Low dose environmental radon exposure and breast tumor gene expression

Our study includes participants from the Nurses’ Health Study (NHS) and the NHSII—both of which are large-scale prospective longitudinal cohorts of registered female nurses in the U.S. The NHS was established in 1976, when 121,701 women aged 30–55 years, completed and returned an initial questionnaire. In 1989, the NHSII was initiated, which enrolled 116,429 women, ages 25–42 years, who completed and returned an initial questionnaire. Participants have been followed via questionnaires mailed biennially to update information on exposure variables and ascertain outcomes [21]. The cumulative follow-up rates for NHS and NHSII were both greater than 90% [22]. Cases of invasive breast cancer were identified by participants’ responses to the biennial questionnaires from the start of follow-up (1976 NHS; 1989 NHSII) through 2012. Following a reported diagnosis, study investigators requested permission to review the participant’s medical records to confirm the diagnosis. Over 99% of breast cancer diagnosis were confirmed upon medical record review. For deceased participants, we contact next-of-kin for record review permission or link to cancer registries. With permission, we linked 96% of breast cancer cases to relevant medical records. The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T. H. Chan School of Public Health, and those of participating registries as required.

We collected participants’ residential addresses from biennial questionnaires updated since initial enrollment in NHS (1976) and NHSII (1989). Residential addresses were geocoded and spatially linked to the Laurence Berkeley National Laboratory U.S. county-level indoor radon exposure model using a geographic information system (GIS) ArcMap 10.3.1 (Esri, Redlands, CA) [17]. The radon exposure model was estimated using a Bayesian mixed-effect regression to predict annual average county-level radon concentrations derived from the short-term Environmental Protection Agency (EPA)/State Residential Radon Survey (SRRS) and long-term National Residential Radon Survey (NRRS) [23, 24]. We calculated radon concentration as a time-varying cumulative average for each participant, where radon concentrations from years prior to diagnosis were averaged and updated biennially until a breast cancer diagnosis [17]. Radon concentration was expressed as pCi/L based on the scale established by the EPA (1pCi/L = 37 Bq/m³ international unit).

We began to collect archived formalin-fixed paraffin-embedded breast cancer blocks for participants with primary incident breast cancer since 1993. Formalin fixed paraffin-embedded (FFPE) tumor blocks were retrieved from the pathology departments of treating hospitals [25]. We were able to obtain tissue microarrays (TMAs) from 5561 NHS and NHSII participants (Dana Farber / Harvard Cancer Center Tissue Microarray Core Facility, Boston, MA) [25, 26]. Prioritizing women with existing genetic and circulating biomarker measurements, we were able to perform tissue gene expression microarrays among 954 breast cancer cases. Participant characteristics were similar among breast cancer cases with and without gene expression measurement [27]. We extracted RNA from multiple cores of 1 or 1.5 mm taken from tumor (n = 1–3 cores) and normal-adjacent (n = 3–5 cores) tissues using the Qiagen AllPrep RNA isolation kit for FFPE. Normal-adjacent tissues were obtained > 1 cm away from the tumor edge. A detailed protocol has been published previously [28‐30] (microarray data accession number: GSE115577). In brief, we measured gene expression using Affymetrix Glue Grant Human Transcriptome Array 3.0 (hGlue 3.0) for samples processed in 2012–2014 and Human Transcriptome Array 2.0 (HTA 2.0) microarray chips (Affymetrix, Santa Clara, CA, USA) for samples processed in 2015–2018. All microarrays were scanned with the GeneChip® Scanner 3000 7G (Affymetrix, Santa Clara, CA, USA). We normalized gene expression data using robust multi-array average (RMA; Affymetrix Power Tools (ATP)), and performed sample quality control using Affymetrix Power Tools probeset summarization based metrics [28, 29]. A total of 2160 FFPE samples were assayed. We excluded samples that failed quality control steps (n = 440), technical replicates (n = 139), samples with unknown IDs (n = 4). In total, 1577 samples (882 tumor tissues and 695 normal-adjacent tissues) from 954 invasive breast cancer cases passed quality control. For genes that were mapped by multiple probes, the most variable probe was selected to represent the gene. The hGlue 3.0 platform included 18,102 genes with unique Entrez ID and the HTA 2.0 platform included 25,023 genes with unique Entrez ID. In our current analysis, we included 17,791 genes that were in common among the two platforms (70% overlapped). We controlled for known technical variabilities (batch / plate) using ComBat—an empirical Bayes method used to control for known batch effects [31]. We further removed genes with low expression (< 25th percentile).

We obtained information on potential risk factors for breast cancer via the biennial NHS and NHSII questionnaires (age, body mass index (BMI), hormone therapy (HT) use) or pathological reports (year of diagnosis). Median household income and region of residence were based on linking geocoded residential addresses with data from the U.S. Census Bureau [17]. ER status was determined via central review of breast tissue microarrays [25]. To calculate pack-year smoking, we multiplied smoking duration in years by packs of cigarettes smoked per day [32]. We restricted our analysis to women with complete exposure and covariate data, resulting in 943 breast cancer cases (874 tumor tissues, 687 normal-adjacent tissues; 1561 total samples; 618 tumor-normal-adjacent pairs).

One-hundred and eighty-four participants were exposed to high levels of cumulative radon (≥2 pCi/L), while 690 participants were exposed to low levels of cumulative radon (< 2 pCi/L). Mean age at breast cancer diagnosis was slightly younger for participants in the high radon group (mean ± SD = 57.6 ± 11.5 years) compared to the low radon group (mean ± SD = 59.7 ± 11.3 years) (Table 1). We observed comparable median year of diagnosis for both groups. In the high radon group, 30% of participants were postmenopausal and had not used HT, 32% of women were postmenopausal and used HT, while 38% were either premenopausal or did not report HT use. The proportion of women who used HT was comparable in the low radon group. Average BMI was 26.3 kg/m² (SD = 4.9) in the high exposure group and was 26.1 kg/m² (SD = 5.1) in the low radon group. Census tract-level median household income was higher in the low radon group (mean ± SD = 64,480 ± 25,064) compared to the high radon group (mean ± SD = 59,950 ± 18,200). In the high radon group, more than 50% of women lived in the Midwest region at the time of breast cancer diagnosis, while 16% of participants in the low radon group lived in this region. The distribution of radon concentrations in the U.S. is shown in Supplementary Figure 1 (ArcMap 10.4 (Esri, Redlands, CA)) and stratified by ER status and tissue types in Supplementary Figure 2.

In ER positive normal-adjacent samples, cumulative radon exposure was associated with differential expression of PLCH2 (negative association), a phospholipase gene involved in cellular proliferation (FDR < 5%). We did not find statistically significant associations between cumulative radon exposure and differential gene expression in ER-positive tumor, ER-negative tumor, or ER-negative normal-adjacent samples that met FDR < 5%. In Supplementary Table 1, we presented the top 10 differentially expressed genes sorted by nominal p value for tumor and normal-adjacent samples. We observed same signs of directionality, similar magnitudes of effect estimates and similar ranges of p-values when we modeled radon as a continuous exposure metric (Supplementary Table 1). Fig. 1 shows differentially expressed genes with relatively large effect size (i.e., ~ top 5%, which was defined as genes with log fold change larger than the average of the 2.5th and 97.5th percentile) and had a nominal p value less than 0.001 in tumor and normal-adjacent samples. In ER positive normal-adjacent samples, 7 of the 11 genes that met FDR < 15% also had large effect size (i.e., ~ top 5%).

As sensitivity analyses, we stratified women to never smokers and women smoked > 16 pack-years (population median of ever smokers). Among the top 10 most significant genes identified in the tumor analyses, we observed same signs of directionality and comparable magnitudes of effect estimate in the two stratified populations (Supplementary Table 2). We also observed comparable results when we restricted radon exposure 5 year prior to breast cancer diagnosis (Supplementary Table 1).

At FDR < 25%, we identified two distinct biological pathways significantly enriched in ER negative breast tumor (P38MAPK signaling pathway: up-regulated) and ER negative normal-adjacent samples (synthesis of phosphocholine: up-regulated) with higher cumulative radon exposure (Fig. 2; Supplementary Figure 3). Genes contributing to the core enrichment for P38MAPK signaling included CREB1, TGFBR1, MAPKAPK2, MKNK1, ATF2, ELK1, MEF2A, MAPKAPK5, HSPB1, RIPK1, MEF2C, RAC1, CDC42, MAPK14, TGFB3, TGFB2, STAT1, and HMGN1, while genes contributing to the core enrichment for phosphocholine synthesis included SLC44A4, PCYT1A, LPIN3, CHKB, SLC44A2, and CHKA (Fig. 3). We summarize our key findings for the single gene analysis and pathway analysis in Supplementary Table 3.