A Novel Workflow for Non-Animal PBK Modelling of UV Filters: Oxybenzone as a Case Study

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Nazanin Golbamaki et al., have presented a draft on A Novel Workflow for Non-Animal PBK Modeling of UV filters: Oxybenzone as a Case Study. Here are some of my comments for the same.

1. The authors used a power parameter (β = 8) for WGCNA without providing a detailed justification.
2. The skin slices used exceeded the OECD TG 428 recommended range (200–400 µm), reaching up to 907 µm. Justify how this affects diffusion kinetics.
3. Please give details of ambient conditions like humidity and airflow for measuring vehicle evaporation.
4. Optical coherence tomography was used, but provide standard deviations or ranges across donors to assess variability.
5. Skin metabolism was performed using Episkin®, while diffusion used excised skin. Discuss in brief how inter-model differences might bias dermal metabolism estimations.
6. Donor skin from abdominal surgery was pooled, but data stratification like age and BMI is missing. Please comment on whether these parameters affect dermal penetration predictions.
7. The model presumes passive diffusion across all layers. Were any active transport processes or efflux proteins considered or ruled out?
8. The PBK model predictions are generally within 2-fold of observed data, but explain why Cmax was underestimated (0.63–0.71 ratio) for Formulation 1.
9. Uncertainty was analyzed using log-normal distribution, but the source of CV% values used for parameter variation was not fully detailed.
10. Mean clinical values were used for validation, but how were outliers or high inter-individual variability (e.g., CV > 150%) treated?
11. The model accurately predicts steady-state Cmax, but discuss why AUC underpredicts by ~20% in Formulation 2.
12. The model significantly overestimated volume of distribution (Vss = 400 L). Please justify.
13. Despite the lower oxybenzone % in lotion, systemic absorption was higher than spray. Is there any impact of evaporation rate, viscosity, and vehicle affinity for the same?
14. Please improve the clarity of figures.

 

Author Response

Reviewer 1:

Nazanin Golbamaki et al., have presented a draft on A Novel Workflow for Non-Animal PBK Modeling of UV filters: Oxybenzone as a Case Study. Here are some of my comments for the same.

Response: Thank you for your valuable feedback. We appreciate your diligent review of our manuscript.

 

1. The authors used a power parameter (β = 8) for WGCNA without providing a detailed justification.

Response: We would like to clarify that our manuscript focuses exclusively on the development and validation of Physiologically Based Kinetic (PBK) models. Specifically, we investigate the dermal absorption of oxybenzone using in silico and in vitro data, leveraging the GastroPlus® software and its Transdermal Compartmental Absorption & Transit (TCAT) module. Our analysis primarily involves sensitivity and uncertainty assessments of the PBK model parameters. We did not employ Weighted Gene Co-expression Network Analysis (WGCNA) or any other gene co-expression network methodologies in this study. Therefore, we believe that the WGCNA power parameter is out of the scope and methods presented in our paper.

 

1. The skin slices used exceeded the OECD TG 428 recommended range (200–400 µm), reaching up to 907 µm. Justify how this affects diffusion kinetics.

Response: We agree with the reviewer that the dermatomed skin was thicker than recommended and we have mentioned this in the paper. However, this parameter does not affect the partition and diffusion coefficients of the dermis. The amount in the RF could be underestimated due to retention in the dermis. Ultimately, the greater thickness of the skin may well affect the extrapolation to the in vivo situation, knowing that blood clearance is effective in the entire dermis and not just up to 200 or 700 µm. Ksys was estimated using the Ibrahim model, and since the results obtained were consistent with PK data, we can conclude that the thickness of skin used did not underestimate skin absorption.

The TCAT module can be customised with the thickness of all skin layers that was measured in the laboratory. Therefore, the study protocol was fully replicated in the software by adding the exact measured thickness as input. This information has been added to the manuscript in lines 553-557.

 

1. Please give details of ambient conditions like humidity and airflow for measuring vehicle evaporation.

Response: The laboratory temperature is regulated at 22°C ± 2°C and the humidity is 50±10% (values checked during the distribution studies). This information has been added to Section 4.2.4. Residual volume fraction – Lines 617-618.

 

1. Optical coherence tomography was used, but provide standard deviations or ranges across donors to assess variability.

Response:  Thank you for this suggestion. We have added the mean (+SD) thickness of the stratum corneum for each human skin donor. For each donor, 3 areas were evaluated with 3 measurements per area. This information has been added to the legend of the new Supplementary Table S4.

 

1. Skin metabolism was performed using Episkin®, while diffusion used excised skin. Discuss in brief how inter-model differences might bias dermal metabolism estimations.

Response: A study has been conducted and published to confirm that the Episkin® model was relevant to assess skin metabolism: " Comparison of xenobiotic metabolizing enzyme activities in ex vivo human skin and reconstructed human skin models from SkinEthic”. The study confirmed that these skin models exhibit metabolic functionalities similar to normal human skin concerning the investigated xenobiotics, making them useful tools for evaluating the local efficiency and safety of cosmetic products. This information has been added to 4.2.5. Skin metabolism data, lines 636-639.

 

1. Donor skin from abdominal surgery was pooled, but data stratification like age and BMI is missing. Please comment on whether these parameters affect dermal penetration predictions.

Response: We have added the ; however, BMI values were unavailable. The BMI can influence skin penetration by altering skin thickness and hydration. However, these two parameters are standardized by dermatoming to a set skin thickness and rehydrating all skin samples for 1 hour at 32°C. The thickness of the different skin layers for each donor are taken in account in the PBK model prediction.

 

1. The model presumes passive diffusion across all layers. Were any active transport processes or efflux proteins considered or ruled out?

Response: We confirm that active transport in the skin was not incorporated into the simulations. Indeed, oxybenzone was not indicated to be a transporter substrate in other assays e.g. Caco-2 A-B/B-A assays (data not shown).

The contribution of active transport in the skin is generally omitted in in silico models, especially considering that ex vivo frozen human skin is extremely unlikely to have functional transporters. Therefore, the simulations will reflect absorption in frozen skin. The latter is a well-accepted model for representing in vivo skin absorption; therefore, by inference, if an in silico model predicts the absorption of a chemical across frozen skin, it is also a good representation of in vivo absorption.

 

1. The PBK model predictions are generally within 2-fold of observed data, but explain why Cmax was underestimated (0.63–0.71 ratio) for Formulation 1.

Response: We acknowledge the underprediction of Cmax for Formulation 1 at all three levels of parameterization. As stated in the manuscript, this value is still within 2-fold of the observed values and is thus within acceptable limits. Of note, it is not expected that the L3 parameterization of the PBPK model will result in a Cmax for a single subject which is exactly the same as the measured Cmax for the overall population in the clinical study (which exhibits some variability – the %CV is 60% for this value).

Table 3 reports the observed vs predicted values of Cmax for the two formulations:

For Formulation 1, the Cmax is slightly underestimated for prediction level 3 – this is optimized using clinical data, which are much more variable for Formulation 1. When running the model for Formulation 1 in 1 individual, all parameterization levels underestimate the Cmax when compared to the mean values observed in the clinical trial. For Formulation 2, the L1 and L2 model underestimated Cmax; however, the L3 model slightly overpredicted Cmax.

When assessing the sensitivity analysis for the two formulations in the full body PBK models, among the parameters related to the whole body in vivo system (excluding the skin), the blood partition coefficient (Rbp), is the most sensitive parameter for both Formulations 1 and 2. The relative sensitivity of parameters related to skin absorption (after Rbp,) for the two formulations are different: For Formulation 1, the order of sensitivity is the vehicle layer partition coefficient followed by the dermis diffusion and dermis partition coefficient; for Formulation 2, the stratum corneum diffusion is the most sensitive, followed by the dermis diffusion and dermis partition coefficient. This highlights how these two Formulation models can be differently affected i.e., by the vehicle layers value in Formulation 1 and by the stratum corneum diffusion in Formulation 2.   

It should be noted that the population analysis shows a higher concordance of Formulation 1 predicted mean Cmax value of 0.0982 µg/mL compared with the observed value (0.089 µg/mL). The same trend was evident in the population analysis of Formulation 2, with similar observed (0.080 µg/mL) and predicted population (0.077 µg/mL) Cmax values. A population analysis considers the variability that is expected for each of the input parameters including the vehicle layer partition coefficient and the stratum corneum diffusion. While traditionally PBK models are built for 1 individual, the population analysis captures the variability that can be observed in vivo. Therefore, the mean values resulting from population analysis are more indicative of the reality and show that the model is actually predicting the mean values which are matching the clinical mean values. We have added this information to the Discussion in Lines 464-471.

 

1. Uncertainty was analyzed using log-normal distribution, but the source of CV% values used for parameter variation was not fully detailed.

Response: We acknowledge these values should be more explicitly highlighted in the text (now in Lines 801-802). The uncertainty is expressed as a %CV and all values are reported in Supplementary Tables S6-S9. When experimental values are measured the related uncertainty is also expressed as a %CV; therefore, this is a recognised method by which uncertainty is evaluated. Further explanation is reported in a Table (now Supplementary Table S10) below for how each uncertainty value was derived for each experimental values or when these were found in the Gastroplus software data (denoted as GP in the table).

 

1. Mean clinical values were used for validation, but how were outliers or high inter-individual variability (e.g., CV > 150%) treated?

Response: All values were kept because we did not know how the data had been removed in the clinical study. We therefore considered the real variability even when it was high. This information has been added to Lines 535-537 of the manuscript.

CV% calculations of Matta et al 2020 (are reported in Reviewer 2, point 4 reply).

Thanks for this comment. We have analysed the clinical data in the Matta et al. 2020 paper, we have plotted these below for clarity:

Regarding the inter-individual variability and outlier treatment, below we provide a summary table of the key PK parameters (Cmax and AUC) for the spray and lotion formulations, with and without the identified outlier (participant AS.12) in the spray data. The table reports CV% of those PK parameter values.

Formulation	Participant Inclusion	Cmax Geometric Mean (ng/mL)	Cmax Geometric CV%	AUC Geometric Mean (ng·h/mL)	AUC Geometric CV%
F1 (Spray)	All participants	85.44	88.18%	1159.63	58.81%
F1 (Spray)	Excluding AS.12	74.24	56.19%	1059.30	44.43%
F2 (Lotion)	All participants	94.24	50.00%	1203.99	36.50%

 

Only one participant (AS.12) exhibited an elevated Cmax. Variability in the spray group is higher than lotion group. Removing this participant reduced the geometric coefficient of variation for Cmax by almost 32 percentage points (from 88.18% to 56.19%), and for AUC by 14 points (from 58.81% to 44.43%). This indicates that AS.12 is an influential outlier strongly affecting the variability metrics. We adopted the same practice as in pharmaceutical industry and included all participants in the clinical data regardless of whether these are outliers. This allows for a good representation of the real-world inter-individual variability.

 

1. The model accurately predicts steady-state Cmax, but discuss why AUC underpredicts by ~20% in Formulation 2.

We believe that an under (or over)-prediction of 20% is still an acceptable accuracy of the simulation. We checked the input parameters and noticed a small factor that could have slightly influenced the result. Steady state (multiple dose) for F2, has been predicted for 2 different models (L2 and L3). In the L2 simulation, the observed underprediction is exactly what we observe in the single dose, for the same individual (maintaining the physiology parameters). These two results correlate well; however, we have updated the graph. The previous graph was obtained for an individual with a different physiology than that of the individual in the single dose output. Keeping those physiologies the same, the Cmax was lower than before. We have updated Figure 6, B, Level 2 simulation.

In the L3 simulations, we acknowledge an underprediction for 1 individual at steady state that is not justified when looking at the simulations for 1 individual in single dose. This issue has been raised with the software developers and cannot be investigated further, as we do not have access to the backend of the software. This highlights one of the disadvantages of using commercial software – some of the equations behind the simulations are unknown.

 

1. The model significantly overestimated volume of distribution (Vss = 400 L). Please justify.

Response: The Vss of 400 L results from the L1 (Level 1) models. These models use only in silico input. The predicted concentrations increase over time, and no half-life can be assigned as the concentrations continue to rise. Therefore, it is not possible to predict AUC_0-infinity with this model. This highlights how L1 models are not predictive of the observed values, and only comparing predicted vs observed for Cmax and AUC_0-23h cannot capture this fact. The half-life and AUC_0-infinty indicate that the L1 models do not predict the PK of oxybenzone in Formulations 1 and 2 very well.   

The Vss of L2 and L3 are reasonable values for a lipophilic compound such as oxybenzone.

 

1. Despite the lower oxybenzone % in lotion, systemic absorption was higher than spray. Is there any impact of evaporation rate, viscosity, and vehicle affinity for the same?

Response: To understand which parameters have the highest impact on the % absorption of oxybenzone in Formulation 1 (spray) and Formulation 2 (lotion), we have conducted a sensitivity analysis of the TCAT model. This indicates a higher influence of the ‘application time’ for the lotion. The application time used in the simulations corresponds to the evaporation time (see paper text lines 631-633). This suggests that TCAT model is highly sensitive to this parameter and that the evaporation of the spray may have played a crucial role in the absorption phase. Both models are sensitive to the ’vehicle layer partition coefficient’. In general, these results suggest that the Spray is more sensitive to the changes in the parameters related to the formulation.

On the other hand, an analysing the clinical results suggests that the Cmax is most sensitive to ’vehicle layer partition coefficient’ for the spray while the most sensitive for the lotion is the ‘stratum corneum layer diffusivity’. This again suggests that the role of the vehicle is more important for the partition from the skin surface to the stratum corneum when using the spray.

Taken together, and considering the high variability of the clinical data, we can refer to the in vitro data (which has less variability) and conclude that although clinical data are showing differences, these can only be observed in individual skin layers in vitro but not when total absorption in all skin layers is compared. Ultimately the in vitro skin absorption suggests that the absorption of oxybenzone in the two formulations is not statistically different (see manuscript text lines 423-427.

 

1. Please improve the clarity of figures.

              Response: We appreciate that the figures were not of the highest quality. Therefore, we have uploaded higher quality versions of the figures using different software.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents a well-designed and comprehensive study on the development and validation of a non-animal PBK model for oxybenzone, with a focus on formulation effects. The work is innovative and aligns with current regulatory demands for animal-free safety assessments. The methodology is robust, and the results are clearly presented. However, some minor revisions are required before accept for publication:

1. The manuscript should explicitly discuss the limitations of assuming constant partition and diffusion coefficients in the TCAT model, especially regarding formulation effects over time.
2. The authors are suggested to provide a brief justification for selecting the Peress equation for evaporation rate calculations and discuss potential alternatives.
3. The lack of statistical significance in ‘’in vitro‘’ skin absorption results between formulations should be discussed in more detail.
4. The manuscript attributes variability in clinical data to the small number of subjects. Could other factors (e.g., application technique, environmental conditions) also contribute? A brief discussion would strengthen this section.
5. Clarify the rationale behind the classification thresholds for reliability (high, medium, low) and how these were derived.
6. Ensure all figures (e.g., Figure 3) are referenced and discussed in the main text.
7. Table 1b is extensive; consider moving less critical details to supplementary materials to improve readability.

Author Response

Reviewer 2:

The manuscript presents a well-designed and comprehensive study on the development and validation of a non-animal PBK model for oxybenzone, with a focus on formulation effects. The work is innovative and aligns with current regulatory demands for animal-free safety assessments. The methodology is robust, and the results are clearly presented. However, some minor revisions are required before accept for publication:

1. The manuscript should explicitly discuss the limitations of assuming constant partition and diffusion coefficients in the TCAT model, especially regarding formulation effects over time.

Response: Thank you for this comment. Unfortunately, the variation of partition and diffusion coefficient over time cannot be easily obtained experimentally. In addition, all in silico models predicting skin penetration also use fixed values for the partition and diffusion coefficients, probably for the same reason. The additional level of difficulty is related to the complexity of the formulation used. For the current study, the exact composition of the formulations was unknown, which makes the measurement of the variation of partition and diffusion coefficient impossible to derive.

 

1. The authors are suggested to provide a brief justification for selecting the Peress equation for evaporation rate calculations and discuss potential alternatives.

Response: The Peress equation does not need the solvent air diffusivity value to calculate the evaporation rate and evaporation time; therefore, we chose this equation to simplify the calculations. This information has been added to lines 628-630.

The different equations are shown below.

1. The lack of statistical significance in ‘’in vitro‘’ skin absorption results between formulations should be discussed in more detail.

Response: Thank you for this suggestion. The clinical differences can be related to factors such as variability in systemic parameters such as Rbp. Such parameters do not play a role in the in vitro skin absorption. Also the application conditions in clinical settings are less practical than in vitro (individuals had to stay without moving for an entire day for the sun screen product not to be removed from the surface of the skin), this could have changed the variability as well as the skin surface area differences in the BMI range that was chosen for the clinical study. We have added a paragraph on this in lines 428– 443.

 

1. The manuscript attributes variability in clinical data to the small number of subjects. Could other factors (e.g., application technique, environmental conditions) also contribute? A brief discussion would strengthen this section.

Response: Thank you for highlighting this point. The application conditions are critical in a clinical study. Friction caused by clothing in particular can contribute to the variability of what actually remains on the body after applications. The clinical study does not specify the application conditions, just the amount applied to 75% of the body surface. We have added this information to the paragraph in lines 428– 443.

 

1. Clarify the rationale behind the classification thresholds for reliability (high, medium, low) and how these were derived.

Response: Uncertainty is a subjective assessment of how reliable the input parameters are. A formal uncertainty analysis as suggested in the WHO PBPK guidance is difficult to perform with a bottom up PBK model as the ratio of median to 95th percentile reflects a measure of variability rather than true uncertainty as could be obtained if the PBK model parameters were fitted to an observed dataset. The uncertainties of parameter that was not measured with in vitro data, were higher compared to the one measured initially by in vitro measurement. Moreover, as mentioned in the SCCS guidance of 2021, the use of estimated values in further modelling might increase uncertainties associated with a model. Based on this assumption, the uncertainty of parameters that were optimized on the ex vivo data, were increased (from low to medium).

We have added this information to the section on uncertainty – lines 371– 380.

 

1. Ensure all figures (e.g., Figure 3) are referenced and discussed in the main text.

Response: We have moved the Materials and Methods and re-labelled the Figures and Tables accordingly.

 

1. Table 1b is extensive; consider moving less critical details to supplementary materials to improve readability.

Response: Thank you for this suggestion. We have moved Table 1 to the Supplementary materials.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed assigned comments at satisfactory level.