A new method of examining gene expression patterns called landscape transcriptomics may help pinpoint what causes bumble bees stress and could eventually give insight into why bee populations are declining overall, according to a study led by researchers at Penn State. The team published their findings in the journal Molecular Ecology.
This study was the first test of the emerging field of landscape transcriptomics, recently envisioned by an interdisciplinary team led by scientists in Penn State's College of Agricultural Sciences. The team hypothesized that it would be possible to collect animals and plants from the wild and determine which stressors they experienced based on specific patterns or signatures in their gene expression profiles.
In this study, the scientists used an artificial intelligence approach known as machine learning to evaluate gene expression profiles of individual bumble bees. The researchers found that the method could accurately identify genetic signatures of stressors, such as excessive heat and cold, in bees both in the lab and in the wild.
Gabriela Quinlan, who led the study when she was a postdoctoral scholar in the College of Agricultural Sciences, said the findings suggest landscape transcriptomics could be used to accelerate conservation efforts for at-risk species.
"This is a really big step in demonstrating this new strategy for identifying at-risk populations and showing how these machine learning models can be used both in the lab and out in the field," she said. "We also found directly applicable insights such as sets of genes that are associated with different stressors in bumble bees, which we didn't have before."
Bees - like many other species of plants and animals - are currently in decline around the world, the researchers said, suggesting that something is causing enough stress to trigger population declines. However, understanding exactly what those stressors are can be challenging.
That's where landscape transcriptomics comes in, according to Christina Grozinger, Publius Vergilius Maro Professor of Entomology and director of the Huck Institutes of the Life Sciences.
"It's like forensic biology, where you can look at an organism's gene expression patterns and identify a signature or fingerprint that relates to the stress it's experiencing," Grozinger said. "Landscape transcriptomics should allow us to identify stressed populations of target species much more rapidly than traditional approaches, which require collecting and analyzing many samples over long periods of time."
While previous studies have found that it was possible to detect transcriptional signatures of specific stressors in organisms reared and treated in the lab under very controlled conditions, the research team wanted to see whether the method would still be feasible in organisms living in the wild.
"In typical laboratory studies, we use organisms from the same genetic background, the same age, reared in the same way, and which are exposed to stressors at tightly controlled levels and periods of time," Grozinger said. "But collecting organisms from the wild, we know nothing about them or the stressors they've experienced, so we were curious if we could still see these specific stress-related transcriptional fingerprints."
For this study, the researchers first did an experiment in the lab in which they exposed bumble bees to different types of stressors, including heat, cold and immune challenges. They then extracted the bees' RNA - the genetic material used to build proteins and help regulate biological functions - and sent the samples to the Penn State Genomics Core Facility for high throughput sequencing.
This gave the researchers information on the number of RNA strands corresponding to each gene, which represents the gene expression patterns. The RNA profiles from individuals that were exposed to different stressors were then used to train a machine learning model, which is a type of artificial intelligence, to recognize which patterns of gene expressions were associated with each type of stressor.
A strength of this study, Quinlan said, was the approach they took to training their machine learning model to recognize the different genetic signatures for each stressor.
"We know that in nature, organisms are going to be subjected to multiple stressors at the same time, so how do we differentiate one stressor versus another?" she said. "We used an algorithm that takes all the inputs from all the genes and finds the patterns that emerge when the bees were experiencing each stressor. We were able to get a model with 92% accuracy, which we could then use to assess wild bees."
For their second experiment, the team collected wild bees from two sites: one in the Arboretum at Penn State and another in a forested, more mountainous area.
The sites were chosen to increase the likelihood that the bees would be exposed to different stressors - those at the Arboretum had access to abundant floral resources and lots of sun, while the bees from the second site experienced more shade and might not have as many flowers to forage.
After analyzing the transcriptomes of these bees, the researchers found that the model was once again very accurate in predicting which stressors the bees were experiencing. However, the researchers found that these signatures didn't last long in the bees' RNA. In bees collected the morning after a heat wave, when environmental temperatures had gone back to normal, their genetic signatures for heat stress were no longer visible.
But this also meant that the researchers could glean precise details about how bees go about their day. For example, they discovered that many bees exhibited signatures of starvation stress when they were collected in the morning as compared to the evening, which might give insight into how bees forage and their motivation to forage.
Quinlan said in the future, additional work could train these models to pick up on stressors that bees have experienced for longer periods of time.
"The initial motivation for this study was to see longer-lived signatures so we can learn what is stressing these bees long term and might be contributing to population decline," she said. "Transcriptomics is very acute in that it picks up on what's going on in the moment, so it might be a matter of training these models to pick up on these longer term, more subtle differences."
Heather Hines, associate professor of biology and entomology, also co-authored this study.
The Penn State College of Agricultural Sciences Strategic Networks and Initiatives Program and the U.S. National Science Foundation Postdoctoral Research Fellowship in Biology Program supported this research.
