Combining In Silico Tools with Multicriteria Analysis for Alternatives Assessment of Hazardous Chemicals: Accounting for the Transformation Products of decaBDE and Its Alternatives

Transformation products ought to be an important consideration in chemical alternatives assessment. In this study, a recently established hazard ranking tool for alternatives assessment based on in silico data and multicriteria decision analysis (MCDA) methods was further developed to include chemical transformation products. Decabromodiphenyl ether (decaBDE) and five proposed alternatives were selected as case chemicals; biotic and abiotic transformation reactions were considered using five in silico tools. A workflow was developed to select transformation products with the highest occurrence potential. The most probable transformation products of the alternative chemicals were often similarly persistent but more mobile in aquatic environments, which implies an increasing exposure potential. When persistence (P), bioaccumulation (B), mobility in the aquatic environment (M), and toxicity (T) are considered (via PBT, PMT, or PBMT composite scoring), all six flame retardants have at least one transformation product that can be considered more hazardous, across diverse MCDA. Even when considering transformation products, the considered alternatives remain less hazardous than decaBDE, though the range of hazard of the five alternatives was reduced. The least hazardous of the considered alternatives were melamine and bis(2-ethylhexyl)-tetrabromophthalate. This developed tool could be integrated within holistic alternatives assessments considering use and life cycle impacts or additionally prioritizing transformation products within (bio)monitoring screening studies.


Methodology of the two MCDA methods
There are two major kinds of MCDA methods: synthesizing criterion methods and synthesizing preference relational systems. 1 Synthesizing criterion methods, including the multi-attribute utility theory (MAUT) 2 , require all criteria to be converted into comparable scales. A trade-off weighting factor would be assigned to each criterion representing the relative significance for aggregation.
Synthesizing preference relational systems, such as the Elimination Et Choix Traduisant la REalité (ELECTRE III) method 3 , conduct pair-wise comparison for each criterion, and present a matrix showing the extent to which an alternative outranks the other alternatives with importance coefficients representing the significances of different criteria.
In this study for the MAUT approach, each criterion was scaled from 0 (worst) to 1 (best) based on the average result of all models for that criterion. For each criterion, the 0 (worst) and 1 (best) levels were set based on the distance between the worst case of studied chemicals to an "Utopia" level, or a set most hazardous level to the best case of studied chemicals. After the scaling, partial scores for P, B, T, and M properties were calculated by multiplying each criterion relevant to an aspect with the same weight. Final scores were calculated by treating the composite criteria of either PBT, PMT or PBMT as equally important (Table S10).
For ELECTRE III, the calculation was done in a manner consistent with other publications. 1,4,5 There are three important thresholds for each criterion in ELECTRE III: indifference thresholds (q), preference thresholds (p) and veto thresholds (v). In practical terms, when the difference between alternatives A and B with respect to criterion j is less than q, the weight of criterion j is not considered in the comparison between A and B, and if the difference is larger than p, the full weight of criterion j should be awarded to the superior alternative. A sliding scale exists between p and q, but if the difference is as large as a defined v, the ELECTRE method eliminates the underperforming alternative from contention. 3,4 In this study, q and p were set based on data uncertainties, which was done in our previous study 6 by considering the variation of modelling outputs from different models for the same criteria. After setting the thresholds, pair-wise comparisons were conducted on each assessment criterion. Final scores were calculated by treating PBT or PBMT equally important and presented in Table S11-S24.

S4
Orange ( Green The cut-off values between yellow and green for persistence were used for Figure 1, Stage C to prioritize persistent transformation production c For MAUT, each criterion were scaled from 0 (worst) to 1 (best). As was done in the previous study 6 , for classification T criteria, the scale was set from all models give positive results (marked as "P" in the table) to all negative results (marked as "N"); for P and B criteria, the scale was set from our worst chemical (marked as "max") to an ideal level; for M and quantification T criteria, the scale was set from a set worst level to our best chemical (marked as "max"). MAUT calculations are presented in Table S7. d For ELECTRE III, there are three important thresholds for each criterion: indifference thresholds (q), preference thresholds (p) and veto thresholds (v). A difference between alternatives A and B less then q means the weight of this criterion is not considered in the comparison between A and B, and if the difference is larger than p, full weight should be awarded. A sliding scale exists between p and q, but if the difference is as large as v, the ELECTRE method eliminates the underperforming alternative from contention.These thresholds were set according to the previous study 6 considering model uncertainties. ELECTRE III calculations are presented in Table S8-S21. High occurrence frequency was defined as not lower than "likely" for CTS, given a Site of Metabolism (SOM) score of more than 300 by Meteor Nexus, or not lower than "neutral" for EAWAG-BBD/PPS. For 2 nd and 3 rd generations, only trasnformation products both transformed from compounds with high occurrence frequency and fit the cut-offs of high occurrence frequency themselves were difined as with high occurrence frequency. b Replicates not removed S8   Table S5. Structures of transformation products with high occurrence potential selected for the alternatives assessment  Yes Yes ammelide [8][9][10] Yes Yes cyanuric acid [8][9][10] Yes Yes NH3 8,9 Yes No CO2 8,9 No No Microbial (soil and water) biuret 8,9 Yes Yes allophanate 8,9 No No Nitrification in soil nitrate 8,11 No  15,[18][19][20] No No Aquatic plants tetraBDE 15 No No Aquatic plants triBDE 15 No No Photodegradation polybrominated dibenzofurans (PBDFs) 17,21 Table S3; for logDow, color are marked as: red=logDow<6; orange=logDow in the range of 6-7.5; yellow=logDow in the range of 7.5-10; green= logDow>10 Figure S1. Heat map of the transformation products where red means that the transformation product is considerably worse than its parent compound for each hazard criteria, yellow means that the hazard level is similar, and green means that the transformation product is considerably better than its parent compound. A "considerable difference" was defined as the difference between transformation product and its parent compound larger than the veto thresholds in ELECTRE III (Table S2).