Screening Chemical Alternatives with Limited Ecotoxicity Data To Support Safe and Sustainable by Design Innovation

Abstract

Enabling effective early-stage screening of chemicals is essential for advancing safety and sustainability, as highlighted by the European Commission’s Safe and Sustainable by Design framework. However, such screening is often constrained by the limitedavailability of ecotoxicity data. Here, we investigate how chemical screening can be performed under data-limitedconditions. We analyzed 132 chemicals with extensive chronic ecotoxicity data and derived Environmental Risk Assessment (ERA) and Life Cycle Assessment (LCA) metricsEnvironmental Threshold Concentrations (ETCs) and Ecotoxicity Effect Factors (EEFs). Eight data-reduction scenarios were compared with a data-rich reference, varying (i) acute versus chronic endpoints, (ii) taxonomic coverage (single species vs three trophic levels: algae-fish-crustaceans), and (iii) calculation method. Across ERA and LCA, the scenario including three trophic levels and chronic data showed the lowest uncertainty and best agreement with the reference. Single-species scenarios introduced substantial uncertainty, particularly for specifically acting chemicals. Replacing chronic with acute data increased uncertainty only moderately. For ERA, ETCs from the lowest toxicity value aligned more closely with the reference than ETCs from species sensitivity distributions. For LCA, the calculation method had limited influence on EEFs. Our approach can support transparent decision-making for the safe and sustainable design of chemicals, where large uncertainties are unavoidable.

Introduction

Chemical pollution poses significant threats to ecosystems, highlighting the urgent need to address impacts already in the design phase by developing safer and more sustainable chemicals. The European Commission’s Safe and Sustainable by Design (SSbD) framework promotes this innovation shift by integrating safety and sustainability considerations into chemical innovation. The SSbD framework buildsamong other approacheson low-tier Environmental Risk Assessment (ERA) and Life Cycle Assessment (LCA), which provide screening and ranking of chemical alternatives to support design decisions within SSbD. ,

ERA and LCA provideamong othershazard-related insights following distinctive yet complementary perspectives. Both rely on laboratory-based ecotoxicity data, such as No-Observed Effect Concentrations (NOECs) and Effect Concentrations that cause a 50% reduction in the assessed effect on the exposed species (EC50). − ERA applies these data within regulatory frameworks that specify minimal requirements for taxonomic coverage and standardized data analysis methods for defining Environmental Threshold Concentrations (ETCs), − whereas LCA uses less stringent requirements for defining Ecotoxicological Effect Factors (EEFs). ,,

During the early (re)­design of chemicals, ecotoxicity data are essential inputs for low-tier assessmentsincluding screening and ranking, which aim to identify potential environmental concerns before chemicals enter the market. To support these assessments under data-limited conditions, New Approach Methodologies (NAMs) are increasingly applied to predict chemical hazards using in vitro bioassays or in silico models based on chemical structure or physicochemical properties. , These predictive approaches offer clear advantages for novel substances and early innovation stages. NAMs, however, generally remain complementary to, rather than replacements for, empirical ecotoxicity data. , Hence, scarcity in experimental ecotoxicity data remains a critical bottleneck in chemical innovation, constraining the reliability of screening assessments and precautionary decision-making. − Studies have investigated various approaches that enable the evaluation of chemical hazards with limited ecotoxicity information. − The findings of these studies consistently demonstrate that outcomes are highly sensitive to methodological choices and underlying assumptions. As a result, there is a growing recognition of the need for structured and transparent guidance to support decision-making under uncertainty and to ensure consistent interpretation of screening results. −

The future of chemical safety and sustainability is inherently related to the capacity to empower the screening of chemicals at a very early stage of their development. Our paper contributes to this ambition by providing insights into the use of limited experimental ecotoxicity data to support the SSbD assessments of chemicals. We also investigated factors that may affect these insights, such as the Mode of Action (MoA) of the chemicals assessed or the adopted method to quantify ETCs and EEFs. We selected 132 data-rich chemicals and simulated eight data-reduction scenarios, reflecting various data limitations. For each data-reduction scenario, we derived ecotoxicological effect metrics for ERA and LCA and compared the outcomes with the data-rich reference scenario. Based on our findings, we identified data-limited scenarios that come closest to the reference scenario, aiming to support screening assessments. Our study can also help the harmonization of the underlying data and methods used to derive ecotoxicological effect metrics in ERA and LCA. Such harmonization is one of the key ambitions of the SSbD framework and is also central to the EU’s One Substance, One Assessment initiative, which aims to streamline chemical evaluations across regulatory domains. ,

Materials and Methods

Ecotoxicity Data Collection and Curation

The initial data set includes 7,081 chemicals with EC50s and NOECs, sourced from the REACH-IUCLID database and Oginah et al. From the data set, we initially filtered the chemicals with experimentally determined ecotoxicity data. , Then, we identified the final set of data-rich chemicals using at least six NOECs, covering at least algae, fish, and crustaceans (AFC).

The final data set comprises 132 chemicals, of which 50 are classified as specifically acting, 27 as narcotic, and 55 as unclassified, based on the consensus MoA classification derived from existing databases and the scientific literature. − The complete data set is reported in S 1, and additional details on the data collection and curation procedures are provided in section S2.

Definition of Scenarios

To assess the sensitivity of the ERA and LCA metrics to data availability, we selected key ecotoxicity data characteristicsspecies coverage, type of endpoint, and trophic levels (AFC)consistent with regulatory frameworks from the Environmental Chemical Agency (ECHA). These scenarios reflect the data configurations most commonly encountered in a regulatory context and allow systematic exploration of uncertainty under data-limited conditions.

First, we defined the reference scenario (Sc_REF in Table ) reflecting conditions that meet the data requirements for constructing the species sensitivity distribution (SSD). Subsequently, eight additional scenarios (Table ) were defined by systematically varying these characteristics to represent data-limited conditions, which are typically relevant to the chemical innovation context.

1. Description of the Scenarios and Their Characteristics.

	Characteristics of the input ecotoxicity data
Name scenario	Number of input data	Type of endpoint	Number of trophic levels	Type of trophic levels
Sc_REF	≥6	NOEC	≥3	Algae, Amphibian, Crustacean, Fish, Insect, Mollusca, Plant, Rotifera
Sc_AFC_NOEC	3	NOEC	3	Algae, Fish, Crustacean
Sc_A_NOEC	1	NOEC	1	Algae
Sc_F_NOEC	1	NOEC	1	Fish
Sc_C_NOEC	1	NOEC	1	Crustacean
Sc_AFC_EC50	3	EC50	3	Algae, Fish, Crustacean
Sc_A_EC50	1	EC50	1	Algae
Sc_F_EC50	1	EC50	1	Fish
Sc_C_EC50	1	EC50	1	Crustacean

^aOne single data point for each of the following three trophic levels: Algae, Fish and Crustacean. The first choice of species was:Daphnia magna, Oncorhynchus mykiss, and Raphidocelis subcapitata , according to EFSA Requirements. If one of these three species was missing, itwas replaced by another species from the same taxonomic group, choosing the onewith the lowest concentration if more were available.

Derivation of Environmental Risk Assessment Metrics

The ecotoxicological threshold concentration (ETC) represents a conservative estimate of the concentration below which no adverse effects are expected. We applied eq provided by ECHA with a few adjustments, as described below and in S2.

$$ETC\text{\_}SS{D}_{x,s}(orETC\text{\_}LT{V}_{x,s})=\frac{TO{X}_{x,s}}{A{F}_s}$$ 1

TOX is the toxicity value (in μg/L), AF is the assessment factor (5, 10, 100, or 1000, depending on the toxicity data availablein line with ERA practices), x is the chemical, and s is the scenario. For each of the eight data-limited scenarios, we derived two ETCsETC_SSD and ETC_LTV depending on the TOX value. TOX was defined as either the Hazardous Concentration affecting 5% of species (HC5), the lowest NOEC, or the lowest EC50. For the Sc_REF, a single ETC was derived based on the HC5. The HC5 was derived from the SSD, characterized by its median (μ), standard deviation (σ), and extrapolation factor (Ks):

$$H{C}_{x,p,s}={10}^{({\mu }_{x,s}-K{s}_{p,s}*{\sigma }_{x,s})}$$ 2

where p is the fifth percentile (0.05) of a log-normal SSD.

The SSD of a chemical is fitted to measured or extrapolated chronic EC10-equivalent values. For the remainder of the present study, it was assumed that the SSD based on NOEC values is equal to the SSD based on chronic EC10 values, given the observations of Oginah et al. For HC5 derivation from EC50 data, we first extrapolated EC50 to NOEC values following Saouter et al. In single-species scenarios, the σ (eq ) was fixed at the 95th percentile of the σ from Sc_REF as a conservative estimate of interspecies differences in toxicity, and the Ks_fifth was set to its asymptotic value corresponding to the “inf” number of ecotoxicity data (1.645). Three σ-95th values were derived, corresponding to the three MoAsspecifically acting, narcosis, and unclassified (Figure S1 in S2). Finally, the confidence interval was quantified for Sc_REF by using Ks to derive the lower and upper bounds of the 90% confidence interval following Aldenberg & Jaworska (S1).

Derivation of Life Cycle Assessment Metrics

The ecotoxicity effect factor (EEF) represents the change in the affected fraction of species due to a change in the concentration of a chemical. We applied eq to derive the EEF:

$$EE{F}_{p,x,s}=\frac{P_{x,s}}{HC{p}_{x,s}}$$ 3

where p is the fifth, 20th, or 50th percentile (0.05, 0.2, or 0.5) of a log-normal SSD, HC the hazardous concentration, x the chemical, and s the scenario.

For each scenario, we derived three values of the EEF – i.e., EEF5, EEF20, and EEF50corresponding to the fifth, 20th, and 50th percentiles of the SSD of a chemical, respectively. The three EEF values represent different assessment points of interest. The HC20 is the metric that is used in the UNEP-SETAC recommendation and the environmental footprint method and is, therefore, always used as the reference scenario. The HC50 is the central portion of the curve with the lowest uncertainty, which is used as the basis for the EEF derivation in a number of life cycle impact assessment methods. , We also included the HC5, as this metric is usually used in ERA.

A similar approach to ETC_SSD was used to derive EEFs in single-species scenarios, with a few key differences. The σ of the SSD was set to the 50th percentile of the σ from Sc_REF to represent the most likely value of interspecies differences in toxicity, and three Ks were applied depending on the LCA metric (1.645 for the EEF5, 0.842 for the EEF20, and 0 for the EEF50). Confidence intervals for Sc_REF were calculated using the same approach described in the “Derivation of Environmental Risk Assessment Metrics” section.

Indicators to Quantify Differences across the Scenarios

We quantified the response ratio (RR), which represents the magnitude of the difference in the estimated metrics relative to the reference scenario (Sc_REF). For each chemical and scenario, the RR was calculated as the 10-based logarithmic ratio between the metric value in each data-limited scenario and the metric in Sc_REF (eqs and ).

$$RR{(ETC)}_{ScX}={\mathrm{Log}}_{10}\left(\frac{ET{C}_{Sc\text{\_}REF}}{ET{C}_{ScX}}\right)$$ 4

$$RR{(EEF)}_{ScX}={\mathrm{Log}}_{10}\left(\frac{EEF{p}_{ScX}}{EEF{20}_{Sc\text{\_}REF}}\right)$$ 5

The equations are formulated to ensure consistent interpretation of the RR values across the two methodologies (LCA and ERA) because of their opposite hazard definitions: in LCA, higher EEF values correspond to a greater hazard, while in ERA, higher ETC values correspond to a lower hazard. An RR of 0 indicates a perfect match between a data-limited scenario and the data-rich Sc_REF scenario. A positive RR indicates a more conservative estimatei.e., the scenario yields a lower ETC or a higher EEF than Sc_REF. Conversely, a negative RR indicates a less conservative estimatei.e., the scenario yields a higher ETC or a lower EEF than Sc_REF. Results for all chemicals were summarized as boxplots.

Based on the RR values, we defined indicators that compare each data-limited scenario with the Sc_REF, which are summarized in Table . Three indicators were used to assess the consistency relative to Sc_REF for both ETCs and EEFs. Variability among chemicals was defined as the 5th–95th percentile of the RR range, with higher values indicating greater variability. The Confidence Interval consistency quantifies the percentage of chemicals whose metric values fall within the confidence interval (CI) of the corresponding metric in Sc_REF. The Quartile alignment evaluates the alignment in the chemical ranking between each scenario and Sc_REF. For this purpose, chemicals were ranked based on their metric values and grouped into quartiles (i.e., Q1 to Q4), with Q1 and Q4 representing the quartiles with the least and the most hazardous compounds, respectively. We assumed that the shift from one quartile to another marks a relevant change in practical decision-making in the innovation context. For ERA, the Percentage of underestimation and Conservativeness within 2 orders of magnitude were added as two additional indicators to assess ETCs’ performance. For LCA, Closeness to the reference within 1 order of magnitude was used to assess EEFs’ performance.

2. Description of the Indicators Evaluated across the Defined Scenarios, Including Their Definition and Corresponding Use in ERA and LCA.

Indicator name	Indicator abbreviation	Description of the indicator	ERA	LCA
Variability among chemicals	5th–95th RR width	5th-95th percentile range of the RRs of different compounds. The lower the range, the better.	Yes	Yes
Confidence Interval consistency	CI consistency	Percentage of chemicals for which ETCs and EEFs fall within the confidence interval of the corresponding value in the reference scenario. The higher the percentage, the better.	Yes	Yes
Quartile alignment	Q alignment	Percentage of chemicals with the same rankingin terms of quartile allocationsto the reference scenario. The higher the percentage, the better.	Yes	Yes
Percentage of underestimation	RR < 0	Percentage of chemicals with underestimated hazard relative to the hazard in the reference scenario. The lower the percentage, the better.	Yes	No
Conservativeness within 2 orders of magnitude	0 < RR < 2	Percentage of overestimated chemicals with hazard within 2 orders of magnitude relative to the hazard in the reference. The higher the percentage, the better.	Yes	No
Closeness to the reference within 1 order of magnitude	|RR|<1	Percentage of chemicals with metric values within 1 order of magnitude relative to the reference. The higher the percentage, the better.	No	Yes

Results

Response Ratios of Ecotoxicological Threshold Concentrations

Figure summarizes the response ratios (RRs) for ETC_LTV and ETC_SSD across all scenarios using NOEC (Figure a) and EC50 (Figure b) data. Results for the full chemical data set are shown in Figure , while MoA-specific patterns are provided in Figure S2. The red dotted line indicates the reference scenario (Sc_REF), where RR > 0 reflects more conservative estimates, and RR < 0 reflects underestimated ETCs.

1.

Boxplots of the RR values (X-axis) for the ERA metrics across the investigated scenarios (Y-axis) with (a) NOEC and (b) EC50 data. The left and right lines of the box are the 25th and the 75th percentiles, respectively, and the line inside the box represents the median (50th percentile). The whiskers represent the 5th and 95th percentiles. The red dotted lines denote RR = 0, representing perfect agreement with the reference scenario. The Y-axis labels indicate the number of chemicals included in the analysis of the variability for each scenario. This number varies because the scenario definitions in Table impose specific conditions that not all chemicals satisfy. [EC50 = Effect Concentration with 50% of the population affected, ETC = Ecotoxicological Threshold Concentration, LTV = Lowest Toxicity Value, NOEC = No Adverse Effect Concentration, RR = Response Ratio].

When ETC is derived with the lowest toxicity value (ETC_LTV), the percentage of underestimation (RR < 0) ranges from 29–52% across the NOEC-based scenarios, with the AFC scenario performing best (Table ). EC50-based scenarios show a similar pattern, with RR < 0 for 16–54% of the chemicals. In both cases, using single-species data markedly increases the percentage of underestimation compared to the corresponding AFC scenario. The conservative RRs for the ETC_LTV mostly remain within 2 orders of magnitude (41–70%). Variability in the RR across chemicals, expressed as the 5th–95th percentile range, is the lowest for the NOEC-based AFC scenario. Single-species scenarios consistently show the widest spread in RRs. EC50-based scenarios follow the same pattern (Table ). CI consistency is relatively stable across scenarios, with 47–55% of ETC_LTV values for all chemicals falling within the CI of the reference ETC. Figure further disaggregates the CI consistency for ETC_LTV values in NOEC-based scenarios. The blue bars represent the 5th–95th percentile range in Sc_REF, while the points show scenario-specific ETC_LTV values by MoA. Single-species scenarios produce more dispersed estimates than Sc_AFC_NOEC, with greater dispersion observed for specifically acting chemicals in Sc_C_NOEC. Narcotic and unclassified chemicals show more compact distributions. Despite their wider spread, specifically acting chemicals exhibit higher CI consistency (60%) than narcotic (40%) or unclassified chemicals (53%) (Tables S3–S5). They also show wider RR distributions (5th–95th width of 2.9) and higher probabilities of underestimation (26–68%) compared with narcotic (13–58%) and unclassified chemicals (0–48%).

3. Summary of Indicators for the Metrics of the Environmental Risk Assessment (ETC) Derived with the Lowest Toxicity Value (LTV) and Species Sensitivity Distribution (SSD) Methods, across the Eight Scenarios^a .

Metric	ETC_LTV				ETC_SSD
Scenario	5th–95th RR Width	CI consistency	RR < 0	0 < RR < 2	5th–95th RR Width	CI consistency	RR < 0	0 < RR < 2
Sc_AFC_NOEC	2.6	53%	29%	70%	3.4	59%	40%	53%
Sc_A_NOEC	4.1	49%	44%	56%	5.6	23%	7%	27%
Sc_F_NOEC	4.2	54%	52%	47%	5.8	19%	9%	27%
Sc_C_NOEC	5.1	55%	52%	48%	7.2	16%	11%	29%
Sc_AFC_EC50	3.1	41%	16%	70%	4.4	51%	41%	45%
Sc_A_EC50	5.3	47%	47%	48%	5.4	22%	10%	26%
Sc_F_EC50	4.3	59%	54%	41%	5.6	21%	5%	34%
Sc_C_EC50	4.2	53%	52%	46%	5.4	12%	5%	41%

^aThe indicators include variability among chemicals (5th–95th RR width), confidence interval (CI) consistency, percentage underestimation (RR < 0), and percentage of conservativeness within two orders of magnitude (0 < RR < 2) relative to the reference scenario.

2.

Confidence interval consistency of the ETC_LTV (x-axis) for the NOEC-based scenarios relative to the reference scenario for the 132 chemicals (y-axis). The blue (circle) dots represent the ETC for the reference scenario (Sc_REF) with the related 5th–95th confidence interval (blue bars). The dots with various colors and shapes represent the ETC_LTV in the NOEC-based scenarios, differentiated by their MoA. Hazardous chemicals go from lowest (top) to highest (bottom) ETC_LTV.

When ETC is derived using SSDs (ETC_SSD), the percentage of underestimation is generally lower compared to that of ETC_LTV, ranging from 7–40% for NOEC-based scenarios and 4–41% for EC50-based scenarios. In contrast to ETC_LTV, single-species scenarios show the lowest percentage of underestimation (4–11%), while the percentage of conservative RRs within 2 orders of magnitude is substantially lower (27–45%) than for ETC_LTV (41–70%). The ETC_SSD exhibits slightly higher variability in the RR across chemicals than does the ETC_LTV. CI consistency is more variable (12–59%), and single-species scenarios show less than 50% agreement within the reference CI. When ETC_SSD is derived, the percentage of underestimation decreases markedly for specifically acting chemicals (0–47%), but this improvement is also accompanied by lower CI consistency (14–19%) and more frequent overestimation (Figure S4 and Table S4). Similar reductions in underestimation are observed for unclassified (0–44%) and narcotic chemicals (4–38%). Variability increases substantially with ETC_SSD, especially for specifically acting chemicals, which show very wide RR distributions (3.5–9.8). Narcotic (2.6–3.6) and unclassified chemicals (1.9–4.7) show spreads comparable to the full data set. Patterns are consistent between NOEC- and EC50-based scenarios (EC50 results in Figure S3).

Response Ratios of Ecotoxicity Effect Factors

Figure presents RRs for EEF5, EEF20, and EEF50 across all scenarios using the NOEC (Figure a) and EC50 (Figure b) data. The red dotted lines mark |RR| < 1 to the reference scenario (Sc_REF), which indicates the closeness to the reference within 1 order of magnitude. Results are reported for the full chemical data set, while MoA-specific results are presented in Figure S6.

3.

Box plots of the RRs (X-axis) for the LCA metrics across the scenarios (Y-axis) with (a) NOEC and (b) EC50 data plus extrapolation. The bottom and top lines of the box are the 25th and the 75th percentiles, respectively, and the line inside the box represents the median (50th percentile). The whiskers represent the 2.5th and 97.5th percentiles. The red dotted lines denote |RR| < 1, representing agreement between data-limited scenario results and the data-rich reference scenario within 1 order of magnitude. The Y-axis labels indicate the number of chemicals included in the analysis of the variability for each scenario. This number varies because the scenario definitions in Table impose specific conditions that not all chemicals satisfy. [EC50 = Effect Concentration 50%, EEF = Ecotoxicity Effect Factor, NOEC = No Adverse Effect Concentration, RR = Response Ratio].

Focusing on EEF20, NOEC-based EEFs yield 63–89% of chemicals within 1 order of magnitude of the reference, whereas EC50-based scenarios show lower values (49–71%). The AFC scenario consistently produces the highest percentage of chemicals with |RR| < 1. Among single-species scenarios, Sc_A performs worst, with only 63% (NOEC) and 49% (EC50) of chemicals within 1 order of magnitude. MoA stratification (Tables S6–S8) reveals consistent differences among chemical groups. Specifically acting chemicals have a lower percentage of being within 1 order of magnitude of the reference (44–78%) than unclassified (53–96%) and narcotic chemicals (55–93%), with the largest deviations in Sc_C_NOEC (Figure ). These MoA differences are reduced in the EC50-based scenarios (Figure S7) and are observed similarly for EEF5 and EEF50 (Figures S8–S9). Variability among chemicals, expressed as the 5th–95th percentile width of the RR distribution, ranges from 1.8 to 4.6 across scenarios. EC50-based scenarios generally exhibit greater variability than their NOEC counterparts, except in Sc_C. Variability by MoA also differs: specifically acting chemicals show substantially wider RR distributions (5th–95th width 1.9–8.9), whereas narcotic (1.4–3) and unclassified chemicals (1.3–5) show narrower distributions. No systematic differences between NOEC- and EC50-based scenarios are evident across the three MoA categories. CI consistency for NOEC-based scenarios ranges from 46% to 86% (Table ); EC50-based scenarios follow a similar pattern except for Sc_AFC, where CI consistency decreases from 86% (NOEC) to 66% (EC50).

4.

Confidence interval consistency of the EEF20 (x-axis) for the NOEC-based scenarios relative to the reference scenario for the 132 chemicals (y-axis). The blue (circle) dots represent the EEF20 for the reference scenario (Sc_REF), and the dots with various colors and shapes represent the EEF20 in the scenarios, differentiated by their MoA. Hazardous chemicals go from highest (top) to lowest (bottom) EEF20.

4. Summary Statistics of the Response Ratio (RR) Distributions for the Three Ecotoxicity Effect Factors (EEF5, EEF20, and EEF50) across the Eight Scenarios .

Metric	EEF5			EEF20			EEF50
Scenario	5th–95th RR Width	CI Consistency	|RR| < 1	5th–95th RR Width	CI Consistency	RR| < 1	5th–95th RR Width	CI Consistency	|RR| < 1
Sc_AFC_NOEC	3.5	76%	69%	1.8	86%	89%	2.4	76%	87%
Sc_A_NOEC	3.5	54%	57%	3.5	46%	63%	3.3	38%	73%
Sc_F_NOEC	3.6	58%	57%	3.5	48%	70%	3.5	44%	76%
Sc_C_NOEC	4.5	58%	70%	4.6	54%	70%	4.6	61%	70%
Sc_AFC_EC50	4.3	58%	51%	2.5	66%	71%	2.2	48%	86%
Sc_A_EC50	4.2	48%	44%	4.2	48%	49%	4.3	40%	58%
Sc_F_EC50	3.8	59%	58%	3.7	51%	64%	3.7	38%	67%
Sc_C_EC50	3.8	58%	61%	3.6	54%	68%	3.5	47%	72%

^aThe indicators include variability among chemicals (5th–95th RR width), confidence interval (CI) consistency, and closeness (|rr| < 1) relative to the reference scenario.

Shifting the effect percentile alters this pattern systematically. Deriving EEF5 reduces the percentage of chemicals within 1 order of magnitude compared to the reference, while deriving EEF50 increases that percentage (Table ). Variability is broadly comparable between EEF5 and EEF50, although Sc_AFC_EC50 displays a wider range for EEF5 (4.3) than for EEF50 (2.2), suggesting greater stability at higher percentiles (Table ). Conversely, CI consistency is generally higher for EEF5 than for EEF20 or EEF50, with exceptions in Sc_AFC_NOEC, Sc_AFC_EC50, and Sc_C_NOEC. These shifts reflect consistent changes in RR distributions: EEF5 produces more conservative estimates, while EEF50 shifts results toward less-conservative values (Figure ). Notably, Sc_C_NOEC shows stable performance across EEF5, EEF20, and EEF50, with similar percentages of chemicals within 1 order of magnitude, indicating relatively robust SSD behavior compared with Sc_A_NOEC and Sc_F_NOEC (Table ).

Variation in the Chemical Ranking

Figure illustrates the variation in the chemical ranking when EEF20 is derived under the data-limited scenarios (right side of each Sankey diagram) relative to the reference scenario (left side). Rankings are expressed by quartile (Q1: most hazardous; Q4: least hazardous). Differences in quartile allocation between data-limited and reference scenarios (marked as “up” and “down” in the figure) indicate the probability of misclassifying priority chemicals. The percentage of chemicals retaining the same quartile allocation is shown on the right side of each diagram. Analogous results for other LCA metrics and ERA metrics are reported in Tables S11–S12.

5.

Shift in the chemical ranking from one quartile to another quartile for the EEF20 (LCA metric), against the reference scenario (Sc_REF). Q1 represents chemicals with a relatively high hazard, and Q2, Q3, and Q4 represent lower hazards. Scenario-related changes are shown by shifts from data-rich reference assessments (left) to the data-limited scenarios (right). No shifts (marked in blue) imply that hazard rankings of chemicals are insensitive to data-limited conditions. The percentage of no shifts compared to the reference is reported on the right side for each quartile of each scenario, and the percentage in bold represents the average in the scenario. Higher-hazard shifts (marked in green) represent chemicals for which the hazards are estimated as higher under data-limited conditions. Low-hazard shifts (marked in green) represent the opposite.

For EEF20, the percentage of chemicals with unchanged quartile allocation ranges from 39% (Sc_A_EC50) to 67% (Sc_AFC_NOEC), with only Sc_AFC_NOEC and Sc_F_NOEC exceeding 50% consistency (Figure ). Most ranking changes occur between adjacent quartiles: the percentage of chemicals shifting by at most one quartile lies between 30% and 52%, while the remainder corresponds to shifts of two (2–14%) or more quartiles (0–5%) (Table S9). Deriving EEF5 or EEF50 produces modest reductions in ranking consistency: the percentage maintaining the same quartile ranges from 38% to 58% for EEF5 and from 38% to 59% for EEF50 (Table S9). Comparable patterns are observed for ERA metrics (ETC_LTV and ETC_SSD; Table S10). Across metrics and scenarios, Sc_A_EC50 exhibits the largest percentage of misclassification, with the highest percentage of chemicals shifting by two or more quartiles. Overall, data-limited scenarios increase the risk of misclassifying priority chemicals, and the magnitude of ranking variation depends on both the effect percentile used to derive the EEF and the underlying data scenario.

Discussion

Key Findings for Environmental Risk Assessment

We quantified how alternative data-limited scenarios affect the probability of under- or overestimating the chemical hazard compared to a data-rich situation. We found that scenarios that preserve taxonomic coverage across three trophic levels (algae-fish-crustaceans, AFC) are the most consistent with the data-rich reference scenario, indicating that maintaining taxonomic coverage is an important driver of robust hazard estimates. When the three trophic groups are available, the lowest toxicity value (LTV) approach produces lower uncertainty than the Species Sensitivity Distribution (SSD) approach, while end point (NOEC vs EC50) has limited influence on the ETC estimates. These findings are consistent with previous work showing that SSDs derived from acute toxicity data across three trophic levels can approximate SSDs from more comprehensive data sets, , and with the observation that LTV methods often yield narrower uncertainty bounds than SSD methods. At the same time, our results echo warnings that acute data are not universally suitable for deriving chronic values for some chemical groups. For groups of chemicals with distinct chronic and acute modes of actionfor example, pharmaceuticalsacute data often do not predict chronic effects reliably, and extrapolation can be misleading. , In such cases, it is advisable to avoid relying on extrapolated acute data or to apply MoA-specific adjustments and sensitivity checks.

When only single-species data are available, ETC_LTV tends to underestimate the hazard, increasing the risk of failing to identify priority chemicals. By contrast, ETC_SSD is highly conservative with single-species data, reducing false negatives but generating a large number of substantial ETC overestimates (more than 2 orders of magnitude). This conservativeness of ETC_SSD in data-limited situations results from inflating the SSD standard deviation by using the 95th percentile of the reference scenario interspecies variability across the chemicals. Thus, ETC_SSD is appropriate when the screening prioritizes precautionary flagging, but it can produce many false positives that may distort prioritization and resource allocation in an innovative context. MoA differentiation clarifies where these trade-offs matter the most. Specifically acting chemicals show substantially greater interchemical variability and sensitivity to methodological choice than narcotic or unclassified chemicals. Under single-species ETC_LTV, specifically acting chemicals are more likely to be underestimated because default assessment factors do not capture their higher interspecies sensitivity. ,, Conversely, ETC_SSD captures this variability through larger SSD standard deviations and wider confidence intervals (the 95th percentile SSD standard deviation for specifically acting chemicals in our data set is markedly higher than that for narcotics or unclassified chemicals), thereby reducing underestimation but increasing the likelihood of large overestimates. These results support recent proposals to tailor assessment factors to chemical MoA, reducing systematic bias , and aligning with empirical evidence of greater interspecies variation for specifically acting chemicals. ,

Comparison with data-rich PNEC values demonstrates the regulatory relevance of our assessment. In a subset of 73 chemicals meeting ECHA data requirements, ETC_LTV reproduced PNEC values for 34–52% of the chemicals. For the chemicals with different hazard estimates, ETC_LTV underestimated the PNEC by up to ∼2.6-fold (Figure S5). This comparison quantifies the potential loss of safety margin when relying on LTV under data-limited conditions and indicates the possibility of exploring additional safety factors to preserve acceptable protection levels.

Key Findings for Life Cycle Assessment

Our analysis of LCA metrics focuses on the stability of relative differences among chemicals rather than absolute conservativeness. Effect factor working points, therefore, need to preserve rank order under data-limited circumstances to support comparative assessment and prioritization. Across our scenarios, those that preserve coverage of three taxonomic groups (algae-fish-crustacean, AFC) are most closely aligned with the data-rich reference scenario. AFC scenarios minimize variability in EEF estimates and produce the highest agreement in quartile-based rankings. The NOEC-based AFC scenario showed lower uncertainty than the corresponding EC50 scenario, supporting current LCA practices of deriving EEF20 from chronic and extrapolated acute data. As with the ERA metrics, however, the reliability of acute-to-chronic extrapolation is chemical-domain-dependent, and its suitability must therefore be judged with particular caution for groups where acute and chronic effects are poorly correlated.

Comparing EEF5, EEF20, and EEF50 highlights a trade-off between robustness and protection, the choice of which depends on the decision context. EEF50 corresponds to the most robust point of the SSD and therefore maximizes ranking robustness, while EEF5 is more precautionary and aligns better with ERA practice but increases uncertainty because it covers the SSD tails. While EEF20 is a pragmatic compromise and is the recommended working point in LCA, , our results show that the choice of working point can influence prioritization outcomes (Figure S10). The crustacean-only NOEC scenario is a notable exception, showing stable alignment with the reference across all three working points.

MoA effects are also observable in LCA, but they are less pronounced than in ERA because the assumed SSD standard deviations in the single-species scenarios are narrower (median sd ∼1.20 for specifically acting, 0.89 for narcotic, 0.87 for unclassified chemicals). Nevertheless, specifically acting chemicals still show larger EEF variability and a greater potential for misclassification. This supports the value of incorporating MoA information where available. ,

Application Potential for Safe and Sustainable by Design

In early-stage screeningparticularly within the EU’s Safe and Sustainable by Design (SSbD) frameworkregulatory data requirements are rarely met, and the primary goal is the identification of major concerns, trade-offs, or patterns that can guide subsequent data generation for chemical innovation. , To support such data-limited assessment, we propose a preliminary practical tool (in S4) that operationalizes the request of this study and helps address the initial bottleneck posed by the limited ecotoxicity data. The tool consists of three interconnected building blocks (Figure ): goal and scope (e.g., screening objectives, acceptable conservativeness, chemical information), input ecotoxicity data (e.g., type and number), and methods. The goal and scope include chemical information, such as the MoAif knownand explicit decision rules on acceptable probabilities of over or underestimation. These elements typically require expert input.

6.

Structure of the tool to define possible entry points for early-stage screening of chemical alternatives is based on three interconnected building blocks: goal and scope, input ecotoxicity data, and methods. Knowledge of any two building blocks enables refinement of the third using the Excel-based tool in Supporting Information S4 (represented by circular icons).

Combining the goal and scope with the available input ecotoxicity data, the tool directly identifies the suitable method in the scenario combination(s). Predefining the goal and scope and the methods for the assessment determines the input ecotoxicity data requirements, while fixing input ecotoxicity data together with the method determines the possible goal and scope (examples of application are reported in S3.6). This flexibility reflects the diverse entry points and constraints typical of early-stage chemical (re)­design.

It is important to note that when using the tool to identify the entry points for a screening assessment, the additional probability of under/overestimation of the estimated metric may be affected when ecotoxicity data are derived from predictive tools. Advances in QSAR modeling, machine learning, and artificial intelligence might enable the derivation of LCA/ERA metrics from minimal experimental data. , However, the integration of predictive data may alter the uncertainty profiles identified in the present study, which, hence, should preferably be explicitly evaluated when using a scenario.

The proposed approach provides a structured basis for selecting entry points in SSbD screenings. However, this approach is not intended to replace higher-tier regulatory assessment but is rather intended to support transparent, context-dependent SSbD decisions, where uncertainty is unavoidable and expert judgment remains essential. Moreover, assessors should tailor method choices and uncertainty to the specific decision context, always keeping expert judgment as essential , not only to select appropriate assumptions and pathways but also to interpret uncertainty and ensure that screening outcomes are used transparently in SSbD assessment and decision-making.

For the implementation of the tool and overall use of the scenario in screening within the SSbD context, the choice of the scenario should match the screening objective. In rankings, regulatory alignment, and comparative LCA, the algae-fish-crustacean combination could be preferred because it best reproduces data-rich behavior and minimizes rank shifts. Single-species scenarios are suggested to be used only as a preliminary step to flag outcomes at coarse performance classes (quartiles) and not as fine-grained ranks. MoA-informed metrics should be applied where they are known. When MoA is unknown, conservative estimates could be considered, or MoA could be derived through predictive software. , Finally, chemical domains (e.g., pharmaceuticals) should be flagged where acute-to-chronic extrapolation is unreliable. All in all, our approach can support the safe and sustainable design of chemicals when large uncertainties are unavoidable.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.6c02685.

• Complete data set of the investigated chemicals and the main results (S1, Excel) (XLSX)

• Excel-based tool to identify potential entry points for the screening assessment (S4, Excel) (XLSX)

• Description of the methodology used for the derivation of the LCA and ERA metrics for each scenario, including the construction of the SSD curve and assumptions introduced (S2, PDF) and complete representation of the obtained results of the analysis of the variation across scenarios in terms of metrics (for ERA and LCA) and in terms of ranking (S3, PDF) (PDF)

#.

S.H., L.P., and A.M.J.R. contributed equally to this work.

The present study has been developed in the context of the Collaborative Doctoral Partnership Agreement No. 35334 between the European Commission Joint Research Centre and Radboud University.

The authors declare no competing financial interest.