IARC Workshop on the Key Characteristics of Carcinogens: Assessment of End Points for Evaluating Mechanistic Evidence of Carcinogenic Hazards

Abstract

BACKGROUND: The 10 key characteristics (KCs) of carcinogens form the basis of a framework to identify, organize, and evaluate mechanistic evidence relevant to carcinogenic hazard identification. The 10 KCs are related to mechanisms by which carcinogens cause cancer. The International Agency for Research on Cancer (IARC) Monographs programme has successfully applied the KCs framework for the mechanistic evaluation of different types of exposures, including chemicals, metals, and complex exposures, such as environmental, occupational, or dietary exposures. The use of this framework has significantly enhanced the identification and organization of relevant mechanistic data, minimized bias in evaluations, and enriched the knowledge base regarding the mechanisms of known and suspected carcinogens. OBJECTIVES: We sought to report the main outcomes of an IARC Scientific Workshop convened by the IARC to establish appropriate, transparent, and uniform application of the KCs in future IARC Monographs evaluations. METHODS: A group of experts from different disciplines reviewed the JARC Monographs experience with the KCs of carcinogens, discussing three main themes: a) the interpretation of end points forming the evidence base for the KCs, b) the incorporation of data from novel assays on the KCs, and с) the integration of the mechanistic evidence as part of cancer hazard identification. The workshop participants assessed the relevance and the informativeness of multiple KCs-associated end points for the evaluation of mechanistic evidence in studies of exposed humans and experimental systems. Discussion: Consensus was reached on how to enhance the use of in silico, molecular, and cellular high-output and high-throughput data. In addition, approaches to integrate evidence across the KCs and opportunities to improve methodologies of mechanistic evaluation of cancer hazards were explored. The findings described herein and in a forthcoming IARC technical report will support future working groups of experts in reporting and interpreting results under the KCs framework within the JARC Monographs or in other contexts. https://doi.org/ 10.1289/EHP15389

Headnote

BACKGROUND: The 10 key characteristics (KCs) of carcinogens form the basis of a framework to identify, organize, and evaluate mechanistic evidence relevant to carcinogenic hazard identification. The 10 KCs are related to mechanisms by which carcinogens cause cancer. The International Agency for Research on Cancer (IARC) Monographs programme has successfully applied the KCs framework for the mechanistic evaluation of different types of exposures, including chemicals, metals, and complex exposures, such as environmental, occupational, or dietary exposures. The use of this framework has significantly enhanced the identification and organization of relevant mechanistic data, minimized bias in evaluations, and enriched the knowledge base regarding the mechanisms of known and suspected carcinogens.

OBJECTIVES: We sought to report the main outcomes of an IARC Scientific Workshop convened by the IARC to establish appropriate, transparent, and uniform application of the KCs in future IARC Monographs evaluations.

METHODS: A group of experts from different disciplines reviewed the JARC Monographs experience with the KCs of carcinogens, discussing three main themes: a) the interpretation of end points forming the evidence base for the KCs, b) the incorporation of data from novel assays on the KCs, and с) the integration of the mechanistic evidence as part of cancer hazard identification. The workshop participants assessed the relevance and the informativeness of multiple KCs-associated end points for the evaluation of mechanistic evidence in studies of exposed humans and experimental systems.

Discussion: Consensus was reached on how to enhance the use of in silico, molecular, and cellular high-output and high-throughput data. In addition, approaches to integrate evidence across the KCs and opportunities to improve methodologies of mechanistic evaluation of cancer hazards were explored. The findings described herein and in a forthcoming IARC technical report will support future working groups of experts in reporting and interpreting results under the KCs framework within the JARC Monographs or in other contexts. https://doi.org/ 10.1289/EHP15389

Introduction

Cancer hazard identification is a critical step to inform decisions on preventive actions.!· Since 1971, the Monographs Programme of the International Agency for Research on Cancer (IARC) of the World Health Organization (WHO) has conducted hazard evaluations of agents that are suspected to be carcinogenic. International working groups of scientists review the relevant literature and conduct evaluations according to well-established and rigorous procedures described in the Preamble to the IARC Monographs.>· Three streams of scientific evidence are considered for each evaluation: a) studies of cancer in exposed humans, b) studies of cancer in experimental animals, and с) mechanistic evidence. Consensus conclusions on these independent streams of evidence are integrated to reach an overall classification regarding an agent's carcinogenicity. To date, 128 agents have been classified as carcinogenic to humans (IARC Group 1), 95 as probably carcinogenic (Group 2A), 323 as possibly carcinogenic (Group 2B), and 500 as not classifiable as to its carcinogenicity to humans (Group 3) (https://monographs. iarc.who.int/).

Consideration of potential mechanisms of carcinogenic exposures has supported the overall evaluation of carcinogenicity since the earliest JARC Monographs volumes.>® An analysis conducted in a previous workshop organized by the JARC Monographs programme investigated the available data on mechanisms of known human carcinogens (Group 1) and identified 10 key characteristics (KCs) of carcinogens"8: KC1 "is electrophilic or can be metabolically activated to an electrophile"; KC2 "is genotoxic"; KC3 "alters DNA repair or causes genomic instability"; KC4 "induces epigenetic alterations"; KC5 "induces oxidative stress"; KC6 "induces chronic inflammation"; KC7 "is immunosuppressive"; KC8 "modulates receptor-mediated effects"; KC9 "causes immortalization"; and KC10 "alters cell proliferation, cell death, or nutrient supply."

Until the most recent update of the Preamble in 2019, evidence on carcinogen mechanisms was formally used to modify the overall evaluation based on evidence of cancer in humans and cancer in experimental animals. Following the implementation of the revised Preamble, mechanistic evidence is now considered simultaneously alongside the other two streams of evidence, reflecting the increased availability, diversity, and applicability of mechanistic information to identify cancer hazards.·

To consistently and transparently review the available mechanistic literature, a framework based on the 10 KCs of carcinogens710 has been applied in evaluating agents across 24 JARC Monographs volumes. For every IARC Monographs evaluation, the working group reviews the available scientific information that is identified using specific search terms for each of the KCs. The data include studies of exposed humans (epidemiological studies) and of experimental systems in vivo and in vitro, including human primary cells and tissues. The literature is evaluated for quality and informativeness.· Notably, the KCs framework does not require an a priori hypothesis of how an agent may cause cancer and, thus, enables a more systematic evaluation of diverse agents."-%1°

After extensive use of the KCs during the past 8 y, the JARC Monographs programme considered it timely to review their utility and consider their furtherance. Thus, the IARC convened a scientific workshop with the goal of ensuring appropriate, transparent, and uniform application of the KCs in future JARC Monographs evaluations.

On 23-28 July 2023, 24 experts across various disciplines from eight countries met in Lyon, France, with IARC scientists to discuss a) the interpretation of end points forming the evidence base for the KCs in exposed humans and in experimental systems, b) the incorporation of data from novel assays of the KCs, and c) the integration of the mechanistic evidence for the KCs as part of cancer hazard identification. The recommendations aim to ensure consistency in applying the KCs in JARC Monographs evaluations, as well as the flexibility to allow the incorporation of emerging scientific methodologies and concepts into the KCs. This article summarizes all topics of the workshop and the major outcomes of the discussions. A follow-up technical report will provide expert opinions mainly on the reporting and interpretation of end points relevant to the KCs.

Methods

IARC Monographs Scientific Workshop participants encompassed broad experience and expertise related to the 10 KCs, including relevant methodologies, computational and alternative approaches, and molecular epidemiology. This broad spectrum of expertise ensured a multilateral perspective on the challenges and opportunities to improve the application of the KCs framework. Several of the experts had experience in JARC Monographs evaluations that had used the KCs framework, and some had participated in the 2012 IARC Workshop on tumor site concordance and mechanisms of carcinogenesis.®

The workshop participants were screened to avoid any conflict of interest.? One expert with a conflict of interest but with critical knowledge and experience deemed important for the workshop was allowed as an invited specialist. His role was to participate in the discussion without contributing to any section of the technical report. Representatives from governmental agencies and observers with relevant scientific credentials were admitted with the specific role of observing without influencing the discussion or drafting any section of the technical report.

The workshop participants were invited to reflect upon the three themes: a) interpretation of end points, b) incorporation of new data, and с) integration of the mechanistic evidence, using as a basis of discussion a retrospective analysis of the mechanistic evidence evaluated in the JARC Monographs with the KCs. The analysis provided by the IARC secretariat included information available in summary sections 5 and 6 of the IARC Monographs from volume 112 to volume 135 (Excel Table S1). In the summary sections of the JARC Monographs, results pertinent to evaluating the mechanistic evidence on carcinogenicity were summarized to indicate how the evaluation was reached.

Discussion

The commentary highlights the experts' recommendations and conclusion on the three themes, relative to the analysis of the agents evaluated with the KCs; the interpretation of endpoints; the incorporation of data from high-content assays; and the integration of the evidence. In addition, it summarises the consensus of several discussions.

Comprehensive Analysis of the KCs of Carcinogens for Agents Reviewed since IARC Monographs Volume 112

According to the IARC Monographs Preamble,' the evaluation of the KCs uses a standard language to evaluate the strength of the mechanistic evidence. The evidence is considered strong when the results in several different systems are consistent and the overall database is coherent. The results are considered limited if the evidence is suggestive but, for example, the studies cover a narrow range of experiments, relevant end points, or species; there are unexplained inconsistencies in the studies of similar design; or there is unexplained incoherence across studies of different end points or in different experimental systems. The mechanistic evidence is classified as inadequate if few or no data are available; there are unresolved questions about the adequacy of the studies' design, conduct, or interpretation; or the available results are negative. The available mechanistic evidence in exposed humans, human primary cells or tissues, and other experimental systems is described separately, forming the whole body of evidence.?

The workshop participants noted that across the 98 agents evaluated in 24 IARC Monographs, references were identified for all 10 KCs, and most of the agents had data related to several KCs. This was noted as an advantage of the KC framework because it allows the identification of patterns and associations in the mechanisms of carcinogenic agents, as well as data gaps.'! The workshop participants agreed that the ability to review the mechanistic evidence using an unbiased approach is important to direct further research on cellular and molecular mechanisms of carcinogens.

Of these 98 agents, 48 exhibited strong mechanistic evidence supporting one or more of the KCs (Excel Table S1). The mechanistic evidence was the sole determinant of the overall evaluation in a few cases, such as arecoline and crotonaldehyde!? and, more recently, for perfluorooctane sulfonic acid, known as PFOS.!· This led in each case to a classification of Group 2B. The remaining 45 agents, with strong mechanistic evidence, exhibited sufficient evidence for cancer in animals or cancer in humans or both, or additional limited evidence for cancer in humans. Agents with strong mechanistic evidence exhibited consistent and coherent evidence for between one and seven KCs, with an average of three KCs per agent (Figure 1A; Excel Tables S1 and S2). More than half (63%) of these agents exhibited consistent and coherent evidence for KC2 ("is genotoxic"), considered a major driver of carcinogenesis for which mechanistic end points are often available. Notably, no agent with strong mechanistic evidence exhibited consistent and coherent evidence for KC9 ("causes immortalization") (Figure 1B; Excel Table S3). This is not surprising because "causes immortalization" is a characteristic commonly observed for viruses," which were not among the agents evaluated since volume 112, and the available literature investigating cell immortalization induced by chemicals is sparse. In addition, the workshop participants noted that the most frequent challenge for working groups conducting the evaluation of KC9 appeared to be a general lack of evidence directly and specifically pertaining to the multistep process of immortalization in human cells for the agent(s) of interest.

An analysis of the types of test system showed that all the agents with strong mechanistic evidence had consistent and coherent data provided by experimental systems, in addition to for one agent, namely, occupational exposure as a firefighter, for which the data were derived largely from exposed humans. In general, for agents with consistent and coherent mechanistic evidence in exposed humans or evidence likely to be operative in humans, the database was always complemented by evidence in experimental systems or in human primary cells or tissues or both (Figure 1C). The source of data was not distributed equally across the KCs. Some KC-associated end points were observed largely in experimental systems, such as for KC2, KC5, KC6, and KC10, whereas others were observed more frequently in exposed humans, such as for KC4 (Figure 1D; Excel Table S4). The results reflected the availability of the techniques and research tools used, as well as the relevance of the specific end points within each KC.

Interpretation of End Points Forming the Evidence Base for the KCs in Exposed Humans and in Experimental Systems

A significant advantage of the KCs framework is that it encompasses a wide range of end points of known relevance to carcinogenesis because the end points were identified by examining the agents classified as Group 1 by the [АКС Monographs." A description of the end points that best define each KC and the assays available to measure KC end points have been previously described. " The workshop participants deliberated on the different degrees of relevance of the end points forming the evidence for the KCs. The relevance of each end point is an important determinant of the overall informativeness and importance of the findings under review for a specific KC. As per the Preamble,' experts participating in IARC Monographs evaluations consider the relevance of an end point along with factors affecting the study quality (e.g., exposure accuracy in exposed humans) and validity (e.g., reliability and sensitivity of the biological assays). The judgment of a working group is an essential component in the [ARC Monographs evaluations, as well as in determining the relevance of KC-associated end points.

The workshop participants discussed a variety of considerations that may play a role in establishing the relevance of a KCassociated end point, including its specificity and how well the end point explains the biological processes underlying the KC, the extent to which the end point has been associated with carcinogenesis or cancer risk, the test system in which the end point is observed, and the persistence of the alterations. The interpretation of some of the KC-associated end points has long been recognized, for example, for micronucleus formation, chromosomal aberrations, or other KC2-associated end points, and they are well established across test systems, with widely recognized testing protocols and standards.® For other KCs-associated end points, considerations may be more nuanced. For example, to establish the relevance of some KC6 (chronic inflammation)-associated end points, it is important to account for the length or persistence of the effects, including acute inflammation that repeatedly occurs over a long period (e.g., with repeated exposures). The workshop participants also discussed and confirmed that certain chronic inflammatory (nonmalignant) diseases can be considered relevant end points for KC6, especially when the chronic inflammatory disease is associated with cancer (including site concordance).

In some cases, additional information provided in conjunction with alterations in KC-associated end points can strengthen the conclusions. The workshop participants concluded that the evidence of the KCs should reflect mechanisms pertinent to how carcinogens cause cancer. For example, under KC8, higher relevance may be inferred when downstream effects are observed, or when it is possible to establish the directionality of the effect and its involvement in carcinogenesis, or by analogy to established carcinogens that exhibit these KCs. As in the case of polycyclic aromatic hydrocarbons (PAHs), a downstream effect of aryl hydrocarbon receptor (AhR) binding is the induction of specific cytochrome P450 (CYP) enzymes. The recognition of degrees of relevance of KCassociated end points can support the assessment of the strength of the mechanistic evidence. A thorough analysis of the relevance of various KC-associated end points by test system will be available in an [АКС Monographs technical report.

During the workshop, participants discussed the interpretation of the KC-associated end points in molecular epidemiological studies. The interpretation and synthesis of the mechanistic evidence among the KCs in observational studies of exposed humans has been challenging for IARC Working Groups. There are multiple considerations to account for in observational studies. These include quality of exposure assessment; exposure range and intensity; proximity in time of the most recent exposure to biological sample collection; comparability of the control group(s); sample size; quality of assay; magnitude of the biomarker effect and its statistical strength; potential confounding by co-exposures, as well as by demographics and lifestyle-related factors; appropriateness of the statistical analysis; potential that a study has produced a false positive finding; availability of replication data; and possibility of publication bias.

The workshop participants concluded that if end points used to assess the 10 KCs in humans have been tested for their ability to prospectively predict the risk of one or more types of cancer in humans, then this should be considered in evaluating the strength of the evidence. However, it is recognized that this is not always possible. End points measured in certain types of tissue samples can also potentially provide insight into the tissue specificity of an exposure's biological effects (e.g., biomarkers measured in urothelial cells and bladder cancer; certain types of airway samples and lung cancer; epidermal cells and skin cancer; and blood and certain hematologic malignancies).··15-16

An important aspect of evaluating mechanistic evidence is the coherence across mechanistic data observed in molecular epidemiological studies and those in experimental systems. The workshop participants discussed whether the strengths and limitations of the end points differed among the test systems. For example, in molecular epidemiological studies, blood was frequently used as a convenient surrogate for the target tissue(s) with little information on correlations between the two specimens. Studies in experimental systems can help corroborate such observations. Indeed, the coherence of findings across multiple test systems increased the strength of the evidence that an agent exhibited a specific KC, for example, coherence of findings in studies related to KC7 identified for perfluorooctanoic acid (PFOA).'3

Incorporation of Data from High-Content Assays in the KCs

In previous ZARC Monographs evaluations, data from high-content assays that may be informative for multiple KCs have been used mainly as supporting evidence under the KCs framework.·!7 The workshop experts discussed how to improve the interpretation and incorporation of these data for mechanistic evaluations. Transcriptomics, metabolomics, mutational signatures, in silico data, and chemical high-throughput screening data (e.g., Tox21, including ToxCast) were selected for discussion because they reflect the most frequent types of high-content assay data contributing to recent IARC Monographs.

Transcriptomics and metabolomics. Transcriptomics and metabolomics data hold substantial promise for cancer hazard identification because they contribute valuable information to understanding a broad range of molecular mechanisms of carcinogens, fill data gaps, and provide supporting evidence for the chemical grouping of agents. There was consensus among the experts to encourage incorporating omics data into the KCs framework. To improve the contribution of omics data to cancer hazard identification and to overcome some of the challenges of interpretation, the workshop participants proposed criteria to assess the quality of transcriptomic and metabolomic studies. These criteria to be elaborated in the forthcoming technical report, will facilitate the identification of the most informative studies.

In addition, there were discussions on the best strategies to incorporate omics data into the strength-of-evidence evaluation for the KCs. The workshop participants acknowledged that many omics end points can be associated with the KCs; as such alterations of these end points can strengthen the evidence that an agent exhibits one or more of the KCs. To establish such associations, working groups could map enriched pathways (based on altered transcripts or metabolites) to specific KCs by considering the relevance of the biological process(es) underlying an omics end point for a particular KC. In the future, it would be desirable to systematically incorporate data from high-quality, standardized public data repositories into ARC Monographs evaluations; however, this would require dedicated resources for data extraction, processing, and analysis. The workshop participants agreed that omics data should be integrated more systematically into the KCs framework to provide additional evidence for cancer hazard identification.

Mutational Signatures

Studies identifying mutational signatures of specific agents with cancer risk are increasing in number, and so is their relevance for cancer hazard identification.'®2° However, the association of mutational signatures with specific exposures requires a comprehensive understanding of the mechanisms underlying the generation of these signatures. This includes consideration of the different experimental models used, technical aspects of the study design, and data analysis and interpretation. Combined with mutational signature analysis using clinical and epidemiological sample collections, the derived signatures provide a link between the genotoxic modes of action of exposures and cancer risk. Workshop participants discussed criteria to assess the quality of mutational signature studies and the best strategies to incorporate the data into the strengthof-evidence evaluation for the KCs. The experts concluded that the currently characterized mutational signatures can be associated with biological processes relevant mainly to three KCs: KC2, KC3, and КС5. In addition, case studies have been developed by the workshop participants to describe mutational signatures associated with different exposures (or biological processes) acting directly or indirectly and to exemplify the diversity of possible associations found in the literature.

In Silico Toxicology

In silico toxicology uses computational models based upon different methodologies [e.g., statistical machine learning (artificial intelligence (AI)] methods and expert rule-based (alert) systems) to predict the toxicity of a chemical from its molecular structure and chemical and biological properties. The workshop participants agreed that many computational models focus on the prediction of assay end points forming the evidence base for the KCs in exposed humans and experimental systems and can be similarly assessed according to end point relevance, validity, and reliability of the biological assays. For example, quantitative structure-activity relationship (QSAR) modeling predicted the mutagenic potential of cupferron.?! During the workshop, the application of QSAR models and alerts supporting KCs 1, 2, and 8 were discussed; however, it was acknowledged that direct prediction of human carcinogenicity as an overall end point is not currently feasible.??

The experts agreed on the need for acceptance criteria. They also recommended that, when using results from computational models, high sensitivity (the ability to predict toxic compounds) and high negative predictivity (the ability to avoid false negatives) are desirable.

High-Throughput Screening Assays, Such as ToxCast/ Tox21

High-content and high-throughput in vitro data can serve as an additional or supportive source of mechanistic evidence.??3 JARC Monographs Working Groups pioneered the inclusion of data from the US Environmental Protection Agency (EPA) ToxCast/ Tox21 high-throughput screening (HTS) program to supplement other mechanistic evidence,' although large-scale screening programs measuring a variety of end points were designed to evaluate large chemical libraries to prioritize chemicals for additional toxicity testing rather than to identify the hazard of a specific chemical or chemical group. Since IARC Monographs volume 110, working groups have used an evolving mapping of ToxCast assay data to relevant KCs. An open-source software tool called kc-hits has since been developed to facilitate the process of summarizing, analyzing, and presenting the ToxCast data relevant to the KCs.?·

The strengths and limitations of the several different approaches to summarizing ToxCast assay results were discussed, and recommendations to improve the use and interpretation of data were provided. The experts recognized that to improve the use of ToxCast data for cancer hazard identification, efforts should be undertaken to identify or develop new HTS assays (e.g., to increase the applicability and coverage of the KCs), to better characterize the biological basis of each assay in the mapping, to update and reevaluate the current mapping, and to consider the potential for using machine learning and Al to enhance the use of these data in the Monographs.

Integration of the Mechanistic Evidence as Part of Cancer Hazard Identification

The workshop participants noted that a beneficial feature of the KCs framework is that several of the KCs are functionally related, adding to the value of each individual KC. The workshop participants explored the potential links among the KCs and discussed strategies to better describe the integration of the mechanistic evidence when data are available. They recognized that linkages among some KCs should be described via underlying KC-associated end points, rather than on the level of KCs. For example, carcinogens that are electrophilic can form DNA adducts (KC1), which can be mis-repaired or left unrepaired (KC3), possibly causing mutations (KC2). Oxidative stress (KC5), which can be caused by chronic inflammation (KC6), can result in oxidative damage to DNA (KC5), which, if not repaired (KC3), can ultimately result in mutations (KC2) that can lead to cancer. The technical report will provide some recommendations regarding which end points could be ascribed to each of the KCs and when they may be relevant to more than one KC. Linkages among KCs can generally strengthen the biological plausibility. For example, evidence of chronic inflammation (KC6) exhibited as the increase of inflammatory markers observed in several molecular epidemiological studies of humans exposed to air pollution becomes stronger with the additional evidence of a significant association between fine particulate matter [PM <2.5 um in aerodynamic diameter (PM; 5)] levels and the frequency of epidermal growth factor receptor (EGFR) mutant lung cancer incidence, both of which are examples of KC2 (PM; is mutagenic, and EGFR mutations are reflect genotoxicity).

In addition, it is important to note that multiple KCs may display additive or synergistic effects, which may considerably strengthen the overall mechanistic evidence of carcinogenicity. The workshop participants concluded that the overall strength of the mechanistic evidence can also be increased (or decreased) by coherence (or lack thereof) of data from multiple end points that may represent multiple KCs. Coherence can be based on "established" causal relationships between end points, such as within a plausible model of carcinogenesis.

Incorporating these features is challenging and may be attainable only in limited circumstances where the available published scientific literature has specifically tested a hypothesis regarding a temporal relationship among mechanistic events. One example showing a "causal relationship" is the longitudinal evidence that, besides genotoxicity and other KCs, tobacco smoke induces epigenetic alterations (KC4) as a plausible causal mechanism of lung carcinogenesis," although analysis of genetic variants as other factors, through Mendelian randomization, did not clearly confirm the observation.?" For other well-studied carcinogens, such as benzene, the precise sequence of mechanistic events may be unknown."

There is no requirement in the Monographs Preamble to establish the temporal, sequential or causal features of how a particular agent operates. As such, each KC can be seen as independent with a distinct role because the conclusion that an agent exhibits a KC means that this agent has chemical or biological properties similar to those of another agent known to be carcinogenic to humans (Group 1), indicating the utility of that KC in supporting a cancer hazard classification. When applying the KCs framework in the context of cancer hazard identification, the mechanistic evidence can be associated with a single KC or with multiple KCs. The evidence for KCS should be interpreted with caution unless found in conjunction with that of other KCs.?

KCs and Other Mechanistic-Type Frameworks

The workshop participants addressed how the KCs framework can be viewed in relation to two other mechanistic-type frameworks being used in different hazard and risk assessment contexts: the mode of action (MoA) proposed by the WHO/International Programme on Chemical Safety (IPCS),?· and the adverse outcome pathways (AOP) developed at the Organisation for Economic Co-operation and Development (OECD).? There was consensus on the utility of the three frameworks in carcinogen assessment. At the same time, the importance of clarifying the purpose and context of use of each framework was noted. The workshop participants deliberated about the importance of these frameworks to convert scientific data into reliable and relevant evidence streams that address many different assessment objectives, processes and contexts.39 The KCs framework, as employed by the IARC Monographs programme, was originally conceived and designed to support the scientific review and evaluation process to identify carcinogens. The MoA and AOP frameworks are conceptually similar to one another, although the ontology underpinning the AOPs is more elaborated and formally described. They both share the principle that a biological or toxicological process can be described by a sequence of discrete, causally linked key events (KEs). The MoA framework, as originally conceived and applied, relies on formulating a mechanistic MoA hypothesis (described by a chain of KEs) as a first step in assessing an actual agent. On the other hand, AOPs have been developed in an agent-agnostic way to design new toxicity testing strategies. The KC and AOP frameworks are similar in that KCs and their associated end points, as well as AOPs and their associated KEs, are defined independently from their use in assessing any particular agent. Therefore, both ontologies can be considered as knowledge bases.

As such, the workshop participants considered that the three frameworks represent different ways of organizing mechanistic information for use in assessments; therefore, each has a different conceptual underpinning and means of application in practice. The workshop participants considered that the KCs framework has a strong empirical basis, being derived from data on known human carcinogenic agents and designed for a retrospective analysis with procedures to minimize bias in assessing the published mechanistic literature to inform an evidence-based decision. In contrast, MoAs and AOPs are established through a hypothesis-driven process to derive a sequence of causally linked KEs leading to a health effect of concern. Because of these differences, the workshop participants considered that individual KCs are not equivalent to individual KEs in a MoA or AOP, and thus, direct comparisons among them can be misleading. Nonetheless, the knowledge captured in KCs can inform the development of AOPs·! and new approach methodologies (NAMs) for toxicological testing and assessment. Likewise, the knowledge captured in AOPs can be useful in KC-based carcinogenicity assessments.

АГ Tools and Systematic Literature Review

The amount of scientific mechanistic literature available for evaluating agents has increased substantially over the years. The KCs framework has proven useful in evaluating agents with very large datasets, such as the literature for cobalt compounds,·? ВРА,ЗЗ or more recently PFOA and РЕО$.!3 Nevertheless, the increasing number of mechanistic studies is recognized as one of the main challenges of the JARC Monographs process. The workshop participants recommended exploring the use of systematic evidence mapping and Al at the initial levels of the assessment process. Its incorporation could improve the evaluation, maximize the impact of the IARC Monographs in cancer hazard identification, and naturally speed up the review process. For example, a systematic evidence map (SEM) can be helpful to describe the evidence landscape and to eventually inform a subsequent detailed assessment. The SEM findings typically include a visual presentation, and it is increasingly common to include a website with filter functions to facilitate the use of SEM results. Unlike systematic review or detailed assessment, the question for a SEM to answer can be broad. Some SEMs aim to identify available evidence in broad groups of mechanisms (e.g., which of the KCs have data), and some go a step further to extract data at various levels to provide a more granular map (e.g., for each KC, how many studies are from investigations in exposed humans, human cells in vitro, or experimental systems in vivo; within a specific KC, what end points have been studied).

Al, including machine learning and text mining, has also been used in the cancer mechanistic literature search···° and as a general literature screening tool for years; however, its use in data extraction for mechanistic studies is relatively new. Its use in prioritizing agents for JARC Monographs evaluation has been reported. · In general, the earlier steps of the evaluation process are more automated than the later steps, and to our knowledge, no AI tool has yet been able to judge mechanistic study quality or synthesize evidence. Several tools have been published to support literature review with different levels of automation. Examples include CRAB3·! or Dextr·? and the living systematic review approach.'

There was consensus on the definition of use, refinement, and clarification of the relevance of specific KCs-associated end points as instrumental in characterizing the informativeness of the mechanistic studies evaluated in the JARC Monographs. In addition, new evidence from emerging assays and strategies to integrate the evidence across the KCs can support a more complete mechanistic understanding, especially considering the reduced reliance on animal models and the promotion of NAMs.

The aims and output of this workshop are entirely consistent with the remit of the JARC Monographs Preamble,' which also calls for the KCs and their end points to be updated as science evolves. These and other considerations are expected to contribute to ensuring consistency across assessments of JARC Monographs and to assisting the scientific community interested in cancer hazard identification.

Acknowledgments

The authors are grateful to International Agency for Research on Cancer (IARC) colleagues Jennifer Nicholson, Solene Quennehen, Niree Kraushaar, Mathieu Rose, Noemi Joncour, and Sandrine Ruiz for their logistics support in the preparation and conduct of the workshop. The authors also acknowledge Dr. Dipak Panigrahy's participation in the IARC Scientific Workshop as an invited specialist. Dr. Panigrahy reports receiving substantial personal consultancy fees from several law firms in connection with expert testimony for plaintiffs in cases related to radiation, continuous positive airway pressure (CPAP) machines, and chemicals (hexavalent chromium, arsenic, trichloroethylene, tetrachloroethylene, styrene) and cancer.

The TARC Scientific Workshop reported in this publication was supported by the National Cancer Institute and the National Institute of Environmental Health Sciences of the National Institutes of Health (NIH) under award no. ROICA033193 (to M.K.S.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Where authors are identified as personnel of the IARC/World Health Organization (WHO), the authors alone are responsible for the views expressed in this article. They do not necessarily represent the decisions, policies, or views of the TARC/WHO. The views expressed by J.M.F., J.J., and D.M.D. are those of the authors and do not necessarily reflect the views or policies of the US Environmental Protection Agency. The mention of trade names or commercial products does not constitute endorsement or recommendation for use.

Sidebar

Address correspondence to Federica Madia, International Agency for Research on Cancer, 25, Ave. Tony Garnier, CS 90627 Lyon, France, Email: madiaf @iarc.who.int

Supplemental Material is available online (https://doi.org/10.1289/EHP15389).

The authors declare they have no conflicts of interest related to this work to disclose.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.

Received 17 May 2024; Revised 10 October 2024; Accepted 6 January 2025; Published 3 February 2025.

Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehpsubmissions @niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

References

References

1. Mehta SS, Morin |, Osborn К, Lemeris CR, Conti M, Lunn RM. 2023. An approach to assessing the influence of environmental and occupational cancer hazard identification on policy decision-making. Environ Health Perspect 131(12):125001, PMID: 38088579, https://doi.org/10.1289/EHP12681.

2. Schubauer-Berigan MK. 2023. Invited perspective: good measure-assessing the impact of cancer hazard identification on policies for cancer prevention. Environ Health Perspect 131(12):121302, PMID: 38088578, https://doi.org/ 10.1289/EHP 14099.

3. IARC (International Agency for Research on Cancer). 2019. Preamble. https:// monographs.iarc.fr/wp-content/uploads/2019/07/Preamble-2019.pdf [accessed 21 January 2025].

4. Samet JM, Chiu WA, Cogliano V, Jinot J, Kriebel D, Lunn RM, et al. 2020. The IARC Monographs: updated procedures for modern and transparent evidence synthesis in cancer hazard identification. J Natl Cancer Inst 112(1):30-37, PMID: 31498409, https://doi.org/10.1093/jnci/djz169.

5. Cogliano VJ, Baan RA, Straif K, Grosse Y, Secretan B, El Ghissassi F. 2008. Use of mechanistic data in IARC evaluations. Environ Mol Mutagen 49(2):100-109, PMID: 18240161, https://doi.org/10.1002/em.20370.

6. Baan RA, Straif K. 2022. The Monographs Programme of the International Agency for Research on Cancer. A brief history of its preamble. ALTEX 39(3):443-450, PMID: 34164695, https://doi.org/10.14573/altex.2004081.

7. Smith MT, Guyton KZ, Gibbons CF, Fritz JM, Portier CJ, Rusyn I, et al. 2016. Key characteristics of carcinogens as a basis for organizing data on mechanisms of carcinogenesis. Environ Health Perspect 124(6):713-721, PMID: 26600562, https://doi.org/10.1289/ehp.1509912.

8. IARC. 2019. Tumour Site Concordance and Mechanisms of Carcinogenesis. Baan RA, Stewart BW, Straif K, eds. IARC Scientific Publications, No. 165. Lyon, France: International Agency for Research on Cancer.

9. Guyton KZ, Rusyn 1, Chiu WA, Corpet DE, van den Berg М, Ross MK, et al. 2018. Application of the key characteristics of carcinogens in cancer hazard identification. Carcinogenesis 39(4):614-622, PMID: 29562322, https://doi.org/10. 1093/carcin/bgy031.

10. Smith MT, Guyton KZ, Kleinstreuer N, Borrel A, Cardenas A, Chiu WA, et al. 2020. The key characteristics of carcinogens: relationship to the hallmarks of cancer, relevant biomarkers, and assays to measure them. Cancer Epidemiol Biomarkers Prev 29(10):1887-1903, PMID: 32152214, https://doi.org/10.1158/ 1055-9965.EP1-19-1346.

11. Rusyn |, Wright FA. 2024. Ten years of using key characteristics of human carcinogens to organize and evaluate mechanistic evidence in IARC Monographs on the identification of carcinogenic hazards to humans: patterns and associations. Toxicol Sci 198(1):141-154, PMID: 38141214, https://doi.org/10.1093/toxsci/ kfad134.

12. IARC. 2021. Acrolein, crotonaldehyde, and arecoline. IARC Monogr Identif Carcinog Hazards Hum 128:1-335, PMID: 36924508.

13. Zahm S, Bonde JP, Chiu WA, Hoppin J, Kanno J, Abdallah M, et al. 2024. Carcinogenicity of perfluorooctanoic acid and perfluorooctanesulfonic acid. Lancet Oncol 25(1):16-17, PMID: 38043561, https://doi.org/10.1016/S1470-2045 (23)00622-8.

14. 1АВС. 2023. Occupational exposure as a firefighter. IARC Monogr Identif Carcinog Hazards Hum 132:1-730, PMID: 37963216.

15. Vermeulen R, Bell DA, Jones DP, Garcia-Closas M, Spira A, Wang TW, et al. 2017. Application of Biomarkers in Cancer Epidemiology. In: Cancer Epidemiology and Prevention, 4th ed. Thun M, Linet MS, Cerhan JR, Haiman CA, Schottenfeld D, eds. New York, NY: Oxford University Press.

16. IARC. 2011. Molecular Epidemiology: Principles and Practices. IARC Scientific Publication No. 163. Rothman N, Hainaut P, Schulte P, Smith M, Boffetta P, Perera F, eds. Lyon, France: International Agency for Research on Cancer.

17. 1АВС. 2017. Some organophosphate insecticides and herbicides. IARC Monogr Eval Carcinog Risks Hum 112:1-452, PMID: 31829533.

18. Hollstein M, Alexandrov LB, Wild CP, Ardin M, Zavadil J. 2017. Base changes in tumour DNA have the power to reveal the causes and evolution of cancer. Oncogene 36(2):158-167, PMID: 27270430, https://doi.org/10.1038/0nc.2016.192.

Degasperi A, Zou X, Amarante TD, Martinez-Martinez A, Koh GCC, Dias JML, et al. 2022. Substitution mutational signatures in whole-genome-sequenced cancers in the UK population. Science 376(6591):science.abl9283, PMID: 35949260, https://doi.org/10.1126/science.abl9283.

Kucab JE, Zou X, Morganella S, Joel M, Nanda AS, Nagy E, et al. 2019. A compendium of mutational signatures of environmental agents. Cell 177(4):821-836.e16, PMID: 30982602, https://doi.org/10.1016/j.cell.2019.03. 001.

IARC. 2021. Some aromatic amines and related compounds. IARC Monogr Identif Carcinog Hazards Hum 127:1-267, PMID: 35044736.

Tice RR, Bassan A, Amberg A, Anger LT, Beal MA, Bellion P, et al. 2021. In silico approaches in carcinogenicity hazard assessment: current status and future needs. Comput Toxicol 20:100191, PMID: 35368437, https://doi.org/10.1016/j. comtox.2021.100191.

Chiu WA, Guyton KZ, Martin MT, Reif DM, Rusyn |. 2018. Use of high-throughput in vitro toxicity screening data in cancer hazard evaluations by IARC Monograph Working Groups. ALTEX 35(1):51-64, PMID: 28738424, https://doi.org/10.14573/ altex.1703231.

Reisfeld B, de Conti A, El Ghissassi F, Benbrahim-Tallaa L, Gwinn W, Grosse Y, et al. 2022. kc-hits: a tool to aid in the evaluation and classification of chemical carcinogens. Bioinformatics 38(10):2961-2962, PMID: 35561175, https://doi.org/ 10.1093/bioinformatics/btac189.

Hill W, Lim EL, Weeden CE, Lee C, Augustine M, Chen K, et al. 2023. Lung adenocarcinoma promotion by air pollutants. Nature 616(7955):159-167, PMID: 37020004, https://doi.org/10.1038/s41586-023-05874-3.

Fasanelli F, Baglietto L, Ponzi E, Guida F, Campanella G, Johansson M, et al. 2015. Hypomethylation of smoking-related genes is associated with future lung cancer in four prospective cohorts. Nat Commun 6:010192, PMID: 26667048, https://doi.org/10.1038/ncomms10192.

Battram T, Richmond RC, Baglietto L, Haycock PC, Perduca V, Bojesen SE, et al. 2019. Appraising the causal relevance of DNA methylation for risk of lung cancer. Int J Epidemiol 48(5):1493-1504, PMID: 31549173, https://doi.org/10. 1093/ije/dyz190.

Sonich-Mullin C, Fielder В, Wiltse J, Baetcke К, Dempsey J, Fenner-Crisp P, et al. 2001. IPCS conceptual framework for evaluating a mode of action for chemical carcinogenesis. Regul Toxicol Pharmacol 34(2):146-152, PMID: 11603957, https://doi.org/10.1006/rtph.2001.1493.

Villeneuve DL, Crump D, Garcia-Reyero № Hecker М, Hutchinson TH, LaLone CA, et al. 2014. Adverse outcome pathway (АОР) development |: strategies and principles. Toxicol Sci 142(2):312-320, PMID: 25466378, https://doi.org/10.1093/ toxsci/kfu199.

Carusi A, Wittwehr C, Whelan M. 2022. Addressing Evidence Needs in Chemicals Policy and Regulation. EUR30941EN. Luxembourg: Publications Office of the European Union.

Rana I, Nguyen PK, Rigutto G, Louie A, Lee J, Smith MT, et al. 2023. Mapping the key characteristics of carcinogens for glyphosate and its formulations: a systematic review. Chemosphere 339:139572, PMID: 37474029, https://doi.org/ 10.1016/j.chemosphere.2023.139572.

IARC. 2023. Cobalt, antimony compounds, and weapons-grade tungsten alloy. IARC Monogr Identif Carcinog Hazards Hum 131:1-594. https://publications. iarc.fr/618 [accessed 21 January 2025].

Ricker K, Cheng V, Hsieh CJ, Tsai FC, Osborne G, Li K, et al. 2024. Application of the key characteristics of carcinogens to bisphenol A. Int J Toxicol 43(3):253-290, PMID: 38204208, https://doi.org/10.1177/10915818231225161.

Korhonen A, Séaghdha DO, Silins I, Sun |, Hógberg J, Stenius U. 2012. Text mining for literature review and knowledge discovery in cancer risk assessment and research. PLoS One 7(4):e33427, PMID: 22511921, https://doi.org/10. 1371/journal.pone.0033427.

Baker $, Ali 1, Silins 1, Pyysalo $, Guo Y, Hógberg J, et al. 2017. Cancer Hallmarks Analytics Tool (CHAT): a text mining approach to organize and evaluate scientific literature on cancer. Bioinformatics 33(24):3973-3981, PMID: 29036271, https://doi.org/10.1093/bioinformatics/btx454.

van Dinter R, Tekinerdogan B, Catal C. 2021. Automation of systematic literature reviews: a systematic literature review. Inf Softw Technol 136:106589, https://doi.org/10.1016/j.infsof.2021.106589.

Cierco Jimenez R, Lee T, Rosillo N, Cordova R, Cree IA, Gonzalez A, et al. 2022. Machine learning computational tools to assist the performance of systematic reviews: a mapping review. BMC Med Res Methodol 22(1):322, PMID: 36522637, https://doi.org/10.1186/s12874-022-01805-4.

Khalil H, Ameen D, Zarnegar A. 2022. Tools to support the automation of systematic reviews: a scoping review. J Clin Epidemiol 144:22-42, PMID: 34896236, https://doi.org/10.1016/j.jclinepi.2021.12.005.

dos Santos AO, da Silva ES, Couto LM, Reis GVL, Belo VS. 2023. The use of artificial intelligence for automating or semi-automating biomedical literature analyses: a scoping review. J Biomed Inform 142:104389, PMID: 37187321, https://doi.0rg/10.1016/j.jbi.2023.104389.

Barupal DK, Schubauer-Berigan MK, Korenjak M, Zavadil J, Guyton KZ. 2021. Prioritizing cancer hazard assessments for /ARC Monographs using an integrated approach of database fusion and text mining. Environ Int 156:106624, PMID: 33984576, https://doi.0rg/10.1016/j.envint.2021.106624.

Ali I, Dreij К, Baker $, Hôgberg J, Korhonen A, Stenius U. 2021. Application of text mining in risk assessment of chemical mixtures: a case study of polycyclic aromatic hydrocarbons (PAHs). Environ Health Perspect 129(6):067008, PMID: 34165340, https://doi.org/10.1289/EHP6702.

Walker VR, Schmitt CP, Wolfe MS, Nowak AJ, Kulesza K, Williams AR, et al. 2022. Evaluation of a semi-automated data extraction tool for public health literaturebased reviews: Dextr. Environ Int 159:107025, PMID: 34920276, https://doi.org/10. 1016/j.envint.2021.107025.

Schmidt L, Finnerty Mutlu AN, Elmore R, Olorisade BK, Thomas J, Higgins JPT. 2021. Data extraction methods for systematic review (semi)automation: update of a living systematic review. F1000Res 10:401, PMID: 34408850, https://doi.org/ 10.12688/f1000research.51117.2.