Exploratory analysis to identify the best antigen and the best immune biomarkers to study SARS-CoV-2 infection

Background Recent studies proposed the whole-blood based IFN-γ-release assay to study the antigen-specific SARS-CoV-2 response. Since the early prediction of disease progression could help to assess the optimal treatment strategies, an integrated knowledge of T-cell and antibody response lays the foundation to develop biomarkers monitoring the COVID-19. Whole-blood-platform tests based on the immune response detection to SARS-CoV2 peptides is a new approach to discriminate COVID-19-patients from uninfected-individuals and to evaluate the immunogenicity of vaccine candidates, monitoring the immune response in vaccine trial and supporting the serological diagnostics results. Here, we aimed to identify in the whole-blood-platform the best immunogenic viral antigen and the best immune biomarker to identify COVID-19-patients. Methods Whole-blood was overnight-stimulated with SARS-CoV-2 peptide pools of nucleoprotein-(NP) Membrane-, ORF3a- and Spike-protein. We evaluated: IL-1β, IL-1Ra, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12p70, IL-13, IL- 15, IL-17A, eotaxin, FGF, G-CSF, GM-CSF, IFN-γ, IP-10, MCP-1, MIP-1α, MIP-1β, PDGF, RANTES, TNF-α, VEGF. By a sparse partial least squares discriminant analysis we identified the most important soluble factors discriminating COVID-19- from NO-COVID-19-individuals. Results We identified a COVID-19 signature based on six immune factors: IFN-γ, IP-10 and IL-2 induced by Spike; RANTES and IP-10 induced by NP and IL-2 induced by ORF3a. We demonstrated that the test based on IP-10 induced by Spike had the highest AUC (0.85, p  <  0.0001) and that the clinical characteristics of the COVID-19-patients did not affect IP-10 production. Finally, we validated the use of IP-10 as biomarker for SARS-CoV2 infection in two additional COVID-19-patients cohorts. Conclusions We set-up a whole-blood assay identifying the best antigen to induce a T-cell response and the best biomarkers for SARS-CoV-2 infection evaluating patients with acute COVID-19 and recovered patients. We focused on IP-10, already described as a potential biomarker for other infectious disease such as tuberculosis and HCV. An additional application of this test is the evaluation of immune response in SARS-CoV-2 vaccine trials: the IP-10 detection may define the immunogenicity of a Spike-based vaccine, whereas the immune response to the virus may be evaluated detecting other soluble factors induced by other viral-antigens.


Introduction
COronaVIrus Disease-2019 (COVID-19) pandemic is caused by the novel coronavirus designated as severe acute respiratory syndrome coronavirus (SARS-CoV)-2 [1] belonging to β-Coronovavirus genus. Its genome contains 14 open reading frames (ORFs) and encodes 27 different proteins, including spike (S), envelope (E), membrane (M) and nucleocapsid (NP) proteins [2]. The majority of people with COVID-19 develop mild (40%) or moderate (40%) symptoms, 15-20% develop a severe disease needing oxygen support and 5% have a critical disease with complications such as respiratory failure, acute respiratory distress syndrome (ARDS), sepsis and septic shock, thromboembolism, and/or multi-organ failure [3][4][5]. SARS-CoV-2 infection induces an immune response in the host characterized in severe COVID-19 cases by a decrease of lymphocytes number and a great increase of cytokines [6]. Currently, the mechanisms that lead to disease exacerbation remains largely undetermined. Thus, there is an urgent need to improve our understanding of the immunology of this disease to find correlate of protection or to monitor the course of the infection.
In this study, we analyzed in a whole-blood-cytokine platform, the best approach to evaluate the SARS-CoV-2-T-cell response to the structural (N, S and M) [19] and accessory protein (ORF3a) [23,24] of SARS-CoV-2. We aimed to identify (i) the best antigen to induce the SARS-CoV-2 specific T-cell response; (ii) the best subset of biomarkers to identify COVID-19-patients.

Identification of plasma biomarkers for distinguishing COVID-19 from NO-COVID-19-individuals
Demographical and clinical information of the enrolled subjects are shown in Table 1. We stimulated the wholeblood of with SARS-CoV-2-specific peptide pools of NP (NP Pool1 and NP Pool2), Membrane, ORF3a, and Spike. Then, we evaluated by luminex the plasma level of 27 analytes. Among the different stimuli, the Spike and NP Pool1 peptides, belonging both to SARS-CoV-2 structural proteins, were the most recognized antigens by COVID-19-patients (Table 2). Spike peptide pool was the most immunogenic stimulus, modulating the highest number of cytokines, chemokines and growth factors ( Table 2).
Applying a supervised sPLS-DA we aimed to identify the most important soluble factors, analyzing at the same time the luminex results and the different SARS-CoV-2-peptides pool stimulations (Fig. 1). Although the difference was not fully discriminative, the distribution of COVID-19 and NO-COVID-19-subjects in the space were quite separated (Fig. 1A). Evaluating the loading weights of each selected variable on each component, the mean level of production for the most important selected variables was maximal in COVID-19-patients within the component 1 (Fig. 1B), whereas the mean level of production was maximal in the NO-COVID-19 within the component 2 (Fig. 1C). Overall, the accuracy of the classification was high for both components (> 92%) (data not shown). Since the component 1 was represented mainly by factors upregulated in COVID-19-patients, we focused on this component. Then, we identified the six variables with the highest weight in the construction of component 1 (Fig. 1B-C): IL-2, IFN-γ and IP-10 induced by Spike, regulated on activation, normal T cell expressed and secreted (RANTES) induced by NP Pool1, IP-10 induced by NP Pool2, and IL-2 induced by ORF3a stimulation (hereafter referred as Spike IL-2, Spike IFNγ, Spike IP-10, NP Pool1 RANTES, NP Pool2 IP-10, and Keywords: SARS-CoV-2, COVID-19, Biomarkers, T-cell, Immunity, IP-10, Whole-blood, Immune response, Spike, IFN-γ Page 3 of 16 Petruccioli et al. J Transl Med (2021) 19:272 ORF3a IL-2). Next, we evaluated, within the six variables signature associated to COVID-19, the proportion of response to each stimulus: IP-10 proportions induced by Spike and NP Pool2 were the most represented in COVID-19-patients (Fig. 2).

Comparison of AUC of the six immune factors
The selected six immune factors of component 1, as expected, had significant quantitative higher levels in COVID-19 compared to controls for: IL-2, IFN-γ, IP-10 induced by Spike (p = 0.0018; p = 0.0175; p < 0.0001; respectively), NP Pool1 RANTES (p = 0.001), NP Pool2 IP-10 (p = 0.027) and ORF3a IL-2 (p = 0.039) ( Fig. 3; Table 2). ROC curve analysis of these factors showed that the highest AUC was related to IP-10 Spike (AUC 0.85; p < 0.0001; Fig. 4). Then, we generated a combined-test based on the six immune factors previously selected (Fig. 4). The combined-test showed a significantly further increase of AUC (AUC 0.94; p < 0.0001) compared to the AUCs of the other single tests except for IP-10 and IL-2 induced by Spike (Fig. 4B). Since IP-10 Spike test showed the highest AUC, we compared it with all the other AUCs and we did not find any significant differences among the different tests (Fig. 4C).

Impact of the clinical characteristics of patients on the COVID-19 signature
We investigated if any clinical characteristic of COVID-19-patients had an impact on the level of the six selected variables (Table 3). We found that age (p = 0.001), cortisone (p = 0.042) and severity of the disease (p = 0.015) had a significant impact on NP Pool1 RANTES. NP Pool2 IP-10 was modulated by symptoms (p = 0.036), IgM index (p = 0.003) and IgM score (p = 0.017). Finally, ORF3a IL-2 was modulated, by the number of days from the symptoms onset (p < 0.0001) and IgM index (p = 0.038). Similarly, Spike IL-2 was modulated by number of days from the symptoms onset (p = 0.001), IgM index (p = 0.028) and IgM score (p = 0.036).
Differently, Spike IFN-γ and Spike IP-10 were not significantly modulated by any of the clinical characteristics considered.

Evaluation of IP-10 in different cohorts of COVID-19-patients
We demonstrated that Spike IP-10 had the highest AUC (0.85, p < 0.0001; Fig. 4) and that the clinical characteristics of the COVID-19-patients did not affect IP-10 production (Table 3). Based on these results, we further evaluated the production of IP-10 in a new study population of NO-COVID-19 and COVID-19-patients stratified according to the hospitalization status and symptoms onset (Table 4). To verify the consistency of our findings, we used a different experimental setting: IP-10 was detected using a routine approach as the enzyme-linked immunosorbent assay (ELISA) and Spike peptides were obtained from a commercial source (Miltenyi). IP-10 production significantly increased after Spike stimulation in the cohort A of "hospitalized COVID-19-patients enrolled between 1 and 14 days after symptoms onset" (p = 0.0014) and in the cohort B of "not hospitalized COVID-19-patients" (p = 0.0002), (Fig. 5A-B). ROC analysis demonstrated a high and significant AUC in cohort A and cohort B (AUC: 0.8167; p = 0.0020; AUC: 0.9056; p = 0.0005)    ( Fig. 5C-D). The specificity of the test to identify COVID-19 was 88.89% for both COVID-19-cohorts; the sensitivity was 66.67% for cohort A and 70% for cohort B (Fig. 5C-D).

Discussion
In this study, by a multivariate exploratory analysis we found the best antigen and the best biomarker to distinguish COVID-19-and NO-COVID-19-individuals. To achieve our goal, we used a whole-blood-platform [10] with a luminex read-out. By the sPLS-DA, we identified a COVID-19 signature based on six immune factors.
Our results showed that Spike IFN-γ, Spike IP-10, Spike In fact, to corroborate the reproducibility of our results, we performed a validation study testing Spike peptides from a commercial company and using a more feasible routine approach such the IP-10 ELISA. We demonstrated that IP-10 had a good accuracy to identify hospitalized COVID-19-patients in the first two weeks after symptoms onset and not-hospitalizedpatients enrolled 35-100 days after symptoms onset. IP-10 is a chemokine mainly secreted by monocytes, fibroblasts and endothelial cells in response to IFN-γ that attracts activated T-cells to foci of inflammation [25]; it has already been described as a potential biomarker for other infectious disease, such as tuberculosis and HCV [26][27][28][29][30] and may be easily measured in condition of immune-depression [30]. In acute COVID-19-patients, IP-10 production is a promising surrogate marker of impaired immune responses [13]. In our study IP-10 production induced by Spike stimulation was the only parameter not affected by any clinical characteristics. We reported that IP-10 identified SARS-CoV-2 infection in the acute phase of disease and in COVID-19-recovered subjects. This result has a double scientific implication. Firstly, it supports the specificity of the immune response to viral-peptides in different clinical conditions; secondly, it suggests a possible application of the "IP-10 and Spike whole-blood test" as a potential additional tool for diagnostic and immune response evaluation of COVID-19-patients during the acute phase of the disease. These findings are in agreement with other cytokine releasebased tests applied for the diagnosis of several infectious diseases [31][32][33][34]. Moreover, an additional possible application of this whole-blood based cytokine assay is the evaluation of immune response in SARS-CoV-2 vaccine trials. In this context, the IP-10 detection may define the immunogenicity of a Spike-based vaccine, whereas the immune response to the virus infection may be evaluated detecting other factors as RANTES induced by NP. Previous reports focused on the pre-existing immune response to SARS-CoV-2 in the general population, demonstrating that ORF1-specific T-cells were detected in SARS-CoV-2 unexposed donors [19,35]. Differently, in recovered COVID-19-subjects, the T-cells mainly recognized the structural proteins [19]. In our study, we observed few modulations of immune factors among COVID-19 and NO-COVID-19 individuals in response to the peptides of accessory protein ORF3a; these data indicate that both groups have a similar immune response and suggest a minor contribution of ORF3a in the immune-specific response in acute-hospitalized COVID-19-patients. In line with previous evidence, the majority of immune modulations concerned to stimulations with structural proteins such as NP and Spike. As already reported [10,16,17] we observed a production of both inflammatory and anti-inflammatory cytokines and chemokines in response to the structural protein of SARS-CoV-2.
More than 90% of seroconverters COVID-19-individuals shows an immunological memory of T-cell compartment [36] and antibody response, for several months after infection [36,37]. However, we need more longitudinal studies to understand exactly if the immune memory response remains stable over time. Considering that the early prediction of disease progression could be useful to assess the optimal treatment strategies, an integrated knowledge of the T-cell and antibody response lays the foundation to develop biomarkers to monitor the course of COVID-19 disease.
The limits of the present study are related to the low amount of patients evaluated. However, five different viral antigens and 27 markers were concomitantly evaluated and validated in different cohorts making the here generated evidence robust. Moreover, in the control group of NO-COVID-19 individuals, it would have been useful to include subjects with acute respiratory diseases, as Influenza. Indeed, it has been demonstrated that serum or plasma IP-10 is increased in several respiratory infections, as tuberculosis [26,38] or influenza [39]. However, in 2020 and 2021 so far, in Europe the Influenza Virus positivity in sentinel specimens remained below the epidemic threshold due to the use of massive vaccination, masks and lockdown rules [40]. Further studies will help understanding if the coinfection of COVID-19 and other acute infectious diseases may have an impact of the SARS-CoV-2-specific IP-10 signature. Nevertheless, in a recent study [10] we showed that NO-COVID-19 patients with respiratory disease such as tuberculosis and bacterial pneumonia did not show IFN-γ-specific response to Spike stimulation. Similarly, in the present study, we did not find IP-10-specific response to Spike in NO-COVID-19 individuals. Interestingly, the NO-COVID-19 group included seven subjects with active tuberculosis under therapy and 5/7 in the acute phase of the disease as they were enrolled within 7 days of diagnosis and of starting the anti-TB specific therapy. These evidences support the specificity of our data even if generated with a low number of control patients.
In conclusion, we demonstrated the potential application of a whole-blood based platform that allowed the selection of the best antigen and best read out to evaluate the immune response to SARS-CoV-2 infection. We also identified IP-10 detection induced by Spike stimulation, as a good in vitro setting to distinguish COVID-19 from NO-COVID-19-individuals.

Peptide pools and stimuli
For the exploratory study, SARS-CoV-2 peptide pools of 15-mers (55 peptides) at 2 µg/mL, covering the whole NP (Pool1 and Pool 2), M, ORF3a proteins and 40.5% of the Spike protein, were used as reported [42]. For the validation study, SARS-CoV-2 PepTivator ® Peptide Pool of the Spike protein at 0.1 µg/mL (Miltenyi, Biotec, Germany) were used. Stimulated whole-blood was overnight incubated at 37 °C, 5% CO 2 , plasma was collected and stored at − 80 °C until used.

SARS-CoV-2 serology
SARS-CoV-2 specific IgM and IgG levels were measured by ELISA according to manufacturer's instructions (DIESSE Diagnostica Senese S.p.a., Monteriggioni, Italy). The ratio between the optical density (OD) of the sample and that one of the cut-off reagent (index) was calculated. The samples were scored positive (index > 1.1), doubtful (index between 1.1 and 0.9) and negative (index < 0.9).

Cytokines, chemokines and growth factors evaluation
Bio-Plex Pro Human Cytokine 27-plex Assay panel and the MagPix system (Bio-Rad, Hercules, CA, USA) were used to evaluate in harvested plasma: cytokines, chemokines and growth factors (IL-1β, IL-1RA, IL-2, In the validation study, IP-10 was measured in plasma using Human CXCL10/IP-10 Quantikine ELISA (R&D Fig. 5 IP-10 modulation in a second cohort of COVID-19 patients. IP-10 production was measured by ELISA in plasma collected after stimulating whole-blood with Spike peptides. A, B The horizontal lines represent the median of IP-10 production; statistical analysis was performed using the Mann-Whitney test, and p value was considered significant when ≤ 0.05. C, D The graphs represent the AUCs obtained by the ROC analysis comparing the NO-COVID-19 subjects with three cohorts of COVID-19 patients. A, C Hospitalized COVID-19 patients enrolled 1-14 days after symptoms onset. B, D Not-hospitalized COVID-19 patients enrolled 35-100 days after symptoms onset. IP interferon-γ inducible protein; CI confidence interval; AUC area under the curve Chi-squared test for categorical variables; receiver-operator characteristic (ROC) analysis for evaluating the area under the curve (AUC) and the diagnostic performance; Spearman Rank Correlation to measure the strength of association between two variables and the direction of the relationship (positive or negative). We performed a multivariate exploratory analysis, sparse partial least squares discriminant analysis (sPLS-DA), to identify the most important soluble factors discriminating COVID-19-from NO-COVID-19-individuals. The sPLS-DA performed a variables reduction, generating latent components to synthetize the data information. For the sPLS-DA analysis, we considered in the model all the 135 analytes simultaneously (5 different stimuli, 27-factors each) limiting the components construction to the first 20 most important variables identified by the method. Data were analyzed with the R-package MixOmics. We performed a logistic regression analysis to evaluate the potential ability of a minimal subset of variables to classify COVID-19 from NO-COVID-19-patients; AUC and p values were reported.