Abstract:
Cell-to-cell heterogeneity in gene expression can elevate antibiotic resistance in one microbe while other cells remain susceptible. These transient forms of drug resistance are often stochastic and dynamic, leading to single-cell level differences in resistance that change with time. To date, methods for quantifying these effects have relied on careful observations of native expression patterns. In this talk, I will discuss a novel approach for controlling gene expression dynamics in single cells that can be used to precisely drive expression in thousands of cells in parallel. Using recent advances in the fields of machine learning and control theory, we train a deep neural network to accurately predict the response of an optogenetic system in E. coli cells. We then use the network in a deep model predictive control framework to impose arbitrary and cell-specific gene expression dynamics on thousands of single cells in real time, applying the framework to generate complex time-varying patterns. We also showcase the framework’s ability to link expression patterns to dynamic functional outcomes by controlling expression of an antibiotic resistance gene. These approaches offer powerful methods that can be used to quantify and control cell-to-cell heterogeneity in antibiotic resistance, providing a detailed view into strategies bacteria can use to evade drug treatment.
Bio:
Mary Dunlop is an Associate Professor of Biomedical Engineering and Dorf-Ebner Distinguished Faculty Fellow at Boston University. She holds additional affiliations in Bioinformatics and Molecular Biology, Cell Biology & Biochemistry. She graduated from Princeton University with a B.S.E. in Mechanical and Aerospace Engineering and a minor in Computer Science. She then received her Ph.D. from the California Institute of Technology, where she studied synthetic biology with a focus on dynamics and feedback in gene regulation. Her lab engineers novel synthetic feedback control systems and studies naturally occurring examples of feedback in gene regulation. In addition, her research has focused on understanding the role of cell-to-cell heterogeneity in bacterial systems. In recognition of her outstanding research contributions, she has received many honors including election as an AIMBE Fellow, the NSF Transitions Award, ACS Synthetic Biology Young Investigator Award, DOE Early Career Award, and NSF CAREER Award. She is also the recipient of several teaching awards, including Boston University’s Biomedical Engineering Professor of the Year Award and the College of Engineering Teaching Excellence Award.
Summary:
Focus: modeling and control of cell gene expression
Observation experiment:
Cells exposed to a short pulse of antibiotic
Most cells die but a few may live and recolonize the whole area
In this experiment this is not a genetic difference: survivors are not more resilient to repetition of antibiotic pulse
Difference is different expression of the same genes
Significant diversity between cells
Causes:
During cell division key regulatory proteins are divided unevenly between the daughter cells
Feedback in gene regulatory networks
Delays between sensing and transcription
Difference in nutrients, temperature, etc.
Experimentally observing dynamic changes in gene expression
Expose different cells to same stimuli, observe different responses and their gene expression
Try to relate gene expression to response to stimuli
Example: when GadX gene is being expressed, cell is more resilient to antibiotics
Goal:
Model this relationship
Control gene expression dynamics to direct cell behavior
Active control of cell response to light
Look at cell response to different color light
Feedback control strategy
Look at the impact of light exposure to cell’s production of GFP (Green Fluorescent Protein) levels
Adjust light exposure to get a target GFP time series pattern
Explored traditional control algorithms
Bang-bang control: simple but noisy
Proportional Integral Derivative: PID
Model Predictive Control: MPC
Based on an expert-defined mathematical model of the cell’s response dynamics
Sampling used to infer the treatment that will produce desired outcome
One-size-fits all model doesn’t account to unknown attributes of cell state/response
Model may be expensive to evaluate
Deep Model Predictive Control (DeepMPC)
Replace expert-defined model with a neural network
Can be more flexible and can incorporate additional data sources (e.g. cell length, image sharpness)
Experiment: 16k single-cell traces
Model
Input 8 different features of the cell (image sharpness, cell area, etc.)
LSTM Encoder
MLP Decoder
Predicted GLP time series under given light patterns
Evaluation: given past time series, predict the future GLP behavior
Accuracy of DeepMPC a little better than the expert model
But is much faster
Used this predictive model for cell control
DeLTA: DeepLearning for Time-lapse analysis
https://gitlab.com/delta-microscopy/delta
Imaging library for tracking cell state
Controlling cells with a customized objective for each cell
Can generate complex activation patterns across cells
E.g. make the cells play a movie; works for ~20 hours
Cells start to degrade over time, their growth drops off until they die
Observation: fast-growing cells are error-prone
Optogenetic control of Antibiotic Resistance
Control green/red light pattern to control how well a cell survives antibiotic exposure
Prediction of complex dynamics
Bistable systems are hard to control for
Model tends to aim for the average of the two states, which is ineffective
Solved by replacing decoder (MLP) with convolutional decoder: probability distributions
Can predict the different possible states