The role of ambient particle radioactivity in inflammation and endothelial function in an elderly cohort

Abstract

Background:

The mechanisms by which exposure to particulate matter might increase risk of cardiovascular morbidity and mortality are not fully known. However, few existing studies have investigated the potential role of particle radioactivity. Naturally occurring radionuclides attach to particulate matter and continue to release ionizing radiation after inhalation and deposition in the lungs. We hypothesize that exposure to particle radioactivity increases biomarkers of inflammation.

Methods:

Our repeated-measures study included 752 men in the greater Boston area. We estimated regional particle radioactivity as a daily spatial average of gross beta concentrations from five monitors in the study area. We used linear mixed-effects regression models to estimate short- and medium-term associations between particle radioactivity and biomarkers of inflammation and endothelial dysfunction, with and without adjustment for additional particulate air pollutants.

Results:

We observed associations between particle radioactivity on C-reactive protein (CRP), intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion moleducle-1 (VCAM-1), but no associations with fibrinogen. An interquartile range width increase in mean 7-day particle radioactivity (1.2 10⁻⁴ Bq/m³) was associated with a 4.9% increase in CRP (95% CI: 0.077, 9.9), a 2.8% increase in ICAM-1 (95% CI: 1.4, 4.2), and a 4.3% increase in VCAM-1 (95% CI: 2.5, 6.1). The main effects of particle radioactivity remained similar after adjustment in most cases. We also obtained similar effect estimates in a sensitivity analysis applying a robust causal model.

Conclusions:

Regional particle radioactivity is positively associated with inflammatory biomarkers, indicating a potential pathway for radiation-induced cardiovascular effects.

INTRODUCTION

It is well established that exposure to particulate air pollution (PM) is associated with increased risk of cardiovascular morbidity and mortality.^1–3 One hypothesized pathway for these observed effects is by the induction of systemic oxidative stress and inflammation.^1,3 Short- and medium-term exposures to PM with an aerodynamic diameter less than 2.5 micrometers (PM_2.5) have been associated with increases in pro-inflammatory cytokines, including interleukin (IL)-1β, IL-6, and TNF-α; acute-phase proteins, including C-reactive protein (CRP) and fibrinogen;^4–9 and biomarkers for endothelial function, intercellular adhesion molecule-1 (ICAM-1), and vascular cell adhesion molecule-1 (VCAM-1).^9–12 However, the specific properties of PM that initiate oxidative stress and the inflammatory response are still not fully understood.

A previously overlooked characteristic of PM in the epidemiology literature is its role as a vector for radioactive isotopes. Previous research has evaluated the radiometric composition of particles using measures of gross beta radiation, gross alpha radiation, and specific radionuclides. These studies have demonstrated that alpha and beta radiation measured from particles is primarily due to the decay of long-lived radon progeny.^13–18 Radon, a gas formed by the decay of radioactive materials in the earth’s crust, is the primary source of naturally occurring radioactivity exposure in the U.S. After formation, radon emanates from the soil, diffuses into air, and decays to radioactive progeny. These progeny quickly form charged clusters and attach to aerosols.^19–22 Spatial and temporal variations in particle radioactivity are therefore driven by the generation of radon and the transport of attached radon progeny through the atmosphere. Factors impacting temporal variations in particle radioactivity include meteorologic variables that influence radon emanation and exhalation (e.g., temperature, relative humidity, and pressure), aerosol dispersion and deposition (e.g., temperature, wind speed, and precipitation), and the origin and movement of air masses.^{13–15,23–27} Spatial variation is primarily driven by regional meteorology and radon potential of the soil.^15,27–29

Attached radon progeny continue to decay after inhalation and deposition, releasing alpha, beta, and gamma radiation into the lungs.^30,31 Previous research has demonstrated that exposure to radon and alpha particles can induce pulmonary inflammation and oxidative damage in animal and in vitro models.^32–35 Therefore, it is important to evaluate whether particle radioactivity could cause inflammation and endothelial dysfunction. We hypothesized that short-term (7 day) and medium-term (14, 21, and 28 day) exposures to particle radioactivity would be associated with changes in CRP, fibrinogen, ICAM-1, and VCAM-1 in a closed longitudinal cohort of elderly men.

METHODS

Study Population

We studied participants in the Normative Aging Study (NAS), a longitudinal study of aging established in 1963 by the U.S. Veterans Administration.³⁶ Initial enrollment was limited to community-dwelling men from the Greater Boston area who were free of any known chronic medical conditions (Figure 1). Participants reported to the study center after an overnight fast and smoking abstinence, once every 3 to 5 years. Each visit included a questionnaire, physical exam, and laboratory collection. Measurements were conducted between 1999 and 2013. We excluded visits where participants had a CRP level above 10 mg/L, as an elevated CRP level often indicates an infection, and any participant who did not live in New England. This study was approved by the Institutional Review Boards of the participating institutions.

Figure 1:

Map of the study region including the location of monitors and study participants. Residential locations have been masked by the addition of random noise to preserve confidentiality.

Note: NAS, Normative Aging Study.

Particle Radioactivity

We used gross beta radiation (Bq/m³) to represent total particle radioactivity. Gross beta radiation is measured as part of the EPA’s RadNet program, which monitors background radiation levels in air, precipitation and drinking water around the U.S.³⁷ RadNet currently includes 140 continuously operating air sampling sites across the U.S, where total suspended solids (TSP) are collected by a high-volume air filter on a 4-inch diameter polyester fiber filter.³⁸ Filters are collected approximately twice per week and sent to the National Analytical Radiation Environmental Laboratory for measurement of gross beta radiation after a waiting period of approximately a week. While RadNet monitors do not follow a standardized collection and measurement schedule, collection duration and time between collection and measurement are similar across stations. For each monitor, we assigned all days within each sampling period to the beta concentration measured from that sample, creating a pseudo-daily time series. On days where one sample was completed and another sample began, we took the mean of the two measured concentrations.

Our study used gross beta concentrations from the following RadNet stations: Boston MA, Concord NH, Hartford CT, Providence RI, and Worcester MA. Not all stations had data for all dates included in the study period. Therefore, we calculated the regional beta concentration as a daily mean of the monitors’ standardized beta concentrations. This method prevents missing days from one monitor from adding false variability to the calculated mean daily value. Specifically, we: 1) calculated daily deviations from each monitor’s overall mean; 2) standardized the daily deviations by dividing by the monitor’s overall standard deviation; 3) calculated a daily regional standardized deviation by averaging the daily standardized deviations of all monitors; 4) multiplied the daily regional deviation value by the regional standard deviation calculated across all monitors; and 5) added back the overall mean of all monitors. Because beta concentrations are log-normally distributed, we performed all calculations on log-transformed values of beta and then transformed back to original units.

We calculated short- and medium-term exposure windows using moving averages for 7, 14, 21, and 28 days. These exposure windows were chosen to represent the period in which inflammatory responses are hypothesized to occur and to be comparable with previous literature.^39,40 We did not evaluate same-day exposures because beta concentrations are collected over a period of several days, and therefore lack the temporal resolution for this time window to be meaningful.

Air Pollution

Ambient particle concentrations were measured on the roof of the Countway Library of Medicine at Harvard Medical School, located in downtown Boston approximately 1 km from the study center. Subjects lived a median distance of 23.6 km from the monitor (Figure 1). Daily concentrations of PM_2.5 were collected by the Harvard Impactor Sampler.⁴¹ Particle number (PN) was monitored continuously using a condensation particle counter (CPC, TSI Inc. Model 3022a, Shoreview, MN). Black carbon (BC) was measured continuously using an aethalometer (Magee Scientific Corp, Model AE-21, Berkeley, CA). We calculated moving averages of 7, 14, 21, and 28 days for each pollutant.

Biomarkers

Participant biomarkers were measured at each visit. We measured plasma ICAM-1 and VCAM-1 using an enzyme-linked immunoabsorbent assay method (R&D Systems, Minneapolis, MN). We measured high sensitivity CRP concentrations using an immunoturbidimetric assay on the Hitachi 917 analyzer (Roche Diagnostics, Indianapolis, IN). We measured fibrinogen using MDA Fibriquick, a thrombin reagent.

Statistical Methods

We assessed potential associations between particle radioactivity and biomarkers using mixed effects models with a random intercept for each subject. Biomarkers were log-transformed to improve the normality of residuals. Our primary statistical model can be written as:

$$Y_{\mathit{\text{ij}}}={\beta }_1+{\beta }_2{X}_{\mathit{\text{PRij}}}+{\beta }_3{X}_{3ij}+\dots +{\beta }_p{X}_{\mathit{\text{pij}}}+b_i+{ϵ}_{\mathit{\text{ij}}}$$ (1)

where Y_ij is the biomarker logarithm for subject i at measurement j, β₂ is the main effect of particle radioactivity, β₃…β_p are the effects of measured covariates X₃ through X_p, b_i is a subject-specific intercept, and ϵ_ij is the within-subject error. The inclusion of a subject-specific intercept accounts for longitudinal correlation in measurements from the same subject and provides unbiased estimates of fixed effects, even when some subjects only have one measurement. Based on previous analyses of these outcomes in the NAS, we chose the following subject-specific covariates: continuous age, body mass index, and pack-years; categorical smoking status (current, former, and never); and an indicator for statin use and physician-diagnosed diabetes.^4,9,42 We controlled for daily mean temperature and humidity, measured at Boston Logan International Airport. We controlled for seasonality by including a sine and cosine term with a period of 365.24 days, which allows for a seasonal pattern following a cosine function with a period of one year and a variable amplitude and horizontal shift.⁴³

Previous studies have found an association between PN, BC, and PM_2.5 and our outcomes.^5,8–11 These are all properties of PM, and may be correlated with particle radioactivity. Therefore, we created two-pollutant models to ensure that any observed associations with particle radioactivity were not due to correlation. These two-pollutant models can be written as:

$$Y_{\mathit{\text{ij}}}={\beta }_1+{\beta }_2{X}_{\mathit{\text{PRij}}}+{\beta }_3{X}_{3\mathit{\text{ij}}}+{\beta }_4{X}_{4\mathit{\text{ij}}}+\dots +{\beta }_p{X}_{\mathit{\text{pij}}}+b_i+{ϵ}_{\mathit{\text{ij}}}$$ (2)

where Y_ij is the logarithm of biomarker for subject i at measurement j, β₂ is the main effect of particle radioactivity, β₃ is the effect of the co-exposure X₃ (either PN, BC, or PM_2.5), β₄…β_p are the effects of measured covariates X₄ through X_p, b_i is a subject-specific intercept, and ϵ_ij is the within-subject error. As secondary analyses, we evaluated the primary associations between PN, BC, and PM_2.5 and biomarkers, with and without adjustment for particle radioactivity. We used the same moving average for the co-exposure as for particle radioactivity in all co-exposure models. We also considered ozone as a possible correlated air pollutant. However, the correlation between daily ozone exposures, measured by the EPA’s Air Quality System, and particle radioactivity exposures was very low (r = 0.040); therefore we did not include it in our analyses.

We conducted several sensitivity analyses. First, we evaluated our assumption of linear effects of particle radioactivity by creating a series of generalized additive mixed models with penalized cubic regression splines for particle radioactivity to allow for non-linear associations, and compared the results to our primary models using Akaike Information Criterion (AIC) values.^44,45 Second, we assessed the impact of potential selection bias in our study by applying stabilized inverse probability weights (IPW), as healthier men are more likely to return for follow-up visits. These weights were calculated using logistic regression as the inverse probability of a second, third, and subsequent visits given each participant’s age, BMI, smoking status, alcohol consumption, cumulative pack-years, hypertension status, cholesterol, and diabetes.^46,47 Third, we re-ran our models limiting participants to those living within 100 km of a beta monitor and assigned exposures from the nearest RadNet monitor, rather than using regional values.

Finally, we used a Rubin causal model to check whether our estimated associations were biased by covariate imbalance across exposure levels.⁴⁸ We defined high particle radioactivity as particle radioactivity concentrations at or above the median. We used logistic regression to estimate propensity scores, which represent the probability of being exposed to high particle radioactivity given background covariates. We retained only visits with overlapping propensity scores between the exposed and unexposed; we discarded all outlying visits. We then matched high particle radioactivity days to low particle radioactivity days using propensity scores and a caliper width of 0.2.⁴⁹ This creates a dataset with groups of exposed and control days with similar distributions of covariates. Because we matched on dichotomized levels of particle radioactivity, we still used adjusted regression models to estimate continuous particle radioactivity associations in this smaller balanced dataset.^50–52

Associations are reported as the percent change in biomarker concentrations per an interquartile range width (IQRW) increase in the given exposure. The IQRWs for each exposure window are included in the supplemental material as eTable1. We performed all analyses with the statistical software R, version 3.5.3.⁵³ Linear mixed effect models were run using the nlme package, version 3.1.137; generalized additive models used the mgcv package, version 1.8.27.^54,55

RESULTS

Our study included 752 male participants, with a mean age of 75. Every participant had at least one measurement of CRP and VCAM-1; 751 participants had at least one ICAM-1 measurement and 654 had at least one fibrinogen measurement. 200 participants (27%) had one visit, 146 (19%) had two visits, 138 (18%) had three, 178 (24%) had four, 88 (12%) had five, and two (0.27%) had six. Tables 1 and 2 present descriptive characteristics of the study population.

Table 1:

Characteristics of the study population reported at study baseline (n = 752) and over all visits (n = 2,070).

Covariate	First Visit	All Visits
Age (years; Mean ± SD)	72 ± 7	75 ± 7
BMI (kg/m2; Mean ± SD)	28 ± 4.1	28 ± 4.1
Pack-years (Mean ± SD)	21 ± 25	19 ± 23
Diabetes (N, (%))	141 (19)	407 (20)
Smoking (N, (%))
Current	36 (4.8)	89 (4.3)
Former	486 (65)	1317 (64)
Never	230 (31)	664 (32)
Statin Use (N, (%))	266 (35)	1013 (49)
Environmental Covariates (Mean ± SD)
Daily mean temperature (°C)	13 ± 8.4	12 ± 8.8
Daily mean relative humidity (%)	68 ± 16	68 ± 16
Particle Radioactivity (10−4 Bq/m3)	2.7 ± 1.1	2.6 ± 1.1
PM2.5 (μg/m3)	11 ± 6.7	9.9 ± 6.4
BC (μg/m3)	0.94 ± 0.44	0.79 ± 0.40
PN (104 #/cm3)	3.0 ± 1.4	2.3 ± 1.2

Note: SD, standard deviation; BMI, body mass index; PM_2.5, particulate matter with an aerodynamic diameter less than 2.5 micrometers; BC, black carbon; PN, particle number.

Table 2:

Biomarker levels for the study population at study baseline (n = 752) and over all visits (n = 2,070), presented as median (25^th percentile, 75^th percentile).

Biomarker	First Visit	All Visits
Fibrinogen (mg/dL)	332 (296, 378)	328 (285, 376)
CRP (mg/L)	1.6 (0.76, 3.0)	1.4 (0.71, 2.7)
ICAM-1 (ng/dL)	288 (247, 332)	275 (237, 323)
VCAM-1 (ng/dL)	987 (811, 1210)	978 (792, 1220)

Note: CRP, C-reactive protein; ICAM-1, intercellular adhesion molecule-1; VCAM-1, vascular cell adhesion molecule-1.

Particle radioactivity showed a moderate correlation with PM_2.5 (r=0.47) and BC (r = 0.25) and a low correlation with PN (r = 0.089) (eTable 2). Distributions of gross beta concentrations were similar across all monitors (eTable 3 and eFigure 1). Final daily regional beta concentrations showed a high correlation (> 0.8) with all five monitors.

We observed strong positive associations between particle radioactivity and ICAM-1 and VCAM-1 at all moving averages of exposure. An IQRW increase in mean 7-day particle radioactivity (0.12 mBq/m³) was associated with a 2.8% increase in ICAM-1 (95% CI: 1.4, 4.2) and a 4.3% increase in VCAM-1 (95% CI: 2.5, 6.1). The same IQRW increase in 7-day particle radioactivity was also associated with a 4.9% increase in CRP (95% CI: 0.077, 9.9), but this association was lower for longer exposure windows. We did not see any primary associations between particle radioactivity and fibrinogen.

The main effects of particle radioactivity remained similar after adjustment for PN, BC, and PM_2.5 in most cases. The addition of PN to our models slightly increased the association between particle radioactivity and all biomarkers. Models that included BC showed a slightly larger association between particle radioactivity and CRP and VCAM-1, but a slightly attenuated association between particle radioactivity and ICAM-1. We observed the largest change in particle radioactivity associations for ICAM-1, where the addition of PM_2.5 to the model reduced associations between particle radioactivity and ICAM-1 from strongly positive to null. Figure 2 shows results from both the single- and two-pollutant models.

Figure 2:

Associations of an interquartile range width (IQRW) increase in particle radioactivity with biomarker outcomes, with and without adjustment for additional particulate air pollutants.

Note: CRP, C-reactive protein; ICAM-1, intercellular adhesion molecule-1; VCAM-1, vascular cell adhesion molecule-1; BC, black carbon; PM_2.5, particulate matter with an aerodynamic diameter less than 2.5 micrometers; PN, particle number.

In our secondary analyses, we found some positive associations between PM_2.5, BC, and PN and ICAM-1 and VCAM-1, but no associations with CRP or fibrinogen. Associations with PN remained similar after adjustment for particle radioactivity. The associations between PM_2.5 and ICAM-1 were strong and insensitive to adjustment for particle radioactivity, while the associations between PM_2.5 and VCAM-1 were attenuated after adjustment for particle radioactivity. The only primary association seen for BC was with ICAM-1 (Figure 3).

Figure 3:

Associations of an interquartile range width (IQRW) increase in particulate air pollution exposures (PM_2.5, BC and PN) with biomarker outcomes, with and without adjustment for particle radioactivity.

Note: CRP, C-reactive protein; ICAM-1, intercellular adhesion molecule-1; VCAM-1, vascular cell adhesion molecule-1; BC, black carbon; PM_2.5, particulate matter with an aerodynamic diameter less than 2.5 micrometers; PN, particle number.

Results from our non-linear generalized additive mixed models suggest that the association between particle radioactivity and biomarkers of inflammation is adequately represented as linear. Almost all non-linear models fit a cubic spline for particle radioactivity with less than two degrees of freedom, although there was some suggestion of non-linearity in our models for VCAM-1. The only non-linear model with improved AIC was the model of the association between a 28-day moving average of particle radioactivity and VCAM-1. A plot of the estimated effect estimate of particle radioactivity shows a fairly linear effect until very high levels of particle radioactivity, at which point the effect becomes negative (eFigure 2).

When we used inverse probability weighting to check for survival bias, associations between particle radioactivity and most biomarkers were similar to our primary models, although the effect of particle radioactivity on fibrinogen was more negative and the short-term effects of particle radioactivity on ICAM-1 were reduced (eFigure 3 and 4).

Limiting our analysis to participants matched to beta monitors within 100 km reduced our study population from 752 to 560 and the total visit number from 2070 to 1214. While the short-term association between particle radioactivity and CRP was reduced, the association between particle radioactivity and ICAM-1 and VCAM-1 increased (eFigure 5 and 6).

Unlike our primary dataset, the matched dataset used in our Rubin causal analysis was balanced across covariates⁵⁶ (eFigure 7 and 8). Descriptive statistics are provided in eTable 4 and 5. Although we used 7-day moving average exposures of PM_2.5, PN, and BC in our propensity score, other moving averages also showed an acceptable balance. The association between particle radioactivity and CRP was strengthened and other results were generally consistent with our primary analysis, although confidence intervals were wider due to reduced sample sizes. Notably, we did not see the same large drop in the particle radioactivity-ICAM-1 association like in our primary analysis. Figure 4 displays the primary and causal models with and without adjustment for PM_2.5 (full results in eFigure 9 and 10).

Figure 4:

Associations of an interquartile range width (IQRW) increase in particle radioactivity with biomarker outcomes, with and without adjustment for PM_2.5. Results are shown for our primary models (n = 2070) and causal models (n = 836), which used a smaller dataset constructed by propensity-score matching.

Note: CRP, C-reactive protein; ICAM-1, intercellular adhesion molecule-1; VCAM-1, vascular cell adhesion molecule-1; PR, particle radioactivity; PM_2.5, particulate matter with an aerodynamic diameter less than 2.5 micrometers.

DISCUSSION

In this study of 752 participants from the NAS cohort, we investigated whether short-term and medium-term exposures to particle radioactivity are associated with biomarkers of inflammation and endothelial dysfunction. We found positive associations between particle radioactivity and ICAM-1 and VCAM-1 across all exposure windows, as well as a short-term association between particle radioactivity and CRP. Associations between particle radioactivity and other particulate pollutants with ICAM-1 and VCAM-1 were similar across different exposure windows. This may be partially due to a strong correlation between the different moving averages, which we observed for all pollutants.

Effect estimates from this study are similar in magnitude to those from previous studies. An analysis by Bind et al. (2012) estimated associations between PM_2.5, PN, and BC and inflammatory biomarkers in the NAS cohort for 2000–2009, and similarly found that PN had the strongest associations with ICAM-1 and VCAM-1. A recent study by Li et al. (2018) examined the association between particle radioactivity and inflammatory biomarkers in a separate cohort from greater Boston and found some positive associations, although the association between particle radioactivity and ICAM-1 was negative.

Our effect estimates for particle radioactivity were robust to adjustment for BC and PN, suggesting that associations with particle radioactivity are independent of these particle characteristics. This makes sense given their different sources. Particle radioactivity, quantified in this study as gross beta concentrations, primarily represents the total concentration of long-lived radon progeny attached to particulate matter and is both long-lived and regional.^14,17,18,57 In contrast, PN and BC represent traffic-related air pollution, with PN primarily representing fresh emissions and BC representing both aged and fresh traffic particles.^4,9 These different particle measurements may be capturing separate toxic properties of PM that operate independently from each other.

It is more difficult to disentangle the roles of particle radioactivity and PM_2.5. In our standard models, we see a drop in the association between particle radioactivity and ICAM-1 after adjusting for PM_2.5, but no change in the association between particle radioactivity and VCAM-1. Conversely, the PM_2.5-ICAM-1 association remains similar after adjustment for particle radioactivity, but the PM_2.5-VCAM-1 association decreases after adjustment. These changes in association are not unexpected, given the strong correlation between PM_2.5 and particle radioactivity and the regional nature of both characteristics.

The distribution of PN, BC, and PM_2.5 were not balanced across participants exposed to high and low levels of particle radioactivity in our dataset. Regression analyses are unable to reliably adjust for covariates if the distribution of these covariates is substantially different across levels of exposure.^52,58 Therefore, we applied a Rubin causal analysis as a sensitivity analysis, using propensity scores to match individuals exposed to high particle radioactivity to individuals exposed to low particle radioactivity. This created a pseudo-randomized experiment where the distributions of measured covariates were similar between the two groups.^48,51,59 Results from this analysis should be better estimates of the independent causal effects of particle radioactivity on biomarkers for the range of particle radioactivity exposures included in our matched dataset. In these models, particle radioactivity associations again appear independent of PN or BC. While the association between particle radioactivity and ICAM-1 was still reduced after adjustment for PM_2.5, this decrease was not as extreme as in our primary models. The associations between PM_2.5 and both ICAM-1 and VCAM-1 were also reduced after adjustment for particle radioactivity. These results suggest that, while our primary models may have been influenced by covariate imbalance, it is still difficult to fully disentangle the independent associations of particle radioactivity and PM_2.5 with ICAM-1 and VCAM-1.

Gross beta concentrations in our study are similar to previous studies,^15,16,60,61 with an average daily beta concentration of 0.27 mBq/m³. These concentrations are lower than other naturally occurring radiation exposures. For example, the National Council on Radiation Protection and Measurements (NCRP) estimated average indoor radon concentrations in the U.S. to be 46 Bq/m³ and outdoor radon concentrations to be 15 Bq/m³.⁶² However, it is critical to distinguish between exposure and dose. The primary source of beta radiation in our study is ²¹⁰Pb, a long-lived radon progeny. Previous studies have found a strong correlation between beta and alpha radiation measured from particulate matter samples,^14,23,26 suggesting that high concentrations of beta radiation also indicate the presence of other radon progeny, such as the alpha-emitter ²¹⁰Po. Radon progeny are responsible for most of the total radiation dose from radon. The NCRP estimated the average annual effective dose from radon progeny in the U.S. to be 2.07 mSv, compared to only 0.05 mSv for radon gas.⁶² Therefore, attached radon progeny may contribute a substantial radiation dose even at very low exposures.

Acute radiation exposures can cause a variety of cancer and non-cancer outcomes,^63,64 but the potential risks associated with low-level radiation exposures are less understood. The best established is the effect of radon exposure on lung cancer, which has been demonstrated at both occupational and residential levels.^65–68 There is limited research on the effect of low-level radiation exposures on non-cancer health effects. One meta-analysis found excess risk for circulatory disease at low and moderate levels of ionizing radiation in occupational cohorts.⁶⁹ Radon concentrations were associated with chronic COPD mortality in the American Cancer Society Prevention Study II,⁷⁰ all-cause mortality among Medicare beneficiaries,⁷¹ and COPD admissions in an ecologic study in Spain.⁷² We also found that radon concentrations modified PM_2.5-mortality associations in a time-series study of 108 U.S. cities.⁷³ Recent cohort studies have found associations between particle radioactivity and blood pressure and lung function;^40,74 eTable 6 summarizes existing cohort studies of particle radioactivity and health outcomes.

There is some evidence for low-dose radiation effects in both animal and human models. Rats exposed to radon showed evidence of bronchoalveolar fluid (BALF) inflammation and IL-6 mRNA expression in both BALF and peripheral white blood cells.³² In a separate study, radon exposure resulted in a dose-dependent increase in 8-OHdG levels in lung tissue and an increase in reactive oxygen species (ROS) in BALF.³³ Human fibroblasts exposed to alpha-emitting particles had increases in ROS production,³⁴ and human pulmonary epithelial cells exposed to alpha-emitting particles demonstrated up-regulation of gene pathways that included those associated with inflammatory and respiratory diseases.³⁵

Our results add to the limited existing literature by demonstrating that particle radioactivity concentrations are associated with biomarkers of inflammation and endothelial dysfunction. This suggests a potential biologic mechanism for effects of particle radioactivity on non-cancer health outcomes.^40,73

There are some important limitations to our current study. Similar to other studies of the same cohort, our study uses air pollution measurements from a single site to estimate air pollution exposures.^4,40,42,75 This induces exposure error, which is likely larger for PN and BC than PM_2.5 because of the local nature of PN and BC.⁴ Any exposure error is non-differential and likely biases our results to the null, as has been shown in previous studies.⁷⁶

There are several sources of potential measurement error in our beta exposures. We used a regional average of particle gross beta radiation exposures as our primary exposure metric. This is appropriate because beta radiation is primarily attributable to long-lived radon progeny^15–18,77 and its spatial distribution is determined by regional meteorology and radon potential.^15,27–29 In our study, beta distributions were similar across all monitors and were strongly correlated with the regional beta concentration. The use of a regional exposure was also supported by the results of our sensitivity analysis using nearest-monitor exposures rather than regional concentrations, which were similar to our primary results. Any exposure error from the use of regional beta concentrations is likely to be non-differentially misclassified. Gross beta concentrations were also collected over several days rather than daily. This introduces additional non-differential exposure error and may further attenuate our results. Finally, particle radioactivity exposures were collected separately from PM_2.5, PN, and BC. Therefore, there is a possibility of residual confounding even after adjustment. Future studies should consider collecting all PM measurements, including particle radioactivity, from the same samples. This would allow us to more clearly understand the relationship between different properties of PM.

Additional validation of our exposure metrics is required. While previous studies using similar methods have found that most beta radiation is attributable to ²¹⁰Pb, none of these studies have used RadNet data.^15–18,77 The RadNet system also does not collect concentrations of individual radionuclides from the same samples used to measure gross beta concentrations. A future study could independently measure gross beta, gross alpha, and individual radionuclide concentrations from filters and use these concentrations to validate RadNet measurements and confirm the different sources of gross beta radiation.

Finally, it is possible that our study has unmeasured sources of confounding, like environmental tobacco smoke exposures and potential medical radiation exposures. However, these variables could only act as confounders if they are correlated to beta radiation exposures, which we think is unlikely. The study population is comprised entirely of elderly white males, and our results may not be generalizable to the general population.