Assessing the Relative Toxicity of Different Road Salts and Effect of Temperature on Salinity Toxicity: LCx Values versus No-Effect Concentration (NEC) Values

Freshwater biota are at risk globally from increasing salinity, including increases from deicing salts in cold regions. A variety of metrics of toxicity are used when estimating the toxicity of substances and comparing the toxicity between substances. However, the implications of using different metrics are not widely appreciated. Using the mayfly Colobruscoides giganteus (Ephemeroptera: Colobruscoidea), we compare the toxicity of seven different salts where toxicity was estimated using two metrics: (1) the no-effect concentrations (NEC) and (2) the lethal concentrations for 10, 25 and 50% of the test populations (LCx). The LCx values were estimated using two different models, the classic log-logistic model and the newer toxicokinetic–toxicodynamic (TKTD) model. The NEC and both types of LCx values were estimated using Bayesian statistics. We also compared the toxicity of two salts (NaCl and CaCl2) for C. giganteus at water temperatures of 4 °C, 7 °C and 15 °C using the same metrics of toxicity. Our motivation for using a mayfly to assess salinity toxicity was because mayflies are generally salt sensitive, are ecologically important and are common in Australian (sub-)alpine streams. The temperature ranges were chosen to mimic winter, spring and summer water temperatures for Australian (sub-)alpine streams. Considering 144-h classical LCx values, we found toxicity differed between various salts, i.e., the lowest 144-h LC50 (8 mS/cm) for a salt used by a ski resort was half that of the highest 144-h LC50 from artificial marine salts and CaCl2 applied to roads (16 mS/cm). The analytical grade NaCl (as shown by 144-h LC50 value at 7 °C) was substantially more toxic (7.3 mS/cm) compared to analytical grade CaCl2 (12.5 mS/cm). Yet for NEC values, there were comparably fewer differences in toxicity between salts and none between the same salts at different temperatures. We conclude that LCx values are better suited to compare the difference in toxicity between substances or between the same substance at different test temperatures, while NEC values are better suited to estimating concentrations of substances that have no effect to the test species and endpoint measured under laboratory conditions.

Salinity is increasing in many freshwater ecosystems because of a range of anthropogenic causes including agriculture, mine effluent (Cañedo Argüelles et al. 2013;Sauer et al. 2016;Niedrist et al. 2021) and in cold regions, the application of salts for deicing, i.e., to prevent the build-up of snow and ice (Shenton et al. 2021). The application of deicing salts involves applying salts to roads, typically sodium chloride (NaCl) and/or calcium chloride (CaCl 2 ). The concentrations and/or ratios of major ions that make up salinity are becoming increasingly recognized as important as the total salinity concentration (Cañedo-Argüelles et al. 2016) to the toxicological effects on freshwater organisms. Consequently, the ecological and toxicological effects of road deicing may differ depending on whether NaCl, CaCl 2 or both are used. Additionally, those studies that have assessed the toxicity of different salts (Mount et al. 2019) have tended to use high purity, i.e., analytical grade salt (Mount et al. 1997), and it is feasible there may be differences in toxicity between the salts applied in deicing applications and these 1 3 high-purity salts used in toxicity tests, as the former would be expected to have lower purity.
Toxicity tests are a widely used experimental method to measure the toxicity of substances for individual test-species organisms. The results of toxicity tests can be expressed using a range of metrics, which have been debated (e.g., Landis and Chapman 2011;deBruyn and Elphick 2013). When the measured response of a test organism is mortality, the concentration lethal to x% of the test population is the most commonly used metric (LC x , see Table 1). For example, the LC 50 is the concentration of a toxicant in a solution that is lethal to 50% of test organisms (Burton. 1977). LC x values are calculated for a predetermined exposure period. The duration of acute tests with macroinvertebrates is usually between 24 and 96 h. For stream macroinvertebrates exposed to salinity, 72-h LC 50 values are commonly used (e.g., Kefford et al. 2012;Castillo et al 2018). Toxicity tests are typically conducted within ± 1-2 °C of the desired test temperature, although temperature is acknowledged to alter toxicity (Jackson and Funk 2019;Macaulay et al. 2020). Consequently, the LC 50 gives the concentration lethal to 50% of the test population for the indicated period of exposure at the test temperature indicated and makes no assessment of the response of the test organisms at other periods of exposure or test temperatures.
An alternative metric to LC x is the no-effect concentration (NEC) estimated with a toxicokinetic-toxicodynamic (TKTD) model accounting for both time and concentration as explanatory variables (Jager et al. 2011;Kon Kam King et al. 2015). The core goal of TKTD models is to allow the description of species responses to time-variable exposure conditions. NEC estimates the maximum concentration that has no effect on the test population studied and the response (or endpoint) measured. Such a response may be mortality (as with LC x ) but could alternatively measure sublethal responses, such as inhibition of a physiological function or the response of a whole organism (e.g., growth or reproductive output). Unlike LC x values, the NEC is time independent (Kon Kam King et al. 2015). LC x values are always associated with some level of mortality as it is not possible to have x = 0. Moreover, as x approaches zero, uncertainty in the estimate of the LC x value increases giving imprecise estimates. In contrast, the NEC is an estimate of the maximum concentration that has no effect, e.g., zero mortality from the toxicant, i.e., no mortality beyond which occurs in the control.
The statistical method by which LC x values are classically estimated, e.g., log-logistic models, uses only the data from the exposure period that the LC x is reported. Consequently, when using a log-logistic model to estimate LC x values, not all data that are typically available to the investigators are used. For example, in estimating 96-h LC x values the investigators would typically have survival data at 24, 48, 72 and 96 h, but when using a log-logistic model they would only use the data from 96 h. TKTD model (Jager et al 2011;Kon Kam King et al 2015) that is used to estimate NEC values can also estimate LC x values, but it uses data, potentially allowing TKTD to produce better estimates of LC x values. We shall refer to LC x values estimated by a log-logistic model as the classic LC x values or estimated by a classic model, and those LC x values estimated by a TKTD model as TKTD LC x values. The TKTD LC x values thus depend on all the TKTD model parameters.
Another potentially important factor is water temperature during salinity exposure. Recently, for example, Jackson and Funk (2019) observed that between 5 and 25 °C, four mayfly (Ephemeroptera) species tended to have higher 96-h LC 50 values (i.e., are more tolerant) when tested at lower water temperatures. Salinity levels from deicing salts are typically at a maximum in months when water temperatures are much lower than that used in standard toxicity tests. Thus, Jackson and Funk (2019) suggest that standard toxicity tests may  (Sutcliffe 1984), and these low metabolic rates at cold temperatures slow toxicant uptake, including major ions, relative to higher temperatures (Orr and Buchwalter 2020) and thus slow down the effect of the toxicant. If this occurs, then we would expect higher LC 50 values from a fixed exposure period at colder water temperatures, but these higher values may not necessarily indicate reduced toxicity at colder water temperatures.
Here, we have three aims. First, to compare the toxicity of various salts (artificial marine salts (AMS), analytical grade NaCl and CaCl 2 and four salts (RMS NaCl, RMS CaCl 2 , EWSS, EGSS) ( Table 1) used in deicing operations near Perisher, New South Wales (NSW), Australia) to the mayfly Colobruscoides giganteus (Ephemeroptera: Colobruscoidea). Second, we compare toxicity of two salts (analytical grade NaCl and CaCl 2 ) at different water temperatures, again using C. giganteus. Our motivation was comparing the toxicity of the different salts or salts and temperature combination, rather than determining environmentally safe concentrations or estimating the effects of salt concentrations in the environment (Shenton et al. 2021). For that end, concentrations of salt used were chosen based on their acute and lethal effect on C. giganteus and not based on environmental concentrations in our study region. Third, we examined the relative merits of using three separate approaches for assessing toxicity: (a) LC x values estimated by a loglogistic model, (b) LC x estimated by a TKTD model and (c) NEC values estimated by the same TKTD model; we do not aim to determine variability between tests. We determined if patterns in relative toxicity between the salts and water temperature were consistent across the NEC values estimated from the TKTD model and classically estimated LC x values and LC x values estimated from the TKTD model.
The NEC and LC x have little similarity with each other mathematically, and they refer to different effects, i.e., the NEC: no effect and the LC x : an x% effect. Nevertheless, on conducting a toxicity test there is a choice as to whether to estimate NEC, LC x or both values. Similarly, if both NEC and LC x values are estimated, there is a choice regarding which to use to aid decisions about the environmental management of chemicals. To make such choices, it helps to understand how NEC and LC x values compare between similar chemicals and the same chemical when toxicity tests are conducted under different conditions. As our interest in the current paper was in comparing the toxicity of salts used for road deicing, we compared NEC and LC x values between the different salts and the same salts at different water temperatures. To the best of our knowledge, comparing NEC and LC x values for similar chemicals has not been attempted before when studying salt-sensitive organisms, such as C. giganteus.

Field Collections
A range of alpine stream macroinvertebrates was collected during the colder months when salts are typically applied (June-September) during 2018 from six sites at 1400-1700 m elevation, near Perisher, Kosciuszko National Park (Supplementary Table S1). Stream water temperature during these collections ranged between 0.4 and 5.2 °C. During transport, the macroinvertebrates were kept in aerated site water that contained some detritus and sediment. Rapid toxicity tests (RTT) were started in the laboratory within 24 h of collection, and RTT methods are reported in detail in Moulding (2018) and Shenton et al. (2021). The reason RTTs were chosen was to determine what alpine species would be acceptable for conventional toxicity test style LC x testing, as a large number of individual species are needed for these tests.
Following the RTTs, C. giganteus was chosen to make more precise estimations of the toxicity of various salts and the effect of water temperature on salinity tolerance. This species was chosen because it is a mayfly (Ephemeroptera), and mayflies are generally considered salt sensitive (Kefford 2019), are ecologically important and are common in (sub-)alpine Australian streams. We specifically chose C. giganteus because of its high abundance in field collections and high long-term survival in laboratory under controlled conditions; see also Bray et al. (2021). All C. giganteus came from one site, Rock Creek (S36.41, E148.41, elevation 1700 m). Rock Creek is close to its headwaters, has minimal urban disturbance, has no deicing salt input (Supplementary Table S1) and consistently has low salinity levels (always < 0.1 mS/cm but typically < 0.01 mS/cm) (Shenton et al. 2021).

Experimental Methods
We conducted two experiments using conventional toxicity tests (Stefania et al. 2010). The first experiment aimed to investigate differences in toxicity between seven salts ( Table 1). The second experiment investigated the effect of water temperature on the toxicity of two of these salts: analytical grade of both sodium chloride (NaCl) and calcium chloride (CaCl 2 ).
All tests were conducted with 144 h of exposure to salt solutions with C. giganteus (seven individuals for experiment 1; and eight individuals for experiment 2) being placed into each beaker with 600 ml of treatment solution and two replicate beakers for each treatment or control. All beakers were aerated using air stones with no substrate added. All controls and dilution solutions were dechlorinated Canberra tap water (≈ 0.1 mS/cm). Test concentrations (0.1-16 mS/cm) were selected to display 100% survival, partial survival and 100% mortality of C. giganteus during the experiments. This was to allow for comparisons of the toxicity of the salts or salt/temperature combination while providing valid dose-response curves (Schäfer et al. 2012). Generally, a geometric series of concentrations (in this case mS/cm) is commonly used in toxicity tests. Generally, a geometric series of concentrations (in this case mS/cm) is commonly used in toxicity tests. Our non-geometric experimental design did not result in concentrations doubling as is common with many toxicants (Environment Canada 2005).
No individuals were fed during the tests, illumination was 8 h per day, and water was not refreshed during experiments. All toxicity tests were inspected, and any dead individuals removed daily. If an individual C. giganteus was suspected of being dead, it was softly prodded with a pipette, while the individual was still submerged in treatment solution. If no movement of the individual was observed after prodding, if it failed to move, it was then collected in a teaspoon (with treatment solution) to witness any small movements, e.g., of legs or mandibles. If the individual was confirmed dead, it was placed into 80% ethanol, and any twitching was noted. In the rare cases where twitching occurred, it was considered alive and became a 'missing' individual from the experiment at subsequent observations. LC x (x = death of 10, 25 and 50% of the test population) values were estimated for exposure periods (24, 48, 72, 96, 120 and 144 h).
Greater replication, i.e., > 2 replicates, would result in a more robust dataset; however, the numbers of C. giganteus available during field collections allowed for only two replicates. We found low variability between replicates; therefore, we decided to proceed with two replicates. Thus, we decided to test more concentrations of each salt or salt/ temperature combination and not perform large replication, rather than testing few concentrations with many replicates. Furthermore, increasing the number of replicate beakers would have been logistically impossible both in terms of the time needed to collect C. giganteus and temperature-controlled space to conduct the test. Moreover, the number of replicate containers (e.g., beakers, fish tanks) per treatments is controversial in ecotoxicology. Some have argued for reducing the number of replicate containers to one and obtaining statistical power with a so-called regression design (Warne and Van Dam 2008). Newer statistical designs which increase the number of tested doses at the expense of fewer replicates are also favored with Omics-type response variables (Larras et al. 2018).

Experiment 1: Determining the Difference in Toxicity of Different Salts to Colobruscoides giganteus
A single conventional toxicity test was used to compare relative toxicities of seven different salts to C. giganteus at 7 °C ± 1.0 °C. The seven salts were: AMS (Ocean Nature), analytical grade CaCl 2 and NaCl and four types of salt applied for deicing in and around Perisher. These salts were CaCl 2 (flakes) and NaCl (granulate) used by the government authority responsible for deicing operations, the Roads and Maritime Services of NSW (RMS, hereafter these salts are referred to as RMS CaCl 2 and RMS NaCl, respectively) and two types of salts used by the Perisher Environmental Team: evaporated gourmet sea salt (EGSS) and evaporated washed sea salt (EWSS) (both 'Mermaid' brand). AMS was included to be able to compare the results with other published toxicity results (e.g., Kefford et al. 2012). Analytical grade NaCl and CaCl 2 were used as these salts (at lower purity) are widely used as deicing salts, including near Perisher (Shenton et al. 2021). The salt used by the RMS and the Perisher Environmental Team was used to determine the toxicity of the salt applied to the environment. The electrical conductivity (EC) of treatment solutions was 0.1 (control), 4, 8, 12 and 16 mS/cm.

Experiment 2: Determining the Effects of Water Temperature on Salinity Toxicity to Colobruscoides giganteus
A conventional toxicity test was also used to compare the sensitivity to analytical grade NaCl and CaCl 2 of C. giganteus at a cold temperature (4 °C ± 1.0 °C), a mid-temperature (7 °C ± 1.0 °C) and a high temperature (15 °C ± 1.0 °C). The cold temperature mimics typical winter water temperatures observed at Rock Creek. The mid-temperature environment was chosen to mimic observed early spring water temperatures, and the high-temperature environment was used to mimic observed late spring/early summer water temperatures (Shenton et al. 2021, unpublished data). The EC of treatments for this experiment was 0.1 (control), 5, 6, 7, 8 and 12 mS/cm. Experiment 1 had 0% survival at 16 mS/cm; consequently, this salinity level was removed from experiment 2.

Data Analysis
In both experiments, GUTS-SD (General Unified Threshold Survival) (Jager et al 2011;Jager & Ashauer 2018) was used to determine LC x and NEC results. GUTS-SD is a type of TKTD model that allows the use of time-resolved data collected during each experiment (i.e., survival data at 24, 48, 72, 96, 120 and 144 h). LC x values were also estimated using a 2-parameter log-logistic model (Ritz et al. 2015).
Parameter inference for both models was performed in the Bayesian framework using the R package 'morse' (Baudrot et al. 2018;Delignette-Muller et al. 2017). We compared the GUTS-SD and GUTS-IT models on each dataset, and although goodness-of-fit diagnostics sometimes favored one or the other model, general results were qualitatively similar regardless of the model chosen. (Detailed goodness-of-fit diagnostics are given in the supplementary material (text S1), with the code to replicate the analysis.) Posterior distributions for the LC x could be calculated directly from the posterior distribution on the parameters for the log-logistic model and the GUTS-SD model. Because of the use of the Bayesian framework, uncertainty in estimates of LC x and NEC values is given in 95% credibility intervals (not to be confused with 95% confidence intervals that are commonly used with a frequentist framework).

Experiment 1: Determining the Difference in Toxicity of Different Salts to Colobruscoides giganteus
Using the classical LC x model, a log-logistic model, LC x values generally decreased with increasing exposure period (Fig. 1). However, there were some exceptions. For example, the median estimate of the LC 10 for EWSS increased marginally from 120 to 144 h (7.7 and ~ 9 mS/cm, respectively). In the absence of mortality, classical LC x values for short exposures, e.g., 24 h, could not be estimated and only a lower bound was available.
Classical LC x values did show that some salts had clearly different toxicity, with these differences being best indicated at prolonged exposure, i.e., 120 h and 144 h (  was less toxic than other salts tested within our experiments despite only approximately a twofold difference and a median estimate of 144-h LC 50 of about 16 mS/cm. In comparison, EWSS had a value for this parameter at about 10 mS/cm. A ranking of the classical LC 50 mean values at 144 h suggests that EGSS is the most toxic salt (8 mS/cm), followed by EWSS (10 mS/cm), RMS NaCl (11 mS/cm), CaCl 2 and NaCl (12 mS/cm). AMS (16 mS/cm) and RMS CaCl 2 appear to be the least toxic salt (16 mS/cm) where medians and means were effectively identical (Table 2).
When applying the TKDT model, estimates of LC x values were more variable for AMS and RMS CaCl 2 than all other salts, especially at the shorter exposure periods, i.e., 24 and 48 h (Fig. 2).
TKTD-modeled LC x values always declined with increasing exposure period, and the upper 95% credibility interval was always defined, i.e., it was never infinity (Fig. 2). TKTD-modeled LC x values converge to the NEC as a direct mathematical consequence of the GUTS model (Kooijman and Bedaux. 1996). With this model, for short exposure periods, e.g., 24 and 48 h, estimated LC x values were typically greater than the experimental salinity levels used because of limited mortality in all treatments with short exposure periods. There were also differences in the toxicity of some salts apparent even after brief exposure periods. For example, at 24 h of exposure AMS had a higher LC 50 (e.g., median LC 50 value of 99 mS/Cm) than EWSS (e.g., median LC 50 value of 39 mS/cm) (Supplementary table S3) with no overlapping of their 95% credibility intervals. Nevertheless, credibility intervals did narrow with increasing exposure period across all salts tested (Fig. 2).
NEC values and associated credible intervals for salts applied within KNP at 7 °C showed that EWSS was less toxic (8.2 mS/cm) compared to other applied salts RMSCaCl 2 (2.5 mS/cm), RMS NaCl (1.25 mS/cm) and EGSS (~ 2.75 mS/cm) when considering their statistical differences (Fig. 3). Of salts tested here, NaCl and RMS NaCl have a lower median NEC (~ 1 mS/cm and 1.5 mS/cm, respectively) compared to EWSS (~ 8.2 mS/cm) showing a marked difference. Salt types AMS (~ 2.4 mS/cm), CaCl 2 (~ 5.1 mS/cm) and RMS CaCl 2 do have larger credible intervals than NaCl and RMS NaCl that are considered the most toxic when using the NEC (Fig. 3). The other salts had fewer precision estimates of their NEC values, and it is likely that these salts all had NEC values that encompass estimates for the NEC values of EWSS, NaCl and RMS NaCl.

Experiment 2: Determining the Effects of Water Temperature on Salinity Toxicity to Colobruscoides giganteus
As with the previous experiment with the seven salts, classically estimated LC x values for CaCl 2 and NaCl at the three temperatures (4 °C, 7 °C, 15 °C) tended to decrease with increasing exposure period but there were some exceptions (Fig. 4). The 72-h LC 50 values of CaCl 2 at 7 °C decreased between 96 and 120 h, e.g., median estimates of ~ 16 and ~ 12.6 mS/cm, respectively) (Fig. 4). Again, as with the previous experiment, the upper credibility intervals for brief  Table S3 exposures (e.g., 24 and 48 h) LC x values were at infinity, thus providing very poor estimates of toxicity.
In terms of patterns in toxicity between temperatures, the classical LC x values show that CaCl 2 had similar toxicity at 4 °C and 7 °C (e.g., median estimates of 144-h LC 50 of 14 and 12 mS/cm, respectively), but was more toxic at 15 °C (e.g., median estimates of 144-h LC 50 of 8 mS/cm).
In contrast, NaCl toxicity tended to increase with increasing water temperature. See for example 144-h LC 50 values (Fig. 4) Table S3).

and 144-h LC 50 values (Supplementary
The classic LC x values with long exposure (i.e., 120 and 144-h) showed that NaCl was less toxic than CaCl 2 at 7 °C (e.g., median estimates of 120-h LC 50 of 22 mS/cm), relative to 4 °C and 15 °C (e.g., median estimates of 120-h   4 Classical (log-logistic) modeled lethal concentrations for x% of the test population or LC x values (mS/cm) to analytical grade CaCl 2 and NaCl and temperature (4 °C, 7 °C, 15 °C). The multiple estimates for each x represent estimates (from left to right 24, 48, 72, 96, 120 and 144 h). x = LC x 10, 25, 50 (from left to right). Median indicated by the horizontal line within box, 95% credibility intervals indicated by boxes. The y-axis is curtailed at 17 mS/cm, and in those estimates without a median plotted, the upper 95% credibility interval continues to infinity. A full listing of all LC x estimates shown here is given in Supplementary Table S2 1 3 LC 50 of 4 and 16 mS/cm, respectively) (Fig. 4, Supplementary Table S3). At 4 °C and 15 °C, the credibility intervals overlapped between the salts for the 144-h LC 50 values. The difference in their toxicity was minimal at 4 °C and 15 °C compared to 7 °C (Fig. 4).
As with the previous experiment, the TKTD-modeled LC x values always decreased with increasing exposure period and the upper credibility interval was always defined (i.e., less than infinity). For CaCl 2 , credibility intervals were much narrower at 15 °C relative to those at 4 °C and 7 °C (Fig. 6). For CaCl 2 , credibility intervals were much narrower at 15 °C relative to those at 4 °C and 7 °C (Fig. 6). For NaCl, credibility intervals were slightly wider at 4 °C relative to those 7 °C and 15 °C.
Similar to the classical LC x values, there was no evidence of difference in the CaCl 2 LC x values between 4 and 7 °C  with overlapping credibility intervals (Fig. 5). At 15 °C, CaCl 2 was more toxic (e.g., median estimate of 144-h LC 50 of 8.4 mS/cm) than at either of the other temperatures tested (median estimate of 144-h LC 50 of 16 mS/cm for both 4 °C and 7 °C). NaCl was least toxic at 4 °C (e.g., median estimate of 144-h LC 50 of 9.7 mS/cm), and this salt had similar toxicity at 7 °C and 15 °C (median estimate of 144-h LC 50 of 7.6 and 7.0 mS/cm for 4 °C and 7 °C, respectively, and without overlapping credibility intervals between the estimated LC 50 values for 4 °C and 15 °C).
All estimations of NEC values for both salts at all three temperatures had NEC values with overlapping credibility intervals (Fig. 6). Considering median estimates, NEC values were lower for NaCl (indicating greater toxicity) at 7 °C compared to CaCl 2 (e.g., CaCl 2 with a NEC value of ~ 5 mS/ cm and NaCl with a NEC of ~ 1.2 mS/cm) than at both 4 °C and 15 °C (both salts = or < 1.25 mS/cm) (Fig. 6).

Discussion
The different methods of toxicity estimation, i.e., classically modeled (i.e., log-logistic) LC x values (Figs. 1 and  4), TKTD-modeled LC x values (Figs. 2 and 5) and TKTDmodeled NEC values (Figs. 3 and 6), provided different estimates of toxicity. Classical LC x estimates were incalculably high for short exposure periods with only the lower credibility interval estimated and the upper credibility intervals stretching to infinity. This is a function of the chosen salinity concentrations having limited mortality at 24 and often 48 h. No doubt with the inclusion of higher salinity concentrations than we used in our experiments, better estimates could be obtained for classic 24-and 48-h LC x values. In contrast, the TKTD-modeled LC x values always had upper credibility intervals defined below infinity. Where a toxicity test aims to make estimates of LC x values across a wide range of exposure periods (e.g., 24 to 144 h, as in the current paper), there are considerable advantages of using a TKTD model. LC x values regardless of exposure period (Delignette-Muller et al. 2017) only take simple dose-response models into consideration and are still often used, despite limitations of only exploiting data observed at the end of the experiment. TKTD models can be advantageous compared to the classic modeling methods to determine the LC x . In comparison, TKTD models simulate the time course of processes leading to toxic effects on organisms (Ashauer and Escher 2010;Jager et al. 2011). Furthermore, when the TKTD model is used it allows a better fit for any exposure period or concentration providing a more accurate NEC (Kon Kam King et al. 2015). Finally, the Bayesian approach is especially helpful when approaching the sparsest of datasets (Delignette-Muller et al. 2017).
Classical LC x values tended to, but did not always, decrease with increasing exposure period, while TKTD LC x values always decreased with increasing exposure period. Toxicity depends on both concentration and duration of exposure. With an increase in the duration of exposure, toxicity will either be unchanged (if there is no more mortality) or toxicity will increase with further mortality. However, the extent to which LC x values should decrease, associated with increased duration of exposure, cannot be quantified. This could be considered a shortfall of classical LC x values. However, TKTD-modeled LC x values for brief exposure periods were typically greater than the concentrations used in the experiments, unlike the classic LC x values. This is because the TKTD model extrapolates the short-term effects of the treatments beyond the data collected. Whether this is acceptable is debatable, but, in our view, sometimes extrapolation of data is necessary and the most meaningful way of analyzing and inferring from data collected (Jager. 2021). This is especially pertinent when it is logistically impossible to collect a larger more robust dataset (Colwell et al. 2004). It could be considered that the TKTD model is using more information and providing more estimations of the toxicity of brief exposures, which the classic LC x model does not. Similarly, without conducting further experiments, we cannot empirically understand the effects of the salinity levels beyond those used in the experiment. However, caution must be used in extrapolating in data analysis, and in this instance in interpreting LC x values from the TKTD model beyond the experimental concentrations used.
When credible intervals do not overlap, there is a substantial difference. The LC x values, whether estimated classically or by the TKTD model, showed substantial differences in toxicity between some salts and the same salt at different water temperatures. The estimated NEC values, however, had overlapping credibility intervals across all salt types or temperatures, with one exception. This exception was that EWSS had a higher NEC value than all other salts for RMS NaCl appearing to be more toxic and CaCl 2 being the least toxic of other salts (Figs. 3 and 6). Regardless of this exception, the NEC values were generally less effective at detecting differences in toxicity between the salts or temperatures than both LC x models. LC x values appear to be better for comparing toxicity between substance or the same substance at different test temperatures. For aiding in the setting of environmental quality guidelines, NEC values are likely to be more useful, especially when using one of the more sensitive taxa within a studied environment (C. giganteus), although environmental quality guidelines should also consider sublethal toxicity and indirect effects of toxicants (Bray et al. 2019).

Experiment 1: What is the Difference in Toxicity of Different Salts to Colobruscoides giganteus
The current study is the first that we are aware to compare the toxicity of analytical grade salts with those used by road deicing operations. We observed statistical differences between the toxicity of NaCl-dominated salts, for example, in terms of classical mean LC 50 values EGSS (8/mS/cm) was considerably more toxic than all other NaCl-dominated salts (10-16 mS/cm), twofold RMS CaCl2 (16 mS/cm) at 144 h ( Table 2). In contradiction to both LC x models, the NEC showed that EWSS was significantly less toxic (~ 8 mS/cm) than all other salts tested (all ~ or < 3 mS/cm) (Fig. 3). All other salts showed no differences in toxicity between each other when the NEC was applied as indicated by overlapping credibility intervals. Further investigation is required, however, to best evaluate toxicity across different taxa, and sublethal toxicity effects in natural or seminatural (e.g., mesocosms), especially where multiple species are interacting (Clements and Kotalik 2016;Bray et al. 2019). It is highly plausible that such studies, because of abiotic and biotic conditions, could find the effect of deicing salts at lower concentrations than is indicated here by single-species laboratory test conditions.

Experiment 2: Does Water Temperature have an Effect on the Salinity Toxicity to Colobruscoides giganteus?
Temperature affects numerous biological processes and modifies the toxicity of many environmental contaminants, including salts, in freshwater ectotherms (Orr and Buchwalter 2020;Verberk et al. 2020). For example, Jackson and Funk (2018) found over a fourfold increase in toxicity for the mayfly species Neocloeon triangulifer (Baetidae) using temperature ranges (5, 7.5, 10, 12.5 and 15 °C). Our results support this trend, at least for classical 96-h LC 50 values for NaCl (13 mS/cm at 4 °C, 11 mS/cm at 7 °C and 10 mS/cm at 15 °C). Furthermore, the subantarctic isopods Limnoria stephenseni had major changes in LC 50 's that can be primarily attributed to a specific temperature increase when exposed to copper (Proctor et al. 2017;Proctor 2019). Similarly, the subantarctic marine copepod Harpacticus sp., isopod Limnoria stephenseni, flatworm Obrimoposthia ohlini and bivalve Gaimardia trapesina also had increased sensitivity to copper with small increases in temperature (2-4 °C) (Holan et al. 2019).
Our results support this trend, at least for classical 96-h LC 50 values for NaCl (13 mS/cm at 4 °C, 11 mS/cm at 7 °C and 10 mS/cm at 15 °C). For CaCl 2 , however, we found no difference in toxicity (as indicated by both LC x models). The classic LC x values with long exposure (i.e., 120 and 144-h) showed that NaCl was considerably more toxic than CaCl 2 at 7 °C, while at 4 °C and 15 °C the differences in toxicity between these salts were less. At 4 °C and 15 °C, the credibility intervals overlapped between the salts for the 144-h LC 50 values. The difference in the toxicity of the two salts was minimal at 4 °C and 15 °C compared to at 7 °C.
Increased temperature is thought to increase toxicity because of an increase in ion transport rates and a higher energy expense during osmoregulation (Orr and Buchwalter 2020;Verberk et al. 2020). However, differences in some LC x values for different salt and/or temperature combinations do not as accurately describe differences in long-term toxicity such as those indicated by NEC values. The similarity in the NEC values for NaCl and CaCl 2 at different temperatures means that we cannot exclude the hypothesis that both salts are similarly toxic at all three of the tested temperatures, but that a response takes longer at the colder temperatures because of decreased metabolism and ion uptake (Orr and Buchwalter 2020) at colder temperatures. Therefore, comparisons between different salt, temperatures, or other stressors with LC x values need to be treated with caution given they may not necessarily indicate real differences in long-term toxicity.

Why are there Differences in Toxicity Indicated by LC x and NEC?
We showed that differences in toxicity between salts, or between the same salt at different exposure temperatures, were indicated by one or both LC x models; however, this was not reflected with NEC values. This appears to be because there was similar overall toxicity of the different salt or exposure temperatures, but that for some salt or at some exposure temperatures, it took longer exposures for the similar toxicity to be apparent. NEC values are time independent, that is, they estimate a threshold of toxicity that has no effect on the test population for any length of exposure for the relevant endpoint (Kon Kam King et al. 2015), in our case mortality. This is because any LC x values are an estimate of the concentration that cause the indicated proportion of mortality (i.e., x%) for the relevant exposure period. In the case of LC 50 values, 50% is a substantial amount of mortality, which if such a level of mortality were to occur to natural populations could have profound consequences for their persistence. While it is possible to estimate LC x values that are more protective, e.g., LC 5 values, they still represent a concentration where an ecologically relevant effect (mortality) is occurring. Moreover, any LC x value is representative of the toxicity for a given exposure period, e.g., 72 h, and makes no estimation of what effect would occur if exposure were longer, other than longer exposure would have the same or more of an effect. NEC values are therefore more meaningful for determining environmental effects than LC x values and are widely suggested as an improvement to the LC x (Proctor 2019) at least when the exposure period is lengthy, and especially when studying biota from cold regions where toxicity test duration should have a lengthy exposure period, up to 42 days (Proctor 2019).
NEC vs. LC x comparisons are limited in the current scientific literature, but see Proctor (2019). Proctor (2019), similar to our results, found that increased temperature lowered the LC x value of a subantarctic isopod, Limnoria stephenseni, which increased the effect of copper exposure. Like our results, Proctor (2019) also found no/little differences in NEC values (for copper) between temperatures. She also observed that at long exposure periods (~ 40 days) LC x values between temperatures become similar. She hypothesizes that this is a result of polar taxa requiring extended exposure to illicit an acute response as they respond slower to toxicants than warmer temperature taxa.
When applying the NEC, for example a few deaths late in the exposure period (as was the case with EWSS), it can make a big difference to the precision of the NEC estimate. When the aim of a toxicity test, using lethality as the endpoint, is to estimate NEC values, exposure periods should be long enough that there is little or no more mortality at the later exposure periods (as in the case of EWSS). Where other sublethal endpoints are used, exposure needs to be long enough that additional effects on measured endpoints are no longer occurring.

Short-Term and Long-Term Salinity Inputs and Outputs
Salinization as a result of deicing salting operations typically has highly variable temporal salt concentrations, frequently with short-term spikes in salinity (Shenton et al. 2021). Relating LC x values, e.g., 96 h, with fixed exposure periods to salinity is problematic because LC x values are highly dependent on the exposure period. As NEC values are time independent, they may be better suited for typical real-world exposure from deicing operations. Deicing salts may store in the land and leach out into streams throughout the entire year, including in our study region (see Shenton et al. 2021;Moulding 2018), so that chronic exposure may also be relevant. The highest salinity spikes in our study region (and other alpine areas) occurred in winter or early spring because of the deicing activities (Shenton et al. 2021;Moulding 2018), when water temperatures are generally low. To determine the longer-term effect of salinity at such low water temperatures requires lengthy experiments. The NEC may be a powerful tool for predicting long-term survival of organisms from relatively short-term tests and would provide a more environmentally sound estimate that can take other variables such as landscape salinity output into consideration to understand any long-term effects from outputs in summer and autumn.

Conclusions
Our results suggest that when studies aim to estimate LC x values across a range of exposure periods, better estimates will generally be obtained using a TKTD model because this model uses all available information. This is seen by the classical model estimating upper credibility interval and the median estimate of LC x values at infinity, while for the TKTD model obtained finite estimates. In contrast, for studies that aim to estimate the concentration of a substance that, with long-term exposure, will not produce toxicity, NEC values will generally be better because this estimate is time independent and is generally lower value and more accurate than comparable LC x values; therefore, NECs offer greater protection for biodiversity.
Our study supports the idea that deicing operations should consider the toxicity of salts used, and examination of this should not be limited to analytical grade salts, as toxicity varied between NaCl-dominated salts. Careful selection of salts in mixtures known to have relatively less toxicity has the potential to reduce the environmental harm of deicing salts. Toxicity was variable and largely similar across most salts examined, but the least toxic salts were RMS CaCl 2 (least toxic by classical 144-h LC 50 results with a median of ~ 16.29 mS/cm) ( Table 2) and EWSS (least toxic by NEC with a median value of ~ 7.75 mS/cm) (Fig. 3).