In a recent paper in Molecular Ecology Resources, Fitzpatrick et al. used a combination of common garden experiments, genome sequencing, and machine learning analyses to understand how genomic offsets (a measure of maladaptation) can be used to predict how organisms might respond to future environmental change. They found that genetic offset was negatively associated with growth and was a better predictor of performance than the difference in sampling site and common garden environmental variables were alone. See the full article for more details on how these trends aligned with panels of putatively informative and randomly selected SNPS, and the interview with lead author Matthew Fitzpatrick below for even more insight into this exciting work.

What led to your interest in this topic / what was the motivation for this study? My research focuses on spatial modeling of biodiversity and involves forecasting how climate change may impact natural systems. Demand for such forecasts continues to grow given the threats facing biodiversity. However, a major – and often overlooked – challenge is assessing forecasting models, which is really important given their potential to (mis)inform conservation. 

The motivation for this study was to test a type of genomics-based forecast founded on an idea that my coauthor Steve Keller and I developed a few years ago that we termed “genetic offsets”. Genetic offsets are in essence a forecast of climate maladaptation based on existing relationships between (adaptive) genomic variation and climate gradients. We tested how well genetic offsets correspond to biological responses to rapid climate change – in this case by transplanting trees from their home climate to a common garden experiment and measuring their response.

What difficulties did you run into along the way? There were all sorts of challenges one might expect with setting up and running common gardens experiments in two countries, which as the modeler on the project I, thankfully, was largely isolated from. We were lucky to have Raju Soolanayakanahally on our team to help with common garden logistics in Canada, along with Steve’s lab running the Vermont common garden. Additionally, there was the challenge of how best to evaluate the population genomic data for signatures of local adaptation prior to the genetic offset modeling. This can always be a challenge to ensure you’re minimizing the effects of population structure and false positives. Steve and his former postdoc Vikram Chhatre approached this from several angles to make sure we had a robust set of selection outliers. From the modeling perspective, we had to be creative about fitting and summarizing a very large number of machine learning models.  

What is the biggest or most surprising innovation highlighted in this study? We found pretty solid evidence that genetic offsets can serve as a meaningful estimate of the degree of expected maladaptation of populations exposed to climate change. It was nice to get some confirmation of our idea, but what was really surprising was that sets of randomly selected SNPs predicted performance of trees as well as or slightly better than did our set of carefully selected candidate SNPs, which was the opposite of what we expected. We’ve seen some other evidence in our simulation studies that also suggest SNPs from the genomic background can be predictive of maladaptation, although the reasons for this are still being investigated.

Moving forward, what are the next steps in this area of research? Ours is a single study on a single species of tree. Many more tests are needed in other study systems before we can fully understand the situations in which genetic offsets can serve a useful purpose. Also, our study tested genetic offsets derived from the machine learning method Gradient Forest, but Gradient Forest is just one of several statistical methods that can be used to estimate offsets. An important next step in my lab is to perform similar testing using another promising method known as generalized dissimilarity modeling.

What would your message be for students about to start developing or using novel techniques in Molecular Ecology? Take good notes and document the process! You will thank yourself later. If you are developing a new method, it is important to thoroughly test it to be sure you understand how it behaves in different circumstances and to make clear its intended uses before publishing on it. And last, teach others to use your method! 

What have you learned about methods and resource development over the course of this project? I thought I knew a lot about Gradient Forest and its behavior, but this study – and another we have in review on testing genetic offsets using simulated data – taught me that methods do not always behave the way we might expect or hope. And even when we have simulated “truth known” data, it can be difficult to understand why methods are behaving a certain way.

Describe the significance of this research for the general scientific community in one sentence. This study shows that for some organisms it may be possible to use genetic data to inform climate change impact assessments.

Describe the significance of this research for your scientific community in one sentence. This study provides evidence that spatial patterns of adaptive genomic variation along climatic gradients can be used to estimate the magnitude of expected maladaptation of populations exposed to rapid climate change through time.  

Fitzpatrick MC, Chhatre VE, Soolanayakanahally RY, Keller SR. 2021. Experimental support for genomic prediction of climate maladaptation using the machine learning approach Gradient Forests. Molecular Ecology Resources. https://doi.org/10.1111/1755-0998.13374.