Association of plasma level of high-mobility group box-1 with necroptosis and sepsis outcomes

The role of high-mobility group box-1 (HMGB1) in outcome prediction in sepsis is controversial. Furthermore, its association with necroptosis, a programmed cell necrosis mechanism, is still unclear. The purpose of this study is to identify the association between the plasma levels of HMGB1 and the severity and clinical outcomes of sepsis, and to examine the correlation between HMGB1 and key executors of necroptosis including receptor-interacting kinase 3 (RIPK3) and mixed lineage kinase domain-like- (MLKL) proteins. Plasma HMGB1, RIPK3, and MLKL levels were measured with the enzyme-linked immunosorbent assay from the derivation cohort of 188 prospectively enrolled, critically-ill patients between April 2014 and December 2016, and from the validation cohort of 77 patients with sepsis between January 2017 and January 2019. In the derivation cohort, the plasma HMGB1 levels of the control (n = 46, 24.5%), sepsis (n = 58, 30.9%), and septic shock (n = 84, 44.7%) groups were significantly increased (P < 0.001). A difference in mortality between high (≥ 5.9 ng/mL) and low (< 5.9 ng/mL) HMGB1 levels was observed up to 90 days (Log-rank test, P = 0.009). There were positive linear correlations of plasma HMGB1 with RIPK3 (R2 = 0.61, P < 0.001) and MLKL (R2 = 0.7890, P < 0.001). The difference in mortality and correlation of HMGB1 levels with RIPK3 and MLKL were confirmed in the validation cohort. Plasma levels of HMGB1 were associated with the severity and mortality attributed to sepsis. They were correlated with RIPK3 and MLKL, thus suggesting an association of HMGB1 with necroptosis.

Data collections. Clinical data, including patient demographics, reason for ICU admission, laboratory data, and severity of illness scores were collected by a trained research coordinator at the time of enrollment using the electronic health records. The sequential organ failure assessment (SOFA) 18 , simplified acute physiology score 3 (SAPS 3) 19 , and acute physiology and chronic health evaluation II (APACHE II) scores 20 were used to evaluate the disease severity. The primary outcome measure was the 28-day mortality. In-hospital mortality and 90-day mortality were measured as secondary outcomes.
The third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) was used to diagnose sepsis and septic shock 1 . Patients enrolled before the publication of Sepsis-3 were reviewed and re-categorized according to the new definition. The control group was defined as the patient group that did not meet the Sepsis-3 criteria.
Measurement of plasma HMGB1, RIPK3, and MLKL. In accordance with the purpose of the SMC-RoCI registry, the study protocol requires the collection of 19 mL whole blood samples from all the patients within 48 h of ICU admission. Centrifuged samples were stored at −80 °C for future use. The details of the study protocol and sample process are available from a previously published study 16 . Plasma HMGB1, RIPK3, and MLKL levels were measured with the enzyme-linked immunosorbent assay with a commercially available kit   23 . We used the log-rank test to compare the differences in the 90-day survival. The correlation analyses of HMGB1 with RIPK3 and MLKL were performed with linear regression based on the Pearson's correlation and Spearman's rank correlation coefficients. Statistical analysis was performed using SPSS (version 20.0, IBM, Chicago, IL, USA), and P < 0.05 was considered statistically significant.
Validation cohort of 77 patients consisted of 24 (31.2%) patients with sepsis and 53 (68.8%) patients with septic shock. The plasma levels of HMGB1 in patients with septic shock were significantly higher than those in sepsis patients [4.0 ng/mL (2.4-5.8) vs. 5.0 ng/mL (4.0-6.5), P = 0.023]. The validation cohort was then classified into the groups of high (n = 21) and low (n = 56) HMGB1 levels according to the optimal cut-off level of 5.9 ng/ mL determined from the derivation cohort (Table 3). Although the baseline characteristics of age, gender, comorbidities, clinical status on ICU admission, or severity of illness scores were similar in all the groups, the 28-day mortality was higher in the high-HMGB1 group (33.3% vs. 3.6%, P < 0.001). In addition, in-hospital and 90-day mortalities significantly differed between the two groups ( Table 3, Fig. 5). The positive correlations of plasma HMGB1 levels with those of RIPK3 (R 2 = 0.7394, P < 0.001) and MLKL (R 2 = 0.584, P < 0.001) were also observed in the validation cohort (see Supplementary Figures).

Discussion
Based on data analyses from 188 critically-ill patients and additional 77 patients with sepsis from the prospective SMC-RoCI registry, we observed that the plasma levels of HMGB1 were associated with the severity and mortality of sepsis. In addition, plasma levels of HMGB1 showed a positive linear relationship with those of RIPK3 and MLKL, key executors of RCD mechanism of necroptosis. The results of our study suggest that plasma HMGB1 may be a potential biomarker for severity assessment and mortality prediction of sepsis. Furthermore, the study and its outcomes suggest that necroptosis is likely a major source of plasma HMGB1 in sepsis.
The results of previous studies on the relationship between plasma HMGB1 and sepsis mortality are contradictory. Huang et al. 24 and Ueno et al. 25 demonstrated that high-plasma HMGB1 levels were associated with higher mortality in patients affected with sepsis and in septic shock patients who underwent polymyxin B hemoperfusion. However, other studies observed a negative relationship between the HMGB1 levels and prognosis 10,26,27 . Recent systematic review and meta-analysis of eight studies on sepsis patients found marginal differences in HMGB1 levels between survivors and nonsurvivors (pooled mean difference 1.2 ng/mL; 95% confidence interval (CI) 0.0-2.4; P = 0.05) 12 . In our study, a significant difference in mortality between patients with high-and low-HMGB1 levels determined by Youden's index was observed up to 90 days.
There are several possibilities for these conflicting results. First, a difference should be recognized in the statistical method used in our study compared with the previous studies. Unlike previous studies that compared plasma HMGB1 levels between survivors and nonsurvivors, we calculated Youden's index first to obtain the best www.nature.com/scientificreports/    www.nature.com/scientificreports/ cut-off point for the prediction of the 28-day mortality, and then compared the mortalities between the groups. Considering that there are certain advantages and disadvantages among statistical methods in evaluating the performance of biomarkers, we decided to calculate the optimal cut-off level to identify the discriminatory power as well as the clinical applicability of HMGB1 in sepsis rather than compare the levels between survivors and nonsurvivors 28,29 . Furthermore, we confirmed the result in the validation cohort using the same cut-off level from the derivation cohort. Another factor, in addition to the statistical difference that needs to be considered pertains to the chronological changes of the HMGB1 levels. In their study of 42 patients with septic shock, Gibot et al. noticed increases in HMGB1 levels between days 1 and 3 among nonsurvivors, whereas a progressive but statistically nonsignificant decrease was noted in HMGB1 levels among survivors 30 . In the same study, the level of HMGB1 at day 3 (and not that at day 0) was associated with increased mortality. Given the fact that HMGB1 is a late mediator of sepsis 31,32 , this result concurs with several studies advocating that delayed measurement of serial changes in HMGB1 may be predictive of mortality 11,33 . Moreover, in a study of sepsis patients with various sources of infection, the kinetics of HMGB1 release showed different patterns according to the primary source of infection 26 . These findings suggest that the timing of the sampling and source of infection may affect the significance of HMGB1 in the prediction of mortality owing to sepsis. However, we cannot confirm whether this is the reason for the conflicting results of the previous studies. In fact, withdrawal of blood was performed within 24 h of ICU admission in our study which may not reflect the delayed phase of HMGB1 elevation. In addition, late elevation was regarded as a consequence of active release from immune cells, while immediate surge may occur from cell damage or necrosis. Therefore, elevated HMGB1 in our cohort may by a manifestation of cell death. Given that our knowledge on the origin of HMGB1 in sepsis is limited, we are not able to draw inferences with certainty at this time. Additional studies identifying the kinetics of plasma HMGB1 are essential to better understand the role of HMGB1 as a biomarker and its meaning in the pathogenesis of sepsis. Finally, heterogeneity of sepsis may have contributed to the discordance. In a recent study on multiple observational cohorts of sepsis, Seymour et al. classified sepsis patients into novel phenotypes based on clinical and laboratory variables 34 . Each phenotype demonstrated distinct patterns of organ failure and clinical outcomes as well as distributions of biomarkers. Disparity in proportion of specific subtypes of sepsis patients among previous studies and ours have resulted in conflicting significance of HMGB1 in mortality prediction. Further research on the association of HMGB1 and sepsis phenotypes is necessary to determine the utility of HMGB1 as a biomarker. One of the most notable findings of our study is the strong association of plasma levels of HMGB1 with those of RIPK3 and MLKL, the key mediators of necroptosis 14,15 . HMGB1 can be actively secreted by immune cells, such as macrophages, dendritic cells, or monocytes in a delayed manner (> 6 h) as well as passively released through necrosis or RCDs other than apoptosis in an immediate manner 35 . Nevertheless, the source of extracellular release of HMGB1 is not yet clearly confirmed in sepsis. In vitro studies and animal models show a delayed surge of HMGB1 after lipopolysaccharide or endotoxin stimulation which suggests active secretion from immune cells as one of the mechanisms of HMGB1 release in sepsis 32,36 . By contrast, a few recent studies have reported that HMGB1 is released from caspase-independent necrotic-like cell deaths of macrophages and myoblasts, respectively 37,38 . The results support the hypothesis of necroptosis as a mechanism of an HMGB1 shift in sepsis. However, its association with RIPK1/RIPK3-dependent necroptosis has not been investigated. To our knowledge, we have demonstrated, for the first time, the association of plasma levels of surrogate markers of necroptosis activity RIPK3 and MLKL, with that of HMGB1 in patients with sepsis. The strong correlation presents compelling evidence that the early rise of HMGB1 may be mainly attributed to necroptosis for reasons other than other RCDs, cell necrosis, or activation of immune cells. Nonetheless, we are aware that this result does not provide direct evidence that extracellular HMGB1 originates from necroptosis in sepsis. Various mechanisms of extracellular HMGB1 secretion and complex pathways of necroptosis as well as possibility of HMGB1-mediated www.nature.com/scientificreports/ necroptosis activation limits definite conclusions 39 . Additional investigations, including single cell study and gene expression analysis are necessary to confirm this speculation. There are several limitations associated with our study. First, our study included patients referred from a single institution. Thus, the results may not be generalized to other settings or hospitals. Second, based on our study protocol, enrollment of a study patient and sampling of blood were completed within 24 h of ICU admission and 48 h of enrollment, respectively. Therefore, patients with high severity and early mortality may not have been included in the study. Interpretation of results requires caution given that there may be a selection bias. Third, the validation cohort of our study was drawn from a second prospective registry. Therefore, additional confirmatory cohort may be required.
In summary, plasma levels of HMGB1 were associated with severity and prognosis of patient with sepsis. Plasma HMGB1 was correlated with that of RIPK3 and MLKL suggesting its association with necroptosis.

Data availability
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.