Benzene toxicity and risk assessment, 1972-1992: implications for future regulation.

Acute and chronic exposure to benzene vapors poses a number of health hazards to humans. To evaluate the probability that a specific degree of exposure will produce an adverse effect, risk assessment methods must be used. This paper reviews much of the published information and evaluates the various risk assessments for benzene that have been conducted over the past 20 years. There is sufficient evidence that chronic exposure to relatively high concentrations of benzene can produce an increased incidence of acute myelogenous leukemia (AML). Some studies have indicated that benzene may cause other leukemias, but due to the inconsistency of results, the evidence is not conclusive. To predict the leukemogenic risk for humans exposed to much lower doses of benzene than those observed in most epidemiology studies, a model must be used. Although several models could yield plausible results, to date most risk assessments have used the linear-quadratic or conditional logistic models. These appear to be the most appropriate ones for providing the cancer risk for airborne concentrations of 1 ppb to 10 ppm, the range most often observed in the community and workplace. Of the seven major epidemiology studies that have been conducted, there is a consensus that the Pliofilm cohort (rubber workers) is the best one for estimating the cancer potency because it is the only one with good exposure and incidence of disease data. The current EPA, OSHA, and ACGIH cancer potency estimates for benzene are based largely on this cohort. A retrospective exposure assessment and an analysis of the incidence of disease in these workers were completed in 1991. All of these issues are discussed and the implications evaluated in this paper. The range of benzene exposures to which Americans are commonly exposed and the current regulatory criteria are also presented.


Introduction
The health hazards of benzene have been the subject of relatively widespread concern at three different times during this century. During the 1920s, acute and chronic toxicity among workers exposed to high levels was reported by numerous researchers and clinicians (1)(2)(3). During the 1970s, data indicated that benzene posed a carcinogenic hazard to workers (4)(5)(6)(7). In the 1980s, attention focused not only on exposure to petroleum and chemical workers but also on the concentrations to which the public was exposed. In particular, benzene exposure via indoor air, tobacco usage, and self-service gas stations were of special interest (8,9). The presence of benzene in gasoline and its production during combustion in automotive engines will almost certainly be well studied in the 1990s. This paper presents a synthesis of recent work involving the many important issues that must be incorporated into a health risk assessment of benzene. The various assessments that have been performed over the past 20 years are evaluated and compared. The paper is organized in the following manner: a) physical and chemical properties of benzene, b) acute and chronic toxicity of benzene, c) relevant epidemiology studies that address the chronic hazard, d) summary of risk assessments that have been performed on benzene, e) current ambient exposure to benzene and the possible risks, and]) guidelines or regulations.

Physical and Chemical Properties of Benzene
Benzene is an aromatic hydrocarbon with the molecular formula C6H6 and a molecular weight of 78.1 (10). Under standard conditions, it is a colorless liquid with an aromatic odor. Benzene is volatile (vapor pressure of 100 mm Hg at 26°C), has a Henry's law constant of 5.5 x 10-3 atm m3/ mole (11), and is highly flammable. The limits of flammability in air are 1.5-8.0% by volume and flashpoint of -11.1°C. Benzene is relatively soluble in water (up to 1.8 g/L at 25°C) and is miscible with a variety of organic solvents. Its density, 0.8737 g/mL at 25°C, is lower than that of water, so that undissolved benzene floats on top of water. The pure liquid freezes at 5.553°C and boils at 80.10C (12).

Sources and Uses
Benzene occurs naturally in petroleum and is formed during the combustion of biomass (13,14). Benzene, derived from petroleum,], is a major industrial chemical. According to Kiefer (15), approximately 1.6 billion gallons ofbenzene were produced in the United States in 1990. The major uses of benzene are as a component of gasoline and as a feedstock for the manufacture of synthetic, organic chemicals.
In the past, benzene was used extensively as a solvent (2). In the late nineteenth century, benzene facilitated the rapid development of the rubber industry because of its ability to dissolve rubber and its ease of evaporation during the manufacture of formed or coated rubber products. It played a similar role in the high-speed printing processes. Many industries have used benzene as a solvent or as a starting material for chemical .syntheses (16). Due to toxicity concerns, benzene is infrequently used as an industrial solvent (15).

Fate and Transport
The majority of benzene lost to the environment is as a vapor (17). Benzene in the environment tends to partition into the air (18). Once in the air, benzene is photodegraded with a half-life of approximately 1 week (19). Because it is a constituent of gasoline and is slightly soluble in water, benzene has often been detected in groundwater as a result of leaking underground storage tanks. The environmental half-life at low concentrations ofbenzene in soil and groundwater under aerobic conditions is approximately 10 to 20 days (20). Under anaerobic conditions, benzene is resistant to degradation and can be present for many years (19).

Occurrence
Benzene is a ubiquitous airborne contaminant in both rural and urban areas. Table 1 presents some concentrations of benzene typically found in the United States. Benzene in ambient urban air is primarily due to automobile use. Other sources of benzene such as industrial releases play a relatively minor role in determining the ambient levels but can influence airborne concentrations in areas near a point source (19). For example, the ambient concentration of benzene in the vicinity of gas stations has been reported to be approximately 0.005 to 0.008 ppm (23). Levels as high as 0.1 ppm (21,22) have been reported.
Concentrations of benzene in the breathing zone of individuals pumping gas at self-service stations range from about 0.1 to 1 ppm (21,(23)(24)(25)(26). Benzene has sometimes been a groundwater and soil contaminant at hazardous and municipal waste sites (27).  (21,24) Toxicology of Benzene The widespread use ofbenzene as a solvent for nearly 50 years has encouraged researchers to study its acute and chronic effects on animals and humans. Numerous reviews have discussed various aspects of benzene toxicology (4,(28)(29)(30)(31)(32). Various scientific bodies including the International Agency for Research on Cancer (IARC) (13,33), the National Academy of Sciences (NAS) (34), the National Cancer Institute (NCI) (35), and the National Institute for Occupational Safety and Health (NIOSH) (36,37) have reviewed the literature published through 1977.

Acute Toxicity in Animals
A summary of the acute effects of benzene in animals is presented in Table 2. As shown, the oral LD50 values in the rat vary from 3.4 to 5.6 g/kg depending on the age and strain (38). The LD50 values are reasonably consistent among species (39). Benzene in the rabbit eye is a moderate irritant and causes conjunctival irritation and slight, transient corneal injury (40). The LC50 ranges from 10,000 ppm in mice and rats to 53,300 ppm in cats. Adult rats and mice are often more resistant to the effects of benzene than are young animals (41).

Acute Toxicity in Humans
Humans have demonstrated a relatively high tolerance for acute exposure to benzene. That is, persons can be exposed to concentrations of benzene up to 1000 ppm for periods up to 1 hr without apparent serious adverse effects (42). Early in the century, this characteristic encouraged the use of benzene as a commercial solvent. The primary acute responses to benzene involve adverse effects on the central nervous system (CNS), including dizziness, giddiness, exhilaration, nausea, vomiting, headache, drowsiness, staggering, loss of balance, narcosis, coma, and death (30,40). Individual case reports of acute benzene intoxication have appeared in the literature since the early 1900s (1)(2)(3)(43)(44)(45)(46). Table 3 presents a summary of published case reports of toxicity associated with various airborne concentrations of benzene (47). These reports indicate that concentrations of benzene up to 1000 ppm are tolerable for short periods and that concentrations between 500 and 1000 ppm are not self-limiting if workers are acclimated to the odor. Moreover, for those workers particularly resistant to its hematopoietic effects, airborne concentrations of benzene in the range of 200-250 ppm (8hr time-weighted average) may not always cause obvious acute or subchronic toxicity. These characteristics help explain why so many workers developed significant blood dyscrasias and other adverse effects before 1960. Table 2. Results of acute toxicity studies in animals exposed to benzene (232 (43,44,62,238) Chronic Toxicity in Animals An increased incidence of neoplasms has been found in rats and mice exposed to high concentrations of benzene by inhalation or ingestion (48)(49)(50)(51)(52)(53). The sites with elevated tumor rates include the zymbal gland, oral cavity, preputial gland, lung, and ovaries. Tbmors of the zymbal gland are the most consistent tumor type reported in rodent bioassays (48,50,54,55). In some studies with mice, tumors of the hematopoietic system were found but only in female mice after inhalation exposure (49,56) and only in male mice after oral dosing (55). The reasons for the differences in response between species is not known; however, interspecies differences in metabolism and pharmacokinetics of benzene via different routes of exposure may be partly responsible (30,32,57). Thus far, no animal species tested has clearly been demonstrated to develop the same types of chronic adverse effects as those seen in humans (acute myelogenous leukemia) (52).

Chronic Toxicity in Humans
Chronic occupational exposure to benzene has been known to cause hematological effects since the nineteenth century. Santesson (58) described cases of benzene toxicity in workers fabricating bicycle tires in the late 1800s. Many cases of benzene toxicity were reported near the turn of the century, and Alice Hamilton warned in her landmark paper, "The Growing Menace of Benzene (benzol) Poisoning in American Industry," (2) that a serious problem was at hand. She suggested that many workers could die due to benzene-induced anemia if exposures were not better controlled.
Excessive exposure to benzene results in suppression of the production of both red and white blood cells, i.e., pancytopenia (59). More severe cases, which are generally associated with a marked decrease in the number of cells in the bone marrow, are classified as aplastic anemia. One important point is that these are not distinct diseases but rather a continuum of changes reflecting the severity of bone marrow damage. Among the studies that have demonstrated this wide range of hematological responses due to benzene toxicity are those of Goldwater and his colleagues (60-62), as well as many others (36,(63)(64)(65)(66)(67)(68)(69)(70)(71)(72)(73)(74)(75)(76).
Early work by Greenburg (1) showed that many persons chronically exposed to benzene can tolerate concentrations of about 100 to 250 ppm for 8 hr/day for many years. Table 4 presents some of the results of his studies. The concentrations in the work areas he studied in which benzene was the only solvent present ranged from 0 to more than 4000 ppm. It is not known whether the values represent timeweighted averages, although he stated that the concentrations given are "representative of general room air and/or air at a worker's station" (1). Hematotoxicity, defined as a 25% decrease in leukocyte count (1), was the most consistent adverse effect observed and was usually related to the concentrations of benzene vapor. In the 1930s-1950s, because air sampling methods were rather insensitive and dermal exposure was often significant, red and white blood counts were monitored as a means of managing the employee risk due benzene to exposure (47,77,78). In retrospect, the reliance on biological monitoring rather than air sampling probably saved many lives because it inherently accounted for differences in individual susceptibility and dermal uptake. Despite a fairly broad awareness ofbenzene toxicity, Elkins (79) found that exposure to concentrations ranging from 200 to 700 ppm over several years sometimes occurred, and these resulted in some fatalities (cause of death not stated) in the leather processing industry during the 1940s.
Leukemia. The thought that exposure to benzene might produce certain leukemias and especially acute myelogenous leukemia (AML) was put forward in the 1950s, but clear evidence was lacking until the 1970s. AML is a cancer in which there is an abnormal proliferation of the myeloid stem cells, which are believed to be the common progenitor for mature circulating blood cells (i.e., erythrocytes, thrombocytes, and leukocytes) (54). This disease is mostly observed in adults and has an increasing incidence with age, peaking in the sixth or seventh decade (80). There have been more than 10 studies of 5 cohorts that have shown an association between AML or its variants and benzene exposure (70,(81)(82)(83)(84)(85). Particularly meaningful is the finding by several researchers of the progression of aplastic anemia in a benzene-exposed individual through a preleukemic phase into frank acute leukemia (46,70,86,87).
Chronic studies of benzene in animals have confirmed its hematotoxicity (28,88,89). Although numerous cases of leukemia have been attributed to benzene exposure, there has been a reluctance by some researchers to identify benzene as the causative factor because no animal model for benzene-induced leukemia has been well established. In addition, it is plausible that benzene produces this effect only when other predisposing factors are present. Aflatoxin is an example of a chemical that appears to be carcinogenic in only a subset of the general population; that is, cancer is seen almost exclusively in those persons who have been chemically exposed to aflatoxin and have had hepatitis (90).
Other Effects. Benzene has been shown to be embryo/ fetotoxic in animals, as evidenced by increased incidences of resorptions, reduced fetal weights, skeletal variations, and altered fetal hematopoiesis (91,92). Because these effects only occurred after relatively high doses, benzene has not been generally considered a significant developmental hazard to humans. There is insufficient evidence to indicate that benzene is teratogenic or overtly embryotoxic in animals or humans at vapor concentrations of 10 ppm for 8 hr/day (93).
There are no data to suggest that benzene produces a developmental hazard to humans at most workplace or environmental concentrations. Although some epidemiological studies have implicated benzene as a developmental toxicant in humans, their limitations are too great to be conclusive (19). The confounding variables include exposure to multiple substances, lack of incidences in control populations for the end points ofinterest, problems in identifying the exposed populations, and lack of data on the degree of exposure. aBased on work conducted in 1925 by Greenburg (1). Suppression was defined as a 25% decrease from normal count. bAll measurements of airborne benzene were collected within the workers breathing zone with charcoal tubes during summer months and were representative of both general room concentrations and work station concentrations.
'Workers were exposed to mixed solvents including benzol. "1Measurements were taken in the winter months.

Benzene Metabolism and Pharmacokinetics
The pharmacokinetics of benzene have been well studied in animals. The elimination rate via exhalation has been studied in dogs (94), rabbits (95), mice (96), and rats (97). Schrenk et al. (94) exposed dogs to 800 ppm benzene by inhalation. It was determined that the degree of elimination and the concentration in exhaled air was related to the duration of exposure because of the tendency of benzene to accumulate in body fat. Parke and Williams (95) administered [14C]-benzene orally to rabbits and mice and recovered approximately 43% of the administered dose as unmetabolized benzene in trapped, exhaled air. Andrews et al. (96) administered benzene to mice subcutaneously and recovered 72% of the dose in the air. Simultaneous treatment with both benzene and toluene (96,98) or benzene and piperonal butoxide (99) increased the excretion of unchanged benzene in the breath. Co-administration of these chemicals is believed to inhibit benzene metabolism, leaving more unchanged benzene available for excretion through the lungs.
Benzene is eliminated via excretion of its metabolites in the urine of experimental animals and only a small amount is excreted in feces. Exhalation is the major route of excretion of unmetabolized benzene (13). A biphasic pattern of elimination occurs in rats exposed to 500 ppm for 6 hr, with half-times for expiration of 0.7 and 13.1 hr (97). The initial half-life (t,12) of 0.7 hr was similar for blood, bone marrow, and other organs; the half-life in fat was 1.6 hr (97). Such compartmental analyses are significantly less useful now that physiologically based pharmacokinetic models for benzene have been developed.
The pharmacokinetics ofbenzene have also been studied in humans (98,(100)(101)(102)(103)(104). Benzene absorbed by any route is eliminated by exhalation of unmetabolized benzene from the lungs and by metabolism of benzene in the liver and, to a lesser extent, by the bone marrow. Metabolites of benzene are excreted in the urine. The fraction of benzene metabolized depends on the route of exposure and the size of the dose. The fraction ofbenzene excreted in the expired air ranges between 12 and 50% (100,101,105). The respiratory elimination in humans was reported to be triphasic (101). The initial compartment has a half-life of about 1 hr, similar to the value determined in rats (97). The second, slower phase has a half-life of 3 hr, and the third has a halflife of > 15 hr. No differences in respiratory elimination were observed between men and women ( Table 5). Gordon et al. (106) has reported a biphasic elimination rate where, as expected, the second phase was proportional to the long-term body burden of benzene.

Metabolism
The metabolism ofbenzene has been extensively studied in animals and humans (107,108). The major pathways of benzene metabolism have probably been identified (Fig. 1). Benzene is metabolized to a number of chemicals that are excreted in conjugated and nonconjugated forms. Metabolism occurs primarily in the liver, but in other organs as well, including the bone marrow. Irons et al. (109) have described the metabolism of benzene in rat bone marrow that was perfused in vitro. In these studies, benzene metabolites accumulated in the marrow much as they do in vivo. The rate of benzene metabolism in marrow is much lower than in the liver and is likely related to the low level of mixed-function oxidase activity in bone marrow. In the liver (89), benzene may be converted via a cytochrome P450-mediated pathway (110) to benzene oxide, which can be transformed by epoxide hydratase to the 1,2dihydrodiol. This step leads to catechol formation (111). Benzene oxide can also rearrange nonenzymatically to phenol, which is metabolized to hydroquinone (112). There is also evidence that the major hydroxylated metabolites of benzene can be formed by direct hydroxylation involving free radical insertion (113). Quantitatively, the major metabolic product of benzene is phenol. There are a large number of metabolites that are formed including glucuronidated and sulfated products. The net effect of the metabolism of benzene is that the nonpolar benzene molecule becomes water soluble and excretable in the urine (59).
The rate of metabolism of benzene in animals and humans is similar (allometrically scaled) and produces many of the same metabolites. However, significant differences in benzene metabolism among some species have been reported. For example, Sabourin et al. (114) and Medinsky et al. (57) have reported significant differences between benzene metabolism in rats and mice. While rats metabolized more benzene (orally administered) on a weight basis than mice, mice tended to form greater amounts of hydroquinone and benzoquinone (107,114). Brodfuehrer et al. (115) has also shown that while metabolism of '4C-labeled benzene by liver slices and microsomal preparation occurred at similar rates in tissues from mouse, rat, and man, covalent binding of 14C-relative metabolites to microsomal protein differed significantly with man > mouse > rat.
The metabolism of benzene is also dose dependent. Medinsky et al. (57) and Sabourin et al. (116) reported that metabolism in humans was likely to be dominated by the hydroquinone pathway for long-term exposures to concentrations below 10 ppm and by phenyl conjugates at concentrations greater than 10 ppm.
The metabolism of benzene is also affected by coexposure to other aromatic compounds. Purcell et al. (117) has demonstrated that co-exposures of benzene and toluene result in a noncompetitive inhibition of both compounds' metabolism in rats. This competition explains the observation that co-exposure to toluene protects against the hematopoietic effects of benzene in animals (118,119), and reduces the amount ofbenzene metabolized in humans (120). This finding suggests that when estimating the human risk of exposure to benzene in petroleum products that contain both chemicals, the approach should attempt

Physiologically Based Pharmacokinetic Models
In the last decade, our understanding of the role of metabolism in the prediction of toxicity has been improved by the development of mathematical models that quantitatively describe the uptake, metabolism, target organ concentration, and elimination of chemicals in animals and humans (121)(122)(123). Such physiologically based pharmacokinetic models (PBPK) models are useful in understanding the effect of route of administration on intemal doses, extrapolation between species, and for providing a basis for predicting the amount of the toxic metabolites formed and their concentration at the target organ.
Travis et al. (124) have successfully described the pharmacokinetics of benzene in rats, mice, and humans using a PBPK model. Their model is based on five tissue groups, including the liver (principal metabolic organ), fat, bone marrow, moderately perfused organs (e.g., brain, heart, kidney, and viscera) and the muscle. The bone marrow was added as a tissue compartment because it is almost certainly the target organ for benzene's toxic effects, and because it is a potential site of metabolism. In their model, benzene was assumed to be eliminated only by exhalation, through Michaelis-Menten metabolism in the liver and, to a lesser extent, in the bone marrow. Metabolic products of benzene were not included in the model. Model results have been successfully fitted to empirical data on inhalation, gavage, and intraperitoneal and subcutaneous injection in mice, rats, and humans. The model by Tlavis et al. has been used to study benzene binding to blood proteins (125), to evaluate benzene elimination data from breath monitoring in humans (103), to determine the influence of soil on the absorption of benzene from the GI tract (126), and to evaluate the metabolism of benzene when coadministered with toluene (117). Medinsky et al. (57) and Bois et al. (127) have separately developed PBPK models for benzene and its major metabolites. Medinsky et al. used published values of physiological parameters and developed kinetic data by fitting the model to data from studies in laboratory animals (27,114). The model accounted for phenol, hydroquinone, benzoquinone, muconaldehyde, and other metabolites. These models have been used to investigate the differences in carcinogenic end points between species and the relationship between route of metabolism and the administered dose.
The Bois et al. (127) model is based on the same compartments and metabolites but differs from the Medinsky et al. (57) model by using a range of values rather than a single value for the physiological parameters and kinetic rate constants. The values for those model parameters are determined by using a Monte-Carlo technique to fit data from animal studies (127,128). Spear et al. (129) used these models to fit experimental data from three benzene animal studies using inhalation (97,116) and gavage (116) to a fivecompartment model. They concluded that the model did not adequately represent experimental outcomes of the three studies and that, in general, parameterization of PBPK models to obtain fits to experimental data is complex and problematic (129). The Bois et al. model has also been used to interpret the results of the benzene and phenol National Cancer Institute (NCI) bioassays (130) and the relative significance of short-term exposures to benzene (131).

Hazard Identification Disease End Points
Occupational exposure to benzene as a solvent and as a component of petroleum products has been reported to be associated with a number of different types of cancer including leukemias, lung cancer, and Hodgkin's disease ( Table 6).
The cancers most clearly associated with workplace exposure to benzene are the leukemias. Leukemias are a heterogeneous group of neoplasms arising from the malignant transformation of hematopoietic cells. Although several agencies have concluded that benzene exposure can increase the incidence of all forms of leukemia (13,(132)(133)(134), acute myelogenous leukemia (AML) and related leukemias (collectively referred to as acute nonlymphocytic leukemias) are the primary leukemias associated with benzene exposure. Lamm (135) and Wong (85) have indicated that AML and its variants are the only cancers consistently associated with exposure to benzene.
While AML may have the strongest association with benzene exposure, other cancers of the hematopoietic system, such as chronic myelogenous leukemia (CML) and multiple myeloma (MM) have also been associated with benzene exposure (37,81,136). Both MM and CML occur when there is a proliferation of cells from the same myeloid stem cells as AML. This suggests that all three types of leukemia could be caused by changes in the initially committed myeloid stem cell. Goldstein (59) has noted that the evidence for the association between the benzene and MM and CML is not yet conclusive. Rinsky et al. (137) suggested that MM's association with benzene exposure may be hidden by the longer latency period for the disease. However, the failure of the most recent update (138) to identify additional cases of MM in the Pliofilm cohort makes this hypothesis less tenable.
Development of quantitative estimates of latency is a necessary aspect of low-dose extrapolation of epidemiology data. Heretofore, most estimates of latency for benzene have been based on radiation-induced leukemias (145,146). The length ofthe latency has varied from assessment to assessment, but the best information on latency for benzene is probably found in the Pliofilm cohort (147,205).

Genotoxicity
Benzene has tested negative for mutagenic potential; however, positive results in Salmonella typhimurium have been reported in the presence ofexogenous metabolic activation (148,149,173). Metabolites of benzene have also been identified as mutagenic (150).
There are conflicting data on the ability of benzene to form adducts with DNA. Snyder et al. (108) reported DNA adducts in bone marrow using a 32P-postlabeling assay after repeated, high oral doses (1 mL/kg), while Reddy et al. (151) was unable to find benzene-induced DNA adducts using similar methodology. Subramangan et al. (152) reported that bone marrow is capable of metabolizing phenol to polymeric products which can bind tightly to DNA. These findings suggest that positive results of the radiolabeled binding studies could be due to incomplete purification of DNA. Finally, two benzene metabolites have been reported to form DNA adducts after in vivo administration (153).
Benzene has been reported positive for various tests for clastogenicity after relatively high (in vivo) doses. Oral administration of 220-880 mg/kg benzene in mice caused micronuclei formation (154,155). In mice, exposure to 3100 ppm benzene for 4 hr caused increases in sister chromatid exchanges (SCE) and chromosomal aberrations (156). Erexson et al. (150) has reported slight but statistically significant increases in SCEs and micronuclei in mice after a single 6-hr exposure to benzene at concentrations as low as 10 ppm. The same investigators reported similar small increases in micronuclei in rats after exposure to 1 ppm. Au et al. (157) reported the induction of chromosomal aberrations in spleen lymphocytes of mice exposed for 6 weeks to 2 > 40 ppb benzene. The data from the two experiments reported in this study do not agree, and the more reliable experiment (as judged by the authors) does not show a dose response over a 25-fold dose range. The animal data indicate that benzene can cause dosedependent clastogenesis in cells of hematopoietic origin in rodents at concentrations as low as 1 ppm. The relationship between this clastogenic response and leukemogenesis in these species is unclear because benzene does not produce leukemia in either rodent species.
The incidence of clastogenic effects in humans exposed to benzene in the workplace has been examined by several investigators. Chromosomal aberrations (CAs) or aneuploid cells have been found in circulating lymphocytes of workers who exhibited signs of hematotoxicity (75,144,158,159). Clear evidence for clastogenicity in humans exposed to low levels of benzene is not available. Picciano (160) reported increased chromosomal aberrations in workers believed to be exposed to 2-10 ppm benzene for several years. However, a group of female workers exposed to up to 40 ppm for up to 20 years did not exhibit any increases in CAs or SCEs even in a subgroup that had decreased blood counts at the time of testing (161). Two other studies of refinery workers and shoemakers (162)(163)(164) found no increases in SCEs in workers with exposures up to 100 ppm.
Although these studies are difficult to interpret due to possible confounding factors such as smoking, inadequate exposure information, and lack of a dose-response relationship, it is reasonable to conclude that benzene exposure at levels sufficient to cause hematotoxicity may produce clastogenic effects in human lymphocytes. The likelihood that these effects can occur at current occupational exposure levels has not been established. The relevance of clastogenic effects to human leukemia is unclear, although such effects are relevant to one of several hypothesized mechanisms for benzene leukemogenesis discussed later.

Mechanisms of Action
Many studies indicate that benzene must be metabolized to produce its major toxic effects (31,89,(165)(166)(167)(168). However, the mechanism for benzene toxicity remains unclear. In addition, the mechanisms that produce chronic hematopoietic effects, clastogenic effects, and leukemia are not necessarily the same (169). Also, uncertainties remain regarding which metabolite(s) is (are) responsible for specific toxic effects, and it appears there may be complex interactions between metabolites contributing to the in vivo and in vitro toxicity (170).
The search for a mechanism for benzene leukemogenesis is greatly hampered by the absence of an animal model for benzene-induced leukemia. Unlike most other human carcinogens, where the evidence for an association between the chemical and cancer has been confirmed in animal studies, no good animal model for benzene leukemogenesis has been identified. Cronkite et al. (56) reported that benzene could induce myelogenous leukemia in the CBA/ca mouse; however, these results require additional confirmatory studies.
In the absence of an animal model, plausible mechanisms for benzene's leukemogenic and other carcinogenic effects have been suggested based on indirect evidence such as the reaction of specific benzene metabolites with DNA and animal models of the hematopoietic effects of benzene metabolites. Perhaps the most widely proposed mechanism is that benzene metabolites initiate cancer by direct reaction with cellular DNA, creating DNA adducts that represent heritable damage to the somatic cell line. The trans, trans-muconaldehyde (171), p-benzoquinone (31), or even the reaction product of p-benzoquinone and glutathione (172) could be the causative agent. As discussed earlier, the evidence for significant formation of DNA adducts after in vivo administration of benzene is equivocal. Cytogenetic effects of benzene resulting in chromosomal damage may also play an important role in the mechanism of benzene toxicity and leukemogeneses (174,175).
A major problem with the proposed DNA adduct mechanism is the failure of phenol to induce cancer in rodents. The NCI bioassay of phenol is negative (176), yet phenol is the metabolite through which many of the suggested binding metabolites, such as hydroquinone and p-benzoquinone, are formed. Bois et al. (130) concluded that the negative phenol bioassay cannot be explained on the basis of a first-pass effect as suggested by Casidy and Houston (177,178).
An alternative hypothesis of benzene leukemogenesis is that the by-products of metabolism of benzene rather than the initial metabolites themselves are responsible for the observed toxicity. This hypothesis is based on observations that benzene is metabolized by the cytochrome P-450TTE1 (179,180) and that further metabolism of benzene metabolites can involve a variety of peroxidases (181). These enzymes can produce a variety of free radicals and activated oxygen species that may contribute to benzene toxicity (182). This hypothesis, as well as the literature on the oxidative metabolism of benzene, has been reviewed by Subrahmanyan et al. (183). It is possible that both direct and indirect mechanisms may be responsible for the myleotoxicity and leukemogenesis of benzene. Cox (170) has discussed these and several other proposed mechanisms in a recent paper (Thble 7).
The mechanism of benzene's hematotoxicity has been investigated in animal models by a number of authors. Benzene metabolism has been shown to be necessary to produce hematotoxicity. Benzene's erythropoietic effects, as measured by inhibition of 59Fe uptake, is apparently the result of a number of metabolites, and the interaction of the metabolites has been shown to be signiflcant (88,108). In vivo measurements of benzene hematotoxicity have been demonstrated to be a function of the toxicity of benzene's metabolites on dividing cells (184) and have been duplicated by the co-administration of the benzene metabolites of hydroquinone and phenol (185).
Bone marrow toxicity has been shown to depend greatly on the timing of the exposure. When phenol and quinol were coadministered to mice by a "continuous" dosing regimen, bone marrow cellularity decreased initially, with gradual recovery beginning at the second week despite lable 7. Possible mechanisms for the leukemogenisis of benzene (170).

Hypothesis 1
Benzene metabolites such as trans, trans-muconic acid or p-benzoquinone initiate cancer by reacting with cellular DNA, creating DNA adducts that represent hertiable, carcinogenic damage to the somatic cell line. The resulting mutated cells fail to respond normally to regulatory signals instructing them to differentiate instead of proliferating. This hypothesis is the most common in the literature.

Hypothesis 2
The compensating proliferation of stem cells created by cytotoxic effects of benzene metabolites on partially differentiated cells increases the likelihood of carcinogenic damage, assuming that stem cells are at greater risk of carcinogenic damage while they are actively proliferating than while they are in their normal, quiescent state. Hypothesis 3 Cytotoxic damage to the stromal microenvironment, including stromal macrophages, impairs its ability to regulate stem cell proliferation and differentiation. Initiated and/or malignant stem cells are allowed to proliferate uncontrollably, whereas normally their division would be suppressed or they might be stimulated to differentiate into harmless lineages. Thus, according to this theory, leukemia is expressed as a result of failure by the stromal microenvironment to produce normal regulatory signals telling stem cells to differentiate instead of proliferate. Hypothesis 4 Cytotoxic damage to the immune system, including lymphocytes and stromal macrophages, allows tumor cells that would normally be detected and killed to survive and proliferate instead, leading to a variety of carcinogenic end points. Hypothesis 5 Benzene metabolites initiate cancer through their effects on stem cell chromosomes. Chromosomal aberrations induced by hydroquinone or p-benzoquinone, for example, may activate oncogenes (by carrying them to an active site as a result of a deletion or translocation) or inactive anti-oncogenes.
continued treatment (155). When the mice were treated by a "discontinuous" regimen, however, bone marrow cellularity decreased profoundly, with no evidence of recovery during the treatment period. This apparent paradox, in which the lower total dose administered in the discontinuous protocol was more toxic than the higher dose ofthe continuous regimen, supports the view that the toxic effects ofbenzene metabolites on the marrow are cell-cycle dependent. Supportive evidence for the importance of timing on benzene toxicity is also provided by the studies of Luke et al. (186,187), who found that a 3-day exposure regimen produced more micronucleated polychromatic erythrocytes than did 5 days of exposure. Benzene has been shown to activate protein kinase C, an enzyme playing a pivotal role in signal transduction, which is involved in cell transformation and tumor promotion (188). A series of studies has shown that benzene affects the function of the cellular and hormonal regulators of blood formation, in particular, the function of the stromal cell (168,189). These effects appear to be due to a selective suppression of interleukin 1 (IL-1) released by macrophages, which, in the case of the benzene-associated suppression ofpre-B lymphocytes, results in a reduction of IL-1-dependent release of L-4 by marrow fibroblasts (190).

Epidemiology
The epidemiology of benzene is best defined by five cohorts. Table 8 presents a summary of their relative strengths and weaknesses. These studies all showed an increase in leukemia in the exposed population, however, most of the studies lack accurate estimates of exposure and were confounded because persons were exposed to several chemicals in addition to benzene.

Shoe Workers Study
Aksoy (6) conducted a case study of 34 shoe workers who were admitted to the hematology department of Istanbul Medical School with a diagnosis of leukemia between 1967 and 1975. The total number of shoe workers in Istanbul was estimated to be 28,500. Aksoy used data on the leukemia mortality in the general population of Western nations as his control population. He later indicated that the leukemia incidence rate of 2.5 to 3/100,000 for the general population of Turkey, which would normally be considered the appropriate control group, was not used in the analysis because vital statistics in Turkey could not by relied upon to draw scientific conclusions. The greatest shortcoming in this study is the very poor knowledge about the degree of exposure to benzene and other chemicals. For example, it must be assumed that workers were exposed to a number of different chemicals, including a mixture-of volatile hydrocarbons, curing agents, and dyes. The author indicated that the concentration of benzene vapor ranged between 15 and 30 ppm outside working hours and between 150 and 210 ppm during working hours, with maximum concentrations reaching 210-650 ppm when adhesives containing benzene were used.

Chemical Workers
Wong (191) conducted a prospective mortality study of a group of 4602 male chemical workers from seven plants who were occupationally exposed to benzene for at least 6 months between 1946 and 1976. They were compared to a group of 3074 male chemical workers from the same plants who had no occupational exposure to benzene. The relative risk of leukemia could not be determined in the exposed group compared to the unexposed group because no deaths from leukemia were observed in the internal control population. It should be noted that none of the leukemia deaths in the exposed exhort were of the acute myelogenous type, the cancer that has been most commonly associated with benzene exposure in other occupational studies.

Dow Studies
Ott et al. (192) studied the mortality of 594 white males who were exposed to benzene in three production areas of the Dow Chemical Company in Michigan. Benzene concentrations in the chlorobenzene, alkylbenzene, and ethyl cellulose areas were based upon industrial hygiene monitoring data. Benzene concentrations ranged from 0 to 937 ppm, although the estimated time-weighted average (TWA) benzene concentrations ranged from 0.1 to 35.5 ppm in the various job categories. A total of three cases classified as leukemia was observed from 1940 to 1973 versus 0.8 expected (based upon incidence data from the Third National Cancer Survey).
Bond et al. (83) conducted a 9-year follow-up ofthe Ott et al. (192) study and evaluated some additional workers. In their analysis, 956 Dow Michigan division employees potentially exposed to benzene were studied. Four leukemia deaths were observed versus 2.1 expected (based upon U.S. white male mortality rates), and this increase was not statistically significant. However, all four leukemias were of the myelogenous type. When the mortality for myelogenous leukemia was compared with that which would be expected based upon National Cancer Institute Surveillance Epidemiology End Results (SEER) data, a statistically significant excess was observed (4 observed versus 0.9 expected; p = 0.011). Recently, Bond (193) noted that a fifth leukemia case occurred in an individual whose death certificate was classified to pneumonia in the Ott et al. study, and this subject also had AML. This study has been considered less-than-optional for understanding benzene carcinogenicity because exposure to many other chemicals occurred. However, it may be useful for defining the upper bound of the possible risk.

Chinese Worker Study
Yin et al. (84) carried out a retrospective cohort study in 1982-83 among 28,460 benzene-exposed workers from 233 factories and 28,257 control workers from 83 factories in 12 large cities in China. All-cause mortality was significantly higher among the exposed (265/100,000 person-years) than for the controls. The standardized mortality ratios (SMR) were evaluated for leukemia (SMR = 5.74), lung cancer (SMR = 2.31), primary hepatocarcinoma (SMR = 1.12), and stomach cancer (SMR = 1.22). Leukemia was evaluated for females only. Leukemia occurred among some workers with as little as 6 to 10 ppm average exposure and 50 ppm-years cumulative lifetime exposure. Among the 30 leukemia cases identified in the exposed cohort, the proportion of subjects with acute lymphocytic leukemia was substantially lower and the proportion with acute nonlymphocytic leukemia was higher than the general population. The major shortcoming of this study is the lack of reliable exposure data, adequacy of the nosological methods, and the virtual certainty that exposure to other chemicals occurred.

Ohio Pliofilm Study
Infante et al. (7) and Rinsky et al. (81,136,137) have examined the leukemia mortality experience of a group of workers who were exposed to benzene in the manufacturing of rubber hydrochloride (Pliofilm) at three facilities in two cities in Ohio between 1934 and 1974. This cohort may be the most intensely and carefully studied group in the history of occupational epidemiology (7,47,77,78,81,(135)(136)(137)(194)(195)(196)(197). This group, originally described by Infante et al. (7), contains approximately 1800 white male workers who had been employed in the manufacture of rubber hydrochloride for at least 1 day between 1940 and 1975. In this cohort, nine cases of leukemia (AML) and four multiple myeloma cases were observed by Rinsky et al. (136). Follow-up studies of the cohort have identified five additional cases of leukemia (138). Plioflim Cohort. Since this cohort is the one most often used to estimate potency, it is important to understand its strengths and weaknesses. While other cohorts have larger numbers of workers or have reported more cases of leukemia, no other cohort contains such good information on the duties and exposure histories ofthe workers, nor the quality of medical surveillance. In addition, exposure to other compounds is much less of a problem in this cohort than in others. Detailed individual job histories are available on all workers, and the manufacturing process remained largely unchanged over the 40 years. Relatively large amounts of air sampling data are available to characterize exposure levels for several time periods.
Estimates of Exposure. While the information on the exposure ofthe rubber workers is generally far superior to other epidemiology studies, there are significant shortcomings. Most of the exposure of the Pliofilm workers apparently occurred during the 1940s, when minimal industrial hygiene data were collected. In addition, benzene exposure for specific individuals is limited to the time they spent in Pliofilm production. The work forces at the two plants differed in their potential for exposure to benzene and other chemicals while performing non-Pliofilm tasks. The two Akron plants were part of a large industrial tire-building complex, and the St. Mary's plant was in a small town in western Ohio.
Over the past 9 years, three different methods have been used to estimate the likely retrospective exposure of these workers. In the first method, Rinsky et al. (136) assumed that for a given job, worker exposure was constant over the 40 years of operation unless specific data indicated otherwise. Thus, for instance, workplace concentrations of benzene in 1945 were assumed to be the same as in 1966. In the second method, Crump and Allen (197) proposed an alternative analysis of the industrial hygiene data as described in a report for the Occupational Safety and Health Administration (OSHA). They developed both an estimate of cumulative exposure and peak exposure for each of the employees. To estimate exposure for jobs during time periods for which no data existed, they used all data for a particular job to calculate a percentage of the threshold limit valve (TLV) for that time and then applied that percentage to earlier periods when no data were available. Thus, if the data in the 1960s indicated that the average exposure to benzene in a particular department was 15 ppm or 60% of the prevailing TLV of 25 ppm, then the benzene concentration in that same department in 1945 was assumed to have the same relationship to the prevailing TLV for 1960 (100 ppm) or 60 ppm. In short, they assumed that the workplace concentration during those time periods (relative to the TLV) would be similar to those periods when it was not measured. Thble 9 shows how the benzene TLV changed during the years of Pliofilm manufacture. Since the Crump and Allen methods were related to the TLV, their estimates of worker exposure for the early years were much greater than those of Rinsky et al. (136).
Paustenbach et al. (47) conducted the third retrospective exposure assessment of these workers. Their estimates attempted to quantitatively account for a) uptake due to short-term, high-level exposure to vapors, b) uptake due to background concentrations in the manufacturing facility, c) absorption through the skin, d) morbidity and mortality data on workers in the Pliofilm process, e) the installation of industrial hygiene engineering controls, f) extraordinarily long work weeks during the 1940s, and g) data indicating that airborne concentrations of benzene were underestimated due to inaccurate monitoring devices and the lack of adequate field calibration of these devices. The Paustenbach et al. (47) analysis suggested that Crump and Allen (197) overestimated the exposure of workers in some job classifications and underestimated Table 9. Recommended occupational exposure limits for benzene (81,204 1939 1944 1949 1954 1959 1964 1969 1974 1936 1941 1946 1951 1956 1961 1966 Year  1939 1944 1949 1954 1959 1964 1969 1974 1936 1941 1946 1951 1956 1961 1966 Year Year FIGURE 2. Estimates of exposure histories for selected Pliofilm work categories (47). them for others, and that Rinsky et al. (81,137) almost certainly underestimated the exposure of nearly all workers. Their analysis concluded that worker exposure at the St. Mary facility was 2to 3-fold lower (on average) than at the two Akron facilities. In addition, short-term, high-level exposure to benzene vapors and dermal exposure was found to make a significant contribution (about 30%) to the total absorbed dose. One of the key findings of Paustenbach et al. was that the three facilities probably had significantly different airborne concentrations of benzene; especially during the 1940s and 1950s. Figure 2 graphically compares the results of the three exposure estimation methods for three different job categories at the two different manufacturing locations.
The Paustenbach et al. (47) estimates indicate that the highest exposures to benzene occurred during the 1940s in the Akron I plant. Before the 1940s, the shorter work week and lower production volume probably kept the doses relatively low. After the 1940s, the improved engineering controls at the St. Mary's and Akron II plants and the increasing concerns about the toxicity of benzene (as reflected in the decreasing TLVs) apparently brought about reductions in the workplace concentrations. During this period, St. Mary workers had greatly reduced levels of exposure due to a reduction in Pliofilm production at that site.
Paustenbach et al. (47) acknowledged that because there were virtually no measurements taken during the 1940s, there is significant uncertainty in their estimates. However, they were confident that their estimates are more likely to be accurate than prior estimates for several reasons. First, they found and used new information not considered by the earlier investigators. Second, based on what is known about the incidence of leukemia at the three plants (based on the cumulative doses), the correlation of disease with the dose predicted by Paustenbach et al. was better than using either the Crump and Allen or the Rinsky et al. methods (Table 10). Finally, Paustenbach et al. take issue with the claim (199) that data on which the revised estimates were based had been fully analyzed during the earlier OSHA hearing on establishing a permissible exposure limit (PEL). Several key findings in Paustenbach et al. were in fact not identified in the OSHA regulatory action. Among these are the St. Mary's shutdown during the 1940s, the exceedingly long work schedules, and an attempt to address dermal uptake.

Petroleum fransportation and Distribution Workers Study
A large epidemiology study of workers exposed to petroleum products (containing benzene) has recently been completed (85). The study reported a statistically nonsignificant excess of AML in land-based distribution workers that was not correlated with total hydrocarbon exposure. A shortcoming of the study is that it did not determine benzene concentrations but only exposure to total petroleum hydrocarbons. An estimate of the concentration of benzene vapors could be made with some degree of confidence; however, this analysis has yet to be conducted.
Rushton et al. (198) and Schnatter et al. (200) also recently reported the results of similar studies of petroleum product distribution workers in the United Kingdom and Canada. Both reports indicated that AML SMRs were slightly elevated; however, the increases were not statistically significant.

Evaluating Benzene's Cancer Dose-Response Curve
It is unlikely that any specific case of environmental leukemia can be directly related to exposure to very low concentrations of benzene (or any chemical), nor is it likely that an animal experiment (alone) can describe the rate of response at low doses. Consequently, low-dose extrapolation models must be used to predict the probability of disease. The method or model used to estimate the leuke-  aWhite male "wetside" employees through 1987.  (51,53,54,55) and Rinsky (191) et al. (81); Ott et al. (192); Wong et al. (191) Disease end point All leukemias All leukemias Acute myelogenous All leukemias leukemia and aplastic anemia Low-dose extrapolation Linear Linear (95th confidence Linear-quadratic Conditional log logistic model limit) Risk-specific dose (10-6) 37 ppta 25 ppta 1-10 ppba 100 pptb aASsumes exposures of 24 hr/day for a 70-year lifetime. bAssumes that persons are exposed for 8-hr/day, 5 days/week, for 30 years. mia risks due to exposure to low doses of benzene has a significant impact on regulations. Changes in the incidence of disease or in the estimates of exposure tend to have much smaller effect on the risk estimates than the lowdose model selections.
The impact of model selection is well illustrated by comparing the currently published cancer potency estimates for benzene. The different potencies predict significantly different acceptable concentrations for the same risks. These differences are in part due to the choice of epidemiology data but are largely a function of the model selected. Table 11 presents a summary of the assumptions used in the various models and the concentration of benzene, which corresponds to a predicted excess lifetime cancer risk of 1 in a million.
Current Federal policies require that human data be used instead of animal data if the human data are sufficiently robust. The epidemiology study most often used to estimate the human risk of exposure to benzene -is the rubber workers cohort (7,81,136,137). Many different extrapolation models have been applied to these data to identify safe levels of exposure. As shown in Table 11, linear models were used by the Environmental Protection Agency (EPA) and OSHA in the early 1980s. A modified form of the linearized multistage was used by Crump and Allen (197) and by the EPA (145). Thorslund et al. (146) also developed a revised estimate in 1988. In 1987, Rinsky et al. used a conditional log-logistic model to predict the risk. As of 1993, none of these models has been identified as clearly superior to the others.

Early Regulatory Assessments
Benzene was one of the first epidemiologically based risk assessments conducted by the EPA (133). The Agency based its evaluation upon the disease data on the Pliofllm rubber workers and the Dow cohorts. Cumulative dose, defined simply as the product of concentration and exposure duration, was the index of exposure. The EPA used a linear nonthreshold model to estimate risk. The results, based on these studies, indicated that the 95% upperbound estimate of the probability of developing leukemia in excess of background rates following a working lifetime of exposure to 1 ppm ofbenzene was 14.9 per 1000 based on the rubber worker data and 46.4 per 1000 based on the Dow data (133).
White et al. (82,202) conducted the next assessment of benzene. A cumulative dose was again used to predict risk based on the Pliofilm and Dow cohorts. In deriving their exposure measures, it was assumed that occupational exposures were at the TLV concentrations at the time of the exposure. In their analysis, workers with fewer than 5 years of work experience were excluded. A simple one-hit model was used to predict the response. For exposure to 1 ppm, 8 hr/day, for 30 years, they predicted 3-11 excess leukemia deaths per 1000 persons (based on the Pliofilm cohort) and 3-10 excess leukemia deaths per 1000 persons (based on the Dow cohort).
In 1984, Crump and Allen (197) developed an assessment based on the work of Bond, the Pliofilm cohort, and the original Dow cohorts (192). They also evaluated a number of dosimetrics for predicting risks, including cumulative dose, weighted cumulative dose, window dose, and peak exposure dose. The cumulative dose is the same exposure measure used by EPA in 1979 and by White et al. in 1982 (82). The weighted cumulative dose gives no weight to exposures within the most recent 2.5 years and progressively less weight to exposures that occurred over the past 7.5 years. The window dose considered exposure in the 10-year window defined by 2.5 and 12.5 years in the past and the peak exposure dose measured all cumulative exposures that exceeded 100 ppm. The time intervals of 2.5, 7.5, and 12.5 years were selected based on the leukemia rates and the time of their appearance in persons exposed to radiation due to the bombing in Nagasaki and Hiroshima.
Crump and Allen (197) used two mathematical approaches, the relative risk and the absolute risk models, to evaluate the relationship between leukemia risk and exposure. The relative risk model assumes that incremental risk due to exposure to benzene will be proportional to the background mortality in the group. The absolute risk model assumes that the additional risk associated with dose is the same for all age groups. A relative risk model using cumulative dose and based on the combined rubber workers and Dow cohorts predicted 7.5 excess leukemia deaths per 1000 workers exposed to 1 ppm for 40 hr/week for 30 years. An absolute risk model predicted two excess leukemia deaths per 1000 workers for the same dose.
The authors expressed a preference for the cumulative or weighted cumulative exposure approach because of the leukemia experience in Japan. They noted "window exposures allow the risk [of leukemia death associated with benzene exposure] to disappear completely after 15 years, which appears to be at variance with the Japanese data [on leukemia latency]" (197). The utility of the peak exposure dose was examined by the authors in further analysis of the Pliofilm cohort. In a separate analysis of persons with 200-plus ppm-year cumulative exposure, workers exposed to benzene at levels less than 100 ppm were found to have higher relative risks than workers with exposures greater than 100 ppm. In light of this finding, this approach was not considered credible.

EPA Assessment
Over the past 20 years, the EPA has developed quantitative unit cancer risk estimates for nearly 150 known or suspect carcinogens using a single methodology (203). However, the standard procedure was not used when the agency evaluated benzene (145). Instead, EPA adopted the models developed originally by Crump and Allen. EPA derived a cancer risk estimate from data obtained in three different epidemiologic studies on workers exposed to benzene vapors (81,191,192). Using an average derived from the application of several models, EPA predicted a risk of2.6 x 10-2 for exposure to 1 ppm benzene for a 24 hr/ day, 70 year-lifetime [equivalent to 0.029 mg/kg/day for lifetime exposure]. Using this estimated potency value, the air concentration associated with an excess risk of 10-6 was 40 ppt. Since this estimate was developed, two updates of the cohort have been released [Rinsky (136) and Paxton (138)]; however, EPA has not yet had an opportunity to revise its risk assessment to incorporate the results.

Rinsky et al.
Rinsky et al. (136,137) developed an assessment of occupational exposure based on a conditional logistic model and additional follow-up data on the rubber worker cohort. The conditional logistic model differed from the assessment of White (82) and Crump and Allen (147) in several ways. Specifically, the assessment was based only on the rubber worker cohort and used a novel doseresponse model. In their assessment, Rinsky developed detailed estimates of worker exposure that differed from the estimates made by Crump and Allen (197). Based on the results of the analysis, they recommended lowering the occupational standard for benzene to 0.1 ppm. In 1990, the American Conference of Governmental Industrial Hygienists (ACGIH) TLV committee proposed lowering the TLV from 10 to 0.1 ppm (204). They plan to reach a final decision on the most appropriate value in 1993 or 1994. Thorslund et al. A detailed reanalysis of the benzene cancer potency factor was performed in 1988 by Thorslund et al. (146,14 7). This reanalysis was based on the rubber workers cohort and considered data from the 1981 update of the disease incidence. The reanalysis also attempted to estimate workers' exposures on an individual basis and did not lump workers into dose groups. Finally, Thorsund et al. examined rigorously the dose-response relationship in the cohort. The objectives of the evaluation were to test whether the update of the cohort and better use of existing exposure data would result in a significantly different estimate of potency and whether the data from the rubber worker cohort was best fit by a linear or nonlinear doseresponse model.
The effect of the various modifications in the reanalysis are presented in Table 12. Two modifications (taken together) had a relatively major effect on the linear model's estimates: use of a more refined latency distribution to obtain the exposure weighting function, and use of rigorous statistical techniques that relied on individual exposure estimates and time-to-tumor data, to estimate the transition rate. Thorslund et al. (146) concluded that using these more plausible assumptions, the linear model's estimate of potency was overestimated by approximately one order of magnitude. Thorslund et al. (146) also concluded that there was significant evidence that the dose-response curve for benzene and AML was better fit by a nonlinear rather than a linear model. They recommended using a quadratic model. Thorslund et al. (146) also developed a model which had both a linear and a quadratic term. The results of the model were dictated by the quadratic term at high (occupational) exposure levels and by the linear term at low (environmental) exposures. Both of these models Linear-quadratic model 1.00 x 10-resulted in lower estimates of potency than the linear model (Table 12). Thorslund et al. (146) concluded that the quadratic equation is the best one for predicting the leukemia risk observed in the Pliofilm cohort. This conclusion was based on a) the empirical fit of the data to the quadratic equation and b) biological evidence that a two-hit mechanism (i.e., two molecules of benzene or its metabolites must interact with a cell's DNA to initiate cancer) is more likely than a one-hit mechanism. The linear-quadratic model, which is based on a conservative assumption about how background factors may influence the probability that benzene can produce leukemia, was also presented as a reasonable maximum upper bound of benzene risk.
Because the dose-response model used by a regulatory agency has significant policy and regulatory implications, a conference was convened in Georgetown in 1989 to evaluate the reanalysis (147,205). At that workshop, participants considered the linear quadratic model for estimating the leukemogenic risks of benzene to be more appropriate than the one-hit model. The participants also concluded that a constant relative risk model was not applicable for benzene-induced acute leukemias because the Rinsky cohort does not show a constant relative risk for acute leukemias as the cohort ages. Furthermore, they concluded that the use of the absolute risk rather than relative risk model is consistent with the EPA's policy regarding the interpretation of animal data. The Georgetown workshop (205) recommended that the human data be fitted to other two-stage models and that a sensitivity analysis be performed on models such as the two stage, quadratic, and linear-quadratic, as well as the 1985 EPA model. Austin et al. (206) surveyed the evidence for benzene toxicity, examined previous risk estimates, and provided an updated risk assessment using the rubber worker and Dow cohorts. These authors modeled the number of excess leukemia deaths associated with various exposure levels. Their model was based on the assumption that "the proportional excess leukemia mortality observed during the follow-up period will continue until all cohort members have died" (206). Using this risk estimation model and a target cumulative exposure of 30 ppm-years (1 ppm, for 40 hr/week for 30 years), 51-83 excess leukemia deaths per 1000 exposed were predicted based on the Pliofilm cohort and 47 excess leukemia deaths per 1000 exposed were predicted based on the Dow cohort.

Reanalysis of Rubber Worker and Dow Studies
Brett et al. Brett et al. (207) conducted a risk assessment that was to a large degree a sensitivity analysis of Rinsky et al. (136). They agreed that the rubber worker cohort provided the best basis for estimating the risk of leukemia and that no other epidemiology study is as appropriate. They showed that the Rinsky model was fairly sensitive to the estimates of employee exposure. For example, when the exposure estimates developed by Crump and Allen were applied to their model, the risks associated with 1-ppm (working lifetime) exposures decreased significantly. Brett et al.'s reevaluation of the data indicates that past assessments may have overestimated the cancer risk by a factor of 3-24. Based on the data of Rinsky et al. (136) and exposure matrices of Crump and Allen (197), a risk estimate of 7.9 excess leukemia deaths per 1000 workers exposed to 10 ppm for 45 years and for those persons exposed to 1 (207). The study also concluded that because no cases of leukemia are observed in workers who started employment after 1950, the observed increase in leukemia may be the result of a threshold response to very high levels that occurred in the early years of the manufacturing process.

Human Exposure Dermal Uptake
Benzene is readily absorbed through the skin of man (208,209) and animals (210)(211)(212). In general, studies in humans suggest a dermal uptake rate in the forearm of 0.4 mg/cm2-hr (209). Paustenbach et al. (47) suggested that under certain occupational circumstances, dermal absorption can make a signifiecant contribution (about 10-25%) of the worker's total uptake.
In contrast to liquid absorption, the dermal uptake of benzene from contaminated soil has been modeled by McKone et al. (213) and Burmaster et al. (214). They showed that benzene is poorly absorbed from a soil matrix and that even under ideal conditions (extensive contact for long periods of time), less than 10% of benzene in soil will be absorbed by the skin.

Occupational Exposure
Occupational exposure to benzene involves airborne concentrations that are 100to 1000-fold greater than environment levels. Occupational exposure to benzene has decreased greatly over the past 40 years as a result of lowering occupational exposure limits like the TLV and by discontinuing the use of benzene as a solvent. Currently, significant exposure to benzene is limited to the workers in the petroleum, coal (coking operations), and synthetic organic chemical industries. Under the current OSHA PEL, workplace exposures may not exceed 1 ppm. In the petroleum industry, 8-hr TWA exposures are generally kept below 0.3 ppm (215). Dermal exposure to benzene in the petroleum industries is generally controlled under OSHA requirements; however, dermal exposure to gasoline and benzene in small workplaces (less than 10 employees) can be significant due to less stringent industrial hygiene practices.

Community Exposure
Due to the ubiquitous presence of benzene in the environment, the general population is exposed daily to varying levels of benzene. Iypical sources of exposure include cigarette smoking, use of certain consumer products, inhalation of indoor and outdoor air, pumping gasoline (self-service gas stations), and riding in an automobile (8). During the 1980s, EPA performed extensive investigations (the TEAM studies) of the general population's exposure to volatile organic compounds such as benzene (8,9,216,217). The TEAM studies found that the most important source of exposure for the general population is mainstream cigarette smoke inhaled by smokers. The source accounted for 39% of the total uptake of benzene in the U.S. population (9). In addition, environmental tobacco smoke (ETS) contributes an additional 5% of the nation's exposure. The second largest source (20%) of nationwide exposure is attributed to various personal activities, which include exposures related to automobile use. Atmospheric emissions (auto exhaust and industrial emissions) account for only about 20% of total exposure (Fig. 3).
A number of sources sometimes considered important, such as petroleum refining operations, petrochemical manufacturing, oil storage tanks, urban-industrial areas, service stations, certain foods, groundwater contamination, and underground gasoline leaks, appear to be relatively unimportant on a nationwide basis (9).
The TEAM studies finding that smoking is the largest source of human uptake of benzene is consistent with chemical analyses of cigarette smoke. Loforth et al. (219) measured an average of 500 pg of benzene produced from an individual cigarette. Wester et al. (220) and Wallace and Pellizzari (217) also reported that average benzene breath levels of smokers are significantly elevated over nonsmokers.
Elevated levels of benzene in indoor air represent an additional source of exposure. Wallace et al. (8,9,216,217) reported the results of the EPA TEAM studies that benzene levels in the home are generally elevated over outdoor levels. These elevated concentrations are largely due to ETS, use of petroleum products, and other sources (9,221).
Although benzene is relatively soluble in water (1.8 g/L at 25°C), the potential for human exposure through water consumption is limited. Benzene's volatility and biodegradability results in low levels in surface water (17,18). Levels of benzene in public drinking water supplies are nondetectable generally (>0.5 g/L). Benzene can be a contaminant in shallow groundwater due to leaking underground storage tanks, but the number of individuals affected by such contamination is small. Under the current EPA drinking water standards, benzene levels in public water supplies cannot exceed 5 jg/L. Benzene has been reported as a low-level component of a large number of foods, (35,222). NCI (35) estimated that many Americans currently ingest as much as 250 pLg/day from their diet. However, a recent study performed by the American Petroleum Institute contradicts these findings. The study examined foods that the literature indicated to contain significant amounts of benzene. Benzene levels in these foods were found to be either very low or nondetectable. Based on these results, the American Petroleum Institute (API) concluded that benzene is unlikely to occur in food at levels of toxicological significance (201). This finding is consistent with analyses of the TEAM studies data that concluded that 99% of benzene exposure occurs via inhalation (9). Travis (18) also concluded that based on the physicochemical properties of benzene, diet was unlikely to be a significant source of exposure.
Although benzene is now rarely used as a bulk solvent (15), trace levels of benzene can still occur in certain consumer products. Wallace (9) reported that 400 of approximately 5000 materials and products tested by the National Aeronautics and Space Administration (NASA) were found to emit benzene vapors in concentrations ranging from 0.01 ppm to 140 ppm. These products include paints, adhesives, pens, rubber products, carpeting, liquid detergent, furniture wax, and building materials. A survey of products containing volatile organics found that a similar percentage of products contained low but measurable amounts of benzene (223). TEAM data have shown that the major sources of emission of benzene are not necessarily the most significant sources of human exposure. For example, indoor air concentrations of benzene are significantly higher than outdoor concentrations (217). Figure 3 presents a comparison of the air concentrations in the Los Angeles area and the sources of exposure (218). Autos and industry contribute (96%) to the outdoor airborne levels of benzene, but human uptake of benzene is determined primarily (79%) by cigarettes, indoor air, and personal factors. Table 13 presents an analysis by Wallace (9) on the magnitude of risks offered by exposures to various sources of benzene. His estimates are based on the cancer potency models used currently by EPA (145). For comparison purposes, the risks from acceptable levels of benzene in public drinking water supplies exposures also has been included.

Risk Characterization
These estimates of population risks indicate that while occupational exposure may involve relatively high airborne concentrations of benzene, the bulk of benzene update and risk results from nonoccupational exposures. Smoking is the dominant source of benzene exposure and risk. However, Wallace estimated that nonsmoking, nonoccupational benzene exposures result in some 400 cancer cases per year. As Table 11 indicates, estimates of risk from benzene vary considerably among published risk assessments. Using the linear-quadratic model of Thorslund et al. (146) for example, results in an estimate of nonsmoking, nonoccupational risks of less than 10 per year in the United States.

Regulatory Issues
Occupational Exposure Limits Recommendations to limit exposure to benzene in the workplace came soon after the hazard posed by benzene toxicity was recognized. Serious episodes of benzene toxicity in the rotogravure printing industry in New York (224) were in part responsible for the setting of occupational standards for benzene, which were recommended by the newly formed ACGIH in the late 1930s and early 1940s.  120 150 Driving car (9) 40 40 Filling gas tank (9) 10 5 Occupational (9) 10,000 10 Other personal (9) 150 200 Drinking water at the 10 0.01 (17) current MCL (229) The ACGIH establishes TLVs that are voluntary guidelines and do not carry the force of law; however, many industries have historically attempted to comply with them. Table 9 shows how benzene standards changed over time. In 1971 the OSHA standard was reduced to 10 ppm and by 1974, when OSHA and the National Institute of Occupational Safety and Health (NIOSH) suggested additional regulations, the 10 ppm TWA standard was complemented with a 25-ppm ceiling value (36). In 1976, OSHA and NIOSH reached the conclusion that benzene was a leukemogen and that a 10-ppm standard was not sufficient to protect workers (207). In 1977, OSHA first proposed lowering the occupational benzene standard from 10 ppm to 1 ppm on the basis of its qualitative assessment of the leukemia risk. Their permanent standard was not, however, supported by the U.S. Supreme Court on the basis that OSHA had not demonstrated that a significant risk existed at the previous 10-ppm standard (149). In 1987, OSHA set the standard at 1 ppm for an 8-hr day (TWA) and 5 ppm for 15 min (134). NIOSH (132) recommended that the occupational exposure standard for benzene be revised to a 10-hr TWA of 0.1 ppm with a 15-min ceiling value of 1 ppm.
In 1991, the ACGIH proposed that benzene be listed on the Chemical Substances TLV Notice of Intended Changes of 1990-1991 at 0.1 ppm as a TWA with a skin notation and designation as a Al carcinogen (confirmed human) (204). The recommended TLV of 0.1 ppm is less than the concentration associated with genetic damage in animals (150) and is less than the concentrations associated with genetic damage in human beings (161). Because calculations show that the dermal absorption of benzene can contribute substantially to the total absorbed benzene dose (208,209), the skin designation was retained.
This proposed reduction is under considerable debate. Paxton et al. (138) has shown that the original carcinogenic risk assessment (137) considered by ACGIH likely overestimates carcinogenic risks. Currently, Infante and others question this conclusion and believe the 0.1-ppm value is necessary to provide adequate protection.

Air Toxics Regulations
The EPA has regulated industrial emissions of benzene under the Clean Air Act's provision for the National Emission Standard for Hazardous Air Pollutants (NESHAP) in 1981 and again in 1988 and 1989. These regulations have focused on the petroleum, chemical, and steel industries (225-227).
Under the revised Clean Air Act of 1990, emissions of benzene and other toxins from industrial sources will be controlled using technology-based standards. In addition, the maximum concentration of benzene in gasoline will be limited to 1% by volume. In addition to the direct regulation of benzene emissions, regulations that control hydrocarbon emissions also have reduced the amount of benzene emitted by stationary sources.

Drinking Water
In 1985, the EPA Office of Drinking Water proposed setting the maximum contaminant level (MCL) for benzene in drinking water at 0.005 mg/L (228). This standard was promulgated in 1987 (229). Because of benzene's known carcinogenicity, EPA's Office of Drinking Water has set the maximum contaminant level goal (MCLG) for benzene in drinking water at 0 (229). The MCLG cannot be enforced. The Office of Drinking Water used the EPA estimates of benzene's carcinogenic potency to estimate the benzene concentrations in drinking water that correspond to carcinogenic risks of 10-4, 10-5, and 10-6. The concentrations were 68, 6.8, and .068 ,ug/L, respectively (229). In the 1980s, the National Research Council (NRC) (230) derived a short-term seven-day standard of 12.6 mg/ L for benzene in drinking waters using the data of Wolf et al. (40).

Conclusion
Although several hundred papers have been written over the past 50 years about the health hazards posed by benzene, we are not yet able to precisely identify the cancer potency factor for humans. There are several reasons for this: first, insufficient time has passed to allow a complete determination of long-term effects in the key occupational cohorts; second, we are not certain which lowdose extrapolation model is likely to give the most accurate estimate of risk. As has been noted in the assessment of many carcinogens, the estimates generated by currently accepted dose-response models are generally the maximum plausible ones and the actual risk at low doses may be zero. A third problem is the absence of an animal model of benzene leukemogenesis, which prevents the direct investigation of the mechanism for benzene's carcinogenic effects in humans.
Fortunately, more and better information is being developed on the metabolism ofbenzene in humans and animals, and it is hoped that PBPK models will allow us to quantitatively understand the likely human health hazard at low doses. Information on metabolism and indirect information on the likely mechanism of action, coupled with better estimates of exposure for the cohorts being studied, would be able to provide more accurate estimates of the cancer risk at doses to which the general population is being exposed currently. Until then, it is likely that the EPA and other agencies will choose to regulate benzene in ambient air based on the incidence data from the rubber worker's cohort, the most recent exposure estimates for that group, and models that are considered most appropriate for benzene and/or for leukemia.