Predicting Competition Dynamics from Growth Curve Data

Given at the 8th ILANIT/FISEB conference, Feb 22, 2017.

Fitness is not well estimated from growth curves of individual microbial isolates in monoculture. Rather, competition experiments must be performed to better infer relative fitness. However, competition experiments require unique genotypic or phenotypic markers, and thus are difficult to perform with isolates derived from a common ancestor or non-model organisms. I will present Curveball, a new computational approach for predicting competition dynamics and inferring relative fitness from growth curve data. We validated this approach using growth curve and competition experiments with bacteria. By integrating several growth phases into the fitness inference, Curveball offers a holistic approach to fitness inference from growth curve data.

Transcript

1. What makes Racehorse fast?
   Predicting Competition Dynamics
   from Growth Curve Data
   Yoav Ram
   Stanford University
   Photo by Noah Silliman
   

   View Slide

2. Fitness
   The ability of a genotype to increase in
   frequency relative to other competing
   genotypes in the population
   2
   

   View Slide

3. Who’s fitter?
   3
   

   View Slide

4. How much fitter???
   4
   

   View Slide

5. Now who’s fitter?
   5
   

   View Slide

6. Growth phases
   6
   Also…
   Death phase?
   Diauxic shift?
   Hall et al., MBE 2014
   

   View Slide

7. Competition experiments
   • Growth in a mixed culture
   • Captures all relevant growth
   phases
   • Fitness inferred from change
   in strain frequencies
   7
   Modified from Elena & Lenski, 2003
   Strain 1 Strain 2
   

   View Slide

8. Competition experiments
   8
   Ara- Ara+
   Strains must have a genotypic
   or phenotypic marker
   • LTEE: arabinose utilization
   phenotype
   Lenski et al., Am Nat 1991
   • Fluorescence and flow cytometry
   Gallet et al., Genetics 2012
   • Lineage tracking with deep
   sequencing
   Bank et al., Evolution 2013
   Levy et al., Nature 2015
   

   View Slide

9. Competition experiments
   Strains must have a genotypic or
   phenotypic marker
   Laborious and Costly
   More so for non-model organisms
   9
   Ara- Ara+
   

   View Slide

10. Competition experiments
    Problem:
    Laborious and Costly
    Our Solution:
    Theoretical framework that predicts
    competition results
    10
    Ara- Ara+
    

    View Slide

11. Prediction and validation
    11
    • Two E. coli strains with fluorescent
    proteins (GFP, RFP)
    • Growth in mono- and mixed
    culture
    • Measure OD over time
    • Fit growth model
    • Predict competition results from
    fitted growth models
    • Use fluorescence to measure
    frequencies in mixed culture
    • Compare predictions to
    measurements
    Frequency
    Time (hr)
    Time (hr)
    Density (OD595
    )
    

    View Slide

12. Prediction and validation
    12
    • Two E. coli strains with fluorescent
    proteins (GFP, RFP)
    • Growth in mono- and mixed
    culture
    • Measure OD over time
    • Fit growth model
    • Predict competition results from
    fitted growth models
    • Use fluorescence to measure
    frequencies in mixed culture
    • Compare predictions to
    measurements
    Frequency
    Time (hr)
    Time (hr)
    Density (OD595
    )
    

    View Slide

13. Prediction and validation
    13
    • Two E. coli strains with fluorescent
    proteins (GFP, RFP)
    • Growth in mono- and mixed
    culture
    • Measure OD over time
    • Fit growth model
    • Predict competition results from
    fitted growth models
    • Use fluorescence to measure
    frequencies in mixed culture
    • Compare predictions to
    measurements
    Frequency
    Time (hr)
    Time (hr)
    Density (OD595
    )
    

    View Slide

14. Growth curves
    14
    DH5α vs. TG1
    - model fit
    ● experiment
    

    View Slide

15. Competition
    15 Time (hr)
    DH5α vs. TG1
    -- model prediction
    ● experiment
    

    View Slide

16. DH5α vs. TG1 without lag phase
    16
    - model fit
    ● experiment
    Growth curves
    

    View Slide

17. DH5α vs. TG1 without lag phase
    17
    -- model prediction
    ● experiment
    - model fit
    ● experiment
    Growth curves Competitions
    

    View Slide

18. Applications
    • Complex growth curves
    • Null model for
    detection of direct
    interactions
    • Interpret fitness
    differences
    • Predict adaptive
    evolution
    18
    Monod, Ann Rev Microbiol 1949
    Monod, Nobel lecture 1965
    

    View Slide

19. Summary
    1. Fit growth models to growth curves
    2. Predict competition results
    3. Infer fitness
    Preprint:
    Ram et al. Predicting microbial relative growth in a
    mixed culture from growth curve data. bioRxiv,
    doi:10.1101/022640
    Website: http://curveball.yoavram.com
    19
    Frequency
    Time (hr)
    

    View Slide

20. Acknowledgments
    20
    Hadany Lab / Tel-Aviv University
    Lilach Hadany
    Uri Obolski
    Eynat Dellus-Gur
    Berman Lab / Tel-Aviv University
    Judith Berman
    Maayan Bibi
    CONTACT
    yoav@yoavram.com
    @yoavram
    github.com/yoavram
    www.yoavram.com
    Israeli
    Ministry
    of Science &
    Technology
    

    View Slide