n Impact on three Defined Approaches (DA)

Three defined approaches were chosen (Figure 5) and the impact of the interchangeability of the inchemico and in vitro assays assessed. The calculated interchangeability values of 85% and 74% forKE2 and KE3 were used to estimate the probability that choosing another assay covering the samekey event would result in a different overall prediction from each DA (an example of the calculation isshown for the BASF 2 / 3 DA in Table 1, with predictions that have changed highlighted in purple).

The interchangeability of in vitro key event-based skin sensitisation assays and how it impacts the uncertainty of predictions from defined approachesMartyn Chilton & Donna Macmillan Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS

References: 1. OECD. The Adverse Outcome Pathway for Skin Sensitisation Initiated by Covalent Binding to Proteins Part 1: Scientific Evidence. 2. OECD. Test No. 442C. 3. OECD. Key Event Based Test Guideline 442D. 4. OECD. Key Event Based Test Guideline 442E. 5. Urbisch, D. et al. Regul. Toxicol. Pharmacol. 71, 337–351 (2015). 6. Takenouchi, O. et al. J. Appl. Toxicol. 35, 1318–1332 (2015). 7. Macmillan, D. S. & Chilton, M. L. Regul. Toxicol. Pharmacol. 101, 35–47 (2019).

n IntroductionDue to a global move away from animal tests to predict thetoxicity of chemicals there are a growing number of OECD-validated in chemico and in vitro assays which measure skinsensitisation. These assays are linked to particular key events inthe Adverse Outcome Pathway (Figure 1) which outlines thesequence of events required to elicit skin sensitisation initiatedby covalent binding to proteins1.

Currently there are OECD Test Guidelines for the Direct PeptideReactivity Assay2 (DPRA) which measures the molecularinitiating event (covalent binding to proteins), the KeratinoSens™and LuSens assays3 which measure key event 2 (keratinocyteactivation), and the h-CLAT, U-SENS™ and IL-8 Luc assays4

which measure key event 3 (dendritic cell activation). Currentlythere are no in chemico or in vitro assays designed to measurekey event 4 or the adverse outcome.

Where multiple assays cover the same key event, there is thequestion of how often assays produce concordant results, as wellas well as considering the intrinsic reproducibility associated withany scientific experiment. This concordance, known asinterchangeability, will impact the uncertainty of both theindividual assay predictions as well as any predictions resultingfrom defined approaches that are based on these test results.

n Method

Figure 1: The skin sensitisation Adverse Outcome Pathway (AOP) and OECD-validated assays which measure specific key events (KE) in the AOP.

Figure 2: Number of chemicals with data for OECD-validated assays in the in chemico / in vitro dataset used for interchangeability analysis. Dataset compiled from publicly available data.

BASF 2 / 35 Lhasa Defined Approach7Kao Sequential Testing Strategy (STS)6

n Interchangeability - key event 2 (KE2)

n Interchangeability - key event 3 (KE3)

• 88 compounds have data for both U-SENS™ and IL-8 Luc (Figure 4)

• 92% of those have the same outcome in each assay

• 8% (7 chemicals) produce different outcomes

• U-SENS™ is more predictive than IL-8 Luc when compared to in vivo data

• 118 compounds have data for both h-CLAT and IL-8 Luc (Figure 4)

• 78% of those have the same outcome in each assay

• 22% (26 chemicals) produce different outcomes

• Both assays predict around half of these correctly when compared to in vivo data

• 139 compounds have data for both h-CLAT and U-SENS™ (Figure 4)

• 74% of those have the same outcome in each assay

• 26% (32 chemicals) produce different outcomes

• Both assays predict around half of these correctly when compared to in vivo data

Figure 3: Venn diagram illustrating the overlap of KeratinoSens™ and LuSens data and the interchangeability of the assays. For discordant results, the table indicates which assay correctly (green), or incorrectly (red) predicts the in vivo outcome. Orange cells indicate that the assay is correct for one species but not another.

Figure 4: Venn diagram illustrating the overlap of h-CLAT, U-SENS™ and IL-8 Luc data and the interchangeability of the assays. For discordant results, the table indicates which assay correctly (green), or incorrectly (red) predicts the in vivo outcome. Orange cells indicate that the assay is correct for one species but not another..

• 72 compounds have both KeratinoSens™ and LuSens data (Figure 3)• 85% of those have the same outcome in each assay• 15% (11 chemicals) have different outcomes• KeratinoSens™ predicts slightly better than LuSens when compared to LLNA data and human data,

respectively

Assay results KE2 = constantKE3 = constant

KE2 = changeKE3 = constant

KE2 = constantKE3 = change

KE2 = changeKE3 = change Overall probability of a change in

predictionKE1 KE2 KE3 Probability Prediction Probability Prediction Probability Prediction Probability Prediction

+ + +

0.85

* 0.

74 =

0.6

29

+

0.15

* 0.

74 =

0.1

11

+

0.85

* 0.

26 =

0.2

21

+

0.15

* 0.

26 =

0.0

39

– 0.039

+ + – + – + + 0.111

+ – + + + – + 0.221

– + + + – – – 0.111 + 0.221 + 0.039 = 0.371

+ – – – + + + 0.111 + 0.221 + 0.039 = 0.371

– + – – – + – 0.221

– – + – + – – 0.111

– – – – – – + 0.039

Average: 0.186

n Conclusions• The interchangeability of assays measuring KE2 and KE3 are high (85%, KE2; 74% - 92%, KE3)

for the relatively small dataset used in this analysis.• When comparing the discordant results against in vivo data none of the assays significantly out-

perform each other. • The probability that a prediction from the BASF 2 / 3 would result in a different outcome when using

another assay is 19%, for the Kao STS it is 13% and for the Lhasa DA it is 11%.• The cursory analysis described here provides an illustration of the additional uncertainty that

interchangeability can add to three well-known DAs.

n Further work• The data will be analysed further and results published:

• Can applicability domain information from assays covering the same key event be gained from the analysis of this dataset?

• Does covering multiple key events confer additional confidence in skin sensitisation predictions as opposed to multiple assays covering a single key event?

• How does the assay order impact on the outcomes of defined approaches?

Figure 5: Defined approaches chosen to assess the impact of interchangeability.

Two metrics were then calculated: the average probability thata prediction would change across all possible combinations ofassay results, and the worst-case probability of a predictionchanging for a single combination of assay results. These rangefrom 11% - 19% and 26% - 37% across the DAs (Table 2).

Table 1: Assessing the impact of assay interchangeability on a defined approach.

Table 2: Average and worst-case probabilities that assay interchangeability would change the overall prediction of a defined approach.

Defined approach Average probability

Worst-case probability

BASF 2 / 3 0.186 0.371

Kao STS 0.130 0.260

Lhasa DA 0.108 0.351

Abstract ID: 0162