This work is still preliminary and has yet to undergo peer-review. This page is also under construction. However, I am always open to questions and comment on the research. Thank you for visiting!

Return to Main Webpage

Background

For ectothermic organisms, those which cannot create their own body heat, the temperature of the environment can have important impacts upon their physiology, ecology and behavior. If conditions are too hot or too cold, then organisms may be unable to complete important aspects of their life history and may suffer non-lethal effects of stress. Just like you may have a preferred setting on the thermostat, organisms can also have preferred temperatures; temperatures which suit their physiological needs and theoretically improve organism fitness.

AquaMaps (2019, October). Computer generated distribution maps for Trematomus bernacchii (Emerald rockcod), with modelled year 2050 native range map based on IPCC RCP8.5 emissions scenario. Retrieved from https://www.aquamaps.org.

Fish of the Antarctic are unusually in being able to survive in sub-freezing temperature. To prevent the formation of ice within their bodies, Antarctic fish have evolved a suite of physiological responses to allow the to live in such frigid temperatures. It is though that these physiological responses are energetically expensive. Furthermore, the water temperature of McMurdo sound is exceptionally stable, thought to hardly change over millions of years. Due to global climate change, the temperature of the Southern Ocean and McMurdo Sound are increasing more rapidly than nearly anywhere on Earth, and how fish will respond to this change in the thermal landscape is unknown.

Our project had two goals. The first was to determine, for the first time, the temperature preference of two Antarctic fish species. The Emerald Rockcod (Trematomus bernacchii) and the Sharp-spined Notothen (Trematomus pennellii) are common, benthic species crucial to the depauperate food-web of the Ross Sea. The second goal of our project was to demonstrate the effectiveness of our Temperature Preference Apparatus (TPA) in a remote setting. An initial version of the TPA was designed by Myrick et al. (2004) and provides a stable thermal gradient within which the fish may select a preference. For our work in the Antarctic we miniaturized the device to better suite the size of our fish and the requirements of working in a remote setting.

Juvenile T. bernacchii, K. Zillig

Hypotheses

We hypothesized that Antarctic fish species would prefer temperature which would minimize energy expenditure or that would reduce stress. Living in Antarctica, these two species have evolved to prevent the formation of ice crystals within their body, and therefore we speculated that living at sub-zero temperatures would be energetically stressful and fish may exhibit thermal preferences warmer than -2°C (The typical temperature of the Ross Sea) and perhaps even above freezing (0°C) thereby avoiding the formation of ice crystals.

Methods

The temperature preference apparatus (TPA) consists of a ring-shaped channel surrounded by concentric rings of mixing chambers. Water of three different temperatures is differentially supplied to the various mixing chambers. Water was partially recirculated with only the cold-water basin receiving new water from McMurdo sound. The medium and warm basins were heated using aquarium heaters. The TPA produces a stable thermal gradient which does not confound temperature with tank architecture (e.g., depth or corners) which may serve as to attract or repel a fish, and thereby confound results. Fish position within the ring was recorded by overhead GoPro cameras and then footage was analyzed using ethovision (Noldus tm).

Temeprature Preference Device, Numbers indicate specific ‘wedges’ used for determining fish location in subsequent video analysis.

Temeprature Preference Device, Food dye was added to visulize the stability of the water currents.

We tested two species of Antarctic rockcod, T. bernacchii (n=29) and T. penellii (n =30). Fish were gathered from McMurdo sound and kept in the Crary Laboratory for between 2-5 days before being tested. Fish were fasted for the 24hr prior to beginning at TPA trial. Fish were placed in the coldest location in the ring and then given 20 minutes to acclimate to apparatus before data was collected, data during this period were not included in analysis. Trials then ran for 135 minutes. Each fish was trialed twice, with the fish transferred to a second TPA upon completing its first trial, to asses the repeatability of the temperature preference metric. After the second trial fish were weighed (units) and measured (units).

The fish’s location throughout the ring is then matched with the temperature of that location. A fish’s thermal preference is the MEDIAN temperature experienced by the individual during the trial. The median was chosen to reduce the effect of outliers as the fish transited the ring. We also measured the movement of the fish, counted as the number of transitions between neighboring wedges. Location of the fish in the ring was assessed 10 times per second.

Statistics were conducted in R using the package brms and images were modeled ggplot2 and tidybayes. We used a linear mixed effects model implemented in a Bayesian framework to determine the mean temperature preference for both species. We used stepwise model selection and selected the lowest WAIC model for further discussion. The species level temperature preference was calculated as the mean of the individual medians.

Results

First, well look at the simple mass and length data of our experimental animals.

Table 1: Demographics of Experimental Fish. N is the nubmer of fish from each species which were tested, each fish was tested twice. Other traits are reported as means and standard deviations of the raw data.
Species N= Mass (g) Length (mm) Condition Factor Obs. Temperature Preference (°C)
T.bernacchii 58 0.63 ± 0.08 41.29 ± 1.48 0.89 ± 0.05 0.08 ± 1.33
T.pennellii 60 0.72 ± 0.1 41.7 ± 1.57 0.99 ± 0.06 0.86 ± 1.34

Mass (A) and Condition Factor (B) of Experimental Animals

We calculated Fulton’s Condition Factor using a the fishes mass and length. T. pennellii is both heavier and has greater condition factor than T. bernacchii. Both traits can reflect energetic state, and therefore may influence a fish’s temperature preference (assuming that persisting at sub-zero temperatures is energetically burdensome).

We conducted a straightforward t-test to look for population differences.

mass.m1 <- lm(data = TPA.dat.1 %>% 
     filter(RUN_NUM == 1), ## Mass traits are the same across runs, so we limit to just one run
   MASS ~ SPECIES)

anova(mass.m1)
## Analysis of Variance Table
## 
## Response: MASS
##           Df  Sum Sq  Mean Sq F value    Pr(>F)    
## SPECIES    1 0.12738 0.127379  14.299 0.0003757 ***
## Residuals 57 0.50776 0.008908                      
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
### p < .001

cond.fac.m1 <- lm(data = TPA.dat.1 %>% 
                    filter(RUN_NUM == 1),
                  COND_FAC ~ SPECIES)

anova(cond.fac.m1)
## Analysis of Variance Table
## 
## Response: COND_FAC
##           Df  Sum Sq  Mean Sq F value    Pr(>F)    
## SPECIES    1 0.14444 0.144441  45.534 8.362e-09 ***
## Residuals 57 0.18082 0.003172                      
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Comparisons of Individual Fish Performance between the first and second trial.

Each facet of this plot is a specific species and each pair of points is an individual fish. The line traces how that fishes temperature preference changed between the first fish and the second trial.

We used the package ‘brms’ to fit Bayesian models to our data. Models estimating the temperature preference were built and selected using step-wise model selection, an abbreviated version is provided here.

Model_8 <- brm(
  family = gaussian,
  MEDIAN_PREF_S ~ 1 + SPECIES*MOVEMENT_S+RUN_NUM+ACC_TEMP+COND_FAC_S+
    (1|FISH_ID)+(1|CALIB_FAC), # random intercepts prescribed to each fish and to that day's specific temperature calibration data
  ## weakly regularizing priors
  prior = c(prior(normal(0,1), class = Intercept),
            prior(normal(0,1), class = b),
            prior(normal(0,10), class =sigma)),
  data = model.dat,
  warmup = 1000, ## Number of burn-in samples 
  iter = 4000, ## Total nober of model iterations
  chains =4,
  cores = 4,
  file = paste0("~/kenzillig.github.io/kenzillig.github.io/assets/rds//TPA_MEDIAN_M10.4.2.1.rds")
)
#conditional_effects(Model_8)

#summary(Model_8)
Model_8 <- add_criterion(Model_8, c("waic","loo")) ## calcualte the WAIC and LOO for this model
## Compare the WAIC scores of each model.
w <- loo_compare(Model_1, Model_2, Model_6, Model_3, Model_4,Model_5, Model_7, Model_8, criterion = "waic")
plot_model_waic(w)

Model Selection Plot

The above is just a plot of Widely Applicable Information Criterion (WAIC) for several statsical models attempting to determine the relationship between temperature preference and fish species. It identifies two models (10.4 and 10.1) as having the lowest. We will be using model 10.4 for further investigation.

Models including fish mass, temperature of in-lab acclimation, and interactions between trial number and species were tested but were not found to increase model fit, so these predictors were discarded. The final model included predictor variables an interaction between fish movement within the chamber, fish species, trial number and condition factor.

Random intercepts for each individual fish (repeated measures) and the TPA’s particular temperature calibration (which fluctuated slightly day to day) were included. These random effects respectively accounted for 37% and 27% of the residual variation.

We used the package emmeans to estimate the Temperature Preference of each species during both the first or second trial.

Temperature Preference of Two Species of Antarctic Rockcod

It appears that T. Pennellii have a higher temperature preference than T. bernacchii. Additionally, both species exhibit temperature preferences greater than their normal environment. It is also evident that the there is little effect of trial run on the temperature preference of either species of Antarctic rock cod.

We can use the posterior distributions to evaluate whether the two species exhibit the same temperature preference.

Pairwise Treatment Comparisons

Based upon these treatment contrasts we don’t find an effect of trial number. The top and bottom contrasts (which compare the performance of a single species in the first or second trial) heavily overlap zero, indicating that differences between runs is non-significant. We do see support for differences between the two species of rock cod. The point-intervals estimate the mean difference between treatments (point) along with the 89% (thin) and 50% (thick) credible intervals. If the point interval overlaps zero, we conclude that the treatments are not significantly different.

In this plot both of a fish’s two trials are represented, each by a black dot. The red point intervals capture the model estimated mean for each species along with the 89% (thin) and 50% (thick) credible intervals. The Red horizontal line indicates the highest temperature available within the temperature preference apparatus while the blue dashed line represents the coldest temperature available. The dotted line at -2°C indicates the typical temperature of the Ross Sea.

Table 2: Model Estimated Temperature Preference (mean and standard error)
Species Temperature Preference (°C)
T.bernacchii -0.05 ± 0.45
T.pennellii 1.05 ± 0.44

Qualitatively, we find that the number of fish selecting the coldest region of the TPA increases in the second trial, with more (n=7) T. bernaccii selecting the coldest region vs. T. pennellii (n= 4). There was one individual T. pennellii which selected the hottest wedge of the TPA (~5.6°C).

Conclusion

Both species exhibited temperature preferences above the normal background temperature of McMurdo Sound, consistent with our hypothesis of fish selecting warmer temperature to mitigate energetic burden. Additionally, we did find a slight, but significant negative effect of condition factor, indicating that fish with higher condition factor preferred colder temperatures. This results further supports or hypothesis of fish preferring warmer temperatures to reduce energetic burden.

Warmer thermal preferences may indicate that Antarctic species will respond positively to predicted warming in McMurdo Sound, especially if increased warming leads to reduced energetic burdens. However, temperature preference is a singular, organisms-specific trait, whether it is predictive of a species responses to broader ecological changes and extend physiological acclimation remains to be determined.

Potential for Field Deployment

Our design of the TPA used primarily recirculating water source and aquarium heaters. Prior iterations required flow-through supply of cold, medium and hot water sources. Our modification on the design allowed us to deploy the TPA at McMurdo Sound and could be used to measure temperature preference in field locations around the world. Being able to determine temperature preference of fish in the wild, or near-wild, can allow new questions about the interaction of ecology and temperature preference to be tested.

Acknowledgements

A huge thank you to Andrew Naslund, Amanda Frazier, Dr. Milica Mandic, and Dr. Anne Todgham. Additionally this work could not have been completed without the support of the McMurdo Station Crary Lab Support Staff who addressed the multitude of challenges associated with conducting work in the Antarctic. Thank you to Steve Rupp, Rob Robbins, and Mandy Frazier who braved the frigid waters to collect our research specimens. Thank you to Liz Kauffman and Curt LaBombard at the Crary Laboratory. Finally, thank you to the fish of McMurdo sound without whose sacrifice we would be unable to conduct this work.

Return to Main Webpage

or

My other ICBF Poster