From Reduction to Replacement: Challenges and Proposed Solutions for Animal-Free Safety Assessment of Hydrocarbon UVCB Substances

Highlights

• •UVCB substances derived from petroleum and renewable feedstocks are a category of complex substances spanning a continuum of hydrocarbon distributions, with well-established hazard profiles.
• •Read-across and predictive models developed based on historical datasets have been extensively used to reduce animal testing.
• •Full replacement of animal testing remains challenging because of UVCB complexity and difficult-to-test properties.
• •Eight key scientific, technical and regulatory challenges limiting animal-free UVCB assessment are identified and practical short-, mid- and long-term solutions are proposed to address these challenges.
• •A roadmap aligned with the European Commission initiative to phase out animal testing is presented using hydrocarbon UVCB as a case study.

  The transition to animal-free chemical safety assessment is expected to follow a different timeline for UVCB substances as compared to most industrial chemicals.

Abstract

The European Commission (EC) developed a Roadmap to phase out animal testing, with milestones and practical steps across 15 different European chemical legislations (EC 2026). Currently though, the number of animals used for compliance testing in the EU remains high, particularly for high-volume chemicals such as Petroleum Substances, and is expected to increase with the introduction of new hazard classes under Regulation (EC) 1272/2008 (CLP). A reduction in animal testing for these substances can be achieved today by grouping, read-across and composition-activity models leveraging historical data and an in-depth understanding of substance composition and manufacturing processes. However, full replacement with New Approach Methodologies (NAM) remains challenging due to the complexity and physico-chemical properties of hydrocarbon substances from conventional and renewable feedstocks which are UVCB substances (Unknown, Variable composition, Complex reaction products or Biological materials). This manuscript outlines the current efforts to reduce animal testing in environmental and human health assessment of hydrocarbon UVCB and highlights key challenges limiting broader application of in silico and in vitro approaches. To address these challenges, we propose a pragmatic and readily implementable roadmap for the transition to animal-free UVCB assessment, adapting the different elements of NAM frameworks including analytical characterization, in silico predictions, in vitro testing, and in vitro-in vivo extrapolation. Short-term actions focus on enhancing read-across and increasing regulatory acceptance of existing approaches, mid-term priorities include improving technical readiness of in vitro testing and in silico approaches and the long-term goal aims at integrating predictive models and the development of an animal-free UVCB framework for regulatory acceptance. This transition will require tailored approaches and is expected to follow a different timeline compared to other industrial chemicals.

Introduction

Across the chemical industry, evolving regulatory requirements may substantially increase the number of animals needed for chemical safety testing. In 2020, Concawe (Conservation of Clean Air and Water in Europe) estimated that the REACH (Registration, Evaluation, Authorisation and Restriction of Chemicals) testing programs for 185 Petroleum Substances (PS) (∼4,500 unique registrations) would consume one million animals (Ketelslegers et al., 2020). In 2023, Rovida et al. revised this number upward, indicating that up to 24 million animals would be needed if grouping and read-across strategies for PS were not accepted (Rovida et al., 2023). Since 2009, approximately 382,000 fish have been used for testing substances under REACH, and amendments introduced in 2022 could require up to 680,000 fish for long-term studies (Knight et al., 2025). Moreover, tens of millions of animals could be required for EU REACH information requirements for endocrine disruptor (ED) assessments (Burden et al., 2024).

Regulatory developments globally are supporting a shift towards animal-free testing, applying New Approach Methodologies (NAM) and NAM-based assessment frameworks. Recently, the European Commission (EC) published a roadmap, with milestones and practical steps, to phase out animal testing across 15 chemical legislations (EC, 2026). It has been recognized that a regulatory paradigm shift is needed in chemical safety assessment to ensure non-animal methods are the default (EPAA, 2025). Multiple areas require changes to enable this paradigm shift, specifically, i) development of fit-for-purpose NAM (Parish, 2020), ii) improvement of tiered exposure-informed assessment approaches combining several NAM (Ball et al., 2022; Doe et al., 2025), iii) validation and implementation of NAM (OECD, 2005, under revision), iv) capacity development at Contract Research Organisations globally and, ultimately, v) policy changes to support regulatory implementation. Further, alignment across multiple stakeholders is needed on common endpoints and methods ensuring regulatory stability and predictability.

With emerging roadmap concepts and NAM-based frameworks for the transition to animal-free safety assessments, it has become clear that different sectors have distinct regulatory and technical needs and will progress towards an animal-free state at varying pace (Mitchell et al., 2025). In this manuscript, we focus on the specific considerations and adaptations that are needed for difficult-to-test substances, including UVCB (substances of Unknown or Variable composition, Complex reaction byproducts, or Biological materials) which represent 20% of the registrations in the EU, by number (ECHA, 2017b). They also represent a large portion by volume of registered substances: in 2022, 889 million tons of PS were registered under REACH (manufactured or imported into the EU) (Concawe, 2026).

PS are complex substances derived from crude oil distillations (atmospheric and vacuum) consisting of complex and variable mixtures of hydrocarbons whose exact composition depends on the crude source and refinery processing steps (Clark et al., 2013). With growing demand for lower emissions and more circular products, new non-crude oil feedstocks (i.e. alternative feeds) are increasingly being utilized to manufacture hydrocarbon products through similar processes (Lampi et al., 2024). Compositional variability is also a consideration for hydrocarbon substances derived from alternative feeds such as biologically derived oils, waste oils and fats, pyrolysis oil from biomass and recycled materials (Concawe, 2025). The presence of constituents ranging in physicochemical properties, and variable constituent composition in PS has already required the development of fit-for-purpose testing and evaluation methods that reduce animal testing today. To develop a UVCB roadmap, we used hydrocarbon (HC) substances as a representative case study to help inform the transition for other UVCB substances, including flavoring agents, fragrances, animal fats and their derivatives, vegetable oils and their derivatives, natural oils and extractives, and biofuels (Clark et al., 2013). We apply the term PS when referring to previous work on substances from conventional crude oil derived hydrocarbon feedstocks and the broader term HC UVCB to cover substances from conventional and renewable feedstocks, as well as co-processed substances.

In this context, our review summarizes the current use of NAM in HC UVCB chemical safety assessment, the challenges limiting their wider implementation, and highlights the adaptations needed to facilitate broader application of NAM and their integration into UVCB-adapted NAM-frameworks to further reduce and ultimately replace animal testing for this substance class. Throughout the manuscript, we distinguish between applications aimed primarily at Human health (HH) assessment, Environmental (ENV) assessment, and cross-cutting approaches relevant to both domains, since the maturity of NAM, the underlying hazard paradigms, and regulatory requirements differ between these areas.

Current approaches to reduce animal testing for human health and environmental safety assessments of HC UVCB

Several approaches have been historically used by the petroleum industry to support the hazard assessment of PS and HC UVCB and, where scientifically justified, reduce reliance on new animal studies. These include read-across from representative similar substances, evaluation of unusually toxic components or fractions, screening/indicator tests benchmarked against historical in vivo data and quantitative composition/activity relationship models to predict outcome based on composition (McKee and White, 2014). The use of such approaches in a regulatory context requires case-by-case justification and depends on the quality of compositional characterization, the endpoint under consideration, and the strength of the supporting weight of evidence.

The scientific basis for applying these approaches to PS has been developed through extensive data generation programs which have led to the following conclusions: i) grouping can be appropriate where similarity in source, manufacturing process and composition support similarity and if compositional variability can be adequately considered, ii) baseline toxicity is the predominant mode of action for environmental assessments and iii) specific constituents or bioactive fractions may contribute to hazards above baseline toxicity (Lampi et al., 2024). The following paragraphs describe the current approaches in more detail.

Analytical characterization

Analytical characterization relies on a range of spectroscopic and chromatographic techniques, described in detail elsewhere (Roman-Hubers et al., 2023; Concawe, 2024a). The resulting compositional data underpin grouping strategies, read-across approaches, and QSAR (Quantitative structure–activity relationship) use. Different analytical techniques result in varying resolution of the HC UVCB. The most comprehensive technique is two-dimensional gas chromatography (GCxGC), which allows to separate constituents in complex substances. Based on GCxGC results the substances can be divided into discrete blocks according to a combination of carbon number and HC class in the so-called Hydrocarbon Block Method (HBM) (King et al., 1996).

Screening and indicator tests

For HH assessment, indicator assays to identify refinery streams with carcinogenicity potential have been developed on large historic PS datasets and an established relationship between mutagenicity, carcinogenicity and aromatic content, namely: IP346 (Concawe, 1994, 2016), the modified American Society of Testing and Materials (ASTM E1687-10) (2014) and a refined carcinogenicity prediction model developed by Goyak et al. (2016). For ENV assessment, substances can also be screened with biomimetic solid-phase microextraction (BE-SPME) which uses fibres coated with polydimethylsiloxane (PDMS) to characterise the bioavailable fraction that would potentially cause environmental toxicity based on comparison to empirically derived acute and chronic thresholds (Redman et al., 2014c, 2018).

In silico approaches

For HH endpoints, composition-based modelling approaches have been explored to support the assessment of PS. The examples discussed here are derived from data generated and compiled through the petroleum industry’s participation to the US EPA High Production Volume Challenge (2026) High Production Volume (HPV) Challenge Program (www.petroleumhpv.org). Historic mammalian in vivo datasets for high-boiling PS, primarily from dermal exposure studies, were used in multi-variate correlation analysis to evaluate the relationship between Polycyclic Aromatic Hydrocarbon (PAH) content and toxicity, known as the “PAH hypothesis”, and to develop predictive models for systemic, reproductive and developmental toxicity (Nicolich et al., 2013; Roth et al., 2013; Murray et al., 2013; Gray et al., 2013).

For ENV toxicity endpoints, it is well-accepted that HC exhibit baseline toxicity in aquatic organisms (Verhaar et al., 1992) which has led to the development of modelling approaches. The Target Lipid Model (TLM) is combined with an oil solubility model to predict aquatic toxicity of HC UVCB in the PetroTox model (Di Toro et al., 2000; McGrath et al., 2005; Kipka and Di Toro, 2009; McGrath and Di Toro, 2009; Redman et al., 2012; Redman et al., 2017; McGrath et al., 2018; Fanelli et al., 2026). The TLM can also be combined with an equilibrium partitioning model, to predict environmental toxicity to soil and sediment organisms (Redman et al., 2014a). Both in silico approaches require composition data based on the HBM.

Complementary to composition–activity models, evidence on species sensitivity supports further reduction of vertebrate testing for environmental toxicity endpoints; for HC acting via baseline narcosis, algae and invertebrates are typically the most sensitive aquatic species, and fish testing could often be avoided (Teixidó et al., 2020; Hughes et al., 2021; McGrath et al., 2018).

Grouping and read-across approaches

Grouping of PS was initially established to streamline hazard classification and labelling through category-based read-across, under the former EU Substance Directive by Comber and Simpson (2007). PS with similar feedstocks and manufacturing processes could be grouped, allowing shared hazard classification (McKee and White, 2014). Read-across extends this concept by predicting the properties of untested substances within a category based on demonstrated compositional similarity (Concawe, 2024b). At least five samples per registered substance are assessed to cover the variability between manufacturing sites and sampling timepoints, in line with the European Chemicals Agency (ECHA) 2022 advice on the Read-Across Assessment Framework (RAAF) for UVCB substances (ECHA, 2022).

For HH endpoints, a HC space map approach is being applied to account for compositional variability within a defined category (Concawe, 2022). The HBM, and computational similarity analysis informs sample selection for in vivo bridging studies (OECD TG422) to test the respective hazard hypothesis (Concawe, 2022).

For ENV toxicity endpoints, the HBM is also applied to account for PS variability within a defined category, utilizing in silico tools to predict the range of ecotoxicity within the category. The in silico data are combined with BE-SPME screening data across all substances in the category to inform the relevant testing hypothesis (Saunders et al., 2024; Martin-Aparicio et al., 2024).

Challenges in the transition to animal-free approaches for HC UVCB

NAM workflows are being proposed to assess chemical safety using non‑animal, mechanistic test methods (Ball et al., 2022; Doe et al., 2025; Rivetti et al., 2025). Developing a universal NAM framework is challenging due to differences in data requirements, uses, exposure levels and regulatory requirements across different sectors (Mitchell et al., 2025). However, the overall workflows and their core building blocks are comparable across frameworks. Typically, Tier 0 includes problem formulation, substance characterization, review of available data and identification of structural/compositional analogues for read-across. The placement of in silico approaches varies across published frameworks, with some incorporating them into Tier 0 as part of problem formulation and existing data evaluations, while others place them into Tier 1 to support hazard and Mode of Action hypothesis development. In the framework we present in this manuscript, Tier 1 comprises in silico and in vitro screening, reflecting the role of composition-activity approaches informing hazard hypotheses for complex substances and UVCB substances, while Tier 2 involves more targeted NAM testing and kinetic modelling. Exposure considerations can be added, depending on the context of use, but are not discussed in this review due to the large number of diverse uses and differences in the HH and ENV exposure assessment.

A generic NAM workflow is shown in Figure 1 together with the specific challenges in adopting the different elements of such a framework for UVCB substances. Those challenges arise from the complexity and variability of the chemical compositions and the diverse physico-chemical properties of UVCB substances and their constituents. HC UVCB are difficult-to-test substances, meaning that their behaviour in the test system makes it difficult to prepare, maintain, or analytically verify the intended exposure concentration under standard OECD (Organisation for Economic Co-operation and Development) test conditions (OECD, 2019). Moreover, full characterization and/or quantification of all the constituents within a substance may not be possible. Constituents may also span a wide range of physico-chemical properties, including limited to no water solubility, high hydrophobicity, volatility, and sticky or waxy physical form, with several of these properties potentially occurring within the same substance.

Fig. 1.

Challenges in applying tiered animal-free assessment frameworks to HC UVCB substances.

The figure presents a generic tiered NAM workflow and highlights eight challenges when adapting the workflow to the animal-free safety assessment of HC UVCB substances.

Together, these features create five linked barriers to animal-free safety assessment of HC UVCB: (1) incomplete compositional characterization, (2) uncertainty in the regulatory acceptance of the proposed grouping and read-across approaches, (3) limited applicability of current in silico models beyond baseline toxicity, (4) technical constraints in dosing, in vitro exposure characterization and bioavailability in in vitro systems, and (5) limited in vitro-in vivo extrapolation (IVIVE) due to lack of Physiologically Based Pharmacokinetic / Physiologically Based Toxicokinetic (PBPK/PBTK) models for complex, variable substances.

Table 1 organises these barriers into eight distinct challenges (C1–C8) with their associated implications. These are further described below.

Table 1.

Overview of HC UVCB-specific challenges, their implications and proposed solutions towards animal-free chemical safety assessment classified by area and timeline of application.

Proposed solution (S)	Area Timeline
Challenge 1 (C1): Substance complexity Complex, variable composition and diverse physicochemical properties make handling, characterization, and testing difficult; no single constituent represents the whole substance, and standard dosing may not deliver representative exposure.
Develop constituent-based or full-substance approaches and justification depending on the context of use.	T/G · Short-Mid
Develop a strategy for methodological adaptation across all elements of animal-free assessment frameworks.	T · Long

Challenge 2 (C2): Analytical characterization Limited compositional characterization and constituent quantification hinder cross-substance comparison and reduce confidence in QSAR development and application.
Adress limitations in detailed constituent quantification through method refinement & justification.	T · Short-Mid
Develop tailored approaches for identification of likely constituents and for quantification of those expected to drive hazard where applicable.	T · Mid

Challenge 3 (C3): Grouping and read-across approaches Structural similarity is difficult to demonstrate for UVCBs, leaving read-across justifications uncertain.
Provide more detailed guidance on regulatory acceptance criteria for UVCB read-across; update ECHA UVCB RAAF (ECHA 2022) and align with OECD Guidance on Grouping of Chemicals (OECD, 2025).	G · Short
Further leverage and critically evaluate UVCB-specific approaches for structural/compositional similarity and biological similarity.	T · Short-Mid
Strengthen similarity assessments through integrated analysis of composition-bioactivity datasets.	T · Mid
Provide example cases of accepted or rejected read-across for UVCB substances; based on improved criteria and methods, apply read-across broadly.	R · Mid
Implement non-animal methods to support biological similarity more broadly.	T · Long
Enable routine use of animal-free bridging studies/concepts through regulatory acceptance.	R · Long

Challenge 4 (C4): In silico limitations Existing QSARs, descriptors, and profilers do not reliably reflect whole-substance hazard, and regulatory acceptance of established UVCB approaches remains uncertain.
Evaluate PetroTox against the OECD QAF (2023) and identify relevant performance benchmarks to strengthen validity across species and substances.	R · Short
Develop and/or expand UVCB-specific descriptor strategies leveraging curated datasets; systematically assess QSAR applicability for regulatory-mature models.	T · Mid
Generate and curate high-quality compositional and bioactivity datasets from ongoing and historical testing programs to support similarity assessment and QSAR development.	T · Mid
Investigate full-substance interactions and develop tailored composition–activity and predictive bioactivity models for different modes of action, combined with kinetics models.	T · Long
Update OECD QAF Framework with UVCB-specific criteria; obtain regulatory acceptance for UVCB-tailored QSAR solutions.	G/R · Mid-Long

Challenge 5 (C5): Solvent extraction-based in vitro approaches Previously developed extraction-based approaches for grouping and bioactivity profiling remain unused; practice has reverted to in vivo bridging studies.
Further evaluate whether bioactive fractions are sufficiently representative for prioritization, grouping and hazard prediction.	T · Short
Demonstrate relevance of the modified Ames (ASTM E1687) to replace the standard Ames assay (OECD TG 471) where applicable.	R · Mid
Generate targeted evidence to support regulatory confidence; demonstrate the composition-in vitro-in vivo relationship (PAH hazard hypothesis).	R · Mid

Challenge 6 (C6): Representative in vitro dosing and exposure delivery Volatilization, sorption, poor solubility, instability, and media binding may prevent representative delivery of UVCB constituents to in vitro systems.
Integrate established in vitro assays into UVCB testing programs to assess assay applicability, endpoint coverage, and exposure delivery.	T · Short–Mid
Develop qualified UVCB dosing approaches with defined applicability domains, including full-substance, fraction-based, solvent-assisted, passive, sealed, or combined approaches.	T · Mid
Evaluate whether selected dosing approaches deliver a representative and biologically relevant UVCB fraction.	T · Mid
Adapt in vitro test batteries using UVCB-compatible dosing approaches and identify remaining substance-coverage limitations and potential impacts.	T · Mid
Define regulatory acceptance criteria for alternative UVCB dosing approaches that deviate from standard OECD TG.	R · Mid
Update OECD GD 23 and GD 286/GIVIMP to address difficult-to-test UVCB substances and representative in vitro exposure delivery.	G · Mid

Challenge 7 (C7): In vitro dose quantification, analytical verification, and IVIVE/QIVIVE support Nominal concentrations may not reflect the biologically relevant dose; without verified dose metrics, potency estimates and IVIVE/QIVIVE remain uncertain.
Combine in vitro testing with analytical verification and mass-balance approaches to quantify assay exposure over time.	T · Mid
Establish UVCB-relevant dose metrics, such as measured, freely dissolved, cell-associated, or time-weighted concentrations.	T · Mid
Develop workflows to estimate or measure bioavailable concentrations in selected in vitro systems.	T · Mid
Adapt existing multi-media mass-balance models which are currently developed for single substances for UVCB in vitro systems.	T · Mid
Apply verified dose metrics in UVCB-tailored in vitro test batteries to support potency estimation and bioactivity-based PoD.	T · Mid-Long
Generate analytically verified dose-response data to support development of predictive replacement models and IVIVE/QIVIVE.	T · Long
Develop guidance for reporting dose metrics, analytical verification, mass balance, uncertainty, and assumptions for IVIVE/QIVIVE.	G · Mid-Long
Extend OECD GD 286/GIVIMP and relevant OECD guidance with UVCB-specific sections on dose metrics and analytical verification.	G · Mid-Long

Challenge 8 (C8): Kinetics readiness Constituent-level toxicokinetic data and UVCB-adapted PBPK/PBTK models are insufficient, making extrapolation from in vitro effects to external exposure difficult.
Compile measured constituent-level TK data and define data generation strategies, as needed, to develop UVCB-adapted PBPK/TK models and frameworks, informed by OECD GD 331 (PBPK models for regulatory purposes).	T · Short-Mid
Evaluate uncertainty in constituent-based predictions for full-substance bioavailability; define a validation strategy for UVCB models in the absence of full-substance data and describe uncertainties.	T · Mid
Use modelled bioavailability for reverse dosimetry and as supporting evidence for grouping/similarity; develop predictive models with TK and TD components linking kinetics, bioavailability, and bioactivity.	T · Long
Evaluate UVCB kinetic modelling approaches against OECD GD 331 and propose amendments to accommodate UVCB substances.	G · Long

Area: T = technical solution · G = guidance · R = regulatory adoption; Timeline: Short = short-term, strengthen currently available approaches · Mid = mid-term, build technical readiness for different NAM framework elements · Long = long-term, integrate elements into an UVCB-adapted NAM framework; Abbreviations: C = challenge; S = solution; QSAR = Quantitative structure–activity relationship; UVCB = substance of unknown or variable composition, complex reaction products or biological materials; ECHA = European Chemicals Agency; RAAF = Read-Across Assessment Framework; QAF = QSAR Assessment Framework; PAH = Polycyclic aromatic hydrocarbon; TG = Test Guideline; GD = Guidance Document; GIVIMP = Good In Vitro Method Practices; PoD = Point of Departure; TK = Toxiokinetics; PBPK = Physiologically based pharmacokinetic; TD = Toxicodynamic.

C1: Substance complexity

As introduced above, HC UVCB substances combine variable composition with diverse physico-chemical properties. No single constituent adequately represents the whole substance, and standard dosing methods may fail to provide representative exposure (Deglin et al., 2026). These characteristics underpin the challenges described in the following sub-sections.

C2: Analytical characterization

Although more advanced analytical techniques have been developed over time, exact quantification of every constituent concentration remains a challenge for substances as complex as HC UVCB. Many analyses rely on relative quantification (i.e., concentrations in percent units of the whole substance) to interpret, visualize, and compare compositions. It must be emphasized that challenges of constituent resolution and adequate characterization rise with the increasing compositional complexity for substances with higher-boiling points, higher carbon number ranges (>C30) and higher molecular weight constituents (Concawe, 2022; Concawe, 2024a; Lyon et al., 2024). Complete characterization remains elusive and is generally only feasible for lower boiling point substances with fewer limitations in analytical resolution. These analytical limitations may hinder demonstration of structural similarity, which is a key regulatory requirement for grouping and read-across justifications.

C3: Grouping and read-across approaches

The challenge of demonstrating structural similarity for UVCB substances was recognized in the initial ECHA RAAF (2017a), and advice on how to establish structural similarity of UVCB substances was further developed by ECHA in 2022 (ECHA, 2022). The recently updated OECD Guidance on Grouping of Chemicals provides additional recommendation and examples on how to establish read-across via a category approach for UVCB substances (OECD, 2025a). Nevertheless, the criteria used by regulatory authorities to accept read-across have not been entirely clarified, and animal testing is still being required in some cases without a clear justification (van Velthoven et al., 2026). In a recent systematic evaluation of ECHA decisions on Testing Proposals (2008–2023) it was reflected that read‑across is widely used, but acceptance is far from guaranteed (Roe et al., 2025). UVCB substance read‑across submissions have failed due to insufficient characterization for target and/or source substances or disagreement in the interpretation of available data. Supporting data from non-guideline “bridging” studies to establish similarity was not found satisfactory (Roe et al., 2025). Where such supporting evidence has not been considered sufficient, in vivo bridging studies have therefore remained an important element of current regulatory practice for establishing biological similarity for HH endpoints.

C4: In silico limitations

Although the examples discussed below originate from HC UVCB, the main challenges for the development of UVCB QSAR models more broadly are i) aligned, fit-for-purpose descriptors, ii) an approach to account for potential interactions between constituents iii) applicability domain limitations of existing QSAR models and iv) curated and accessible databases of composition and bioactivity data to validate existing models or develop UVCB-specific ones. The limitations of in silico methods for HC UVCB differ between HH and ENV assessment. ENV assessment applies a tailored baseline toxicity model with PetroTox whereas HH assessment relies on a smaller number of composition-activity models with more restricted applicability domains. PetroTox uses the HBM and assumes strict additivity of toxicity, which has been established for aquatic toxicity of HC substances. Beyond strict additivity of effects, when contributions from different, structurally related constituents vary, multivariate regression analysis can help identify the relative contributions of different drivers of effect. The models developed by the American Petroleum Institute (API) under the US HPV Program to predict mammalian toxicity based on aromatic ring content are examples of such composition-activity models (Nicolich et al., 2013; Roth et al., 2013; Murray et al., 2013; Gray et al., 2013). However, their broader application is constrained by the applicability domain defined by the model training set.

Expanding these models, or developing new composition-activity relationships, requires large datasets that combine robust analytical characterization with biological test data. Although extensive toxicity datasets are available (McKee et al., 2014; Redman et al., 2017), the analytical composition of many historically tested substances was not characterized to current standards. This lack of sufficiently detailed compositional information may limit the use of historical datasets for model development and validation. Therefore, further development of such models will likely depend on fit-for-purpose in vitro testing strategies that can generate endpoint-relevant biological data while reducing reliance on additional whole-substance in vivo studies.

C5-C8: In vitro testing

The difficulties associated with testing PS also extend to in vitro methods. Volatility, sorption, poor solubility and binding to media and test systems cause constituent losses; nominal concentrations become unreliable and biological interpretation is compromised. Beyond these general exposure issues, four related but distinct challenges need to be addressed in parallel.

C5: Solvent extraction-based in vitro approaches (C5)

Several in vitro approaches that use solvent extraction have been developed as dosing strategy to enrich the toxicologically relevant bioactive fraction of PS. The DMSO extraction approach reflects the hypothesis that the most severe HH hazards of high-boiling PS are driven by a bioactive fraction rather than the full substance. One example of a category-based application is the Cat-App project, which was designed to evaluate whether PS could be assessed using manufacturing category and compositional information together with targeted extraction-based testing. Cat-App evaluated 141 PS from 16 manufacturing categories and showed high concordance between solvent-extract responses and in vivo hazard potential (Concawe, 2020; House et al., 2021; 2022).

The hypothesis that in vitro biological activity signatures can support grouping and read-across for PS has also been examined in several programs (Kamelia et al., 2017; 2018; 2019 a,b). In addition, Luo et al. (2020) investigated the in vitro bioavailability and cellular exposure of DMSO extracts, confirming selective PAH extraction and reproducible exposure profiles.

Despite this supporting evidence, ECHA considered the uncertainties associated with testing DMSO extracts, rather than whole substances, to be insufficiently resolved, particularly regarding how extraction may influence toxicological outcomes in vitro and in vivo (ECHA, 2021). Consequently, current regulatory practice has largely reverted to in vivo bridging studies for HH endpoints as highlighted in the read-across challenge section.

C6: Representative in vitro dosing and exposure delivery

Beyond solvent extraction, alternative dosing methods have been explored to improve in vitro testing of HC UVCB by enabling exposure to a broader range of constituents. However, establishing stable and analytically confirmed exposure remains technically challenging because of limited solubility, sorption to test-system components, volatilization, and matrix-binding effects (Deglin et al., 2026). Reviews of dosing approaches and case studies highlight the need to select methods based on substances properties and assay endpoint with closed glass systems and passive dosing offering advantages where feasible (Trac et al., 2021; Cordova et al., 2023; Wennberg et al., 2026). Standard in vitro set-ups for both HH and ENV endpoints are likely not achievable for HC UVCB as exposure control will require dedicated adaptations.

C7: In vitro dose quantification, analytical verification, and IVIVE/QIVIVE support

Nominal in vitro doses can misestimate potency because of substance loss in the test system, a challenge also documented across the Tox21 database (Fischer et al., 2017). Determining the bioavailable concentration, freely dissolved fractions and appropriate dose metrics is critical for IVIVE (Groothuis et al., 2015; Kramer et al., 2012; Luo et al., 2020). To characterize and quantify substance- and concentration-specific differences between nominal and bioavailable concentrations, analytical verification and mass-balance approaches are needed for both ENV and HH in vitro systems. The bottleneck for UVCB substances are the limitations for analytical dose verification, especially in high-throughput test systems, and multi-media mass-balance models, which are currently developed for single substances and not for UVCB substance in vitro testing.

C8: Kinetics readiness

Another unmet need for successful extrapolation of in vitro NAM effect levels to human-relevant external exposures is the availability of UVCB-adapted PBPK/PBTK models. Constituent-level toxicokinetic information, including absorption, distribution, metabolism and elimination, remains limited for many chemicals, and generating such data across numerous variable constituents is technically demanding (Wambaugh et al., 2025; Deglin et al., 2026).

Proposed solutions for the transition to animal-free approaches for HC UVCB

The challenges in the adoption of animal-free assessment methods and frameworks for UVCB substances need to be addressed and, in some areas, work is already ongoing. For the challenges laid out, solutions are being proposed as listed in Table 1. Within each challenge, actions are ordered by timeline: short-term, strengthen currently available approaches, mid-term, build technical readiness for different NAM framework elements and long-term, integrate elements into an UVCB-adapted NAM framework. They also follow a recurring T/G/R logic: technical development, guidance updates and regulatory adoption. More details and references to ongoing efforts are being provided in this section with solutions (S) for each challenge (C1-C8).

S-C1: Substance complexity

For substance complexity, the short- to mid-term action is to select, on a case-by-case basis, the most appropriate representation of the substance for a given assessment and provide the supporting scientific justification: constituent-based, block-based (HBM) or full-substance approaches (Deglin et al., 2026). In the long-term, a coherent methodological adaptation strategy across all elements of the animal-free assessment framework is needed, so that the choice of representation is consistent within a given dossier or substance category.

S-C2: Analytical characterization

Limitations in the detailed constituent quantification as outlined in the report from a 2024 Concawe workshop need to be addressed through method refinement and justification of the impact of incomplete constituent characterization and quantification on chemical safety assessment. These efforts should be complemented, as needed, by tailored approaches for identification and quantification of the constituents that are most likely to drive the hazard (Roman-Hubers et al., 2023; Concawe, 2024a; Lai et al., 2022; Lyon et al., 2024). Dissemination and training will help to create a common understanding of the limitations and their impact on chemical safety assessment. Analytical characterization is fundamental to further NAM implementation and work needs to be completed in the short- to mid-term.

S-C3: Grouping and read-across approaches

Grouping and read-across have high potential to reduce animal testing numbers for UVCB substances, especially with the currently limited availability of viable in silico and in vitro approaches. However, there is a lack of clarity on the acceptance criteria for current approaches put forward by registrants for HH and ENV assessment which creates unnecessary rejection and rework. The immediate short-term action is therefore to obtain more detailed regulatory guidance, updating the ECHA UVCB RAAF (ECHA, 2022) and aligning with the OECD Guidance on Grouping of Chemicals (OECD, 2025a). To expand successful approaches is it important that examples of cases of accepted and rejected UVCB read-across are shared and published.

For HH endpoints specifically, the combination of in vivo bridging studies with omics technologies has been shown to strengthen the demonstration of biological similarity, without requiring new dedicated animal studies (Sperber et al., 2019; Kamp et al., 2024). The applicability of such sensitive high-content omics techniques to complex substances and different in vivo study designs needs to be further explored and additional efforts are required to establish and refine study designs for UVCB substances. This will improve confidence in determining whether the results of such studies can support further reduction or the waiving of animal studies in the short- to mid-term.

In the mid-term, curated compositional and bioactivity datasets as further detailed in the in silico solutions section should be integrated with quantitative similarity metrics (e.g. Lester et al., 2023) and made available for case-by-case read-across justifications. Data integration more broadly is necessary because complementary data streams can be used to identify statistically and mechanistically meaningful relationships between composition, biological response and hazard, thereby strengthening the scientific basis for similarity assessment and read-across (Wehr et al., 2025; Zhou et al., 2025).

In the long-term, in vitro bridging studies combining in vitro profiling (e.g. Cat-App, omics-based signatures) with in silico similarity metrics should routinely replace in vivo bridging, provided regulatory acceptance can be granted.

S-C4: In silico limitations

The maturity of approaches differs substantially between ENV and HH applications. ENV assessment already benefits from established models such as PetroTox, whereas for HH assessment a limited set of composition-activity models for specific endpoints and applicability domains is available. Consequently, in the short-term, the most mature HC UVCB-specific in silico model, PetroTox, should be evaluated against the OECD (Q)SAR Assessment Framework (QAF, OECD, 2023) to identify relevant performance benchmarks and to strengthen regulatory confidence for its use across species and substances.

Because current QSAR approaches for HH are constrained by descriptor selection, full substance interactions and limited high-quality compositional datasets, a systematic assessment of QSAR approaches for UVCB substances and recommendations on how to expand their use are needed in the mid-term. This can be based off recommendations developed by Belfield et al., (2023): expand UVCB-specific descriptor strategies and systematically assess QSAR applicability domains for regulatory-mature models. Advancing in silico solutions for HH is of high importance as experimental work cannot describe all the possible combinations of constituents and effects.

Ongoing HH testing programs under REACH are generating new in vivo test data and high-quality analytical data on HC UVCB, further refined HC substances and constituents. Large collections of standardized toxicological studies generated under regulatory frameworks constitute a unique resource for investigating mechanisms of toxicity and composition-activity relationships and to train predictive models. However, current access conditions associated with confidential business information prevent their systematic aggregation and reuse outside the regulatory context. Wider implementation of FAIR (Findable, Accessible, Interoperable, Reusable) data principles (Wilkinson et al., 2016) could enable broader use and should be achieved in the mid-term.

In the long-term, tailored composition-activity and predictive bioactivity models for different modes of action reflecting full substance hazard need to be combined with kinetic models. PetroTox can be used as guiding principle, but concentration addition may not be applicable to all HH endpoints.

Updating the OECD QAF (OECD, 2023) with specific recommendations will provide the guidance necessary for regulatory acceptance of UVCB-tailored QSAR solutions in the mid- to long-term.

S-C5: Solvent-extraction-based in vitro approaches

A large body of evidence supports the use of DMSO as a biocompatible vehicle for solvent-extraction and enrichment of the bioactive fraction for HH in vitro studies used to prioritize substances for testing when the PAH hypothesis is applicable (Kamelia et al., 2017; 2018; 2019a,b; Concawe, 2020; Luo et al., 2020; House et al., 2021; 2022). The immediate short-term action is to further evaluate whether the resulting bioactive fraction is sufficiently representative for prioritization, grouping and hazard prediction for HH, building on the work cited above.

The modified Ames assay (ASTM E1687), originally developed for high-boiling PS and extensively characterized from a technical perspective (Blackburn et al., 1984; 1986), could replace the standard Ames assay according to OECD TG 471 as a standard information requirement where applicable, provided its relevance and reliability are confirmed across a broader range of UVCB substances in the mid-term (Kamelia et al., 2024).

In the mid-term, targeted evidence is needed to build regulatory confidence, particularly by showing that in vitro testing of bioactive fractions from HC UVCB of varying complexities reliably reflects and predicts in vivo outcomes for HH (ECHA, 2021).

S-C6: Representative in vitro dosing and exposure delivery

Where regulatory accepted in vitro alternatives exist, for example for skin/eye irritation, skin sensitization and genotoxicity, their applicability to HC UVCB should be critically evaluated (Greminger et al., 2024). For methods for which applicability has yet to be demonstrated, integration into ongoing compliance testing programs should be considered in the short- to mid-term. The generation of in vitro data alongside in vivo data would help to build confidence in their use for substance prioritization and to support grouping and biological similarity assessments. However, this is only possible if a suitable dosing solution is already available for the given test system, substance and application.

In the mid-term, qualified dosing methods with defined applicability domains should be developed, including recommendations on remaining gaps, needs and combinations of dosing approaches for HH and ENV to cover testing of the full substance (Wennberg et al., 2026). Multi-stakeholder initiatives are already advancing this work: the Better in vitro Dosing (BID) Consortium is developing a framework to dose and assess chemical exposure in in vitro assays for difficult-to-test chemicals.

Based on this work, in the mid-term, previously proposed in vitro test batteries should be explored for broader application with revised dosing solutions. Capability for high-throughput data generation and dose analytics should be built in parallel.

Regulatory uptake is required for success of this strategy and acceptance criteria should be developed in the mid-term, together with an update of the OECD GD on Aquatic Toxicity testing (OECD, 2019), expanding it to in vitro systems, and the OECD Guidance on Good in vitro method Practices (GIVIMP, OECD, 2025b) to explicitly address UVCB and difficult-to-test substances.

S-C7: In vitro dose quantification, analytical verification, and IVIVE/QIVIVE support

For UVCB substances, establishing appropriate dose metrics (freely dissolved concentrations, cell-associated concentrations, mass-balance-corrected doses) and UVCB mass-balance models for selected HH and ENV in vitro systems are the core mid-term technical actions (Fischer et al., 2017; Groothuis et al., 2015; Kramer et al., 2012; Luo et al., 2020). As introduced in the challenges section, multi-media mass-balance models are currently developed for single substances (Fischer et al., 2017; Armitage et al., 2021) but not for UVCB in vitro systems; adaptation for HC UVCB is a mid-term technical action to underpin robust dose metrics and IVIVE.

Workflows to estimate or measure bioavailable concentrations should be integrated with dosing method selection so that dose analytics is not treated as a separate exercise. They should be tested on UVCB substances of increasing complexity.

Building on this dosing and assay work, in the long-term, bioactivity-based NAM toolboxes combining in vitro assays with exposure modelling should be developed and evaluated to derive protective points of departure (PoDs) for UVCB risk assessment, as has already been demonstrated for well-defined substances and certain use cases (Cable et al., 2025; Paul Friedman et al., 2025).

The data generated through these programs will support the development of predictive models that reduce reliance on in vivo bridging and read-across. This long-term goal needs to be critically re-evaluated while working through the short- and mid-term actions. In the future, predictions can be evaluated against measured in vitro data which we expect to be generated at a much higher scale than in vivo data today. The applicability domain of these models needs to be defined and re-evaluated as needed.

On the guidance side, OECD GD 286 (GIVIMP, OECD, 2025b) and other relevant OECD guidance should be extended in the mid- to long-term with dedicated sections on mass-balance reporting, biokinetic-informed dose expressions (freely dissolved / cell-associated concentrations) and analytical verification requirements for UVCB testing. OECD TG relying on nominal concentrations should be complemented with an UVCB substance annex.

S-C8: Kinetics readiness

Another important step in implementing in vitro testing is the ability to extrapolate from internal dose or effect dose in vitro to an external exposure. Given the sparse constituent-level toxicokinetic (TK) data available today for HC constituents (Wambaugh et al., 2025), initial efforts should focus on defining relevant dose metrics for in vitro systems, representing UVCB substance composition through the HBM or representative constituents, and compiling measured TK and partitioning data for those blocks. This would allow simple reverse-dosimetry models to be developed and tested for selected substance categories in the mid-term before expanding towards more comprehensive mixture PBPK/PBTK models. UVCB-adapted PBPK/PBTK frameworks, informed by OECD GD 331 (PK models for regulatory purposes) can then be built on exiting frameworks such as the EPA High-throughput Toxicokinetic (HTTK) framework (Pearce et al., 2017).

In the mid-term, a validation strategy is needed for UVCB PBPK/PBTK models in the absence of full-substance data, together with an explicit description of uncertainties. Uncertainty in constituent-based predictions of full-substance bioavailability should be quantified and communicated.

In the long term, modelled bioavailability can be used for reverse dosimetry and as supporting evidence for grouping and similarity assessment. Predictive models with both TK and toxicodynamic (TD) components can then support full risk characterization workflows.

From a regulatory guidance perspective, UVCB-adapted PBPK/PBTK models should be evaluated against OECD GD 331 (OECD, 2021) and amendments should be proposed to accommodate UVCB substances as necessary.

A roadmap towards phasing out animal testing for UVCB and difficult-to-test substances

The previous sections identify the scientific and regulatory solutions needed to address the challenges associated with the application of NAM to HC UVCB substances. However, scientific and technological progress alone is insufficient to achieve routine application in regulatory decision-making. For NAM to effectively replace animal testing for HC UVCB substances, their development and qualification must be accompanied by a coordinated pathway that translates scientific innovation into regulatory practice. To support this transition, we applied the principles of the EC Roadmap towards phasing out animal testing (EC, 2026) to the specific needs of HC UVCB substances and translated it into UVCB-specific implementation actions. We defined three pillars to connect scientific development, regulatory acceptance, and coordinated implementation: (i) science and NAM development (ii) regulatory standards and acceptance, and (iii) coordination, capacity building and governance.

Table 2 summarizes the priority actions under each pillar across and the short, mid and long term. The pillars are interlinked, and progress in all three areas is required. Scientific work develops and qualifies UVCB-adapted approaches, while regulatory work defines the evidence and reporting requirements for evaluating the resulting evidence. Governance supports the coordination and prioritization of these activities, as well as training and implementation monitoring.

Table 2.

Proposed roadmap towards animal-free (HC) UVCB chemical safety assessment.

	Timeline
Pillar 1: Science and NAM development for UVCB Building the technical basis: fit-for-purpose analytical methods, in silico tools, in vitro assays and PBPK/PBTK models tailored to HC UVCB complexity.
Consolidate UVCB-adapted analytical, in silico and in vitro method development into a coherent scientific foundation for regulatory use and identify priority regulatory use cases and technical gaps.	Short
Generate targeted validation and qualification data, including but not limited to bridging datasets, to demonstrate the reliability and regulatory relevance of UVCB-adapted NAM for defined use cases.	Mid
Deliver mature, integrated animal-free assessment frameworks that combine toxicodynamic and toxicokinetic evidence and demonstrate relevance to full-substance assessment.	Long

Pillar 2: Regulatory standards and acceptance Translating technical progress into updated OECD test guidelines, guidance documents and ECHA/agency practice for UVCB substances.
Prioritize short-term guidance updates for enhanced UVCB read-across, existing QSAR models under the QAF, and already applicable in vitro methods and clarify the associated regulatory evidence and reporting elements.	Short
Update relevant guidance to reflect qualified UVCB dosing approaches, adapted in vitro test methods including OECD GD 23 and GD 286/GIVIMP, and UVCB-specific QSAR frameworks under the QAF. Publish case examples for UVCB read-across and other UVCB-adapted approaches.	Mid
Achieve broader and consistent regulatory adoption of animal-free UVCB assessment frameworks, including integrated NAM test batteries, predictive models, and PBPK/PBTK-supported reverse dosimetry under REACH and equivalent frameworks by using established templates for method qualification and reporting.	Long

Pillar 3: Coordination, capacity-building and governance Ensuring alignment across regulators, industry and academia through defined responsibilities, shared case studies and transparent implementation tracking.
Establish a dedicated UVCB working group under the EC roadmap governance structure to coordinate stakeholder input, prioritize method gaps, define responsibilities, and maintain a shared implementation tracker.	Short
Deliver targeted training and knowledge-sharing activities, including workshops, joint case studies and shared databases, to build regulatory capacity and a common understanding of UVCB-adapted NAM. Align with the roadmap's data-sharing and transparency commitments.	Mid
Track implementation through UVCB-specific indicators, including qualified case studies, guidance updates, regulator training uptake, regulatory acceptance and use of NAM-based approaches in UVCB registration dossiers, and documented reductions in animal testing. Periodically review progress and update roadmap priorities as needed.	Long

Timeline: Short = short-term, focus on immediate actions for reduction of animal testing today · Mid = mid-term, expand current approaches and increase NAM use in UVCB assessment · Long = long-term, full replacement of animal testing for UVCB substances; Abbreviations: NAM = New Approach Methodologies; UVCB = substance of unknown or variable composition, complex reaction products or biological materials; PBPK = Physiologically based pharmacokinetic; PBTK = Physiologically based toxicokinetic; HC = hydrocarbon; OECD = OECD Organisation for Economic Co-operation and Development; ECHA = European Chemicals Agency; QSAR = Quantitative structure–activity relationship; QAF = QSAR Assessment Framework; REACH = Registration, Evaluation, Authorisation and Restriction of Chemicals; GD = Guidance Document; GIVIMP = Good In Vitro Method Practices; EC = European Commission

In the short term, the focus should be on using and strengthening approaches that are already sufficiently mature. In the medium term, targeted evidence should be generated to qualify promising approaches for defined regulatory use and evidence from successful application should inform guidance and regulatory practice. In the long term, qualified components should be integrated into animal-free UVCB assessment frameworks. Progression through these stages should be based on technical performance, reproducibility, a defined applicability domain, exposure relevance, the ability to address whole-substance hazard, transparent characterization of uncertainty, and the relevance to the regulatory application. For UVCB substances, the resulting evidence should also support conclusion about the whole substance while accounting for variable composition and test-material complexity. Regulatory feedback throughout the process should inform adjustments to priorities, address remaining evidence gaps, and guide the progression of UVCB-adapted approaches towards regulatory implementation.

Conclusions

In the transition to animal-free safety assessments for both HH and ENV endpoints, difficult-to-test substances such as HC UVCB pose unique challenges for analytical characterization, in silico modelling, in vitro testing and kinetic modelling. Although substantial reductions in animal testing are already achievable through composition-activity models, grouping and read-across, and screening or indicator tests, these approaches alone are not sufficient to enable full replacement of animal testing, particularly for complex toxicological endpoints.

Further progress will require targeted research, fit-for-purpose test method development, and adaptations that address the specific properties of HC UVCB substances. Key priorities include improved analytical characterization, better dose metrics and dosing approaches for in vitro testing, development of PBPK-PBTK/IVIVE capability for complex mixtures, and predictive models with clearly defined applicability domains. Cross-sector collaboration is crucial to increase efficiency, especially for shared challenges such as data standardization, confidential data sharing, model validation and regulatory acceptance. Moreover, early alignment between research initiatives and regulatory requirements will be critical to ensure that new methods are developed with clear protection goals and practical regulatory use in mind.

Despite these challenges, HC UVCB provide an important test case for developing and evaluating UVCB-adapted animal-free assessment approaches, supported by extensive historical data, long-standing experience with complex substance testing, and recent advances in analytical characterization and NAM technologies.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work, the authors used Microsoft 365 Copilot to assist with language editing and to improve the clarity, readability, and sentence structure of the manuscript. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

CRediT authorship contribution statement

Daniela Holland: Writing – review & editing, Writing – original draft, Investigation, Conceptualization. Louise Camenzuli: Writing – original draft, Investigation, Writing – review & editing. Lenny Kamelia: Writing – review & editing, Writing – original draft, Investigation, Visualization. Maria Blanco-Rubio: Writing – review & editing, Writing – original draft, Investigation, Conceptualization.

Declaration of competing interest

The authors of this article are employed by companies that manufacture petroleum products. Daniela Holland and Louise Camenzuli were employed by ExxonMobil who provided funding for this study via ExxonMobil Biomedical Sciences Inc. (EMBSI), a wholly owned subsidiary of ExxonMobil Corporation and Maria Blanco-Rubio and Lenny Kamelia were employed by Shell Global Solutions International B.V.

Data availability

No data was used for the research described in the article.