3TCRζ cells, and a double-negative population of monocytes and B cells that do not express TCRζ (A and data not shown). We documented a relationship between TCRζ and CD3ϵ expression, with TCRζ T cells expressing the highest levels of CD3ϵ, and TCRζ or TCRζ cells expressing levels of surface CD3ϵ that for some donors were at least an order of magnitude lower (A). This relationship varied considerably between donors but was not related to cell viability based on annexin-V staining (data not shown). Analysis of CD3TCRζ cells confirmed expression of TCRζ in both CD56 and CD16 NK cell subsets; CD3CD56 cells were found to be uniformly TCRζ, whereas the CD3CD16 NK cell subset was TCRζ (data not shown; Nicola Dalbeth and Margaret Callan, unpublished data, May 2005). This would be consistent with the physical association between CD16 and the ζ chain. The relationship between antigen engagement and reduced levels of TCRζ reported previously suggested that TCRζ expression should discriminate between subsets of naïve and memory T-cell subsets defined by conventional cell surface markers. To this end, we stained healthy donor PBLs with antibodies specific for CD45RA or CD45RO to define subsets of naïve and memory T cells, respectively, and then compared the expression of these markers on TCRζ and TCRζ cell subsets. Histogram plots from a representative experiment, shown in B, demonstrated enrichment of CD45RA T cells in the TCRζ population and many CD45RO T cells residing within the TCRζ subset. Nevertheless, there was considerable overlap between subsets, findings supported by analysis of multiple donors (C). Indeed, as many as 40% of TCRζ T cells expressed CD45RO, whereas a similar proportion of TCRζ T cells expressed CD45RA. More extensive phenotyping failed to demonstrate uniform loss of CD62L or CCR7 expression on all TCRζ T cells when compared to the TCRζ subset, indicating that low expression of TCRζ does not define differentiating effector memory T cells per se, but rather analysis of TCRζ expression might provide additional functional information besides that provided by conventional cell surface phenotyping. To examine the functional consequences of TCRζ expression on TCR responsiveness, we stained fresh PBLs from healthy donors with carboxy-fluorescein diacetate succinimidyl ester (CFSE) and tracked the proliferative capacity of CD4 and CD8 T-cell subsets in response to stimulation with OKT3, or OKT3 and anti-CD28 in combination. At the indicated times cells were harvested, stained for TCRζ expression, and analyzed by CFSE dye dilution. Although little proliferation was detected within the first few days of TCR stimulation, dividing T cells could be detected at subsequent time points, and these were uniformly TCRζ (A arrows). In contrast, the majority of TCRζ and TCRζ T cells failed to proliferate throughout the period of stimulation. Note that populations of nonproliferating TCRζ CD4 and CD8 T cells, representing less than 10% of the total, were observed at day 4 but not at subsequent time points. The reasons for this are unknown, although it is conceivable that these T cells undergo programmed cell death after nonproductive TCR engagement. By day 6, expression of TCRζ in proliferating cells declined progressively with each cycle of cell division. Although this result was predicted, the extent to which proliferation was restricted to a subset of CD4 or CD8 T cells expressing high levels of TCRζ was unexpected. These findings could not be explained by suboptimal levels of TCR stimulation, because higher concentrations of OKT3 and anti-CD28 antibodies induced similar patterns of proliferation with respect to TCRζ expression. It followed from this that after TCR stimulation in vitro, antigen-reactive T cells should reside in the TCRζ compartment. To test this directly, we analyzed fresh PBLs from HLA-B*0702 CMV-reactive donors after stimulation with specific CMV peptide and tracked antigen-specific CD8 T cells with fluorescent HLA-B7/p65 MHC-peptide tetramers. In a representative experiment, the proportion of tetramer-positive T cells increased from less than 3% prior to stimulation to almost 10% after 8 days of culture (B). By gating on TCRζ and TCRζ T-cell subsets, we confirmed that the majority of peptide-specific CD8 T cells reside within the TCRζ population after 8 days of stimulation (1.27% tetramer-positive TCRζ cells versus 26.4% tetramer-positive TCRζ cells, C); note the downward shift in tetramer staining consequent on down-regulation of the TCR complex after stimulation (B-C right panels). Together these experiments demonstrated that TCRζ staining can be used to identify populations of tetramer-positive T cells that have undergone productive engagement with specific antigen and suggest that circulating ζ cells represent polyclonal populations of antigen experienced lymphocytes in vivo. A characteristic feature of effector T cells is their capacity to produce cytokines. Current paradigms of T-cell differentiation suggest that T cells undergo a complex series of epigenetic events to acquire stable transcriptional competence for cytokine gene expression. These events depend on signals integrated from both the antigen T-cell receptor and cytokine receptors that induce the expression and activation of lineage-specific transcription factors. We therefore explored whether TCRζ T cells that had already engaged antigen in vivo were competent for cytokine gene expression when compared to TCRζ cells from the same donor. To test this, fresh PBLs were stimulated in vitro for 6 hours with phorbol ester and calcium ionophore. This stimulus, in contrast to anti-TCR antibodies, allowed us to assess competence for cytokine gene expression by activating T cells through pathways independent of TCRζ expression. Monensin was added to cultures for the last hour of stimulation prior to harvesting and staining using specific antibodies for CD3ϵ and TCRζ in combination with the indicated anticytokine antibody. Representative histogram plots are shown for each cytokine ( left panel), alongside data derived from 12 healthy donors ( right panel). The TCRζ subset (shaded histograms) was enriched for cells expressing TNF-α and IFN-γ when compared to the TCRζ population from the same donor (A-B). By contrast, the dominant IL-10–producing cell subset was found to reside within the TCRζ subset (C), indicating that cytokine expression in general was not confined to the TCRζ subset. A similar trend was seen for IL-4, but levels of expression were too low for a meaningful statistical analysis. The results described thus far indicated that circulating TCRζ T cells were enriched for cells that had previously engaged antigen and that were primed for cytokine production. However, the proliferative responses of T cells to TCR engagement shown in indicated that the TCRζ subset was refractory to membrane proximal TCR signals. To define alternative signaling pathways of effector function, we prepared bulk populations of TCRζ T cells by stimulating elutriated T cells with inflammatory cytokines such as TNF-α, which down-regulates TCRζ in murine T cells, using a combination of IL-2, IL-6, and TNF-α sufficient to support T-cell differentiation in vitro in the absence of specific antigen and accessory cells. After 7 days of cytokine stimulation a significant proportion of T cells were TCRζ (67.3% compared to 12.2% for unstimulated T cells; A). We then tested the capacity of TCRζ T cells to produce IFN-γ or IL-10 either in the absence of stimulation or following costimulation by CHO cell transfectants expressing CD80 or CD86; CHO cells expressing empty vector were used as a negative control. In the absence of CHO cells, stimulation with the cytokine cocktail alone was sufficient to induce IFN-γ production (B). Whilst costimulation with CHO/vector cells reduced IFN-γ production by TCRζ T cells, stimulation with CHO cells expressing either CD80 or CD86 was equally efficient in augmenting IFN-γ production. Neither cytokine nor costimulatory signals were sufficient to induce T-cell IL-10 production. To test whether TCRζ T cells could reciprocally activate monocytes, TCRζ or TCRζ T cells were fixed prior to incubation with elutriated monocytes and supernatants harvested 24 hours later and assayed for monocyte-derived TNF-α production. Monocytes or T cells alone produced little or no TNF-α, whereas LPS stimulation induced abundant TNF-α (C right panel). Although coculture of fixed TCRζ cells with monocytes failed to elicit TNF-α production (C left panel), cytokine-stimulated TCRζ T cells induced monocytes to produce TNF-α in a dose-dependent fashion when T cell–monocyte ratios were increased from 3:1 to 7:1. TNF-α production was attenuated when T-cell–monocyte contact was blocked with a Transwell insert (data not shown). These data demonstrated that although TCRζ cells are relatively refractory to TCR-induced proliferation, they are by no means senescent or inert, being capable of inflammatory cytokine expression in response to cytokine receptor or costimulatory signals. TCRζ cells are also potent activators of monocytes through cell contact-dependent pathways. This effector phenotype, being reminiscent of RA synovial T cells, prompted us to compare expression of TCRζ in paired PB and SF T cells from patients with an acute flare of inflammatory arthritis who attended a clinic. TCRζ expression profiles were notable in several respects. First, a significant proportion of SF T cells had negligible levels of TCRζ expression, equivalent to that detected in the CD3ϵTCRζ double-negative cell population, and yet surface expression of CD3ϵ was spared (A). Second, although the numbers of CD3TCRζ NK cells were lower in SF than in PB, expression of TCRζ was also reduced. Furthermore, lower numbers of TCRζ T cells were detected in PB of patients with an acute disease flare (B). This was associated with accumulation of the TCRζ subset in SF when compared to PB from the same patient ( < .03), suggesting selective migration of these effector cells to inflamed joints and their exit from the circulating pool; increased numbers of TCRζ T cells in RA synovial tissue (ST) was consistent (B). If circulating TCRζ T cells represented populations of antigen-experienced effector cells captured prior to their migration to inflamed joints, then blocking cell migration should lead to accumulation of this subset in blood. We took advantage of the fact that TNF blockade inhibits leukocyte trafficking by tracking PB TCRζ cells in 17 patients with chronic, active RA (DAS28 ≥ 5.1) before and after treatment with anti-TNF. We documented considerable variability in expression of PB TCRζ T-cell numbers following anti-TNF therapy. However, when data were stratified according to validated EULAR clinical response criteria we found that at 30 weeks the best clinical outcomes were associated with an accumulation in the percent CD3TCRζ T cells at 14 weeks (C); this relationship held true for CD4 but not CD8 T cells, whereas no such association was found between clinical response to anti-TNF and other cellular parameters, such as total lymphocyte counts, or the percent CD3 or CD45RO T cells (data not shown). We reasoned that accumulation and persistence of PB TCRζ effector T cells in anti-TNF responders might explain why disease flares occur on withdrawing anti-TNF therapy, after which TCRζ T cells would migrate back to the synovial joint compartment and reinitiate a local inflammatory response. Ethics precluded further investigation in patients, and so we opted to study transendothelial migration of TCRζ T-cell subsets in vitro. Preliminary experiments demonstrated that by 24 hours about 20% to 25% of PBLs from healthy donors migrate across TNF-stimulated endothelial monolayers cultured in the Transwell. Flow cytometric analysis demonstrated that cells migrating to the lower chamber expressed lower levels TCRζ. This was true for both T-cell and NK cell subsets (A), and the expression profiles for upper and lower chambers were quite comparable to those we had observed for PB and SF (A). Transmigration did not perturb TCRζ expression because the proportions of TCRζ and TCRζ T cells were similar to those in PB cultured for the same period in the absence of endothelial cells or Transwell, nor was it dependent on proliferation because CFSE fluorescence of migrating and nonmigrating T cells was comparable (data not shown). Even though there are many more TCRζ than TCRζ T cells in the population added to the upper chamber, a much higher proportion of TCRζ cells migrated when compared to their TCRζ counterparts (∼70% versus 30%), differences more striking for CD4 than CD8 subsets (B). Migration of PBLs could be further enhanced by the addition of CXCL-10, CCL5, or CXCL-12 to the lower chamber, chemokines known to be up-regulated in RA synovial joints (C). Subset analysis revealed enhanced migration of TCRζ T cells in the presence of CXCL-12 (D), whereas the migratory profiles of TCRζ cells were substantially higher and barely augmented in the presence of exogenous chemokine (D). Indeed, for the TCRζ subset transmigration in response to CXCL-12 was reduced when compared to cells migrating in the absence of exogenous chemokine. Taken together, these results suggest that antigen-experienced TCRζ T cells have likely acquired in vivo the propensity to migrate in part through expression of chemokines and their cognate receptors at the cell surface. Finally, we asked whether TCRζ cells accumulating in PB retain the capacity to migrate despite anti-TNF therapy. To this end, the migratory competence of PB cells obtained from RA patients with active disease at baseline was compared with cells acquired after 14 weeks of anti-TNF treatment. Although the results demonstrated a trend toward reduced migration of CD8TCRζ T cells and a significant reduction in migration of the CD3TCRζ NK cell subsets, the capacity of CD3 and CD4TCRζ T-cell subsets to migrate across activated endothelium in vitro was not influenced by anti-TNF treatment in vivo (). We believe that these results provide a rational explanation for disease flares associated with treatment withdrawal and have implications for both the design and monitoring of therapeutic regimens aimed at prolonged periods of biologic drug-free remission. It seems likely that T cells expressing low levels of TCRζ define polyclonal populations of antigen-experienced T cells in vivo based on their cell surface phenotype, tetramer staining, cytokine expression, proliferative responses, and migratory competence. Even though their accumulation in inflamed tissues appears consistent with this, we cannot yet account for the heterogeneity of this TCRζ population. This subset will include populations of T cells that have very recently engaged antigen, perhaps explaining their refractoriness to TCR restimulation, but also T cells in which loss of TCRζ expression persists over time. Our data seem consistent with this because during an acute inflammatory response TCRζ cells are depleted from the PB compartment and enriched in inflamed tissue; TNF blockade reverses this. Furthermore, transendothelial migration experiments demonstrated an enhanced propensity for circulating TCRζ T cells to migrate when compared to their TCRζ counterparts, implying that migration is associated with, and perhaps even facilitated by, loss of TCRζ expression in peripheral lymphoid organs. We suggest that on migration to tissues in vivo this subset would be exposed to an inflammatory milieu where hypoxia, reactive oxygen intermediates, depletion of essential nutrients, inflammatory cytokines, and activated cells of myeloid lineage each contribute to chronic down-regulation of TCRζ expression. Therein, aberrant TCR signals and their functional sequelae may be perpetuated. We still do not fully understand which extracellular cues sustain the effector function of TCRζ T cells and how these apparently dysfunctional cells contribute actively to pathogenic processes, besides the apparent imbalance of TCR-dependent arrest of cell motility and competing chemokinetic “go” signals promoting adhesion and migration. One possibility is that TCRζ-deficient TCR/CD3 complexes are capable of transmitting signals to induce and sustain inflammatory cytokine gene expression through compensatory up-regulation of the FcϵRIγ signaling subunit, as first described in intraepithelial lymphocytes of TCRζ-deficient mice. Indeed, T cells from TCRζ-deficient mice are fully competent to differentiate into IFN-γ–producing Th1 cells, a finding consistent with the enrichment for IFN-γ and TNF-α expression in human TCRζ T cells (). On the other hand, the relative absence of IL-10–expressing T cells in the same population could point to an intrinsic requirement for intact TCR signaling to maintain immune homeostasis through the generation or function of regulatory T-cell subsets, as suggested in studies of patients with diabetes treated with anti-CD3. Appropriate antigen-dependent expansion of adaptive CD4CD25CD62LFoxp3 regulatory T cells from a pool of differentiating effector T cells might fail for the same reason. An alternative possibility is that the chronic phase of effector function is maintained independently of antigen stimulation, as suggested by experiments in which TCRζ T cells produce IFN-γ following engagement of costimulatory pathways, while at the same time stimulating monocytes to produce inflammatory cytokines such as TNF through cell contact-dependent pathways. Moreover, attenuation of TCRζ-dependent signaling will perturb pathways of antigen-dependent activation-induced cell death. It follows from this that TCRζ T cells would persist in tissues over extended periods of time, further amplifying effector pathways mediated through cell contact with infiltrating macrophages, B cells, and resident stromal cells. Although such sustained T-cell effector responses might usefully serve to sustain immunity to foreign pathogens, they are clearly harmful in autoimmunity. A direct link between the loss of integrity of signaling from the TCR/CD3 complex and the generation of a repertoire of arthritogenic T cells has recently been reported in SKG mice. A spontaneous mutation in ZAP-70, which uncouples membrane proximal TCR signal transduction, impairs thymic selection leading to a peripheral repertoire of autoreactive T cells, which, among other tissues, targets synovial joints manifesting as a symmetrical and destructive polyarthritis. There are striking similarities between CD4 T cells from SKG mice and TCRζ T cells, not least of which being their profound hyporesponsiveness to TCR engagement, skewed T-helper cell differentiation favoring expression of TNF-α, IFN-γ, and (in our hands) IL-17, in combination with a robust capacity to activate monocytes. These T-cell–monocyte interactions likely play a major role in promoting the chronic, excessive cytokine drive characteristic of both autoimmune arthritis in SKG mice and RA. Given the strong predilection to migrate to inflammatory tissues coupled with their effector phenotype, TCRζ T cells should now be considered a valid cellular target for treating a wide range of chronic inflammatory diseases.

Our recent findings of a large variation in plasma membrane Ca pump (PMCA) activity of normal human red blood cells (RBCs) prompted the question of whether this variation was related to cell age and, if so, whether an age-related decline in PMCA activity might account for some observed changes in aging RBCs. The PMCA is the only active Ca transporter in human RBCs. At saturating [Ca] levels, its maximal Ca extrusion rate, , is about 10 to 15 mmol(340 g Hb)h, whereas the normal pump-leak Ca turnover is only about 50 μmol(340 g Hb)h. Thus, in physiological conditions the mean pump activity is a small fraction of its mean . This balance, however, is far from uniform among the RBCs, whose PMCA activity showed a broad unimodal distribution pattern, with a coefficient of variation of about 50%. The cells with the weaker Ca pumps would have higher [Ca] levels than those with stronger pumps. Because RBC [Ca] levels control several transport and biochemical processes affecting volume regulation, metabolism, and rheology, the resulting variations in [Ca] may generate marked differences in RBC composition. There were early indications that PMCA activity decreased with RBC age, but there has been no systematic study of the age-activity relation over the full range of pump variation observed. Here we devised a method to segregate RBCs according to their PMCA strength to be able to measure directly the age of the segregated RBCs. We used the fraction of glycated hemoglobin, Hb A1c, as a reliable measure of RBC age. Because protein glycosylation is a nonenzymic process linear with time and glucose concentration, the Hb A1c fraction is a reliable age marker for normal human RBCs, which, unlike those of diabetic patients, are exposed to relatively stable glucose levels in vivo. In the course of investigating the age-activity relation of the PMCA we observed an unexpected reversal of the main pattern of RBC behavior, as detailed below, which could be explained by the presence of a small fraction of aged RBCs unable to dehydrate by K permeabilization. Such RBCs had been described before, but their age had not been measured. We now demonstrate that they have a high Hb A1c content, suggesting that they represent a normal terminal condition of senescent RBCs. For a dense, low-Na, aging RBC to change to a lesser dense, higher-Na, terminal condition in the circulation, it must undergo a change in membrane permeability whereby net NaCl gain exceeds net KCl loss, leading to progressive net salt and water gain. A likely candidate for such a process was the nonselective cation permeability pathway originally described by Crespo et al as “prolytic” (following a nomenclature originally used by Ponder to describe effects of lysins on RBC homeostasis). The prolytic pathway exhibited a sufficiently high Na permeability to exceed the extrusion capacity of the Na pump, leading to the swelling and eventual lysis of the affected RBCs. We undertook a detailed investigation of the main properties of the “prolytic” pathway, which we designate here “Pcat,” applying methods that examine the responses of the entire RBC population to the factors that activate this pathway. p r o v a l f o r t h e s e s t u d i e s w a s o b t a i n e d f r o m t h e i n s t i t u t i o n a l r e v i e w b o a r d s o f t h e U n i v e r s i t y o f C a m b r i d g e , U n i t e d K i n g d o m , a n d f r o m t h e A l b e r t E i n s t e i n C o l l e g e o f M e d i c i n e , B r o n x , N Y . shows the patterns of yield recovery for RBCs that had fully extruded their Ca load (δ less than 1.117, harvested from the top of the DEP oil), or had not yet done so (δ more than 1.117, harvested from pellets), at each time interval after temperature activation of PMCA-mediated Ca extrusion. The results are typical of 9 experiments with blood from 6 donors. The selected results () exemplify the 2 patterns observed, differing mainly in the rate of increase in the yield of light RBCs (filled symbols). This rate reflects differences in the fraction of fast-pumping RBCs in samples from different donors, with faster-climbing yields of light RBCs indicating a greater fraction of high- cells. The time required for full Ca extrusion varied from 2 to 4.5 minutes in these experiments. Even at the earliest sampling time (30 seconds), the fraction of cells with δ less than 1.117 was never 0 but varied between 1.5% and 5%. With an initial Ca load of about 500 μmol(340 g Hb), the RBCs would require pumping rates of over 60 mmol(340 g Hb)h to fully extrude this load in less than 30 seconds. We see below that such high-pumping cells account for only part of the RBCs in these early light fractions. The relation between PMCA activity and RBC age is shown in . The percent Hb A1c is plotted as a function of the yields of light (δ less than 1.117) and dense (δ more than 1.117) RBCs. The 3 results shown illustrate the patterns observed in 9 experiments. All experiments showed a clear divergent pattern toward both higher and lower Hb A1c values at decreasing yields of dense and light cells, respectively, differing only in the extent of reversal in the Hb A1c decline of the light cells at the lowest yields of light RBCs. We analyze first the significance of the divergent patterns and then consider the reason for the reversal. The patterns in are precisely those expected from a progressive decline in the PMCA activity with RBC age. The RBCs harvested from pellets at the latest times after pump activation are those with the weakest pumps, represented by the points approaching the lowest yields (δ more than 1.117) with the highest Hb A1c values, indicating their older age. The quasilinear trend of increasing Hb A1c values of dense cells with decreasing yields shows that the decline in PMCA activity is an approximate linear function of RBC age. The mirror image variation in Hb A1c with yield of the light RBCs (δ less than 1.117) demonstrates the reciprocal trend, a progressive increase in PMCA activity with increasing RBC youth. The reversal of the downward trend in the Hb A1c values observed at the lowest yields of light RBCs (δ less than 1.117) was seen in all experiments, varying only in magnitude (compare A-C). A “contaminant” presence of high-Na“calres” RBCs, unable to dehydrate to densities above 1.117 g/mL by K permeabilization, was an obvious candidate to explain the observed reversal, if those RBCs had high Hb A1c levels. Measurements in 14 blood samples from 7 different donors showed Hb A1c levels (mean ± SEM) of 8.19% ± 0.9% and yields of calres RBCs of 3.2% ± 1.1%. The mean Hb A1c in the unfractionated controls was 5.77% ± 0.06%. The Na content of calres cells was measured in 5 samples from a single donor from blood drawn on different days; the Na content of the calres cells was (mean ± SEM) 51.6 ± 4.3 mmol(340 g Hb), whereas that of the unfractionated cells was 10.0 ± 0.1. Thus, calres RBCs from healthy donors had the high Hb A1c and Na contents, which could explain the observed reversal as a variable mix of low and high Hb A1c cells. Together with the early observations, the present results show that senescent RBCs reverse their progressive densification process toward a low-density, low-K, high-Na condition, requiring a change in their normal cation permeability properties. In preliminary experiments with the profile migration method, we confirmed many of the results of Crespo et al: the elevated permeability of Pcat to Na and K and the lack of effect of many transport inhibitors (ouabain, amiloride, bumetanide, furosemide, tetrodotoxin, chlorpromazine, trifluoroperazine, oligomycin, mepacrine, and tetracaine). We also found that quinine was partially inhibitory (50% inhibition at 1 mM), that clotrimazole and vanadate were slightly stimulatory, and that iodoacetic acid and N-ethyl maleimide (NEM), known inhibitors of a nonspecific, voltage-dependent channel of RBC membranes (NSVDC), had no effect. These results are not detailed here. We focus rather on results that provide additional new information about Pcat properties. A illustrates the screening test applied to expose Pcat using the profile migration protocol. A sudden Ca load produced uniform dehydration of the RBCs, seen as a rapid, marked, and largely parallel left shift in the osmotic fragility curve; after 10 minutes, the RBCs have attained near-maximal dehydration. If they retained their normal Na permeability, they would remain in this dehydrated state indefinitely, as seen in D, in which 120 mM external NaCl was isosmotically replaced by 240 mM sucrose. In high-Na media (A), however, the osmotic fragility curves showed a slow and persistent right shift, eventually leading to the lysis of many RBCs. Unlike the largely uniform initial left shift, the right shifts occurred with progressively lower slopes of the osmotic fragility curves; thus, all the RBCs underwent rehydration, but their rehydration rates showed a continuous variation, with no sharp discontinuities. Thus, the extent of Pcat activation was variable among the RBCs, causing some to lyse in 1 to 3 hours and others after 20 hours (A-C, E-G). The clear distinction between dehydration and rehydration processes shown here in the presence of SCN could not be seen so distinctly in the absence of SCN, when dehydration, rate-limited by the Cl permeability, was much slower. However, the long-term rehydration pattern was identical with and without SCN (not shown), indicating that prolonged exposure to 10 mM SCN had no effect on the rehydration response. To investigate whether the observed variations in Pcat were related to RBC age, Hb A1c was measured in RBCs exposed to the dehydration-rehydration protocol of A before and after prolonged incubations (6 to 22 hours). If RBC age correlated with the rehydration rate, the Hb A1c levels of the unlysed cells recovered after prolonged incubations would have differed from the original mean level. In 4 experiments with blood from 3 donors, the differences between mean Hb A1c percents in controls and unlysed cells recovered at late times, after lysis of 25% to 42% of the cells, were 0, 0.1, 0, and -0.1, averaging 0.00% ± 0.04% (SEM). Thus, at least for most of the RBCs, age was not a significant determinant of the observed differences in hydration rate and hence of Pcat activation. These results do not exclude the possibility that small subpopulations of aged or young RBCs were among those lysing first due to a high Pcat response. Comparison of A and B shows that after similar dehydration the rate of rehydration was much faster when intracellular Ca was increased. Rapid extraction of the RBCs' Ca load by addition of Na-EGTA in large excess over [Ca] at t = 10 minutes, after the cells had dehydrated, markedly reduced the rate of rehydration in both high-Na (B) and high-K media (C) but did not inhibit it completely. B also illustrates the response of RBCs dehydrated with valinomycin in the absence of Ca loads. Thus, the minor level of Pcat activation observed after Ca extraction cannot be attributed to a residual irreversible effect of a previous Ca load. The similarity of dehydration-induced rehydration rates of Ca-free cells in high-Na (B) and high-K (C) media suggests that the Pcat-mediated permeability for these 2 cations is similar. Microscopic observation on timed samples of profoundly dehydrated RBCs incubated in culture media at 37°C after ionophore and Ca extraction showed slow and uneven rehydration, with many initially crenated RBCs regaining biconcave or spheroidal shapes. In all these experiments, rehydration occurred in the absence of contaminant white cells and platelets. In the experiments of E-H, the RBCs were suspended in 90K-SCN medium, which prevented their dehydration following Ca loads (E-G) or valinomycin treatment (H). All Ca-loaded RBCs showed a hydration response that was not significantly modified by charybdotoxin (F) or ouabain (G), indicating that inhibition of Gardos channels or Na pumps had minimal effects on this response. On the other hand, the valinomycin-treated RBCs showed a barely detectable hydration response (H). The results in E-H thus show that Pcat can be activated by increased [Ca] alone, without dehydration, and that K permeabilization itself has no effect on Pcat. In the conditions of A, rehydration started with the cells initially profoundly dehydrated. When cells dehydrate by the net loss of KCl, the concentration of Cl in the isotonic effluent (about 150 mM) is higher than in the cells (about 100 mM). The resulting dilution of the intracellular Cl concentration increases the [Cl]/[Cl] ratio, and the anion exchanger rapidly restores the equality [Cl]/[Cl] = [H]/[H] with consequent intracellular acidification. In addition, the increase in impermeant anion concentration in the dehydrated RBCs hyperpolarizes the cells. Therefore, rehydration in the conditions of A starts with the cells dehydrated, acidified, and hyperpolarized. On the other hand, in the initial conditions of E, cell volume, pH, and membrane potential were all at physiological levels at the start of hydration. Because hydration proceeded comparably in both conditions, Ca activation of Pcat appears not to be much affected by dehydration, pH, or membrane potential, although subtle effects cannot be ruled out from these data. The measured changes in mean RBC Na and K contents during the dehydration-rehydration protocol of A are shown in A. The RBC K content fell sharply in the first 10 minutes and leveled off by 90 minutes, whereas the rate of Na gain was fairly constant throughout rehydration. These are average values for the RBC population as a whole and do not reflect the heterogeneity of the response, clearly discernible from the changing slope of the right-shifting osmotic fragility curves during rehydration (A). From the data in A it is possible to obtain a rough estimate of the value of the Na permeability, P, through Pcat, an important parameter for comparison with other permeation pathways in the RBC. Computation of the mean P through Pcat can be done by simulating the experimental conditions with a model of proven reliability and by searching for the P-Pcat values that would fit best the experimental data in A. The computations were performed with the Lew–Bookchin model, and the results are shown in B. The P-Pcat value required to generate the agreement apparent in the comparison of A and B was 0.015 h, a relatively minor 10-fold increase in the normally low electrodiffusional component of the RBC Na permeability, a value 3 orders of magnitude lower than the electrodiffusional K permeability through Ca-saturated Gardos channels and within the same order of magnitude as the cation permeability generated by the interaction between Hb S polymers and the membrane of sickle RBCs, “Psickle” (reviewed by Lew and Bookchin). The experiment of shows the effect of increasing Ca concentrations on the rehydration response. A shows the volume distribution of ionophore A23187-treated RBCs in media with EGTA and no added Ca and therefore no activation of Gardos channels. It can be seen that when [Ca] is not elevated there is a barely detectable right shift of the osmotic fragility curves with time. Increasing Ca concentrations triggered increasing rehydration responses (B-D) reflecting higher levels of Pcat activation. The range of [Ca] values over which Pcat activation increases in all the RBCs was far above physiological [Ca] levels; at the [Ca] levels of 50 μM used in most of the experiments reported here, the Pcat response was below its maximal Ca-saturated value. The cation selectivity of Pcat was investigated by isosmotic replacement of NaCl (A, LK) with choline chloride (B, choline) or with N-methyl-D-glucamine chloride (NMDG) (C). Rehydration occurred in all conditions, indicating that choline and NMDG were permeable through Pcat. In 1 experiment we investigated the Mg permeability through Pcat. After Ca loading, the ionophore A23187 was extracted with albumin at 4°C and the RBCs resuspended in iso-osmotic MgCl with 10 mM NaSCN and 1 mM vanadate to inhibit the PMCA and retain the Ca load. The results showed an almost identical rehydration pattern as that of C, indicating that Pcat is permeable to Mg. These results suggest that Pcat has a wide ionophoric path in the open state with poor cation selectivity. The present results demonstrate that the activity of the plasma membrane Ca pump declines monotonically with RBC age, that the subpopulation of high-Na“calres” RBCs represents old RBCs near the end of their circulatory life span, and that RBCs have a poorly selective cation permeability pathway, Pcat, activated primarily by elevated [Ca]. These results have important implications for understanding the changes in RBCs during physiological senescence and also for the interpretation of the statistical distributions of some classic RBC indices widely used in clinical laboratory tests. The mechanism of programmed RBC senescence has been the subject of intense research for a few decades. Experimental evidence strongly supports terminal macrophage recognition and removal of RBCs following well-characterized antigenic changes in outer surface protein domains. There is also abundant evidence of gradual age-related RBC changes, including reduced enzymic activities, cross-linking of cytoskeletal components, cumulative oxidative damage, and loss of lipid asymmetry. The generally increasing density of aging RBCs has long been recognized, and density fractionation has been used extensively to separate age cohorts. Yet the mechanism of densification has seldom been addressed, and little is known about the composition of aging RBCs. The condition documented here for the high-Hb A1c calres RBCs is one of increased Na permeability, not the often-suggested terminal condition of profoundly dehydrated RBCs. The present findings of a monotonic decline in PMCA activity with RBC age and of the advanced age of calres RBCs suggest a working hypothesis for the mechanism of the homeostatic changes in aging RBCs: the decline in PMCA activity would drive the pump-leak Ca balance toward higher [Ca] levels, which when exceeding the threshold activation of Gardos channels would induce a gradual net loss of KCl and water, leading to progressive RBC densification. With PMCA variations of, for example, 60 and 2 mmol(340 g Hb)h, estimated pump-leak steady-state [Ca] levels may vary from about 15 to 80 nM in young and old RBCs, respectively, leading to low-level Gardos channel activation in aging RBCs. Toward the final days in the circulation, the progressively increased [Ca] levels, together with RBC dehydration, would activate Pcat, allowing the influx of Na at a rate exceeding the balancing capacity of the Na pump, eventually leading to net NaCl gain in excess of KCl loss and osmotic cell swelling toward a low-density terminal state (similar to the density of reticulocytes). Valres and calres RBCs, with elevated Na contents, were originally detected in density-fractionated samples of sickle cell anemia and normal blood, sharing the same light-density fraction as reticulocytes of 1.070 to 1.090 g/mL. This is well beyond the prelytic density range of 1.050 to 1.060 g/mL, suggesting that intravascular lysis may not play a significant role in the normal circulatory removal of RBCs. However, the actual density spread of circulating calres RBCs deserves further investigation. Elevated [Ca] was shown to be a much more powerful activator of Pcat than RBC dehydration, which, without elevated [Ca], was only a weak stimulus of Pcat. Permeation of very large cations such as NMDG () suggests that Pcat presents a large-conductance, wide ionophoric path in the open state. At uniformly high Ca loads, Pcat was activated in all the RBCs of each blood sample but to different extents, with a unimodal pattern of variation. This variation was not age-related for the bulk of the RBC population, but the precision limits of the measurements did not exclude enhanced Pcat sensitivity in up to 5% to 6% of RBCs. Pcat activation was observed in the absence of contaminant white cells or platelets and is therefore strictly a RBC response. Although Pcat has all the attributes required to explain the density reversal process leading to calres-valres cell formation, there is no conclusive evidence for its activation in vivo. The conditions that activate Pcat experimentally, elevated [Ca] and RBC dehydration, do occur in vivo to an extent that would appear sufficient to elicit a very slow net sodium gain and density reversal, with the oldest RBCs reaching a light terminal density similar to that of reticulocytes. Finally, let us consider the implications of RBC densification with age to the statistical distribution of the hemoglobin concentration (HC) in normal RBCs. The distributions of Hb contents (CH) and volumes (V) of human RBCs have coefficients of variation of about 13%. If these 2 distributions were independent, the coefficient of variation (CV) of the HC, representing the ratio of Hb content to volume in each RBC, would be much larger than 13%. But the measured CV of the HC is only about 7%, indicating the biologic priority of uniformity of HC during RBC production. The measured CH distribution probably changes little after RBC release from the bone marrow, with perhaps minor Hb loss by exovesiculation. On the other hand, the measured V distribution is a composite of its original distribution at birth and its progressive reduction as the RBCs become increasingly dense with age. Therefore, the V and HC distributions of day-1 RBC cohorts must be much narrower than those measured for the whole RBC population comprising about 120 different day cohorts. This conclusion is supported by early flow cytometric measurements of HC distributions on density-separated RBCs, in which the CV of each one of 8 density fractions was about 3% (Table 2, column 2). Because each fraction represented a roughly 15-day age cohort, this result suggests that the coefficient of variation of the HC distribution of RBCs released from the bone marrow must be much less than 3%, an amazing degree of uniformity for a process delivering 10 billion RBCs per hour to a healthy adult's circulation.

Since the discovery that it acts as an endothelium-dependent vasodilator [], acetylcholine has been used to document the presence and degree of endothelial cell dysfunction in a variety of disease states. In the context of atherosclerotic disease, intracoronary injection of acetylcholine is associated with impaired vasodilation or frank vasoconstriction, leading to the suggestion that it serve as a test for early detection of coronary artery disease [–]. Likewise, the vasodilatory response to brachial or femoral artery injection of acetylcholine is compromised in the presence of coronary artery disease [] and hypertension []. These approaches have been used to track attempts to improve endothelial function in patients with coronary disease [] and peripheral-arterial disease [–]. However, the need for an intra-arterial injection has led such tests to be primarily confined to research settings. In an effort to avoid the invasive nature of intra-coronary, intra-femoral, or intra-brachial injection, investigators have sought other means of acetylcholine delivery. Laser Doppler flowmetry (LDF) of the microvasculature has identified vasodilation in response to iontophoretic application of acetylcholine in healthy subjects and documented that this response is compromised in patients with diabetes []. While iontophoresis offers a less invasive method of drug delivery than intra-arterial injections, it, too, is far from ideal. In addition to the slight risk of electrical burn from the technique [], electric current itself has been shown to be a potent vasodilator [], whose effects are impaired in some vasculopathic subjects []. Even when data are modified to account for a potential current-induced effect, the current remains a potential confounder in such studies []; its effect varies among specific vehicles and drugs as well as spatially over the area of drug delivery [,]. In order to avoid these limitations, we prepared and tested a means for LDF monitoring of acetylcholine responses via a translucent, non-iontophoretic, transdermal delivery system. In addition, a commercially available translucent patch for delivery of nitroglycerin (Minitran™ Patch, 3M, Minnesota) was adapted for LDF monitoring of local changes in microvascular perfusion without remote or systemic effects. The primary endpoint of the present investigation was whether non-iontophoretic topical application of acetylcholine would induce local vasodilation greater than placebo. We also sought a preliminary understanding of whether the acetylcholine-induced vasodilation was comparable to that achieved with nitroglycerin. To our knowledge, this is both the first report of the use of non-iontophoretic translucent patches for continuous LDF monitoring of drug effects during topical application and the first published report of the non-iontophoretic, transdermal effects of acetylcholine on the microcirculation. With Institutional Review Board (IRB) approval, 10 healthy volunteers were recruited for LDF measurements of forehead perfusion at sites of transdermal application of acetycholine, nitroglycerin, and placebo. For each session, after informed consent was obtained, subjects lay on a bed in a semi-recumbent position in a temperature-controlled room (22 ± 1°C). A three-lead electrocardiogram and a non-invasive brachial artery blood pressure cuff were applied. Local forehead skin oils were decreased by wiping lightly with an alcohol swab, followed by wet and dry gauze. Per agreement with our IRB, the application of acetylcholine to the highly vascular forehead in this preliminary investigation was limited to doses of acetylcholine that did not exceed 10 percent of that used safely in other clinical settings. We initially tested 10 percent of the standard 20 mg dose of an acetylchloline-containing preparation (Miochol-E™, Norvartis Ophthalmics, East Hanover, New Jersey) that is used to induce pupillary constriction during cataract surgery. While initial (unpublished) trials by our investigative team found that a 5 percent solution (20 mg/0.4 ml) of Miochol-E™ induced a 100 percent to 300 percent increase in local blood flow after topical application to the forehead under an LDF probe, the presence of mannitol in the Miochol-E™ preparation may have affected local vascular changes. This prompted our decision to dissolve pure acetylcholine chloride powder in water. To achieve a concentrated solution for delivery of a dose ≤2 mg (≤10 percent of that in the Miochol-E™), 100-mg acetylcholine chloride (Spectrum Chemical, New Brunswick, New Jersey) powder was freshly mixed with 0.6 ml high pressure liquid chromatography (HPLC) grade water to a final volume of approximately 0.7 ml. Then 0.02 ml of this 143 mg/ml acetylcholine solution (containing 2.86 mg of drug) was spread on a double stick clear disk (diameter: 12.6 mm, area: 124.7 mm). The study sites were delineated by placing double-stick discs (3M Health Care, Neuss, Germany) with an overall diameter of 38.1 mm and a central hole diameter of 8.7 mm on the forehead. These discs served as an adherent base onto which laser Doppler probes were placed, as well as to define the area (59.4 mm) for drug application. Drug delivery was limited to the area over the central hole of the double-stick disc (59.4 mm), or approximately one-half of the total drug in solution. The anticipated dose of 1.43 mg was ≤10 percent of the 20 mg dose used for intraocular injection and allowed for safe, unanticipated delivery of a slightly greater quantity of drug. The dry side of the acetylcholine patch was adhered to the end of an LDF probe (PF 5010 with Probe Model 407, Perimed, Järfälla, Sweden). A sham drug patch was made in the same way with 0.02 ml of HPLC-grade water instead of acetylcholine solution. For transdermal delivery of nitroglycerin, we used a portion of a commercially available translucent nitroglycerin patch (Minitran™ nitroglycerin patch, 3M, St. Paul, Minnesota) which delivers at a homogenous rate of 0.03 mg/hr/100 mm. A section of a standard 20 cm patch (designed to deliver a clinical dose at a rate of 0.6 mg/hr) was obtained by using a standard hole-punch to cut a 6 mm diameter (28.3 mm area). The delivery of 0.008 mg/hr from this reduced area was well within the 10 percent safety limit. The patch then was placed onto a double-stick disc and placed over an LDF probe in a similar manner to the acetylcholine patch. The three LDF probes with patches attached were placed on a double-stick disc at three forehead sites so as to enable undisturbed monitoring and drug delivery. LDF monitoring was performed continuously at each site for a maximum of 20 minutes or until a vasodilatory plateau was maintained for ≥3 minutes. Data were collected using Chart for Windows (ADInstruments, Colorado Springs, Colorado). All LDF sensors were calibrated using a motility standard (Perimed, Sweden) twice during the three-month study period. The local microvascular effects of transdermal acetylcholine and nitroglycerin were tested in 10 healthy volunteers (seven males, three females; mean age: 36.1; range: 19 to 56 years) without known vascular disease. An investigator blinded to the status of the subject and to the nature of the study site assessed the laser Doppler tracing at each study site. Our trials demonstrated that within 10 seconds of probe application, a steady baseline interval was consistently obtainable and that a progressive drug-induced rise in flow with both acetylcholine and nitroglycerin was noted to begin no sooner than 30 seconds after probe application. This enabled a 20-second baseline period to be obtained, without the need to remove the probe for subsequent drug application. The drug-induced increase in flow in the present study was determined by comparing the mean during the lowest 10-second interval during the baseline period with the mean during the maximum 10-second interval after a plateau was reached. If no distinct rise in blood flow was evident, then the highest 10-second interval that occurred beyond the first two minutes of readings was used. All data are presented as the mean with 95 percent confidence intervals (CI). Since the primary goal was to test whether the means of delivery and testing introduced herein generate and delineate a significant vasodilatory response compared to placebo, the primary endpoint was the percent change in LDF voltage after transdermal application of each active drug vs. placebo in the study population. Data were analyzed with the Wilcoxon Signed Rank Test (WSRT), using SPSS for Macintosh (SPSS Inc., Chicago, Illinois). Power analysis was based on reports that iontophoretic delivery of acetylcholine generated increases in skin blood flow ranging from 160 percent to 710 percent, depending on concentration of drug and iontophoretic current []. Given these data and our preliminary trials, we anticipated at least a 100 percent increase in blood flow (with a standard deviation of 50 percent) at the sites of active patches vs. a 20 percent difference (with a standard deviation of 10 percent) at sites of placebo application. In order to identify the difference between the active and placebo patches with an alpha of 0.025 (simple Bonferroni correction of p = .05 given two comparisons) and a beta of 0.8, we calculated that a sample size of seven subjects would be required. For values < 0.05, the actual value is reported; p values greater than 0.025 were considered not significant. With the realization that our study design was not powered to assess inter-drug comparisons, a secondary endpoint was assessed. The effects of the single dose of acetylcholine were compared to the single dose of nitroglycerin in the healthy volunteers, with the differences analyzed using WSRT. Acetylcholine and nitroglycerin both induced a marked rise in LDF voltage within two minutes of drug application in each subject. This was evident not only with respect to mean flow, but also with respect to the amplitude of the pulsation coincident with each heart beat. Mean blood flow readings increased by 406 percent (245 percent to 566 percent) and 36 percent (26 percent to 46 percent), respectively, at the acetylcholine and placebo sites (p = .005 by WSRT for acetylcholine vs. placebo); and they increased by 365 percent (179 percent to 550 percent) at the nitroglycerin site (p = .005 by WSRT nitroglyerin vs. placebo; p = NS nitroglycerin vs. acetylcholine) []. Mean baseline pulse amplitude at the acetylcholine site was 0.13 V (0.06 to 0.19) at baseline vs. 0.43 V (0.25 to 0.61) during drug effect (p = .005 by WSRT). For nitroglycerin, baseline pulse amplitude was 0.09 V (0.05 to 0.13) vs. 0.25 V (0.13 to 0.36) during drug effect (p = .005 by WSRT, p = NS for nitroglycerin vs. acetylcholine). Mean placebo pulse amplitude did not change significantly, going from 0.17 V (0.09 to 0.24) to 0.16 V (0.06 to 0.26) during drug effect (p = NS by WSRT) []. The lack of significant systemic effect was suggested by the lack of vasodilation at the placebo site; mean post-drug and baseline flow and amplitude values did not change significantly. The slight increase in flow at the placebo site seen in is attributable to our method of comparing the highest 10 second interval after drug application with the lowest 10 second interval at baseline. LDF readings have some temporal variation even under baseline conditions, a feature that would be emphasized by looking at maximal differences. A clinically trivial drop in mean blood pressure was recorded from 81 mmHg (75 to 86) at baseline to 78 mmHg (73 to 83) after drug absorption (p = .03 by WSRT). There was no significant change in heart rate: 1.00 Hz (0.90 to 1.11) at baseline vs. 1.04 Hz (0.94 to 1.15) after drug absorption (p = NS). We have demonstrated that microvascular dilation comparable to that achieved with iontophoretic delivery [,] can be induced via topical application of acetylcholine and nitroglycerin and that such effects easily can be monitored via LDF directly over a translucent delivery vehicle. The local vasodilatory effect of transdermal nitroglycerin has long been appreciated; to the best of our knowledge, a comparable response to non-iontophoretic transdermal acetylcholine has not been reported. Its rapid breakdown by plasma cholinesterase makes it well-suited to transdermal application for local testing of the microvasculature as well as possible vasodilatory therapy without significant systemic effect. The robust response to non-iontophoretic acetylcholine delivery in the present study may be attributable, in part, to our choice of the forehead. Parasympathetic innervation is more prominent in this central region than in the vasculature of the periphery [–], and the vasodilatory response to iontophoretic delivery of acetylcholine has been shown to decrease in the distal periphery as compared to more proximal sites []. Prior research by our team [,] and others [] noted that perfusion in central skin such as that of the forehead tends to be maintained in the context of vasoconstrictive challenges that cause decreased perfusion in the periphery. It has been suggested that this may be a consequence of homeostatic cholinergic mechanisms designed to preserve organ blood flow [,]. It should be noted that the patch technique described herein may be prone to the spatial and temporal variation that characterizes LDF readings in general. The translucent delivery systems enabled us to obtain a short baseline reading in the initial period of drug application, thereby avoiding an impact of spatial variation on baseline-to-drug comparisons. However, the technique would be improved by a delayed-release patch that would allow for a more prolonged baseline measurement. With the LDF probes employed in this study, it has been determined that one to three arteriolar-capillary networks are typically interrogated in the approximate 1 mm volume of tissue that is monitored []. Spatial variability may be reduced with a larger LDF probe and local mapping with a moving probe [] or a laser Doppler scanner. Whether the test described will be useful for microvascular assessments of the endothelium or for therapeutic induction of local vasodilation remains to be determined. With further elaboration, this technique may constitute a useful and minimally invasive way to interrogate the microvasculature in general, including its responses in various disorders and the microcirculatory changes induced by therapeutic interventions.

Standard pellet diet and oil-free diet were purchased from Pranav Agro Industries Ltd., Pune, India. Groundnut oil (SVS groundnut oil, trade name) was purchased from the local market. The oil was stored in a jar at 4 °C. The standard pellet diet contained 8 percent oil. Since the oil-free diet contained 2 percent oil, 94 g of this diet was mixed with 6 g of groundnut oil to make an 8 percent oil content. Streptozotocin was procured from Sigma-Aldrich, St. Louis, United States, and glibenclamide from Hoechts, Frankfurt, Germany. All other chemicals used were of analytical grade and obtained from E. Merck, Darmstadt, Germany, and Hi-media, India. Adult albino female Wistar rats with body weight of 180 to 200 g bred in Central Animal House, Department of Experimental Medicine, Rajah Muthiah Medical College, Annamalai University, were used in this study. The feed and water were provided ad libitum to the animals. The animals were made diabetic with an intraperitoneal injection of STZ at a dose of 40 mg/kg b-wt dissolved in citrate buffer (0.1 M, pH 4.5). STZ-injected animals exhibited massive glycosuria and hyperglycemia within a few days. Diabetes was confirmed in the overnight-fasted rats by measuring blood glucose concentration 96 hr after injection with STZ. The rats with blood glucose above 240 mg/dl were considered to be diabetic and used for the experiment. Blood glucose was estimated by the method of Sasaki et al. []. Hb and HbA were estimated by the methods of Drabkin and Austin [] and Sudhakar and Pattabiraman [], respectively. The activities of glucokinase, glucose-6-phosphatase and fructose-1, 6-bisphosphatase were assayed by the methods of Brandstrup et al. [], Koide and Oda [], and Gancedo and Gancedo [], respectively. Plasma vitamin C, vitamin E, GSH, TBARS, and HP were estimated by the methods of Roe and Kuether [], Baker et al. [], Ellman [], Nichans and Samuelson [] and Jiang et al. [], respectively. TC [], HDL-C [], and TG [] were measured. LDL-C was calculated by Friedwald's formula []. Results were expressed as means ± SD, for six rats in each group. Data were analyzed using one-way analysis of variance (ANOVA), and group means were compared with Duncan's Multiple Range Test (DMRT) [] using SPSS-10. Table 1 shows the effect of groundnut oil on body weight and blood glucose in normal and STZ-diabetic rats. A significant weight loss was observed in the diabetic control group. The weight loss was minimal in the oil-treated group, but significant improvement of weight was observed in the group treated with standard drugs. A drastic increase in blood glucose level was found in the diabetic control group. A small but significant reduction in blood glucose level was found in the diabetic animals fed groundnut oil. The reduction was highly significant in the drug-treated group. Table 2 illustrates the effect of substitution of oil on Hb and HbA levels in normal and STZ-diabetic rats. Hb level decreased significantly, while an increase in HbA was observed in diabetic rats when compared with normal rats. Significant improvement in Hb levels and a decrease in HbA were observed in diabetic animals treated with groundnut oil and glibenclamide, though it was more prominent in the drug-treated group. Table 3 shows the effect of groundnut oil on carbohydrate metabolic enzymes in tissues of normal and STZ-diabetic rats. The hexokinase activity decreased, whereas the activities of gluconeogenic enzymes such as glucose-6-phosphatase and fructose-1, 6-bisphosphatase increased in the liver and kidney of diabetic rats when compared with normal rats. A significant increase of hexokinase and reduction in glucose-6-phosphatase and fructose-1, 6-bisphosphatase activities were observed in the groups treated with groundnut oil and glibenclamide when compared with diabetic controls. Table 4 illustrates the effect of groundnut oil on TBARS and HP in the plasma of normal and STZ-diabetic rats. A significant increase in TBARS and HP was observed in diabetic rats when compared with normal rats. TBARS and HP decreased significantly in the plasma of diabetic rats fed with groundnut oil and treated with glibenclamide when compared with diabetic control rats. The decrease in the levels of TBARS and HP was more remarkable in glibenclamide-treated rats. Table 5 shows the effect of groundnut oil on antioxidants in the plasma of normal and STZ-diabetic rats. A significant reduction of plasma vitamin C and GSH and increase of vitamin E were observed in diabetic rats when compared with control rats. Significant increase in vitamin E and GSH were found in diabetic rats fed with groundnut oil when compared with diabetic rats. The levels of vitamin C and GSH significantly increased while vitamin E decreased in glibenclamide-treated rats when compared with diabetic controls. An increase in vitamin E also was observed in normal rats fed with oil when compared with the control group. Table 6 shows the effect of groundnut oil in the diet on plasma TC, VLDL-C, LDL-C, HDL-C, TG, and TC/HDL-C ratio in STZ-diabetic rats. In our study, diabetic rats had elevated levels of TC, VLDL-C, LDL-C, and TG and decreased levels of HDL-C when compared with normal rats. Diabetic rats fed with groundnut oil showed a small but significant reduction in the levels of TC, VLDL-C, LDL-C, and TG and elevation in HDL-C level when compared with diabetic controls, but diabetic animals treated with glibenclamide showed better improvement. STZ is a commonly employed compound for the induction of diabetes mellitus in experimental rats []. It causes DNA strand breaks in pancreatic islets, stimulates nuclear poly (ADP-ribose) synthetase, and thus depletes the intracellular NAD⁺ and NADP⁺ levels, which inhibits proinsulin synthesis and induces diabetes []. The decrease in body weight in diabetic rats shows that the loss or degradation of structural proteins is due to diabetes, and the structural proteins are known to contribute to the body weight []. In diabetic rats fed with groundnut oil, the weight loss was minimized, which may be due to the reduction of blood glucose. Groundnut oil-fed diabetic rats showed a significant reduction in blood glucose level. The reduction may be due to the presence of MUFA. Rasmussen et al. have reported reduction in peak plasma glucose concentration with the consumption of a MUFA-rich diet []. An earlier report from our laboratory shows that dietary substitution of sesame oil showed a better reduction of blood glucose 322.61 ± 9.49 to 222.02 ± 8.27) than groundnut oil in STZ-diabetic rats []. Insulin generally has an anabolic effect on protein metabolism in that it stimulates protein synthesis and retards protein degradation []. Previous reports have shown that protein synthesis is decreased in all tissues due to decreased production of ATP and absolute or relative deficiency of insulin [], which may be responsible for the decreased level of Hb in diabetic rats. HbA comprises 3.4 percent to 5.8 percent of total Hb in normal human red cells, but it is increased in patients with overt diabetes mellitus []. It was found to increase in diabetic patients up to 16 percent [], and the level of HbA is monitored as a reliable index of glycemic control in diabetes []. Elevated levels of HbA and reduced levels of Hb observed in our study reveal that diabetic animals had prior high blood glucose level. Groundnut oil-fed diabetic rats showed an increase in Hb level and decrease in HbA level, which may be due to the reduction of the blood glucose level. In experimental diabetes, enzymes of glucose metabolism are markedly altered. Persistent hyperglycemia is a major contributor to such metabolic alterations that lead to the pathogenesis of diabetic complications, especially microvascular diseases []. One of the key enzymes in the catabolism of glucose is glucokinase, which phosphorylates glucose to glucose-6-phosphate []. In our study, the glucokinase activity was decreased in the liver of diabetic rats, which may be due to the deficiency of insulin. Groundnut oil-fed diabetic rats showed an elevated activity of glucokinase, which may be associated with reduced blood glucose. Insulin decreases gluconeogenesis by decreasing the activities of key enzymes, such as glucose-6-phosphatase, fructose-1, 6-bisphosphatase, phosphoenolpyruvate carboxykinase, and pyruvate carboxykinase []. Glucose-6-phosphatase is an important enzyme in homeostasis of blood glucose as it catalyzes the terminal step both in gluconeogenesis and glycogenolysis []. Fructose-1, 6-bisphosphase is one of the key enzymes of gluconeogenic pathway. It is present in liver and kidney but absent from heart, muscle, and smooth muscle. In our study, the increased activities of glucose-6-phosphatase and fructose-1, 6-bisphosphatase in liver and kidney of diabetic rats may be due to insulin deficiency. In oil-fed diabetic rats, the activities of these two enzymes were significantly reduced, which is responsible for the improved glycemic control. Diabetes mellitus has been reported to generate reactive oxygen species (ROS). ROS, such as free hydroxyl radicals (֗OH) and superoxide (O2֗ˉ ), can cause lipid peroxidation []. Membrane lipid peroxidation results in loss of PUFA, decreased membrane fluidity, and loss of enzyme and receptor activity. The products of lipid peroxidation are capable of interacting with DNA and cause oxidative damage []. In our study, the lipid peroxidation markers such as TBARS and HP were significantly increased in the plasma of STZ-diabetic rats as reported earlier []. The levels of TBARS and HP were significantly decreased in diabetic rats fed with oil, which may be associated with decreased blood glucose and the presence of vitamin E in the oil. Oxidative stress occurs when there is an imbalance between free radical reaction and the scavenging capacity of the antioxidative defense mechanism of the organism []. The nonenzymatic antioxidants such as GSH, vitamin C, and vitamin E are interrelated by recycling process []. Glutathione is the most important non-protein compound-containing thiol group, which acts as a substrate for glutathione transferase and glutathione peroxidase involved in preventing the deleterious effect of oxygen radicals []. In our study, diabetic rats showed a significant decrease in the level of GSH, which may be due to increased utilization. In oil-fed diabetic rats, a significant improvement in GSH was observed. This could be due to the decreased utilization of GSH. Vitamin C is one of the most powerful natural antioxidants []. It is capable of regenerating α-tocopherol from the tocopheroxyl radical that is formed upon the inhibition of lipid peroxidation by vitamin E []. Vitamin C has been reported to contribute up to 24 percent of the total peroxyl radical-trapping antioxidant activity (TRAP) []. In our study, vitamin C was decreased significantly in the plasma of diabetic rats as reported earlier []. Groundnut oil-fed diabetic rats did not show any variation in vitamin C levels. Among lipid soluble antioxidants, α-tocopherol plays a central role as it controls radical-induced lipoprotein lipid peroxidation []. In our study, vitamin E level was also significantly elevated in the plasma of diabetic rats as reported earlier []. The increased level of α-tocopherol could be due to the increased release from membrane damage by ROS. Groundnut oil-fed diabetic rats showed a significant elevation of vitamin E in the plasma of diabetic rats. The increased level might be due to the presence of vitamin E in the oil. The levels of serum lipids are usually elevated in diabetes mellitus and such an elevation represents a risk factor for coronary heart disease []. Diabetic rats fed with groundnut oil showed a small but significant reduction in levels of TC, VLDL-C, LDL-C, and TG and elevation in HDL-C levels when compared with diabetic controls. This could be due to the presence of MUFA and PUFA in the oil. There have been numerous studies in humans and animals that have demonstrated that oils containing saturated fatty acids raise serum TC, TG, and, in particular, LDL-C levels, while those enriched in unsaturated fatty acids lower TC, TG, and LDL-C [,]. Diets high in monounsaturated fatty acids have been found to be relatively hypocholesterolemic or hypotriacylglycerolemic, respectively [,]. In conclusion, our results show that groundnut oil substitution in the diet influences blood glucose, lipid profile, lipid peroxidation, and antioxidants beneficially in STZ-diabetic rats.

Kikuchi’s disease, or necrotizing histiocytic lymphadenitis, is a rare disease, characterized by the presence of enlarged and inflamed lymph nodes. Literature reviews have estimated that up to 80 percent of the patients are of Far Eastern descent, with the disease showing a preponderance toward women []. Only six of the 108 patients surveyed in a study of Kikuchi’s disease diagnosis were African-American and showed an age range from 11 to 75 years [], while the typical age of presentation is in the third to fourth decades of life. Our patient, a 51-year-old African-American male, is therefore unusual as a patient with this disease. The epidemiology of Kikuchi’s disease is widespread, spanning the globe from Japan, where it was first described in 1972, to the United States, and including Europe, the Middle East, and South America [-]. The first documented cases outside of Japan were described by Pileri et al., depicting cases in West Germany, Iran, Italy, Korea, and Spain []. This was shortly followed by the first cases reported in the United States [,]. There has been no strong genetic predisposition established for this disease. Rare familial cases have been reported primarily from Japan and Saudi Arabia [,]. Here, we report a case study of a 51-year-old African-American male with Kikuchi’s disease and describe a novel and successful tapered steroidal therapeutic regime. Our patient had no pertinent travel history. He was raised in New York City and knew of no family member who had an illness with symptoms similar to his. A 51-year-old African-American male presented to the emergency room with a history of intermittent fever, chills, poor appetite, nausea, vomiting, and cough with productive white sputum for the past 10 days. He also complained of progressively worsening right-sided facial swelling, particularly around the parotid region, and severe peri-orbital edema, all of which had originated four weeks prior to presentation. He had noticed a seven- to eight-pound weight loss within those four weeks as well. He worked in a homeless shelter as an administrator. He was diagnosed and treated for tuberculosis in 1985. Upon physical examination, the patient was febrile, 104.1°F, had significant peri-orbital edema, along with enlarged and tender lymph nodes in the right parotid region, pre-tragal area, and the anterior cervical region. His laboratory results are given in Table 1, revealing marked neutropenia and lymphopenia. italic #text xref #text The lymph node architecture was extensively effaced by diffuse necrosis that was predominantly paracortical and spared occasional reactive follicles. Necrotic areas contained apoptotic bodies and were surrounded by a cellular proliferation, which included variable proportions of round, pale cells with blastic morphology consistent with plasmacytoid monocytes, transformed lymphocytes, and predominant histiocytes, often with crescentic nuclei. Neutrophils were absent. Immunophenotyping disclosed that the cells surrounding necrotic areas were mainly activated T-cells, expressing either CD4 or CD8 antigen, and some small CD20 positive B-cells. The immunostain with HECA-452 (directed against cutaneous lymphocyte antigen) highlighted numerous transformed lymphocytes and plasmacytoid monocytes. The latter, along with many macrophages, also expressed PG-M1 (against macrophage-restricted CD68 epitope) []. Within the remaining lymphoid areas, CD20 positive B-cells and CD3 positive T-cells were distributed as expected in reactive lymphoid tissue. A PCR from extracted DNA for immunoglobulin heavy-chain and T-cell receptor gamma chain gene rearrangements failed to reveal a monoclonal cell population. hybridization for Epstein-Barr virus-encoded RNA was negative. The histological findings, together with the immunologic and molecular studies, supported a reactive lymphoid process consistent with the diagnosis of necrotizing histiocytic lymphadenitis (Kikuchi’s disease). The etiology of Kikuchi’s disease is not entirely known. It has been linked to sequela of infection by human herpes virus 6, cytomegalovirus (CMV), and even human T-lymphotropic virus 1 [-]. Recent reports had suggested links between Kikuchi’s disease and HHV-8 or Epstein-Barr virus (EBV). These, however were discredited by George et al. []. Some case reports have linked Kikuchi’s disease to systemic lupus erythematosus (SLE) as well, as patients who attributed their symptoms to Kikuchi’s disease went on to develop SLE. This mitigates the hypothesis that Kikuchi’s disease may be an autoimmune disease. Serologic tests confirmed that our patient was not infected by any of these pathogens. Although his EBV capsid antibody titers were above the normal limit, signifying a past infection, the hybridization for EBV-encoded RNA was negative, indicating he was not actively infected. The clinical presentation of Kikuchi’s disease is very similar to malignant lymphoma, tuberculosis, and systemic lupus erythematosus [-]. Laboratory analysis revealed leukopenia and lymphopenia without eosinophilia or basophilia. Elevated erythrocyte sedimentation rate, C reactive protein, and serum lactate dehydrogenase and transaminases were observed [] (Table 1). He had had a prior infection with tuberculosis, his sputum was repeatedly cultured for acid-fast bacilli and was negative, and he had no evidence of SLE. The pathologic hallmark of Kikuchi’s disease is the presence of an enlarged lymph node with paracortical necrotic foci, which are devoid of neutrophils and surrounded by plasmacytoid monocytes, immunoblasts and crescentic histiocytes. Immunohistochemistry was helpful in identifying characteristic plasmacytoid monocytes. The latter are nonphagocytic natural type 1 interferon, producing cells more likely involved in cytotoxic immune reactions. Recent studies using CD68 and HECA-452 antibodies, on paraffin-embedded sections, revealed that these antibodies, together with CD4, marked the plasmacytoid monocytes [,]. This co-expression of CD68 (PG-M1) and HECA-452 in the plasmacytoid monocytes has been previously reported in Kikuchi’s disease. These histological findings are sufficiently distinctive to permit an accurate diagnosis of Kikuchi’s disease, provided systemic lupus erythematosus has been excluded by the appropriate serologic tests. Malignant lymphoma, especially T-cell non-Hodgkins lymphoma, can be mistaken for Kikuchi’s disease. Loss of pan T-cell antigens by immunostains and determination of the monoclonality of T cells by molecular studies are necessary for confirming the diagnosis of T-cell lymphoma. It is possible to use fine needle aspiration cytology (FNAC) to confirm the diagnosis of Kikuchi’s disease, but the focal involvement can be completely missed, as illustrated by the equivocal FNAC in our patient. Excisional biopsy eventually was undertaken in our case, which consisted of obtaining diagnostic tissue and a representative picture of its architecture, along with removal of the swollen mass or masses. Cross-sectional imaging findings in patients with Kikuchi disease have been described as the presence of clusters of many small or mildly enlarged lymph nodes that appeared abnormal, mostly not because of their size but because of the increased number. On CT and MR examination, the lymph nodes were uniform in attenuation and intensity. The cervical lymph nodes are commonly involved, but supraclavicular, axillary, mediastinal, celiac, peripancreatic, and inguinal chain lymph nodes have been reported. The abnormality of the right parotid region in our patient can be attributed to inflammation of the right parotid lymph nodes. The course of Kikuchi’s disease is relatively benign and self limited [,,]. Nevertheless, many therapeutic regimens have been suggested for shortening its course by reducing swelling and fever spikes. The swelling in our patient was clinically noteworthy, particularly in the right peri-orbital region, where he was unable to open his eye. While we do not advocate the use of any therapy to treat Kikuchi’s disease, steroid therapy is recommended by some to alleviate the related symptoms, such as the marked swelling. Steroid therapy, particularly prednisone, can benefit patients suffering from symptoms such as severe edema and fever, although our patient became afebrile two days prior to the start of his steroid regimen [,]. One case report described the use of chloroquine to alleviate symptoms over a five-month period []. Another case study described the use of minocycline, which alleviated symptoms in 10 days, encouraging the idea that this disease might be the result of microorganism, rather than an autoimmune etiology []. These, however, were the results of single case reports with no follow-up studies to our knowledge. We discourage antibiotic use for Kikuchi’s disease, particularly as no causative microorganism has been identified and the detrimental side effects these regimens can have, as witnessed in our patient. We encourage the use of a tapered steroid regimen, as it was efficacious and of short duration. Our patient was placed on a steroidal regimen, previously described, and responded dramatically with complete alleviation of the peri-orbital swelling within two weeks; the patient also maintained his afebrile state throughout the entire course. The proposed steroid therapy with a dosage regimen is unique to our knowledge in the realm of Kikuchi’s disease management. Our case report reinforces the idea that although Kikuchi’s disease does predominantly affect young women, it can appear at all ages, irrespective of gender. The constellation of clinical findings consisting of regional lymphadenopathy, fever, marked leukopenia in the presence of characteristic histiocytic necrotizing lymphadenitis, and pathological analysis provide the diagnosis. Although many treatment regimens have been recommended, there has not been any established therapy for this disease, nor have any therapeutic trials been undertaken to our knowledge.

Recently, much attention has been focused on the antioxidant defense system in oxidative stress and cardiovascular diseases. Natural antioxidants and polyunsaturated fatty acids contained in dietary sources are candidates for the prevention of oxidative damage and cardiovascular diseases []. Polyunsaturated fatty acids are essential for normal growth and development and may play an important role in the prevention and treatment of coronary heart disease, hypertension, diabetes, and arthritis and other inflammatory and autoimmune disorders. Clinical and epidemiological studies have shown the cardiovascular protective effects of oils rich in polyunsaturated fatty acids (PUFA) [,]. In particular, these substances have been reported to lower blood pressure and prevent the development of hypertension [,]. Sesame seeds and oil have long been categorized as traditional health food in India and other East Asian countries. Sesame oil has been found to contain considerable amounts of the sesame lignans: sesamin, episesamin, and sesamolin. Sesame oil also contains vitamin E (40 mg/100 g oil), 43 percent of polyunsaturated fatty acids, and 40 percent monounsaturated fatty acids. The lignans present in sesame oil are thought to be responsible for many of its unique chemical and physiological properties, including its antioxidant and antihypertensive properties [–]. In the present study, we evaluated the effect of sesame oil (rich in antioxidant lignans, vitamin E, and unsaturated fatty acids) in hypertensive patients on medication with either hydrochlorothiazide or atenolol as antihypertensive therapy. The present study consists of patients of both sexes in the age group 35 to 60 years with mild to moderate hypertension, medicated with diuretics (hydrochlorothiazide) or β-blockers (atenolol), who were recruited from the Department of Medicine at Rajah Muthiah Medical College and Hospital, Annamalai University, and Prof. Maniarasan Memorial Polyclinic, Chidambaram, Tamilnadu, India. The criterion for hypertension was systolic blood pressure greater than or equal to 140 mm Hg and diastolic blood pressure greater than or equal to 90 mm Hg, recorded on at least three different occasions after they had rested for 10 minutes supine. Patients with secondary hypertension, hypertension associated with diabetes mellitus, chronic alcoholism, female patients on oral contraceptives, pregnant females, and lactating mothers were excluded from the study. All the subjects gave informed consent to undergo the investigations, and the Ethical Committee of Rajah Muthiah Medical College, Annamalai University, Tamilnadu, India, approved the study. A detailed clinical history and physical examination were performed at baseline, and the following measurements were taken: blood pressure; anthropometric measurements, such as height, weight, and body mass index (BMI); lipid profile (total cholesterol (TC), high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C), and triglycerides (TG)); electrolytes (Na⁺, K⁺); lipid peroxidation (TBARS); and enzymic and non-enzymic antioxidants in blood. The patients were advised to continue their antihypertensive drugs as usual. The patients were on medication with hydrochlorothiazide or atenolol for one year prior to the enrollment in the study. The patients were supplied 4 to 5 kg of sesame oil (Idhayam gingelly oil) for a four-member family per month, which constitutes approximately 35 g of oil/day/person. The patients were asked to use sesame oil as the only edible oil for 45 days. At the end of the 45th day, the investigations were repeated. Finally, the patients were asked to switch over to whatever original oil they had been taking before the enrollment of the study for another 45 days. Mostly they were using either sesame oil, groundnut oil, or palm oil interchangeably. All the measurements were repeated at the end of the 90th day of our experiment. The patients were told to strictly adhere to the study protocol. Those who could not follow the protocol until the end of the experiment for any reason were excluded. To avoid much difference in dietary patterns and caloric changes, the same patients have been subjected to substitution of sesame oil and withdrawal of sesame oil substitution. Body weight was measured, using a level balance, to the nearest 0.1 kg. Body height was measured without footwear to the nearest 0.5 cm. BMI was calculated as weight (in kg)/height (in m). Blood pressure was measured by using standard mercury sphygmomanometer. Fasting blood samples were collected on entering the study (0 days), at the end of 45 days, and after 90 days (i.e., after substitution and withdrawal). Lipid profile, electrolytes, lipid peroxidation, and enzymatic and non-enzymic antioxidants were estimated at the three experimental periods: baseline, after sesame oil substitution, and after withdrawal of sesame oil. All the biochemical determinations were carried out in the Biochemistry Laboratories at Department of Biochemistry, Faculty of Science or Rajah Muthiah Medical College and Hospital, Annamalai University. TC [], HDL-C [], and TG [] concentrations in plasma were determined by standard enzymic methods with a semiautoanalyser (Bayer RA 150, Germany) using commercially available kits (Biocon, Germany). LDL-C was calculated using Friedwald equation []. Sodium and potassium [], TBARS [], enzymic antioxidants such as superoxide dismutase (SOD) [], catalase (CAT) [], glutathione peroxidase (GPx) [], and non-enzymic antioxidants such as vitamin C [], vitamin E [], β-carotene [], and reduced glutathione (GSH) [] also were estimated. Student’s t test was applied for comparison between two related samples; values for continuous variables are expressed as means ± SD. Table 1 shows blood pressure and anthropometric measurements at baseline, sesame oil substitution, and withdrawal of sesame oil. Replacement of sesame oil as cooking oil in hypertensive patients brought their systolic and diastolic blood pressure to normal in a statistically significant fashion. Significant reduction in body weight and body mass index also was noted. After the withdrawal of sesame oil substitution, the values rose again. Table 2 shows the plasma lipid profile at baseline, after sesame oil substitution, and after withdrawal of sesame oil. No significant alterations were seen in TC, HDL-C, LDL-C, and the TC/HDL-C ratio. TG levels decreased significantly and then rose, following sesame oil substitution and withdrawal, respectively. Table 3 shows the plasma levels of electrolytes at baseline, after sesame oil substitution, and after withdrawal of sesame oil. Plasma sodium levels decreased significantly and then rose, following sesame oil substitution and withdrawal, respectively. Potassium levels increased significantly upon sesame oil substitution and subsequently decreased, but within normal limits. Table 4 shows the levels of TBARS, enzymic and non-enzymic antioxidants at baseline, after sesame oil substitution, and after withdrawal of sesame oil. Significant reduction in TBARS was noted, and the values were almost maintained even after withdrawal of sesame oil. Plasma CAT and erythrocyte membrane bound SOD activities significantly increased, while erythrocyte membrane bound GPx activity decreased gradually from sesame oil substitution to withdrawal. Significant elevations of vitamin C, vitamin E, ß-carotene, and reduced glutathione were observed, and the levels decreased once sesame oil substitution was stopped. In the present study, substitution of sesame oil lowered systolic and diastolic blood pressure remarkably in hypertensive patients. Studies reported that sesamin, a lignan from sesame oil, exerts antihypertensive action by interfering with renin-angiotensin system, as the lignan is more effective on the renin-independent DOCA (Deoxycorticosterone acetate) -salt hypertension than on the renin-independent 2K (two kidney), 1C (one clip) renal hypertensive model [,]. In another study using the rat aortic ring, sesamin produced Ca²⁺ antagonistic vasodilatory activity []. This pharmacological action, at least in part, may contribute to its antihypertensive activity. Natural antioxidants and polyunsaturated fatty acids show protective function against hypertension []. Supplementation of vitamin E reduced blood pressure in mild hypertensive patients and was associated with a remarkable decrease in systolic and diastolic blood pressure []. The fatty acid composition of dietary fat is a key determinant of membrane fatty acid composition []. As PUFA substitution increases the fluidity of the bilipid layers, the distensibility of biomembranes may increase. The blood pressure-lowering effect of sesame oil may be due to its richness of antioxidant lignans (sesamin, episesamin, sesamol, and sesamolin), vitamin E, and unsaturated fatty acids. The risk of hypertension increases progressively with higher levels of body weight or BMI and parallels the degree of obesity. The association between BMI and blood pressure consistently has been shown in numerous studies []. Numerous studies consistently have documented that for those who are already overweight, weight loss significantly reduces blood pressure and the incidence of subsequent hypertension. Large, randomized trials of weight reduction in adults with hypertension have shown significant reductions in blood pressure in response to weight loss []. Studies suggest that polyunsaturated fatty acid increases the plasma levels of leptin, which, in turn, would facilitate the reductions of weight []. Polyunsaturated fatty acids in sesame oil also may play a role in the reduction of body weight in our study, which in turn may reduce the blood pressure. The reduction of body weight and body mass index in our study mainly may be due to sesame oil substitution, since the values increased once the sesame oil substitution was withdrawn. Prior studies in rats have been shown that sesame lignans (sesamin and/or episesamin) lower serum and liver cholesterol concentrations by inhibiting absorption and synthesis of cholesterol []. We did not find a cholesterol-lowering effect in hypertensive patients on medication with diuretics or ß-blockers. This may be due to the negative effect of diuretics and ß-blockers on lipids. Recently, the Scientific Advisory of the American Heart Association reported that high monounsaturated fatty acids diets tend to lower triglyceride concentrations []. We found that substitution of sesame oil as edible oil lowered plasma triglyceride concentrations. Reports suggested that antihypertensive compounds modulate the Na⁺-K⁺ pump and thereby maintain the electrolytes levels in hypertensive patients. Cardiac output is influenced by blood volume, which is greatly dependent on body sodium. Thus, sodium excretion is central to blood pressure modulation. Decreasing sodium excretion increases fluid volume and leads to high cardiac output. Potassium can influence cell membrane stabilization and vascular smooth muscle relaxation []. In our present study, we found that plasma levels of sodium decreased while potassium levels increased upon the substitution of sesame oil. However, the mechanism of reduction of sodium and elevation of potassium upon sesame oil substitution is not known. Thiobarbituric acid reactive substances, a measure of lipid peroxidation, decreased significantly upon sesame oil substitution. It has been reported that sesamolin, a lignan present in sesame oil, reduced lipid peroxidation in rats []. Sesamin and sesamolin may potentiate the effect of vitamin E and they themselves act as antioxidants, which, in turn, may reduce lipid peroxidation. In our study, plasma levels of TBARS did not change even after withdrawal of sesame oil substitution. Perhaps the lignans stored in the body may be responsible for this. The role of the antioxidant defense system, which includes superoxide dismutase (EC 1.15.1.1; Cu/Zn SOD), catalase (EC 1.11.1.6; CAT), and glutathione peroxidase (EC 1.11.1.9; GSH-Px), in protection against oxidative insults is well characterized, and it has been suggested that this antioxidant defense system may be influenced by nutrition []. Enzymatic antioxidants, such as SOD and CAT, play an important role in the conversion of ROS to oxygen and water. SOD is a well-known scavenger enzyme preventing the cell from oxidative stress. CAT is an important antioxidant enzyme whose physiological role is to detoxify HO into oxygen and water and thus limit the deleterious effects of reactive oxygen species. Cells maintain their vital functions against oxidative damage with the help of a system that involves GPx, SOD, CAT, glutathione reductase, some trace elements, and vitamins A and E. The increase of SOD and CAT may be due to decreased utilization, since lipid peroxidation levels are low. GPx probably decreased due to the decreased synthesis, since lipidperoxidation levels were low. Vitamin E has been recognized as one of the body’s major natural antioxidants. Sesame oil contains 40 mg of vitamin E per 100 g of oil []. Vitamin E has several potentially cardio-protective effects: It decreases lipid peroxidation and spares glutathione [,]. Vitamin E has been shown to lower blood pressure in spontaneously hypertensive rats []. In the present study, plasma levels of vitamin E increased upon substitution, which could be due to the greater availability of vitamin E in sesame oil. Elevation of vitamin C upon the substitution of sesame oil could be due to the decreased utilization or due to increase in the levels of GSH, because vitamin C and GSH are synergistic antioxidants []. Epidemiological reports show that carotenoids may play a preventive role in cardiovascular disease []. Plasma levels of ß-carotene rose significantly upon the substitution of sesame oil, which could be due to the sparing action of vitamin E and sesame lignans. In conclusion, substitution of sesame oil, as the sole edible oil, lowered blood pressure in hypertensive patients who were taking diuretics and ß-blockers. Sesame oil also has beneficial effects on the levels of triglyceride, electrolytes, lipid peroxidation, and antioxidants.

Motility in most bacterial species depends on a sophisticated molecular machine called the flagellum. The flagellar apparatus is made of dozens of different proteins and thousands of individual subunits. The bacterial flagellum is actually a mechanical nanomachine with a rotation frequency of 300 Hz, an energy conversion rate of nearly 100%, and the ability to self assemble (, ; ; ). Various efforts have been made to identify all components required for bacterial motility, resulting in a list of more than 60 proteins in (, ). Functionally, these proteins can be subdivided into several subsets: the chemotaxis system connects environmental stimuli to the direction of flagellar rotation and thus direction of movements. The chemotaxis system is connected to the basal body complex, which anchors the flagellum in the inner membrane and also incorporates a type-III-secretion system necessary for the self-assembly process of the flagellum. Two motor proteins, MotA and MotB, convert an ion gradient (for most bacteria a proton gradient) into rotational energy of basal body components; these components are connected to the rod structure and then via a flexible hook to the filament of the flagellum, which operates like a propeller. However, whereas the overall structure has been known for decades, many of the mechanistic details responsible for the assembly and operation of the motor have yet to be worked out. In fact, it remains unclear whether all the protein components of the flagellar apparatus have been identified. Similarly, whereas at least 51 protein–protein interactions (PPIs) have been described in the literature (), many more interactions are likely to be required for assembly and proper operation. Despite the vast body of literature about bacterial motility, there have been few systematic attempts to identify the components of the flagellar apparatus and their function besides genome sequencing. Systematic analysis of hundreds of completely sequenced genomes, for example, has predicted many additional motility genes based on their location in flagellar operons or gene clusters, yet their actual roles in motility often remain unknown. In this study, we systematically identified genes essential for bacterial motility by testing the swarming capability of 3985 gene deletion strains of (). In addition, we integrated data from similar screens carried out for () and mutant screens of (; ) and (). Second, we screened all motility proteins recovered from the literature for PPIs. We reasoned that unknown motility proteins can be discovered by interactions with known flagellar and chemotaxis components. Protein interactions were identified by screening the proteomes of two small distantly related bacteria, and , using comprehensive array-based yeast-two-hybrid screens (; ). In addition, we compared our data to protein interaction data of () and (). Finally, we integrated these experimental data sets with predictions of functional associations from the STRING database (; ). The result is a list of known and new flagellar components, including 23 novel motility proteins ( and ). Many features of the bacterial flagellum have changed over the course of evolution. This is reflected in the surprisingly different composition and protein interaction patterns in the flagella of different species, which may reflect adaptations to species-specific motility needs (compare and ). While the overall conservation allows us to predict ∼18 000 interactions for 64 proteomes of flagellated bacteria, it remains to be seen how many of them are functional. Several systematic mutant screens have been performed to find genes involved in bacterial motility (; ; ; ). To generate a comprehensive motility mutant data set for the gram-negative model bacterium we have used the gene deletion library constructed by Baba (). These mutants were plated out in arrays of 24 colonies on swarming agar and tested for swarming (). Of 3985 mutants tested 159 deletions showed a swarming defect (). Interestingly, a similar screen in yielded a similar number of 146 motility mutants () (). Thus about 4% of the nonessential genes in both species show an effect on motility under the conditions tested. Among them are 43 (30%) and 48 (27%) genes previously annotated as motility genes in and , respectively (). The other mutants with motility phenotypes are significantly enriched for proteins involved in ‘motor activity' and ‘macromolecule metabolism' (). Many of these genes may be required to provide energy to the flagellar motor or may be indirectly involved in the assembly of the flagellar apparatus, for example in restructuring the peptidoglycan to allow penetration of the cell wall. Unexpectedly, only 7 of the () mutant genes that were previously not known to have a motility function, had a homolog with a phenotype in () ( and ). Thus, there appear to be many proteins with a species-specific role in motility. Examples of such proteins are discussed further below. Decades of research have identified many components of bacterial flagella and their motors (). We have used most of these motility proteins in two-hybrid screens in both and All known motility proteins were tested as fusions to the Gal4-DNA binding domain (baits) in a systematic array-based yeast-two-hybrid screen against a whole genome prey library (i.e. fusions with the Gal4 activation domain) of These screens identified 176 PPIs for (TPA, and and ). Similarly, the motility proteins were tested for interactions with most of the proteins in systematic LexA-based Y2H screens (), and a comparable number of 140 high-confidence interactions (CJE HCF) was found among 690 total interactions (CJE All) ( and ). Additional motility protein interactions were filtered from the Y2H interaction map of (HPY, ), and from a complex purification study of () (ECO SPK, ECO SAI, see Materials and methods) (). Pairwise comparisons of these various interaction data sets revealed only a limited overlap ranging from 2.5% for the (ECO SPK) versus the data to 25.0% for versus CJE HCF (). Overall, ECO SPK has the weakest pairwise similarities. Thus the overlap between the different data sets appears to reflect both phylogenetic relationships as well as methodological differences between yeast two-hybrid and complex purification data sets. As might be expected, interactions between motility proteins are common in the motility interaction maps ( and ). An overview of the number of proteins (nodes) and their interactions (edges) and additional properties of these networks can be found in . Finally, for a comparative analysis of motility interactions, we carried out a comprehensive review of the literature for published flagellar PPIs using PubMed and found 51 unique interactions (). Of these 51, 39 had interologs in and 38 in , but only 9 and 5 were reproduced in and CJE ALL, respectively (). Only one interaction is common to both and CJE ALL screens, and thus a total of 13 interactions were recovered in either of our screens. That is, sampling of the two species recovered 33% of all published flagellar interactions. One reason for this relatively low coverage may be that most previous studies used different methods that may be better applicable to flagellar proteins. All literature comparisons can be found in . The diversity of information on different genomes, proteins, phenotypes and so on makes it difficult to keep track of all details. Therefore, we generated an integrated motility network, which combines a diverse set of interaction networks as well as phylogenetic and phenotyping data (). This network combines protein–protein interactions of , , and as well as interactions curated from the literature with motility phenotyping data from and . It also displays clusters of orthologous groups (COGs) rather than individual proteins, which reduces complexity and improves the quality of links. These links represent direct interactions, indirect interactions (if proteins do not interact directly, but via a bridging protein), and literature interactions. Out of all interactions, 73% connect known motility COGs. In addition, 45% were predicted by STRING (highest confidence: S>0.9) to be strongly associated. These numbers indicate that this integrated network is more reliable and biologically relevant than individual networks. In addition, links among orthologous groups can usually be transferred to proteins of other species. However, because of the stringent filtering not all interactions are included in this network. Another striking connection is the conserved MotB–FliL interaction in and . For , FliL is thought to be involved in sensing of the actual flagellum status (). Here, we found evidence that this sensing is mediated by a direct interaction with the motor apparatus (). A major goal of this study was to find novel flagellar components among the many proteins of still unknown function. In addition, we suspected that there must be previously characterized proteins whose role in motility remained unknown. Indeed, 28% and 33% of the interactions found in and , respectively, connected a known motility protein to a conserved hypothetical protein (), suggesting that there are still unidentified proteins with a motility function. To identify potential novel bona fide motility proteins, we used our integrated data and identified 23 hitherto uncharacterized proteins (). For example, members of the orthologous group COG1664, such as TP0048 and HP1542 (), show interactions to the FliC–FliS cluster. Additional evidence for a role in motility comes from the double mutant of the orthologs and , which also shows reduced motility. TP0658 ( in ), a previously uncharacterized protein, was found to interact with all three flagellin proteins (FlaB1-B3) of . The deletion mutants of TP0658 orthologs in both () and (CJ1075) show a highly reduced motility phenotype (; ). We have recently shown that TP0658 and appear to stabilize flagellin in the cytoplasm, thus exhibiting properties of a chaperone (). TP0658 and its orthologs thus appear to be flagellar assembly factors or factors involved in export of the filament protein FliC. TP0561 is another hitherto uncharacterized protein that appears to be involved in flagellar protein export based on its interaction pattern; it interacts with multiple components of the export machinery such as FliR, FliL, FliQ, and FlhB (, bottom left). In addition, a mutation in TP0561 results in significantly reduced motility. Due to its motility phenotype in TatD is an exceptional case. Its homolog TP0979 interacted with FliE. In , has two paralogs, and , which are functionally unassigned, genomically unlinked, and show 29% and 24% amino-acid sequence identity to TatD, respectively. As is localized in an operon with genes of the twin-arginine transport (Tat) system, a transport function of TatD was anticipated. But even a strain with all three TatD paralogs deleted did not show a Tat-related transport deficiency, leaving the question for TatD's function unanswered (). As we found a very small, but significant, increase in motility for the single mutant, we tested the previously described triple mutant (all TatD paralogs deleted) for motility (). The triple mutant was constructed in a MC4100 strain background, a strain known to be nonmotile, presumably due to a point mutation in FlhD, a known master regulator for motility (). Unexpectedly, the triple mutant showed a slight rescue of motility. We investigated the relation between the FlhD point mutation and the TatD paralogs by expressing a functional FlhD construct both in the MC4100 strain and the triple mutant. A strong synergistic effect of FlhD expression and the triple mutation on motility was observed pointing to a regulatory antagonism of FlhD and TatD paralogs (). These findings indicate that TatD and/or its orthologs (COG0084) have a negative role in motility, perhaps mediated by its DNAse activity (). Overall, most parts of the flagellum are well conserved in motile bacteria. Nevertheless, evolutionary adaptation of several components can clearly be identified and range from the duplication of proteins, for example of flagellins, to the complete loss or gain of components, for example of export chaperones (). Here, we will give three examples for such evolutionary processes on the interaction level. Whereas the flagellar apparatus is a well-defined nanomachine, it does not act in isolation. Besides the obvious link to the chemotaxis pathway, we noticed several interactions with proteins of other function (). For example, a link of motility proteins with ‘nucleobase, nucleoside, nucleotide, and nucleic acid metabolism' (GO:0006139) is found in both and interaction sets, as well as the mutant phenotyping data. NrdB (ribonucleoside-diphosphate reductase), the key enzyme for the conversion of ribonucleosides into desoxy-ribonucleosides, interacts with two flagellar proteins, FliC and FlgB (). A functional link of NrdB to motility is provided by a study by , where the authors found a reduction in flagellar protein expression upon nrdB deletion. Although the authors assumed a link on the transcriptional level, an additional post-translational link, as indicated by a direct protein interaction, becomes likely. The functional link between electron transport (electron transport chain) and motility via a proton gradient (or sodium gradient) is well known and also reflected by an association with ‘transport' (GO:0006810) in a motility gene expression data set (FlhD) (). We found a direct interaction between NuoC, (NADH dehydrogenase I) and FliM both in the and data sets (). This enzyme forms complex I of the electron transport chain and converts the oxidation of NADH into an electrochemical proton gradient. At least in these two species ( does not have an electron transport chain), motility might be optimized by increasing the local proton concentration. Motility is known to be regulated by environmental stimuli such as nutrients and this is reflected by the over-representation of ‘response to stimulus' (GO:0050896) proteins among flagellar interactors. For example, motility is controlled by the second messenger cyclic-di-GMP, which is produced by the enzymatic activity of so-called GGDEF domains (). Here, we find a conserved interaction of the GGDEF COG, COG2199, with FliC in and (), pointing to an important regulatory role of this interaction. FliA, the sigma factor for several flagellar operons, interacts with two subunits of the RNA-polymerase (rpoB and rpoC) in the and the interaction sets (). In addition, in the same species, an interaction of FliA with the glutamyl-tRNA synthetase, GltX, was found, suggesting a regulatory role of this interaction. Given the amazing complexity of the bacterial motility system, we wondered whether our interaction data and phenotypes can contribute to the understanding of its evolution. As a first step into that direction, we first constructed a phylogenetic supertree of 30 species based on 35 flagellar protein families (). Our flagellum supertree strongly supports the monophyly of spirochetes, as well as γ and β, ɛ, and α proteobacteria. These relationships are similar to the previously reported phylogenies, for example, an rRNA tree () and a tree which was based on 31 highly conserved protein families (). This shows that the flagellar system evolved together with other cellular systems and not independently. Evolution of the flagellum is also consistent with the fact that neither any flagellar proteins nor any of their interactions is conserved. In fact, our data set predicted 173 interactions for , of which we found only 49 (). This indicates that protein interactions may be evolutionarily less conserved than generally believed. An evolutionary model also predicts that core proteins, which have been associated with the flagellum, should be tightly integrated, and thus have more interactions than peripheral proteins, which have been only recently recruited to the flagellar machinery. Indeed, we did find a weak, but statistically significant linear relationship between the number of interactions of an orthologous group and its conservation ratio among flagellated bacteria (=0.43, <0.005; ). Therefore, our analysis supports the evolution of the flagellum from core components by adding additional ones over time (). In this study, we pursue an integrative systems biology approach to assemble a comprehensive picture of bacterial motility. Motility interaction data sets for and and a genome-wide motility data set for are presented. Our data are combined with functional and interaction data from multiple species to reconstruct an integrated network of bacterial motility. Insights into the internal structure of the flagellum, its connections to other functional classes, and on potentially novel components of the flagellum have been obtained. Due to the size of our data set, we were able to analyze only a few selected interactions in more detail. We confirmed the presence of anti-sigma factors (FlgM) in and possibly based on their interactions with a flagellum-specific sigma factor. Recently, we have assigned a new function to TP0658 (now called FliW), a conserved protein of previously unknown function. We could show that this protein acts as a molecular chaperone and/or assembly factor of the bacterial flagellum (). The bacterial flagellum represents an interesting entity to study the evolution of complex biological machines. For an evolutionary view of the flagellum on the protein level, we constructed a phylogenetic supertree solely based on flagellar protein sequences. As anticipated, this tree closely recapitulates phylogenetic relationships identified, employing traditional phylogenetic marker molecules such as rRNAs. Whereas it is generally believed that the motility machinery evolved from an ancient type III secretion system, the detailed steps leading to current structures have yet to be defined. A prediction from this theory would be that the conserved core proteins should exhibit more interactions than peripheral proteins. Indeed, proteins which are well conserved and part of all flagellar complexes have more conserved interactions (e.g., FliC, FliG, FliY, FliM, FliA, Mcp, CheW, and CheY) than proteins which are found only in a subset of motility complexes (e.g., FlhF or FlgJ; see and ). Similar to protein sequences and structures, interactions among proteins are often conserved in the course of evolution. In fact, the phylogenetic relationships of different species are partially reflected by the phylogenetic interaction profile of the integrated network (). Finally, we could thus use our interaction data sets to predict interactions in other bacterial species. To obtain only high-confidence predictions, we used our integrated motility network, that is, all interactions found in more than one species or supported by other evidence from the literature, and predicted ∼18 000 interactions for 64 flagellated bacteria (). It remains to be seen which of these interactions do indeed occur and what specific role they play in each of these organisms. We collected motility genes from three major classification systems: KEGG (including motility, chemotaxis, and flagellar assembly; ), TIGR (including chemotaxis and motility; ), and GO (including GO:0019861 flagellum; ). Data were compiled in March 2007. In total, we identified 293 proteins (in , and ) with at least one classification evidence. Among those, 89% were classified by KEGG, 75% by TIGR, and 65% by GOA. As KEGG provides the most comprehensive classification, we have used the KEGG motility collection throughout this study and refer to its proteins as ‘known motility proteins' (). COGs were taken from the NCBI COG database (downloaded March 2006) and complemented by COGs from the STRING version 3 database (including nonsupervised COGs) (). A systematic single-gene knockout collection of of 3985 individual mutant strains () was tested for altered motility by a swarming assay. Each gene mutation was tested in two independent strains as provided by the Keio collection. Strains were grown to saturation in LB medium at 37°C (mutants with growth defects were not considered for the motility assay) and transferred to Omnitrays (Nunc) with swarming agar (LB medium with 0.25% Agar) in a 24 colonies per plate format by pin replication with a Biomek 2000 laboratory robot (Beckman-Coulter). The swarming diameters of the mutant strains were compared after ∼8 h incubation at 37°C and mutants with reproducible reduced motility were retested in individual swarming assays. The swarming behavior of each mutant was classified as wild type, reduced (reduction by at least 50%), or nonmotile (reduction by at least 90%), as measured by the diameter of the bacterial colony (). Additionally, we constructed a few individual gene deletions of and and tested them for motility: gene disruptions of , , and were performed in , as described by . The mutants of , , and (double mutant) were obtained by phleomycin–cassette integration, as described by . The PCR primers used are listed in . Information on gene deletions affecting motility in , , and was taken from the literature (; ; ; ). Forty-nine proteins, which are part of our KEGG motility collection, were selected (). Bait fusions (Gal4-DNA-binding domain) of these proteins were constructed by Cre-loxP mediated recombination of pUni clones (; ) with two bait vectors: pAS1 and pLP-GBKT7Amp (created by replacing Kan by Amp in pLP-GBKT7 (Clontech). A systematic whole-genome prey library for was created by transferring all ORFs from their original pUni-vector vector () to our prey vector, pLP-GADT7 (Clontech) by Cre-LoxP-mediated recombination. All prey and bait clones were then individually transformed into Y187 (MATa) and AH109 (MATα) (; ) yeast strains, respectively, by a standard LiAc protocol. Prey strains were arrayed onto 384-well formatted Omnitray-agar plates (Nunc) and each bait strain was individually tested against the whole prey array using a previously described array-based Y2H procedure (). Interactions involving the 46 proteins assigned to the motility category in the KEGG database () () were identified in proteome-wide two-hybrid screens, using a pooled matrix approach as described previously () (). The motility PPI set ‘' (HPY) was generated by selecting all interactions of known motility proteins from (also see motility filtering below). Note that tested only 261 bait fusion proteins (out of 1590 ORFs) against a random prey library. To mine all the published PPIs of the known flagellum components, we carried out a comprehensive literature review for flagellum PPIs, using the PubMed query ‘(flagellum OR flagella) AND (interaction OR interact OR interacts OR bind OR binds)' on 13 January 2004. This analysis yielded ∼700 abstract/articles from which 51 unique PPIs between flagellar components were manually curated (). Motility-related protein interactions of were derived from , who conducted a comprehensive complex purification study using a His-tagged ORF clone library. A total of 2667 out of 4339 proteins were successfully analyzed and their interacting partners were identified by MALDI-TOF in this study. Complex purification studies do not provide binary interaction data, but only lists of proteins that co-purified with the used bait protein involving both direct and indirect interactions. provided their results according to the spoke model. We used two models to predict direct interactions for flagellum/chemotaxis proteins (one among the pair of interacting proteins is a known flagellum/chemotaxis protein) from the complex data. The ‘ECO SPK' interaction set assumes binary interactions between bait proteins and their co-purifying proteins (SPOKE model). The ‘ECO SAI' interaction set is based on a model that has been proposed by to infer complexes from multiple overlapping purifications. Similar to the matrix model, it predicts PPIs among all proteins. However, the difference is that PPIs are weighted according to the pair's propensity to associate with each other relative to what would be expected from their frequency. Based on the cumulative percentage distribution of socio-affinities, we defined the top 25% of PPIs to be highly associated (socio-affinity score >5). Both ECO sets have been filtered for interactions of known motility proteins (see motility filtering below) (). Data from Butland (Nature 433: 531, 2005) have not been considered in this analysis as only one flagellar protein, FliY, appears to have worked as a bait in this study. ext-link xref italic sup #text xref #text The confidence score of the StringDB ( score) is the approximate probability that a predicted link exists between two enzymes in the same metabolic map in the KEGG database. Confidence limits are as follows: low confidence ( score>0.15) 20% (or better); medium confidence ( score>0.4) 50%; high confidence ( score>0.7) 75%; highest confidence ( score>0.9) 95% (from ). Myc-tagged TP0974 and HA-tagged TP0709 were cloned into vectors of the pBAD series () and were co-transferred into BL21 (DE3) . Protein expression was induced with 0.2% (w/v) L-Ara for 3 h at 37°C. The co-immunoprecipitation was performed with anti-Myc antibodies (Santa Cruz). All interaction data from this study can be retrieved from the IntAct database () under the following accession numbers: EBI-1190357 ( data set) and EBI-1190361 ( data set).

A central focus of synthetic biology is constructing isolated gene regulatory networks in living cells in order to determine their dynamical behavior in a given environment. The complexity of naturally occurring gene networks makes this determination difficult, and many investigators have concentrated on smaller artificial ‘circuits' in an attempt to understand fundamental principles (; ). These smaller networks are more amenable to computational modeling, construction and modification. In addition, the systematic construction of genetic circuits from a set of quantitative design principles will accelerate progress towards therapeutic applications. Many cellular functions rely on underlying regulatory processes that are highly dynamic. In the context of interconnected gene regulatory and signalling networks, the persistence of mRNA and protein molecules has a large effect on the overall behavior of the cell. This effect often manifests in the underlying network dynamics, where degradation rates can play a central role in mitigating crucial timing events. Naturally occurring components are often adapted for the construction of artificial gene circuits. These components are usually stable, and this stability constrains the ability to monitor highly dynamic circuits and to construct circuits that possess desired response characteristics. For example, GFP is often used as a reporter in native and artificial synthetic genetic networks, due to its ability to be expressed in various hosts without interfering with cellular function. However, GFP is stable in most cells on a time scale of hours to days (). Once stable GFP is expressed, it is cleared from the system only through growth and subsequent dilution, making measurements at higher temporal resolution difficult to interpret. Temporal resolution can be improved by decreasing the half-life of fluorescent proteins. However, such an increase in temporal responsiveness is not without cost, since it reduces detectable signal. Ideally, one could control the stability of a network component, tuning it to balance the needs of signal detection and dynamical resolution. There have been several studies that use native protein degradation systems to make network components less persistent. used the ssrA system in to target proteins to degradation pathways. In bacteria, polypeptides that stall during translation (e.g., under starvation conditions) have an 11-amino-acid tag added to their C-terminus by a small ssrA molecule. This tag is specifically recognized by the ClpXP proteasome, and tagged proteins are degraded (; ). More recently, it was demonstrated that degradation in could be increased with a modified tag that binds a helper molecule (). In , a similar system was described that made translational fusions between a given protein and a domain of , a yeast protein that is degraded quickly (). It has also been shown that modification of the N-degron signal sequence can lead to impressive destabilization of reporters down to a half-life of 2 min (). While all of these systems produce more dynamic proteins, none focused on developing tunability over a wide range of degradation rates. In addition, the overexpression of native degradation components in the same organism for the purpose of increasing degradation can have undesired pleiotropic effects. We have constructed a strain (CGD699) that allows tunable degradation of a tagged protein. To accomplish this, we expressed a modified ClpXP protease in yeast under the control of a repressible promoter. Proteins that are tagged with the ssrA tag are quickly degraded, and this degradation rate is controlled by the induction level of ClpXP. We integrated the two genes ( and ) that code for the ClpXP protease into the yeast genome (, and Materials and methods). We found that the gene needed to be modified with 10 silent mutations to be expressed in yeast (see Materials and methods). These two genes were placed under the control of two separate copies of the modified promoter (), at two different loci in the yeast genome. Additionally, we integrated , a mammalian-enhanced version of (), controlled by a wild-type promoter. The wild-type version of the promoter exhibits constitutive expression, while the promoter is repressed by LacI in the absence of IPTG. Addition of IPTG to the medium results in ClpXP production and degradation of a tagged protein. To demonstrate the utility of this approach, we also integrated a yEGFP gene tagged with an 11-amino-acid ssrA tag (AANDENYALAA), under the control of the promoter into CGD699. This promoter is fully induced by 0.5% w/v galactose and repressed by 2% w/v glucose. CGD699 cells grown in the presence of galactose produce GFP and are fluorescent, and GFP production ceases if the carbon source in the media is switched from galactose to glucose. Observations of growth rate and morphology indicate that the exogenous proteases cause no deleterious cellular effects over a wide range of IPTG levels (). While coexpression of tagged yEGFP and the degradation machinery resulted in almost complete loss of fluorescence, coexpression of untagged yEGFP with the degradation machinery showed no significant drop in fluorescence (). This confirms that the degradation effect is specific to tagged proteins. We further investigated the ability of CGD699 cells to degrade tagged yEGFP in response to IPTG induction by microscopy in a microfluidic chamber. Prior to imaging, the cells were grown in log phase for 24–36 h in media containing 0.5% galactose and the experimental IPTG concentration. These cells were then loaded into a yeast microfluidic chamber () that was designed to allow temporal switching of inducer (). The growth media was switched to media containing IPTG and 2% glucose to shut off GFP production from the promoter. shows a series of bright-field and corresponding fluorescent images for cells exhibiting active degradation in the 1 mM IPTG condition (also see ). In cells with but without incorporation, the half-life of GFP was 104±19 min. This half-life is the result of growth-related dilution (note that the doubling time of the batch culture grown on galactose was approximately 125 min; see ). The induction of varied expression of the ClpXP protease allows the reduction of the half-life of GFP to as low as 22 min at 5 mM IPTG. Previous studies have used exponential fits to characterize the half-life of the reporter. We found that modeling decay as arising from a set of enzymatic Michaelis–Menten reactions led to excellent agreement between model and experiment (see below). However, in order to systematically compare with the previous fluorescent reporter degradation studies, we chose to first analyze the fluorescence trajectories with exponential fits that were reasonably accurate (). From these fits, we were able to calculate the mean half-life for each concentration of IPTG, as shown in . The half-life decreases from a value of 91 min for no IPTG to 22 min for media containing 5 mM IPTG. In addition to the exponential fitting, we used the experimental data to determine how the degradation process should be modeled. For this, we assumed that both the and promoters are constitutive, and that yEGFP-ssrA, ClpXP and mLacI have reached equilibrium before the introduction of glucose. In this case, the transcription of the promoter stops once glucose has been added and its mRNA begins to decay exponentially from its steady-state value. However, the translation of the mRNA continues, leading to an ODE describing the dynamics of the concentration of yEGFP-ssrA: where is the concentration of yEGFP-ssrA, γ is the degradation rate of the mRNA, is a constant related to the transcription and translation rates of , δ is the dilution rate and and are the Michaelis–Menten constants describing the ClpXP-mediated enzymatic decay of yEGFP-ssrA. is the steady-state initial concentration. Because the constant is proportional to the concentration of the enzyme (ClpXP), it will increase with increasing levels of IPTG (which activates ClpXP production). Therefore, increasing the concentration of IPTG will increase the degradation rate of the substrate (yEGFP-ssrA). shows a comparison between single-cell data from three trials (symbols) and time series obtained from a numerical best-fit algorithm of equation 1 (solid lines). The excellent agreement implies that ClpXP-mediated degradation is a Michaelis–Menten enzymatic process, as opposed to first-order decay, which is often assumed in models describing signalling or gene-regulatory networks. We have adapted the prokaryotic ssrA tagging system (; ) for use in eukaryotic yeast cells. The CGD699 strain allows for the tunable degradation of any protein by adding a short amino-acid tag. In reporter systems, the competing requirements of signal detection and dynamical resolution can be balanced without the need for additional cloning procedures. This system has several advantages over previously described systems. The degradation is tunable over a range of IPTG concentrations. The degradation tag is small and presumably unlikely to interfere with protein function. The small size of the tag simplifies construction of tagged genes by PCR amplification or use of a tagging vector, and many genes can be tagged in parallel. Several variant ssrA tags have been described (; ), allowing another level of control. Induction of ClpXP was non-toxic, as we found no significant change in growth rate or morphology at different IPTG concentrations. In addition, although a yeast homologue to ClpX has been identified (), it is located in the mitochondria and acts as a chaperone and not a protease. No homologue to ClpP has been found in yeast. The degradation components are stably integrated into the yeast genome, which allows it to be adapted for use with any number of tagged targets, either integrated or expressed from a plasmid. Due to the simple nature of the system, it might be portable into other genetically tractable eukaryotic models. There is evidence that bacterial ClpX and ClpP can interact with mammalian mitochondrial ClpX and ClpP (), but the restricted subcellular localization of these mammalian proteins should limit interaction. Increased degradation might be achievable by coexpressing the gene, which facilitates ssrA tag binding to the ClpXP protease (). Although we have only used the CGD699 strain with a fluorescent reporter protein, it should be applicable to any protein that can support the short C-terminal ssrA (). We expect that this system will prove useful for a variety of applications where increased temporal resolution is desired, both as a tool in the construction of small-scale synthetic networks (; ), and in systems biology studies, which seek to unravel network connectivity through perturbation of a given component (; ). cells were grown at 30°C with 300 r.p.m. shaking in SD dropout media containing appropriate amino-acid supplements. This was supplemented with 20 mg/l tryptophan, histidine and uracil and 100 mg/l leucine, as appropriate. DH5α cells were grown at 37°C with shaking at 300 r.p.m. in LB medium supplemented with 100 μg/ml ampicillin to maintain plasmids. DNA manipulations were carried out by standard techniques (). () was cloned from pRS31-yg () into an ssrA tagging vector that attached an ssrA tag (AANDENYALAA) to the C-terminal end, and then cloned back into pRS31-yg and integrated into the locus of strain K699. and were cloned from 2.300 by genomic PCR using Phusion polymerase (New England Biolabs). was placed behind the LacI-repressible promoter on pRS3GFPi (), replacing the gene. The promoter–gene–terminator cassette was then cloned into pRS404 and incorporated into the locus. Yeast-optimized was incorporated into the locus via pRS406 using a similar procedure. Mammalian-enhanced () was placed under the wild-type promoter on pRS3GFPa, replacing the gene. The promoter–gene–terminator cassette was then cloned into pRS405 and incorporated into the locus. Integrations were confirmed via selective auxotrophic plates. The resulting strain was designated CGD699, and is available upon request. Other strains were constructed by integrating the expression cassettes in a different order. Our initial attempts to construct this strain resulted in fluorescent signal but no corresponding drop in fluorescence when and were induced. RT–PCR results showed that the mRNA was transcribed, but Western blotting results showed that the ClpX was not translated (data not shown). We suspected that codon bias between the two organisms was a factor, and optimized for yeast expression. Sources of possibly deleterious codon bias were identified by comparing codon usages between and using the Codon Usage Database (). Arginine codons were commonly problematic. We introduced 10 silent mutations in two rounds of multiple mutagenesis using a previously described method (), and using Phusion polymerase (NEB). We omitted the megaprimer step, instead transforming single-stranded DNA directly into ultracompetent XL-10 Gold cells (Stratagene). The following primers were used (mutations induced are set bold): Primer 1: GGTCGCGGTATACAACCATTACAAAGTGCGCAACGGCGATACCAGCAATGGCGTCGAG Primer 2: GCTGCTGGCTGAAACGCTGGCGTGCTGGATGTTCCGTTCACCATGGCCGAC Primer 3: CTATCGCTAAGAAAGCGATGGCGAAAACCGGTGCCCGTGGCCT. A microfabricated chemostatic growth chamber that contained an on-chip media switch was used to measure ClpXP-mediated GFP decay within single cells. The microdevices were fabricated using well-documented poly-dimethylsiloxane (PDMS, Dow Corning Sylgard 184) replica molding techniques (). Chambers were fabricated to 4 μm in depth to facilitate quantitative fluorescence imaging of single cells (). Before experimentation, cells were grown from plates in appropriate auxotrophic media containing 0.5% galactose, 2% raffinose and the experimental IPTG condition for at least 24 h in batch culture. Each culture was maintained in log-phase growth. Yeast cells were loaded into microfabricated growth chambers at an OD of 0.4–0.6 and allowed to grow for 1 h in loading media. They were then transferred to media containing 2% (w/v) glucose and no galactose, the experimental IPTG concentration and a red fluorescent tracer dye (sulforhodamine 101, Sigma).

Analyses of several yeast genomes have confirmed the presence of a whole-genome duplication (WGD) in the clade including the bakers' yeast (; ). One of the most puzzling aspects of any WGD event is the question of what immediate selective advantage it conferred upon its possessor. Such an advantage would have been necessary to counteract the disadvantages of reproductive isolation (, ) and increased metabolic costs (; ) experienced by a post-WGD organism compared to its peers. Here, we try to place the genome duplication into the larger picture of the evolutionary history and ecology (; ) of this species. Several authors have speculated that WGD enhanced 's ability to metabolize glucose (; ; ) and/or to grow anaerobically (; ; ). We provide evidence that the preservation of some duplicate gene pairs created by the WGD was related to their contribution toward high glycolytic flux. We further consider the possibility that this selection was active soon after the WGD and may be the reason for its survival. In the presence of oxygen, most eukaryotes fully oxidize glucose to carbon dioxide and water using the TCA cycle, driving mitochondrial ATP synthesis with the accumulated reduced coenzymes. When oxygen is limited, a fermentative pathway is used instead, so that, in yeasts, glucose is converted to ethanol (). is unusual in that it prefers to ferment glucose into ethanol even in the presence of oxygen (the Crabtree effect; ; ), despite this pathway's energetic inefficiency. This phenotype is part of a suite of adaptations that allow to maintain very high growth rates when glucose is in excess (). On the basis of comparative genetics and genomics, some of these Crabtree-related adaptations can be dated to prior to the WGD and some to after it. For instance, the gene in seems to have acquired a role in the regulation of respiration since the split with the non-WGD species (; ). The alcohol dehydrogenase genes and in are the product of a gene duplication also post-dating the WGD (). The two resulting gene products allow to efficiently use glucose through fermentation (). The product of is primarily responsible for producing ethanol from acetaldehyde, while 's gene product is optimized to catalyze the reverse reaction (). On the other hand, a regulatory circuit that represses pathways that metabolize other sugars when glucose is abundant is conserved in . This circuit includes the MIG1 repressor (; ) and the glucose-sensing proteins (RAG4 in and SNF3 and RGT2 in ) that initiate signal cascades that in turn alter gene expression in response to glucose (, ; ). and are WGD paralogs of each other and orthologous to in . Notably, the two paralogs appear to have undergone functional divergence since duplication, with the former signaling low glucose concentrations and the latter higher concentrations (, ). Of course the primary metabolic enzymes of glycolysis, fermentation and respiration are ancient and widely distributed in yeasts (), and the ability to ferment glucose under anaerobic conditions also predates WGD (; ). These observations suggest that the yeast lineage leading to has been characterized by a long period of natural selection for rapid growth on substrates such as glucose. Several lines of evidence suggest that WGD may have played a role in this selection. A survey of over 40 yeast species both with and without the WGD indicates that the ability to grow anaerobically on minimal media, the presence of a Crabtree effect and the ability to generate petite mutants are all strongly associated with yeasts possessing the WGD (). Another study also found a general, though weak, trend for higher rates of ethanol production in post-WGD yeasts (e.g., and ) than in non-WGD yeasts (). There is also an excess of energy metabolism genes surviving in duplicate from this event (). In this paper we propose that the WGD had an important impact on gene dosage and that this dosage change had a knock-on effect on the lineage of post-WGD yeasts (including ) uses glucose. We propose three linked hypotheses relating glucose metabolism to the yeast WGD. First, we suggest that the loss of duplicate copies of other genes after WGD increased the concentrations of glycolytic enzymes (which survived in duplicate). Second, we propose that the inherent kinetics of fermentation and respiration meant that this increase in enzyme concentration gave rise to an increased preference for fermentation in the partially polyploid yeast. Finally, we propose that this yeast had a selective advantage because it was able to use glucose more rapidly than its ancestors and hence out-compete other yeasts when glucose was in excess. In the sections below, we briefly introduce each hypothesis in turn. #text In this work, we propose a link between whole-genome duplication, increases in enzyme concentrations and the preference of modern to ferment glucose in the presence of oxygen. We find evidence for the preferential retention of duplicate genes from WGD in the glycolytic pathway, as well as evidence that, at least for the kinetic constants measured in modern , increases in enzyme concentrations both tend to increase glycolytic flux and to favor ethanol fermentation over oxidative respiration. We also confirm previous work showing that organisms with fast but inefficient metabolisms can have a selective advantage over their more efficient kin under certain conditions. Collectively, these results tend to support the hypotheses proposed, although of course much remains to be done in order to fully understand the effect of WGD on metabolism and on glycolysis in particular. If the above hypotheses are borne out by further analysis, they will help integrate a number of facts regarding the biology of yeasts including , such as the origins and phylogenetic distribution of the Crabtree effect (), the evolutionary rationale for the patterns of duplicate gene retention in yeast () and the nature of ecological competition among microbes (; ). It will also be interesting to study the effect of WGD events in other taxa on glycolysis. Duplicate copies of the glycolytic genes in vertebrates have been studied, but uncertainties in phylogenetic reconstructions involved made it difficult to determine if observed duplications among these genes owed their origins to WGDs (), leaving open the possibility of future analyses with gene order data to clarify the nature of the genes surviving from these WGDs. The chief question raised by the above data is whether dosage selection for increased glycolytic flux was itself the reason for the survival of this WGD in the first place. This intrinsically attractive hypothesis is very difficult to test, and data that both support and undermine its plausibility can be found in the literature. Speaking against this possibility is the view that the uniform doubling in gene content through WGD should not change relative enzyme concentrations. Thus, tetraploids of wild-type cells have relative gene expression profiles that are essentially identical to diploid cells (). Other data, such as that regarding cell size, is ambiguous with respect to the idea of dosage selection. If cell volume does not also double after WGD, even identical relative gene dosages can still yield changes in absolute enzyme concentrations. Artificial tetraploid strains of actually showed more than doubling of cell volume relative to diploids, in theory implying a general in enzyme concentrations (), which would tend to speak against our hypotheses. As a further complication, it is currently an open question whether this WGD was a true duplication of chromosomal content (autopolyploid) or a hybridization between two related species (allopolyploid). In the case of allopolyploidy, others have argued that the scaling of gene expression after hybridization will in general not be uniform and linear (). On the other hand, there are known cases where duplications are associated with apparent selection for gene dosage. These include aneuploidies observed in vineyard and deletion mutant strains of (; ; ) as well as in clinical and laboratory isolates of (; ; ). In the case of deletion mutants in , several aneuploid chromosomes were observed to carry a close homolog of the deleted gene and in some cases showed clear growth advantages as a result (). Drug-resistant strains of carry multiple copies of chromosomes or chromosome arms where genes conferring drug-resistance reside (e.g., ; ; ). It is also suggestive that genes that survive in duplicate from a WGD in the ciliate are enriched for highly expressed genes (). Collectively, these points argue that large-scale duplications can indeed be associated with selection for increased dosages of certain genes. Although existing data give only mixed support to the idea of dosage selection preserving a WGD, the hypothesis has some very attractive features. One of the most important is the relatively simple nature of the changes required to produce it. We argue that even if relative gene dosages were unchanged immediately after genome duplication, the rapid gene loss that followed the WGD () would have quickly altered this situation. It is reasonable to argue that those duplicate pairs that survived this loss would experience an increase in relative expression as a result. Other mechanisms of fixing gene duplications, such as neofunctionalization or subfunctionalization, require specific mutations in one or both of the duplicate genes to create a selective pressure for duplicate maintenance. However, when some fraction of genes are under selection for dosage, mutations leading to loss of ANY other genes elsewhere in the genome will be sufficient to yield a selective advantage. Since this possibility massively increases the mutational ‘target' that can yield the beneficial phenotype, it follows that such a beneficial mutation will arise much more rapidly than is the case for neo- or subfunctionalization. Of course, it seems rather unnecessary to duplicate an entire genome in order to change the flux in a single pathway. However, we suspect that the overall picture is more complicated than we have described here, with other duplicate genes (such as the previously mentioned pair of glucose-sensing genes) preserved as part of an adaptation to growth on glucose (, ; ). For example, four enzymes of the pentose phosphate pathway also maintain WGD duplicates in . This pathway has an important role in biosynthesis, and it would be reasonable to expect that increasing growth rates would also require increasing its flux. It is clear that this WGD was only one event in a long process of adaptation leading to modern bakers' yeast. It did, however, have lasting consequences, since a number of the duplicated genes have since evolved new or more specialized roles. For instance, the HXK and PYK gene pairs appear to have partitioned their ancestral functions to roles in high and low glucose levels (; ; ). Equally intriguingly, two glycolytic duplicates may have acquired roles in cell proliferation: in meiosis for () or mitosis for (; ). Under our model, duplicate copies of glycolysis genes were initially maintained for dosage reasons, but subsequent tuning of enzyme expression levels may have later freed one paralog to innovate (). More generally, our argument suggests that the niche inhabited by is only one of many such niches about which we know very little. To properly understand the genetics and metabolisms of yeasts, it will therefore be vital to appreciate the role played by natural selection in adapting each species to a particular niche. Behavior seen in the laboratory, under conditions very different from the wild, can then be understood in the light of this (currently unknown) ecology.

During oogenesis, an egg grows very large and inherits from its mother all the nutrients and cell components needed to proceed through a series of cell cycles after fertilization (). In many types of embryos, these early division cycles rapidly alternate between DNA replication (S-phase) and mitosis (M-phase). Later in development, the synchrony and speed of the first divisions is lost, and gap phases (G1 and G2) are introduced into the somatic cell cycle (). The egg is an extreme case. Its nuclei proceed through 13 very rapid (10–12 min) divisions without cell division. As a consequence, three hours following fertilization, 6000 nuclei share the same cytoplasm (syncytium). The rapidity of these early cycles can be explained by an abundance of maternally supplied cell cycle components. After mitosis 13, the nuclei become cellularized. Some cells arrest in G2, whereas other cells continue to divide at a slower and more variable schedule (; ). The molecular basis of the first 13 rapid, synchronous nuclear division cycles is the subject of mathematical modeling in this paper. The master regulators of the eukaryotic cell cycle are the cyclin-dependent protein kinases (Cdk's). To be active, a Cdk has to be associated with a cyclin partner, which determines the substrate specificity and subcellular localization of the Cdk/cyclin complex. The prototype of a Cdk/cyclin pair is M-phase-promoting factor (MPF), first identified in frog eggs. MPF is a complex of Cdk1 and a B-type cyclin (CycB). Cdk1 subunits are usually present in excess in cells, and therefore do not limit formation of Cdk/cyclin dimers. Cyclin subunits, on the other hand, fluctuate dramatically during the cell cycle and thereby play a major role in determining MPF activity (). In growing cells, cyclin synthesis is regulated at the transcriptional level. In early embryonic cells, transcription is blocked or greatly restricted because they lack G1- and G2-phase (). In both growing and embryonic cells, cyclin level is controlled by proteolysis. Ubiquitylation of CycB by the anaphase promoting complex (APC) targets them to the proteasome for degradation (). In embryos, a protein called Fzy (encoded by the gene) is responsible for targeting CycB to the APC during exit from mitosis (). The Fzy/APC complex is activated by phosphorylation by MPF (), creating a negative-feedback loop (+/−) in the reaction network. The activation of Fzy/APC by MPF is an indirect process (). In addition, Cdk/cyclin activity can be controlled by phosphorylation of the Cdk subunit. Phosphorylation of a tyrosine residue (near the N terminus) inhibits the protein kinase activity of Cdk/cyclin complexes (the phosphorylated form of MPF, P-Cdk1/CycB, is called preMPF). Cdk phosphorylation is catalyzed by the protein kinase Wee1, and the phosphate group is removed by a phosphatase called String in (). Wee1 and String proteins are themselves phosphorylated by MPF, creating a pair of positive-feedback loops. String is activated by MPF (a +/+ loop) and Wee1 is inactivated by MPF (a −/− loop). In sea urchin and frog embryos, the first 12 cell cycles are known to be driven by a cytoplasmic clock that causes periodic degradation of the CycB subunit of MPF as cells exit mitosis (). The resultant oscillations of MPF activity control both nuclear divisions (M-phase) and the characteristic surface contractions that persist even after enucleation (). In contrast, in , observed that total CycB level and MPF activity remain high (not oscillating) during the first eight cycles. After cycle eight, small fluctuations appear in both CycB level and MPF activity with increasing amplitude. Even though CycB degradation might appear negligible during these early cycles, introduction of a nondegradable form of CycB into a embryo blocks mitotic cycles, which underlines the importance of CycB degradation at certain stages of the cell cycle (; ). The apparent paradox surrounding CycB degradation during embryogenesis can be resolved by recognizing that CycB degradation occurs only locally, in the vicinity of the mitotic spindle (; ). In this paper, we use mathematical modeling to explore whether the hypothesis of local CycB degradation gives an adequate description of CycB patterns during the first 13 nuclear division cycles of the embryo. After introducing compartmentalization and local degradation, we show that it is possible to simulate the key features of early embryonic cell cycles in . The model also reproduces the effects of alpha-amanitin treatment, loss-of-function mutations and overexpression mutations. Our model for cell cycle regulation in the early embryo, inspired by of , is diagrammed in full in . The model tracks the interactions of MPF, Fzy, Wee1 and String in the cytoplasm and in the nuclei. In , Fzy/APC is localized at centrosomes, kinetochores and spindles (; ; ). Consequently, we assume that CycB subunits of MPF undergo regulated destruction only in the nuclear compartments. In our model, MPF activates Fzy/APC through an ‘intermediary enzyme' (IE), because some evidence indicates that the activation step is indirect (). These interactions comprise a delayed negative-feedback loop that generates local oscillations of MPF activity in the nuclear compartments. Meanwhile, Wee1 and String are also regulating MPF activity in both compartments by phosphorylating and dephosphorylating (respectively) a tyrosine residue on the Cdk subunit of MPF. Using the basic principles of biochemical kinetics, we translate the diagram into a set of ordinary differential equations (). The equations, which describe the time-rates of change of the fluctuating protein species in the diagram, contain a number of unknown rate constants that must be estimated by fitting the model to the available data (mutant phenotypes, responses to drug treatments, and so on). The parameter values reported in are suitable for simulating the experiments described in this paper. We distinguish two compartments in the embryo (): cytoplasm and nuclei. What we call the ‘nuclear compartment' is not the volume enclosed by the nuclear envelope, because the nuclear envelope breaks down during mitosis. Despite the loss of a barrier between nucleus and cytoplasm, we assume that CycB degradation during M-phase occurs only in a limited region (our ‘nuclear compartment') in the vicinity of the mitotic spindle. Hence, in our model, the nuclear compartment persists in separation from the cytoplasmic compartment throughout the nuclear division cycles. At telophase, the number of nuclear compartments doubles (or, nearly so), as described later. We will assume that the transport coefficients for MPF, Wee1 and String between nucleus and cytoplasm do not change as the nuclear envelope breaks down and reassembles each cycle. In the Discussion, we will present evidence that our conclusions are not significantly changed by the more likely assumption that intercompartmental transport increases during mitosis. The volume of the nuclear compartment is assumed to be the product of the number of nuclei () and the volume of a single nuclear compartment (), that is =. Consequently, cytoplasmic volume is =(1−ɛ), where =total egg volume (constant) and ɛ=/≪1. Next, we need to specify how individual proteins are distributed between nuclei and cytoplasm. For example, MPF is present in both cytoplasm and nuclei, where its concentrations are denoted [MPF] and [MPF], respectively. where and are the fluxes of MPF (number of molecules per unit time) into and out of a single nucleus. We assume that =[MPF] and =[MPF], where and are permeability constants and is the surface area of a single nucleus (assumed constant). For proteins that are actively accumulated in nuclei, ≫. where IN, OUT, SYN, and so on stand for the rate expressions for transport and chemical reactions. Similar equations hold for preMPF. For MPF and preMPF, we assume that =0. Wee1 and String are also distributed between cytoplasm and nuclei, and these proteins can be phosphorylated and dephosphorylated in both compartments. As a consequence, we consider four different forms of each protein (unphosphorylated and phosphorylated forms in the cytoplasm and in the nuclei), where the cytoplasmic forms (Stg, StgP, Wee1 and Wee1P) interact with MPF and preMPF, whereas the nuclear forms (Stg, StgP, Wee1 and Wee1P) interact with MPF and preMPF. For String and Wee1, we can write balance equations for the total concentrations : By assumption [Wee1] is constant in the model, so we can compute Wee1P from the equation: In the model, we assume that nuclei enter M-phase when MPF activity abruptly increases, and they divide when MPF activates Fzy/APC (i.e., when Fzy activity increases through 0.5). At each cycle, the nominally doubles. However, to account for the few nuclei that do not divide, we multiply the existing at each division by 1.95 rather than by 2. If we assume that the concentrations of protein species do not change at the point of discontinuity (when increases to 1.95 ), then there is a sudden ‘creation of matter' needed to populate the new nuclei with all their protein components. For proteins that turn over (synthesis and degradation), this new material is quickly equilibrated and the excess removed by protein degradation, with no appreciable effect on the solution of the differential equations during the 13 cycles. But for conserved species (e.g., Wee1), there is a steady increase of total concentration, as a little new material is created at each division. The increase starts to become noticeable at the 12 division. The problem can be remedied by rescaling concentrations after nuclear division according to the prescriptions (see Materials and methods section for details): shows the results of a numerical simulation of the model with localized cyclin degradation. For this simulation, we assume no synthesis or degradation of String, that is, [String]=constant. The initial concentration for MPF in the cytoplasm is high (one arbitrary unit) and this activity suffices to keep Wee1 inactive and String active in the cytoplasm (at least initially). The first oscillations of the nuclear concentration of MPF are very rapid. However, in later cycles, they slow down, and finally arrest in G2-phase of the 15th cycle with low MPF activity. The first 13 cycles are so rapid in compared to , because they are driven by nuclear import of pre-formed cytoplasmic CycB rather than by synthesis of CycB. In the early cycles, MPF enters the nucleus so quickly that nuclear Wee1 cannot inhibit it and, as a result, the rapid accumulation of MPF drives the nucleus into mitosis. These early cycles are driven by the negative-feedback circuit involving MPF, IE and Fzy/APC. As the and the volume of the nuclear compartment nearly double after each mitosis, the concentration of MPF+preMPF in the cytoplasm decreases. The drop in the cytoplasmic concentration slows down the transport of MPF into the nuclei and thus slows down the oscillation. As Wee1 kinase activity in the nucleus becomes more and more comparable to the nuclear entry rate of MPF, MPF complexes can be inhibited through tyrosine phosphorylation by Wee1. MPF activity must reach a certain threshold to switch Wee1 kinase off and to activate String. As reaching this threshold takes some time, the period of the oscillation increases further. Eventually, total cyclin level in the cytoplasm is so low that nuclear MPF concentration cannot surmount the threshold level to enter mitosis. Hence, Wee1 remains active and String inactive, and MPF oscillations cease. In , the ‘embryo' has an extended cycle 14 and arrests in cycle 15. (The cycle number () for which the model arrests depends sensitively on parameter values.) In this section, we use one-parameter bifurcation diagrams to characterize the mitotic control system with localized cyclin degradation (; ). A bifurcation diagram describes the stability of a dynamical system as some key parameter (the parameter) is changed. As bifurcation parameter, we introduce ‘cycle number' , defined by =1.95. With this definition, is the number of nuclei in the egg during cycle number . When computing a bifurcation diagram, is treated as a real number, even though increases stepwise during simulations. For the bifurcation diagrams in , the state of the control system (on the vertical axis) is characterized by either nuclear MPF or total CycB (to give two different views of system state). For a large , the control system has a single, stable steady state (solid line) with low MPF concentration. At this steady state, MPF is tyrosine phosphorylated both in the cytoplasm and in the nuclei. This stable steady state represents a G2 arrest. When is small, the control system is in a region where a branch of unstable steady states (dashed line) is surrounded by stable limit cycle oscillations (filled circles). These oscillations have large amplitude of [MPF], but small amplitude of [CycB] (). As increases after every oscillation, we move from left to right ( increasing) on the bifurcation diagram as the embryo develops. Between cycles 11 and 12 (), the limit cycle oscillations undergo a pair of ‘cyclic fold' bifurcations, which indicate a qualitative change in the oscillation mechanism. For ⩽11, the oscillations are driven by the negative-feedback loop (cyclin degradation) alone. For ⩾12, the positive-feedback loops (involving MPF phosphorylation) contribute to the oscillatory mechanism. These positive-feedback loops become more and more significant as increases, eventually creating alternative stable steady states between cycles 14 and 15. The stable steady state with low activity of MPF blocks the oscillations at a SNIC bifurcation (saddle-node-invariant-circle, where the period of oscillation tends toward infinity) at =14.33. Hence, the MPF control system completes cycle 14 and arrests at a stable steady state in interphase of cycle 15. This is a problem, because eggs arrest in interphase of cycle 14. It could be corrected by adjusting some parameters to move the SNIC point to 13 <<14, but there are other more serious problems to be solved, along with this one, in the next section. Interestingly, the bifurcation diagram does not change very much in the absence of String (), that is, a maternal loss-of-function mutation. The reason for this insensitivity is the existence of a second protein phosphatase, Twine, which overlaps the function of String (). (Twine activity is represented in the model by the parameter ′.) In the absence of String, the amplitude of MPF oscillations is slightly reduced, and the SNIC bifurcation point moves to a slightly smaller value of (=13.22). Consequently, MPF oscillations stop one cycle earlier (in cycle 14) than for [String]=0.8. Although the simulation in is superficially similar to observations of MPF fluctuations in fruit fly embryos (), there are significant quantitative differences: the number of division cycles is incorrect (easily fixed), and the total amount of String in the embryo is not constant (less easily fixed). String protein level in embryos is low at fertilization, rises for seven or eight cycles and drops gradually to zero at interphase 14. As String level determines the position of the SNIC bifurcation (where oscillations are suppressed), we need to extend the model to changing levels of String protein. To this end, we take into account that mRNA is stable until cycle 14 and is degraded abruptly in the first 20 min of interphase 14 (). Two distinct mechanisms seem to be responsible for mRNA degradation (; ), one operating on maternal mRNA and the other one mediated by zygotically induced genes. To describe String dynamics, we append to the model () three differential equations: for mRNA (equation), for the mRNA of an unknown factor (equation) that is responsible for degrading mRNA, and for the corresponding protein level (equation). We assume that (mRNA for factor ) is synthesized in the embryo at a rate proportional to . The protein is produced at a rate proportional to . The rate constants in equations and are chosen so that there is no detectable synthesis of until about the 10th cycle, and then its level rises sharply in cycles 11–13. (Other assumptions might be more reasonable: for example, may not be synthesized during the very rapid cycles 1–6, when the DNA is likely unavailable for transcription. In that case, the rate constants would have to be readjusted to maintain the sharp rise in in cycles 11–13.) We assume that the gene is not transcribed during early embryonic cycles, so the differential equation for mRNA (equation) has only degradation terms. The first term represents degradation by maternal products, whereas the second term is degradation by . Hence, String message level drops slowly at first and then increasingly faster in cycles 11–13. String protein is assumed to be synthesized at a rate proportional to its mRNA (Stg) and degraded with first order kinetics. Hence, total String protein level rises during the first 9–10 cycles and then decays away after its message is destroyed. With these amendments to the model, a better description of the early cell cycles in is achieved (). The levels of String message and protein change as observed in Edgar's experiments () and the embryo arrests solidly in interphase of cycle 14. In the simulation (), oscillations in String phosphorylation become noticeable around cycle 8, whereas, in experiments, fluctuations in String phosphorylation are already observed in cycle 5 or 6 (). If this model is basically correct, it should account not only for all features of nuclear division in wild-type embryos, but also for subtle differences in phenotypes of mutant embryos. In this case, the relevant mutants are deletions and overexpressions of , , , and . Simulations of these mutants are summarized in . Deletion of breaks the negative-feedback loop on which the oscillations depend and is, of course, lethal to embryonic development (). The model predicts that a half-dose of ( heterozygote) is similar to wild type, whereas overexpressing may produce an extra nuclear division. Overexpressing speeds up the cycles and adds an extra nuclear division (; ). Conversely, a half-dose of causes delays in cycles 10–13 (; ). Deletion of has no effect on the early mitotic cycles, as long as is in place (see : deletion halts in cycle 14), but deletion of both genes is lethal to the egg (). Interestingly, deletion of and a half-dose of cause the early cycles to stop one division earlier than normal (in cycle 13) in the model and in experiments (). In the bifurcation diagram for this mutant, the SNIC bifurcation is moved to =12.57 (i.e., the egg arrests in cycle 13). Overexpressing and/or is tricky to interpret. We might expect such mutations to speed up the cycles and perhaps add an extra nuclear division (arrest in cycle 15). An extra division is observed in a small fraction (3–5%) of embryos (). In simulations, extra peaks of MPF are observed, but the later cycles are of reduced amplitude and it is not clear whether they could effectively drive a complete mitosis or not. The reason for the reduction in amplitude of MPF oscillations is clear from the bifurcation diagram (). The SNIC bifurcation is lost and the later oscillations arise from a supercritical Hopf bifurcation (small amplitude limit cycle oscillations). The bifurcations in this case can be very complex, but the end result in simulations is a damped oscillation of MPF. The gene is not essential for early nuclear divisions (). In simulations, the -deletion mutant (like and overexpression mutants) arrests via damped oscillations at a supercritical Hopf bifurcation (). Simulation of a -overexpressing mutant (6 × ) predicts nuclear-division arrest in cycle 12 (). showed that if an embryo is treated before cycle 6 with alpha-amanitin (an inhibitor of RNA polymerase), then 88% of the nuclei arrest at interphase of cycle 15 instead of cycle 14. If the same treatment is performed later, then the extra cycle is not observed (; ). These results are nicely reproduced by simulations. In , we let the model run up to =55 min (just before cycle 6), then set =0 (i.e., no further synthesis of message), then continue the simulation. The concentration of never gets very large, and the degradation of mRNA is slower. As a result, total String protein is slightly higher, allowing an extra cycle to occur. In contrast, when we simulate the same treatment at =75 min (after cycle 6), concentration is higher, mRNA drops faster and no extra cycle is observed. Early development of embryos has been thoroughly studied by developmental geneticists and molecular biologists. Prompted by beautiful data on gene expression patterns, many groups have proposed models for pattern formation in post-blastoderm embryos (; ; ). Until now, however, no one has systematically explored the role of protein interactions during the first 13 nuclear division cycles in the syncytial, undifferentiated embryo. These cycles are remarkable because of their great speed (approx 10 min cycle time) and because they are not associated with large changes in observable activity of MPF, as seen for example in the embryos of sea urchins and frogs. Although the bulk cytoplasm of the syncytial egg contains a massive amount of active MPF, the activity of MPF in the vicinity of chromosomes may fluctuate dramatically, first of all because active MPF must be imported into the nuclei during interphase, and secondly because MPF activity may be cleared from chromosomal regions by rapid CycB degradation on anaphase spindles. Many authors have previously forwarded this hypothesis, or something similar, to explain the curious features of nuclear division in the early embryo of (; ). We have explored this hypothesis in terms of a mathematical model that distinguishes nuclear compartments from the bulk cytoplasm. Nuclear import of active MPF drives DNA replication, nuclear envelope breakdown, spindle assembly and congression of chromosomes to the metaphase plate. Local degradation of CycB at the spindle drives chromosome segregation and nuclear envelope reformation. Simulations of the model agree with salient features of wild-type embryos, as summarized in of and our . Wild-type embryos arrest in interphase of cycle 14, but mutant embryos may arrest sooner or later (or may not cycle at all), depending on their genetic make-up. The model is consistent with these subtle effects on nuclear division cycles, as summarized in . Lastly, we have shown that the model accounts for differential effects of alpha-amanitin treatment before and after cycle 6. Our model does not attempt to describe the complex regulatory signals introduced at the mid-blastula transition, when many zygotic genes are newly transcribed. italic #text The models are defined in terms of ordinary differential equations and simulated with the software XPPAut () (). An ‘ode' file for simulating is provided in the . To compute , some numerical constants must be changed, as described in the ode file. When modified as described therein, the ode file can also implement the NEB model described in the Discussion. A second ode file is supplied for computing bifurcation diagrams (as in ) using XPPAut. In addition, our model is available as an SBML file from the BioModels database (, temporary accession MODEL1509031628). Define and to be the concentrations of species in the nucleus and in the cytoplasm before nuclear division (at time ), and and to be the concentrations after division (at time +Δ). Then: Insisting that , we find that and must be adjusted at nuclear division as follows:

One of the most characterized protein degradation systems in is the AAA+ protease family, which includes ClpXP and ClpAP. These proteases recognize and degrade -tagged proteins (). The tag and its variants were fused to many proteins () to reduce their half-lives for various synthetic circuits (; ; ). Such synthetic biological circuits enable the testing of operating principles governing biological networks and the exploration of potential applications that are not limited by natural systems (; ; ; ; ; ; ; ). A modified version of the tag has also been engineered recently to allow controllable degradation (). data () showed that tagged proteins display a first-order degradation kinetics, suggesting a relatively high , whereas data () showed a much lower (75 nM) which leads to a zeroth-order kinetics. This discrepancy raised the issue of protein degradation reaction order used in designing synthetic circuits and prompted us to examine the kinetics of protein degradation in detail. Interestingly, we observed that the degradation kinetics for the native tag in single cells was in fact zeroth-order. However, population measurements showed that the degradation kinetics was first-order. This discrepancy is another example of single-cell behavior masked by population average. Using both experimental and theoretical analyses, we showed that the discrepancy was caused by the long-tailed distribution of the initial protein level. Moreover, through simulation and mathematical analysis, we demonstrated that the difference between the single-cell and the population measurements would exist even when all the degradation processes were synchronized. Therefore, the discrepancy was not a result of the asynchronous dynamics. Furthermore, theoretical analysis showed that the kinetic form of protein degradation can have a profound effect on the parametric robustness of biological circuits (). The importance of accurate kinetic model for protein degradation in predicting circuit properties had also been highlighted (). Through computational analysis, we showed that when the protein degradation kinetics approaches the zeroth-order limit of the Michaelis–Menten kinetics, the parametric robustness of synthetic oscillators can be significantly enhanced. Therefore, the zeroth-order tag used in the gene-metabolic oscillator reported previously () may improve the robustness of the oscillation. To determine the kinetic order of protein degradation, we fused the corresponding codons of two different versions of the tags, AANDENYA (LAA) and AANDENYA (ASV), to the coding sequence of the green fluorescent protein (GFP) and expressed them under an IPTG-inducible promoter in glucose medium. The fluorescence property of GFP provides a convenient way for measurement, both at the population and the single-cell level. After resuspension in acetate to wash away IPTG and to induce a time lag in growth, the protein degradation in the nondividing cells was measured using quantitative time-lapse fluorescence microscopy. The LAA tag is naturally found in and the ASV tag is a modified version of LAA, which has a longer half-life (). Interestingly, the degradation dynamics of the LAA-tagged GFP for individual cells displayed a zeroth-order kinetics (), similar to the data reported previously (). This result indicates that the protein level is significantly higher than the of the protease in a Michaelis–Menten kinetics. However, when measured in a bulk solution, the degradation dynamics exhibited a first-order kinetics (), which indicates that the protein level is much smaller than the . To eliminate the potential effect of residual IPTG, we added chloramphenicol, a translation inhibitor, to the media during protein degradation measurements. The discrepancy between single-cell and population kinetics still existed (), indicating that residual IPTG was not the cause for the observed phenomenon. In addition, a quantitative western blot showed that the initial GFP–LAA level at population level is higher than 10 μM (data not shown), which is much larger than the of ClpXP (75 nM) for LAA-tagged protein (). Thus, it is reasonable to observe that the degradation kinetics of LAA-tagged protein is zeroth-order. In contrast, GFP tagged with ASV, which has a longer half-life, displays first-order degradation kinetics at both the single-cell and the population levels (). It is possible that the LAA- and ASV-tagged proteins may be degraded by different proteases. The more dominate protease for the degradation of LAA-tagged protein is ClpXP, whereas ClpAP plays a minor role (). The last three amino acids of the tag have been shown to be the binding site for ClpX (). Therefore, proteases other than ClpXP may also be responsible for the degradation of GFP–ASV. Using the Keio-knockout collection (), we showed that the Δ strain showed more proteolytic activity than the Δ strain (). This result suggests that ClpAP plays a more significant role in the degradation of ASV-tagged proteins. This finding may explain the difference in the single-cell degradation kinetics between the LAA- and ASV-tagged proteins. To prove the cause of the kinetic discrepancy between single-cell and population measurements, we found that the initial protein level distribution from the LAA experiment displays a long-tailed distribution with a range over 160-fold (). Averaging the single-cell data from the LAA experiment produced a first-order degradation kinetics, consistent with the measurements performed in bulk solutions (). The large initial protein distribution could be caused by plasmid instability. We tested this possibility by measuring the percentage of cells that still retains the plasmid after induction with IPTG. After 2, 3, and 4 h of induction, we plated approximately 200 cells onto a LB plate without any antibiotics and allowed the cells to growth overnight. We picked 100 colonies to test for ampicillin resistance (the antibiotic resistance in the GFP expression plasmid) and all 100 colonies grew in the presence of ampicillin. Therefore, plasmid lost during the course of induction does not seem to be the cause of the long-tailed distribution observed in our experiments. To avoid overrepresentation of the cells expressing low levels of GFP, we excluded the cells with low GFP level (first bar in ) in the analysis. This elimination decreased range ratio to ∼10 (first bar in ), but did not alter the masking effect by population diversity (). Excluding more low-GFP cells further reduced the range distribution and gradually eliminated the discrepancy between single-cell and population measurements. The non-Gaussian, long-tailed distribution of proteins was also observed elsewhere (). The long-tailed distribution is proposed to be caused by the noise in molecular partitioning during cell division and the noise in protein synthesis rate (). To explain how a population of cells exhibiting zeroth-order kinetics at the single-cell level can create a first-order kinetics at the population level, we developed a stochastic model based on the Gillespie algorithm () to simulate the proteolysis (see , for more detail). We first tested the hypothesis that initial protein distribution resulted from protein expression noise is the cause of this discrepancy. The initial protein level, , varies for different cells following either an exponential or a normal distribution, and the degradation kinetics for each cell is modeled using the Michaelis–Menten kinetics with a smaller than the protein level (zeroth-order degradation). One thousand cells were used in each simulation. When follows the exponential distribution, the population average of the zeroth-order degradation kinetics from single-cell becomes first-order, provided that the range of distribution is sufficiently large (). This is not true, however, for normally distributed , regardless of the range of the distribution (). We also investigated the role of protease distribution on the discrepancy. We performed the simulation with an exponentially or a normally distributed protease levels, but with no initial protein distribution. Given a large enough range ratio, both exponential and normal protease distributions caused the population degradation dynamics to appear first-order (). These results demonstrate that a wide distribution in either the protein expression level or the protease level could distort the protein degradation kinetics in a population average. We constructed an analytical model to examine the effect of initial protein level distribution on the population kinetics (see for more detail). For simplicity, we let the zeroth-order protein degradation rates be identical in each cell. where (, , ) is the concentration of the protein, is the initial protein concentration, is the degradation rate and is the time. We first allowed the initial protein level, , to follow an exponential distribution ranging from to . When averaged over all cells, the degradation kinetics is separated into three regions: (i) no cell has reached the zero protein level (⩽), (ii) some cells have reached the zero protein boundary (⩽⩽), and (iii) all the proteins are degraded in all cells (⩾). The first region will always be zeroth-order, and the third region will always be a flat line (slope equals to zero). The second region, however, is exponential in time. The size of the second region depends on the range of distribution. In the limiting case where the distribution is infinitely wide, → 0 and → ∞, then we have which is exactly an exponential form with the initial concentration corresponding to the mean of the initial GFP distribution. Thus, the range of the exponential distribution defines the deviation from the first-order kinetics. Because the range of distribution of the initial GFP level observed in our single-cell measurement is over 160-fold (), it is sufficient to give rise to the first-order kinetics that we obtained from bulk measurement. When the initial protein level follows a normal distribution, the population average contains an exponential term raised to , plus other time-dependent terms (). Therefore, normally distributed initial protease level will not yield exact first-order population degradation kinetics. We also derived an analytical solution for the case where the degradation rate, which is related to the protease level, follows an exponential or a normal distribution (). The resulting analytical solution of the population dynamics yields a nonlinear combination of exponential terms. A combination of initial protein and protease distribution can also create an apparent first-order population degradation kinetics. We also investigated the effect of protein degradation on synthetic oscillators. We used the metabolator () as an example, which is a synthetic gene-metabolic oscillator that integrates transcriptional regulation into the metabolism to generate oscillation. The metabolator is consisted of a flux-carrying network with two interconvertible metabolite pools: Acetyl-CoA (AcCoA) and Acetyl-phosphate (AcP) (). These two pools of metabolites are catalyzed by two enzymes, phosphotransacetylase () and acetyl-CoA synthetase (). The expression of and are negatively and positively regulated by AcP, respectively. The oscillation dynamics of the metabolator is driven by the glycolytic flux. The integration of genetic and metabolic control is a hallmark found in many natural oscillators (; ; ; ). In the original metabolator model (), protein degradation was described as a first-order process, which was lumped together with the term describing dilution by cell growth. To investigate the effect of protein degradation kinetics, we modified the original model by including a Michaelis–Menten kinetics for protein degradation, with a of 75 nM () (see ). Thus, this current model includes a protein degradation kinetics that falls in the zeroth-order regime and a first-order protein dilution due to cell growth. When the zeroth-order degradation rate, , is equals to zero, the current model reduces to the original model where only the first-order degradation term exists. Linear stability analysis was then used to map the parametric loci that give rise to Hopf bifurcation (See for details). shows that by including the zeroth-order process, the parameter space of oscillation significantly increased. When =0, the -axis represents the range of first-order degradation rate, , at which oscillations will occur with the original model. Increasing significantly enlarges the parameters space that leads to oscillation. We also explored the effect of the zeroth-order degradation kinetics on other parameters of the metabolator. We chose the parameters for the zeroth-order degradation based on literature data (; ). The phase diagram in shows that the zero-order degradation kinetics gives a larger oscillatory region than the first-order degradation kinetics. A point on the phase diagram is chosen so that it falls within the zero-order boundary, but outside of the first-order boundary. Using this set of parameters, the model with first-order degradation kinetics reaches a stable steady state (no oscillation), whereas the model with zero-order degradation displays oscillation (). This work demonstrates a discrepancy between single-cell and population measurements. Typically, single-cell measurements are important when multimodal distribution or asynchronous dynamics exists. We showed that even single-modal and synchronous single-cell dynamics can be masked by population heterogeneity, depending on the initial protein or the protease distribution. When the initial protein or protease level exhibits a wide exponential distribution, the zeroth-order protein degradation in single cell will appear to be first-order as a population average. In the example shown here, GFP–LAA level follows an exponential distribution with a wide dispersion (160-fold), which is sufficient to explain the first-order kinetics observed in population. Coupling with protease distribution can also create first-order population kinetics from zeroth-order single-cell kinetics. The phenomenon described here also can explain differences between and protein degradation measurements. For example, the value of ClpAP () has been estimated to be 10-fold higher than the value measured (). This difference may be caused by the masking effect of population heterogeneity. Therefore, when considering the differences between and measurement, population heterogeneity might also need to be accounted for. Zeroth-order kinetics had been shown to generate ultrasensitivity in enzymatic systems such as isocitrate dehydrogenase () and glycogen phospholyase (). The ultrasensitivity created by zeroth-order kinetics had been proposed to play an important role in developmental threshold () and the responses to morphogen gradients in embryonic ventral ectoderm (). Here, we show that zeroth-order protein degradation expands the parameter space for oscillation in the metabolator (). Interestingly, the LAA -tagged protein degradation, which was used in the metabolator (), displays a zeroth-order kinetics when measured in single cells and may enhance the robustness of the synthetic circuit. However, the exact model for oscillation could not be ascertained until all kinetics involved are determined. Zeroth-order degradation is generated when the level of protein is significantly higher than the of the protease. Hence, the protein level does not have to be high to achieve zeroth-order degradation. In addition, protein localization can further enhance the local concentration of the protein relative to the protease, thus allow the degradation kinetics to occur in the zeroth-order regime even when the number of protein per cell is lower than the . Zeroth-order degradation by ClpXP was also observed in the degradation of CtrA, a master regulator in cell-cycle regulation (). The activity of CtrA is regulated by proteolysis and phosphorylation (; ; ). A recent experiment has shown that CtrA degradation by ClpXP displays a low value which is comparable to its counterpart in , when the adaptor molecule SspB is present (). Therefore, CtrA can potentially be degraded with a zeroth-order kinetics inside and enhance the cell-cycle oscillation as well. GFP and GFP was cloned into pZE12-luc between I and I restriction sites through PCR cloning, with the degradation sequence flanked at the end of the reverse primer and transformed into DH5αZ1 (both pZE12-luc and DH5αZ1 are gifts of ). For measuring the degradation dynamics, pZE12-gfp was then transformed into BW25113, which contains a native copy of lacI, along with pTB114, a pCL1920-derived plasmid with a copy of lacI. Overnight cultures in M9 minimal media supplemented with 0.5% (w/v) glucose, 1 mM MgSO, 1 μg/ml vitamin B1, 100 μM CaCl, 100 μg/ml ampicillin, and 50 μg/ml of spectinomycin, were diluted into 10 ml of fresh media in a shake flask at initial OD 0.1 and grew for about 20 min before addition of 2 mM IPTG at 37°C. After 1–3 h of induction, 1 ml of cells was harvested, (OD∼0.2–0.8) washed once, and resuspended in M9 minimal media with 20 mM of acetate with or without chloramphenicol. During this period, no protein synthesis occurred and GFP degradation kinetics was measured. The cells were then transferred to an agar pad containing M9 minimal media with 20 mM acetate, supplements and antibiotics, and seal with a coverslip, as described in . Time-lapse microscopy was performed using a Nikon TE2000-S microscope with a × 60 DIC oil immersion objective. Images were captured using a Cascade:650 from Roper Scientific controlled through Metamorph software. The temperature of the samples was maintained at approximately 37°C by an objective heater. Brightfield (0.1 s) and epifluorescence (0.1 s) images were captured every 1–5 min, with both light sources shuttered between exposures. The nutrient downshift caused a growth lag for around 3 h (), thus remove the effects of cell division from proteolysis. In a typical experiment, 100–200 cells were monitored. The single-cell protein degradation kinetics was extracted by automated image analysis software. The time-series bright field images were used to generate a region mask for each cell. To create the mask, we first utilized the background flattening and shading correction function by the Metamorph software. The corrected images were then converted to binary images with a predefined threshold and segmented by custom-made software in MATLAB (the Mathworks Inc.). An area filter was applied to all segmented region to remove small areas caused by noise. An iterative algorithm was applied to track individual cells along time-series images and enumerate each tracked region. Finally, the numbered region masks were applied to the fluorescent images to obtain the mean fluorescent intensity of each cell at each time point. GFP expressed cultures for the bulk degradation kinetics experiments were prepared in the same way as in the experiments using time-lapsed microscopy. The cells were in the same media and initial cell density as the microscopy experiment and induced with 2 mM of IPTG until the OD is around 0.2–0.8. When the cells were ready for the experiment, the cells were washed once and either diluted or concentrated to OD∼0.3. A 200 μl portion of the diluted culture was transferred to 96-wells plates (Corning) with black walls and clear bottom. For each experiment, 4–8 wells were used. A 50 μl portion of silicon oil was added to each well to prevent evaporation (). The plate was then incubated in the microplate reader (Molecular Device, SpectraMax Gemini XS) at 37°C with high-intensity continuous shaking. Fluorescence measurements were taken every 3–5 min. The deterministic model of metabolator was simulated with Matlab and Mathematica. The phase diagram was generated with MatCont (). The stochastic modeling was performed using Matlab. The detail of the model can be found in .

By crosslinking actin filaments and interacting with transmembrane receptors and cytosolic signaling proteins, filamins play important roles in regulating the dynamics of the actin cytoskeleton and integrating cellular mechanics and signaling (). Vertebrate filamins are non-covalent dimers of 240–280 kDa subunits composed of an N-terminal actin-binding domain formed from two calponin homology domains followed by a rod region composed of 24 tandem immunoglobulin-like domains (IgFLN1–24) (; ; ). Dimerization is mediated via the C-terminal IgFLN24 (). Flexible hinges between IgFLN15 and 16 and IgFLN23 and 24 result in a V-shaped flexible actin crosslinker capable of stabilizing orthogonal networks with high-angle F-actin branching (). In addition to crosslinking F-actin, filamins act as scaffolds for a growing list of transmembrane receptors, signaling and adapter proteins (; ; ). In general, these interactions are mediated by the C-terminal domains, IgFLN 16–24, enabling filamin to complex multiple partners in close proximity to one another, potentially enhancing signal transduction (). Humans and mice each have three homologous filamin genes encoding the proteins filamin A, B and C; of these, filamin A (FLNa) is the most abundant and widely expressed (; ). In mice, FLNa expression is essential for proper cardiac and vascular development (; ), FLNb is required for skeletal and microvascular development () and FLNc is necessary for normal myogenesis (). In humans, heterozygous null FLNa alleles result in defective neuronal migration causing periventricular heterotopia (), while certain FLNa missense mutations cause familial cardiac valvular dystrophy () and putative gain-of-function mutations result in a spectrum of congenital malformations generally characterized by skeletal dysplasias (). Mutations in FLNb cause a class of diseases with abnormal vertebral segmentation, joint formation and skeletogenesis () and an FLNc mutation causes an autosomal dominant myofibrillar myopathy (). The diversity in phenotypes associated with different filamin mutations reveals that filamins perform a variety of essential functions and the current evidence suggests that in many cases specific disease phenotypes will result from disruption of specific interactions between IgFLN domains and their binding partners. Despite identification of more than 39 vertebrate filamin-binding proteins (; ), relatively little is known about how binding is regulated, how the IgFLN domains are arranged with respect to one another or how the arrangement of IgFLN domains modulates the ligand-binding activity of adjacent domains. We previously identified IgFLNa21 as the major binding site in FLNa for integrin adhesion receptors (). Integrins, αβ-heterodimers that span the plasma membrane, connect the extracellular environment to the actin cytoskeleton (). Thus, filamin–integrin complexes could provide a mechanical and biochemical link through which the dynamic actin cytoskeleton could respond to external cues. Indeed, modulation of integrin–filamin binding through both gain-of-function and loss-of-function mutations in integrin β-tails modulates cell migration () and alternative splicing of filamin genes, which results in deletions of portions of the rod domain, enhances integrin binding and affects myogenesis (; ; ). The structure of IgFLNa21 in complex with a β7 integrin peptide confirmed that IgFLNa21 is a β-sandwich composed of two β-sheets, similar to other human and IgFLNs (; ; ; ; ). The integrin β7 peptide binds to the CD face of the IgFLNa21 β-sandwich and this may represent a general mechanism for IgFLN domain–ligand interactions as other IgFLNs also bind their respective ligands at the CD face (). Integrin binding to IgFLNa21 can be inhibited by phosphorylation of the integrin tail or by other integrin tail binding proteins that compete with filamin (); however, whether filamin's ligand-binding activity is itself regulated remained unclear. To date, the only reported multi-domain structures of filamin are of two- and three-domain fragments of filamin (ddFLN), where the domains form an elongated zigzag chain (; ). ddFLN is different from vertebrate filamins in that it contains only six IgFLN domains and dimerizes in an end-on antiparallel fashion rather than the proposed parallel or v-shaped arrangement for human filamins. To determine how adjacent human filamin domains are oriented, we investigated an integrin-binding three-domain fragment of FLNa, IgFLNa19–21. X-ray crystallography shows that within this three-domain protein, IgFLNa21 and IgFLNa19 are very similar to one another and to other IgFLNs, whereas IgFLNa20 is partially unfolded and its first strand binds the integrin-binding CD face of IgFLNa21. NMR and biochemical analyses indicate that the IgFLNa20–21 domain pair inhibits integrin β-tail binding and mutations perturbing the IgFLNa20–21 interaction enhance integrin binding. Analysis of other domain pairs suggests that this may be a general feature of filamin–ligand interactions. To study the domain arrangement of the major integrin-binding site within filamin, we crystallized a fragment containing human FLNa domains 19–21 (IgFLNa19–21). Diffraction data to 2.5 Å resolution were used for the crystallographic calculations (). The asymmetric unit of the crystal contained two molecules; accordingly, two copies of partial poly-Ala models for IgFLNa21 and IgFLNa19 were initially positioned in the asymmetric unit by the molecular replacement program Phaser (). In the final model, both copies of IgFLNa19 and IgFLNa21 could be completely built, but loops BC and DG of IgFLNa20 in chain A and 56 residues in chain B could not be included because of missing electron density (). The final -factor values for the model remained moderate (=25.3%, =29.8%) apparently because of disorder in the crystals resulting in missing electron density and high -factor values, especially in IgFLNa20 (). Despite the disorder in IgFLNa20, its partial model could be validated by locating anomalous selenium signals in their appropriate positions of IgFLNa20 in chain A in crystals grown from SeMet-labeled protein (). As chain A is better resolved, and since non-crystallographic symmetry restraints were used in the refinement (), the structure of chain A was used in all further analyses. The domain arrangement of the IgFLNa19–21 fragment is unexpected and different from other immunoglobulin-like domain structures determined thus far (). The three domains form an elongated shape but the domain order along the long axis of the fragment is not sequential. Instead, IgFLNa19 is followed by IgFLNa21 and then IgFLNa20. The β-strands of IgFLNa19 and IgFLNa21 are arranged roughly along the long axis of the fragment, whereas the main part of IgFLNa20 is located across the loops of IgFLNa21 roughly perpendicular to the long axis. While the N-terminus of the fragment is at one end, the C-terminus is in the middle. This arrangement is only possible because IgFLNa20 does not have a complete immunoglobulin-like fold and interacts with IgFLNa21 in an unusual way. The first part of IgFLNa20 is separated from the rest of IgFLNa20 and, as discussed in more detail later, forms an additional β-strand next to the CFG β-sheet of IgFLNa21 ( and ). The remainder of IgFLNa20 lies on top of IgFLNa21, interacting mainly with the BC loop of IgFLNa21. As a consequence of β-strand A being separated, the half of the IgFLNa20 immunoglobulin sandwich that should include β-strands ABED, as seen in IgFLNa19 and IgFLNa21 (), is rather distorted in IgFLNa20 (). While IgFLNa20 has an unusual fold, IgFLNa19 and IgFLNa21 are very similar to one another () and to previously published IgFLN structures (; ; ; ; ; ). IgFLNa19 and IgFLNa21 can be superimposed with an r.m.s.d. of 1.50 Å for 89 Cα atoms, with the biggest differences observed in the BC and DE loops (). We previously reported the structure of IgFLNa21 bound to a peptide from the integrin β7 cytoplasmic tail (PDB code 2BRQ) (), and comparison of the two IgFLNa21 structures reveals good alignment of the β-strands (r.m.s.d., 0.60 Å for 40 Cα atoms) but BC and DE loops differ (overall r.m.s.d. 1.67 Å for 91 Cα atoms) (). These differences can be attributed to the contacts with IgFLNa20 (loop BC) in the three-domain structure and the presence of a covalently bound glutathione molecule in 2BRQ (loop DE). The crystallographic results showed an unexpected domain arrangement of IgFLNa19–21. To test if the whole arrangement of IgFLNa19–21 is stable without the various interactions provided by crystal contacts, we performed a molecular dynamics simulation of a single IgFLNa19–21. As domain–domain movements usually occur on a nanosecond scale (; ), the extension of the simulation to 10 ns should be long enough to detect at least the beginning of substantial domain movements. However, during the simulation, further domain–domain packing is observed that stabilize the IgFLNa19–21 structure (). When compared to the position of IgFLNa21, IgFLNa19 shows only slight movement, while the removal of crystal contacts releases IgFLNa20 to move closer to the head of IgFLNa21 (). Overall, the simulation results suggest that the domain arrangement of IgFLNa19–21 is rather stable as seen in the crystal. We have previously shown that integrin β-tails bind to the CD face of IgFLNs, with the integrin tail forming a β-strand that extends the β-sheet formed by the CFG strands (). In the current structure, the first strand of IgFLNa20 extends the IgFLNa21 CFG β-sheet in a manner analogous to the integrin peptide and completely covers the integrin-binding site (). Despite little primary sequence similarity between the IgFLNa21-binding portions of the integrin β7 tail and IgFLNa20 (), when bound to IgFLNa21 they adopt very similar structures () (r.m.s.d. 0.74 Å for nine Cα atoms) and bury comparable areas (727 A for IgFLNa20 and 667 A for β7) of accessible surface on the CD face of IgFLNa21. As was observed for the β7 integrin–IgFLNa21 complex, the first strand of IgFLNa20 forms hydrogen bonds to strand C of IgFLNa21 (). We have predicted that the specificity of this kind of IgFLN–ligand interaction is mainly determined by the hydrophobic interactions between the binding partner and the side chains of IgFLN β-strand D side chains (). The common hydrophopic interaction shared between integrin β7 tail and IgFLNa20 is Ile (IgFLNa residue 2144 and β7 residue 782) that is sandwiched between Leu2283 and Phe2285 of IgFLNa21 β-strand D (). Other interactions of IgFLNa20 seem to be less optimal than those of the integrin tail. In particular, the four Arg residues (2146–2149) of IgFLNa20 appear to form suboptimal interactions and are quite uncommon for a β-strand (). As described above, molecular dynamic simulations predict that the interaction between the first part of IgFLNa20 and the CD face of IgFLNa21 is stable, but to test this in solution NMR experiments were performed. The interaction between IgFLNa21 and IgFLNa20 was validated by solution-state NMR in two ways: (i) Addition of IgFLNa20 to N-labeled IgFLNa21 gave selective shifts concentrated on the CD face of IgFLNa21 (). This pattern is very similar to that observed with the β7 integrin () and confirms that the CD face of IgFLNa21 is involved in the binding of IgFLNa20. Furthermore, the shifts induced by a truncated IgFLNa20 protein (residues 2167–2235), which lacks the residues forming the first β-strand, were very much reduced (), demonstrating that this region of IgFLNa20 is required for binding IgFLNa21. (ii) Since there is only one Ala–Leu pair in the IgFLNa19–21 sequence (Ala2272 and Leu2271), selective labeling (1-C-Leu, N-Ala) of IgFLNa19–21 allowed Ala2272 in strand C to be uniquely identified in the complicated spectrum of the triple domain; this resonance was observed to be considerably shifted in IgFLNa19–21 at 37°C (δ=8.080 p.p.m., δ=124.918 p.p.m.) compared to that in IgFLNa21 alone (δ=8.921 p.p.m. and δ=124.970 p.p.m.). Taken together, these NMR experiments give strong support to the interaction of the N-terminus of IgFLNa20 with the CD face of IgFLNa21 as seen in the crystal structure. IgFLNa20 and integrin β-tails bind to the same site on IgFLNa21 (), suggesting that they may compete for binding to IgFLNa21. NMR analysis indicates that when free in solution integrin β7 tails bind IgFLNa21 with a higher affinity than IgFLNa20 does, and that β7 tails can displace free IgFLNa20 from IgFLNa21—evidenced by induction of a new pattern of shifts following addition of β7 tails (). However, in intact filamin, IgFLNa20 is tethered to IgFLNa21, thus increasing the effective local concentration and the occupancy of the interaction. To assess the impact of IgFLNa20 on integrin binding to IgFLNa21, we compared the binding of GST-IgFLNa21 and the two domain fragment GST-IgFLNa20–21. GST-IgFLNa21 bound to recombinant β7 integrin tails in a dose-dependent manner (). When quantified by scanning densitometry, curve fitting analysis indicated an apparent of 0.7±0.1 μM, in good agreement with our previous data (). Notably, binding of the two-domain fragment to β7 was much lower (). A reliable calculation of the binding affinity for this interaction was not possible because saturation of GST-IgFLNa20–21 binding to β7 could not be achieved. Structural analyses show that the first strand of IgFLNa20 is responsible for the interaction with IgFLNa21. We therefore generated a truncated IgFLNa20–21 protein (residues 2152–2329) lacking the first 13 amino acids of IgFLNa20, which normally form the interacting strand. This protein displayed a significant increase in β7 binding compared to wild-type IgFLNa20–21 (). To specifically disrupt the first strand interaction with IgFLNa21 without deleting a large stretch of amino acids, we substituted Ile2144 with Glu. Structurally Ile2144 corresponds to Ile782 in the integrin β7 tail (), which is important for β7 integrin binding to filamin (), and occupies a hydrophobic pocket on IgFLNa21 that is important for integrin binding (). We predicted that the introduction of a large charged residue at this site should destabilize the interaction, and observed that IgFLNa20–21 (I2144E), like the N-terminal truncation, displayed enhanced β7 integrin binding in comparison to wild-type IgFLNa20–21 (). Thus, disruption of the IgFLNa20–IgFLNa21 interaction can enhance integrin binding, presumably through exposure of the integrin binding CD face on IgFLNa21. The experiments described above were performed using short, bacterially expressed recombinant fragments of filamin. To verify the results in the context of filamin expressed in cultured cells, we compared the ability of integrin β-tails to pull down wild-type and mutated filamin from cell lysates. FLNa lacking IgFLNa20 (FLNaΔ20) exhibited enhanced binding to β7 integrin tails (), consistent with an inhibitory role for IgFLNa20. Similar results were obtained using both untagged and GFP-tagged FLNa (); this effect was not limited to β7 integrins, as removal of IgFLNa20 also enhanced binding to β1A tails (), consistent with the general ability of β-integrin tails to bind to the CD face of IgFLNa21 (). The binding of FLNa (I2144E), containing the point mutation that destabilizes the IgFLNa20–IgFLNa21 interaction, was also assessed (). GFP-FLNa (I2144E) displayed enhanced binding to β7 and β1A integrin tails in pull-down assays from cell lysates (). Thus in the context of full-length filamin, IgFLNa20 masks the major integrin-binding site in IgFLNa21 and a single point mutation is sufficient to expose the integrin-binding site and enhance integrin–filamin interactions. The observation that an intramolecular interaction between two adjacent IgFLNa domains reduces integrin binding to filamin suggests that this interaction may form part of a regulatory mechanism controlling filamin association with ligands. have shown that naturally occurring FLNa and FLNb splice variants (var-1) exhibit enhanced binding to a variety of integrin β-tails. These splice variants lack a 41-amino-acid sequence encompassing the C-terminal part of IgFLNa19 and the N-terminal part of IgFLNa20, including the first strand of IgFLNa20. We have confirmed the previously reported increase in β1A integrin-binding activity of FLNa var-1 proteins using EGFP-tagged FLNa19–21 var-1 expressed in cultured cells and shown that binding to β7 tails is also increased (). We have also shown increased binding of purified bacterially expressed GST-FLNa19–21 var-1 protein to both β7 and β1A integrin tails (). Alternative splicing may therefore be one mechanism by which the inhibitory intramolecular IgFLNa20–IgFLNa21 interaction is regulated to control filamin's ligand-binding activities. The preceding data indicate that IgFLNa20 negatively regulates ligand binding to IgFLNa21. Structural analysis of IgFLNa21 and IgFLNa19 reveals that both domains are very similar (), and mutagenesis and NMR analysis suggest that IgFLNa19 and IgFLNa21 both bind integrin β-tails in a very similar fashion (). We therefore tested whether IgFLNa18 could negatively regulate integrin binding to IgFLNa19. GST-IgFLNa19 bound β-integrins in a dose-dependent manner similar to IgFLNa21, but in comparison the two-domain construct GST-IgFLNa18–19 displayed severely reduced binding (). Thus, the auto-inhibition of ligand binding to IgFLN domains by the preceding even-numbered IgFLNa domains may be a more general phenomenon. h a v e d e s c r i b e d t h e m o l e c u l a r s t r u c t u r e o f a t h r e e - d o m a i n f r a g m e n t o f h u m a n F L N a . T h i s r e v e a l s a n u n e x p e c t e d a r r a n g e m e n t w i t h d o m a i n s i n a n o n - s e q u e n t i a l o r d e r c o n t r a r y t o p r e v i o u s p r o p o s e d m o d e l s , a n d t h i s a r r a n g e m e n t i s p o s s i b l e b e c a u s e I g F L N a 2 0 a d o p t s a n u n u s u a l s t r u c t u r e . T h e s t r u c t u r e a l s o r e v e a l s a m e c h a n i s m o f a u t o - i n h i b i t i o n , l i m i t i n g a c c e s s i b i l i t y t o t h e i n t e g r i n - b i n d i n g s i t e i n f i l a m i n . T h i s a u t o - i n h i b i t o r y m e c h a n i s m m a y b e e x t e n d e d t o o t h e r l i g a n d - b i n d i n g s i t e s i n f i l a m i n . F i n a l l y , l o s s o f a u t o - i n h i b i t i o n i n f i l a m i n s p l i c e v a r i a n t s p r o v i d e s a m o l e c u l a r e x p l a n a t i o n f o r t h e i r e n h a n c e d i n t e g r i n - b i n d i n g a c t i v i t y . Recombinant His-tagged integrin cytoplasmic tail model proteins were produced and purified as previously described (). IgFLNa19 (amino acids 2046–2141), IgFLNa21 (amino acids 2236–2329), IgFLNa20–21 (2142–2329), IgFLNa18–19 (amino acids 1955–2141), IgFLNa19–21 (amino acids 2046–2329) and IgFLNa19–21 var-1 (amino acids 2046–2125 then 2168–2329) were generated by polymerase chain reaction and subcloned into pGEX (Amersham) or EGFP (BD Biosciences) vectors for expression of GST or EGFP fusion proteins. Point mutations and deletions were introduced by QuikChange™ site-directed mutagenesis (Stratagene). All inserts were verified by DNA sequencing. GST fusion proteins were produced in BL21 cells and purified on Glutathione Sepharose 4 Fast Flow medium (Amersham Biosciences) according to the manufacturer's instructions. The IgFLNa19–21 fragment used for crystallization was cloned in modified pET24d vector containing a His tag followed by tobacco etch virus (TEV) cleavage site (). Protein was produced in BL21(DE3) cells and purified on Ni-NTA Agarose (Qiagen) according to the manufacturer's instructions. The His tag was cleaved with TEV protease (Invitrogen) during overnight dialysis at 4°C to 50 mM NaCl and 50 mM Tris–HCl pH 8.0. Additional purification was achieved by anion-exchange chromatography on Acell QMA matrix (Waters) and gel filtration on HiLoad 16/60 Superdex 75 column (Amersham Biosciences). The IgFLNa19–21 protein was crystallized using the hanging drop vapor diffusion method at 22°C by mixing 1 ml of 30 mg/ml protein solution in 100 mM NaCl, 50 mM Tris and 1 mM DTT pH 8.0 with an equal volume of 1.6 M (NH)SO, 0.1 M citric acid pH 6.1 and 10% dioxane. The crystals were transferred to 0.25 M KBr, 20% glycerol, 1.6 M (NH)SO and 0.1 M citric acid pH 6.1 before freezing under liquid nitrogen. The data for final structure solution were collected at 100 K at European Synchrotron Radiation Facility (Grenoble, France) beam line ID23-1 by using MarMosaic 225 CCD detector (Marresearch GmbH). The data were processed with the XDS program package (). Partial poly-Ala models for IgFLNa19 and IgFLNa21 were derived from IgFLNc24 structure (PDB code 1VO5) by replacing non-identical amino acids with Ala. Two copies of each of the models were initially positioned by molecular replacement program Phaser () and the final model was generated by iterating between manual model building with programs O () or Coot () and TLS+ restrained refinement with Refmac 5.2 (). Tight non-crystallographic symmetry restrains between IgFLNa19 and IgFLNa21 of chains A and B were used in the final refinement. Crystallographic images were generated with PYMOL (DeLano Scientific, San Carlos, CA, USA; ). Further details of the crystallographic data, structure validation and molecular dynamics are given in . Atomic coordinates and structure factors have been deposited in the Protein Data Bank, accession code 2J3S. IgFLNa20 (2141–2235) and IgFLNa20 var-1 (2167–2235) were expressed from a pGEX-6P-2 vector with a Precission protease cleavage site using BL21 codon plus cells (Stratagene) and uniform labeling was achieved by growing in M9 minimal media with N-NHCl and C-glucose. Selective labeling of IgFLNa19–21(1–C-L2271,N-A2272) was produced using a mixture of labeled (1-C-leucine and N-alanine) and non-labeled amino acids (0.1 g for each per liter M9) (). The identity and purity of the products were confirmed by mass spectrometry and SDS–PAGE. The β7-derived peptide PLYKSAITTTINP (N-terminally acetylated and C-terminally amidated) was purchased from EZBiolab (USA). All NMR samples were buffered with 50 mM sodium phosphate (pH 6.10) containing 100 mM NaCl and 5 mM DTT in 90% HO and 10% DO. NMR data were collected at H frequencies of 500, 600 and 750 MHz. The backbone amide N and H chemical shifts for wild-type IgFLNa21 were assigned using a 1 mM U-C,N-labeled protein sample and standard triple resonance experiments (). Gradient enhanced [H,N]-HSQC () experiments were used to carry out titrations for 100 μM IgFLNa21 with varying amounts of β7/IgFLNa20/IgFLNa20 var-1 at 25°C. The [HN,N] chemical shifts for (1-C-L2271,N-A2272)-IgFLNa19–21 were recorded with a 500 μM sample at 37°C on a Bruker cryoprobe-equipped Avance 500 MHz machine using a 3D HNCO pulse sequence (). NMR data processing was carried out with NMRPipe () and SPARKY (). Spectra were referenced to the water proton shift (4.766 p.p.m. at 25°C, 4.623 p.p.m. at 37°C) () with indirect referencing in the nitrogen dimension using a N-H frequency ratio of 0.101329118 (IUPAC). Binding assays using recombinant integrin tail model proteins were performed as previously described (; ). For GFP fusion proteins, or full-length human FLNa or FLNa mutants, Chinese hamster ovary (CHO) cells were transiently transfected with 3 μg of expression vector using Lipofectamine™ (Invitrogen), cells were harvested 24–48 h later, lysed as described previously () and binding assays were performed. Anti-filamin mAb1680 (Chemicon) and anti-GFP (Rockland) antibodies were purchased.

Adolescence is a time characterised by marked behavioural, hormonal and physical changes (Feldman and Elliott, ; Coleman and Hendry, ). Adolescents develop a capacity to hold in mind more multidimensional concepts and are thus able to think in a more strategic manner (Peterson, ). In addition to improvements in such ‘executive functions’ (Anderson ., ), during adolescence there seems to be a qualitative shift in the nature of social thinking such that adolescents are more self-aware and self-reflective (Elkind, ; Steinberg, ). In this study, we investigated the development of the neural circuitry underlying the ability to predict the actions that result from self-related intentions during adolescence. Recent structural MRI studies have demonstrated that the brain undergoes considerable development during adolescence. In particular, the prefrontal cortex (PFC) undergoes the most pronounced course of structural development, while development of superior temporal cortex, including the superior temporal sulcus (STS), is most protracted (Sowell ., ; Gogtay ., ; Toga ., ). These MRI studies demonstrate that in PFC, there is an increase in grey matter up to the onset of puberty and a subsequent rapid decrease in grey matter density from just after puberty and throughout adolescence, continuing into early adulthood. While grey matter development in the PFC follows a sharp inverted U-curve, grey matter in the superior temporal cortex/STS steadily declines during adolescence and well into adulthood, reaching maturity relatively late (Gogtay ., ; Toga ., ). At the same time, there is an increase in cortical white matter density from puberty, throughout adolescence and into adulthood (Giedd ., ; ; Reiss ., ; Sowell ., ; Barnea-Goraly ., ). Results of earlier post-mortem investigations of human brain development suggest that the cortical changes detected using MRI, especially in PFC, mainly reflect two cellular processes occurring during adolescence: (i) synaptogenesis, which is followed by synaptic pruning, and (ii) axonal myelination (Yakovlev and Lecours, ; Huttenlocher, ; Huttenlocher ., ). It has been hypothesised that these maturational processes fine-tune neural circuitry in the PFC and other cortical regions, and thus increase efficiency of the cognitive systems they subserve (see Blakemore and Choudhury, for review) Based on the finding that PFC and superior temporal cortex/STS undergo structural development during adolescence, it was hypothesised that the functioning within these regions would also show developmental change during this time period. Many high-level cognitive abilities rely on these brain regions, including mentalising (or Theory of Mind; Frith and Frith, ). Mentalising refers to the inferences that we naturally make about other people's intentions, beliefs and desires, which we then use to predict their behaviour. It includes the understanding that intentions relate to actions. A number of neuroimaging studies, using a wide range of tasks, have reported activation in a highly circumscribed ‘mentalising network’, comprising the medial PFC, the STS and temporo-parietal junction (TPJ), and the temporal poles adjacent to the amygdala (Fletcher ., ; Brunet ., ; Castelli ., ; Gallagher ., ; Vogeley ., ). Lesion studies have also implicated the frontal cortex (Stone ., ; Channon and Crawford, ; Happé ., ; Rowe ., ; Stuss ., ; Gregory ., ; though see Bird ., ) and STS/TPJ (Samson ., ; Apperly ., ) in mentalising. Signs of social competence develop during early infancy, such that by around 12 months of age, infants can ascribe agency to a system or entity (Spelke ., ; Johnson, ). The understanding of intention emerges at around 18 months, when infants acquire joint attention skills, for example, follow an adult's gaze towards a goal (Carpenter ., ). These early social abilities precede more explicit mentalising, such as false belief understanding, which usually emerges by about 5 years of age (Barresi and Moore, ). While normally developing children begin to pass theory of mind tasks by about 5 years, the brain structures that underlie mentalising undergo substantial development beyond early childhood. We hypothesised that the functioning of brain areas involved in mentalising tasks may change during adolescence. Functional imaging of the adolescent brain provides an opportunity to investigate this development. Previous neuroimaging studies that have investigated functional brain development during adolescence have focussed mainly on executive function tasks. Some have shown that activation of frontal cortex increases with age (e.g. Rubia ., ; Adleman ., ; Kwon ., ), while others report decreased frontal activation with age (e.g. Gaillard ., ; Blakemore and Choudhury, ; Durston ., ). The same discrepancy in findings with respect to frontal activity applies to studies that have investigated the development of social cognitive processes during adolescence. However, there is some indication that, for social cognitive tasks, activity in the frontal cortex increases between childhood and adolescence, and then decreases between adolescence and adulthood. For example, female subjects (but not male subjects) showed increased activation in dorsolateral PFC in response to fearful faces between childhood and adolescence (Killgore ., ). A recent study reported increased activity in PFC (bilaterally for girls; right sided for boys) in response to fearful faces between age 8 and 15 years (Yurgelun-Todd and Killgore, ). In contrast, another study of face processing found that attention to a non-emotional aspect of fearful relative to neutral faces was associated with increased activity in orbitofrontal cortex in adolescents compared with adults (Monk ., ). A recent fMRI study investigated the development of communicative intent using an irony comprehension task and found that children (aged between 9 and 14 years) engaged frontal regions (medial PFC and left inferior frontal gyrus) more than did adults in this task (Wang ., ). In summary, functional imaging studies have reported mixed findings with respect to changes in frontal activity with age, but there is a hint that activity during certain social cognition tasks might increase during childhood and decrease between adolescence and adulthood. Here, we employed a mixed factorial design with the factors (i) (intentional causality physical causality) and (ii) (adults adolescents). In the Intentional Causality condition, the subject's task was to answer blocks of questions posing scenarios that involved their own intentions and consequential actions. The Physical Causality condition involved answering questions about the causal link between physical events and their consequences. In a previous study looking at adults only, we found that the Intentional Causality task, relative to the Physical Causality task, activates regions associated with mentalising (medial PFC, STS and temporal poles) and self-reflection (medial PFC and posterior cingulate/precuneus) (den Ouden ., ). The objective of this study was to investigate whether the adult brain and the adolescent brain process this intentional causality task differently. A total of 19 right-handed, female adolescents (mean age: 14.79; age range: 12.12–18.06 years) and 11 right-handed female adults (mean age: 28.43; age range 22.40–37.76 years) with no history of psychiatric or neurological disorder took part in the study. To ensure a consistent level of verbal intelligence across all participants, the British Picture Vocabulary Scale (BPVS; Dunn ., ) was administered individually to each participant. Furthermore, adult subjects were university students or graduates, and the adolescent subjects were from a selective private school in London at which the vast majority (about 95%) of pupils go on to do undergraduate degrees and higher. The school teachers confirmed that each adolescent subject performed above average on national SAT tests. Written informed consent was obtained prior to the study from all participants, and from a parent or guardian of participants aged 16 and under. The study was approved by the UCL National Hospital for Neurology and Neurosurgery Ethics Committee. The experiment was split into two 11 min sessions in which subjects were presented with a series of descriptions of a scenario followed by a question relating to this scenario. Each block consisted of three scenario/question trials. In half the blocks, scenarios pertained to intentions and consequential actions []: e.g. : ‘You are at the cinema and have trouble seeing the screen’; followed by : ‘Do you move to another seat? Likely or Unlikely?’ In the other blocks, the scenarios pertained to natural occurrences and consequential events []: . : ‘A huge tree suddenly comes crashing down in a forest’; followed by : ‘Does it make a loud noise? Likely or Unlikely?’ In each block, the scenario stimulus was presented for 4.7 s, and was immediately followed by the question stimulus. The question stimulus was presented for 4.7 s, during which time subjects were asked to respond by pressing one of two buttons on a keypad corresponding to ‘likely’ and ‘unlikely’. The scenarios and questions were matched between the two conditions in terms of number of characters, words and clauses. In addition to the two conditions described earlier (IC and PC), a baseline condition was included in which subjects were asked to fixate on a black cross on a white background for a duration of 30 s. There were eight repetitions of each of the three conditions. Block order was counterbalanced within and between subjects. Each subject was trained on the task for approximately 4 min prior to scanning. Stimulus presentation was programmed in Cogent () running in Matlab 6.5, which recorded subject responses. A 1.5 T Siemens Sonata MRI scanner was used to acquire both 3-D T-weighted fast-field echo structural images and multi-slice T*-weighted echo-planar volumes with blood oxygenation level dependent (BOLD) contrast (TR = 3.6 s). For each subject, functional data were acquired in two scanning sessions of approximately 11 min each in which 195 volumes were acquired. The first five volumes were discarded to allow for T1 equilibrium effects. Each functional brain volume was composed of 40 2-mm axial slices with a 1-mm gap, and in-plane resolution of 3 × 3 × 2 mm positioned to cover the whole brain. The acquisition of a T1-weighted anatomical image occurred after the two sessions for each participant. The total duration of the experiment was approximately 35 min per subject. #text For imaging data analysis statistical parametric mapping was used, implemented in SPM2 []. For each subject, a set of 380 fMRI scans was realigned to correct for interscan movement and stereotactically normalised using sinc interpolation (Friston ., ), with a resolution of 3 × 3 × 3 mm, into the standard space defined by the Montreal Neurological Institute (MNI) template (Evans ., ) smoothed with a Gaussian kernel of 6 mm full-width half maximum to account for residual inter-subject differences. The analysis of the functional imaging data entailed the creation of statistical parametric maps representing a statistical assessment of hypothesised condition-specific effects (Friston ., ). The scans corresponding to the instruction phase of each block were excluded from the analysis. Condition-specific effects were estimated with the General Linear Model with a delayed boxcar wave-form for each condition. Low-frequency sine and cosine waves modelled and removed subject-specific low-frequency drifts in signal, and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The resulting parameter estimates for each regressor at each voxel were then entered into a second level analysis where ‘subject’ served as a random effect in a within-subjects ANOVA. The main effects and interactions between conditions were then specified by appropriately weighted linear contrasts and determined using the -statistic on a voxel-by-voxel basis. Statistical analysis at the second level was performed for each group separately to examine the main effect of the two experimental conditions compared with the baseline condition, and the main effect of intentional causality [(IC)–(PC)]. Since we had no predictions about differential activation in PC, no PC–IC contrasts were computed. To compare the two age groups directly, we investigated the interaction between group (adult adolescent) and causality task using the contrasts [(adultIC–adultPC)–(adolescentIC–adolescentPC)] and [(adolescentIC–adolescentPC)–(adultIC–adultPC)]. In addition, the effect of age on the neural processing of IC–PC was investigated using a regression function at the second level with age as the covariate of interest. Statistical contrasts were used to create an SPM { t } , which was transformed into an SPM { Z } and thresholded at < 0.05 (corrected on the basis of the theory of random Gaussian fields for multiple comparisons across the whole brain volume examined). We report regions that survive whole brain correction (or, where we had an hypothesis for their activation, small volume correction) at < 0.05. Every subject made a response to every causality question. For RTs, a between-subjects repeated measures ANOVA revealed that subjects from both groups were significantly faster to answer IC questions than PC questions (F (1, 28) = 89.29, < 0.0001; ). There was no overall significant difference between the two groups (F (1, 28) = 1.92; > 0.05), nor was there a significant interaction between group and condition (F (1, 28) = 0.52; > 0.05). There was no significant correlation between age and RTs for either condition ( > 0.05). Response types (i.e. whether subjects chose likely or unlikely in their response to each scenario) were compared by calculating the percentage of ‘likely’ responses in both conditions for both groups. For the IC condition, 47% of the adults’ responses and 49% of the adolescents’ responses were ‘likely’. For the PC condition, 38% of the adults’ responses and 35% of the adolescents’ responses were ‘likely’. A between-subjects repeated measures ANOVA revealed that the percentage of ‘likely’ responses was greater in the IC condition than in the PC condition for both groups (F (1, 28) = 33.24; < 0.0001). There was no overall significant difference between the two groups (F (1, 28) = 0.39; > 0.05), nor was there a significant interaction between group and condition (F (1, 28) = 2.01; > 0.05). xref fig #text italic xref fig table-wrap #text There was a significant interaction between group and condition in the medial PFC. The right medial PFC (MNI coordinates: 12 42 21) was activated significantly more in adolescents than in adults during IC compared to PC ( = 3.52; < 0.05 SVC) ( upper left panel). Investigation of the parameter estimates for each condition relative to baseline for both groups revealed that the interaction was driven by increased activity in medial PFC in the IC condition in adolescents relative to adults ( upper right panel). Note that parameter estimates are negative because activity in the fixation baseline condition has been subtracted from activity in each experimental condition. There was also a significant interaction between group and condition in the right STS (63 −33 −9; Z = 3.35; < 0.05 SVC; upper left panel). However, for the STS the interaction was driven by higher activity in IC relative to PC in adults, compared with adolescents ( upper right panel). No brain regions showed a significant interaction between age group and causality in the direction of PC relative to IC. The regression analysis showed that activity in the medial PFC (15 45 18) showed a significant negative correlation with age ( = −0.45; < 0.05; lower panel). The correlation between the right STS (63 −33 −9) and age was not quite significant ( = 0.34; = 0.062). The aim of the present study was to investigate adolescent development of the neural network involved in thinking about intentions. Previous studies have suggested a role for a network of areas (medial PFC, STS, TPJ and temporal poles) in this type of mentalising task (Brunet ., ; Gallagher ., ; Vogeley ., ). In the current study, subjects responded to scenarios relating either to their own intentions and consequential actions (intentional causality) or to physical events and their consequences (physical causality). We investigated how activity during these tasks in the adult brain compares with activity in the adolescent brain. Our results showed that both groups recruit the mentalising network (medial PFC, STS/TPJ and temporal poles) during IC relative to PC. However, adolescents activated the medial PFC part of this network to a significantly greater extent than did adults during IC relative to PC. In contrast, the right STS was activated more by IC than PC for adults only. Our results suggest that activity shifts from anterior to posterior regions of the mentalising network during adolescence. Both groups of subjects took significantly less time to respond to the IC questions than to the PC questions (). The difference in reaction times could not have been due to any difference in the structural features of the stimuli since these were matched. Instead, this effect may be due to an inherent difference in cognitive processing demands for the two types of question. These results are in line with previous findings demonstrating a tendency for normally developing children to show better performance on questions about intentions than questions about the physical world (Baron-Cohen ., ). One possibility is that, perhaps because we have ‘direct’ information about intentions, the understanding of intentions is more intuitive, and requires less explicit reasoning and ‘working out’ than does the understanding of physical causality. There were no significant RT differences between the two groups. We also analysed the response types made by each group. For both conditions, a scenario was followed by a question about how likely or unlikely a particular consequence to the scenario is (see ‘Methods’). Both groups gave more ‘likely’ responses to the suggested consequences to the IC scenarios than to the suggested consequences to the PC scenarios. The reason for this is not clear, but it is possible that subjects view the suggested consequences to the IC scenarios as more likely because people are relatively flexible in their response to events. In contrast, events in the PC condition were limited by physical and natural laws. There were no significant response choice differences between the two groups. r a i m w a s t o i n v e s t i g a t e h o w t h e n e u r a l s y s t e m a s s o c i a t e d w i t h i n t e n t i o n u n d e r s t a n d i n g c h a n g e s f r o m e a r l y a d o l e s c e n c e t h r o u g h t o a d u l t h o o d . T h e r e s u l t s o f o u r f M R I s t u d y r e v e a l e d t h a t , w h e n t h i n k i n g a b o u t i n t e n t i o n a l c a u s a l i t y ( r e l a t i v e t o p h y s i c a l c a u s a l i t y ) , a d o l e s c e n t s r e c r u i t m e d i a l P F C t o a g r e a t e r e x t e n t t h a n d o a d u l t s , a n d a d u l t s u s e p a r t o f t h e r i g h t S T S m o r e t h a n d o a d o l e s c e n t s . T h i s s u g g e s t s t h a t t h e n e u r a l s t r a t e g y f o r t h i n k i n g a b o u t i n t e n t i o n s c o n t i n u e s t o d e v e l o p d u r i n g a d o l e s c e n c e a n d e a r l y a d u l t h o o d . T o o u r k n o w l e d g e , t h i s i s t h e f i r s t i m a g i n g s t u d y t o p r o v i d e e v i d e n c e t h a t p a r t s o f t h e m e n t a l i s i n g n e t w o r k c o n t i n u e s t o d e v e l o p a f t e r e a r l y c h i l d h o o d . W h i l e n o r m a l l y d e v e l o p i n g c h i l d r e n p a s s t h e o r y o f m i n d t a s k s b y a b o u t a g e 5 y e a r s , o u r d a t a s u g g e s t t h a t t h e m e n t a l i s i n g n e t w o r k c o n t i n u e s t o b e c o m e r e f i n e d d u r i n g a d o l e s c e n c e .

The acclaimed goal of systems biology is quantitative understanding of functional interactions between the multiple cellular components to eventually predict network, cell and organism behavior (; ). Beyond intuition, quantitative understanding inevitably requires computational models to capture the enormous numbers of molecular components that interact in a highly nonlinear manner within interlinked information and biochemical networks (; ; ). For most cellular networks, such as signaling or protein–protein interaction networks, however, we do not even know all involved components that need to be represented in a model. Hence, much of the current focus is on experimental (; ) and computational (; ; ; ) identification of missing components and their interactions to establish the network topology as a prerequisite for mechanistic modeling. Metabolic networks are a notable exception because their interaction topology is well established in several cases; that is, we know most reactions, the enzymes that catalyze them, the genes that encode the enzymes and how they interact stoichiometrically within a biochemical network. As incomplete as this knowledge may be, it is currently far beyond that of basically any other cellular network and allows to construct metabolic models that represent almost entire microbial genomes (). With up to 1000 biochemical reactions, these genome-scale models allow to predict network capabilities, for example, by using flux balance analysis (FBA) (). Successful FBA applications include prediction of gene deletion lethality (; ; ), end points of adaptive evolution () and optimal metabolic states (). In contrast to dynamic models that require detailed, typically unavailable kinetic parameters, constraint-based modeling with FBA permits steady-state analysis of large-scale networks without large fitted parameter sets (; ; ). To identify optimal solutions in the vast solution space, FBA objective functions are defined to solve the system of linear equations that represent the mass balance constraints. While different objectives were proposed for different biological systems (; ; ; ; ; ), by far the most common assumption is that microbial cells maximize their growth (see below for further explanation). Since the identified optimal solutions are often inconsistent with the biological reality, the solution space is further restricted through additional constraints that reflect thermodynamic, kinetic or biochemical knowledge. Another problem is that, depending on the shape of the solution space, multiple intracellular flux distributions (alternate optima) may underlie the exact same optimal value that is identified by the objective (e.g., the maximum biomass yield) (; ). This space of steady-state flux solutions has been explored for biological meaning (; ; ; ) and to identify candidate network states (; ), but was largely ignored in many FBA studies that examined arbitrary, single optimal solutions (; ; ). In a fully complementary approach, C-experiments are used to determine intracellular flux states that reveal operation of metabolic networks (; ; ; ). In some instances, such experimental flux data were used to further restrict the FBA-computed flux solution space. For lack of experimental data, however, only one or two arbitrary flux distributions were considered (; ). Attempts to actually predict intracellular fluxes by FBA methods are few and either unverified () or tested for a single case (; ; ). With the recent availability of large-scale experimental flux data from various microbes (; ; ; ; ), a more systematic analysis of the correlation between the feasible flux space and the realized fluxes is now possible. Here we examine the predictive capacity of 11 linear and nonlinear network objectives, by evaluating the accuracy of FBA-based flux predictions through rigorous comparison to C-based flux data from grown under six environmental conditions. By systematically testing all permutations of 11 objective functions with or without eight additional constraints, we identify the most appropriate combination(s) to predict fluxes by FBA. More generally, we thus assess whether assumed optimality principles of evolved network operation are generally valid or whether specific objectives are necessary for environmental conditions that require different metabolic activity. To predict intracellular fluxes through the presently known reactions of central carbon metabolism, we constructed a highly interconnected stoichiometric network model with 98 reactions and 60 metabolites that supports the major carbon flows through the cell ( and ). FBA-based fluxes are typically expressed as relative fluxes that are normalized to the specific glucose uptake rate. Typically, this reference flux is known, hence absolute fluxes can be calculated by re-scaling. Due to linear dependencies in the network, the systemic degree of freedom is restricted to a limited number of reactions that define the entire flux solution. For our network, 10 reactions are sufficient to describe the actual systemic degree of freedom, as identified by calculability analysis (; ). These fluxes were expressed as split ratios at pivotal branch points in the network, where each of the 10 reactions that consume a cellular metabolite is divided by the sum of all producing reactions ( and ). Qualitatively identical results were obtained when repeating all reported simulations directly with the 10 absolute fluxes instead of the 10 split ratios (data not shown). Dividing a specific consumption flux by all producing fluxes scales to unity, an unbiased comparison of the 10 network fluxes with often-different magnitudes is possible. Moreover, it enhances intuition and biological interpretation because, wherever possible, the ratios were chosen to represent metabolic flux ratios that are obtained from C-experiments () ( and ). For example, split ratio R1 represents the fraction of the intracellular glucose-6-phosphate (G6P) pool that is metabolized through phosphoglucoisomerase (Pgi), relative to the summed production via G6P-dehydrogenase (Zwf), glucokinase (Glk) and the phosphotransferase system (Pts) ( and ). The experimentally determined split ratios () can be subdivided into three groups: (i) R1, R4, R5, R6 are always active, (ii) R3 is inactive under all considered conditions and (iii) R2, R7–R10 are conditionally active. Optimal solutions in this underdetermined system of linear equations were identified by FBA with 11 linear and nonlinear objective functions to identify optimal solutions, some of which are combinations of pairs of objectives (). Depending on the shape of the solution space, linear optimization frequently leads to alternate optima; that is, alternate sets of feasible flux distributions with an identical optimal value (; ). To quantify the overall variance of fluxes, we first determined the absolute range of variation for the individual split ratios by maximizing and minimizing each flux separately. For example, maximization of biomass yield (which is synonymous to the frequently used term of maximization of growth rate ()) results in ranges of the split ratios R1, R4, R6 and R7, but unique values for the remaining six split ratios (). Maximization of ATP yield without further constraints, in contrast, is a much better defined example with unique values for all 10 split ratios (). Beyond objective functions, these flux variabilities are not only dependent on the chosen objective but also the network structure, and were also shown to exist in genome-scale models (; ). To further constrain the solution space, we imposed eight additional constraints on network operation (). The choice of objective functions and constraints widely predefine the degree of freedom in terms of specific pathway usage (data not shown), hence appropriate objective/constraint combinations can potentially be used to approximate metabolic behavior. We systematically assessed the predictive capability of FBA by comparing all objective/constraint permutations to C-detected flux distributions from six growth conditions, including glucose- and ammonium-limited chemostat cultures and batch cultures with excess nutrient supply ( and ). For each of the 99 different optimization problems, the maximum and minimum Euclidean distance between and flux solutions was evaluated by simultaneously considering the 10 split ratios. Confidence intervals of the experimental flux ratios were considered by the standardized Euclidean distance, which weights the distance between prediction and data by the square of the corresponding standard deviation δ. The resulting value describes the overall deviation of the predicted flux distribution (or range of flux distributions) with respect to the corresponding experimental reference solution, and is henceforth referred to as predictive fidelity (). This predictive fidelity depends on two factors: (i) the minimal possible standardized Euclidean distance to the results and (ii) the potential variance of the fluxes that arises from alternate FBA optima. First, we determined the predictive fidelity of the 99 objective/constraint combinations for unlimited batch growth on glucose under aerobic, anaerobic and nitrate-respiring conditions (; ). Obviously, the agreement is specific for each case, since it depends (i) on the particular objective/constraint combination that defines the shape of the solution space and (ii) the experimental reference flux distribution (, and ). Since minimization of glucose consumption and maximization of ATP yield per reaction step gave almost identical results as maximization of biomass yield and maximization of ATP yield, respectively, only the latter two are discussed (). The results obtained by minimization of reaction steps, minimization of the redox potential and minimization of ATP producing fluxes were considerably worse than those obtained with the other objectives, hence are not discussed in the following (). Without requiring additional constraints, the highest predictive fidelity for aerobic batch cultures was obtained by maximizing ATP yield per flux unit, yielding a unique flux prediction that is closest to the experimental data (). Since this nonlinear optimization function is non-convex, it bears the danger of identifying only local optima. To confirm that indeed a global optimum was identified, we first reformulated the original objective function as a convex function that contains a linear and a nonlinear, but convex term (see Materials and methods for details). This new nonlinear but convex function cannot be optimized , since it needs weighting of both function terms. Thus, in a second step, we performed a sensitivity analysis around the previously identified optimal solutions. Since no iteration resulted in a solution with a higher optimal value, we have strong indication that indeed global optima were identified (). Of the remaining objective functions, only the maximization of the ATP yield objective achieved similar predictive fidelities when combined with particular constraints. Maximization of biomass yield, in contrast, suffered from alternate optima over a wide range of constraints (). Although unique results are feasible by invoking a P-to-O-ratio (moles of ATP produced per mole of oxygen) of unity, the predictive fidelity is still inferior to the one obtained with the maximization of ATP objectives. The predictive fidelity is a general criterion for the predictive accuracy. It cannot, however, identify the individual metabolic functions that are responsible for the deviations. To elucidate whether these are based on large errors in single ratios or on small errors in many ratios, we plotted and ratios as scatter plots where perfect predictions fall on the bisecting diagonal (). For aerobic batch cultures, acetate secretion (R9) in combination with a sound predictive fidelity is one main discriminating variable that was only captured by the maximization of ATP yield per flux unit (). In combination with oxygen constraints, the maximization of ATP yield mimics the maximization of ATP yield per flux unit objective (, ). Minimizing the overall intracellular fluxes inherently leads to acetate secretion, however, at the cost of deviations in other ratios, in particular for Entner–Doudoroff activity (R2) (data not shown). Akin to aerobic cultures, the maximization of ATP yield per flux unit was the only objective that achieved reasonable predictions without further constraints for anaerobic nitrate-respiring batch cultures ( and ). Improved predictions are possible, however, by setting a particular constraint on the nitrate respiration rate for the maximization of biomass yield objective ( and ). Largely independent on the invoked constraints the anaerobic batch culture was well predicted by all considered objective functions ( and ). This behavior is due to the low degree of freedom in the absence of an external electron acceptor and to the experimental uncertainty of some of the fluxes (see ), which allow for a relatively high predictive fidelity even if the actual agreement is low. In contrast to unrestricted nutrient supply in batch cultures, a single, defined nutrient limits the rate of growth in continuous chemostat cultures (; ). To evaluate predictions for the rather different metabolic behavior under such conditions, we used experimental flux data from slowly (0.1 h) and rapidly (0.4 h) growing chemostat cultures under glucose- () and ammonium limitation () ( and ). Largely independent of additional constraints, the maximization of ATP or biomass yield objectives approximated all chemostat cultures best ( and ). Maximization of ATP producing fluxes resulted in similar predictive fidelities as maximization of ATP yield (). Although mathematically distinct, both objectives maximize ATP production and thus lead to similar flux predictions. Alternate optima occurred only in one case for the biomass objective and can be overcome, as for aerobic batch cultures, by constraining the P-to-O-ratio to unity. The relative independence of the predictive fidelity on constraints demonstrates that these objectives provide somewhat robust predictions for metabolism under nutrient limitation. Nevertheless, various specific objective/constraint combinations were also capable of describing the different conditions, in particular the maximization of ATP yield per flux unit, in combination with all eight constraints for both carbon-limited chemostat cultures ( and ). In sharp contrast to batch cultures, however, the maximization of ATP yield per flux unit objective was basically useless without further constraints (). As a main discriminating variable of good objectives for aerobic batch cultures, the well-known phenomenon of acetate overflow was only captured when maximizing the ATP yield per flux unit (). Maximizing the overall ATP yield mimicked this behavior only when combined with oxygen uptake constraints. Since particular combinations of oxygen uptake and P-to-O ratio constraints and the frequently used maximization of biomass yield objective should achieve the same effect (; ), we performed a sensitivity analysis by determining the predictive fidelity and acetate production for step-wise increases of the oxygen uptake constraint for four P-to-O ratios ( and and ). As for maximization of the overall ATP yield, only fine-tuning of the network by invoking particular combinations of P-to-O ratio and oxygen constraints resulted in reasonable flux predictions and acetate secretion rates for the maximization of biomass yield objective function. However, the predictive fidelity was very sensitive to changes in the parameters, such that only a narrow range of oxygen uptake constraints enforced a good fit for a given P-to-O ratio, for example unrealistically low oxygen uptake rates of 5–7 mmol/g h at a P-to-O ratio of 2 ( and ). For maximization of the ATP yield objective, in contrast, the predictive fidelity was relatively insensitive to the exact value of the oxygen uptake and the P-to-O ratio constraint, with a critical threshold for the oxygen uptake constraint of around 15 mmol/g h; that is, the upper bound of experimentally observed values in glucose batch cultures (; ; ). Since such metabolic parameters often vary between strains or with small environmental differences, a certain robustness of predicted flux solutions with respect to the chosen constraints is a highly desirable property for constraint-based modeling. Hence, both ATP objectives are clearly of superior robustness for the prediction of fluxes in aerobic batch cultures. To systematically analyze the predictive capability for individual flux ratios, we defined the specific agreement ρ (). This value considers the deviation between each single pair of and values for all 10 split ratios and is weighted by the ratio of the corresponding accuracies, that is the possible range of the computational split ratio divided by the experimental standard deviation. To obtain an overview of the most difficult to predict fluxes, independent of the specific objective and constraints, we applied cluster analysis to the Euclidean distance among specific agreements ρ ( and and ). The 2700 considered data points represent the 10 split ratios for all objective/constraint combinations considered in and under each environmental condition. Most difficult to predict in four out of the six conditions were the fluxes into glycolysis (R1) and acetate secretion (R9), although sometimes in different combinations (). Exceptions were the C-limited chemostats, were fluxes through the glyoxylate shunt (R7) and the TCA cycle (R6) were most difficult to predict (). Thus, some fluxes are clearly more difficult to predict than most others, but those problematic ones often change with the environmental conditions. The key question addressed here is whether intracellular fluxes in metabolic steady state can be predicted from network stoichiometry alone by invoking optimality principles. Our systematic and statistically rigorous comparison of FBA-based flux predictions from 99 objective/constraint combinations to fluxes from C-experiments demonstrated that prediction of relative flux distributions is, within limits, possible. Since no single objective predicted the experimental data for wild-type under all conditions, the pivotal element is to identify the most relevant objective for each condition. For unlimited growth on glucose in oxygen or nitrate respiring batch cultures, by far the best objective function was nonlinear maximization of the ATP yield per unit of flux, which is a combination of the linear maximization of overall ATP yield and minimization of the overall flux. In some cases, similar predictions could be achieved by combining the overall maximization of ATP and biomass yield objectives with particular oxygen uptake constraints. Under nutrient scarcity in nutrient-limited continuous cultures, in contrast, the linear maximization of ATP or biomass yield were clearly superior. As a result of the low degree of freedom in non-respiring batch cultures, all objective functions lead to equally well flux predictions. As an unexpected key result, model preconditioning through additional and potentially artificial constraints is not necessary if the appropriate objective function is chosen for a given condition. Invoking additional constraints for suitable objectives achieved only subtle improvements or avoided alternative optima in few cases. When combined with particular constraints, even suboptimal objectives could be forced to yield equally accurate predictions for some conditions; in the sole case of nitrate respiration even better predictions. Setting of these additional constraints, however, is condition- and objective specific, thus requires considerable knowledge to be biologically meaningful. Except for subtle differences in predictive fidelity, all major conclusions are independent of using normalized ratios instead of absolute flux values (data not shown). We explicitly considered the fundamental FBA problems of alternate optima and experimental accuracy by size reduction through calculability analysis and by including confidence regions for fluxes in the standardized Euclidean distance, respectively. An important question is whether or not our results are model dependent. To address this point, we verified the key conclusions with two genome-scale models of metabolism (; ). Although specific properties such as flux variability clearly depend on the network structure of the particular stoichiometric network model (see below for details), the above-identified objectives also achieved the best predictions for fluxes in the central carbon metabolism with either genome-scale model (data not shown). Clearly, alternate optima occurred also in genome-scale network models (see Materials and methods for details). Independent of the model size, however, variability can be avoided such that uniquely defined solutions are obtained when low P-to-O ratios are assumed or when all internal proton fluxes are balanced. With the proton-balanced genome-scale model of , for example, such unique flux solutions can be obtained. The solution space spanning all alternate optima has previously been scanned for biological meaning (; ; ), and identified interesting correlations with levels of gene expression that point to evolutionary constraints on how tight certain reactions need to be regulated (). A potential complication is that the solution space itself is not, or not entirely, an inherent network feature, but also a function of the arbitrarily chosen objectives and constraints. An important distinction must be made between FBA-based prediction of the typically investigated general physiology (i.e., extracellular uptake and production rates and the growth rate) and the here attempted prediction of the underlying intracellular flux distribution, which has several-fold more variables. Hence, there is no immediate contradiction between good prediction of growth physiology obtained by maximizing the growth yield () or minimizing the redox production rate (), and their here demonstrated limited capacity to predict the underlying flux state. In some cases, alternate flux optima are responsible for the apparent discrepancy and in others it is primarily the specific combination of chosen constraints that explain why good physiology predictions were achieved with suboptimal objectives. By demonstrating that intracellular fluxes can be approximated with experimentally validated optimality assumptions, we go beyond flux prediction algorithms such as minimization of metabolic adjustment () or regulatory on/off minimization () that require a reference flux distribution in the wild type to predict fluxes in mutants. Carefully chosen objectives achieve intrinsically good prediction not only of growth physiology, but also of intracellular fluxes in wild-type without preconditioning the system through additional constraints apart from the experimentally determined growth rate. Since the network model was identical in all cases, the identified optimality functions potentially reflect the evolved regulatory processes that realize the particular flux states under different environmental conditions. Under nutrient scarcity in chemostat cultures, metabolism normally supports efficient biomass formation with respect to the limiting nutrient (). Based on our results, this operational state appears to have evolved under the objective to maximize either the ATP or biomass yield (synonymous to the frequently used maximization of growth rate objective). Under conditions of unlimited growth in aerobic or nitrate-respiring batch cultures, in contrast, energy production is clearly not optimized because cells secrete large amounts of acetate instead of using the more efficient respiratory chain. What then is the biological interpretation of the more appropriate maximization of the ATP yield per flux unit? Optimization of this objective is realized by maximizing ATP production (the nominator) and by minimizing the overall intracellular flux (the denominator). Hence, small networks with yet high, albeit suboptimal catabolic ATP formation are identified, which has three potential biological consequences. Firstly, resources are economically allocated because expenditures for enzyme synthesis are, on average, greater for longer pathways. Secondly, suboptimal ATP yields dissipate more energy and thus enable higher catabolic rates because the difference between the free energies of substrates and products must be used for both, energy conservation by synthesizing ATP (increase the yield) and energy dissipation to drive the chemical reaction (increase the rate) (; ). Thirdly, at a constant catabolic rate, a small network results in shorter residence times of substrate molecules until they generate ATP and probably other cofactors. The relative contribution of these consequences to the evolution of network regulation is unclear, but simultaneous optimization for ATP yield and catabolic rate under this optimality principle identifies a trade-off between the contradicting objectives of maximum overall ATP yield and maximum rate of ATP formation (). Under nutrient scarcity, in contrast, the metabolic state is closer to an optimal yield of ATP (or biomass) at the cost of the rate of formation. The constructed stoichiometric model of contains all presently known reactions in central carbon metabolism with 98 reactions and 60 metabolites (). To apply FBA, the reaction network was automatically translated into a stoichiometric matrix () by means of a parser program implemented in Matlab (MATLAB, version 7.0.0.19920 (R14), The MathWorks Inc., Natick, MA). corresponds to the stoichiometric matrix ( × ) and ν ( × 1) to the array of metabolic fluxes with ν as lower and ν as upper bounds, respectively. The above equations represent the conservation law of mass that is fundamental to constraint-based modeling. For all herein presented stoichiometric analyses, maximization of biomass yield is synonymous to the frequently used maximization of growth rate objective (). This is because stoichiometric models are sets of linear balance equations that are inherently dimensionless, hence maximization of the biomass reaction optimizes the amount of product (i.e., the yield) rather than a time-dependent rate of formation. The P-to-O ratio constraint was implemented by omitting the energy-coupling NADH dehydrogenase I (Nuo), cytochrome oxidase (Cyo) and/or cytochrome oxidase (Cyd) components of the respiratory chain. For a ratio of unity, Cyd and Nuo were set equal to zero. Under anaerobic conditions, electron flow is only possible via the NADH oxidases Nuo or NADH dehydrogenase II (Ndh) to fumarate reductase (Frd), hence coupled to succinate fermentation. For nitrate respiration, the terminal oxidase nitrate reductase (Nar) was used instead of Cyd or Cyo (). For the genome-scale analysis we used two recently reconstructed models of metabolism (; ). growth was simulated on glucose minimal medium for all six environmental conditions. ADP remained unbalanced, since otherwise formation of adenosine would be carbon-limited. For the proton-balanced model of , severe alternate optima occurred in central carbon metabolism given an unlimited proton exchange flux between the cell and the medium and a P-to-O ratio of 2, that is the upper bound of the biologically feasible range of P-to-O ratios (). To prevent the unlimited production of ATP equivalents through the ATPS4r reaction under this condition, all external protons involved in the respiratory chain and the transhydrogenase reaction were balanced (specifically, we balanced the external protons around the reactions ATPS4r, TDH2, CYTBD, CYTBO3, NO3R1, NO3R2, NADH6, NADH7, NADH8). A P-to-O ratio of 2 was implemented by assuming both the transport of four protons through CYTBO3 and NADH6 across the membrane and the diffusion of four protons through ATPS4r for the formation of one ATP equivalent. Linear optimization was used to identify optimal solutions for the objectives maximization of biomass or ATP yield, minimization of glucose consumption, minimization of the redox potential and minimization as well as maximization of ATP producing fluxes. The mathematical definition for all 11 objective functions is given in . While identification of a global optimal value is guaranteed, alternate optima occur frequently. Nonlinear optimization such as the minimization of the overall intracellular flux and maximization of biomass or ATP yield per flux unit do not produce alternate optima. Minimization of the overall intracellular flux always identifies a global optimum because the underlying optimization problem is quadratic and thus convex. Since such convexity cannot be assumed for maximization of biomass or ATP yield per flux unit, we used the general nonlinear solver of the programing package LINDO (Lindo Systems Inc., Chicago, IL) with 100 randomly chosen starting values to find global optima for these two non-convex nonlinear optimization problems. We implemented two independent approaches to validate our results. In a first approach we randomly changed the value of 5% of the variables by 10% iteratively 100 times for all constraint permutations (data not shown). Flux distributions with a higher objective function value were not identified in any of the iterations. respectively, corresponding to maximization of (2a) ATP and (2b) biomass yield per flux unit, respectively ( contains the mathematical definitions of all objective functions). ε represents a small value that characterizes the unique trade-off between ATP (biomass) maximization and minimization of the flux norm. We assessed the objective function value of ATP (biomass) yield maximization per flux unit for different ε values according to ε=ε (1±0.5), where ε was set equal to the objective function value, which was identified previously with the non-convex nonlinear objective function of max ATP (biomass) yield per flux unit, for the particular environmental condition ( and ). Given ε=ε, the previously found flux distribution yielded the optimal solution for every environmental condition ( and ). For all other ε values, independent optimizations only lead to suboptimal solutions. Hence, in the present case we have strong indication that global optima are actually identified. where for each flux , =1 stands for a non-zero, that is active flux in ν and =0 otherwise, and and are thresholds for determining non-zero fluxes (, and ), with κ and ε specifying the relative and absolute ranges of tolerance, respectively ( and ). The definitions of the objective functions of linear minimization of reaction steps and nonlinear maximization of ATP yield per reaction step () can be taken from . The constraints of the original linear programing problem with respect to steady state of mass balances and enzyme reversibilities were maintained ( and ). For κ and ε, we chose minimal values that still resulted in reasonable running times of the mixed-integer solver (specifically, we chose κ=0, ε=0 and κ=0, ε=1 for linear minimization of reaction steps and nonlinear maximization of ATP yield per reaction step, respectively). Optimality of the solution obtained by the mixed-integer nonlinear optimization was verified by randomly changing 5% of the integer values 10 times iteratively for the six conditions without additional constraints (data not shown). Instead of comparing all computationally (ν) and experimentally identified fluxes (ν) in the network, we focused on those that are sufficient to describe the complete systemic degree of freedom because most fluxes are linearly dependent. This minimal subset of fluxes was identified by calculability analysis (; ) from the null space of the stoichiometric matrix and allowed the calculation of all unique reaction rates in the underdetermined network. To reduce the considerable difference in magnitude of different fluxes in the network, their rates were expressed as split ratios of divergent fluxes ( and ), hence they are scaled to values between zero and unity. Error propagation was used to take the standard deviation of each of the experimentally determined split ratios into account. A default error of 5% was assumed for inactive flux ratios. Secretion of succinate, pyruvate and formate was not considered for calculability analysis, since the corresponding rates are negligible. Non-carbon fluxes such as respiration were also neglected. Generally, there are two basic principles to quantify the agreement between data series; that is, correlation coefficients that measure linear dependencies and the geometric distance (; ). Since we were interested in the similarity between multiple computational and experimental results rather than their linear dependency, we used the Euclidean distance to quantify the overall agreement. The Euclidean distance belongs to the group of L distance measures, which capture the deviation between two points in absolute terms () Our reduction of the overall solution space to a minimal set of 10 linear independent split ratios allowed the direct comparison of complete experimental and computational flux distributions. We defined the term predictive fidelity as the overall agreement between complete experimental and computational flux solutions relative to the specific experimental variance. This explicitly includes putative variability of split ratios due to existence of alternate optima. The global optimal value of different objective functions (e.g., maximization of biomass or ATP yield) is determined in a preliminary optimization step, which defines the computational solution space. In case of linear objective functions, with potential alternate optima, the best and the worst possible agreement of the underlying range of flux vectors and experimental data is subsequently quantified by minimizing and maximizing the standardized Euclidean distance, , respectively: The global optimal value has to hold () as well as the constraints of the original linear programing problem (, and ). marks the standardized Euclidean distance (), where the deviation ε between and ratios, and , respectively (), is weighted by the experimental variance σ (). Finally, the split ratios are a function of intracellular fluxes ν (). Note that both computational and experimental split ratios determine the Euclidean distance, such that even small changes in the and results can result in a completely different behavior. Hence, the predictive fidelity explicitly considers both alternate optima and unique solutions. All optimizations for the calculation of the predictive fidelity were performed with linprog and fmincon (MATLAB, version 7.0.0.19920 (R14)). Iterative calculations with 100 different, randomly chosen starting points were performed when the nonlinear solver fmincon was used. Predictive fidelity ranks the overall agreement between FBA flux predictions and experimental data. For detailed analysis of the predictive agreement in individual split ratios, experimental confidence intervals and theoretical flexibility due to alternate optima had to be taken into account. The absolute distance between each experimental and the mean computational split ratio was weighted by the experimental standard deviation δ and the possible range Δ of the computational split ratios (). The specific agreement ρ was quantified by a standardized variable: By weighting the absolute distance with the experimental uncertainty, the predictive accuracy is taken into account, that is a large absolute deviation is considered less severe if it is associated with a large experimental uncertainty δ. On the other hand, the absolute deviation is considered more severe if it is associated with a large computational uncertainty Δ. A default value of 0.05 was chosen for δ or Δ, respectively, when the values were zero. The hierarchical cluster trees were created with the linkage algorithm (MATLAB, version 7.0.0.19920 (R14)) using the Euclidean distances among all data points. The cophenetic correlation coefficients (i.e., the correlation coefficients of the distance values) () were calculated for all cluster trees to guarantee a faithful representation of the dissimilarities among the 10 ratios for every objective/constraint combination considered. Groups of nodes were assigned where the linkage among the nodes was less than 0.7, when the linkage was normalized to values between 0 and 1.

The transcriptional function of the androgen receptor (AR) is essential for normal male sexual development and drives the onset, and subsequent progression, of prostate cancer (PrCa; ; ). PrCa is the most common solid malignancy in men in the EU, and resulted in more than 85,000 deaths in 2004 alone (). Many of the current biomarkers of PrCa are androgen-regulated genes, including prostate-specific antigen (human glandular kellikrein (KLK3)/PSA), illustrating the enhanced AR activity in PrCa. Androgen ablation is an effective first-line therapy for the treatment of advanced PrCa; however, recurrence is common and is associated with androgen independence (). Despite loss of response to anti-androgens, most advanced PrCa express AR and have an active AR signalling cascade (). Cell line models also suggest that the AR is required for androgen-independent PrCa cell growth (). It is important to identify the main pathways downstream of AR transactivation that might mediate the ‘androgen-independent' function of the AR. These pathways are likely to contribute to the growth and progression of PrCa, and might include new candidate biomarkers or future therapeutic targets. The activated AR is known to bind as a homodimer to androgen-response elements (AREs), which consists of two 6-base pair ‘half-sites' arranged as inverted or direct repeats separated by 3 bp. The -derived (SELEX) consensus sequence for the AR was found to be the inverted repeat 5′-AGAACAnnnTGTACC-3′ (). However, sequence alignment of known AR genomic binding sequences reveals both inverted repeat and direct repeat consensus sequences (). This degeneracy of functional AREs, and also the divergence from the -derived binding sequence, make computational prediction of AR-binding sites in the human genome problematic. Alternative approaches have therefore been taken to identify AR-regulated genes. Several studies have used expression microarray techniques to identify AR transcriptional targets (; ; ; ). Although these approaches have successfully identified androgen-regulated transcriptional events, they cannot discern direct gene targets of the AR from secondary transcriptional events. Therefore, such expression studies neither give insights into the genomic sequences that are bound by the AR nor identify the initiating signals that ultimately produce the large number of downstream androgen-regulated transcriptional events. To identify direct transcriptional targets of the AR, we used chromatin immunoprecipitation (ChIP) with on-array detection (ChIP-chip) in the androgen-responsive LNCaP PrCa cell line. LNCaP cells were androgen deprived before stimulation with a synthetic androgen (R1881) or vehicle (ethanol). Using ChIP, an AR antibody was labelled and hybridized to an array with a tiling coverage of 24,275 gene promoter regions. Data from biological replicate AR ChIP-chip experiments were filtered by signal intensity (>twofold increase in androgen versus vehicle), reproducibility (Wilcoxon -value ⩽0.01) and probe coverage (>four probes contributing to these scores; see the online for details). This analysis identified as potential AR-binding sites 1,532 promoter regions that were more than twofold enriched in androgen-stimulated cells compared with vehicle-treated cells (; see and online). The thresholds for AR ChIP-chip allowed detection of 15 known direct AR target genes identified using literature searches ( online). These 15 known AR targets show a range of AR ChIP-chip enrichment values. For example, the well-studied AR targets in the carbonic anhydrase 3, PSA and pepsinogen C promoters had enrichment scores of 2.26, 2.51 and 4.46, respectively, suggesting that direct AR targets were present even among the lower scoring candidates. There are, at present, several examples in the literature of direct AR gene targets; therefore, to identify a larger number of androgen-responsive genes in our AR ChIP-chip data, we compared our data with gene lists from a published meta-analysis of six gene expression data sets, all of which used the LNCaP cell line to identify androgen-regulated genes (; see the online for details). We identified 92 genes that were enriched by AR ChIP-chip and were also androgen regulated in one or more expression data sets ( online), making them strong candidates for direct transcriptional regulation by the AR in response to androgen stimulation (see the online). To validate our AR ChIP-chip promoter targets, a set of 26 gene promoters with a range of enrichment and significance scores were assessed by independent AR-ChIP and quantitative PCR (). Three known AR-binding sites in the diazepam-binding inhibitor, KLK2 and KLK3 promoters (; ), and 23 new, randomly selected gene promoters were assessed. There was an increase in AR binding at 23 promoters, after 1 h of androgen exposure, suggesting a false discovery rate of approximately 13% in the AR ChIP-chip data (). AR recruitment was also assessed in the DUCaP PrCa cell line at eight candidate promoters. Six of these promoters were enriched in androgen-treated DUCaP cells (), suggesting that many of the AR-binding sites identified by ChIP-chip in LNCaP cells might be common AR targets in PrCa cells. Gene Ontology analysis of the 1,532 genes associated with AR ChIP-chip-enriched promoters showed AR target genes to be involved in protein synthesis, development, secretion, apoptosis and transcription (; online). To assess the clinical relevance of these AR ChIP-chip target genes, we retrieved expression values for these genes from publicly available clinical PrCa expression array data sets (; ; ). Interestingly, we found genes that were overexpressed in primary PrCa and metastatic PrCa compared with benign prostate epithelia, and also genes that were downregulated after androgen ablation therapy (; online). Further comparisons with published expression array gene lists from cell line experiments showed subsets of AR target genes upregulated in response to androgen treatment and genes that were downregulated after RNA interference (RNAi) ‘knock-down' of the AR in LNCaP cells (; online; ; ). The 92 AR target genes, which were also found to be androgen regulated in LNCaP cells, represent the strongest candidates for direct AR transcriptional targets (; online). Several of these androgen-regulated AR gene targets were overexpressed in primary PrCa samples compared with benign samples in two independent clinical expression array data sets (; ; ). AR target genes with Gene Ontology annotations for development and transcription had increased expression in primary PrCa and a subset was also upregulated in metastatic PrCa (). The upregulation of AR targets with transcriptional annotations suggests complex transcriptional changes downstream of the AR that might be activated in PrCa. These analyses show that many of the direct AR target genes identified by ChIP-chip analysis have cancer-related functional annotations, are upregulated in clinical PrCa and might therefore have potential as biomarkers or future therapeutic targets. We examined the 1,532 AR-binding sites identified by ChIP-chip for the presence of ARE-like sequences, to determine the preferred binding sequences of the AR. Although the occurrence of 15-bp ARE sequences was enriched in the AR promoter targets compared with non-candidate promoters (χ test, <2 × 10), only 410 (26.8%) of the 1,532 AR promoter-binding sites contained sequences that resembled the established 15-bp AREs (). The AR ChIP targets, which were identified as androgen-regulated genes in published expression array data, had an equal occurrence of AREs (26 of 92 genes, 28%), suggesting that functional AR target genes might lack the established 15-bp AR-binding sequence. To identify conserved sequences in the AR-bound promoters in an unbiased manner, we used the Nested Motif Independent Component Analysis (MICA) motif recognition software (). Nested MICA analysis using a ‘training set' of 225 AR ChIP-chip-enriched sequences did not identify any over-represented sequences that resembled the established 15-bp ARE sequence. However, several frequently occurring and highly constrained motifs were present among the Nested MICA searches for enriched 6-bp motifs ( online). Among the most frequently occurring non-repetitive motifs were two that resembled one-half of the 15-bp ARE sequences (motifs 2 and 6; ). The more constrained motif 2 AR ‘half-site', which aligned to one-half of the 15-bp ARE sequence (), was used for subsequent sequence analysis. This 6-bp AR ‘half-site' occurred in 1,212 (79.2%) of the AR candidate promoter sequences, including 876 (57.2%) of the AR ChIP-chip sequences that did not contain a 15-bp ARE sequence (), and was enriched in the AR ChIP-chip candidate promoters compared with non-candidate promoters present on the array (χ test, <2 × 10). We used an oligonucleotide pull-down assay to examine AR binding to the 6-bp AR ‘half-site'. As a positive control, the AR was shown to bind to the KLK2 promoter ARE sequence more strongly than a scrambled 15-bp oligonucleotide (). The AR also bound specifically to the 6-bp AR ‘half-site' sequence from the UNQ9419 promoter, but not a scrambled control oligonucleotide (), showing that the AR can bind directly to these 6-bp ‘half-sites'. To examine AR recruitment to 6-bp ‘half-sites', we used AR ChIP and quantitative PCR for the UNQ9419 promoter region, which lacks a 15-bp ARE sequence (). ChIP analysis showed androgen-dependent recruitment of the AR to the UNQ9419 promoter at a level similar to that of the KLK2 promoter (). An androgen treatment time course showed transient UNQ9419 upregulation and sustained upregulation of KLK2 expression (). These data show that the AR can bind directly to 6-bp ‘half-sites' and that AR transactivation might occur in genes adjacent to these 6-bp AR-binding sites. Direct AR binding to 6-bp ‘half-sites' raises important questions about our understanding of AR biology and further work will be required to determine the mechanism by which AR is recruited to these ‘half-sites' (see online for potential models). Further sequence analysis using the Genomatix Matbase program () and Nested MICA identified frequent consensus binding sequences for the avian erythroblastosis virus E26 homologue (ETS) family of transcription factors (; see the online). The ETS-like Nested MICA motif 9 and AR 6-bp ‘half-site' co-occurred in 1,073 (70%) of the AR-enriched promoters (χ test, <2 × 10), suggesting that there might be co-recruitment of ETS transcription factors and the AR to a subset of promoters. To investigate further the association of AR and ETS, we selected the ETS1 transcription factor as a candidate, as the ETS1-binding site was among the most common ETS sequence motifs found in the AR ChIP-chip promoters (as identified by Genomatix Matbase) and ETS1 was recently reported to be overexpressed in PrCa (). Using ChIP, an ETS1 antibody showed that ETS1 was associated with a subset of AR-bound promoters (). The CCNG2 and UNQ9419 promoters contain predicted ETS1-binding sites () and showed androgen-dependent recruitment of both the AR and ETS1 by ChIP (, ). However, the PRAME promoter does not contain a predicted ETS1-binding site () and was not enriched by ETS1 ChIP (). Knockdown of endogenous ETS1 by RNAi resulted in a matched reduction in CCNG2 and UNQ9419 transcripts, but did not affect the level of PRAME transcript (). Conversely, ETS1 overexpression resulted in enhanced transcription of CCNG2 and UNQ9419, but not PRAME (). Confocal microscopy showed that endogenous ETS1 was located throughout LNCaP cells grown in steroid-depleted media and was redistributed to the nucleus together with transfected AR–green fluorescent protein (GFP; ) or endogenous AR () in LNCaP cells on androgen treatment. In an AR luciferase reporter assay, ETS1 transfection enhanced AR transactivation in an AR- and androgen-dependent manner ( online). The AR was co-immunoprecipitated using an ETS1 antibody, suggesting a direct interaction between the AR and ETS1 in LNCaP cells ( online). These data indicate androgen-dependent recruitment of ETS1 to a subset of AR promoter targets and also that endogenous ETS1 is required for expression of these AR target genes (see the online). Recruitment of the AR to 6-bp ‘half-sites' raises the question of how the AR is selectively recruited to its targets in large mammalian genomes. This might occur by co-operative binding with other factors, such as ETS1. As ETS1 is recruited to AR promoter targets in response to androgens, it is unlikely that ETS1 acts as a ‘pioneer factor' for the AR, as has been reported for FOXA1 and the estrogen receptor (; ). The implication that ETS transcription factors transactivate the AR might have an impact on our understanding of the functional effects of TMPRSS2-ETS gene fusions given the prevalence of this rearrangement in PrCa (). LNCaP and DUCaP cells were grown in RPMI (Invitrogen, Paisley, UK) media supplemented with 10% FBS. Cells were transfected with the GeneJuice transfection reagent (Merck, Nottingham, UK), according to the manufacturer's instructions. The pSG5–ETS1 expression construct was a kind gift from Dr A. Bègue (). Cells were grown to 70–80% confluence in phenol-red-free RPMI (Invitrogen) supplemented with 10% charcoal-stripped FBS (HyClone, Logan, UT, USA) for 48 h before stimulation with 1 × 10 M R1881 (synthetic androgen) or an equal volume of ethanol for 1 h. ChIP was carried out as previously described, using 5 μg of AR (N20, Santa Cruz) or ETS1 (C20, Santa Cruz, Biotechnology, Santa Cruz, CA, USA) antibodies (see the online for details; ). Biological replicate hybridizations were carried out for AR ChIP from LNCaP cells stimulated for 1 h with R1881 or ethanol (vehicle), co-hybridizing with labelled total genomic DNA to NimbleGen Systems (Madison, WI, USA) 1.5 kb promoter arrays. Raw data for AR ChIP-chip experiments is available through ArrayExpress (accession E-TABM-233, ). Array analysis was carried out using the limma package in the R statistical software (see the online for details). Nested MICA motif recognition software (v0.7.2) was used to identify conserved sequence motifs in a ‘training' set of 225 highly enriched AR target sequences (see and online for details). Searches for the AR, ETS1 and the conserved 6 bp Nested MICA sequence motifs were carried out on all 1,532 AR ChIP-chip targets, using the position weight matrices shown in online (see the online for details). Real-time quantitative PCRs were carried out in an ABI Prism 7900, using SYBRgreen PCR master mix (Applied Biosystems, Warrington, UK). Reactions were carried out in triplicate and with biological replicates. Primers are shown in online. LNCaP cells lysates were incubated with AR (N20, Santa Cruz), ETS1 (C20, Santa Cruz) or control rabbit IgG antibodies. Immune complexes were isolated on protein-A/G beads, washed four times with RIPA buffer and resuspended in sample loading buffer, before western blotting for AR (N20, Santa Cruz). Oligonucleotide pull-down assays using LNCaP cell lysates were carried out as described previously (see the online for details; ). LNCaP cells on coverslips were treated, transfected and stained for AR (AR-N20, Santa Cruz) and ETS1 (C20, Santa Cruz or 1G11, Novocastra Laboratories, Newcastle upon Tyne, UK) as indicated (see the online for details). is available at online ().

Multicellular organisms must modulate cellular growth and proliferation in response to available nutrients, energy, and growth factor signaling. The regulatory kinase target of rapamycin (TOR) has emerged as a convergence point for the transduction of these signals into appropriate changes in cell metabolism and growth. TOR activity is stimulated by insulin-responsive phosphatidylinositol-3 kinase (PI3K)/Akt signaling, by nutrients such as amino acids, and by high cellular energy levels (for reviews see ; ). In response to these signals, TOR effects changes in a diverse number of downstream processes, including transcription, translation, nutrient import, and autophagy, to achieve the alignment of cellular energy utilization with available resources. Unsurprisingly, improper activation of TOR signaling has been implicated in the development of cancers and hamartoma syndromes and in metabolic diseases such as diabetes and obesity (). For example, mutations in either component of the TSC1–TSC2 complex, which is an upstream inhibitor of TOR signaling, result in the formation of benign tumors in multiple tissues. Although TOR is known to affect a wide range of cellular processes, the relative contribution of these processes and how they interact to result in a directed growth response remain poorly understood. Components of the translational machinery are well established downstream effectors of TOR signaling (). TOR directly phosphorylates eukaryotic initiation factor-4E–binding protein (4E-BP) and ribosomal protein S6 kinase (S6K), thereby facilitating cap-dependent translation and ribosome biogenesis. Although these effects on protein synthesis are likely to contribute substantially to cellular growth capacity, they are unlikely to fully account for the growth effects of TOR. For example, whereas inactivation of TOR results in a nearly complete block of protein synthesis in yeast, this effect is more modest in mammalian cells, with an ∼15–50% decrease in translation rate (; ). In , null mutations in are without effect on cell growth (), and the growth phenotype of -null mutants is significantly less severe than that of mutants (; ; ). Activation of S6K only partially overcomes the growth arrest of mutants in this system. Recent genetic studies in mouse have also shown that the ribosomal substrate of S6K, rpS6, does not appear to be a relevant mediator of the growth effects of this pathway (). These observations have motivated the search for other effector pathways and cellular processes downstream of TOR that might contribute to its effects on cell growth. A growing number of studies of TOR structure, function, and localization point to an important role for TOR signaling in controlling vesicular trafficking. Biochemical studies in yeast have found that TOR localizes to intracellular vesicles and cofractionates with endosomal markers (; ), which is consistent with a role in the endocytic compartment. In addition, TOR has a highly conserved function as a regulator of autophagy, which is a process of cytoplasmic degradation that involves the reorganization of intracellular membranes into autophagic vesicles (; ). Finally, TOR is structurally related to the class III PI3K/Vps34 family of lipid kinases, with well characterized roles in endocytosis. Interestingly, recent studies have identified a novel role for hVps34 in relaying intracellular nutrient status to TOR (; ), indicating that this family of molecules may have common roles in nutrient sensing and membrane trafficking. We report the identification of the clathrin-uncoating ATPase Hsc70-4 in a genetic enhancer/suppressor screen for novel TOR interactors in . Hsc70-4 is a critical regulator of clathrin-mediated endocytosis, and we provide evidence that TOR signaling influences bulk endocytosis, as well as the targeted endocytic degradation of a specific amino acid transporter. Our results suggest that TOR controls growth, in part, by simultaneously down-regulating aspects of endocytosis that inhibit growth and up-regulating potential growth-promoting functions of endocytosis. To better understand the regulation and downstream effects of TOR-mediated growth signaling, we took a genetic approach to reveal novel players in the TOR signaling pathway. We used a tissue-specific TOR overexpression phenotype as the sensitized background for a dominant-modifier ethyl methanesulfonate (EMS) mutagenesis screen. To create this background, element–mediated transposition was used to introduce a copy of the wild-type cDNA, which was preceded by multiple copies of the enhancer, into the genome. -driven overexpression of TOR (TOR) led to an overall reduction in size of the adult eye (), demonstrating the paradoxical decrease in growth previously observed to result from wild-type TOR overexpression (). In addition to the reduction in eye size, ommatidial patterning was disorganized, with some ommatidia appearing fused or pitted. Immunohistochemistry of TOR imaginal discs revealed a delay in morphogenetic furrow progression and missing and disorganized photoreceptor cells, as well as an increased level of cell death (Fig. S1, available at ). The severity of the TOR phenotype was strongly enhanced by a heterozygous-null mutation of the negative TOR regulator () and was suppressed by addition of rapamycin to the media (Fig. S1). These results suggest that the TOR phenotype results from inappropriately high levels of TOR signaling, and they demonstrate the potential utility of this overexpression phenotype as a dosage-sensitive genetic background. We next sought to identify novel factors involved in TOR signaling by screening for the ability of EMS-generated mutations to dominantly modify the TOR phenotype. From ∼60,000 F1 progeny scored, we identified 23 TOR enhancers and 2 suppressors. Within this collection were two independent groups that failed to complement for lethality. We report the analysis of a third chromosome complementation group consisting of two alleles ( and ) isolated as dominant enhancers of the TOR phenotype. Heterozygous mutation of either or caused a further reduction in eye size of TOR flies ( and Fig. S1) and a similar enhancement of MS1096-Gal4-driven TOR misexpression phenotypes in the adult wing (Fig. S1). Through recombination and deficiency mapping we localized the and mutations to the 88E4 genomic region. Complementation tests with lethal mutations in this interval revealed that and disrupt the gene, which encodes a constitutively expressed member of the stress-induced Hsp70 family of ATPases. Independently isolated mutant alleles of also displayed a strong enhancement of the TOR phenotype (Fig. S1). In addition, both the dominant TOR enhancement and the recessive lethality of and could be rescued by a transgene carrying a wild-type copy of (; Fig. S1). We found that the and mutations result in single amino acid substitutions within the ATPase domain of Hsc70-4 (Ser to Phe and Arg to Cys, respectively; ). These residues are completely conserved in fungi, plant, and animal Hsc70-4 orthologues, indicating their likely importance for Hsc70-4 function. Therefore, from this point forth we refer to these mutations as and . Hsc70-4 catalyzes the uncoating of clathrin-coated vesicles, which is an essential late step in clathrin-mediated endocytosis (; ). In addition, Hsp70 family members can act as chaperones to regulate protein folding and stability (). To determine whether either of these functions of Hsc70-4 contributes to its genetic interactions with TOR, we first assayed the ability of other endocytic factors to modify TOR misexpression phenotypes. encodes the homologue of dynamin, which is a GTPase critical for the proper membrane closure and budding of endocytic vesicles from the plasma membrane (). We found that expression of a dominant-negative form of shibire, Shi (), strongly enhanced the TOR phenotype (), suggesting that Hsc70-4 may influence TOR signaling through its role in endocytosis. In contrast, mutations did not lead to increased levels of TOR protein in TOR eye imaginal discs (), indicating that the enhancement of the TOR phenotype by mutants is not caused by increased stability or abundance of overexpressed TOR protein. We also observed no increase in total levels of endogenous TOR protein in mutants (). Interestingly, we found that expression of Shi led to increased localization of TOR to vesicular structures, a subset of which were accessible to an endocytic tracer (; see below), further supporting a link between TOR signaling and endocytosis. To further characterize the interplay between TOR and endocytosis, we tested the effect of TOR signaling on several endocytic markers in the larval fat body. The fat body serves as a nutrient storage organ that is analogous to the vertebrate liver (), and it has recently been shown to act as a nutrient sensor capable of affecting global growth through a TOR-dependent humoral mechanism (). We tested the effects of altered nutrient availability and TOR signaling on the intracellular localization of components of the endocytic machinery, using GFP-fusions to Rab5, Rab7, and clathrin. In fat body cells from fed control animals, we observed GFP-Rab5 localization both at the cell surface and throughout the cytoplasm (). In contrast, larvae subjected to a 5-h starvation displayed a variable redistribution of GFP-Rab5 toward the cell surface, with a less diffuse, punctate pattern often observed in these cells (). More direct alteration of TOR signaling, through overexpression of TSC1 and TSC2, or TOR itself, also affected GFP-Rab5 localization, resulting in the appearance of large aggregates near the cell surface (). The localization of GFP-Rab7 and clathrin-GFP was also altered in response to TOR inhibition (unpublished data). Overexpression of TOR in fat body cells also resulted in increased expression of Hsc70-4, as monitored by a GFP-Hsc70-4 fusion expressed from the endogenous locus (). Together, these data indicate that altered TOR signaling effects an endocytic response in the fat body. To better understand the nature of these endocytic changes, we monitored the effects of TOR signaling on the ability of fat body cells to internalize a fluorescent endocytic tracer, Texas red–conjugated avidin (TR-avidin). In control experiments we found that genetic disruption of or prevented proper endocytic uptake in these cells, as was demonstrated previously in Garland cells (; ). TR-avidin failed to be efficiently internalized in Shi-expressing fat body cells, and instead accumulated near the cell surface, often in large aggregates (). TR-avidin uptake was also blocked in cells that were homozygous for a null allele of (), and, in this case, a decrease in both surface-bound and internalized tracer was observed (). A similar, but less extensive, decrease in TR-avidin uptake was observed in and mutant cells (Fig. S2, available at ). We next used this assay to monitor the effects of altered TOR signaling on bulk endocytic uptake. Mosaic clones mutant for a null allele of showed a nearly complete block of TR-avidin uptake (). Similar effects were observed when TOR signaling was inhibited by overexpression of TSC1–TSC2 or TOR (Fig. S2). To see whether increased activation of TOR could further stimulate endocytic internalization, we generated mitotic clones homozygous for a -null allele. Increased tracer uptake was observed in these cells (). Overexpression of Rheb, which is an upstream activator of TOR, caused a similar increase in TR-avidin uptake (Fig. S2). We also tested whether two targets of TOR, S6K and 4E-BP, affected endocytosis. Cells mutant for a null allele of showed a strong decrease in TR-avidin uptake, suggesting that the effects of TOR on endocytosis are mediated in large part through S6K (). In contrast, overexpression of 4E-BP, whose activity is inhibited by TOR, had no effect in this assay (). As an alternative approach to monitoring endocytosis in vivo, we tested the ability of fat body cells to internalize larval serum protein 2 (Lsp2). Lsp2 is present in the hemolymph during larval development, and is internalized through endocytosis by the fat body during the late third instar period (). Anti-Lsp2 staining revealed a fine punctate appearance throughout the cytoplasm of wild-type cells (). In contrast, Shi-expressing cells displayed large aggregates of Lsp2 protein at the cell surface and an absence of internal Lsp2 staining (). mutant cells were also defective in Lsp2 uptake, and, again, differed from Shi-expressing cells in showing a decrease in Lsp2, both internally and at the surface (). The effects of TOR signaling on Lsp2 internalization were identical to the effects on TR-avidin; Lsp2 uptake was severely reduced in or mutant cells and in TSC1–TSC2–overexpressing cells, and was increased in cells mutant for or overexpressing Rheb (, and Fig. S2). Together, these results demonstrate that the Rheb–TOR–S6K pathway is required for proper endocytic uptake and indicate a novel role for TOR signaling in the positive stimulation of bulk endocytosis. Furthermore, they indicate that Shi expression and loss of result in distinct endocytic phenotypes in the fat body. In addition to the nonselective endocytic uptake described in the previous section, studies in yeast and cultured mammalian cells have shown that TOR can selectively influence the endocytic uptake and degradation of specific nutrient transporters (; ). To investigate the effects of TOR signaling on targeted endocytosis, we generated an antibody to Slimfast, which is a cationic amino acid importer previously shown to positively affect organismal growth through effects on the TOR and PI3K signaling pathways (). We used this antibody to test whether TOR signaling may regulate the endocytic turnover of Slimfast in response to nutrient conditions. Antibody staining of fat body tissue confirmed that Slimfast is localized primarily to the cell surface, with peak levels just below the plasma membrane ( and Fig. S3, available at ). Disruption of endocytosis through Shi expression resulted in a marked increase in Slimfast at the cell surface (), suggesting that endocytosis normally acts to antagonize the surface expression of Slimfast. In contrast, disruption of did not lead to increased surface levels of Slimfast (), which is consistent with the distinct effects of Shi and mutants on bulk endocytic uptake. We next asked whether changes in TOR signaling affect Slimfast levels. Activation of TOR through clonal overexpression of Rheb caused an increase in surface levels of Slimfast, which was similar to the effects of Shi (). The relative effect of Rheb on Slimfast levels was especially pronounced in larvae subjected to a 24-h starvation (Fig. S3). Under these conditions, TOR activity is reduced in wild-type cells, but maintained in Rheb-overexpressing cells (). To test whether Slimfast up-regulation is simply caused by increased protein synthesis in Rheb-overexpressing cells, we added the translation inhibitor cycloheximide during the 24-h starvation period. Under these conditions, Rheb overexpression still led to a modest relative increase in Slimfast levels (Fig. S3), which was consistent with a partial posttranslational effect. Reduction in TOR signaling also had a marked effect on Slimfast staining, as overexpression of TSC1–TSC2 led to a decrease in Slimfast levels (). Thus, the localization and levels of Slimfast are sensitive to TOR activity, with high levels of TOR signaling resulting in accumulation of Slimfast near the cell surface, and decreased TOR signaling leading to Slimfast down-regulation. We next sought to determine whether TOR signaling might mediate these changes in Slimfast levels through effects on the targeted endocytosis of this importer. We tested whether the Slimfast down-regulation that results from TSC1–TSC2 overexpression requires endocytosis by cooverexpressing TSC1–TSC2 and Shi. In the absence of functional endocytosis, TSC1–TSC2 overexpression no longer led to Slimfast down-regulation. Instead, Slimfast protein appeared to be trapped in large aggregates near the surface of these cells (). Similar results were observed when TOR activity was inhibited through starvation. In well fed larvae, partial disruption of endocytosis through expression of dominant-negative Rab5 had little effect on Slimfast levels (). In contrast, Rab5 expression led to a marked persistence of Slimfast at the plasma membrane in starved larvae (). Collectively, these results demonstrate that endocytosis is critical for Slimfast down-regulation in response to reduced TOR signaling, resulting either from poor nutrient conditions or inactivation by the TSC1–TSC2 complex. To further investigate whether the endocytic down-regulation of Slimfast is a specific, targeted process, we examined the effects of altering components of the endocytic sorting/targeting machinery on Slimfast levels. Hrs (hepatocyte growth factor–regulated tyrosine kinase substrate) is an early endosome–associated, ubiquitin-binding protein that is critical for proper endocytic sorting. As shown in , homozygous mutation of led to a marked increase in Slimfast levels near the cell surface. Expression of a dominant-negative version of Nedd4, which is an E3 ubiquitin ligase, similarly caused persistence of Slimfast near the cell periphery (). Monoubiquitination by the Nedd4 family of enzymes is critical for the initial targeting of plasma membrane–localized proteins to clathrin-coated pits for internalization (). Together, these results demonstrate that components of the endocytic targeting/sorting machinery are critical in mediating Slimfast turnover. Our observations indicate that TOR signaling affects both bulk and targeted endocytosis. Interestingly, we note that TOR appears to exert opposing effects on these processes, stimulating bulk endocytosis ( and ) and inhibiting the targeted endocytic degradation of Slimfast (). To test whether TOR's effects on endocytosis might contribute to its role in promoting cell growth, we assayed the growth properties of cells in which endocytosis was disrupted. mutant cells, relative to wild-type cells (). These effects were acutely dosage-sensitive, as heterozygous cells were also larger than wild type (Fig S4). Expression of Shi or of a dominant-negative transgene also resulted in increased cell size (). This cytometric profile of increased cell size and G2 content is similar to that of mutations in negative regulators of TOR such as Tsc1, Tsc2, and Pten (; ), and to that of cells overexpressing Rheb (). Disruption of endocytosis also led to dosage-dependent cell size changes in the fat body. Expression of Shi caused a 1.5-fold increase in fat body cell size (), which is consistent with its growth effects in wing disc cells. In contrast, cells homozygous for mutant alleles of showed either a slight or a threefold decrease in cell size (Δ16 null; ). Given that Shi expression and mutation result in a similar block of endocytic uptake, but show distinct effects on Slimfast expression, these results suggest that the relative contributions to growth of bulk and transporter-mediated nutrient uptake may differ between imaginal disc and fat body cells. In addition to acting downstream of TOR, the potential role of endocytosis in controlling nutrient import suggested that it might also function upstream to regulate TOR activity, which is stimulated by nutrients in general and Slimfast in particular (). Endocytosis has also been shown to affect the levels and activity of the insulin receptor (), which may function upstream of TOR by activating PI3K signaling (). Indeed, we observed an increase in insulin receptor levels similar to that of Slimfast in response to Rheb overexpression (Fig. S3). Therefore, we examined the effects of endocytosis on in vivo markers of PI3K and TOR activity. To test whether endocytosis influences PI3K signaling, we analyzed the effects of Shi expression on the transcription factor FOXO, which is excluded from the nucleus in response to PI3K-dependent phosphorylation (). Clonal expression of Shi in fat body cells resulted in relocalization of FOXO from the nucleus to the cytoplasm (), reflecting increased PI3K activity and, thus, indicating that endocytosis normally exerts a negative effect on PI3K signaling in these cells. As a cellular readout of TOR signaling, we monitored the effects of endocytosis on autophagy, which is a process of cytoplasmic degradation that is inhibited by TOR (; ). Disruption of endocytosis through expression of Shi prevented proper induction of autophagy after starvation (), which is indicative of high levels of TOR signaling in these cells. A similar inhibition of autophagy was observed in -null cells and in cells expressing dominant-negative Hsc70-4 (Fig. S4, available at ). To assess the effects of endocytosis on the kinase activity of TOR, we monitored the levels and the phosphorylation status of known TOR substrates. Inactivation of TOR through mutation or starvation has previously been shown to cause a decrease in S6K phosphorylation at Thr398, as well as an increase in S6K protein levels, through an unknown mechanism (; ). More recently Akt (Ser505) has been identified as a substrate for TOR in association with its cofactor rictor (). Accordingly, in control experiments we found that homozygous larvae showed a decrease in total S6K levels and a slight decrease in Thr398 phosphorylation relative to wild-type controls (); normalized for S6K levels, Thr398 phosphorylation was increased as expected. mutant larvae also showed a strong loss of Akt-Ser505 phosphorylation, which is consistent with recent studies showing that the TSC1–TSC2 complex promotes TOR–rictor signaling to Akt (). We next examined these markers in extracts from transheterozygous mutants, which are viable through the third instar larval stage. Similar to mutants, larvae showed a decrease in overall S6K levels and a reduction in S6K and Akt phosphorylation; in this case, Thr398 phosphorylation relative to S6K abundance was reduced compared with wild type (). The similarity between and mutants was further underscored through chemical and genetic interaction studies. Imaginal disc cells doubly mutant for and showed a synergistic increase in cell size as compared with their single mutant counterparts (). Similarly, using the -FLP system to generate homozygous mutant eye tissue, we observed an exacerbation of -induced tissue overgrowth in the presence of mutation, in some cases resulting in a marked outgrowth of tissue in the anterior portion of the retina (). We also tested whether mutation of affects sensitivity to rapamycin, which inhibits TOR signaling through the formation of an inhibitory binding complex with the intracellular protein FKBP12 (; ). Growth of wild-type is delayed on media containing rapamycin, and this delay is sensitive to the dosage of TOR pathway genes (; ). We found that heterozygous mutation significantly alleviated the rapamycin-induced delay (), which is consistent with increased levels of TOR signaling. Collectively, the cellular, biochemical, and genetic effects of endocytosis are consistent with this process playing a significant role in cell growth and TOR signaling. Inactivation of TOR causes an inhibition of cellular growth, a reduction in cell size, and a suppression of cell cycle progression. In addition to well described changes in protein synthesis and ribosome biogenesis, recent studies have suggested that other cell processes are likely to contribute to these growth effects of TOR. The present study identifies endocytosis as one such process. Our results demonstrate that the clathrin-uncoating ATPase Hsc70-4 interacts genetically with TOR and Tsc1, and that bulk endocytosis is stimulated in cells with activated TOR signaling. Conversely, we find that TOR activity inhibits the endocytic degradation of nutrient transporters such as Slimfast. Together, these endocytic effects of TOR promote both the bulk and targeted uptake of nutrients and other biomolecules required for cell mass increase (). In addition to this direct role in cellular biosynthesis and growth, nutrients also act as potent regulators of TOR signaling. Indeed, Slimfast was previously identified as an upstream activator of TOR (). Our findings that disruption of endocytosis effects cell size, rapamycin sensitivity, and TOR kinase activity are consistent with an additional role for endocytosis upstream of TOR. Mutations that disrupt endocytosis are likely to have both positive and negative effects on nutrient uptake and cell growth because they inhibit bulk endocytic uptake, as well as degradation of nutrient transporters and other signaling molecules. Thus, the overall effects of endocytic disruption on nutrient uptake, cell growth, and TOR signaling are difficult to predict a priori. Our results suggest that both the cellular context and the specific step at which endocytosis is blocked influence the growth response. Thus, in fat body cells, expression of Shi resulted in an increase in cell size, whereas loss of function caused reduced cell size. We note that these changes mirror the effects of these mutations on Slimfast levels; whereas both Shi expression and mutation decreased bulk endocytic uptake, only Shi resulted in increased levels of Slimfast. In contrast, both Shi and mutants led to the increased size of wing imaginal disc cells, suggesting that in these cells the growth-inhibitory effects of endocytic degradation of membrane proteins such as Slimfast predominate over the potential positive effects of increased bulk uptake. Similarly, our results indicate a complex effect of endocytosis on TOR signaling. Partial reduction in levels lead to an increase in TOR signaling, as was evident in the TOR interaction and rapamycin resistance. In contrast, larvae that are homozygous mutant for show a decrease in TOR kinase activity. These results suggest that modest inhibition of endocytosis may increase TOR signaling, whereas a complete block of endocytosis may reduce it. A striking parallel to the inverse regulation of bulk and targeted endocytic processes by TOR can be observed in its effects on autophagy in yeast. Through autophagy, random portions of cytoplasm are nonselectively engulfed within double membrane–bound vesicles for delivery to the lysosome. Activation of TOR causes this nonselective form of autophagy to be suppressed, and, instead, the autophagic machinery engages in a selective type of autophagy known as the cytoplasm–vacuole targeting (CVT) pathway, which is responsible for lysosomal delivery of specific hydrolases (). Thus, TOR acts as a switch between selective and nonselective autophagy. TOR may also be involved in switching between clathrin-and caveolae/raft-mediated endocytosis in higher eukaryotes. A genome-wide survey of protein kinases found that RNAi-mediated inactivation of TOR in HeLa cells inhibited clathrin-dependent processes such as transferrin uptake and vesicular stomatitis virus infection, and stimulated cavelolae/raft-dependent events (). Together, these findings suggest that TOR may control the specificity of membrane trafficking components. In addition, our results show that S6K, which is an important TOR substrate, acts downstream of TOR in promoting bulk endocytosis, but is not involved in the suppression of starvation-induced autophagy. The identification of endocytosis as a TOR-controlled function adds to the growing list of cell processes regulated by TOR, including protein synthesis, ribosome biogenesis, autophagy, metabolic gene expression, and cytoskeletal organization. How these distinct functions interact to achieve a coordinated growth response is only beginning to be understood. One likely mechanism involves the common use of molecular components and cellular substrates by different cell functions, as in the case of selective and nonselective autophagy, bulk endocytosis, and endocytic degradation. Two or more distinct branches of TOR signaling may also act cooperatively to control the same target, as in the case of Slimfast regulation by both translation and endocytosis, or may act in opposition, as previously observed for the role of S6K in limiting autophagy. Finally, distinct TOR complexes may converge on the same targets with opposing effects, as in the regulation of Akt by TOR-raptor versus TOR–rictor complexes (; ). The finding that TOR signaling regulates the levels of Slimfast, which was previously shown to function upstream of TOR, adds another layer of complexity to the TOR signaling network. . Flies were incubated at 25°C on standard cornmeal–yeast medium, unless otherwise indicated. Starvation experiments were performed essentially as previously described (). In brief, larvae were transferred to fresh fly food supplemented with yeast paste, allowed to feed for 24 h, and then transferred to a 20% sucrose solution and starved for the indicated time before dissection. For rapamycin treatment, larvae were cultured in standard fly medium supplemented with 2 μM rapamycin. A 1.5-kb KpnI (blunted)–BamHI fragment containing four copies of the eye-specific enhancer fragment from the gene and the noninducible promoter was isolated from pBD1915 (), and ligated into the unique SalI (blunted) and XbaI sites located immediately upstream of a FLAG-tagged cDNA in pBluescript (). The sequence was excised as a 9.3-kb XbaI–PspOMI fragment and cloned into the transformation vector pCasper4 digested with XbaI and NotI. element–mediated transposition was used to introduce the TOR transgene into flies by standard methods. CyO and TM3 balancers carrying TOR insertion lines were created through transposase-mediated mobilization of an X chromosome TOR insert. #text and were localized to map position 57 centimorgans on the third chromosome by genetic recombination, and this region was further refined to the 87E8-88E6 interval by deficiency mapping. In complementation tests against available lethal mutations in this region, failed to complement multiple element-, EMS-, and x-ray–induced mutations in showed complete failure to complement and , which is an antimorphic allele of , and partial complementation with other alleles. Based on this pattern of complementation and on the strength of TOR interactions, we classify as an antimorph and as a hypomorphic allele of . Sequencing and mutations. Genomic DNA was isolated from and homozygous mutant larvae, and the second exon of (containing the coding region) was PCR amplified using the primers CCATTTTCTCAGTATTACTTCTCCTCTGGC and GAGAACTGTTACTGTATGGTTGCATTGAGG and sequenced using the primers CGAGAAAAGGAAAATTAGAATTGTAAAACACACC, GGAGATCTCTTCGATGGTGCTTACC, CAAGCACAAGAAGGATCTGACCACC, CATTCTGCACGGCGACAAGTCG, and GTCGTCTCTCCAAGGAGGACATC. #text h s - F L P / F R T – m e d i a t e d m i t o t i c r e c o m b i n a t i o n w a s i n d u c e d i n 0 – 8 h e m b r y o s , b e f o r e i n i t i a t i o n o f e n d o r e p l i c a t i o n i n f a t b o d y t i s s u e , t h r o u g h a 2 h , 3 7 ° C h e a t s h o c k i n a n a i r i n c u b a t o r . F o r u n i f o r m G F P m a r k e r e x p r e s s i o n , t h e f a t b o d y d r i v e r s C g - G a l 4 o r f b - G a l 4 w e r e u s e d t o a c t i v a t e U A S - G F P l i n e s o n F R T - b e a r i n g c h r o m o s o m e s . L o s s - o f - f u n c t i o n c l o n e s w e r e m a r k e d b y t h e a b s e n c e o f G F P . i r d i n s t a r l a r v a e w e r e d i s s e c t e d i n P B S , i n v e r t e d , a n d t r a n s f e r r e d t o 1 . 5 - m L t u b e s c o n t a i n i n g 3 . 7 % f o r m a l d e h y d e i n P B S T ( P B S + 0 . 1 % T w e e n 2 0 ) f o r 4 - h ( i m a g i n a l d i s c s ) o r o v e r n i g h t ( f a t b o d y ) f i x a t i o n a t 4 ° C , w i t h g e n t l e a g i t a t i o n . F i x w a s r e m o v e d t h r o u g h f o u r 5 - m i n w a s h e s i n P B S T . x e d c a r c a s s e s w e r e b l o c k e d f o r 2 – 4 h i n P B S T G ( P B S T + 5 % n o r m a l g o a t s e r u m ) b e f o r e i n c u b a t i o n i n P B S T G + p r i m a r y a n t i b o d y a t 4 ° C o v e r n i g h t . C a r c a s s e s w e r e w a s h e d f o u r t i m e s f o r 2 0 m i n i n P B S T a n d b l o c k e d i n P B S T G f o r 1 – 2 h b e f o r e i n c u b a t i o n i n P B S T + s e c o n d a r y a n t i b o d y + 1 μ M H o e c h s t 3 3 2 5 8 a t 4 ° C o v e r n i g h t . C a r c a s s e s w e r e r i n s e d i n P B S T f o u r t i m e s f o r 2 0 m i n . T i s s u e w a s d i s s e c t e d i n P B S a n d m o u n t e d i n F l u o r o G u a r d r e a g e n t ( B i o - R a d L a b o r a t o r i e s ) . t i - S l i m f a s t ( 1 : 4 0 0 d i l u t i o n o f r a b b i t p o l y c l o n a l s e r a r a i s e d a g a i n s t a b a c t e r i a l l y p r o d u c e d f u s i o n b e t w e e n G S T a n d t h e C O O H - t e r m i n a l c y t o p l a s m i c t a i l o f S l i f [ a m i n o a c i d s 5 5 4 – 6 0 4 ] ) , a n t i - F O X O # 3 0 1 5 ( 1 : 3 0 0 ; g i f t f r o m O . P u i g , U n i v e r s i t y o f C a l i f o r n i a , B e r k e l e y , B e r k e l e y , C A ) , a n t i - F L A G M 5 ( 1 : 5 0 0 ; S i g m a - A l d r i c h ) , a n t i - E l a v 9 F 8 A 9 ( 1 : 5 0 ; D e v e l o p m e n t a l S t u d i e s H y b r i d o m a B a n k ) , a n t i B - g a l ( 1 : 5 0 0 ; C a l b i o c h e m ) , a n t i I n R 3 8 6 ( 1 : 2 , 0 0 0 ; g i f t f r o m R . F e r n a n d e z , N e w Y o r k U n i v e r s i t y M e d i c a l C e n t e r , N e w Y o r k , N Y ) , a n t i - L s p 2 ( 1 : 5 0 0 ; g i f t f r o m H . B e n e s , U n i v e r s i t y o f A r k a n s a s , L i t t l e R o c k , A R ) . r c a s s e s w e r e d i s s e c t e d , f i x e d , w a s h e d , a n d b l o c k e d f o r 2 – 4 h i n P B S T G , b e f o r e b e i n g i n c u b a t e d a t 4 ° C o v e r n i g h t i n 0 . 1 6 5 μ M A l e x a F l u o r 5 6 8 p h a l l o i d i n ( I n v i t r o g e n ) i n P B S T G c o n t a i n i n g 1 μ M H o e c h s t 3 3 2 5 8 . C a r c a s s e s w e r e r i n s e d i n P B S T f o u r t i m e s f o r 2 0 m i n b e f o r e f i n a l d i s s e c t i o n , m o u n t i n g , a n d i m a g i n g . #text 5–10 larvae per genotype were bisected and inverted in PBS, and then transferred to a 1.5-mL tube containing 80 μg/ml TR-avidin (Invitrogen) in M3 insect medium (Sigma-Aldrich) containing 5% fetal calf serum, 1× insect medium supplement (Sigma-Aldrich) and penicillin/streptomycin antibiotics (Invitrogen). Carcasses were incubated for 15 min at room temperature with gentle agitation, rinsed two times and washed three times for 5 min with ice-cold PBS + 0.5% BSA at 4°C, and then fixed, washed, and mounted in FluoroGuard Reagent. Confocal images were acquired on a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) equipped with a digital camera (ORCA-ER; Hamamatsu) and a spinning disc confocal system (CARV; BD Biosciences). Axiovision v3.1 acquisition software, Plan-Apochromat 63×, 1.40 NA, and Plan-Neofluar 40×, 0.75 NA, objectives were also used (all from Carl Zeiss MicroImaging, Inc.). All other images were taken with a digital camera (DXM1200; Nikon) attached to an epifluorescence microscope (Axioscope 2; Carl Zeiss MicroImaging, Inc; ACT-1 acquisition software [Nikon], Plan-Neofluar 40×, 0.75 NA, 20×, 0.50 NA, and 5×, 0.15 NA, objectives were also used) or a digital camera (Coolpix 990; Nikon) attached to a dissecting microscope (Stemi 200C; Carl Zeiss MicroImaging, Inc.). All images were processed in Photoshop v7.0 (Adobe). Larvae were transferred to fresh medium supplemented with yeast paste and allowed to feed for 24 h before extraction. Extracts were prepared by homogenizing equal masses of wild-type or mutant larvae directly in 1× SDS sample buffer. Extracts were boiled, run on an 8% SDS-polyacrylamide gel, and transferred to PVDF membrane using the Mini PROTEAN 3 electrophoresis and wet transfer system (Bio-Rad Laboratories). Membranes were blocked in PBST containing 5% dry milk for 4 h before overnight incubation in primary antibody. Primary antibodies were diluted in PBST containing 5% BSA. The primary antibodies used were as follows: Akt, P-Ser505-dAkt, P-Thr398-dS6K (1:1,000; Cell Signaling Technology), dS6K (1:1,000, mouse monoclonal; gift from G. Thomas, University of Cincinnati, Cincinnati, OH), dTOR (1:1,000; gift from D. Pan, Johns Hopkins University, Baltimore, MD), tubulin DMIA (1:5,000; Calbiochem). Membranes were washed four times for 20 min in PBST and blocked for 1 h before 4 h incubation in HRP-conjugated secondary antibody. Membranes were washed four times for 20 min in PBST. SuperSignal ECL substrate solutions (Pierce) were applied to membranes to allow protein detection on BioMax light film (Kodak). Mitotic loss-of-function clones were induced at 3.5 d after laying through a 2-h heat-shock in a 37°C air incubator, and were allowed to grow for 45 h postinduction at room temperature. Overexpression clones were induced at 3.5 d after laying through a 1 h 15-min heat-shock, and were allowed to grow for 48 h at 25°C. Approximately 25 wing imaginal discs per genotype were dissected and processed for FACS analysis, essentially as previously described (). In brief, discs were added to 250 μL of PBS in 5 mL polystyrene round-bottom tubes, and kept on ice until 275 μL trypsin solution (9× Trypsin-EDTA [Sigma-Aldrich], 1× PBS, and 2 μM Hoechst 33342) was added to each tube to allow tissue dissociation into single cells. Samples were mixed on a Nutator (Clay Adams) at room temperature for a minimum of 2 h before flow cytometric analysis using a FACSAria (BD Biosciences). FlowJo v4.4 was used for data analysis. Fig. S1 displays additional effects of TOR overexpression in eye and wing tissues, interactions with additional alleles, and rescue by a genomic Hsc70-4 construct. Fig. S2 displays additional TR-avidin and Lsp2 internalization experiments. Fig. S3 shows Slimfast antibody specificity controls and additional Slimfast and insulin receptor localization data. Fig. S4 displays effects of on autophagy induction and cell size. Online supplemental material is available at .

The DnaB helicase is the main replicative helicase of eubacteria (). The enzyme sits on the 5′ single-stranded DNA (ssDNA) arm of the replication fork () and utilizes the hydrolysis of ATP to fuel the unwinding of double-stranded DNA (dsDNA). Its movement along the DNA proceeds in the 5′–3′ direction () and at 25°C it has an intrinsic rate of ∼ 291 bp per second (). At the replication fork, DnaB forms a critical interaction with the DnaG-primase that stimulates the activities of both the helicase and the primase (). Helicases assemble into a variety of oligomeric forms, ranging from monomers to heptamers and have been classified into several families based on analysis of their sequences (). All helicases share the Walker A and B motifs, whose most conserved residues are implicated in nucleotide binding and hydrolysis (). Specific helicase families are defined by the presence of additional specific motifs. Structural studies of helicases from many families have shown that all these proteins contain a core fold that was first visualized in the crystal structure of RecA (). DnaB belongs to a small family of hexameric helicases whose members function in replication. In addition to DnaB, this family includes the replicative helicases of the T4 and T7 bacteriophages as well as the RepA helicase of plasmid RSF1010. All members of this family contain by definition five conserved motifs (H1, H1a, H2, H3 and H4) that are located in the C-terminal region of the protein (). Motifs H1 and H2 are equivalent to the Walker A and B motifs. The N-terminal region of these helicases is less well conserved. Indeed, in the T7 helicase a domain related to DnaG primase is fused to the N-terminal portion of the helicase. Electron microscopy (EM) has revealed that DnaB family helicases form hexameric ring-shaped structures that exists in equilibrium among these different quaternary states (,), exhibiting either 6-fold rotational symmetry (C), 3-fold rotational symmetry (C) or an intermediate state between the two (CC). While these EM reconstructions predict a movement of the C-terminal region of DnaB, the transition from the C to the C particle is believed to be principally determined by the N-terminal region (,). Fluorescence transfer experiments have suggested that DnaB is positioned at the replication fork with the 5′ ssDNA arm inside the ring while the 3′ ssDNA arm is occluded from the ring, hence providing a mechanism for duplex separation (). Several crystal structures of the T7 gp4 helicase have been determined (). The presence of the same sequence motifs in the helicase domain of the T7 protein and the C-terminal region of DnaB would suggest that the structure of the DnaB C-terminal region resembles that of the T7 helicase domain. To date, the only structural information available for DnaB are the crystal () and NMR () structures of an N-terminal fragment. Here we present the 2.9 Å resolution crystal structure of the full-length DnaB helicase from the thermophile ( DnaB) that shows a larger, complete N-terminal domain exhibiting four different orientations relative to the C-terminal domain. The expression plasmid for DnaB () was transformed into BL21(DE3) cells (Novagen). The transformed cells were grown in LB medium containing 200 μg/ml ampicillin at 37°C. When the cells reached an A of between 0.4 and 0.8, gene expression was induced by addition of isopropyl-β--thiogalactopyranoside (Sigma) to a final concentration of 1 mM. After 3 h, the cells were harvested by centrifugation and quickly flash frozen in liquid nitrogen. The cell paste containing DnaB was resuspended in Buffer B1 containing 20 mM Tris pH 8.0, 50 mM NaCl, 5 mM MgCl and 10% (w/v) sucrose supplemented with complete EDTA-free protease inhibitors (Roche). The cells were lysed in a microfluidizer and then heated at 65°C for 20 min. Following centrifugation, the protein was run over a Q-sepharose column (GE Healthcare) that had been equilibrated with buffer B1. DnaB was eluted using a linear gradient from 50 mM to 1 M NaCl. The pooled fractions were then applied to a Hi-Trap Heparin HP column (GE Healthcare) and eluted in a linear gradient from 0.1 to 1.0 M NaCl. Finally, the protein was gel filtered over a 26/60 Superdex 200 prep grade column (GE Healthcare) that had been equilibrated in B2 buffer containing 10 mM Tris pH 8.0, 10 mM NaCl, 1 mM MgCl and 25 mM Maltose. DnaB factions were pooled and spin concentrated before being flash frozen in liquid nitrogen. The final protein was over 99% pure as judged by SDS-PAGE and coomasie blue staining. The Seleno-methionine (SeMet) substituted DnaB was expressed in B834(DE3) (Novagen) and purified using the same methods that had been used as the native protein with the exception that the reducing agent, dithiothreitol (Sigma), was added at all buffers. DnaB crystals were prepared using the sitting drop method of vapor diffusion, mixing equal volumes of protein sample and well solution. Well solutions contained 1.3 M ammonium sulfate, 10 mM magnesium sulfate, 5% 1,6-hexanediol and 100 mM MES pH 6.5. To provide cryo-protection, the crystals were transferred in 0.5 M increments into a solution of 3.0 M ammonium sulfate, 10 mM magnesium sulfate, 2.5% 1,6-hexanediol, 100 mM MES pH 6.5 and 10% ethylene glycol. Native diffraction data were collected at the National Synchrotron Light Source (NSLS) station X25. SeMet substituted crystals were used to collect multi-wavelength anomalous dispersion (MAD) data at the Argonne Photon Source (APS) station 19-ID. The HKL suite of programs was used for all data integration and scaling (). The crystals belong to space group P321 with unit cell dimensions = 104.7 Å, = 104.7 Å, = 363.3 Å and have a diffraction limit of 2.9 Å resolution where its / drops to 1.3. Data collection statistics are shown in . The positions of 47 Se atoms were determined using the direct methods procedure as implemented in the program SnB (). After MAD phasing using the program SOLVE (), the electron density was solvent flattened and each of the four N-terminal and C-terminal domains were 4-fold averaged with the program RESOLVE (). The maps were further improved by sharpening the data by applying a B factor of 50 to the experimental amplitudes. Cycles of model building in the programs O () and COOT () and refinement in REFMAC () resulted in a structure with an of 24.2% and of 28.5%. The current model contains 1641 out of 1776 amino acids across four molecules of DnaB. Residues 1, 175–182, 326–332 and 441–444 where disordered in the electron density maps and therefore are not included in any of the DnaB chains of the current model. Phasing and refinement statistics are given in and a portion of a sigma-A weighted 2 − electron density omit map can be seen in A. The full-length DnaB structure consists of two domains that are connected by a long flexible linker (A and B). The N-terminal domain or NTD (residues 1–150, 1-168; here and throughout the manuscript residue numbering in will have the prefix ) is an all α-helical structure containing a largely spherical bundle of helices terminated by an extended helical hairpin ( and A). The C-terminal domain or CTD (residues 186–444; 207-471) consists of a predominantly parallel β-sheet flanked by α-helices ( and A). The flexible linker of 34 residues (residues 151–185; 169-206) has an extended structure composed of a single α-helix flanked by two loops that connect the helix to the NTD and the CTD. DnaB does not form a closed hexameric ring quaternary structure in this crystal lattice as does the isolated molecule visualized by EM (,) nor does DnaB form a 6-fold filament, as observed in the original crystal structure of the T7 gp4 helicase domain (). Instead, the four copies of DnaB observed in the crystal asymmetric unit appear to be arranged as individual monomers or dimers (see below). Size-exclusion chromatography of DnaB indicates the protein is hexameric in the absence of nucleotide, even under high salt (1 M NaCl) conditions (Supplementary Figure S1). To determine whether the conditions used for crystallization perturbed the equilibrium between hexamer and monomer, we performed size-exclusion chromatography experiments under conditions mimicking, as closely as possible, those of crystallization. Thus, DnaB was gel filtered in a buffer containing 50 mM MES pH 6.0, 5% 1,6-hexanediol, 10 mM magnesium sulfate and 200 mM ammonium sulfate (Supplementary Figure S1). Using this approach, we were unable to assay higher ammonium sulfate concentrations as the helicase began to interact with the column matrix (data not shown). This gel filtration experiment clearly showed that, within the experimental detection limits, the helicase remains hexameric (Supplementary Figure S1). We therefore assume that crystals of the monomeric or dimeric helicase are formed from the equilibrium with the hexamer, which may or may not favor the monomer at high ammonium sulfate concentrations. Comparison of the positions of the NTD relative to the CTD in each of the four monomers reveals a highly flexible monomer structure. Alignment of the four DnaB molecules found in the asymmetric unit by superposition of their CTDs reveals that each of the linkers takes a different trajectory from the CTD and in turn each NTD adopts a different orientation with respect to the linker (). The flexible joints generated by the two linker region loops result in almost unrestrained positioning of the two domains with respect to one another. The organization of the NTD, linker region and CTD are highly reminiscent of the helicase domain, linker region and primase domain of the gp4 helicase-primase subunit structure (). This flexibility is consistent with previous hydrodynamic studies that predict an extended structure for the DnaB monomer (). Identification of the boundaries of the two DnaB domains has previously been based on its digestion with trypsin (), which cuts the protein into an N-terminal 12 kDa fragment (P12) and a C-terminal 33 kDa fragment (P33). Superposition of the coordinates of the DnaB NTD onto those of the P12 fragment reveals that the structures are highly similar with a root mean squared deviation (rmsd) between them of 1.2 Å over 97 Cα atoms. However, comparison of the two structures reveals that the NTD of intact DnaB is 40 residues larger than the P12 fragment. These additional residues of the DnaB NTD extend the C-terminal helix of P12, which then interacts with a further C-terminal helix to form a helical hairpin subdomain (A). This helical hairpin includes residues that have been shown by mutagenesis experiments to modulate the interaction of the helicase with the DnaG-primase (). Previously these residues had been ascribed to be part of the linker region. The four copies of the DnaB NTD in the asymmetric unit appear to form two independent dimers. Each dimer interface is formed by the helical hairpin subdomains that pack together to produce a four helix bundle (D). Short chain hydrophobic residues dominate the interface that buries 1250 Å of surface area per monomer. The two dimers are highly similar as their superposition results in an rmsd of 1.5 Å over 302 Cα atoms. A recent NMR structure determination of an DnaB fragment protein that included residues C-terminal to 171 (150 in ), which we now know corresponds very well to the NTD of the intact protein, revealed that the C-terminus of the isolated NTD does not form a helical hairpin as is observed in the full-length structure (). Instead this region was found to be unstructured. Although this could be due to inherent differences between the two species, this seems unlikely as secondary structure predictions indicate that this region within the protein should also be helical and form a hairpin (). More likely, the hairpin may be stabilized in the context of the hexamer assembly, possibly in the C3 symmetric particles by the formation of the NTD dimer observed here (D). Recent structures of the C-terminal helicase-binding domain (HBD) of the DnaG-primase have revealed that the fold of this domain is related to that of the P12 fragment of DnaB (). Comparison of the structure of the NTD in the intact DnaB with that of the NMR structure of the HBD () shows that this similarity extends to the additional helical hairpin (). Thus, the C-terminal HBD of DnaG primase and the NTD of DnaB helicase have related folds. The five motifs that define the dnaB family are all located within its CTD. Four of the motifs (H1, H1a, H2 and H3) form the core β-sheet and provide the residues that line the nucleotide-binding pocket (A). The fifth motif (H4) forms a helix, the N-terminus of a core β-strand as well as their connecting loop (A) and has been proposed to contain residues that are involved in DNA binding (). As expected the structure of the CTD of DnaB closely resembles that of the helicase domain of T7 gp4 (rmsd of 1.5 Å over 172 Cα atoms) and that of the RecA core (rmsd of 1.9 Å over 151 Cα atoms) (Supplementary Figure S2). Structural similarity is also observed between DnaB, a 5′–3′ helicase and the papillomavirus 3′–5′ E1 helicase. The CTDs of these two helicases superimpose with an rmsd of 3.0 Å over 102 Cα atoms. The proposal that the CTD of DnaB contains a leucine zipper dimerization motif is not supported by the crystal structure. Analysis of sequence of residues 361–390 in (338–368 in ), that lie upstream of a region rich in basic residues (Eco324-329, residues 299–304 in ), suggested that it forms a leucine zipper reminiscent of the eukaryotic transcription factors, such as cFOS and cJUN (). This sequence overlaps with the H3 motif sequence in the DnaB family. The basic region is highly conserved across DnaB proteins, although it is not found in the T7 gp4 helicase. Site-directed mutagenesis of the conserved basic residues of this region results in significant attenuation of DNA binding, ATPase activity and consequently the helicase activity of DnaB (). Inspection of the CTD fold reveals that the basic sequence RARARR (in ) is located on a helix that lies on the opposite face of the β-sheet from the nucleotide-binding pocket (A). However, the residues implicated in forming a leucine zipper, which is an all α-helical structure, in fact form a helix-loop-strand motif in the core of the RecA fold of the CTD (A). Despite attempts to prepare a crystalline complex of DnaB with its nucleotide substrate, either by soaking crystals or co-crystallizing in solutions containing ATP, ADP or non-hydrolyzable analogs of ATP, no binding of nucleotide was observed. We presume that nucleotide binding was inhibited by the high concentration of sulfate ions (at least 1.0 M) in the crystallization buffers. Indeed, a sulfate ion is bound to the Walker A motif of each monomer in the crystal, presumably mimicking a phosphate of ATP (A). Although the structure of a nucleotide complex has not been achieved as yet, the similarity between the structure of the CTD of DnaB and that of the T7 gp4 helicase domain allows us to use the structure of the nucleotide complex of T7 gp4 helicase (B) to homology model the ATP onto the DnaB structure (C). The resulting homology model is in excellent agreement with that of the T7 gp4 structure. Coordination of the modeled magnesium ion is provided by the side chains of residues T214 ( T238) and D317 ( D343). Residues R397 ( R420) and F430 ( F452) are ideally positioned to stack on the adenine ring of the modeled nucleotide and the γ-phosphate is in contact distance with residues K213 ( K237) and Q361 ( Q384). The T7 gp4 helicase ring is stabilized through interactions between its linker helix in one subunit and a small α-helical pocket, found on the periphery of the RecA core, in the adjacent subunit (,). Although it varies slightly in secondary structure (Supplementary Figure S3), DnaB also contains this hexamerization pocket (residues 240–283, 264-308). Within the lattice of the monomer crystal each linker helix is bound in the hexamerization pocket of an adjacent CTD but in such a way that does not form a closed hexameric ring. However, comparison of DnaB with the T7 gp4 ring structures (,) suggest that the linker region helix interface observed here will be similar in a closed hexameric assembly of DnaB (Supplementary Figure S3). A homology model of the hexameric arrangement of the DnaB CTD can be built by superimposing its CTD onto the hexameric ring structure of the T7 gp4 helicase (). The resulting model is free of any main chain steric clashes and the dimensions and shape are consistent with the EM reconstructions of DnaB from and the bacteriophage SPP1 (,). Structural and mutagenesis experiments with the T7 gp4 helicase have suggested residues within motif H4 and loops I, II and III are likely candidates for DNA binding (). The homology model of the DnaB CTD likewise positions the equivalent elements in DnaB at the center of the modeled ring, and this arrangement is consistent with fluorescence energy transfer experiments done using DnaB (). The highly basic region, whose mutagenesis has been shown to helicase activity (), is found on the N-terminal surface of the modeled hexamer. At the replication fork, the duplex DNA would be located at the C-terminal surface of the helicase ring and the ssDNA would emanate from the N-terminal surface (). The N-terminal surface would also be the location of the NTD in the intact protein. In order to model the locations of the NTDs within a hexamer, we attempted to manually fit the DnaB NTD coordinates into the published EM reconstructions. The reconstructions used where of the C and C symmetrized particles from DnaB () and the C, C, CC symmetrized particles from bacteriophage SPP1 G40P DnaB homolog (). However, no satisfactory placement of the NTD or the observed NTD dimer into any these reconstructions could be found. In each case, the NTD could not be fit within the EM volume without severe steric clashes. Difficulties in fitting the NTD into the low resolution EM reconstructions could arise from several sources. The NTD subdomains could adopt a different conformation with respect to one another in the context of a hexameric assembly than observed. Alternatively, the EM reconstructions could be distorted, either by the inherent difficulties in generating reconstructions of highly polymorphic molecules (such as DnaB) or from negative staining techniques which are known to cause compression of the specimen (,). The crystal structure of the full-length helicase reveals that DnaB is a highly flexible two domain molecule in the absence of the restraints imposed by the formation of the hexameric quaternary structure. We observe a potential dimer interface within the crystal lattice that is mediated by the helicase NTDs. Although dimers of DnaB have not been reported previously, both monomers and trimers have been detected by analytic ultracentrifugation in the absence of magnesium (). Models have also been proposed that implicate the dimerization of the NTD in the transition of the DnaB ring from 6-fold to 3-fold symmetry (,). However, more recent EM reconstructions have discounted dimerization of the NTD within the 3-fold symmetric ring (,). Therefore the significance, if any, of the dimer interface observed here awaits further study. The optimal activity of DnaB during replication is achieved through its interaction with the DnaG-primase. Mutagenesis studies suggest that the interaction interface between the two proteins is extensive and involves residues from the NTD and the CTD as well as their flexible linker (,). However, the structure of the intact protein reveals that the residues that had previously been assigned to the linker region form the C-terminal helical hairpin of the NTD. Several of the DnaB mutations that affect its interaction with primase lie in this hairpin structure () and to our knowledge none of the mutations studied previously lie within the reassigned linker region of the intact protein. Hence, the structure of DnaB in conjunction with the mutagenesis studies suggest that the primase interaction interface of the helicase is formed largely by the redefined NTD. The structure of the NTD of DnaB has a fold that is highly similar to the fold of the DnaG-primase C-terminal HBD. The HBD has been shown to be the only domain of the DnaG-primase that is required for interaction with and stimulation of the helicase () and the NTD of DnaB forms at least a large part of the primase interaction surface. In light of this, it is tempting to speculate that the DnaB NTD and the DnaG HBD could utilize their respective hairpins to interact in a manner similar to that of the NTD dimer observed in the crystal. However, this would require the energetically unfavorable burying of several charged residues from the HBD hairpin with hydrophobic residues from the DnaB NTD hairpin. Therefore, it seems highly unlikely that HBD will interact with DnaB in this manner. Interrogation of the protein structure database reveals that no other known protein structure shares the fold observed in the DnaB NTD and DnaG HBD. These results have lead to the suggestion that the last common ancestor of the DnaB family helicase was a bi-functional helicase-primase molecule. In this scenario the N-terminal domain of primase and C-terminal domain of helicase where separated by a linker, whose fold was related to that of the modern day NTD of DnaB. Duplication of the linker region followed by gene separation has presumably resulted in the current helicase and primase (). Very little is understood about the mechanism by which DnaB unwinds duplex DNA. Homology modeling of the DnaB CTD onto the structure of the T7 gp4 hexamer suggests that DnaB could in principle utilize the ‘binding change’ model of sequential nucleotide hydrolysis put forward on the basis of the T7 gp4 hexamer structure (). However, several differences exist between the DnaB and the T7 gp4 helicases that may imply differences in their respective mechanisms. DnaB contains an NTD that is not present in the T7 gp4 helicase and the helical bundle moiety of this domain has recently been shown to modulate the polarity of the DnaB helicase. DnaB truncation mutants that lack this helical bundle have been shown to poses 3′–5′ helicase activity (). Kinetic studies also suggest that DnaB has three high affinity binding sites for ATP (), which is consistent with the 3-fold symmetry observed by EM (,). However, the model based on the T7 helicase crystal structure predicts four ATP-binding sites and 3-fold symmetric molecules of the T7 helicase have not been detected by EM. The DnaB helicase and the papillomavirus E1 helicase share several common features; both are ring-shaped hexameric helicases that unwind genomic DNA at replication forks and both are formed by an all α-helical NTD (although the structures of the two NTD folds are unrelated) and a CTD that contains a core RecA-like fold (). However, these proteins have distinct biochemical properties as DnaB unwinds DNA in the 5′–3′ direction where as the E1 helicase unwinds DNA in the 3′–5′ direction () and furthermore, these two proteins are structurally distinct. Comparisons of their respective RecA-like folds shows that although the Walker A and Walker B motifs (motifs H1 and H2 in the DnaB family) are similarly positioned, the DNA-binding elements are located in different regions of the fold, shown schematically in A. But, in the quaternary structures, the DNA-binding elements are still found at the center of the ring of each helicase. This is possible because both helicases form hexamers in distinct manners; the plane of the core β-sheet of the E1 helicase are found approximately parallel to the ring axis whereas in DnaB-family helicases, the plane of the core β-sheet is approximately perpendicular to the ring axis (B). A mechanism for ssDNA translocation by the E1 helicase has been proposed based on the crystal structure of the E1 helicase bound to ssDNA (). In this structure, the DNA-binding loops from each subunit of the hexamer form a spiral, with each loop bound to one nucleotide of the ssDNA. The mechanism by which the helicase translocates along the DNA has been proposed to involve each DNA-binding loop carrying a nucleotide of ssDNA across the ring in response to ATP binding and hydrolysis (). A similar spiral method of translocation is attractive for DnaB, and indeed the T7 gp4 helicase structure also displays a spiral conformation of its subunits, suggesting this spiraling mechanism may be common to all hexameric helicases. However, our analysis of the structure of the E1 hexamer and that of DnaB implies that the way in which DnaB and the E1 helicase utilize ATP binding and hydrolysis to move the DNA-binding elements are very likely to differ. The high degree of structural similarity observed between DnaB and RecA supports the previous hypothesis that DnaB originated from a gene duplication of a RecA-like ancestor (). Sequence analysis and the structural comparisons of DnaB and the E1 helicase suggest this event occurred after the divergence of eubacteria from archaea and eukaryotes. The replicative DnaB helicase of bacteria and the replicative helicases of eukaryotes e.g. the E1 helicase and related MCM proteins (), are clearly distinct and thus appear to have evolved separately from a RecA-like ancestor, which was most likely not a helicase. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The occurrence of hyperlipidemia, hyperglycemia, insulin resistance and its metabolic complications such as type-2 diabetes mellitus (T2DM) increases dramatically in the western world. A deeper understanding of the pathogenesis causing these diseases and development of drugs targeting metabolic disorders currently has high priority. Nuclear receptors (NRs), including liver X receptors (LXRs), have been suggested as potential drug targets for the treatment or prevention of T2DM (). LXRα and LXRβ are established regulators of cholesterol and lipid metabolism and activation of LXRs promotes conversion of cholesterol to bile acids, lipid/triglyceride biosynthesis and reverse cholesterol transport from peripheral cells to the liver and subsequent elimination of cholesterol via the gall bladder [reviewed in ()]. A large body of literature establishes an important physiological role of LXR in carbohydrate metabolism. The carbohydrate-response element-binding protein (ChREBP) mediates glucose activated lipogenesis via the xylulose 5-phosphate pathway () and has been identified as an LXR target gene (). Recently, glucose itself was shown to be an LXR agonist activating LXRs at physiological concentrations (). Activation of LXR promoted glucose uptake and glucose oxidation in muscle (). As skeletal muscle constitutes 40% of the human body weight and is the major site for glucose utilization, this observation suggests that LXR might have a considerable impact on overall glucose oxidation in the body. Expression of the insulin responsive glucose transporter GLUT4 in adipocytes was induced by LXR while the basal expression of GLUT4 was lower in LXRα mice compared to wild type mice (,). Increased glucose uptake in adipocytes and muscle cells as well as reduced hepatic gluconeogenesis due to suppressed expression of gluconeogenic genes including PEPCK, G6P and PGC1α were observed in response to treatment with an LXR agonist (,,). Moreover, activation of LXR increased glucose dependent insulin secretion from pancreatic β-cell line cultures () and lead to increased plasma insulin concentrations in mice (). It was also shown that LXRβ mice have less basal insulin levels and, on a normal diet, are glucose intolerant due to impaired glucose-induced insulin secretion (). LXR signaling seems more prominent in disease where, for instance, impaired lipid oxidation was seen in isolated muscle cells from T2DM patients compared to control cells when the muscle cells were treated with an LXR agonist (). Further, improved glucose tolerance was observed in obese C57Bl/6 mice in response to treatment with an LXR agonist, but not in lean C57Bl/6 mice () and similar results were observed in db/db mice, Zucker diabetic and obese rats and ob/ob mice (,,). Improved whole body insulin sensitivity was observed in ob/ob mice upon activation of LXRs, but not in lean mice (). Together, these observations suggest an anti-diabetic role of LXRs. Elk1 is a well-studied member of the ETS family of transcription factors. Elk1 activity is tightly regulated by phosphorylation and dephosphorylation which have been extensively studied in the context of cellular signaling. Elk1 has been shown to be positively regulated by activation of the MAPK pathway including Erk1/2, p38 and JNK, which has been shown to be dysfunctional in T2DM (,). Here we identify a 5′-ETS site and a 3′-Elk1 binding site in the human LXRB gene promoter and show that Elk1 can bind both sites while SRF only binds to the 3′-Elk1 site. We show that binding of SRF and Elk1 to the identified binding sites is important for LXRB transcription. Furthermore, we report that glucose significantly induces transcription via the LXRB gene promoter and that the identified binding sites are important for proper glucose responsiveness. The LXRB gene specific primers 5′-CGGCCTCTCGCGGAGTGAACTACTCCTGTT-3′ and nested 5′-AGGCTGAGCTGGCCTCATCAGTGCCTGGGA -3′ were used to amplify 5′-transcript from full-length cDNA from human testis, ovary and thymus using Marathon ready cDNA kits (Clontech, Mountain View, CA, USA) with the Expand Long Template PCR System (Boehringer Mannheim, Mannheim, Germany) according to the manufacturer's instructions. The PCR products were cloned into the pGEM-T easy vector (Invitrogen, Carlsbad, CA, USA), and the identity of cloned products determined by DNA sequencing. The pcDNA-Elk1 plasmid was generously provided by Dr Robert Hipskind (Institut deGénétique Moléculaire de Montpellier, FRANCE). The SRF plasmids were a gift from Dr Eric Olson (UT Southwestern Medical Center at Dallas, USA). PCR fragments of the human LXRB gene promoter were cloned into the pGL3-Basic luciferase reporter vector (Promega, Madison, WI, USA) using the I and I sites with forward primers (−3839) 5′-ATCACTTTTACCTCATTTAGTCATAAGAGTAAGGCAACAAGGTCA-3′, (−1673) 5′ATCAAAAACAGCATATGCAGTAAAGAAGTCAGCCAGATCCCAGCA-3′ and (−245) 5′-ATCAGGTACCGGCCGCAGGCTCAGAGAAGCGCATGAATGAGCTAA-3′ and reverse (+1163) 5′-ATCACTCGAGGGTGGGGTCACGGAGCAGCCTGTAGAATACAGGGGATTGAGAG-3′ with the restriction enzyme sites underlined. All mutations were introduced using the QuickChange™ XL Site-Directed Mutagenesis Kit (Stratagene, La Jolla, CA, USA). The -245/+1163 construct was further mutated to destroy the putative Ets binding site using primers 5′- GATCTACCCGGTAAACTTTTGGTGAGTTTCCAACTTCCG-3′ and the corresponding reverse compliment. The Elk1 binding site was mutated using 5′-GGCAGCAGCTTCGGCTGGTCCTAAGCGGTTTTTTTGTTCGTCAAGTTTCACGCTCCGCCCCTCTTCCGG-3′ and the reverse compliment primers. DNA sequencing confirmed the identity of all clones. The mouse MIN6 insulinoma cell line was maintained in Dulbecco's modified Eagle's medium (DMEM, 4.5 g/l glucose), (GIBCO-BRL cat no. 41965-039), and the rat INS1E insulinoma cell line was maintained in RPMI 1640, including -glutamine and 11.1 mM glucose, (GIBCO-BRL, cat no. 21875-034). Media were supplemented with fetal bovine serum (INS1E: 10%, MIN6: 15%), 50 μM β-mercaptoethanol and penicillin/streptomycin at a final concentration of 100 U/ml and 100 μg/ml, respectively. MIN6 medium also contained 2 mM L-glutamine while 10 mM HEPES and 1 mM Sodium Puryvate were added to INS1E medium. For serum and glucose starvation, INS1E cells were grown in plain RPMI 1640 containing no serum or glucose. Cells were grown under 5% CO at 37°C. Total 4 × 10 MIN6 and 25 × 10 INS1E cells were seeded in 24-well plates and transiently transfected using Lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's protocol. Each well received 125 ng of reporter vector and 500 ng of expression vector. Empty vehicle vector was added to ensure equal amounts of DNA in each transfection. Cells were transfected for 24 h and thereafter lysed in 25 mM TAE, 1 mM EDTA, 10% glycerol, 1% Triton X-100 and 2 mM DTT. Luciferase activities were measured using a Luciferase Assay Kit (BioThema, Umeå, Sweden) in a luminometer (Luminoscan Ascent, Thermo electron Corporation, Waltham, MA, USA). Cells were grown in 24-well plates, washed with PBS and incubated in TEN buffer (40 mM Tris-HCl, 1 mM EDTA, 150 mM NaCl) for 4 min. Cells were mechanistically removed with a cell scraper and pelleted by centrifugation at 3500 r.p.m. for 2 min at 4°C. Cell pellets were freeze dried on dry ice and resuspended in 50 μl ice-cold buffer C (10 mM HEPES-KOH pH 7.9, 0.4 M NaCl, 0.1 mM EDTA, 5% glycerol, 1 mM DTT, 0.5 mM PMSF). After another round of freeze drying, cell debris was removed by centrifugation for 5 min at 13 000 r.p.m. at 4°C. The supernatant corresponds to whole cell extracts. WT and mutated (‘Mut’) oligos (mutated nucleotides underlined) were; ETS1:5′-GATCTACCCGGTAAACTTCCGGTGAGTTT-3′, Elk1:5′-GGTCCTAAGCGGACCGGAAGTTCGTCAAGTTTCA -3′, Mut ETS1:5′- GATCTACCCGGTAAACTTGGTGAGTTT -3′ and Mut Elk1:5′-GGTCCTAAGCGGGTTCGTCAAGTTTCA-3′. Five microgram of the respective forward and reverse oligos were annealed in 20 mM Tris-HCl pH 7.8, 2 mM MgCl, 50 mM NaCl by heating to 95°C for 5 min and slow cooling by 1.5°C/min for 47 cycles. Oligonucleotide probes were labeled by mixing 0.2 μg annealed oligo with 250 μM non-radioactive dATP, dGTP, dTTP, respectively, 1× Klenow buffer, 20 μCi P labeled dCTP (GE Lifesciences, Piscataway, NJ, USA) and 1 Unit Klenow polymerase. Samples were incubated for 20 min at room temperature (RT) and the reactions terminated by adding 0.5 M EDTA. Probes were purified using G-25 Nick Columns (GE Lifesciences) and the efficiency of labeling determined using the 1214 Rackbeta liquid scintillation counter (LKB Wallac, Markham, Ontario, Canada). For binding reactions, 2 μg of whole cell extracts were incubated with 4 × 10 c.p.m. of radiolabeled oligonucleotide in binding buffer pH 8.0 (10 mM Tris-HCl, 1 mM DTT, 1 mM EDTA, 50 mM KCl, 0.3% BSA, 5% glycerol) including 1 μg poly(dI/dC) and 1× Proteinase Inhibitor Cocktail (PIC). One microgram DNA template was translated in a 50 μl reaction using the TNT® Coupled Reticulocyte Lysate Systems (Promega). From this, 5 μl was used in EMSA binding reactions. Binding reactions were incubated for 20 min at RT and protein–DNA interactions separated by electrophoresis at 240 V for 4 h at 4°C using 8% polyacrylamide gels. The gels were dried and analyzed by autoradiography. In supershift assays, 1 μg of the respective antibodies were added prior to the addition of WCE or IVT protein. INS1E cells were transfected with the LXRB gene promoter containing reporter vectors and expression vectors for 24 h and protein–DNA were crosslinked using 1% formaldehyde for 20 min at RT. Cells were washed and harvested in cold 1× PBS and pelleted. The pellet was resuspended and incubated in cold RIPA buffer (50 mM Tris pH 8.0, 1 mM EDTA, 0.5 mM EGTA pH 8.0, 1% Triton X100, 0.1% Na deoxycholate, 140 mM NaCl and 1 × PIC for 10 min). DNA was sheared by sonication, centrifuged for 10 min at 13 000 r.p.m. at 4°C and the supernatant incubated with 20 μl protein A/G sepharose/agarose (50% slurry in RIPA buffer) on a rotating wheel for 2 h at 4°C. Fifty microliter of the supernatant was immunoprecipitated with 25 μg salmon sperm DNA, 100 μg BSA and 10 μg of Elk1 antibody in RIPA buffer at 4°C on a rotating wheel over night. Twenty-five microliter protein A/G slurry was added and incubated for an additional 1 h. The samples were centrifuged at 5000 r.p.m. for 2 min, the precipitates washed twice with 1 ml TSE I (1% Triton X100, 2 mM EDTA, 20 mM Tris pH 8, 150 mM NaCl), once with LiCl buffer (20 mM Tris-HCl pH 8, 1 mM EDTA, 250 mM LiCl, 1% NP40, 1% Na deoxycholate) and twice with TE buffer pH 7.4 and the protein–DNA complexes were eluted with 100 μl freshly prepared 1% SDS/TE by incubation for 30 min on a rotating wheel at RT following 65°C over night. For input control, 10% of saved samples were treated similarly to the immunoprecipitated samples. Supernatants were purified using QIAQUICK columns (QIAGEN, Hilden, Gemany). Five microliter of the elution was used in each real-time qPCR reaction using primers covering conserved sites in the rat promoter (Forward:5′-AGGCATCTCATTCGGTGGC-3′ and Reverse:5′-GGAAAGGTGACAGACTTCCGG) or the human promoter (Forward:5′-CCGGAAGTTCGTCAAGTTTCA and Reverse:5′-TTGCGTCACGTCCGGAA). Total RNA was prepared from cells using the RNeasy mini kit (QIAGEN) according to the manufacturer's instructions. Here, 0.5 μg total RNA was reverse transcribed into cDNA using SuperscriptII and random hexamer primers (Invitrogen). The concentration and quality of the purified total RNA were determined spectrophotometrically at OD nm and by the OD ratio, respectively. mRNA expression levels were quantified using the ABI 7500 instrument and the SYBR green technology (Applied Biosystems, Foster City, CA, USA). All primers were designed with the Primer Express® Software version 2.0, a program specifically provided for primer design using ABI qPCR instruments. Hundred nanomolar of SYBR green assay primers were used and for each primer pair a dissociation curve analysis was carried out to ensure the specificity of the qPCR amplification. All primer pairs were designed over exon–exon boundaries. All real time qPCR reactions were performed in triplicates. We calculated relative changes employing the comparative C method using 18S as the internal reference gene. INS1E cells were transfected for 4 days with mouse siElk1 and siSRF (both SMRT pool) oligos (Dharmacon, Lafayette, CO, USA) using DharmaFECT™ buffer 4 (Dharmacon) according to the manufacturer's instructions. Importantly, R&D at Dharmacon confirmed that the oligo sequences used in the mouse SMRT pools for Elk1 and SRF matched the rat sequence as well. Non-targeting control (D-001210-01), siLuciferase and siGAPDH were used as controls at corresponding concentrations. After incubation, cells were either used for WCE extraction for western blot analysis or used for RNA preparation and subsequent real-time qPCR analysis of knockdown. Rapid amplification of 5′-cDNA ends (5′-RACE) was performed using different tissue libraries to identify transcription start sites and, consequently, the proximal promoter region of the human LXRB gene promoter. No exact transcriptional start site was observed, rather transcription was initiated within a confined region of the promoter, in keeping with observations from other TATA-less promoters and previous observations for the mouse gene promoter (). We designated the most 5′-transcription start site observed as +1 (). The genomic sequences from mouse, rat, dog and cow were aligned with the corresponding identified proximal promoter region of human LXRB. Using a theoretical transcription factor binding site search [Transcription Element Search System (TESS); ] two highly conserved binding sites were identified, Elk1 and ETS (A). The ETS site is located 5′ of the Elk1 site in the LXRB gene promoter (B). Next, we used EMSA to analyze protein–DNA interactions at the identified binding sites using independent DNA oligos covering these sites depicted in B. Bands representing protein–DNA interactions at both the wild type Elk1 and ETS binding sites were observed using whole cell extract (WCE) and translated (IVT) Elk1 protein (A, lanes 1–5 and 3C, lanes 1–3) and the interactions were abolished when these binding sites were mutated (A, lanes 6–10 and 3C, lanes 4–6). The IVT Elk1 interactions were supershifted using a specific Elk1 antibody or an HA antibody (Elk1 cDNA was HA-tagged), but no supershift was observed using an antibody directed against the transcription factor C/EBPβ indicating a specific binding of Elk1 to this site (B, lanes 1–6). WCE yielded a complex which migrated more slowly compared to the pure IVT Elk1 protein indicating that additional proteins forming larger complexes were responsible for the interaction observed using WCE. A, lanes 1–4 show that IVT Elk1 and SRF bind to the wild type Elk1 binding site, although the Elk1 interaction seems to be stronger. Both proteins were equally expressed in our transcription/translation system (C) suggesting that this is not due to a molar difference for the two proteins, rather, this might simply be due to the composition of the EMSA binding buffer used, favoring Elk1 binding. Both SRF and Elk1 binding was abolished when the Elk1 binding site was mutated (lanes 5–6). Combining both IVT Elk1 and SRF yielded two bands of smaller size than observed using WCE but of the same size as when IVT Elk1 and SRF were used separately (B, lanes 1–4) suggesting that translated Elk1 and SRF do not by themselves form the same complex as seen in WCE. SRF did not interact with the ETS binding site (data not shown). Furthermore, we performed ChIP assays in the INS1 cell line to analyze the interaction of Elk1 and SRF on the transfected human LXRB proximal promoter and the native rat promoter in the INS1 cell line using a non-specific IgG antibody as control. The LXRB gene promoter was transiently transfected into INS1 cells before crosslinking of DNA and proteins. Elk1 was enriched at the identified binding sites and the enrichment was strongly increased upon overexpression of Elk1 before crosslinking (A). Similar results were seen on the endogenous rat gene promoter where endogenous Elk1 was found to be enriched (C and scaled up in the inserted frame) and this enrichment was strongly enhanced upon overexpression of Elk1. No enrichment of Elk1 was seen when the LXRB gene promoter with mutated binding sites for Elk1 and ETS was transiently transfected (B). This indicates that Elk1 is associated with its binding site at the endogenous promoter. No enrichment was seen using primers amplifying the luciferase gene (used as control for the overexpressed reporter gene experiment) or primers amplifying an exon in the gene (used as control for the rat native promoter experiments) (data not shown) indicating that the enrichment is specific for the identified binding sites SRE. Unfortunately, we could not get any of the antibodies directed against SRF to work in the ChIP assay. Next we investigated the effect of knocking down Elk1 and SRF in the INS1 cell line. A significant knockdown of either Elk1 or SRF was observed with siRNA targeting Elk1 or SRF but not with unrelated siRNA used as controls. The efficacy of siRNA knockdown was anlyzed at the RNA level using qPCR for Elk1 (A) and at the protein level using western analysis for SRF (B); in the latter case β-actin was used as a control. No cytotoxicity was observed even at 500 nM siRNA (data not shown). Using WCE from the INS1 cell line after transfection with siRNA targeting either Elk1 or SRF almost completely abolished binding to the Elk1 site in the LXRB gene promoter (C, lanes 4, 5, 9 and 10) while control siRNAs did not affect the protein–DNA interaction at the Elk1 site (lanes 1, 2, 3, 6, 7 and 8). These results suggest that both Elk1 and SRF must be present for adequate transcription factor complex formation at the binding sites in the LXRB gene promoter. In order to characterize the importance of the identified transcription factor binding sites for transcription of the LXRB gene we cloned the −245 to +1163 LXRB gene regulatory region in front of the luciferase reporter gene. We knocked down expression of Elk1 and SRF in the INS1 cell line by siRNA targeting Elk1, SRF or both and then transiently transfected the −245/+1163 LXRB gene promoter construct. A significant reduction in promoter activity was observed when the levels of Elk1 and SRF were reduced (A). Second, the Elk1 site, the ETS site or both sites (the same mutations were used here as the ones which showed abolished binding in and ) were mutated in the −245/+1163 LXRB promoter constructs and transiently transfected into the INS1 cell line. The individually mutated Elk1 or ETS sites reduced promoter activity by 50–60% while the activity was reduced by 90% in the double mutation (B), suggesting that both binding sites are necessary for full activity and that both Elk1 and SRF induce transcription via these binding sites in the LXRB gene promoter. A strong induction in expression of known LXR target genes including the ATP-binding cassette (ABC) transmembrane cholesterol and lipid transporters (ABCA1 and ABCG1) and the lipogenic sterol regulatory element-binding protein 1c (SREBP1c) transcription factor was observed when INS1 cells were treated with an LXR agonist (A) indicating that LXR signaling in these cells is working properly. Furthermore, the INS1 cells showed the expected induction in expression of pyruvate kinase upon treatment with increasing concentrations of glucose () (B) while the endogenous expression of LXRβ was not affected by glucose treatment (C). Neither was the endogenous expression of LXRβ using primary pancreatic β-cells from rat affected with glucose treatment (data not shown). LXRB gene promoter constructs with wild type or mutated Elk1, ETS or mutations of both sites were transiently transfected into INS1 cells and the cells were treated with increasing concentrations of glucose (). As expected and in keeping with B, the basal activities of the mutated constructs were reduced. Surprisingly, a significant concentration dependent induction of the wild type promoter activity was observed with glucose treatment whereas the double mutated Elk1 and ETS construct showed reduced response to glucose. This shows that glucose significantly induces transcription via the wild type LXRB gene promoter. Furthermore, both the ETS site and the Elk1 site are involved in proper glucose response as the activity of the double mutant construct at 20 mM glucose only showed an activity similar to that of the WT promoter under starved conditions. The effect of glucose on the promoter was surprising since endogenous expression of LXRβ in the INS1 cell line was not affected by increasing concentrations of glucose (C). Therefore, we cloned larger 5′-regions of the human LXRB gene promoter as indicated in to look for regions which might cause repression of transcription. Transient transfections of equimolar amounts of the promoter–reporter vectors showed that inclusion of upstream regions significantly reduced the activity of the promoter. The −3839/+1163 construct still mediated a glucose response, but this response was markedly reduced compared to that seen with the −245/+1163 construct (data not shown). We speculate that these repressive elements and others located outside the fragments we have cloned suppress endogenous activation by glucose. Thus, additional cell signaling pathways could be responsible for targeting these repressive functions which apparently overrun the stimulating effects of glucose on the LXRB promoter. Together with the SRF, the ETS family of transcription factors is known to form a transcription complex which can bind the serum response element (SRE). An SRE normally consists of a 5′-binding site which can recruit members of Elk1 and ETS transcription factor family and a 3′-binding site which recruits SRF. SRF binds to the 3′-binding site and associates with a member of the Elk1-ETS-family of transcription factors at the 5′-binding site and consequently the transcription complex occupies both binding sites (). The identified 3′- binding site in the LXRB gene promoter was not a consensus SRF binding site as reported in the literature (). Thus, our identified binding sites do not represent a classical SRE. Nevertheless, in this study we identify two binding sites in the human LXRB gene promoter which are highly conserved between species. Using mammalian cellular systems our results indicate that the ETS site and the Elk1 site are involved in increasing transcription via the LXRB gene promoter. Interestingly, we also show that the promoter is strongly responsive to glucose, partially through mediation by these binding sites in the promoter. Mutating both the Elk1 and ETS binding sites in the LXRB, gene promoter strongly reduced its activity and suppressed its response to glucose indicating that these sites are important for both basal promoter activity and promoter responses to glucose. However, the mutations did not completely abolish the glucose response, indicating involvement of additional transcriptional regulatory factors. Both the INS1 and MIN6 cell lines were analyzed for changes in endogenous levels of LXRβ after treatment of glucose. Both overexpression of Elk1 and/or SRF as well as siRNA targeting both factors was performed in presence or absence of glucose. Surprisingly, none of these treatments had any effect on the endogenous expression of LXRβ. Neither did glucose treatment lead to any changes in recruitment of Elk1 or SRF to the endogenous rat LXRβ promoter in the INS1 cells or the transfected human LXRβ promoter as analyzed by ChIP (data not shown). Immortal cell lines do not always reflect responses in normal cells from where the immortal cells originate. Interestingly, however, neither was any effect of glucose treatment seen on LXRβ expression in primary pancreatic β-cells from rat (data not shown). Therefore, it cannot be excluded that a glucose responsive human LXRB promoter is confined to human cells since these cells have the necessary transcriptional network for the response in question. Unfortunately, human primary pancreatic β-cells are very difficult to obtain so we could not test this notion in a human primary cell system. However, the identified binding sites are highly conserved between species from mouse, rat, cow, dog and human () and, therefore, we do not expect to see any species specific effect using primary islets from human. This is also supported by our observation that Elk1 is enriched at the endogenous gene promoter in rat (B). Rather, we speculate that more complex cell signaling mechanisms including additional transcription factors, which modify the chromatin structure on the native promoter, are necessary for endogenous glucose response. The biological effects of glucose, directly affecting transcriptional regulation of target genes (for instance via ChREBP) or via insulin-mediated signaling, are pivotal for overall energy homeostasis. Therefore, it is conceivable that the glucose-activated transcriptional regulatory pathways are under strict and complex control. Treatment of the MIN6 insulinoma cell line with glucose activates Elk1 by phosphorylation of the Ser residue which could lead to induced expression of Elk1 target genes (). We show that Elk1 with the phosphorylation sites Ser, Ser and Ser mutated to alanine interacts with its binding site in the LXRB gene promoter (A), but has weaker effects on transcriptional regulation of the LXRB gene promoter (data not shown). Thus, phosphorylation of Elk1 could alter the effect of Elk1 on transcription. Furthermore, insulin has been shown to phosphorylate Elk1, thereby inducing its transcriptional activity (). Accordingly, several important signaling events in response to metabolic processes may influence Elk1 activity and potentially alter the expression of LXRβ. Thus, the regulation of Elk1 activity is important for its effect on the targeted promoter. Multiple signaling cascades involved in cell growth and proliferation including the mitogen-activated protein kinases (MAPKs) have been identified as activators of the SRF/TCF complex and factors that form these complexes (). These signals can work via SRF/TCF to regulate gene expression. For instance, a dominant negative Elk1 was shown to inhibit cell proliferation and induce apoptotic cell death () and the TCF is a docking site for the pro-proliferative Wnt signaling cascade via β-catenin, which is known to interact with several members of the NR family and regulate cell proliferation events (). LXRs were recently shown to induce growth arrest and promote apoptosis in the INS1 insulinoma cell line () and several additional studies report that LXRs mediate anti-proliferative effects (). Hence, cell signaling cascades targeting TCF might also elicit anti-proliferative effects by enhancing expression of LXRβ. The role(s) of LXRs in various aspects of glucose metabolism and as mediators of biological effects of glucose render LXRs highly interesting to study in the attempts to define molecular mechanisms behind insulin resistance and diabetes.

Alternative pre-mRNA splicing is a critical mechanism for regulating gene expression in metazoan organisms, and leads to tremendous protein diversity from a relatively small number of genes. A majority of human genes exhibit some form of alternative splicing. In particular, the human genome encodes a complex alternative splicing program that switches alternative exons on and off according to the needs of individual differentiated cell types. Despite intensive study in recent years, the mechanisms regulating the human alternative splicing program are not yet well understood. The complex decision process, involving which subset of exons on the primary RNA transcript (henceforth, pre-mRNA) will get spliced into the mature mRNA isoform, is mediated by a combination of -regulatory elements organized across exons and introns (), quite analogous to the -regulation of transcription. Global identification of splicing regulatory elements has been difficult and has been primarily restricted to exonic elements (), while limited computational information is available on intronic elements (). However, availability of splicing microarrays (), which can interrogate expression levels of exons genome-wide under any particular biological condition, has opened up new possibilities. In this work, we demonstrate that one can now apply analogous computational approaches used for dissecting transcriptional regulation () to decipher the splicing regulatory elements, with genes replaced by exons and promoters by pre-mRNA regions proximal to the splice sites. A new set of approaches based on correlation with expression has been particularly successful in identifying -regulatory elements governing transcription (). Here, the premise is that gene expression results from integration of multiple signals within the promoter region, as mediated by binding of -factors to the -elements. This implies that for an active -regulatory motif, its parameters [occurrence frequencies and position weight matrix (PWM) scores] must be significantly correlated with the expression levels across genes under any specific biological condition. Multiple studies have eseished that, using this strategy, one can identify the motifs that are functional under the tested condition. Furthermore, expression data from a single test condition and a reference condition are often sufficient for the analysis. In addition, unlike clustering-based approaches, interacting combinations of motifs can be inferred with high confidence (,). Finally, a recent study based on linear splines, which model the sigmoidal nature of transcriptional response, shows that such approaches can accurately identify direct targets of -factors binding to the active motifs, even when the motifs are very degenerate (). Target identification in such situations has been quite challenging. Thus, one can delineate the key elements of transcriptional regulatory networks using correlation with expression. This has proven effective in both lower eukaryotes, e.g. yeast (,,), and in mammals (). Here we report the first application, to our knowledge, of the correlation with expression approach for identification of -elements that regulate alternative splicing by integrating pre-mRNA sequence information with the exon microarray data. Specifically, we focused on tissue-specific splicing, as tissue-specific pre-mRNA regions are largely conserved across species (,,), and thus, phylogenetic conservation can be used to evaluate the predictions. We employed an Affymetrix exon microarray () to identify 56 muscle-enriched alternative cassette exons, a number of which are predicted to alter the expression of cytoskeletal related genes. We used both linear regression () and linear splines () to examine whether -elements in introns adjoining these exons correlate with gene-normalized exon expression in muscle. Multiple motifs that demonstrated statistically significant correlation were also found to be conserved in mouse, chicken and frog. In addition, several of these elements have been previously characterized experimentally as regulators of muscle-specific splicing via binding to members of the Fox (), CELF () and PTB () families of splicing factors. Taken together, our study shows that correlation with expression is indeed effective in deciphering splicing regulatory elements, and provides the most comprehensive picture yet available of muscle-specific alternative splicing program in humans. Total RNA from three biological replicates (three separate individuals) of 16 normal adult human tissues was purchased from BioChain (Hayward, CA, USA). Labeled target was generated from ∼200 ng of total RNA and hybridized to a prototype version of the Affymetrix Human Exon Array as described (). The set of microarrays contain ∼1.4 million probesets designed to interrogate, as comprehensively as possible, more than 1 million exon clusters derived from a variety of input sources including annotated genes, cDNA sequences and exon prediction algorithms. Design information and microarray data is available at the GEO database (; accession number: GSE5791). Candidate muscle-enriched probesets were identified using the splicing index approach (,,). Exon-level expression was normalized to the expression level of the parent gene by dividing probeset intensities by the median intensity of probesets from exons supported by RefSeq or Ensembl annotations. Exons that exhibited statistically significant differences in inclusion rate were identified using a student's -test on the gene-level normalized probeset intensities (NI). NI values from the three biological replicates of heart and skeletal muscle tissues were compared as a group to the replicates of 14 other non-muscle tissues as a second group. The magnitude of inclusion rate change (splicing index) was estimated by calculating a log ratio (base 2) of the median muscle NI and the median non-muscle NI (,). After filtering out non-expressed probesets and genes with low expression, probesets with -test -values <0.001 and splicing index magnitudes of >0.5 were considered candidates for muscle-enriched exons. Manual filtering of the initial list was performed to select further for high confidence internal cassette exons, by mapping candidate muscle-enriched probeset to their genomic context using the BLAT tool () at the UCSC genome browser (). Probesets that overlapped annotated alternative transcriptional starts, alternative polyadenylation sites, or regions with alternative 5′ or 3′ splice sites, were removed from consideration in this study. Exon-level probeset intensities were additionally observed using BLIS (Biotique Systems, Inc. Reno, NV), an integrated genome browser that enables exon expression data from the microarray to be viewed in genomic context. Only probesets that showed clear patterns of muscle enrichment were kept for further analysis. Probesets had to demonstrate higher intensity levels in the muscle tissues and have exon-level data for surrounding probesets consistent with exon skipping in a majority of non-muscle tissues. Probesets were subsequently mapped to the May 2004 human genome (NCBI Build 35) using BLAT (). Exact exon boundaries were determined by comparison to EST and mRNA sequences requiring consensus splice sites. For phylogenetic analysis, the orthologous exons were identified in another mammalian genome (mouse; ), in an avian genome (chicken; ) and in an amphibian genome (frog; ) using VISTA alignment tools. Automatic alignment was successful at finding most of the longer alternative exons directly, but in a few cases the alignments were adjusted manually. The upstream 200 nt (U200) intronic region was selected as the base 1 to base 200 adjoining the exon in the upstream direction, while downstream 200 nt (D200) intronic region was selected as the base 1 to base 200 downstream of the exon. Alignments of orthologus introns and exons sequences were generated by LAGAN using default parameters (). The ‘tissue-non-specific alternative’ dataset was derived as described previously () from the European Bioinformatics Institute database of human alternative exons (). ‘Control exon datasets’ were generated from randomly selected chromosomal regions by extraction from RefSeq annotation databases to get exon coordinates. Control groups for the mammalian and chicken genomes were described previously (). The muscle-enriched datasets and the control datasets is available at: A random subset of candidate muscle-enriched exons was selected for validation by RT–PCR, focusing (for ease of amplification) on those ⩽155 nt in length. RNAs from different human tissues, including heart, skeletal muscle and six non-muscle sources, were purchased from Clontech. One microgram of each RNA source was transcribed into cDNA using random hexamer primers in a total volume of 10 μl. Then, 2 μl cDNA was amplified in a volume of 25 μl, using primers located in the flanking constitutive exons (Supplementary Table 2), for 35 cycles under the following conditions: 30 s at 94°C; 30s at 55°C; 45 s at 72°C. The identity of PCR products was confirmed by DNA sequence analysis. Counts of hexamers were obtained in a specific pre-mRNA sequence region (upstream or downstream proximal intron). For each region, a linear model was fitted between the logarithm of ratios of gene-normalized exon expression levels and count of each 6-mer word across a set of exons, : NI is the gene-normalized expression level of exon in muscle, and refers to a reference sample. The reference data was taken as the average NI across all tissues. The coefficients and were obtained by a least squares fit. -values were calculated using an -test, as described previously (). The best fit was obtained for a set of sequences that included the muscle-specific exons (foreground set) and a background set of sequences ( = 300), drawn randomly from a set of manually curated 957 cassette exons across the human genome (). Since we started with a prioritized set of tissue-enriched sequences, a background set was necessary to model the correct dependence of log ratios on word count. such random draws were performed ( = 25), and a linear fit was obtained for each such draw. A geometric mean of the -values from all iterations reflects the overall significance of the word. italic disp-formula inline-formula sub #text xref #text xref #text The human muscle-enriched exon dataset analyzed in this study (Supplementary Table 1) was derived from exon microarray hybridization data using a platform designed to provide a comprehensive genome-wide analysis of annotated and predicted exons (see Methods section). In order to identify motifs that regulate tissue-specific alternative splicing, it is critical to identify a set of alternative exons having similar expression patterns indicative of regulation by a shared splicing program. Therefore, the group of exons studied here was carefully selected by analysis of exon microarray data from a panel of 16 normal adult human tissues. Probesets that exhibited gene-level NI that were significantly higher in heart and skeletal muscle, relative to 14 other tissues, were first identified. For this part of the analysis, we grouped the heart and skeletal muscle exon expression together to enhance the power of the statistical tests. Then a manual filtering process was performed so as to retain only probesets representing cassette exons, and to eliminate probesets corresponding to alternative first and last exons or to exon regions generated from alternative 5′ and 3′ splice sites. The final dataset consisted of 56 muscle-enriched, internal cassette exons. Most of these exons (∼80%) are integral multiples of 3 nt in length, with a median length of 84 nt, consistent with the notion that alternative exons are smaller than average constitutive exon length [∼145 nt (,)]. However, the genes with such alternative exons have a median size of 123 kb, much longer than the average gene length. To explore evolutionary conservation of candidate splicing regulatory elements, we also identified highly conserved orthologs for most of these human muscle-enriched exons in mouse, chicken and frog (Supplementary Table 1). It is important to note that while many of these exons show evidence of alternative splicing in Genbank, most were not previously known to exhibit muscle-enriched splicing and were not identified in the pilot study of muscle-enriched exons by Sugnet . (). Therefore, analysis of this dataset should yield novel insights into the vertebrate muscle alternative splicing program, and should provide an opportunity to explore computationally the regulatory motifs that carry out this program. Muscle-enriched splicing patterns for a random subset of these exons were validated experimentally in the human dataset by RT-PCR (). Although splicing patterns were not absolutely muscle specific, in almost every case the efficiency of exon inclusion was highest in heart and skeletal muscle, confirming the predictions of the exon microarray. Importantly, mRNA and/or EST evidence from the genetic databases (data not shown) demonstrates that the majority of these exons are alternatively spliced in at least one of the other species examined (mouse, chicken or frog), suggesting that the incidence of conserved alternative exons in this specialized dataset is higher than the reported rate for general alternative exons (). Taken together, these results indicate that the muscle-enriched exons constitute a special class of highly conserved alternative exons. Intron sequences flanking orthologous alternative exons in the mouse and human genomes tend to be evolutionarily conserved (), consistent with the observation that -regulatory elements for tissue-specific alternative splicing are often located in those proximal intron regions. We used VISTA genome alignment tools to compare the proximal intron sequences in this muscle-enriched dataset and extended the evolutionary comparison to include chicken and frog. In the proximal 200-nt upstream (U200) and downstream (D200) introns, mouse sequences were highly similar to their human orthologs (median identity of 61 and 58%, respectively), while chicken and frog introns were much less homologous. The full quantitative data are shown in Supplementary Table 3 and representative alignments of exons with relatively high conservation (FXR1), or lower conservation restricted mainly to the exon (LRRFIP1), are displayed in . The reduced overall homology of chicken and frog introns suggests that conserved motifs in these regions are likely conserved specifically for their function as -regulatory elements for muscle-specific alternative splicing, rather than being passively conserved as part of a larger conserved element. Previous studies have demonstrated that the brain-specific alternative splicing factor, NOVA1, modulates the splicing of many components of the neuronal synapse (). We hypothesized that the muscle alternative splicing program might similarly coordinate the expression of a particular class of genes that share a common pathway or cellular process. Using the method described previously (,), to examine the gene ontology (GO) terms associated with each parent gene for the muscle-enriched exons, we found a strong association with cytoskeleton organization and biogenesis, microtubule stabilization and muscle development (Supplementary Table 4). These associations were statistically significant ( < 0.001), suggesting that the muscle alternative splicing program is critical for proper expression of the unique cytoskeleton characteristic of vertebrate muscle. Alternative splicing regulatory elements responsible for tissue-specific splicing are often located in proximal intron sequences (). To search for candidate intronic regulatory motifs for the muscle-specific splicing program, we correlated the frequencies of hexamers in specific intronic regions with the logarithm of ratios of gene-normalized exon expression levels in skeletal muscle, across the 56 muscle-enriched exons in the human dataset. The ratio for any exon was enumerated against its average gene-normalized expression level across all the tissues. Thus, it is similar to the splicing index used above. The -elements exhibiting significant correlation with expression were considered potentially functional in regulating muscle-specific splicing. These were further examined for relative over-representation in introns of muscle-enriched exons, compared to a background set of introns flanking constitutive exons, using a hypergeometric distribution based on word counting in the oligonucleotide sequences. Finally, we examined their spatial conservation through vertebrate evolution () by testing whether the motif is over-represented in the other species using exactly the same statistical measures as used for humans. Here we consider in the downstream 200 nt (D200) of intron sequence as an example. This hexamer represents the binding site for mammalian Fox-1 and Fox-2 splicing factors (), which have identical RRM domains. has been reported as a common motif in proximal introns adjacent to tissue-specific exons. In a few cases, functional splicing assays have confirmed the importance of this motif in regulation of splicing (). In the large group of muscle-enriched exons studied here, we found a highly significant correlation of frequency with muscle expression ( = 6.8−05). The distribution and the linear fit for a single iteration of correlation analysis (see Methods section) are shown in A. Similar analysis shows that muscle expression does not correlate with occurrences in the upstream intron, whereas in the downstream intron the magnitude of correlation decreases with distance from the exon as demonstrated by the increasing -values (B). These dependencies are further corroborated by strong over-representation of this motif in proximal downstream introns in human and other vertebrates (C). Indeed, almost half of the muscle-enriched exons in all four datasets (23/56 in human, 21/54 in mouse, 20/43 in chicken and 19/36 in frog) possessed at least one motif in the first 200 nt of the downstream proximal intron. Together, these results strongly support the hypothesis that is potentially an important regulatory element for muscle-specific alternative splicing, as predicted by the correlation with expression analysis. We extended the above analysis to identify additional muscle-specific -elements in upstream and downstream intron sequences. In the downstream 200 nt sequence, we searched all possible hexamers and identified a total of 35 hexamers that were significantly correlated with expression (⩽0.05) and also over-represented in the human dataset ( ⩽ 0.05, ⩽0.2); nine of these were also over-represented in at least one other species (A and Supplementary Table 5A). Several of these elements have been previously characterized experimentally as regulators of muscle-specific splicing. These motifs fell into three distinct classes: (i) the Fox1/2-binding motif ( = 6.8−05) and two closely related hexamers (); of note, in all four species the majority (58–76%) of GCAUG motifs in the D200 region occurred in the context of the full UGCAUG hexamer. (ii) -rich elements and (correlation -values = 0.032 and 0.005, respectively), that resemble binding sites for the CELF family of splicing factors; and (iii) the novel motif ( = 0.0006) and related hexamers ( = 0.004) and ( = 0.04). The latter class is similar to the element noted in a recent study of a small group of muscle-specific exons in mouse (). The distribution of these elements in flanking introns of exons in the human, mouse, chicken and frog datasets is shown in A. Importantly, this analysis revealed that was the most over-represented hexamer in all four datasets, and both and were also consistently in the top ∼1% of the most over-represented hexamers in these species. For upstream intron sequences (200 nt), we found a total of 27 hexamers that were significantly correlated with expression and also over-represented in human, of which three were over-represented in at least one other species (B and Supplementary Table 5B). Many such elements are strongly pyrimidine-rich, characteristic of binding sites for PTB protein, an inhibitor of splicing for many alternative exons (). In all four species, the muscle-enriched datasets showed strong over-representation of the reported PTB-binding sites, and , in the proximal upstream intron (). was concentrated mainly in the U200 region. was focused even more tightly in the U100 region (B), where it was consistently among the top five over-represented tetramers in all four species. Lesser over-representation of motifs over a broad area of downstream intron sequences was also noted, perhaps consistent with previous findings that optimal splicing repression by PTB requires binding sites both upstream and downstream of the regulated exon (,). Many of the remaining significant hexamers, both for upstream and downstream introns, have low similarity to the previously discovered elements. Although these may represent novel elements, given that splicing elements are often degenerate, they can also be specific examples of known degenerate motifs. Our analysis using degenerate motifs presented below suggests that the latter possibility is more likely. Finally, for some of the major splicing regulatory elements described above, we observed that the profiles of positional over-representation have been conserved through vertebrate evolution: mouse, chicken and frog. This is displayed in for Fox, CELF and PTB-binding sites. Such strong positional conservation of motifs lends additional support to our findings using correlation with expression. The Fox-binding site, , was previously shown to be over-represented downstream of brain-enriched alternative exons (,), raising the possibility that the brain- and muscle-specific alternative splicing programs might exhibit functional similarities by sharing related components of the splicing machinery. To determine which of the candidate muscle -regulatory elements might be shared with brain-specific alternative splicing, and which are unique to the muscle program, we compared the frequency of several key -regulatory motifs in muscle (this study) and brain () datasets. Two elements, and , were clearly muscle specific since their frequencies were consistently higher in the D200–D300 region adjacent to muscle-enriched exons compared with the intronic region downstream of brain-enriched exons (Supplementary Figure 1; positive contrast scores). These motifs were also not over-represented in brain relative to control exons (data not shown). In contrast, the motifs and occurred at even higher frequencies in the proximal introns of the brain-enriched dataset than they did in the muscle-enriched dataset (Supplementary Figure 1; negative contrast scores for in the D200–D400 region, and for in the U100 region). Essentially equivalent distribution patterns were observed in the mouse, chicken and frog datasets (data not shown). These results strongly suggest that tissue-specific alternative splicing programs may utilize a combination of unique and shared -regulatory motifs that will require much additional analysis in the future. Because many splicing factors bind degenerate oligonucleotide sequences in RNA, we extended our analyses to include degenerate motifs through the use of PWMs (,). PWMs are probabilistic representations of degenerate binding sites. Over-represented PWMs in introns of 56 muscle-specific exons in the human dataset were obtained using the DME algorithm (,). We scanned multiple parameter settings of DME in order to obtain a large number of PWMs and reduce bias from DME. To identify the functional PWMs, we assessed their correlation with muscle expression using linear splines (,). Linear splines are among the simplest non-linear variants of linear models. In contrast to many other approaches, they facilitate adaptively learning the cutoff scores of PWMs that discriminate true targets from false targets of -factors. Previous regression approaches have used either maximum score of the PWM () or a global average of PWM scores for all potential binding sites () on an input sequence as the predictor variable. However, realistically, a small number of sites, sometimes >1, are bound by the corresponding -factor. Here we overcame this limitation by including both strength of PWM and the number of putative binding sites in our linear splines approach (see Methods section). Degenerate 6-nt and 4-nt sequences that were over-represented in the proximal downstream intron sequence are shown in A and B and Supplementary Table 6A and B. Notably, all of the top 10 over-expressed PWM hexamer motifs in the D200 region are consistent with the major over-expressed unique motifs identified above. Among these, the six most statistically significant motifs represent close matches to the Fox-binding site, ; two ( and ) are very similar to the novel element; and the remaining motifs ( and ) resemble -rich-binding site for CELF proteins. Analysis of over-expressed 4-mers in the D50 region revealed that the top-scoring motif is . While this motif is included in the Fox recognition sequence, other considerations suggest that it would not be sufficient for Fox binding. Instead, likely represents the -rich sequences, characteristic of some CELF-binding sites. In the U200 region, all of the statistically over-represented motifs were quite pyrimidine-rich relative to the control group (B and C). Further investigation will be required to determine whether these elements are primarily bound by the PTB protein or by additional splicing factor(s). All remaining PWMs that have high significance in upstream and downstream introns exhibit at least partial similarity to the above three elements. For PWMs with only partial similarity, the similarity is observed either at the 5′ or at the 3′ end of the motif, indicating that the remainder of the motif most probably represents the flanking region. For example, for the PWM , the last four bases match the 5′ end of the , and hence, the first two bases are presumably the flanking region of this putative Fox-binding motif. Furthermore, in contrast to previous work (), the new formulation of linear splines used here allowed us to obtain not only the potential target exons, but also the binding sites of the above splicing factors (see Methods section). The results for a representative set of motifs are summarized in the Supplementary Table 7. For the putative Fox-binding motif , we find as the most frequently occurring oligonucleotide sequence, as expected of Fox-binding sites. We have observed similar accuracy in binding site prediction in the context of transcriptional regulation (Das,D., unpublished data). Interestingly, we notice that not all possible combinations of nucleotides of a degenerate PWM are realized in the set of 56 muscle-specific exons. For example, for the candidate CELF-binding motif, , only is predicted as the binding site. These are consistent with the previous observations made in the context of transcriptional regulation (,). In this study we have demonstrated that the correlation with expression approach, applied to global exon expression profiles, represents a powerful new tool for identification of -regulatory motifs for alternative splicing. Using a dataset of high-confidence muscle-enriched alternative exons extracted from human exon microarray data, we correlated motif occurrences in the flanking introns with the splicing index measure of relative muscle enrichment to identify candidate regulatory motifs for the muscle-splicing program. The logic of this strategy is supported by many studies of transcriptional regulation, and a few of splicing regulation (), showing that functional response often correlates with regulatory motif copy number. The analysis presented here demonstrates that the number of Fox splicing factor binding sites correlates strongly with the muscle splicing index (A), consistent with previous reports that Fox proteins can regulate various tissue-specific alternative splicing events. The validity of correlation results were further supported by over-representation analysis, by comparative genomics showing that top scoring correlation motifs are phylogenetically conserved among vertebrate genomes, and by previous experimental studies implicating most of the same motifs in regulation of muscle-specific exon(s). Since tissue-specific alternative splicing is rarely an all or nothing phenomenon (e.g. ), correlation with expression may offer an attractive approach toward understanding complex tissue-specific patterns of alternative splicing. This approach may be particularly effective when PWMs are utilized in the splines-based framework to account simultaneously for both relative affinity and number of motif occurrences, providing insight into both the target exons and binding sites associated with a given motif. Our immediate goal here in this proof of concept study was to examine whether the correlation with expression method can be used to identify splicing regulatory motifs, and consider muscle-specific alternative splicing program as an example of this application. This analysis strongly implicated several classes of known regulatory factors including Fox (), CELF ( and ), PTB ( and ) and putative KH-type splicing factor () as important mediators of muscle-enriched splicing. The current study thus confirms and substantially extends earlier reports that these factors can regulate one or a few muscle-enriched exons by providing significant new computational evidence that they correlate with muscle exons in a much larger dataset. Interestingly, there was a notable lack of novel -elements in the proximal flanking introns that strongly correlate with muscle expression across the entire dataset. This could indicate that much of the fundamental machinery for regulation of generalized muscle-enriched splicing has been identified or, more likely, that additional features need to be incorporated in the algorithms to identify the remaining components. Such features may include wider motifs and motifs located more distally from the regulated exons. It is also possible that there are weaker elements, which may only be revealed when combinatorial interactions among motifs are included in the regression models, or which may be required for spatially or temporally distinct subsets of muscle-enriched exons. To obtain an initial estimate as to which of these factors may be most influential, we extended our study to include PWMs of width 5–7 nt. The results are displayed on our website (). We observe that most motifs have similarities to the known motifs as identified above. There is one motif in D200, GGSYVYW, which seems novel. But since it has much higher -value than others ( = 0.01), it is not readily clear if it is truly functional. Hence, we suspect that inclusion of combinatorial interactions among motifs may be most effective in revealing the novel motifs. One question that needs to be addressed in future studies, as improved measures of binding specificity become available, is the importance of additional splicing factors such as the muscleblind proteins that are already known to influence the splicing of at least a few muscle-specific alternative exons (). A working model that summarizes these findings is presented in . Fox, CELF and -binding factors are proposed as positive regulators of muscle-enriched exons via their binding to the downstream proximal intron. The distribution of binding motifs among individual introns suggests that these factors function independently in some cases, and collaboratively in others, to specify muscle-enriched splicing. For Fox proteins an especially widespread role is suggested by the high absolute abundance of -binding motifs: almost half of the muscle-enriched exons in datasets of all four species possess at least one motif in the D200 intronic region, and some of those lacking a proximal have phylogenetically conserved distal motif(s) (data not shown) analogous to the myosin II heavy chain-B neural specific exon (). It will be interesting in the future to explore how coordination among these and other factors ultimately determines the spatial and temporal details of muscle-enriched splicing events. Based on studies in other systems, PTB is predicted as a negative regulator of splicing, functioning primarily from upstream intronic sites to prevent inappropriate inclusion in non-muscle cell types (,,). Finally, it is important to note that variations of this general model likely pertain to individual exons; in particular, Fox and CELF proteins can also have a negative role in the regulation of exons that are skipped in muscle (,). Future experimental analysis of these splicing factors, using functional splicing assays and targeted disruption of splicing factor activity (), will be required to more fully test the predictions of this model. Some of the -regulatory elements associated with muscle-enriched alternative exons have previously been observed flanking brain-enriched exons: was the most over-represented motif in proximal downstream intron (,,), and was the second most over-represented motif in the U100 region upstream of brain-enriched exons (). These observations suggest general roles for Fox- and PTB-related proteins in regulating tissue-specific splicing, at least for muscle and brain, but raise the question as to how tissue specificity is ultimately determined. Several mechanisms may contribute to determination of temporal and spatial pattern of splicing switches, including tissue-specific differences in transcription and/or alternative splicing of Fox and PTB paralogs (). Differential expression of additional RNA-binding proteins, such CELF proteins and KH-type -binding proteins in muscle, or NOVA1-related proteins in brain, likely also play a role, as may non-RNA-binding co-factors that preferentially interact with paralogs/isoforms of the primary RNA-binding proteins. In summary, normal metazoan development requires not only a transcriptional program, but also an alternative pre-mRNA splicing program to ensure that each gene encodes specific protein isoforms in the appropriate spatial and temporal patterns. Enrichment within the muscle dataset of genes with functions in cytoskeleton organization, microtubule stabilization and muscle development supports the notion that this splicing program is essential for proper expression of the unique muscle cytoskeleton. The exon microarray employed in this study will enhance our ability to track the expression of individual exons during development and differentiation. As we have demonstrated here, this experimental approach is well complemented by the computational approach based on correlation with expression. We anticipate that correlation with exon expression will provide valuable insights into the -regulation of alternative splicing as additional datasets of tissue-specific exons become available for analysis. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

Ever since the dawn of the genomic age, accurate prediction of protein coding (and other) genes has been a central problem of biology (,). New annotations are continually released, even of the best-annotated and most carefully studied genomes. The statistical task of distinguishing true coding genes from non-coding sequences requires evading an array of pitfalls, among them pseudogenes, coding elements of repetitive sequences, short ORFs that are biologically but not statistically significant, and transcription of non-coding sequences. A wide variety of statistical and bioinformatic approaches to gene prediction have been developed, including comparisons with sequences from transcripts and from other species, statistical analysis of ORF length and Bayesian comparison of gene models (). One of the major obstacles to accurate gene prediction in eukaryotes is the presence of spliceosomal introns. Splicesomal introns are genomic sequences that are removed from RNA transcripts prior to translation by a very large RNA–protein complex called the spliceosome. Spliceosomal introns show large variations in intron number per gene, typically exhibit no length or sequence conservation either within or between species, and afford opportunities for alternative splicing, further complicating accurate deduction of a species’ protein arsenal. Here we point out a simple aspect of the expected distribution of spliceosomal intron lengths within a genome, which we hope may be helpful to ongoing and future annotation efforts. Due to their removal from transcripts prior to translation, intron sequences are generally not expected to respect the coding frame and meaning of the surrounding coding sequence. Correspondingly, many predicted introns in the most thoroughly annotated eukaryotic genomes have in frame stop codons, and predicted introns in these genomes are equally as likely to be a multiple of 3 basepairs (bp) (‘3’), and thus to conserve reading frame, as to contain an ‘extra’ one (3 + 1) or two (3 + 2) bp. For genomic sequences without exhaustive databases of transcript (EST and/or cDNA) sequences, prediction of introns is a difficult task. Here, examination of the distribution of intron lengths can provide insights into the possibility of intron over/underprediction. We report distributions of predicted intron lengths for 29 fully sequenced eukaryotic species. We find frequent deviations in the number of predicted 3 introns relative to 3 + 1 and 3 + 2 introns. Some species show a pronounced deficit of 3 introns, others an excess of 3 introns. We discuss five different species that show highly skewed length distributions among predicted introns, and suggest ways in which current annotations might be improved. italic table-wrap #text We studied current genome sequence annotations for 29 different eukaryotic species (). For most genomes with large numbers of introns, there are very similar numbers of 3 + 1 and 3 + 2 introns: among genomes with >300 introns, the percentages of 3 + 1 and 3 + 2 introns are within 2.8% of each other in 23/25 genomes. In stark contrast, the number of 3 introns varies much more widely, falling only within 2.8% of the average of 3 + 1 and 3 + 2 for half (13/25) of the genomes. Species are roughly evenly divided between those that show an excess of 3 introns relative to 3 + 1 or 3 + 2 introns, and those that show a deficit of 3 introns. Two species, and the nucleomorph of , show very different patterns—an excess of 3 + 1 introns in and an excess of 3 + 2 introns in . We next analyzed the sets of predicted introns for several cases that showed pronounced deviations from equal intron numbers in the three classes. Due to the lack of close relatives with sequenced genomes or very large samples of transcript sequences, genes in the genome were largely predicted by homology searches against sequences from deeply diverged eukaryotic species (). Predicted genes have on average 1.4 introns per gene. These predicted introns show a strongly skewed length distribution, with 3 introns accounting for 61.2% of all predicted introns (9573 3 introns, 3037 3 + 1, 3029 3 + 2; an example is shown in ). Such a skew suggests that many predicted 3 introns are not true introns but instead represent exonic sequences. In keeping with this possibility, most 3 introns (75.2%) lack inframe stop codons, in stark contrast to 3 + 1 (29.1%) and 3 + 2 (28.6%) introns. From these two observations it is possible to obtain independent estimates of the number of predicted introns that in fact represent coding sequence (i.e. false positive intron predictions). First, based on the assumption of equal numbers of 3, 3 + 1 and 3 + 2 introns there is a 3 excess of 6540 introns (note that numbers of 3 + 1 and 3 + 2 introns are nearly identical). Second, roughly equal fractions of 3, 3 + 1 and 3 + 2 introns are expected to lack inframe stop codons (for instance 29.1% of 3 + 1 introns and 28.6% of 3 + 2 introns should lack stop codons). There are 2368 stop-containing 3 introns, thus we expect roughly 972 (=2368 × 28.8%/[1–28.9%]) 3 introns without inframe stop codons, 6233 fewer than predicted. Thus, two independent estimates suggest that 6200–6600 (86–90%) of the 7205 predicted 3 stop codon-lacking introns instead represent unspliced coding sequence. A second case suggestive of the reverse problem, of underprediction of 3 introns, is found in the annotation of the largest somatic chromosome of . In this case, there is a striking deficit of predicted 3 introns (185 total) compared to 3 + 1 and 3 + 2 introns. In this case, this deficit is likely due to the short intron lengths in : all predicted introns are less than 36 bp in length. Whereas, long non-coding sequences are likely to contain in-frame stop codons by chance, but short introns may lack stop codons, in which case 3 introns may be mistaken for coding sequence, whereas the presence of a 3 + 1 or 3 + 2 intron may be inferred from the disruption of the coding frame. That many stop codon-lacking 3 introns may have gone unpredicted is underscored by the high frequency of stop codons in the predicted 3 introns (91.3% contain a stop codon), much higher than in 3 + 1 (46.0%) or 3 + 2 (47.7%). If there were 264 3 introns currently incorrectly predicted as coding sequence, the number of 3 introns would be equal to the average of 3 + 1 and 3 + 2, and the fraction of stop codon-containing introns would be similar (37.6% for 3) across classes. A similar pattern is seen in the predictions for the genome, where 3 introns (20% of all predicted introns; 1290 total) are only half as frequent as 3 + 1 or 3 + 2 introns (). Again, a much higher fraction of predicted 3 introns contain stop codons (67.5%) than for 3 + 1 (42.1%) or 3 + 2 (38.5%). If there were an additional 1290 unpredicted 3 introns, there would be equal numbers across the classes, and similar fractions of stop-containing introns (33.8% for 3). The above examples each concern difficulties of predicting introns based on an accurate genome sequence. Alternatively, errors in a genome assembly can lead to overprediction of introns. One example involves the genome of (). Previous analysis of introns in genes thought to have been laterally transferred from prokaryotes showed that many predicted introns were associated with errors in the assembly in which a single base was missing in the assembly relative to the corresponding individual sequencing reads (). Insertion of this missing base into the assembly yielded an ORF that continued through the predicted intronic sequence, suggesting that there is no intron present (e.g. ). Thus, these assembly indels led to frameshifts in coding sequences, which were compensated for by prediction of an intron. Further analysis of the genome suggests that this may be a common problem. Among predicted introns, there is an excess of 3 + 2 introns over 3 or 3 + 1 introns. BLAST searches of the predicted intronic and flanking exonic sequences against individual sequencing reads showed that 23.1% (722/3126) of predicted introns were associated with gaps in the intron or within 120 bp of the intron (). Of the gaps, 98.3% (831/845) were single-base gaps, and 81.3% (687/831) were missing bases in the assembly relative to the sequencing reads (the predominance of missing bases in the assembly is consistent with the excess of 3 + 2 introns, since the added base yields a 3 sequence; the smaller number of extra assembly bases relative to reads is consistent with the smaller excess of 3 + 1 introns over 3 introns). Correction of these apparent assembly errors extended the ORF through the intronic sequence in 79.8% (538/649) of these cases, suggesting that the predicted intronic sequence instead represents coding sequence. In an additional 48 cases (7.4%), correction of the apparent assembly error yields an ORF spanning from within (or upstream of) the predicted intron to the predicted stop codon (or to the next predicted intron boundary in the case of multi-intron genes). These results suggest that at least some 20% of predicted introns are not in fact introns but instead coding sequence. Thus, upwards of 6% of predicted genes in the genome appear to have an assembly error within their sequence. Finally, it is worth noting that there is at least one apparently bona fide case of stiking genome-wide difference between introns of different length classes (,). Among predicted introns in the genome of the nucleomorph of the genome, 70.3% of introns are 3 + 1 (,). However, this is due to the extreme regularity of intron length, with 70.2% of predicted introns having length 19 bp, and 99.1% being of length 18–20 bp (an example is given in ). How accurate is this length distribution likely to be? Introns in are predicted based on maximizing ORF length. On its face, then, we would expect the rates of detection of 19 and 20 bp introns to be similar (since both impose a frameshift). Thus, the great excess of 19 bp introns over 20 bp introns is likely to be a true feature of intron lengths in . On the other hand, detection of 18 bp introns will be far more difficult since these introns will not disrupt the ORF unless they contain an inframe stop codon (which is not likely in a 18 bp sequence). Correspondingly, the fraction of predicted introns containing inframe stop codons is higher for 18 bp introns (89.7%) than for 19 bp introns (48.4%). This suggests that 18 bp introns are underpredicted by perhaps twofold, however underprediction is unlikely to account for the 7.0-fold excess of predicted 18 bp introns over 19 bp introns. The example of raises a larger point of the possible utility of integrating genome assembly with gene annotation. In this case, examination of predicted genes indicated errors in the genome assembly itself, thus integration of the two processes should lead to improvements in both assembly and annotation quality. Such considerations are likely to be all the more important with the increased number of partial and low-coverage genome sequencing projects. Analysis of the apparent coding meaning of preliminary assemblies could identify probable indels in the assembly, and corresponding individual sequencing reads could then be scrutinized in order to correct actual errors. This process lends itself well to automation and we think that such feedback between assembly and annotation could potentially substantially diminish assembly errors in coding regions. In the absence of extensive transcript data, intron prediction often proceeds by statistical comparison of alternative gene models: introns are called when an intron-containing structure has a higher probability given model perameters than does an intron-lacking structure (). In the case of non-3 or stop-containing introns, this is comparatively straightforward, with intron prediction being likely if the resulting coding sequence is significantly longer than the intronless alternative. For non-stop-containing 3 introns, the case is quite different. In this case intron calling involves comparison of two structures with identical 5′ and 3′ flanking coding sequences, thus whether an intron is called will depend only on the extent to which the intervening sequence conforms to expected intron sequences. Here, too much/little sensitivity to intron-like structures will lead to over/under-prediction of introns. One possible actual biological deviation from equal proportions is worthy of discussion. In a subset of alternative splicing events, an intronic sequence from one splicing isoform is retained in another sequence (so-called ‘intron retention’). In such cases, the alternatively spliced intron must be 3 and lack inframe stop codons in order for both isoforms to encode proteins. Thus, frequent alternative splicing could conceivably bias intron lengths towards 3 introns. Four observations suggest that such alternative splicing events are not a major contributor to the skewed distributions reported here. First, the genomes which show the highest frequencies of alternative splicing (for instance ) are among the genomes that show the intron length proportions most nearly equal. Second, in all studied genomes, not more than 5% of introns are known to be alternatively spliced in the manner discussed above, and many of these are not 3 introns, thus this effect is unlikely to drive any more than a small bias even in the most highly alternatively spliced genomes. Third, there is no evidence for frequent alternative splicing in any of the genomes that show pronounced intron length skew. Finally, such an effect specifically predicts an excess of 3 introns, and as such is not a contributor to the other observed skews (3 deficit, 3 + 2 excess). Accurate genome annotation is an extremely difficult problem, requiring balancing of false negatives and positives, and accuracy versus time constraints. Even the best annotation sets are subject to improvement. Evaluation of distributions of predicted intron lengths promises rapid and straightforward detection of a variety of possible systematic biases in gene prediction or even, as in the case of , problems with genome assemblies.

In the recent past, experimental and computational approaches have identified a vast variety of non-protein-coding RNAs (), generally abbreviated as non-coding RNAs (ncRNAs), from both unicellular and multicellular eukaryotes. Many ncRNAs in modern eukaryotes function in RNA–protein complexes within which the RNAs may have direct regulatory roles at the reaction centres (). The size of many ncRNAs is small compared with protein-coding RNAs, and lack of sequence homology often results in difficulties of identifying ncRNAs in distant eukaryotes through purely biological or computational approaches. In this study, our combined experimental and computational approach has been successful in finding novel ncRNAs in the distant eukaryote . Eukaryotic genomes are rich in non-protein-coding sequences. Large-scale cDNA cloning studies have shown that a large proportion of mammalian RNA transcripts do not appear to encode proteins (), and an increasing number of ncRNAs have been shown to be functional ().The origin of ncRNA is likely to date back to the earliest events when life emerged on earth. The theory of the ‘RNA-World’ (,) suggests that self-replicating RNAs are older than protein or DNA. The versatile features of RNA molecules support this hypothesis: first, RNA stores information in the same way as DNA; second, single-stranded RNA molecules are highly flexible to form secondary or tertiary structures, like peptides, they can form enclosed reactive centres and catalyze biological reactions in liquid environment. However, modern natural ribozymes have limited catalytic abilities, as natural ribozymes only perform ligation and/or nucleic acid cleavage reactions. These reactions are normally not limited by the rate of the catalytic reaction (). Therefore, it is assumed that most ancient ribozymes have gradually been replaced by protein enzymes (). On the other hand, the evolution of ncRNAs has been continuous, and functions of ncRNAs have been diversifying throughout the evolution of eukaryotes. Based on structural and functional definition, eukaryotes have several distinct classes of ncRNAs, which form complex RNA-processing networks. shows that each type of RNA often participates in the modification of another type of RNA, and the whole network fits into the general RNA-processing cascade (). It is necessary to provide some brief background on the types of ncRNAs here, because in this study, we have characterized a number of different types of ncRNAs from . Probably the best studied ncRNAs are uridine-rich spliceosomal snRNAs (U-snRNAs). They function in the catalytic centre of major and minor spliceosomes. The major spliceosome that splices the majority of eukaryotic introns, consists of 5 U-snRNAs (U1, U2, U4, U5 and U6) and over 200 proteins (). The minor spliceosome is low-abundant machinery containing U11 and U12 snRNAs instead of U1 and U2, and splices a ‘minor’ (less frequent) class of introns (). Both major and minor spliceosomes may be ancestral to eukaryotes because they have now been identified in animals, plants, fungi and recently some distantly related protists (,). The small nucleolar RNAs (snoRNAs) are involved in rRNA biogenesis. An increasing number of novel snoRNAs have been widely identified and have been reviewed in detail (). Based on their structural motifs, snoRNAs are divided into two classes: C/D-box 2′-O-methylation snoRNAs and H/ACA-box pseudouridylation snoRNAs. The snoRNAs bind near the sites of modification through antisense recognition, and guide protein enzymes to the sites of editing. In addition, the functions of snoRNAs can be extended to acting as general chaperones targeting other nuclear or cellular RNAs (). There are a number of larger ncRNAs (>300 nt) such as the RNase P and RNase MRP RNAs. To date, besides the ribosome, RNase P is the only ribozyme required in both eukaryotes and prokaryotes (). Eukaryotes have another related ribonuclease, RNase MRP, which processes a specific site in the pre-rRNA which is not found in prokaryotes, however, it seems likely that it is present in all eukaryotic lineages (). Structural analysis of RNase P RNAs from phylogenetically diverse eukaryotes reveal a very similar minimum core (). The overall structure of the RNA subunit in RNase MRP is similar to that of RNase P (), and also the MRP enzyme shares a number of proteins with RNase P (). The smallest ncRNAs are micro RNAs (miRNAs) with length ranging from 21–25 nt and function in a variety of gene silencing pathways (). About 800 miRNAs from different animals and plants have been reported (). miRNAs from animals are usually transcribed as long and often polycistronic precursors, and then processed into small hairpin intermediates, which are then cleaved by a conserved protein Dicer () into mature miRNAs. The Dicer protein has been well studied for (). Recently, new experimental and bioinformatic approaches have identified a great number of novel ncRNA candidates from many organisms including: bacteria (), animals () plants () and protists (). The most widely used experimental method for identifying novel RNA candidates is based on size-selected cDNA libraries. Since most mRNAs have lengths greater than 500 nt, it is possible to isolate the majority of ncRNAs by size fractionation on a denaturing PAGE gel. Several methods are available to generate cDNAs from purified RNAs including the addition of poly(C)/poly(A) tail, and adaptor ligation at 5′-end and/or 3′-end, followed by reverse transcription, cloning and cycle sequencing (). Here, we have constructed a cDNA library for ncRNAs from the deep-branching eukaryote , a parasitic diplomonad, is phylogenetically distant to all model eukaryotes (,). This unicellular organism has reduced mitochondria (mitosomes) and lacks hydrogenosomes (). Two spliceosomal introns have been found (,), as well as several spliceosomal proteins () which strongly suggests that has a functional spliceosome. To date, several studies have identified 24 sno-like RNAs and the RNaseP of (). However, there is little systematic research reported for the RNomics of . We have screened our cDNA library, resulting in 31 novel candidates, within which, three are possibly C/D-box snoRNAs, one is possibly an H/ACA box snoRNA, and one is a fragment of the RNase P RNA. A computational study using known 's C/D box snoRNAs has resulted in new putative snoRNAs. In addition, an extended transcript has been found for the RNase P RNA, and two unusual self-cleaving dsRNA candidates have been studied. Given its proposed basal position on the eukaryotic tree (), is evolutionarily distant to all the eukaryotic species, and probably highly reduced. It is not surprising to see that there may be some different RNA processing components in this organism. Future comparison of RNA-processing between and other eukaryotes is very necessary in understanding the evolution of RNA metabolism in reduced organisms (). RNA processing in is expected to have changed in both the RNA and protein components as a result of genome reduction () due to the parasitic nature of this organism. Our study moves towards understanding differences in RNA-processing machinery from that other eukaryotes which to date is largely confined to model, well-studied eukaryotes. Cells were collected from TY1-S-33 growth media at a concentration of 1.4 × 10 cells/ml by centrifugation (10 min, 2500 r.p.m., 4°C). Total RNA was prepared using Trizol reagent (Invitrogen) according to the protocol provided by the manufacturer. The RNA-cDNA mix was treated with RNase A and PCR amplified using Sal-1 and Not-1 primers using a Biometra thermocycler. The PCR product was then double digested by Sal-1 and Not-1 restriction enzymes and ligated into the pSPORT1 vector (Invitrogen), followed by transformation into Top10 cells (Invitrogen). cells were grown on LB agar plates (100 μg/ml Ampicilin) at 37°C overnight. Colonies were PCR amplified using the M13for and M13rev primers (Roche Taq polymerase): The sequences were assembled using DNAMAN 5.2 and DNASTAR 5.0 packages, and were then blasted against the genomic database () as well as the NCBI databases (). Putative snoRNA prediction used the modified Snoscan program (Snoscan-G) in C for Windows (the original source code is available at ). However, the C-box scoring function was modified so that it read user-specified input of the C-box scoring matrix. RNA structures were generated using the RNAfold program from the Vienna-RNA-1.4 Package (), structural alignment was done using RSmatch1.0 converted for Windows (original program is available at ) and FoldalignM ().rRNA sequence alignments for preliminary methylation site analysis were generated using ClustalW (). Assembly of the cDNA sequences resulted in 31 novel ncRNAs, 15 previously known snoRNAs () and 10 out of 48 characterized tRNAs (). Candidates were obtained in the following manner. A total of 616 initial sequences were assembled into 166 contigs and each contig was blasted against the genome database and NCBI databases to screen for easily characterized RNAs. After discarding empty vector contaminants, sequences below the length of 20 nt and contaminant sequences, the remaining 152 contigs (including repeats or duplicates) contained 33 mRNA fragments, 28 known tRNA sequences, 10 5.8S rRNA sequences, 7 LSU rRNA and SSU rRNA fragments, 29 known ncRNAs sequences and 45 unknown sequences. All the unknown sequences were further analysed so that any broken fragments of a single RNA could be reassembled into a complete sequence, leaving 31 novel RNA candidates. Details of candidate sequences and features are listed in Supplementary Data. In order to carry out further computational analysis, 5′-and 3′-extensions (200 nt from each end) were extracted from the genome database for each candidate. Eukaryotic 2′-O-methylation C/D box snoRNAs are characterized by two short sequence motifs near their 5′-and 3′-termini: C-box (‘5′-AUGAUGA-3′’) and D-box (‘5′-CUGA-3′’), which are brought together by a short () terminal stem (). There are one or two 10–20 nt antisense guide elements immediately upstream of the D-box or D’-box, and these elements bind to complementary sequences on rRNAs spanning the methylation sites (). The position of the nucleotide which is methylated is usually the fifth position upstream of the D-box or D’-box (). Since the genome is fully sequenced (NCBI accession number: AACB00000000), it is possible to check our experimentally found RNAs for snoRNA features using potential interactions to rRNA sequences. Once we identify the conserved features of a snoRNA, we can identify more snoRNAs using a computational search. However, to date there are no full-length rRNA large subunit and small subunit rRNA sequences available for . Raw sequence reads from the GiardiaDB () were pulled out individually and assembled using SeqMan. Three contigs were generated, and correspond to the large subunit (LSU), small subunit (SSU) and 5.8S rRNAs, with lengths of 2908, 1449 and 138 nt respectively, and they arrange in the typical eukaryotic rRNA-gene order of SSU-5.8S-LSU, which reveals a site of cleavage by RNase MRP (). The sequences are listed in Supplementary Data. Shortened lengths of the rRNAs are consistent with an earlier study () that 's rRNAs are much shorter than usual eukaryotic rRNAs, and unlike other eukaryotes, does not appear to have the 5S rRNA (), which was also not found during our searches. The snoRNA search was done using modified source code of the Snoscan program, which was originally used to identify a large number of C/D-box snoRNAs from (). We have predicted 3 C/D box snoRNA candidates from the 31 novel candidate sequences. Of the 15 known snoRNAs () that were found in our cDNA library, 14 are C/D box snoRNAs and 1 is an H/ACA box snoRNA. Comparing all the available C/D box, snoRNA sequences revealed that snoRNAs from share common sequence features within the C boxes and D boxes. All but one of the confirmed C/D box snoRNAs has a perfect ‘CUGA’ D-box near the end of the 3′-end, and most C-boxes have a conserved sequence ‘5′-AUGAU-3′’ allowing one mismatch at either 5′-or 3′-end. The C-box sequences also appear more variable as their lengths range between 5 and 7 nt. The C-box scoring function of Snoscan was adjusted to use the consensus sequence. The C’-box is generally missing or poorly identifiable, and the existence of D’-box is not essential. The length between the C- and D-boxes is varying from 28 to 124 nt. In addition, few of the known C/D box snoRNAs have a terminal stem. The general structure of C/D box snoRNAs during rRNA modification is shown in a. Structural alignment was done on all the experimentally found C/D box snoRNAs using RNA structures generated from Vienna-RNAfold program, but the result did not indicate any additional consensus motifs. Therefore, no further structural features were incorporated into modifying the Snoscan program. Our modified Snoscan program, Snoscan-G, identified 13 out of 18 confirmed C/D-box snoRNAs with the following parameters: cutoff total score (), C-box score (2.0) and the maximum distant between C and D boxes (150 bp). The others were not recovered due to poorly defined C-boxes or imperfect D-boxes. This testing indicated that it was possible to identify additional C/D-box snoRNAs from the genome with this computational method. shows the range of scores obtained from experimentally identified snoRNAs. These are considered as standard scores for , thus used to compare with the scores generated for computationally predicted snoRNAs further on. We refer to these computationally predicted snoRNAs as ‘putative’ snoRNAs in order to distinguish them from the ‘candidate’ snoRNAs found experimentally. Due to the short (5 nt) and less conserved C-box, large volume of output was expected. A whole genome search for C/D box snoRNAs using the same parameter settings yielded many non-repetitive putative candidates, which were subsequently analysed through a strict three-step post-scan filtering. Three features of the putative snoRNAs were looked for during the post-scan filtering: All the output sequences from Snoscan-G were compared against the database of open-reading frames (ORFs) downloaded from GiardiaDB () to exclude possible mRNA sequences. These ORF datasets have been expertly compiled using software such as GLIMMER and CRITICA with parameters adjusted for this unique eukaryote. Our search of this database implicitly filtered out putative candidates with obvious coding potential. The status of the genome is such that a large number of ORFs remain hypothetical. Any explicit assessment for coding potential could be on only a subset of highly conserved proteins and would not be representative of the entire proteome. Hence, the use of this database maximizes our exclusion of contaminant mRNAs. Unlike other eukaryotes, has only two confirmed introns (,), and most ncRNAs characterized to date are located between protein-coding genes, with a small number (less than 10) of them located on the minus strand of protein-coding genes. To exclude any ambiguities, only sequences located between protein-coding genes were considered. Sequence searches showed that most of the Snoscan-G outputs had full-length 100% match to ORFs, leaving 423 potential putative snoRNAs. After excluding shorter partial sequences and repetitive sequences with different names, 357 sequences remained. To date, all 13 experimentally confirmed C/D-box snoRNAs that had been detected in the small-scale Snoscan-G testing were also found in this large-scale genome search. It was noticeable that all the experimentally characterized snoRNAs were located in ORF-rich regions of the genome, which could due to the fact that these snoRNAs do not seem to possess their own promoters. Therefore, further screening was done based on genomic location. Only putative sequences that are located near ORFs were selected with those appearing in heterochromatic regions excluded because they are less likely to be transcribed. This screening left 101 putative snoRNA sequences. Strict post-scan filtering based on C-box and D-box sequences was then done so that only sequences with ‘AUGAU’ or ‘GUGAU’ in C-box and ‘CUGA’ in D-box were considered as highly likely putative snoRNAs. In the end there were 60 strong putative snoRNAs. All sequences had distinct C-box and D-box motifs and fulfill the criteria for snoRNAs (,). In addition, they had average Snoscan-G total score of 12.5, which was slightly above the average total score of experimentally identified snoRNAs. The details of candidates are shown in the Supplementary Data. As a control, we generated a random database with its size equivalent to genome using a third-order Markov chain based on 4-mer frequencies () within the genome. A search of this random sequence database yielded 6721 false positives with an average score of 11.8 and a best score of 25.26. As downstream filtering based on genomic location was impossible to carry out on randomized data, only the last step of the three-stage filtering could be performed on this output. Therefore, a parallel comparison between the Snoscan-G outputs and the randomized data outputs was not entirely applicable since the first two steps of the post-scan filtering were the most important and based on genomic information. However, a strict scan was still performed on this output with more stringent parameter settings based on C-box and D-box motifs, as was done in the final stage of post-scan filtering described above, reduced the positives down to 89 non-overlapping ones. Although these outputs contain C-box and D-box motifs, they do not represent comprehensive data for comparisons. In all, the purpose of generating a randomized dataset was to show that post-scan using genomic information was necessary to improve the selection of putative snoRNAs in a distant organism such as . To test if the large number of initial output from the random database was due to special features within the genome, another Snoscan-G was run on a partial yeast genome (with a size similar to genome) using the same parameter settings. There were 1756 non-repetitive outputs. This test showed that the genome has less regional variation in its sequence, and this may result in the observation of more false positives. This testing showed that it was necessary to carry out stringent downstream filtering as was done in our Snoscan-G of the genome to obtain acceptable putative snoRNAs. As an additional analysis, human and yeast C/D box snoRNAs have been mapped onto rRNAs (alignments included in Supplementary Data). Since human and yeast are extremely evolutionarily distant from , most known methylation sites do not have homologues in , apart from two. ncRNA candidate-1 from our cDNA library is predicted to guide methylation of G on SSU-rRNA, which corresponds to the site of modification by human U25 snoRNA. Snoscan-G predicted putative snoRNA U0025 is likely to guide methylation of C on LSU-rRNA, which corresponds to the site of modification by an undetected human snoRNA. However, as these alignments are between such diverse organisms, no extensive conclusions can be drawn at this time. In all, our Snoscan-G in combination of the post-scan filtering has identified 60 C/D-box snoRNA putative snoRNAs based on information from previously experimentally characterized snoRNAs. This approach was tested against two negative controls and showed that the use of -specific information made it possible to screen for functional ncRNAs in this reduced genome. The pseudouridinylation guide H/ACA box snoRNAs have a common secondary structure consisting of two parallel hairpins linked by a hinge. Two conserved motifs box H (ANANNA) and box ACA are located at the hinge and the 3′ tail, respectively, together with the flanking helix, they play important roles in box H/ACA snoRNA accumulation (). However, compared to the single continuous antisense elements in box C/D snoRNAs, the antisense elements of H/ACA box snoRNAs are very short and bipartite (). Almost all the H/ACA box snoRNA adopt the two hairpin model, except one small H/ACA box snoRNA containing only one hairpin described in (). Based on hallmark sequences and structural features, one of the identified potential novel ncRNA (candidate 16, Supplementary Data), is likely to represent a novel H/ACA box snoRNA. It features a single, long stem positioned upstream from the ACA box motif as shown in b. As such, it is strongly reminiscent of archaeal and Trypanosomal H/ACA box snoRNAs, that also feature a single hairpin (). In agreement with the rules applying to eukaryotic H/ACA snoRNAs, the targeted uridine is separated from the H/ACA box by 9–16 nt. Therefore, according to structural modelling, we predict that candidate_16 may guide a pseudouridylation in LSU rRNA. The ribozyme RNase P cleaves the 5′-end of pre-tRNAs. The RNase P RNA was recently identified by sequence similarity search and the RNase P holoenzyme was purified (), and showed that RNase P RNA has the conserved eukaryotic RNase P core structure, and shared extensive similarity with the RNase P RNA of the microsporidian . Both RNAs lack the conserved P3 helix bulge loop, which has been found in all the other eukaryotes studied so far. The RNase P RNA has been found in our library (candidate 9), but surprisingly, the sequence was not terminated at the previously predicted 3′ end, and extended further into the GlsR15 snoRNA (). These two known RNAs have a 24 nt overlap, which is shown in . It is likely that candidate 9 is part of a full-length RNA transcript. To verify this idea, RT-PCR was done using an upstream primer (testP/GlsR15_For) that binds within the RNase P sequence (position 34–53 on the possible full length transcript) and a downstream primer (testP/GlsR15_Rev) which binds within the GlsR15 snoRNA sequence (position 269–289 on the possible full length transcript). RT-PCR results (data not shown) indicate that the RNase P and GlsR15 are indeed transcribed as a single transcript. This rises to a question that whether this transcript is a single functional RNA molecule, or a precursor to give two different RNAs. Structural studies () indicate that the shorter transcript could fold with conserved eukaryotic RNase P motifs. Therefore, the second assumption is preferred. It is possible that an as yet unknown ribonuclease is involved in producing two different RNAs from one precursor. However, this leads to a result that only one of the two RNAs can be generated as a full-length molecule and the other one will be non-functional. A fragment of the variant surface protein (VSP) mRNA was found in the cDNA library. It has been suggested () that antisense regulation controls the expression of VSP genes, and the function of RNA-dependent RNA polymerase (RdRp) is involved to restrict the VSP gene repertoire to a single gene at any one time. Careful sequence mining within the genome observed that there were many tandem repeats sharing short sequence fragments, and these fragments are often complementary to repeated sequences in VSP genes and cysteine-rich protein genes. Blasting a VSP-fragment sequence found in our cDNA library against the genome yielded a potentially functional antisense element. This sequence is a long tandem repeat consisting of nine units, each containing one fragment complementary to the VSP ORF (). RT-PCR was carried out targeting both the ‘+’ and ‘−’ strand of this sequence, and the results showed that both strands were transcribed, to give a double-stranded RNA product. Unlike other tandem repeats of retrotransposons such as LINEs or SINEs, this tandem repeat shows no feature of any known retrotransposon. In comparison, there have been a few studies on unusual repeated sequences in : one study () showed a non-LTR element with site-specific tandem insertions in a chromosomal DNA repeat, and suggested that this element was unlikely to have evolved site specificity unless it did have a function. Another more recent study showed this element was transcribed into a dsRNA (). In addition, there are 22 antisense transcripts identified in the genome (); however, there are no known functions of these transcripts. Our study has revealed a surprising feature shared by two tandem repeats in : one repeat is the experimentally verified dsRNA with fragments complementary to the VSP (Rep-1); and the other is the non-LTR element Genie-1 (). A partial sequence from each element was amplified by PCR with T7-promoter attached primers. The PCR products were transcribed by T7 RNA polymerase to produce dsRNAs. As a control, a single stranded Rep-1 RNA was also produced by elimination of T7 promoter sequence from the reverse primers. Both dsRNAs underwent one self-cleavage at roughly the middle of the sequence (under a basic assay condition with Mg added to water or buffer) (a). The single stranded Rep-1 control did not cleave (b). Timing Mg titration (c) assay and divalent ion assays (d) were performed with the Genie-1 dsRNA. Results showed that the self-cleavage did not happen when Mg concentration was below 1 mM; and self-cleavage only happened at the present of Mg or Co, while Mn and Ca did not have any effect. In addition, addition of EDTA prevented Mg induced cleavage. Further investigation will be necessary to analyse this unusual phenomenon. The aim of this study was to explore the variety of ncRNAs in and obtain a view of ncRNA expression in this genomically reduced deep-branching eukaryote. The scale of this cDNA library is small compared with equivalent studies of ncRNAs in other organisms (). However, studying on a relatively small scale can help getting a comprehensive view of the special features and conserved patterns within this organism, before any large scale studies are attempted. There were previously no systemic studies on the ncRNA composition of . As an extant group of eukaryotes, Diplomonads share very low sequence homology with other eukaryotes, which makes characterization of RNAs extremely difficult. From the 31 novel ncRNA candidates, only 3 can be identified by homology searching as C/D box snoRNAs, the rest have little similarities to known types of ncRNAs. However, comparing the 18 characterized C/D box snoRNAs from has shown that these snoRNAs still share the basic conserved features seen with snoRNAs from other eukaryotes. This makes a computational screen possible. Within the computationally identified putative snoRNAs, we recovered 13 out of our control set of the 18 experimentally characterized snoRNAs. Snoscan-G used looser parameters than the original Snoscan program in order that the experimentally identified snoRNAs (13 in this study) were included in the results. This ensured the sensitivity of the algorithm which was then used for a whole-genome search. However, the large number of false positive hits obtained from the negative control search on a random database, indicated the requirement for other post-scan filtering of putative snoRNA sequences using data unable to be included in the Snoscan-G software. Also, a fairly large result obtained from scan of the yeast genome confirms that the parameter settings for Snoscan-G are less stringent than the original Snoscan program. Comparing putative snoRNA sequences against the ORF database excluded most of the first-round positive hits, and information from genomic locations of the sequences extended the reliability of the putative snoRNAs. Possibly due to its reduced genome, 's snoRNAs are less conserved than those of other eukaryotic organisms; therefore it was necessary to apply less stringent searching criteria. This is because there are as yet no additional -specific sequence features, which can be incorporated into the algorithm. This explains the increase in false positives when large databases are screened. However, combining several filtering steps dramatically reduced the number of positive hits, and at the same time did not result in the loss of any true positives. The remaining putative snoRNAs showed greater similarities to the experimentally identified snoRNAs than the first-round Snoscan-G results before post-scan filtering. Therefore, our computational approach is reliable when used in parallel with an experimental approach speeding up the discovery of novel putative ncRNAs. Blasting the novel RNA candidates against the genome revealed three types of encoding patterns. There is very little known about splicing in . Sequence mining from the genome shows that most of the eukaryotic specific spliceosomal proteins () are present in , as well as the important U5 snRNA (), which functions at the centre of both major and minor spliceosomes. It is common in eukaryotes that the spliceosomal snRNAs are expressed at a high level (), since intron splicing generally occurs at a high rate. However, it seems not the case in . We did not find any sequence in our cDNA library with similarity to any known spliceosomal snRNA. To determine the possible presence of any spliceosomal snRNAs in the library, PCR reaction using the U5 primers (Materials and Methods section) was done on the cDNAs. Results show that U5 snRNA is expressed and present, but in very low quantities. Another puzzling question concerns the U6 snRNA. U6 snRNA is the most conserved spliceosomal snRNA across all the eukaryotes studied to date. U6 snRNAs take part in the actual catalysis during splicing (), and share extensive sequence similarities across eukaryotes. In an early study (), it has been shown that a single pair of PCR primers could detect U6 snRNAs from 17 different species of eukaryotes. As a trial, the same pair of primers was tried on in both genomic PCR and RT-PCR reactions. Despite extensive effort, there is as yet no detectable candidates for a U6 snRNA. It is therefore concluded that our current approach is not powerful enough to solve the puzzle of 's spliceosomal snRNAs. Total 26 out of our 31 novel RNA candidates cannot yet be extensively characterized as belonging to any known class of ncRNA; a feature seen in other species-specific studies (). Structural studies and motif analysis of these RNAs did not show distinct features found in known ncRNAs. A number of these RNAs are GC rich, providing a basis for strong helical structures. Lack of characterization is possibly due to the highly divergent sequences of compared to those of the major eukaryotic groups, and because most computer programs developed for identifying ncRNAs are based on human and yeast. One way to further approach the identification of ncRNA is through more computational studies by incorporating more -specific information into the existing programs, followed by experimental verification of our proposed candidates. Another way is through biochemical studies of central protein components of various RNA processing pathways. These are to be investigated in the future. In conclusion, our cDNA library successfully uncovered 31 novel ncRNAs from , and our computational approach was shown to be a useful method that worked well in parallel with an experimental approach to aid discovery of 60 potential putative snoRNAs in a deep-branching eukaryote. Although it is hard to characterize each candidate ncRNAs found from the cDNA library due to sequence divergence, as far as we can tell, has quite typical eukaryotic RNA processing despite being reduced and with many introns lost. The transcriptional patterns seen in these ncRNAs may help in understanding the mechanism of RNA processing. Future work will continue to be done in investigating the unusual properties of ncRNAs by combined biochemical and computational methods. p p l e m e n t a r y D a t a a r e a v a i l a b l e a t N A R O n l i n e .

The occurrence of non-canonical nucleosides in RNA has been well-characterized (), with ∼100 modified bases having been found in transfer RNA (tRNA) alone (). Modifications occur at the post-transcriptional level, where some modifications are more simple chemical transformations (e.g. methylation) and still others are more complex (e.g. transglycosylation). One enzyme that performs a complex RNA modification (hypermodification) is tRNA-guanine transglycosylase (TGT, EC 2.4.2.29), catalyzing the exchange of the modified base queuine for the anticodon loop wobble position guanine in eukaryl and eubacterial tRNA (). The proposed biochemical pathway for queuine incorporation in eubacterial tRNA is shown in . Although not yet fully understood, it is known that four genes are involved in the eubacterial biosynthesis of the queuine precursor, preQ (). PreQ is incorporated into the tRNA by TGT (). Two subsequent enzymes convert preQ to queuine in the tRNA (). In contrast, eukaryl organisms acquire queuine from external routes such as diet, and this heterocyclic base is incorporated directly by the eukaryl TGT (,). TGT plays a vital role in the pathogenesis of shigellosis (), a disease that causes severe dysentery in humans. The pathogenic strain, , infects the cells of the human gastrointestinal tract following evasion of the host immune system defense mechanisms, such as the engulfment of foreign substances by macrophages. is able to escape from the macrophage endosome via the expression of certain virulence factors (). There are several bacterial genes (including and ) that mediate this escape as well as cell-to-cell spread of the organism (). VirF, encoded on the primary pathogenicity island of , is a potent transcriptional regulator of the AraC family that regulates the expression of these virulence factors () (,). There are several environmental factors that promote the expression of VirF, including oxygen and iron limitation, temperature, pH, osmolarity and post-transcriptional RNA modification (). Durand and Björk () have demonstrated a positive correlation between VirF and TGT by characterizing a mutant strain with an inactivated gene (termed ). In this mutant, the translation of VirF was markedly reduced whereas the levels of mRNA remained unchanged, and as a result, the bacteria were unable to invade host cells. In addition, when transformed with a plasmid containing a functional gene, restoring queuine modification, the mutants exhibited both restored VirF expression and virulence, thus demonstrating a positive connection between mRNA translation and the presence of active TGT (and presumably queuine modification). The role of modified nucleosides in RNA structure and stability has been well-studied (). Agris and Brown () have identified key interactions between modified nucleosides and magnesium ions essential to the secondary structure of tRNA, in addition to facilitating key RNA–protein interactions. Mandal . (,) have characterized binding of small molecules and metabolites to RNA motifs termed ‘riboswitches’. Riboswitches are structural motifs in mRNAs [sometimes in the 5′ untranslated region (UTR) and extending into the start of the open reading frame] that can exist in at least two stable conformations. One of these conformations is stabilized by binding to a small molecule, thus altering the equilibrium between the conformations. One conformation supports translation of the protein while with the other conformation, translation is blocked. In this way, binding of the small molecule changes the conformation of the RNA and modulates its translation. In addition to the 5′ UTR of prokaryotic mRNAs, riboswitches have also been found in the 3′ UTR and introns in several eukaryotic species (). Interestingly, the Breaker lab has just reported the discovery of a riboswitch that responds to the queuine precursor, preQ and appears to regulate the expression of the four genes that are involved in preQ biosynthesis (). It certainly seems possible that base modification of an mRNA could modulate a similar conformational change. It therefore is feasible that such a control mechanism for gene expression might be involved in regulating the expression of virulence factors in pathogenic organisms, as is apparently seen with VirF. The incorporation of modified nucleosides has been characterized more fully for tRNA (and some other RNAs, e.g. rRNA and snRNA), than for mRNA (,,). The most common example of post-transcriptional processing in mRNA is the eukaryotic 7-methylguanosine 5′ cap structure, which aids in the binding to the small ribosomal subunit and is essential for the efficient synthesis of eukaryotic proteins (). To date, the only known function of TGT is to catalyze the modification of tRNA with queuine. Previous work has shown that the eubacterial TGT will recognize a U-G-U sequence in the loop of an RNA hairpin structure that corresponds to the anticodon stem–loop of its cognate tRNAs (,). The eubacterial TGT will also recognize a U-G-U containing hairpin in the context of a dimeric form of a cognate tRNA (). It is conceivable that an mRNA may also be modified directly by TGT, provided the mRNA contained the appropriate recognition elements. An examination of the sequence of mRNA for the presence of U-G-U sequences revealed six unique U-G-U sites (). Mfold analysis of the regions surrounding each of these U-G-U sequences reveal that nucleotides 410–433 could possibly fold into a hairpin structure with the U-G-U sequence in a position in a loop that is analogous to the anticodon loop of TGT-cognate tRNAs. As a first step towards probing the possibility that TGT may modulate the translation of VirF via modification of the mRNA, Michaelis–Menten kinetic analyses were conducted to probe this modification by TGT . We report that the TGT, which has 99% sequence identity to the TGT, does indeed recognize the mRNA as a substrate . Further, we show that this recognition results in the site-specific modification of a single base in the mRNA. Unless otherwise specified, all reagents were ordered from Sigma or Aldrich. DNA oligonucleotides, agarose, dithiothreitol (DTT), T4 DNA ligase and DNA ladders were ordered from Invitrogen. All restriction enzymes and Vent® DNA polymerase were ordered from New England Biolabs. The ribonucleic acid triphosphates (NTPs) and pyrophosphatase were ordered from Roche Applied Sciences. The deoxyribonucleic acid triphosphates (dNTPs) were ordered from Promega. Low-melting Seaplaque agarose was ordered from Cambrex. Gelase™ Enzyme Prep, MasterAmp™ High Fidelity RT–PCR Kit, and Scriptguard™ RNase Inhibitor were ordered from Epicentre. Epicurian coli® XL2-Blue ultracompetent cells were ordered from Stratagene. Amicon Ultra Centrifugal Filter Devices were ordered from Millipore. Whatman GF/C Glass Microfibre Filters and all bacterial media components were ordered from Fisher. The QIAPrep® Spin Miniprep Kit was ordered from Qiagen. Tris–HCl Buffer was ordered from Acros Organics. [H] PreQ was ordered from American Radiolabeled Chemicals, Inc. T7 RNA polymerase was isolated from BL21 cells containing the plasmid pBH161 according to the procedure of Prof. William McCallister, State University of New York, Brooklyn. The TGT was isolated with an amino terminal his-tag as previously described (). The cyano precursor (preQ, ) was synthesized according to the method of Migawa . () by the condensation of chloro(formyl)acetonitrile and pyrimidine (). Reduction of the cyano precursor with tritium gas gave the desired radiolabeled substrate preQ () with a specific activity of 500 mCi/mmol (). The tritium reduction was performed commercially by American Radiolabeled Compounds Co. The plasmid pBDG302, containing the gene, was received from Prof. Glenn Björk (Umeå University, Sweden). The gene was amplified from the plasmid by polymerase chain reaction (PCR) under the following conditions: primers (20 pmol each), pBDG302 (500 ng), Mg (2 mM), dNTPs (1 mM each), Vent DNA polymerase (4 U), brought to a final volume of 50 μl with deionized water. The sample was treated with 30 PCR cycles of the following sequence: 94°C (1 min), 50°C (1 min), and 72°C (2 min), followed by a final extension at 72°C (5 min). Following a double restriction enzyme digest with PstI and EcoRI (40 U each, 20-μl reaction) for 1 h at 37°C, the PCR product and vector were gel-purified from Seaplaque agarose with Gelase™ according to the vendor protocol. The purified gene was then ligated into digested pTZ19R (5:1 volume ratio, 20 μl reaction) following overnight incubation with T4 DNA ligase (2 U) at 17°C. The ligated sample (10 μl) was transformed into 100 μl of Epicurian coli® XL2-Blue ultracompetent cells according to the Stratagene protocol. Cells were grown overnight at 37°C on L-Amp plates (50 μg/ml ampicillin). Individual colonies were isolated, and 3 mL 2xTY (16 g Bactotryptone, 10 g yeast extract, 5 g NaCl/liter of water with 50 μg/ml ampicillin) liquid cultures were inoculated at 37°C with shaking. Plasmid was isolated via miniprep, and the gene sequence was confirmed with DNA sequencing (University of Michigan DNA Sequencing Core Facilities). transcription reactions with pTZvirF were conducted by first linearizing the plasmid at the end of the sequence with the restriction enzyme EcoRI (40 U/100 μl DNA, 500 μl reaction). The sample was ethanol precipitated at −20°C, and the pellet was re-suspended in 250 μl of deionized water. 1 ml transcription conditions were as follows: pTZvirF template (100 μl), transcription buffer (4 mM Tris–HCl, pH 8.0; 2 mM MgCl, 0.5 mM DTT, 0.1 mM spermidine), NTPs (4 mM each), T7 RNA polymerase (2500 U), inorganic pyrophosphatase (2 U) and RNase inhibitor (200 U). The reaction was incubated at 37°C for ∼4 h. The reaction was stored at −20°C following transcription. Best results were obtained when the 1 ml reaction was prepared and redistributed into 100 μl volumes prior to incubation at 37°C. The MasterAmp ™ High Fidelity RT–PCR Kit was used according to vendor protocol to generate DNA, which was confirmed with sequence analysis of the DNA product (). Analysis of the energetically favorable secondary structures within the mRNA sequence was performed using the biophysical web tool Mfold (M. Zuker, ). Sequences of ∼10 nucleotides surrounding either side of the six possible recognition motifs were analyzed by the web tool, and the hairpin structure determined to be most favorable was found between nucleotides 410–433 in the mRNA, which contains the potential recognition sequence UGU. The RNAture Oligonucleotide Analyzer web tool was used to predict a Tm of 71°C for the minihelix hairpin. (Note: This web tool appears to be no longer available.) An Expedite™ Nucleic Acid Synthesis System was used to synthesize this 24-nucleotide sequence (5′-GGAGUAGUCUUUGUCGACUAUUUU-3′) using the vendor's protocols for the synthesis of RNA at the 1 μmol scale. The reagents were from Perkin–Elmer and the RNA amidites were from Glen Research. The extinction coefficient calculated for this RNA minihelix was ε = 265.3 OD/mM. The single nucleotide mutation of guanine 421 to adenine in the mRNA sequence was generated via QuikChange site-directed mutagenesis (Stratagene), producing the new vector pTZvirF(GA). The reactions conditions were as follows: complimentary oligonucleotides with desired mutation (175 ng), pTZvirF(wt) template (800 ng), dNTPs (0.25 mM) and Vent DNA polymerase (2 U), brought to a final volume of 30 μl with deionized water. The sample was treated with 25 PCR cycles of the following sequence: 94°C (30 s), 50°C (1 min), and 72°C (6.5 min). The PCR product was then incubated for 2 h at 37°C with Dpn I (40 U), and addition of NE Buffer 4 was required for proper digestion of wild-type plasmid. The digested sample (10 μl) was transformed into 100 μl of Epicurian coli® XL2-Blue ultracompetent cells according to the Stratagene protocol. Cells were grown overnight at 37°C on L-Amp plates (50 μg/ml ampicillin). Individual colonies were isolated, and 3 ml 2xTY (16 g bactotryptone, 10 g yeast extract, 5 g NaCl/liter of water with 50 μg/ml ampicillin) liquid cultures were inoculated at 37°C with shaking. Plasmid was isolated via miniprep, and the mRNA(GA) mutation was confirmed with DNA sequencing (University of Michigan DNA Sequencing Core Facilities). Assays were conducted by monitoring the incorporation of radiolabeled substrate, [H] preQ, into tRNA, dGECYMH (5′-GGGAGCAGACUdGUAAAUCUGCUCCC-3′) and various mRNA substrates. Samples from transcriptions were concentrated with Amicon Ultra Centrifugal Filters (10,000 MWCO). The concentration of mRNA was determined with a Cary UV-Visible Spectrophotometer, where the approximate concentration of a single-stranded RNA sample A = 1 OD is 40 μg/ml and the molecular weight of mRNA is 252 g/mmol mRNA. In brief, kinetic assays were set up under the following conditions: RNA substrate (various concentrations), [H] preQ (10 μM, 296mCi/mmol stock), TGT (100 or 200 nM, as specified), and HEPES reaction buffer (100 mM HEPES, pH 7.3; 20 mM MgCl; 5 mM DTT) to a final volume of 400 μl. All samples were incubated at 37°C for purposes of equilibration before initiating the reaction with the addition of TGT. Aliquots (70 μl) were removed at various points throughout the reaction and quenched in 3 ml of 5% TCA for 1 h before filtering on glass-fiber filters. Each filter was washed with three volumes of 5% TCA and a final wash of ethanol to dry the filter. The samples were analyzed in a scintillation counter (Beckman) for radioactive decay, where counts were reported in DPM and later converted to picomoles [H] preQ by the following conversion (pmol = DPM × 0.00152, for the [H] preQ stock with a specific activity of 296 mCi/mmol). Initial velocities were determined by converting the slopes of these plots (pmol/min) to units of second, taking into account the concentration of the enzyme and aliquot size. The individual data points from each trial were averaged, and the standard deviation was determined for each concentration of RNA substrate. The average data points (with error bars representing their standard deviations) were plotted. However, all of the individual data points were fit via non-linear regression to the Michaelis–Menten equation and the line for that fit is displayed (). All non-linear regression fits with the Michaelis–Menten equation were determined using Kaleidagraph (Abelbeck Software). To provide micromolar quantities of mRNA for our studies, we generated an transcription clone for the mRNA. The gene was subcloned from the plasmid pBDG302 containing the gene (a gift from Professor Glenn Björk, Umeå University, Sweden) into a plasmid suitable for transcription, generating pTZvirF. mRNA was synthesized via run-off transcription following digestion with EcoRI, linearizing pTZvirF at the end of the gene sequence. The mRNA was physically characterized on an ethidium bromide stained, 1.2% formaldehyde agarose gel.RT–PCR was utilized to generate dsDNA from the transcription product using the same oligonucleotide primers initially designed for subcloning of the gene. Examples of formaldehyde and TAE agarose gels of the mRNA and the dsDNA from the RT–PCR are shown in . The single-stranded mRNA appears to run on the gel at ∼500 bp in comparison to the dsDNA ladder. The size of the gene is 789 bp, and the corresponding mRNA is 789 nucleotides in length. We hypothesize that the mRNA is running at a lower ‘apparent’ molecular weight due to the propensity of mRNA to adopt a variety of conformations, even in an agarose gel. This would explain why the observed molecular weight is a little larger than one half the size of the double-stranded DNA. The RT–PCR product (dsDNA) migrated to the anticipated molecular weight for a DNA sample of ∼800 bp. DNA sequencing of this product matched the mRNA sequence (data not shown). For comparison, Michaelis–Menten kinetic analyses were conducted with the natural RNA substrate tRNA (ECY) and the modified minihelix substrate dGECYMH (the anticodon stem–loop of ECY where the guanosine at position 34 contains a 2′-deoxyribose) with [H] preQ. It has been shown previously that a minihelix RNA consisting of the anticodon arm and loop of a queuine-cognate tRNA is a sufficient substrate for TGT (). Aliquots were taken at various time points over a 15 min incubation of 100 nM TGT, various concentrations of ECY (0.05–1.5 μM) or dGECYMH (0.05–5 μM) and saturating concentrations of [H] preQ. The kinetic constants determined for the incorporation of [H] preQ with ECY and dGECYMH are shown in comparison with the kinetic data for the substrates in . Using the same approach described above, Michaelis–Menten kinetic analyses were conducted with mRNA. Aliquots were taken at various time points over a 1 h incubation of 200 nM TGT, various concentrations of mRNA (0.1–10 μM), and saturating concentrations of [H] preQ (A). Higher concentrations of mRNA were tested to obtain an accurate kinetic profile by characterizing the reaction over a large range of concentrations. In addition to characterizing the wild-type mRNA, a minihelix RNA ( MH) corresponding to the 410–433 hairpin sequence (underlined in ) as well as a full-length mRNA mutant (GA) were studied. The kinetic analyses were performed with 100 nM TGT, various concentrations of MH (0.1–10 μM) and saturating concentrations of [H] preQ (B). The full-length mRNA(GA) was incubated under the same conditions as the wild-type mRNA, but only at concentrations corresponding to and 5x, as determined from the kinetic constants of mRNA(wt) (). A ‘no RNA’ control was also included to determine the background level of radioactivity present in the samples (C). The data were fit by non-linear regression. Both the full-length mRNA(wt) and the minihelix exhibited RNA concentration-dependent incorporation of [H] preQ over time, and the Michaelis–Menten equation provided a good fit for the data. The full-length mRNA(GA), which is full-length mRNA with a single nucleotide mutation at guanine 421, was analyzed at both 2 μM (∼) and 10 μM (∼5 × ) mRNA. The mRNA(GA) mutant showed no detectable activity greater than the ‘no RNA’ control (C). The kinetic constants determined for the mRNA substrates are shown in . Both the mRNA and MH have values in the low micromolar range, even though and for both is lower than the corresponding values for the ECY substrates. VirF is a critical transcriptional regulator responsible for activating virulence genes in . Durand and colleagues () demonstrated the involvement of TGT in modulating the translation of VirF via the observation that a mutant strain of with an inactivated gene (termed showed decreased virulence. VirF protein levels were dramatically lower in the mutant as compared to wild-type, but the mRNA levels showed no detectable difference from wild-type . The lack of VirF protein resulted in a reduction of all downstream virulence gene expression, and thus exhibited a less virulent phenotype than that of the wild-type bacterium. When transformed with a plasmid encoding the gene, both VirF translation and virulence were restored. It had previously been shown that the presence of modified nucleosides enhances translation (,). However, other studies have shown that growth rate and protein translation as a whole are not directly affected by a lack of queuine-modified tRNA (). While this interesting correlation between VirF translation and TGT activity has been known for some time, yet the exact role TGT plays in the translation of this primary virulence factor remains unclear. Our laboratory has previously demonstrated that TGT can modify substrates with more unusual structures than a canonical tRNA fold. We reported that a dimeric form of the ECY serves as a substrate for TGT, with a slightly higher and identical , relative to the normal tRNA (). It had previously been shown from NMR studies that the anticodon arms of the dimer subunits were intact and pointing away from the center of the dimer (). Those studies demonstrate that TGT can recognize a minihelix containing the requisite U-G-U sequence even in the context of a larger RNA structure. TGT is not the first tRNA modification enzyme to demonstrate recognition of alternative RNA structures. Gu and coworkers () have shown that, , the modification enzyme tRNA (mU54)-methyltransferase will methylate 16S ribosomal RNA from in addition to its physiological tRNA substrate, although they found no evidence for this occurring . The enzyme pseudouridine synthase catalyzes the isomerization of specific uridines to the modified nucleoside ψ. A single pseudouridine synthase has been shown to have ‘dual specificity’, recognizing and modifying both tRNA and snRNA (). Ofengand and colleagues (,) have also characterized a critical pseudouridine synthase responsible for site-specific modification with pseudouridine for both tRNA and 23S rRNA in . Such precedence for tRNA modification enzymes to recognize and modify other RNA species and , suggests that mRNA modification mediated by TGT may also occur. As a first step to probe for this possibility, we examined the mRNA sequence for the presence of a U-G-U sequence in a TGT recognition motif. Of the six U-G-U sequences in the mRNA, the one involving bases 410–433 () was predicted by Mfold analysis to be able to fold into a hairpin structure possibly suitable for recognition by TGT. Incubation of mRNA with TGT and radiolabeled preQ revealed that the mRNA is indeed a substrate for TGT . The kinetic parameters (; ) for mRNA were determined from a Michaelis–Menten analysis (A). The mRNA substrates exhibit the same trend in kinetic parameters as the ECY substrates, where the minihelix substrates (dGECYMH and MH) have slightly higher values for both and with respect to the corresponding full-length RNA substrates (). The values of for both ECY and mRNA are very similar, both in the low micromolar range. This is encouraging considering the size and structural difference between the two substrates, where ECY is ∼80 nucleotides in length with a very well-defined tertiary structure common to tRNA and the mRNA is . 800 nucleotides in length and presumably does not have a compact tertiary structure. The Michaelis–Menten plot for the mRNA (A) fits to a single , consistent with a single site of modification within the mRNA. The for the mRNA is ∼ 40-fold lower than that for tRNA. It has previously been shown that altering the position of the U-G-U sequence within the minihelix loop of cognate tRNAs is correlated with a reduction in activity (). All of the biochemical and structural data previously reported is consistent with a covalent intermediate, via Asp264, in the TGT reaction (,). The hairpin loop of the ECY substrate contains seven nucleotides, whereas the hairpin in the mRNA substrate contains only six nucleotides in the loop. This difference in loop length may result in a suboptimal orientation of the guanosine ribose in the U-G-U loop in the mRNA, making nucleophilic attack by Asp264 less likely. This difference in loop length and orientation could account for the reduced that we have observed for the mRNA substrates. Our analysis of the mRNA sequence predicts that there should be a single site of modification, guanine 421. We have taken two approaches to investigate this. In our first approach, a MH corresponding to the predicted hairpin structure within the native mRNA sequence (bases 410–433), was chemically synthesized (B). The stem consists of nine base pairs, where the first four nucleotides in the stem are uridine residues forming wobble pair interactions with three guanines and one Watson–Crick pair with an adenosine. At first glance, the stability of a minihelix with three G-U wobble pairs might be questionable; however, in the context of the mRNA, the ends of the helix may be held in close proximity by other intramolecular interactions. Additionally, the MH by itself has a predicted melting temperature of 71°C, indicating the structure should be stable at physiological and assay temperatures. The MH is a substrate for TGT . The values for both the full-length mRNA and MH (1.8 and 0.87 μM, respectively; ) are very similar, suggesting that the minihelix structure is likely a predominant conformation in the mRNA. Although the recognition of the minihelix by TGT is consistent with the mRNA serving as a substrate for TGT, it does not provide conclusive evidence that guanine 421 is the site of modification in the mRNA. There are six UGU sequences within the mRNA that are possible recognition sites for TGT. Therefore, our second approach was to construct the point mutation, GA, in the full-length mRNA to demonstrate the importance of guanine 421. The mRNA(GA) mutant resulted in a complete loss of activity at two different concentrations of RNA (C), indicating that G is indeed essential for recognition by TGT. Had a second exchangeable guanine existed in the sequence, we would have expected to see a decreased or possibly even unchanged activity of the mRNA. The relationship between the ‘no RNA’ negative control and the mRNA(GA) indicates that guanine 421 is the only exchangeable nucleotide in the mRNA sequence (at least within the concentration ranges tested), and that the kinetic parameters observed for mRNA(wt) are due to specific recognition by TGT and could not be attributed to non-specific interactions with this large nucleic acid molecule. It should be noted that, under the conditions of the assay (C), it appears that the enzyme is undergoing a limited number of turnovers. Two factors may be contributing to this. The first is that our calculations of kinetic parameters assume 100% active enzyme, which is almost certainly an over-estimate. The active enzyme concentration may be as much as 2-fold lower as recent studies suggest that the eubacterial TGT may exist as a homodimer with ‘half-of-sites’ reactivity (). This would effectively double the turnovers per active site. Second, the off-rate for the modified mRNA may be sufficiently slower (relative to that for tRNA) such that the turnover rate may indeed be significantly slowed under these conditions. It remains to be seen if these observations hold under conditions. From the results presented herein, it is clear that the mRNA does act as a substrate for the eubacterial TGT . Although there are six possible UGU recognition motifs, both the mutagenesis and mRNA minihelix studies are consistent with G serving as the sole site of modification within the mRNA. With a value in the low micromolar range, it is very possible that the modification of mRNA may be biologically relevant (e.g. may occur ). These results provide the first ‘proof of principle’ evidence that post-transcriptional RNA modification may regulate mRNA function, as it has long been recognized to do for tRNA. The recent work characterizing the preQ riboswitch revealed that it is a fairly simple hairpin structure (the simplest riboswitch structure characterized to date) (). Such a simple structure could feasibly occur in the coding region of an mRNA species. In fact, the TGT modification site that we have discovered in the mRNA is predicted to occur in a simple hairpin structural motif. Base modification in a hairpin structure in the coding sequence of mRNA could induce a similar structural switch as seen in the preQ riboswitch and thereby influence translation of VirF. Studies to determine the physiological significance of the mRNA modification by TGT that we have observed are currently in progress.