Blood was stored at room temperature for up to 12 hours prior to processing. Peripheral blood mononuclear cells (PBMCs) were isolated, counted using a hemocytometer, and frozen in freezing media at -80°C. When sufficient samples were collected, PBMCs were thawed, washed, and stained with fluorochrome-labelled antibodies. An aliquot of PBMCs from each patient was stimulated ex vivo as described below.

Cells were stained with LIVE/DEAD fixable viability dye (Life Technologies, Carlsbad CA); samples were plated in duplicate and stained with either a 20-antibody panel for T cell markers or for innate and B cell markers. Intracellular markers (CD68, FoxP3, Granzyme B, IL-10, IL-17A, IFNγ, T-bet, TNFα) were applied after samples were fixed and permeabilized. Stained cells were stored at 4°C for up to 7 days. Flow cytometry was performed on a BD LSR Symphony 23-color cell analyzer (BD Biosciences, San Jose CA). The gating strategy can be referenced in the supplement. Results were analyzed using FlowJo software v10.1 (BD Bioscience, San Jose, CA). Cell counts were determined using the white blood cell count when available from the electronic medical record on the day of sample collection.

Levels of plasma cytokines from Day 1 plasma samples, collected and stored at -80°C, were determined using multiplex ELISA (Eve Technologies, Calgary, Alberta, Canada).

The reserved aliquot of PBMCs noted above was stimulated with phorbol myristate acetate (50ng/ml, PMA) and ionomycin (1.5uM, Io) for three hours at 37°C in 5% CO₂. The reaction mixture included Brefeldin A (2ug/ml, BD Biosciences, San Jose, CA) to inhibit cytokine secretion to increase flow cytometry sensitivity for intracellular cytokine production (28, 29). After incubation, these cells underwent staining and analysis as described above. All stimulation assays were performed alongside a control without stimulation, as previously described (30).

Categorical data were expressed as frequency (percent) and continuous data as mean (standard deviation) or median (interquartile range). Analyses were performed using GraphPad PRISM 9. We used Kruskal-Wallis test with Dunn’s correction for multiple comparisons to compare subpopulations in different cohorts with healthy controls. To examine relationships between stimulation (yes/no) and sepsis vs healthy cohort, the stimulated samples were subjected to a mixed effect analysis. For analysis of the relationship between subpopulations within a given cohort with subject age, simple linear regression was used. For plasma cytokine analysis, values out of range were assigned the upper or lower limit of the assay. Cytokine values that failed normality based on D’Agostino and Pearson Test underwent natural logarithmic transformation prior to analysis. Cytokine values were analyzed between the four cohorts via one-way ANOVA with Dunnet correction for multiple comparisons of each cohort to the healthy cohort. The sepsis and healthy cohorts were categorized by age (>5 years or ≦5 years) and two-way ANOVA was used to compare plasma cytokine levels. Age of 5 years was used to divide patients by prior to and following adrenarche (31). Plasma cytokine levels that were significantly different on initial one-way ANOVA were included in this analysis.

Figure 1 represents cytokines that differed among cohorts. Several cytokines were uniquely different between healthy and sepsis cohorts. Cytokines known to activate the adaptive immune system were generally higher in sepsis compared to the healthy cohort, including RANTES (p=0.01), G-CSF (p<0.0001), M-CSF (p<0.0001), IL-6 (p<0.0001), IL-8 (p<0.0001), IL-27 (p<0.0001), and MIP-1β (p<0.0001) (32–36). Markers of an IFNγ response were higher in sepsis than in healthy patients, including IL-18 (p=0.02) which increases IFNγ production and MIG (p<0.0001), an IFNγ inducible chemockine (37–40).

IL-10 level was higher in sepsis (p<0.0001) and AOD (p=0.02) compared to the healthy control cohort. IL-10 promotes regulatory T cells, playing a counter-regulatory role to host inflammatory response (41).

Plasma concentrations of IL-17A (p=0.03) were higher in patients with uncomplicated infection than in healthy controls. IL-17A is produced by T cells and induces macrophage and neutrophil production of cytokines which in turn further activates the innate immune response. Levels of this cytokine in the sepsis and healthy cohorts were not statistically distinguishable. Compared to healthy patients, those with uncomplicated infection had higher levels of PDGF-AA (p=0.02), a cytokine known to dampen the immune response, and MDC (p=0.04), a chemokine known to promote neutrophil recruitment (42, 43).

Cytokines may drive immune cellular response or be reflective of those responses. To investigate the cellular immune response to sepsis, PBMCs were examined by flow cytometry. On day 1, total numbers of neutrophils (CD11b⁺, HLA-DR⁺/CD66b⁺, CD64⁺) and monocytes (CD11b⁺, HLA-DR⁺/CD14⁺) were higher in patients with AOD than the numbers in healthy patients (p=0.004 and p=0.02, respectively). Monocytes can be characterized by CD16 expression for inflammatory responses. Healthy patients had higher percentages of pro-inflammatory monocytes (CD16^-) as a percentage of the peripheral blood mononuclear cell population than did those patients with AOD (p=0.02) or sepsis (p=0.007). Monocytes that were CD206^- and CD163⁺ were present in higher proportion in patients with sepsis (p=0.01) ( Figure 2 ).

There were no differences in the total circulating CD4⁺ or CD8⁺ T cell numbers between cohorts. AOD patients had a lower percentage of CD4⁺ T cells as a percentage of circulating live cells compared to healthy patients (p=0.04). There were no differences in naïve CD4⁺ or CD8⁺ T cells between cohorts. Circulating memory T cells allow for a prompt response to an antigen that the host has previously encountered (44). On Day 1 of admission, there were differences in total circulating numbers of central, effector, and terminal effector memory (TEMRA) CD4⁺ and CD8⁺ T cells (45). Specifically, patients with AOD had less TEMRA (CCR7^-/CD45RA⁺) CD4⁺ T cells than healthy (p=0.02). The healthy and sepsis cohorts had several differences. There were less central memory (CCR7⁺/CD45RA^-) CD4⁺ and CD8⁺ T cells in sepsis compared to healthy (p=0.02 and 0.04, respectively). Sepsis patients had a higher number of circulating TEMRA CD8⁺ T cells compared to healthy (p=0.02). The percentage of TEMRA CD8⁺ T cells as a percentage of CD8⁺ T cells was lower in healthy patients than in sepsis (p=0.009) or in uncomplicated infection (p=0.04). The percentage of central memory CD8⁺ T cells was lower in sepsis compared to the healthy cohort (p=0.02). Figure 3 summarizes these results.

T-bet promotes the differentiation of effector CD8⁺ T cells from central memory T cells (46). In the sepsis cohort, a lower percentage of CD8⁺ T cells expressed T-bet than in healthy controls (12% vs 27%, p<0.001).

The sepsis cohort, compared to healthy, had a higher proportion on Day 1 of CD4⁺ T cells expressing regulatory markers (CD45RA⁺/FoxP3⁺) (22% vs 12% of CD4⁺, p=0.01). The sepsis cohort, compared to healthy, had a lower proportion of CD4⁺ T cells producing IL-10 without ex vivo stimulation (p=0.01). IL-10 limits the host response and is often produced by regulatory T cells. IL-10⁺ regulatory T cells were present in lower proportions in sepsis patients compared to the proportion present in the healthy cohort ( Figure 4 , p=0.04).

Ex vivo cytokine production in response to stimulation represents the cellular capacity to respond to a new challenge. Without stimulation, patients in the sepsis cohort had a lower proportion of CD4⁺ T cells producing IL-10 than did healthy controls. Both healthy and sepsis cohorts had a reduction in CD4⁺IL10⁺ T cells after stimulation; this reduction was more pronounced in healthy patients compared to sepsis patients (stimulation effect 7% of variation p<0.001, cohort effect 18% of variation p=0.001, stimulation by cohort effect 3% of variation p=0.029).

In comparing healthy and sepsis cohorts, older age and sepsis were both associated with higher Granulocyte Colony Stimulating Factor (G-CSF) levels. G-CSF level was influenced by both sepsis and age (<5 years vs >=5years) (interaction p=0.007; age p=0.009; sepsis presence p>0.001). Older age and sepsis presence were both associated with higher IL-6 levels. Sepsis presence and older age influenced the level of IL-6 (age p=0.035; sepsis presence p>0.0001). Older age and sepsis presence were both associated with higher IL-8 levels. Presence of sepsis and older age influenced the level of IL-8 (age p=0.032; sepsis presence p>0.0001). Presence of sepsis and older age influenced the level of IL-27 (interaction 3% p=0.056; age p=0.009; sepsis presence p>0.0001). Platelet derived growth factor (PDGF)-AA (p=0.018) and PDGF-BB (p=0.015) were both affected by the interaction between age and sepsis ( Figure 5 ). PDGF is produced by several cells including regulatory T cells and serves to modulate T cell behavior, enhancing IL-2 production and decreasing the production of several pro-inflammatory cytokines. This effect on cytokines suggests PDGF serves to dampen the immune response (43, 47, 48).

The magnitude of several of the adaptive immune subpopulations on Day 1 samples in the sepsis cohort correlated with patient age ( Figure 6 ). In sepsis, age positively correlated with the percent of CD4⁺ cells that were regulatory (CD45RA⁺FoxP3⁺), r² = 0.32, p=0.002. The regulatory CD4⁺ population in the other cohorts had no correlation with age. Age positively correlated with the percent of central memory CD4⁺ (CCR7⁺CD45RA^-) T cells in sepsis (r² = 0.36, p=0.001) and AOD (r² = 0.46, p=0.01). Healthy patients did not have a correlation with the percent of central memory CD4⁺ T cells and age. Age was inversely correlated with the percent of naïve (CCR7⁺/CD45RA⁺) CD4⁺ T cells in AOD (r² = 0.49, p=0.008) and sepsis (r² = 0.19, p=0.024). Age inversely correlated with the percentage of CD8⁺ T cells positive for IFNγ in sepsis alone (r² = 0.16, p=0.04) and the percentage of CD8⁺ T cells positive for TNFα in sepsis alone (r² = 0.22, p=0.013).

In our study of pediatric sepsis, we found that the magnitudes of the regulatory and central memory T cell responses increase as children age. Age was associated with several differences in peripheral blood cytokine levels, indicating that patients following adrenarche presented with higher levels of pro-inflammatory cytokines. PDGF, a cytokine produced by regulatory T cells to dampen the immune response, was higher in younger sepsis patients (47, 48). Younger patients had a higher proportion of circulating naïve T cells, but also had lower numbers of regulatory and memory T cells. These observations suggest that older children may mount a more pro-inflammatory response to sepsis that may be shaped by T cell immunity and that age is an important variable influencing immune dysregulation in sepsis.

T-bet promotes the differentiation of effector CD8⁺ T cells from central memory cells (46); both CD8⁺ central memory and CD8⁺Tbet⁺ T cells were lower in sepsis patients compared to healthy. Sepsis patients had higher regulatory CD4⁺ T cells but an overall more pro-inflammatory plasma cytokine milieu and higher levels of TEMRA CD8⁺ T cells. Further, CD4⁺IL10⁺ T cells, including IL10⁺ regulatory CD4 T cells, were lower in sepsis compared to healthy patients. These cells were also less functional in sepsis: after stimulating PBMC’s ex vivo with PMA/Io, only healthy samples had a statistically significant reduction in CD4⁺IL10⁺ T cells. This differential response suggests the sepsis patients had already mounted a full response and had reduced their IL-10 production, so therefore were unable mount a full response on subsequent ex vivo stimulation. These observations suggest either a failure of the regulatory T cell compartment in sepsis or a bias toward a more pro-inflammatory response. This may contribute to overall immune dysregulation.

A recent pediatric study found more TEMRA CD4⁺ T cells in sepsis patients, which is comparable to our finding of more TEMRA CD8⁺ T cells in sepsis (49). This study was smaller, limiting comparisons over age, and lacked a healthy cohort. A post-mortem study of sepsis patients demonstrated that spleen and lung tissue showed a decrease in both pro-inflammatory and anti-inflammatory cytokine production and upregulation of inhibitory ligands, though this associated with high levels of several markers of T cell activation. These results suggest multiple mechanisms of immune dysregulation in sepsis characterized by both altered cytokine production and decreased host ability to respond to circulating cytokines (50). Our study on PBMCs and plasma from living sepsis patients showed higher levels of many circulating cytokines, predominantly those that are pro-inflammatory (32–36, 51).

There were less IL-10 producing CD4⁺ T cells in sepsis patients and the ones that were present were less able to respond to ex vivo stimulation than their healthy counterparts, which is somewhat congruent with other sepsis observations of a regulatory T cell compartment failure (52). There are also reports of an augmented regulatory T cell compartment in adults with sepsis and septic shock, so perhaps the regulatory T cell compartment failure we observed is unique to children, or perhaps, our early timepoint captured a more acute phase of the sepsis response (52, 53). These studies failed to examine direct T cell intracellular IL-10 production which may further contribute to variation in our observations. This is an important area of future study as a failure of T cell IL-10 production may contribute to the immune dysregulation observed in sepsis.

Understanding immunologic changes over age may provide insight into the differences in immunity that may shape the immune response to neonatal and pediatric sepsis. Following birth, neonates are exposed to a natural environment allowing T cell memory to develop in response to multiple childhood viral and bacterial exposures. Our observation that memory CD4 T cells increase in sepsis as patients age is somewhat congruent with murine studies that show changes in the T cell memory compartment after exposure to a non-sterile environment (54–58).

Based on our observations, we propose that a possible mechanism of immune dysregulation in early pediatric sepsis involves a primary failure of the CD4 T cell regulatory response leading to an increased CD8 effector T cell response, altering the monocyte response. Through cell to cell and cytokine interactions, this increased CD8 effector T cell response contributes to systemic immune activation and organ dysfunction. This proposed mechanism requires further testing with mechanistic studies but could support clinical changes in the sepsis response that have been noted as neonates age through childhood to adulthood. Augmentation of the regulatory T cell response in older children represents a potential therapeutic target if our results can be replicated.

We performed several analyses with age as a variable to test our hypothesis that age impacts the immune response to sepsis. Interestingly, we found that age positively correlated with the percent of CD4⁺ central memory T cells in both sepsis and acute organ dysfunction. Age was also inversely correlated with the proportion of naïve CD4⁺ T cells in both sepsis and acute organ dysfunction. These findings suggest that a change in T cell phenotypes contributes to organ dysfunction regardless of the presence of infection. A relationship between age and memory and naïve T cell populations and how these variables influence organ dysfunction requires further investigation.

Our observations suggest that older children mount a more pro-inflammatory response to sepsis and that age is an important variable in the immune dysregulation of sepsis. We have also identified several aspects of the immune response which may drive these differences and warrant future study. Sepsis has a varied clinical presentation and disease course based on patient age, with neonates, younger children, older children, and adults all behaving differently. Further study into the differences in immune dysregulation over age may assist with therapeutic investigations targeting specific immune components based on patient age. Further, given the differing clinical variation between neonates, children and adults, specific immunologic manifestations that are detrimental to one group may be beneficial to another. Our work is the first step in understanding these concepts.

Sepsis is a heterogenous disease, and in this limited cohort we were unable to sub-analyze our data based on inciting organism, which is likely an important variable (49). Other variables we intend to explore in future studies but were unable to include in this multivariable model include comorbidities and steroid use. Larger patient pools are needed to compare the correlation between age and immune subpopulations between the four study cohorts. We could not control when in the disease course a child presented to the hospital, so we may be including samples of both early and mid-sepsis time points. Only one time point, Day 1, was analyzed because sufficient Day 3 and Day 7 samples were not available due to either quick improvement or early mortality of pediatric sepsis patients.

A notable strength of this study compared to other pediatric observational studies is our unique use of three comparison groups. We observed that those with sepsis, uncomplicated infection, and acute organ dysfunction without infection share several characteristics. We also observed several important differences that may provide mechanistic insights into distinctions between these processes. A further strength of our study was that our healthy cohort was not hospitalized and unexposed to anesthesia, so we believe is representative of the unstimulated immune state (59). Finally, our use of ex vivo stimulation allowed us to gain functional information concerning the T cell immune response to pair with our in vivo data, extending our ability to comment on potential mechanisms of sepsis biology.