RESEARCH

functional morphology

phenotypic change

biotic interactions

machine learning

I use both a micro- and macro-scopic approach to unearth underlying abiotic and biotic drivers of phenotypic change and its ecological consequences across levels of biotic organization and across time.

Within and among lineage phenotypic evolution

On generational timescales, it has been shown that lineages often change in directions of above average evolvability. It is unknown if this is true over longer timescales or if this pattern persists during periods of cladogenesis.

Here, we employ the fossil and modern record of the clonal organisms, bryozoans. Since every zooid in a colony is genetically identical to each other, we can disentangle genetic from environmental effects that give rise to phenotypic differences within and across colonies. We ask if the direction of greatest variability (i.e., what evolution can act upon) changes through time and across lineages (Balk et al., in prep.).

Some species have different morphs of the zooids, allowing us to investigate questions about modularity and integration in constraining or directing evolution. From this, we can also ask how this variation is responding to environment (phenotypic plasticity) and how it may affect the tempo and mode of a lineage's evolution.

We have taken SEMs 100s of colonies, containing 1000s of zooids, and applied machine learning to extract morphological trait data.

Abiotic drivers of phenotypic evolution

Macroevolutionary studies investigating show that it is easier for a mammal to get smaller than larger, as it has already moved through that morphological space.

I tested this hypothesis at the microscale by investigating rates of body size changes of the bushy-tailed woodrat, Neotoma cinerea. Since N. cinerea is known to decrease size when it is warmer and increase size when cooler, following Bergmann's Rule, we can ask if populations became changed body quicker during periods of warming (smaller) or cooler (larger). We used a 36,000 year record of the Bushy-tailed woodrat across the western United States, consisting of 1000s of size estimates. We found no difference in rates of change, suggesting that these packrats do not follow the macroevolutionary pattern (Balk et al. 2019).

Future research will investigate if the phenotypic space or amount of variation influences amount of phenotypic change, as well as tempo and mode of evolution within a lineage.

Biotic drivers of phenotypic evolution

"Animals increase body size to avoid predation" is a common, yet unsubstantiated, 'fact' in the literature. I approach answering this question by looking at the response of a predator in the fossil record to a prey base that increases in size. I use the mid-Miocene to Pliocene fossil record, a period that encompasses the extinction of the megatooth shark, Otodus megalodon, and sees diversity and size changes in their supposed prey, marine mammals. I first seek evidence of predation intensity on marine mammals through trace fossils of shark-bitten bones to quantify which taxa and size prey were most targeted. To date, I have imaged over ~400 shark bitten bones.

Next, I quantify modern predator-prey body size relationships. These are first-order approximations of how energy flows through a system and of selection pressures a species may experience. I use modern shark diet data from FishBase and data on prey items to construct scaling relationships between shark body size and their prey body size. We find a strong, positive relationships in the maximum prey size and variance of prey body size as shark body size increases. This means that larger-bodied shark species can eat a larger and a wider variety of prey sizes. Changes in diet may cause destabilization in populations and change community structure. Humans, in particular, may contribute to prey abundance and therefore alter available prey items for consumption. We find that possible extinctions of certain prey items will cause drastic changes in shark diet and affect these allometric relationships significantly (Balk et al., in prep.).

We then apply this research to the past by asking if the size range of prey that Otodus megalodon could consume are dramatically reduced.

Future research will investigate other life history changes in prey items that allow them to escape predation.

Mobilizing digital records

Digitized records

As trait data is extracted from images or by researchers, there needs to be a findable and accessible repository. The Functional Trait Resource for Environmental Studies (FuTRES) project is a collaborative project among four universities (University of Oregon, University of Arizona, University of Florida, and Howard University). The key deliverables of FuTRES are a workflow for assembling functional trait data measured at the specimen level, a database to serve that data, and scientific publications demonstrating the utility of the assembled data (Balk et al. 2022).

I work with paleontologist, zooarchaeologists, and biologists to create a template that holds appropriate metadata fields and ontologize trait terms in their dataset. When possible, we use Darwin Core terms for template terms to make data more interoperable. We integrate with existing ontologies, such as the Uberon multi-species anatomy ontology (UBERON) for anatomical terms and the Ontology for Biological Attributes (OBA) for trait terms. All code can be found on GitHub.

Machine learning for image data

Museums are digitizing data: both into databases and by imaging. It is vital to harness this data revolution by creating workflows that are reusable and interoperable as new tools and methods are rapidly developed.

The Biology-Guided Neural Networks is a multi-institute project (Seattle Children's Hospital, Drexel University, Tulane University, Virginia Tech University, National Ecological Observatory Network) seeking to extract traits from specimen images using a machine learning workflow. The project combines machine learning on specimen images, collected by museums and iDigBio, with ontologies to help improve the performance of the machine learning algorithm (Balk et al. 2024).

Google Sites

Report abuse