Data anlalysis for the project hambiTippingPoints

Manuscript:

Preprint available from bioRxiv

Background

The effects of antimicrobial therapies on the human gut microbiome is a critical issue for global health. A typical consequence of antimicrobial therapy is microbiome dysbiosis - the alteration of microbiome structure from a “healthy” state. A human microbiome in dysbiosis can provide growth opportunities for pathogenic and antimicrobial-resistant bacterial strains, which are global threats to human health. Alternatively, an ecologically resilient microbiome resists antimicrobial disturbances and pathogenic invasions. Resilience is an ecosystem’s ability to withstand or rebound from disturbance to its original state. Abrupt shifts in the ecosystem state, known as critical transitions, can occur if a disturbance exceeds the ecosystem’s buffering capacity, which pushes the system to an alternative and often drastically different stable state. Critical transitions are well-known phenomena in large-scale ecosystems - e.g., commercial fisheries collapse due to over-harvesting - but largely unexplored in the field of microbiome research. There is a great need to predict when and how microbiomes may transition to compositionally and functionally alternative states in response to pressures ranging from antibiotic treatment to temperature change, flooding, fires, and other stresses resulting from anthropogenic climate change.

One unique property of microbiomes is that species can rapidly evolve in response to stress. This rapid evolution will occur on the same time scale as ecosystem dynamics that determine critical transitions. Thus, any comprehensive predictive model of critical transitions in microbiomes will need to account for internal evolutionary processes affecting species interactions and overall community state. For example could the evolution of antibiotic resistance in response to antibiotic treatment accelerate the transition of a microbiome to an alternative state? Alternatively, might evolution stabilize internal community dynamics buffering against change? These questions are completely unanswered and are the motivation for this project.

Additional background related to density dependent growth

The Allee effect (Allee et al. 1949) is one mechanism underlying critical transitions in natural systems. The Allee effect describes a phenomenon where the per capita growth rate of a population is maximal at intermediate species densities, reduced at high densities, and negative at low densities. At high densities individuals compete too strongly for resources driving down per capita growth rates. At low densities per capita growth is negative because there are not enough individuals to properly find mates, form groups for hunting or predator avoidance, and other cooperative behaviors.

Antibiotic degradation by commensal bacteria can be a cooperative behavior. For example, there can be a public goods effect where some species can degrade or modify antibiotics which has the downstream effect of shielding more vulnerable strains. There is precedent for this in clinical settings (Wright 2005; Gjonbalaj et al. 2020) when measuring the minimal inhibitory concentration (MIC) of antibiotic for different pathogen isolates. For example as stated in Artemova et al. 2015

However, while the MIC has been used as a single value proxy for fitness, its relationship to evolutionary fitness is often complicated. For β-lactam antibiotics, the oldest and most widely used class of antibiotics (Bonomo et al. 2007), the MIC is frequently subject to the “inoculum effect”: Its measured value is strongly dependent upon the starting cell density of the culture (Brook 1989). This occurs because resistance to β-lactams is often achieved via hydrolytic inactivation of the antibiotic by resistant cells, which can benefit the entire bacterial population by causing overall depletion of antibiotic (Dugatkin et al. 2005; Clark et al. 2009). β-lactams are bactericidal, and therefore, any bacterial population that survives the treatment will often go through a phase of cell death. The dynamics of these populations can be complex (Yurtsev et al. 2013), and since the MIC is sensitive to the initial cell density, the relevance of a high-density MIC measurement to the evolutionary fitness of individual bacteria is unclear (Goldstein et al. 1991).”

In this scenario cell surface beta-lactamases degrade the antibiotic before it can enter the cell. If there are enough beta-lactamase expressing individuals in the population, then this inactivates a large proportion of the total antibiotic dose which provides protection to cells that do not express beta-lactamases.

This is a classic public-goods scenario and is susceptible to Allee effect where the susceptibility of a community to Ampicillin depends on the density of community members expressing beta-lactamases. Ampicillin acts as an irreversible inhibitor of the enzyme transpeptidase which is needed for cell wall biosynthesis. The idea here is that evolution could alter this public-goods scenario by mutating beta-lactamase genes or by making a species more intrinsically resistant (for example a mutation to the transpeptidase that allows the protein to function in the presence of ampicillin). On one hand, any mutation that, for example, increased the efficiency of the beta-lactamase could in essence become a “public” mutation because it would increase degradation of total Ampicillin which also benefits strains or even different species without the mutation. On the other, hand a mutation to, for example, the transpeptidase would be a “private” mutation because it will only benefit the strain with the mutation. We are interested in the dynamic between private and public mutations and how this might affect tipping points.

Theory predicts that we can anticipate catastrophic state shifts in an ecosystem by observing increasingly slow recovery to perturbations as the system approaches a tipping point - an early warning signal termed critical slowing down. Critical transitions can occur if a disturbance exceeds the ecosystem’s buffering capacity, which pushes the system to an alternative and often drastically different stable state. Critical slowing down has been demonstrated experimentally for monocultures of yeast (Dai et al. 2012) and cyanobacteria (Veraart et al. 2011).

The beta-lactamase in the 23 species HAMBI community is encoded by the ampC gene. I have detected AmpC β-Lactamases in the genomes of 17 HAMBI strains: (0105, 1279, 1287, 1292, 1299, 1874, 1896, 1966, 1972, 1988, 1992, 2164, 2443, 2467, 2659, 3031, 3237).

Relevant papers:

• Recovery rates reflect distance to a tipping point in a living system
• Generic Indicators for Loss of Resilience Before a Tipping Point Leading to Population Collapse
• Early warning signals of extinction in deteriorating environments
• Isolated cell behavior drives the evolution of antibiotic resistance
• The inoculum effect and band-pass bacterial response to periodic antibiotic treatment
• Eco-evolutionary feedbacks between private and public goods: evidence from toxic algal blooms

Experiment overview

The repo contains data from an experiment with a synthetic community of 23 bacterial species exposed to the model penicillins class antibiotic ampicillin as a stressor. Here we aimed to ask:

1. Where are the tipping points leading to population collapse in this system?
2. What are the early warning signals / critical slowing down of population collapse?
3. What is the role for rapid antibiotic resistance evolution in position ecosystem collapse or resillience?

Experimental evolution

Prior to the main experiment, 23 species were grown and experimentally evolved to have resistance/tolerance to the antibiotic ampicillin. Species were each grown in monoculture in serial transfer (x% volume transfer every x days) in the presence of ampicillin done my Ida-Marija (Same protocol as in APEX experiment - I will copy this over when Johannes determines the protocol).

Experiment setup

Three different communities were constructed using species with different evolutionary histories and ampicillin sensitivity.

1. EVO1: In this community sensitive and experimental evolved species were mixed. The idea was to add ampicillin evolved species that are typically rare in this community (< 1% average abundance) to test for the effect of this trait in rare species. Amp-evolved species included HAMBI_0097, HAMBI_0262, HAMBI_1279, HAMBI_1299, HAMBI_1842, HAMBI_1988, HAMBI_2159, HAMBI_2164, HAMBI_2494, HAMBI_2659, HAMBI_2792, HAMBI_3031, and HAMBI_3237.
2. EVO2: In this community all 23 species added were ampicillin resistant/tolerant.
3. EVO3: In this community all 23 species were in their clonal/ancestral form.

Experiment treatment conditions

EVO1, EVO2, and EVO3 communities were each grown under three different gradually increasing Ampicillin (AMP) concentrations (endpoints: 5 ug/ml, 50 ug/ml, and 500 ug/ml) + control without AMP. For the first 30 days non-control communities were serially transferred every 24 hours (1/10 dilution, 1 ml total, 96 deep-well plates) to higher AMP concentrations with the AMP endpoints being reached after 30 days - see Figure 1 below. After 30 days, communities were allowed to recover by serially transferring without ampicillin for 41 additional days.

Figure 1: Ampicillin concentration (vertical axis) over time (horizontal axis) for the three different experimental ampicillin endpoint concentrations (5, 50, and 500 ug/ml). The no-ampicillin control is not shown. Note: data plotted in A) is a log-scaled, while in B) there is no scaling.

Measurements and data types

• OD600 measured every 24 hours before serial transfer.
• DNA preserved every 24 hours (-20 C) for amplicon sequencing
• Every odd transfer samples were stored 1:1 w/ 85% glycerol in -80C for later clone isolation

Code

Data and code here is provided under GPL3. Feel free to use or remix as you see fit.

Project structure

1. /R contains R scripts
2. /data contains data that has been processed in some way for later use
3. /_data_raw contains unprocessed data scraped from compute cluster
4. /figs contains figures generated from R scripts

Availability

The rendered project site is available at https://hambitippingpoints-slhogl-623cfec303619cc856140fccfc094718702f7.utugit.fi/. The website has been produced using Quarto notebooks.

This GitLab repository (https://gitlab.utu.fi/slhogl/hambiTippingPoints) hosts the code and data for this project. The rendered webpage can be fully recreated using the code at https://gitlab.utu.fi/slhogl/hambiTippingPoints.

Reproducibility

The project uses renv to create reproducible environment to execute the code in this project. See here for a brief overview on collaboration and reproduction of the entire project. To get up and running you can do:

install.packages("renv")
renv::restore()