Introduction

As we are all well aware, the COVID-19 pandemic, caused by the SARS-CoV-2 virus, challenged the social and economic stability of the world since emerging in late 2019 [1]. It motivated the rapid expansion and development of diagnostic methods for identifying infection, including nucleic acid, antigen, and antibody detection [2]. The deployment of safe and effective vaccines in late 2020 changed the course of the pandemic by reducing the severity of disease and suggested the possibility that immunity would be maintained by a future endemic status [3, 4]. The virus has subsequently mutated to produce variants of concern (VOCs) that have caused waves of breakthrough infections [5]. A third dose of BNT162b2, the mRNA vaccine developed by Pfizer, has been shown to improve immunity to the omicron variant BA.1 by increasing the neutralizing capability of circulating IgGs [6, 7]. Data from a study of a fourth BNT162b2 vaccine dose in Israel suggests that the doubly boosted immune response lasts less than or equal to two months in individuals over 60 years of age [8]. As governments worldwide have begun to embrace an endemic future of COVID-19 it will be important to track immune responses to VOCs in vaccinated individuals to inform decisions about mask wearing, booster doses for healthy and immunocompromised individuals, and development of VOC-specific vaccines [9].

Cross-reactive antibodies against common human coronaviruses (hCoVs), SARS-CoV-1, and MERS are increased after COVID-19 infection [10], and to a lesser extent vaccination [11, 12].The ability to evaluate cross-reactive antibody binding to hCoV antigens, SARS-CoV-1, MERS, and SARS-CoV-2 antigens with a single test could enable further exploration into the relationship between cross-reactivity of the antibodies and disease outcome.

Influenza surveillance around the world currently relies on either RT-PCR or rapid tests. Rapid tests only indicate the presence of Influenza A or B, not subtype or individual strain. RT-PCR can test for subtype and strain but requires primers for each strain that is being tested for [13]. Tracking which strains of influenza are circulating or are closely cross-reactive is important for vaccine development and outbreak prediction [14].

For several years, we have worked to develop array-based methods based on Arrayed Imaging Reflectometry (AIR) for rapidly assessing immunity to upper respiratory viruses, and have published results that agree with established Luminex and ELISA immunoassay platforms [15, 16]. AIR relies on the quantitative perturbation of a near-perfect antireflective condition on a silicon/silicon dioxide chip as targets bind to the arrayed probes. The change in intensity of light reflected off the AIR chip and captured by a charge coupled device (CCD) is quantified and cross-calibrated to a change in thickness due to a label-free binding event [17, 18]. As an imaging technique, AIR can quantify binding of more than 100 targets on an array independently and simultaneously [19]. Our laboratory recently developed an AIR array suitable for monitoring the immune response to SARS-CoV-2, and demonstrated multiplex data consistent with single-analyte ELISA [20]. Here, we have employed a prototype commercial version of AIR, which requires only a few microliters of each serum sample or whole blood, with a significantly expanded array of recombinant proteins (antigens) from upper respiratory viruses. We demonstrate the use of this platform to screen for antibodies against antigens from 34 human respiratory viruses including SARS-CoV-2, SARS-CoV-1, MERS, common hCoVs, and pandemic and seasonal influenza strains of type A and B. This is a longitudinal study of adult subjects who were infected with SARS-CoV-2, received a Pfizer/BioNTech (BNT162b2) or Moderna (MRNA-1273) vaccine, or experienced both infection and vaccination. Vaccinated samples were collected prior to the first dose, one week after the first dose, one week after the second dose. In some individuals we were able to obtain samples monthly for up to 6 months after the second vaccine dose, and two weeks after a breakthrough infection.

This 34-plex array generates 561 combinations of probes, which creates an opportunity to evaluate relationships between antibody responses between probe antigens. Linear regression provides a single numerical measure (coefficient of determination, R² value) of the influence that one variable has on the other. This analysis helps us quantify the impact that the Pfizer and Moderna vaccines, which consist of mRNA encoding the full-length SARS-CoV-2 spike protein [21, 22], had on the immune response to other SARS-CoV-2 antigens, including the D614G mutation, as well as related SARS-CoV-1, MERS, and hCoV proteins. This could help identify cross-reactive binding of antibodies to new variants or identify when vaccines need to be modified to improve immune response. Although the arrays tested here are commercially produced, the technology allows for easy modification to add new Covid-19 or influenza VOCs as they arise.

Materials and methods

Results and discussion

An early 16-plex version of the ARVA was used for initial experiments (Table 1). The samples beginning with ‘SN’ were drawn from subjects who had had an unknown respiratory illness at some time during late 2019 and early 2020. This was early in the pandemic when tests were in short supply, and only SN028 had a PCR-confirmed case of Covid-19. The ‘HD’ samples were acquired at least 14 days after illness from convalescent COVID-19 patients via the University of Rochester Medical Center’s Healthy Donor protocol.

The heatmap in Fig 2 shows antibody binding onto each antigen in the array as a thickness change relative to the control chip (build). We can see varied responses to influenza and common hCoV antigens across individuals, which is expected due to personal health histories and propensity to receive annual flu shots. When looking at the SARS-CoV-2 antigens we see that many, but not all the convalescent patients have antibodies in their serum that bind to N-protein and spike proteins including the receptor binding domain (RBD) as well as full length S1+S2 and individual subunits S1 and S2. This is consistent with work performed on these samples in our laboratory with the preliminary, non-automated version of a SARS-CoV-2 AIR array [23], and with ELISA results on these samples acquired by an independent laboratory [31]. The low SARS-CoV-2 antibody responses in some convalescent patients can be explained by some patients self-reporting illness without a positive PCR test. Assuming that all “SN” samples were uninfected with SARS-CoV-2 with the exception of SN028 (known PCR positive), these values excluding those from SN028 were averaged together into the uninfected group in S3 Fig. All HD patients plus SN028 were grouped into the convalescent group. When comparing average antibody build across all convalescent patients to all uninfected patients, there is significantly more antibody binding onto all SARS-CoV-2 antigens in convalescent serum (unpaired, 2-tailed t-test assuming unequal variance). There is also a significant increase in cross-reactive antibody binding to the SARS-CoV-1 S and OC43 S1S2, but not to MERS S1. The significant increase in cross-reactive antibodies against OC43 after infection with COVID-19 is consistent with the ELISA results from the same convalescent samples [33]. Interestingly, that same group provided evidence that hCoV memory B cells against OC43 were activated in response to infection with SARS-CoV-2.

To understand the quantitative range of the assay we selected five high responding convalescent samples and performed serial serum dilutions of 1:20, 1:40, 1:80, 1:160, and 1:320 (Fig 3). Here the data are reported as simply thickness (Å) where negative control information is not subtracted but is instead plotted independently. The standard deviation of serum samples was calculated from replicate probe spots on each array [21]. Positive samples were determined as those significantly different from control by one-tailed two-sample t-test with p < 0.01. The sample dilutions were all analytically well-behaved and titrated to zero. While the creation of a standard curve to convert thickness change to antibody concentration was not within the scope of this study, these data suggest that it would be possible to measure protein concentration using the ZIVA platform.

An expanded 34-plex ARVA kit with antigens listed in Table 2 and full length names of influenza in S2 Table was used for longitudinal studies of vaccinated individuals (designated VN) in order to track immune response over time. The array generates a large amount of data, making a heatmap the most effective way to gain an overview of the range of responses per sample (Fig 4). The hue indicates thickness change relative to a control chip per antigen on the array per sample, with darker blue indicating a larger thickness change as more antibody bound to antigen.

As expected based on recent studies of time-dependent antibody titers following vaccination [32–35], the thickness change on the SARS-CoV-2 antigens is dynamic. Looking at VN02, we see that this subject had an increase in SARS-CoV-2 spike antibodies two weeks after the first vaccine dose, which is consistent with other vaccine studies [36]. This pattern holds for other subjects (VN- 01, 03, 04, 05, 17, 25, 27) who also donated serum two weeks after the first dose. Additionally, these observations fit with general knowledge of the adaptive immune response. Our results show that the multiplex ARVA array is able to differentiate subjects who had a covid-19 infection (9, 11, 15, and 16) from those who only received the vaccine, by looking at the thickness change caused by antibodies binding to the N-protein. The N-protein is located inside of the viral envelope and interacts with the viral RNA [37]. Since the vaccine mRNA only encodes the spike protein, there isn’t an immune response to N-protein in naïve vaccinated individuals. A few of these subjects (VN- 21, 23, and 30) were taking immunosuppressive medications during the course of the study. While VN23 and VN30 seem to mount immune responses to the vaccine, the same cannot be said of VN21. This is a very small sample size, but suggests that AIR technology and the ARVA array could be used to screen immunosuppressed patients to determine if they need extra vaccine doses, antibody therapy, or if they need to take extra precaution to avoid infection. Likewise, such screening could bring peace of mind to immunosuppressed patients who do generate antibodies in response to vaccination, especially since antibody titer correlates with neutralizing capability and protection against severe disease [38].

Antibody response due to vaccination was assessed by grouping samples acquired before vaccination and samples acquired 1–2 weeks after the second vaccine dose (S4 Fig). There are significant increases in antibodies binding all SARS-CoV-2 antigens except the N-protein as expected, since the vaccines encode the full-length spike protein, and significant cross-reactive antibodies to SARS-CoV-1 RBD and S1, MERS-CoV-S1, and 229E. Recent work profiling cross-reactive antibodies to related coronaviruses after vaccination of naïve individuals found significant antibody cross-reactivity to SARS-CoV-1 and MERS-CoV spike proteins [12]. Cross-reactivity of antibodies toward common cold hCoVs following vaccination against SARS-CoV-2 has also been observed [11]. It seems that most of this cross-reactivity is recognizing conserved epitopes on the S2 subunit of these hCoVs [10, 11]. Our array only included the S1 subunit of NL63 and HKU1. This array included the full-length spike protein for OC43, but many of the responses were below the limit of detection for that protein on this version of the array.

The build onto the influenza antigens is fairly consistent for many subjects, while others have increases in antibody binding. VN09 received a flu shot prior to the fourth serum sample as noted in the figure. While the 2020–2021 quadrivalent flu shot was designed to activate an immune response against A/Guangdong-Maonan/SWL1536/2019 (H1N1) pdm09-like virus, A A/Hong Kong/2671/2019 (H3N2)-like virus, B/Washington/02/2019-like virus (B/Victoria lineage), and B/Phuket/3073/2013-like virus (B/Yamagata lineage) [39], this individual seemed to mount an immune response to other influenza antigens on the array, which is probably cross-reactivity among influenza strains. This may be the case for other subjects as well, but we lack information about flu shots or illness history to be able to confirm.

Even though information about influenza vaccination or infection during the time frame that these samples were collected is not available, there is still an opportunity to look for correlated immune responses to antigens across the array. Linear regression analysis provides an R² value indicating how much influence one variable has on the other. In this context we are using it to determine the correlation between samples, where a perfect correlation has R² equal to 1. Biologically, stronger correlations mean that higher amounts of antibodies binding to the epitope of one protein indicate more antibodies binding to the epitope of the second protein. Linear regression was only performed on thickness change data that was larger than the calculated limit of detection (LOD) for each protein on the array, and only the models with p-values < 0.01 were considered representative of a true relationship between variables. Of note, there were no negative correlations between any of the antigens on the array, indicating that vaccination with these SARS-CoV-2 vaccines doesn’t negatively affect the immune response to other upper respiratory viruses. As a positive control, we saw that the two pandemic strains of influenza isolated in California in the same year (Cal09) had a correlation of 1 (Fig 5A). This is expected because these two proteins have a nucleic acid sequence similarity of 99.9%. We were able to identify how strongly the subtypes, H1N1, H3N2, and influenza B, influenced each other. Proteins of the same subtype showed very strong correlation, with Influenza A Beijing H1N1 and Influenza A California H1N1 07–2009 showing the strongest relationship (R² = 0.96) (Fig 5B). Influenza A Texas H3N2 and Influenza A Hong Kong H3N2 2019 were highly correlated H3N2 subtypes (R² = 0.83) (Fig 5C). Influenza B Washington 2019 and Influenza B Massachusetts were the highly correlated influenza B strains with R² = 0.83 (Fig 5D)

There is a high positive correlation between antibody binding to the 229E full length spike protein and SARS2- S2 subunit (R² = 0.77) on our array (Fig 5E). For medical research, a correlation coefficient between 0.7 and 0.9 is considered a high positive correlation [40]. The alpha-coronavirus 229E has 31% amino acid sequence similarity for the spike protein of beta-coronaviruses SARS-CoV-2 and SARS-CoV-1, particularly in the conserved regions of the spike protein [41]. Titers of 229E antibodies have been shown to increase in convalescent and immunized populations, and antibody binding to 229E decreased 71% after depleting the serum with a monomeric SARS-CoV-2 S2 subdomain [11, 33].

There is also a high positive correlation between SARS-CoV-1 RBD and SARS-CoV-2 RBD (R² = 0.83) (Fig 5F). This is expected because these peptides share 74% amino acid sequence identity [42]. The D614G spike mutation became widespread after March 2020 [43]. Studies of the structure of the S1 and D614G S1 variant indicate that the D614G variant could have more antibody binding due to having a more flexible S1-S2 interface [44]. Antibody neutralization studies have shown slightly more antibody neutralization against the D614G variant [45]. On our array, D614G S1 protein is observed to have more binding overall than the wild-type S1 (S3 Fig), and the correlation between them is very high and positive (R² = 0.93) (Fig 5G). Many antibody-binding responses are the array were not strongly correlated. For example, SARS-CoV-2 RBD and Influenza A Beijing hemagglutinin have a negligible correlation (Fig 5H), which is expected because these viruses are from different phylum.

Comparing antibody responses across all of the proteins on our array from longitudinal samples enables comparisons of antibody duration and waning. Fig 6 presents SARS-CoV-2 RBD and influenza strain A/California/07/2009 hemagglutinin protein antibody levels over time. The individuals included in this analysis met the criteria of having a baseline sample collected immediately before receiving the vaccine and samples collected one week after first dose and one week after second dose of vaccine. Some individuals also went on to have monthly blood draws thereafter. Most of these individuals had not been infected with Covid-19 prior to this study, with the exception of VN11. Circulating SARS-CoV-2 RBD antibody levels in VN11 were initially higher than the other, naïve subjects shown in this figure, and immediately increased one week after receiving the first vaccine dose. The naïve subjects did not show an increase in circulating antibodies against SARS-CoV-2 RBD one week after the first vaccine dose, but had a robust response after the second dose, which is consistent with findings from clinical trials [34, 35]. The waning of SARS-CoV-2 RBD antibodies over time agrees with other studies showing circulating antibody levels returning close to baseline after 6 months or around 200 days [46]. At day 0, the variance in the number of antibodies present against influenza is greater than against SARS-CoV-2 RBD, but the overall level of influenza antibodies was significantly higher. This was expected because these strains of influenza infections and vaccines have been circulating for over a century, and most if not all individuals have been exposed to them throughout their lifetimes. In contrast, SARS-CoV-2 is a new strain of coronavirus. People who receive influenza vaccinations generally retain influenza-specific IgG antibodies for around two years [47], and cross-reactive antibodies to other strains will make this appear higher.

A subset of subjects who received a booster vaccine dose or had a breakthrough infection is shown in Fig 7. Circulating antibody levels increase from first to second dose, followed by a slow decline over the following six months, and a sharp increase after booster or breakthrough infection. Subjects VN04 and VN09 both had breakthrough infections. Although we expected to see antibodies against the N-protein after infection in both samples, this was not the case for VN04. A low humoral response of antibodies against the N-protein in a fraction of individuals after infection has been observed in other studies [48, 49], and younger people with asymptomatic or mild cases tend to produce lower antibody titers against the N-protein [50]. Subject VN09 (a female, under 40, with a mild case of COVID-19) is intriguing because she had a mild PCR-confirmed case of COVID-19 before receiving the vaccine, but didn’t have a robust antibody response to the N-protein until after breakthrough infection. Her second infection seemed to boost her initially low immune response towards the N-protein.

The small volume requirements of the ZIVA AIR system make it possible to do a whole blood assay using only a fingerstick volume of blood (3 μL). An initial test produced clean arrays that were comparable to serum in both background and target reflectivity (Fig 8A). The serum and the blood were from the same individual, who had a breakthrough case of COVID-19 confirmed by a positive commercial antigen test. The thickness change on cytC (negative control antigen) and FluA/HK/H3N2/19 was identical between the serum and whole blood. The serum was collected 5 weeks prior to the whole blood sample, so the decrease in circulating antibodies against SARS-CoV-2in the whole blood sample compared to serum is consistent with that seen in Figs 4, 6, 7 and 8B.

Conclusion

In this study, we have demonstrated that a prototype automated version of the label-free AIR sensor technology is able to profile human antibody responses to 34 antigens from upper respiratory viruses including SARS-CoV-2 from a small (<10 μL) sample of serum or whole blood. Of particular current interest, this approach proved useful in providing insight into immune responses following Covid-19 infection and vaccination. The results were consistent with previous low-multiplex work performed by our lab and with findings elsewhere. Given that AIR is expandable to include 100 or more probes, we can envision future work in which antigens from new SARS-CoV-2 variants are added to the array as they arise. For example, we know now that the booster dose can improve antibody neutralization of the Omicron variant [6, 7], but it is not currently known how well existing immunity will adapt to future variants. We expect our tool will be useful to help predict vaccinated immune responses to newly discovered VOCs before they become widespread. Multiplex AIR technology could also be useful for influenza and coronavirus surveillance and could ease further investigation into the relationship between antibody cross-reactivity and disease outcome. Finally, the ability to use a fingerstick quantity of blood to generate a real-time profile of circulating antibodies could be useful as a clinical diagnostic technique. Studies along these lines are currently in progress in our laboratory.

Supporting information

Acknowledgments

We thank Professor Matthew Brewer, Professor Mark Sangster, and Micah Wiesner for helpful discussions in immunology, virology, data analysis, and programming.