(click an image to read more)

SIR ode equations.jpg
fox tsi posterior function.png
Figure 1.png
eLife Figure 4.jpg
sea lion proud.JPG

Ecological drivers of disease spread

 
 

Where do infectious diseases come from?

How do they spread in nature?

 

What are the forces driving their transmission?

 

A key motivation for my research has been to observe, measure and analyze the ecological drivers of transmission. 

(move your cursor over the box to read more)

Population density, contact rates and transmission

When transmission depends on contacts between individuals, and if the rate of contacts can change with population density, transmission is likely to be density-dependent. Density-dependent transmission is a central concept in disease ecology and epidemiology, but testing it has proven difficult due to the fact that it requires elaborate experiments in which the density of a population is altered while measuring contacts and transmission.

Through field experiments in Tanzania we found that rodents change their number of contacts depending on how many other rodents there are, and that contacts are a sigmoidal function of density: at low densities they maintained a minimum number of contacts, while contacts increased linearly with density until reaching a certain maximum level.

(Borremans et al. 2017 Oikos

We then combined these data with a mathematical model of transmission in which we simulated transmission in a realistically fluctuating rodent population.

This showed us that the way in which contacts change with density can have significant effects on whether a pathogen can persist in the population.

(Borremans et al. 2017 Royal Society Open Science)

 
Fig1 degree_vs_density_enclosuremeans_wi
Fig2%20SIR%20stochastic%20output%204%20c
SIR ode equations.jpg

This work was done at the University of Antwerp (Belgium) with Prof. Herwig Leirs, Dr. Jonas Reijniers and many others, in collaboration with the Sokoine University of Agriculture (Tanzania)

field enclosures.jpg
mastomys grass.JPG

(move your cursor over the box to read more)

sea lion proud.JPG

Environment, immunity and transmission

Disease transmission can be governed by a variety of forces that typically fall into one of two categories: intrinsic or extrinsic drivers.

Intrinsic drivers those that arise from the host-pathogen system itself.  Immunity is a common intrinsic driver, as changes in the proportion of immune individuals can cause transmission rates to vary.  Another potential driver is behavior, for example when contacts between individuals change during seasonal breeding.

Extrinsic drivers are those that are not part of the host-pathogen system.  These can be environmental factors (temperature, humidity, ..) that change pathogen survival and transmission, or the presence of other species (predators, competitors, ...) that affect contacts, survival or health.

Using machine learning, we analyzed a 29-year time series of leptospirosis outbreaks in California sea lions, and found that both intrinsic and extrinsic drivers are important determinants of outbreak size.  The intrinsic driver was the proportion of susceptible individuals in the population (more susceptibles = larger outbreak), while the extrinsic drivers were oceanographic factors related to sea lion movement and mixing, and sea lion health.  

These results also motivated us to set up simulations to get a sense for how climate change might affect leptospirosis outbreaks.  These simulation models suggest that outbreaks will on average become smaller but occasionally more extreme due to larger environmental variation.

(This work is almost written up for publication, stay tuned!)

A crucial part of this work is the development of a method to reconstruct a time-series of the proportion of susceptible sea lions in the population, which is age- and sex-specific.  In one of the boxes below you can find more information about this exciting susceptible reconstruction.

This work was done at UCLA (US) with Prof. Jamie Lloyd-Smith and Dr. Katie Prager, and many others including The Marine Mammal Center and the NOAA Marine Mammal Laboratories.

sea lions beach.png
figure monthly cases population size and
susceptibles vs environment vs outbreak
Srec_sex_age_lepto_drivers_newSrec_all_a
 

COVID-19

 

Collecting data in nature

Estimating time of infection, analyzing serological data

 

When were individuals infected?

How do antibody levels change after infection?

 

The time of infection of individuals in a population is used to reconstruct incidence time series of pathogen transmission.

 

Because it is impossible to sample every individual every day to catch the moment they become infected,

time of infection and incidence must be estimated using sparser data.

Over the last years, significant advances are being made on time of infection estimation methods.

Using wildlife systems I have developed a strong interest in these methods, and have been thinking a lot about how we can advance them. 

(move your cursor over the box to read more)

Estimating time of infection using different sources of information.

Part 1: using experimental data.

When antibody concentrations decay at a predictable rate after infection, the concentration at any given time point can in theory be used to back-calculate when infection occurred.

Such antibody-based back-calculation methods have been developed for HIV (Brookmeyer) and other diseases (Teunis, Kretzschmar, Hens), but never to wildlife systems.

Using arenavirus infection in multimammate mice, we developed a new approach to estimating time of infection.  It uses a Bayesian framework that allows incorporating all sources of information about time of infection, instead of only antibody concentration

Examples of other sources of information: 

  • Pathogen presence/concentration

  • Disease symptoms

  • Disease seasonality

  • Age

We found that combining information dramatically improves time of infection estimates and incidence reconstruction

More information about this work can be found in Borremans et al. 2016.

This work was done at the University of Antwerp (Belgium) with Prof. Herwig Leirs and Dr. Jonas Reijniers, Prof. Niel Hens and Prof. Philippe Beutels.

time since infection different types of
time since infection equation explained.
time since infection antibody plot.jpg
time since infection incidence.jpg
time since infection incidence.jpg
time since infection nasal injection.png

(move your cursor over the box to read more)

island fox sitting.JPG

Estimating time of infection using different sources of information.

Part 2: no experimental data available.

 

This work was done at UCLA (US) with Prof. Jamie Lloyd-Smith and at Hasselt University (Belgium) with Prof. Niel Hens.

 
fox tsi prior peak titers.jpg
fox tsi posterior individual.png
fox tsi posterior function.png
fox tsi equations.jpg
fox tsi posterior peak titers.png

Estimating time of infection using antibody concentrations or other data requires a model that describes how the measured variable changes after infection.  In the case of antibodies, that means having a model that describes antibody decay.  These models can then be used to extract time of infection information from new data.  

Such models are ideally constructed using experimental data for which the time of infection is known.  A common problem however is that such data do not exist for many pathogens and host species, and this problem is holding back the widespread use of time of infection estimation methods.  

We used data on the antibody response against Leptospira sp. in Channel Island foxes to develop an approach that allows the construction of antibody decay models without the need for experimental data.  

The approach uses Bayesian MCMC to incorporate different sources of prior information about peak antibody level and possible infection windows (birth date, last negative sample, ..).

This approach opens up many datasets to time of infection estimation methods, and offers the potential to improve incidence time series for many pathogen-host systems

This work is in its final stages, and I'm very excited about this!

 

Pathogen spillover

When and where should we expect pathogen spillover between species?

How can we improve theory and concepts to better understand spillover processes?

 

Spillover is the process where parasites/pathogens cross the species barrier.

It drives the emergence of new infectious diseases and outbreaks of zoonoses, and can result in horrible pandemics like the 2014 Ebola outbreak in West Africa or the 2020-20?? SARS-CoV-2 crisis.

Spillover is notoriously difficult to study because actual spillover events are almost impossible to detect in real-time.

Almost all research on spillover relies on post-hoc inference using genetics and serology.

And despite large efforts by projects such as PREDICT, we are still far from being able to predict which pathogens will spill over next into humans, and even more elusively when and where this will happen. 

 

Inspired by, amongst others, Jamie Lloyd-Smith's field-advancing ideas on spillover I have been thinking a lot about this topic, and a while ago had the opportunity to contribute to a special issue on spillover of the Philosophical Transactions of the Royal Society that resulted from a super fun workshop organized by Raina Plowright in 2018.

(move your cursor over the box to read more)

On the spillover of pathogens across ecosystems

Spillover has been hypothesized to happen more frequently at the boundaries between ecosystems.

This is largely based on the observation that many spillover events have occurred in fragmented landscapes and in situations where people live in close contact with wildlife.

Evidence for this hypothesis is not entirely conclusive though, as it is tricky to observe rare spillover events.  

We dove into this topic by approaching it from different angles, and borrowing insights from related fields like movement ecology and macroecology.

We asked whether we should indeed expect spillover to be more likely at ecosystem boundaries, and which processes should be the most important drivers.

A central part of this paper is the application of the concept of permeability, borrowed from movement ecology, to pathogens.  In this context, permeability is the degree to which pathogens are able to cross an ecosystem barrier.  This depends on several factors like host movement, ecosystem boundary-area ratio, pathogen prevalence and edge biodiversity effects. 

If you are interested, you can download the paper here:

Borremans et al. 2019 Phil Trans B.

This work was done at UCLA with Prof. Jamie Lloyd-Smith.

Figure 1.png
phil%20trans%20cover_edited.jpg
 

SARS-CoV-2 / COVID-19

How long after developing symptoms can antibodies be detected?

How does the detection probability of viral RNA and antibodies change over time?

What are the antibody kinetics during the early phase of infection?

 

 

Understanding SARS-CoV-2 infection and transmission requires knowledge of the underlying patterns of virus and antibody dynamics.  

If we want to develop good data-based models, we need to quantify when antibodies and viral RNA can be detected and how this changes after infection.

Despite the incredible number of studies being published from the start of the COVID-19 pandemic, these data were difficult to extract and collect from the numerous smaller studies.

We performed a meta-analysis of antibody and RNA dynamics after symptom onset, with the goal of providing robust quantitative estimates that include variation and confidence intervals.
 

(move your cursor over the box to read more)

Meta-analysis of SARS-CoV-2 antibody and RNA detection

We extracted data from 22 published studies that reported original IgG and IgM antibody presence and concentration data, as well as RNA presence data, all relative to time since symptom onset.

These data were collected at different laboratories, using different assays, and from different study populations, which means care had to be taken to integrate these data. 

 

To do this, we developed new statistical methods that provided estimates of mean values as well as the variation around these means. 

This included the development of a stepwise hierarchical bootstrapping method that gives more weight to higher quality data.

The estimates derived in this meta-analysis are proving useful for interpreting antibody and RNA data, and for parameterizing mathematical models of SARS-CoV-2 transmission.

More information: Borremans et al. 2020 eLife.

This work was done at UCLA with Prof. Jamie Lloyd-Smith.

eLife Figure 1.jpg
eLife Figure 3.jpg
eLife Figure 2.jpg
 

Predicting sea lion survival following leptospirosis

When a California sea lion​ is rescued after stranding due to leptospirosis, what are its chances of survival?

Which individual variables correlate with disease severity?

 

Leptospirosis, a kidney disease caused by the bacteria Leptospira interrogans, is one of the leading causes of stranding for California sea lions. At the Marine Mammal Center in San Francisco, the world's largest marine mammal rescue & rehabilitation center, between 20 and 300 sea lions suffering from leptospirosis are treated every year. Mortality rates are high, and over 70% of affected stranded sea lions ends up dying from the disease.

During intense outbreak periods the number of rescued sea lions can increase rapidly, which means that veterinary clinicians need to diagnose the animals as quickly as possible, and assess which individuals should be euthanized. 

In order to facilitate this decision, we have developed a statistical tool that predicts sea lion survival probability using only 2 sources of easily obtainable data. 

This tool was published in the Journal of Zoo and Wildlife Medicine, and is now being used to improve the treatment of sea lions affected by leptospirosis.

sea lion proud.JPG

(move your cursor over the box to read more)

Statistical model to predict sea lion survival

We fitted a "classification and regression tree" model (CART), a machine learning method, to predict the survival of individual sea lions. This approach models the correlation between the outcome variable (survival) and other variables using a classification tree that uses simple binary decisions to predict whether the individual will survive or not.

Different combinations of multiple variables were used to fit the model, including blood chemistry measurements (BUN, Creatinine, Sodium, ..) and behavioral observations (eating behavior). Predictive accuracy (tested using cross-validation) was highest when using only 2 (!) variables (BUN and Creatinine, both known to be indicators of kidney functioning). 

Using these variables, survival could be predicted correctly for 87.6% of cases.

This simple model is highly useful as an objective tool to facilitate the hard decisions that need to be made by veterinary clinicians, and will be particularly useful during large outbreaks that can overwhelm the staff and rescue center.

More information: Whitmer et al. 2021 Journal of Zoo and Wildlife Medicine.

This work was done while at UCLA with Prof. Jamie Lloyd-Smith, in close collaboration with The Marine Mammal Center.

CSL_survival CART.jpg