Skip to main content
Annals of Intensive Care
Download PDF

Development of a biomarker prediction model for post-trauma multiple organ failure/dysfunction syndrome based on the blood transcriptome

Annals of Intensive Care volume 14, Article number: 134 (2024) Cite this article

Abstract

Background

Multiple organ failure/dysfunction syndrome (MOF/MODS) is a major cause of mortality and morbidity among severe trauma patients. Current clinical practices entail monitoring physiological measurements and applying clinical score systems to diagnose its onset. Instead, we aimed to develop an early prediction model for MOF outcome evaluated soon after traumatic injury by performing machine learning analysis of genome-wide transcriptome data from blood samples drawn within 24 h of traumatic injury. We then compared its performance to baseline injury severity scores and detection of infections.

Methods

Buffy coat transcriptome and linked clinical datasets from blunt trauma patients from the Inflammation and the Host Response to Injury Study (“Glue Grant”) multi-center cohort were used. According to the inclusion/exclusion criteria, 141 adult (age ≥ 16 years old) blunt trauma patients (excluding penetrating) with early buffy coat (≤ 24 h since trauma injury) samples were analyzed, with 58 MOF-cases and 83 non-cases. We applied the Least Absolute Shrinkage and Selection Operator (LASSO) and eXtreme Gradient Boosting (XGBoost) algorithms to select features and develop models for MOF early outcome prediction.

Results

The LASSO model included 18 transcripts (AUROC [95% CI]: 0.938 [0.890–0.987] (training) and 0.833 [0.699–0.967] (test)), and the XGBoost model included 41 transcripts (0.999 [0.997–1.000] (training) and 0.907 [0.816–0.998] (test)). There were 16 overlapping transcripts comparing the two panels (0.935 [0.884–0.985] (training) and 0.836 [0.703–0.968] (test)). The biomarker models notably outperformed models based on injury severity scores and sex, which we found to be significantly associated with MOF (APACHEII + sex—0.649 [0.537–0.762] (training) and 0.493 [0.301–0.685] (test); ISS + sex—0.630 [0.516–0.744] (training) and 0.482 [0.293–0.670] (test); NISS + sex—0.651 [0.540–0.763] (training) and 0.525 [0.335–0.714] (test)).

Conclusions

The accurate assessment of MOF from blood samples immediately after trauma is expected to aid in improving clinical decision-making and may contribute to reduced morbidity, mortality and healthcare costs. Moreover, understanding the molecular mechanisms involving the transcripts identified as important for MOF prediction may eventually aid in developing novel interventions.

Background

Trauma is among the leading causes of morbidity, mortality, increased length of stay and healthcare costs [1,2,3]. Multiple organ failure/dysfunction syndrome (MOF/MODS) is one major adverse outcome with a high incidence among trauma patients [4,5,6,7], who experience acute and prolonged immune dysregulation [8,9,10,11,12,13,14,15] and a high incidence of infections [16,17,18]. MOF/MODS is identified as a significant source of mortality and resource consumption in this population [19, 20], suggesting that timely detection of post-trauma MOF/MODS soon after injury to achieve appropriate and efficient delivery of early preventative and management measures is expected to improve patient outcomes and mitigate healthcare costs. Moreover, identifying novel clinical factors and molecular mechanisms associated with MOF to elucidate mechanisms underlying its development is expected to be impactful.

Current clinical practices for diagnosing patients entail monitoring MOF/MODS-specific physiological score systems such as the Denver [21], Marshall multiple organ dysfunction score (MODS) [22] or sequential organ failure assessment (SOFA) [23] scores to detect its onset. Various studies have assessed the ability of common injury severity scores computed soon after admission, including the Acute Physiology and Chronic Health Evaluation (APACHE) II [24], Injury Severity Score (ISS) [25] and New Injury Severity Score (NISS) [26] as predictors of trauma-related MOF/MODS [27,28,29,30] and infections [31, 32]. However, these scores are limited in accuracy and timeliness of outcome detection and are based on gross clinical measures that do not account for individual molecular responses to injury. Indeed, it has been reported previously that injury severity scores and immune responses are not consistent in their ability to predict clinical outcomes post-trauma [33], and it has been noted that novel methods based on molecular biomarkers are needed to improve monitoring MODS [34].

Given that trauma patients are at especially high risk for MOF/MODS, developing novel biomarkers for accurate prediction is imperative. Novel machine learning (ML) algorithms provide immense potential to support the implementation of personalized medicine approaches using genome-wide data to ameliorate deficiencies of current practices involving clinical scores generally across all patients. Injury severity scores are limited as a method for populational assessment, rather than a valid approach for prognostication at the individual level, which gene expression signatures would allow. Such an ML-based approach is expected to maximize the information obtained from each patient and aid in developing accurate prediction methods to improve clinical decision-making, enhance resource allocation and augment the quality and cost-effectiveness of patient care [35,36,37]. It has been noted that biomarkers to improve critical care is needed and that additional studies to determine which combinations of biomarkers can give optimum results are of immense interest [38].

Studies among trauma and burn patients using ML analysis of blood transcriptomic data to develop biomarker panels for the early prediction of infections have consistently shown that they significantly outperformed various injury severity scores [39,40,41]. Moreover, the advantage of the molecular profiling approach is highlighted by the uncovering of novel mechanisms. These studies suggest that applying ML to early blood transcriptomic data is likely a feasible method to develop prediction models for adverse post-trauma outcomes, including MOF/MODS, that are more accurate than clinical scores and aid in elucidating molecular factors involved.

Although there are previous studies that have leveraged transcriptomic data collected in trauma centers to discover transcripts associated with MODS [14], sepsis [42,43,44] and other poor outcomes post-injury [45], they were aimed at identifying differentially regulated transcripts rather than developing early prediction models. Another study that developed MOF prediction models evaluated common cytokines [46], rather than employing an unbiased ML approach. One study used the Least Absolute Shrinkage and Selection Operator (LASSO) and Elastic Net (EN) ML algorithms and identified decreased CD62L and CD63 neutrophil expression and CD63 monocyte expression as predictors of MODS, and showed improvement in performance over NISS [47]. Taken together, no previous study has applied ML analysis of genome-wide transcriptome data from early blood samples to develop and validate prediction models for MOF/MODS and make comparisons with different common injury severity scores. Moreover, in addition to the LASSO and EN penalized regression methods, more recently developed algorithms including eXtreme Gradient Boosting (XGBoost) [48] has yet to be used to develop prediction models for trauma-related outcomes. The SHapely Additive exPlanations (SHAP) [49, 50] scores can be determined to evaluate which features in the model contribute to the outcome prediction, making XGBoost more interpretable, highlighting its advantage. Thus, we employed both LASSO and XGBoost, which is novel and expected to improve the development of prediction models relevant to post-trauma outcomes.

We developed highly accurate early prediction methods for post-trauma MOF outcome, based on genome-wide transcriptomic data collected from early blood samples collected within 24 h of injury. This approach is expected to significantly improve the accuracy of early identification of trauma patients at risk of MOF using blood samples collected at triage for implementing risk stratification strategies to help improve patient outcomes. Potential early interventions for patients found to be at increased risk of MOF may include immediate admission to the Intensive Care Unit (ICU) rather than the Step-Down Unit, more aggressive hemodynamic and culture monitoring, improved resuscitation precision by implementing higher levels of monitoring and more aggressive drainage of infections. MOF risk evaluation may also be included as a factor among others for borderline resuscitation decisions and providing information to the patients’ family. Early MOF risk evaluation may also be included among factors for Crisis Standards of Care or military triage. MOF risk assessment is also expected to be advantageous for standardizing patient group selection in clinical research. Identifying novel molecular markers of MOF is also expected to enhance understanding of underlying mechanisms, which may aid in improving preventive approaches and therapeutics development.

Methods

Study design and population

Patient clinical and transcriptomic data were obtained from the Glue Grant (“Inflammation and Host Response to Injury”) cohort [51], a multi-center cohort that enrolled patients at US Level 1 trauma centers between 2003 and 2009. The sample collection/storage and data generation were performed by the Glue Grant Consortium, and the permission for the access and secondary analysis of de-identified data was obtained from the Massachusetts General Hospital Institutional Review Board (MGH IRB protocol 2002P001743).

Inclusion/exclusion criteria

Among the 2,002 patients in the dataset we obtained from the Glue Grant, 141 patients were included in the analyses. Our inclusion/exclusion criteria were as follows: adults aged ≥ 16 years old (excluding those < 16 years old) who sustained blunt trauma (excluding penetrating injury only or blunt with penetrating injury) with buffy coat collected early, within ≤ 24 h since trauma injury (excluding those without buffy coat collection or those collected after 24 h) and transcriptome data of high RNA quality of ≥ 3 out of 4 (excluding those with RNA quality below 3 and removing chips identified as outliers) (Fig. 1A, B). Where patients had multiple microarrays consistent with the inclusion criteria, the earliest timepoint was used. There were 58 MOF cases and 83 non-cases in total included in the study.

Fig. 1
figure 1

Description of the patient population and study design. A Schematic of the study design and timing of the blood sample collection, and B patients who were included/excluded in the study

Full size image

Study variables

Clinical scores (APACHEII, ISS, NISS, Denver, Marshall) were recorded by participating institutions according to the guidelines outlined by the Glue Grant Consortium. Body mass index (BMI) was calculated from recorded height and weight (weight in kilograms/height in meters2). Patients were assigned to MOF-cases and non-cases, according to the Glue Grant Study’s recorded MOF onset day, indicated as using the criteria of Marshall score without the Glasgow Coma Scale ≥ 6.

Software and packages

R version 4.3.2 was used for the analyses.

Baseline characteristics calculations

For the analyses, MOF-cases (according to the Glue Grant Study’s criteria) were compared with non-cases, and the same criteria were used to develop the outcome prediction model. Baseline characteristics are reported means ± standard deviation (SD), or total numbers with proportions (%), as indicated in the legend (Table 1). Means between MOF-cases versus non-cases were compared by the unpaired equal variance two-tailed t-test. For comparing proportions, the Chi-square test was used for all expected values of 5 or greater, or Fisher’s exact test for an expected value below 5.

Table 1 Baseline characteristics of the study population
Full size table

Microarray analysis

The GCRMA [52] package (version 2.74.0) was used to process CEL files to normalized log2 expression values of probe sets. The arrayQualityMetrics [53] package (version 3.58.0) and principal components analysis with the factoextra [54] package (version 1.0.7) were used to remove outlier chips. Internal control and low abundance probe sets (i.e., log2 expression value < 3 across all samples) were filtered, reducing the number of probe sets from 54,675 to 28,888 for the subsequent analyses. The limma [55] package (version 3.57.10) was used to calculate the log2 fold change values and false discovery rate (FDR)-adjusted p-values to compare MOF-cases versus non-cases. The top 500 most variable probe sets that also showed at least 1.2-fold change (105 probe sets) were used for the subsequent biomarker development.

Machine learning prediction model development

The data was randomly split into 70% training (n = 100) and 30% test (n = 41) sets using the Caret [56] package (version 6.0.94). The Glmnet [57] package (version 4.1.8) was used to implement the least absolute shrinkage and selection operator (LASSO) regression to select probe sets that were predictive of MOF. The expression levels were standardized with the training set as the reference, using the Caret [56] package. The penalty weight, lambda (λ) that minimized the deviance was identified by performing tenfold cross-validation (CV) with 100 repeats. Probe sets were selected according to the hyperparameter values of ⍺ = 1 and λ = 0.0395, which yielded 19 probe sets, mapping to 18 transcripts. A multivariable logistic regression model was constructed for the outcome of MOF onset and the maximum likelihood coefficient estimates were obtained for the model with the 19 probe set predictors selected with LASSO.

To develop the eXtreme Gradient Boosting (XGBoost) model, Bayesian optimization was performed with tenfold CV on the training set to determine the hyperparameter values that maximized the mean test AUROC, using the ParBayesianOptimization [58] package (version 1.2.6) (learning rate = 0.146, maximum depth of a tree = 4, gamma = 1.519, minimum child weight = 2.800, subsample ratio = 0.851, column sample ratio = 0.971, L1 regularization = 0, L2 regularization = 2.829, number of boosting rounds = 45). The SHapely Additive exPlanations (SHAP) scores were found for each probe set with the SHAPforXGBoost [59] package (version 0.1.3), which yielded 42 probe sets, mapping to 41 transcripts with mean SHAP score above 0.

For identifying overlapping transcripts, mapped transcript names between those identified with LASSO and XGBoost were compared, and a Venn diagram was drawn using the ggvenn [60] package (version 0.1.10). From the LASSO model, the two probe sets mapping to the transcripts that were not included in the XGBoost model were removed, and a multivariable logistic regression model was constructed for the outcome of MOF onset and the maximum likelihood coefficient estimates were obtained. Multivariable logistic regression models were also constructed with various injury severity scores (APACHEII, ISS, NISS) and sex. All the models above were initially constructed in the training set and then evaluated in the test set.

The area under the receiver operating characteristic curve (AUROC) with DeLong 95% confidence intervals was calculated using the pROC [61] package (version 1.18.5). The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated using the epiR [62] package (version 2.0.66).

Functional assessment

Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were conducted for this panel of probe sets using the pathfindR [63] package (version 2.3.0). Terms with at least two genes included where plotted using the ggplot2 [64] package (version 3.4.4). A network plot was constructed using the GeneMANIA Cytoscape plug-in [65].

Results

Patient demographics and baseline characteristics show significantly higher injury severity scores and proportion of males among MOF-cases

Baseline demographic and injury characteristics, as well as clinical outcomes of the 141 blunt trauma patients included in the study (Fig. 1A, B) are presented in Table 1. Motor vehicle collisions were the most frequent injury mechanisms, and no significant differences in injury types and characteristics were found between MOF-cases and non-cases (Supplementary Table S1). The overall study population consisted of patients with a mean age of 33.9 ± 11.0 years old, with 37.6% females and 62.4% males. Age was not significantly different between MOF-cases and non-cases. However, the relative proportion of males was significantly higher for MOF-cases compared to non-cases (75.9% male versus 24.1% female MOF-cases and 53.0% male versus 47.0% female non-cases, p = 0.010). Baseline injury severity scores generally tended to be higher for MOF-cases compared to non-cases, comparing APACHEII (28.9 ± 5.7 for MOF-cases versus 27.1 ± 5.7 for non-cases, p = 0.073), ISS (34.3 ± 12.8 for MOF-cases versus 29.9 ± 14.3 for non-cases, p = 0.058) and NISS (38.5 ± 12.1 for MOF-cases versus 35.3 ± 13.7 for non-cases, p = 0.155). The average day to MOF diagnosis since injury among cases was 3.2 ± 2.3 days. On average, the maximum organ-specific MOF scores recorded were highest for the central nervous system score, followed by cardio score, respiratory score, renal score, hepatic score and finally, hematologic score (Supplementary Table S2).

MOF is associated with worse patient outcomes

There were seven total patients who did not survive, who were all among the MOF-cases (12.1% among MOF-cases vs. 0% among non-cases, p = 0.002) (Table 1). The main causes of death for these patients were attributed to MOF, sepsis, shock, head injury, hypoxia and brain death (Table 2). MOF-cases, compared to the non-cases, also had significantly longer days in the ICU (19.7 ± 10.8 among MOF-cases vs. 8.7 ± 6.5 among non-cases, p < 0.001), on the ventilator (15.9 ± 9.3 among MOF-cases vs. 5.9 ± 5.0 among non-cases, p < 0.001) and discharge day since injury (32.3 ± 19.4 among MOF-cases vs. 20.7 ± 14.0 among non-cases, p < 0.001).

Table 2 Causes of death among non-survivors
Full size table

Infections were significantly increased among MOF-cases, and mostly occurred after MOF, rather than before

Among all patients, 63.8% had a record of infections during follow-up, regardless of MOF development (Table 1). Among MOF-cases, 86.2% had an incidence of infection at any time during follow-up, which was significantly higher than 48.2% among non-cases (p < 0.001). When assessing the timing of infection relative to MOF among cases, 22.4% were found to have the first record of an infection before or on the day of MOF diagnosis (n = 6 within 7 days and n = 7 on the same day), whereas 63.8% of patients had the first record of an infection after the MOF onset date (n = 23 at 1–7 days, n = 9 at 8–14 and n = 5 at 15–26 days) (Table 1, Supplementary Table S3). Overall, among MOF-cases, pneumonia (75.0%) was the most common infection type, followed by surgical site infections (42.3%), bloodstream infections (40.4%) and urinary tract infections (23.1%); similarly, among the non-cases, pneumonia (42.5%) was the most common, then surgical site and urinary infections (both 35.0%), then bloodstream infections (22.5%) (Table 3). Comparing infection types detected before and after MOF onset, pneumonia incidence was similar before and after MOF (69.2% before vs. 64.6% after), whereas for other infections, the proportion was larger after MOF onset.

Table 3 Microorganisms and infection types detected (NOS: not otherwise specified)
Full size table

The top three microorganisms detected overall among MOF-cases were Staphylococcus aureus (30.8%), Enterobacter species (26.9%) and Acinetobacter (26.9%), compared to non-cases with Staphylococcus aureus (24.4%), Enterobacter species (22.0%) and Coagulase-negative staphylococci (17.1%) (Table 2). Species with a higher proportion detected in infections before or on MOF onset day versus after included Enterobacter species (30.8% before or on MOF onset day vs. 22.9% after), Haemophilus influenza (15.4% before or on MOF onset day vs. 12.5% after), E.coli (15.4% before or on MOF onset day vs. 4.2% after) and other gram positives (23.1% before or on MOF onset day vs. 8.3% after) (Table 3). Pathogens not detected before or on MOF onset day, but only detected in infections after or among non-cases, included Klebsiella pneumoniae, Serretia marcescens, Coagulase-negative staphylococci, Bacterioides, other gram negatives and fungi species (Table 2). Polymicrobial infections were detected more frequently among MOF cases (60.0%) compared to non-cases (42.5%) (Table 3).

The predictive biomarker model significantly improved early MOF detection compared to the injury severity models

The model developed using LASSO feature selection, included 19 probe sets, mapping to 18 transcripts and the XGBoost model identified 42 probe sets, mapping to 41 transcripts as being important for MOF outcome prediction. There were 16 overlapping transcripts in the LASSO and XGBoost models, suggesting their importance, and a total of 43 transcripts were identified across both models (Fig. 2A). A review of previous literature found that many of the 16 transcripts found by both LASSO and XGBoost have been implicated in traumatic injuries and organ failure/dysfunction, or immune cell functions (Table 4). The model developed using LASSO showed accurate MOF outcome prediction (AUROC [95% CI]: 0.938 [0.890–0.987] in the training set and 0.833 [0.699–0.967] in the test set) (Fig. 2B–D; Supplementary Table S4, S5), and the XGBoost model showed exceptional prediction performance (0.999 [0.997–1.000] in the training set and 0.907 [0.816–0.998] in the test set) (Fig. 2B, C, E; Supplementary Table S4). We further evaluated a simpler multivariable logistic regression model based on the 16 overlapping transcripts and found its performance to be similar to that of the model based on LASSO (0.935 [0.884–0.985] in the training set and 0.836 [0.703–0.968] in the test set) (Fig. 2B, C, F; Supplementary Table S4, S5). The biomarker-based models were notably better performing compared to models based on clinical variables of injury severity scores and sex (APACHEII + sex—0.649 [0.537–0.762] in the training set and 0.493 [0.301–0.685] in the test set; ISS + sex—0.630 [0.537–0.744] in the training set and 0.482 [0.293–0.670] in the test set; NISS + sex—0.651 [0.540–0.763] in the training set and 0.525 [0.335–0.714] in the test set), each showing the lack of ability to predict MOF (Fig. 2B, C, G; Supplementary Table S5).

Fig. 2
figure 2figure 2figure 2

Description of the predictive biomarkers identified and comparisons of the performance of the various MOF outcome prediction models. A Summary of the transcript names that were unique or overlapping between the LASSO and XGBoost models. ROC curves and AUROC [95% CI] of each of the models constructed (LASSO biomarker panel with 19 probe sets mapping to 18 transcripts; XGBoost biomarker panel model with 42 probe sets mapping to 41 transcripts; overlapping 16 transcript panel (i.e. the LASSO panel minus the probe sets mapping to the two transcripts that were not included in the XGBoost panel); APACHEII + sex; ISS + sex; and NISS + sex), where results are shown for the models developed in the (B) training set, and then applied to the (C) test set. (D) The coefficient estimates for the LASSO model, (E) mean SHAP scores for the XGBoost model, (F) coefficient estimates for the overlapping transcripts model, and (E) coefficient estimates for the different clinical models are shown

Full size image
Table 4 Previous literature on the overlapping 16 transcripts with their potential relevance to MOF
Full size table

The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were also notably high for the biomarker prediction model (LASSO model—sensitivity 0.878, specificity 0.949, PPV 0.923 and NPV 0.918 in the training set, and sensitivity 0.765, specificity 0.750, PPV 0.684 and NPV 0.818 in the test set; XGBoost model—sensitivity 0.951, specificity 1.000, PPV 1.000 and NPV 0.967 in the training set, and sensitivity 0.833, specificity 0.706, PPV 0.800 and NPV 0.750 in the test set; overlapping biomarkers model—sensitivity 0.878, specificity 0.932, PPV 0.900 and NPV 0.917 in the training set, and sensitivity 0.765, specificity 0.750, PPV 0.684 and NPV 0.818 in the test set), as compared to the other models based on clinical variables (Table 5). Overall, the XGBoost model showed improved performance compared to the LASSO and overlapping transcripts models. Furthermore, we compared MOF detection overall versus according to subgroups of impacted organs and found no significant difference separately by any specific organ (Supplementary Table S6), suggesting that the prediction models could be useful despite some heterogeneity in the MOF outcome presentation.

Table 5 Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of the various models constructed
Full size table

Molecular functions and pathways associated with the predictive biomarkers

We evaluated the functional annotations of the 43 predictive transcripts identified by both models. Significant enrichment was detected for GO terms related to nuclear envelope, endoplasmic reticulum lumen, cell adhesion and proteolysis (Fig. 3A), and KEGG pathway terms related to signaling (HIF-1 and oxytocin signaling, nucleocytoplasmic transport and RNA degradation), metabolism (glycolysis/gluconeogenesis and fructose and mannose metabolism) and proteoglycans in cancer (Fig. 3B). Moreover, network analysis found that the transcripts are highly interconnected (Fig. 3C).

Fig. 3
figure 3

Evaluation of the molecular roles associated with the 43 predictive transcripts. Enrichment analysis evaluating (A) Gene Ontology (GO) and (B) KEGG pathway terms, and (C) network plot are shown

Full size image

Discussion

The results of our study demonstrate that indeed, MOF is a major adverse outcome among trauma patients and is strongly associated with increased mortality and prolonged length of stay. These observations support the notion that improved management of post-trauma MOF by developing better methods for early risk stratification is expected to improve patient outcomes and alleviate healthcare cost expenditure. It has been shown previously that acute organ dysfunction among sepsis patients is associated with both short-term and long-term mortality [66], which may indicate that the adverse impact of MOF could be larger than observed at the end of follow-up.

Our results show that MOF mostly occurred soon after traumatic injury, suggesting that the immune dysregulation from the impact of the trauma itself is significant, highlighting the importance of our study to analyze early (within 24 h of injury) blood molecular response to injury for subsequent outcomes. While each of the baseline injury severity scores of APACHEII, ISS and NISS tended to be higher among MOF-cases, showing that higher impact of traumatic injury may indeed render patients at increased risk, none of them were effective in MOF outcome prediction. These results support the notion that performing ML analysis of genome-wide transcriptome data to characterize patients’ responses to injury more meticulously, and using this information to develop accurate biomarker prediction models is important.

Our ML analysis identified 18 transcripts by LASSO and 41 by transcripts by XGBoost to be included in the outcome prediction model, with 16 transcripts overlapping between the two models, suggesting that they are likely important as mechanisms related to MOF. As expected, most of the transcripts have been found previously to be linked with traumatic injuries, organ failure/dysfunction and immune responses. As summarized in Table 5, various transcripts identified to be important by both LASSO and XGBoost were previously implicated in traumatic injuries and organ failure/dysfunction (ADK, ARNT/HIF1B, CAMK1D, RPH3A, SLC25A37 and ZFAND2A) and immune cell functions (IKZF1, PFKP, PKNOX1 and TPI1). Other transcripts have not yet been linked to traumatic injuries or inflammatory responses (ADGRB3-DT, CEP68, IRAG1, PDCD4-AS1, YBEY and ZFY), and future molecular studies to test their potential role in responses to traumatic injuries, organ failure or immunity may result in novel understanding of their mechanisms.

The proportion of males among MOF-cases was significantly higher than non-cases, which has been reported previously [67,68,69,70]. Although previous studies have suggested sex hormone signaling [70] or increased Interleukin-6 (IL-6) among males [68] as possible important molecular mechanisms, there has not yet been a study that investigated transcriptome differences, and a future study with a sufficiently large sample size to allow stratified analysis by sex is expected to be highly informative. In our study, X-inactive specific transcript (XIST), a key initiating signal for X-inactivation, was found among transcripts important for MOF outcome. XIST overexpression in the serum of acute pneumonia patients has been reported, showing its relevance in immune response [71]. Various studies have demonstrated that XIST can exert its immunomodulatory functions by binding to miRNAs and acting as a competing endogenous RNA (ceRNA)—its targets include miR-370-3p, a negative regulator of Toll-like receptor 4 (TLR4) [71]; miR-132-3p, which controls the mitogen activated protein kinase 14 (MAPK14) pathway [72]; miR-142-5p, which suppresses Programmed cell death protein 4 (PDCD4) [73]; and miR-133a, which inhibits Suppressor of cytokine signaling 2 (SOCS2) [74]. In these studies, XIST knockdown was found to be protective of LPS-induced apoptosis and inflammation [71]; acute lung injury [72]; acute kidney injury [73]; or myocardial ischemia reperfusion injury [74]. XIST inhibition was also described to mitigate sepsis-induced acute liver injury by suppressing Bromodomain-containing Protein 4 (BRD4) expression [75]. On the other hand, XIST was found to promote burn wound healing by suppressing miR-19b to enhance IL-33 expression and M2 macrophage polarization [76]. Our study also identified the importance of USP9Y, a male-specific transcript, previously found to be overexpressed in myocardial samples with heart failure [77]. Given our findings, further mechanistic studies to understand the impact of biological sex in MOF are expected to be highly informative.

Mortality was found only among MOF-cases, demonstrating the adverse effect of MOF. The main causes of death for these patients were attributed to MOF, or a variety of other causes, including sepsis, shock, head injury, hypoxia and brain death. It is uncertain whether the gene expression signatures this study identified to be relevant to MOF outcome may also be related to other clinical syndromes, and further assessment would be informative.

Recent studies have challenged the notion that bacterial infections are a major trigger of MOF among trauma and surgical patients, and suggested that on the other hand, MOF can also be a major contributor to infections [78]. While we found that the incidence of infections is significantly greater among MOF-cases compared to non-cases overall, MOF tended to precede infections, rather than being a consequence of them, as a previous study also reported [79]. Thus, infections appear to be both a driver and a consequence of MOF in this setting. In addition to MOF-cases, infections were also detected among a notable proportion of non-cases, consistent with previously described post-injury immune dysregulation. Trauma triggers the secretion of danger-associated molecular patterns (DAMPs) to induce a hyper-inflammatory state, termed systemic inflammatory response syndrome (SIRS) associated with early MOF, and as a counter response, also triggers the suppressing/inhibiting DAMPs (SAMPs) to induce the compensatory anti-inflammatory response syndrome (CARS), described as an immunosuppressive state that renders patients highly susceptible to nosocomial infections [9, 80,81,82]. In our study, we further characterized potential differences between MOF-cases and non-cases by specific infection types and microorganisms detected. Notably, we found that pneumonia incidence was higher among MOF cases compared to non-cases, with similar incidence before and after MOF. Surgical site and bloodstream infections were also more frequent among MOF cases compared to non-cases; however, the incidences increased after MOF compared to before. Interestingly, urinary tract infection incidence was lower among MOF-cases compared to non-cases, and moreover, among MOF-cases, the incidence was higher after MOF. One potential caveat is that determining infections requires swabs to be taken in a timely manner and culturing on specific selection plates, and therefore, it is possible that not all infections can be detected accurately. Nevertheless, our results suggest that further studies to characterize the timing of the association between MOF and specific types of infections are expected to provide additional insights into the mechanism of association between MOF and infections.

This study provides proof-of-concept results for the advantage of biomarker development based on ML analysis of genome-wide data to understand the molecular responses to traumatic injury that renders patients at increased risk for adverse outcomes, including MOF. The application of the results presented here in the clinical setting would be to develop a rapid assay to selectively measure the predictive transcripts from routine blood drawn at admission to the hospital, to allow calculating the predicted probability of MOF. Such a method for early patient stratification by adverse outcome risk is expected to enhance clinical decision making and expected to aid in the early implementation of surveillance and intervention strategies to mitigate the risks. As such, it may result in improving patient outcomes and alleviating healthcare costs. Early intervention strategies for patients found to be at increased risk of MOF may include immediate ICU admission, increased hemodynamic and culture monitoring, increased precision of resuscitation and more aggressive drainage of infections. Such MOF risk assessment may also be included as a factor for borderline resuscitation decisions and providing information to the patients’ family. It may also be advantageous for Crisis Standards of Care or military triage, as well as patient selection in clinical research. Given that MOF is known to be a major adverse outcome among trauma patients and is significantly associated with mortality and increased length of stay, as our results also show, novel methods to alleviate the burden of MOF are imperative.

To overcome limitations of this study, future studies to externally validate the biomarker prediction method in new large and diverse trauma patient populations, and mechanistic studies to determine how the transcripts may be related to MOF onset would strengthen our findings. Such validation and mechanistic studies may contribute to the development of novel preventative and therapeutic agents in the future.

Conclusions

Applying ML to analyze genome-wide transcriptome data from early blood samples collected within 24 h of traumatic injury resulted in the development of an accurate prediction model for MOF based on 41 associated transcripts. The biomarker-based prediction models provided a significantly better prediction of MOF compared to those based on injury severity scores and sex (APACHEII, ISS and NISS) or the detection of infections, highlighting the importance of exploring novel molecular medicine approaches for early risk stratification.

Availability of data and materials

The data that support the findings of this study are available from the Glue Grant Study, but restrictions apply to the availability of these data, which were used under approval for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request, with permission of the Glue Grant Consortium.

Abbreviations

ML:

Machine learning

LASSO:

Least absolute shrinkage and selection operator

XGBoost:

EXtreme gradient boosting

APACHEII:

Acute physiology and chronic health evaluation II

ISS:

Injury severity score

NISS:

New injury severity score

References

  1. Heron M. Deaths: leading causes for 2016. Natl Vital Stat Rep. 2018;67(6):1–77.

    PubMed  Google Scholar 

  2. Krug EG, Sharma GK, Lozano R. The global burden of injuries. Am J Public Health. 2000;90(4):523–6.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  3. Hashmi ZG, Schneider EB, Castillo R, et al. Benchmarking trauma centers on mortality alone does not reflect quality of care: implications for pay-for-performance. J Trauma Acute Care Surg. 2014;76(5):1184–91.

    Article  PubMed  Google Scholar 

  4. Cole E, Gillespie S, Vulliamy P, et al. Multiple organ dysfunction after trauma. Br J Surg. 2020;107(4):402–12.

    Article  CAS  PubMed  Google Scholar 

  5. Shepherd JM, Cole E, Brohi K. Contemporary patterns of multiple organ dysfunction in trauma. Shock. 2017;47(4):429.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Fröhlich M, Lefering R, Probst C, et al. Epidemiology and risk factors of multiple-organ failure after multiple trauma: an analysis of 31,154 patients from the TraumaRegister DGU. J Trauma Acute Care Surg. 2014;76(4):921.

    Article  PubMed  Google Scholar 

  7. Ting RS, Lewis DP, Yang KX, et al. Incidence of multiple organ failure in adult polytrauma patients: a systematic review and meta-analysis. J Trauma Acute Care Surg. 2023;94(5):725.

    Article  PubMed  PubMed Central  Google Scholar 

  8. Heffernan DS, Monaghan SF, Thakkar RK, Machan JT, Cioffi WG, Ayala A. Failure to normalize lymphopenia following trauma is associated with increased mortality, independent of the leukocytosis pattern. Crit Care. 2012;16(1):R12.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Huber-Lang M, Lambris JD, Ward PA. Innate immune responses to trauma. Nat Immunol. 2018;19(4):327–41.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  10. Islam MN, Bradley BA, Ceredig R. Sterile post-traumatic immunosuppression. Clin Transl Immunol. 2016;5(4): e77.

    Article  Google Scholar 

  11. Ward PA. Immunosuppression after trauma. Crit Care Med. 2005;33(6):1453–4.

    Article  PubMed  Google Scholar 

  12. Paladino L, Subramanian RA, Bonilla E, Sinert RH. Leukocytosis as prognostic indicator of major injury. West J Emerg Med. 2010;11(5):450–5.

    PubMed  PubMed Central  Google Scholar 

  13. Xiao W, Mindrinos MN, Seok J, et al. A genomic storm in critically injured humans. J Exp Med. 2011;208(13):2581–90.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  14. Cabrera CP, Manson J, Shepherd JM, et al. Signatures of inflammation and impending multiple organ dysfunction in the hyperacute phase of trauma: a prospective cohort study. PLoS Med. 2017;14(7): e1002352.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Manson J, Cole E, De’Ath HD, et al. Early changes within the lymphocyte population are associated with the development of multiple organ dysfunction syndrome in trauma patients. Crit Care. 2016;20:176.

    Article  PubMed  PubMed Central  Google Scholar 

  16. Cuenca AG, Delano MJ, Kelly-Scumpia KM, et al. A paradoxical role for myeloid-derived suppressor cells in sepsis and trauma. Mol Med. 2011;17(3–4):281–92.

    Article  CAS  PubMed  Google Scholar 

  17. Glance LG, Stone PW, Mukamel DB, Dick AW. Increases in mortality, length of stay, and cost associated with hospital-acquired infections in trauma patients. Arch Surg. 2011;146(7):794–801.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Cole E, Davenport R, Willett K, Brohi K. The burden of infection in severely injured trauma patients and the relationship with admission shock severity. J Trauma Acute Care Surg. 2014;76(3):730–5.

    Article  CAS  PubMed  Google Scholar 

  19. Dewar D, Moore FA, Moore EE, Balogh Z. Postinjury multiple organ failure. Injury. 2009;40(9):912–8.

    Article  PubMed  Google Scholar 

  20. Evans JA, van Wessem KJP, McDougall D, Lee KA, Lyons T, Balogh ZJ. Epidemiology of traumatic deaths: comprehensive population-based assessment. World J Surg. 2010;34(1):2014.

    Article  Google Scholar 

  21. Moore FA, Sauaia A, Moore EE, Haenel JB, Burch JM, Lezotte DC. Postinjury multiple organ failure: a bimodal phenomenon. J Trauma. 1996;40(4):501–10.

    Article  CAS  PubMed  Google Scholar 

  22. Marshall JC, Cook DJ, Christou NV, Bernard GR, Sprung CL, Sibbald WJ. Multiple organ dysfunction score: a reliable descriptor of a complex clinical outcome. Crit Care Med. 1995;23(10):1638–52.

    Article  CAS  PubMed  Google Scholar 

  23. Vincent JL, Moreno R, Takala J, et al. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med. 1996;22(7):707–10.

    Article  CAS  PubMed  Google Scholar 

  24. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985;13(10):818–29.

    Article  CAS  PubMed  Google Scholar 

  25. Baker SP, O’Neill B, Haddon W, Long WB. The injury severity score: a method for describing patients with multiple injuries and evaluating emergency care. J Trauma. 1974;14(3):187–96.

    Article  CAS  PubMed  Google Scholar 

  26. Osler T, Glance LG, Hosmer DW. Simplified estimates of the probability of death after burn injuries: extending and updating the baux score. J Trauma. 2010;68(3):690–7.

    PubMed  Google Scholar 

  27. Suka M, Yoshida K, Takezawa J. Association between APACHE II score and nosocomial infections in intensive care unit patients: a multicenter cohort study. Environ Health Prev Med. 2004;9(6):262–5.

    Article  PubMed  PubMed Central  Google Scholar 

  28. Cerra FB, Negro F, Abrams J. APACHE II score does not predict multiple organ failure or mortality in postoperative surgical patients. Arch Surg. 1990;125(4):519–22.

    Article  CAS  PubMed  Google Scholar 

  29. Baez AA, Giraldez EM, Sztajnkrycer MD. Predictive effect of the injury severity score on the development of multiple organ failure among young adults and elderly patients. Ann Emerg Med. 2004;44(4):S127.

    Article  Google Scholar 

  30. Balogh Z, Offner PJ, Moore EE, Biffl WL. NISS predicts postinjury multiple organ failure better than the ISS. J Trauma Acute Care Surg. 2000;48(4):624.

    Article  CAS  Google Scholar 

  31. Hurr H, Hawley HB, Czachor JS, Markert RJ, McCarthy MC. APACHE II and ISS scores as predictors of nosocomial infections in trauma patients. Am J Infect Control. 1999;27(2):79–83.

    Article  CAS  PubMed  Google Scholar 

  32. Jamulitrat S, Narong MN, Thongpiyapoom S. Trauma severity scoring systems as predictors of nosocomial infection. Infect Control Hosp Epidemiol. 2002;23(5):268–73.

    Article  PubMed  Google Scholar 

  33. Yang Y-W, Wu C-H, Tsai H-T, et al. Dynamics of immune responses are inconsistent when trauma patients are grouped by injury severity score and clinical outcomes. Sci Rep. 2023;13(1):1391.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  34. Typpo KV, Wong HR, Finley SD, Daniels RC, Seely AJE, Lacroix J. Monitoring severity of multiple organ dysfunction syndrome: new technologies. Pediatr Crit Care Med. 2017;18(3 Suppl 1):S24-31.

    Article  PubMed  PubMed Central  Google Scholar 

  35. Gavan SP, Thompson AJ, Payne K. The economic case for precision medicine. Expert Rev Precision Med Drug Dev. 2018;3(1):1–9.

    Article  Google Scholar 

  36. Ginsburg GS, Phillips KA. Precision medicine: from science to value. Health Aff. 2018;37(5):694–701.

    Article  Google Scholar 

  37. Johnson KB, Wei W-Q, Weeraratne D, et al. Precision medicine, AI, and the future of personalized health care. Clin Transl Sci. 2021;14(1):86–93.

    Article  PubMed  Google Scholar 

  38. Barichello T, Generoso JS, Singer M, Dal-Pizzol F. Biomarkers for sepsis: more than just fever and leukocytosis—a narrative review. Crit Care. 2022;26(1):14.

    Article  PubMed  PubMed Central  Google Scholar 

  39. Yan S, Tsurumi A, Que YA, et al. Prediction of multiple infections after severe burn trauma: a prospective cohort study. Ann Surg. 2015;261(4):781–92.

    Article  PubMed  Google Scholar 

  40. Tsurumi A, Flaherty PJ, Que YA, et al. Multi-biomarker prediction models for multiple infection episodes following blunt trauma. iScience. 2020;23(11):101659.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Tsurumi A, Flaherty PJ, Que YA, et al. A preventive tool for predicting bloodstream infections in children with burns. Shock. 2023;59(3):393–9.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  42. Sweeney TE, Azad TD, Donato M, et al. Unsupervised analysis of transcriptomics in bacterial sepsis across multiple datasets reveals three robust clusters. Crit Care Med. 2018;46(6):915–25.

    Article  PubMed  PubMed Central  Google Scholar 

  43. Sweeney TE, Wong HR. Risk stratification and prognosis in sepsis: what have we learned from microarrays? Clin Chest Med. 2016;37(2):209–18.

    Article  PubMed  PubMed Central  Google Scholar 

  44. Sweeney TE, Shidham A, Wong HR, Khatri P. A comprehensive time-course-based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set. Sci Transl Med. 2015;7(287):287ra71.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Desai KH, Tan CS, Leek JT, et al. Dissecting inflammatory complications in critically injured patients by within-patient gene expression changes: a longitudinal clinical genomics study. PLoS Med. 2011;8(9): e1001093.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  46. Ishikawa S, Teshima Y, Otsubo H, et al. Risk prediction of biomarkers for early multiple organ dysfunction in critically ill patients. BMC Emerg Med. 2021;21(1):132.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  47. Bravo-Merodio L, Acharjee A, Hazeldine J, et al. Machine learning for the detection of early immunological markers as predictors of multi-organ dysfunction. Sci Data. 2019;6(1):328.

    Article  PubMed  PubMed Central  Google Scholar 

  48. Chen T, Guestrin C. XGBoost: a scalable tree boosting system. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining-KDD ’16 2016;785–94.

  49. Lundberg SM, Lee S-I. A unified approach to interpreting model predictions. Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS’17) 2017;Red Hook, NY, USA: Curran Associates Inc.:4768–77.

  50. Lundberg SM, Nair B, Vavilala MS, et al. Explainable machine-learning predictions for the prevention of hypoxaemia during surgery. Nat Biomed Eng. 2018;2(10):749–60.

    Article  PubMed  PubMed Central  Google Scholar 

  51. Tompkins RG. Genomics of injury: the Glue Grant experience. J Trauma Acute Care Surg. 2015;78(4):671–86.

    Article  PubMed  PubMed Central  Google Scholar 

  52. Wu J, Irizarry RA. gcrma: Background Adjustment Using Sequence Information. R package version 2500 2017;

  53. Kauffmann A, Gentleman R, Huber W. arrayQualityMetrics–a bioconductor package for quality assessment of microarray data. Bioinformatics. 2009;25(3):415–6.

    Article  CAS  PubMed  Google Scholar 

  54. Kassambara A, Mundt F. factoextra: Extract and visualize the results of multivariate data analyses. R package version 107 2020;

  55. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7): e47.

    Article  PubMed  PubMed Central  Google Scholar 

  56. Kuhn M. Building predictive models in R using the caret package. J Stat Softw. 2008;28(5).

  57. Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1–22.

    Article  PubMed  PubMed Central  Google Scholar 

  58. Wilson S. ParBayesianOptimization: Parallel Bayesian Optimization of Hyperparameters. R package version 126 [Internet] 2022; Available from: https://CRAN.R-project.org/package=ParBayesianOptimization.

  59. Liu Y, Just A. SHAPforxgboost: SHAP Plots for “XGBoost.” R package version 013 2023; https://CRAN.R-project.org/package=SHAPforxgboost.

  60. Yan L. ggvenn: Draw Venn Diagram by “ggplot2.” 2023; R package version 0.1.10. https://CRAN.R-project.org/package=ggvenn.

  61. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77.

    Article  PubMed  PubMed Central  Google Scholar 

  62. Stevenson M, Sergeant E. epiR: tools for the analysis of epidemiological data. R package version 2026 2023; https://CRAN.R-project.org/package=epiR.

  63. Ulgen E, Ozisik O, Sezerman OU. pathfindR: an R package for comprehensive identification of enriched pathways in omics data through active subnetworks. Front Genet. 2019;10:858.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  64. Wickham H. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York; 2016. Available from: https://ggplot2.tidyverse.org

  65. Montojo J, Zuberi K, Rodriguez H, Bader GD, Morris Q. GeneMANIA: fast gene network construction and function prediction for Cytoscape. F1000Res. 2014;3:153.

    Article  PubMed  PubMed Central  Google Scholar 

  66. Schuler A, Wulf DA, Lu Y, et al. The impact of acute organ dysfunction on long-term survival among sepsis survivors. Crit Care Med. 2018;46(6):843–9.

    Article  PubMed  PubMed Central  Google Scholar 

  67. Sperry JL, Nathens AB, Frankel HL, et al. Characterization of the gender dimorphism after injury and hemorrhagic shock: are hormonal differences responsible? Crit Care Med. 2008;36(6):1838–45.

    Article  PubMed  Google Scholar 

  68. Sperry JL, Friese RS, Frankel HL, et al. Male gender is associated with excessive IL-6 expression following severe injury. J Trauma Acute Care Surg. 2008;64(3):572.

    Article  CAS  Google Scholar 

  69. Trentzsch H, Lefering R, Nienaber U, Kraft R, Faist E, Piltz S. The role of biological sex in severely traumatized patients on outcomes: a matched-pair analysis. Ann Surg. 2015;261(4):774–80.

    Article  PubMed  Google Scholar 

  70. Trentzsch H, Nienaber U, Behnke M, Lefering R, Piltz S. Female sex protects from organ failure and sepsis after major trauma haemorrhage. Injury. 2014;45(Suppl 3):S20-28.

    Article  PubMed  Google Scholar 

  71. Zhang Y, Zhu Y, Gao G, Zhou Z. Knockdown XIST alleviates LPS-induced WI-38 cell apoptosis and inflammation injury via targeting miR-370-3p/TLR4 in acute pneumonia. Cell Biochem Funct. 2019;37(5):348–58.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  72. Li C, Liu J-H, Su J, et al. LncRNA XIST knockdown alleviates LPS-induced acute lung injury by inactivation of XIST/miR-132-3p/MAPK14 pathway. Mol Cell Biochem. 2021;476(12):4217–29.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  73. Tang B, Li W, Ji T, et al. Downregulation of XIST ameliorates acute kidney injury by sponging miR-142-5p and targeting PDCD4. J Cell Physiol. 2020;235(11):8852–63.

    Article  CAS  PubMed  Google Scholar 

  74. Li Z, Zhang Y, Ding N, et al. Inhibition of lncRNA XIST improves myocardial I/R injury by targeting miR-133a through inhibition of autophagy and regulation of SOCS2. Mol Therapy Nucleic Acids. 2019;18:764–73.

    Article  CAS  Google Scholar 

  75. Shen C, Li J. LncRNA XIST silencing protects against sepsis-induced acute liver injury via inhibition of BRD4 expression. Inflammation. 2021;44(1):194–205.

    Article  CAS  PubMed  Google Scholar 

  76. Pi L, Fang B, Meng X, Qian L. LncRNA XIST accelerates burn wound healing by promoting M2 macrophage polarization through targeting IL-33 via miR-19b. Cell Death Discov. 2022;8(1):1–10.

    Article  Google Scholar 

  77. Fan S, Hu Y. Integrative analyses of biomarkers and pathways for heart failure. BMC Med Genomics. 2022;15(1):72.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  78. Fry D. Infection: cause or result of organ failure? In: Baue AE, Faist E, Fry DE, editors. Multiple organ failure: pathophysiology, prevention, and therapy. New York: Springer; 2000. p. 598–604. https://doi.org/10.1007/978-1-4612-1222-5_61.

    Chapter  Google Scholar 

  79. Minei JP, Cuschieri J, Sperry J, et al. The changing pattern and implications of multiple organ failure (MOF) after blunt injury with hemorrhagic shock. Crit Care Med. 2012;40(4):1129–35.

    Article  PubMed  PubMed Central  Google Scholar 

  80. Vourc’h M, Roquilly A, Asehnoune K. Trauma-induced damage-associated molecular patterns-mediated remote organ injury and immunosuppression in the acutely Ill patient. Front Immunol. 2018. https://doi.org/10.3389/fimmu.2018.01330/full.

    Article  PubMed  PubMed Central  Google Scholar 

  81. Relja B, Land WG. Damage-associated molecular patterns in trauma. Eur J Trauma Emerg Surg. 2020;46(4):751–75.

    Article  PubMed  Google Scholar 

  82. Timmermans K, Kox M, Vaneker M, et al. Plasma levels of danger-associated molecular patterns are associated with immune suppression in trauma patients. Intensive Care Med. 2016;42(4):551–61.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  83. Gebril HM, Rose RM, Gesese R, et al. Adenosine kinase inhibition promotes proliferation of neural stem cells after traumatic brain injury. Brain Commun. 2020;2(1):fcaa017.

    Article  PubMed  PubMed Central  Google Scholar 

  84. Pignataro G, Simon RP, Boison D. Transgenic overexpression of adenosine kinase aggravates cell death in ischemia. J Cereb Blood Flow Metab. 2007;27(1):1–5.

    Article  CAS  PubMed  Google Scholar 

  85. Shen H-Y, Lusardi TA, Williams-Karnesky RL, Lan J-Q, Poulsen DJ, Boison D. Adenosine kinase determines the degree of brain injury after ischemic stroke in mice. J Cereb Blood Flow Metab. 2011;31(7):1648–59.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  86. Xu Y, Wang Y, Yan S, et al. Regulation of endothelial intracellular adenosine via adenosine kinase epigenetically modulates vascular inflammation. Nat Commun. 2017;8(1):943.

    Article  PubMed  PubMed Central  Google Scholar 

  87. Ullah K, Ai L, Humayun Z, Wu R. Targeting endothelial HIF2α/ARNT expression for ischemic heart disease therapy. Biology. 2023;12(7):995.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  88. Bogdanovski DA, DiFazio LT, Bogdanovski AK, et al. Hypoxia-inducible-factor-1 in trauma and critical care. J Crit Care. 2017;42:207–12.

    Article  CAS  PubMed  Google Scholar 

  89. Enescu DM, Parasca SV, Badoiu SC, et al. Hypoxia-inducible factors and burn-associated acute kidney injury—a new paradigm? Int J Mol Sci. 2022;23(5):2470.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  90. Elzière L, Sar C, Ventéo S, et al. CaMKK-CaMK1a, a new post-traumatic signalling pathway induced in mouse somatosensory neurons. PLoS ONE. 2014;9(5): e97736.

    Article  PubMed  PubMed Central  Google Scholar 

  91. Zhu X, Li H, You W, et al. Role of Rph3A in brain injury induced by experimental cerebral ischemia-reperfusion model in rats. CNS Neurosci Ther. 2022;28(7):1124–38.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  92. Feng Z, Fan Y, Shi X, et al. Dysregulation of iron transport-related biomarkers in blood leukocytes is associated with poor prognosis of early trauma. Heliyon. 2024;10:e27000.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  93. Speir RW, Stallings JD, Andrews JM, Gelnett MS, Brand TC, Salgar SK. Effects of valproic acid and dexamethasone administration on early bio-markers and gene expression profile in acute kidney ischemia-reperfusion injury in the rat. PLoS ONE. 2015;10(5): e0126622.

    Article  PubMed  PubMed Central  Google Scholar 

  94. Wang J-H, Nichogiannopoulou A, Wu L, et al. Selective defects in the development of the fetal and adult lymphoid system in mice with an ikaros null mutation. Immunity. 1996;5(6):537–49.

    Article  CAS  PubMed  Google Scholar 

  95. Nichogiannopoulou A, Trevisan M, Neben S, Friedrich C, Georgopoulos K. Defects in hemopoietic stem cell activity in ikaros mutant mice. J Exp Med. 1999;190(9):1201–14.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  96. Hoshino A, Okada S, Yoshida K, et al. Abnormal hematopoiesis and autoimmunity in human subjects with germline IKZF1 mutations. J Allergy Clin Immunol. 2017;140(1):223–31.

    Article  CAS  PubMed  Google Scholar 

  97. Gandhirajan A, Roychowdhury S, Kibler C, et al. SIRT2-PFKP interaction dysregulates phagocytosis in macrophages with acute ethanol-exposure. Front Immunol. 2023. https://doi.org/10.3389/fimmu.2022.1079962.

    Article  PubMed  PubMed Central  Google Scholar 

  98. Peng J, Li P, Li Y, et al. PFKP is a prospective prognostic, diagnostic, immunological and drug sensitivity predictor across pan-cancer. Sci Rep. 2023;13(1):17399.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  99. Wang S, Park SH, Lim JS, Park Y-Y, Du L, Lee J-H. Phosphofructokinase 1 platelet isoform induces PD-L1 expression to promote glioblastoma immune evasion. Genes Genomics. 2022;44(12):1509–17.

    Article  CAS  PubMed  Google Scholar 

  100. Lee J-H, Liu R, Li J, et al. Stabilization of phosphofructokinase 1 platelet isoform by AKT promotes tumorigenesis. Nat Commun. 2017;8.

  101. Chen J, Zou L, Lu G, et al. PFKP alleviates glucose starvation-induced metabolic stress in lung cancer cells via AMPK-ACC2 dependent fatty acid oxidation. Cell Discov. 2022;8(1):1–16.

    Article  Google Scholar 

  102. Scherlinger M, Pan W, Hisada R, et al. Phosphofructokinase P fine-tunes T regulatory cell metabolism, function, and stability in systemic autoimmunity. Sci Adv. 2022;8(48):eadc9657.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  103. Zhuang G, Meng C, Guo X, et al. A novel regulator of macrophage activation. Circulation. 2012;125(23):2892–903.

    Article  CAS  PubMed  Google Scholar 

  104. Yang X, Ye C, Zheng H, Dai C, Zhu Y. Systemic analyses of the expression of TPI1 and its associations with tumor microenvironment in lung adenocarcinoma and squamous cell carcinoma. Dis Markers. 2022;2022:6258268.

    PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

We acknowledge the Glue Grant (“Inflammation and Host Response to Injury”) Consortium for the dataset we analyzed.

Funding

NIH/NIAID R03 AI151499 to AT and NIH/NIAID R56 AI155505 to LGR.

Author information

Authors and Affiliations

  1. Department of Surgery, Massachusetts General Hospital and Harvard Medical School, 50 Blossom St., Their 340, Boston, MA, 02114, USA

    Ivan Duran, Ankita Banerjee, Colleen M. Ryan, Laurence G. Rahme & Amy Tsurumi

  2. Department of Mathematics and Statistics, University of Massachusetts at Amherst, Amherst, MA, 01003, USA

    Patrick J. Flaherty

  3. Department of Intensive Care Medicine, Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland

    Yok-Ai Que

  4. Shriners Hospitals for Children-Boston®, 51 Blossom St., Boston, MA, 02114, USA

    Colleen M. Ryan, Laurence G. Rahme & Amy Tsurumi

  5. Department of Microbiology and Immunology, Harvard Medical School, 77 Ave. Louis Pasteur, Boston, MA, 02115, USA

    Laurence G. Rahme

Authors
  1. Ivan Duran
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Ankita Banerjee
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Patrick J. Flaherty
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Yok-Ai Que
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Colleen M. Ryan
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Laurence G. Rahme
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Amy Tsurumi
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

Study concept and design—IV, AB, LGR, PJF, YQ, CMR, AT; data analysis—IV, AB, AT; data interpretation—IV, AB, LGR, PJF, YQ, CMR, AT; drafting of the manuscript—IV, AT; critical revision—IV, AB, LGR, PJF, YQ, CMR, AT; funding – AT, LGR.

Corresponding author

Correspondence to Amy Tsurumi.

Ethics declarations

Ethics approval and consent to participate

Consent to participate was not required, as this study entailed analyzing data from the Glue Grant (“Inflammation and Host Response to Injury”) cohort and did not enroll new patients. Permission for the secondary analysis of de-identified data was obtained from the Massachusetts General Hospital Institutional Review Board (MGH IRB protocol 2002P001743).

Competing interests

ID, AB, PJF, YQ, CMR, AT report no conflict of interest. LGR has a financial interest in Spero Therapeutics.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Duran, I., Banerjee, A., Flaherty, P.J. et al. Development of a biomarker prediction model for post-trauma multiple organ failure/dysfunction syndrome based on the blood transcriptome. Ann. Intensive Care 14, 134 (2024). https://doi.org/10.1186/s13613-024-01364-5

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s13613-024-01364-5

Keywords