In the largest case–control study of beryllium-exposed workers to date, we evaluated the relationship between quantitative beryllium exposure estimates in combination with HLA-DPB1 E69 genotype in determining the risk for BeS and CBD. We noted increased exposure associated with CBD as compared with control subjects, which was evident whether considering self-reported exposure assessments or quantitative exposure reconstructions. However, no exposure–response relationship was apparent for BeS, even with inclusion of genetic risk factors. For both CBD and BeS, E69 conferred increased odds as has been shown in other studies (19
). We found that the odds of BeS and CBD appear to be greater among carriers of the non-*02 HLA-DPB1 E69 alleles, and among HLA-DPB1 E69 homozygotes even after adjusting for beryllium exposure. Most importantly, we found evidence supporting individual contributions to CBD risk by increasing exposure and genetic susceptibility via E69 with no significant gene–environment interaction. Last, we provided evidence suggesting an exposure response for CBD and lack thereof for BeS after adjusting for E69 genetic risk factors.
The finding of an exposure–response relationship for CBD has implications for standard setting in the workplace, at a time when OSHA is reconsidering revising the currently out-of-date beryllium exposure standard. In this study, the odds of CBD were associated with higher lifetime weighted average and cumulative exposures, whereas increasing exposure was not a risk factor for BeS. This confirms the previous work by Viet and colleagues (36
) at the RFETS facility showing significant relationships for both cumulative and mean exposures and CBD, but not BeS. Also, confirming previous reports (10
), we identified CBD case subjects at low apparent exposures, with three CBD case subjects reporting no known beryllium exposure and an additional five reporting never having worked in areas or tasks where reconstructed exposures exceeded 0.02 μg/m3
. Thus, this study, although clearly showing a higher prevalence of CBD at higher exposure levels, fails to demonstrate a threshold for the development of CBD. In contrast, for BeS, there was no evidence of either an exposure threshold or an exposure–response relationship as evidenced by the tendency for BeS case subjects to be overrepresented in the lowest lifetime weighted average exposure quartile and with more than one-third of BeS case subjects never having worked in areas or processes where reconstructed exposures exceeded 0.02 μg/m3
. This frequent occurrence of BeS among workers with only minimal known exposure combined with evidence of an exposure–response relationship for CBD has important implications for worker protection both in terms of medical surveillance and removal from exposure. The findings could imply that workers who are sensitized and have not incurred sufficient beryllium exposure to develop CBD may remain sensitized indefinitely. In addition, the findings suggest that even minimally exposed workers should be screened, using the BeLPT, to detect BeS and facilitate early removal from exposure.
This study also confirms the individual contributions of exposure and genetics (E69 status) to the development of CBD. In our current study, carriage of any E69 allele conferred about eightfold increased odds of CBD and each unit increase in lifetime weighted average beryllium exposure increased the CBD odds approximately twofold. This 8-fold increased odds for carriage of any HLA-DPB1 E69 variant is within the confidence limits of the 12-fold increased odds from the initial gene–environment study (20
). In comparing the risks from genetics and exposure, our current findings suggest that, in terms of CBD odds, carriage of any single E69 allele even in extremely low exposures incurs similar odds as exposure to an average beryllium concentration of 4 μg/m3
for those without an E69 allele.
The increased odds for carriage of any E69 allele appears to be differentially distributed when considering E69 genotype, with carriers of only a single copy of an *02 allele only at 3-fold increased odds, those with non-*02 genotypes at nearly 12-fold increased odds, and those with two E69 allele copies at more than 20-fold increased odds. Increased risks of BeS and CBD have been reported previously for carriers of non-*02 alleles and homozygotes (21
). It has been shown previously that carriage of E69 itself is critical to allow binding of beryllium or beryllium-bound peptides to the HLA-DP molecule (42
). The biological basis for the observed increased odds of BeS and CBD among carriers of non-*02 alleles is unclear, although it may be related to the previously identified more electronegative charge on the β chain of the HLA-DP molecule encoded by some of the non-*02 alleles (29
), or it could be related to yet-to-be-determined differences in antigen or T-cell receptor–MHC binding affinity or kinetics. An important limitation of the allele grouping strategy used in this analysis is that the increased odds of BeS and CBD conferred by the non-*02 E69 alleles is likely driven by one or more of the alleles in the group and is not a true group effect. We are also limited in our ability to draw firm conclusions about the differential risks by genotype, due to the wide and overlapping confidence intervals for the genotype odds ratios resulting from the limited number of individuals in each of the strata. Larger studies evaluating the effects of HLA-DPB1 genotypes in combination with exposure and HLA-DRB1 are necessary to more precisely evaluate gene–environment effects.
In terms of policy development, exposure reduction has the potential to provide a greater public health benefit than preemployment genetic testing. As has been presented previously (44
), the low prevalence of BeS and CBD among those exposed and the high carrier frequency of the E69 allele combine to produce an unacceptable positive predictive value for using the E69 marker to determine eligibility for employment in the beryllium industry. Results from this study continue to support this assertion. From this study, considering the greatest genetic risk factors, non-*02 E69 genotype or E69 homozygosity, for the odds of CBD and assuming a generous CBD prevalence rate of 5%, a non-*02 genotype frequency of 15%, and a 4% frequency of E69 homozygotes, the positive predictive value of genetic testing is only 23% for the non-*02 genotype and only 59% for E69 homozygotes. This low positive predictive value implies that for every 100 individuals denied employment because of this genetic trait, the majority of them would not have developed CBD. Exposure reduction, on the other hand, reduces the odds for all exposed, regardless of E69 status, and might reduce the progression from BeS to CBD.
Using a weighted logistic regression, the models from our case–control study can be extrapolated to project the probability of CBD for workers at RFETS given the facility prevalence of CBD of 1.7% identified in a stratified sample by Kreiss and colleagues (10
), and assuming the population characteristics of the participants in this study were representative of all workers at the site (). The probability of CBD predicted by our model at a lifetime weighted average exposure of 0.2 μg/m3
is 0.4% for those with E69-negative genotypes, 1.5% for those with a single *02 allele, 5.4% for those with a single non-*02 E69 allele, and 9.1% for E69 homozygotes. Assuming the genotype frequencies of the workers at the entire site are similar to those of the participants in this study, the composite probability of CBD for all workers at the site would be 1.5% at a lifetime weighted average exposure of 0.2 μg/m3
. From an occupational exposure limit point of view, this suggests strict compliance with an exposure limit of approximately 0.8 μg/m3
(assuming a log-normal distribution and a geometric standard deviation of three), using an upper tolerance limit approach as described by Mulhausen and colleagues (46
) would result in a CBD prevalence of about 1.5% in an exposed population. This estimate is much higher than the 1-in-200 (0.5%) odds of CBD at an occupational exposure limit of 2 μg/m3
suggested by Viet and colleagues (36
Our multiple logistic regression model for BeS suggested increased odds for those working fewer than 5 years at RFETS. Although there have been reports of BeS occurring within a short period of time after first exposure (15
) and others have reported similar protective effects (49
), this study would likely not detect early BeS as most of the case subjects were first screened many years after first exposure to beryllium. Only 20% of BeS case subjects in this study were diagnosed as current workers. These case subjects were diagnosed on average 17 years after starting work at the facility, with all diagnosed more than 6 years after starting work at the facility. It is more likely that these increased odds for short-term workers is an artifact of study design, as our study did not include frequency matching for the number of years worked. It is also possible that this effect was a result of the increased participation by long-term workers in the control group. However, this finding suggests that workers exposed for only a short time are at risk of BeS.
Misclassification of disease status and exposure could have impacted our results. The main source of disease misclassification was the inclusion of 17 unconfirmed BeS case subjects. All of the unconfirmed BeS case subjects met the definition of BeS with repeat abnormal BeLPTs; however, as they had not undergone complete medical evaluations, many of them likely had CBD rather than BeS. Analyses of significant exposure effects, using this mixed group of BeS case subjects, would likely be biased away from the null as CBD was found to be associated with exposure. Thus, the lack of significant association between exposure and BeS in this mixed group is noteworthy.
One of the strengths of this study is its detailed exposure reconstruction, in which the use of individual interviews accounts for the large variation of work composition within a single job classification. This attention to exposure at the individual level was likely one of the reasons that this study was able to identify an exposure–response relationship for CBD, whereas others using grouping strategies at the job classification level have failed. It is unclear whether the small differences in CBD odds identified at exposure levels less than 0.2 μg/m3 were accurate or the result of imprecision in exposure reconstruction at the lower levels. Furthermore, cumulative exposures may have been overestimated for case subjects, as exposures accrued until the date of BeS or CBD diagnosis, which was likely much later than the date of disease development. The use of reported time percentages to calculate average and cumulative exposures likely resulted in lower exposure estimates than would have been assigned using methods relying on grouping strategies at the job classification level. The use of industrial hygiene data from other time periods and facilities in the development of the task exposures likely resulted in misclassification on an absolute micrograms per cubic meter scale, but less misclassification on a relative scale for comparing study participants. In assigning exposure estimates to tasks rather than individuals, the misclassification on both the absolute and relative scales should have been nondifferential. In spite of these potential misclassifications, we did find exposure–response relationships for CBD in our multiple logistic regression models. Future studies will be needed to address interactions with other genes in the HLA region and the effects of exposure on CBD severity as higher exposures may be more important with increasing CBD-related impairment. The use of contemporary exposure data in future studies will be critical to determine the CBD risk at average exposures less than 0.2 μg/m3.