how to apply expert review to ec3 predictions in derek … llna ec3 % similarity % sensitiser...

Derek Nexus

SkinSensitisation

How to apply expertreview to EC3 predictions in

Defined as the process of a chemical, known as a hapten, causing an allergic reaction which leads to allergic contact dermatitis (ACD)1.

ACD develops in 2 stages:

Stage 1 is induction where the hapten forms a complex with the skin protein and initiates a cascade ending with proliferation of allergen specific T cells.

Stage 2, elicitation, occurs after subsequent contact with the same allergen. This leads to the hapten-skin complex triggering the same allergen specific T cells, inducing inflammatory cytokines and ACD.

This mechanism of skin sensitisation has been studied extensively and has a well-defined Adverse Outcome Pathway (AOP)2.

An AOP is the sequence of events that leads to a particular in vivo outcome of interest and the following AOP (Figure 1), published by the OECD in 2012, covers each event in the pathway of skin sensitisation caused by covalent binding of chemicals to skin proteins in detail.

There are 4 key events. The 1st key event, the molecular initiating event (MIE), is the binding of the chemical hapten to skin protein. The 2nd and 3rd key events are cellular responses by keratinocytes and dendritic cells. The 4th key event is defined as the organ response where T-cells are activated and then finally the adverse outcome of skin sensitisation is reached.

Skin sensitisation

ADME Cellular response Organ

responseAdverseoutcome

Molecular Initiating

Event (MIE)

Penetration of test chemical

Transformation of test chemical

1st key event 2nd key event 4th key event 3rd key event

Lipophilicity Pre-hapten

Pro-hapten

Hapten binding Stress response T-cell activationActivation of DC Skin sensitisation

Figure 1. Skin sensitisation Adverse Outcome Pathway (adapted from OECD 2012)2.

3

The murine local lymph node assay (LLNA)3 and guinea pig maximisation test (GPMT)4 are the most widely used in vivo assays to identify sensitising chemicals. However, some chemicals are intrinsically more potent than others and it has been suggested that factors such as chemical reactivity and hydrophobicity may play a role in determining the potency of specific classes. Consequently, differentiating only between sensitising and non-sensitising chemicals can lead to sub-optimal risk assessments, which in some cases may be over-protective or too lax.

In this respect, the LLNA offers a significant advantage over the GPMT. While the GPMT is typically used only to identify chemicals that represent a sensitising hazard, the LLNA provides information on the potency of an identified sensitiser, thereby permitting an assessment of both hazard and potency. Potency is inferred from the LLNA using the test concentration required to cause a 3-fold increase in the stimulation index (SI) relative to vehicle controls which is referred to as the EC3 value5. EC3 values have proven reliable measures of potency and several studies have shown a good correlation between EC3 values and known sensitisation potency in humans6-8.

Two EC3 classification schemes are in common use (Table 1 and Table 2): the European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) and a Globally Harmonized System of Classification and Labelling (GHS). Both of these classification schemes assign categories to chemicals based on their potency (EC3 value) to aid risk assessment.

In vivo assays

EC3 classifications

GHS Classification

ECETOC Classification

EC3 (%) Category

Weak sensitiser≥10

Moderate sensitiser≥1-<10

Strong sensitiser≥0.1-<1

Extreme sensitiser<0.1

EC3 (%) Category

1A - Strong sensitiser≤2

1B - Other sensitiser>2

Table 1 and Table 2. ECETOC and GHS EC3 classification systems.

in silico model(s)

in chemico assay(s)

in vitro assay(s)

Prediction of hazard potential and / or potency

Data Integration Procedure

Despite the success of the LLNA to detect skin sensitisers, regulatory changes and animal welfare concerns have resulted in a significant drive to develop non-animal alternatives. As of early 2017, there were three in chemico/in vitro assays for skin sensitisation validated by ECVAM and associated Test Guidelines from the OECD published:

• DPRA9 - an in chemico assay which assesses the 1st key event, the molecular initiating event, in the skin sensitisation AOP

• KeratinoSens™10 - an in vitro assay which assesses the 2nd key event in the skin sensitisation AOP

• h-CLAT11 - an in vitro assay which assesses the 3rd key event in the skin sensitisation AOP

However, it is generally accepted that no single non-animal assay will fully replace in vivo tests and the focus has now turned to combining results from multiple in chemico and in vitro assays and in silico tools in Integrated Testing Strategies (ITS, also known as Data Integration Procedures (DIP)) (Figure 2). Lhasa Limited has published on using Derek Nexus as part of an in silico/in chemico/in vitro ITS to predict skin sensitisation hazard (sensitiser/non-sensitiser)12 and is working to improve its skin sensitisation endpoint by developing in silico predictions of skin sensitisation potency.

Non-animal assays and ITS/DIP

Figure 2. Schematic of ITS/DIP (Integrated Testing Strategies/Data Integration Procedures).

The model is built on a dataset consisting of over 1000 high quality LLNA studies.

The initial prediction given by the model may be improved by applying expert review for the following reasons:

• There is significant variability associated with the LLNA (approx. 3-fold).

• The alert domain used to calculate the alert may have a wide spread of EC3 values.

• An outlier may have a disproportionate effect on the EC3 prediction

• A subset of the alert domain may be more relevant to a prediction of a specific query compound.

The model utilises a k-Nearest Neighbours (kNN) approach. Initially, both compounds (query and dataset) must activate the same skin sensitisation structural alert in Derek to be considered as a Nearest Neighbour. These Nearest Neighbours are then arranged by Tanimoto similarity generated by an in-house radial fingerprinting method. The minimum number of neighbours required is 3, otherwise no prediction is given, and the most similar neighbours, up to 10th place, are considered (Figure 3). The predicted EC3 value is the weighted average of all the valid neighbours. This approach has been validated using an internal test set (n = 45) and an external data set of proprietary member data (n = 103)13.

An ideal in silico tool would provide reliable predictions of skin sensitisation hazard but also be able to accurately predict potency. With this goal in mind, an EC3 prediction model has been developed based on publicly available LLNA studies and mechanistic (alert) domains as provided by Derek Nexus, an expert knowledge base system for toxicity predictions developed by Lhasa Limited.

This workbook provides a number of worked examples of performing expert assessment on EC3 predictions in Derek Nexus. All expert calls made on the examples are solely the opinion of experts at Lhasa Limited and are made here for guidance purposes only.

Expert assessment of EC3 predictions

Figure 3. Example of 10 Nearest Neighbours used in a typical EC3 prediction.

Expert assessment can help improve the quality of EC3 predictions given by Derek Nexus.

10

100

1EC3 %

0.1

0

Weak sensitiser

Moderate sensitiser

Tanimoto score

Strong sensitiser

Extreme sensitiser

Nearest Neighbours NOTused in EC3 prediction

10 Nearest Neighbours used in EC3 predictions

Nearest neighbours Query compound

In silico EC3 predictions using Derek

6

Example

(E)-1-(cyclohex-1-en-1-yl)but-2-en-1-one

3.13 (BioByte Corp, version 5.9)

150.22 (Lhasa Limited, version 1.0)

480: alpha,beta-Unsaturated ketone

Look at EC3 data in the alert activated by the query compound

Highlight similar/dissimilar compounds and/or outlier(s)

Analyse compound(s)

Suggest addition/removal from EC3 calculation based on chemical and mechanistic knowledge

Assess new EC3 prediction

0.0048%

O=C(/C=C/C)C1=CCCCC1SMILES

logP

MW

Alert

Predicted EC3

01

L

H

A

S

A

7

LLN

A E

C3

%

Similarity %

Sensitiser Non-sensitiser Not Used Predicted

1E-40.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

0.001

0.01

0.1

1

10

100

LLN

A E

C3

%

Similarity %


1E-40.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

0.001

0.01

0.1

1

10

100

Highlight similar/dissimilar compounds and/or outlier(s) Majority of compounds in alert space have an EC3 value between 1% - 10%.

One compound in particular is significantly more potent than any other compound in the alert space (EC3 = 3x10-4%).

Look at EC3 data in the alert activated by the query compound 13 data points in alert space.

EC3 values dispersed over multiple ECETOC and GHS categories.

Number of similar compounds used in the calculation: 10/13Predicted LLNA EC3: 0.0048% (extreme sensitiser) - [Derek EC3 Model - 1.0.6]

Number of similar compounds used in the calculation: 10/13Predicted LLNA EC3: 0.0048% (extreme sensitiser) - [Derek EC3 Model - 1.0.6]

8

Similar Compound

Comment:

LLNA EC3: 3.0E-4% (extreme sensitiser)

Similarity to Q compound: 22%

Source: LhasaIndex: 673Structure ID: Structure-673File: Lhasa DataLLNA EC3: 3.0E-4% (extreme sensitiser)Reference: Gerberick, Toxicological Sciences, 2007, 97,417-427.

Remove from calculation

Add comment...

Analyse compound(s) of interest This compound has a cyclopropenone group, not present in any other compound in this alert space.

Small carbocycles like this have high ring strain due to bond angle deviation from the standard tetrahedral 109.5°.

This increases the reactivity of this alpha,beta-unsaturated ketone substantially and may explain the potency of this compound compared to others within the same alert space.

Furthermore, this compound has only been tested in one LLNA study - result may not necessarily be reproducible.

Suggest addition/removal of compounds based on chemical and mechanistic knowledgeDue to:

• the potency of this compound and

• the potential for increased reactivity compared to others within the same alert space

removal of this compound should give a more accurate EC3 prediction.

9

LLN

A E

C3

%

Similarity %


1E-40.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

0.001

0.01

0.1

1

10

100

GHS 1B

GHS 1A

Notes:

Assess expert-reviewed EC3 predictionExpert-reviewed EC3 prediction more in line with other compounds in the same alert space.

Original EC3 prediction: 0.0048%

Expert-reviewed EC3 prediction: 2.7%

Predicted EC3 is relatively close to GHS boundary.

User may prefer to use conservative approach and treat chemical as 1A sensitiser.

Number of similar compounds used in the calculation: 9/13Predicted LLNA EC3: 2.7% (moderate sensitiser) - [Derek EC3 Model - 1.0.6]

10

Example 02

3-(4-ethylphenyl)propanal

2.9 (BioByte Corp., version 5.9)


419: Aldehyde

4.4%

C1=C(C=CC(=C1)CC)CCC=OSMILES

logP

MW

Alert

Predicted EC3



Analyse compound(s)



L

H

A

S

A

11

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.01

0.1

1

10

100

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.01

0.1

1

10

100

Highlight similar/dissimilar compounds and/or outlier(s) Cluster of 5 chemicals with high Tanimoto similarity to query compound.

Small cluster of two compounds also used in default Lhasa EC3 prediction.


EC3 values mainly weak or moderate (based on ECETOC categories).



12

Analyse compound(s) of interest Cluster of 5 chemicals - subset of the query compound. Alpha methyl group adjacent to the aldehyde shouldn’t impact chemical reactivity so no need to remove these from the calculation.

Cluster of 2 chemicals - one carbon shorter in chain length. Shouldn’t impact chemical reactivity of aldehyde.

Alpha,beta-unsaturated ketone - May undergo alternative mechanism due to presence of Michael acceptor.

Suggest addition/removal of compounds based on chemical and mechanistic knowledge

Keep cluster of 5 chemicals with high Tanimoto similarity to query compound as they are extremely similar.

Small cluster of two compounds kept as reduction in alkyl chain length should not impact aldehyde reactivity.

Final three used in default prediction removed:

• Alpha,beta-unsaturated ketone because of alternative mechanism.

• Final two as not required with 7 very similar compounds in EC3 prediction.

LLNA EC3:5.9% (weak sensitiser)Similarity: 59%

LLNA EC3:18% (weak sensitiser)Similarity: 99%

LLNA EC3:6.3% (moderate sensitiser)Similarity: 57%


LLNA EC3:0.06% (strong sensitiser)Similarity: 36%






Similar Compounds

LLNA EC3:5.9% (weak sensitiser)Similarity: 59%










Similar Compounds

13

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.01

0.1

1

10

100

Notes:

Assess expert-reviewed EC3 predictionExpert-reviewed EC3 prediction more in line with other compounds in the same alert space.



Predicted EC3 is borderline between weak and moderate ECETOC categories.

User may prefer to use conservative approach and treat chemical as a moderate sensitiser.


strong

extreme

moderate

weak

14

Example 03

HO OH

5-methylbenzene-1,3-diol



440: Resorcinol or precursor

2.0%

OC1=CC(C)=CC(O)=C1SMILES

logP

MW

Alert

Predicted EC3



Analyse compound(s)



L

H

A

S

A

15

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.1

1

10

100

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.1

1

10

100

Highlight similar/dissimilar compounds and/or outlier(s) One compound very similar to query.

2/6 data points are experimental non-sensitisers.

One compound is slightly more potent than any other compound in the alert space (EC3 = 0.49%).





16

Analyse compound(s) of interest Small cluster of non-sensitisers - Anisole motif present instead of two hydroxyl groups. May have an impact on chemical reactivity.

Alpha,beta-unsaturated ketone - May undergo alternative mechanism due to presence of Michael acceptor.

Suggest addition/removal of compounds based on chemical and mechanistic knowledge

Cluster of non-sensitisers removed from EC3 prediction as anisole may be responsible for reduced activity of these compounds compared to the query compound.

Alpha,beta-unsaturated ketone removed from EC3 prediction as it may undergo an alternative mechanism due to presence of Michael acceptor.

Other three compounds in alert domain kept and used in EC3 prediction.

LLNA EC3:non-sensitiserSimilarity: 24%






Similar Compounds

OH

OH

CI

OH

OH

OH

OH

OH

HO

OH

BrOH







Similar Compounds

OH

OH

CI

OH

OH

OH

OH

OH

HO

OH

BrOH

17

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.1

1

10

100

Notes:

Assess expert-reviewed EC3 predictionExpert-reviewed EC3 prediction has changed GHS category from 1A to 1B.



Predicted EC3 is moderate sensitiser although is close to ECETOC category border.


strong

moderate

weak

18

Example 04

Cl

Cl

Cl

O

O

NH

3-((2,3,4-trichlorophenyl)amino)oxetan-2-one



411: Ring-strained amide, ester, thioamide or thioester

1.8%

C1(C(CO1)NC2=CC=C(C(=C2Cl)Cl)Cl)=OSMILES

logP

MW

Alert

Predicted EC3



Analyse compound(s)



L

H

A

S

A

19

Similarity %


0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0

LLN

A E

C3

%

0.1

1

10

100




Similar Compounds

O

OOH

N

S

NH2

O

O

OH

N

S

NH

O

O

O


In the absence of similar compounds in the alert space:

The remaining expert review steps cannot be carried out (see below).

How much confidence can be had in an EC3 prediction based on these three Nearest Neighbours?

Look at EC3 data in the alert activated by the query compound Only 3 data points in alert space.



Analyse compound(s) of interest

Suggest addition/removal of compounds based on chemical and mechanistic knowledgeAssess expert-reviewed EC3 prediction

X

XX

20

References

1. Kimber I, Basketter DA, Gerberick GF, Dearman RJ, 2002. Allergic contact dermatitis. Int. Immunopharmacol. 2, 201–211. DOI: 10.1016/S1567-5769(01)00173-4

2. OECD, 2012. The Adverse Outcome Pathway for Skin Sensitisation Initiated by Covalent Binding to Proteins Part 1: Scientific Evidence. DOI:10.1787/9789264221444-en

3. OECD. 2010. Test No. 429: Skin Sensitisation: Local Lymph Node Assay. DOI: 10.1787/9789264071100-en

4. OECD. 1992. Test No. 406: Skin Sensitisation. DOI: 10.1787/9789264070660-en

5. Basketter DA, Blaikie L, Dearman RJ, Kimber I, Ryan CA, Gerberick GF, Harvey P, Evans P, White IR, Rycroft RJG. 2000. Use of the local lymph node assay for the estimation of

relative contact allergenic potency. Contact Dermatitis 42, 344–348. DOI: 10.1034/j.1600-0536.2000.042006344.x

6. Basketter DA, Wright ZM, Warbrick EV, Dearman RJ, Kimber I, Ryan CA, Gerberick GF, White IR. 2001. Human potency predictions for aldehydes using the local lymph node

assay. Contact Dermatitis 45, 89–94. DOI: 10.1034/j.1600-0536.2001.045002089.x

7. Basketter DA, Clapp C, Jefferies D, Safford B, Ryan CA, Gerberick F, Dearman RJ, Kimber I. 2005. Predictive identification of human skin sensitization thresholds. Contact

Dermatitis, 53, 260–267. DOI: 1034/j.1600-0536.2001.045002089.x

8. Gerberick GF, Robinson MK, Ryan CA, Dearman RJ, Kimber I, Basketter DA, Wright Z, Marks JG. 2001. Contact allergenic potency: correlation of human and local lymph node

assay data. American journal of contact dermatitis : official journal of the American Contact Dermatitis Society 12, 156–161. DOI: 10.1053/ajcd.2001.23926

9. OECD. 2015. Test No. 442C: In Chemico Skin Sensitisation: Direct Peptide Reactivity Assay (DPRA). DOI: 10.1787/20745788

10. OECD. 2015. Test No. 442D: In Vitro Skin Sensitisation: ARE-Nrf2 Luciferase Test Method. DOI: 10.1787/9789264229822-en

11. OECD. 2016. Test No. 442E: In Vitro Skin Sensitisation DOI: 10.1787/9789264264359-en

12. Macmillan DS, Canipa SJ, Chilton ML, Williams RV, and Barber CG. 2016. Predicting skin sensitisation using a decision tree integrated testing strategy with an in silico model and

in chemico/in vitro assays. Regulatory Toxicology and Pharmacology 76, 30-38. DOI: 10.1016/j.yrtph.2016.01.009

13. Canipa SJ, Chilton ML, Hemingway R, Macmillan DS, Myden A, Plante JP, Tennant RE, Vessey JD, Steger-Hartmann T, Gould J, Hillegass J, Etter S, Smith BPC, White A,

Sterchele P, De Smedt A, O’Brien D, Parakhiai R. 2017. A quantitative in silico model for predicting skin sensitisation using a Nearest Neighbours approach within expert-derived

structure-activity alert spaces. Journal of Applied Toxicology, in press. DOI: 10.1002/jat.3448.

21

Notes

shared knowledge shared progress

Company Registration Number 01765239. Registered in England and Wales. VAT Registration Number GB 396 8737 77. ISO 9001: CERTIFIED

Lhasa Limited Registered Office Granary Wharf House, 2 Canal Wharf, Leeds LS11 5PS Registered Charity (290866) Reference 2/17

+44 (0)113 394 [email protected]

how to apply expert review to ec3 predictions in derek … llna ec3 % similarity % sensitiser...

Documents