Even though four years have passed since the outbreak of the COVID-19 pandemic, it continues to pose a global health challenge. The clinical features of SARS-CoV-2 infected patient are well documented, ranging from asymptomatic, mild, and moderate to critically ill patients who present acute respiratory distress syndrome (ARDS), septic shock, metabolic acidosis, coagulopathy, organ failure, and eventually death (1–3). The fate of the COVID-19 disease is decided by T cells, which have been an important component of the cell-mediated immune responses (4–7). Despite a general reduction in the lymphocyte count, the CD4+ and CD8+ T lymphocyte subsets are hyperactivated (1). In mild COVID-19, SARS-CoV-2 induces a memory and Th1 phenotype with CD4+ T cells expressing higher levels of IFN-γ, IL-2 and TNF and a strong response of CD8+ T cells with high expression of granzyme A, B, and perforins (8). By contrast, in severe COVID-19 patients, the immune responses are marked by exhausted T cells displaying high levels of terminal differentiation/senescence markers (immune checkpoints like Programmed Death-1 (PD-1), T cell immunoglobulin and mucin domain-containing protein 3 [TIM-3), T cell immunoreceptor with Ig and ITIM domains (TIGIT)], with proinflammatory CD25+ CD8+ T cells of reduced cytotoxic potential dominating the T cell profile (9, 10). The immunological profile in ARDS is further exacerbated by the cytokine storm, which mainly arises from deficiencies in immune regulatory mechanisms. Regulatory T cells (Tregs) are central in maintaining the immune homeostasis due to their immunosuppressive potential of inhibiting excessive immune responses. Interestingly, despite a significant expansion of Tregs in severe cases compared to mild ones, their functionality is deranged. They exhibit similarities with tumor-infiltrating Tregs, characterized by a distinct transcriptional signature marked by upregulation of FoxP3, overexpression of both suppressive effectors (IL-10) and proinflammatory molecules (IL-32), and apoptotic features (11–13). As a feedback loop, components of the proinflammatory storm (IL-6 and IL-18) contribute to and potentiate this disturbed Treg phenotype (12).

The up-listed dysfunctions in immune cells characterizing the severe COVID-19 patients are associated with alterations in immune checkpoints (ICPs). ICPs comprise paired receptor-ligand molecules with co-stimulatory, inhibitory or dual effects on immune activity, defense, regulation or self-tolerance (14, 15). The up-regulation of inhibitory ICP expression on various immune cells, such as T and B lymphocytes and NK cells, and their increased interaction with specific membrane or soluble ligands in tissue endothelium, antigen-presenting cells (APCs), and plasma are tightly linked to their exhaustion and depletion in a variety of viral infections, including COVID-19 (16, 17). Of note, soluble isoforms of immune checkpoints (sICPs) may result from the protease-mediated shedding of the membrane ICPs(mICPs) or by alternative mRNA splicing, and likely form a complex circulating immune regulatory system (18, 19). On one hand, the blood levels of sICPs may reflect the extent of expression of their membrane-bound counterparts, but interacting with their specific membrane receptors or ligands, these ICPs may potentiate or inhibit the activity of various immune cells. Indeed, a storm of sICPs was associated with severe COVID-19 and their levels on hospital admission were suggested they might serve as biomarkers of disease progression and outcome (19, 20). Thus, understanding the profile of soluble immune checkpoints in COVID-19 patients with distinct forms of disease severity, ranging from mild to moderate and severe, and their potential interlinks with multiple inflammatory mediators would provide a broad picture of the complex immune processes underlying the disease pathogenesis, and potentially predict the individual pattern of evolution and outcome. These observations would be crucial in the management of the SARS-CoV-2 infected patients and for applying the most effective therapeutic strategy.

Based on these considerations, we set up the evaluation of a novel sICPs panel composed of commonly investigated sICPs (including sCD40 with its ligands sCD40L and Gal-9; sCD25; sCD27) as well as molecules recently found and relatively underexplored (such as sCD30, s4-1BB, sTIM-1, sB7-H3 and sCD163), in relation to previously investigated inflammatory mediators (3) and clinical characteristics of SARS-CoV-2-infected patients (for instance, the vaccination status, the presence of comorbidities, and the administered therapy during hospitalization). Here we have identified distinct baseline sICPs profiles at hospital admission characterizing the 3 forms of disease severity (mild, moderate and severe), the Delta vs. Omicron variant infections, and the non-survivors’ group of COVID-19 patients. While sCD40 is central in all defined profiles, the variations in distribution of four sICPs (sCD30, s4-1BB, sTIM-1, sB7-H3) and their interactions with additional paraclinical and clinical variables help distinguishing between the different subgroups of COVID-19 patients. Applying an artificial neural network analysis to our complex patients’ data, retrospectively revealed the importance of administering targeted anti-inflammatory therapy (Tocilizumab or Anakinra) in defining the outcome of patients with high baseline of sTIM-1 levels.

Immune checkpoints receptors and ligands are molecules with stimulatory, inhibitory or dual functions in immune cells which maintain the immune surveillance and prevent excessive activation, while ensuring the efficient and effective defense responses against pathogens or cancer cells (14, 26).

Most of the soluble isoforms of immune checkpoints (sICPs) are believed to result from the proteolytical cleavage of membrane-bound proteins or by alternative splicing of mRNA (18) with the role to compete with and counteract the activity of the membrane-bound forms, detailed in Supplementary Table 11 . The clinical relevance of sICPs remains largely unknown, despite emerging evidence on their dysregulation in patients with viral infections (including the SARS-CoV-2 infection) (17, 20), autoimmune diseases (27) or cancer (18, 28). Resulting from the proteolytical shedding of their membrane bound counterparts upon inflammation, sICPs function, in a first instance, as a mirror of hyperinflammation and activity of metalloproteinases, but also reflect the profile of distinct subsets of immune cells and their checkpoint surface expression. Additionally, while some sICPs maintain the biological activity of their membrane isoforms, others act as molecular decoys or form soluble complexes by interacting with their soluble ligands or receptors (18). Understanding the distinct profile of sICPs associated with various forms of infection is key in decoding the molecular mechanisms of disease pathogenesis which can further provide crucial insights for devising the most effective therapeutic approach.

Here we analyzed the variations in soluble checkpoints receptors and ligands in relation to COVID-19 severity, SARS-CoV-2 variant infection and mortality and compared them to standardized proinflammatory molecules and additional hematological and serological biomarkers. We included in our analysis a list of commonly investigated sICPs (sCD40 with its ligands sCD40L and Gal-9; sCD25; sCD27) as well as molecules newly found and relatively less studied in the literature, such as sCD30, s4-1BB, sTIM-1, sB7-H3 and sCD163. We will continue to discuss the main findings of our study related to each mentioned soluble molecule.

The CD40-CD40L interaction is crucial for immune regulation and activation. This interaction triggers bidirectional signaling of both antigen-presenting cells (APCs, such as dendritic cells, macrophages, B cells) and T cells, leading to upregulation of co-stimulatory molecules and cytokine secretion. Indeed, activation of CD40 on dendritic cells promotes their maturation and enhances their B7 ligands expression, while CD40 signaling in monocytes and macrophages induces the secretion of pro-inflammatory cytokines such as IL-1β and IL-6 (29, 30). CD40 on B cells is of paramount importance for proper activation of B cells and the generation of humoral immune responses. CD40L (also known as CD154) was initially identified on stimulated CD4+ T cells, but now it is known to have a broader immune cell expression, including stimulated CD8+ T cells, mast cells, basophils, and platelets (31, 32). The membrane expression of CD40L on platelets is linked to their activation and further stimulates leukocytes and endothelial cells, by direct engagement of CD40L to CD40 and P selectin to its ligand PGSL-1, to produce proinflammatory cytokines and adhesion molecules (33). Importantly, direct binding of SARS-CoV-2 virions to platelets initiates multiple signaling intracellular pathways, including the AKT-PKC cascade which upregulates the platelet membrane-bound CD40L expression and the release of soluble CD40L molecules stored in platelet granules upon their activation (34, 35). Both soluble isoforms of CD40 and CD40L may also be generated by the shedding of their extracellular domains through the action of specific metalloproteinases such as ADAM10 and ADAM17 (also known as TNF-alpha converting enzyme, TACE) activated upon hyperinflammation conditions (36, 37). Thus, the CD40-CD40L interaction is believed to significantly contribute to monocyte activation, cytokine storm and hypercoagulation associated with severe COVID-19 (29). We here reported increasing levels of sCD40 dependent on disease severity, from mild to moderate and severe COVID-19. In our cohort, sCD40 showed a positive correlation with the pro-inflammatory biomarker IL-6 in mild cases, correlation that diminished in moderate and severe cases, suggesting multiple intricated molecular mechanisms despite increased levels of both biomarkers. Importantly, the non-survivors’ group presented at hospital admission a high association between sCD40 with both MCP-1 and sTREM-1, two well-known biomarkers of monocytes and macrophage overactivation during sepsis and hyperinflammation (3). Interestingly, despite no significant correlation between sCD40L and sCD40 in the non-survivor group, sCD40L showed a high association with HGF and moderate with platelet number and fibrinogen, possibly pointing towards a sepsis hypercoagulable profile characterized by activated platelets releasing higher amounts of both HGF and soluble CD40L. Indeed, this observation is in line with the previously described role of sCD40L as biomarker of unfavorable outcome in patients with severe sepsis or septic shock. These associations in the non-survivors’ group may thus indicate a mechanism of interaction between platelets and monocytes that resulted in the release of high amount of soluble CD40. Additionally, only sCD40 and not sCD40L showed a linear increase pattern from mild to moderate and severe cases (sCD40L did not differentiate between the moderate and severe groups), suggesting additional mechanisms, and possibly, additional cellular contributors in generating the pool of soluble CD40, such as fibroblasts, epithelial and endothelial cells stimulated by hyperinflammatory cytokines characterizing the severe patient cohort.

CD30, another member of the tumor necrosis factor receptor superfamily, is generally strongly expressed in classical Hodgkin lymphoma (cHL) and anaplastic large cell lymphoma, a rare subtype of NHL, with a variable expression observed in T-cell, NK cell or B cell lymphomas (38, 39).

In non-malignant scenarios, CD30 is mainly expressed on a limited subset of activated T cells, B cells, NK cells and regulatory T cells providing pro-survival and anti-apoptotic roles (39, 40). In inflammatory diseases, the extracellular region of the membrane-bound CD30 is easily cleaved by protein hydrolases into soluble fragments (sCD30) (41), thus generally being considered a marker of a subset of activated effector T cells that produce high amount of IFN-γ and IL-5 that exhibit enhanced help activity for B cell Ig production and maintains the Th1/Th2 balance state (42, 43). These CD30+ CD4+ T cells are essential also for sustaining the follicular germinal center responses with the generation of B cells with antigen receptors of high affinity (44). In our cohort, a strong association between sCD30 and CD40 was observed only in mild cases suggesting proper T-B cell responses, while this association reduced in moderate cases and even more in severe patients. However, recent studies have highlighted CD30 as a marker of activated effector regulatory T cells, which are critical for the regulation and/or suppression of memory T and B cells responses (45). These observations are important since, despite a significant increase in the serum levels of both sCD40 and sCD30, sCD30 showed a clear dissociation from sCD40 in non-survivors and a clear association with sCD25, a soluble fragment released from the membrane bound CD25 which is highly expressed on regulatory CD25+ FOXP3+ Tregs, suggesting an increase in the Tregs population within those individuals. These observations are in line with the previous studies reporting a rising percentage of deranged CD25+ FOXP3+ Tregs among CD4+ T cells in severe compared to mild COVID-19 patients which returned to normal levels only in the recovering individuals (4, 11–13, 46). However, compared to sCD30, sCD25 significantly discriminated between mild, moderate and severe patients, suggesting additional cellular sources. For instance, the proinflammatory CD25+ CD8+ T cells with lower cytotoxic activity are expanded in severe COVID-19 patients, being an important source of sCD25 and cause of delayed clearance of virus (9, 47–49). Interestingly, the initial correlation seen between sCD40 and sCD30 in mild cases was replaced among non-survivors, by the association between sCD40 with s4-1BB and sCD27, the soluble forms of the counterpart surface markers of memory CD8+T cells. These observations taken together, indicate a shift of the immune profile towards Tregs and pro-inflammatory CD8+ T cells which suppress the proper T-B cell responses and delays the virus clearance. Indeed, similar to a previous study (17), we confirmed that both s4-1BB and sCD27 were linked to a higher disease severity rate.

Next, we noticed that, in the non-survivors’ group, Gal-9 markedly associated with all sCD40, sCD27 and s4-1BB. Gal-9 is expressed by a variety of immune cells including T and B cells, Tregs, neutrophils, monocytes, macrophages, dendritic cells and are secreted via non-classical pathways or freely released upon cell death with multiple biological functions depending on the interacting receptor CD40, PD-1, TIM-3, 4-1BB (50–52). In a variety of inflammatory conditions and infectious diseases, Gal-9 suppressed the B cell receptor signaling (53) and was commonly associated with exhausted T cells and impaired cytotoxic NK cells (54). Furthermore, Gal-9 through interaction with PD-1 and TIM-3 inhibits T cell proliferation and induces cell-death (55, 56). We, as others (51), found significantly higher levels of Gal-9 in patients with severe disease (ranging from 5.4 ng/mL to 59.7 ng/mL) compared to mild/moderate cases (ranging from 4.1 ng/mL to 23.4 ng/mL) with the highest values observed in the non-survivors’ subgroup (ranging from 19.0 ng/mL to 59.7 ng/mL) and positive association with pro-inflammatory/sepsis biomarkers (sTREM-1, MCP-1, IL-6, NLR, PLR, VSH) and with neutrophils and eosinophils numbers. This last result was expected, since immune activation following SARS-CoV-2 infections causes Gal-9 shedding from neutrophils and eosinophils (51). Interestingly, Treg cells also express high levels of Gal-9 and the Gal-9/Tim-3 signaling pathway further promotes the induction of Tregs, suggesting additional mechanisms for Gal-9 to exhibit suppressive immune responses (57). Additionally, Gal-9 binding to 4-1BB in a region distinct from the binding site of its natural ligand (4-1BBL), augments 4-1BB signaling and activity in T cells and NK cells to produce proinflammatory molecules (TNF-α, MCP-1) Thus, the Gal-9 induced release of TNF-α and MCP-1 activates metalloproteinases (e.g., ADAM10, ADAM17) which further mediates the ectodomain shedding of 4-1BB from T cells (58) and the scavenger receptor CD163, who’s expression is induced by the immunosuppressive IL-10 on the membrane of M2 macrophages, cells with anti-inflammatory activity (59). Indeed, s4-1BB showed strong associations with MCP-1 and sCD163 at baseline in the non-survivors’ group from our study. However, shedding of 4-1BB further limits the Gal-9 pro-inflammatory activity (58). These observations are important, since the suppressive immune activity of both Gal-9 and sCD163 in association with high sCD40 basal levels were mainly revealed by the neural network models corrected for the therapeutic strategy, age, vaccination status and the presence of comorbidities.

Further, the neural network analysis identified TIM-1 to bring the highest contribution among the investigated checkpoint molecules in discriminating between survivors and non-survivors. Our analysis further revealed important correlations between sTIM-1 with sCD40, MCP-1, Gal-9, sTREM-1 and suPAR. Among non-immune cells, TIM-1 glycoprotein is expressed by lung and kidney epithelial cells where it serves as a phosphatidylserine receptor and thus, as an alternative receptor to ACE-2 for SARS-CoV-2 (60). The soluble form is released from the membrane bound TIM-1 by proteolytical cleavage mediated by ADM-10 and ADAM-17 (61).

Not surprisingly, sTIM-1 and sCD40L strongly correlated in mild cases of infections, since SARS-CoV-2 modulates their cellular expression of infected epithelial cells and platelets (34, 35). However, this correlation was inversed in the non-survivors’ group, possibly indicating the existence of additional cellular sources for sTIM-1, including immune cells such as Th2 cells, NK cells, and regulatory B cells within these individuals (62, 63). Since sTIM-1 still binds phosphatidylserine, it may serve as a negative regulator of cellular TIM-1 (61). For instance, TIM-1 mediates T cell trafficking by directly binding the adhesion receptor P-selectin during inflammation (64). Blocking TIM-1 signaling in B cells enhances the type 1 interferon response within these cells, leading to increased B cell activation and antigen presentation, along with heightened co-stimulation capabilities (62, 63). Importantly, in our patient cohort, all the cases with initially high sTIM-1 levels (over 300 pg/mL) who received specific anti-inflammatory therapy (Tocilizumab or Anakinra) survived, pointing towards the beneficial effect of this therapy in influencing the final outcome of selected patients. The proposed ANN and Random Forest models highlighted the role of soluble TIM-1 and Gal-9 in predicting mortality among COVID-19 patients, effect that was diminished when anti-interleukin therapy was added into these models. This finding suggests that the administration of Tocilizumab or Anakinra may have been particularly beneficial for patients with initially elevated soluble sTIM-1 and Gal-9 levels, increasing their chance of survival. This observation becomes even more relevant when considering the strong association between sTIM-1 and D-dimers, indicating a heightened risk for coagulopathy or thrombotic events in patients with initially elevated sTIM-1 levels. Thus, high sTIM-1 and Gal-9 levels may reflect an intensified inflammatory response, and by modulating the immune response, this targeted therapy could help reduce inflammation, leading to better clinical outcomes for these patients. When comparing the Delta and Omicron variant infections, few differences in the checkpoint ligands’ profile emerged. First, a reduction in the sCD40L and Gal9 levels were noticed in the Omicron variant group, which was associated with a lower mortality rate among affected individuals. Alongside this decrease, there was a concurrent rise in sB7-H3 levels, a newly investigated molecule in literature. sB7-H3 is a functionally active form that is released by the ectodomain shedding of membrane-bound CD276 (mB7-H3) from the surface of monocytes, dendritic cells and activated T cells by various metalloproteinase (65). B7-H3 has been described as a checkpoint molecule with dual function, serving both as a stimulatory factor for T cell activation and IFN-γ production, as well as exhibiting inhibitory effects on T cell proliferation and Th1 differentiation (66–68). This is important since in our analysis, sB7-H3 appeared to play a primarily inhibitory role by limiting overactivation and/or exhaustion of T cells in individuals infected with the Omicron-variant.

Our analysis is however limited to a relatively small sample size and the focus on only ten distinct soluble immune checkpoint molecules. This highlights the need for future studies to cross-validate or ensemble these models on larger and/or different datasets, which could further enhance prediction accuracy and reduce the risk of overfitting. With the aim of exploring a series of less studies ICPs, this study primarily focused on examining stimulatory checkpoint molecules, while including only few inhibitory ones, specifically sTIM-1 and Gal-9. Future studies should also incorporate more well-known inhibitory molecules, such as LAG-3, GITR, PD-1, and CTLA-4, whose alterations are commonly associated with viral infections and COVID-19, such as LAG-3, GITR, PD-1, or CTLA-4 (69). For instance, LAG-3 expressed primarily on exhausted CD8+ T cells, acts by inhibiting T cell proliferation and cytokine production, thereby acting as a compensatory mechanism to excessive immune activation. Elevated levels of soluble LAG-3, along with sTIM-3, s-GITR, sPD1, and sCTLA-4 were consistently higher in patients with severe and critical COVID-19, and these levels were negatively correlated with the absolute counts of CD4+ and CD8+ T cells (17, 19). Importantly, Gal-9, the ligand of TIM-3 mirrored these findings, further validating the importance of our study.

This study is of importance as it provides baseline values for a series of soluble immune checkpoint receptors and ligands for which only limited data exist in literature, and identify the biomarkers associated with disease severity and mortality in COVID-19. For instance, our study confirms the association of Gal-9 and sCD25 with disease severity and identifies sCD40 to play a central role in defining the distinct immune profiles characterizing the patients with mild, moderate and severe COVID-19 at hospital admission (see the diagram in Figure 9 ). Our study also points out the differences in the serum profiles of soluble immune checkpoint ligands between Delta and Omicron variant infections, and identifies associations of sCD40L and Gal-9 with sB7-H3 and sCD163 levels in subjects infected with the Omicron variant ( Figure 9 ). Despite the limitation of not offering information in dynamics, our study further provides correlations and interaction models with previously documented hyperinflammatory molecules (suPAR, sTREM-1, HGF, MCP-1, IL-1β, IL-6, ferritin) (3), alongside hematological and coagulation profiles (revealed by NLR, PLR, ESR, fibrinogen, D-dimer), markers of multiple organ injury (LDH, creatinine, urea, AST, ALT, total bilirubin) and data regarding therapeutic strategies, comorbidities, and vaccination status. Nevertheless, this study also points out the importance of including neural network models in complex data analysis for identifying the key elements and molecular-therapeutic interactions that underlines the overall paraclinical and clinical patient’s profile.