We conducted a prospective observational cohort study of patients aged ≥18 years presenting to the Intensive Care Unit (ICU) at University College London Hospitals (UCLH) between 1^st August 2021 and 26^th January 2023 who had blood cultures taken for suspicion of bacterial infection. Exclusion criteria for this study included patients with: i) underlying hemato-oncologic diagnoses, ii) severe anemia, ii) a contra-indication to blood transfusion, iii) pregnancy and iv) those not expected to survive beyond 24 hours of admission. Patient demographics, clinical data (physiology, diagnoses), laboratory data, and clinical outcomes were recorded from electronic healthcare records. Patients were followed up to hospital discharge/death. This was a sub-study of a larger study investigating biomarkers in sepsis (REC reference 20/LO/1024). As a reference, venous blood samples were taken from healthy volunteers (HV) between August 2021 and January 2023. Approval was obtained from the University College London (UCL) REC; reference 19181/001.

Whole blood was collected in CPT™ (8ml), K2 EDTA (4ml), and SST Advance™ vacutainers (4mls, all Becton Dickinson (BD), Wokingham, UK) and processed within 30 minutes. The CPT vacutainers were centrifuged at 1500rcf for 15minutes, the PBMC layer extracted, washed twice in PBS, resuspended in freezing media (fetal bovine serum (FBS) (Gibco, Thermo Fisher (TF), Cambridge, UK) with 10% DMSO (Dimethylsulfoxide, Sigma-Aldrich (SA), Gillingham, UK), and cooled to -80 °C using a Mr Frosty™ isopropyl alcohol gradient (TF). After 24–48 hours, the PBMCs were transferred for long term storage in liquid nitrogen. Frozen PBMCs were rapidly defrosted in batches using media, washed twice in media and rested for 1 hour prior to stimulation. Serum was stored at -80 °C.

Lymphocytes were stimulated with CD3/28 beads (Miltenyi Biotec (MB), Woking, UK) at a concentration of 4:1 (beads: PBMCs) for 72 hours at 37°C, 5% CO₂ in 11 healthy volunteers and 22 ICU patients.

Cells were acquired an ID7000 spectral cell analyzer (Sony Biotechnology Inc, Weybridge, UK) running ID7000 software (version 1.2, Sony). Alignment check beads were used prior to running each experiment. Spectral references for each fluorochrome were added using either single stain labelled compensation beads (TF) or heat-killed cells (60°C for 10minutes, viability stains) with appropriate negative controls. FMO (fluorescence minus one) samples were used to identify cell populations. The stopping gate was set at 10,000 events for either classical⁺ monocytes or CD4⁺ lymphocytes.

Cells were identified initially by forward and side scatter, singlets (side scatter area by height) and viable. Monocytes were then identified as being HLA-DR⁺, CD11b^-, CD14⁺⁺/CD16^-. CD4⁺ lymphocytes were identified as being CD3⁺CD19^- and CD4⁺CD8^-. ( Supplementary Figure 1 ).

Fluorochrome panels were designed to investigate cellular immunophenotypes. ( Supplementary Table 1 ) Dilutions were determined using dose titrations based on manufacturers recommendations.

CRP was measured by the hospital laboratory. Serum AREG, IL-1β, IL-6, IL-10, TNF-α, PD-1, and PD-L1 were measured using Duoset ELISA kits (R&D Systems, Abingdon, UK) as per manufacturer instructions. Samples were diluted 1:2 in reagent dilutant. Optical densities were acquired on a SPECTROstar Nano microplate reader (BMG Labtech, Aylesbury, UK).

Statistical analysis was carried out using GraphPAD prism software (Version 10, GraphPad, San Diego, CA). Categorical data was compared using Chi-squared test. For continuous data, a two-tailed Mann Whitney U was used, when two groups were being compared. Where more than two groups were being compared, Kruskal-Wallis test with Dunns uncorrected test for multiple comparisons for three groups, and multivariate analysis was performed and volcano plots generated using a corrected p-value (-log10) with a False Discovery Rate (FDR) of 5% calculated using the two-stage step-up method of Benjamini, Krieger and Yekutieli. The sensitivity and specificity of AREG in predicting mortality was assessed using an area under the receiver operator curve (AUROC). The association of AREG with mortality was assessed by log-rank test and presented as a Kaplan Meier curve.

We evaluated serum AREG levels in 42 critically ill patients who were admitted to the ICU with sepsis (all patients bacterial infections; 2 with co-existing malaria and one with COVID-19). The median time from ICU admission to blood sampling was 2.5 (1–5) days. A total of 9/42 patients (21%) died in hospital. The median patient age was 57 (range 43-72) and 64% were male ( Table 1 ). At ICU admission, Sequential Organ Failure Assessment (SOFA) scores were significantly higher among patients who subsequently died in hospital, compared to those who survived (p=0.004).

In addition, peripheral blood mononuclear cells (PBMC) were available from a cohort of 35 patients with sepsis, 13 of whom were non-survivors (37%) ( Supplementary Table 2 ). This allowed us to apply high-throughput multi-parameter flow cytometry assays, inclusive of EGFR and AREG antibodies, to assess these cellular traits (either as a percentage frequency/median fluorescence intensity (MFI)) between groups. The median patient age was 62 (range 51-71) and 70% were male. ICU admission SOFA sore was significantly higher among non-survivors compared to survivors (p=0.01).

In order to examine baseline cellular expression levels of AREG and EGFR, we included samples from 20 healthy volunteers (HV, 33 (23–40) years, 45% male) ( Supplementary Table 3 ).

We first examined whether serum AREG levels might be elevated in adult sepsis, compared to healthy donors, as we had previous observed in pediatric infections (1). To that end, we observed that serum AREG levels were significantly higher in patients admitted with infection to ICU compared to healthy volunteers (p<0.0001; Figure 1a ).

We next compared serum AREG levels between sepsis survivors and non-survivors. Strikingly, we observed that ICU non-survivors had significantly higher levels of AREG on admission (p<0.001), day 1 (p=0.003), and day 5 (p=0.031) compared to survivors ( Figure 1b ). Moreover, when combined with 14 other routinely collected clinical analytes ((CRP, creatinine, lactate, platelet count, lymphocyte count, monocyte count, neutrophil count, leukocyte count) and a panel of a priori selected cytokines associated with sepsis (IL-1β, TNF-α, IL-6, IL-10, soluble PD-1, soluble PD-L1)), only serum AREG could differentiate sepsis non-survivors from survivors (mean rank diff 14; -log10 (q value) = 2.4) ( Figure 1c ). Indeed, elevated serum AREG was associated with hospital mortality (AUROC = 0.87, p<0.001), and outperformed the SOFA (Sequential Organ Failure Assessment) score (AUROC = 0.80, p=0.006) ( Figure 1di-ii ). C-reactive protein was not associated with hospital mortality (AUROC = 0.71, p=0.08), implying a potential disconnect between parameters of biochemical inflammation and tissue injury. An AREG cut-off value of 350pg/mL (derived from the Youden’s index) on ICU admission was associated with in-hospital mortality (p<0.05 on log-rank) ( Figure 1e ).

Having identified elevated circulating levels of AREG in sepsis, and higher levels being associated with mortality, we next sought to identify which immune cell types, if any, might be a cellular source of AREG. Thus, we next examined the frequency of AREG⁺ cells in PBMC obtained from healthy volunteers and sepsis patients ex vivo and after activation in vitro ( Figure 2 ). Intriguingly, we observed a modest, yet statistically significant, increase in the frequency of AREG-positive CD4⁺ lymphocytes (p<0.0001) and CD8⁺ lymphocytes (p=0.0002), in ICU patients compared to healthy volunteers. We observed no significant differences between sepsis survivors and non-survivors, nor did we observe any significant differences in the frequency of AREG⁺ cells across any of the groups for myeloid cells (classical monocytes, intermediate monocytes, non-classical monocytes). These data imply that lymphocytes are a prominent source of AREG in the circulation in sepsis.

Similar trends in lymphocyte populations were seen on assessment of MFI, and monocyte populations demonstrated a reduction in AREG MFI in sepsis compared to HV ( Supplementary Figures 2 , 3 ). On CD3/CD28 bead stimulation, increases in AREG MFI were seen in CD8⁺ lymphocytes.

To establish whether circulating immune cells in adult septic patients could directly be influenced by AREG, a prerequisite to our hypothesis that EGFR-axis signaling contributes to sepsis immune dysregulation, we assessed surface levels of EGFR in healthy volunteers and patient PBMCs. The frequency of CD4⁺ lymphocytes (p<0.05), CD8⁺ lymphocytes (p<0.0001) expressing EGFR was higher among ICU patients compared to healthy volunteers. No differences in EGFR expression were seen in CD19⁺ lymphocytes, or monocyte subsets ( Figure 3 ).

Given that we observed higher T cell frequencies of EGFR expression in sepsis versus healthy volunteers, we hypothesized that EGFR-expression within lymphocytes may require prior cellular activation. To recapitulate this in vitro, we next exposed HV or ICU sepsis patient PBMCs to CD3/CD28 beads and assessed changes in EGFR percentage and MFI from baseline. We observed an increase in the percentage of CD4⁺ (p<0.01) and CD8⁺ (p<0.001) lymphocytes expressing EGFR ( Figure 3 ).

On assessment of MFI, there were no differences in EGFR expression between HV and ICU patients ( Supplementary Figures 4 , 5 ). However, changes to CD4⁺ and CD8⁺ lymphocyte EGFR MFI on CD3/CD28 stimulation were similar to changes in percentage positive cells.

Given the capacity for CD4⁺ and CD8⁺ lymphocytes to upregulate AREG/EGFR expression upon activation, we next assessed correlations between these markers and other prototypic indices of cellular activation, exhaustion or functional capacity. CD4⁺ lymphocyte EGFR expression positively correlated with cellular traits including: CD279 (PD-1) (r=0.80; p=8.7x10^-7), CD25 (IL-2R) (r=0.86; p=2.1x10^-7), Tbet (r=0.96; p=3.2x10^-12), NFκB (r=0.74, p=8.5x10^-5)and CD152 (CTLA-4) (r=0.88, p=7.9x10^-8) as well as cytokines IL-10 (r=0.91; p=4.0x10^-9), IL-4 (r=0.86; p=2.3x10^-7) and IFN-γ (r=0.83; p=1.7x10^-6). In contrast, CD4⁺ lymphocyte EGFR expression negatively correlated with IL-17A (r=-0.69; p=0.0004) ( Figure 4 ). AREG expression correlated with CD4⁺ lymphocyte CD95 (r=0.50; p<0.05), and CD279 (r=0.55; p<0.01) ( Supplementary Figure 6a ).

In CD8⁺ lymphocytes, EGFR expression correlated with CD152 (r=0.79; p=2.8x10^-5), CD279 (r=0.75; p=2.5x10^-4), CD274 (PD-L1) (r=0.71; p=3.2x10^-6), Tbet (r=0.62; p=9.2x10^-5), HLA-DR (r=0.61; p<0.01), and CD194 (CCR4) (r=0.54; p<0.05) ( Figure 4 ). There were no correlations between CD8⁺ lymphocyte AREG and other functional proteins ( Supplementary Figure 6b ).