Long-Term Exposure to Air Pollution and COVID-19 Vaccine Antibody Response in a General Population Cohort (COVICAT Study, Catalonia)

The COVID-19 cohort in Catalonia (the COVICAT study) evaluates the health impact of the COVID-19 pandemic on the population in Catalonia, Spain, and builds on five preexisting adult cohort studies.⁹ We limited this analysis to the vaccine response in the largest cohort within COVICAT, the GCAT-Genomes for Life cohort.³¹ The GCAT cohort started participant recruitment in 2015 and includes middle-aged (40–65 years of age at baseline) residents of Catalonia. Most enrolled participants were blood donors invited through a public agency, the Blood and Tissue Bank (BST). The last prepandemic follow-up was done in 2018–2019 ($n=\mathrm{9,308}$). In early summer 2020 post-lockdown, participants completed an online COVICAT questionnaire or responded to a computer-assisted telephone questionnaire; a random sample of participants also provided blood samples up to mid-November 2020.^9,32 Residential address at the time of the prepandemic questionnaire was geocoded.

We recontacted eligible participants a year later in spring 2021 after COVID-19 vaccine administration began in Spain (Figure S1). Participants were asked to respond to a questionnaire (online or via telephone) and provide a blood sample. Blood sampling in 2021 was offered to 2,404 participants, including all participants with a seropositive or undetermined serostatus in 2020 (response rate, $N=507$, 44.0%) and to a random sample of seronegative participants in 2020 (response rate, $N=575$, 46.2%). People were aware of their 2020 serology results. A total of 1,090 participants provided blood samples and completed the questionnaire. We excluded from this analysis individuals who were not vaccinated ($n=120$), those vaccinated with Janssen COVID-19 vaccine due to the small number of people receiving this (one dose) vaccine ($n=16$), one participant $<40$ years of age, and 26 participants with incomplete information (questionnaire 2021, serology 2020, or air pollution data). This left 927 participants for this analysis.

Ethical approval was obtained by the ethics committees at the Hospital Universitari Germans Trias i Pujol (CEI no. PI-20-182) and the Parc de Salut Mar (CEIM-PS MAR, no. 2020/9,307/I). All participants provided informed consent and had consented to be recontacted during the first follow-up.

We linked participants’ prepandemic addresses to estimates of long-term exposure to the following air pollutants: particulate matter with an aerodynamic diameter of $\le 2.5\mathrm{\mu }\mathrm{m}$ (${\mathrm{PM}}_{2.5}$), black carbon (BC), nitrogen dioxide (${\mathrm{NO}}_2$), and ozone (${\mathrm{O}}_3$). Air pollution estimates for the period 2018–2019 were based on models developed by the Effects of Low-Level Air Pollution: A Study in Europe (ELAPSE) project (http://www.elapseproject.eu/), which have been described in detail.³³ We applied Europe-wide hybrid land-use regression models incorporating air pollution monitoring data, satellite observations, dispersion model estimates, land use, and traffic variables as predictors. We based the models for ${\mathrm{PM}}_{2.5}$, ${\mathrm{NO}}_2$, and ${\mathrm{O}}_3$ (warm season) on 2010 measurements in the AirBase database that is maintained by the European Environmental Agency. We used European Study of Cohorts for Air Pollution Effects (ESCAPE) monitoring data to develop models of BC.³⁴ The model was evaluated using 5-fold hold-out validation in random subsets of the monitoring data stratified by type of measurement and region of Europe. Models explained 66% of measured spatial variation for ${\mathrm{PM}}_{2.5}$ in annual average concentrations in hold-out validation; the corresponding fraction for BC was 52%; for ${\mathrm{NO}}_2$_, 58%; and for ${\mathrm{O}}_3$, 63%. Participants were assigned the annual average 2010 concentration based on predicted surfaces ($100\times 100\mathrm{m}$) from the ELAPSE model. We then applied a temporal correction to estimate exposures for the years 2018 and 2019 following protocols for temporal extrapolation developed in the ESCAPE project. Although 2010 models have been validated, we did not have validation against measurement data from 2018–2019. We used daily time-series data from the official routine monitoring network and calculated the ratios between the 2018–2019 period and 2010 for ${\mathrm{NO}}_2$, nitrogen oxides (${\mathrm{NO}}_{\mathrm{x}}$), ${\mathrm{PM}}_{2.5}$, and ${\mathrm{O}}_3$. Because BC is not measured at routine monitoring stations, we used ${\mathrm{NO}}_{\mathrm{x}}$ to temporally adjust for BC values given that it is a primary combustion pollutant from traffic emissions with similar pollutant behavior. We used the average of 2018–2019 levels for each pollutant as our main exposure metrics.

Blood samples collected at both follow-up periods (2020 and 2021) were processed within 24 h of collection and frozen, and anti-SARS-CoV-2 antibody levels in plasma were analyzed in one batch at the ISGlobal Immunology laboratory in Barcelona. The levels [median fluorescence intensity (MFI)] of IgG, IgM, and IgA were assessed by high-throughput multiplex quantitative suspension array technology against a panel of five SARS-CoV-2 antigens: the spike full length protein (S) and the receptor-binding domain (RBD) (both fused with C-terminal 6xHis and StrepTag purification sequences and purified from supernatant of lentiviral-transduced CHO-S cells cultured under a fed-batch system³⁵), the subregion S2 (purchased from SinoBiological), the nucleocapsid (N) full length (FL), and the specific C-terminal (Ct) region (both expressed in E. coli and His tag-purified).³⁶ Assay performance was previously established as 100% specificity and 95.78% sensitivity for seropositivity 14 d after symptoms onset (15). Antigen-coupled microspheres were added to a 384-well $\mathrm{\mu }\text{Clear}$ flat bottom plate (Greiner Bio-One) in multiplex (2,000 microspheres per analyte per well) in a volume of $90\mathrm{\mu }\mathrm{L}$ of Luminex buffer [1% bovine serum albumin (BSA), 0.05% Tween 20, 0.05% sodium azide in phosphate-buffered saline (PBS)] using the 384 channels Integra Viaflo semi-automatic device (96/384, 384 channel pipette). Hyperimmune pools were used as positive controls prepared at 2-fold, eight serial dilutions from 1:12.5. Prepandemic samples were used as negative controls to estimate the cutoff of seropositivity. Ten microliters of each dilution of the positive control, negative controls, and test samples (prediluted 1:50 in 96 round-bottom well plates), were added to a 384-well plate using an Assist Plus Integra device with a 12-channel Voyager pipette (final test sample dilution of 1:500 for all isotypes, and a second dilution at 1:5,000 for IgG to assess the response to S proteins in vaccinated participants, avoiding signal saturation). To quantify IgM and IgA, test samples and controls were pretreated with antihuman IgG (Gullsorb) at 1:10 dilution, to avoid IgG interferences. Technical blanks consisting of Luminex buffer and microspheres without samples were added in 4 wells to control for nonspecific signals. Plates were incubated for 1 h at room temperature in agitation (Titramax 1000) at $900\text{\hspace{0.17em}rpm}$ and protected from light. Then, the plates were washed three times with $200\mathrm{\mu }\mathrm{L}/\text{well}$ of PBS-T (0.05% Tween 20 in PBS), using BioTek 405 TS (384-well format). Twenty-five microliters of goat antihuman IgG-phycoerythrin (PE) (GTIG-001; Moss Bio) diluted 1:400, goat antihuman IgA-PE (GTIA-001; Moss Bio) 1:200, or goat antihuman IgM-PE (GTIM-001; Moss Bio) 1:200 in Luminex buffer were added to each well and incubated for 30 min as before. Plates were washed and microspheres resuspended with $80\mathrm{\mu }\mathrm{L}$ of Luminex buffer, covered with an adhesive film, and sonicated 20 s on a sonicator bath platform before acquisition on the Flexmap three-dimensional reader. At least 50 microspheres per analyte and per well were acquired, and MFIs were reported for each isotype–antigen combination. Assay positivity cutoffs specific for each isotype–antigen combination were calculated as 10 to the mean plus 3 standard deviations (SDs) of ${\log }_{10}$-transformed MFI of 128 prepandemic controls. Results were defined as undetermined when the MFI levels for a given isotype–antigen combination were between the positivity threshold and an upper limit at 10 to the mean plus 4.5 SD of the ${\log }_{10}$-transformed MFIs of prepandemic samples, and no other isotype–antigen combination was above the positivity cutoff. We defined overall serostatus by isotype and by antigen. Results for IgM are informative for the evaluation of short-term effects in antibody response and although measured in the whole study population are shown only for those participants having one dose at the time of the 2021 study visit, sampled within 1 month post vaccination.

We retrieved data on the number of doses, date of administration, and trade names of vaccines for each participant from the electronic health records of the Epidemiological Surveillance Emergency Service of Catalonia of the Department of Health. Participants had received the following vaccines: Comirnaty (BNT162b2, mRNA; BioNTech-Pfizer), Spikevax (mRNA-1273; Moderna), and Vaxzevria (ChAdOx1 nCoV-19; AstraZeneca).

We detected a heterologous prime-boost approach in 11 people (Vaxzevria as a first shot followed by Comirnaty). Thus, participants’ vaccine type is categorized according to the type of their first dose.

We used a two-part strategy to identify participants infected with SARS-CoV-2 prior to the 2021 study visit a) positive viral detection test (polymerase chain reaction or antigen test) prior to sample collection in 2021, self-reported in study questionnaires, or identified through record linkage with the SARS-CoV-2 test registry from the Epidemiological Surveillance Emergency Service of Catalonia of the Department of Health,³⁷ and b) seropositivity based on our antibody data using the following criteria: seropositivity in the prevaccination 2020 serology sample, or seropositivity to N-antigen in 2021 sample, given that the available vaccines do not contain or produce N-antigen.

Information on basic characteristics (age, sex, and educational level) was available from earlier contacts and verified in the COVICAT questionnaire (available in Spanish at http://www.gcatbiobank.org/media/upload/arxius/COVICAT/encuesta%20COVICAT.pdf). In this analysis, we used self-reported information on several variables, including educational level, smoking, alcohol consumption, medical history diet, symptoms related to COVID-19, height, and weight. Medical history included prior diagnosis of any chronic disease, asking for a list of several major diseases that required a visit to the doctor or medical treatment in the last 6 months, such as cardiovascular (hypertension, heart attack, angina pectoris), respiratory (asthma, chronic obstructive pulmonary disease), diabetes, kidney, immune-related (autoimmune diseases, HIV, or other immunodeficiency), digestive, or gynecological diseases, as well as cancer, mental health diseases (anxiety, depression, or other diseases), and addictions, along with an open question on any other disease. Diet was assessed with the 14-point Mediterranean Diet Adherence Screener (MEDAS) used in the Prevención con Dieta Mediterránea (PREDIMED) study that has been validated against a classic food frequency questionnaire.³⁸ Self-reported symptoms related to COVID-19 included fever, cough, dyspnea, fatigue, headache, muscle/joint pain, loss of odor/taste, nausea, vomiting, and rash. We collected changes in residential address from the prepandemic period through the 2021 follow-up. We linked prepandemic residential addresses to census tract-level deprivation index based on the 2011 census³⁹; the index uses six indicators, specifically, unemployment, manual work, temporary employment, insufficient education at $>16$ years of age, young age, and dwellings without access to internet. We also linked residential address with population density and degree of urbanization of the census tract of residence using information from the 2011 census.⁴⁰

In all analyses, we used antibody levels that were ${\log }_{10}$ transformed due to their skewed distribution. We applied linear regression models to assess the association between air pollution levels and the ${\log }_{10}$-transformed antibody levels, and results were expressed as percentage change in the geometric mean and 95% confidence intervals (CIs). Air pollutants were modeled continuously and estimates per interquartile range (IQR) of each pollutant were reported. Antibody responses to vaccination measured through IgM levels included only participants sampled within 1 month post first dose vaccination, whereas analyses on IgG/IgA levels included all participants irrespective of sampling time post vaccination. We considered the following variables as potential confounders: age (continuous), sex (male, female), highest attained educational level (primary or less, secondary, university), and socioeconomic status according to area of residency (in quantiles). Time since last vaccination ($<31$, 31–60, 61–90, 91–120, $>120$ d), type of vaccine and number of doses are strongly related to vaccine antibody response, and we also included those variables in all models. We finally adjusted for factors that have been associated with response to COVID-19 or other vaccines, including smoking, diet, prior chronic diseases, and mental health.^23–26 We expected an attenuated effect among those with previous SARS-CoV-2 infection and therefore defined a priori stratified analysis by infection status. Evaluation of effect modification by infection status through likelihood ratio tests indicated $p_{\text{Interaction}}<0.1$ for all air pollutants and the S and S2 spike antigens. In the infected strata, we adjusted models for severity of the infection based on self-reported symptoms and from hospital records (0 symptoms, 1–3 symptoms, $\ge 4$ symptoms, hospitalized). Participants with missing covariates were excluded from the complete-case analysis models. We used generalized additive models to explore the relationship between days since vaccination and IgG antibody levels among participants without prior infection according to air pollution levels (with low defined as below the median vs. high defined as above the median of the distribution for each pollutant). We performed all statistical analyses using Stata/SE (version 16; StataCorp LLC.).

The flow chart in Figure S1 shows the participants contacted, those recruited, and those excluded. Of the 1,090 participants recruited who gave blood samples, 927 participants were included in this analysis (Figure S1). The mean age was 57 y (range: 44–72 y), and 58% were female (Table 1). We evaluated differences in air pollution exposure, sociodemographic, and clinical variables between the 1,090 persons who gave blood samples in 2021 (participants) and the 1,314 who were contacted but did not attend the study visit in 2021 (nonparticipants). The average ${\mathrm{NO}}_2$ levels for participants and nonparticipants was similar ($36.8{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ vs. $36.6{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$, respectively) and for ${\mathrm{PM}}_{2.5}$ were nearly identical ($16.6{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$) (Table S1). Participants were slightly older (55.8 years of age) than nonparticipants (55.3 years of age). There were minor, nonstatistically significant differences for socioeconomic status/education, sex, and prior infection based on serological tests in 2020 (Table S1).

Pollution levels at residence had the following correlations: 0.8 between ${\mathrm{NO}}_2$ and ${\mathrm{PM}}_{2.5}$, $\sim 0.8$ for ${\mathrm{NO}}_2$ and BC and 0.7 for ${\mathrm{PM}}_{2.5}$ and BC, and $\sim -0.8$ for ${\mathrm{NO}}_2$ and ${\mathrm{PM}}_{2.5}$ with ${\mathrm{O}}_3$ (Table 2). Mean exposure during 2018–2019 in the study population was $35.1{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ for ${\mathrm{NO}}_2$ and $16.4{\mathrm{\mu }\mathrm{g}/\mathrm{m}}^3$ for ${\mathrm{PM}}_{2.5}$ (Table 3).

Among vaccinated participants, 319 (34.4%) had one dose and 605 (65.6%) had completed two doses (Table 1). The first dose was either Comirnaty–Pfizer/Bionetch (45.5%) or Vaxzevria–AstraZeneca- (42.3%), and a smaller fraction of people were vaccinated with Spikevax–Moderna (12.2%). Vaxzevria was administered as a first dose followed by Comirnaty in 11 people. Median time since the last vaccination was 28 d (IQR: 14–54 d, minimum: 1 d, maximum: 160 d). Median time between the first and second dose was 21 d for Comirnaty (IQR: 21–22 d, minimum: 14 d, maximum: 42 d), 28 d for Spikevax (IQR: 28–29 d, minimum: 27 d, maximum: 36 d), and 83 d for Vaxzevria (IQR: 78–95 d, minimum: 63 d, maximum: 121 d).

There was no clear association of air pollutant levels with the type of vaccine administered (Table 3). Recipients of the AstraZeneca vaccine had only slightly higher average exposure levels of ${\mathrm{NO}}_2$ ($35.6\mathrm{\mu }\mathrm{g}/\mathrm{m}³$) and ${\mathrm{PM}}_{2.5}$ ($16.5\mathrm{\mu }\mathrm{g}/\mathrm{m}³$) compared with those receiving the Pfizer ($34.9\mathrm{\mu }\mathrm{g}/\mathrm{m}³$ and $16.3\mathrm{\mu }\mathrm{g}/\mathrm{m}³$, respectively) and Moderna vaccines ($34.3\mathrm{\mu }\mathrm{g}/\mathrm{m}³$ and $16.2\mathrm{\mu }\mathrm{g}/\mathrm{m}³$, respectively).

Among participants without prior infection, antibody responses to vaccination measured through IgM levels (participants within 1 month post first dose vaccination) and IgG levels (any time post vaccination, all participants) were negatively associated with long-term air pollution; no associations were observed for IgA (Table 4). The decrease in IgM levels was between 5% and 14% per IQR increase in ${\mathrm{NO}}_2$ and ${\mathrm{PM}}_{2.5}$, and it was statistically significant for most antigens. The decrease in IgM was slightly smaller for BC, whereas no associations were observed for ${\mathrm{O}}_3$. For IgG, IQR increases in ${\mathrm{NO}}_2$, ${\mathrm{PM}}_{2.5}$_, and BC were associated with statistically significant decreases of the S and S2 antigens. An IQR increase in ${\mathrm{PM}}_{2.5}$ was associated with an $\sim 8\%–10\%$ decrease in IgG levels; for RBD, percentage change $=-8.1$ (95% CI: $-15.9$, 0.4); for S, percentage change $=-9.9$ (95% CI: $-16.2$, $-3.1$); and for S2, percentage change $=-8.4$ (95% CI: $-13.5$, 3.0). Similar associations were observed for BC, whereas for ${\mathrm{NO}}_2$ the decrease in IgG antibody response was slightly smaller (Table 4). An inverse association for IgG was observed for exposure to ${\mathrm{O}}_3$. Estimates for IgG and air pollutants adjusted only for age and sex are shown in Table S2. Associations adjusted for the same variables as in the Table 4 models but also including lifestyle factors (smoking and alcohol consumption), the diet MEDAS score, prior mental health diseases, and prior medical history of chronic diseases gave similar estimates (Table S3) as those shown in Table 4 for more limited adjustments. The direction of the air pollution–antibody response associations was similar in women and men, but they were stronger in men (Table S4). There were no statistically significant interactions ($p_{\text{Interaction}}<0.1$) observed by sex for ${\mathrm{NO}}_2$, ${\mathrm{PM}}_{2.5}$_, or BC, although there were for ${\mathrm{O}}_3$ exposure. Finally, we evaluated whether the association of air pollution with vaccine response depended on type of vaccine (Table S5). Results were similar for the three main vaccines administered.

Among participants without prior infection, differences in IgG antibody response between high and low air pollution exposures (defined as above or below the median exposure for ${\mathrm{NO}}_2$ and ${\mathrm{PM}}_{2.5}$) were small but persisted over time since vaccination after both the first (Figure 1) and the second dose (Figure 2). After the first dose, the peak in IgG RBD and S responses occurred later, and lower levels were observed in people exposed to high levels of ${\mathrm{PM}}_{2.5}$ and ${\mathrm{NO}}_2$ (Figure 1). After the second dose, we identified a faster decrease of IgG levels with time in noninfected participants with high air pollution exposure (Figure 2).

Among infected participants, there were no associations observed between air pollution and antibody response to vaccines for any of the air pollutants examined (Table 5). The largest decrease in IgG levels per IQR increase in ${\mathrm{PM}}_{2.5}$ was observed for RBD ($-8.7\%$) but results were not statistically significant.