SARS-CoV-2 serology in 4000 health care and administrative staff across seven sites in Lombardy, Italy

Lombardy is the Italian region most affected by COVID-19. We tested the presence of plasma anti-SARS-CoV-2 IgG antibodies in 3985 employees across 7 healthcare facilities in areas of Lombardy with different exposure to the SARS-CoV-2 epidemic. Subjects filled a questionnaire to self-report on COVID-19 symptoms, comorbidities, smoking, regular or remote working, and the exposure to COVID-infected individuals. We show that the number of individuals exposed to the virus depended on the geographical location of the facility, ranging between 3 and 43%, consistent with the spatial variation of COVID-19 incidence in Lombardy, and correlated with family interactions. We observed a higher prevalence of females than males positive for IgG, however the level of antibodies was similar, suggesting a comparable magnitude of the anti-spike antibody response. IgG positivity among smokers was lower (7.4% vs 13.5%) although without difference in IgG plasma levels. We observed 11.9% of IgG positive asymptomatic individuals and another 23.1% with one or two symptoms. Interestingly, among the IgG positive population, 81.2% of subjects with anosmia/dysgeusia and fever were SARS-CoV-2 infected, indicating that these symptoms are strongly associated to COVID-19. In conclusion, the frequency of IgG positivity and SARS-CoV-2 infection is dependent on the geographical exposure to the virus and primarily to family rather than hospital exposure.

Frequency of IgG positivity and correlation with geographical area. Anti-S1 and anti-S2 SARS-CoV-2 IgG specific antibodies were detected via an indirect chemiluminescence immunoassay. According to kit manufacturer, the test discriminates among negative (< 12AU/mL; with 3.8 as the limit of IgG detection), equivocal (12.0-15.0 AU/mL) and positive (> 15.0 AU/mL) subjects. Of 3985 enrolled subjects, 3462 (87%) were negative, 76 (2%) equivocal and 447 (11%) positive (which together with equivocal summed up to 523 (13%)) (Supplementary Table S1). We decided to group together equivocal and positive subjects as they behaved very similarly both as separate groups or as grouped together (Positive-IgG ≥ 12.0 AU/ml) in subsequent analyses (for instance in correlation with the number of symptoms, as shown in Supplementary Fig. S2). We cannot exclude inclusion of false positives.
The number of individuals exposed to the virus reflected the geographical area where the healthcare facility was located (Fig. 1). Two hospitals based in Bergamo (Humanitas Gavazzeni and Castelli) were the most affected as between 35 and 43% of the subjects was IgG positive. In the other sites the prevalence of IgG antibodies ranged between 3% (HMC) and 9% (ICH) (Supplementary Table S2). We applied a hierarchical (multilevel) model adjusting for clustering of individuals within healthcare facilities when assessing overall relationships between locations, professional status and the analyzed variables. Humanitas Gavazzeni in Bergamo was significantly different from all other facilities except for Castelli in Bergamo (Table 1 and Supplementary Fig. S3, p < 0.0001, OR and 95% CI indicated in the table). Interestingly, HMC and HMD which had the lowest incidence of IgG positive individuals (3.0% and 3.8%, respectively), are located in Varese, which is one of the less COVID-19 affected province in Lombardy (https:// www. lomba rdian otizie. online/ coron avirus-casi-lomba rdia/). In agreement, the percentage of patients hospitalized for COVID-19 was higher in Bergamo at Gavazzeni (56%), intermediate in Milan at ICH (36%) and low in Varese at HMD (10%) (not shown).
Frequency of IgG positivity and correlation with gender, BMI and smoking. IgG positivity (≥ 12 AU/mL) was different between males (11.4%) and females (14%) ( Supplementary Table S1), and, as shown in Table 2, this difference was retained in the multilevel logistic analysis (OR 0.73; 95% CI 0.57-0.94, p = 0.016). Thus, the observed effect was associated to a difference in gender susceptibility to infection or differences in exposure or to the ability to mount an antibody response rather than to a bias of the recruitment itself.
Interestingly, 23.9% of subjects were smokers (n = 951) and IgG positivity among smokers was lower (7.4% vs 13.5%) (OR 0.45; 95% CI 0.34-0.60, p < 0.0001, Supplementary Fig. S4) confirming the trend effect observed with the multilevel logistic analysis (OR 0.48; 95% CI 0.38-0.61, p < 0.0001, Table 2). On the contrary, there was no statistically significant difference according to BMI ( www.nature.com/scientificreports/ Frequency of IgG positivity and correlation with age. We then evaluated whether there was a difference in the positivity to IgG according to age. We found that there was a higher number of 31-50 years old IgG positive (≥ 12 AU/mL) individuals (Fig. 2a), but then when analyzing the frequency of positivity at the different age ranges we observed an age dependent reduction of IgG positive individuals (Fig. 2b). However, this age-dependency was primarily due to the female rather than the male population, particularly for subjects either young (20-40 yo) or older than 60 yo (test for heterogeneity of trend effect age between male and female  www.nature.com/scientificreports/ p = 0.0024 Fig. 2b). In older than 60 yo, IgG positivity dropped from 12% in males to 5% in females (Table 2 and Fig. 2b).
Frequency of IgG positivity and correlation with viral positivity at time of testing. Following the guidelines of the national health system, 516 out of 523 subjects with IgG ≥ 12 AU/ml underwent a nasopharyngeal swab for SARS-CoV-2 RNA viral detection. We found that 39 (7.6%) individuals resulted positive for viral RNA detection (Table 3). However, in 31 of these (79.5%), the subjects tested negative for at least one  Frequency of IgG positivity and correlation with professional status. We then evaluated the proportion of IgG positive individuals (≥ 12 AU/mL) across the different professional workers and found that there was no statistically significant difference when taking into account age, gender and their interaction (i.e. gender depending on age), smoking habits, BMI and location (

Frequency of IgG positivity and correlation with symptoms, comorbidities and vaccinations.
We then evaluated whether there was a correlation between the number and typology of self-reported symptoms and the frequency of antibody response both as number of individuals (Fig. 3a,c) and as percentage on the whole population (Fig. 3b,d). As shown in Fig. 3, when analyzing individuals deemed to be positive on the basis of the amount of IgG (≥ 12 AU/mL) we observed a two phases decay: first a similar frequency (b) or number (a) of subjects with 0 to 7 concomitant symptoms, and then a drop of individuals at higher number of concomitant symptoms. This distribution followed a sigmoidal, four parameter logistic curve whereby X is the number of symptoms (R 2 = 0.97). By contrast, in the population with IgG < 12 AU/mL we found a higher number (c) or frequency (d) of individuals with 0, 1 or 2 symptoms and the distribution was following an exponential curve (R 2 = 0.9975). The multilevel logistic analysis showed that, besides pneumonia, among the symptoms, fever, anosmia/dysgeusia (loss of smell or taste) and chest pain were those that best characterized the IgG positive population, particularly when collated (Table 6). Indeed, dividing the population according to the number of symptoms, IgG positivity increased from 11.9% in the absence of symptoms to 42.8% in the presence of 5 symptoms or more (Table 7), and the value of the AUC derived by the multivariable model was 79%, while the combination of just fever versus anosmia/dysgeusia had an AUC of 78% ( Supplementary Fig. S5). Fever with anosmia/dysgeusia, was present in 165 subjects (4.1%) of which 134 (81.2%) were associated with positive serology. Although sensitivity was 25.6% (134 with these symptoms /523 positive subjects), specificity was 99.1.% (3431 without these symptoms / 3820 negative subjects), yielding a positive likelihood ratio of 28.  The curve that best interpolated the data was a sigmoidal, four parameter logistic curve whereby X is the number of symptoms (R 2 = 0.97). (c, d) Distribution of the IgG negative individuals (IgG < 12 AU/mL) as number of individuals (c) or percentage of the population (d) versus the number of self-reported symptoms. The curve that best interpolated the data was exponential (R 2 = 0.9975). www.nature.com/scientificreports/ www.nature.com/scientificreports/ These results indicate that there are symptoms that best characterize the paucisymptomatic COVID-19 population and that when individuals present with fever and anosmia/dysgeusia they are likely to have SARS-CoV-2 infection.
When we analyzed if there was any correlation between IgG positivity and comorbidities or vaccination, we could not detect any and the odds ratios were all not statistically significant, however, the number of individuals with more than one comorbidity was low (8.5% of IgG positive individuals) making it difficult to draw major conclusions (Supplementary Tables S4a,b IgG plasma levels and population characteristics. The frequency of IgG positive (≥ 12 AU/mL) individuals clearly reflected the increased exposure to the virus across the analyzed geographical areas, correlated with COVID-19-related symptoms and showed a higher proportion of positivity within young females and non-smoker subjects. However, an advantage of using a quantitative assay for IgG testing is that it is also possible to assess the magnitude of the anti-spike antibody response. Hence, we analyzed the data also in relation to IgG plasma levels. As we took into account a possible temporal confounding factor of symptoms detection and serological test, we compared the IgG plasma levels of swab positive or swab negative individuals with that of the ascertained COVID-19 population (between the months of March and April). We could not detect any statistically significant difference among IgG ≥ 12 subjects with nasopharyngeal swab positive or negative and the COVID-19 populations, suggesting that the positivity to the swab does not correlate with a higher IgG plasma level (Fig. 4a). However, we observed that physicians (including anesthesiologists and physiotherapists) had a lower level of plasmatic IgG as confirmed by statistically different odds ratios (Supplementary Table S6  www.nature.com/scientificreports/ and odds ratios in Supplementary Fig. S6) this may reflect the use of PPE which might have reduced the viral load and hence also the development of an IgG immune response.
IgG plasma levels BMI, gender and smoking. As we observed a higher proportion of positivity (IgG ≥ 12 AU/mL) within females that may indicate either a higher exposure to the virus, a higher incidence of infection or a higher ability to mount an immune response, we evaluated whether this difference was paralleled by increased IgG plasma levels. We found out that there was no difference between IgG plasma levels of males versus females, suggesting a similar magnitude of the anti-spike antibody response (Fig. 4b, Supplementary  Table S6). However, when assessing a difference of IgG plasma levels across age ranges, there was a trend towards an increased level of plasma IgG towards the older age (p = 0.0002, total; p = 0.0085, females; p = 0.008, males) (Fig. 4c). Interestingly, while smoking seemed to inversely correlate with the frequency of IgG positive individuals, it did not have any effect on IgG plasma levels ( Supplementary Fig. S7a) even in relation to the number of smoked cigarettes per day ( Supplementary Fig. S7b and Supplementary Table S6). On the contrary, BMI, which was not influencing the frequency of IgG positive individuals, was affecting the plasma concentration of IgG: the higher the BMI, the higher the level of IgG (Supplementary Table S6, p = 0.0009).
IgG plasma levels and symptoms. As Table S7; CI are reported in the table for the different symptoms and combinations thereof, p < 0.014). By contrast, the level of IgG in the population with values < 12 AU/mL was constant, regardless of the number of symptoms (Fig. 5b). When we analyzed the levels of IgG in relation to symptoms, we found that those subjects reporting fever, anosmia/dysgeusia, or pneumonia had all significant odds ratios (Fig. 5c). In addition, subjects with fever, cough, muscular pain, asthenia, anosmia/dysgeusia, dyspnea, chest pain, tachycardia or pneumonia, they all had higher IgG levels than those without symptoms (Supplementary Table S7, p values and CI are reported in the table for the different symptoms). This confirms that presence of symptoms correlates with higher IgG levels as shown in Fig. 5a. Further, combination of fever with anosmia/dysgeusia and/or dyspnea characterized populations with higher IgG plasma levels (Supplementary  Table S7, p < 0.0001, CI are reported in the table for the different symptoms and combinations thereof), also when considering the whole population regardless of IgG positivity (Fig. 5d, p < 0.0001 in a LR test for global null hypothesis ordinal logistic analysis). When considering the level of IgG ≥ 12 the distribution of IgG plasma levels was very similar in subjects with these symptoms (Fig. 5e), however we found statistically significant differences in odds ratios when combining fever with anosmia/dysgeusia and fever with anosmia/dysgeusia and dyspnea (Fig. 5f, p = 0.0004 in a LR test for global null hypothesis ordinal logistic analysis). No statistically significant differences in IgG plasma levels were observed in relation to comorbidities. In steatosis/cirrhosis we had only two cases in which we found an inverse correlation but this should be confirmed on a higher number of individuals (p = 0.0211, CI 12.30-12.70, Supplementary Table S8). Moreover, we observed an interesting inverse correlation between pneumococcal vaccination and IgG plasma levels (p = 0.03, CI 15-29.40, Supplementary Table S9).

Discussion
Here we report a comprehensive analysis of nearly 4000 individuals from different healthcare facilities representative of dramatically different levels of SARS-CoV-2 exposure in Lombardy, the most affected region by COVID-19 in Italy. We observed a range of positivity which strongly correlated with the geographical area of viral exposure from 3% in the Varese area to 43% in the Bergamo area, which were respectively the least and most COVID-19-affected Lombardy provinces. The proportion of IgG positive females was higher than that of males in all of the analyzed sites. Although differences in exposure cannot be completely accounted for, the effect of gender remained even after adjustment for site of work and other factors associated with an increased risk of susceptibility. This is in contrast with a recent meta-analysis showing that in USA and Europe, but not in East Asia, the prevalence of IgG antibodies was moderately higher among the male health care workers 13 . However, we found a lower proportion of IgG positive individuals in females older than 60 years old in age-matched males, suggesting that females are more likely to be infected-or to induce an IgG response-when young, and less likely at ages higher than 60 yo. This is quite intriguing as males and females are equally affected by COVID-19, but males have a worse prognosis 14 , suggesting that seroprevalence cannot provide any correlation with the clinical outcome. Through the use of a quantitative antibody test of IgG we were also able to assess the magnitude of the anti-spike antibody response. We found that younger males (below 40 yo) displayed reduced IgG plasma levels than older males (from 41 yo onwards). This is in line with a recent report in COVID-19 patients showing that younger patients developed lower titers of IgG 15 and another one showing that higher IgG titers in the plasma of convalescent individuals were associated to older age and male sex 16 . Thus, also in the healthy population, younger males exposed to the virus develop a reduced antibody response. Moreover, the difference of IgG response between females and males in relation to age remains quite intriguing and we still have to understand its relevance with the higher incidence of COVID-19 in males 17 .Hence, it is very important when analyzing the serology to SARS-CoV-2 to take into account both age and sex.   www.nature.com/scientificreports/ Our study differs from the one reported by Sood and colleagues in Los Angeles County 18 as we used a quantitative antibody test and analyzed a large hospital population which ranged from healthcare professionals, researchers and administrative staff from 7 different facilities. Indirectly, we show that it is rather the environment than the hospital professional exposure which dictates the probability of contracting SARS-CoV-2 infection. Indeed, we show a higher percentage of IgG positive individuals in areas with a higher incidence of COVID-19 and among subjects who had been in contact with COVID-19 affected relatives. Further, we observed a similar proportion of IgG positive individuals among healthcare professionals functioning at hospital and administrative staff working from home. We were also able to pinpoint 11.9% of IgG positive individuals which were completely asymptomatic and another 23.1% of paucisymptomatic subjects with 1 or 2 symptoms. This percentage differs from that observed in two recent studies analyzing SARS-CoV-2 infection via viral testing across different population types 19,20 . Among healthcare workers, 57% of the subjects were asymptomatic/paucisymptomatic 19 while in a wider population in the Italian municipality of Vo' Euganeo, that underwent lockdown soon after detecting the first case of COVID-19, the percentage of asymptomatics was 42.5 20 . This suggests that there is still a good proportion of individuals that probably do not develop antibodies, or develop them at low/undetectable levels. This would be in line with recent studies showing that IgG positivity is short-lived 21 and relatives of COVID-19 patients carry T cells specific for SARS-CoV-2 but do not seroconvert, suggesting that also IgG positivity may not detect all individuals exposed to the virus 22 . However, if we expand our analysis to healthcare workers without symptoms but with IgG > 3.8 AU/mL, and not simply those IgG ≥ 12 AU/mL, which are considered 'positive' by the manufacturer of the test, the percentage of positivity raises to 40%. This may indicate that there may be different exposures to the virus which result in a wide range of levels of antibody production, including those between 3.8 and 12 AU/mL, but this could be a speculation. As in 283 subjects with IgG 3.8-12 AU/mL we performed a viral test, and it was negative, this indicates that they probably were exposed to the virus earlier and succeeded in eliminating it without a strong antibody response. Thus, it would be interesting to evaluate which serology test was used and which cut-off value was considered for determining IgG positivity in studies where IgG responses were negative, but T cell responses were present 22 . In addition, some of SARS-CoV-2 viral peptides activated T cells also from unexposed individuals suggesting that some cross-reactivity could occur, maybe in response to previous infections with common cold coronaviruses 23 .

Scientific Reports
Among the symptoms, those that characterized most the IgG positive population were fever and anosmia/ dysgeusia. Indeed, 81.2% of individuals presenting both anosmia/dysgeusia and fever resulted SARS-CoV-2 infected, indicating that these symptoms are strongly associated to COVID-19.
Selected vaccines such as BCG have been suggested to increase pathogen-agnostic off-target resistance to infectious agents 24 . However, a recent report showed no differences in incidence of COVID-19 in BCG vaccinated versus non vaccinated patient population 25 . In line with this, we did not observe a correlation between IgG positivity and BCG vaccination.
Intriguingly, we observed an inverse correlation between induction of an antibody response and smoking habit. This may indicate either a lower incidence of SARS-CoV-2 infection in smokers, or their inability to induce an immune response to the virus. The former is more likely as a large study (114,545 individuals) in Israel recently showed that the risk of SARS-CoV-2 positive testing in smoking individuals is reduced by half 26 . Similarly, Norden et al. reported an inverse association between national smoking rates and COVID-19 mortality 27 and this was confirmed in a meta-analysis showing an unexpectedly low prevalence of current smoking among patients with COVID-19 in China 28 . When analyzing the concentrations of IgG in plasma, we found no differences between males and females, nor in smoking habits. However, we observed that the presence of some symptoms such as fever, cough, muscular pain, asthenia, anosmia/dysgeusia and clinical manifestation such as pneumonia correlated with higher levels of IgG than in subjects without these symptoms. In addition, we found that the presence of comorbidities did not affect IgG plasma levels except for cirrhosis, but the frequency of individuals with multiple comorbidities was low (8.5% of IgG positive individuals with > 1 comorbidities). Subjects with cirrhosis had an inverse correlation with IgGs but the number of subjects was very low (2 out of 522) and this finding needs to be confirmed. The significance of this is unknown, but the finding that cirrhotic patients have higher risk of severe COVID 29 may suggest that this condition may affect the correct development of antibodies. Finally, we found that although very few subjects had anti-pneumoccocal vaccination, these had a reduced level of IgG, while there was no difference among those vaccinated with influenza vaccine. A recent study has shown an inverse correlation between nasopharyngeal swab positivity and flu and pneumococcus vaccinations suggesting protection in vaccinated individuals 30 . Another report on the Italian population, based on regional and not on individual data, has shown an inverse correlation between COVID-19 cases and coverage of flu vaccinations, suggesting again a protective role of flu vaccination 31 . It is possible that flu vaccination, by inhibiting concomitant infection, may result in a better prognosis of the SARS-CoV-2 infected individuals which becomes evident only when analyzing COVID-19 cases and not asymptomatic or paucisymptomatic individuals. Why individuals vaccinated against pneumococcus should produce less antibodies remains to be evaluated. We found that transmission seemed to occur primarily when in contact with infected family members. This may be due to the lack of use of PPE in the household. Indeed, it has to be considered that the use of PPE may have impacted on transmission as in the hospital it was implemented starting from February 1st 2020, consistently we observed a reduction of IgG plasma levels in physicians. This finding is in line with a recent report on testing the serology of healthcare workers and their families 32 .
In conclusion, our study has some limitations, as it is monitoring healthcare workers, and thus it does not reflect a comprehensive population as a good proportion of professionals are women and there is an underrepresentation of the elderly due to working age limit. However, we show that antibody testing can identify the individuals that were exposed to SARS-CoV-2 and is a powerful tool to retrospectively evaluate viral diffusion, even in asymptomatic individuals. Another limitation of the study is that we may have missed the population of subjects which do not develop an antibody response or that is reduced over time. As our study is ongoing we www.nature.com/scientificreports/ will assess the evolution of the IgG response over a planned follow up of one year, allowing for measuring the duration of the antibody response. Should a second wave of SARS-CoV-2 infection occur, the wide range of IgG serology in the different sites will be particularly valuable as it will allow us to assess the role of antibodies in viral protection. The results presented here suggest that hospital health care professionals, researchers and administrative staff can provide invaluable information to assess variables affecting the immune response to SARS-CoV-2 as a snapshot and during the follow-up.

Methods
Clinical study. The study has been approved by the international review board of Istituto Clinico Humanitas for all participating institutes (clinicaltrial.gov NCT04387929). Accrual was on a voluntary basis: it started on April 28th 2020 and more than 65% of employees (n = 3985) participated as of May 16th 2020. An international medical school (Humanitas University, Pieve Emanuele (MI)) and 6 different hospitals participated to the study: Istituto Clinico Humanitas (ICH), Rozzano (MI); Humanitas Gavazzeni, Bergamo; Humanitas Castelli, Bergamo; Humanitas Mater Domini (HMD), Castellanza (VA); Humanitas Medical Care (HMC), Varese (VA); Humanitas San Pio X, Milano (MI). All participants signed an informed consent and filled a questionnaire (Supplementary Data S1) before blood collection. We analyzed COVID-19-associated symptoms and clinical manifestations as tachycardia and/or pneumonia (the latter confirmed by TC). We considered "asymptomatics" subjects without any symptoms; "paucisymptomatics" individuals that developed 1 or 2 symptoms; "symptomatics" individuals with more than 3 symptoms. Clinical manifestations (tachycardia and pneumonia) were always associated with at least other 2 symptoms. In order to be tested, subjects had to fill in the questionnaire. Only after filling the questionnaire in its entirety, would the individuals be scheduled for blood sampling. Hence, we obtained a 100% complete data. None of the participants were enrolled at the time of symptoms. Thus, when the serological test was performed, they were either asymptomatics or the symptoms had disappeared. If they resulted positive for IgG (IgG ≥ 12 AU/mL) they underwent a nasopharyngeal swab for SARS-CoV-2 viral RNA detection to exclude being actively infected. We analyzed the mean of serology date versus the appearance of COVID-19 symptoms and found that they were tested around 43 ± 17 days after symptoms. IgG determination. For the determination of IgG anti SARS-CoV-2 the Liaison SARS-CoV-2 S1/S2 IgG assay (DiaSorin, Saluggia (VC), Italy) was used 33 . The method is an indirect chemiluminescence immunoassay for the determination of anti-S1 and anti-S2 specific antibodies. Intra-and inter-assay coefficient of variation are < 1.9% and < 3.7% respectively. In the datasheet of the test, in a cohort of 304 COVID-19 patients 97.5% of the samples above 15 AU/mL and 86% of those with IgG > 12 AU/mL resulted positive for neutralizing antibodies with a titer > 1:40 in the Plaque Reduction Neutralization Test (PRNT).
Statistics. This was a cross sectional study aimed at determining the frequency of plasma anti-SARS-CoV-2 IgG antibodies in nearly 4000 (3985) employees of 7 different healthcare facilities located in Lombardy.
The serological determination was offered to all the employees of the involved sites and the anticipated refusal rate was assumed to be 10-15%. The overall IgG positivity was assumed to be around 10-15% (between 400 and 600 subjects).
Primary endpoint was the number of test positive subjects (523 out of 3985). Given the study size, the study was able to estimate the overall positivity with a width of 95% confidence interval equal to 2% and the positivity for subgroups of at least 200 patients with a width of 95% confidence interval equal to 9%. No power analysis was performed to calculate the sample size.
As secondary endpoints we aimed to investigate the factors associated to plasma anti-SARS-CoV-2 IgG positivity.
Basic demographic characteristics, including age, sex, smoking status, comorbidities, previous vaccinations and symptoms were recorded and their association with IgG positivity was tabulated in contingency tables and described by means of relative and absolute frequencies.
The association between these characteristics and IgG positivity was assessed by univariate logistic model and measured with the odds ratio (OR) and 95% Confidence Interval (CI) for each factor. A multivariable analysis, was also performed in order to assess the effect of characteristics of interest after adjustment for the other variables.
Furthermore, since our data have clustered structures as patients were clustered within hospitals, a multilevel model was also used to analyse data. The binary outcome (positive vs. negative serological assessment) was chosen and analysis was approached with a generalized linear model with the logic link, which is the logistic regression model. In our multilevel analysis there was a patient level and an hospital level.
We produced two models. In the first one we want to assess the effect of role, gender, age, BMI and cigarette smoking on the risk of positive serology. We estimated a random intercept model for hospital profiling and treat the hospital effects as random effects only. The role of subjects was considered related to the hospital attributes, while the subject level attributes were age (grouped in > 60, ≤ 60 years) and gender, BMI (≤ 20, ≤ 25 www.nature.com/scientificreports/ and number of smoked cigarettes/day (no, ≤ 10, ≤ 20 > 20). BMI and cigarette smoking were considered as continuous variables. The others variables are all categorical. The interaction term of age*gender was included in the model. In the second one we explored the relationship between vaccinations, symptoms and serological positivity, after adjustment for the previous factors, still considering a hierarchical model. This was obtained adding the symptoms, defined as binary variables, to the previous models. Summary statistical measures were the odds ratio with relative confidence interval. SAS Glimmix procedure (SAS 9.4 release) was used for the purpose of analysis.
The same approach was used also for analysis of plasma levels. SAS Mixed procedure (SAS 9.4 release) was used for the purpose of analysis.
Kruskal-Wallis and non-parametric tests for trend (Cuzick's and Mantel-Haenzel test, when appropriate) have been used for multiple comparisons using Prism 8 Graphpad. For plasma levels, odds ratio and relative 95%CI based on Wald Test presented in the figures were calculated using logistic model applied to ordinal data.

Data availability
Data supporting the findings of this study are available within the paper and its Supplementary Information files. The dataset generated during and/or analysed during the current study is available from the corresponding author on reasonable request.