Research Overview

Predicting the phenotypic impacts of a mutation is a major goal in biology and medicine. But the paths linking genotype to phenotype are difficult to navigate. For one, some phenotypes impact others, so the impacts of mutation can stretch out across networks of related traits, percolating from impacts at the molecular level to higher orders of biological organization. Moreover, the impacts of mutation, and the networks of traits through which they spread, change across contexts like the environment or genetic background. Comprehensively mapping genotype to phenotype is akin to untangling a ball of knotted strings, only to realize the task needs to be repeated many times.

We implement and develop high-throughput phenotyping methods to simultaneously quantify the phenotypic impacts of many mutations across many contexts. We interpret these data in different ways. In some projects, we quantify the correlations between phenotypes to infer the network through which a mutation’s influence travels and how that network changes across contexts. In other projects, we deconvolute these high-dimensional data into an abstract map that uses shared patterns of mutant behavior across contexts to make phenotypic predictions.

Some of our work is driven by mechanistic hypotheses about how basic properties of cells change the phenotypic impacts of mutation in predictable ways, while other work focuses on large collections of adaptive mutations generated by laboratory evolution experiments. Our overall goal is to build predictive maps from genotype to phenotype and, in so doing, to generate insights about the map’s structure (e.g., is it modular or interconnected?) and tools to study this map that will be broadly useful to the biology community.

Building genotype-phenotype maps of drug resistance

Strategies in evolutionary medicine aim to thwart the evolution of drug resistance by using combinations of drugs that demonstrate “collateral sensitivity,” which is when mutants that resist Drug ‘A’ increase susceptibility to Drug ‘B’. Problematically, it often turns out that resistance and susceptibility are not inextricably linked, and some rare mutations exist that can disrupt these phenotypic correlations. Most technologies cannot quantify the prevalence of collateral sensitivity. We utilize a laboratory evolution platform that is perfect for doing so in the model organism S. cerevisiae. Rather than identifying the most resistant mutants in Drug ‘A’, this platform utilizes thousands of barcoded lineages to explore all mutations that resist Drug ‘A’. Then we quantify their resistance to Drug ‘B’ (and C, D, E…you get the idea). Characterizing these tradeoffs reveals trends and limits that define a fitness landscape. These landscapes have multiple peaks, corresponding to multiple ways to resist a drug that come with different tradeoffs (e.g., some may come without collateral sensitivity). Once we create abstract genotype-phenotype maps, we can make predictions about how organisms will evolve when exposed to new conditions.

Relevant Papers:
-Drug resistant mutants have different fitness tradeoffs link
-Our review on the challenges of predicting evolution and studying adaptive mutations link
-Rare mutations (and other contextual changes) can disrupt correlated phenotypes link
-Abstract genotype-phenotype map example link

Quantifying the context-dependent costs of misfolding

Protein misfolding toxicity is a major factor in neurological disease. The high load of misfolded proteins in tumors has also been called cancer’s ‘Achilles heel.’ But to understand how misfolding impacts health, we need to quantify how cell fitness declines as misfolded proteins accumulate. A hypothesis as to why misfolded proteins are toxic is that they steal resources (eg, chaperones), implying toxicity depends on the availability of resources. Our goal is to make accurate predictions about the context-dependent costs of misfolding mutations by quantifying the relationship between the number of misfolded proteins in a cell and cell fitness. We do so through a combination of novel methods, some of which utilize DNA barcodes and CRISPR gene-editing, some of which hunt for subpopulations of cells that are sensitive to misfolding, and some of which involve technologies that we developed ourselves. Deciphering how the cost of misfolding scales with the number of misfolded proteins will improve predictions about the combined costs of multiple misfolding mutations. It will also reveal when and how these extremely common, but typically small-effect, mutations contribute to evolutionary processes.

Relevant Papers:
-A novel system to identify mutations that cause protein misfolding link
-Our review of mechanistic hypotheses underlying context-dependence link
-Quantifying the toxicity of misfolded proteins link
-Chaperone levels go up in lock step with the number of misfolded proteins link

Devloping high-throughput phenotyping technology

Quantifying the myriad phenotypic impacts of genetic change at the molecular, cellular and organismal levels, and doing so over and over in many different contexts requires incredibly high-throughput phenotyping methods. We develop phenotyping technology, having thus far filed 3 patents, all of which list multiple lab members as inventors. Some of these methods pertain to new single-cell sequencing technology that allows combining DNA and RNA sequencing of the same cell. Other methods involve precisely quantifying the stability of thousands of protein variants at the same time. In addition to developing new methods, we also implement, modify, and rigorously test technologies created by other labs. For example, we have modified an inexpensive and highly multiplexable single-cell RNA sequencing platform for yeasts. We’ve also performed 79 technical replicates of a method to measure organismal fitness in order to determine how to achieve higher precision. We care about reproducibility, so we share what we learn through open science campaigns like #1BigBatch, preprints, and special issues. Many of our methods employ DNA barcodes, represented in our lab logo to the left.

Relevant Papers:
-Optimizing an existing single-cell RNA sequencing platform for yeasts link
-A novel system to identify mutations that disrupt protein stability link
-Designing an ultra-precise experiment to measure fitness link
-A special issue on Best Practices in Microbial Evolution link

Breaking the (growth) law

Countless studies have shown that the rate of cell growth correlates with many other phenotypes, including the amount of ribosomes being produced and cell size. These correlations, often called “growth laws”, are powerful because they mean you can predict some phenotypes from others. But we’ve shown that these laws can be broken. For example, ribosome levels do not track growth rate when growth rate is depressed by misfolded proteins. Our goal is to better understand when the growth laws predict phenotype and under what contexts these laws are broken. Some of our work in this area involves single-cell RNA sequencing to find conditions where ribosome levels do not scale with growth rates. Other work involves evolution experiments that select on both cell size and cell growth rate in an effort to tease apart these two linked traits. Will we be able to break the laws? We hope so!

Relevant Papers:
-Ribosome levels do not scale with growth when growth is slowed by misfolded proteins link

Tracing mutations impacts through regulatory networks

A theme in many of our studies is our goal to predict some phenotypic responses from others, for example, we want to predict the toxicity of a misfolded protein from its degree of misfolding, or predict growth rate from ribosomal content. We also want to know the limits of these predictions and under what contexts they fail. Some of the most complex networks of related traits are gene regulatory networks. These often can reveal how the impacts of mutation fan out at the molecular level affecting the abundances of many transcripts. We’ve optimized an RNAseq method to report on correlated transcript levels across thousands of single yeast cells. We will use these data to infer regulatory networks and how they change across contexts. This work will uncover the paths linking genotype to phenotype to phenotype to phenotype and how these paths can change across contexts.

Relevant Papers:
-Inferring networks of related traits that predict which traits are jointly affected by mutation link
-Optimizing a single-cell RNA sequencing technology to investigate gene regulation in yeasts link

Predicting patterns of epistasis in biological networks

Some components of regulatory networks can modify the impact of mutations to other components. For example, proteins that repress the activity of other proteins may also tend to repress the impacts of mutations. What are other rules that dictate whether one protein represses or enhances the impacts of mutations in another? To investigate rules that arise from regulatory relationships, my lab focuses on the network of proteins that are regulated by the protein-folding chaperone, Hsp90. Hsp90 modifies the impacts of mutations in many genes, sometimes dampening these impacts but more often exaggerating them. My lab’s goal is to predict which impacts of mutation will be repressed v. enhanced by Hsp90. In doing so we strive to reveal generic rules that explain how regulatory network structure contributes to epistasis.

Relevant Papers:
-Simple regulatory relationships predict direction of epistasis link
-Hsp90 interacts with mutations in diverse ways link