Bioenergetics modelling to analyse and predict the joint effects of multiple stressors: Meta-analysis and model corroboration
Benoit Goussen, Cecilie Rendal, David Sheffield, Emma Butler, Oliver R. Price, Roman Ashauer
BBioenergetics modelling to analyse and predict the joint effects ofmultiple stressors: Meta-analysis and model corroboration
Benoit Goussen a,b, ⁎ , Cecilie Rendal b , David Shef fi eld b , Emma Butler b , Oliver R. Price b,1 , Roman Ashauer a,2 a Environment Department, University of York, Heslington, York YO10 5DD, UK b Safety and Environmental Assurance Centre, Colworth Science Park, Unilever, Sharnbrook, Bedfordshire, UK
H I G H L I G H T S • Joint effects of natural and chemicalstressors impact many ecological appli-cations. • Prospective ecological risk assessmentrequires an understanding of these ef-fects. • We lack full understanding of how or-ganisms react to a combination ofstressors. • We used a Dynamic Energy Budgetmodel to predict the effect of multiplestressors. • We successfully showed the plausibilityof our approach and validated ourmodel. G R A P H I C A L A B S T R A C T a b s t r a c ta r t i c l e i n f o
Article history:
Received 25 January 2020Received in revised form 30 July 2020Accepted 3 August 2020Available online 4 August 2020Editor: Ralf B Schaefer
Keywords:
Bioenergetics modellingMultiple stressorsMeta-analysisRisk assessmentToxicokinetics toxycodynamicsCase-studyDynamic Energy Budget
Understanding the consequences of the combined effects of multiple stressors — including stress from man-madechemicals — is important for conservation management, the ecological risk assessment of chemicals, and manyother ecological applications. Our current ability to predict and analyse the joint effects of multiple stressors is insuf- fi cient to make the prospective risk assessment of chemicals more ecologically relevant because we lack a full under-standing of how organisms respond to stress factors alone and in combination. Here, we describe a Dynamic EnergyBudget (DEB) based bioenergetics model that predicts the potential effects of single or multiple natural and chemicalstressors on life history traits. We demonstrate the plausibility of the model using a meta-analysis of 128 existingstudies on freshwater invertebrates. We then validate our model by comparing its predictions for a combinationof three stressors (i.e. chemical, temperature, and food availability) with new, independent experimental data onlife history traits in the daphnid Ceriodaphnia dubia . We found that the model predictions are in agreement with ob-served growth curves and reproductive traits. To the best of our knowledge, this is the fi rst time that the combinedeffects of three stress factors on life history traits observed in laboratory studies have been predicted successfully ininvertebrates. We suggest that a re-analysis of existing studies on multiple stressors within the modelling frame-work outlined here will provide a robust null model for identifying stressor interactions, and expect that a better un-derstanding of the underlying mechanisms will arise from these new analyses. Bioenergetics modelling could beapplied more broadly to support environmental management decision making.© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Science of the Total Environment 749 (2020) 141509 ⁎ Corresponding author at: ibacon GmbH, Arheilger Weg 17, 64380 Roßdorf, Germany.
E-mail address: [email protected] (B. Goussen). Current address: RB, Dansom Lane, Hull, HU8 7DS, United Kingdom. Current address: Syngenta Crop Protection AG, Basel, Switzerland.https://doi.org/10.1016/j.scitotenv.2020.1415090048-9697/© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Contents lists available at ScienceDirect
Science of the Total Environment . Introduction
Multiple stressors drive environmental change on a global scale, in-cluding the effects of the loss of biodiversity on ecosystem functioning(Vorosmarty et al., 2010; Rockstrom et al., 2009; Baert et al., 2016). Spe-cies are exposed to a multitude of natural and anthropogenic stressors,including man-made chemicals (Schwarzenbach, 2006; Schäfer et al.,2016). When factors in the ambient environment of a given organismfall outside their speci fi c tolerance range, they become stressors. Conse-quently, organisms may alter their life history traits in response to thesenatural environmental stressors (e.g. low dissolved oxygen concentra-tions, food limitations, non-optimal temperatures, predation threats,or parasitism) (Boersma and Vijverberg, 1995; Hall, 1964; Homer andWaller, 1983; Seidl et al., 2005a; Connolly et al., 2004; Filho et al.,2011), including potentiation of effects or physiological mitigation(Hanazato, 1996; Penttinen and Holopainen, 1995; Seidl et al., 2005b;Coors and De Meester, 2008). This change in response to naturalstressors has the potential to affect how organisms respond to chemicaltoxicants (Holmstrup et al., 2010; Laskowski et al., 2010).Understanding the joint action of stressors can be a help inunravelling the combined adverse effects of multiple stressors that af-fect many ecosystems, and further, can be a contributing factor in un-derstanding the qualities that new chemicals should have to exhibithigh functional performance but limited effects on non-target environ-mental species (Schäfer and Piggott, 2018).Prospective risk assessments of chemicals are widely used as a toolto help prevent potential risks to ecosystems and to develop more sus-tainable products. Current environmental risk assessment approachesare typically aimed at assessing single chemicals, and although mecha-nistic understanding of the biological and chemical processes that gov-ern adverse effects is often incorporated into assessments, mostassessments still rely on single chemical and single species interactions.A key challenge is to increase the ecological relevance of the prospectiverisk assessment of chemicals by tackling the effects of multiple stressorson multiple species (Scienti fi c Committee on Health and EnvironmentalRisks SCHER, 2013; van den Brink, 2008; Relyea and Hoverman, 2006;Rohr et al., 2006). Speci fi cally, our ability to predict the impact of multi-ple simultaneous environmental and chemical stressors is hampered bya lack of understanding of how these factors interact to alter life historytraits (Holmstrup et al., 2010; Galic et al., 2018; Crain et al., 2008).The current prospective environmental risk assessments (ERAs) ofman-made chemicals has been developed to be protective rather thanpredictive and is widely recognised to lack ecological realism (Forbesand Calow, 2013; Goussen et al., 2016; Forbes et al., 2009; Ashaueret al., 2011). This is in part rooted in the way standard endpoint dataare derived. In ERAs, organisms are tested under laboratory experi-ments at strictly-controlled — and often optimal — conditions (e.g. adlibitum food intake, optimal temperature, high levels of dissolved oxy-gen) in order to reduce the effects of confounding factors and isolatethe effect of the chemical. Even when studies have assessed the com-bined effects of natural and chemical stressors (Holmstrup et al., 2010;Chandini, 1988; Hamda et al., 2014; Cedergreen et al., 2016; DeConinck et al., 2013; Pieters and Liess, 2006) they did not result in a gen-eral mechanistic theory or predictive models. This is the consequence ofthe fact that ERAs must simplify some of the environmental complexityto be applied on a daily basis. Consequently, only limited efforts havebeen made to develop tools like mechanistic and predictive models ac-counting for multiple stressors. However, we need such tools, becauseaccounting for the effects of multiple stressors — both chemical and nat-ural — and allowing extrapolating to untested species or combinations ofstressors and stress intensities would improve the ecological relevanceof ERAs.When the observations of multiple-stressor experiments deviatefrom the joint effect predicted by a given null model (i.e., the referencemodel), researchers typically infer that an interaction — speci fi cally syn-ergism or antagonism — must have occurred (see Holmstrup et al., 2010; Laskowski et al., 2010; Crain et al., 2008 for meta-analyses). In thesecases, the terms synergy and antagonism are used as black box termsto describe an unknown effect that has caused a greater or smaller re-sponse than expected. However, in many cases, the underlying nullmodels do not re fl ect the true system, even in the absence of interac-tions because they are poorly de fi ned, and lack a mechanistic basis(Schäfer and Piggott, 2018). Thus, it is possible that studies reportingsynergy and antagonism in the face of the multiple stressors could be re- fi ned to con fi rm and better understand the reported effects.Furthermore, while the null models that are typically applied at or-ganism level may consider the statistical distribution of sensitivitiesand the observed stressor-effect relationships, they rarely model theunderlying processes (e.g. physiology), even though the central role oforganism level-responses in stress ecology has long been understood(Maltby, 1999). Thus, we argue that a step change is needed in the anal-ysis and prediction of the effects of multiple stressors. Here we proposea generalised, mechanistic framework for predicting the joint effects ofstressors on the life history traits of a given individual organism. Bymodelling stressor-induced changes in energy allocation, we proposeto predict the impact of a multiple-stress environment on an organism'sgrowth and reproduction, and therefore on population dynamics.Organismal development is a thermodynamic process. Organismsrequire energy to maintain their basal metabolism, move, create bodymass, reach sexual maturation, and reproduce (Dajoz, 2006). Bioener-getic models, such as the Dynamic Energy Budget (DEB) model(Kooijman, 2010) are coherent models of organism physiology, and in-tegrate various physiological processes using a consistent, quantitative,and rigorous framework. DEB models describe how energy is assimi-lated from food and used for the organism's maintenance, growth, mat-uration, and reproduction following the κ rule (van der Meer, 2006),where a fraction of the energy ( κ ) is allocated to maintaining homeosta-sis (i.e. somatic maintenance) and growth processes. The remaining en-ergy (1 − κ ) is allocated to processes associated with maturation andreproduction. DEB models can pinpoint which life history processesare perturbed by anthropogenic (e.g. chemical) or natural stressors(Freitas et al., 2011; Goussen et al., 2015; Jager and Zimmer, 2012;Ashauer and Jager, 2018), or combinations of these stressors (Pieterset al., 2006; Jager et al., 2010), where the resulting impacts on growth,maturation, and reproduction are easily translated into changes at thepopulation level (Beaudouin et al., 2015). In that way, the results fromhighly standardised laboratory toxicity tests can be extrapolated to eco-logically relevant, fi eld conditions (David et al., 2018).We can now see that bioenergetics modelling with DEB modelscould offer a coherent framework to understand and predict effects ofmultiple stressors acting individually or jointly on organisms and so in-crease the ecological relevance of ERAs. What we do not yet know ishow well this idea works. Therefore, the aims of the present studywere: i) to develop a bioenergetics-based modelling framework to pre-dict the effects of single or multiple natural and chemical stressors onlife history traits and use the extracted and normalised data from previ-ously published studies to assess the model's plausibility, and ii) to cor-roborate the model by comparing its predictions with novelexperimental data.
2. Materials and methods
As freshwater invertebrates have a long history as model test organ-isms, the relatively large amount of data available in the published liter-ature allows us to interrogate of the role that the interaction ofenvironmental and chemical stressors plays in altering the life historytraits. To this end we performed a meta-analysis of the scienti fi c litera-ture (see details in Supporting Information (SI), Table S1).For each study collated from the literature, a control (or optimum)condition was chosen, which correspond to the condition with the B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 trongest life-history trait values (e.g. largest body size, most reproduc-tive output). Typically, these are conditions with ad libitum food avail-ability, high oxygen levels, absence of extremes of temperature, andno addition of toxic compounds. Data extracted from each of the studieswere scaled by the optimal value (e.g. largest body size or maximum re-productive output) for the corresponding study to facilitate comparison.
To assess the ability of DEB models to predict effects of multiplestressors, we used a DEB-based model for
Daphnia magna to simulatehow single stressors and multi-stressor combinations would affect thelife-history traits of organisms. We then compared these simulationsto the meta-analysis results. We simulated the effect of single stressorsusing the DEBkiss model, which is a simpli fi D . magna using a model pre-viously established parameter values (Jager, 2017) (see SI Sections 2.1.4and following).For each individual stressor of interest, the relevant DEBkiss modelparameter values (e.g. parameters linked to food for the food levelstressor) and variables were varied in a range that corresponded tothe range of the data collected in the meta-analysis. Simulations wererun for 30 days, which was suf fi cient time to reach the maximum (as-ymptotic) body length under optimal conditions. The outputs analysedwere (i) the body length at the end of the simulation under a wide rangeof sub-optimal conditions (i.e. stress) divided by the simulated maximalbody length the organism can reach under optimal conditions (i.e. nostress) and (ii) the cumulative reproduction (i.e. offspring per female)at the end of the simulation under a wide range of sub-optimal condi-tions (i.e. stress) divided by maximal cumulative reproduction (i.e. off-spring per female under no stress conditions). This procedure allowedus to obtain normalised endpoints that can be compared across differentstudies of different species in the meta-analysis. The simulation resultsand the data from the meta-analysis are then plotted as overlays acrossthe range of the individual stressor (Fig. 1).The effect of the combination of two stressors was also simulatedwith DEBkiss using the same strategy as for the effect of each individualstressor of interest and parameters for D . magna , with the exception thatMonte Carlo simulations were performed using the ranges of the modelparameters or input variables (e.g., food availability, temperature, pres-ence of predator kairomones, cadmium levels) that corresponded tothose collected for the two stressors of interest. The outputs of thesesimulations were used to plot raster and contour plot overlays (Fig. 2). To test and corroborate our multiple-stressor response model, an ex-tensive set of laboratory experiments were performed using the cladoc-eran
Ceriodaphnia dubia . A brief summary of the design is given here;detailed information are available in the Supporting Information. Specif-ically, we developed a high-throughput experimental method for mea-surements of growth, reproduction, and mortality in
Ceriodaphniadubia . The short life cycle of C . dubia allows for chronic toxicity assaysto be performed in 7 –
10 days (compared to
Daphnia magna at21 days). To assess the effects of food availability and temperature ongrowth and reproduction, we used a factorial study design in which C . dubia were exposed to one of four different temperatures (15, 20, 25,and 30 °C) and one of three different concentrations of Chlorella vulgaris (0.5 × 10 , 4 × 10 , and 15 × 10 cells mL − ). In addition a commercialfood supplement was provided to all batches (GP 5-50), which containsmarine fi sh, krill (23%), fi sh roe, soy lecithin, yeast autolysate, micro-algae, fi sh gelatine, squid meal, hydrogenated vegetable fat, vitamins and minerals, antioxidants. In spite of the conventional duration of7 –
10 days for the C . dubia toxicity test, our experiment was conductedfor 21 days, to ensure that asymptotic growth and reproduction werereached for organisms cultured under optimal conditions.Two scenarios were investigated in combination with the effects ofphenol on growth and reproduction: (A) one in which organismswere exposed to phenol under conditions of optimal temperature andfood availability and (B) one in which they were exposed to phenolunder sub-optimal temperature and reduced food availability. In sce-nario A, organisms were exposed to phenol (nominal concentrationsof 0.0, 0.32, 0.56, 1.0, 1.8 and 3.2 mg L − ) under standard (optimal) cul-turing conditions (25 °C and a feed level of 3.4 × 10 cells mL − C . vulgaris + GP 5-50). In scenario B, organisms were exposed to thesame concentrations of phenol as Test A, but at a slightly lower temper-ature (20 °C) and a signi fi cantly reduced concentration of C . vulgaris (1 × 10 cells mL − + GP 5-50). A bioenergetic model based on the DEB theory was calibrated andused to perform independent simulations. A full description is availablein the Supporting Information Sections 2.3 and 2.4. Brie fl y, a DEBkissmodel was calibrated in two separate steps. In the fi rst one, the calibra-tion was performed on growth and reproduction data obtained from anexperiment with food and temperature stress but without chemicalstress. In the second step, the toxicity parameters were calibrated ondata from an experiment with chemical stress without food or temper-ature stress. Without further calibration, the DEBkiss model was thenused to predict the growth and reproduction over time for C . dubia ex-posed to either optimal food and temperature conditions and phenol(i.e. scenario A) or to a non-optimal temperature, food level, and phenolconcentration (scenario B).
3. Results and discussion
The effects of temperature, food level and oxygen on growth and re-production in different freshwater invertebrate species are too sparseand too variable to extract any information on the differences betweenthe species (Fig. 1). Limited information can be extracted from the rela-tionship between the temperature and growth and reproduction(Fig. 1A, B) because the data points are all in the 70% to 100% range.The relationship is more de fi ned for food levels (Fig. 1C, D) and oxygen(Fig. 1E, F), where a pattern can be seen in the empirical data. Overlaidwith this empirical data is the bioenergetics (DEB) model prediction foreffects of temperature, food level and oxygen on growth and reproduc-tion in D . magna . The model line goes through the data in all panels ofFig. 1, which tells us that the model for D . magna - scaled in the sameway as the experimental data in order to facilitate comparison - is con-sistent with empirical data for a total of 15 different species, althoughthe empirical data is very patchy and variable. Thus, to improve theoryand model, we need more data to better cover the whole range ofstressor values, especially for temperature, and we need less variabledata. The in fl uence of temperature on metabolic rate has been well docu-mented (Kooijman, 2010). Our meta-analysis included 10 studies withdata on the effects of temperature stress on the maximal length and cu-mulative reproduction for 7 species (Fig. 1A, B) and revealed that datawere primarily collected well within the temperature tolerance rangesof the species studied.Few data points fall outside the boundaries of the tolerance ranges.This is most likely because researches avoid the experimental B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 omplications expected at temperature extremes. At low temperatureslong experimentation times are required, and at high temperatures,high mortality rates are expected.Our analysis also shows that the temperature relationship used inour model for D . magna is consistent with the data on temperatureand life-history traits obtained for 7 species. In the DEB model simu-lation of reproduction is compromised by even a small deviationfrom the optimal temperature (Fig. 1B). In contrast to reproduction,growth is maintained over a wider temperature range, followed by asteep reduction at the boundary of temperature tolerance (Fig. 1A).Additional experiments should consider adding data points that areclose to the maximum and minimum temperature tolerance valuesof each species because these data are lacking (Fig. 1A, B). Thiswould facilitate better model calibration and extrapolation to awider range of temperatures. The meta-analysis of the literature yielded 179 data points from 13species related to effect of food density on growth and reproduction.We found that a slight decrease in food density from the optimumfood level induces only a small effect on maximal body length and cu-mulative reproduction (Fig. 1C, D). In contrast, low food density leadsto a large impact on both growth and reproduction (Fig. 1C, D). The de-cline in relative reproduction is more pronounced and appears alreadyat higher food levels than the decline in body length (Fig. 1C, D). Thatsuggest that a reduction in food availability affects reproduction beforeit affects growth and that this effect is more pronounced with decreas-ing food levels. This pattern from our meta-analysis is a good matchwith the results we obtained using our DEB model for D . magna . Al-though the effects of food availability on growth and reproduction aregenerally well captured by the DEB model, the large degree of variability Fig. 1.
Effect of environmental stressors on body length and reproduction of freshwater invertebrates. Data from the meta-analysis are presented by points. Black lines represent DynamicEnergy Budget (DEB) simulations of
Daphnia magna responses. Body length and reproduction data are presented relative to the values of the optimum condition achieved under non-stressed conditions (see meta-analysis methods for de fi nition).4 B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 bserved for reproduction data across food levels (Fig. 1D) means thatlittle general knowledge can be gained here. This could imply that wecurrently do not know how reliable DEB model predictions at differentfood levels are or it could imply that life-history traits vary considerablyin response to food levels below ad libitum. Optimal food density for agiven species can be highly variable depending on the study. For in-stance, while D . magna has previously been considered to be fed adlibitum at food densities of 0.75 to 2 mg C L (Heugens et al., 2006;Pieters et al., 2005), it has been suggested that some of these densitiesmight actually be food-limiting (Augustine et al., 2011). The meta-analysis shows that the three freshwater invertebrate spe-cies where data were available are tolerant to reductions in dissolvedoxygen levels down to approximately 25% of the optimal concentration(Fig. 1E, F; where optimal is as de fi ned for the meta-analysis, typicallyranging between 7 and 9 mg L − ), below which the impact of dissolvedoxygen on both growth and reproduction increases rapidly. In otherwords, organisms can compensate for sub-optimal environmental con-ditions to a critical cut-off point, after which compensatory mechanisms become ineffective. Our modelling results are consistent with the pic-ture from the meta-analysis (Fig. 1E, F), which con fi rms that our simplemodelling approach accounts for the effect of low levels of dissolved ox-ygen on life history traits of freshwater invertebrates. However, it is im-portant to note that the onset of the effects appears at slightly higheroxygen levels in the empirical data compared to the model. Future stud-ies could explore if this mismatch could be eliminated by using the prin-ciple of the synthesising units (Kooijman, 2010) instead of the verysimple modelling approach used here. The synthesising unit conceptcould play an important role in increasing the coherence of the modelwith the empirical data as it is a generalised unit that processes incom-ing substrate (e.g. food, compound) in order to yield one or more prod-ucts (Muller et al., 2019). Food and temperature play a role in determining both the energyavailable to and the energy required by a given organism, via their ef-fects on metabolic rates. They are thereby among the main drivers of
Fig. 2.
Joint effects of environmental and/or chemical stressors on length and reproduction. Data from the meta-analysis are presented as points of increasing size (i.e. greater effects arerepresented by larger sizes). DEB simulations are represented by response surface and contour plots. Effects are presented as relative to the amount of growth (i.e. body length) orreproduction achieved under non-stressed environmental conditions. Stressor levels are represented as relative to an optimal condition (e.g. 0% represents an absence of foodcompared to the optimal condition, and 100% represents 100% of the optimal food availability). The chemical modelled for panels C and D is cadmium. See Supporting Information forraw data. 5
B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 evelopment. The meta-analysis indicates that these two stressorsjointly affect both growth and reproduction, although reproduction isimpacted to a greater degree (Fig. 2). A limitation is that the meta-analysis data are quite sparse outside of or close to the temperature tol-erance ranges of the studied organisms. Data are also scarce for interme-diate food levels at non-optimal temperatures. According to DEB theory,although growth remains relatively unchanged within the temperaturetolerance range, steep changes occur at the boundaries of this range. Incontrast, the negative effects of low food availability appear only at verylow food levels. Given the scarce and unevenly distributed meta-analysis data relating to body length (Fig. 2A), it is dif fi cult to assesswhether the pattern predicted by DEB is reasonable. The data on repro-duction (Fig. 2B) at deviations from optimal conditions indicates ashallower gradient towards impacts compared to the data for growth(Fig. 2A), but the impact starts at smaller deviations from optimum con-ditions. While this pattern is well captured by the DEB model, only a fewdata points are available where the predicted gradient is the steepest.The DEB theory assumes that reduced food availability reduces theamount of energy available, while temperature acts on metabolic rate.Both reproduction and growth appear to be impacted by the interac-tion between food and temperature. For reproduction the impact occursat small deviations from optimum conditions with a shallow gradient(Fig. 2B). For growth the impact occurs at stronger deviations fromthe optimum but with a steep gradient (Fig. 2A). Both food and toxic chemicals have the potential to impact the lifehistory traits of organisms. The example presented here is based onthe joint effect of cadmium and food availability on growth and repro-duction (Fig. 2) (Table S1). A limitation of the empirical meta-analysisdata is that data are scarce, especially data for high chemical concentra-tions and intermediate food levels. Our model simulates a limitation inthe intake of available energy (caused by a reduction in food availabil-ity) coupled with the physiological mode of action of the chemical,which in this case, impacts the assimilation of energy from food(Margerit et al., 2016; Baillieul et al., 2005). Other physiological modesof action can be modelled in similar fashion. The meta-analysis results(empirical data and model) highlight the interactive effects of the twostressors on both growth and reproduction, and shows that the jointstressors appear to exert stronger effects on reproduction than ongrowth (i.e. body length). This trend is particularly evident (in boththe empirical meta-analysis data and the model) as a result of increasingconcentrations of the chemical in question. More speci fi cally, while aslight increase in concentration appears to have almost no effect ongrowth, our model shows that it can reduce reproductive output by asmuch as 75%.Assessing this type of combination of potential stressors is of greatinterest for environmental risk assessments, as these stressors canalter the effects of each individual stressor. Furthermore, modellingthe interactions of these stressors allows for the extrapolation of chem-ical effects from an environment with a constant concentration of food(e.g. laboratory conditions) to an environment with fl uctuations infood availability (e.g. fi eld conditions). To better evaluate the ability of DEB theory to predict the combinedeffect of natural and chemical stressors on growth and reproduction,modelling should be performed in conjunction with physiological casestudies. The following case study outlines laboratory studies using thedaphnid C . dubia exposed to three stressors: temperature stress, foodstress, and chemical (i.e. phenol) stress. An in-depth analysis of thecase study experimental results can be found in the Supporting Infor-mation Section 4. We fi rst assessed the effects of combined food and temperaturestress, by testing three food levels at four temperatures. Statistical anal-ysis shows that temperature affected both growth and reproductivetraits in C . dubia , and speci fi cally impacted growth rate, fi nal length,and cumulative reproduction (Fig. 3). Reproductive traits, in particular,proved to be sensitive to temperature and were also highly impacted byreducing food levels at all temperatures. As the experimental tempera-ture increased, only the lowest food level resulted in additional impactson growth and reproductive traits. At the optimal temperature (25 °C),growth was impacted in a dose-dependent manner based on food level,while reproductive traits were only impacted at the lowest food level. Asimilar pattern was found at the highest temperature tested (30 °C), al-though the average fi nal length of C . dubia was smaller than that of or-ganisms exposed to their optimal temperature. The observed effects ofthe joint stressors (food density and temperature) on growth and repro-ductive traits in C . dubia were well captured by the DEB model with atendency to overpredict the effect on reproduction at higher tempera-ture and lower food level as shown in Fig. 3. The temperature and feed-ing parameters obtained are presented in Table S4. Next, chemical (i.e. phenol) stress was introduced into the experi-ment. We designed an experiment using optimal food levels and tem-perature and exposure to a range of phenol concentrations to infer thetoxicity related model parameter values for phenol. Exposure to phenolunder otherwise optimal conditions (25 °C, 3.4 × 10 cells mL − of C . vulgaris ) resulted in a dose-dependent response on both growth and re-production (Fig. S2). As was the case with food and temperature stress,the effects of phenol on C . dubia were well captured by the DEB model,with a no-effect concentration (NEC) of 0.142 mg L − , a concentrationtolerance ( c T ) of 1.524 mg L − , and a dominant rate constant ( k e ) of10 d − , which corresponds to an almost immediate internalisation ofthe compound in the organism (Fig. S2 and Table S3). Based on the above simulations of experimental results, all DEBkissmodel parameters (i.e. those related to basic organism physiology, tem-perature, food, and chemical toxicity) were fi xed, and a DEBkiss simula-tion was used for two scenarios exposed to a range of phenolconcentrations: (A) one based on the standard set of optimal conditionstypically used in a laboratory (e.g. 25 °C, 3.4 × 10 cells mL − of C . vulgaris ), and (B) the other scenario with mild temperature and foodstress (20 °C, 1 × 10 cells mL − C . vulgaris ), which exempli fi es extrapo-lation to a more environmentally realistic situation. The results of thissimulation were then compared with the experimental results. Asgrowth and reproduction were already impacted by the non-optimalenvironmental conditions, the experimental results showed that phenolhad a dose-dependent impact on both growth and reproduction, with astatistically greater impact under non-optimal conditions (see SI for sta-tistical analysis). Likewise, these patterns were accurately predicted bythe simulation (Fig. 4), thereby demonstrating the extrapolation fromstandard environmental conditions to non-standard conditions involv-ing complex combinations of stressors. It is important to note that survival was also impacted by both envi-ronmental conditions (food level and temperature) and phenol toxicity.
Ceriodaphnia dubia exposed to varying levels of food densities and tem-peratures exhibited high mortality at low temperatures for all food den-sities as well as at the lowest feed density for all temperatures(Table S5). In addition, mortality at day 21 was greater than 80% andas high as 100% at the highest temperature. Phenol also induced mortal-ity in a dose-dependent manner, where we observed a greater impactunder non-optimal conditions (e.g. up to 100% mortality at>1.8 mg L − ) than under optimal conditions (e.g., up to 66% mortality B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 t 3.2 mg L − ). In the current study, only the sublethal effects have beeninvestigated. The survival effect could be analysed by adapting a modellike the General Uni fi ed Threshold model of Survival (Jager et al., 2011;Jager and Ashauer, 2018) to account for a combination of environmentaland chemical stressors. Accounting for the impact of other environmental variables that canbecome stressors — such as pH, nitrates, or nitrites — can be important inenvironmental risk assessments, particularly if one is interested in thelife history traits of organisms in the “ impact zone ” (i.e. downstream)of wastewater discharge points (Finnegan et al., 2009). This is particu-larly true in many parts of the world where untreated wastewater isroutinely discharged into surface water. Stressors such as pH or nitratecan severely alter the life history traits of organisms (Soucek andDickinson, 2016; Rendal et al., 2012; Davis and Ozburn, 1969; Aliboneand Fair, 1981) and could be included in bioenergetics models in the same manner as chemical toxicants. Furthermore, although life historytraits can also be altered by biotic stressors that are widely encounteredin the natural environment (e.g. pressures owing to competition, para-sitism, or predation), these stressors are not usually accounted for inclassical ERA frameworks, and the scienti fi c understanding of how toquantitatively model their interactions with chemical stress is limited.In a comparable manner, the current understanding on how to quanti-tatively integrate environmental factors with high potential impact onthe biodiversity such as habitat destruction or hydro-morphological al-terations in a multiple-stressor environment requires further scienti fi cdevelopment. Such factors would require models focusing on a higherlevel of organisation such as individual based models. The presentstudy, although it focuses on individual level modelling, can serve as aguide to developing models for these additional stressors. Here, we show that the DEB model — which includes both environ-mental and chemical stressors — can accurately predict patterns in theeffects that result not only from exposure to a single stressor (Fig. 1), Fig. 3.
The joint effects of temperature and food density on growth and reproductive parameters in C . dubia . The upper graphs represent growth (i.e. length) over time, and the lowergraphs represent the cumulative reproductive outputs over time for four different experimental temperatures and three food densities (0.5 × 10 , 4 × 10 , and 15 × 10 cells ml − ).Our experimental data are represented by squares, triangles, and lozenge (one point per surviving organism at each time-point), while the dynamic energy budget model simulations( fi tted) are represented by the lines. 7 B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 ut also a mixture of two (Figs. 2 & 3) or three (Fig. 4) stressors. Ouranalysis also highlighted gaps in the existing experimental data, partic-ularly in the stressor ranges that result in steep model responses (e.g.borderline temperatures; Fig. 1A and B). Further work is required to as-sess the predictive capabilities of bioenergetics modelling for a widerrange of stressor combinations, especially those with three or morestress factors. Mapping the relative importance of stressors in real envi-ronments would also inform risk management, conservation, and miti-gation strategies.
4. Conclusions
This study has demonstrated that DEB theory can be used to repro-duce patterns in empirical data for the joint effects of multiple environ-mental and chemical stressors on the life history traits of freshwaterinvertebrates. However, very few studies address the impact of multiplestressors for environmental conditions outside of the neutral zone (optimal range). These types of data are critical for further model devel-opment, and to challenge and re fi ne the patterns predicted by DEB the-ory. Generation of more datasets similar to our study on C . dubia can beused to follow a pattern-oriented modelling strategy (Grimm andRailsback, 2011), as we did here.Despite the limitations in data availability, this investigation has suc-cessfully demonstrated the plausibility of using a bioenergetics-basedmodelling framework to predict the effects of multiple natural andchemical stressors on life history traits. The model was corroboratedby comparing its predictions for a combination of chemical, tempera-ture, and food-related stressors with independent experimental dataon C . dubia life history traits. To the best of our knowledge, this is the fi rst time the effects of the interaction of three stress factors on life his-tory traits have been successfully predicted.Bioenergetics modelling based on DEB theory could be a powerfultool for multiple-stressor research, chemical risk assessments, and envi-ronmental management, although further work is needed to establish if Fig. 4.
Model predictions of combined stress effects compared to independent experimental data. Predictions (lines) of the combined effects of phenol, temperature, and food restriction onbody length and reproduction in C . dubia using dynamic energy budget modelling. The points represent independent experimental data (i.e. model not fi tted; one point per survivingorganism at each time-point). The upper graphs represent growth (i.e. length) over time, and the lower graphs represent the cumulative reproductive outputs over time for fourphenol concentrations and both optimal (blue squares and lines) and suboptimal (green points and lines) food levels and temperature. (For interpretation of the references to colourin this fi gure legend, the reader is referred to the web version of this article.)8 B. Goussen et al. / Science of the Total Environment 749 (2020) 141509
EB based bioenergetics modelling of multiple stressors also works forother freshwater species and additional compounds, and ultimately be-yond freshwater invertebrates.We suggest that existing and future multiple-stressor experimentsshould be reanalysed within the modelling framework outlined hereto account for bioenergetics, as this reanalysis will provide a robustnull model for identifying stressor interactions such as synergy and an-tagonism. Lastly, we expect that, data from many existing studies thatconcluded synergetic or antagonistic interactions could be re-analysedusing the null model proposed here, and conclusions would probablyneed to be revised based on the improved mechanistic insight fromthe DEB model. Indeed, DEB based bioenergetics can already explainmany interactions between stressors and reveal mechanisms behindsynergism and antagonism effects.
CRediT authorship contribution statementBenoit Goussen:
Conceptualization, Methodology, Software, Formalanalysis, Investigation, Data curation, Writing - original draft, Writing -review & editing.
Cecilie Rendal:
Methodology, Formal analysis.
DavidShef fi eld: Formal analysis.
Emma Butler:
Investigation.
Oliver R.Price:
Conceptualization.
Roman Ashauer:
Conceptualization, Method-ology, Writing - original draft, Writing - review & editing.
Declaration of competing interest
The authors declare that they have no known competing fi nancialinterests or personal relationships that could have appeared to in fl u-ence the work reported in this paper. Acknowledgements
This study was funded by Unilever (MA-2014-00701). We thank ourcolleagues who helped with the meta-analysis literature prioritisationas well as two internal and two anonymous external reviewers thatgreatly helped improve this manuscript.
Appendix A. Supplementary data
Further details are available in the Supporting Information, includingall data used in the meta-analysis, a discussion of alternative tempera-ture relationships, case study experiments with C . dubia incl. statisticaldata analysis, details on model parameterisation, modelling of ingestionand assimilation rates and more meta-analysis results on food levelsand predation threat. Supplementary data to this article can be foundonline at https://doi.org/10.1016/j.scitotenv.2020.141509. References
Alibone, M.R., Fair, P., 1981. The effects of low pH on the respiration of Daphnia magnaStraus. Hydrobiologia 85, 185 – – fi sh, Danio rerio. Comp. Biochem. Physiol. A Mol.Integr. Physiol. 159, 275 – fi c scope for growth in Daphnia magna Strauss. Comp. Biochem. Phys-iol., Part C: Toxicol. Pharmacol. 140, 364 – fi sh population dynamicsaccounting for energy dynamics. PLoS One 10, e0125841.Boersma, M., Vijverberg, J., 1995. Synergistic effects of different food species on life-history traits of Daphnia galeata. Hydrobiologia 307, 109 – – – – – – ’ écologie. Sciences SUP. Sciences de la vie. Cours. Dunod, Paris,France, p. 8.David, V., et al., 2018. Modelling historical mesocosm data: application of a fi sh bioener-getics model in semi-natural conditions. Ecol. Freshw. Fish https://doi.org/10.1111/eff.12418.Davis, P., Ozburn, G.W., 1969. The pH tolerance of Daphnia pulex (Leydig, emend., Rich-ard). Can. J. Zool. 47, 1173 – – – fl ow labora-tory simulation of stream water quality changes downstream of an untreated waste-water discharge. Water Res. 43, 1993 – – e80.Forbes, V.E., et al., 2009. Ecological models in support of regulatory risk assessments ofpesticides: developing a strategy for the future. Integr. Environ. Assess. Manag. 5,167 – – – – – – – fi ciency on life his-tory characteristics and fi lter screens of Daphnia. J. Plankton Res. 18, 757 – – – – fi ed dynamic energy budget model for analysingecotoxicity data. Ecol. Model. 225, 74 – – fi ed threshold model ofsurvival-a toxicokinetic-toxicodynamic framework for ecotoxicology. Environ. Sci.Technol. 45, 2529 – – — a meta-analysis and case studies. Sci. Total Environ. 408, 3763 – – – B. Goussen et al. / Science of the Total Environment 749 (2020) 141509 an der Meer, J., 2006. An introduction to Dynamic Energy Budget (DEB) models withspecial emphasis on parameter estimation. J. Sea Res. 56, 85 – š cek, T., Nisbet, R.M., 2019. Inhibition and damage schemes within thesynthesizing unit concept of dynamic energy budget theory. J. Sea Res. 143, 165 – – – fl uence of food limitation on the effects of fenvalerate pulse ex-posure on the life history and population growth rate of Daphnia magna. Environ.Toxicol. Chem. 24, 2254.Pieters, B.J., Jager, T., Kraak, M.H.S., Admiraal, W., 2006. Modeling responses of Daphniamagna to pesticide pulse exposure under varying food conditions: intrinsic versusapparent sensitivity. Ecotoxicology 15, 601 – – – – – – – – fi c Committee on Health & Environmental Risks SCHER, 2013,. Scienti fi c Commit-tee on Emerging & Newly Identi fi ed Health Risks SCENIHR, Scienti fi c Committee onConsumer Safety SCCS, “ Addressing the New Challenges for Risk Assessment ” .European Union https://doi.org/10.2772/37863.Seidl, M.D., Paul, R.J., Pirow, R., 2005a. Effects of hypoxia acclimation on morpho-physiological traits over three generations of Daphnia magna. J. Exp. Biol. 208,2165 – – fl uence of chloride on the chronic toxicity of sodiumnitrate to Ceriodaphnia dubia and Hyalella azteca. Ecotoxicology 25, 1406 – –561.10