Infectious fungal diseases (IFDs) pose a significant global public health threat, causing 1.5 million annual deaths and affecting over 1 billion individuals, with 5.7 billion people at risk. Climate change-induced factors, antifungal resistance, and neglect by public health officials have exacerbated IFD challenges. To address this escalating concern and mitigate unnecessary mortality, rapid, accurate, and cost-effective diagnostics are required. This study utilizes universal polymerase chain reaction and digital high-resolution melt (U-dHRM) coupled with machine learning, for classification of clinically significant fungal pathogens and novelty detection of previously unknown fungal pathogen species. The study examines the impact of signal preprocessing filters and ramping rates on machine learning accuracy for classification and area under the curve of the receiver operating characteristic curve (ROC AUC) for novelty detection. The pipeline failed to sufficiently classify similar Aspergillus species curves, suggesting future exploration of alternative gene regions. Additionally, a micromanipulator device is developed to extract DNA from specific U-dHRM chip wells for the identification of novel fungal pathogens in clinical samples. Notably, the study identifies the multi-drug resistant Trichosporon asahii in a patient previously diagnosed and treated for Aspergillosis and passed away. Despite failure to significantly improve the effectiveness of the machine learning pipeline, this study still sheds light on future research directions for the automation of fungal pathogen classification and novelty detection, and enables the detection of rare/novel pathogens of clinical importance from patient samples.

Outbreaks of infectious diseases are rising around the world. In addition to outbreaks, several known infectious diseases have a significant impact on mortality in the US and around the world. In this doctoral work we use a combined experimental/computational approach to study the performance and limitations of digital High Resolution Melt (dHRM), an infectious disease screening platform that was previously established in the Fraley lab. This allows comparisonswith other microbiological technologies and helps shine some light on its potential use-cases. First, we developed a computational framework for estimating the resolving power of dHRM technology for defined sequence profiling tasks. By deriving noise models from experimentally generated dHRM datasets and applying these to in silico predicted melt curves, we enable the production of synthetic dHRM datasets that faithfully recapitulate real-world variations arising from sample and machine variables. Second, we present an advancement in universal microbial high resolution melting (HRM) analysis that is capable of accomplishing both known genotype identification and novel genotype detection. Specifically, this novel surveillance functionality is achieved through probabilistic modeling of sequence-defined HRM curves, which is uniquely enabled by the large-scale melt curve datasets generated using our high-throughput digital HRM platform. Our hope is that in the future, the dHRM platform can translate into a near-point of care, cost-effective tool for infectious disease screening.

Bacteria identification plays an important role in confirming diagnosis of bacterial disease and tracing the source of an outbreak for public health. However, the standard method for bacteria identification which relies on the colony feature of the bacteria cultured from the patient’s specimen is time consuming and lack of accuracy and sensitivity. New technology such as qPCR based sequencing and microarray showed their benefit in this area but are still suffering problems regarding to cost, complexity and accuracy. Therefore, this thesis tries to advance the high resolution melt (HRM) technology, a molecular technique to detect difference in double-stranded DNA, to realize bacteria serotyping. Two methods: intercalating dyes and multiplex PCR were investigated for this purpose. To our excitement, multiplex PCR of V1 to V6 region and V6 region of the gene 16S ribosome RNA of Salmonella followed by HRM can distinguish 5 serotypes of Salmonella under 3 hours, including Newport, Typhimurium, Heidelberg, Enteritidis and Choleraesuis. We further applied this technology to achieve antibiotic resistance detection and bacteria speciation in one test and successfully detected the kanamycin resistance from E. coli. This multiplex PCR combined with HRM will be further combined to digital PCR to build a fast, easy to use, highly specific, highly sensitive and absolute quantitative tool for bacteria diagnosis.

Human cytomegalovirus and Herpes simplex viruses 1 and 2 are double stranded DNA viruses that establish lifetime latency in a host upon initial infection(1–3). Primary infection or reactivation of the latent virus in a pregnant woman can transmit the virus in utero or during birth to the fetus causing high neonatal morbidity(4,5). Majority of these infections are asymptomatic at birth, but may present later with potential lethal infection and disseminated disease (HSV) or long term neurodevelopmental sequelae including sensorineural hearing loss (HCMV)(1,4,5). The infection can, however, be symptomatic where it may present as non-specific neurologic conditions for HCMV or sepsis-like condition for HSV or severe HCMV cases(4,6), which may not result in the physician ordering immediate testing for these viruses(4).

An early detection of HCMV/HSV is thus necessary to decrease morbidity for asymptomatic infections. Their specific detection against other disease-causing agents is also required to inform correct patient treatment for non-specific symptoms or polymicrobial infections. An ideal test should be rapid to aid in clinical decision-making and broad based to include these viruses with other common neonatal infectious agents with similar symptoms and for which tests are commonly ordered. This work discusses the development of a rapid quantitative PCR based viral assay for HSV and HCMV that can be multiplexed with sepsis or other infectious disease panels with its probe-free chemistry, in order to allow early and specific detection of these viruses in neonates.

Cancer is a leading cause of death worldwide. During metastasis, cells from a primary tumor escape, invade local tissues and spread to distant organs. Even though this process represents the highest risk for cancer related death, a lot remains unknown about the principles behind its beginning and progression. The main barrier that cancer cells face at each step of the metastatic process is a complex 3D extracellular matrix (ECM), which they need to bind to, pull on and degrade. In this dissertation I use 3D culture systems to mimic a native 3D ECM in vitro, to quantitatively study the role of the ECM structure and physical properties on the development of invasive phenotypes and gene expression programs.

In chapter 1, I identified a switch to a highly persistent and polarized cell motility in a high density environment and discovered that cells undergoing this migration formed multicellular structures. Using RNA sequencing I found an upregulated 70 gene module that together with the motility response has been describes as vasculogenic mimicry (VM). I show evidence that the identified transcriptional program predicts survival in patient data across nine distinct tumor types, suggesting it may represent a conserved metastatic response.

In chapter 2, I investigate the biophysical mechanisms of cell-ECM interactions. I find that matrices with low degradability limit cell adhesion. Cells in these matrices display hallmarks of anchorage independent growth, including oxidative stress. They show decreased expression of several mTOR target genes and downregulation of TCA cycle and pyruvate metabolism pathways. Lastly, I show that as part of the response to low adhesion state, cells upregulate key NOTCH signaling molecules.

In chapter 3, I propose the use of in vivo models to deconstruct the role of collagen density in promoting cancer cell metastasis. I present preliminary data and future directions for these in vivo models.

The data presented in this dissertation provides a multiscale view of the role of the collagen microenvironment in the development of cancer cell invasive motility and metastatic phenotypes. The discoveries presented hereby highlight the value of using 3D culture assays in combination with quantitative analysis to uncover biological phenomenon with high translational potential.

A major component of cancer’s complexity lies in its heterogeneity. Because cancer heterogeneity can manifest across multiple spatiotemporal and biological scales1–4, comprehensive characterization is challenging. Assays that allow controlled experimentation to determine the mechanisms behind cancer heterogeneity remain underdeveloped. Many existing technologies rely on exploitation of predetermined characteristics5–7, which precludes exploration of phenotypes that are ill-defined. Other platforms begin with blind genomics, using post-sequencing experiments for validation8,9. These approaches inherently require significant differences in the genomics underlying heterogeneity to elucidate potential mechanisms, since unsupervised clustering relies on deploying mathematics to obtain clear separations10–12. While successful when involving multiple cell types, since their biological profiles lie in separable state spaces6,8,13,14, unsupervised analysis has limitations when evaluating more similar cells, such as within-cell-type heterogeneity, as it becomes difficult to separate profiles that are highly alike. A study on this specialized scope has not yet been done, probably due to the challenges of parsing a more subtly heterogeneous sample.This dissertation describes the development and application of a platform that enables detailed interrogation of within-cell-type heterogeneous, 3D collective cancer cell migration phenotypes. Collective migration is a process where multiple cells coordinate their movement15,16. Recent studies point to the importance of collective cell migration in cancer metastasis17–20. Although various phenotypes have been identified, factors that regulate collective behavior are not fully understood15,16,21. Using a flexible, microscopy-based platform, I identified and isolated cells that exhibit invasive and non-invasive collective migration. By first defining functional subpopulations, I deployed a supervised approach to determine the mechanisms behind this aspect of heterogeneity. I found that the collectively invasive phenotype is associated with upregulated proliferation, increased stress responses, and the ductal carcinoma subtype, while the collectively non-invasive phenotype was associated with immune-related processes and the luminal carcinoma subtype. Functional perturbation of differentially expressed genes resulted in shifts in migration phenotypes. These results validate the platform I developed for identifying mechanisms of within-cell-type heterogeneity. Furthermore, the results demonstrate a link between migration regulation, stress response, proliferation, and immune response and indicate potential value in exploring how collective invasion may be controlled through these associated modules.

Cellular plasticity is essential for healthy function during development and healing. However, it is the dysregulated plasticity of neoplastic cells that gives cancer its advantageous heterogeneity and resistance to therapy. Plasticity also endows cancer cells with the ability to mimic and aberrantly activate fundamental developmental processes such as collective cell migration (CCM) in the form of collective invasion, which initiates the metastatic cascade that is the primary cause for cancer-related deaths. Remarkably, the initiation of this process is strongly associated with changes in the extracellular matrix (ECM) environment of the tumor, which functions as both a physical barrier and dynamic signaling cue for cell migration. Namely, simultaneous increases in the density, stiffness, and alignment of Collagen I fibers in the tumor ECM mark a key transition. However, challenges associated with modeling this process limit our capacity for identifying new approaches that target these plastic cancer cells. In this dissertation, we develop ECM engineering and dynamic measurement techniques to enable study of the biophysical cell-cell and cell-matrix interactions that drive collective cancer cell invasion. We then use these newfound insights to develop a phenotypic screening model that successfully identifies a cytodifferentiation therapy of plastic Triple Negative Breast Cancer (TNBC) cells.

In chapter I, we advance a matrix engineering approach to tune Collagen I fibrillar architecture independently of known modulators of collective cell behavior such as ECM density and stiffness. We find that tighter fibril networks are less degradable and adhesive, inducing cellular confinement that prompts a switch from single cell migration to CCM. Chapter II integrates multiple modalities of biophysical imaging with in silico modeling to dissect the underlying cell-cell and cell-matrix interactions of the two distinct CCM behaviors that emerge from single populations of cancer cells in confining matrices: collective rotational migration and collective invasive migration. We find that the predominant phenotype of invasive CCM relies on a combination of proteolytic-based matrix remodeling and physical remodeling while the less common rotational CCM relies solely on mechanical-based matrix remodeling. Contrary to the current paradigm, we also find that epithelial characteristics such as cadherin-based cell-cell junctions or establishment of a Laminin-rich basement membrane are non-essential for either CCM behavior. In chapter III, we conduct a phenotypic screen of TNBC cells within confining Collagen I matrices to identify cytodifferentiation therapies targeting CCM, a trait that is associated with enrichment for Cancer Stem Cells (CSCs). Phenotypic screening identified two structurally similar leads that abrogate collective invasion, induce CSC cytodifferentiation and Mesenchymal-to-Epithelial Transition, and increase sensitivity to traditional chemotherapies. Importantly, this cytodifferentiation response is not observed in 2D-treated cells or in non-confining 3D matrices, highlighting our simple 3D phenotypic screening assay as an improved method for identifying relevant cytodifferentiation therapies for TNBC. Taken together, we provide a bottom-up description of how understanding and designing cell-matrix interactions can give rise to physiologically relevant CCM behaviors that provide a novel paradigm for identifying therapeutics that target cancer plasticity.

Viral, fungal, and bacterial infections often share similar or overlapping symptoms that complicate diagnosis and impede appropriate and timely treatment. Infections can spread to the bloodstream and cause sepsis, a severe and quickly progressing condition that necessitates rapid identification and treatment of the underlying pathogen. However, most emerging pathogen detection technologies for sepsis are follow up tests after a culture yields a positive result or have narrow adoption due to limited coverage of pathogen targets, low sensitivity, and high costs. This dissertation advances universal digital PCR and high resolution melting (U-dHRM) by developing novel methods to integrate multiple cross-kingdom universal primers to accurately identify known and novel fungal, viral, and bacterial pathogens with high sensitivity and low cost directly from low volume (1mL) samples in a single test.In Chapter 1, to expand U-dHRM’s capabilities by establishing a universal fungal assay, we targeted bronchoalveolar lavage fluid (BAL) as an ideal sample type suited for clinical validation which commonly encounters additional clinically relevant genera in patients suspected of invasive mold infection (IMI). By demonstrating U-dHRM’s speed, ability to simultaneously identify and quantify clinically relevant mold pathogens in polymicrobial samples, and detect emerging opportunistic pathogens to possibly aid treatment decisions and improve patient outcomes we show promise as a separate or combination diagnostic approach to standard mycological tests. In Chapter 2, we expand the bacterial database and address additional strategies for U-dHRM to discriminate species with multiple strains that present unique clinically relevant diseases. A 2-step rapid assay for syphilis detection and typing by eubacterial species distinction and hypervariable gene targeting is described using qPCR and HRM. In Chapter 3, we advance U-dHRM with a cross-kingdom multiplex assay for neonatal infections, the most critical patient population U-dHRM is ideally suited for with the potential to overcome low starting blood and low expected pathogen loads. We detect and differentiate a panel of the most common neonatal pathogens in a single fast assay that has the potential to advance the field of diagnostics to yield more timely and accurate treatment.

Sepsis is a life-threatening condition that results from a severe immune response to a bloodstream infection. Identifying sepsis-causing organisms rapidly and accurately within a clinically relevant time-frame remains a significant challenge. To properly identify sepsis-causing pathogens, the ideal diagnostic should be: (a) rapid, (b) broad-based, (c) capable of polymicrobial detection, (d) highly sensitive and specific, (e) minimally invasive, (f) easily integrated into clinical workflow, (g) able to detect antibiotic resistance determinants, (h) and able to identify new and unknown pathogens. The current gold standard for pathogen detection is blood culture which has limited sensitivity and detection capabilities and requires a significant amount of time. Other technologies have been developed to address some of the ideal sepsis criteria; although many meet several of the criteria for the ideal sepsis diagnostic, none have successfully fulfilled them all. We have developed a novel device that meets each of these criteria; it can identify sepsis-causing bacteria at a clinically relevant load for neonatal sepsis within four hours through an integrated system which incorporates universal PCR amplification, High Resolution Melt across a 20,000 well microfluidic chip and a machine learning algorithm. The bacterial DNA is separated by digitization across the picoliter-sized wells and the 16S gene is targeted and amplified using universal primers. This device fingerprints each well simultaneously and compares them to a library of characterized curves by utilizing a machine learning algorithm. This technology could be a valuable clinical addition as an ideal sepsis diagnostic.