n/a

Adeno-associated viruses (AAVs) are a powerful tool in gene therapy, offering a safe and highly effective method for delivering therapeutic genes to treat human disease. AAV-based gene delivery vehicles, or vectors, have shown success in both preclinical studies and clinical trials, with six therapies having received U.S. FDA approvals. Several properties of AAVs make them an advantageous vector for delivering DNA for clinical gene therapy: their minimal lack of integration into the host genome, lack of pathogenicity, and ability to promote high expression of the delivered therapeutic. However, issues still arise over the efficiency and specificity of AAV gene delivery, as well as possible immune responses. Naturally-occurring AAVs have the ability to transduce a wide range of tissue and cell types, reducing delivery efficacy to the intended tissue, requiring higher dosages to be administered, and increasing the potential for off-target effects. In addition, the presence of neutralizing antibodies against AAVs significantly limits the effectiveness of any AAV-based treatment or re-administration of the therapeutic. To overcome this, engineering the AAV capsid, or virus shell, is necessary to improve both improve its delivery efficiency to target cell types and prevent recognition of our body’s neutralizing antibodies.

Several strategies have been implemented to engineer the AAV capsid and thus alter its properties and function. Directed evolution is a technique that was designed to mimic natural evolution by subjecting a large library of AAV variants to iterative rounds of selection against a selective pressure to isolate variants with the desired properties. Directed evolution strategies have been applied to the AAV cap gene, which encodes the proteins that comprise the capsid, to generate millions of novel AAV capsids and identify variants with enhanced transduction towards specific cell types, evade neutralizing antibodies, and/or improve packaging capabilities. In Chapter 2, we use the process of directed evolution to identify novel AAV variants that successfully transduce wild-type Schwann cells in human or mouse. The ability to deliver therapeutics to Schwann cells provides an opportunity to preventatively and actively treat genetic diseases of the peripheral nervous system, including Neurofibromatosis Type I (NF1) and Charcot-Marie Tooth disease. Through this work, we have subjected millions of AAV variants to iterative rounds of selection in human and mouse Schwann cells to identify variants with improved transduction towards human or mouse Schwann cells. We have identified two groups of capsids: human Schwann cell-tropic capsids with increased resistance to antibody neutralization, and mouse Schwann cell-tropic capsids that also demonstrate decreased localization to the liver compared to wild-type AAVs.

While directed evolution strategies offer a high-throughput method of screening millions of AAV variants for enhanced properties, there is a significant lack of knowledge about the composition of many AAV DNA libraries due to sequencing limitations. The ability to fully sequence these DNA libraries offers insight into the evolutionary process and allows for earlier detection of successful variants. However, complete characterization is hindered by current sequencing methodologies. Short-read sequencing, or next generation sequencing (NGS), is limited to 300-500bp reads, unable to sequence individual variants whose mutations span the entire 2.2kb cap open reading frame. Long-read sequencing methodologies suffer from high error rates and limits on the number of variants that can be sequenced. In Chapter3, we develop a novel synthetic, long-read sequencing methodology, called Barcode-Labeled Short Reads (BLaSR), to overcome the major barriers that current technologies encounter. We harness the random mechanism of the Tn5 transposon (also known as a ‘jumping gene’) to insert random barcodes into a diverse AAV plasmid library such that each AAV variant contains insertions with a unique, identifiable barcode. This allows for the generation of barcode-linked short reads that can be sequenced with NGS technologies, so that sequencing reads with the same barcode can be traced to the same AAV variant and thus assembled into the full AAV variant cap sequence. We have engineered a robust pipeline to ensure highly efficient barcoded-Tn5 insertion into the AAV genome and have characterized BLaSR according to error rate, read depth, and Tn5 insertion bias. We also applied BLaSR to a diverse, randomly shuffled AAV capsid library and estimate the full assembly of thousands of variants. Further work is ongoing to completely assemble and characterize the sequenced shuffled library, as well as sequence the shuffled library after a round of selection in adult human brain tissue to identify enriched variants.

For many decades, electron dynamics and lattice dynamics have been studied by performing separate experiments, requiring different instrumentation. In general, electronic dynamics are most often studied by using visible pump probe spectroscopy, while lattice dynamics can be followed by core level spectroscopy performed at synchrotrons. With the advent of high harmonic generation (HHG), it became possible to generate femtosecond to attosecond bursts of extreme ultraviolet (XUV) radiation that are both spatially and temporally coherent. Since high harmonic generation can be performed on a table-top setup, it has become possible to follow both electronic and lattice dynamics concurrently, without the need for a synchrotron. In this dissertation, visible pump-XUV probe transient absorption spectroscopy is used to elucidate the excited state electronic and lattice dynamics of two materials relevant for solar energy conversion, namely a-Fe2O3 and single crystalline Si thin films.

In this dissertation, the first chapter deals with the pertinent information necessary to understand both the core-level spectroscopy that can be used to glean information about the electronic and lattice dynamics of solids, and the solid-state physics which govern the electronic and lattice dynamics, with special focus on the theoretical models used in these fields. The second chapter outlines the experimental apparatus that is used to perform the pump probe experiments on solid state samples, as well as several upgrades made to it during the course of this thesis. The third chapter details experiments performed on a-Fe2O3 that provide insights into its excited state dynamics following photoexcitation by multiple pump wavelengths (400 nm, 480 nm, 520 nm, 560 nm), including polaron formation and its impact on the photoconversion efficiency of photoelectrochemical cells. The fourth chapter details experiments performed on single crystalline Si thin films following photoexcitation with 800 nm pump pulses. This experiment highlights the strengths of HHG pump probe spectroscopy by giving insights into the electron-phonon scattering mechanisms that follow photoexcitation across the bandgap.

A probabilistic formulation for semantic image annotation and retrieval is proposed. Annotation and retrieval are posed as classification problems where each class is defined as the group of database images labeled with a common semantic label. It is shown that, by establishing this one-to-one correspondence between semantic labels and semantic classes, a minimum probability of error annotation and retrieval are feasible with algorithms that are 1) conceptually simple, 2) computationally efficient, and 3) do not require prior semantic segmentation of training images. In particular, images are represented as bags of localized feature vectors, a mixture density estimated for each image, and the mixtures associated with all images annotated with a common semantic label pooled into a density estimate for the corresponding semantic class. This pooling is justified by a multiple instance learning argument and performed efficiently with a hierarchical extension of expectation-maximization. The benefits of the supervised formulation over the more complex, and currently popular, joint modeling of semantic label and visual feature distributions are illustrated through theoretical arguments and extensive experiments. The supervised formulation is shown to achieve higher accuracy than various previously published methods at a fraction of their computational cost. Finally, the proposed method is shown to be fairly robust to parameter tuning.