New Approach Methodologies in Food Safety: A Survey and Commentary on Industry Adoption, Barriers, and the Path to Regulatory Adoption

Abstract

New approach methodologies (NAMs) are increasingly recognized as essential tools for modernizing regulatory science, yet their adoption in food safety risk assessment is not well characterized. To address this gap, we conducted a landscape analysis of NAMs use across the food industry. An anonymous survey was conducted between February and April 2026 with responses from seventeen international companies. Fourteen respondents reported using NAMs in the previous five years. Conventional cell-based assays, in silico toxicodynamic models, threshold of toxicological concern frameworks, and physiologically based pharmacokinetic (PBPK) modeling represented the most adopted approaches. More advanced platforms such as organ-on-chip systems were not routinely used. NAMs were predominantly applied to internal screening and prioritization, with ten companies reporting either experience with or interest in regulatory submissions incorporating NAMs-derived data. Novel food ingredients were the primary application area. Despite the interest in NAMs, significant barriers persist to their use in regulatory submissions, including challenges in data interpretation, limited availability of formally validated or qualified methods, high cost, and lack of specialized expertise to generate and interpret data. Regulatory acceptance emerged as the dominant factor influencing both internal adoption and submission decisions. Respondents indicated strong willingness to engage regulators through voluntary data-sharing mechanisms under appropriate “safe-space” conditions. These findings indicate that NAMs use in the food sector is established, but remains constrained and that progress toward broader regulatory integration will depend on fit-for-purpose guidance, clearly defined contexts of use, and improved frameworks for evidence integration.

Who Is Using NAMs in Food Safety and How?

Regulatory agencies in the United States, European Union, and beyond are increasingly focused on integrating new approach methodologies (NAMs) into safety assessment practice (Lu et al., 2025; Schmeisser et al., 2023). This shift is further accelerated by legislative and policy mandates from the regulatory agencies (Empl et al., 2026; European Commission, 2026; Tieghi et al., 2025; US FDA, 2025) that aim to replace, reduce and refine animal testing, increase the use of new methods from computational and in vitro science, and include more mechanistic data into evaluations of the expanding universe of novel substances and drug modalities (Wu et al., 2026).

There are already many examples of the use of NAMs in drug safety evaluations, both for internal decisions at the pharmaceutical companies (Beilmann et al., 2025) and in regulatory submissions (Avila et al., 2023; Yang et al., 2022; Yao et al., 2025). There have also been concerted efforts to assess non-animal models for various toxicity testing endpoints required for agrochemicals and antimicrobial products that have resulted in formal Organization for Economic Cooperation and Development (OECD) test guidelines and acceptance by Federal Agencies in the United States (Daniel et al., 2025; Raabe et al., 2025). NAMs have been integrated into the standard safety assessment of food ingredients in some areas, such as evaluation of genotoxicity and allergenicity (EFSA Panel on Nutrition Novel Foods and Food Allergens et al., 2024; Hanlon et al., 2026). However, only a few examples of how NAMs are used for comprehensive hazard and risk assessments of chemicals in food are available (Wood et al., 2025) reflecting that, despite the use of NAMs for chemical hazard characterisation (e.g. under REACH/ECHA), food safety assessments still often rely on traditional safety studies, despite encouragement from the regulators (Cattaneo et al., 2026). The scientific advances and recognition of opportunities for NAMs have already led to an EU food industry commitment to adopt and drive the transformation to accelerate the regulatory uptake of NAMs in food safety risk assessment through greater consistency, clearer guidance and strengthened dialogue between regulators and applicants (FoodDrink Europe, 2026).

To characterize current industry practices and perspectives, an anonymous survey was conducted between February and April of 2026. The survey was designed to mirror, to the extent possible, previous surveys conducted by the International Consortium for Innovation and Quality (iQ) Microphysiological Systems (MPS) affiliate (Baker et al., 2024). A working group consisting of the authors of this manuscript adapted the iQ MPS surveys and tailored them to the food sector. The final survey used in this analysis consisted of 36 questions (Supplemental File 1). The survey was distributed to multinational companies producing or using food ingredients, raw material food commodities, and finished products. These companies are members of The Institute for the Advancement of Food and Nutrition Sciences (IAFNS), a non-profit organization that catalyzes science for the benefit of public health, and Association for Sustainable Food Safety (ASFS), a technical association that develops sustainable strategies on chemical food safety topics and is a voice for the food industry in service of delivering a safe food supply for consumers. The survey was conducted anonymously using Qualtrics, and the results (Supplemental Files 2–3) capture responses from seventeen companies of varying sizes, from 1,000 to 10,000 employees (four) to more than 10,000 employees (thirteen). While we do not know which companies submitted responses, we are confident that the information collected provides a global view on the use of NAMs by the food industry because we have received confirmation of survey completion from companies headquartered in North America, Europe, and Asia & Oceania.

Fourteen of the responding companies reported using NAMs within five years. Activities relied on a mixed resourcing model: most (eleven out of fourteen) companies contracted studies to for-profit third parties, while five generated data in-house, five collaborated with academic or non-profit laboratories, and five engaged in NAM data generation through trade associations or consortia. Organizational structures ranged from centralized (three companies) to segmented (seven) to hybrid (four) models. The number of personnel involved in NAM-related work varied widely but is very small, considering the size of the companies – from fewer than five (eight companies) to 5–10 (five) to up to fifty (one) individuals per organization. The hybrid model aligns with practices observed in the pharmaceutical sector and reflects both the breadth of NAM approaches and the practical constraints in building internal expertise, as well as differing needs across the food value chain. It also underlines the importance of collaborative networks between industry, academia, and regulatory bodies for sustaining momentum in NAM development and deployment.

Which Biological Systems and Endpoints Are Being Used?

There is no universally accepted definition of NAMs, and definitions vary by sector (National Academies of Sciences Engineering Medicine, 2023). In drug safety, NAMs typically refer to advanced in vitro testing platforms such as bioprinted tissues and organ-on-chip models (Marx et al., 2025) In food safety, NAMs more broadly include in vitro, in silico, and integrated approaches (Marx-Stoelting et al., 2026). For this survey, NAMs were defined as “new alternative methods [those that do not have an OECD test guideline] to assess safety and may include in vitro, in silico, and in chemico approaches, or an integration of multiple approaches, i.e., IATA, that help reduce the use of animals in safety testing.”

Most respondents to the survey (eleven of fourteen) reported using conventional cell-based assays, in silico toxicodynamic models (e.g., Quantitative Structure-Activity Relationship (QSAR) modeling or threshold of toxicological concern (TTC), nine) and PBPK approaches (seven). The use of more technically advanced NAMs was less common: complex in vitro models such as spheroids were used by only three companies, and organ-on-chip systems by just one. Additional NAMs mentioned in free-text responses included read-across, co-culture models, ToxTracker assays, omics-based methods, and integrated animal models designed to capture multiple endpoints within a single study. This distribution confirms that food-sector adoption of NAMs is currently focused on relatively established, operationally accessible methods with a track record of publications and/or of integration into tiered risk assessment strategies rather than on more advanced in vitro methods. It is also clear that the food industry, as well as some regulatory agencies, also considers recently developed short-term animal studies as acceptable NAMs because they can be used to both derive points of departure and better understand the molecular mechanisms of toxicity (U.S. EPA, 2023). In addition, TTC and read across have been widely adopted in areas where there is limited data, but human exposure is anticipated to be very low and/or structural similarities allow for the bridging of data. Read-across and TTC also tend to be areas where an assessment may be needed but is not for seeking regulatory approval. Still, the use of more sophisticated NAMs for evaluation of food additives and novel foods is hampered by the uncertainties of acceptance and questions about scientific maturity of various new models as compared with familiarity with advantages and limitations of animal studies.

Genotoxicity and mutagenicity endpoints were the most used by food companies (eight of fourteen), followed closely by gut models (seven of fourteen). Models that can evaluate immune and inflammatory endpoints, liver, kidney, nervous system, skin sensitization, and developmental and reproductive toxicity were used by two to four companies. Cardiovascular and inhalation models were used by only one company each, which could be due to potential differences in capabilities as some companies operate in multiple sectors beyond just food, and most respondents indicated they were not using multi-organ integrated in vitro systems (one out of fourteen).

When asked about future priorities, models for the gastrointestinal system ranked highest, followed closely by genotoxicity and mutagenicity, and liver. This profile reflects the nature of food safety science needs: substances are primarily encountered through oral exposure, gut physiology is central to bioavailability and local toxicity, and genotoxicity testing remains a foundational regulatory requirement. Kidney, developmental and reproductive toxicity (DART) and immune system/inflammatory response NAMs were in the second tier of priorities with other organ systems being not highly prioritized. The emphasis on immune function and gut biology also speaks to growing scientific interest in the gut microbiome, allergenicity, and systemic immune responses as relevant outcomes for novel food ingredients. By contrast, pharmaceutical companies in comparable surveys have consistently prioritized liver, kidney, gastrointestinal tract, central nervous system, and cardiovascular endpoints – organs and tissues that are central to drug metabolism, systemic safety, and pharmacological activity. Both sectors share interest in gut and liver biology, but the scientific and commercial drivers behind that interest differ meaningfully.

Interestingly, expectations of the biological NAMs throughput were modest. Most food-sector respondents (twelve of fourteen) reported a desire for fewer than twelve experimental conditions in a single experiment, with two companies wishing the throughput is 12–24 conditions. This finding likely reflects focus on specific ingredients vs. larger screening exercises and may also factor in reliance on in silico and conventional in vitro approaches before proceeding to more complex NAMs. Pharmaceutical companies (Baker et al., 2024) generally considered that up to 24 replicate conditions is sufficient, consistent with the lower-throughput nature of advanced MPS platforms.

When choosing which NAMs to use, food-sector companies consistently identified validation status and regulatory acceptance as important criteria (Figure 1). Regulatory acceptance was the most important factor, with nearly all respondents stressing the need for confidence that submitted NAMs data will be accepted by authorities. Scientific quality was the second major priority, including validation against recognized methods, reliability, reproducibility, and standardization. Relevance to the specific regulatory purpose and practical considerations, such as cost, throughput and the availability of public guidance on regulatory positions for NAMs adoption, were also important, but secondary factors.

Figure 1.

Word cloud of the answers to survey question “What are the most important factors that you would consider when submitting NAMs data to a regulatory agency?” The word cloud was prepared using freely available online tool: https://www.jasondavies.com/wordcloud/.

Among companies using in silico models, government agency open-access tools were the most commonly used source (ten of fourteen respondents), followed by commercial fee-for-service platforms or academic publications (six each), and in-house developed models (three). The prominence of government-provided tools is noteworthy: it suggests that accessibility, institutional credibility, and perceived regulatory familiarity all factor into model selection decisions, not just scientific performance alone.

Taken together, these findings reinforce a consistent pattern across both pharmaceutical and food ingredient sectors: model choice and in-house uptake is driven not necessarily by the novelty of the NAM but by confidence in its utility; specifically, confidence that a method is scientifically fit for purpose, that its outputs are interpretable, and that it can meaningfully contribute to decision-making in a recognized regulatory context.

How Are NAM Data Being Used?

We observed that NAM-derived data in the food sector were used most commonly for internal decision-making and screening or prioritization purposes, each reported by ten and nine, respectively, of the fourteen NAM-using companies. Internal uses included screening substances for further testing, understanding potential mechanisms of action, and supporting read-across justifications. Fewer respondents mentioned using NAMs for characterization of toxicokinetic parameters, validation of the observation in animal studies, or in support for TTC approach (even though many respondents reported using TTC). An important nuance is that testing or screening individual substances (e.g., a pure chemical in an in vitro assay or computational model) may not adequately reflect how that substance behaves in a finished food product, which is a complex mixture of many components interacting with each other, for which a NAM-based approaches may not be easy to implement.

Ten of fourteen respondents reported collecting NAM data for regulatory submissions with eight indicating that they have included such data in regulatory filings. Different agencies were identified for such filings including the FDA, Health Canada, EFSA, FEMA, and ECHA. This level of regulatory submission activity distinguishes the food sector from its pharmaceutical and cosmetic counterparts in important ways – food safety assessments still largely rely on in vivo–derived dose–response metrics for establishing health-based guidance values (e.g., under EFSA). In a comparable pharmaceutical survey (Baker et al., 2024), no respondents reported submitting advanced MPS data in regulatory filings in the prior five years, albeit such submissions have started since that time (Ingber, 2026). The food sector’s submission experience most likely reflects the wider and more mature mix of methods included under the NAM umbrella, particularly in silico tools and some conventional in vitro assays, which have a longer track record of regulatory use compared with cutting-edge MPS devices. The conventional in vitro cell-based assays used, however, may be predominantly the traditional genotoxicity assays or assays to screen for endocrine activity and to a lesser extent assays to quantify other systemic endpoints, which may not have been adopted in regulatory submissions yet. It also suggests that NAM-based assessments are already a working reality in some food regulatory contexts, even if full replacement of animal studies for complex systemic endpoints remains a long-term aspiration.

Despite relatively widespread adoption of NAMs, considerable practical barriers to internal implementation remain. The most frequently reported challenge was difficulty interpreting NAMs data, cited by nine of fourteen companies. Difficulty with finding validated in vitro NAMs and generating sufficiently robust data packages for model qualification were each reported by eight companies. High cost was cited by seven respondents, and lack of qualified personnel by six. These barriers cluster around a core challenge: the outputs of most NAMs differ conceptually from those of traditional animal studies. Rather than producing apical endpoints, such as observable adverse outcomes at the organism level, most NAMs generate mechanistic, pathway-level, or narrowly focused toxicokinetic data. Translating this information into conclusions about hazard or risk requires interpretive frameworks, reference datasets, and weight-of-evidence reasoning that are still being developed. Validation against actual in vivo outcomes and qualification for specific decision contexts remain technically and organizationally demanding tasks, and the absence of standardized protocols compounds these difficulties.

Novel food ingredients were the most common current area of NAM use among food-sector respondents (thirteen of fourteen), followed by proteins (seven), food contact materials (six), and botanicals (five). Novel food ingredients were also the leading category for anticipated future NAM use, and they ranked among the most frequently cited challenging categories, as they may be used at high doses and can be also a complex mixture rather than a single substance. Additional difficult substance categories included complex mixtures, contaminants, chemicals that are hard to test using conventional methods, and endocrine-active substances. Respondents also highlighted functional foods, gut microbiome-related questions, immune function, allergenicity, repeated-dose toxicity, and complex mixture assessment as areas where NAMs could add particular value.

This landscape reflects several converging pressures: the rapid expansion of novel food ingredients derived from biotechnology and fermentation, the inherent complexity of botanical and macromolecular substances, and the recognized limitations of standard rodent bioassays for assessing immunotoxicity, developmental neurotoxicity, and systemic effects of substances encountered chronically at low doses. These are precisely the areas where traditional toxicology has the greatest difficulty producing human-relevant answers and where NAMs have a strong scientific rationale for offering improvements.

Regulatory Acceptance: Unlocking the Potential of NAMs

Regulatory considerations emerged as the prominent determinant of NAM adoption across the survey with companies identifying regulatory acceptance as the dominant factor in submission decisions, prioritizing it above even scientific quality metrics such as validation and reproducibility. Among companies that had not included NAM data in regulatory filings, key concerns were acceptance of models and acceptance of the resulting data (six of fourteen companies). Internal consideration of insufficient confidence in the data was cited by four respondents. A persistent discrepancy exists within the food sector between guidance and regulatory practice: while some food safety authorities recommend the use of NAMs in safety assessments, regulatory agencies often do not accept these approaches and instead require traditional in vivo studies.

Companies also raised more nuanced concerns that go beyond simple acceptance or rejection. Respondents expressed uncertainty about how regulators would interpret NAM data, particularly negative findings, and whether human-relevant results in cell-based models that diverge from animal data would be treated as valid or would be discounted in favor of conventional endpoints from animal studies. Concerns that positive findings may be more readily accepted than negative ones points to a potential asymmetry that could distort how NAM results are used in practice (Wetmore et al., 2019). More fundamentally, respondents described the paradox of being asked to validate highly human-relevant methods against animal data that were themselves never formally validated for human relevance.

Despite these concerns, eight of fourteen respondents indicated they would submit blinded NAM data to regulatory agencies if there were no associated program risks, such as market access or delays in time to market, through a “safe space” mechanism, a proportion strikingly similar to the affirmative responses recorded in pharmaceutical-sector surveys. This convergence suggests that industry willingness to engage regulators is substantial, but that the current regulatory environment does not yet provide sufficient predictability and protection to make voluntary submission routine.

Why Have Some Companies Not Adopt NAMs?

Three of the seventeen survey respondents reported no NAM use in the previous five years. All three identified the absence of formal regulatory guidance, such as OECD test guidelines or equivalent frameworks, as a primary reason for non-adoption. One respondent additionally indicated that key internal decision-makers viewed NAM technology as unproven and unlikely to meet near-term expectations. Notably, no respondents cited lack of resources or prior regulatory rejection of NAM data as reasons for not adopting NAMs.

Despite not currently using NAMs, these companies identified meaningful potential applications, including repeat-dose systemic toxicity testing (as a replacement for 28-day and 90-day animal studies), screening for mechanism of action, supporting read-across, and assessing novel food ingredients from emerging production technologies such as precision fermentation. These needs align closely with broader recognition that repeated-dose toxicity and complex food-relevant materials remain key gaps in NAM development.

The Path Forward: From “Name the Assay we Should Use” to “Define the Decision Context”

A central conceptual challenge for regulatory integration of NAMs is that these methods are not simple one-for-one replacements for animal studies. NAMs represent compendia of approaches and evidence-integration strategies that generate different types of data, and no single assay is likely to serve as a universal substitute for a traditional in vivo endpoint. The regulatory question therefore shifts from “Which assay replaces this animal test?” to “What combination of evidence is sufficient for this decision?” There are examples of validated in vitro and in silico methods that have largely replaced animal tests (e.g., for skin sensitization) but these assays are typically used for hazard classifications and labeling, not for risk assessment (Islas-Robles et al., 2025). In food safety risk assessment, derivation of a point of departure remains central for establishing health-based guidance values.

NAM use also creates practical tension – industry seeks clarity on acceptable methods, while regulators are cautious about prescribing specific assays across all contexts. Regulators, however, are often reluctant to prescribe specific assays across all contexts, because the scientific value of a NAM depends heavily on its context of use, biological domain, performance characteristics, and integration with other evidence streams. Prescriptive lists risk both premature standardization and exclusion of valid approaches Additionally, developers of NAMs do not always design methods aligned with regulatory decision needs (Rusyn, 2025). An example is the long process for reaching qualification status by one of the first MPS to attempt “regulatory acceptance” – the Emulate liver MPS (Ingber, 2026).

Therefore, evidence from this and other surveys supports movement towards fit-for-purpose, tiered framework for regulatory NAM use. This opinion is shared by authorities and the regulators (Cattaneo et al., 2026; Cronin et al., 2025). In such frameworks, NAMs are first applied in established contexts: screening, prioritization, read-across, mechanistic investigation, toxicokinetic modeling which would build confidence through case studies. This approach has been embraced by the cosmetics industry through the Next Generation Risk Assessment (NGRA) framework (Carmichael et al., 2022). In the analysis of ECHA compliance decisions (Roe et al., 2026), it was found that simpler read-across hypotheses achieved higher regulatory success, especially when substantiated by toxicokinetic and other NAMs data. More complex decisions, such as repeated-dose systemic toxicity assessment or reproductive toxicity, need to draw on more complex arguments and weight-of-evidence approaches, integrated approaches to testing and assessment (IATAs), or risk assessment paradigms that combine multiple NAM outputs (Caloni et al., 2022).

From this perspective, progress depends less on defining specific assays and more on clearly defining contexts of use, evidentiary expectations, and interpretation frameworks. By developing targeted case studies, industry can demonstrate that combinations of NAMs yield sufficiently robust data for regulatory confidence and decision making. Progress accelerates when guidance evolves from mandating specific tools to establish distinct decision contexts and confidence thresholds, which fosters scientific innovation while preserving regulatory integrity.

To enable this, food safety authorities and regulators should provide fit-for-purpose guidance, context-of-use qualification pathways, and illustrative case studies. Low-risk “safe space” (European Commission, 2026) data-sharing mechanisms would help build the shared evidence base needed for regulatory confidence for voluntary data sharing, analogous to the “safe harbor” arrangements used in drug safety evaluations (Liberti et al., 2017). Equally important is making clear how mechanistic and pathway-level NAMs evidence will be interpreted alongside, or instead of, traditional animal data, and how human-relevant results that differ from animal findings will be handled rather than automatically discounted.

Ultimately, the food and pharmaceutical sectors are moving in the same strategic direction: toward greater reliance on human-relevant, mechanism-informed, non-animal methods (Table 1). They currently occupy different positions along that path, shaped by differences in the methods available, the nature of the substances being assessed, and the maturity of regulatory frameworks in each domain. The food sector’s broader and more varied NAM portfolio, combined with its demonstrated experience in regulatory submission of established methods, provides a foundation to build on. Closing the gap between current practice and full regulatory integration will require coordinated effort across industry, regulatory agencies, and the scientific community, but the direction of travel is clear, and the momentum is real.

Table 1.

Comparisons of major topics on the use of NAMs between the pharmaceutical and food industries.

Topic	Pharmaceutical Industry Approach (Baker et al., 2024; Beilmann et al., 2025; Pognan et al., 2023)	Food Industry Approach (this study)
Types of NAMs used	Stronger emphasis on advanced systems such as microphysiological systems (MPS), organ-on-chip, iPSC-based models, and mechanistic investigative toxicology tools, although adoption is still limited by qualification, reproducibility, and cost	Using a mix of established and emerging NAMs: conventional cell-based assays, QSAR/in silico toxicodynamic models, TTC, and PBPK/toxicokinetic models across contexts; advanced systems like organ-on-chip remain uncommon
Main contexts of use	NAMs are used largely for mechanistic investigation, hazard profiling, translational safety questions, and specific discovery/development decisions; advanced MPS remain primarily internal and are not yet routine in regulatory filings	NAMs are used primarily for screening, prioritization, mechanism understanding, read-across, and support for novel ingredient assessment; established NAM approaches (e.g., QSAR, TTC) are already being used in some regulatory submissions
Regulatory maturity and submissions	Regulatory acceptance remains a major barrier; few companies report submitting advanced MPS data, and companies request qualification pathways, case examples, and safe-harbor data-sharing options	Regulatory acceptance is also the primary bottleneck, but compared with advanced pharma MPS, food companies report more regulatory use of NAM-derived data, likely reflecting greater reliance on established in vitro and in silico approaches
Strategic goals for increased NAM adoption	NAMs as fit-for-purpose tools for defined questions, especially where animal models are poor or non-relevant for certain modalities, human-specific targets, or mechanistic toxicology questions	Similarly moving toward fit-for-purpose use, shifting away from identifying single replacements for animal studies and toward defining context of use, evidence integration, and confidence thresholds for food safety decisions

Highlights.

• NAM use is established for food safety applications, but uneven across companies

• Adopted NAMs comprise cell assays, QSAR/TTC, and PBPK models

• NAM data mainly support screening, prioritization, and read-across

• Regulatory acceptance remains the main barrier to wider NAM adoption and use

• Fit-for-purpose guidance and safe-space data sharing are needed to increase use

List of Abbreviations

NAMs

New approach methodologies

PBPK

physiologically based pharmacokinetic

OECD

Organization for Economic Cooperation and Development

iQ

International Consortium for Innovation and Quality

MPS

Microphysiological Systems

IAFNS

Institute for the Advancement of Food and Nutrition Sciences

ASFS

Association for Sustainable Food Safety

QSAR

Quantitative Structure-Activity Relationship

TTC

Threshold of toxicological concern

DART

Developmental and reproductive toxicity

Data availability

Data for the analyses described in the article is made available in full through Supplemental information