Scaling-up NLP Pipelines to Process Large Corpora of Clinical Notes.

INTRODUCTION: This article is part of the Focus Theme of Methods of Information in Medicine on "Big Data and Analytics in Healthcare". OBJECTIVES: This paper describes the scale-up efforts at the VA Salt Lake City Health Care System to address processing large corpora of clinical notes through a natural language processing (NLP) pipeline. The use case described is a current project focused on detecting the presence of an indwelling urinary catheter in hospitalized patients and subsequent catheter-associated urinary tract infections. METHODS: An NLP algorithm using v3NLP was developed to detect the presence of an indwelling urinary catheter in hospitalized patients. The algorithm was tested on a small corpus of notes on patients for whom the presence or absence of a catheter was already known (reference standard). In planning for a scale-up, we estimated that the original algorithm would have taken 2.4 days to run on a larger corpus of notes for this project (550,000 notes), and 27 days for a corpus of 6 million records representative of a national sample of notes. We approached scaling-up NLP pipelines through three techniques: pipeline replication via multi-threading, intra-annotator threading for tasks that can be further decomposed, and remote annotator services which enable annotator scale-out. RESULTS: The scale-up resulted in reducing the average time to process a record from 206 milliseconds to 17 milliseconds or a 12- fold increase in performance when applied to a corpus of 550,000 notes. CONCLUSIONS: Purposely simplistic in nature, these scale-up efforts are the straight forward evolution from small scale NLP processing to larger scale extraction without incurring associated complexities that are inherited by the use of the underlying UIMA framework. These efforts represent generalizable and widely applicable techniques that will aid other computationally complex NLP pipelines that are of need to be scaled out for processing and analyzing big data.

The Dim protein family: from structure to splicing.

The spliceosome is a dynamic macromolecular machine that catalyzes pre-mRNA splicing through a mechanism controlled by several accessory proteins, including the Dim proteins. The Dim protein family is composed of two classes, Dim1 and Dim2, which share a common thioredoxin-like fold. They were originally identified for their role in cell cycle progression and have been found to interact with Prp6, an essential component of the spliceosome, which forms the bridge of U4/U6.U5-tri-snRNP. In spite of their biological and structural similarities, Dim1 and Dim2 proteins differ in many aspects. Dim1 bears distinctive structural motifs responsible for its interaction with other spliceosome components. Dim2 forms homodimers and contains specific domains required for its interactions with partners. This originality suggests that although both proteins are involved in pre-mRNA splicing, they are likely to be involved in different biological pathways. In the present article we review the structure and function of the Dim proteins.

A non-covalent peptide-based strategy for siRNA delivery.

The major obstacle to clinical development of siRNAs (short interfering RNAs), like for most of the nucleic-acid-based strategies, is their poor cellular uptake and bioavailability. Although several viral and non-viral strategies have been proposed to improve siRNA delivery, their applications in vivo remain a major challenge. We have developed a new strategy, based on a short amphipathic peptide, MPG, that is able to form stable nanoparticles with siRNA. MPG-based particles enter the cell independently of the endosomal pathway and can efficiently deliver siRNA in a fully biologically active form into a variety of cell lines and in vivo. This short review will discuss the mechanism and the potency of the MPG strategy for siRNA delivery both in vitro and in vivo.

Cell-penetrating peptides: tools for intracellular delivery of therapeutics.

The main problem of therapeutic efficiency lies in the crossing of cellular membranes. Therefore, significant effort is being made to develop agents which can cross these barriers and deliver therapeutic agents into cellular compartments. In recent years, a large amount of data on the use of peptides as delivery agents has accumulated. Several groups have published the first positive results using peptides for the delivery of therapeutic agents in relevant animal models. These peptides, called cell-penetrating peptides (CPPs), are short peptides (fewer than 30 residues) with a net positive charge and acting in a receptor- and energy-independent manner. Here, we give an extensive review of peptide-mediated delivery systems and discuss their applications, with particular focus on the mechanisms leading to cellular internalization.

Combination of a new generation of PNAs with a peptide-based carrier enables efficient targeting of cell cycle progression.

The design of potent systems for the delivery of charged and noncharged molecules that target genes of interest remains a challenge. We describe a novel technology that combines a new generation of peptide nucleic acids (PNAs), or HypNA-pPNAs, with a new noncovalent peptide-based delivery system, Pep-2, which promotes efficient delivery of PNAs into several cell lines. We have validated the potential of this technology by showing that Pep2-mediated delivery of an antisense HypNA-pPNA chimera directed specifically against cyclin B1 induces rapid and robust downregulation of its protein levels and efficiently blocks cell cycle progression of several cell lines, as well as proliferation of cells derived from a breast cancer. Pep-2-based delivery system was shown to be 100-fold more efficient in delivering HypNA-pPNAs than classical cationic lipid-based methods. Whereas Pep-2 is essential for improving the bioavailability of PNAs and HypNA-pPNAs, the latter contribute significantly to the efficiency and specificity of the biological response. We have found that Pep-2/HypNA-pPNA strategy promotes potent antisense effects, which are approximately 25-fold greater than with classical antisense oligonucleotide directed specifically against the same cyclin B1 target. Taken together, these data demonstrate that peptide-mediated delivery of HypNA-pPNAs constitutes a very promising technology for therapeutic applications.

A peptide carrier for the delivery of biologically active proteins into mammalian cells.

The development of peptide drugs and therapeutic proteins is limited by the poor permeability and the selectivity of the cell membrane. There is a growing effort to circumvent these problems by designing strategies to deliver full-length proteins into a large number of cells. A series of small protein domains, termed protein transduction domains (PTDs), have been shown to cross biological membranes efficiently and independently of transporters or specific receptors, and to promote the delivery of peptides and proteins into cells. TAT protein from human immunodeficiency virus (HIV-1) is able to deliver biologically active proteins in vivo and has been shown to be of considerable interest for protein therapeutics. Similarly, the third alpha-helix of Antennapedia homeodomain, and VP22 protein from herpes simplex virus promote the delivery of covalently linked peptides or proteins into cells. However, these PTD vectors display a certain number of limitations in that they all require crosslinking to the target peptide or protein. Moreover, protein transduction using PTD-TAT fusion protein systems may require denaturation of the protein before delivery to increase the accessibility of the TAT-PTD domain. This requirement introduces an additional delay between the time of delivery and intracellular activation of the protein. In this report, we propose a new strategy for protein delivery based on a short amphipathic peptide carrier, Pep-1. This peptide carrier is able to efficiently deliver a variety of peptides and proteins into several cell lines in a fully biologically active form, without the need for prior chemical covalent coupling or denaturation steps. In addition, this peptide carrier presents several advantages for protein therapy, including stability in physiological buffer, lack of toxicity, and lack of sensitivity to serum. Pep-1 technology should be extremely useful for targeting specific protein-protein interactions in living cells and for screening novel therapeutic proteins.

UBA domains mediate protein-protein interactions between two DNA damage-inducible proteins.

The Saccharomyces cerevisiae genes RAD23 and DDI1 were identified in a screen for multicopy suppressors of the temperature-sensitivity of a mutant allele of S. cerevisiae PDS1. Pds1 is a regulator of anaphase that needs to accumulate and then be degraded by the ubiquitin-proteasome pathway at the metaphase-anaphase transition for cells to progress normally through mitosis. Both the Rad23 and Ddi1 pds1 suppression phenotypes depend on a shared motif known as a UBA domain found in a variety of proteins associated with ubiquitin metabolism. UBA domains were found to be essential for homodimerization of Rad23 and heterodimerization between Rad23 and Ddi1, but not for homodimerization of Ddi1. This observation, coupled with the findings that Rad23 and Ddi1 UBA domains bind ubiquitin and that dimerization of Rad23 blocks ubiquitin binding, suggests a possible mechanism for regulating Rad23 and Ddi1 function.

UBA domains of DNA damage-inducible proteins interact with ubiquitin.

Rad23 is a highly conserved protein involved in nucleotide excision repair (NER) that associates with the proteasome via its N-terminus. Its C-terminal ubiquitin-associated (UBA) domain is evolutionarily conserved from yeast to humans. However, the cellular function of UBA domains is not completely understood. Recently, RAD23 and DDI1, both DNA damage-inducible genes encoding proteins with UBA domains, were implicated genetically in Pds1-dependent mitotic control in yeast. The UBA domains of RAD23 and DDI1 are required for these interactions. Timely degradation of Pds1 via the ubiquitin/proteasome pathway allows anaphase onset and is crucial for chromosome maintenance. Here, we show that Rad23 and Ddi1 interact directly with ubiquitin and that this interaction is dependent on their UBA domains, providing a possible mechanism for UBA-dependent cell cycle control. Moreover, we show that a hydrophobic surface on the UBA domain, which from structural work had been predicted to be a protein-protein interaction interface, is indeed required for ubiquitin binding. By demonstrating that UBA domains interact with ubiquitin, we have provided the first indication of a cellular function for the UBA domain.

Translocating peptides and proteins and their use for gene delivery.

A dramatic surge in the development of peptides for gene delivery in vitro and in vivo has been witnessed in the past decade. A better understanding of the structural and mechanistic properties of peptides has been an important step for the rational design of optimal peptide-based gene delivery systems. Research has focused on the design of short synthetic peptides that overcome both extracellular and intracellular limitations of other gene delivery systems by binding reversibly and condensing DNA, specifically targeting cells and/or tissues, rapidly releasing plasmids into the cytoplasm and mediating efficient nuclear translocation.

The SU glycoprotein 120 from HIV-1 penetrates into lipid monolayers mimicking plasma membranes.

Increasing evidence suggests that the HIV envelope binds through its surface (SU) gp120 not only to receptors and coreceptors, but also to other components of the cellular membrane where the glycolipids appear to be good candidates. To assess the ability of HIV-1 SU gp120 to penetrate into phospholipid membranes, we carried out a study of the interactions between a recombinant SU gp120 from HIV-1/HXB2 and artificial lipid monolayers mimicking the composition of the outer leaflet of the lymphocytes and which were spread at the air-water interface. We show that the protein, in its aggregated form, has amphipathic properties and that the insertion of this amphipathic species into lipids is favored by the presence of sphingomyelin. Furthermore, cholesterol enhances the penetration into mixed phosphatidylcholine-sphingomyelin monolayers. Coexistence of different physical states of the lipids and thus of domains appears to play a major role for protein penetration independently of the presence of receptors and coreceptors.

Modulation of the typical multidrug resistance phenotype by targeting the MED-1 region of human MDR1 promoter.

Multidrug resistance of cancer (MDR) is the major cause of failure of chemotherapy. The typical MDR phenotype is due to the overexpression of membrane proteins among which the main representative is P-glycoprotein (Pgp) encoded by the MDR1 gene. Many attempts to modulate MDR by chemosensitizers have been unsuccessful in human therapy due to their intrinsic toxic effects. In an effort to modulate the MDR phenotype efficiently we designed an antisense and a transcriptional decoy strategy targeting the TATA-less human MDR1 gene promoter. The choice of the start point of transcription in a multiple start site window is related to an upstream MED-1 cis-element, the sequence and configuration of which are specific to human MDR1 gene expressed in Pgp-overproducing cancer cells. A 12mer antisense oligodeoxynucleotide (ODN) and a 12mer double-stranded ODN, both containing the MED-1 sequence, were designed and efficiently vectorized into the nucleus with the chimerical MPG peptide. A synthetic cellular model (NIH-EGFP) and highly resistant human CEM/VLB0.45 leukemia cells, significantly responded to transfection with the ODN/MPG complex. The level of EGFP fluorescence in NIH-EGFP cells decreased, and thus its production, and viability of CEM/VLB0.45 cells decreased by 63% in the presence of vinblastine, revealing that their resistance to the anticancer drug was reversed. These results open new insights into transcriptional decoy and anti-gene therapies of MDR cancers that overproduce Pgp. Gene Therapy (2000) 7, 1224-1233.

An essential phosphorylation-site domain of human cdc25C interacts with both 14-3-3 and cyclins.

Human cdc25C is a dual-specificity phosphatase involved in the regulation of cell cycle progression in both unperturbed cells and in cells subject to DNA damage or replication checkpoints. In this study, we describe the structure-function relationship of an essential domain of human cdc25C that interacts with 14-3-3 proteins. We show that this domain is a bi-functional interactive motif that interacts with cyclins primarily through their P-box motif in addition to 14-3-3 proteins. Characterization of the structural features of this domain by NMR and circular dichroism reveals two distinct alpha helical moieties interconnected by a loop carrying the 14-3-3 binding site. Moreover, the helical folding is induced upon binding to 14-3-3, suggestive of a conformational regulation of this domain of cdc25C through interactions with partner proteins in vivo. Combining our structural and biochemical data, we propose a detailed model of the molecular mechanism of cdc25C regulation by differential association with 14-3-3 and cdc2-cyclin B.

The thumb domain of the P51-subunit is essential for activation of HIV reverse transcriptase.

The biologically relevant and active form of human immunodeficiency virus reverse transcriptase is a heterodimer produced in a two-step dimerization process. Dimerization involves first the rapid association of the two subunits, followed by a slow conformational change yielding a fully active form. In the present study, we demonstrate that the interaction between the thumb domain of p51 and the RNase-H domain of p66 plays a major role in an essential conformational change required for proper folding of the primer/template and the tRNA-binding site, for maturation and for activation of heterodimeric reverse transcriptase. A synthetic peptide derived from the sequence within the thumb domain of p51, which forms the interface with the RNase-H domains of p66, binds heterodimeric reverse transcriptase with an apparent dissociation constant in the nanomolar range and selectively inhibits activation of heterodimeric reverse transcriptase with an inhibition constant of 1.2 microM. A detailed study of the mechanism of inhibition reveals that this peptide does not require dissociation of heterodimeric RT for efficient inhibition and does not affect subunit association, but interferes with the conformational change required for activation of heterodimeric reverse transcriptase, resulting in a decrease in the affinity of reverse transcriptase for the tRNA and an increase in the stability of the primer/template/reverse transcriptase complex. We have previously proposed that the dimeric nature of reverse transcriptase represents an interesting target for the design of antiviral agents. On the basis of this work, we propose that the conformational changes involved in the activation of reverse transcriptase similarly represent an important target for the design of novel antiviral compounds.

Analysis of biomedical text for chemical names: a comparison of three methods.

At the National Library of Medicine (NLM), a variety of biomedical vocabularies are found in data pertinent to its mission. In addition to standard medical terminology, there are specialized vocabularies including that of chemical nomenclature. Normal language tools including the lexically based ones used by the Unified Medical Language System (UMLS) to manipulate and normalize text do not work well on chemical nomenclature. In order to improve NLM's capabilities in chemical text processing, two approaches to the problem of recognizing chemical nomenclature were explored. The first approach was a lexical one and consisted of analyzing text for the presence of a fixed set of chemical segments. The approach was extended with general chemical patterns and also with terms from NLM's indexing vocabulary, MeSH, and the NLM SPECIALIST lexicon. The second approach applied Bayesian classification to n-grams of text via two different methods. The single lexical method and two statistical methods were tested against data from the 1999 UMLS Metathesaurus. One of the statistical methods had an overall classification accuracy of 97%.

Synthesis of a template-associated peptide designed as a transmembrane ion channel former.

We describe the design and the Fmoc/tBu solid phase synthesis of a 20 residue long peptide containing five regularly distributed lysines. Cyclization of this peptide was achieved using BOP as coupling agent. After side-chain deprotection, all the basic residues were iodoacetylated and then allowed to react either with a C-terminal free COOH peptide or with peptides bearing a cysteamide group. The final pentameric templates were identified by mass and amino acid analysis which gave data compatible with the expected values.

Design and synthesis of a peptide derived from positions 195-244 of human cdc25C phosphatase.

We have designed, synthesized and purified a 51 amino acid peptide derived from an essential domain of human cdc25C phosphatase. In vivo, differential phosphorylation of this domain regulates either the induction of mitotic processes, or the checkpoint arrest of eukaryotic cells in response to DNA damage. Peptide synthesis was achieved using the stepwise Fmoc strategy and resulted in an important yield of highly pure peptide. The final peptide was identified by amino acid analysis, electrospray mass spectrometry and nuclear magnetic resonance, which revealed that one of the two methionines within the peptide was oxidized into its sulphoxide derivative We investigated whether this 51 amino acid peptide folded into secondary structures in solution by circular dichroism and observed the formation of alpha helices in TFE. Finally, we verified that this peptide could bind to its biologically relevant 14-3-3 partner in vitro by fluorescence spectroscopy.

A new potent HIV-1 reverse transcriptase inhibitor. A synthetic peptide derived from the interface subunit domains.

The biologically relevant and active forms of human immunodeficiency viruses type 1 and 2 reverse transcriptase found in infectious virions are heterodimers produced in a two-step dimerization process. Dimerization involves first the rapid association of the two subunits, followed by a slow conformational change yielding a fully active form. We have shown that the dimeric nature of reverse transcriptase represents a important target for the design of a new class of antiviral agents. In this work, we propose a new strategy for its inhibition by targeting protein/protein interactions during viral formation in infected cells. From the screening of peptides derived from the tryptophan cluster at the interface of the connection subdomain, we have designed a short peptide (10 residues) corresponding to residues 395-404, which can block dimerization of reverse transcriptase in vitro and in infected cells. This peptide is highly efficient in abolishing the production of viral particles, without any adverse toxic side effects, when transduced into human immunodeficiency virus type 1-infected cells together with a new peptide carrier.

A novel potent strategy for gene delivery using a single peptide vector as a carrier.

We have shown previously that a peptide, MPG, derived from the hydrophobic fusion peptide of HIV-1 gp41 and the hydrophilic nuclear localisation sequence of SV40 large T antigen, can be used as a powerful tool for the delivery of oligonucleotides into cultured cells. Now we extend the potential of MPG to the delivery of nucleic acids into cultured cells. In vitro, MPG interacts strongly with nucleic acids, most likely forming a peptide cage around them, which stabilises and protects them from degradation in cell culture media. MPG is non-cytotoxic, insensitive to serum and efficiently delivers plasmids into several different cell lines in only 1 h. Moreover, MPG enables complete expression of the gene products encoded by the plasmids it delivers into cultured cells. Finally, we have investigated the potential of MPG as an efficient delivery agent for gene therapy, by attempting to deliver antisense nucleic acids targeting an essential cell cycle gene. MPG efficiently delivered a plasmid expressing the full-length antisense cDNA of human cdc25C, which consequently successfully reduced cdc25C expression levels and promoted a block to cell cycle progression. Based on our results, we conclude that MPG is a potent delivery agent for the generalised delivery of nucleic acids as well as of oligonucleotides into cultured cells and believe that its contribution to the development of new gene therapy strategies could be of prime interest.

Effects of phosphorylation of threonine 160 on cyclin-dependent kinase 2 structure and activity.

We have prepared phosphorylated cyclin-dependent protein kinase 2 (CDK2) for crystallization using the CDK-activating kinase 1 (CAK1) from Saccharomyces cerevisiae and have grown crystals using microseeding techniques. Phosphorylation of monomeric human CDK2 by CAK1 is more efficient than phosphorylation of the binary CDK2-cyclin A complex. Phosphorylated CDK2 exhibits histone H1 kinase activity corresponding to approximately 0.3% of that observed with the fully activated phosphorylated CDK2-cyclin A complex. Fluorescence measurements have shown that Thr160 phosphorylation increases the affinity of CDK2 for both histone substrate and ATP and decreases its affinity for ADP. By contrast, phosphorylation of CDK2 has a negligible effect on the affinity for cyclin A. The crystal structures of the ATP-bound forms of phosphorylated CDK2 and unphosphorylated CDK2 have been solved at 2.1-A resolution. The structures are similar, with the major difference occurring in the activation segment, which is disordered in phosphorylated CDK2. The greater mobility of the activation segment in phosphorylated CDK2 and the absence of spontaneous crystallization suggest that phosphorylated CDK2 may adopt several different mobile states. The majority of these states are likely to correspond to inactive conformations, but a small fraction of phosphorylated CDK2 may be in an active conformation and hence explain the basal activity observed.

Characterization of the interactions between human cdc25C, cdks, cyclins and cdk-cyclin complexes.

We have overexpressed and purified human dual-specificity phosphatase cdc25C from a prokaryotic expression system at high levels and in a soluble, active form, and have studied and quantified its potential to interact with cdks, cyclins and preformed cdk-cyclin complexes by fluorescence spectroscopy and size-exclusion chromatography. Our data indicate that human cdc25C forms stable complexes, through hydrophobic contacts, with cdk and cyclin monomers, as well as with preformed cdk-cyclin complexes. In vitro, cdc25C interacts with cyclin monomers with high affinity, with tenfold less affinity with cdks, and with intermediate affinity with cdk-cyclin complexes. Moreover, changes observed in the intrinsic fluorescence of cdks, cyclins and cdk-cyclin complexes upon interaction with cdc25C are indicative of concomitant conformational changes within cdks and cyclins. From our results, we propose that in vitro, in the presence of monomeric cdks and cyclins, cdc25C forms stable ternary complexes, first through a high affinity interaction with a cyclin, which may then help target cdc25C towards a cdk. We discuss the biological relevance of our results and propose that a similar, two-step mechanism of interaction between cdc25C and cdk-cyclin complexes may occur in vivo.