Occupational exposure to beryllium in French industries

Abstract Beryllium (Be) is a metal mainly used in the form of alloys, with copper (Cu) and aluminium (Al) in the metal industry. Be is an extremely toxic element which must be handled under strictly controlled conditions to avoid health hazards to workers. Exposure to Be can be responsible for Chronic Beryllium Disease, a pulmonary disease preceded by sensitization to the element, and for lung cancer. The goals of the current study were to investigate Be exposure in France, to determine the airborne Be occupational exposure levels, the associated impregnation of employees through their urinary Be levels and the factors that might affect them, and finally to study a possible relation between biomonitoring and airborne data. Seventy-five volunteer subjects were thus atmospherically and biologically monitored in five French companies involved in Cu or Al casting, Al smelting, CuBe machining or AlBe general mechanical engineering. Airborne exposure was quite low with only 2% of measurements above the current French Occupational Exposure Limit (2 µg/m3); the population potentially most exposed was foundry workers. Impregnation with Be was also low with only 10% of quantified urinary Be measurements above the current German BAR value (0.05 µg/L). Using a Bayesian statistical modelling approach, the mean subject-specific urinary excretion of Be was found to increase significantly with the mean subject-specific exposure to airborne Be. From this relationship, and based on the current French OEL-8 hr, a Biological Limit Value of 0.08 µg/L (= 0.06 µg/g creatinine) could be proposed.


Introduction
Beryllium (Be) is a metal with interesting physicochemical characteristics (mechanical strength, corrosion resistance, thermal conductivity, permeability to X-rays, light weight, and high melting point), which explains its increasing use in advanced industries such as aeronautics, aerospace, nuclear, military, electronics, metallurgy, scientific and technical instrumentation, and manufacture of ceramics, medical equipment, dentures, watches, and jewellery. Annual consumption is estimated at approximately 40-60 tons. [1] Be is mainly used in three forms: Be metal, Be alloys (from 2-60% Be, with aluminium, nickel, copper, nickel-chromium, nickelcobalt, and others) and Be oxide. [2] In France, Be ore is neither extracted nor processed. Be is imported as pure Be oxide, alloys and scrap containing Be.
The toxicity of Be has been well-known for many years. Chronic Beryllium Disease (CBD) (preceded by sensitization) and lung cancer are often reported. The acute form of beryllium disease, caused by short-term exposures to high concentrations of Be, is only rarely encountered today. [3,4] Be is listed as a Group 1 (proven) carcinogen by the International Agency for Research on Cancer [5] and a Category 1B (known or presumed) carcinogen by the European Union. [6] In France, a register of cases of CBD began in 2010; 29 cases are listed today. [7,8] According to recent studies, 2-15% of Be-exposed workers will develop CBD. However, cases of CBD appear to be considerably under-diagnosed. The number of workers directly at workstations where they are exposed to Be in France lies in the range 9,400-14,400. [7] The sectors of industry in which workers are most exposed to Be are the industries of metallurgy and casting, precision machining, and scientific and technical instrumentation.
Several countries have set indicative occupational exposure limits (OELs): 2 mg/m 3 and 0.2 mg/m 3 are thus the 8-hr time-weighted average exposure limits for exposure to Be and its compounds in France and in the United States (OSHA Permissible Exposure Limit), [9] respectively. The American Conference of Governmental Industrial Hygienists (ACGIH V R ) recommends a time-weighted average Threshold Limit Value (TLV V R -TWA) of 0.05 mg/m 3 [10] and the European Scientific Committee on Occupational Exposure Limits (SCOEL) proposed in 2017 an OEL of 0.02 mg/m 3 . [4] Recent studies have indicated that the 8-hr-OEL at 2 mg/m 3 is not sufficiently protective against the risk of Be sensitization and CBD, whose latency can exceed 30 years. [12,13] In 2010, collective expertise of ANSES (French Agency for Food, Environmental and Occupational Health & Safety) on the evaluation of effects on health and methods for measuring occupational exposure levels for Be and its compounds, recommended a pragmatic 8-hr-OEL of 0.01 mg/m 3 should be set with the skin being indicated as the exposure route. [14] Generally, it remains difficult to establish a relationship between exposure and biological effects on employees. Many research programs are ongoing and are focusing on the identification of vulnerable populations, exposure levels, dermal exposure analyses, effects of low exposure, and genetics. Research areas have also been geared towards identifying the most predictive indicators of specific Be sensitization and chronic beryllium disease in the history and assessment of occupational exposure to Be (size, number, surface area, speciation, bioavailability). [15][16][17][18][19][20] Regarding biomonitoring of Be, several biological indicators of exposure have been identified in the literature, including blood Be, urinary Be, and Be in exhaled air condensates. Two biological endpoints of effects have also been described in the literature, namely the lymphocyte proliferation assay (Be-LPT), in blood first and then in bronchoalveolar lavage fluids, that are used as indicators of Be sensitization. [21] A team of experts from ANSES studied the advantages and limitations of each biological indicator. According to them, urinary Be (BeU) was selected as the most relevant for biological monitoring of occupational exposures to Be and its compounds. [21] No studies with workers correlated BeU with the occurrence of health effects (in particular carcinogenic effects and chronic berylliosis). However, relationships between BeU and airborne Be concentration (BeA), on a natural scale, have been reported [22] or calculated from data available on graphic (ATSDR 2002 from Stiefel, et al. [1980] and Zorn et al. [1986]). [23][24][25] However, because of the limited number and variability of these data, they are difficult to use. Also, Morton et al. [26] found that BeU measured in urine taken at the end of the shift and sampled at weekends from aluminum smelting workers was 47% higher than that measured in urine samples taken at the beginning of the week. The increase in average values ranged from 4.1-6.1 ng/L. [26] However, uncertainties regarding toxicokinetic data persisted and those data were too limited to establish biological limit values. Further studies were needed to better define these relationships.
The main objectives of this study were to: investigate Be exposure in French industry, to identify both the nature of the exposures and the exposed populations; determine the levels of occupational exposure to airborne Be and the associated impregnation of employees through urinary Be levels and the factors that might affect them; provide new insight into the toxicokinetics of Be; and study the relationship between the urinary excretion of Be and exposure to airborne Be, and, where applicable, to provide an estimation of a Biological Limit Value.

Sampling sites and volunteer selection
The most relevant processes in terms of exposure to Be were casting of non-ferrous metals (production or recycling of Cu-Be alloy), the sector of copper transformation, and, to a lesser extent, metal machining and aluminium production. In this study, five different factories were visited: a CuBe foundry (Company A), an AlBe mechanical engineering company (Company B), a CuBe parts machining company (Company C), an aluminium smelter (Company D), and an aluminium foundry (Company E). The workers were followed for 5-10 days.
Company A was a foundry in which cylindrical billets of CuBe alloys (Be content of 0.5% or 1.85% by weight) were produced. Cast founders, metal park workers and one blacksmith were monitored. Production of CuBe alloys was limited to one day during this campaign.
Company B was a company assembling AlBe alloy parts (up to 62% Be). Studied operations consisted in mechanical or adhesive bonding of AlBe parts followed by adjustment and alignment of the system. All these operations were carried out in a clean room or under local exhaust ventilation.
Company C was a company machining/manufacturing CuBe parts (2% Be). The first operation consisted in machining CuBe ingots, followed, on a suction table, by deburring and engraving the workpieces. Finally, the CuBe workpieces were fitted into their positions in the equipment and processed to their specified dimensions by hand machining without any local exhaust ventilation.
Company D was an aluminium smelter at which aluminium was produced by electrolysis of alumina. Be is an impurity of alumina and tends to be concentrated with fluorides in the electrolyte bath during aluminium production. Several operations were monitored: anode replacement and bath feeding by bridgecrane operators (fitted with personal respiratory equipment), cleaning operations using jackhammers for cleaning anodes, cathodes (lining), and melting bowls (operators fitted with personal respiratory equipment for cathodes and bowls), as well as various operations at the workplace (process management, bath sampling, anode transport for cleaning, and maintenance).
Company E was an aluminium foundry. Be was added as AlBe ingots (containing 5.4% Be by weight) to molten aluminium for specific applications before moulding. The final Be content in the alloys was limited to a few mg per kg. Four specific operations were monitored: preparation of loads from AlBe ingots, AlBe casting, dross recycling (using a rotary furnace) and disposal, moulded ingot transfer by bridge crane (bridge-crane operators).
For each company, we defined Similar Exposure Groups (SEGs), such as groups of workers with the same pattern of exposure to Be, due to the similarity and frequency of the tasks performed, of the materials used, and of the work processes implemented.
The study includes computer processing of personal data, and a request was submitted to that effect. Volunteers were specifically informed, and they were asked to fill in a questionnaire to identify their professional and personal habits.

Reagents and solutions
All of the chemicals used in the study were of analytical grade or higher. Nitric acid was used to prepare 0.2% and 2% HNO 3 (v/v) with ultrapure water. Be standard solutions for ICP-MS calibration were prepared by diluting a 10g/L Be standard stock solution (140.061.041, SCP Science, Quebec, Canada) with 2% v/v HNO 3 . An internal standard solution containing either 10 mg/L of Li, Sc, Y, In, Tb, Bi (VAR-IS, Inorganic Ventures, Christianburg, VA) or 100 mg/L of Sc (140.051.211, SCP Science, Quebec) was prepared by diluting a 1,000 mg/L internal standard stock solution with 2% v/v HNO 3 . The internal standard was added to all samples and standard solutions.

Apparatus
Inductively Coupled Plasma Mass Spectrometer (ICP-MS, Varian 820-MS, Palo Alto, CA), with an external sample introduction assembly with a Peltier-cooled spray chamber, a concentric glass nebulizer, a peristaltic pump mounted outside the torch box and an SPS3 auto sampler, was used for airborne and urinary samples. A PFA Nebulizer and a spray chamber, as well as Pt skimmer and sampler cones, were used for the airborne and urinary samples (containing HF). A discrete dynode electron multiplier detector provided nine decades of dynamic range in an all-digital pulse design. The Varian 820-MS system also featured a Collision Reaction Interface (CRI) providing fast, flexible, interference-free analysis using simple collision and reaction gases.
Inductively Coupled Plasma-Atomic Emission Spectroscopy (ICP-AES, Spectro Ciros, Kleve, Germany) and Atomic Absorption Spectrometry (AAS, Varian, AA280Z, Palo Alto, CA) were used for metal concentration analyses for the airborne and urinary samples from Company A, respectively.
Specific devices used for aerosol characterization provided a real-time recording of particle concentration and size.

Assessment of individual airborne exposure
For an 8-hr work shift, each of the workers was equipped with a battery-charged personal breathing air sampling pump that was set at a flow rate of 2 L/ min, and that was connected to a closed-face cassette. Airborne particles were collected on quartz fiber filters that were 37 mm in diameter for Company A to prevent the risk of filter overloading (considering dust level in this company) and on mixed cellulose ester filters (Millipore, Burlington, MA) for the other companies.
Particle digestion was carried out at room temperature in an HF-HNO 3 mixture for the quartz fiber filters [27] and HClO 4 -HF-HNO 3 mixture for the cellulose filters [28] directly in the cassette to take into account wall deposits. [29] In addition, to take into account the solubility of the Be-containing particles present in workplace air, an extraction protocol for inorganic Be species was applied in air samples specifically collected on cellulose ester membranes disposed in a cassette having a diameter of 37 mm using CAThIA sampling devices. [30,31] This method investigates aerosol equivalent solubility of the Be salts, Be metal, or calcined Be oxide without making it possible to identify the chemical species of Be. Extraction protocols and results for Companies A, D, and E have been described elsewhere. [32] Airborne measurements were performed during the working shifts only.

Collection and preparation of biological samples
To assess the exposure of employees, urinary Be among potentially exposed employees was measured from urinary samples collected occasionally. Volunteers collected their urine using a protocol with care to avoid contamination (collection outside the "polluted" area, without work clothes, and after handwashing).
Urine collections were performed for at least 5 consecutive working days concomitantly with airborne measurements. All voids were collected both at the workplace and at home. Work shift cards containing all the activities were established for each employee. This protocol included the entire working week, and in some cases the weekly rest periods (2 days before or after the week of urinary collections; not necessarily the weekends). Late workweek urines were, however, not systematically collected.
In each case, the urine was collected in suitable vials that were previously washed with nitric acid (10%). To avoid contamination, the volunteers were informed about collection procedures (handwashing, removing work clothes). Samples were aliquoted, acidified if necessary and frozen in a field laboratory installed in the company. The samples were then transferred for analysis. The set of measurements were carried out in a controlled-atmosphere room. Samples were collected in pre-cleaned polyethylene bottles and were stored at -20 C.
Creatinine levels were determined in all urine samples using an automated alkaline picrate method (Jaffe method) on a Daytona (Randox, Ireland) analyzer. BeU measurements were expressed per gram of creatinine (/g creat.) to reduce the possible effects of diuresis. All measurements with a creatinine concentration lying outside the range 0.5-3.0 g/L were discarded because they were overly diluted or overly concentrated, respectively.

Be assays in urine
This field study took place over 5 years. Over that time, analytical methods for determining BeU (and also BeA) were continuously improved. This explains why method quantification limits were quite different between companies: ranging from 0.002-0.1 mg/L for urine and from 0.11-100 ng/filter for air samples (the contemporary limits for each company are summarized in Table 1).
The most efficient method of analysis (0.002 mg/L) has been described in detail in a previous article. [33] Briefly, a chelate of Be acetylacetonate formed from Be(II) in human urine was pre-concentrated on an SPE C18 cartridge and eluted with methanol. After drying the eluate, the residue was solubilized in nitric acid and analyzed by AAS and/or ICP-MS. The procedure meets all the required validation criteria in terms of linearity limit, precision (intermediate precision, repeatability and reproducibility), accuracy, and sensitivity.
Quality control materials such as standard or certified reference materials were routinely used to ensure the accuracy of the instrument and of the analytical procedure. External quality assurance was performed by participation in international comparison programmes and quality assessment schemes.

Statistical analysis
The Geometric Mean (GM) and the Geometric Standard Deviation (GSD) were calculated for each SEG of each studied company, taking into account data below the limits of quantification (LOQ), with a "Tobit" regression model including a subject randomeffect (based on maximum likelihood estimation), to take into account the between-worker variability.
The kinetic profile of Be was assessed by describing urinary excretion, urinary Be concentrations and airborne Be exposure measurements over time.
Modeling of urinary Be as a function of airborne Be For this modeling, we considered only exposed workers (N ¼ 39). Owing to the fact that Be urinary Table 1. Descriptive statistics of Beryllium air exposure samples and Beryllium urinary measurements (with 0.5  excretion half-life was greater than 48 hr (see "Results" section), the modelling of the relationship between urinary and airborne Be was based on the respective chronic exposures. The model was based on the following four assumptions.
The geometric means of the airborne measurements depended on the SEG described in Table 1. The relationship between the urinary and airborne measurements was the same, whatever the SEG. For each worker, the airborne exposure measurements during the week were representative of his/ her chronic airborne exposure. For each worker, the measurements of the urinary beryllium concentrations depended only on his/her chronic airborne exposure and on a random within-worker component.
From a statistical point of view, and for each SEG, we modelled the airborne exposure measurements by a log-normal distribution whose (geometric) mean depended on the chronic airborne exposure and whose (geometric) standard deviation was constant across the workers. Denoting by X ij the log-transformed airborne exposure measurements for subject i, this can be written in following equation: We modeled the urinary measurements by a lognormal distribution whose (geometric) mean depended only on the chronic airborne exposure and whose (geometric) standard deviation was constant across the workers. We assumed that the urinary measurements (on a logarithmic scale) depended linearly on airborne measurements.
Denoting by Y ij the log-transformed urinary measurements for subject i, this can be written as: In detail, the model, adapted from [34] , estimated the relationship, on a log scale, between the subjectspecific mean urinary Be Y i (dependent variable) and the subject-specific mean airborne Be X i (independent variable) using a Bayesian framework.
Thus, the relationship modelled was between the mean of all urine samples of the week (several per day) and the weekly mean of all airborne samples (one per shift). These means were modeled using all obtained measurements including measurements below the LOQ for both urinary and airborne Be. This model took into account between-worker and within-worker variability for both airborne and urinary measurements. Informative priors were used for the within-worker and between-worker airborne exposure variability parameters, based on published literature. [35] The details of this model are presented in Appendix A.
Finally  Table 1 summarizes descriptive statistics about Be airborne and urinary measurements. These are presented by company and by SEG (job function) within a company.

Air sampling and solubility of Be in aerosols
Airborne exposure was quite low with only four measurements (2% of measurements) above the French OEL (2 mg/m 3 ) (three in Company A and one in Company D). While most of the air samples for Companies B and C were below quantification limits with 97% and 88%, respectively, only 14% were not quantified for the other companies.
On average, exposure levels were highest in Company A. The casting of CuBe alloys in Company A lasted only one day in the middle of the week of the investigation. The metal (ingots and recycled chips of Cu and CuBe) was loaded into the two melting furnaces. Once the desired composition of the alloy was reached, the furnaces were switched successively into the channels that conveyed the molten metal to the casting head of the continuous casting machine. As a result, only some cast founders were directly involved in the casting of CuBe alloys. Aerosol concentrations determined from real-time monitoring using Electrical Low Pressure Impactor (ELPI, Dekati) were high (from 10 5 to 5.10 6 particles/cm 3 ). Airborne particles consisted in more than 95% sub-micronic particles (i.e., <1mm), which represented only approximately 10% of the total mass. More than 80% of the particles were less than $100 nm. Operators were mainly exposed to "Be oxide"-like particles [32] and none of them wore protective respiratory equipment. Be airborne measurements were all quantified, and were found higher for the "cast founder" group than for the "metal park worker" and "blacksmith" groups.
Data interpretation for Company D was complicated by the number of work stations studied, the variability of the solubility of Be and the size of particles found in the work atmosphere, and the wearing of personal protective equipment for particular activities within the same task; the GSD at 4.45 reflects the dispersion of the measurements. The lining activity, for instance, generated a very large amount of coarser aerosols with the presence of insoluble Be compounds mixed with more soluble phases. [32] Activities using a jackhammer, such as lining or fusion bucket cleaning generated similar exposure levels. For these two "similar" activities, operators were equipped with personal protective equipment. Furthermore, the lining operator used a remote jackhammer allowing him to stand back from the dust transmission area. The particle size distributions, determined using a Handheld 3016 particle counter (Lighthouse), showed a predominance in number of fine particles of aerodynamic diameter less than 0.5 mm for bowl cleaning as well as workshop operations, whereas lining operations generated coarser particles (>2 mm).
Very low total concentrations of Be were detected in Company E except for the Dross Recycling Operator group and, to a minor extent, for the Cast Founder group. The distribution of the particle size determined using the Handheld 3016 particle counter mostly consisted in predominance in the number of fine particles of an aerodynamic diameter which was less than 0.5 mm as for Company D. Differences in Be solubility were found between the tasks investigated. During casting, Be present in the aerosol was quite readily soluble whereas for dross recycling operations (melting and disposal) solubility was low, equivalent to the solubility of Be oxides. [32] During dross recycling and disposal, the personal concentration of airborne Be increased significantly for recycling operators (0.16 ± 0.02 mg/m 3 ) relative to other operations such as dross removal from the melting bath (0.05 ± 0.02 mg/m 3 ).

Urine samples
Fourteen percent of the total urinary measurements, which corresponds to a number of 234 measurements, were discarded because of their creatinine concentrations (outside the range 0.5 g/L to 3.0 g/L). Over the remaining (2,600) measurements, BeU was quite low; only 81 measurements (6%) were above the current German BAR urinary value (0.05 mg/L, equivalent to 0.035 mg/g creat. considering a mean creatinine concentration of 1.4 g/L in the population). Of these, 77 measurements came from Company A and 4 from Company D. All of the measurements in Companies B and C were below the LOQ (¼ 0.02 mg/L).
For Company A (CuBe foundry), 13 workers were monitored. Seventy-five percent of the BeU measurements were below the LOQ (¼ 0.1 mg/L). The "Cast founder" group was more exposed since 41% of urinary measurements were above 0.1 mg/L (geometric mean of 0.048 mg/g creat.).
For Company D (Al smelting), 15 workers were monitored. The geometric mean of the BeU excretion was 0.005 mg/g creat. Only 8% of urinary Be measurements were below the LOQ (0.002 mg/L), "jackhammer operator" and "bridge-crane operators" groups had the highest BeU geometric mean, with 0.0092 and 0.0054 mg/g creat., respectively.
For Company E (Al foundry), 13 workers were monitored. The level of BeU was even lower than for Company D whose geometric mean was 0.0020 mg/g creat. Fifty-three percent of the BeU measurements were below the lower LOQ (¼ 0.002 mg/L). As previously, the "bridge-crane operators" group had the higher BeU.

Assessment of the urinary Be half-life
Analysis of all of the kinetics (data not shown) did not reveal any excretion profile related to airborne Be exposure. Indeed, no excretion peak was measured in the hours following the exposure. Moreover, urinary excretion of Be did not decrease during the 2-day unexposed periods (periods without any airborne Be exposure) following the working week. It could be concluded from these analyses that urinary Be measurements do not reflect recent exposure and that the urinary excretion half-life is greater than 48 hr.

Modeling of BeU as a function of BeA
In the following, Companies B and C have not been included in the statistical analysis because all of the urinary and almost all of the airborne measurements were below their respective LOQs.
Therefore, the Bayesian model was applied to a dataset including Be measurements (BeA and BeU) of the 39 exposed workers of Companies A, D and E. This studied dataset comprised 1,302 measurements of BeU (in mg/g creat.) and 180 measurements of BeA. Of these, 488 BeU (37%) and 19 BeA (10%) were below the LOQs. Table 2 shows the main parameter estimates for the modelling, with their 95% Bayesian confidence interval (CI). Convergence of the posterior distribution of the different estimated parameters was verified using the Brooks-Gelman-Rubin diagnostic graphs.
The linear slope between subject-specific mean logtransformed BeU and mean log-transformed BeA was estimated at 0.59 with a 95% confidence interval (95% CI) [0.410.77] not containing zero, attesting to the statistical validity of this relationship. The exponentiated intercept of this linear regression model was estimated at 0.047 mg/g creat. It corresponds to the BeU estimation at 1 mg/m 3 BeA. The relationship is illustrated in Figure 1.
From the French 8-hr-OEL equal to 2 mg/m 3 , a BLV estimated at 0.06 mg/g creat. was derived from this relationship, with a 95% confidence interval [0.03; 0.13]. The between-worker geometric standard deviation (GSD) of airborne exposure Be data, BW GSD A , was lower than the within-worker GSD, WW GSD A . Conversely, the between-worker GSD of urinary Be data, BW GSD U , was greater than WW GSD U . It should be noted that these GSD U s are for a constant airborne exposure.

Discussion
A previous assessment survey of occupational exposure to Be in 95 facilities belonging to 37 sectors of activity was conducted in France from late 2004 to the end of 2006. [36] At that time, more than 15% of the airborne measurements exceeded the French 8-hr-OEL (2 mg/m 3 ) (and 50% above the ACGIH TLV value of 0.05 mg/m 3 ). Exposures related to activities requiring the use of hot alloys in foundries and electrometallurgy (Al production) were the highest (5.4 mg/m 3 in average); followed by those from the radio, television, and communications production industry (2.4 mg/m 3 ). In the metallurgical industry, 75% and 40% of airborne measurements exceeded 0.05 mg/m 3 and 2 mg/m 3 , respectively. In the metal-working industry where the number of exposed employees is greater, exposure levels were lower (0.19 mg/m 3 in average), but 30% of the measurements exceeded 0.05 mg/m 3 .
The present study initiated in the early 2010s focused on five French companies in which high exposure levels could be expected, namely companies working in the following fields: Cu or Al foundry work, Al smelting, CuBe machining or general mechanical engineering. Although heterogeneous between companies (BeA could extend over 3-4 decades), and sometimes between SEGs within a company, airborne exposure was quite low, with "only" 2% of measurements above 2 mg/m 3 . The population potentially most exposed was still foundry workers. Although perfectible, a significant improvement in working  Table 2. Estimated parameters and BLV using French OEL, ACGIH TLV-TWA,OSHA PEL, and recommended SCOEL OEL. BW GSD A is the between-worker geometric standard deviation of airborne exposure Be data. WW GSD A is the within-worker geometric standard deviation of airborne exposure Be data. BW GSD U is the between-worker geometric standard deviation of urinary Be data. WW GSD U is the within-worker geometric standard deviation of urinary Be data. conditions with a reduction in airborne exposure levels can nevertheless be noted. The dispersion of the urinary data was also noted not only between companies but also within companies. Considering the data as a whole, a relationship could be obtained between subject-specific mean log concentrations of urinary and airborne beryllium (Figure 1). On average, the employees of Company A who were the most exposed to airborne Be excreted more urinary Be than the employees of Company D, who, in turn, excreted more than the employees of Company E, who were known to be the least exposed to airborne Be.
Within each company, the relationship between BeU and BeA became weaker. For the same level of BeA, the BeU could vary by up to three orders of magnitude. Several hypotheses related to the activities can be put forward to explain these differences (difference in solubility in the species of Be handled, whether or not personal respiratory protection is used) but they do not really make it possible to explain such amplitude. Special attention has therefore been paid to the elimination kinetics of Be.
From the individual analysis of the kinetics of urinary Be, the hypothesis of a short half-life of Be is excluded. Urinary Be does not reflect a very recent exposure but rather exposures over preceding weeks or months. Consequently, BeU can be considered as a marker of chronic exposure and not of an immediate exposure. This result correlates well with measurements taken simultaneously in similar French companies in a companion study. The authors of that study found a significant relationship between certain biomarkers of pulmonary inflammation in Exhaled Breath Condensate (EBC) and indexes both of cumulative exposure and of duration of exposure to Be. [37] Thus, the Be urinary concentrations at the end of the shift, like at any other time-point, reflect more the chronic exposure than the exposure of the given day.
Considering the measurements of the 39 exposed workers of Companies A, D, and E, this finding (BeU half-life higher than 48 hr) made it possible to adapt a model to quantify the relationship between BeU and BeA. To do that, a statistical method based on Bayesian theory was used, making it possible to use measurements below the LOQ for both urinary and airborne data, and integrating random effects in order to quantify both between-worker variability and within-worker variability. This method has shown its benefits compared to methods based on maximum likelihood which substitute or delete some of the data below LOQ for airborne exposure.
For urinary data, the Bayesian modelling estimated a between-worker variability greater than the withinworker variability ( BW GSD U > WW GSD U ) for a given airborne exposure within a company. This result was consistent with our analysis of historical data for metals collected by our laboratory (unpublished data), performed by using a mixed model including the airborne exposure: the between-worker variance component was indeed greater than the within-worker variance component adjusted on airborne exposure. The reason might be the inter-individual differences in physiology and metabolism, whereas the individual urinary excretion depends mostly on the past airborne exposures. Thus, this between-worker variability must be taken into consideration when analyzing biomonitoring data, involving repeated measurements workerby-worker. Conversely, for airborne data, the Bayesian model estimates a between-worker variability lower than the within-worker variability ( BW GSD A < WW GSD A ). Indeed, for the airborne exposure model, within-worker variability, due to day-to-day differences is usually greater than the between-worker variability which depends mostly on worker-specific practices. [38] However, several aspects are worth taking into consideration prior to estimating the robustness of the model. First, the model does not include the solubility of airborne exposure (itself linked to the speciation of Be and probably also to the size of the inhaled aerosols). And as already shown, Be solubility differs depending on the company. It is then tricky to include and test fixed-effect solubility in the model, associated with airborne Be level, since solubility is confused with company. In the same way, it was not possible to integrate other factors like how often the workers wear their personal protective equipment, the seniority of their positions or their smoking statuses in this analysis. Indeed, there is currently not enough data to integrate these variables into the model.
Another weakness of this model is to estimate how representative the chronic exposure is based on the measurements over the week. And, unfortunately, it is not possible to verify this hypothesis. However, considering the significance of the estimated relationship between urinary and airborne measurements, this hypothesis may seem reasonable.
Finally, the assumption of linearity between the log of urinary measurements and the logarithm of airborne measurements is questionable for low urinary beryllium values, since they are of the order of nonoccupationally exposed population values (German BAR value ¼ 95 th percentile of a non-occupationally exposed population ¼ 0.05mg/L). Indeed, at such low urinary concentrations of beryllium, it is reasonable to question the occupational nature of the exposure. In order to explore the influence of low exposures, a sensitivity analysis at these low values was performed, with the same model, excluding the data from Company E, which was the company with the lowest airborne exposure (data not shown). It should be noted that our estimate of BLV at 0.06 mg/g creat. based on the current French 8 hr-OEL, has a wide 95% confidence interval [0.03-0.13], related to the fact that all airborne exposures are well below the French 8 hr-OEL and that the estimation is therefore based on extrapolation. We can note that, based on the TLV-TWA of ACGIH, the BLV was estimated at 0.007 mg/g creat. (95% CI [0.005;0.009]); based on the PEL of OSHA, the BLV was estimated at 0.015 mg/g creat. (95% CI [0.01; 0.023]); and based on the recommended OEL of the SCOEL, the BLV was estimated at 0.004 mg/g creat. (95% CI [0.003;0.005]) ( Table 2).
Until now, the limited amount of workplace data and the significant differences between studies did not make it possible to establish a BLV in France. Recently, a BRV (Biological Reference Value $ German BAR) (< 7 ng/L) was proposed by ANSES, in 2018, [21] but it should be noted that that value corresponds to the limit of detection used in the only study considered representative for the general French population conducted by Hoet and colleagues in 2013. [39] That value was not intended to protect health effects but rather to provide support for the interpretation of worker exposure levels.
Our proposal is slightly higher than the German BAR value (0.05 mg/L, i.e., 0.035 mg/g of creatinine if we consider an average concentration of 1.4 g/L of creatinine in the population) or the BGV (Biological Guidance Value) of 0.04 mg/L for urinary beryllium recently recommended by the Scientific Committee on Occupational Exposure Limits (SCOEL). [11] However a single spot measurement of BeU cannot directly be compared to our estimated BLV as it is based on the average BeU over the working week. Nevertheless, from our estimate of the he within-worker urinary variability ( WW GSD U in Table 2), we can deduce that there is a 95% chance that the average urinary value of the worker will lie in the range U -29% of U to U þ 41% of U, i.e., of that measurement. Only the latter value could then be referred to a given BLV.
However, a better strategy would be to repeat the BeU measurements over a working week, to better evaluate the chronic exposure of a worker. The mean of these measurements would evaluate the worker's chronic exposure, and could be compared with the proposed BLV. Finally, since surface contamination can promote Be incorporation into the body and because Be is classified as a Category 1 skin sensitizer under the European Regulation and because the consequences of sensitization by skin contact can cause the occurrence of general immunoallergic pathologies of particular concern, the cleaning of surfaces and clothes and personal hygiene must absolutely not be neglected. In addition to urinary measurements, skin wipes (face þ wrist) and surface measurements should be considered as an efficient preventive tool which deserves concern.
Conclusion Seventy-five volunteers were atmospherically and biologically monitored in five French companies involved in Cu or Al foundry work, Al smelting, CuBe machining, or general mechanical engineering. Airborne exposure was low with only 2% of measurements above the current French Occupational Exposure Limit (2 mg/m 3 ); the population potentially most exposed was foundry workers. Impregnation with Be was also low with only 10% of quantified urinary Be measurements above the current German BAR value (0.05 mg/L).
Using a Bayesian statistical modelling approach, the mean subject-specific urinary excretion of Be was found to increase significantly with the mean subjectspecific exposure to airborne Be. From this relationship, and based on the current French 8-hr-OEL, a Biological Limit Value of 0.08 mg/L (¼ 0.06 mg/g creat.) could be proposed. This value is however conservative as it was estimated at 0.15 mg/g creat. when excluding the lowest exposed company. Guillaume Antoine, Alain Boulet, Nathalie Carabin, Fr ed eric Cosnier, Thibaut Durand, Jean-Marie Elcabache, Virginie Matera, Mathieu Melczer, Samuel M€ uller, V eronique Oury, and Anne-Marie Lambert-Xolin.

Funding
Funding for this project was entirely based on the core funding of the INRS which is the sole employer of the authors. The authors declare no conflict of interest relating to the material presented in this article. Its contents, including any opinions and/or conclusions expressed, are solely those of the authors.