1983 research outputs found
Sort by
Inferring dynamic extracellular matrix composition of the thymus: towards a biomimetic scaffold for T cell culture
No full textImmunotherapy is a pioneering approach using T lymphocytes (T cells) to fight cancer and immune deficiency, which together affect millions of patients in the US. Current clinical therapies require patient- or donor-derived T cells, leading to manufacturing delays and limited supply. While stem cell-derived T cells offer a scalable strategy to meet growing demand and increase accessibility to therapy, generating mature and clinically usable T cells through this method is challenging and cell yields remain low.
In the body, T cell maturation relies on cell migration through thymic structures that change chemically and physically over time; the extracellular matrix (ECM) has a role in shaping these evolving niches. Existing solutions to producing “off the shelf” engineered T cells use mouse thymic epithelial cell lines and native ECM to adapt key in vivo components of the T differentiation process. However, these solutions are not fully synthetic, limiting their scalability and potential therapeutic use. Additionally, these models of the thymic ECM overlook developmental shifts in thymic structure that might advance an understanding of T cell maturation.
Therefore, dynamic biological and chemical properties in the thymic microenvironment are important in efficiently producing mature T cells in vivo. This project constructs a proof-of- concept biomimetic platform for immune cell differentiation from stem cells based on these changing characteristics of the thymic extracellular matrix. Using single-cell RNA sequencing data to quantify gene expression levels, the extracellular matrix composition of the thymus at different developmental stages is computationally characterized, and characteristic ratios of key extracellular matrix components (fibronectin, collagen, and laminin) are identified at these different stages. These ratios are then used to engineer a stage-specific alginate-based hydrogel for cell culture, designed to replicate thymic tissue stiffness and viscoelasticity with tunable properties that reflect different developmental stages. Using this stage-specific platform, researchers can investigate how T cell differentiation, activation, and toxicity are influenced by the thymus's developmental stage. Additionally, this platform could allow for the identification of the optimal stage of thymic development to mimic in T cell culture scaffolds, enabling more scalable T cell expansion.Engineering Sciences S
Learning-Based Methods for Recovering Visual Structure
No full textExtracting explicit geometric structure from image data is a prerequisite for understanding visual scenes. This process manifests in 2D as the recovery of curvilinear boundaries that delineate objects. Similarly, in the 3D realm, it involves the derivation of scene surfaces from a set of multiple images. These transformations from images to structured representations are notoriously difficult inverse problems complicated by sparse data, misleading local evidence, and geometric complexity. Historically, approaches to this challenge have tended to bifurcate into two distinct paradigms: "geometry-first" methods that rely on rigorous, but restrictive, mathematical priors; and "learning-first" methods that prioritize data-driven scalability but lack interpretability and struggle to generalize beyond their training sets. This dissertation explores two cases of \textit{structured differentiability}, a synthesis of these paradigms that aims to overcome their limitations by embedding geometric objectives and inductive bias directly into differentiable, learning-based formulations.
First, in the context of boundary detection in 2D images, we introduce a lightweight network that employs a differentiable, geometry-aware attention mechanism to resolve ambiguities and recover from measurement noise. Our model decomposes an image into a field of geometric primitives, thereby preserving the geometric precision of geometry-first methods, while leveraging the inference speed and data-driven scalability of neural networks.
Second, we address the challenge of novel view synthesis, where the goal is to recover underlying surface geometry and appearance from a set of images to predict novel viewpoints. We build upon fast and effective splatting-based methods, which represent scene structure as millions of discrete primitives defined by their shape, color, and opacity. To overcome the limitations of traditional methods, which depend on manual tuning, we propose a probabilistic reformulation of 3D Gaussian Splatting. Rather than relying on the heuristic split-and-prune strategies traditionally used to manage surface primitives, we define a continuous, learnable probability distribution from which primitives are sampled. This transforms the allocation of geometry from a set of rigid, discrete rules into a fully differentiable process, allowing gradient descent to naturally concentrate representational capacity where it is needed most.
Collectively, these contributions demonstrate that coupling data-driven learning with geometrically grounded, differentiable objectives reconciles the interpretability of explicit modeling with the empirical power of deep learning, yielding recovery processes that are efficient, interpretable, and robust to real-world ambiguity.Engineering and Applied Sciences - Engineering Science
Essays on Health and Economic Outcomes over the Life-Course
No full textThis dissertation studies how health and health policies impact economic outcomes like earnings, employment, disability application, and wealth accumulation throughout the life course. The first two chapters focus on the role of childhood health in shaping life-long economic inequality. The third paper focuses on outcomes after disability onset in midlife and the role of employer-sponsored health insurance.
Chapter 1 investigates the role of childhood health on economic outcomes over the life-cycle using uniquely detailed data from the Health and Retirement Study with a linkage to Social Security administrative earnings records. I first present several key facts about the role of childhood health in long-term economic outcomes: childhood health is associated with substantial gaps in earnings and employment throughout the prime working years; gaps peak in midlife when poorer childhood health is associated with a 7.5 percentage point reduction in employment and a reduction of $6,000 in earnings; and these gaps translate to substantial gaps in wealth in retirement. I then implement a quasi-experimental approach using variation in access to the measles vaccine upon its introduction to test for a causal effect of childhood health on later-life economic outcomes. While I find no evidence for a causal link between the measles vaccine and later life outcomes, my results reflect a need to better understand the specific channels through which childhood health effects persist.
Chapter 2 extends on work in Chapter 1 and investigates the role of childhood health in shaping racial economic inequality over the life course. Many of the same structural and systemic barriers that impact racial economic disparities have also produced disparities in child health among Black Americans, and childhood health has consistently been shown to be a predictor of labor market and human capital outcomes later in life. Using data from the Health and Retirement Study with a linkage to Social Security Administration earnings records, I find that poorer childhood health is associated with disproportionately large reductions in earnings and employment for Black respondents over the life course compared to White respondents. In Oaxaca-Blinder decompositions, I find that up to 2.4% of the earnings gap and 12.2% of the employment gap in my sample can be attributed to childhood health, with the contribution being greatest in midlife. Conversely, childhood health has little role in shaping Black wealth or the Black-White wealth gap, likely reflecting the greater influence of other factors in shaping these outcomes. These results highlight a small, but potentially important impact of childhood health as well as the need to continue addressing structural barriers to racial economic equality in the U.S.
Chapter 3 (with Nicole Maestas and Kathleen Mullen) studies the potential for “employment lock”, a phenomenon where individuals are reluctant to leave their current job due to fear of losing their employer-sponsored health insurance coverage, among workers experiencing the onset of a new disability. We use prospective longitudinal data on newly disabled older workers from the Health and Retirement Study to examine the effect of employer sponsorship of health insurance (ESHI) on post-onset employment and disability insurance claiming. We compare outcomes of workers with ESHI and no access to another coverage source immediately prior to onset with outcomes of two comparison groups: individuals with ESHI who also have access to an alternative coverage source, those who are covered by coverage source other than ESHI prior to onset. We find limited evidence of “employment lock” only when restricting to newly disabled older workers. We find no evidence that ESHI impedes application for Social Security Disability Insurance, but ESHI is associated with higher rates of disability receipt for newly disabled older workers.Health Polic
Base Editing–Enabled Technologies and Multiplex Genome Editing
No full textBase editing is a precision genome editing technology that enables targeted single-nucleotide changes without requiring double-strand DNA breaks. This dissertation presents several distinct advances in multiplex base editing and base editing–enabled technologies in mammalian systems. First, we demonstrate large-scale editing of transposable elements, which can be targeted with a single guide RNA due to their repetitive nature. Using catalytically inactive Cas9 base editors, which minimize editing-associated cytotoxicity, we achieve several thousand edits per cell. Second, we develop Genomic Sequence Encryption (GSE), a cryptographic framework that uses multiplex base editing and pooled guide RNAs to encode information across more than one hundred distinct genomic loci. We implement GSE in mammalian cell lines and stem cells, establishing a robust method for introducing a high number of edits in both bulk populations and individual stem cells. We devise an enrichment strategy that enables the isolation of stem cells carrying more than two dozen distinct precision edits across a single diploid genome with minimal screening. This represents a significant advancement in the scale of simultaneous precision editing achievable in primary or stem cells, and, in the context of GSE, paves the way for encrypted genomic signatures in living animals. Lastly, we develop reprogrammable ADAR sensors, a programmable RNA-sensing platform that links endogenous transcript detection to protein translation through A-to-I RNA editing. Together, these contributions expand the scope of base editing by enabling large-scale genome modification, secure biological information encoding, and transcript-responsive regulation in mammalian cells.Engineering and Applied Sciences - Engineering Science
Development of CD70 CAR-NK Cells with TGF-β Sensor-Controlled and Tumor-Directed IL-12 Expression against Clear Cell Renal Cell Carcinoma
No full textChimeric antigen receptor (CAR)-engineered T and NK cell therapies have shown promising success in treating hematological malignancies, yet their efficacy in solid tumor remains limited due to poor infiltration and reduced expansion and persistence in the tumor microenvironment (TME). To address these major challenges, we engineered a novel TGF-β sensor-controlled and tumor matrix-directed IL-12 secreting CD70 CAR-NK cell and evaluated it against clear cell renal cell carcinoma (ccRCC). In vitro studies confirmed that CD70 CAR-NK cells exhibited more potent tumor antigen-specific cytotoxicity, and both secretory and collagen-binding IL-12 enhanced killing efficiency by two folds against ccRCC cell lines. IL-12 secreting CD70 CAR-NK cells also demonstrated increased IFN-γ production and degranulation. Mice study indicated that these engineered NK cells effectively suppressed tumor growth in vivo. To further counteract TGF-β-induced immunosuppression within the TME and ensure controlled IL-12 release for improved clinical safety, we designed a novel TGF-β sensor based on SMAD-binding motifs. We showed that the TGF-β sensor successfully upregulated EGFP reporter protein expression in a TGF-β dose-dependent manner. Positioning the sensor on the reverse strand with IL-12 secretion domain precisely regulated IL-12 release upon TGF-β1 stimulation with minimal leakage. In vitro studies revealed that TGF-β sensor-controlled IL-12 secreting CD70 CAR-NK cells displayed superior tumor-killing capacity comparable to IL-12 secreting CD70 CAR-NK cells even at a low transduction rate. Overall, this study described a novel strategy to overcome the immunosuppressive TME and enhance tumor-specific killing in CAR-NK cell therapy while ensuring safety, supporting it as an effective therapeutic approach against ccRCC.Graduate Educatio
Interrogating sequence-structure-function relationships in DNMT3A
No full textDNMT3A is a complex chromatin-modifying enzyme with a critical role in mammalian development and cancer biology. As one of two human de novo DNA methyltransferases, DNMT3A is responsible for establishing and maintaining proper gene expression profiles that allow the cell to differentiate properly during development. Dysregulation of this process in hematopoietic stem cells can lead to improper gene expression programs and diseases such as clonal hematopoiesis and Acute Myeloid Leukemia, as well as developmental disorders such as Tatton Brown Rahman syndrome. The role of DNMT3A in the genome has been extensively studied, and progress has been made towards understanding the role of DNMT3A at this genomic scale. However, at a molecular scale, many mysteries regarding the precise biochemical mechanisms by which DNMT3A performs its functions and is altered in disease remain. Specifically, the precise molecular mechanism by which the AML hotspot R882H mutation in DNMT3A causes its dominant negative effect remains unclear. In this dissertation, I will describe my work applying biochemical, genetic, and biophysical techniques to unravel the molecular basis of DNMT3A activity and its perturbation in disease.
In chapter 1, I will begin by giving an overview of what is currently known about DNMT3A biology. Then, I will briefly review literature on the evolution and biochemistry of protein-protein interfaces, with a focus on homo-oligomers. This body of work provides important context for the oligomeric behavior of DNMT3A and particularly the R882H mutation. Then, I will provide an overview of deep mutational scanning, broadly defined. Deep mutational scanning is a powerful genetic technique that, driven by increases in the affordability of DNA sequencing and synthesis technology as well as implementation of CRISPR-based methods, has recently diversified in both questions addressed and methods.
In chapter 2, I use base editor scanning to look for activating mutations to DNMT3A. In doing so, I uncover several activating mutations. These include a mutation at the ADD-MTase autoinhibitory interface, a cryptic splice site in the DNMT3A gene that causes a four amino acid deletion in the ADD-MTase linker, and a C-to-Y mutation in the ADD domain that based on existing structures is unlikely to perturb the known ADD-based autoinhibitory mechanism. In particular, this C-Y mutation may inform drug discovery efforts to design and understand small molecule DNMT3A activators.
In chapter 3, I use deep mutational scanning, biochemistry, and Hydrogen-Deuterium Exchange Mass Spectrometry (HDX-MS) in collaboration with Shaunak Raval and Malvina Papanastasiou to uncover the molecular mechanism of the hotspot R882H mutation. First, we show that the R882H mutation causes the mutant protein to form aberrant oligomers, explaining its dominant negative effect. Then, we use a rescue mutant of DNMT3A R882H, L859F, to propose a model in which the R882H mutation pre-orders the residues that form the RD interface of DNMT3A into their binding-competent state, promoting oligomerization. L859F instead opposes this pre-ordering, rescuing the oligomeric state of the protein. This work both solidifies the oligomerization-promoting effect of the R882H mutation and provides evidence for its biophysical mechanism, which will be useful in drug discovery efforts to treat diseases associated with this mutation. Furthermore, this work describes the first report for this type of aberrant biochemical mechanism in cancer, providing a mechanistic framework that may be operative in other disease contexts as well.
In chapter 4, I discuss unpublished work exploring the role of the N-terminal regulatory domains in DNMT3A oligomerization and activity. This preliminary data suggests that the PWWP domain may make contacts with the MTase domain that are important for enzyme function, and that these contacts may also be involved in the ability of DNMT3A to form the RD interface. I also show that these properties seem to be exclusive to the DNMT3A PWWP domain, through experiments subbing the DNMT3B PWWP domain for that of DNMT3A. Finally, in chapter 5 I discuss conclusions and future directions to build upon this dissertation.Chemical Biolog
Cathodoluminescent Probes for Multicolor Electron Microscopy
No full textCathodoluminescence (CL) microscopy offers a promising approach to nanoscale analysis, enabling detection of optical emission from a sample while leveraging the high resolution of electron microscopy (EM). However, achieving multicolor single-particle CL imaging remains a significant challenge. Here, we establish lanthanide nanoparticles (LNPs) as a model system for multicolor CL imaging. We identify the critical limitation that precluded multicolor CL imaging—nonlocal signal caused by stray electrons—and mitigate these nonlocal excitations to demonstrate multicolor single-particle CL imaging.
To be viable for multicolor CL imaging applications, LNPs must be available in multiple emission colors. Therefore, having achieved single-particle CL imaging, we use this method to study the photophysical properties of LNPs and expand their multiplexing capability. We determine the dependence of LNP brightness on lanthanide ion concentration, develop a method to measure CL excited state lifetimes of LNPs, and study energy transfer between lanthanide ions. Next, we combine multiple lanthanide elements to engineer unique LNP colors and use them for seven-color CL imaging.
Applying CL probes as bioimaging labels would enable simultaneous visualization of cellular structures (via EM contrast) and specific biomolecules (via CL contrast) at the nanoscale resolution of EM. However, achieving this is challenging because LNP synthesis yields hydrophobic nanoparticles, limiting their utility as bioimaging labels. To address this challenge, we functionalize LNPs with DNA to produce hydrophilic LNPs. We show that their single-particle CL emission is retained after DNA functionalization and after common EM sample preparation steps, and demonstrate nanoscale, multicolor CL imaging of DNA-functionalized LNPs in a biological sample.
Finally, we explore the viability of small-molecule fluorescent dyes as CL labels. We show that these dyes can be excited by an electron beam and emit CL signal. We demonstrate three-color CL imaging using dye-loaded polymer beads, and two-color CL imaging of mammalian cells with dye-labeled organelles, illustrating the potential of small-molecule fluorescent dyes for CL bioimaging. Together, this work establishes CL as a useful contrast mechanism for high-resolution, multicolor electron microscopy and represents a significant step toward the application of cathodoluminescent probes for simultaneous imaging of cellular structures and biomolecules.Biology, Molecular and Cellula
Elucidating Cellular Dependencies of O-GlcNAc Transferase Structure and Function
No full textO-GlcNAc transferase (OGT) is an essential enzyme and the most conserved glycosyltransferase in humans. Previous studies of OGT at both the cell and organismal level have modulated OGT with genetic deletion, pharmacological inhibition, or RNA interference. These studies utilized methods where phenotypes were scored following prolonged genetic perturbation or probes that exhibited off-target effects. Moreover, the authors attributed phenotypes associated with knockout or knockdown of OGT as being solely due to a lack of glycosyltransferase activity. However, OGT has numerous binding partners and has been observed in multiprotein complexes, suggesting noncatalytic roles that may regulate cellular physiology.
In this thesis, I use next-generation chemical and genetic tools to investigate how changes in OGT’s catalytic and non-catalytic functions alter organellar and cellular physiology. First, I introduce this complex enzyme and previous attempts to define its role in cells. In Chapter 2, I collaborate with a graduate student to investigate how truncations of the tetratricopeptide repeat (TPR) domain of OGT alters its subcellular localization and subsequently cellular viability. In Chapter 3, I present a three-arm chemical genetic screen that separates OGT’s catalytic and non-catalytic synthetic lethal partners. In Chapter 4, I elucidate how a chemical probe that specifically inhibits OGT alters mitochondrial physiology on the scale of hours, suggesting OGT’s essentiality is linked to its role in mitochondrial homeostasis. Finally, in chapter 5 I conclude that OGT’s catalytic functions drive its essentiality and suggest future directions for OGT studies in cells.Chemical Biolog
Quantum algorithms and quantum error correction with neutral atoms
No full textQuantum computers have the potential to solve certain problems exponentially faster than classical computers. However, realizing this potential in practice is a major challenge, as it requires precise control over large-scale quantum systems operating at extremely low error rates. Quantum error correction (QEC) provides a path to achieving such error rates, but its substantial resource overhead poses a significant practical barrier. In addition, identifying which computational problems offer a provable quantum advantage—and which remain fundamentally intractable—remains an open question. This thesis presents progress towards addressing both of these challenges.
The first part of this thesis describes advances that substantially reduce the resource overhead of QEC. We begin by presenting realizations of logical circuits in dynamically reconfigurable arrays of neutral atoms. By jointly decoding the logical qubits, we reduce the cost of implementing such transversal Clifford circuits by a factor proportional to the code distance. We then develop new theories of fault tolerance to extend these savings to universal quantum computation with magic state inputs. Finally, we introduce techniques for fast correlated decoding, enabling practical implementations of these improvements in experimental hardware. Collectively, these advances accelerate progress toward large-scale computation and reduce the cost of QEC by over an order of magnitude.
The second part of the thesis explores the relative power of quantum and classical computation in combinatorial optimization, a class of problems that is ubiquitous in science and engineering and foundational to the theory of computational complexity. We experimentally implement the optimized quantum adiabatic algorithm (QAA) in neutral atom arrays and observe evidence of a superlinear speedup over classical simulated annealing on certain hard problem instances. To interpret these results, we develop a theoretical framework to compare the performance of QAA with a broad class of classical Markov chain Monte Carlo algorithms. We identify conditions under which a quantum quadratic speedup is achievable and propose modifications to the QAA to reliably realize this advantage. Together, these contributions advance both the practical implementation and theoretical understanding of quantum algorithms.Physic
Microbial Evolution through the Lens of Metagenomics and Archaeogenetics
No full textMicrobial evolution is fundamental to fields ranging from industrial bioprocesses to agriculture and medicine, governing everything from pathogen emergence and transmission to environmental ecosystems and biotechnological advancement. However, microbiology’s historical reliance on culture-based methods and primarily clinically focused studies has limited our ability to fully integrate ecological and evolutionary perspectives, leaving critical gaps in knowledge regarding the genomic structure, diversity, and evolutionary dynamics of microbial populations. Host-microbial associations and microbial community interactions in particular drive several key biological events across Earth’s history - from organelle origins to evolutionary arms race dynamics. Recent advances in sequencing technologies have revealed a startling expansive picture of the genomic diversity of symbiotic microbes – including the enigmatic Candidate Phyla Radiation (CPR) – a group of abundant yet largely uncultured, ultra-small bacteria that are found ubiquitously from deep-sea hydrothermal vents to the human microbiome. In this dissertation, I integrate metagenomic, pangenomic, and paleogenomic approaches to investigate microbial eco-evolutionary dynamics. Altogether, this work illuminates how genomic plasticity facilitates microbial adaptation and diversification, with broad implications for the ecological understanding of microbial relationships and insights into host-microbial co-evolution.
In my first chapter, I utilize a novel statistical framework that combines longitudinal metagenomic sampling with clonal sequencing to track the strain-level population dynamics of industrial yeast (Saccharomyces cerevisiae) lineages across two Brazilian bioethanol refineries over two industrial seasons. The results show diverging evolutionary trajectories: one plant’s yeast community is characterized by the stable dominance of a domesticated starter lineage, whereas another plant experiences invasion by foreign but closely related strains. These findings highlight how ecological forces, such as competition and invasion, influence microbial communities within industrial processes in relation to industrial operational stability.
In my second chapter, I develop a scalable, integrative computational pipeline that combines metagenomic assembly and pangenomics to characterize the genome structure of over two thousand Parcubacteria (OD1) genomes in their environmental context, including novel genomes recently discovered from deep-sea anemones. Utilizing this approach, I demonstrate that Parcubacteria – despite having extremely reduced core genomes with limited metabolic capabilities – retain a flexible and modular accessory genome across clades, structured by phylogeny rather than habitat specificity. With a core genome that encodes primarily informational systems, DNA recombining and repair mechanisms, environmental sensing, and a conserved type IV pili, the Parcubacteria core genome reflects a host-dependent interaction-focused lifestyle. In contrast, the accessory genome exhibits remarkable modularity, with lineage-specific gene clusters encoding diverse specialized functions in secondary metabolism, stress responses, and signal transduction systems, with complete turnover between clades. Strikingly, co-occurrence analyses also show extensive genomic plasticity driven by insertion sequence (IS) elements – especially the IS21 family, and reveal structured mobility of antibiotic resistance genes (ARGs) and accessory gene clusters, highlighting ongoing gene transfer and mobility without disrupting core genomic integrity. Altogether, the results showcase a unique evolutionary strategy for small genomes where genome reduction is coupled with modular genomic innovation. Altogether, these findings redefine genome reduction paradigms by illustrating how Parcubacteria leverage dynamic accessory content and interaction alongside its genomic minimalism for flexible specialization and ecological persistence.
My third chapter employs a combination of archaeogenetics and Bayesian tip-calibrated phylogenetics to reconstruct the evolutionary history and habitat transitions of the prominent CPR lineage Saccharimonadia (TM7), a globally distributed bacterial group that is also commonly found in human microbiomes. Leveraging ancient DNA derived from archaeological samples from ancient human populations and Neanderthals spanning a range of 100,000 years, as well as oral microbiome samples from underrepresented traditional farming and hunter-gatherer communities, I curated a dataset of 4,317 genomes, establishing one of the most comprehensive temporal datasets available for a CPR bacterial group. I identified at least seven independent habitat transitions from environmental reservoirs into mammalian hosts, with distinct lineage diversifications driven by host colonization and subsequent specialization in oral or gut biofilms. Bayesian evolutionary analyses were used to calculate substitution rates, with TMRCA (Time to the Most Recent Common Ancestor) dating diversification events into the Pleistocene epoch. Notably, my analyses also uncovered several previously unrecognized human-associated lineages that persist within ancient and non-industrialized human populations, indicating underrepresentation linked to human lifestyle or subsistence differences. Altogether, these findings underscore the significance of ancient DNA approaches in providing high-resolution evolutionary modeling for tracking long-term microbial evolutionary trajectories, and revealed how evolutionary events shape host-specific diversity patterns observed in modern Saccharimonadia populations.Biology, Organismic and Evolutionar