Application of adverse outcome pathway networks to integrate mechanistic data informing the choice of a point of departure for hydrogen sulfide exposure limits

Abstract

Acute exposure to hydrogen sulfide initiates a series of hallmark biological effects that occur progressively at increasing exposure levels: odor perception, conjunctivitis, olfactory paralysis, “knockdown,” pulmonary edema, and apnea. Although effects of exposure to high concentrations of hydrogen sulfide are clear, effects associated with chronic, low-level exposure in humans is under debate, leading to uncertainty in the critical effect used in regulatory risk assessments addressing low dose exposures. This study integrates experimental animal, observational epidemiology, and occupational exposure evidence by applying a pathway-based approach. A hypothesized adverse outcome pathway (AOP) network was developed from 34 studies, composed of 4 AOPs sharing 1 molecular initiating events (MIE) and culminating in 4 adverse outcomes. A comparative assessment of effect levels and weight of evidence identified an AOP leading to a biologically-plausible, low-dose outcome relative to the other outcomes (nasal lesions, 30 ppm versus olfactory paralysis, >100 ppm; neurological effects, >80 ppm; pulmonary edema, >80 ppm). This AOP (i.e. AOP1) consists of the following key events: cytochrome oxidase inhibition (>10 ppm), neuronal cell loss (>30 ppm), and olfactory nasal lesions (defined as both neuronal cell loss and basal cell hyperplasia; >30 ppm) in rodents. The key event relationships in this pathway were supported by moderate empirical evidence and have high biological plausibility due to known mechanistic understanding and consistency in observations for diverse chemicals.

Keywords:

• Hydrogen sulfide
• adverse outcome pathway (AOP) network
• occupational exposure limit
• olfactory nasal lesions
• olfactory paralysis
• evidence integration
• human relevance
• mode of action
• critical effect

Introduction

Hydrogen sulfide (H₂S) is a naturally occurring constituent in crude oil and natural gas and is emitted from sources such as volcanoes, sewers, and stagnant water. As such, occupational exposures can occur in oil and gas operations, but also in a myriad of other scenarios, including in sewage treatment plants, during agricultural tasks such as manure handling and during certain manufacturing processes (pulp and paper production, tanneries, viscose rayon production, etc.) (ACGIH 2010; ATSDR 2016). In acute, high exposure scenarios, this irritating, colorless gas induces a series of hallmark biological effects that occur progressively at increasing exposure levels, starting with effects like conjunctivitis (∼5–50 ppm) and olfactory paralysis (∼100 ppm) and progressing to “knockdown” (reversible acute central neurotoxicity; ∼500 ppm), pulmonary edema (∼200–750 ppm), and apnea (immediate paralysis of breathing; ∼1000 ppm) (Reiffenstein et al. 1992; Milby and Baselt 1999; Guidotti 2015). Human experience indicates that perception of “rotten eggs” at an approximate threshold of 0.1 ppm serves as an early warning signal (SCOEL 2007) to avoid/limit exposures.

The derivation of a safe dose to protect against these effects can be based on human evidence; however, human studies often have limitations (e.g. limited exposure characterization, acute versus chronic considerations). In the case of H₂S, human respiratory irritation is a key consideration to protect against short-term exposures (i.e. via the STEL), but human evidence assessing repeated, environmentally relevant exposures is lacking. Thus, olfactory nasal lesions observed in rats after subchronic inhalation exposure are often cited as hydrogen sulfide’s critical effect, due to the consistency with decreased olfactory function noted in humans (EPA US 2003; Health Council of the Netherlands 2006; SCOEL 2007; Hartwig 2013) (Figure 1). Nonetheless, there are significant anatomical and functional differences across species, such as breathing patterns and nasal airflow characteristics, that create uncertainty in extrapolation of rodent nasal outcomes to humans (Harkema et al. 2006; Chamanza and Wright 2015), leading some organizations to select oxygen uptake and venous blood lactate concentration changes at 5 ppm as the critical effect (ACGIH 2010; ATSDR 2016) (Figure 1). The divergent choices made across organizations exemplifies that a range of factors contribute to the uncertainty inherent to occupational exposure limit setting (Dourson et al. 1996; Deveau et al. 2015), which can include the physiological relevance of the animal model to of the human condition and functional adversity of the apical outcome.

Figure 1. Global reference values for hydrogen sulfide. A comparison of reference values (occupational (prolonged, short term) or community exposure) are shown in the top panel (A). The critical effect on which each value is based is shown in the bottom panel (B). IRIS: Integrated Risk Information System; ATSDR: Agency for Toxic Substances and Disease Registry; ACGIH: American Conference of Governmental Industrial Hygienists; DECOS: Dutch Expert Committee on Occupational Standards; EU SCOEL: European Union’s Scientific Committee on Occupational Exposure Limits; MAK: Maximale Arbeitsplatz Konzentration (MAK) (or Maximum Workplace Concentration) Commission; NIOSH: National Institute for Occupational Safety and Health; OSHA: Occupational Safety and Health Administration.

This review hypothesizes that consideration of the biological mechanisms leading to apical outcomes can increase the scientific justification supporting the selection of a point of departure, building on a previous case study in which a comparative weight of the evidence (WOE) assessment was conducted to identify an operant mode of action (MOA) for individual chemicals (Becker et al. 2017). Specifically, biological pathway analyses are applied as a means to identify a data-informed and biologically plausible point of departure for assessing human health risks associated with chronic exposures at environmentally relevant concentrations of H₂S. The biological pathway analyses were conducted in a 2-step process, where mechanistic evidence was organized into an AOP network to facilitate identification of outcomes occurring at the lowest effect levels, followed by a comparative weight of evidence assessment in accordance with the World Health Organization/International Program on Chemical Safety’s Mode of Action/Species Concordance framework (Meek et al. 2014) to increase confidence (and reduce uncertainty) that the chosen effect is both human relevant and health-protective.

Materials and methods

Development of an AOP network

Using selected advisory agency reviews (EPA US 2003; SCOEL 2007; ACGIH 2010; ATSDR 2016), the biological effects data from the primary sources cited within these reviews were organized into key events leading to the hallmark adverse outcomes defined in the reviews as relevant for regulatory risk assessment. This is not a systematic review but relies on the extensive reviews conducted by credible organizations as a starting point.

Starting with the regulatory review, effects noted in the primary sources were categorized initially by level of biological organization. Next, effects were than sorted by tissue or organ system (e.g. nasal tissue, pulmonary tissue) and organized into linear pathways leading, following AOP development best practices (Villeneuve et al. 2014b; OECD 2018). Linear pathways were hypothesized based on professional judgment, linking potential key events identified in a tissue or organ system using a general understanding of biology (e.g. epithelial cell loss observed in pulmonary tissues might lead to accumulation of extravascular fluid in the lungs). The effects chosen as apical outcomes were those effects occurring at either tissue/organ or individual level of biological organization.

Throughout this article, the naming conventions for the MIEs, KEs, and AOs described in this AOP network will be consistent with the naming shown in Figure 2. The AOPs are described in brief narrative form. For each AOP, the primary references and the basis for each effect level are documented in Supplementary Tables 1–4 to facilitate independent assessment of the interpretations and conclusions drawn in this analysis.

Figure 2. Naming conventions for the AOP network describing the mechanisms leading to five hallmark biological effects of H2S exposure. memory/motor impair: memory/motor impairment; neurodegen: neurodegeneration.

Assessment of each AOP in the network

The confidence in each AOP was assessed with regards to biological plausibility and empirical evidence following the recommendations in the OECD AOP Developer’s Handbook (OECD 2018). That is, biological plausibility was defined for each key event relationship (KER) by asking is there is a mechanistic (either structural or functional) relationship between the upstream and downstream key events consistent with established biological knowledge, where high biological plausibility is indicated by extensive previous documentation of a broadly accepted mechanistic basis, moderate is indicated by documentation of analogous biological relationships with incomplete scientific understanding, and weak is indicated by statistical association without function or structural understanding. The level of empirical support for a KER was assessed by asking if upstream key events occur at lower doses and earlier time points than downstream key events and/or if the frequency and incidence of an upstream KE is greater than downstream KE at the same dose, where high evidence is indicated by dependent change in both KEs following exposure to a wide range of stressors and extensive evidence with few inconsistencies, moderate indicated by demonstrated dependent change following exposure to a small number of stressors and some inconsistencies in the evidence, and low indicated by few or no studies documenting dependent change and/or lacking evidence. Based on the intended chemical-specific application, essentiality of the KEs was addressed as part of the chemical-specific assessment, rather than as part of the general AOP weight of the evidence assessment.

Application of the AOP network in risk assessment

The AOP network was used to identify the health outcomes occurring at the lowest exposure levels to be considered as candidate health-protective critical effects. For each AOP containing a candidate outcome, a comparative weight of evidence assessment was done, summarizing the H₂S-specific data to assess dose-temporal concordance of the pathway as a whole (versus individual KER assessments) to ultimately identify the most sensitive pathway best supported by empirical evidence for H₂S.

Results

Development

Cytochrome oxidase inhibition leading to nasal tissue outcomes

AOPs leading to nasal tissue outcomes were developed based on data indicating that the respiratory tract is sensitive to H₂S exposure, as noted in a number of regulatory reviews (EPA US 2003; SCOEL 2007; ACGIH 2010; ATSDR 2016). Biomarkers of effects in nasal and lung tissue have been noted after inhalation exposure to H₂S in a variety of studies, ranging from markers of cytotoxicity such as serum chemistry changes (e.g. lactate dehydrogenase, alkaline phosphatase), morphological changes (e.g. increased cellularity of nasal and bronchoalveolar lavage fluids), and histopathological changes (e.g. necrosis and exfoliation of respiratory and olfactory mucosal cells). Based on effects in the respiratory tract, two nasal-specific AOPs are proposed, initiated by inhibition of cytochrome oxidase and leading to distinct adverse outcomes: nasal lesions (AO1: defined by both neuronal cell loss and basal cell hyperplasia, where H₂S-specific evidence is limited to animal studies), olfactory paralysis (AO2: loss of olfactory function, where H₂S-specific evidence is limited to human experience) (Figure 2).

Description of key events: nasal tissue outcomes

Oxidative phosphorylation is the key cellular pathway through which cells obtain energy, in the form of adenosine triphosphate (ATP). This pathway is mediated by a series of enzymes, including cytochrome oxidase, that are embedded in the inner mitochondrial membrane and generate an electrochemical gradient across this membrane (Kühlbrandt 2015; Cogliati et al. 2018). Due to the critical role in maintaining cellular energy, inhibition of any of the enzymes, including cytochrome oxidase (MIE), that are involved in oxidative phosphorylation can result in cellular stress responses such as a disruption to Ca + homeostasis and reactive oxygen species formation (Imaizumi et al. 2015; Bergman and Ben-Shachar 2016). The MIE can be measured in vitro as the rate of oxidation of reduced ferricytochrome c in tissue homogenates obtained from animals treated in vivo (Dorman et al. 2002).

In an environment of depleted energy, loss of viable cells (KE1) can occur as one of two distinct types. The first, apoptosis, is considered an active process, due to its dependence on the availability of ATP. Apoptosis includes the following sequential events: shrinkage, nuclear disassembly, and fragmentation of a cell into discrete bodies with intact plasma membranes. The second type, necrosis, is a more passive death pathway, characterized morphologically as a gain of cell volume, swelling of organelles, plasma membrane rupture, and loss of intracellular contents (Fink et al. 2005; Kroemer et al. 2009). In the nasal tissue, inhaled contaminants can damage distinct cell types, possibly due to increased deposition resulting from a combination of regional airflow and phys-chem properties of the inhaled substance. Alternatively, it is possible that certain cell populations (e.g. respiratory epithelium, olfactory epithelium, olfactory neurons) are uniquely susceptible to certain chemicals (Monticello et al. 1990; Harkema et al. 2006). KE1 can be measured as part of tissue histopathology visual inspection (i.e. reduction in thickness of the epithelial layer) or by immunohistochemical staining of tissue sections using cell-specific markers (Brenneman et al. 2000; Teeguarden 2017).

Of chemicals that target the olfactory epithelium, general categories of tissue damage have been described in animals, including degeneration, regeneration, post-degenerative atrophy, inflammation, respiratory metaplasia, and basal cell hyperplasia. The earliest change is generally degeneration of olfactory mucosa, usually by the loss of either sensory neurons (such as those in experimental studies on hydrogen sulfide; Brenneman et al. 2000) and 3-methylindole (Kim et al. 2010)). Olfactory neurons are unique in their ability to regenerate; however, under continued insult, this capacity can be exceeded, ultimately resulting in adaptive tissue remodeling or hyperplasia often with respiratory epithelia or basal cells (AO1) (Monticello et al. 1990; Hardisty et al. 1999; Teeguarden 2017). AO1 is defined as both neuronal cell loss and basal cell hyperplasia determined by visual histopathological assessment after in vivo rodent exposure (Hardisty et al. 1999).

A distinctly observed nasal tissue effect after exposure to inhaled chemical stressors is olfactory dysfunction (AO2), which can be characterized in humans as hyposmia (reduced sense of smell), anosmia (loss of odor perception), dysosmia (distorted odor perception), or paralysis (temporary loss of odor perception) (Stevenson 2010). Olfactory dysfunction can be reported clinically as change to the sense of smell or detected in rodent studies measured as change in behavior to avoid an odor or decreased in time to find food (Youngentob et al. 1997; Kim et al. 2010).

Cytochrome oxidase inhibition leading to neurological outcomes

The rationale for consideration of neurological effects was based on reports both in the clinical literature and in animal studies after exposure to H₂S. Typical effects reported in humans include recurrent seizures, persistent headache, nausea, vomiting, fatigue, hearing impairment, movement disorders (e.g. spasticity, ataxia), altered psychological states, memory impairment, and vision impairment, among others (ACGIH 2010; Rumbeiha et al. 2016). In experimental animal studies, neurological effects after repeated exposure to H₂S include changes in neuronal activity as measured by electroencephalographic activity (Skrajny et al. 1996), motor activity deficits (Struve et al. 2001), and functional deficits such as those measured by maze performance (Partlo et al. 2001; Struve et al. 2001). Based on these reported observations, we propose an AOP initiated via inhibition of cytochrome oxidase (MIE), which induces neuronal cell loss (KE1) in the brain, progressing to neurodegenerative lesions (KE2) and memory and/or motor deficits (AO3) (Figure 2).

Description of key events: neurological outcomes

In addition to the biological description of the upstream key events in olfactory tissues discussed above, inhibition of cellular energy production (MIE) and neuronal cell loss (KE1) is relevant in neurological tissues as well, due to high energy requirements (Rugarli and Langer 2012; Bal-Price et al. 2018).

As a term, neurodegeneration combines two words: “neuro” in reference to nerve cells and “degeneration” in reference to progressive damage. As a diagnosis, neurodegeneration (KE2) is used to describe conditions that present with a loss of nerve structure and function and/or by neuronal cell loss. Multiple biological pathways can lead to neurodegeneration, including protein misfolding and aggregation, mitochondrial dysfunction, disrupted axonal transport, and neuroinflammation (Bal-Price et al. 2018; Wolfe 2018). Common clinical diseases (Alzheimer’s, Parkinson’s, Hungtindon’s) are defined by both neurodegeneration and progressive memory or motor loss (AO3) (Bal-Price et al. 2018).

Cytochrome oxidase inhibition leading to pulmonary outcomes

The final biological system considered as part of this network the pulmonary system. In human experience, pulmonary edema is relatively common after acute H₂S exposure (Wasch et al. 1989; Tvedt et al. 1991; Schneider et al. 1998; SCOEL 2007). AOP4 is proposed to be initiated at the molecular level by inhibition of cytochrome oxidase, leading to cellular hypoxia and subsequent cell death in the pulmonary epithelium, ultimately leading to pulmonary edema (Figure 2).

Description of key events: pulmonary outcomes

Normal lung physiology relies on circulation of both air and blood, which are relegated to their own conduits. The barrier between capillaries and the bronchoalveolar spaces consists of an epithelial cell layer. This barrier is semi-permeable, setting up a balance between the filtrate removed from pulmonary circulation and the fluid absorbed by the lymphatic system. Under normal conditions, this leaky epithelial membrane allows only a small volume of into the interstitial space. Under conditions of trauma, such as surgical ischemic/reperfusion injury, epithelial cell death (KE3) occurs, increasing capillary permeability and leading to excessive fluid translocation to the interstitial space (Baumann et al. 2007; Murray 2011; Assaad et al. 2018). KE3 can be detected in lung tissue histological and ultrastructural assessment (Lopez et al. 1988; Prior et al. 1990).

Pulmonary edema (AO4) is defined as the accumulation of extravascular water in the lungs. Two main types of pulmonary edema are characterized based on the preceding biological events: cardiogenic, which develops from increased capillary pressure due to heart failure, and noncardiogenic, which is associated with increased capillary permeability (Murray 2011; Assaad et al. 2018). Noncardiogenic pulmonary edema has been reported in association with central neurologic insult, in which increased capillary permeability arises due to localized ischemic insult in brain trigger zones (Baumann et al. 2007; Davison et al. 2012). AO4 can be measured as the ratio of lung weight immediately after removal from the thoracic cavity and after drying, as an estimate of amount of fluid in the lung (Prior et al. 1990).

Assessment

Biological plausibility

The early key events in AOP1-3 are well established across a broad chemical range (Table 1). Disruption of oxidative phosphorylation via cytochrome oxidase inhibition (MIE) initiates mitochondrial dysfunctional and cell death (KE1), as observed for chemicals such as rotenone (Moon et al. 2005), methylmercury (Yee and Choi 1996), and paraquat (Tawara et al. 1996). For neurons in particular, cell survival (KE1) is critically dependent on the integrity and functionality of mitochondria (MIE) as the main source of cellular energy. The specialized functions of neurons require a significant amount of energy, to maintain membrane potential and neurotransmitter loading and release from synaptic vesicles (Kann and Kovács 2007; Rugarli and Langer 2012).

Table 1. Assessment of biological plausibility for individual key event relationships.

AOP	Key event relationship (adjacency of KEs)	BP^a	BP rationale^b
			High: Extensive understanding of the KER based on previous documentation and broad acceptance	Moderate: The KER is plausible based on analogy to accepted biological relationships, but scientific understanding is not completely established	Weak: There is empirical support for a statistical association between KEs but the structural or functional relationship between them is not understood
AOP1	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)	High	Established biological knowledge across a range of chemicals (rotenone, methylmercury, paraquat) (Kann and Kovács 2007; Imaizumi et al. 2015) Established mechanistic basis (disrupted cellular energy production in the form of ATP leads to cell death, specifically in neurons which have a high energy demand) (Rugarli and Langer 2012; Imaizumi et al. 2015; Bal-Price et al. 2018)
	KE1→ AO1 Neuronal cell loss leads to olfactory nasal lesions (Direct)	High	Established biological knowledge across a range of chemicals (unique susceptibility of individual cell types in olfactory tissues leading to regenerative hyperplasia) (Monticello et al. 1990; Hardisty et al. 1999; Harkema et al. 2006) Established mechanistic basis (loss of neurons leading to regenerative hyperplasia) (Teeguarden 2017)
	AO1 → AO2 Olfactory nasal lesions leads to rhinitis (Direct)	Low	No biological knowledge for this association other than H2S (olfactory lesions to rhinitis) No established mechanistic basis beyond the H2S experimental evidence (Dorman et al. 2004)
AOP2	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)	High	See identical KER in AOP1
	KE1 → AO3 Neuronal cell loss leads to olfactory paralysis (Direct)	Moderate	Established biological knowledge across chemicals (e.g. sulfur mustard gas and nickel sulfate) (Dorman et al. 2004; Jia et al. 2010; O'Neill et al. 2011) Proposed mechanistic basis (relationship between olfactory dysfunction and damage to neurons) (Huart et al. 2013; Boesveldt et al. 2017)
AOP3	MIE → AO4 Inhibition of cytochrome oxidase leads to memory/motor impairment (Indirect)	High	Established biological knowledge for a series of clinical outcomes (Alzheimer’s, Parkinson) (Ohta and Ohsawa 2006; Chin-Chan et al. 2015) Established mechanistic basis (disrupted cellular energy production in the form of ATP leads to clinical outcomes) (Burté et al. 2015)
	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)	High	Established biological knowledge across a range of chemicals (rotenone, methylmercury, paraquat) (Kann and Kovács 2007; Imaizumi et al. 2015) Established mechanistic basis (disrupted cellular energy production in the form of ATP leads to cell death, specifically in neurons which have a high energy demand) (Rugarli and Langer 2012; Imaizumi et al. 2015; Bal-Price et al. 2018)
	KE1 → KE2 Neuronal cell loss leads to neurodegeneration (Direct)	High	Established biological knowledge for a series of neurodegenerative clinical outcomes that are characterized by loss of neurons (Alzheimers, Parkinson) (Godoy et al. 2014; Chin-Chan et al. 2015; Bal-Price et al. 2018) Established mechanistic basis (loss of neurons leading to histopathological lesions and/or tissue atrophy in animals) (Ohsawa et al. 2008)
	KE2 → AO4 Neurodegeneration leads to Memory/motor impairment (Direct)	High	Established biological knowledge across a series of clinical outcomes (normal aging; Alzheimer's) (Mattson 2004; Ohsawa et al. 2008) Established mechanistic basis (neurodegeneration correlates with onset of cognitive impairment (human clinical and animal performance tasks like object recognition/Morris water maze) (Mattson 2004; Ohsawa et al. 2008; Bal-Price et al. 2018)
AOP4	MIE → AO5 Inhibition of cytochrome oxidase leads to pulmonary edema (Indirect)	High	Established biological knowledge for varied trauma/clinical outcomes (gunshot wounds; head injuries; epilepsy) (Baumann et al. 2007) Established mechanistic basis (localized ischemic insult leads to increased permeability of pulmonary capillaries) (Davison et al. 2012)
	MIE → KE3 Inhibition of cytochrome oxidase inhibition leads to pulmonary epithelial cell loss (Direct)	Moderate	Proposed biological knowledge for cell loss in general but not specifically for pulmonary system (Kann and Kovács 2007; Imaizumi et al. 2015) Established mechanistic basis (disrupted cellular energy production in the form of ATP leads to cell death) (Kann and Kovács 2007; Imaizumi et al. 2015)
	KE3 → AO5 Pulmonary epithelial cell loss leads to pulmonary edema (Direct)	Moderate	Proposed biological knowledge for at least one other chemical (3-methylindole) (Durham and Castleman 1985) Established mechanistic basis (noncardiogenic pulmonary edema characterized by damage to capillary endothelial cells and alveolar epithelial cells, resulting in fluid leakage into airways) (Murray 2011)

KE: key event; BP: Biological plausibility.

^aDefining question for BP: Is there a mechanistic relationship between the upstream KE and the downstream KE consistent with established biological knowledge?

^bCriteria to assess high/moderate/weak BP defined in (OECD 2018).

Moving to downstream key events in AOP1, nasal cell-specific susceptibility to individual cell stressors is a well-documented phenomena, and those chemicals targeting neurons (KE1) (e.g. methyl bromide, zinc sulfate, chlorine, rotenone) induce lesions (AO1) described by necrosis of the olfactory epithelium and regenerative basal cell hyperplasia (Monticello et al. 1990; Hardisty et al. 1999; Harkema et al. 2006; Sasajima et al. 2015; Teeguarden 2017).

Specific to nasal outcomes, AOP1, which starts with inhibition of cytochrome oxidase and leads to olfactory nasal lesions, is considered to have high biological plausibility, based on known mechanistic understanding and the consistency in observations for diverse chemicals for the 2 KERs (Table 1).

Considering the unique KER in AOP2, chemicals are observed that both target olfactory neurons (KE1) and induce a loss of olfactory function (AO2) in humans. For example, wartime experience has shown that sulfur mustard can induce anosmia in humans; mechanistically, rodent inhalation studies have shown that sulfur mustard analogues induce oxidative damage to the olfactory epithelium resulting in neuron loss (O'Neill et al. 2011). Occupational exposures to nickel have reported associations with hyposmia and anosmia, which has been hypothesized to be induced by nickel-induced apoptosis of olfactory sensory neurons observed after intranasal instillation of nickel sulfate in mice (Jia et al. 2010). Studies assessing the cellular basis leading to loss of smell have suggested that disrupted regeneration of neurons in the olfactory epithelium are associated both with clinical symptoms (Huart et al. 2013; Boesveldt et al. 2017) and with mechanistic animal studies, assessing a variety of chemical agents such as sulfur mustard, nickel, and 3-methylindole (Jia et al. 2010; Kim et al. 2010; O'Neill et al. 2011).

Specific to nasal outcomes, AOP2, which starts with inhibition of cytochrome oxidase and leads to olfactory paralysis, the reviewed evidence supports a conclusion of moderate biological plausibility for AOP2, based on the limited experience of chemicals inducing both neuronal cell loss and degrees of olfactory dysfunction (Table 1). Although some instances have been identified, the relationship is neither extensively documented nor has broad scientific acceptance.

Considering KER leading to neurological outcomes (AOP3), the indirectly adjacent KER between inhibition of oxidative phosphorylation (MIE) and the neurological symptoms of memory/motor loss (AO3) have been implicated in a number of clinical outcomes, including Parkinson’s disease (Brown and Borutaite 2004), Alzheimer’s disease (McManus et al. 2011), and neurodegeneration (Wallace 1999). These outcomes are defined clinically by loss of neurons in specific regions of the brain (e.g. dopaminergic neurons in the substantia nigra) (KE1) or by neurodegeneration (KE2) (Fahn 2003; Wolfe 2018). Of note, the evidence supporting the relationship between mitochondrial dysfunction mediated by inhibition of oxidative phosphorylation to neurological effects (i.e. parkinsonian-like motor deficits, AO3) has already been characterized in an OECD-endorsed AOP with moderate to strong biological plausibility for each assessed KER (Bal-Price et al. 2018).

Specific to neurological outcomes, the reviewed evidence supports a conclusion of high biological plausibility for AOP3, based both on the biological understanding gained through clinical disease defined by loss of neurons leading to memory/motor loss and mechanistic understanding across a diverse chemical space for all KERs (1 indirect and 3 direct, Table 1).

Considering the KEs leading to pulmonary outcomes (AOP4), clinical evidence derived from human literature characterizes pulmonary edema (AO4) as arising from localized ischemic insult (MIE), which results in increased capillary permeability, permitting extravascular fluid to leak into the lungs (Baumann et al. 2007). This general etiology of pulmonary edema is not derived from the H₂S literature, but rather from non-chemical specific clinical medicine. Further, noncardiogenic pulmonary edema (AO4) is often associated clinically with central neurologic insult, presumably due to the subsequent disruption to cellular respiration (MIE).

Considering the adjacent KERs, it has been demonstrated that disruption of cellular respiration via inhibition of enzymes responsible for oxidative phosphorylation (MIE) leads to cell death (KE3) (Imaizumi et al. 2015). However, there is a lack of established biological knowledge that this relationship occurs in pulmonary tissues specifically. Similarly, limited studies were found in which a relationship was observed between pulmonary cytotoxicity (KE3) and pulmonary edema (AO4), despite a reasonable mechanistic explanation that leakage of fluid into the lungs can occur after death of either alveolar cells or capillary endothelial cells.

Specific to pulmonary outcomes, the reviewed evidence supports a conclusion of moderate biological plausibility, due to the limited chemical space in which most of these relationships are observed (Table 1). Although the relationship between MIE and AO4 is considered high due to the demonstrated clinical evidence and mechanistic understanding, the 2 direct KERs in AOP3 are considered to have moderate biological plausibility since these relationships have not been evaluated or observed for many chemicals.

Empirical evidence

In all of the individual pathways, the upstream KERs (inhibition of cytochrome oxidase leading to neuronal or epithelial cell loss) are considered to be supported by strong experimental evidence (Table 2), with multiple chemical stressors inducing a reduction in cell viability after inhibition of cytochrome oxidase in a dose- and temporally consistent manner. As an example, H₂S inhibited cytochrome oxidase approximately 10–20% in rats treated with 10–30 ppm for 4 h versus control rats after acute exposure (Khan et al. 1990; Dorman et al. 2002). Histopathological detection of cell death was only observed at higher concentrations (i.e. necrosis observed in 1/5, 3/5 and 4/5 rats at 80, 200, and 400 ppm, but not at 30 ppm (Brenneman et al. 2002)) or longer exposures (subchronic exposure: necrosis observed at 30 and 70 ppm with an severity score of 1.4 and 2.4 (max: 3), respectively; Brenneman et al. 2000).

Table 2. Assessment of empirical evidence for individual key event relationships.

AOP	Key event relationship (adjacency of KEs)	EE^a	EE rationale^b
			High: Multiple studies showing dependent change in both exposure to a wide range of specific stressors (extensive evidence for temporal, dose–response, and incidence concordance) and no or few critical data gaps or conflicting data	Moderate: Demonstrated dependent change in both events following exposure to a small number of stressors and some evidence inconsistent with expected pattern	Weak: Limited or no studies reporting dependent change in both events following
AOP1	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)	High	Rationale: The inhibition of cytochrome oxidase and neuronal cell loss has been documented to occur in a dose- and temporal-specific way after exposure to a range of chemical stressors considered to be inhibitors of mitochondrial complex IV (potassium cyanide and sodium azide (Ricci et al. 2003; Liang et al. 2006; Selvatici et al. 2009); beta amyloid peptides (Wei et al. 2000; Veereshwarayya et al. 2006)
	KE1 → AO1 Neuronal cell loss leads to olfactory nasal lesions (Direct)	Moderate	Rationale: Olfactory nasal lesions have been described after repeated inhalation exposure to a variety of chemical stressors that induce cell death to various cells in the olfactory epithelium ((Hardisty et al. 1999; Teeguarden 2017)), including H2S (Brenneman et al. 2000), vinyl acetate (BOGDANFFY et al. 1994), and acetaldehyde (Dorman et al. 2008). While a number of chemical stressors are noted, dose-, incidence/severity- and/or time-dependence is not always clear
	AO1 → AO2 Olfactory nasal lesions leads to rhinitis (Direct)	Weak	Apparent dose-related change between olfactory nasal lesion development (> 30 ppm (Brenneman et al. 2000)) and rhinitis (>80 ppm (Dorman et al. 2004)) after subchronic exposure in rats; possibly only a statistical association
AOP2	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)	High	See evidence summary for identical KER in AOP1
	KE1 → AO3 Neuronal cell loss leads to olfactory paralysis (Direct)	Weak	Rationale: An association between olfactory dysfunction (measured as avoidance response or time to find food) and olfactory neuron loss has been shown for a number of chemicals, including 3-methylindole (Kim et al. 2010) and butyric acid (Sasajima et al. 2015); however most studies show similar dose levels for both responses; so limited information available to assess the dependency of response. Direct evidence to assess temporal concordance is not available; responses in animals have been measured after 3–4 h acute exposure while olfactory paralysis happens after minutes (GLASS 1990)
AOP3	MIE → KE1 Inhibition of cytochrome oxidase leads to neuronal cell loss (Direct)		See evidence summary for identical KER in AOP1
	KE1 → KE2 Neuronal cell loss leads to neurodegeneration (Direct)	Weak	Rationale: Neuron cell death at the cellular level can accumulate over time until lesions to specific brain regions are visible via tissue-level detection methods. Evidence is primarily indirect, as the neuronal cell death detected in methods designed to find active process such as apoptosis or necrosis may only be a portion of cumulative cell death (Gorman 2008; Chi et al. 2018)
	KE2 → AO4 Neurodegeneration leads to Memory/motor impairment (Direct)	High	Rationale: Experimental support is summarized by Bal-Price et al. (2018), in which dopaminergic neurodegeneration has been linked to the onset of motor dysfunction based on assessment of chemicals stressors that target mitochondrial complex I. Similarly, a dose-dependent relationship between the degree of neurodegeneration and the severity of memory loss in Alzheimer’s disease
AOP4	MIE → KE3 Inhibition of cytochrome oxidase inhibition leads to pulmonary epithelial cell loss (Direct)	Moderate	Rationale: Dose-dependent relationships between inhibition of cytochrome oxidase and epithelial cell death (not from the pulmonary system) has been documented to occur in a dose- and temporal-specific way after exposure to a range of chemical stressors considered to be inhibitors of mitochondrial complex IV (e.g. potassium cyanide and sodium azide to oligodendrocytes (Ziabreva et al. 2010) and keratinocytes (Gandhirajan et al. 2018)
	KE3 → AO5 Pulmonary epithelial cell loss leads to pulmonary edema (Direct)	Moderate	Rationale: Limited studies have assessed the dependency of epithelial cell death and pulmonary edema after acute chemical exposure, although a general associations between markers of apoptosis and pulmonary edema due to acute chemical exposure, in both experimental animal studies (Barazzone et al. 1998; Romashko et al. 2003), development of 3 D lung models (Huh et al. 2012), and in human patient case studies (Albertine et al. 2002)

KE: key event; EE: empirical evidence.

^aDefining question for EE: does the upstream KE occur at lower doses and earlier time points than the downstream KE; is the incidence or frequency of the upstream KE greater than that for the downstream KE for the same dose of tested stressor?

^bCriteria to assess high/moderate/weak EE defined in (OECD 2018).

The empirical evidence underpinning the downstream KERs in each of the AOPs ranges is less extensive, characterized as weak to moderate based on the targeted literature searches (Table 2).

Limitations and/or conflicting evidence

Inconsistencies were identified through consideration of animal studies aimed at investigating the hypothesis that olfactory neuron loss (KE1) can affect odor detection (AO2). Although the studies above did note chemical-induced olfactory functional deficits in animals for sulfur mustard, nickel and 3-methylindole, destruction of approximately 95% of rat olfactory epithelium by methyl bromide did not negatively impact performance in an odor differential response task when group performance was assessed (Youngentob et al. 1997). Some individual rats did perform significantly better with an intact olfactory epithelium. It is possible that rodents have a higher ability to maintain near-normal odor detection during continual damage than humans, due to differences in due to differences in the relative percentage of nasal epithelium committed to olfaction (humans, 3% versus rodents, 50%) (Youngentob et al. 1997; Harkema et al. 2006; O'Neill et al. 2011).

Application

AOPs are tools to organize biological effects data, to link early key events with adverse outcomes that are relevant for regulatory risk assessment. Application of the AOP in risk assessment decision-making requires a fit between the level of confidence supporting these linkages and the type of decision being made. In this case, the objective is to identify a critical effect that is protective against all likely effects and is relevant to humans. For H₂S, mapping the effects cited in existing regulatory risk assessments identified pathways leading to tissue-specific outcomes, each with differing levels of confidence (concluded based on empirical evidence and biological plausibility). To identify a critical effect with high confidence that can be used to derive occupational exposure levels, individual AOPs were assessed to identify the effect that occurs comparatively at the lowest dose with the highest confidence.

Chemical-specific WOE assessment

Because the problem formulation for this analysis is specific to H₂S as a chemical stressor, a comparative WOE assessment was done specifically using H₂S data. Regarding nasal outcomes, AOP1 integrates empirical evidence from twelve studies. Assessing the key event relationships in AOP1, consistency in temporal- and dose-responsiveness is observed between key events (Figure 3(A)). That is, later key events are observed at doses/time points exceeding those at which the early key events occur (detailed list of individual studies/effect levels can be found in Supplementary Table 1). Specifically, the most upstream key event, MIE, was observed at the lowest dose, 10 ppm (Khan et al. 1998), compared to downstream KEs: KE1, 30 ppm (Brenneman et al. 2000; Dorman et al. 2004); AO1, 30 ppm (Brenneman et al. 2000); and AO2, 80 ppm (Dorman et al. 2004). However, direct evidence to assess the KEs in AOP1 collectively in a single study was not identified. Based on the consistency in this dose-temporal pattern across the AOP, together with the high biological plausibility, AOP1 (olfactory nasal lesions) is considered to have higher confidence for use as a potential point of departure for H₂S exposure limits.

Figure 3. Dose–response and temporal concordance analysis of (A) AOP1 (nasal tissue outcomes: olfactory nasal lesions), (B) AOP2 (nasal tissue outcomes: olfactory paralysis), (C) AOP3 (neurological outcomes), and (D) AOP4 (pulmonary outcomes). +: 0–33% inhibition or mild severity; /++: 34–66% inhibition or moderate severity; /+++: 67–100% inhibition or massive severity; *observed and reported qualitatively.

The second pathway related to nasal outcomes, AOP2, integrates empirical evidence from 13 studies, where 11 studies that provide supporting data for MIE and KE1 and were conducted in experimental animals (detailed list of effect levels per study in Supplementary Table 2). The remaining 2 studies report effects observed in humans, with estimated doses based on case reports (Beauchamp et al. 1984; van Aalst et al. 2000). Limited conclusions can be drawn on the dose-concordance between neuronal cell loss (KE2), measured in experimental animal studies, and olfactory paralysis (AO2), due to the uncertainty in quantitative effects levels at which olfactory paralysis occurs in humans. The minimal doses at which neuronal cell loss is observed appear to be lower than the minimal doses at which olfactory paralysis is observed (Figure 3(B)), however, this assessment is complicated by the timing at which these effects are observed after exposure. Temporally, olfactory paralysis has been reported to occur within minutes (Beauchamp et al. 1984); however, experimental animal data were not identified that evaluated neuronal cell loss (the upstream key event) at this time frame. Experimental animal studies only made histopathological assessments ≥24 h after acute exposure (Lund and Wieland 1966; Brenneman et al. 2002; Anantharam et al. 2017), limiting the ability to directly assess the essentiality of these KEs collectively. Considering the lack of data by which the dose-dependence of the key events in this pathway can be assessed after H₂S exposure, AOP2 (olfactory paralysis) has lower confidence for use as a potential point of departure for H₂S exposure limits.

Regarding neurological outcomes, AOP3 integrates H₂S-specific empirical evidence from 14 studies supporting the key events in this AOP. Assessing the key events in AOP2 together, the empirical evidence is concluded to be moderate. While each individual key event shows dose-responsive pattern itself, there is a gap in available evidence in the mid-dose range and at least one noted inconsistency, where memory/motor deficits appear to occur at lower doses than neurodegeneration (Figure 3(C); more detail provided in Supplementary Table 3). Specifically, no evidence of neurodegeneration was observed at 80 ppm after 90 days of treatment in mice and 2 strains of rats (Sprague Dawley and Fischer 344) (CIIT 1983a, 1983b, 1983c), despite the observation of decreased learning performance (treated animals made more tries to complete radial arm maze task versus control, p = 0.01) at this dose (Partlo et al. 2001) after 5 days of exposure. As noted, there are no studies assessing neuronal cell loss in CNS tissues and neurodegeneration between 80 ppm and 500 ppm, limiting concordance and essentiality conclusions. Given the noted inconsistency in dose-temporal patterns, AOP3 (neurological outcomes) has lower confidence for use as a potential point of departure for H₂S exposure limits.

Regarding pulmonary outcomes, AOP4 integrates H₂S-specific empirical evidence from 14 studies supporting the key events in this AOP. After acute exposures in animals, H₂S produced cytotoxicity in the pulmonary epithelium at high doses as indicated by changes in cellularity of bronchoalveolar gavage fluid and histological changes such as necrosis and exfoliation of mucosal cells lining the airways (Lopez et al. 1987, 1988; Prior et al. 1990; Khan et al. 1991; Wu et al. 2011). Pulmonary edema observations ranged from slight congestion in the lungs after 1 h exposure to 78 ppm H₂S (Kohno et al. 1991) to moderate/massive pulmonary edema after 4 h exposure to >200 ppm H₂S (Prior et al. 1990; Green et al. 1991). Although the severity of pulmonary edema is observed with increasing dose-responsiveness, the key events relationships do not display the expected dose- and temporal-responsiveness; that is, the downstream KE (pulmonary edema) is observed at doses below that which the upstream KE (pulmonary cell loss) occurs: 80–100 ppm (Figure 3(D); Supplementary Table 4). AOP3 is supporting by low empirical evidence, suggesting that pulmonary edema induced by H₂S is not initiated via pulmonary cell loss. Given this inconsistency and lack of direct evidence to assess essentiality of the KEs, AOP4 (pulmonary outcomes) has lower confidence for use as a potential point of departure for H₂S exposure limits.

Comparative assessment of effect levels

Based on the H₂S-specific data, the minimum effect level at which each adverse outcome occurred was compared to identify the effects occurring at the lowest doses for consideration as health-protective critical effects. As noted earlier, the series of key events leading to H₂S’s hallmark effects are thought to start with the inhibition of cytochrome oxidase (MIE), a key component in cellular respiration. The molecular events are proposed to lead to a series of tissue-specific downstream events (e.g. nasal lesions (AO1), olfactory paralysis (AO2), memory/motor impairment (AO3), pulmonary edema (AO4)). At high doses (>400 ppm), mortality occurs. Of the networked tissue- and individual-level effects occurring at doses below those at which mortality is the key concern, the key events in AOP1 are induced by H₂S at the lowest doses: cytochrome oxidase inhibition (MIE, 10 ppm), neuronal cell loss (KE1, 30 ppm), and olfactory nasal lesions (AO1, 30 ppm). Comparatively, the other potential adverse outcomes occur at higher effect levels: olfactory paralysis (>100 ppm), pulmonary edema (≥80 ppm), neurodegenerative lesions (≥500 ppm) (Figure 4). Comparison of effect levels suggests that the key events in AOP1 are likely candidates for consideration as the critical effect to be used in the derivation of H₂S exposure limits.

Figure 4. Comparison of lowest effect levels across the AOP network. Lowest effect levels are reported for key events). Dotted lines: low or moderate biological plausibility for KER. Solid lines: high biological plausibility for KER.

Comparative assessment of confidence

Three different measures of confidence were assessed: biological plausibility of the KERs, empirical evidence of the KERs, and a comparative WOE assessment for the AOPs as a whole. Considering all three measures, the pathway leading to nasal outcomes, AOP1, is comparatively supported by strong WOE, where individual KERs are considered to have high biological plausibility and moderate empirical evidence based on multiple chemical stressor data and the pathway demonstrates an expected pattern of dose-/temporal-concordance after exposure to H₂S (Table 3). Each of the remaining AOPs has gaps or inconsistencies in at least one measure of confidence; for example, AOP2 is considered to have lower confidence based on the chemical-specific assessment of dose-/temporal-concordance, due to the lack of data evaluating the anticipated timing at which neuronal cell death (KE2) can be detected relative to the observations of olfactory paralysis (AO3). AOP3 and AOP4 are considered to have lower confidence, also due to the H₂S-specific pathway WOE assessment, due to inconsistent dose-response across KE in the pathway (i.e. downstream AO are observed at lower doses than upstream KEs).

Table 3. Comparative assessment of confidence measures for the AOPs.

	Biological plausibility of the KERs		Empirical evidence of the KERs		H2S-specific WOE of the AOP
	Decision^a	Rationale	Decision ^a	Rationale	Decision	Rationale
AOP1	High	Extensive understanding based on multiple stressors with accepted mechanistic understanding; related AOP exists in wiki	Moderate	Associations between KE for multiple stressors but dose-dependence not extensively documented	Higher	Dose-temporal concordance for minimal effect levels
AOP2	Moderate	Proposed mechanistic basis between loss of sensory neurons and olfactory dysfunction. Does not have broad acceptance; inconsistencies noted	Weak	Upstream KE observed within minutes while downstream KE only evaluated after hours; direct evidence not available	Lower	Lacking data to assess temporal concordance
AOP3	High	Extensive understanding based on multiple stressors with accepted mechanistic understanding; related AOP exists in wiki	Weak	Neuron cell death accumulates over time and may not be detected during single assessment of tissue neurodegeneration after repeated exposure; direct evidence limited	Lower	Gap in mid-dose range data. Inconsistent dose-temporal concordance (AO observed at lower doses than upstream KE)
AOP4	Moderate	Relevance to pulmonary epithelial cells not clear (beyond H2S)	Moderate	Direct evidence in pulmonary epithelial cells lacking (beyond H2S)	Lower	Inconsistent dose-temporal concordance (AO observed at lower doses than upstream KE)

Decision is based on the KER with lowest confidence in a pathway as defined in Tables 1 and 2.

Rationale for selection of a point of departure

For the purposes of deriving an OEL, all studies assessing potential health effects due to H₂S exposure were organized into a proposed adverse outcome pathway (AOP) network, to incorporate the mechanistic data into the decision-making process and to increase confidence in the selection of the critical effect that will serve as the point of departure. Initially, the lowest effect levels of all possible tissue- and individual-level key events were compared to identify candidate critical effects that would be protective of the range of possible adverse outcomes, finding that the KE in AOP1 leading to olfactory nasal lesions may be appropriate as protective points of departure. Next, weight of evidence considerations were used to distinguish which outcome would provide greater confidence for use in risk assessment decision-making. Based on this qualitative assessment, the AOP leading to nasal lesions was assessed to have high biological plausibility, moderate empirical evidence supporting the KERs, and high dose-temporal concordance across the entire pathway. Comparatively, the weight of evidence is strongest for nasal lesions, according to each measure (biological plausibility, KER empirical evidence, and pathway dose-temporal concordance), lending confidence to the selection of these KE as potential points of departure in the derivation of H₂S exposure limits.

Discussion

One step in the development of occupational exposure limits is the selection of the critical effect, or point of departure, associated with exposure to a chemical. This choice can vary across organizations and may reflect both problem formulation (e.g. is the exposure limit relevant for intermittent use or continuous use over an 8-h work day) and risk policy (e.g. is higher weight placed on animal studies with controlled experimental conditions or human experience) (Deveau et al. 2015). In practice, this can lead different organizations to select distinct outcomes, as is the case with H₂S, partially contributing to a range of global exposure limits for H₂S (0.001–10 ppm) (as depicted and referenced in Figure 1). To move toward harmonization, new evidence integration methods provide a means to incorporate mechanistic information into the selection of a point of departure and increase confidence in this selection via added biological context. To this end, an adverse outcome pathway approach was applied to integrate mechanistic information, demonstrating consistency in both response across experimental conditions and quantitative effect levels across related biological events.

Using this approach, an AOP network proposed based on H₂S-specific effects, organized around one MIE, cytochrome oxidase inhibition, as the often-cited biological event triggering multiple outcomes in distinct tissues (olfactory, nervous, pulmonary) (Beauchamp et al. 1984; Reiffenstein et al. 1992; ACGIH 2010; Guidotti 2015). The KERs in the AOP network were assessed in terms of biological plausibility and empirical evidence based on multiple chemical stressors that activate the same MIE (inhibition of cytochrome oxidase), including potassium cyanide, sodium azide, and beta amyloid peptides. Finally, the likelihood that H₂S acts via each AOP was assessed via consideration of the dose-/temporal-concordance of the key events collectively. Based on the 3 measures of confidence, the AOP leading to olfactory nasal lesions was assessed to have the strongest WOE in comparison to the other 3 AOPS, where the KERs were considered to be have high biological plausibility and supported by moderate empirical evidence and the KEs collectively exhibited a consistent dose-/temporal-pattern (Table 3). Consequently, the KEs in this AOP (cytochrome oxidase inhibition, neuronal cell loss, and olfactory nasal lesions) are viable candidates as potential points of departure for the derivation of an occupational exposure limit for H₂S.

Although the AOP network is based on a single MIE, it’s unlikely that this is the sole interaction responsible for the diverse biological signaling mediated by H₂S, considering that H₂S interacts with a network of enzymes and is recognized as an endogenous neuromodulator (Reiffenstein et al. 1992; Zhang and Bian 2014). Hypothesized pathways initiated by alternate molecular events were considered but not included here due to limited supporting mechanistic data. For example, the release of various neurotransmitters, including glutamate, gamma-aminobutyric acid (GABA) and noradrenaline (Savolainen et al. 1980; Kombian et al. 1988; Hannah et al. 1989; Warenycia et al. 1989; Nicholson et al. 1998) were considered as potential MIEs that might lead to CNS toxicity, an outcome observed after acute, high-exposure conditions (Milby and Baselt 1999; Guidotti 2015). However, the studies identified in our search did not evaluate a common set of neurotransmitters in specific tissue regions, preventing identification of specific key event relationships. It is possible that a targeted systematic review could be conducted to tease out key events with more specificity.

The AOPs developed in this analysis do not clearly reflect the human effects (i.e. oxygen uptake and venous blood lactate concentration) that are cited as the critical effect basis for H₂S exposure limits recommended by some organizations that have set exposure limits (ACGIH 2010; ATSDR 2016). These effects in human volunteers were observed in a series of 5 publications assessing physiological responses to H₂S over short time periods during exercise, conducted following similar but distinct protocols (Bhambhani and Singh 1991; Bhambhani et al. 1994, 1996a, 1996b, 1997). Individually, the studies identified statistically significant responses, but examination of the five studies collectively reveals no pattern of reproducible adverse effects in the range of 0.5–10 ppm. That is, changes in oxygen uptake are observed but not consistently; Bhambhani and Singh (1991) found increased oxygen uptake at 5 ppm under conditions of exhaustive exercise, but no significant changes in two different studies at the same dose under moderate exercise (Bhambhani et al. 1994, 1996b). At 10 ppm, decreased oxygen uptake was identified with moderate exercise (Bhambhani et al. 1997). Its plausible that increased blood lactate reflects a shift in the ratio of cellular energy production via oxidative phosphorylation relative to glycolysis, consistent with the MIE. In this respect, increased blood lactate is more consistent with a biomarker of MIE rather than a key event in and of itself. Beyond this observation, the current analysis did not identify a mechanistic linkage between the physiological responses observed in the human volunteer studies and the adverse outcomes noted in animal studies or to the biological hallmarks indicative of human exposure to H₂S.

Biological response pathway assessments have proven to be a useful tool in chemical risk assessment in wide-ranging applications, including clarification that carcinogenic effects in animals are qualitatively or quantitatively not possible in humans (Holsapple et al. 2006; Proctor et al. 2007), informing chemical grouping for the purposes of cumulative risk assessment (Pistollato et al. 2020), and providing the mechanistic context for integrated assessment and testing approaches (Patlewicz et al. 2014). Although sometimes referred to interchangeably, a key distinguishing factor between adverse outcome pathway (AOP)-based analysis and mode of action (MOA)-based analyses is that AOPs are intended to be “chemically agnostic” whereas MOAs are chemical-specific (Villeneuve et al. 2014a). In other words, AOPs are descriptions of general biological responses that should hold true regardless of chemical stressor, while MOAs focus on the likelihood of a specific chemical to initiate and progress through a defined set of key events, taking into account toxicokinetic/toxicodynamic considerations that impact the target-tissue dose required to “tip” a biological system from one key event to the next. The current analysis applies these frameworks in a step-wise process to utilize the strength of each framework, first relying on the AOP framework to organize all evidence (human, animal, mechanistic) into a series of hypothesized key event relationships describing biological pathways and then using the MOA framework to comparatively assess the weight of the evidence supporting each pathway to assess the fit between confidence in the pathway and application in decision-making. The AOP development phase was done iteratively, where the network was hypothesized initially using only H₂S data but additional, targeted literature searches were conducted to inform biological plausibility and empirical evidence assessments of KERs.

One potential limitation of AOP development is lack of clarity in the evidence identification process, especially when development and assessment are done in multiple rounds of literature searching as described above. Even though AOPs are considered chemically agnostic in nature, the KERs within an AOP are hypothesized and assessed using data specific to one chemical or a group of chemicals. Clear documentation of the literature search strategy as part of the AOP development process has been highlighted as an opportunity to increase the potential utility of AOPs in regulatory decision-making (Kleinstreuer et al. 2016; Leist et al. 2017), and collaborative initiatives are ongoing to align on best practices in the application of systematic literature review tools during AOP development (de Vries et al. 2021, accepted in press). The current study relied on existing systematic reviews on a well-studied chemical, however, in practice, these reviews are older and do not reflect more recently published studies. In future studies, systematic literature review tools such as SWIFT-Review, would offer improved transparency for the follow-up searches on individual KERs.

One final point that warrants discussion is that the conclusions drawn in this study may be considered contradictory to a MOA-based assessment of similar tumors (nasal neuroblastomas and nasal respiratory epithelial adenomas) induced by naphthalene inhalation exposure in rodents. For H₂S, the MOA-based conclusion drawn here is that the key events leading to olfactory nasal lesions in rodents are relevant to humans, while the cytotoxicity-induced nasal tumors occurring after naphthalene exposure are concluded not to be relevant to human risk assessment due to quantitative MOA differences between species (Rhomberg et al. 2010; Piccirillo et al. 2012; Bailey et al. 2016). Two key reasons support that these conclusions are not truly contradictory. First, naphthalene’s cytotoxic and tumorigenic responses are dependent on metabolic activation, primarily oxidation of naphthalene via CYP2F (Bogen et al. 2008). On the other hand, the fast-acting effects associated with H₂S, such as near-immediate olfactory paralysis and knockdown, do not appear to be associated with metabolism. Second, there is limited human epidemiological evidence consistent with nasal or lung tumor risk for naphthalene (Piccirillo et al. 2012), whereas there is ample human evidence indicating H₂S induces nasal tissue effects in humans (i.e. olfactory paralysis). Therefore while these chemicals do have common KEs within their MOAs, they fundamentally function by different MOAs due to toxicokinetic differences.

Conclusions

A biological pathway-based approach is a means to incorporate different lines of evidence into a single framework to increase confidence in the identification of a biologically plausible effect that can be used to derive a health-protective exposure limit. In this analysis, animal and human data from 34 studies identified from existing systematic literature reviews on H₂S were used to propose an Adverse Outcome Pathway (AOP) network, composed of 4 AOPs sharing 1 molecular initiating event (MIE) and 4 potential outcomes. Potential health-protective critical effects (cytochrome oxidase inhibition, neuronal cell loss, olfactory nasal tumors) were identified through consideration of effect levels to distinguish lower- from higher-dose outcomes, followed by a more focused assessment of dose-temporal concordance specific to the H₂S-data.

Acknowledgments

The authors would like to gratefully acknowledge Bruce Copley and Felix Boachie (former epidemiologist and current industrial hygienist, respectively, at ExxonMobil Biomedical Sciences, Inc.) for initial discussions on organizing hydrogen sulfide effects data into the AOP framework; Jessica Ryman-Rasmussen (policy advisor at American Petroleum Institute) for discussions on the topic of hydrogen sulfide exposure limits; James Freeman, Robert Barter (retired and current toxicologists, ExxonMobil Biomedical Sciences, Inc.) and members of ExxonMobil’s Occupational Exposure Limit committee for their encouragement to write this manuscript; Christine Palermo (toxicologist at ExxonMobil Biomedical Sciences, Inc.) for her review and comment on the draft manuscript which helped in particular to clarify the weight of evidence considerations; to Bette Meek (Associate Director, Chemical Risk Assessment at the University of Ottawa) and Brigitte Landesmann (scientist at the European Commission, Joint Research Centre) for review of the preprint version of this manuscript and comments related to the application of the AOP framework in decision-making; and to the external reviewers selected by the Editor and anonymous to the Authors whose comments were valuable in revising and refining the manuscript.

Declaration of interest

The authors are current (KOG) or retired (RJL) employees of ExxonMobil Biomedical Sciences, Inc., a separately incorporated but wholly owned affiliate of Exxon Mobil Corporation. This company processes crude oil and natural gas that contains hydrogen sulfide as a naturally occurring component. The manuscript was written as part of the authors’ normal employment and was the sole responsibility of the authors. No external funding was obtained for manuscript preparation. None of the authors has appeared in any legal or regulatory proceedings related to the contents of this paper.

Supplemental material

Supplemental data for this article can be accessed here.