Rising Stars: Women 2019


Atrial Fibrillation (AF) is the most prevalent sustained heart rhythm disorder, contributing significantly to global health care costs and rates of mortality and morbidity. For AF patients who do not respond to drugs, the treatment of choice is catheter ablation to isolate arrhythmia triggers and eliminate the substrate for arrhythmia perpetuation. However, ablation outcomes are dismal due to extensive structural remodeling of the atria which confounds all known strategies for identifying ablation targets. Thus, there is an urgent need for new strategies to swiftly and accurately identify optimal AF ablation targets. The overall aim of my research is to develop a novel, non-invasive approach for carrying out custom-tailored catheter ablation procedures based on computational simulations in AF patients. Successful completion of this research could set the stage for a shift towards a new AF treatment paradigm featuring tightly-coupled interaction between computational and clinical teams.


Long-lasting auditory deficits following temporary developmental hearing loss

Permanent hearing loss that begins during an early childhood critical period can lead to severe speech and language deficits, yet even a transient period of hearing loss can also be detrimental to auditory function. In fact, the severity of these deficits correlates with hearing loss duration, suggesting that the central nervous system remains vulnerable to sensory deprivation well into adolescence-an age range that is seldom considered in experimental studies. To facilitate effective strategies for intervention and remediation of hearing deficits that originate in the nervous system, it is essential to utilize animal models to address how auditory experience, both normal and degraded, can shape the neural mechanisms that support perceptual skills. Therefore, I use a favorable animal model, the gerbil, to determine how long duration developmental hearing loss impacts auditory processing. The experimental approach uses earplugs to temporarily induce mild-to-moderate hearing loss, and a behavioral task to assess sensitivity to modulations in stimulus amplitude, a sound cue that is foundational to all natural sounds, including speech. Finally, I measure auditory cortex neuron functional properties to determine whether long periods of temporary hearing loss induce long-lasting changes that could serve as a candidate for remediation.


Close loop control of cognitive flexibility by intermittent high frequency stimulation of the dorsal striatum.

Neuropsychiatric disorders are the number one cause of disability in the United States. Pharmacotherapies modulating the neurotransmitter system have limited benefit and sometimes serious side effects. This has stimulated interest in neuromodulation techniques such as deep brain stimulation which aims to modulate disrupted brain networks associated with psychiatric disorders. Understanding how to drive the pathological brain network towards a healthier state is essential for the success of these neuromodulation therapies. Furthermore, changes in cognition/behavior from the normal to pathological state are not always well defined and are highly variable across psychiatric patient populations, making assessment of an intervention non-trivial. DBS programming in movement disorders for setting effective stimulation parameters is largely guided by visual feedback of the improvement in symptoms such as tremor and mobility. In psychiatric disorders, the effect on symptoms cannot be immediately assessed by visual inspection and can take months to be effective.Such challenges necessitate the identification of a quantitative measure, such as brain rhythms associated with disrupted cognitive function, which may serve as a target for designing intervention strategies.

Dysregulated cognitive flexibility and conflict resolution are core features of multiple psychiatric disorders such as mood and anxiety disorders. Cognitive control in the setting of response conflict is often studied through cognitive tasks where subjects must suppress a natural response to follow a less intuitive rule. In a recent study, we showed that chronic ventral capsule/ventral striatum DBS enhances cognitive control, that this enhanced control is correlated with pre-frontal theta oscillations, and that the increased theta power is in turn correlated with clinical recovery. For achieving an on-demand stimulation paradigm, we investigated if short bursts of high frequency intermittent stimulation in the striatum can significantly improve cognitive control in human subjects performing a conflict task. We show that we can not only improve cognitive control but can also use this data to design closed loop stimulation for specifically targeting a specific underlying cognitive state estimated from the task response. Furthermore, we model simultaneously recorded local field potential and task behavior to find neural features than can be effectively used for decoding hidden cognitive states and modulate them by external stimulation. We demonstrate a neural decoder based closed loop control of cognitive flexibility in 3 human subjects. Finally, we establish that there is a shift in the neural encoding of cognitive control with and without external intermittent striatal stimulation. These results bring us closer to the design of a closed loop brain stimulation paradigm for improving cognitive control in patients with addiction, anxiety and mood disorders.


Engineered Heart Slice Preparation for Drug Studies and Disease Modeling

Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) hold great promise for cardiac studies, representing a large, patient-specific pool of cells that can be used for drug discovery and screening and for modeling cardiomyopathies. However, clinically-relevant findings using hiPSC-CMs require electrophysiological and contractile studies on multi-cellular functional syncytia of these cells. Such studies are often performed on cell monolayers, but more complex models that incorporate extracellular cues present in the native heart will be necessary to recapitulate the behavior of cardiac tissue. My work has centered on the development of engineered heart slices (EHS), preparations of decellularized myocardium repopulated with hiPSC-CMs that incorporate the biochemical and topographical cues of the cardiac extracellular matrix and exhibit structural and functional improvements over standard monolayer cultures. EHS exhibit multicellular, aligned bundles of elongated cardiomyocytes with organized sarcomeres, anisotropic conduction of action potentials, and electrophysiological functionality for more than 200 days. Importantly, EHS have several advantages over monolayers for 1) studying the effects of ion channel modulating drugs and 2) modeling arrhythmogenic cardiomyopathy (AC), a disease in which genetically‐encoded mutations of the desmosomes that interconnect cells lead to aberrant electrical conduction and arrhythmias and for which multicellular syncytial models are needed to study underlying disease mechanisms. When subjected to an assay to measure the response of drugs at different doses and pacing rates, EHS were less sensitive than monolayers to blockers of potassium and calcium channels, reflecting a response closer to that of cardiac tissue. Further, AC‐associated transcription patterns were present in EHS, reflecting an induction of the disease phenotype that was not seen in monolayers, and resulting in a preparation appropriate for functional electrophysiological studies of AC. Thus, the EHS preparation is a tissue-like model that enables long term culture, structural and functional improvement, higher fidelity drug response, and enhanced disease models of hiPSC-CMs.


Engineering stem cell-derived tissues for hindbrain and spinal cord therapeutics

Neurotrauma to the central nervous system can lead to a debilitating lifestyle for individuals and care givers. While decades of research have focused on regenerative strategies to provide functional recovery, no cure has been developed. One route of regeneration is through cell replacement strategies to repair circuity following a traumatic event. Stem cells offer a “blank slate” starting material for cell replacement therapies because they can differentiate into any cell found in an organism when exposed to the appropriate cues.  My research has focused on developing stem cell-derived spinal cord and hindbrain cells and tissues that can be used as therapeutics and models of development.

In one example of engineering a cell therapy, I discovered how to differentiate V2a interneurons, an excitatory interneuron found in the hindbrain and spinal cord, from mouse and human embryonic stem cells (ESCs). V2a interneurons are a candidate cell replacement therapy for spinal cord injury as they receive motor signaling from the hindbrain and relay information downstream to motoneurons in the spinal cord. To study how neurons interact when differentiated together, I developed a hindbrain organoid model from human ESCs that contains multiple neurons that are involved in the breathing circuit. While these discoveries have advanced the field of stem cell engineering, the differentiation strategies can be improved by comparing to in vivo tissues. To elucidate how in vivodevelopment occurs in the developing hindbrain, I am using single cell transcriptomic analysis to define the gene network that drives a homogenous progenitor population to differentiate into over 40 different neuronal nuclei throughout the cerebellum and brainstem. In tandem, I will compare hindbrain organoid models to in vivodevelopment to identify how in vitromodels may fail and use the findings to design improved models. Together, my research goals are to elucidate the gene networks that drive fate neuronal decisions to ultimately develop cell replacement strategies and in vitro models to improve therapies for neurotrauma to the hindbrain and spinal cord.


Image-Based Deformable Motion Compensation in Cone-Beam CT

Cone-beam CT (CBCT) is increasingly prevalent in minimally invasive interventional radiology (IR) procedures in the abdomen to provide 3D image guidance and quality assurance. One established IR technique that leverages CBCT is trans-arterial chemoembolization (TACE), in which hepatocellular carcinoma is treated by delivering embolic particles with chemotherapeutic drugs to the tumor through the hepatic artery. Identification of the complete vascular structure, including small capillary vessels, is crucial to concentrate the chemotherapeutic effect to a confined target area while sparing adjacent healthy tissue. CBCT 3D guidance capabilities offer better identification of small tumors and feeder vessels than conventional 2D techniques and can enable superselective embolization with substantial effect on the treatment. However, CBCT requires moderately long scanning time (~5-30 sec), making it strongly susceptible to patient motion. During awake interventional procedures such as TACE, patients can have difficulty holding their breath (introducing respiratory motion), difficulty staying still (introducing skeletal motion), and /or show involuntary organ motion from peristalsis and gas movement in the abdomen. Consequently, motion artifacts are present in a large percentage of CBCT images.

My research focuses on developing an image-based motion compensation method that leverages the autofocus concept to compute the 4D spatio-temporal deformation field through maximization of an image sharpness metric in CBCT images. Being purely image-based, this method does not require additional input outside of the raw imaging data (e.g., respiratory or cardiac gating and / or external monitoring of the patient).  Studies have demonstrated robustness of the algorithm to a broad range of motion amplitudes, frequencies, data sources (i.e., simulation, mobile C-arm, and clinical C-arm) and other confounding factors in experimental data (e.g., truncation and scatter), supporting translation of the method to clinical studies in interventional body radiology.


Engineering the Microenvironment for Heart Muscle Cell Mechanobiology

During development and disease, the heart undergoes biophysical and biochemical changes. For example, after a heart attack, damaged myocardium regions change stiffness and extracellular matrix (ECM) composition. Cardiomyocytes (CMs) differentiated from human induced pluripotent stem cells (hiPSC-CMs) hold great potential as a model to expand our knowledge of human heart muscle cells and their interactions with the surrounding microenvironment.Polyacrylamide hydrogels are a common mechanobiology substrate with a tunable physiological relevant stiffness range and allow for functional quantitative measurements such as traction force microscopy. By engineering microphysiological systems with tunable biophysical and biochemical properties, our aim is to understand the CM mechanobiology of cell-ECM and cell-cell interactions.

In order to probe CMs interactions with surrounding ECM and cells, we need to understand the in vitrocell-substrate interface. First, we analyze the cell adhesion receptor expression profile of hiPSC-CMs by integrin antibody fluorescent labeling and flow cytometry. Knowledge of the specific integrin types can be leveraged to rationalize cell binding motifs in a biomaterial design for CMs. Protein attachment to substrate can achieved via a physisorption or chemisorption linker and protein-substrate adhesion can regulate cell behavior. We study CM response to different linking approaches used to attach ECM proteins onto polyacrylamide hydrogel substrates. With a better designed cell-substrate interface specifically for hiPSC-CMs, we can tune microenvironment properties and quantify cell-ECM and cell-cell interactions throughout cardiac development and in disease states.


Clusterless decoding methods for understanding internally generated hippocampal sequences

Millisecond-timescale patterns of neural activity are the substrate for the computations that underlie complex cognitive processes. In the hippocampus, for example, internally generated sequences of hippocampal place cell activity that occurred during a recent experience are often replayed in a time-compressed manner during a phenomenon called sharp wave-ripples (SWRs) that last 100-200 milliseconds. To understand of the causal relationship between these patterns and the learning and memory processes they support, we need decoding algorithms to identify these sequences as they occur and manipulate targeted circuits based on their content in real-time. Previously, a key assumption is that the neural signals have been accurately sorted into single units before the decoding algorithm is applied; however, spike sorting remains a time-consuming, difficult task. A continuing theme in my research has been the development of clusterless decoding methods, which maintain the accuracy of previous methods, but avoid the clustering problem of spike sorting entirely. The first clusterless decoding algorithm for real-time applications I developed was tested and validated with tetrode recordings from rat hippocampus, which makes it possible to identify and classify the representational content of SWRs for content-dependent closed-loop manipulation. Moreover, clusterless decoding methods have proven to be an extremely powerful data analysis tool as well: taking advantage of the high temporal decoding resolution afforded by these methods, my collaborators have discovered a novel internal prospection mechanism in the hippocampal circuits. We envision the need for clusterless decoding methods to broaden and enhance our understanding of a wide range of neural systems.


Differential effects of anti-EGFR treatment on cells in the colorectal cancer microenvironment and its consequence of drug resistance

Drug resistance remains a major problem in the treatment of most cancers. In metastatic colorectal cancer, anti-EGFR therapy in combination with chemotherapy is standard of care for a subset of patients. Genetic alterations which result in non-response have been identified, including mutations in KRAS which is screened for prior to treatment. However, the underlying mechanism of resistance in as many as 10-30% of patients remains unknown.  We identified a mechanism by which cells surrounding cancer, known as cancer-associated fibroblasts, promote resistance to anti-EGFR therapy. The cytokines colorectal cancer-associated fibroblasts secrete are altered after these cells are exposed to drug. An increase in a particular cytokine – EGF – was discovered to be at levels sufficient to cause cancer cell resistance to anti-EGFR therapy. In addition to identifying a novel mechanism of resistance to anti-EGFR treatment in colorectal cancer, this work emphasizes the importance of considering the impact of targeted therapies on cells types found throughout the body.


Machine Learning Models to Understand and Improve Health Care Delivery

Electronic Health Record (EHR) data are increasingly used in large-scale retrospective analyses to better understand healthcare delivery and patient health. For example, machine learning methods using EHR data have the promise to aid in clinical decision-making by surfacing only the information most important to a given patient. On the other hand, they face considerable challenges in generalizing across care settings, healthcare institutions, EHR system changes, and health policy shifts.

My research uses machine learning to address both the promises and challenges of building and using these models. We have used machine learning methods to enable automated summarization of high-dimensional structured health record data into low-dimensional topic representations of clinical notes, like the summaries written by care team members during the course of care. In addition, I aim to understand the limitations to how well EHR data (and correspondingly, machine learning models that make use of it) generalize. In prior work, we’ve shown that transferring a machine learning model from one EHR system to another, even in the same hospital, results in degraded performance. Effective use of machine learning in clinical settings requires us to better understand the care processes and clinical workflows behind EHR data. My current work uses machine learning to investigate this question. As EHR system design and health policy continue to evolve, a better understanding of these processes is necessary to ensure safe generalization of deployed machine learning models.


Pathomechanisms of Peripheral Neuropathy Caused by PMP22 Copy Number Variation

Charcot-Marie-Tooth disease, Type 1A (CMT1A) and Hereditary Neuropathy with Liability to Pressure Palsies (HNPP) are demyelinating peripheral neuropathies that result from copy number variation of the same gene: Peripheral Myelin Protein 22 (PMP22). CMT1A patients have an additional copy of the gene and HNPP patients have a loss of a copy. These disorders cause sensory and motor deficits to the peripheral nervous system that dramatically affect patient quality of life and burden the healthcare system. Although it is clear that precise levels of PMP22gene expression are required for proper myelination, the function of the encoded PMP22 protein and whether this function is dysregulated in myelinating Schwann cells in CMT1A and HNPP remain unclear. PMP22 has been suggested to function as an adhesion protein given that it is homologous to Claudin proteins and localizes to tight junctions in epithelial cells. The presence of leaky myelin lamellae in HNPP model mice and our recent findings demonstrating increased substrate adhesion in CMT1A patient fibroblasts support this hypothesis. Current studies are focused characterizing the mechanism of altered adhesion and utilizing mouse models to determine if it contributes to CMT1A and HNPP pathogenesis.


Dynamical constraints on neural population activity

Brief description: Animals’ behaviour and neural activity are often highly structured in time. Several lines of evidence suggest that activity of populations of neurons can be well described in terms of dynamical systems and that temporal dynamics are a key signature of neural computations. In using a dynamical systems perspective to describes the time dependence of neural activity, we characterise the constraints under which observed neural activity evolves. Dynamical constraints on neural activity can result from local network properties or may be adaptively shaped based on contexts. My research focuses on characterising dynamical constraints on activity in the motor cortex and how these constraints are shaped by movement preparation. We use a brain-computer interface to manipulate neural activity during movement preparation to causally test how motor preparation shapes dynamics.


Dr. Jenna Mueller is a postdoctoral associate in biomedical engineering who works with the Center for Global Women’s Health Technologies at Duke University to develop low-cost devices and therapies to improve the management of cervical cancer in low and middle-income countries (LMICs). She worked with a multidisciplinary team to develop the Pocket colposcope, a low-cost, portable device to screen women for cervical pre-cancer at the primary care setting, and conducted studies to demonstrate its impact in 1000 women in 8 countries. Additionally, she is spearheading a new program to develop low-cost therapeutics for cervical pre-cancer and received a K99 award from the NIH to do large animal trials before moving to patient studies.

​Jenna received a B.S. degree in bioengineering with a minor in global health technologies from Rice University, and completed both an M.S. and Ph.D. in biomedical engineering at Duke University. Her graduate work focused on developing optical systems and automated algorithms to image tumor margins during surgery to improve the accuracy of cancer excision and reduce the likelihood that a patient needs to return for re-excision surgery.


New neural activity patterns emerge with long-term learning

Learning has been associated with changes in the brain at every level of organization.  However, it remains difficult to establish a causal link between specific changes in the brain and new behavioral abilities.  To establish such a link, one would need to know which neurons drive behavior, as well as the relationship between the activity of those neurons and the behavior.  Then, any observed change in behavior could be attributed to an observed change in the neural activity.  A brain-computer interface enables us to link changes in neural activity directly to learning because the relationship between neural activity and behavior is known exactly, and only the neurons we record directly influence behavior.  With a brain-computer interface framework, we induced learning and showed that new neural activity patterns emerge over several days.  We demonstrated that these new neural activity patterns cause the new skilled behavior.  Thus, the formation of new patterns of neural population activity can underlie the learning of new skills.


The Neuronal Correlates and Computations underlying Bilateral Somatosensation

We actively use both of our hands to interact with our environment. When manipulating objects, estimating object size and shape, and comparing two objects, tactile information obtained through both the left and right hand is key to somatosensation. It is well established that tactile information collected by one hand initially reaches the neocortex solely in the opposite cerebral hemisphere. However, somatosensory cortex, the portion of neocortex dedicated to the sense of touch, is also known to be connected across hemispheres. My research investigates the function of such interhemispheric connections in bilateral somatosensation. In particular, my work focuses on 1) identifying the populations of neurons contributing to the integration of tactile signals arising from both sides of the body, 2) on understanding the computational principles governing bilateral tactile signal integration, and 3) on establishing the role of bilateral integration in sensorimotor behaviors involving bilateral touches. My research is conducted in the mouse whisker system, as rodents mainly rely on whisker-mediated bilateral somatosensation to explore and navigate through their environment. Furthermore, the organization of the somatosensory neural circuits in rodents and primates is highly similar. My work combines laminar multi-electrode silicon array recordings in awake head-fixed mice to measure the spiking activity of populations of individual neurons in response to bilateral touches, with optogenetic identification and modulation of interhemispheric neural networks, and simple machine learning analytic approaches to relate neuronal activity to the encoding and perception of bilateral touches. My research aims to provide a detailed comprehension of the neuronal circuits and the computations contributing to the sense of touch involving both sides of the body, and to further the understanding of the function of the neuronal connections linking the two cerebral hemispheres.


Methods and data integration approaches for improved reconstruction of gene networks

We now have the ability to a) sequence every base of the human genome, b) quantify the expression of every human gene, and c) obtain measurements for multiple genomic modalities for an individual patient. Developing effective computational and statistical models to utilize this data for successful bench-to-bedside applications is still a huge challenge. A primary concern for nearly all modalities of genomic data is that it is inherently noisy. This noise introduces biases in the data which is sometimes correlated with outcomes or variables of interest, thus leading to inaccurate conclusions. Additionally, limited statistical power is another major hurdle in computational genomics. Low probability of finding true signal, overestimation of effect sizes and lack of reproducibility are some of the major issues that result from underpowered studies. While we have the ability to measure expression for over 30,000 genes, the number of gene expression samples available is usually limited to a few hundred samples, insufficient to make consistent and statistically robust conclusions.

Amidst noisy and underpowered genomic data – I build robust statistical and machine learning models to identify functional relationships between genes, and their implications in human disease. Groups of genes are known to interact with each other to perform distinct biological processes. Networks are often used to model such interactions among entities in complex systems. Identifying patterns of relationships between genes can improve our understanding of genetic and molecular basis of gene regulation. It can also provide insights into the cascade of molecular events critical for disease manifestation in humans under a variety of conditions. Gene-gene relationships can be modeled as an undirected graph (gene co-expression network) where nodes represent genes, and edges between nodes represent potentially functional biological interactions between genes. My research tackles challenges for network reconstruction that include: a) methods for artifact correction in genomic data, b) build hierarchical transfer learning model to improve accuracy and power for inference of gene co-expression networks by utilizing large-scale publicly available gene expression data, and c) applications of the transfer learning framework to reconstruct context-specific co-expression networks across multiple tissues and diseases.


Synthetic genetic feedback to control single cell heterogeneity in reprograming

Cell fate programming can be achieved through the forced overexpression of key transcription factors (TFs), directing cells to new states. The ability to program cell fate not only provides novel insights into the nature of cell state plasticity, but also has applications in regenerative medicine. Perhaps the most prominent example of cell programming is the process of reprogramming, whereby somatic cells are induced to a pluripotent stem cell state through the overexpression of four key transcription factors. The resultant induce pluripotent stem cells (iPSCs) can give rise to all cell types in the human body in a patient-specific manner. While reprogramming has been extensively studied, the relationship between the expression level of the key TFs and the resulting single cell reprogramming trajectory has remained elusive.

We hypothesize that the process of reprogramming human somatic cells, which is traditionally hindered by low levels of efficiency, can be improved by guiding single reprogramming cells on a trajectory of TF overexpression in which factor stoichiometry is well controlled. To do this, we apply a combined synthetic biology and barcode-based lineage tracing strategy to uncover the reprogramming route of successful human cells. Preliminary results suggest that the pluripotent state is maintained by a “sweet spot” of TF expression that, when controlled during reprogramming, results in improved outcomes.

These results will not only provide a better understanding of human reprogramming, but also serve as a case study for other cell fate programming systems, providing key insights into the role of key TF manipulations on driving cell fate as well as how cell fate trajectories can be predictably controlled using a synthetic biology approach.


Decomposing cell identity for transfer learning across platforms, tissues, and species

The rise of single cell RNA sequencing (scRNAseq) has lead to the discovery of new cell types and challenged traditional morphological based classification and signatures derived from bulk RNA sequencing (RNAseq). Thus, it is becoming increasingly important to have new methods to determine the identifying and consequential features of cells. I have proposed 1) that cell identity should map to a reduced set of shared dimensions, the unique combination of which defines the cell and 2) that these shared dimensions can be used to learn meaningful relationships across diverse data sets including different systems, different technologies, and different species. To this end, my postdoctoral work has focused on developing machine learning and statistical methods to test this hypothesis. Application of these tools to scRNAseq dataset of ~125k cells across mouse retina development demonstrated how feature dimensions can capture shared aspects of biology in retina datasets from different sequencing technologies, retina ATAC data, scRNAseq from other tissues in the mouse, and both mouse and human cortical development. This implementation of transfer learning via dimension reduction represent a platform for in silico experimentation and hypothesis generation where knowledge from multiple data sets can be leveraged to inform the selection of meaningful feature dimensions for biological validation. Thus, these methods represent a key breakthrough for the study of human diseases where information must be gleaned across multiple model systems.


Bridging spatial scales in the statistical analysis of neural recordings

Technological advances are rapidly expanding the breadth of high-resolution neural recordings. The number of simultaneously recorded neurons, for example, has doubled roughly every seven years for the last fifty years, and probes developed in the last five years have far outpaced this rate. At the same time, the invasiveness of these technologies can limit their clinical applicability in humans. In order to translate the insights gained from these massive high-resolution recordings to much lower resolution but less invasive recordings, we need models that capture the propagation of electrical and magnetic fields from the scales of synapses and neurons to the scales of local patches of cortex and scalp recordings. This propagation depends on both static anatomical features and state-dependent dynamical features like active pathways, neuromodulation, and layer-specific patterns of coherent activity. My research involves using prior knowledge about these features in specific brain states to build statistical models that bridge spatial scales. Such models can explain apparent discrepancies in signals recorded at different scales, test theories of meso-scale computational principles, and improve our understanding of the micro-scale dynamics that underlie macro-scale biomarkers of pathological states.


Nontraditional Antibacterials: Combining Antibacterial Entities into One Effective Treatment

The growing prevalence of multidrug-resistant (MDR) bacteria is having major impacts on global human health. In the US alone, about 2 million people are infected by and about 23 thousand people die from MDR infections each year. Scientific research and clinical communities are developing several new treatments to combat this growing occurrence of MDR bacteria. Some of these treatments involve the combination of multiple antibacterial entities into one effective treatment entity. My research aims to validate the hypothesis that conjugating an antimicrobial polymer to an antibacterial nanoparticle will be able to synergistically increase their activity against harmful bacterial cell growth. The polymers of my study, the butyl poly(oxonorborene)s or PONs, are polymers with selective, tunable, broad-spectrum activity against bacteria; including against some bacterial strains for which we are observing increasing cases of MDR. Previous work by the Lienkamp group at the University of Freiburg has shown that conjugating PONs to silicon wafers, gold wafers, polymer matrices, and gold nanoparticles can improve the activity of the polymers against bacterial targets. Concurrently, research on the antibacterial properties of luminescent semiconductor nanoparticles, or quantum dots (QDs), have identified several QDs with light-activated antibacterial activity. Thus, this work investigates and compares the intended antibacterial activity and off-target activity of the free PONs polymer series, free QDs, and PONs-QDs conjugates. Thus far we have observed increased anti-E. coligrowth and decreased off-target activity from the light activated PONs-QDs compared to free PONs and QDs.  The combination of increased intended activity with decreased off-target activity illuminates the potential for these conjugates to be used for the desired antibacterial applications, and with a smaller concern for off-target effects than we would have with free PONs application.


Development of lipid-based drug delivery systems triggered by intrinsic overexpressed enzyme

With the development of biomedical science and chemistry, numerous cytotoxic molecules with the potential to kill highly proliferative cancer cells have been designed and synthesized. However, most of these molecules lack selective antitumor effects, resulting in severe side-effects and limited clinical applications. Well-designed drug delivery systems can enhance drugs’ bioavailability by releasing them specifically to target sites. As the main constituents of biological cell membranes, lipids have been widely employed in drug delivery systems due to their lack of toxicity and high bioavailability. However, lipid-based drug delivery systems have common issues such as low stability upon blood dilution and insufficient drug release at the target sites. The intrinsic overexpressed enzymes in the disease regions have the potential to overcome these issues: lipid-based drug delivery systems can be optimized so that they remain integrated under normal physiological conditions but quickly disassemble to deliver the active compounds at target sites where the enzyme is overexpressed. Secretory phospholipase A2 (sPLA2), a lipolytic enzyme, has been found to be overexpressed in a variety of cancers, and its level is strongly related to cancer stage and tumor metastasis. As an interfacial enzyme, catalytic activity of sPLA2 relies on the physicochemical properties and structural organization of lipid substrates. The variety of structures and compositions of lipid aggregates makes them versatile for application in drug delivery systems; however, this variety also renders the design of sPLA2-triggered lipid-based drug delivery systems challenging. Based on a fundamental understanding of the molecular interactions of lipids with sPLA2, my study focuses on establishing mechanism-based methods for designing and optimizing the composition and structures of sPLA2-triggered lipid-based drug delivery systems.