The study population was a subgroup of a 99-patient cohort recruited at the anesthesiology-led interdisciplinary surgical ICU of a tertiary care hospital (University Medical Center Mannheim) between June 2017 and May 2019 as reported (39). Briefly, for cohort inclusion F.S.C. and J.J.S. evaluated consecutive patients on the first two days of their ICU admission daily at 2 PM for the presence of SIRS and Sepsis-1/2, that is SIRS driven by an infection (3, 4). Sequential Organ Failure Assessment (SOFA) scores were calculated as described (40). Patients were included at 3 PM on either day two if SIRS was present since ICU admission or on day one if they presented with sepsis but neither severe sepsis nor septic shock. The reason to not also include patients with SIRS already on day one was to more reliably exclude the presence of sepsis on inclusion in these patients and to dependably refer to them as post-injury sepsis cases if a sepsis label was subsequently assigned during follow-up. Patients were excluded in the case of chronic glucocorticoid therapy, recent organ transplant, end-stage renal failure, and neoadjuvant therapy.

EDTA-anticoagulated blood was sampled from a central venous catheter at eight-hour intervals starting at 3 PM on the day of study inclusion for a follow-up of five days or until ICU discharge or death. Samples were kept on ice, and plasma was obtained by centrifugation within 30 minutes of sampling, aliquoted, and immediately stored at -80°C.

F.S.C., J.J.S., and M.T. retrospectively adjudicated upon the presence of SIRS and all three stages of Sepsis-1/2 at all sampling time points during follow-up considering the available medical records. For 33 out of 1138 time points (2.9%), no blood sample and no definite adjudication results could be obtained. In the following, adjudication results are also referred to as expert labels and are italicized. For patients admitted due to polytrauma, T.S. retrospectively determined Injury Severity Score (ISS) values on ICU admission. All other clinical characteristics were retrieved from medical records and are reported for the time of study inclusion.

The analyses commenced in August 2018. At that time, samples were available from 63 patients with completed follow-up and physician-adjudication ( Figure 1 ). The differentiation of patients with sepsis from patients with non-infectious SIRS is particularly challenging in the critically ill, notwithstanding that severe sepsis and septic shock also require rapid diagnosis. Yet, in the first part of our study our focus was on patients, who transitioned from SIRS to sepsis on consecutive labeling time points, i.e., within eight hours, and identified nine such post-injury sepsis cases among the 63 hitherto included patients. Remarkably, these cases were all admitted due to polytrauma. Altogether 20 out of the 63 patients had SIRS due to acute polytrauma, out of which 12 developed sepsis according to Sepsis-1/2, specifically pneumonia, and nine had a direct transition from SIRS to sepsis as mentioned. Out of these and out of the eight polytrauma patients remaining infection-free during follow-up, seven case-control pairs were formed minimizing the overall mean age difference, which was 4.57 years (range 0 – 11 years).

We first considered changes over time in the concentrations of soluble immune checkpoints and established makers of inflammation and organ dysfunction in blood plasma from the seven selected polytrauma patients, who transitioned from SIRS to sepsis during follow-up. Starting with the first sample obtained on study inclusion, i.e., 3 PM on ICU-day two, the second sample was included at 7 PM on day three and then at intervals of 24 hours until the first sepsis label. Subsequent samples for up to a total of five or six time points were chosen depending on plasma availability. Soluble immune checkpoints were additionally determined in plasma from the seven paired infection-free polytrauma controls on corresponding reference time points. Figure 2 identifies the specific time points, expert labels, and patients for the 76 included samples.

In the second part of our analysis, starting at the end of the study period in May 2019, 56 unselected non-polytrauma patients with samples and expert labels available for the complete follow-up period were included ( Figure 1 ). During follow-up, 24 of them remained infection-free and 13 were assigned at least one incident label for sepsis, namely, four for sepsis, three for severe sepsis, and six for septic shock. On study inclusion, eleven had a label for sepsis, six had a label for severe sepsis and two for septic shock instead of sepsis, i.e., for eight patients the retrospective adjudication implied a higher degree of infection severity than the original evaluation on study inclusion. The samples obtained on the inclusion of the non-polytrauma patients were assayed for selected immune checkpoints. The results are referred to as study baseline levels.

Sixteen soluble immune checkpoints and 61 blood markers of inflammation and organ dysfunction were determined in cryopreserved blood plasma on a MAGPIX system (Luminex, Austin, TX, USA) with MILLIPLEX map kits (Merck Millipore, Burlington, MA, USA) according to the manufacturer’s instructions. Kits and analytes are listed in Supplementary Table 1 . Technical duplicates were averaged and analyzed with MILLIPLEX Analyst software (Merck Millipore).

We interrogated the GeneCards database (www.genecards.org) (41) to obtain information on the functions of plasma proteins and immune checkpoint receptor-ligand interactions, and additionally The Human Protein Atlas database (www.proteinatlas.org) (42) for cellular sources of plasma proteins.

In all our seven polytrauma cases that developed sepsis according to Sepsis-1/2, infectious foci occurred in the lung. Among the microbiology test samples obtained during follow-up, bronchoalveolar lavage of four cases tested positive for Gram-negative and two additionally for Gram-positive bacteria that were also detected in the blood of one case (details on microbiology test results are given in Supplementary Text 1 ).

Demographics and clinical scores as well as hospital LOS for pneumonia cases and controls remaining free of infection during follow-up were highly similar. Patients were male except for one case. For cases and controls, the mean age ± standard deviation was 46.9 ± 15.5 and 47.7 ± 18.5 years (p = 0.927), and the mean hospital LOS was 43 ± 20 and 52 ± 43 days (p = 0.616), respectively. Hospital survival was 100% in both cases and controls. The average Charlson Comorbidity Index was 3.5 times higher in the control group (p = 0.076). There was no record of diabetes and pre-existing respiratory disease but cardiovascular disease and alcoholism in one case and two controls each (p = 1.000 for both conditions). There were also no statistically significant group differences for the ISS (p = 0.943), SOFA score (p = 1.000), the Simplified Acute Physiology Score II (p = 0.715), the Richmond Agitation-Sedation Scale (p = 0.516), and the Core-10-Therapeutic Intervention Scoring System (p = 0.101). Details on clinical scores are given in Supplementary Table 3 .

Between ICU admission on day one and study inclusion on day two, three cases and controls each had received any type of blood component therapy (see Supplementary Table 4 for details). Transfusions had overall been two times more frequent in controls than cases. Namely, red blood cell concentrates had been transfused in one case and three controls, fresh frozen plasma in two cases and controls each, platelet concentrates in one case and two controls, and one control had received fibrinogen replacement therapy. None of these differences was statistically significant.

On study inclusion, all polytrauma patients were mechanically ventilated, and five cases and controls each received catecholamines. Table 1 summarizes additional selected clinical study baseline characteristics, i.e., on study inclusion. Altogether, we compared demographics, admitting department, chronic conditions, blood gas analysis, clinical chemistry, hematology, vital signs, hemodynamic and respiratory function indices, and clinical scores (see Supplementary Table 5 for a comprehensive summary as median with interquartile range (IQR) values). Among all of these characteristics, only a higher fraction of inspired oxygen (FiO₂) in cases (mean = 0.36 ± 0.06) than in controls (mean = 0.31 ± 0.02) reached statistical significance (p = 0.035). Medical conditions of polytrauma cases and controls on study inclusion thus appear overall highly similar.

For simplicity, the prefixed “s” to indicate “soluble” is omitted from the immune checkpoints’ protein symbols, and markers of inflammation and organ dysfunction are collectively referred to as cytokines. Figure 3 summarizes the time series of blood plasma levels for 16 immune checkpoints and 51 cytokines for the seven polytrauma cases, who developed pneumonia during follow-up, as a heat map. The majority of the variance is seen between individual patients, which is consistent with a fair grouping of the samples by patient in a PCA ( Supplementary Figure 1 ). Four subclusters (labeled 1–4) could be distinguished. Subclusters 1 and 2 featured two (TIM-3 and CD40) and one (HVEM) immune checkpoints, respectively, together with mainly, but not exclusively, myeloid cell-expressed/dependent cytokines and markers of polymorphonuclear neutrophil degranulation and cell death. Respective cytokines in subcluster 1 included IP-10, IL-10, IL-1 receptor antagonist (IL-1RA), MIP-1α, MIP-1β, tumor necrosis factor α (TNF-α), and in subcluster 2, IL-6, MIF and IL-8. Notably, most of them display chemotactic activities, namely, IP-10, IL-1RA, MIP-1α, MIP-1β, MIF, IL-8, and additionally complement C5a in subcluster 1. Markers of degranulation and cell death in subcluster 1 were LDLR, HMGB1, granzyme B, perforin, and myeloperoxidase (MPO), and in subcluster 2, HSP70 and ARG1. Subcluster 3 combined 11 out of the 16 immune checkpoints, mainly due to overall low and high levels in patients 9 and 66, respectively. Subcluster 4 was largely characterized by markers of endothelial activation/vascular trauma, i.e., G-CSF, PDGF, tissue factor, fractalkine/CX3CL1, VEGF, sRAGE, GM-CSF and sVCAM-1, as well as coordination of acute tissue inflammation and adaptive immune response, i.e., IL-2, IL-12p70, IL-1β, eotaxin/CCL11, eotaxin-3/CCL26, IL-15, IL-4, MIG/CXCL9, IFN-γ, IL-13 and AGP. It additionally contained the immune checkpoints CD27 and CD28.

Table 3 lists clinical study baseline characteristics for the unselected non-polytrauma patients ( Figure 1 ), that is surgical admissions to our ICU (see Supplementary Table 8 for a summary of all characteristics considered as medians with IQR values). In contrast to the polytrauma patients, they showed several subgroup differences detailed in Supplementary Text 2 . Briefly, the non-polytrauma patients were overall older than the polytrauma patients and contained more men than women in the incident sepsis compared to the sepsis-free subgroup. Mortality was highest in the incident sepsis subgroup. Baseline disease severity was highest in this subgroup as judged by SOFA, catecholamine therapy, and systolic blood and mean arterial pressure. In agreement with the presence of infection, C-reactive protein (CRP) and proportions of neutrophils were elevated and lymphocytes were reduced in the sepsis on-inclusion subgroup. There were no subgroup differences in pre-existing chronic conditions and mean hospital LOS. Microbiology test results are given in Supplementary Text 1 .

Figure 7 summarizes the comparisons of the study baseline levels of HVEM and ICOS across all polytrauma and non-polytrauma subgroups with correction for multiple testing. The absence of any difference among the three larger non-polytrauma subgroups is notable. The main differences consisted of overall lower levels of ICOS in the polytrauma cases than in all non-polytrauma subgroups.

The recognition that immunosuppression is a major risk factor for infection and sepsis in the critically ill has put a spotlight on immune checkpoints as potential predictive and diagnostic sepsis markers in this patient population (52). Many immune checkpoints are detectable in soluble form in peripheral blood. We asked whether their concentrations in patients admitted to our interdisciplinary surgical ICU with trauma-induced SIRS were associated with the development of infection-triggered SIRS, i.e., sepsis according to Sepsis-1/2, during the first week. The emulation of this clinical challenge, i.e., the early differential diagnosis of sepsis in this setting, with high fidelity in an observational study is complicated by the strong heterogeneity of the patient population and the presentation of the syndrome. Moreover, we previously determined that 93% of the patient days in our ICU were associated with a SOFA > 2 compared to 68% with SIRS, regardless of infection (8).

The first part of the current study was focused on the particularly challenging differential diagnosis of an infection as a trigger of systemic inflammation in patients with pre-existing SIRS, which we found to be common in polytrauma patients developing pneumonia during the first week in the ICU ( Figure 1 ). We included polytrauma admissions, who experienced a direct transition from SIRS to sepsis. For seven such incident cases, we identified seven polytrauma patients remaining infection-free during follow-up that were highly similar with regard to clinical characteristics. These included the severity of the traumatic injury, degree of organ dysfunction, and other risk factors, as well as treatment intensity ( Supplementary Table 5 ). Starting on ICU-day two, one blood plasma sample was included daily for analysis until the onset of sepsis (pneumonia) and additional samples thereafter ( Figure 2 ). Therein, 16 immune checkpoints were quantified by multiplex bead array. For each of them, the absolute concentrations determined were on essentially the same levels as reported previously by Wu et al. (2021) for polytraumatized patients on day three in the ICU using the same commercial assay (53), which supports the consistency of this analytical method.

We first considered the blood plasma immune checkpoint profiles in the context of the profiles of 51 established blood markers of inflammation and organ dysfunction detectable in the polytrauma post-injury pneumonia cases ( Figure 3 ). TIM-3 as well as CD40 in subcluster 1, both reportedly upregulated on sepsis monocytes (22, 26–28), and HVEM in subcluster 2, previously shown to mediate enhancement of in vitro bactericidal activities of monocytes and neutrophils (54), indeed clustered with markers of myeloid cell activation including cytokines, mainly chemokines, and markers of degranulation and cell death. Notably, eleven checkpoints essentially defined the distinct subcluster 3. The remaining two checkpoints, the T cell-expressed co-stimulatory molecules CD28 and CD27, were contained in subcluster 4 together with proteins mainly originating from activated endothelium/vascular trauma and inflamed tissue and active in the transition to an adaptive immune response. This clustering result suggests that immune checkpoint profiles in blood after polytrauma are related to distinct submechanisms of the inflammatory process.

Next, we subjected the blood levels of the 16 immune checkpoints in both the polytrauma cases and time-matched samples from infection-free controls ( Figure 2 ) to a cluster analysis obtaining two well-separated subclusters, A and B ( Figure 4 , Figure 5 ). Here, CD27 from subcluster 4 of the above case-restricted analysis ( Figure 3 ) merged with the smaller subcluster A, containing TIM-3 and CD40 from subcluster 1 as well as HVEM from subcluster 2, and CD28 with the main subcluster B, otherwise comprised of the eleven subcluster 3 immune checkpoints. A comparison of the relative variance of immune checkpoint levels confirmed the impression from the heat map in Figure 4 that levels were overall more stable in cases than controls ( Supplementary Figure 4 ). Apparently, more pronounced and dissimilar dynamics for CD27 and CD28 in the controls compared to cases accounted for their divergent subcluster memberships in Figure 4 against the first case-restricted cluster analysis in Figure 3 .

PCA, based on the immune checkpoint plasma concentrations, confirmed higher overall similarities of samples within than between the 14 polytrauma patients ( Supplementary Figure 5 ). But it also suggested a fair separation of cases and controls. To gain further insight into pneumonia-associated differences in the time trends of immune checkpoint levels in cases and controls we used hierarchical linear models accounting for intra- and inter-individual variability in repeated measurements. The results corroborated a difference only for CD27 and consisted in a steeper estimated increase before the time of pneumonia onset in controls than in cases ( Supplementary Table 7 ). Throughout follow-up, relatively high positive slopes in the controls, reaching statistical significance for TIM-3, was indeed a defining feature of subcluster A compared to negative control slopes in subcluster B (except for LAG-3) reaching statistical significance for PD-L1 ( Figure 6 , Table 2 ). In both subclusters, immune checkpoint levels in cases overall trended towards lesser changes over time than in controls. The significant rise in TIM-3 in polytrauma cases, despite no significant difference from the controls, may correspond to elevated levels of this protein in pediatric burn patients developing nosocomial infection (38).

Another common feature of subcluster A-immune checkpoints, in addition to positive control slopes, consisted of higher levels in controls than in cases throughout follow-up. This was most pronounced for HVEM and was reflected by a significantly higher estimated study baseline value (intercept) both when considering all measurements ( Table 2 ) and the measurements before pneumonia onset only ( Supplementary Table 7 ). In the latter subanalysis, this was also the case for ICOS, where controls however trended towards a negative slope, potentially mitigating the difference between case and control levels over time. HVEM, by contrast, if any slightly increased over time in controls and appeared relatively stable in cases (see also Supplementary Figure 4 ).

Among all the above-described characteristics of immune checkpoint blood levels in polytrauma patients, the consistently reverse relation between HVEM levels and pneumonia incidence makes HVEM a candidate predictive marker for pneumonia in these patients. Higher measured study baseline levels for HVEM in controls than cases reached statistical significance in an unadjusted analysis. This and overall similar case and control levels of HVEM’s main binding partner BTLA in our study, however, contrast reportedly elevated levels in sepsis for HVEM on monocytes and granulocytes and for BTLA on CD4-positive T cells, monocytes and granulocytes (23, 55), as well as higher post-mortem HVEM levels in the respiratory epithelium and unchanged levels on lung macrophages (24). Contrarily, the similarly reverse relation between sepsis incidence and T cell positivity for HVEM after polytrauma reported by Wakeley et al. (2020) (25) matches our finding for HVEM blood levels. In their report, five out of seven post-injury sepsis cases developed pneumonia compared to all seven in our study. This may point towards the lung as a source of elevated HVEM in the controls and a role of HVEM in lowering the risk of secondary infection in this tissue as further discussed below.

The biology of HVEM as both a ligand and a receptor (56) and its role in sepsis (57) were reviewed previously. Briefly, HVEM is widely expressed in adult tissues with the highest levels in the spleen, blood lymphocytes, and myeloid cells. Its membrane-bound binding partners BTLA, CD160, and LIGHT (lymphotoxin-like, inducible expression, competes with herpes simplex virus glycoprotein D for HVEM, a receptor expressed by T lymphocytes), as well as soluble LTα (lymphotoxin α), are expressed by lymphoid cells and LIGHT additionally by myeloid cells. The binding of HVEM to LTα and LIGHT leads to NFκB activation and immune stimulation in the HVEM-expressing cell. As a ligand, HVEM however induces immunosuppression in BTLA- and CD160-expressing cells. In the same naïve T cell, inhibitory signaling through BTLA overrules concomitant stimulatory signaling through HVEM upon receptor-ligand engagement.

Given its complexity, the net outcome of HVEM signaling is highly context-dependent. Its role in fine-balancing immune tolerance and activation, unsurprisingly, makes HVEM a two-edged sword in sepsis. In various mouse models of intestinal and respiratory bacterial infections, protective mucosal barrier function depended either on the stimulatory HVEM-LIGHT axis or the inhibitory HVEM-CD160 axis but did not involve BTLA in either case (57). Notably, siRNA-mediated knockdown of upregulated pulmonary HVEM in a double-hit mouse model of acute lung injury (ALI) by hemorrhagic shock followed by polymicrobial abdominal sepsis dampened myeloid cell-mediated lung damage and improved survival (58). The latter points towards a relatively high importance of the immune stimulatory activity of HVEM in the lung.

We cannot directly extrapolate from our patient subgroup differences in blood levels of immune checkpoints to their activation status in blood cells or tissues. Yet, controlled ectodomain shedding is fundamental to the regulation of cell surface receptors and ligands (59). It is reasonable to assume that it contributes to these differences and adds to the high positive correlations of the blood levels for immune checkpoint receptor-ligand pairs within subcluster B in cases and controls ( Supplementary Figure 2 ). Levels of BTLA from subcluster B, however, showed no correlation with its ligand HVEM from subcluster A. As mentioned above, both elevated HVEM levels in T cells (25) and, even more so, in blood (this study) were associated with the development of secondary pneumonia after polytrauma. The essential role of LIGHT but not BTLA in the reported HVEM-dependent protection of respiratory mucosal barrier function in an animal model of Chlamydia psittaci respiratory tract infection (60) would be consistent with an involvement of the immune-stimulatory HVEM-LIGHT axis rather than inhibitory HVEM-BTLA axis. Immune-mediated lung damage after polytrauma and hemorrhage, in particular, is driven by thrombo-inflammation and involves the complement cascade and infiltration of the lung with activated myeloid cells (61, 62). The clustering of elevated blood HVEM with blood markers of these inflammatory processes, such as complement C5a and MPO (subclusters 1 and 2 in Figure 3 ), thus indicates that HVEM may escape from the inflamed lung into the blood after polytrauma (this study). This may particularly be exemplified by patient 9 showing the overall highest plasma levels of subcluster 1 and 2 proteins ( Figure 3 ).

Taken together, HVEM contributes to myeloid cell-mediated lung damage in the double-hit ALI model (58) and to respiratory mucosal barrier integrity, through the HVEM-LIGHT axis, in the Chlamydia psittaci infection model (60). Both reports are consistent with an interaction between HVEM, upregulated in the respiratory epithelium (24) and on T cells (25), and LIGHT on infiltrating myeloid cells. This both enhances antimicrobial defense but also drives lung inflammation. Wakeley et al. (2020), nevertheless, explained higher HVEM levels on T cells in their polytrauma controls than in post-injury pneumonia cases with a role of HVEM signaling in lymphocytes in maintaining an appropriate immune response after critical tissue trauma (25). This interpretation may also, at least partially, apply to the corresponding HVEM blood profile in our polytrauma patients.

We feel however that, in addition to these pro-inflammatory properties of HVEM, we also need to consider its immunosuppressive functions that may balance its lung damage potential. Interestingly, Mintz et al. (2019) reported that, in a mouse model, B-cell-expressed HVEM at the immunological synapse restricts T cell help during the germinal center (GC) response through engaging BTLA on T cells (63). This limits T cell receptor signaling and assembly of preformed intracellular CD40L (CD40 ligand) into the synapse. Reduced recruitment of CD40 on B cells, in turn, then destabilizes the synapse. Therefore, it is tempting to speculate that T follicular helper (Tfh)-like cells, that infiltrate the inflamed lung and express Tfh-markers CD40L and IL-21 (64), interact with naïve B cells in bronchus-associated lymphatic tissue through this particular HVEM-BTLA axis after polytrauma. Despite unchanged blood levels of BTLA, this interaction may still lead to the release of HVEM and CD40 from the B cells as a counter-regulatory and lung protective mechanism and accounting for the elevated HVEM and CD40 levels in blood. These two, indeed, correlate positively in cases and controls ( Supplementary Figure 3 ). On that note, the increasing levels of CD27 in our polytrauma controls compared to cases until the time of pneumonia onset ( Supplementary Table 7 ) likely reflect continued activation of T cells (65) and thus potentially, the need for their effective control to limit immune-mediated lung damage. Uncontrolled T cell and endothelial activation may dominate, for instance, in patient 15 featuring the overall highest levels of the corresponding markers in subcluster 4 ( Figure 3 ).

We further assessed whether the development of sepsis according to Sepsis-1/2 within the first week in the ICU was associated with low HVEM and potentially ICOS in blood at the study baseline only in polytrauma patients or also in our unselected surgical admissions (non-polytrauma patients) ( Figure 1 ). Besides subgroups of non-polytrauma patients without and with an expert label for sepsis during follow-up, corresponding to our polytrauma case and controls, we further considered surgical admissions with sepsis already present on inclusion. Notably, among the non-polytrauma subgroups the incident sepsis subgroup was characterized by the highest proportion of men, severity of illness judged by SOFA and macrocirculatory abnormalities, and mortality ( Table 3 ), consistent with sepsis development. The highest disease severity in this subgroup may explain the absence of differences in HVEM and ICOS compared to the sepsis on-inclusion subgroup ( Figure 7 ). Yet, the absence of differences in both these two subgroups compared to the sepsis-free ICU controls is notable.

We cannot fully exclude that the retrospective adjudication on the presence of sepsis by clinical experts may have been affected by hindsight bias, i.e., by the apprehended benefit for the patient of the therapeutic measures taken and of the clinical outcome achieved (66). For the timely identification of sepsis onset, however, this strategy is more appropriate than relying solely on clinical criteria for consensus definitions of sepsis. The latter was intended for different purposes, namely the inclusion of patients into trials to treat hyperinflammation (Sepsis-1) and retrospective epidemiological studies (Sepsis-3).

The major limitation of this study was the small number of almost exclusively male polytrauma patients in our cohort from only a single ICU. The limited statistical power precludes us from rejecting the absence of differences in study baseline blood levels in soluble immune checkpoints for candidates other than HVEM. However, delivery of care in our ICU over a reasonable enrollment period followed a consistent standard reducing the variance in iatrogenic risk factors for pneumonia and sepsis. Highly similar clinical characteristics support the comparability of the polytrauma post-injury pneumonia cases and infection-free controls.

A strength was the availability of close-meshed blood plasma samples with expert annotation on the presence of SIRS and sepsis according to Sepsis-1/2 between ICU-days two and six. This facilitated our trend analysis. Although we cannot extrapolate from our results to events beyond the follow-up period, we had previously shown that half of the overall sepsis cases in our ICU occur by day six (8) and in our polytrauma patients in particular by day seven (67). We also consider our conservative assessment of statistical significance by applying the correction for multiple testing and the absolute rather than relative quantitation of blood plasma concentrations as strengths.