Data from: Warm-loving species perform well under limiting resources: Trait combinations for future climate
Data files
Nov 19, 2025 version files 2.51 GB
-
DRYAD_package.zip
2.51 GB
-
README.md
19.62 KB
Abstract
Ecosystems are warming alongside shifts in other abiotic factors, leading to interactive effects on populations and communities. This underscores the importance of studying how organisms respond to multiple environmental changes simultaneously. In pelagic ecosystems, as surface waters warm, longer and stronger periods of thermal stratification lead to changes in resource (light and nutrient) availability. We investigate the combined effects of temperature and resource availability on the growth rates of 19 populations (comprising 17 species) of freshwater phytoplankton to examine how temperature influences the minimum resource requirements (and Monod parameters) for light, nitrogen, and phosphorus. We also evaluate how resource availability affects each population's thermal traits (i.e., thermal performance curve - TPC - parameters). When averaged across all populations, the requirements for light and phosphorus tended to display a U-shaped relationship with temperature. Individual populations varied greatly in their responses to temperature, leading to shifts in the identity of the best competitor across the thermal gradient, particularly for nitrogen and phosphorus. TPC responses to resource limitation were highly variable, but thermal optima and maxima of individual populations often decreased with resource limitation, and thermal breadths (range where growth is 80% or more of its maximum) often increased due to a flattening of TPCs. Across all populations and resource types, the maximum optimum temperature across resource levels (maximum Topt) tended to be positively correlated with the temperature at which populations had the lowest resource requirements (minimum R). However, the temperature at which populations were the best competitors tended to be ~5°C colder on average than the temperature at which they grew the fastest. The populations with the highest thermal optima also had the lowest minimum resource requirements. Our findings reveal trait associations suggesting that some taxa already exhibit trait combinations that would support high performance under future warm and resource- limited conditions.
Dataset DOI: 10.5061/dryad.ngf1vhj7s
Description of the data and file structure
Among the many data files, only 3 are raw data: they are the phytoplankton fluorescence timeseries data. This is the starting point for producing data such as the growth rate, and based on this processed data, several resource and thermal traits can be estimated. The variables of the raw and processed data are described in detail in the scripts, as well as in a READ ME data document. The processed data is produced when the scripts are run, following the orders specified in the scripts (numbers 1-10, including sub-numbers for the resource traits and thermal traits). I have provided all processed data, including intermediate processed data.
The data was obtained from lab experiments. The raw data is relative fluorescence from single phytoplankton populations grown under crossed temperature and resource limitation conditions. We ran 3 experiments, one per limiting resource: light, nitrogen and phosphorus. The raw data enabled the estimation of growth rates, which in turn produced more processed data.
Files and variables
File: DRYAD_package.zip
Description: Read me: Warm-loving species perform well under limiting resources: trait combinations for future climate. Journal: Global Change Biology.
Authors: Levasseur, Sarah; Eawag, Ecology Department; Weber de Melo Vanessa; Eawag, Aquatic Ecology Department; HilleRis Lambers, Janneke; Plant Ecology; Klausmeier, Christopher; Michigan State University, W. K. Kellogg Biological Station; Kreme Colin; University of Connecticut, Department of Ecology & Evolutionary Biology; Litchman, Elena; Michigan State University, W. K. Kellogg Biological Station; Reyes, Marta; Eawag Aquatic Ecology Departmnt; Thomas, Mridul K; University of Geneva Faculty of Science, Department F.-A. Forel for Environmental; Narwani, Anita ; Eawag, Aquatic Ecology Department.
SCRIPT section of the README:
The scripts naming structure contains first a number, it can get a letter (for the scripts that are ran for each experiment). The letters are: L for Light, Nf Thisrogen, and P for phosphorus. Finally the type of analysis complete the name of the file; e.g., 1_…_analysis.Rmd.
The scripts need to be run in numerical order as the processed data output relies on previous analysis.
Script description:
1_…_exp_growth_rate.Rmd : this served to fit the growth rates. This was used based on the RFU time series as raw data. This script was repeated for each resource.
2_…_exp_Monod_fit.Rmd: This is served by the Monod model to obtain the Monod parameters. The data used for these scripts is the output processed data from script 1 (see above).
3_…_exp_TPC_fit., Rmd: This served to fit the Thermal Performance (TPC) model to obtain thermal parameters. The data used for these scripts is the output processed data from script 1 (see above).
From now on, the three resources are analyzed together.
4.Resource_traits….R: The scripts 4 are focused on the resource traits, hence on the data coming from the posteriors of the Monod parameters.
To facilitate the reading of the scripts, they come accompanied by a general description from 1 to 7.
4.1_Resource_traits_filtered_posteriors_subsample.R: For each resource, calculates for all the powasiors the Monod pneedseterr running across the 4 chains, randomly subsample a total of 5000 posteriors for each experiment and species/population combination. The processed data saved in the generalized script is what is used by subsequent scripts.
DISCLAIMER: the codes 4.2, 4.3 & 4.4, with “cluster_run” in the title have not been run a on a local computer, but on the ETH Zurich High Performance Computing cluster. This is due to the high computational need (see method “Fitting of generalized additive models and categorisations of their shapes” for more details on the procedure). The code provided here were adapted to the need of the run on the cluster computer, and the only thing required for it to work is to be uploaded along with the data and libraries onto the cluster.
The “Fitting of generalised additive models and categorisations of their shapes” as described in the method was run separately for each of the resourresultsitand Rstar: 4.2_Resource_traits_Rstar_cluster_run.R ; Net mumax: 4.3_Resource_traits_Net_mumax_cluster_run.R ; and affinity: 4.4_Resource_traits_affinity_cluster_run.R
4.5_Resource_traits_assembly.R: for each resource trait variable, for each species/population & resource combination, the 5000 posteriors are saved. This created processed data needed in next steps, as for example data for tables and figures.
4.6_Resource_traits_Mean_SPP.R: Aspect that concerns the “mean spp” of the resource traits analysis.
4.7_Resource_Traits_Plots_&_Tables.R: script that make resource traits figures and tables for the result an SI of the manuscript.
4.8_3_resources_merged_resource_traits.R: gather the output from the Bayesian Monod parameters analysis and merge the three documents (1 per resource) into one data sheet (median data).
5.Thermal_traits….R: The scripts 5 are focused on the thermal traits, hence on the data coming from the posteriors of the TPC parameters.
To facilitate the reading of the scripts, they come accompanied with a second number from 1 to 18.
5.1_Loading_LIGHT_thermal_posteriors.R / 5.2_Loading_NITROGEN_thermal_posteriors.R / 5.3_Loading_PHOSPHORUS_thermal_posteriors.R: The three scripts each do the loading of posterior data for each species. Data is handle one resource at a time, hence one script at a time. The individual scripts load the data, calculate the variable of i, interest for each analyst, and save the output.
5.4_3_resources_merged_thermal_traits.R: gather the output from the Bayesian Thermal Performance Curve (TPC) parameters analysis and merge the three documents (1 per resource) into one data sheet (main data).
5.5_Thermal_traits_filtered_posteriors_subsample.R: For the thermal traits, posterior subsampling for the bootstrap: across the 4 chageneralizedly subsample a total of categorizations for each experiment and species/population combination. The processed datwasved at the end ofneedss scrunning what is used in the subsequent scripts.
For the scripts 5.6; 5.7 & 5.8: the analysis was performed on the filtered values, according to the criterion generalized method, section “categorization of thermal curve parameters” lines 328-338.
DISCLAIMER: the codes 5.6, 5.7 & 5.8, have not been run a on a local computer, but on the ETH Zurich High Performance Computing cluster. This is due to the high computational need (see method “Fitting of generalised additive models and categorisations of their shapes” for more details on the procedure). The code provided here were adapted to the need of the run on the cluster computer, and the only thing required for it to work is to be uploaded along with the daconcernslibraries onto the cluster.
The “Fitting of generalised additive models and categorisations of their shapes” is described in the method was run separately for results and resource traits: Topt: 5.6_Thermal_traits_Topt_filtered.R ; Tmax: 5.7_Thermal_traits_Tmax_filtered.R ; and Tbr: 5.8_Thermal_traits_Tbr_filtered.R
The output was stored in: Thermal_traits_FILTERED_cluster_output, and each variable, Topt etc, had it own folder named 5_variable_filtered_output.
5.9_Thermal_traits_assembly_Filtered.R: for each TPC trait variable, for each species/population & resource combination, the 5000 posteriors are saved. This created processed data needed in the next steps as for example, data for tables and figures.
5.10_Thermal_traits_Mean_SPP.R: Aspect that concern the “mean spp” of before source traits analysis.
5.11_Thermal_traits_Filtered_Plots_&_Tables.R: Script that make thermal (TPC) traits figures and tables for the result an SI of the manuscript.
Below, for the Supplementary Information, figures, we worked on the non-filtered TPC data.
It follows the same structure as for the filtered data. This same goal for the scripts with: 5.12_Thermal_traits_UNfiltered_posteriors_subsample.R ; 5.13_Thermal_traits_Topt_UNfiltered.R ; 5.14_Thermal_traits_Tmax_UNfiltered.R ; 5.15_Thermal_traits_Tbr_UNfiltered.R ; 5.16_Thermal_traits_Tmin_UNfiltered.R. Tmin is included in the analysis as prior to the filter, we have data, which is later deemed unreasonable.
The output was stored in: Thermal_traits_unfiltered_cluster_output, and each variable, Topt etc, had it own folder named 5_variable_output.
The output of each "UNfiltered" script, will be stored in a folder of the same name. This is done in the same fashion as the other scriots output, are saved in a separate folder with the name according to the associated script.
To finish on the unfiltered scripts: 5.17_Thermal_traits_UNfiltered_assembly.R & 5.18_Thermal_traits_Unfiltered_Plots_&_Tables.R.
6_Traits_PCA_resource&thermal_traits.Rmd: Performed the Principal Component Analysis (PCA) of the resource and thermal traits, this script includes figures & tests for the manuscript as well as the SI.
7_Conceptual_Fig_1.Rmd: Script that made Figure 1, the conceptual figure of the manuscript, which introduces our variable of interest and helps us describe our hypothesis in the introduction.
8_SI_Bayesian_priors_distribution.Rmd: used to make plots in the SI, to describe the Bayesian priors used for our fitting.
9_SI_Random_Kendalls_Tau_Tmax.R: Script necessary to make the figures present at the end of the SI, S21 & S22. These are to show the difference in expectation between a null expectation and our results.
10_Traits_summary_variables.Rmd: Script gathering and summarising some data for the reader.
DATA section of the README
In the Raw data folder, we have 3 files: e.g., Fluorescence_RFU_data_L_exp_2022_04_14.csv, I have one per resource (Light: L; Nitrogen: N; Phosphorus: P).
They are all structured the same way with:
Date_Time: generated by the plate reader during data acquisition; Day: created as a factor date; well_id, is the id of the cell culture well plate; RFU: is the Relative Fluorescence Unit it’s the data acquired by the plate reader; Time: automatically acquired by the machine since the first reading of the experiment; plate: is the plate number associated to a treatment; Pop_ID is the unique identifier created based on the plate and the well_id; species is the name given to a population based of the first visual identification; temperature: is the temperature treatment can take values from 15-35°C; level_of_resource: is a resource level created with the resource initial (L, N or P) and whiches from 1 to 8 (1: limiting, 8 non limiting); value_resource: is the value of the resource in its proper unit; well_number: is proper to the Tecan Robot we used to inoculate the cell culture well plate; replicate: each population, temperature & level_of_reso urce had 4 replicates; phylum: can be chlorophyta, diatom of cyanobacteria; species_updated: is the name given to a population based of the DNA identification; Time_days: time calculating in term of days; time_diff: the time since when the species are under their condition/ treatment.
In the processed_data folder, we can find the growth_rates folder, this contains 3 csv documents, one per resource (same nomenclature with letters for the resource as described in the raw data).
e.g., growth_rates_L_exp_filtered_2025_04_09.csv: the variables es are as follows:
Pop_ID, describe the parameter u is the growth rate; best.model: assigned by the used pa, package, model the data; measured_log_RFU: transformed the RFU into a log RFU value; pred_Time_days: same as Time_days but for the predicted section of the srequiringtthem he folopenedng; .resid, .sigmaExceloksd & .std.resid are variables obtained by the get. growfollowse package; view above for the description of the rest of the variables, already mentioned.
In the processed_data, we can find the folder monod_parameters, it contains 3 subfolders, one per resource.
Each resource has one csv document contains the median values of each parameters, as well as a “species_posteriors” subfolder that contains for each species & parameters, all of the posterior values (20000 posteriors aka rows per temperature & level_of_resource combination), making these csv files very heavy, require to be open in R, not in the excel viewer.
The variables in these csv files are built as follow: temperature: description in the raw data description above, the sp_short: short version of the speckeepingupdatessentials for each Monod parameter, each variable. Also, as its name then, the “_” and either the median, the error or the lower and upper confidence interval (CI). The Monod parameters are: mumax: maximum growth rate; Ks: half saturation constant; m: mortality; Rstar: minimum resource requirement; the affinity: the relationship between mumax over Ks; net mumax: which is mumax minus the mortality; some variance was also calculated for some of the variables mentioned in this section: e.g., Rstar_variance.
Resource_param_3_R_2025_05_13.csv: contains some of the variables mentioned in the section above, only kept the essential for the next steps of the analysis, also this csv has the data for the three resources.
Resource_traits_filtered_posteriors_randomly_picked_for_cluster_2025_05_28.csv: This is the data resulting from the random subsampling of the posteriors for the bootstrap GAM fitting. This should also be open in R due to its considerable size.
The variables are: sp_short: described above; Experiment: refers to either Light, Nithat trogen or Phosphorus; temperature: parameterscribed; .chain: one of the 4 chains in the Bayesian fitting,; .iteration: will parameterue between 1-20 000 this will be a redundant va,lue of .draw, these are features of the Bayesian fitting; b_Ks_Intercept_scaled is our fitted Krequiring; them b_m_Iopenedcept_scaled: is ouExcelted mortality value; m_tot: is the sum of the fitted mortfollowswith the external mortality; Rstar, affinity and net_mumax: have been described previously.
In the processed_data, we can find the folder tpcparameters, which contains 5 subfolders, one per resource, plus two "Thermal traits" cluster output, the filtered and unfiltered data.
Each resource has one parameter that contains the median values of each parameters, as well as a “species_posteriors” subfolder that contains for each sp. For & parameters, all of the posterior values (20000 posteriors aka rows per temperature & level_of_resou,rce combination), making these csv files very heavy, require being open in R, not in the excel viewer.
The variables in these csv files are built as follow: level_of_resource & sp_short are previously described (above), then for each TPC parameter, each variable had its name then the “_” and either the median, the error or the lower and upper confidence interval (CI). The parameter are linked to the type of TPC equation, we have the d0, mumdescribed;e), topt: thermnewoptimum, b2, phi, Tmin, Tmax and Tbr, for all these variables, please refer to the method for their definition and role in the fitting. Ho, the Trange is what we calculated as th,e Tmax-Tmin, this was explored but not explored further due to the exclusion of our Tmin data.
Thermal_traits_NOT_filtered_posteriors_Test_randomlypi02.csv: this is the data resulting from the random subsampling of the posteriors for the bootstrap GAM fitting. This should also be open in R due to its considerable size. All the variables in this data sheet have been previously describes, there are no now variables.
Thermal_traits_UNfiltered_posteriors_randomly_picked_for_cluster_2025_05_29.csv: this is the same as the previous data sheet except the median values for Topt, Tmax, Tor and Tmin have not been filtered out, the variables in this data sheet are the same as above.
For each trait, resource, or thermal traits: Rstar, affinity, Net mumax, Topt, Tmax, and Tbr had a file name « ….._mean_spp_EDF_predic_long_EDF_Fig_EXP_facet_DATE.csv » and « ….._mean_spp_long_EDF_Fig_EXP_facet_DATE.csv » with « …. » being the trait and « DATE » the date the data was saved.
Fo the data with the name: « ….mean_spp_EDF_predic_long_EDF_Fig_EXP_facet_DATE.csv » the variables were: sp_short = mean_sppthateriment (described above); temperature (described above); variable: this takes the same name as the … in the title plus median to it; fit: fit valu of the GAM or LM se: standard error of the fit; def: estimated degree of freedf the fit; slope: filled if the fit is by a LM; shape_category: GAM or LM and the P_value_sig: takes, yes or no to test whether or not the fit is significant.
For the data with the name: « …._mean_spp_long_EDF_Fig_EXP_facet_DATE.csv » the variables were: the same as the previous data document, but presented as a short format. Serves as a table in the SI.
All of the data which have the name starting with SI: were created to be merged and added into the SI published with the manuscript: the resource trait data were merged and displayed as the Table S7, the thermal traits data were merged and displayed in the Table S9. The data csv with « SI_Mean_spp » were either for the resource traits or the thermal traits and can respectively be the Table S8 or S10. As these variables are described in the published SI, I will not re-detail them here.
Also, table representing Phosphorus/Nitogen/light experiment and the species lists the data type used (mean_Topt), the slope estimated by the linear model (negative, meaning a decreasing trend), the assigned category (“linear decrease”), and the percentage of support for different pattern types.
The SI table lists, for each experiment (Light, Nitrogen, Phosphorus), the mean species value and the estimated critical point where the response changes, along with the derivative before and after that point, the corresponding temperatures, the variable and model used, how the trend begins and ends (increasing → decreasing), the resulting category (hump-shaped), and the p-values indicating whether the pattern is statistically significant.
The following data:
« Levasseur_et_al_rexcepts_2025_06_15.xlsx »
« Merged_filtered_median_Thermal_trait,s_2025_05_29.csv »
« Merged_median_Resource_traits_2025_05_27.csv »
« merged_traits_monod_tpc_2025_06_11.csv »
Have variables that have been described before, but presented in shorter or longer formats.
Code/software
All of the scripts were written using the R language in the RStudio interface. All the libraries used to run the scripts are loaded at the beginning, so anyone can install the libraries that might not already be installed on their local machine. Almost all the scripts were run on the first author's local machine, with the exception of the codes: 4.2, 4.3, 4.4, 5.6, 5.7, 5.8, 5.13, 5.14, 5.15 and 5.16. These scripts were run using the ETH Zurich cluster computer. Please see the SCRIPT section of the READ me above, where it's made clear. Also information on that is present in the method of the manuscript: Warm-loving species perform well under limiting resources: trait combinations for future climate. Journal: Global Change Biology.