1983 research outputs found
Sort by
Fracture of Complex Hydrogels: Dynamic and Microstructural Effects
No full textHydrogels are a unique class of soft materials composed of three-dimensional polymer networks swollen with large amounts of water. This hybrid solid–liquid composition enables hydrogels with tissue-like mechanical properties, permeability to small molecules, and exceptional designability. By adjusting the combination of solvent, polymer, and network topology, hydrogels can be tailored into functional materials that meet diverse application requirements. These properties make hydrogels highly attractive for a wide range of applications, particularly in biomedical engineering, soft robotics, and bioelectronics.
Understanding the mechanical behavior of hydrogels is essential for guiding material design, evaluating failure criteria, and ensuring reliable performance in practical applications. It also provides valuable inspiration for the development of other soft polymeric materials such as elastomers and thermoplastics. In this thesis, we investigate the mechanical behavior of hydrogels, with a primary focus on their fracture properties. The complex conditions near the crack tip make the fracture process highly sensitive and show interesting phenomena when the chain topology, external stimuli, or loading conditions are altered at the crack tip.
A fundamental feature of hydrogels, and many other polymers, is viscoelasticity, which exhibits both elastic and viscous behavior depending on the timescale of deformation. Viscoelasticity originates from the molecular architecture of polymer networks. The flexibility of covalent C–C bonds allows for segmental motion, while interchain interactions dissipate energy and enable stress relaxation, with recovery to some degree upon unloading. In swollen systems like hydrogels, polymer–solvent interactions further influence chain mobility and relaxation dynamics by facilitating or hindering molecular motion. Additionally, chain entanglements act as transient constraints on motion, contributing to energy dissipation and delayed elastic recovery. Together, these factors result in the complex, time-dependent mechanical response that defines viscoelastic materials. While viscoelasticity is important for global deformation, its role becomes even more critical when considering how cracks initiate and propagate. In particular, the fracture behaviors of hydrogels are fundamentally shaped by their time-dependent mechanical response, which governs how energy is dissipated near the crack tip and how the material resists crack propagation. In these systems, energy dissipation around the crack tip is not only governed by intrinsic material toughness but also by the rate of deformation and solvent dynamics. A faster loading rate can decrease the apparent fracture toughness by decreasing energy dissipation through viscoelastic relaxation mechanisms. Similarly, the solvent viscosity in hydrogels modulates chain mobility and effective chain length, with higher solvent viscosity leading to much decreased resistance to crack propagation. In certain conditions, though hydrogels can show nearly perfect elasticity, their fracture can still show nonelastic process due to the complex and non-uniform crack tip zone. One striking phenomenon observed under these conditions is crack branching, where rapid deformation results in the formation of multiple crack paths. This behavior underscores the complex interplay between material structure, loading conditions, and environmental factors in defining the fracture response of viscoelastic materials.
In addition to passive viscoelastic effects, polymer networks can be engineered to actively respond to mechanical stress through mechanochemically triggered reactions. One such example is the incorporation of disulfide bonds (-S–S-), which can undergo dynamic exchange reactions under triggers like UV light and free radicals. When materials containing these bonds undergo crack initiation, localized stress near the crack tip can be relaxed when the dynamic reaction is activated, effectively redistributing stress and limiting crack growth, as the network is able to reorganize and reform in response to mechanical damage. This mechanochemical coupling introduces a powerful design strategy: leveraging molecular reactivity not only to enhance material toughness but also to enable the material to adaptive, self-protective behavior under extreme conditions.
Building on the concept of responsive materials, we introduce a triggerable crosslinking strategy using ferric citrate to strengthen natural polymer matrices. In this work, a complex crosslinker, ferric citrate, is applied to coordinate with chitosan, forming a robust film. The resulting ferric citrate-crosslinked chitosan film exhibits significantly improved mechanical strength, enhanced acid resistance, and recyclability. This system demonstrates a promising approach for the development of sustainable and functional biopolymer materials, where the crosslinking is not only reversible but also tunable based on environmental pH. By integrating a metal–ligand coordination chemistry with sodium citrate, we achieve a recyclable and chemically resistant material platform suitable for applications in packaging, filtration, or biomedical devices.
To further enhance the mechanical performance of hydrogels, we developed a method for fabricating hierarchically structured fibrous hydrogels using Wet Rotary Jet Spinning (WRJS). This scalable and rapid technique enables the production of continuous microfibers with controlled alignment and diameter. Taking polyvinyl alcohol (PVA) as a model system, we fabricated a fibrous PVA hydrogel with superior mechanical strength and flaw tolerance. The fibers, with diameters below 10 µm, were rapidly produced in large quantities by spinning PVA solution into a coagulation bath, followed by a salting-out process that stabilizes the fibrous structures. Inspired by the mechanical architecture of spider silk and other natural hierarchical structures, the resulting hydrogel combines high extensibility, strength, and crack resistance, attributed to the aligned, hierarchical fibrous network that dissipates energy effectively under stress. This hierarchical design offers a compelling route to engineering tough, scalable hydrogels with structural features mimicking natural load-bearing tissues, like tendons.
This dissertation presents a comprehensive investigation into the fracture mechanics of hydrogels, with particular emphasis on the complex mechanical behaviors near crack tips. Through a combination of experimental studies, material design, and theoretical insights, we explore how viscoelasticity, dynamic crosslinking, and hierarchical structure influence crack propagation and energy dissipation. The findings contribute to a deeper understanding of time-dependent fracture in soft materials and provide new strategies for designing tough, responsive, and multifunctional hydrogel systems suitable for a broad range of applications.Engineering and Applied Sciences - Engineering Science
Integrative analysis of transcriptomics, epigenetics, and copy number to assess lineage and dynamics of tumor clones during cancer progression
No full textReconstructing dynamic biological processes, such as cancer evolution, is challenged by sparse clinical sampling and incomplete multi-omic measurements. This dissertation develops and applies integrative computational strategies to recover interpretable cell-state dynamics under these constraints, focusing on multiple myeloma (MM), chronic lymphocytic leukemia (CLL) progression to Richter's syndrome (RS), and anti-BCMA CAR-T cell therapy.
First, to model cellular dynamics from static snapshots, I contributed to scDiffEq, a neural stochastic differential equation framework that infers cellular drift and diffusion from scRNA-seq. This approach improves trajectory inference and fate prediction. My contributions included designing benchmark criteria and developing novel simulation strategies using binned CytoTRACE pseudotime, which validated the method's robustness to sparse sampling density.
Second, to reconstruct evolutionary time using genetic lineage, I developed Numbat-Multiome. This method unifies copy number variation (CNV) inference from both scRNA-seq and scATAC-seq data. By integrating coverage and allelic imbalance signals within a shared genomic binning scheme, the method accurately detects diverse events (F1 > 0.9), including copy-neutral loss-of-heterozygosity. This enables the reconstruction of subclonal phylogenies to serve as a lineage anchor for multi-omic regulatory analysis.
Applying these frameworks, I dissected regulatory heterogeneity in multiple myeloma. By profiling chromatin accessibility (scATAC-seq) across 36 patient samples, I identified differentially accessible regions and transcription factor programs associated with disease progression. In a separate study on anti-BCMA CAR-T therapy, single-cell multiome profiling of post-infusion cells characterized the coupled transcriptional and epigenetic states underlying T-cell exhaustion and linked CAR promoter accessibility to CAR expression.
To trace clonal evolution during the transformation of CLL to RS, I integrated scRNA-seq, mitochondrial scATAC-seq, and scDNA-seq from pilot cases to map clone-specific regulatory rewiring. Furthermore, I analyzed STAG-seq data from six CLL samples, which jointly profiles targeted genotype and transcriptome in the same cells. This analysis enabled the direct and unambiguous assignment of transcriptional programs and immune cell states to specific genetic subclones.
Together, these contributions provide a lineage-anchored, time-aware framework for studying tumor evolution and therapy response. By coupling mutation-defined ancestry with multi-omic regulatory readouts and neural dynamical models, this work delivers practical tools and conceptual clarity for inferring cancer cell-state transitions from limited and static clinical data.Biomedical Informatic
Everything Is a Matrix: Minimizing Data Movement and Parameter Count Across the Machine Learning Stack
No full textMachine learning has revolutionized natural language processing, computer vision, and beyond. Yet as machine learning models scale in size and capability, the demand for computational resources likewise grows, exposing new challenges in efficient and scalable deployment. Extracting maximal performance from existing hardware is therefore vital to unlocking the next wave of progress in artificial intelligence.
In many modern workloads, matrix operations dominate resource consumption, sometimes accounting for more than 99% of the workload [1]. Thus, we will focus on matrices as the central unit of optimization. This thesis presents an array of novel techniques to reduce memory footprint, accelerate computation, and improve overall hardware utilization. We demonstrate substantial efficiency gains are achievable by rethinking how data is computed, stored, and compressed, with a special focus on matrices, the core computational structure underpinning both scientific computing and neural networks.
First, we address dense matrix multiplication by introducing CAKE, a method that partitions computation into optimally shaped blocks to minimize memory bandwidth bottlenecks (Chapter 2). We extend this method to tensor contractions with any number of loops with mCAKE (Chapter 3). Then, for neural networks exhibiting moderate sparsity, the Rosko framework (Chapter 4) exploits outer-product structure to efficiently skip zero-valued computations and enables the creation of hardware-compatible sparsity patterns through structured pruning.
Next, we investigate efficient representations of weight matrices of neural networks using Singular Value Decomposition (SVD) (Chapter 5), enabling both memory savings and accelerated inference. Building on this, we explore low-rank model compression, where the compact forms of decomposed weight matrices facilitate efficient training and adaptive fine-tuning (Chapter 6). We then introduce blockwise knowledge distillation techniques (Chapter 7) that allow highly compressed, SVD-based student models to learn directly from their full-rank teacher counterparts, preserving both efficiency and model accuracy. Lastly, we demonstrate a privacy-preserving framework for distributed inference that splits computation between local devices and cloud servers, ensuring user data labels remain on-device while leveraging powerful cloud-based feature extractors (Chapter 8).
Together, these contributions meaningfully advance the efficiency and scalability of both conventional scientific workloads and the latest state-of-the-art AI models.
Reference: [1] A. Ivanov, N. Dryden, T. Ben-Nun, S. Li, and T. Hoefler, “Data movement is all you need: A case study on optimizing transformers,” 2020.Engineering and Applied Sciences - Computer Scienc
Unraveling the molecular mechanisms of neuronal and astrocytic proteostasis in human models of Alzheimer’s disease
No full textCellular proteostasis is the integrated regulation of protein synthesis, folding, trafficking, and degradation that sustains the integrity of the proteome. By balancing these processes, cells preserve homeostasis under both basal conditions and during periods of physiological stress. This regulation is especially critical in post-mitotic cells such as neurons, as well as in other brain cell types including astrocytes. In the brain, disruption of the proteostasis network promotes the accumulation of misfolded or aggregated proteins that drive the pathogenesis of neurodegenerative disorders, including Alzheimer’s disease (AD). Of particular relevance to neurodegeneration is the degradation arm of proteostasis, as impairments in autophagy and proteasomal pathways result in the accumulation of toxic protein aggregates that are hallmarks of disease.
The autophagy-lysosome pathway (ALP) and the ubiquitin-proteasome system (UPS) represent the two major degradative mechanisms that sustain cellular proteostasis. Together, these systems form the core of the protein quality control network, ensuring the clearance of damaged, misfolded, or aggregated proteins, as well as dysfunctional organelles. While the UPS primarily targets short-lived soluble proteins, the ALP provides both bulk and selective degradation routes capable of removing larger protein aggregates and organelles, a particularly important process for mitigating the toxic effects of aggregation prone proteins in neurodegenerative diseases. While autophagy can be both a bulk and selective degradation process, selective autophagy itself depends on molecular chaperones and adaptor proteins that identify substrates and recruit them to degradative machinery. Dysfunction in these pathways has been strongly implicated in AD pathogenesis, highlighting the importance of understanding how some of these adaptor proteins and chaperones might regulate ALP activity and influence disease pathophysiology.
Optineurin (OPTN) is one such autophagy adaptor protein with established roles in selective autophagy. Pathogenic mutations in OPTN have been linked to amyotrophic lateral sclerosis, frontotemporal dementia, and glaucoma, but its contribution to AD and neuronal function remains unclear. To investigate the role of OPTN in neuronal proteostasis and AD, we utilized induced pluripotent stem cell (iPSC)-derived neuron (iN) and astrocyte (iA) models. Analyses revealed a negative correlation between OPTN and specific pTau epitopes in neurons, as well as a decrease in OPTN protein abundance in brain tissues of individuals with AD. Given these findings, we generated OPTN knockout (KO), heterozygous (HET), and wildtype (WT) iNs and iAs using CRISPR/Cas9 editing in two genetic backgrounds. Loss of OPTN in iNs increased specific pTau proteoforms without substantially affecting autophagy processes or mitochondrial respiration. Despite no clear effect on mitochondrial function, several mitochondrial proteins, including OXCT1, were enriched in an unbiased analysis of the OPTN interactome in iNs, as well as proteins involved in intracellular trafficking. Proteomic analyses further identified intracellular Clusterin (CLU), an AD risk gene, as significantly upregulated in OPTN KO iNs, suggesting OPTN may influence its intracellular processing. Our model system demonstrates modest roles for OPTN in certain neuronal biological processes and potential implications for AD pathogenesis. These findings also suggest that OPTN may exhibit functional redundancy with other autophagy adaptor proteins in human neurons, leading to relatively mild phenotypic changes with complete loss of OPTN.
Another important form of selective autophagy relevant to neurodegenerative disease is chaperone-assisted selective autophagy (CASA). Bcl-2-associated athanogene 3 (BAG3) is a mediator of CASA, and given the genetic and pathological links of BAG3 to proteostasis and neurodegenerative diseases, we investigated how BAG3 contributes to cellular function and Alzheimer's disease (AD) in both human neurons and astrocytes. We first utilized a large panel of iPSCs from deeply phenotyped cohorts to interrogate genetic contributions to baseline autophagic flux and UPS activity in human neurons, and protein turnover was assessed using SILAC-based quantitative proteomics. Across our panel of neurons, we observed substantial inter-individual differences in autophagic flux, which was inversely correlated with UPS activity. This reciprocal relationship extended to tau homeostasis, where higher autophagic flux resulted in reduced accumulation of aggregated, phosphorylated tau. Proteomic analyses revealed that global protein turnover dynamics stratified based on degradation pathway activity and could predict pathway-specific substrate dependencies. Interestingly, BAG3 emerged as a dynamically regulated autophagy chaperone, responsive to pharmacological inhibition of both the UPS and ALP. BAG3 knockout in neurons decreased autophagic flux and increased levels of high-molecular-weight phosphorylated tau. Notably, familial APP AD mutations and Aβ exposure induced BAG3 expression in neurons, while elevated BAG3 levels in human brain tissue were associated with higher neuropathological burden and disease progression.
While an elevation of BAG3 was observed in the AD brain, it was unclear which brain cell type might be contributing most to this upregulation. We found that in human brain and iPSC models, BAG3 was most highly expressed in astrocytes. Further, BAG3 loss in our iPSC model system caused greater proteomic disruption in astrocytes than in neurons. In the absence of BAG3, astrocytes showed reduced autophagy, diminished lysosome abundance and activity, and decreased proteasome function. To uncover molecular binding partners of BAG3 that might influence these phenotypes, we performed co-immunoprecipitation, revealing interactions with HSPB8 and other heat shock proteins, proteasome regulators (PSMD5, PSMF1), and the retromer component, VPS35. Integration of BAG3 KO transcriptomic and proteomic datasets pinpointed AD-relevant proteins under post-translational control of BAG3, which included GFAP, BIN1, and HSPB8. HSPB8 levels were markedly reduced in BAG3-deficient astrocytes with overexpression partially rescuing its levels. Loss of astrocytic BAG3 impaired Aβ clearance in co-culture with APP/PSEN1 mutant neurons, directly linking BAG3 to a disease-relevant astrocyte function. Finally, analysis of postmortem brain tissue revealed BAG3 marks a stress-responsive astrocyte subtype in the brain of aged individuals with AD.
Collectively, these studies define complementary and cell-type specific contributions of OPTN and BAG3 to proteostasis in the human brain and AD. They reveal how adaptor proteins and chaperones regulate neuronal and glial protein quality control, highlight BAG3 as a central regulator responsive to genetic and pathological stress, and establish mechanistic links between proteostasis dysfunction and AD pathogenesis. Together, this work advances our understanding of proteostasis networks in the brain and identifies potential therapeutic nodes within these pathways for combatting neurodegenerative disease.Biological and Biomedical Science
Robust Causal Inference Methods for Electronic Health Record-Based Studies with Missing Eligibility and Calendar Time-Varying Treatment Effects
No full textElectronic health records (EHR) are seen as useful alternatives to randomized controlled trials when the latter are infeasible due to financial, ethical, or logistical constraints. Unfortunately, EHR exist to record clinical activity and assist with billing, and thus information is not collected with research in mind. When using EHR to study comparative effectiveness, there are many factors that a researcher can not control: treatments are not randomly assigned, information on certain patient covariates may be unavailable, when to begin follow-up is not always clear, and which patients receive treatment and why may change over time. As such, rigorous statistical methods which contend with these factors, often simultaneously, are necessary when conducting EHR-based studies.
In Chapter 1, we consider the problem of selection bias due to missingness in covariates which define study eligibility in target trial emulations. We illustrate the dangers of naively excluding patients missing certain eligibility-defining covariates and propose a solution based on a novel missing at random assumption using inverse probability weighting. Our solution integrates seamlessly within a larger framework for dealing with common sources of bias in sequential target trial emulations, such as confounding, non-adherence, and censoring.
Next, in Chapter 2, we extend the ideas of Chapter 1 and propose a robust and efficient estimator of the causal average treatment effect on the treated, defined in the study eligible population, in cohort studies where eligibility-defining covariates are missing at random. The approach facilitates the use of flexible machine-learning strategies for component nuisance functions while maintaining appropriate convergence rates for valid asymptotic inference, and displays robustness to various degrees of model misspecification in the component nuisance functions.
Finally, in Chapter 3, we formalize sequential target trial emulations for continuous outcomes and propose a statistical framework to describe both how and why causal effects vary over treatment initiation time in EHR-based studies. Our approach projects doubly robust, time-specific treatment effect estimates onto candidate marginal structural models and uses a principled model selection procedure to best describe how effects vary by treatment initiation time. We further introduce a novel summary metric, based on standardization analysis, to quantify the role of covariate shift in explaining observed effect changes and disentangle changes in treatment effects from changes in the patient population receiving treatment.
The statistical methods developed in this dissertation are motivated by real EHR-based studies of bariatric surgery at Kaiser Permanente. Throughout, we use these data to both illustrate and validate the methods introduced in this work.Biostatistic
Positivity in Cluster Algebras and Their Generalizations
No full textThe theory of bf cluster algebras gives us a combinatorial framework for understanding the previously opaque nature of certain algebras. Each cluster algebra is generated by its cluster variables, which can be obtained via the recursive process of mutation. One remarkable property of cluster algebras is Laurent positivity, which means that every cluster variable can be written in a nice form; specifically, as a Laurent polynomial with positive integer coefficients in the initial cluster variables. Laurent positivity for cluster algebras unifies positivity phenomena in a variety of contexts, including Teichmuller theory, Gromov-Witten theory, string theory, and tropical geometry.
Laurent positivity was conjectured by Fomin and Zelevinsky when they introduced cluster algebras in 2002, but the proof remained elusive for over a decade. There have since been two proofs: a combinatorial approach by Lee and Schiffler, and a geometric approach by Gross, Hacking, Keel, and Kontsevich using a novel connection to scattering diagrams. Scattering diagrams themselves are powerful tools, originating from mirror symmetry, where they track how certain geometric invariants (Gromov--Witten invariants and Donaldson--Thomas invariants) change under varying stability conditions. Every cluster algebra is associated with a cluster scattering diagram that encodes algebraic relations between cluster variables, making them a useful tool in cluster algebra theory.
The work in this dissertation unifies these methods, aiming to deepen our understanding of positivity in both cluster algebras and scattering diagrams. In Chapter 3, which is joint work with Kyungyong Lee and Lang Mou, we prove positivity for generalized cluster algebras of all ranks, confirming a 2014 conjecture of Chekhov--Shapiro. We achieve this by giving a directly computable, manifestly positive, and elementary but highly nontrivial formula describing rank 2 generalized cluster scattering diagrams. This formula enumerates a new class of Dyck path objects, called tight gradings, implying positivity of the scattering diagrams in rank 2. In Chapter 4, which is joint work with Kyungyong Lee, we construct an explicit bijection between broken lines on scattering diagrams and compatible pairs on Dyck paths, which both play crucial roles in the proofs of cluster algebra positivity. In Chapter 5, we give a new expansion formula for quantum cluster variables using colored subpaths of Dyck paths, leveraging a connection we make to the compatible pair framework.Mathematic
Evaluating the Pharmacokinetics of Topically Applied Small Molecule Drugs in Skin using Stimulated Raman Scattering Microscopy
No full textUnderstanding the pharmacokinetics of therapeutic compounds is crucial to the development of safe and effective therapeutics, including topical compounds, which are applied to the skin and other exterior sites of the body. Topical drugs are used to treat numerous dermatological conditions, ranging from cosmetic concerns to autoimmune conditions. Topical drug efficacy depends on topical drug pharmacokinetics in the skin, but cutaneous pharmacokinetics can be especially challenging to evaluate due to the barrier function and complex structure of the skin. Because of this, several aspects of cutaneous pharmacokinetics are not well understood, such as drug uptake to specific skin regions over time. While a variety of methods can be used to measure or visualize topical drug uptake in skin, many of these lack sufficient spatiotemporal resolution to evaluate cutaneous pharmacokinetics over time, especially in specific skin regions. Previous works have shown the value of using stimulated Raman scattering (SRS) microscopy as a tool to study topical drug uptake, since SRS imaging is rapid, chemically-specific, non-invasive, and SRS signal is proportional to drug concentration.
This work describes the development of SRS imaging methods to evaluate topical drug pharmacokinetics, helping reveal previously poorly understood aspects of topical drug uptake and cutaneous pharmacokinetics, such as the effects of perfusion on cutaneous pharmacokinetics, how pharmacokinetics in the stratum corneum relate to pharmacokinetics in deeper skin layers, and the temporal dynamics of topical drug delivery to the sebaceous glands. Measuring uptake of tazarotene with SRS in in vivo and ex vivo mouse ears in a paired experiment revealed that differences in pharmacokinetics were observed in the presence of perfusion. Using adaptive optics to correct for wavefront aberrations and overcome SRS imaging depth limitations allowed the capture of ruxolitinib uptake data in the stratum corneum, viable epidermis, and dermis. Finally, developing an SRS imaging method to detect drug uptake in sebaceous glands and performing corrections to account for light attenuation variations and lens effects enabled the measurement of tazarotene uptake in sebaceous glands, representing a novel approach of assessing topical drug delivery over time to structures located deep in the skin. The methods developed here present promising avenues for future studies and further illumination of cutaneous pharmacokinetics and topical drug uptake.Biology, Molecular and Cellula
Developing a chemogenetic oxidative damage model to phenocopy dry age-related macular degeneration
No full textThere is no cure for dry age-related macular degeneration (AMD), and research is challenging due to a lack of good animal models that accurately recapitulate the human disease. Because humans are the only species that develop the full form of AMD, studying its pathogenesis remains particularly challenging. To address this, we developed a chemogenetic oxidative stress model that mimics aspects of dry AMD using Cre-inducible D-amino acid oxidase (DAAO) transgenic mice. In this system, DAAO generates hydrogen peroxide upon exposure to its substrate, D-alanine, allowing spatial and temporal control of oxidative stress. By crossing DAAO mice with cell type specific Cre drivers, we induced oxidative damage selectively in the retinal pigment epithelium (RPE), rod photoreceptors, or vascular endothelial cells. We combined visual function tests (optomotor response and ERG), multimodal imaging (OCT, fundus, and fluorescein angiography), histology (flatmounts, H&E, and immunofluorescence), electron microscopy, and RNA-seq to characterize this model in RPE cells. Oxidative damage targeted to the RPE led to multiple AMD-like phenotypes, including vision loss, RPE and photoreceptor degeneration, hyperreflective foci, epithelial-mesenchymal transition, macrophage infiltration, mitochondrial damage, basal infolding loss, vacuolization, and complement activation. Additionally, we found that this phenotype could be prevented by Nrf2 overexpression by AAV-Best1-Nrf2 subretinal injections in neonates.
Additionally, by targeting the oxidative damage to various cell types, RPE, rod photoreceptors and vascular endothelial cells, we examined the autonomous and non-autonomous effects of oxidative stress in the eye. Inducing oxidative damage in the RPE caused degeneration of RPE cells and photoreceptors, and in severe cases, vascular leakage, indicating that oxidative stress in the RPE can elicit both intrinsic and extrinsic degenerative effects. In contrast, we found that targeting oxidative damage to the endothelial cells did not lead to vision loss, or degeneration of RPE or photoreceptors. However, when oxidative damage was directed to rod photoreceptors, low D-alanine concentrations led to RPE degeneration without rod loss, indicating non-autonomous oxidative damage in the eye. Taken together, these findings demonstrate that this chemogenetic model phenocopies dry AMD when targeted to the RPE and provides a powerful system to study cell-autonomous and non-autonomous oxidative damage in the eye.Biological and Biomedical Science
Elucidating novel chemical and genetic mechanisms of LSD1-HDAC1/2-CoREST complex regulation
No full textProtein complexes have a wide range of functions that can be modulated through both enzymatic activity and protein-protein interactions. In efforts to design small molecule therapeutics that target specific protein functions, molecules with diverse mechanisms of action have been developed. Traditionally, drug design has focused on inhibiting an enzyme’s catalytic activity, either by directly blocking its active site or by designing an allosteric modulator. However, beyond catalytic function, enzymes often serve as scaffolds that mediate interactions with other proteins—interactions that influence the overall function of a protein complex within cells. Discovering new small molecule modalities that selectively target one function of a protein can optimize therapies for various disease indications, while also offering insight into novel mechanisms of protein regulation and disease pathogenesis.
In this thesis, I investigate the mechanisms of action of two recently developed small molecule modalities targeting the LSD1-HDAC1/2-CoREST (LHC) complex and uncover how mutations in an E3 ligase substrate adaptor lead to a new mode of LHC complex dysregulation in cells. In Chapter 1, I provide an overview of the LHC complex and key structural studies that have elucidated its fundamental functions and biological roles in cells and human development. I then review its involvement in disease and therapeutic applications, which has driven efforts to identify small molecules that target the complex. Finally, I introduce the concepts of targeted protein degradation—a therapeutic strategy recently applied to targeting the LHC complex—and genomic screening methodologies that we have used to evaluate novel mechanisms of protein regulation.
In Chapter 2, I describe our work elucidating the mechanism of action of the small molecule T-448, an LSD1 inhibitor that selectively targets enzymatic activity while preserving LSD1’s interactions with transcription factors. Through mass spectrometry and structure-activity relationship studies with T-448 analogs, we found that T-448 forms a covalent drug-FAD adduct in LSD1’s active site, which subsequently undergoes Grob fragmentation to yield a compact formyl-FAD adduct. This adduct preserves LSD1’s scaffolding function with transcription factors such as GFI1/GFI1B, thereby reducing hematological toxicity and making T-448 a more promising candidate for treating neurological disorders. Additionally, we show that this conversion from drug-FAD to formyl-FAD can serve as a resistance mechanism in AML cells. Using CRISPR suppressor scanning, we previously identified a loop deletion mutation distal to the catalytic site that confers resistance to certain LSD1 inhibitors through this mechanism. Altogether, this work highlights how small molecule design can target specific LSD1 functions and how distal loop mutations can impact drug mechanism of action.
In Chapter 3, I detail our investigation into the mechanism of action of another LSD1-targeting molecule, UM171. It was initially identified in a phenotypic screen and later shown to induce degradation of LSD1 and CoREST via the E3 ubiquitin ligase substrate adaptor KBTBD4, however the direct binding partners of UM171 remained unknown. Using fluorescence-based cellular assays and biochemical binding studies, we determined that HDAC1/2 is the direct binding partner of UM171 within the LHC complex. UM171 acts as a molecular glue and increases the affinity between LHC and KBTBD4. We validated this by solving a cryo-EM structure of the KBTBD4-UM171-LHC complex and unexpectedly identified a second molecular glue—inositol hexakisphosphate—at the binding interface. The structure revealed that the KBTBD4 dimer engages a single copy of the HDAC1/2-CoREST complex asymmetrically. Additionally, base editing scanning was performed and confirmed UM171’s binding sites. This was the first study to elucidate the binding mode of a molecular glue that engages a Cullin3 E3 ligase substrate adaptor, revealing the mechanism behind LSD1 and CoREST degradation.
In Chapter 4, I move beyond small molecule regulation to explore how mutations can also modulate protein-protein interactions. In medulloblastoma, complex insertion-deletion and substitution mutations have been found in a single loop in the Kelch domain of KBTBD4. In patient-derived xenograft models harboring these mutations, CoREST and LSD1 levels are depleted and knockout of mutant KBTBD4 rescues these protein levels. Through a series of biochemical assays, we also discover that the mutant KBTBD4 have increased binding affinities to LHC. To understand the scope of these gain-of-function mutations, we performed a deep mutational scan of the loop mutated in medulloblastoma. Surprisingly, a wide array of mutations promoted CoREST degradation and further analysis was performed to determine the types of mutations that had the strongest phenotype. Structural studies were performed to determine how these KBTBD4 variants were engaging the LHC substrate and revealed that mutant KBTBD4 mimics the UM171-induced binding mode observed in Chapter 3. This structural insight prompted us to test known HDAC active-site inhibitors as a potential therapeutic strategy to disrupt the mutant KBTBD4-LHC interaction. Through this study, we show that KBTBD4 cancer mutations chemically and functionally mimic UM171 and that deep mutational scanning can identify mutations that drive substrate degradation. We also demonstrate that active-site inhibitors can be repurposed to disrupt pathogenic protein-protein interactions.
Overall, this thesis showcases multiple ways to perturb the LHC protein complex, via distinct small molecule modalities and neomorphic mutations that induce novel protein interactions leading to degradation. These studies not only teach us how we can differentially regulate the LHC complex for different therapeutic applications but also serve as an instructive example of how we can use both chemical and genetic methodologies to induce similar phenotypic effects. This work opens the door to applying high-throughput genomics to discover novel modulators for other protein complexes and cellular pathways in various disease contexts.Chemistry and Chemical Biolog
Multiplexed auditory synchronization: Measuring distraction in listeners with normal hearing and with tinnitus
No full textDespite numerous sounds presented to our ears, people typically want to choose to listen to only one. Some unwanted sounds can be distracting and harder to suppress while others are just background noise. Along with environmental sources of sound, individuals with tinnitus have an additional competitor: an unflagging auditory percept with no external generator. For some individuals with tinnitus, this phantom sound impinges on them constantly; for others, the phantom fades harmlessly into the background unless they attend to it. The difference in how these groups experience tinnitus may be related to their abilities to suppress distracting sounds more generally.
To test this hypothesis, we developed a novel paradigm for measuring behavioral and neurophysiological measures sensitive to the level of distraction provided by competing sounds. Participants are presented with a target stimulus organized around nested timescales, including temporal fine structure (~500 Hz), envelope (~25-80 Hz), envelope changes (~7 Hz), and embedded context (~0.5 Hz). EEG is recorded to capture synchronization to the features across these timescales as participants make perceptual judgments about the embedded context. These target stimuli are presented alongside two different types of distractors—melodic and noise distractors—which share low-level acoustic features but differ in the amount of distraction they produce.
Normal hearing participants are more distracted by the melodic distractors than the noise distractors. While the inclusion of the distractor reduces synchronization across all timescales, only the slowest synchronization to envelope changes, the envelope change following response (ECFR), is sensitive to the level of distraction. The ECFR is reduced when the target is paired with the melodic distractor compared to the noise distractor and is also reduced when participants make errors in perceptual judgment about the target stimulus. Results in participants with tinnitus are in line with results from participants with normal hearing. When behavioral and ECFR results are compared between a group with lower tinnitus burden and a group with greater tinnitus burden, the groups do not differ.
Our novel paradigm simultaneously provides information about synchronization to up to 9 different features of auditory stimuli at nested timescales. This paradigm has yielded the ECFR, a newly described synchronization measure that is sensitive to the level of auditory distraction. Such a measure may prove useful in a variety of populations who have difficulty suppressing awareness of unwanted sounds. In participants with tinnitus, this measure was not associated with the level of tinnitus burden, suggesting that suppression of externally presented distractors and internally generated phantom sounds utilize different resources.Speech and Hearing Bioscience and Technolog