Introduction

Between 1946 and 1990, about 230,000 tons of uranium were mined by the former WISMUT company in East Germany for the Soviet nuclear program (Wesch et al. 1999). The miners were not only exposed to high doses of ionizing radiation and crystalline silica but in many mines also to arsenic. An arsenic content of up to 8% in the ore together with lacking protective measures under Soviet supervision posed a serious health hazard for many workers.

Medical records of about 450,000 miners and collected tissues of 20,793 miners comprise a comprehensive archive of valuable cancer data, the scientific importance of which equals that of the records of the atomic bomb survivors (Kahn 1993). The pathological institute of the WISMUT company possessed an extensive tissue repository now being a part of the WISMUT archive (Mohner et al. 2006). A histopathological database of the tissue depository has been developed, implementing the re-classification of about 5,545 lung cancers by three independent pathologists according to WHO guidelines (Wesch et al. 1999). Enormous efforts were undertaken to quantitatively assess exposure to ionizing radiation, quartz dust, and arsenic, including reconstruction of historical workplaces (Lehmann et al. 1998; Bauer 2000; Dahmann et al. 2008). Linking these WISMUT databases now allows to mine key cancer data and to evaluate the effects of lung carcinogens quantitatively.

The International Agency for Research on Cancer (IARC) has classified arsenic as a Group 1 carcinogen for the development of skin and lung cancer in humans (IARC 1973, 1980, 1987). So far, major research has concentrated on skin cancer induced by arsenic-contaminated drinking water (Yu et al. 2006). Other sources of arsenic exposure and their effect on lung cancer have been investigated to a lesser extent. The knowledge of the mechanisms of action of arsenic and the nature of the putative pulmonary target cells is scarce. Furthermore, it is not yet clear if arsenic is a complete carcinogen of the lung. Being a very heterogeneous tissue, the lung gives raise to three major types of cancer, namely small cell lung cancer (SCLC), squamous cell carcinoma (SqCC), and adenocarcinoma (AC), the latter two belonging to the group of non-small cell lung cancers (NSCLC). The histological tumor types are likely to be of different cellular origin (Kim et al. 2005). Whether the corresponding cell types represent specific targets for certain carcinogens is controversial (Taeger et al. 2006a). There are only a few studies showing an arsenic-induced shift in the distribution of histological types of lung cancer (Guo et al. 2004; Newman et al. 1976; Tokudome et al. 1988; Axelson et al. 1978; Wicks et al. 1981; Pershagen et al. 1987). Furthermore, the low statistical power and weak study design of the existing data do not allow a sound conclusion on a relation between arsenic exposure and a specific type of lung cancer, a view shared by the IARC (IARC 1980; Ives et al. 1983).

In analogy to the role of arsenic in the development of skin cancer and based on our previous results (Taeger et al. 2006a, b), we hypothesized that arsenic may influence the distribution of lung cancer types. Particularly, the induction of SqCC may be closely associated with arsenic exposure. Therefore, the present study aimed to investigate whether and how airborne arsenic may shift the distribution of the major histological types of lung cancer among miners. In contrast to previous studies, quantitative assessment of occupational exposure to arsenic and co-existing carcinogens, extensive reference pathology of the cell type, large size of the study, and statistical modeling were anticipated to allow for a better assessment of the role of arsenic in lung cancer development. We considered radiation and quartz dust as co-carcinogens in uranium mining and stratified by silicosis as potential effect modifier.

Material and methods

Study population

Histopathological data and other relevant information for male German uranium miners who died from lung cancer during 1957–1990 were retrieved from the database of the WISMUT tissue repository. The majority of miners worked underground. That database has been described previously by Wesch et al. (1999). In brief, the database contains information about the underlying cause of death and accompanying diseases classified according to the International Classification of Diseases (ICD-9) (WHO 1977). The histopathological type of lung cancer was determined from the tissue slides independently by three German pathologists. From the database, taking into account the information of the WISMUT job-exposure matrix (JEM), we selected male lung carcinoma cases (ICD-9 162) with exposure to arsenic, quartz dust, and radon. Eligible for analysis were miners with one or more years of exposure to at least one of these carcinogens. The analysis was restricted to SCLC, SqCC, or AC as the major cancer types. Only cases where at least two of the three pathologists agreed unambiguously on the classification of the histological type were included. Data on age at death, silicosis as assessed during autopsy, duration of exposure, and time since last exposure were extracted. A total of 1,786 lung carcinoma cases with complete information on all variables fulfilled the above criteria. Smoking information was available as dichotomous variable only. A worker was classified as smoker if smoking was mentioned at least once during employment or in the medical records of the WISMUT health data archive.

Exposure assessment

Over a period of more than one decade, a WISMUT JEM has been developed for the assessment of exposure in a detailed manner regarding job tasks, mineshaft, geology, and other factors. Great efforts were undertaken to evaluate the annual exposure to radon and its progeny, quartz dust, and arsenic, based on available measurements, model calculations, and reconstruction of historical workplaces (Lehmann et al. 1998; Bauer 2000; Dahmann et al. 2008). For each miner, cumulative arsenic exposure was measured as arsenic years, i.e., the accumulated exposure to arsenic in μg/m3 as annual shift average times duration of exposure in years. Cumulative exposure to respirable quartz dust was measured as quartz years in a corresponding manner, i.e., the lifetime exposure to the quartz concentration of the alveolar fraction in mg/m3. Cumulative radon exposure was given in working level months (WLM). A working level (WL) is equivalent to any combination of short-lived radon daughters in 1 liter of air that will result in the ultimate emission of 1.3 × 105 MeV of potential alpha particle energy. A WLM equals the exposure to 1 WL for 170 h. Individual job tasks and the number and duration of annual shifts of a miner were taken into account.

Statistical analyses

To investigate the hypothesis that the distribution of the three major histological types of lung carcinoma changes with the levels of arsenic exposure, we first calculated the observed fractions of the three major histological types of lung carcinoma with 95% confidence limits by different exposure levels of arsenic (>25th quartile; 25–50th quartile; 50–75th quartile; >75th quartile) for all miners, silicotics, and non-silicotics. In addition to this analysis at group level, we estimated the individual probability of each miner for dying of a specific histopathologic type relative to the remaining types of lung cancer at the corresponding year of death. We used a polytomous (multinomial) logistic regression model to estimate the individual probability of miners for dying from the three major cancer types at various arsenic exposure levels, conditional on fixed levels of the other exposures and covariates (Hosmer and Lemeshow 2000). These probabilities are estimates of the relative fraction of dying from a specific type of lung cancer adjusted for covariates. This method has been applied for occupational settings by Schmidt and Strauss (1975). The method of Liao (2000) was applied to calculate 95% confidence limits for the estimated fractions.

We used SAS/STAT software version 9.2 (SAS Institute Inc., Cary, NC) for all calculations. The procedure LOGISTIC with the option LINK = GLOGIT was applied to select the best-fitting polytomous logistic regression model. The following covariates resulted in the lowest Akaike’s Information Criterion (AIC = 3738.017): cumulative arsenic years (μg/m3 × years), cumulative quartz years (mg/m3 × years), cumulative radon exposure (WLM), age at death (years), duration of exposure (years), and time since last exposure (years) for each exposure variable. The estimates of the parameters of these variables were used to estimate the relative fraction of dying from different tumor types at various arsenic exposure levels, conditional on fixed levels of the other exposures and covariates. The 25th and 75th percentiles of the co-exposure distributions to quartz and radon defined these exposure levels. The median of the other explanatory variables was used correspondingly. The modeled fractions were dependent on the arsenic levels. We additionally stratified the study population by silicosis and smoking status. For the 36 non-smokers no meaningful logistic regression was applicable due to small numbers.

Results

The study population consisted of 1,786 miners with exposure to arsenic, quartz dust, and radon who died from a major cell type of lung cancer between 1957 and 1990. Table 1 describes the exposure characteristics of the study population by lung cancer type. The predominant cell type was SCLC (n = 818), followed by SqCC (n = 511), and AC (n = 457). Miners with SCLC died on average at younger age (60.1 years) than miners with SqCC (62.8 years) or AC (62.0 years) and had a shorter time since last exposure for all carcinogens than miners with NSCLC. The opposite was observed for the duration of exposure, i.e., miners with SCLC had longer duration of exposure. Cumulative exposure to arsenic was highest among miners with SqCC, followed by AC. AC occurred at lower levels of quartz and radon exposure with a longer time since last exposure than SCLC and SqCC. The fraction of miners with silicosis was 40% in workers who developed SCLC, 33% in miners with SqCC, and 27% in cases with AC. Out of 1,786 miners, 503 were identified as ever smokers and 36 were likely non-smokers, but for the majority of miners smoking information was missing.

Table 1 Exposure characteristics of German uranium miners by major cell type of lung cancer
Full size table

The estimated coefficients of the arsenic exposure of the polytomous regression models are given in Table 2. Regardless of the silicosis status the coefficients for SCLC versus SqCC keep their algebraic negative sign, meaning that SCLC, in comparison to SqCC, is decreasing for one unit (μg/m3 × years) of additional arsenic exposure, whereas for SCLC versus AC and SqCC versus AC the algebraic sign changes from positive to negative with silicosis status (non-silicotics to silicotics).

Table 2 Estimated coefficients of the polytomous logistic regression models for the entire study population (N = 1,786), non-silcotics (N = 1,117), and silicotics (N = 669)
Full size table

The observed fractions of histological types of lung carcinoma in quartiles of arsenic exposure by silicosis status are given in Table 3. No clear pattern of increasing or decreasing observed fractions can be found. For non-silicotics, the fractions remained nearly unchanged, but for silicotics strong fluctuations in fractions between the quartiles for SqCC and AC can bee seen.

Table 3 Observed fractions with 95% confidence intervals of major cell types of lung cancer at various quartiles of exposure to arsenic (<25th quartile, 25–50th quartile, 50–75th quartile, >75th quartile)
Full size table

The results of the polytomous logistic regression model that estimates the fractions of dying from a certain cell type of lung cancer at various arsenic levels conditional on co-existing but fixed levels of exposures to quartz and radon are given in Fig. 1. Silicosis turned out to be a major determinant of the distribution of cell types by exposure to arsenic. Arsenic exposure was related with non-small cell cancer. SqCC fractions increased with rising arsenic levels in non-silicotics by about 16% at all levels of co-exposure to quartz or radon. The highest SqCC fraction was found under high exposure to arsenic, radon, and quartz. No substantial impact of arsenic on SqCC occurrence could be found among silicotics. The AC fractions did not vary by arsenic exposure when analyzing the total study population but varied in the subgroups stratified by silicosis. Among non-silicotics, the AC fractions decreased, whereas in silicotics the AC fractions increased with rising arsenic levels. The fractions of AC occurrence were lowest at high levels of radon and quartz dust, being very low in non-silicotics. The SCLC fractions decreased with increasing arsenic exposure by more than 10% at all levels of co-exposures in miners with silicosis. In miners without silicosis, the decrease was less pronounced. Analyzing ever smokers separately no different patterns as described above were found (data not shown).

Fig. 1
figure 1

Estimated fractions of the occurrence of the major cell types of lung cancer by arsenic level for the entire study population and stratified by silicosis at fixed levels of covariates (25 and 75th percentiles of exposure to quartz and radon and median of temporal covariates)

Full size image

Discussion

Arsenic has been classified as a human carcinogen. Skin and lung are the major target organs affected by different routes of exposure. Whereas arsenic-induced skin cancer has been well investigated (Yu et al. 2006), less is known about the target cells and their interplay with co-existing carcinogens in arsenic-induced lung cancer. So far, exposure data for airborne arsenic at workplaces are scarce. Based on the comprehensive histopathological and exposure information on German uranium miners, we analyzed the role of arsenic in lung cancer regarding to cell type, with radon, quartz dust, and silicosis as cofactors.

Although observed fractions show no clear pattern of changes in the distribution of cell types with various categorical levels of arsenic exposure, the estimation of the individual probabilities of each miner shows an explicit pattern. The latter analysis incorporates additionally the effect of quartz and radon as co-carcinogens. The algebraic sign of the coefficients of the polytomous logistic regression models reflects these patterns. These coefficients are not statistically significant. However, it cannot be expected that one unit (μg/m3 × years) change in arsenic exposure imposes a significant change.

In a preceding analysis, we found that radon exposure was more closely associated with the development of SCLC (Taeger et al. 2006a), which confirmed observations in other studies (Kreuzer et al. 2000). Here, we found a decrease in the relative occurrence of SCLC with increasing arsenic exposure that was even more pronounced in miners with silicosis. Thus, arsenic appears to be associated with NSCLC. Regarding the debate whether arsenic is a complete carcinogen further investigations are needed.

However, silicosis was a major determinant of the distribution of NSCLC cell type by arsenic exposure. In miners without silicosis, SqCC fractions were elevated with rising arsenic levels on the expense of AC. On the one hand, in silicotics, AC occurred at a relatively high fraction and more frequently with arsenic exposure on the expense of SCLC where SqCC fractions remained relatively unchanged. These results correspond to the observed fractions of the histological subtype of lung cancer by arsenic exposure, but incorporate time-dependent co-variables in the model.

This result raises the question whether the increase in AC occurrence reflects a residual confounding by dust exposure, or another mode of action of arsenic. For example, a possibly synergistic mode of action of arsenic and silicosis could be discussed.

Many studies suffer from uncontrolled confounding due to small size or lacking data on co-existing exposures. In our relatively large study the potential confounding was controlled by estimating the fractions of the occurrence of cell type conditional on different but fixed exposure levels of radon and quartz dust. Nevertheless, highly correlated data are difficult to disentangle.

A limitation of the study is poor smoking data. However, the arsenic effect seen here is probably not strongly confounded by smoking because only a very small number of workers were non-smokers. However, a co-carcinogenic effect of smoking cannot be ruled out. Further, a competing mortality from silicosis has to be taken into account (Taeger et al. 2008).

Another general concern in using histopathological data is histological heterogeneity of lung tumors (Travis et al. 1999), diagnostic disagreement among pathologists (Stang et al. 2006) and the usually unknown source population (Taeger et al. 2008). Deceased miners with lung diseases, including lung cancer and silicosis, had a high probability of autopsy due to compensation claims. Here, we analyzed lung cancer cases where at least two out of three pathologists agreed upon the cell type and restricted the analysis to the major cell types. Applying the more rigid criteria of full agreement between all three pathologists resulted in similar risk estimates but in a loss of statistical power (data not shown).

We analyzed the data with the a priori hypothesis that arsenic exposure may be associated with SqCC. The hypothesis was based on the known effects of arsenic in skin and recent findings for the liver (Ramirez et al. 2000). Arsenic accumulates in the skin and causes increased expression of cytokeratins, hyperkeratosis, and skin cancer, the latter frequently being SqCC (Yu et al. 2006; Centeno et al. 2002; Tapio and Grosche 2006). Arsenic has a high affinity to thiol groups, including those attached to proteins, which might contribute to its accumulation in certain tissues. In liver cells, arsenic generates cross-links between cytokeratins and DNA and increases the cytokeratin expression (Ramirez et al. 2000; Liu et al. 2006). The cells of the normal bronchial epithelium express a combination of different cytokeratins, while squamous cell metaplasia and SqCC show a shift in distribution and expression pattern towards an overexpression of some cytokeratins (Fisseler-Eckhoff et al. 1996; Tsubokawa et al. 2002). The cells in the central parts of the lung might be generally more susceptible to arsenic exposure. Alternatively, because of the relatively good solubility of arsenic compounds, most of the inhaled arsenic may be dissolved in the central parts and therefore, the bronchial mucosa may experience the highest exposure to arsenic within the lung, the preferred location for SqCC.

In our hypothetical scenario arsenic may accumulate locally in the bronchial mucosa and in a first step cause cellular proliferation and transformation to squamous epithelium. Squamous metaplasia is also frequently seen in smokers, and smoking is associated with a higher risk for the development of SqCC or SCLC than with AC (Simonato et al. 2001). Therefore, a synergistic effect between arsenic and smoking may be expected (Lau and Chiu 2006; Chen et al. 2006). An increased expression of cytokeratins would facilitate the accumulation of arsenic within the cells leading to further cross-links, DNA damage, and other defects. The subsequent genetic and epigenetic “reprogramming” events would finally lead to the development of SqCC.

How arsenic changes the differentiation of epithelial, progenitor or stem cells is still not known. The lung is a heterogeneous tissue where terminally differentiated cells are assumed to develop from different stem or progenitor cells (Giangreco et al. 2007). According to the cancer stem cell hypothesis, cancer is driven by transformed stem cells or de-differentiated cells with stem cell-like properties (Polyak and Hahn 2006). Regarding putative cancer stem cells, NSCLC and SCLC are likely of different origin (Kim et al. 2005). If radon, quartz dust, and arsenic are associated with different histological types of lung cancer, one could conclude that these carcinogens act on different progeny lineages or on cells with different degrees of stemness within a certain lineage (Taeger et al. 2006a). Consequently, arsenic may act on stem or progenitor cells of the bronchial epithelium by altering the expression of genes involved in key regulatory pathways. Indeed, arsenic has been shown to inhibit protein tyrosine and serine/threonine phosphatases, regulatory enzymes that often carry a thiol group in their active center, and to alter the expression of many kinases implicated in signaling pathways (Andrew et al. 2003; Cavigelli et al. 1996; Huang et al. 1995; Tapio et al. 2005). In line with this, large-scale expression analysis revealed that specific lung developmental pathways are reactivated in NSCLC (Borczuk et al. 2003). Proof of the hypothesis is feasible; however, it requires molecular investigations of tumor samples of the WISMUT archive (Wistuba et al. 2002).

Regarding the effect of arsenic on AC in workers with silicosis, little is known about the molecular mechanisms relating silicosis with lung cancer. Future work will include elucidation of the role of silicosis and quartz dust in more detail with the WISMUT databases and tissues. The latter will be used to investigate molecular patterns on the genomic, epigenetic, transcriptional, and proteomic level to address the question of synergistic action of radiation, quartz dust, and arsenic in lung carcinogenesis.

Conclusion

The analysis of the WISMUT cancer and exposure data provides insight into the putative cellular targets and the combinational effect of different lung carcinogens. The results allow the deduction of hypotheses for experimental studies to investigate the underlying modes of action. Here, we found an association of arsenic exposure with NSCLC in general. In particular, SqCC was more frequently associated with high arsenic exposure in miners without silicosis whereas in miners with silicosis arsenic exposure increased the occurrence of AC at all levels of coexisting exposure to radon and quartz. These results indicate a cell type characteristic specific effect of arsenic in the development of lung cancer.