Microfluidic Immuno‐Serolomic Assay Reveals Systems Level Association with COVID‐19 Pathology and Vaccine Protection

How to develop highly informative serology assays to evaluate the quality of immune protection against coronavirus disease‐19 (COVID‐19) has been a global pursuit over the past years. Here, a microfluidic high‐plex immuno‐serolomic assay is developed to simultaneously measure50 plasma or serum samples for50 soluble markers including 35proteins, 11 anti‐spike/receptor binding domian (RBD) IgG antibodies spanningmajor variants, and controls. This assay demonstrates the quintuplicate test in a single run with high throughput, low sample volume, high reproducibilityand accuracy. It is applied to the measurement of 1012 blood samples including in‐depth analysis of sera from 127 patients and 21 healthy donors over multiple time points, either with acute COVID infection or vaccination. The protein analysis reveals distinct immune mediator modules that exhibit a reduced degree of diversity in protein‐protein cooperation in patients with hematologic malignancies or receiving B cell depletion therapy. Serological analysis identifies that COVID‐infected patients with hematologic malignancies display impaired anti‐RBD antibody response despite high level of anti‐spike IgG, which can be associated with limited clonotype diversity and functional deficiency in B cells. These findings underscore the importance to individualize immunization strategies for these high‐risk patients and provide an informative tool to monitor their responses at the systems level.


Introduction
Coronavirus disease-19 (COVID- 19)  has become a serious worldwide public health emergency and the ongoing evolution of the SARS-CoV-2 virus still poses enormous challenges for global pandemic control. [1][4] Although vaccinations are highly effective against symptomatic disease and notably reduce fatality rates, [5] patients with hematologic malignancies such as chronic lymphocytic leukemia (CLL), multiple myeloma, and autoimmune diseases under immunosuppressive treatments may not mount adequate neutralizing antibody responses after receiving vaccinations. [6,7]Thus, as COVID-19 variants continuously emerge, there is still an unmet need for new immuno-serolomic assays to systematically evaluate the effectiveness of immune protection against COVID-19 in these vulnerable populations.
However, most of the currently available methods only provide low-plex (<10) detection of either SARS-CoV-2 specific serum antibodies [8] or biomarkers of immunocompetence, [9][10][11] without capability of evaluation a patient's immune status at the systems biology level.While some fluorescent bead-based methods offer high-plex evaluation of cytokines, the throughput is very low, and high-volume input is required.Recently, several microfluidic chips have been developed and have demonstrated superior detection performances than traditional methods such as enzyme-linked immunosorbent assays, [12] chemiluminescent immunoassays, [13] and lateral flow assays. [14]For example, Swank and colleagues developed and validated a nanoimmunoassay device to detect the presence of anti-spike IgG antibodies with throughput of 512 to 1024 samples in parallel, [15] and Rodriguez-Moncayo et al. reported a 4-plex SARS-CoV-2 serology platform that can measure 50 serum samples per assay. [16]These platforms routinely show high detection sensitivity (>95%) and specificity (>90%) with ultralow-volume whole blood or plasma samples as input (<2 μL).However, due to the limitation of current microfluidic pattern design, the level of multiplexing in these devices remains low (<10) and most assays are limited to the detection of only antibody responses.As new SARS-CoV-2 variants continue to emerge, there is a constant demand for assays that offer high-plex co-profiling of antibody binding against all variants of concern.Moreover, considering the orchestrated immune responses induced by infection or vaccination, a device capable of simultaneously detecting antibodies and immune mediators would be highly desirable.
Here we report a portable microdevice designed to perform highly multiplexed measurements of anti-SAR-CoV-2 antibodies and immune mediator proteins (up to 50 in total) in microliters of human serum with a high sample throughput of up to 50 samples per run per device.A "microfluidic patterning chip" was prepared in advance to perform microchannel-guided immobilization of capture antibodies or viral antigens, creating a 1D protein stripe barcode array on a glass slide. [17]When patient serum samples are ready to measure, a "microfluidic test chip" with a set of parallel microchannels allows for loading up to 50 serum samples over the microarray for simultaneous detection of serum proteins and SARS-CoV-2 IgG antibodies.We successfully co-measured 48 human serum samples plus one labeling control to detect a panel of 35 soluble proteins and 11 anti-SARS-CoV-2 IgG proteins in a single assay.A total of 1012 serum samples were successfully measured, demonstrating high robustness of our high-plex immuno-serolomic assay.In this work, we performed an in-depth analysis using 366 samples covering 127 patients with hematologic or other types of cancers, autoimmune disorders either with SAR-CoV-2 infection or pre-or post-COVID vaccination, and 21 healthy donors pre-or post-vaccination.We observed substantial differences in anti-spike/receptor binding domian (RBD) IgG binding antibody (bAb) responses and immune context between patients with hematologic malignancies or receiving treatments compromising immune status and Non-Heme Cancer patients and healthy controls in response to both COVID infection and vaccination.As expected, our assay revealed limited IgG bAb response in those with a weakened immune system as compared to healthy donors and provided a comprehensive assessment of multi-factorial immune responses.We also found an impaired RBD-specific binding ability in hematologic malignancy patients with COVID infection that could be associated with a limited diversity of B cell clonotypes, which was confirmed by single-cell B cell repertoires (BCR) sequencing.

Development and Calibration of High-Plex Immuno-Serolomic Assay
To develop a high-plex immuno-serolomic assay with high versatility and low complexity, we utilized a single-layer microfluidic design strategy that employs polydimethylsiloxane (PDMS) "microfluidic patterning chip" and a poly-L-lysine coated glass slide (PLL slide) for creating high-density protein barcode array and a "PDMS microfluidic test chip" for multiplexed measurement of protein targets from a large number of samples (Figure 1a).This microfluidic flow patterning method as described in our previous work [17] was adapted to fabricate high-density barcode array chips.In total, 35 capture antibodies, including antibodies against cytokines/chemokines, angiogenesis markers, neutrophil activation markers, endotheliopathy markers, and 11 SARS-CoV-2 recombinant antigens were immobilized on the PLL slide by using the first PDMS chip, which contains 50 parallel microfluidic channels with 5-turn serpentine patterns in each channel (Figure 1b).After removing the first PDMS chip, the immobilized functional microarray slide can be stored in a −80 °C freezer until use.When serum samples are ready to measure, a second PDMS test chip with 50 microchannels perpendicular to those in the first PDMS chip was placed on the microarray slide.This unique design allows simultaneous measurement of 50 serum samples with 5 replicates for each sample, minimizing the detection error and/or bias caused by nonspecific antibody bindings or debris (Figure 1c).After flowing the serum samples over the surface of the microarray, the second microfluidic chip was removed and a cocktail mixture of biotinylated detection antibodies and fluorescent dye-conjugated anti-human IgG antibodies was loaded onto the microarray slide.Finally, the fluorescent images were obtained by using a three-laser microarray scanner (Genepix 4200) and a software suite has been developed to quantify fluorescence intensities of the corresponding protein targets (Methods).
To quantitatively evaluate the detection ability of our device, we conducted titration tests using recombinant proteins, anti-SARS-CoV-2 Spike Abs, and the US SARS-CoV-2 serology standards.Serially (2-fold) diluted antigen or SARS-CoV-2 Ab solutions in 1X phosphate-buffered saline (PBS) were loaded in each inlet of the microfluidic test device and flowed onto the barcode array.The relative intensity against recombinant proteins or SARS-CoV-2 Spike Ab was adjusted by log2 normalization after subtracting the background threshold.We first utilized Human SARS-CoV-2 Serology Standard Spike IgG and Nucleocapsid IgG generated by the NCI Frederick National Laboratory for Cancer Research (FNLCR) to evaluate the ligand binding activity of our assay, demonstrating a detection range of 0.01-1000 Binding Antibody Units (BAU) mL −1 (Figure 1d).Then, commercially available SARS-CoV-2 Abs were used to estimate the concentrations of IgG Ab against spike or RBD antigens of SARS-CoV-2 wild type and other variants, including alpha (B.1.1.7),beta (B.1.351),gamma (P.1), and delta (B.1.617.2).We found that titration curves from all recombinant antigens merged together with notably high Pearson's R values (Figure 1e; Figure S1, Supporting Information), indicating comparable binding affinities of commercial antibodies to the panel of antigens.There are no standard values available in these antibodies to convert the fluorescent intensities of theses variants to BAU mL −1 .However, we were able to choose the titration curve of wild-type spike antigens as a reference and convert all the other scanning intensities to concentrations.A detection capacity ranging from 0.001 μg mL −1 to 1 μg mL −1 was achieved for these IgG bAbs (Figure 1e).Similarly, the titration curve for a 35-protein panel was generated and a dynamic range of 0.001-10 ng mL −1 was obtained (Figure 1f).Overall, these data indicate that our high-plex immuno-serolomic assay could quantify concentrations of both SARS-CoV-2 antibodies and plasma protein markers with good sensitivity and dynamic range.

Performance Evaluation of the Assay
We systematically evaluated the performance of our newly developed high-plex assay.We first demonstrate the throughput capac-ity of the assay with 48 serum samples obtained from patients.After fabricating a microarray by immobilizing SARS-CoV-2 recombinant antigens and capturing antibodies against a panel of immune mediator proteins, 8-μl volumes of a serum sample from each patient/donor were loaded onto the device and drawn from the inlets to the outlets through the barcode array for 1 hour using a controlled house vacuum system.To differentiate the location of the starting column and row in the raw fluorescent images, we introduced the fluorescein isothiocyanate (FITC)-conjugated bovine serum albumin (BSA) and FITC-conjugated anti-mouse IgG Ab to the first channel of two PDMS microfluidic chips, respectively.Allophycocyanin (APC, 635 nm emission) and phycoerythrin (PE, 532 nm emission) were employed to measure circulating plasma proteins and anti-SARS-CoV-2 bAbs, respectively.In the scanned fluorescent image of the assay region (Figure 2a), each row represents one tested sample and each column is the signal of a specific protein or anti-SARS-CoV-2 antibody.We observed high consistency of the florescence patterns between the 5 replicates among all the 48 samples, indicating the robustness of our assay.For data quantification, we exported the fluorescence intensity and converted it to concentration or BAU mL −1 using the titration curves (Figure S2a,b, Supporting Information).
To confirm that there is no crosstalk or leakage between channels during the reagent flowing, we coated antibodies generated from other species (human, goat, or rabbit) for PDGF-BB, IL-17A, IL-9, IL-32, Granzyme B, and sCD40L, whereas the remaining 29 are mouse-derived capture antibodies.After applying FITCconjugated anti-mouse IgG Ab to the first row of the second PDMS device, no signal was detected for the above-mentioned proteins as well as the SARS-CoV-2 spike or RBD antigens, suggesting minimum crosstalk (Figure 2b).To assess device-todevice reproducibility, we repeated the tests independently using 5 barcoded array chips fabricated at different times (Figure 2c), and the results show tight correlation of average Pearson's R = 0.99 (p-value <0.0001) (Figure 2d).
To explore the accuracy and reliability of our device regarding serological detection, we compared our high-plex assay with a commercially available assay from Quest Diagnostics.Our multiplex assay had a Pearson's R of 0.79 (p-value <0.0001) when correlated with the commercial, semiquantitative assay for anti-SARS-CoV-2 Spike IgG bAbs for 98 serum samples (Figure 2e).We then evaluated all the anti-spike and RBD antibodies in our serology panel using 101 blinded samples provided by NCI FNLCR including sera from patients with COVID infection and healthy controls (Figure S3a-c, Supporting Information).Independent analysis by scientists at the NCI FNLCR revealed that overall average sensitivity and specificity of IgG reactivity are 87.3% and 86.4%, respectively (Figure 2f).Specifically, the detection sensitivity of IgG against spike antigens was higher than against the RBD (mean 96.6% vs mean 77.8%), whereas the specificity of RBD detection was greater in comparison with that of spike antibodies (mean 92.2% vs mean 80.6%).The best overall performance was obtained by the detection of IgG against alpha (B.1.1.7)spike with 100% sensitivity and 95.9%, specificity.Moreover, for both anti-spike and RBD antibodies, we found that curves of protein panels.Each titration curve was plotted with a hyperbola equation in nonlinear regression.The data presented is the mean value of 5 replicates from a single assay.Scatter plots show means ± SEM.PDMS, polydimethylsiloxane; Ab, antibody; BSA, bovine serum albumin; FITC, fluorescein isothiocyanate; APC, allophycocyanin; PE, phycoerythrin; BAU, Binding Antibody Units.later-stage variants were generally detected at higher sensitivity and lower specificity than wild type.Nevertheless, concentration levels were not affected by differences in sensitivity/specificity (Figure 2g).Consistent with this observation, we found that the antibody binding affinities for the nucleocapsid protein were significantly lower than for wild-type Spike antigen (Figure 2h), suggesting sub-optimal performance of this assay in the current configuration to distinguish vaccination from natural infection (Figure S3d, Supporting Information).
We further conducted comparative studies to validate the detection performance of the immune mediator protein panel.A total of 20 serum samples were tested using both our assay and a fluorescent bead-based commercial assay (Eve Technology).For 25 cytokines that were evaluated in both assays, an overall Pearson's R of 0.8 was achieved (Figure 2i).In addition, serum samples from four healthy controls, four Intensive Care Unit (ICU)-survived patients, and four deceased patients were obtained and evaluated using our high-plex device.Although few significant differences were observed due to the limited cohort size, our data showed a clear trend toward elevated concentrations of IL-8, GM-CSF, resistin, G-CSF, IFN-, IL-6, IL-17A, lipocalin-2, MIP-1, and TNF- in ICU-survived or ICU-deceased patients compared with controls (Figure S4a, Supporting Information), which is highly consistent with the results from a third party commercial assay in a previously published study that profiled these proteins in the same patient populations (Figure S4b, Supporting Information). [18]hus, all these fully validated our microfluidic immuneserology assay using the commercial serology test by Quest Diagnostics, the blind samples from NCI FNLCR, and the commercial multiplex protein assay.Compared with other methods (Table 1), our technology is versatile with capability of 46-plex detection of both SARS-CoV-2 bAbs and functional proteins, while demonstrating satisfactory performances of high throughput, low sample volume input, minimal crosstalk, high reproducibility, and high accuracy.Notably, this high-plex assay has been used to measure 1012 serum samples to monitor immune responses to natural infection or vaccination and this cohort is still expanding (Figure 2j).The highly consistent measurements from standard samples and the real-world clinical serum samples have demonstrated the robustness of our microfluidic high-plex immuneserology assay (Figure S5, Supporting Information).

Association of Immune Mediator Proteins in 148 Patients and Donors upon Vaccination
We conducted an in-depth assessment of 366 serums from 148 patients or donors using our high-plex immuno-serolomic assay image shows a fluorescence signal from 5 serum samples.FITC conjugated BSA was introduced to differentiate first column.b) The evaluation of potential channel cross-talks.Antibodies generated from other species were coated for the indicated proteins, and the remaining 29 are mouse-derived capture antibodies.FITC-conjugated anti-mouse IgG Ab was introduced to the first row.c) Device-to-device reproducibility evaluation using 5 barcoded array chips prepared at different time points.d) The Pearson correlations across all 5 chips.e) The correlation plot between our high-plex assay and commercialized Quest diagnosis in the measurement of 98 serum samples.f) The sensitivity and specificity of each evaluated anti-spike/RBD antibody.101 preliminary blinded samples from NCI-Frederick National Laboratory for Cancer Research were used.g) The concentration comparisons between anti-spike or anti-RBD antibodies.h) The sensitivity and specificity of nucleocapsid in the test of 101 preliminary blinded samples.i) The Pearson correlation between our assay and commercialized Eve Technology assay in the measurement of 20 cytokines from 20 serum samples.j) Unsupervised clustering of all the 1012 samples measured using our assay (only for visualization).The concentrations of the 35 proteins and 10 anti-spike/RBD antibodies were used to perform the clustering.Each point represents one measured sample.P values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).Scatter plots show means ± SEM.Ab, antibody; BSA, bovine serum albumin; FITC, fluorescein isothiocyanate.
to evaluate its performance in clinical samples.Patients were divided into six groups based on their clinical demographics, immune status or baseline illness, and vaccination record (Figure S6a, Supporting Information).Within our cohort, 38 patients were infected with acute COVID-19, 26 of them were patients with hematologic malignancies (Heme COVID+), and the remaining 12 were patients with non-hematologic cancers (Non-Heme Cancer COVID+).An additional 89 patients donated blood pre-and post-vaccination.Of these 23 had cancer and were known not to have had COVID infection prior to their blood draw (Non-Heme Cancer pre-Vax and Non-Heme Cancer post-Vax), and 66 patients had autoimmune diseases for which they received B cell depletion therapy (Autoimmune B cell-depleted pre-Vax and Autoimmune B cell-depleted post-Vax).Additionally, we included 21 healthy donors preand post-vaccination (HD pre-Vax and HD post-Vax) to establish baseline controls.Basic demographic information of these patients and donors is provided in Table S1 (Supporting Information).In regard to vaccine type, BNT162b2 (Pfizer-BioNTech, Germany), mRNA-1273 vaccine (Moderna, USA), or Ad26.COV2.S (Johnson & Johnson, USA) were administrated to patients/donors.In total, 366 serum samples were collected and analyzed.
Among all 35 immune mediators, we observed higher concentrations of PDGF-AB, IL-10, IL-12p70, and IL-13 compared to other protein markers (Figure S6b, Supporting Information).To gain insights into key differences in these immune function proteins between patients with and without COVID infection and pre-versus post-vaccination, we compared the measurements of the 35 proteins across all groups and identified substantial differences in expression patterns.In contrast to minimal associations in healthy donors before vaccination (Figure S7a, Supporting Information), we found several subsets of proteins whose expression levels are highly correlated in the patient groups.Specifically, in the Non-Heme Cancer Pre-Vax group, positive associations between T cell functional markers, including GM-CSF, TNF-, IL-6, IL-16, IL7, and IL-15, dominated the correlation pattern (Figure 3a); patients with auto-immune diseases who received B-cell depletion therapy showed concerted dysregulation in a larger number of proteins in our panel albeit with lower correlation coefficients (Figure 3b).This suggests that our device and our protein panel can capture highly relevant protein signatures and associations unique to the responses of these vulnerable patient populations.Interestingly, we found a negative correlation between Resistin (RETN) and other markers of neutrophil activation including G-CSF and Lipocalin-2, and similar anticorrelation was also present in two angiogenesis signatures (PDGF-AB and PDGF-BB), implying complex immune dysregulation in these immunocompromised patients.After COVID vaccination, a concerted immune response emerged in healthy donors (HD post-Vax group) (Figure S7b, Supporting Information), indicating that our device can detect systemic modulation of the immune responses post-vaccination.In Non-Heme Cancer patients, compared to HD, we found more modest changes in protein associations between pre-and postvaccination samples (Figure 3c).Finally, as would be expected in immunocompromised patients, immunization did not induce coordinated immunological responses in Autoimmune (B cell-depleted) patients as reflected in an unchanged correlation pattern in the post-Vax compared to pre-Vax setting in this patient population (Figure 3d).
We next sought to quantitively compare functional protein levels pre-and post-vaccination.As expected, in the healthy donor group, post-vaccination samples showed significantly higher concentrations of soluble proteins related to pro-inflammatory, T cell effector, and endotheliopathy pathways, whereas proteins that regulate angiogenesis, neutrophil activation, and macrophages remained mostly unchanged compared to prevaccination samples (Figure 3e).Non-Heme Cancer patients and patients with autoimmune disorders receiving B-cell depleting therapy had no significant changes post-vaccination (Figure 3f,g).It has been found that the markers of angiogenesis (VEGF-A, PDGF-AA, and PDGF-AB/BB) are elevated in hospitalized patients compared with non-critical COVID-19 infection, [19] the markers of neutrophil activation (RETN, LCN2, HGF, IL-8, G-CSF) are among the most potent discriminators of critical illness, [18] and uncontrolled activation of macrophages is responsible for acute respiratory distress syndrome (ARDS) in COVID-19 patients. [20]Our data indicate that vaccination elected detectable systemic immune modulation but does not induce a COVID-like illness in healthy individuals or even patients with cancer or autoimmune diseases on B-cell depleting therapy.
In our vaccination cohort, serum samples were collected at multiple timepoints, which are two weeks after the first dose of vaccination (Visit 1), 0-3 days before the second dose of vaccination (Visit 2), and one week after the second dose of vaccination (Visit 3).We longitudinally evaluated the protein levels at different timepoints in each group.At Visit 1, there was no significant elevation of either pro-inflammatory or T cell effector markers in healthy donors (Figure 3h,i).The peak level of these proteins was observed at Visit 2, and TNF-, IL-15, and IFN- dominated the elevation (Figure S8a,b, Supporting Information), which is in line with previous reports. [21]By contrast, endotheliopathy markers peaked at Visit 3 (Figure 3j; Figure S8c, Supporting Information), representing a late-stage response to vaccination.These trends were observed when we examined the change of signals from pre-to post-vaccination in five donors although it did not reach statistical significance (Figure S9a, Supporting Information).Again, we did not find kinetic changes in the two patient groups, and the concentration of pro-inflammatory proteins postvaccination in Autoimmune (B cell-depleted) patients was even significantly reduced when compared with pre-vaccination level (Figure 3h; Figure S9c, Supporting Information).Taken together, these data suggest that SARS-CoV-2 vaccination has minimal effect on the immune mediator protein production in these immunosuppressed patients.

Natural SARS-CoV-2 Infection Induces a Higher Magnitude Of The Pro-Inflammatory Signatures Compared to Vaccination
Utilizing the data from our protein panel, we next asked what the immunologic profile difference is following natural COVID-19 infection as compared to that induced by vaccination.To this end, we performed unsupervised clustering of 302 patient serum samples based on the concentration of the 35-protein panel, leading to the identification of five proteomically distinct subpopulations (Figure 4a).Notably, Uniform Manifold Approximation and Projection (UMAP) representation color-coded by patient conditions showed that the cluster 2 subset contains mostly samples from COVID-19-infected patients (Figure 4b).The differentially expressed proteins defining each cluster indicated that samples in cluster 2 have upregulated levels of protein markers regulating inflammatory and effector T cell functions (Figure 4c).The average concentration of pro-inflammatory and T cell effector proteins was consistently higher in the two COVID+ patient groups as compared to baseline HD pre-Vax samples, whereas the significance level was reduced when comparing vaccinated groups to the baseline group (Figure 4d).
We next examined if SARS-CoV-2 infection conferred a different functional protein response to hematologic or nonhematologic cancer patients by interrogating the protein correlations within these two infected groups.In Heme COVID+ patients, a group of cytokines regulating inflammatory response or T cell activation, including IL-9, IL-21, IL-22, IL-12 p70, RANTES, GM-CSF, TNF- and IL-32, were positively correlated with each other (Figure 4e).These proteins were also significantly correlated with molecules involved in T cell cytotoxicity and macrophage activation, such as IFN-, MIP-1, and MCP-1.While several functional T-cell cytokines were also positively correlated for Non-Heme Cancer COVID+ patients, the relationships were much weaker (Figure 4f).Prior work demonstrated that patients with severe COVID-19 had increased positive associations of proteins linked to cytokine release syndrome (CRS), which overlapped with the immune mediators identified above. [22]Thus, our data suggest that patients with hematologic cancers with COVID infection were more likely to exhibit higher disease severity than Non-Heme Cancer COVID+ patients.
Leveraging the protein data detected by our high-plex assay panel, we attempted to develop a model to differentiate the severity of COVID-19 infections.Patients were classified into three groups on the basis of their level-of-care requirements: admission to ICU, regular hospital admission (non-ICU hospitalization), or no hospitalization.The concentrations of eight proinflammatory proteins or ten proteins regulating T cell effector functions were integrated into a logistic regression model using level-of-care as the predictive variable.Stratified five-fold cross-validation was used to assess the discriminative power of this model.Overall, a sensitivity of ≈72% to separate ICU from no hospitalization patients with a false-positive rate of <2% was achieved using the pro-inflammatory panel [receiver operator characteristic area under the curve (ROC AUC) = 0.73 ± 0.23] (Figure 4g).The sensitivity was reduced to 55% in the classification of ICU versus non-ICU hospitalization (Figure S10a, Supporting Information) and further reduced in the separation of non-ICU hospitalization versus no hospitalization (Figure S10b, Supporting Information).By contrast, the T cell effector panel could provide an improved sensitivity (≈59%) to discriminate patients with non-ICU hospitalization from those without hospitalization (Figure S10c, Supporting Information).However, this panel could not differentiate between ICU and no hospitalization (Figure S10d, Supporting Information).
To identify specific molecules that could discriminate natural infection from vaccination, we evaluated each of the proteins in our panel and compared them across groups.Among all the comparisons (Figure S11, Supporting Information), IL-16, IL-17A, and IL-21 were found to be significantly elevated in Heme COVID+ patients compared with vaccinated groups (Figure 4h).Furthermore, the Heme COVID+ group also had significantly higher levels of the neutrophil activation marker IL-8, and the angiogenesis marker TGF-1 than the Non-Heme Cancer COVID+ group (Figure S11, Supporting Information), implying elevated severity in these patients with hematologic cancers. [18,23]By contrast, the concentration of the immune mediator sCD40L and PDGF-AB, a potent inducer of CCL2 gene expression, were significantly higher in vaccinated groups (Figure 4i).Other liter-ature has reported that these two proteins were inversely correlated with severity of COVID-19, [24,25] further suggesting increased severity signatures in infected patients.Additionally, the chemokine RANTES (CCL5) has been found to be significantly elevated from an early stage of the infection in patients with mild but not severe disease. [26]Our evaluation identified a significantly higher concentration of RANTES in post-vaccination groups than in infected groups, suggesting similar response characteristics between vaccination and mild infection (Figure 4i).Validation of previously identified cytokine signatures in COVID infected versus vaccinated patients further support the accuracy of our high-plex protein assay.In addition, the regression model based on above identified six proteins unambiguously separated Heme COVID+ patients from the other groups (Figure S12, Supporting Information).Collectively, our data confirm upregulation of a group of pro-inflammatory proteins that mark higher severity of disease in patients with active COVID-19 infection, distinguish ICU admission from mild disease that does not require hospitalization, and support the observed clinical safety of vaccination compared to infection.

A Second Dose Vaccination Boosts Kinetic Elevation of SARS-CoV-2 IgG bAb Levels, but not in Patients with B Cell Depletion Therapy
We next move to the analyses of measuring circulating IgG antibody binding against five protein variants from each of the spike and RBD domains in the serum samples.We created a "concentration heatmap" of the anti-spike and anti-RBD antibodies in each patient/donor group, showing higher levels of bAbs in patients with COVID-19 infection or groups post-vaccination, except for patients with autoimmunity treated with B cell depletion therapy (Figure S13a, Supporting Information).Next, quantitative comparisons were conducted to examine dynamic bAb changes at different timepoints after receiving vaccination.As expected, the primary vaccination effectively induced antibody responses against wild type and all the other variants in healthy donors, whereas only spike IgG bAbs showed significant increases in Non-Heme Cancer patients (Figure S13b,c, Supporting Information).The bAb levels were sustained two weeks after the first dose of vaccination (Visit 2) (Figure 5a,b).After the secondary dose vaccination, these responses were boosted significantly in both healthy donors and Non-Heme Cancer patients, indicating the beneficial effect of the second shot.No differences were observed between the types of vaccine in this study (Figure S14, Supporting Information).In both groups antibodies effectively bounds all variants at both Visit 1 and Visit 3 (Figure 5c,d).Interestingly, we found significantly increased RBD IgG binding against B.1.1.7,B.1.351and P.1 variants compared with the wild-type protein in healthy donors while anti-spike IgG binding was uniform across variants (Figure S15a, Supporting Information).Variant recognition by anti-spike and anti-RBD antibodies showed no differences in Non-Heme Cancer patients (Figure S15b, Supporting Information).By contrast, both primary and secondary vaccination failed to induce robust anti-spike or RBD antibody concentrations in autoimmune patients with B cell depletion therapy (Figure 5a,b; Figure S13b,c, Supporting Information), attributable to the elimination of circulating B-cells in this patient population.Antibody correlations were low across variants after the first dose of vaccination, especially for the RBD antibodies which improved at Visit 3 after second vaccination but not to the same levels as in the other two donor/patient groups (Figure 5e).
At Visit 3, we observed that anti-spike IgG levels of all the healthy individuals were above 0.5 ug mL −1 , whereas the concentrations of several Non-Heme Cancer patients remained nearly zero, suggesting the necessity of evaluating and comparing the response rate (proportion of individuals with concentration larger than the cut-off value) between patient groups and different timepoints.The cut-off value (50 BAU mL −1 ) of anti-spike IgG bAb provided by the World Health Organization (WHO) [27] is equivalent to 0.3 μg mL −1 in our assay (Table S2, Supporting Information), revealing protective antibody concentrations against nearly all variants in healthy donors at one week after the second dose (Visit 3) (Figure 5f).A lower response rate was noted in the Non-Heme Cancer patients at all three timepoints as compared to the healthy donor group, and the difference was further increased compared with B cell-depleted patients (Figure 5g), especially for anti-RBD antibodies (Figure S16, Supporting Information).The high-plex capacity of our device to simultaneously detect IgG antibodies and secreted proteins allowed us to examine the relationships between alterations in these functional proteins and the levels of IgG bAbs in the same serum or plasma samples.No significant correlations between protein responses and anti-spike/RBD IgG titers were detected (Figure S17, Supporting Information).Together, these data point to a profound reduction of the second dose vaccination-induced IgG response in patients with non-hematologic cancer, and a nearly complete loss in immunocompromised patients with B cell depletion therapy.

Hematologic Malignancy Patients with COVID Infection Exhibit Impaired RBD-Specific Binding
We next expanded our characterization to the humoral antibody response following natural COVID-19 infection in patients with hematologic or non-hematologic cancers.As expected, the binding of SARS-CoV-2 nucleocapsid (N) protein, an RNA-binding protein critical for packaging of the viral genome, [28] was found to be substantially higher in the two infected groups as compared to vaccination groups (Figure 6a).The convalescent antibody level in Heme COVID+ patients was significantly lower than the vaccination-induced antibody level in healthy donors, whereas negligible differences were found in the Non-Heme Cancer patients as compared to donors (Figure 6b).Consistently, the response rate showed a significant increase in the Non-hematologic cancer group (Figure 6c).These observations suggest that most patients with non-hematologic cancer could achieve certain levels of COVID IgG antibody seroconversion, whereas hematologic cancers and their treatment might notably impair the antibody response.No differences were observed in the comparisons between IgG titers against wild type and other variants, exclusive of enhanced binding to B.1.617.2 variant in Heme patients (Figure S18, Supporting Information).Additionally, we found significantly increased anti-RBD IgG levels in convalescent Non-Heme Cancer patients as compared to their non-infected counterparts after primary vaccination.However, the differences were no longer detectable at the later timepoints, further indicating major benefits of the second dose vaccination (Figure S19, Supporting Information).
Consistent with their reduced ability to elicit effective IgG responses, the correlations between spike and RBD binding were largely absent in COVID-infected Heme patients (Figure 6d), similar to the pattern we observed in the B-cell depleted group (Figure 5e).By contrast, significantly positive associations were noted for Non-Heme Cancer patients.Again, no correlations were seen between circulating proteins and IgG antibodies in the two infected groups (Figure S20, Supporting Information).We next longitudinally evaluated IgG bAb titers at two timepoints, in ICU or after recovery, from a hematologic cancer COVID+ (CLL) patient and a Non-Heme Cancer COVID+ patient.Similar antispike/RBD IgG bAb titers were observed at two timepoints for the Non-Heme cancer patient (Figure 6e), whereas the hematologic cancer patient had significantly lower anti-spike/RBD IgG bAb titers during the ICU admission with improved antibody levels at time of recovery (Figure 6f).However, this hematologic cancer patient's RBD antibody response either during the ICU stay or at time of recovery was nearly undetectable, suggesting a lower diversity of antibody repertoire and thereby compromised humoral immune protection against COVID-19 in some hematologic cancer patients despite the presence of anti-spike IgG antibody.

Single-cell BCR and Transcriptome Sequencing Identifies a Limited Clonotype Diversity and Functional Deficiencies in B Cells from Heme COVID+ Patient
The observed differences in RBD binding between COVIDinfected patients with hematologic or other types of cancer prompted us to further investigate the underlying immune response mechanisms.By combining single-cell RNA sequencing (scRNA-seq) and co-measurement of BCR repertoires, we characterized the heterogeneity of circulating B cells from the above evaluated hematological cancer patient and Non-Heme Cancer patient (in ICU and after recovery) (Figure 6e,f).In total, we profiled the transcriptome and BCR clonotype of 5856 single cells from the three samples.Due to the clonal expansion and accumulation of malignant long-lived B cells in CLL patients, [29] a majority of cells (≈78%; 4526 cells) were derived from the Heme patient (Figure S21b, Supporting Information).Integrating and clustering all the scRNA-seq data points identified 7 subpopulations and no major batch effect was observed (Figure S21a,b, Supporting Information).High CD5 expression has been recognized as one of the signatures in leukemic B cells in CLL. [30]Consistently, we observed a notably higher expression of the encoding gene CD5 in cells from the Heme cancer patient compared to cells from the Non-Heme cancer patient (Figure S21c, Supporting Information).On the basis of CD5 and FCER2 (encoding CD23) expression, used for differential diagnosis of CLL, [30] we classified all the Heme B cells into Leukemic or Non-Malignant B cell groups (Figure S21d,e, Supporting Information).After filtering out leukemic-B cells, 1803 non-malignant B cells from the Heme cancer patient were included in the downstream analysis to compare with the other patient.
Consistent with our hypothesis, we found a much lower diversity of the immune repertoire in remaining non-malignant B cells from the Heme cancer patient indicating an inability to develop antibody response to diverse antigens (Figure 7a; Table S3, Supporting Information).By contrast, high clonal diversity was observed in Non-Heme Cancer patients during the time both in the ICU and after recovery.The frequency of the most common clonotype was only ≈0.3-0.5%, and no difference existed for the top 5 clonotypes.To further explore the association between sample source and B cell states, UMAP clustering of all the filtered B cells was performed and 5 new clusters were identified (Figure 7b; Figure S21f, Supporting Information).A previous report identified an expansion of late-stage B cells in patients with severe symptoms as compared to the healthy controls and patients with mild symptoms. [31]We checked the same marker gene expression and found that cells in cluster 3 exhibited elevated levels of plasma cell markers (CD27 and CD38), [32] whereas cells in cluster 1 were highly enriched for IGHD expression, a marker for naïve B-cells (Figure 7c).The cell proportion of Heme sample in cluster 1 was found to be notably lower than that of Non-Heme samples while the proportion in cluster 3 was higher (Figure 7d), suggesting a more severe COVID infection signature in Heme patients.Furthermore, considering the expression levels in the three cell groups, a significantly decreased IGHD and increased CD27/CD38 expression in Heme cells were also observed (Figure 7e).
To identify the unique transcriptomic signatures of each sample, we performed differentially expressed genes (DEGs) analysis among the three samples.Comparing B cells between Heme and Non-Heme (ICU) leads to the identification of top upregulated genes such as HBB, HLA-B, NOSIP, and top downregulated genes such as LTB, MS4A1, and CD69 (Table S4, Supporting Information).Ingenuity Pathway Analysis (IPA) [33] was performed to explore the biological pathways associated with the DEGs of each comparison.We found that Coronavirus pathogenesis pathway was significantly activated in B cells from Heme as compared to Non-Heme (ICU) (Figure 7f), which further confirmed that SARS-CoV-2 could induce higher severity levels in immunocompromised patients.The top symbolic molecules regulating this pathway include BCL2, DDIT3, EEF1A1, EIF2A, FAU, and FOS (full gene list provided in Table S5, Supporting Information).By contrast, a group of pathways that regulate B cell survival, proliferation, differentiation, and plasticity, including mTOR signaling, IL-15 signaling, PI3K signaling, and Signaling by Rho Family GTPases, were collectively inhibited in Heme cancer B-cells (Figure 7f).Additionally, the oxidative phosphorylation that provides metabolic support for cell activities was also found to be downregulated in Heme cancer B-cells.Similar results were obtained when comparing Heme with Non-Heme (recovery) cells (Figure S22a, Supporting Information).Furthermore, we investigated the underlying mechanistic pathways in Non-Heme cells at ICU stage or after recovery.As expected, B cell activities and functions at ICU stage were highly upregulated in comparison with the recovery timepoint (Figure S22b, Supporting Information).Thus, these single-cell sequencing data confirmed a markedly reduced clonotype diversity and functional deficiency in B cells from Heme COVID+ patients, which might have contributed to the observed lack of anti-RBD-specific response and likely compromised immune protection.

Discussion
The COVID-19 outbreak has highlighted the need for easily accessible and adaptable assays to guide intervention plans, especially in high-risk groups.SARS-CoV-2 infection has particularly poor outcomes in patients with hematologic cancers and in patients with compromised B-cell immunity. [34,35]e therefore designed a portable high throughput multiplexed immune-serological assay to co-profile antibody response signature and immune mediators induced by natural infection or vaccination specifically in patients with hematologic malignancies or immunosuppressed conditions compared to patients with non-hematologic cancer or healthy donors.Our microchip assay combines high sample throughput of up to 50 patient sera per chip and a high degree of multiplexing for simultaneous analysis of up to a total of 50 soluble proteins and IgG antibodies.Our assay is low-cost (PDMS chips on PLL slide), fault-tolerant (5 replicates/sample), reproducible (average Pearson's R = 0.99 using five independent assays/devices) and characterized by high sensitivity and specificity (≈87%) using low sample volumes (8 μL) as input.It has been validated with 1012 serum samples from patients with altered immunity and donors collected over the past two years of the pandemic.
To investigate response to infection and vaccination in immunocompromised patients versus general population and in different groups of immunocompromised patients, we analyzed 366 serum samples comprising six study cohorts: patients infected with SARS-CoV-2 with or without hematologic cancer, non-hematologic cancer patients pre-and post-vaccination, autoimmune disease patients under B cell depletion treatment pre-and post-vaccination, and healthy donors pre-and postvaccination.For each serum sample, we simultaneously measured 35 plasma/serum proteins (divided into 6 functional categories) and SARS-CoV-2 IgG antibodies against 11 different viral antigens (anti-nucleocapsid and anti-spike/RBD of wild-type virus and of other four variants).First, as expected, we noticed that the first two doses of vaccination did not induce an adequate immunological response in patients with a weakened immune system, supporting the need for additional strategies including booster doses for these patients.Specifically, we observed that immunization induced an increase in the levels and in the correlative expression of immune proteins in healthy donors, but the impact was significantly lower in non-hematologic and autoimmune patients, which is consistent with other reported studies. [36]hen we expanded our analysis to COVID+ samples.We confirmed that natural SARS-CoV-2 infection leads to a significantly stronger pro-inflammatory signature than SARS-CoV-2 vaccination, and we demonstrated that the expression of proinflammatory proteins can successfully distinguish ICU admission versus no-hospitalization.Of note, hematologic cancer patients were unambiguously characterized by elevated levels of pro-inflammatory cytokines IL-16 and IL-17A, and of T cell effector cytokine IL-21.As previously reported, [22] high levels of these cytokines are associated with severe disease.In contrast, the concentration of sCD40L, PDGF-AB, and RANTES was significantly reduced in hematologic cancer patients, further implying a higher risk of COVID-19 severity in these patients. [24,25]The evaluation of these six markers may therefore develop a multivariant model to predict clinical outcomes and tailor treatment strategies. [37]he analysis of anti-spike and RBD IgG bAbs supported the accuracy of our device by detecting enhanced humoral response in healthy donors one week after the second dose of COVID-19 vaccination.In contrast, as observed also for circulating immune proteins, vaccination triggered a blunted immune response in immunocompromised patients, in particular in autoimmune patients undergoing B cell depletion treatment.These data highlight the need to establish effective therapeutic strategies in autoimmune patients by balancing the benefit of booster doses, the side effect given the already dysregulated immune condition, and the administration of B cell targeting drugs. [38]he impairment in the immune response was even more striking in COVID+ hematologic cancer patients with significantly lower levels of anti-nucleocapsid and anti-spike Abs at the time of active SARS-CoV-2 infection.Moreover, for the first time, we observed that the levels of anti-RBD Abs were significantly lower in Heme COVID+ patients suggesting a limited diversity of BCR repertoire and potentially compromised immune protection even if anti-spike bAb levels were adequate.We, therefore, performed single-cell BCR sequencing comparing Heme versus Non-Heme Cancer COVID+ samples that confirmed this finding.We also observed several functional deficiencies in the immune system of these patients.Specifically, regarding nonmalignant B cells in the Heme sample, we reported: i) low clonotype diversity; ii) expansion of late-stage B cells associated with severe symptoms; [32,39], and iii) enrichment in COVID-19 pathogenesis pathway and down-regulation of B cell activation pathways.Interestingly, we observed a high level of circulating IL-21 in COVID+ hematologic cancer patients, which is the signature cytokine produced by T follicular helper (Tfh) cells that promote B cell response and maintain germinal center (GC) formation and maturation in secondary lymphoid organs. [40]It seems that despite the effort to elicit Tfh response upon SARS-CoV-2 infection [41] as characterized by elevated IL-21 level, these patients could not develop adequate B cell immunity presumably due to defective GC reaction, especially considering pre-existing disease-related GC dysregulation. [42]e recognize that the current serology panel in our device does not include the latest emerging Omicron (B1.1.529)variants although these could be easily integrated into our microchip.More studies are required to define the immune response in hematologic cancer patients not only during infection but also during the response to vaccination to help protect this highly vulnerable group of patients.The patient cohort needs to be enlarged to evaluate the effect of other clinical variables.Finally, we still need to examine T cell response to disentangle the complex mechanisms implicated in COVID-19 severity and immunity.Nevertheless, our newly developed high-plex assay allowed to simultaneously profile COVID-19-related plasma/serum proteins and antibodies, leading to the identification of unique signatures in immunocompromised patients, in particular hematologic cancer patients, that may be used for predicting COVID-19 prognosis and adjusting therapies for these vulnerable populations.

Experimental Section
Fabrication of PDMS Device for High-Plex Immuno-Serolomic Assay: Conventional soft lithography was applied to obtain a polydimethylsiloxane (PDMS) microfluidic device.A negative photoresist of SU-8 (SU-8 2025 and SU-8 2050, MicroChem) and the chrome photomasks (Range Photomasks, Lake Havasu City, AZ) were used for the molds of 25-μm-wide and 50-μm-wide microfluidic channel device.≈25 μm-thick and ≈50 μmthick SU-8 film was spin-coated on a 4-inch silicon wafer, and the molds were then fabricated following the manufacturer's recommendations.After stir mixing GE RTV PDMS part A and part B at a 10:1 ratio, the PDMS precursor was poured onto the mold and fabricated following the degassing and curing process.The solidified PDMS slab was peeled off, and inlet and outlet holes were punched with a 2 mm in diameter biopsy puncher, which could hold up to 13 μL of solution. [43]thical Statement: This study was approved by Yale Human Research Protection Program Institutional Review Boards.Informed consent was obtained from all enrolled patients and healthcare workers.The reported results on human blood specimens were carried out under IRB-approved protocols at Yale University.
Sample Collection: Peripheral blood was collected in red top tubes containing no anti-coagulant, allowed to clot, then centrifuged at 1500 g for 10 min at 4 °C within 2 h of being drawn, the supernatant was then collected as serum.Peripheral blood samples collected in lavender Ethylenediaminetetraacetic acid (EDTA) coated tubes were layered over Ficoll and centrifuged at 1800 rpm for 20 min with the brake off.Plasma samples were collected from the top layer after centrifugation; peripheral blood mononuclear cells (PBMNCs) were collected from the layer below the plasma, washed with phosphate-buffered saline (PBS) counted, and then cryopreserved in dimethyl sulfoxide (DMSO) with 10% fetal bovine serum (FBS).Plasma and serum samples were stored until analysis at −80 °C, and PBMNCs were stored in liquid nitrogen.
Workflow of Multiplex Immuno-Serolomic Assay: A pair of PDMS devices containing 50 parallel microfluidic channels was placed on the same poly-L-lysine coated slide.The first PDMS device was used for preparing the high-density barcoded array chips adapted by flow patterning method, and the second microfluidic device was employed to flow serum samples on the barcoded arrays.The channel width of the first and second PDMS microfluidic devices was 25 and 50 μm, respectively.A 5-turn serpentine pattern was designed at the center of the first microfluidic device, and the second PDMS device had orthogonally aligned 50 parallel microfluidic channels, which can flow the loaded serum samples through the 5 replicate barcoded arrays.
For generating barcoded arrays, the first microfluidic device was placed on a clean poly-l-lysine coated slide (PLL slide, Electron Microscopy Science, Cat# 63478-AS) and baked for 2 h at 75 °C oven to strengthen the bonding.Afterward, 4 μL of capture Ab (≈4 μg mL −1 ) or SARS-CoV-2 recombinant antigen (≈0.2 mg mL −1 ) in 1x PBS (Gibco) was loaded in each channel and then withdrawn using a house vacuum system for ≈3 min.Information on all the antibodies used in this work is provided in Table S6 (Supporting Information).To differentiate the location of barcoded arrays in the fluorescent image, fluorescein isothiocyanate (FITC, 488 nm emission) conjugated bovine serum albumin (BSA, 1 mg mL −1 , Thermo Fisher Scientific) in 1x PBS was introduced to the first microfluidic channel.Following 4 h of incubation at room temperature, the capture antibodies or SARS-CoV-2 recombinant antigens were immobilized on the PLL slide.Then PLL slide with PDMS chip was soaked into the 1% BSA in 1x PBS (BSA solution) and PDMS chip was detached form the PLL slide.After removing excess capture antibodies or SARS-CoV-2 recombinant antigens with three times pipetting using BSA solution, the PLL slide was blocked for 1 h at room temperature using 1 ml of BSA solution on the surface.Finally, the PLL slide was sequentially dipped into 1x PBS, 1:1 ratio of 1x PBS and deionized water (DI water), and DI water to rinse off remaining salts, and then dried using nitrogen gas.Functionalized barcoded array chips were stored in the −80 °C freezer until use (within 1 week).
To evaluate circulating proteins or SARS-CoV-2 IgG bAb in serum, the second microfluidic device was carefully aligned and attached to the first barcoded array chip.Thermal bonding was not applied to avoid degradation of immobilized panels by high temperature.A total of 8 μL serum sample from patient/donor was loaded in each channel.As similar, FITC conjugated anti-mouse IgG antibody (2 μg mL −1 , Cell Signaling Technology) was loaded in the first microfluidic channel to differentiate the location of loaded samples in fluorescent image as well as recognize the presence of mouse-derived antibodies on PLL slide.The serum samples were withdrawn for 20 min using the house vacuum system, and the flow direction was exchanged three times every 20 min.The circulating proteins or SARS-CoV-2 IgG Abs can be captured by the barcoded PLL slide.Afterward, the second microfluidic chip was removed from the PLL slide after soaking in BSA solution, and a cocktail of biotinylated detection antibody (≈1 ng mL −1 per each protein panel) and phycoerythrin (PE, 532 nm emission) conjugated anti-human IgG antibody (10 ng mL −1 , Abcam) was loaded onto the barcoded array slide after removing excess serum using BSA solution.Following 45 min of incubations at room temperature, allophycocyanin (APC, 635 nm emission) conjugated streptavidin (4 ng mL −1 , Thermo Fisher Scientific) and 1 ml of BSA solution were applied for 20 min and 30 min at room temperature sequentially.Finally, the fluorescent image was obtained using Genepix 4200A scanner (Molecular Devices) in 5-μm resolution after desalination and drying steps described above.
Scanned Fluorescence Signal Analysis: Three channels, 488 nm (FITC), 532 nm (PE), and 635 nm (APC) laser, were used to collect fluorescence signals, and each image scanning was conducted under the same power and gain values.Fluorescence intensities were evaluated with an image analysis tool in Genepix Pro 6.1 software (Molecular Devices), and the mean photon counts were extracted by aligning a 20 × 20 μm 2 square array template from not only intersection of panel and serum but also a region without a microfluidic channel for a serum sample.5 square arrays from a region without a second microfluidic channel were used as on-chip control to provide a threshold for each panel in each replicate, and the threshold of each panel was defined as the mean value of the 5 square plus 3 times standard deviation.Only the values higher than threshold were log2 transformed after subtracting threshold, whereas values below the threshold were set as 0. If >2 out of 5 replicates were lower than the threshold, the data point was considered low quality and also set as 0. Afterward, fluorescence intensity data were converted to concentration or BAU mL −1 unit using the titration curve shown in Figure 1d-f.
For quantitative evaluation, a titration test was conducted using a serial 2-fold dilution of recombinant antigens or SARS-CoV-2 Abs for the 46-plex panel, and each curve was fitted by applying hyperbola equation in a nonlinear regression model using Prism 9 (GraphPad).The value of Kd and Bmax for each protein or antibody was calculated by the equation below.
where B max is the maximum number of binding sites, K d is the ligand concentration that binds to half the receptor sites at equilibrium, respectively.B max and K d values of each protein or antibody were listed in Table S7 (Supporting Information).
Cytokine detection using Eve Technology: A total of 20 serum samples were measured with a 71-plex laser bead technology-based platform (Eve Technology, Canada) to cross-validate the protein detection accuracy of the assay.There were 25 shared proteins in the commercialized panel and the panel.Briefly, 50 μl of serum was added according to the supplier's instructions.Different concentration of red and infrared fluorophore dyes creates uniquely-colored bead sets, and each bead was incorporated with different capture antibody.Once target analyte was captured, detection antibody and streptavidin conjugated with phycoerythrin dye (streptavidinphycoerythrin) were introduced for evaluating the analyte on each bead.Dual-laser system (Bio-plex 200) was used to obtain fluorescent signals from beads and streptavidin-phycoerythrin.Then, the results were quantified according to the standard curve.For the correlation analysis, the data point with zero concentration in the assay or out of detection range in the Eve Technology assay was removed.
Single-cell RNA Sequencing and BCR Sequencing: Cryopreserved PBMCs were thawed and the dead cell removal kit (Myltenyi Biotec) was applied to sort out dead cells.To enrich B cells, PBMCs were stained with anti-CD19 MicroBeads (Miltenyi Biotec) and loaded on a MACS Column that is placed in the magnetic field as per the manufacturer's instructions.B cells from each sample were then labeled with hashtag multiplexing antibodies (BioLegend) according to the manufacture's protocol and pooled together.Chromium Next GEM Single Cell 5' Reagent Kits v2 (10x Genomics) were employed to perform the multi-omics profiling of enriched B cells.Briefly, stained cells were loaded onto the Chromium Next GEM Chip K, Gel Beads-in-emulsion (GEMs) were generated, and cells were lysed followed by the capture of poly-adenylated mRNA onto beads.The transcripts were reverse transcribed inside each GEM, and the barcoded fulllength cDNA was PCR-amplified.Then, the amplified cDNA from mRNA and from cell surface hashtags were separated by size selection for generating V(D)J, 5ʹ gene expression libraries, and hashtag libraries, respectively.Finally, the constructed libraries were sequenced on the paired-end 150 bp Novaseq platform (Illumina).
Single-Cell RNA-Seq Data Processing: The sequenced data were aligned and quantified using the CellRanger multi pipeline (version 7.0.0,10x Genomics) against the GRCh38 human reference genome.Cells from each demultiplexed sample were first filtered based on two metrics: 1) the number of detected genes per cell must be between 200 to 5000; 2) the proportion of mitochondrial gene counts (UMIs from mitochondrial genes / total UMIs) must be <10%.Then, the gene expression data were normalized using Seurat sctransform. [44]To perform batch effect correction, the Seurat v4 anchor-based integration workflow was used with the default parameter setup. [45]Finally, the "integrated" data assay was reduced to two dimensions using UMAP for visualization, with 30 computed PCs as input.The unique cell barcodes of filtered cells were used to match and identify the BCR V(D)J gene sequences from the annotated contig files.
Ingenuity Pathway Analysis: Ingenuity Pathway Analysis (IPA, QIA-GEN) was used to understand the underlying signaling pathways. [33]Here, DEGs distinguishing each sample, the corresponding fold change value, p-value, and adjusted p-value of each gene were loaded into the dataset.Ingenuity knowledge base (genes only) was used as reference set to perform Core Expression Analysis.B cell-related signaling was selected from identified top canonical pathways to represent major functional profiles of each sample.The z-score was used to determine activation or inhibition level of specific pathways.
Statistics: All the statistical analyses were performed with Prism 9 (GraphPad).Two-sided Mann-Whitney test was used to compare the specific observations between two groups.A p-value <0.05 was considered statistically significant.Scatter or bar plots show means ± SEM.

Figure 1 .
Figure 1.High-plex immuno-serolomic assay design and titration test.a) Schematic workflow.The first PDMS "microfluidic patterning chip" with 50 parallel microchannels was placed on the poly-l-lysine coated glass slide (PLL slide), then purified antibodies or SARS-CoV-2 recombinant antigens flowed into the microchannels.After removing the first PDMS, the second "microfluidic test chip" was placed on the same PLL slide.Then, serum samples were added, and the captured proteins or SARS-Cov-2 binding antibodies were detected via a surface-bound immune-sandwich assay.Finally, scanning fluorescent images were obtained and analyzed.b) Photographic image of the first PDMS microchip.c) Integrated device design after overlaying the first and second PDMS.The protein and SARS-CoV-2 serology panels evaluated in this work are listed on the right side.d) The titration curves of US SARS-CoV-2 serology standard (Frederick National Laboratory).e) The titration curves of SARS-CoV-2 anti-spike/RBD antibodies.f) The titration

Figure 2 .
Figure 2. Performance validation and assay for 1012 patient samples.a) Representative fluorescent image of high-plex immuno-serolomic assay in the measurement of 48 serum samples in a single run.Red and yellow signals represent the plasma protein and SARS-CoV-2 IgG Ab, respectively.Enlarged

Figure 3 .
Figure 3. Vaccination-induced immunological functional protein response in each patient and donor group.a-d) Correlation matrices of the 35 proteins evaluated in our high-plex immuno-serolomic assay panel for Non-Heme cancer patients pre-vaccination (a) and post-vaccination (c), or in autoimmune patients with B cell depletion therapy pre-vaccination (b) and post-vaccination (d).Only significant correlations (<0.05) are represented as dots.Pearson's correlation coefficients from comparisons of protein concentrations across all the patients in a specific group are visualized by color intensity.Proteins were ordered by hierarchical clustering.e-g) Comparisons of the average concentration of each functional protein category between pre-and post-vaccination in healthy donors (e), Non-Heme cancer patients (f), and autoimmune (B cell-depleted) patients (g).h-j) Comparisons of the average concentration of proteins regulating pro-inflammatory pathways (h),effector T cells (i), and endotheliopathy (j) between pre-vaccination and different time points post-vaccination.Visit 1, two weeks after the first vaccine dose; Visit 2, 0-3 days before the second vaccine dose; Visit 3, two weeks after the second vaccine dose.p-Values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).ns, not significant.Scatter plots show means ± SEM.

Figure 4 .
Figure 4. SARS-CoV-2 natural infection and vaccination induce different magnitudes of pro-inflammatory signatures, correlating to the level of care for patients.a) Unsupervised clustering of 302 patient serum samples based on the concentrations of the 35-protein panel.UMAP is used to visualize the data.b) UMAP representation split by patient conditions.Samples from COVID-19 infected patients are localized in cluster 2. c) Differentially expressed proteins that define each cluster.d) Comparisons of the average concentration of proteins regulating pro-inflammatory and T cell effector pathways between different patient/donor groups.e,f) Correlation matrices of the 35-protein panel for e) Heme COVID+ patients and f) Non-Heme cancer COVID+ patients.Only significant correlations (<0.05) are represented as dots.Pearson's correlation coefficients from comparisons of protein concentrations across all the patients in specific groups are visualized by color intensity.Proteins were ordered by hierarchical clustering.g) ROC curve for level-of-care prediction based on the pro-inflammatory protein concentrations of COVID-19-infected patients.A binomial logistic regression was used to fit the model, and a stratified fivefold cross-validation was implemented to compute the ROC and AUC.h,i) Comparisons of the concentration level of specific proteins between COVID-19 infected groups and vaccinated groups.p-values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).Scatter plots show means ± SEM.ICU, Intensive Care Unit; ROC, receiver operator characteristic; AUC, area under the curve.UMAP, Uniform Manifold Approximation, and Projection.

Figure 5 .
Figure 5. Vaccination-induced circulating IgG antibody response in each patient and donor group.a,b) Comparisons of the concentration of a) antispike and b) anti-RBD IgG binding against wild type or other variants between different time points post-vaccination.c-e) Correlation matrices of the 10 SARS-CoV-2 serology panels at c) Visit 1 and Visit 3 in healthy donors (c) d) Non-Heme cancer patients and e) autoimmune (B cell-depleted) patients.Only significant correlations (<0.05) are represented as dots.Pearson's correlation coefficients from comparisons of IgG antibody concentrations across all the patients in a specific group are visualized by color intensity.Antibodies are listed in order of the variants' appearance.f) Comparisons of the IgG antibody response rates between different time points post-vaccination in each group.g) Comparisons of the IgG antibody response rates between patient/donor groups at each timepoint post-vaccination.Visit 1, two weeks after the first vaccine dose; Visit 2, 0-3 days before the second vaccine dose; Visit 3, two weeks after the second vaccine dose.P values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).Scatter plots show means ± SEM.

Figure 6 .
Figure 6.COVID-19-infected hematologic malignancy patients exhibit limited RBD-specific binding.a) Comparison of the SARS-CoV-2 nucleocapsid (N) protein binding between COVID-infected and post-vaccinated groups.b) Comparisons of the concentration of anti-spike and anti-RBD binding against wild type and subsequent variants between COVID-infected and post-vaccination groups.c) Comparison of the IgG antibody response rates between COVID-infected patients with hematological cancer or other types of cancer.d) Correlation matrices of the 10 SARS-CoV-2 serology panels in COVIDinfected Heme patients and Non-Heme cancer patients.Only significant correlations (<0.05) are represented as dots.Pearson's correlation coefficients from comparisons of IgG antibody concentrations across all the patients in specific groups are visualized by color intensity.Antibodies are listed in order of the variants' appearance.e,f) SARS-CoV-2 IgG binding antibody titers in ICU or after recovery from one Heme COVID+ e) and one Non-Heme cancer COVID+ patient f).p-Values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p< 0.001, **** p < 0.0001).Scatter plots show means ± SEM.BAU, Binding Antibody Units; ICU, Intensive Care Unit.

Figure 7 .
Figure 7. Single-cell RNA and BCR sequencing identify low clonotype diversity and functional deficiency in B cells from the patient with hematologic cancer.a) The percentage of top 5 clonotypes in each sample.b) UMAP clustering of B cells from COVID-infected Heme patients and Non-Heme cancer patient (two time points, in ICU and after recovery) identified 5 distinct subpopulations.c) Dotplot of selected B cell marker gene expression in each cluster.The size of each circle represents proportion of single cells expressing the gene, and the color shade indicates normalized expression levels.d) Cell proportion of each sample in each identified cluster.e) Comparison of expression levels of marker genes in each sample.f) Corresponding canonical pathways regulated by the highly differentially expressed genes between Heme patients versus Non-Heme cancer patients (ICU).Pathway terms are ranked by -log 10 (P value).The side-listed gene names represent top 6 symbolic molecular markers related to the pathway.A statistical quantity, called z score, is computed and used to characterize the activation level.z score reflects the predicted activation level (z < 0, downregulated; z > 0, upregulated; z ≥ 2 or z ≤ −2 can be considered significant).P values were calculated with a two-tailed Mann-Whitney test.(* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).Bar plots show means ± SEM.BCR, B cell repertoires; ICU, Intensive Care Unit.UMAP, Uniform Manifold Approximation, and Projection.

Table 1 .
Comparison of our immuno-serolomic assay with other methods.