Our prospective observational study included 260 subjects, birth to ≤21 years of age, enrolled 1 April 2020, through 1 May 2022, at sites in Connecticut, United States and Cali, Colombia. After obtaining IRB approval (#21-004), subjects were enrolled in two cohorts: an experimental cohort (Cohort A) and a reference cohort (Cohort B). The experimental cohort was comprised of subjects who were positive for SARS-CoV-2 infection per antigen or PCR testing and hospitalized for COVID-19 symptoms (subgroup A1); hospitalized and meeting the Center for Disease Control’s criteria for MIS-C (subgroup A2); and non-hospitalized subjects testing positive for SARS-CoV-2 by antigen or PCR, (subgroup A3). The reference cohort was comprised of SARS-CoV-2 negative subjects who were: hospitalized with a diagnosis of Kawasaki disease (subgroup B1); hospitalized because of an acute viral infection (subgroup B2); or healthy controls undergoing routine ambulatory surgery (subgroup B3). To ensure correct subgroup assignment, cases were independently reviewed and adjudicated by three pediatric specialists. Demographic data, health history, current symptoms, and in-patient treatments and diagnostic testing results were collected at baseline. Subjects will be followed for up to 4 years via survey for health and SARS-CoV-2 vaccine status.

Patient saliva samples were stored at −80°C for at least 24 h prior to processing. The samples were placed in the biosafety cabinet and allowed to thaw at room temperature for approximately 20 min. Once thawed, the tubes were cleaned with alcohol wipes and centrifuged at 200 g for 15 min at 4°C. For samples to be analyzed with the grating-coupled fluorescent plasmonic biosensor chip, 500 μL of supernatant was aliquoted and 35 μL of protease inhibitor dissolved in 1 mL PBS was added (Pierce™ protease inhibitor tablet, Thermo Fisher Scientific, MA, USA) was added. For samples intended to be used in 16S RNA PacBio sequencing, 250 μL of supernatant was aliquoted and heat-inactivated for 30 min at 56°C. Once processed, samples were stored at −80°C until use.

Sera were isolated from patient blood samples by resting the blood for approximately 30 min after collection to allow for coagulation. The non-coagulated fraction was transferred into a 15 mL tube and centrifuged at 1000 g for 15 min. Supernatant was then transferred to a new tube and centrifuged at 1000 g for 5 min. Supernatants were aliquoted into 200 μL cryovials and stored at −80°C until transferred to participating laboratories.

Biosensor chips were fabricated and processed as described previously with the following modifications (Molony et al., 2012; Rice et al., 2012; Chou et al., 2020; Cady et al., 2021). Capture antibodies were diluted to 250 μg/mL with PBS (Table 1) and printed on the GCFP chips using a 0.35 mm diameter microarray pin and a SpotBot II microarray printer (ArrayIt, CA, United States) or an XactII microarray printer (LabNEXT, NJ, United States). Each capture antibody was spotted to create five individual regions of interest (ROI) per analyte on each biosensor chip. During printing, microarrays were kept at a relative humidity of 70% and at ambient temperature (AT) (−25°C). After printing, chips were allowed to dry at a relative humidity of 70% and AT (−25°C) for 1 h. Chips were then transferred to a 50 mL polypropylene tube containing desiccant and stored at AT for up to 4 weeks, with no significant signal loss (data not shown).

Biosensor chips were processed as previously described with the following modifications (Molony et al., 2012; Rice et al., 2012; Chou et al., 2020) (Figure 1). Chips were fitted with plexiglass windows and gaskets to allow fluid to pass over the surface of the chip. The chip was then blocked with 80 μL of Superblock™ T20 (PBS) blocking buffer (Thermo Fisher Scientific, MA, United States) in a static incubation for 1 h. Each chip was then washed with PBS-T at a flow rate of 0.5 mL/min for 3 min using a peristaltic pump. PBS doped with recombinant protein (20 ng/mL) or patient samples (80 μL of serum or saliva samples diluted with PBS to 1.5 mL) were then recirculated over the chip for 90 min at 0.5 mL/min for 3 min. Each chip was then washed with PBS-T at 0.5 mL/min for 3 min. Biotinylated secondary antibody mixtures were created with all detection antibodies at a concentration of 200 ng/mL and then recirculated over the chip for 90 min. The chip was then washed with PBS-T at 0.5 mL/min for 3 min and then PBS containing 500 ng/mL of streptavidin-AlexaFluor 647 (Thermo Fisher) was recirculated over the chip for 1 h. The chip was then washed for a final time with PBS-T at 0.5 mL/min for 3 min and 70 μL of PBS was injected into the flow cell. ROIs were then identified using Ciencia software, Enhanced fluorescence Reader V2.3, and fluorescence intensity for each ROI was collected using a GCFP reader (Ciencia, Inc.). A GCFP detection ratio was used to normalize each ROI value as previously described (Cady et al., 2021)

DNA was extracted using the Complete Lyse & Purify kit (Shoreline Biome, Farmington, CT, United States). Saliva samples were thawed on ice, and between 50 μL and 200 µL of saliva was pelleted by centrifuged at 5,500 rpm for 10 min. Pellets were resuspended in nuclease-free water, then purified according to the manufacturer’s protocol. Extracted DNA was amplified with the StrainID Amplify kit according to manufacturer protocol and as previously described (Graf et al., 2021). Briefly, 10 µL of DNA and 10 µL of PCR mix (Shoreline Biome, United States) were added to each well with barcoded primers. The PCR reaction was carried out on a thermocycler (BioRad, Hercules, CA, United States) according to manufacturer protocol. Amplicons were screened on a QIAxcel (Qiagen, Germantown, MD, United States) Advanced system using the Fast Analysis protocol. Samples were then pooled together based on band intensity. Samples were cleaned using the GeneRead Size Selection Kit (Qiagen, United States) according to manufacturer protocol and resuspended in 50 µL of elution buffer. After verifying that both pools were pure, amplicons were pooled and sequenced on a PacBio Sequel IIe. Sequences were demultiplexed with SBAnalyzer, then split into groups of approximately 35 samples. DADA2 software (https://benjjneb.github.io/dada2/) was used to call ASVs, and taxonomy was assigned using the Athena database (Graf et al., 2021; Callahan et al., 2016). Data analysis was performed in QIIME2 and R (McMurdie and Holmes, 2013; Bolyen et al., 2019). Visualizations were made with GraphPad Prism and microViz (Barnett et al., 2021). The code used is available at https://github.com/brandon-osullivan/Code-for-Maltz-Matyschsyk-et-al-2022.

Cytokines/Chemokine levels in the serum samples were measured in duplicate using the Luminex^® 200^™ instrument and Milliplex^®MAP kits from EMD Millipore (Cat #HSTCMAG-28SK, HCYTA-60K, HCYP2MAG-62K and HCYP4MAG-64K) according to manufacturer’s protocol. A 96 well plate provided with the kit was first washed with 200 µl per well of wash buffer. The wash buffer was then discarded, and 25 µl of serum sample was added to the plate in duplicate along with 50 µl of standards and controls as provided with each kit. Assay buffer (25 µl) was then added to the sample wells followed by 25 µl of premixed magnetic beads (provided with the kit) to each well. The plate was covered with a dark lid and placed on a plate shaker (200 rpm on a Barnstead 4625 Titer plate shaker) overnight at 4°C in a dark room. On the following day, the plate was washed 3X with 200 µl of wash buffer using BioTek ELx405^™ microplate washer with magnetic capture: after washing, detection antibodies (50 µl) were added to each well and the plate was incubated for 1 h on a plate shaker covered with a foil. Streptavidin-Phycoerythrin (50 µl) was added to each well and the plate was incubated for 30 min on a plate shaker covered with foil. Finally, the plate was washed 3 × with 200 µl of wash buffer using BioTek ELx405^™ microplate washer and analyzed using the Luminex^® 200^™ instrument (Calibrated each week with Luminex 200 Calibration and Performance Verification kits: Cat # LX2R-CAL-K25, LX2R-PVER-K25) with 150 µl of Sheath Fluid present in each well. Standard curves were generated using the Luminex xPONENT^® software and the concentrations of cytokines/chemokines in the serum samples were calculated using these standard curves in pg/mL.

Specimens were assessed for the presence of antibodies reactive with SARS-CoV-2 using an MIA as previously described (Yates et al., 2021). Briefly, recombinant SARS-CoV-2 N protein antigen (Native Antigen Company, United Kingdom) and the receptor binding domain (RBD) of the SAR-CoV-2 spike protein (MassBiologics, MA, USA) were covalently linked to the surface of fluorescent microspheres (Luminex Corporation, TX, USA). Additional target antigens (S1 or S2 domain (Native Antigen, United Kingdom) of SARS-CoV-2 were included in the multiplexed microsphere assay. Serum samples (25 μl) and antigen-conjugated microspheres (25 μl) were mixed and incubated before washing and further incubation with phycoerythrin-conjugated antisera. The antisera used were chosen to specifically recognize, as indicated, total antibodies (pan-Ig), or, individually IgM, IgA, IgG, IgG1, IgG2, IgG3, IgG4. After washing, the microsphere fluorescence intensity (MFI) was quantified with a FlexMap 3D Luminex analyzer (Luminex Corporation, TX, United States). Results were either direct MFI values with reactivity based upon a defined cutoff MFI, or, were normalized by comparison to the MFI of the negative controls and expressed as the ratio between the two (P/N). Positive reactivity is determined by a result that is ≥6 SD above the cutoff; results that fall between 3 SD and 6 SD are considered “Indeterminate”.

We set out to optimize the initial biosensor chip configuration by comparison to an established Luminex assay. First, we performed a dose response curve examining the limit of detection for the 11 analytes that were evaluated in both the GCFP assay and the Luminex data using parameters that had been employed with the first-generation GCFP chip. Chips were spotted as described above with the capture antibodies for the following analytes: TNF-α, IL-7, IL-6, IL-4, IL-21, IL-2, IL-1β, IL-17A, IFN-δ, CCL-3, CCL20. Recombinant proteins (R&D Systems, MN, USA; Shenandoah Biotechnology, Inc, PA, United States) were then diluted to 5 ng/mL, 1 ng/mL, 200 pg/mL, and 1 pg/mL and recirculated over the chip as described above. Detection ratios were calculated, and the limit of detection was determined for each matched pair set by GCFP. We then optimized the capture and secondary antibodies to reach values detectable with the Luminex MIA assay.

Since MIS-C and COVID-19 are relativity new disease states, we set out to develop our first generation GCFP biosensor chip based on literature available early in the pandemic (Cabrero-Hernandez, 2020; Bar-Meir et al., 2021; Consiglio et al., 2020; Ramaswamy et al., 2020; Whittaker et al., 2020; Buszko et al., 2021; Filbin et al., 2021). These studies showed an array of cytokines, chemokines, tissue damage markers, and inflammatory indicators that differed significantly in children with MIS-C, COVID-19, and Kawasaki (Table 1). Our goals with the first-generation chip were to: 1) identify commercially available matched pair antibodies for our ROIs to differentiate MIS-C from COVID-19, Kawasaki, and healthy controls and 2) develop a GCFP assay to simultaneously evaluate if these markers could be used as a disease-specific biosignature. We validated 42 commercially available matched pair antibodies by ELISA using recombinant proteins (Supplementary Figure S1). Nine commercially available kits that did not meet our ELISA standards when performed following manufacture protocols, and were omitted from further analysis (Supplementary Figure S1).

Once matched-pair kits were validated using ELISA, we then tested their ability to perform a sandwich-based immunoassay on the gold-coated nanoscale grating surface chip using IL-6, IL-8 and both positive (Alexafluor-647 and biotin-BSA) and negative (PBS) controls (Figures 1, 2A). When compared to PBS (negative control), the fluorescent intensity within the ROIs for IL-6, IL-8 and positive controls was significantly higher, indicating 20 ng/mL of IL-6 and IL-8 recombinant proteins could be detected with GCFP technology. Next, we examined the specificity of the ROIs by spotting all the capture antibodies in Table 1 onto the chip and recirculating 20 ng/mL of recombinant CCL5, galactin-3 and CCL7 diluted in PBS over the biosensor chip (Figure 2B). When compared to PBS (negative control ROIs), the fluorescent intensity within the ROIs for CCL5, galactin-3 and CCL7 were significantly higher, indicating that the GCFP assay was able to capture and detect these recombinant proteins at cognate ROIs and that there was no significant cross-reactivity between these recombinant proteins and unrelated ROIs. To evaluate the use of human salvia as the sample matrix, CCL5 and CLL7 (20 ng/mL) were spiked into commercially available pooled human saliva (Innovative Research, MI, United States) (Figure 2C). Spiked samples in pooled saliva matrix produced higher values than when the same analytes were diluted in PBS owing to the presence of endogenous CCL5 and CCL7 in the saliva pool. A heat map was created using the detection ratio for each set of 5 ROIs (Figure 2D). The heat map shows the detection ratio and indicates positive detection of all recombinant proteins when normalizing to the negative control ROIs on the same chip.

The first-generation GCFP chip was then evaluated to identify a candidate biomarker signature of disease using data from each of our 6 cohorts. Patient saliva and serum were analyzed with the biosensor chips and the fluorescent intensities were captured using GCFP reader (Ciencia, Inc., CT, United States) (Figure 1). For each patient sample, an image was produced from the reader; Ciencia software was used to identify ROIs, and then the fluorescent intensities for each ROI were measured and the corresponding detection ratios calculated. Figure 3 shows representative saliva data generated from a randomly selected patient from cohort B2. The image generated from patient ID 210041005 shows that five capture antibodies and the two positive controls (Biotin-BSA and alexafluor-647 labeled BSA) have a positive signal (Figure 3A). Cystatin C, Marapsin, CRP, Galectin-3, and IL-8 ROIs were each significantly different from the negative control, indicating that these analytes are present at high levels in this patient’s saliva (Figure 3B). The fluorescent intensities were then normalized to the negative control and a heat map was generated to visualize the biomarker signature (Figure 3C). Cystatin C, Marapsin, CRP, Galectin-3, and IL-8 all had high detection ratios. Saliva from the six cohorts (A1 = 15, A2 = 9, A3 = 6, B1 = 4, B2 = 13, B3 = 7) were similarly analyzed, and results are displayed as heatmaps (Figures 4A,B). Figure 4A shows the mean for each analyte from all the patient samples in each of the 6 cohorts. From this initial small data set we can begin to identify potential biomarkers for A2 (CXCL10), A3 (sDC25) and B1(IL-1 β and IL-2). Interestingly, when comparing individual patients from each cohort we can see variation in these biomarker signatures, suggesting variation in disease state presentation (Figure 4B). From the serum cohorts, 56 samples (A1 = 13, A2 = 12, B1 = 12, B2 = 12, B3 = 7) were analyzed with the first-generation biosensor chip configuration and heat maps were generated for each group and for each individual patient (Figure 5A) shows the mean for each analyte from all the patient samples run in each of the 5 cohorts and indicates potential biomarker signatures for each cohort. In A2 (MIS-C group) biomarker sCD25 was found to be a potential unique biomarker for this group, while IL-21 was a unique biomarker for B1 (Kawasaki). Again, variation between individuals’ biomarker signatures within the different cohorts indicates heterogeneous disease presentation (Figure 5B).

We noted some analyte outliers within individual saliva samples (Figure 4B). To further analyze the saliva from these samples, we performed 16S-23S’ rRNA PacBio sequencing. For this study, we singled out sequencing data for the samples that were run over the first-generation biosensor chip and compared the relative abundances of different bacterial species. To identify bacterial species that were more abundant in the samples that were analyte outliers, we applied a 99% confidence interval, and confirmed outliers with the ROUT method. These data revealed a difference in relative abundance of several bacterial species, some of which could be linked to immunocompromised status, oral hygiene or other diseases (Figure 6). For example, sample 2100410014 from cohort A3 had an inflammatory profile that more closely resembled a sample from the MIS-C or Kawasaki cohorts. This sample also had highly elevated levels of P. endodontalis, a known oral pathogen, with a relative abundance of 19.26% of the overall community (median = 0.0%) (Figure 6B). In sample 2100410113, another saliva sample that was an outlier in the biomarker analysis, we found that the community was composed primarily of Streptococcus sp. A12 (63.45%, median = 0.74%) (Figure 6C). Sample 2100430008 had high levels of Gemella sanguinis (17.61%, median = 0.59%) and Streptococcus sp. A12 (8.00%, median = 0.74%) (Figures 6C,D). Sample 2100440031 had raised levels of Porphorymonas endodontalis (0.64%, median = 0.0%) and G. haemolysans (4.65%, median = 0.84%) (Figures 6B,E). In sample 2100410044, the majority of the community was Streptococcus salivarius (61.67%, median = 2.05%) (Figure 6F). Sample 2100410011 had elevated levels of Streptococcus pneumoniae (6.77%, median = 0.0%) (Figure 6G). 2100410072 also had elevated levels of S. pneumoniae (2.20%, median 0.0%), as well as above average levels of Gemella haemolysans (7.26%, median = 0.84%) (Figures 6E,G). In sample 2100410018, we observed elevated levels of Streptococcus sp. ChDC B345 (28.39%, median = 1.34%) (Figure 6H). Our last GCFP analyte outlier, sample 2100410034, had elevated levels of S. salivarius (18.09%, median = 2.05%) and Schaalia odontolytica (17.07%, median = 3.36%) (Figures 6F,I).

Concurrent with these microarray studies, candidates for inclusion on a second generation GCFP sensor chip were evaluated. We examined 249 serum samples (180 from the United States and 69 from Colombia) from five different cohorts (A1, A2, B1, B2, and B3). Patient samples were analyzed for levels of 26 cytokines/chemokines using the Luminex xMAP^® multiplex assay. IL-4, IL-6, IL-10, and IL-13, known to be involved in Th₂ type immune response and promoting B cell differentiation to plasma cells, were significantly elevated in the A2 (MIS-C cohort) when compared to most other cohorts (Figure 7A). Furthermore, chemokines IL-8 and CXCL11 were also found to be expressed at higher levels in the A2 cohort. By comparison, the first-generation GCFP chip detected higher IL-8 for cohorts A1, A2, B1 and IL-4 in cohorts A1 and A2 (Figure 5A), while it did not detect IL-6 for any of the cohorts. It is important to note that MIA may use different proprietary antibodies or conditions or recognize different epitopes on these targets. The results from the Luminex studies were representative of a higher sample size than the GCFP analysis, which may also contribute to differences in the results. MIA data did confirm IL-10, IL-13 and CXCL11 as possible new targets for the second-generation chip.

The antibody response to the SARS-CoV-2 nucleocapsid and spike components (full spike, RBD, S1, S2) were measured using a MIA to separately detect anti-viral IgM, IgA, and all four IgG subclasses. Figures 7B–G shows that responses in the MIS-C cohort were substantially different from standard COVID-19 infections and from healthy controls. Note that some sera in the control cohort had antibodies to either the spike proteins or nucleocapsid, indicating previously undetected SARS-CoV-2 infection or an unreported vaccination. The predominant IgG subclasses produced in response to COVID-19 infection are IgG₁ (Yates et al., 2021; 2021); both the COVID-19 and MIS-C cohorts had prominent IgG₁ response centered around the RBD/S1 components of SARS-CoV-2 spike protein. The MIS-C cohort generally made a stronger IgG₁ response as compared to the COVID-19 cohort. Notably, the MIS-C cohort made a significantly higher IgA response to SARS-COV-2 spike domains as compared to the COVID-19 cohort (Figure 7G), indicating SARS-COV-2 spike antigen as a possible target for the second generation GCFP chip.

We compared US serum samples with Colombia serum samples using both GCFP biomarker signature data and cytokine microbead assay (Figure 8A; Table 2). In cohort A2 (MIS-C), the biomarkers detected by the GCFP chip in US serum samples produced a more complex and robust signature when compared to the Colombia serum samples (Figure 8A). We also compared US patient saliva samples with Colombia saliva samples: biomarker signatures produced by GCFP microarray had a trend similar to that observed with serum samples (Figure 8B). Correspondingly, MIA performed on serum samples showed lower levels of cytokine in Cohort A2 Colombia samples compared to US samples (Table 2). Taken together these data indicate a difference in immune response between these two populations.

To compare MIA data directly with the biomarker signature generated on the first-generation chip, 11 analytes that had been included in both assays and detection ratios were log transformed to visualize the data by cohort and as individual patients (Figure 9). The heat map reveals that most of the analytes were detected within both biomarker signatures generated from MIA and GCFP for the cohorts with the exception of IL-4, IL-6, CCL20 and IL-21. When we compare individual samples, similar signature patterns were found in the following samples for both MIA and GCFP; 2100410146, 2100410150, 2100410042, 2100410049, 2100410131, 2100410168, 2100410184, 2100440014, 2100410153, 2100410114, 2100410208, 2100440008, 21004400062, 2100410094, 2100410139, and 2100410165. When noting the quantified analyte levels (pg/mL) in the microbead assay dataset (Supplementary Table S1) it also becomes clear that some of the analytes (IL-4, IL-6 and CCL20) were not detected on the GCFP chip while others, like IL-1 β , were detected on the chip but not within the microbead assay. To address this, we performed a dose response curve on the first-generation chip for analytes that were measured both by MIA and GCFP using PBS that was doped with different amounts of recombinant proteins (10ng, 5ng, 1ng, 200pg, 1 pg/mL) (Figure 10A). These data indicate that for many of the analytes analyzed by both MIA and GCFP, the first-generation chip has a detection limit of 1 ng/mL while MIA could detect levels as low as 1 pg/mL, which could explain the analytes that were undetected by GCFP. We further optimized the GCFP chip to improve the assay’s limit of detection (Figure 10B). We first increased the amount of secondary antibody to 10-fold the amount recommended by the manufacturers for ELISA. We did not see an improvement with this change (Figure 10B). We then increased the amount of capture antibody immobilized on the GCFP chip by twofold (500 μg/mL) and we were able to see an improvement in limit of detection for most analytes, indicating that manipulating capture reagents can further improve detection limits. Future studies for the next-generation chip will not only be focused on adding other reported analytes now known to be important for MIS-C, but also optimizing each analyte reagent set to allow for detection in the pg/mL range.