Cells' ability to efficiently replicate their genomes is essential for regulating chromosomal division and maintaining chromosome integrity.  Defects in any of these cellular processes may cause genomic instability, potentially leading to cancer.  The Chaos3 allele in the yeast Saccharomyces cerevisiae is a single base pair change causing an amino acid substitution in the Mcm4 protein.  Mcm4, a component of the replicative helicase, is recruited to replication origins to unwind double stranded DNA and initiate replication.  Chaos3 is in a region of MCM4 that is highly conserved across eukaryotes; while mutations in conserved regions are generally non-viable, Chaos3 is a viable allele that causes genomic instability, leading to elevated cancer rates in mice.  In S. cerevisiae, Chaos3 decreases early firing of the autonomously replicating sequences (ARS) where DNA replication begins.  Chaos3 does not affect all early firing ARSs in the genome; rather, a large proportion of origins near centromeres, thereby delaying replication of those centromeres.  Essential for chromosome segregation, the centromere is the location where spindle fibers attach to pull apart sister chromatids during cell division.  I hypothesize that this delay in centromere replication results in chromosomal instability, including the loss of a chromosome.  I am using CRISPR guided cutting directed by a customizable guide RNA to replace centromeric adjacent ARS510, that has decreased firing levels in Chaos3, with unaffected, early firing ARS305, to see if firing levels in the mutants are affected based on ARS chromosome location (i.e., proximity to a centromere) or ARS sequence.  If replacing ARS510 with ARS305 restores early origin firing in this region this will confirm the Chaos3 mutation affects specific ARS sequences rather than ARS location on the chromosome.  Furthermore, if centromere replication delays are the cause of genomic instability in Chaos3, this ARS replacement should rescue chromosome loss.

Meningiomas, the most common type of primary human brain tumor, arise from the thin fibrous membrane that covers the brain and spinal cord. Most grow slowly and are diagnosed when they disrupt brain function or lead to persistent headaches. While many meningiomas can be cured by surgery, ~20% of them cannot be fully resected or display increased growth, invasion and destruction of adjacent brain and skull. Effective control or eradication of these ‘High Grade II/III’ meningiomas is clinically challenging. To identify new agents and treatment measures, our project uses both computational and experimental approaches in concert to identify new and potentially better therapies. As part of this effort, we are using PISCES, a machine learning model, together with augmented drug and radiation combination datasets to predict potential new therapy synergies. The best predictions from PISCES will then be tested experimentally in our cell line model versus standard-of-care treatments.  My presentation summarizes work to characterize genomic, drug and ionizing radiation sensitivity data on IOMM-Lee, a Grade III human meningioma cell line disease model. We detail how A.I.-driven analyses of IOMM-Lee and related meningioma datasets led us to test new drug pairs and drug-radiation combinations predicted by PISCES to be more effective in killing IOMM-Lee tumor cells. This translational cellular disease model and project are part of a long-term effort to develop better ways to rapidly and efficiently identify and validate new treatment options for brain tumors and other human cancers that can be taken directly to clinical trial.

One of the core problems in the analysis of protein tandem mass spectrometry data is the peptide assignment problem: determining, for each observed spectrum, the peptide sequence that was responsible for generating the spectrum. Two primary classes of methods are used to solve this problem:database search and de novo peptide sequencing. State-of-the-art methods for de novo sequencing employ machine learning methods, whereas most database search engines use hand-designed score functions to evaluate the quality of a match between an observed spectrum and a candidate peptide from the database. We hypothesize that machine learning models for de novo sequencing implicitly learn a score function that captures the relationship between peptides and spectra, and thus may be re-purposed as a score function for database search. Because this score function is trained from massive amounts of mass spectrometry data, it could potentially outperform existing, hand-designed database search tools. To test this hypothesis, we re-engineered Casanovo, which has been shown to provide state-of-the-art de novo sequencing capabilities, to assign scores to given peptide-spectrum pairs. We then evaluated the statistical power of this Casanovo score function, Casanovo-DB, to detect peptides on a benchmark of three mass spectrometry runs from three different species. Our results show that, at a 1% peptide-level false discovery rate threshold, Casanovo-DB outperforms existing hand-designed score functions by 35% to 88%. In addition, we show that re-scoring with the Percolator post-processor benefits Casanovo-DB more than other score functions, further increasing the number of detected peptides.

Eukaryotic DNA is wrapped around nucleosomes to be packaged into a cell’s nucleus. Nucleosomes are made up of a combination of canonical histones (H2A, H2B, H3, and H4) and histone variants. Histone variants are evolutionarily derived from their canonical counterparts and can replace canonical histones to perform specialized chromatin functions. Given their crucial and widespread functions, mutations of histone genes are correlated with poor prognosis in cancers. Histones and their variants are typically evolutionarily conserved in sequence and function. Therefore, lineage-specific differences in histone repertoires present unique opportunities to understand their functional consequences on genomic organization and biological processes. Here, we study two such changes in the H2A repertoires of budding yeast and fruit flies. Most eukaryotes including humans have an H2A repertoire of canonical H2A and two variants– H2A.X important for DNA damage response (DDR) and H2A.Z essential for gene regulation. However, in yeast, H2A.X entirely replaced H2A, and in flies, H2A.X and H2A.Z are fused into a single variant H2Av. We take two approaches to discover the adaptive advantages of these changes. First, we are studying the evolutionary origins and diversification of H2A repertoires in yeast. We find that many basally branching fungi have a canonical H2A, suggesting that only some yeast have lost canonical H2A. Second, we are recreating the fly and human H2A repertoire in S. cerevisiae. We find that while yeast with a fly-like H2A repertoire have a DDR, it is dramatically reduced compared to wild-type yeast. This raises the intriguing possibility that a fly-like repertoire might lead to a trade-off of compromised DDR in fly genomes. We are now analyzing how DDR-dependent processes like meiosis are altered in fly-like yeast. By leveraging the power of evolution and yeast genetics our work will reveal the biological consequences of unexpected histone innovation.

Alphaviruses are arthropod-borne viruses that are responsible for febrile illness, chronic arthralgias, and premature deaths worldwide. Yet, there are no existing vaccines or therapeutics for the treatment of alphaviral diseases. Despite the limited size and coding capacity of alphavirus RNA genomes, most alphaviruses can adapt to multiple, evolutionarily divergent vertebrate and insect host species. As such, alphavirus RNA genomes and proteins carry host-specific adaptive features to compensate for the differences between hosts, such as body temperature (28°C in insects vs. 37°C in vertebrates) and methods of host immune response. Previous experiments have shown that continuous passaging of the dual-host alphavirus Sindbis virus (SINV) in Adedes albopictus (C6/36) cells results in a gain of fitness in said cells and a loss of fitness in human embryonic kidney (HEK293T) cells, resulting in a mosquito-adapted SINV (SINV^M). Using a modified long-read sequencing method (MrHAMER), we identified an assortment of fixed mutations that were serially acquired over the course of mosquito cell adaptation. Interestingly, multiple synonymous mutations were mapped to the 5' end of the SINV RNA genome, which is known to adopt functionally important RNA structures necessary for virus replication and genome packaging. Additionally, non-synonymous mutations were acquired within the viral structural genes. In this study, I aimed to understand the functional consequences of these two classes of SINV^M mutations with regards to viral fitness in insect and vertebrate cells. Furthermore, I assessed if the phenotype of these mutants were influenced by host body temperature by incubating the infected HEK293T and Vero cells at both 28°C and 37°C. We envision that studying these mutations will allow us to better understand the selective pressures influencing alphavirus evolution and potentially identify host-specific viral determinants of infection. Ultimately, this knowledge will allow us to identify ways to intervene at different stages of the alphaviral transmission cycle.

Understanding how DNA folds in three dimensions is crucial for deciphering cellular function. Chromosomal contacts are interactions between different DNA regions. These contacts hold key information about tissue-specific characteristics, such as gene expression and regulation. However, current predictive models for genome folding primarily focus on within-chromosome interactions, largely ignoring variations across tissues and the role of interactions between chromosomes (trans-contacts). To address these issues, we developed TwinC, a machine learning model that predicts trans-contact maps from pairs of nucleotide sequences. To build TwinC, we used a convolutional decoder coupled with an encoder architecture that can be configured to employ transformers, convolutional networks, or a hybrid approach. Preliminary results suggest that the convolutional architecture achieves performance comparable to Orca, the current state-of-the-art in sequence-to-contact predictions. TwinC is trained and evaluated on contacts measured in two human tissues and one mouse tissue. We are experimenting further with other encoder architectures, fine-tuning the model, and investigating how it generates its predictions. This research will provide valuable insights into the underlying biological mechanisms responsible for chromosomal contacts and lead to an improved, high-performance model for predicting trans-contacts.

Tumors generally spread and remain in one individual; however, transmissible cancer differs. Detected among Tasmanian devil and dog populations, transmissible cancer is defined as whole cancer cells spreading among individuals in a population. Recently, Bivalve Transmissible Neoplasia, a transmissible cancer found among bivalve species, was discovered among soft-shell clams (Mya arenaria) on the East Coast of the USA and Canada; however, it has not been reported in any populations of soft-shell clams on the West Coast. Preliminary data showed that BTN was present in at least one West Coast soft-shell clam population. We focused on testing for BTN among soft-shell clams sampled from multiple sites in Washington State and determining which sub-lineage of BTN can be found among the West Coast soft-shell clams. I extracted DNA from the hemolymph of 37 soft-shell clams from Similk Bay to test for the presence of BTN using quantitative PCR (qPCR). The results indicated that soft-shell clams from Similk Bay tested negative for BTN despite an exposed population nearby in South Skagit Bay. I also analyzed DNA from soft-shell clam BTN from South Skagit Bay and Triangle Cove, Washington, to identify specific genetic markers to diagnose the specific sub-lineage the Soft-shell clams derived from - East Coast, USA or Prince Edward Island, Canada. An SNV in the COI gene specific to the Canadian sub-lineage of BTN was not found, and insertion sites of the Steamer retrotransposon specific to the USA sub-lineage of BTN were present among the West Coast soft-shell clams BTN samples, showing that the BTN on the West Coast arose from the USA sub-lineage. Given these results, analysis of more sites will provide information on how far the USA BTN sub-lineage has spread among West Coast areas and possibly how neoplastic cells can transmit among softshell clams across a body of water.

Evolution is a challenging topic that high school students often struggle to grasp. The yEvo project, a collaboration between genetics labs and high school biology classrooms, seeks to address this issue by giving students an active hands-on experience in evolution using yeast as a model. Over the course of a school year, yEvo students expose yeast to a selective pressure (such as antifungal drugs), and the natural mutation rate creates beneficial mutations that increase in frequency due to selection. Whole-genome sequencing and computational analysis of the ancestor versus evolved populations reveals changes in the genome that were selected over the course of the experiment. However, the resulting raw mutation data is difficult for students to interpret, thus motivating a need for an intuitive, interactive method to visualize sequencing results. Built using R shiny, a first draft of the yEvo mutation browser consisted of a chromosome diagram with each classroom’s mutation data mapped, a pie chart of mutation types, and a gene viewer depicting the altered sites in each mutated gene. Our work focuses on upgrading the user interface to make it more intuitive, optimizing the backend data processing for more streamlined data filtering, and adding features that allow users to upload and interact with their own datasets. This tool will enable students to more easily interpret the results of their evolution projects. Students will be able to view their classroom data, compare it to other datasets with the same condition, visualize mutation clusters in the genome, and see how each mutation changed the genes. In the future, the yEvo browser can be outsourced and used as a framework for data visualization for other model organisms that can serve to benefit the genetics community as well as educators.

While recent advances in sequencing technology have increased the detection of missense variants in human genes, the functional impact of ~99% of these variants is unknown. Improving our understanding of variant effects will make precision medicine more effective, allowing us to define, test, diagnose, and treat genetic diseases better. One way to understand what variants do is to measure and read out each variant’s effect in cell-based assays. However, there are nearly 9 billion possible single nucleotide variants in the human genome. To measure variant effects in a comprehensive manner, scalable experiments are necessary. Previously, the Fowler lab developed landing pad (LP) vectors to conduct scalable cell-based assays over entire gene variant libraries. However, the current LP design suffers from rapid silencing in cell culture, a phenomenon in which cells deactivate the expression of transgenes, greatly limiting the scope of variant assays. Stem cells are known to rapidly silence most exogenous sequences during differentiation, but they are a target model as LPs embedded in stem cells would allow us to study the effect of gene variation in cells specific to the related disease. Through previous work at the Fowler lab and others, we have characterized a set of promoters, enhancers, insulator sequences and other elements that will provide stable and robust expression of transgenes over time. We hypothesize that combining the LP with this set of elements will allow us to stably express variant libraries regardless of cell context. To test this hypothesis, we have designed a nested LP delivery system that allows us to compare different LPs side-by-side in the same genomic context. By comparing expression levels of transgenes over time, we expect to find that LPs enhanced with a set of stabilizing elements will express stronger signals over longer periods of time compared to LPs that are not enhanced.

De novo sequencing of peptides is a key technique in proteomics mass spectrometry for detecting and quantifying proteins from biological samples without the use of a reference protein database. This is necessary when working with organisms with low proteome coverage in extant databases, such as in metaproteomics, and in circumstances where the space of possible peptide sequences is extremely large, as in immunopeptidomics. Currently, all state-of-the-art de novo sequencing methods employ deep learning models. However, the majority of the available data used to train these models comes from mass spectrometry experiments that use the enzyme trypsin to digest proteins into peptides. Consequently, these methods exhibit a strong learned bias toward peptides generated by tryptic digestion, and degraded performance when applied to spectra generated by alternative digestion enzymes. This bias limits the models’ ability to generalize to other settings, where the use of alternative digestion enzymes can preserve sequence features that are lost from tryptic digestion. We approach this problem by modifying the state-of-the-art deep learning model, Casanovo, to incorporate an explicit representation of the digestion enzyme as part of its input. We then train Casanovo on a set of spectra digested using a wide variety of enzymes, providing the enzymatic context to the model to facilitate learning the diverse spectral and peptide patterns inherent in non-tryptic digests. We observe that our enzyme-aware model learns enzyme-specific digestion rules and shows significant improvements over the control model on enzymatically diverse data. The enhanced generalizability of our model will enable proteomics researchers to improve the robustness of their de novo sequencing workflows in settings that employ non-tryptic digestion.

In order to design effective countermeasures against HIV, we must first understand the forces that drive it to evolve resistance within hosts. While linkage patterns in genetic data are potentially a powerful tool to quantify the relative contributions of multiple evolutionary forces (mutation, recombination, selection) acting during an HIV infection, the severe viral population bottlenecks accompanying drug therapy complicate these patterns. To interpret genetic linkage in the context of such major changes in population structure (which are themselves driven by specific mutations), we develop a simulation framework for viral evolution in which genetics and population structure influence each other. This framework overcomes limitations from both dynamical modeling, in which patterns of linked variation are ignored, and from population genetic modeling, in which population structure is predetermined. Using few parameters, we are able to reproduce linkage patterns and population bottlenecks that broadly conform to those observed in vivo. As a case study to demonstrate this model’s utility, we consider a recent hypothesis that viral recombination is suppressed during population bottlenecks due to diminished opportunities for coinfection. In simulating populations with and without recombination suppression during population contraction, we show that this effect measurably changes genetic diversity in rebounding populations, but is less visible when examining simulated viral loads or resistance mutations alone. Then, using this model as a null expectation for linkage patterns, we assess if the linkage structure in HIV populations treated with bNAbs is consistent with density-dependent recombination in vivo. Collectively, our work demonstrates that, by generating realistic null expectations of linkage under complex changes in population structure, we can employ linkage patterns as a powerful source of information for evaluating viral evolutionary hypotheses.

Budding yeast cultures grown in limited sulfate conditions are overtaken by cells with an inverted triplication of the gene SUL1, which encodes for a sulfate transporter. The extra copies of the sulfate transporter provide a selective advantage because these cells outcompete other yeast cells for the limiting resource. To explain the mechanism behind this type of amplification the Brewer and Dunham Labs proposed a model (Origin Dependent Inverted Repeat Amplification or ODIRA), which requires both a DNA replication origin and inverted repeats flanking SUL1. ODIRA starts with a DNA replication error involving replication fork regression that leads to an extrachromosomal DNA intermediate. This intermediate then replicates and recombines into the genome, producing the observed amplification. Because similar triplications are observed in the human genome, including in human disorders, the mechanism of ODIRA offers insights into human genome evolution and disease. While the yeast research is consistent with ODIRA, we still do not know which proteins are responsible for the process. I am testing whether the genes RAD5 and RAD54 — involved in fork regression and strand switching, respectively — are involved in ODIRA. To do so, I am measuring the ODIRA frequency in strains with each gene deleted compared to a wild-type control. If either gene deletion leads to a statistically significant change in ODIRA frequency compared to the wild-type strain, I can conclude this gene is involved in ODIRA. To measure ODIRA frequency, I grow the deletion strains under selection for DNA recombination events and use whole chromosome gel electrophoresis and Southern blotting to detect ODIRA events. Preliminary data analysis suggests that there is a reduction in ODIRA events when either RAD5 or RAD54 is deleted, indicating that these genes are likely needed for ODIRA. These results may provide insight into how inverted triplications may arise.

To perform its function as a tumor suppressor, breast cancer 1 (BRCA1) must dimerize with BRCA1-associated RING domain protein 1 (BARD1). Due to this critical interaction, pathogenic BARD1 variants are also associated with increased breast and ovarian cancer risk. Genetic testing has identified many rare single-nucleotide variants (SNVs) that cause missense amino acid substitutions in BARD1. Currently, 93% (1,692 of 1,819) of BARD1 missense SNVs are classified as a variant of uncertain significance (VUS) in ClinVar. A VUS classification prevents clinicians from using genetic test results to guide patient care. Consequently, there is a strong need to functionally assess BARD1 SNVs to help resolve VUS. We applied a multiplex assay for variant effect called saturation genome editing (SGE) to functionally assess all possible 12,000 SNVs and 2,300 3-base deletions in BARD1. In SGE, we use CRISPR-Cas9 to edit all possible SNVs into a region of BARD1 in haploid HAP1 cells. BARD1 is essential for cell growth, therefore cells edited with loss-of-function variants become depleted from the population. We track which SNVs become depleted from the population by sequencing. We then generate functional scores for each variant by calculating the change in the abundance of a variant in the original SNV library versus its abundance in the cell population after 13 days in culture. Thus far, I have generated reagents for all 14,300 variants and 2,400 have completed the full experimental pipeline. Functional scores for the functionally critical BRCA1 interaction domain show depletion of 94% stop-gain, 48% splice-site, and 21% missense variants relative to 5% synonymous and 6% intronic variants. This ultimately demonstrates SGE’s ability to accurately identify functionally normal and loss-of-function BARD1 variants. Generating functional scores for all possible BARD1 variants will provide the functional evidence needed for reclassifying BARD1 VUS and definitive test results for providers treating patients with BARD1 variants.

The human genome is associated with numerous variations, some of which can be pathogenic. Any variations larger than 50 base pairs (bp) are considered structural variants (SVs), and SVs impacting copy number of the genome are termed copy number variants (CNVs). By studying CNVs, we can understand the underlying mechanisms of damage and repair within DNA, which can help work towards prevention and cure in other fields, such as cancer genomics. My project investigates a CNV on chromosome 2 of a pediatric patient, who presents with tetralogy of Fallot, global development delay, and multiple other congenital anomalies. Using array comparative genomics hybridization (aCGH), we identified a 3.2 megabase (Mb) complex genomic rearrangement (CGR) spanning the 2q31 region of the proband’s chromosome 2. Detailed analysis of the short-read whole genome sequencing (WGS) allowed us to locate the exact coordinates of each junction, or the beginning and end points of each extra copy. PacBio long-read genome sequencing and optical genome mapping detected the same junctions and facilitated confirmation of the overall structure. The CGR can be characterized as a duplication-triplication-duplication-triplication-duplication (DUP-TRP-DUP-TRP-DUP), meaning this segment of the genome contains a segment of alternating 1 and 2 extra copies of this region. The CGR is de novo, or not inherited from the proband’s parents. Due to the nature of this variant, it is likely to be impacting the phenotype of our patient. To establish a genotype-phenotype correlation, I did a literature review of patients with overlapping CGRs, comparing their phenotypes to our proband, as well as reviewing any known disease-associated genes in the region using the online catalog of human genes and genetic disorders (OMIM). Patient and disease comparison revealed the extreme rarity of our patient’s CGR, leading us to believe the phenotype of the proband results from impact of multiple genes in the affected region.

In the 20th century, the discovery and widespread use of antibiotics became humanity's primary weapon against pathogenic bacteria. Overuse of antibiotics has unfortunately given rise to antimicrobial resistance, weakening us in this evolutionary arms race. Proteolysis-targeting chimeras (PROTACs) have been proposed as a strategy for development of novel therapeutics. By tagging target proteins with ubiquitin, targeted protein degradation (TPD) can occur via a eukaryote's own molecular machinery. Due to prokaryotes lack of ubiquitin, research has shifted to the development of PROTAC-like molecules to achieve proteolysis and cell death in bacteria, called BacPROTACs. However, to eventually experiment with these small molecules and see their mechanism of TPD, off-target effects, and changes in host and bacterial proteomes, we must be able to profile the degradation of proteins in an unbiased manner. Proteomics and mass spectrometry can identify and measure thousands of proteins simultaneously, enabling systems-level and mechanistic understanding of these novel therapeutics. In order to ensure reliability, reproducibility, and overall accuracy in analyzing a cell’s proteome, an optimized proteomics workflow is imperative. Our lab sought to understand the downstream impacts of different sample preparation protocols, differing in the material used to capture precipitated protein. Here, I present an evaluation of three proteomic sample preparation methods used on triplicates of reduced and alkylated aliquots of human cell lysate: "SP3" (single-pot solid-phase sample preparation), uses magnetic carboxylate beads as a substrate for protein aggregation; "SP4", omits a capture substrate in favor of centrifugation; and "S-Trap/S-Tip," captures protein on a borosilicate glass fibers filters. Following a Trypsin and LysC digest and desalting, samples will be run on an Orbitrap Eclipse mass spectrometer. Analysis and subsequent optimization with more complex human samples will ensure the selection of a protocol providing the highest proteomic coverage for further research.

Most eukaryotes use histones to package the genome. However, many animals package their sperm genomes using specialized DNA-binding proteins called protamines, which package DNA in sperm more tightly to fit inside the sperm head. Based on the transcriptional silencing role of protamines, we hypothesize that protamines can suppress meiotic drivers, which kill other sperm to bias their own transmission. Previously, we found that one protamine gene, Mst77F, is required to suppress meiotic drivers on the Y-chromosome in Drosophila melanogaster. Since drive is generally deleterious for the transmission of autosomal alleles (due to lower male fertility for example) theory predicts that multiple suppressors of drive will arise in populations; Mst77F may represent just one such suppressor. We hypothesized that multiple natural variants in distinct genetic loci interact with and impact meiotic drive in Drosophila melanogaster. To identify these natural variants, I crossed wildtype flies to knock out flies and generate hemizygous Mst77F flies carrying genetic backgrounds from four different populations. I measured the fertility and drive strength by crossing a single hemizygous male from each cross to five wild type females. In all cases, I found that the sex ratio was skewed to favor male offspring, indicating they all carry X-linked targets. However, I did not identify any dominant genetic variation associated with the drive strength, indicating Mst77F might be the major suppressor of this drive. I am conducting reciprocal crosses to determine whether Y chromosomes from different populations carry the same strength of drive. In the future, I will extend my analyses to other genetic backgrounds. My study contributes to a better understanding of the pervasive effects of meiotic drive in natural populations and unexpected functions of protamines.