Text mining is a powerful approach to efficiently identify and extract information from large amounts of text. Its application to biomedical literature promises to allow researchers the ability to process hundreds and thousands of articles in a way that was never possible before. This new technology, coupled with the increasing amount of published literature and open access literature sources may usher in a new era of meta-research in which computational algorithms can generate new scientific hypotheses from existing information.

Many obstacles, however, stand in the way of truly automated methods for text mining. The difficulty in obtaining full text literature, specialized jargon in scientific research, and the ambiguity of biological entity terms are but a few of the challenges that need to be overcome. That being said, semi-automated text mining methods can still greatly aid researchers by identifying topics of interest, reducing the number of articles to read, and targeting relevant information in articles. The judicious and dedicated use of semi-automated methods has the possibility of having a great impact in efficiently distributing the task of manual reading and processing of scientific literature.

Our contribution to semi-automated methods for biomedical text mining center on the identification and extraction of point mutation information. Point mutations are an integral aspect of protein research, as they are the vehicle of diversity and the key to functional changes in proteins. They are also represented in a format that lends itself to text mining and can be referenced to the growing numbers of biological sequence databases. This dissertation focuses on the ability to parse literature for point mutations and extract their functional effects. We show that, using statistical, graph theoretical, and machine learning methods, we can efficiently transform information that was previously embedded in the text into information that is computationally stored and processable.

The left-handed parallel β-helix (LβH) is a structurally repetitive, highly regular, symmetrical fold, formed by coiling of elongated β-sheets into helical `rungs'. This canonical fold has recently received interest as a possible solution to the fibril structure of amyloid. However, sequence and structural requirements, and stability and folding of the LβH are not fully understood. To analyze this protein fold and investigate the possibility of its involvement in amyloid formations, combined approaches using computational and experimental methods were applied. (1) Sequence characteristics of the LβH were determined by analyzing known structures. A genome-wide sequence analysis was undertaken to determine the prevalence of this unique protein fold across the genomes. Molecular dynamics (MD) simulations were used to demonstrate the stabilizing effect of successive rungs and the hydrophobic core of the LβH. (2) An in-vivo folding assay using a known LβH protein was developed to investigate the residue tolerance and effect of mutations at structurally critical regions of the β-helical structure. Tyrosine fluorescence spectroscopy was also used to study the thermodynamic stability of the LβH protein. It showed that the β-helix could tolerate non-hydrophobic residues at interior positions, and proline residues in the β-helical domain were shown to be important, but not critical, in folding of the β-helix. A recombinant protein consisting of the LβH with a prion fragment was designed, constructed and tested. The results suggest that the amyloid prone PrP fragment may fold into a β-helix. (3) Modeling of possible monomeric subunits of the insulin amyloid fibril was conducted using β-solenoid folds, namely the β-helix and β-roll. While both the β-helix and β-roll models agreed with currently available biophysical data from misfolded insulin, MD simulations showed that the β-roll model was relatively more stable. Insulin monomeric subunit models were incorporated into plausible models of insulin fibrils. Simulated X-ray fiber diffraction patterns of these models demonstrated that the model fibers constructed from a β-solenoid subunit provided the reasonable fit to available experimental diffraction patterns.

Background

Results

Conclusions

Synthetic prions were produced in our laboratory by using recombinant mouse prion protein (MoPrP) composed of residues 89-230. The first mouse synthetic prion strain (MoSP1) was inoculated into transgenic (Tg) 9949 mice expressing N-terminally truncated MoPrP(Delta23-88) and WT FVB mice expressing full-length MoPrP. On first and second passage in Tg9949 mice, MoSP1 prions caused disease in 516 +/- 27 and 258 +/- 25 days, respectively; numerous, large vacuoles were found in the brainstem and gray matter of the cerebellum. MoSP1 prions passaged in Tg9949 mice were inoculated into FVB mice; on first and second passage, the FVB mice exhibited incubation times of 154 +/- 4 and 130 +/- 3 days, respectively. In FVB mice, vacuolation was less intense but more widely distributed, with numerous lesions in the hippocampus and cerebellar white matter. This constellation of widespread neuropathologic changes was similar to that found in FVB mice inoculated with Rocky Mountain Laboratory (RML) prions, a strain derived from a sheep with scrapie. Conformational stability studies showed that the half-maximal GdnHCl (Gdn(1/2)) concentration for denaturation of MoSP1 prions passaged in Tg9949 mice was approximate to4.2 M; passage in FVB mice reduced the Gdn(1/2) value to approximate to1.7 M. RML prions passaged in either Tg9949 or FVB mice exhibited Gdn(1/2) values of approximate to1.8 M. The incubation times, neuropathological lesion profiles, and Gdn(1/2) values indicate that MoSP1 prions differ from RML and many other prion strains derived from sheep with scrapie and cattle with bovine spongiform encephalopathy.

Gerstmann-Strfiussier-Scheinker (GSS) disease is a dominantly inherited, human prion disease caused by a mutation in the prion protein (PrP) gene. One mutation causing GSS is P102L, denoted P101L in mouse PrP (MoPrP). In a line of transgenic mice denoted Tg2866, the P101L mutation in MoPrP produced neurodegeneration when expressed at high levels. MoPrPSc(P101L) was detected both by the conformation-dependent immunoassay and after protease digestion at 4degreesC. Transmission of prions from the brains of Tg2866 mice to those of Tg196 mice expressing low levels of MoPrP(P101L) was accompanied by accumulation of protease-resistant MoPrPsc(P101L) that had previously escaped detection due to its low concentration. This conformer exhibited characteristics similar to those found in brain tissue from GSS patients. Earlier, we demonstrated that a synthetic peptide harboring the P101L mutation and folded into a beta-rich conformation initiates GSS in Tg196 mice (29). Here we report that this peptide-induced disease can be serially passaged in Tg196 mice and that the PrP conformers accompanying disease progression are conformationally indistinguishable from MoPrPSc(P101L) found in Tg2866 mice developing spontaneous prion disease. In contrast to GSS prions, the 301V, RML, and 139A prion strains produced large amounts of protease-resistant PrPSc in the brains of Tg196 mice. Our results argue that MoPrPSc(P101L) may exist in at least several different conformations, each of which is biologically active. Such conformations occurred spontaneously in Tg2866 mice expressing high levels of MoPrPC(P10IL) as well as in Tg196 mice expressing low levels of MoPrPC(P101L) that were inoculated with brain extracts from ill Tg2866 mice, with a synthetic peptide with the P101L mutation and folded into a beta-rich structure, or with prions recovered from sheep with scrapie or cattle with bovine spongiform encephalopathy.