CorPuScula (Cor for coronary, Pu for pulmonary and S for sanguis or blood) is a longitudinal repeated-measures panel study conducted in 2000–2002 by the Institute and Outpatient Clinic for Occupational, Social and Environmental Medicine, Ludwig-Maximilians-University (LMU) in Munich, Germany. The study included 52 seniors and 50 children.

UFPs (≤100 nm in diameter) were measured as particle number concentration (PNC, particles/cm³). The measurements for the seniors were carried out with a condensation particle counter (TSI 3022A, lower particle size detection limit >7 nm) which was located approximately 3 km away from the care home. For children, a particle counter (TSI 3025A, lower particle size detection limit >3 nm) was placed on the school grounds, which is located on the Strehleranger road. The calendar daytime mean concentrations and individual hourly and lagged exposures prior to the examinations were calculated from hourly means.

Particulate matter PM₁₀ and PM_2.5 concentrations were measured using a Low Volume Sampler (LVS3; Leckel brand) located in the park of the Wohnstift Augustinum and the courtyard of the Strehleranger elementary school for both the seniors and the children's cohorts, respectively. Measures of fine PM were collected for 24 h but only the concentrations between 9:30 am until 9:30 pm were used as these were the main times in which the participants were active outdoors. The daytime mean concentrations were then calculated and used in the analyses as lags 0, 1 and 2.

Exposure to other air pollutants and meteorological parameters (i.e., temperature, water vapor pressure, precipitation and wind speed) were collected at the measuring station Johanneskirchen (L8.12). The station was located 9 km north and 7 km south of the senior home and the school, respectively. We obtained air pollution data from the station for nitrogen dioxide (NO₂, μg/m³) and ozone (O₃, μg/m³). Meteorological variables for the senior cohort were obtained from the measuring station of the Meteorological Institute of the LMU in Munich (MIM), which is approximately 8 km from the care home. For the children's cohort, we obtained the meteorological data from the MIM's measuring station Theresienstrasse (approximately 7 km away). The meteorological data included hourly mean values for air temperature (°C), water vapor pressure (hPa), precipitation (mm) and wind speed (m/s). We calculated the corresponding daytime mean values for the measured air pollution intervals (9:30 am to 9:30 pm).

Every other week, HRV was measured using a long-term ECG monitoring system from the company Customed (Custo Flash 200). This monitor used 2 channels alongside 5 electrodes. Raw ECG data was processed and cleaned prior to HRV analyses by an experienced doctor. Quality control was performed for the 10-min measurement interval, and extra systoles were identified and removed from the dataset.

For data collection, participants lay relaxed on the bed for at least 5 min before and during the entire 10 min of the ECG examinations. They were instructed not to speak or move and to breathe calmly and evenly during the examination. We extracted the time-domain measures standard deviation of normal-to-normal intervals (SDNN) representing both the SNS and PNS activity, and the root mean square of successive differences between normal heartbeats (RMSSD) that is more strongly influenced by the PNS than SDNN. The frequency-domain measures yielded information about the total power in the heart's rhythm explained by the variability of the normal-to-normal intervals. These include the low-frequency power (LF, 0.04–0.15 Hz) influenced by both SNS and the PNS, the high-frequency power (HF, 0.15–0.4 Hz) reflecting the PNS, and the frequency ratio of LF/HF to estimate the balance between SNS and PNS activity.

We analyzed associations between UFP, PM_2.5/PM₁₀ and HRV outcomes using linear mixed-effects regression models separately in adults and children. The time variable in the models was the date of measurement. The models also included the personal identification number as random effect. The autocorrelation structure included the time covariate and the personal identification number as a grouping factor within the observations that were assumed correlated. We used both unweighted (homoscedastic) as well as weighted (heteroscedastic) models where we allowed variances to differ by person, with a control for time trend (35). The weights were calculated as a ratio of the stratum standard errors to the within-group standard error. We assessed the adequacy of the heteroscedastic fit by re-examining plots of the standardized residuals vs. the fitted values by stratum to check if the standardized residuals had the same variability in each stratum. Also, the fit was tested with the ANOVA method. 95% confidence intervals (CIs) and p-values were computed using a Wald t-distribution approximation.

For each pollutant (UFP, PM₁₀ and PM_2.5), we calculated lagged exposures including both single-day lags (i.e., lag 0, lag 1 and lag 2) as well as individual lags for PM₁₀ (3, 12 and 24 h) and UFP (1, 3, 12 and 24 h).

The 5 outcomes (SDNN, RSMMD, LF, HF and LF/HF) were analyzed separately. In regression analyses, we first estimated the crude association between the lag periods for single pollutants with each outcome. Second, we estimated the same associations adjusting for possible confounders as found in previous studies (21, 36). To identify potential confounding and minimal sufficient adjustment sets, we constructed a directed acyclic graph (DAG) using the DAGitty program (37) separately for seniors and children. For seniors, the second model included age (years), sex (male/female), alcohol [did they drink on the day of the examination (yes/no), repeated measure], BMI (kg/m²), day of the week, precipitation, temperature, and water vapor pressure (see Supplementary Figure S2A). For children, the second model included age (years), sex (male/female), BMI (kg/m²), day of the week, precipitation, temperature, and water vapor pressure (see Supplementary Figure S2B). We did not adjust for multiple comparisons because the tests were highly correlated, thus violating the assumption of independence (see Supplementary Table S1 for correlations of outcome variables and the results section for correlations of exposure variables).

Meteorological variables (i.e., temperature, precipitation, and water vapor) were assessed for linearity using models without exposure variables, which resulted in modelling temperature and water vapor pressure with natural splines using 4 knots and precipitation with no splines. The best fitted lags for meteorological variables, which varied by outcome and group, were used (see Supplementary Table S1).

We then conducted two and multipollutant models by simultaneously adding the best fitted concentration (lag 0–2) of NO₂ and O₃ to the main models to examine the robustness of the effects after further controlling for the exposure (see Supplementary Table S2). Also, for models where the main exposure was UFP, we further adjusted for PM_2.5 and vice versa.

We also ran further sensitivity analyses to assess the robustness of our results. For the seniors, we investigated the inclusion of the intake of heart medication (yes/no if they were prescribed a heart meds at the time) before every examination. We also repeated analyses including season as an extra meteorological factor (4 seasons: winter, spring, summer, and autumn). For the children, we investigated the inclusion of having hay fever symptoms (yes/no) at the day of the examination as well as season.

For seniors and as examined in prior studies (38), we evaluated potential effect modification by inflammation markers (i.e., C-reactive protein as a continuous variable) and for blood pressure (hypertensive vs. not hypertensive). We conducted effect modification analyses through addition of a multiplicative interaction term between the exposure and the covariate of interest into the main model.

The models were estimated using restricted maximum likelihood (REML) and “nlminb” optimizer [R-package “state” (39)]. We employed the R functions varIdent() to define the variance function structure and lme() to fit the models [R-package “nlme” (35, 40)]. All statistical analyses were conducted in R version 4.3 (R Core Team) (39).

In UFP single pollutant models among seniors (Figure 1 and Supplementary Table S3), we observed an increase in SDNN at lags 0 and 1 day. In contrast, at short term exposures to PM₁₀, we observed decreases in SDNN (lags 12 and 24 h, and lag 1 day). No associations were found between UFP exposure and RMSSD. PM₁₀ (12 h lag) was associated with a clear decrease in RMSSD. In relation to the frequency-domain outcomes, UFP exposure was associated with consistent increases in both LF and HF powers at all lags, whereas for PM₁₀, only very acute exposure was associated with a transient increase in LF and HF. UFP exposures (lags 12 and 24 h and lags 1 and 2 days) were positively associated with the LF/HF ratio, but not PM₁₀ was not. No associations were observed between exposure to PM_2.5 and any outcomes.

In UFP models, we observed robust results after adjusting for NO₂ and O₃ throughout the 5 different outcomes. Results were also robust to adjustment for PM_2.5 with the exception of a more pronounced increase in SDNN and LF at lag 0 and lag 1, thus also affecting the estimates at lag 0 and 1 in multipollutant modelling (Figure 2 and Supplementary Table S4). In PM₁₀ models, results were robust to adjustment for NO₂ and O₃ in two and multipollutant modelling (Supplementary Figure S3).

Among children no clear associations were found between UFP and fine PM and the time-domain outcomes (Figure 3 and Supplementary Table S5). While point estimates for UFP were consistently below 0 for LF, all CIs were large and overlapped the null. In contrast, no patterns could be observed for UFP and HF. Similarly for PM₁₀, consistent but weak decreases were seen for LF but no pattern for HF. Short-term exposures to UFP (12 and 24 h) were positively associated with the LF/HF ratio. PM_2.5 and PM₁₀ were both negatively associated with the ratio at lag 2.

In UFP models, results were robust to adjustment for PM_2.5 and O₃ throughout the 5 different outcomes. Models were also robust to adjustment for NO₂ with the exception of a more pronounced increase in SDNN and LF/HF ratio at lag 12 and 24 h thus also affecting the estimates in multipollutant modelling (Figure 4 and Supplementary Table S6). In PM₁₀ models, results were robust to adjustment for NO₂ and O₃ in two and multipollutant modelling (Supplementary Figure S4).

Sensitivity analyses showed that UFP and PM₁₀ results were robust to adjustment for heart medication and season among seniors (Supplementary Figures S5, S6). Among children, model results for UFP and PM₁₀ were largely unchanged after adjusting for season and hay fever (Supplementary Figures S7, S8). There were no clear differences in the effects when it came to analyses by hypertension (Supplementary Figures S9, S10) or CRP (results not shown) among seniors in any of the exposures.

Among seniors, we observed positive associations between SDNN, LF, HF and LF/HF ratio and short-term exposure to UFP (hourly and daily lags) in contrast to decreases in SDNN and RMSSD after exposure to PM₁₀. Generally, the association between fine PM and all 5 HRV outcomes was less clear. For two and multipollutant modelling, we generally found UFP associations were robust to adjustment. Among children, we observed inconsistent associations between UFP and all HRV outcomes and mostly null associations between PM_2.5 and PM₁₀ and HRV outcomes. Results were robust for multipollutant modelling in general, however, adjustment for NO₂ enhanced some patterns for the association of UFP with SDNN and LF/HF (UFP lag 12 and 24 h).

In terms of short-term exposure to UFP, and in contrast to our findings, a recent meta-analysis by Zhang et al. (22) showed decreases in SDNN, RMSSD, LF and LF/HF within 6 h after exposure. Moreover, they found that daily lags on the same day or preceding days were not associated with HRV which is in contrast with our results of a positive association with SDNN, LF, HF and LF/HF. In terms of short-term exposure to fine PM, previous studies on the general population show inconsistent associations with HRV outcomes. Some observed decreases in time-domains (21, 41) and frequency-domain outcomes (20, 27, 42, 43), but also others observed increases in both domains (43–45) or no associations at all (46–48) as seen in our study.

Further, a very limited number of studies have looked at the exposure of ambient short-term UFP and fine PM on HRV outcomes among children. A study in California demonstrated that adolescents residing in neighborhoods characterized by higher concentrations of PM_2.5 had a decrease in HRV specifically in HF band in response to social stress (49). Saenen et al. (30), examined the association between fine PM exposure at home and school on SDNN, RMSSD, HF and LF among school children aged between 9 and 12 years old and found a decrease in LF with an increase in outdoor PM₁₀ exposure as well as decreases in RMSSD, HF and LF with an increase in indoor PM₁₀ exposure. In contrast, we observed no consistent and clear associations between fine PM exposure and HRV. Chen et al. (29) investigated the short-term effects of indoor UFP on children aged between 11 and 14 years old in schools in Beijing and found decreases in SDNN at lag 1 and decreases in SDNN, LF and HF at lag 2. Similarly, we found decreases in LF at most hourly and daily lags.

Our results for two and multipollutant models showed mostly robust results. A review on ambient air pollution and HRV conducted by Buteau and Goldberg (50) observed some decreases in HRV after the exposure to NO₂ ranging between 5 min to 5 days prior measurements. In our study and among seniors, the correlations between pollutants were high. Even though UFP was correlated with the co-pollutants, including NO₂, the robustness in our results indicates that there was no major collinearity influencing the reliability of the two and multi-pollutant models and that the associations seen between UFP and HRV were independent of fine PM, NO₂ and O₃. Moreover, among children, we observed a weaker correlation between UFP and the co-pollutants compared to the seniors. We adjusted our analyses for NO₂ as surrogate for traffic related air pollution (TRAP), thus reducing confounding of the air pollutant and HRV associations by sources of traffic. However, the potential for residual confounding by traffic remains due to the imperfect correlation between NO₂ and TRAP.. In terms of adjusting for O₃ and as observed by a recent systematic review, it is important to consider adjustment for O₃ since an association has been observed between short-term exposure to O₃ and a decrease in HRV (51). Moreover, in a study conducted on HRV in a two-pollutant model including PM_2.5 and O₃, it was concluded that PM_2.5 influenced HRV rather than O₃ (31). We obtained comparable conclusions where, after the adjustment of O₃ in two and multipollutant modelling, the results were robust leading to the affirmation of the effects of UFP and fine PM.

It is important to note that half of the seniors were on heart medication and the previous literature observed a decrease in HRV activity among people with heart disease (31, 52). This might explain the low SDNN in comparison to other studies in the general population. Upon adjusting for heart medication, the estimates were robust.

It is important to note that the placement of the pollutant measurement stations may lead to pollutant-specific exposure misclassification. In particular, among seniors, the fine PM measurement sampler was located in the park of the care home whereas for UFP, the counter was placed 3 km. away from the care home. For NO2 and O3, the measuring station was located 9 km north of the care home. We therefore expect a higher degree of exposure misclassification for UFP and NO2. However, simultaneous measurements of UFP at the care home showed that the temporal concentration changes at both locations were very similar. Among children, the exposure measurement does not account for exposures of children at home but only at school, potentially contributing to the inconsistent results in children.

Previous studies have demonstrated that exposure to UFP and fine PM causes both inflammation and oxidative stress that can contribute to cardiovascular disease and eventually cardiovascular mortality (53). A number of experiments have shown the biological mechanisms behind the association of UFP and fine PM with cardiovascular diseases at the molecular level. The very first contact of the particles is with the nasal and pulmonary receptors in the respiratory epithelium that will lead to a disturbance of the autonomic dysfunction and perturbations in the heart rate. Secondly, oxidative stress initiates an inflammatory response by the release of biologically active intermediates. This localized imbalance might also lead to an oxidative stress response and eventually lung inflammation. HRV is an indicator of the ANS that depends on the SNS and PNS for transferring information and where both systems interplay to regulate HRV (53). An imbalance in the function of the SNS and PNS could have opposite effects on HRV leading either to an increase or a decrease in HRV. More specifically, the effects of an exposure to UFP and fine PM could be mediated either through the SNS leading to decreases in HRV or through the PNS leading to increases in HRV (33). The SDNN and LF band reflect both SNS and PNS activities while the RMSSD and the HF band reflect PNS activity. We therefore consider the LF/HF as a ratio between SNS/PNS.

Among seniors we observed a general decrease in SDNN, RMSSD and HF after the exposure to PM₁₀ which would suggest a sympathetic response. In line with this, we also observed increases in the LF/HF ratio upon the exposure to UFP. This suggests greater SNS than PNS responses, which coincides with previous studies where greater SNS responses played an important role in sudden heart attacks or arrhythmia (22). The same response was observed among children with increase in the LF/HF ratio upon the exposure to UFP leading to the risks of arrythmia and heart disease (22).

This study has several strengths, firstly the ability to consider a range of exposure windows for all air pollutants and thus the assessment of temporal exposure patterns, both hourly and daily lags. Second the availability of data on co-pollutants allowed us to adjust for those in two and multipollutant modelling. Thirdly, the design of Corpuscula allowed for repeated-measures. While the number of participants was small, we were able to analyze a large number of observations per participant, which was enough to detect some associations among the seniors.

One main limitation is the small number of observations we had among children in terms of UFP measures and when adjusting for multi-pollutants. Because of the large number of missing values, the weighted models did not converge, and we had to report unweighted results, thus the confidence intervals were very wide and we were not able to detect clear associations. Another limitation was that we did not have personal exposure measures for NO₂ and O₃, since these co-pollutants were not measured at the school or the care-home but instead 7–8 km away from the actual research area. Thus, our ability to detect accurate associations for these pollutants in multipollutant modelling might have been reduced. A further limitation is that the study was carried among seniors and children, and the results may not be transferable to other age groups or to the general population. Nevertheless, these are populations traditionally thought to be particularly vulnerable to the effects of air pollution.