Fluorochrome-conjugated antibodies (as listed in Supplementary Table 1), BD Cytofix/Cytoperm™ Fixation/Permeabilization Solution set, human BD Fc block™, BD GolgiPlug™ and BD Horizon Brilliant Stain Buffer Plus were purchased from BD Biosciences (San Jose, CA). UltraComp eBeads™ Compensation Beads were purchased from ThermoFisher Scientific (Waltham, MA). Recombinant human IFN-γ was purchased from PeproTech (Rocky Hill, NJ). LAG-3Ig was kindly provided by Immutep Pty Ltd. Bovine Serum Albumin solution (35% in D.PBS) and LPS were purchased from Sigma Aldrich (St. Louis, MO).

Thirteen patients who were admitted to the intensive care unit (ICU) of the Nepean Hospital (Kingswood, NSW, Australia) between December 2017 and February 2019, were recruited to this study. The inclusion and exclusion criteria are as follows:

Inclusion criteria: (1) age ≥18 years old; (2) presence of shock, likely of infectious origin defined as day 1 Sequential Organ Failure Assessment (SOFA) score cardiovascular > 2; (3) presence of at least 2 organ dysfunctions as defined by aggregate day 1 or day 2 SOFA score >4; (4) no treatment limitations articulated prior to study enrolment.

Exclusion criteria: current or recent use of known immune suppression medication such as steroids (>5 mg prednisone or equivalent) methotrexate, mycophenolate or recent use of immune checkpoint inhibitors.

Ten age-matched healthy volunteers (age ≥18 years old) were also recruited as controls.

Venous blood was collected into tubes containing EDTA. For patients with sepsis, blood was collected on days 1–3 post ICU admission. PBMCs were prepared from whole blood by density gradient centrifugation using Ficoll-Paque Plus (Bio-strategy Pty Ltd) and cryopreserved. Briefly, whole blood was diluted with an equal volume of Dulbecco's Phosphate-Buffered Saline (D.PBS) (without Ca²⁺ and Mg²⁺) and layered on top of an equal volume of Ficoll-Paque Plus in a 50 ml Falcon tube. This was followed by centrifugation using a swinging-bucket rotor at 400 × g for 30 min at room temperature (RT) with the brake off. After centrifugation, PBMCs at the interphase between Ficoll-Paque Plus and plasma, was transferred into a new 50 ml Falcon tube and washed twice with D.PBS. Each wash was followed by centrifugation at 300 × g for 10 min. For cryopreservation, PBMC cell pellet after the final centrifugation was resuspended in fetal bovine serum (FBS) and then 20% Dimethyl Sulfoxide (DMSO) diluted in FBS was added dropwise to the cell suspension to make the final concentration of DMSO to 10%. One milliliter (ml) of cell suspension was transferred to 2 ml cryovial in a Corning® CoolCell™ Cell Freezing Container (Sigma-Aldrich), which was left at −80°C freezer overnight before transferring to liquid nitrogen for long-term storage. On the day of experiment, cryopreserved PBMCs were thawed and treated with DNase I to prevent cell clumping (26). Briefly, frozen PBMCs were thawed rapidly in a 37°C water bath, followed by washes first with RPMI1640 (Lonza) with 2% FBS and 10 mM HEPES, pH 7.0–7.6 and then with D.PBS. PBMCs were then treated with 100 units/ml DNase I in the presence of 5 mM MgCl₂ for 15 min at RT, during which cells were counted with the CytoFLEX Flow Cytometer (Beckman Coulter, Brea, CA). After DNase I treatment, cells were washed once with D.PBS and then resuspended in cell culture medium (RPMI1640 with 10% FBS, 2 mM L-Glutamine, 50 units/ml Penicillin/50 μg/ml Streptomycin at 1 × 10⁶ cells/ml) for culturing in BD Falcon® 5 ml polypropylene round bottom tube (in vitro Technologies) at 37°C, 5% CO₂.

PBMCs were cultured under three conditions (two aliquots per condition and up to 1 × 10⁶ per aliquot): untreated, plus IFN-γ (100 ng/ml) (27) or plus LAG-3Ig (1 μg/ml) (22) for 2 h. BD GolgiPlug (1 μg/ml) for detection of intracellular cytokines was added to all the cultures which were then cultured with or without LPS (100 ng/ml) (27) for a further 4 h. Cells were harvested by pipetting at the end of stimulations and subjected to cell count using the CytoFLEX Flow Cytometer before proceeding with flow cytometry staining.

After 6 h of culture, ~1 × 10⁶ PBMCs were washed twice by resuspending cells in 4 ml D.PBS and centrifugation at 300 × g, RT for 5 min. Washed cells were then transferred into a 96-well V-bottom plate for the following staining procedure. Cells were first stained with 100 μl of BD Horizon™ Fixable Viability Stain 440 UV diluted in D.PBS (1:1,000), for 10–15 min at RT in the dark as per manufacturer's instructions. Cells were washed twice with 250 μl of staining buffer (D.PBS with 0.5% Bovine Serum Albumin) with centrifugation as above before surface staining. For surface staining, cells were incubated with 2.5 μg of human Fc Block diluted in staining buffer for 10 min at RT. Fluorochrome-conjugated surface antibodies as listed in Supplementary Table 1 (amount of antibody used for each test was pre-titrated) was then added along with BD Horizon™ Brilliant Stain Buffer Plus (10 μl per test) in a final staining volume to 100 μl. Samples were incubated for 30 min at 4°C in the dark, then were washed four times with 250 μl staining buffer with centrifugation as above. Washed cell pellets were fixed and permeabilized with 100 μl BD Cytofix/Cytoperm™ Fixation/Permeabilization Solution for 20 min at 4°C in the dark. Fixation/Permeabilizaiton buffer was removed by two washes with 250 μl of freshly prepared 1 × BD Perm/Wash buffer. All centrifugation steps after permeabilization were performed at 500 × g for 5 min at RT. The cell pellet was resuspended in 200 μl of staining buffer and stored overnight at 4°C in the dark. The following day, cells were pelleted by centrifugation (500 × g, 5 min, RT) and fixed/permeabilized again as described above. For intracellular staining, cells were incubated with 100 μl of fluorochrome-conjugated antibody mixture as listed in Supplementary Table 1 (amount of antibody used for each test was pre-titrated) diluted in 1 × BD Perm/Wash buffer with BD Horizon™ Brilliant Stain Buffer Plus (10 μl per test) for 30 min at 4°C in the dark. Cells were then washed four times with 1 × BD Perm/Wash buffer by centrifugation (500 × g, 5 min, RT), resuspended in 200 μl of staining buffer to be analyzed on a BD FACSymphony™ A5.2 cell analyzer (BD Biosciences). Samples were stained and analyzed in five batches and application settings were applied for each acquisition to ensure consistency in the instrument performance.

Flow data were analyzed using CATALYST (28), explored using tidyverse (29) and visualized using ggplot and pheatmap (30) in the R statistical environment (31).

Briefly, flow cytometry files (fcs files) were pre-processed [compensated, gated for single and viable cells and quality control checked using FlowAI (32)]. Data was then scaled and exported for analysis using R (v4.0). FlowSOM (33) is an algorithm for clustering cells with Self-Organizing Maps (SOMs), based on the behavior of chosen markers on all cells. We used FlowSOM to cluster cells (max k = 50, using 20 × 20 SOM grid) by surface markers including CD3, CD4, CD8, CD19, CD25, CD127, CD11c, CD14, CD16, CD56, HLA-DR, which were referred to as “type markers”. The type markers are lineage markers that were used to define cell types within PBMCs. Clusters were then manually annotated based on dominant cell frequency. Besides the type markers, we chose eight markers, including IL-2, IL-6, IL-10, IFN-γ, TNF-α, IL-17A, CCL-4, and HLA-DR, to be the “state markers,” because expression levels of these state markers give indications on the functional state of the cell. Uniform Manifold Approximation and Projections (UMAPs) were generated based on type or state marker expression, limited to 3,000 cells per sample. We also performed Differential State (DS) analysis using a modified version of diffcyt (34). For DS, cluster medians for state markers were extracted, background corrected (>0 in more than half the number of samples in the smallest group) and then compared using the limma eBayes test (35) with customized contrast matrices and multiple testing (corrected p-value cut-off of 0.05). Marker list overlaps were visualized using euler diagrams (36).

Figure 1 outlines the experimental design and workflow. PBMCs isolated from the whole blood of either healthy donors or patients with septic shock, underwent six different pre-treatment and/or stimulation conditions before they were stained and subjected to flow cytometry analysis.

In order to compare immune cell composition of PBMCs between healthy controls (HC, n = 10) and sepsis patients (sepsis, n = 13), flow cytometric data from untreated PBMCs (no treatment or LPS stimulation) from both HC and sepsis groups, were concatenated and FlowSOM was used to cluster cells by “type markers” as detailed in the materials and methods. Median expression levels of all the type markers in both HC and sepsis groups were shown on the UMAPs (Figure 2A). Based on dominant cell frequency, 50 meta-clusters were defined (Figure 2B). The meta-clusters were then merged into seven different immune cell subsets: CD19⁺ B cells, CD14⁺ monocytes, CD3⁺CD4⁺ T cells, CD3⁺CD8⁺ T cells, CD11c⁺ mDCs, CD3⁻CD56⁺ NK cells, and lineage negative (lin_neg) cells, based on the presence or absence of specific type markers. Relative proportions of the seven immune cell subsets in PBMCs were compared between HC and sepsis (Figure 2C). Our results demonstrate the stark differences between the immune profile of HC and Sepsis subjects. There was a trend toward increased proportion of CD14⁺ monocytes and decreased proportions of CD4⁺, CD8⁺ T cells, and NK cells in sepsis as compared to HC. Proportions of the seven immune cell subsets in individual sample were shown in Figure 2D. Heterogeneity within the sepsis patients was observed. Among the seven immune cell subsets investigated, proportions of CD14⁺ monocytes, CD4⁺ T cells, CD8⁺ T cells, mDCs, and NKs were significantly altered in sepsis compared to HC (Figure 2E).

Median expression levels of all the “type” and “state markers” (as detailed in the materials and methods) on total PBMCs after 4 h of LPS stimulation without pretreatment with IFN-γ or LAG-3Ig, were compared across all the meta-clusters between HC and sepsis as displayed on UMAPs (Figure 3A). We observed an overall trend toward decreased HLA-DR expression on CD14⁺ monocytes meta-clusters in sepsis as compared to HC, which is one of the hallmarks for sepsis. Median expression levels of all the markers in the seven immune cell subsets were compared between HC and sepsis (Supplementary Figure 1). To explore if there was any presence of subgroups in the heterogeneity observed in Figure 2D, we performed hierarchical clustering analysis, an algorithm used to group similar objects into clusters with each cluster distinct from each other. Median expression levels of all the type and state markers as in Figure 3A, were used for the hierarchical clustering analysis. Eight out of the 10 HC samples formed a tight cluster whereas sepsis samples were segregated into two subgroups which we named “Hyper” and “Hypo” subgroups (Figure 3B). The Hyper subgroup displayed an overall upregulation of surface markers CD25, CD11c, CD14, HLA-DR, and intracellular cytokines and chemokine as well as down regulated expression of CD3, CD8, CD127, and CD4 compared to HC. In the Hypo subgroup, expression of CD25, CD11c, CD14, HLA-DR, and intracellular cytokines and chemokines was intermediate between HC and the Hyper subgroup whereas expression of CD3, CD8, CD127, and CD4 was similar to that in the HC.

To further characterize the immune status of the two subgroups of sepsis, we performed differential state (DS) analysis by comparing Hyper and Hypo subgroups to HC in the expression levels of the state markers. Expressions of the state markers on 50 meta-clusters of PBMCs in response to LPS stimulation were analyzed. When compared to HC, both Hyper and Hypo subgroups were shown to have significant changes in the levels of the state markers across the 50 meta-clusters, 155 of which were shared between the two subgroups (Figure 4A). Significant differences between Hyper and Hypo subgroups were also observed. There were 59 unique immune alterations for the Hyper subgroup and 57 unique immune alterations for the Hypo subgroup. Differential expression of the state markers in Sepsis compared to HC were expressed as Log2 fold changes (LogFC). Representative meta-clusters from each immune cell subset, which had significant LogFC in either of the eight state markers, were shown in Figure 4B comparing the Hypo subgroup to HC (Hypo_vs._HC) and Figure 4C comparing the Hyper subgroup to HC (Hyper_vs._HC). An expanded version of Figures 4B,C can be found in Supplementary Figures 2A,B. Major differences between Hyper and Hypo subgroups were observed in IL-6, TNF-α, IL-17A, HLA-DR, and CCL-4 expression levels. IL-6 production in CD8⁺ T cells was decreased in the Hyper but was increased in the Hypo subgroup. Monocyte expression of TNF-α production was upregulated in the Hyper but downregulated in the Hypo subgroup. IL-17A production was upregulated in CD19⁺ B cells and CD8⁺ T cells from the Hyper but not in the Hypo subgroup. We also observed a more than 2 × fold increase in HLA-DR expression on NKs from the Hyper subgroup, which was absent in the Hypo subgroup. CCL-4 upregulation was observed across most of the immune cell subsets in Hyper subgroup. In the Hypo subgroup, however, CCL-4 expression was only increased slightly in CD4⁺, CD8⁺ T cells, and NKs. On the other hand, IL-10 and IFN-γ expression levels did not vary too much between these two subgroups. In light of the differences in the immune status between the Hyper and the Hypo subgroups, we sought to determine if the clinical characteristics of these two subgroups also differ. C-reactive protein (CRP), serum lactate, whole blood count (WBC), SOFA score, and Acute Physiology and Chronic Health Evaluation (APACHE) III score were compared between the Hyper and the Hypo subgroups (Figure 4D). We did not observe significant differences in these clinical characteristics between the two subgroups. Detailed information on the age and gender of these patients in addition to the above-mentioned clinical data can be found in Supplementary Table 2.

To determine if Hyper and Hypo subgroups would respond differently to IFN-γ or LAG-3Ig, we performed DS analysis within each subgroup by comparing median expression of the state markers under “No treatment + LPS” (LPS) to that under either “IFN-γ pretreatment + LPS” (IFN-γ) or “LAG-3Ig pretreatment + LPS” (LAG-3Ig) conditions. As shown in Figures 5A,B, immune cell subsets from the Hypo subgroup of sepsis showed significant responses to either IFN-γ or LAG-3Ig. Within each cell subset, the meta-clusters which demonstrated significant responses to either treatment were squared out in red. In response to IFN-γ + LPS, three meta-clusters (#29, 34, and 42) within CD14⁺ monocytes from the Hypo subgroup had higher TNF-α expression and meta-cluster #34 also had higher IL-2 expression as compared to no treatment LPS group (Figure 5A). Two meta-clusters (#23 and 31) within CD4⁺ T cells from the Hypo subgroup also responded to IFN-γ by producing more IL-10 as compared to no treatment LPS group. Figure 5B showed responses of the Hypo group to LAG-3Ig treatment as compared to no treatment LPS group. All four meta-clusters of CD4⁺ T cells (#22, 23, 31, 37, and 40), two meta-clusters within CD8⁺ T cells (#11 and 45) responded to LAG-3Ig by producing more IL-10. Seven meta-clusters within CD14+ monocytes (#18, 19, 29, 33, 34, 42, and 44) produced more IFN-γ and IL-2 in response to LAG-3Ig. An increased CCL-4 expression was observed in meta-cluster #21 within NK cells in response to LAG-3Ig. On the other hand, responses of the Hyper subgroup to either IFN-γ or LAG-3Ig were less prominent compared to the Hypo subgroup. The Hyper subgroup demonstrated no significant responses to IFN-γ treatment (data not shown) and minimal responses to LAG-3Ig as shown in Supplementary Figure 3.