Computational Derivation of Core, Dynamic Human Blunt Trauma Inflammatory Endotypes

Abstract

Systemic inflammation ensues following traumatic injury, driving immune dysregulation and multiple organ dysfunction (MOD). While a balanced immune/inflammatory response is ideal for promoting tissue regeneration, most trauma patients exhibit variable and either overly exuberant or overly damped responses that likely drive adverse clinical outcomes. We hypothesized that these inflammatory phenotypes occur in the context of severe injury, and therefore sought to define clinically distinct endotypes of trauma patients based on their systemic inflammatory responses. Using Patient-Specific Principal Component Analysis followed by unsupervised hierarchical clustering of circulating inflammatory mediators obtained in the first 24 h after injury, we segregated a cohort of 227 blunt trauma survivors into three core endotypes exhibiting significant differences in requirement for mechanical ventilation, duration of ventilation, and MOD over 7 days. Nine non-survivors co-segregated with survivors. Dynamic network inference, Fisher Score analysis, and correlations of IL-17A with GM-CSF, IL-10, and IL-22 in the three survivor sub-groups suggested a role for type 3 immunity, in part regulated by Th17 and γδ 17 cells, and related tissue-protective cytokines as a key feature of systemic inflammation following injury. These endotypes may represent archetypal adaptive, over-exuberant, and overly damped inflammatory responses.

Keywords: biomarker; critical illness; inflammation; network analysis; systems biology.

Figure 1

Patient-specific Principal Component Analysis followed by unsupervised hierarchical clustering of circulating inflammatory mediator data yields three distinct trauma survivor sub-groups. Patient-specific Principal Component Analysis was carried out for the Group of 227 trauma survivors using data on circulating inflammatory mediators obtained at three time points within the first 24 h of hospital admission. These data allowed patients to be clustered hierarchically using unsupervised methods as described in the Materials and Methods (A), resulting in three patient sub-groups: Group 1 (blue; n= 85 patients), Group 2 (red; n= 41 patients), and Group 3 (green; n = 101 patients). (B–D) Significant clinical outcome differences among Groups 1–3. (B): Group 2 (n = 41, 1.4 ± 0.9 days) showed significantly fewer days on ventilation over a time course of 8 days as compared to Group 1 (n = 85; 2.3 ± 0.4 days) and Group 3 (n = 101; 2.5 ± 0.6 days); p = 0.02. (C): The requirement for mechanical ventilation was significantly different across Group 1 (n= 85; 44 on vs. 41 off ventilation), Group 2 (n = 41; 12 on vs. 29 off ventilation), and Group 3 (n = 101, 39 on vs. 62 off ventilation) over a time course of 8 days; p = 0.0127. (D): Group 2 (n = 41) showed significantly lowered Marshall MODScores over a time course of 8 days as compared to Group 1 (n = 85), and Group 3 (n = 101); p = 0.0126.

Figure 2

Dynamic Network Analysis (DyNA) suggests distinct dynamic inflammatory programs in trauma patient sub-groups. Dynamic Network Analysis was carried out on the systemic inflammatory mediator data of Groups 1–3, and network complexity was quantified as described in the Materials and Methods. (A) Group 1 had a consistently higher network complexity than the other two sub-groups, while Group 2 network complexity stayed consistently low without any peaks. Group 3 network complexity stayed low at the same level as Group 2 for the first 6 days before increasing towards the level of Group 1 at day 7. (B) The total network connections for each inflammatory mediator in each trauma survivor sub-group were tallied. Group 1 showed the highest degree of connectivity per mediator over a time course of 7 days, with the most connected mediators being IL-5, IL-15, IL-1RA, MIP-1β, and IL-1β. Group 2 and Group 3 showed similar degrees of individual mediator connectivity though at a lower level than in Group 1. The most connected mediators in Group 2 were IL-21, IL-22, IL-33, IL-17E/IL-25, and IL-1β, whereas the most connected mediators in Group 3 were IL-17A, MIP-1α, IL-1RA, IL-15, and IL-33.

Figure 3

Dynamic Bayesian Network (DyBN) inference suggests distinct early inflammation programs in trauma patient sub-groups. Dynamic Bayesian Network inference was carried out on the systemic inflammatory mediator data of Groups 1–3 as described in the Materials and Methods. All groups contained a core motif consisting of IL-23 and IL-17E/IL-25, in which IL-23 was a central node. (A) Group 1 contained only the central motif of IL-23 and IL-17E/IL-25. (B) Group 2 showed the most complex inflammatory network, consisting of MCP-1 driven by MIG, along with of IL-22 driven by IL-23 and feedback by MIG. (C) Group 3 included MIG as a downstream mediator driven by IL-23.

Figure 4

Principal component analysis of trauma patient sub-group data suggests a role for Type 3 immunity, in part regulated by Th17 cells, in the circulating inflammatory response to traumatic injury. Principal component analysis was carried out on the systemic inflammatory mediator data of Groups 1–3 as described in the Materials and Methods. (A) In Group 1, IL-22, IL-33, IL-23 IL-17E/IL-25, IL-13, and IL-10 were the principal characteristics. (B) In Group 2 patients, IL-1β, IL-22, IL-13, IL-4, IL-33, and IL-17A were the principal characteristics. (C) In Group 3, IL-10, IL-13, IL-22, IL-4, IL-33, and IL-1β were the principal characteristics.

Figure 5

Fisher Score analysis points to Th17-related immune mediators the main differentiators among trauma patient sub-groups. Fisher Score analysis was carried out on the systemic inflammatory mediator data of Groups 1–3 as described in the Materials and Methods. The mediators IL-22, IL-33, and IL-17E/IL-25 were the mediators that best segregated among Groups 1–3, with IL-9, IL-21, IL-23, IL-1β, and IL-4 being the next best segregators.

Figure 6

Spearman correlations of IL-17A vs. GM-CSF, IL-10, or IL-22 suggest differential presence of IL-17A–producing T cell subsets in trauma patient sub-groups. Spearman Correlations were carried out using the data on IL-17A, GM-CSF, and IL-10 from days 0 to 7 post-admission in Groups 1–3. Significant correlations between IL-17A and GM-CSF were inferred to suggest the presence of pathogenic Th17 cells, significant correlations between IL-17A and IL-10 were inferred to suggest the presence of non-pathogenic Th17 cells, and significant correlations between IL-17A and IL-22 were inferred to suggest the presence of γδ 17 T cells. Group 1 showed no correlation between either IL-17A and GM-CSF (r = −0.05, p = 0.25) (A) or between IL-17A and IL-10 (r = 0.0737, p = 0.0738) (B). In contrast, Group 2 showed a positive correlation between IL-17A and GM-CSF (r = 0.30, p < 0.0001) (D) and a positive correlation between IL-17A and IL-10 (r = 0.24, p < 0.0001) (E). Group 3 showed only a positive correlation between IL-17A and GM-CSF (r = 0.24, p < 0.0001) (G) and no correlation between IL-17A and IL-10 (r = 0.02, p = 0.55) (F). Significant correlations between IL-17A and IL-22 suggest the presence of γδ 17 T cells in all groups: (C) Group 1 (r = 0.12, p = 0.004), (F) Group 2 (r = 0.16, p = 0.007), (I) Group 2 (r = 0.25, p < 0.0001).

Figure 7

Dynamic Spearman correlations of IL-17A vs. GM-CSF, IL-10, or IL-22 suggest distinct trajectories of IL-17A–producing T cell subsets in trauma patient sub-groups. Spearman Correlations were carried out over 1-day time intervals using the data on IL-17A, IL-22, GM-CSF, and IL-10 in Groups 1–3. Significant correlations between IL-17A and GM-CSF were inferred to suggest the presence of pathogenic Th17 cells, significant correlations between IL-17A and IL-10 were inferred to suggest the presence of non-pathogenic Th17 cells, and significant correlations between IL-17A and IL-22 were inferred to suggest the presence of γδ 17 T cells. (A) In Group 1, non-pathogenic Th17 cells reached significant (p < 0.05) r values between days 3 and 5 post-admission and were inferred to predominate over pathogenic Th17 cells from day 2 to 7. (B) In Group 2, both pathogenic and non-pathogenic Th17 cell subsets showed similar r values, with a steady rise during the time course of 7 days after admission. (C) In Group 3, pathogenic Th17 cells appeared to predominate over non-pathogenic Th17 cells up to day 2. Thereafter, pathogenic and non-pathogenic Th17 cell subsets appeared to follow similar dynamics through day 7. Significant correlations between IL-17A and IL-22 were observed only in Group 1 (days 3–4) and Group 3 (all days). For exact r- and p-values, see Table S1 .