Journal Logo

Papers of the 141st ASA Annual Meeting

High Dimensional Multiomics Reveals Unique Characteristics of Early Plasma Administration in Polytrauma Patients With TBI

Wu, Junru MD*,†,‡; Moheimani, Hamed MD, MPH†,‡; Li, Shimena MD, MSc†,‡; Kar, Upendra K. PhD, MBA†,‡; Bonaroti, Jillian MD†,‡; Miller, Richard S. MD§; Daley, Brian J. MD; Harbrecht, Brian G. MD; Claridge, Jeffrey A. MD#; Gruen, Danielle S. PhD†,‡; Phelan, Herbert A. MD**; Guyette, Francis X. MD, MPH††; Neal, Matthew D. MD†,‡; Das, Jishnu PhD‡‡; Sperry, Jason L. MD, MPH†,‡; Billiar, Timothy R. MD†,‡

Author Information
doi: 10.1097/SLA.0000000000005610

Abstract

The evolution of resuscitation strategies over the past 3 decades has reduced the early mortality for critically injured patients with hemorrhagic shock.1,2 The inclusion of blood plasma in the resuscitation of patients at risk for hemorrhagic shock either as part of component therapy or through administration of whole blood is now standard practice.3,4 The Prehospital Air Medical Plasma (PAMPer) Trial, a randomized study designed to assess the benefit of early plasma resuscitation, found that prehospital administration of thawed plasma (TP) during air medical transport significantly reduced 3 and 30 days mortality only in patients with traumatic brain injury (TBI).5,6

The reasons for the selective survival benefit of early TP in TBI patients are unknown; however, post hoc analysis of patient subsets have suggested that the effects of early plasma are seen only in TBI patients with a pronounced systemic response to injury as seen in TBI patients with significant nonbrain injuries.6,7 Current theories include the possibility that TP reduces endothelial injury at the blood–brain barrier and/or prevents further intracranial bleeding by improving coagulation.8–10 We recently showed that TP administration preserved circulating lipid levels11 and was associated with unique circulating protein patterns compared with patients that did not receive prehospital TP.7 This suggests that exogenous plasma is metabolized differently in patients with TBI as part of their injury complex. However, the proteomic database had very few nonsurvivors, which limited its usefulness for assessing causal factors.

Here, we hypothesized that causal mediation analysis of a multi-omic dataset derived from blood analysis of PAMPer patients would provide mechanistic insights into the selective benefit of early TP in severely injured TBI patients. Using a new proteomic database, levels of 7211 proteins (SomaLogic Inc.) were assessed in 149 patients that included survivors and nonsurvivors (n=62) in plasma from blood samples collected upon admission, the first blood draw after prehospital TP administration. We added to this novel proteomic dataset, mediation analyses of our previously reported lipidomic and metabolomic databases on these same patients.7,11 Linear and multivariate latent-factor regression analyses coupled with mediation analysis yielded multiple proteins and lipids causally related to prehospital TP in reducing early mortality in TBI patients. These findings should guide mechanistic research into the optimal management of trauma patients at risk for hemorrhagic shock and TBI.

METHODS

Patient Enrollment and Data Collection

We analyzed the plasma samples collected upon admission in a subset of patients from PAMPer Trial.5 Samples from 149 patients representative of the overall cohort were selected from 194 patients had previously metabolomic and lipidomic datasets created.7,11 The criteria for patient enrollment and characteristics of the patients have been reported previously.5 Briefly, trauma patients who had experienced at least 1 prehospital episode of hypotension (systolic blood pressure <90 mm Hg and heart rate >108/min, odds ratio systolic blood pressure <70 mm Hg) were randomized to receive standard care fluid resuscitation (crystalloid with or without packed blood cells) or 2 of during air-medical transport followed by standard care fluid resuscitation. TBI was diagnosed at each enrollment site from the initial head computed tomography scan. Endotypes were previously generated from the clustering analysis of circulating endotheliopathy biomarkers, cytokines, and metabolite levels on admission.7 Twenty-nine low or no-injury trauma patients transported by air medical transport in STAAMP trial,12 sampled upon arrival, were included as minimally injured controls.

This study was approved by the IRB of University of Pittsburgh. Detailed information and study protocol for PAMPer Trial are available on https://clinicaltrials.gov/ct2/show/NCT01818427. The PAMPer Trial received approval for emergency exception from informed consent (EFIC) from the Human Research Protection Office of the US Army Medical Research and Material Command.

Multiplexed Proteomics Assay

We adopted a multiplexed, aptamer-based platform (SOMAscan assay) capable of measuring levels of 7596 human proteins in plasma. Briefly, 150 μL of EDTA-treated plasma was thawed, aliquoted and sent on dry ice to SomaLogic Inc. (Boulder, Colorado). The assay is based on protein-capture reagents constructed by chemically modified single stranded DNA. The SOMAmer mimic amino acid side chains and can specifically bind to the targeted proteins with a similar structure compared with antibody-antigen interaction.13 Intensity of each interaction were recorded by DNA microarray and was used to represent the concentration of proteins. Raw data were evaluated by quality analysis of hybridization and calibration metrics. A total of 7211 (94.9%) passed the accepted range for the hybridization and calibration scale factors (0.4–2.5 and 0.8–1.2, respectively).

Causal Mediation Analysis

Mediation is the process through which a certain exposure or intervention leads to an outcome. Causal mediation analysis is a statistical method that calculates how much of the total effect of an exposure on the outcome is caused by a certain mediator. Here, causal mediation analysis (implemented in R package mediation)14 was conducted to screen for possible features causally related to the effects of prehospital plasma in reducing 3-day or 30-day mortality in trauma patients. A linear regression model was fitted to filter features (202 proteins, 568 lipids, and 149 metabolites) associated with treatment arms (P<0.10). Then, filtered features were tested for the mediation effect in the framework of simple mediation. Nonparametric bootstrapping (1000×) was used to estimate the 95% confidence intervals and P values. Aiming for an optimal screening test on the limited sample size, we allowed a higher type II error with lower type I error. Candidate features were called based on the following criteria: (1) total effect were associated with a reduction in mortality (odds ratio <1). (2) Mediated effect were significant (P<0.05 or proportion >30%).

Essential Regression (ER) Analysis

ER is a novel method that is able to integrate multiomic datasets and effectively identify causal latent factors significantly linked to an outcome.15 Latent or hidden-factors are variables that are composites of measured variables. They can revealed through mathematical algorithms that model measured and observable data features. Such factors may correspond to a physical yet unmeasured characteristic, or an abstract underlying category/concept. In multi-omic analysis, latent factors may correspond to biological pathways or mechanisms. ER reveals significant latent factors through a regression model that is data-distribution free and does not require certain assumptions that can limit the use of regular regression models. Moreover, it has the advantage of being data dependent and having the ability to provide inferential insights without the need of incorporating any prior knowledge.

In this work, ER was applied to a multiomic dataset of 8218 unique observable features in endotype 2 (E2)-TBI patients for the outcome of 3-day mortality. In addition to the commonly used least squares method, we ran the analysis using the Dantzig estimation. This method utilizes a regularization technique (L1) that helped us more robustly identify a sparse subset of latent factors (~pathways/mechanisms) that possess the most significant causal relationship with the outcome (3-day mortality). A 5-fold cross-validation with 50× repeats were applied to fine-tune the model. Hyperparameter delta was chosen using a separate cross-validation. Through causal mediation analysis, the identified latent factors were further assessed as effect mediators of TP.

Pathway Enrichment Analysis

Pathway enrichment analysis is a statistical technique that provides mechanistic insights by identifying biological pathways that are enriched (over-represented) in a gene list of interest, more than what we would expect by chance alone. Here, this method was conducted using R package Clusterprofiler (v 3.11).16 The names of 64 proteins identified from causal mediation analysis (proportion >30%) were transformed into Entrez ID and the Reactome Database was used to search for related pathways. The P value of enriched pathways was adjusted by Benjamini-Hochberg method. Pathways with enrichment of at least of 2 proteins with adjusted P value <0.05 were considered to be significant.

Statistical Analysis

Kaplan-Meier curves (K-M curves) and log rank P values were computed for survival analysis. Pearson χ2 test or Kruskal-Wallis tests were used for categorical variables or continuous variables in the contingency table of clinical layer data. For near-normally distributed data, multiple group comparisons were conducted by analysis of variance test with post hoc analysis by Tukey HSD test. For skewed distributed data, multiple group comparisons were conducted by Kruskal-Wallis test with post hoc analysis by Dunn test. Differential proteins between the 2 treatment arms were generated by Wilcoxon rank sum test and P value adjustment was performed using the Benjamini-Hochberg method.

RESULTS

Overview of the Cohort and Analysis Workflow

A subset of PAMPer patients representative of the overall cohort (n=149) were separated based on previously described endotypes that correspond to systemic response patterns (E1: low response, E2: high response),7 the presence or absence of TBI (NBI: no brain injury) and treatment arm (STD: standard-of-care vs prehospital TP). Demographic and clinical patient information is shown in Table 1 and the overall analysis workflow is depicted in Figure 1. Briefly, we adopted a 3-step strategy by integrating 2 separate regression methods and causal mediation analysis14,17 (see also the Methods section). In an analysis of each individual omics database, linear regression was used to filter the features associated with TP administration, which were then subjected to mediation analysis. In a complimentary analysis, ER15 was used to identify the latent factors associated with outcomes from a multiomic dataset that combined the proteomic, lipidomic and metabolomic databases. The framework for the causal mediation analysis consisted of TP, molecules/ factors in the patients’ circulation and 30-day/3-day mortality as predictor, mediators and outcome, respectively.

TABLE 1 - Demographic Information for Trauma Patients, Grouped by TBI, Endotypes and Treatment Arms
E1-NBI STD E1-NBI TP E1-TBI STD E1-TBI TP E2-NBI STD E2-NBI TP E2-TBI STD E2-TBI TP P P P P
(N=21) (N=16) (N=25) (N=18) (N=14) (N=11) (N=24) (N=20) P E1-NBI E1-TBI E2-NBI E2-TBI
Age, y 36 (±43) 53 (±26) 48 (±41) 51 (±45) 55 (±29) 45 (±25) 41 (±33) 39 (±28) 0.780
Sex 14 (66.7%) 13 (81.3%) 20 (80.0%) 13 (72.2%) 11 (78.6%) 6 (54.5%) 21 (87.5%) 13 (65.0%) 0.427
ISS 16 (±12) 18 (±6.5) 29 (±10) 23 (±17) 27 (±13) 34 (±26) 38 (±14) 34 (±19) <0.001 0.572 0.633 0.307 0.381
GCS 14 (±3.0) 15 (±8.3) 3.0 (±3.0) 3.0 (±11) 12 (±10) 12 (±12) 3.0 (±0) 3.0 (±2.5) <0.001 0.492 0.413 0.901 0.399
HR 110 (±16) 120 (±16) 120 (±16) 110 (±39) 120 (±26) 120 (±62) 120 (±28) 130 (±19) 0.371
Shock Index 1.4 (±0.40) 1.5 (±0.09) 1.7 (±0.53) 1.4 (±0.19) 1.8 (±0.74) 1.5 (±0.70) 1.6 (±0.24) 1.6 (±0.42) 0.297
PH time 40 (±23) 42 (±14) 39 (±24) 45 (±32) 40 (±14) 40 (±13) 43 (±16) 46 (±11) 0.819
PH blood 9 (42.9%) 3 (18.8%) 8 (32.0%) 2 (11.1%) 8 (57.1%) 2 (18.2%) 12 (50.0%) 6 (30.0%) 0.048 0.231 0.217 0.118 0.300
PH crystalloid 1300 (±1900) 780 (±1200) 1000 (±1700) 250 (±1200) 150 (±1600) 250 (±1300) 950 (±1500) 500 (±1100) 0.182
PH PRBC 0 (±2.0) 0 (±0) 0 (±1.0) 0 (±0) 1.5 (±2.8) 0 (±0) 0.50 (±2.0) 0 (±1.0) 0.020 0.183 0.249 0.016 0.113
Transfusion 24 h 3.0 (±10) 7.0 (±7.3) 4.0 (±10) 0 (±2.0) 26 (±27) 16 (±14) 12 (±12) 9.0 (±13) <0.001 0.455 0.007 0.366 0.473
PRBC 24 h 3.0 (±6.0) 5.0 (±6.0) 3.0 (±5.0) 0 (±2.0) 19 (±21) 10 (±8.5) 7.0 (±5.5) 4.5 (±6.8) <0.001 0.263 0.019 0.395 0.210
Plasma 24 h 0 (±3.0) 0.50 (±2.5) 1.0 (±4.0) 0 (±0) 4.0 (±8.5) 4.0 (±5.5) 2.5 (±6.3) 3.5 (±4.0) 0.006 0.633 0.050 0.877 0.501
Platelets 24 h 0 (±1.0) 0 (±0) 0 (±2.0) 0 (±0) 1.0 (±2.8) 1.0 (±2.0) 1.0 (±1.0) 1.0 (±2.0) 0.002 0.438 0.024 0.630 0.586
Crystalloid 24 h 4900 (±4000) 5600 (±3200) 5400 (±3700) 4100 (±1900) 6600 (±2900) 5300 (±1300) 4900 (±3100) 4200 (±4900) 0.239
INR 1.4 (±0.50) 1.2 (±0.11) 1.2 (±0.31) 1.2 (±0.12) 1.8 (±0.78) 1.2 (±0.29) 1.5 (±0.52) 1.3 (±0.27) <0.001 0.014 0.460 0.023 0.006
Coagulopathy 9 (42.9%) 11 (68.8%) 10 (40.0%) 7 (38.9%) 11 (78.6%) 10 (90.9%) 19 (79.2%) 12 (60.0%) 0.004 0.218 1.000 0.775 0.291
Mortality 3 d 2 (9.52%) 0 (0%) 4 (16.0%) 4 (22.2%) 7 (50.0%) 5 (45.5%) 12 (50.0%) 4 (20.0%) 0.001 0.592 0.904 1.000 0.081
Mortality 30 d 5 (23.8%) 1 (6.25%) 10 (40.0%) 7 (38.9%) 7 (50.0%) 6 (54.5%) 19 (79.2%) 7 (35.0%) <0.001 0.324 1.000 1.000 0.008
Pearson χ2 test was used for calculating P value of categorical variables. Kruskal-Wallis test was used for calculating P value of continuous variables. Post hoc analysis by Dunn test of 4 subgroup Comparisons (STD vs TP in 4 subgroups) were applied for variable of P<0.05 from Kruskal-Wallis test.
GCS indicates Glasgow coma score; HR, heart rate; ISS, injury severity score; PH, Prehospital; PRBC, packed red blood cells.

F1
FIGURE 1:
Scheme of the analysis workflow. The figure depicts the 3 steps of our analysis. We first performed feature selection through linear and multivariate (essential regression) regression strategies performed in parallel on a multiomic dataset (step 1). This was followed by mediation analysis to identify the causal features that drive the impact of thawed plasma on survival in trauma patients from the PAMPer Trial (step 2). In this step, 1-way regression models determined the direct association between administration of thawed plasma with 3 day/30 day mortality (A) and the indirect effect of thawed plasma through molecules or latent factors on 3 day/30 day mortality (B and C). Last, in step 3, we performed formal statistical mediation analysis to determine whether molecules or latent factors mediated the association between thawed plasma and mortality.

Linear Regression Coupled to Mediation Analysis Identifies Circulating Features that Explain the Protective Effects of TP in E2-TBI Patients

Consistent with our previous analysis,7 patients classified as E2 (n=69) had higher circulating levels of systemic storm related biomarkers upon admission including endotheliopathy markers, proinflammatory mediators and intracellular constituents (markers of cell death) than E1 patients (n=80) (Figs. 2A, B). Also consistent with our published results,6,7 stratified K-M curves demonstrated an early survival separation for all TBI (P=0.039) and E2-TBI (P=0.0035) patients that received 2 units of TP prehospital with no separation observed in the overall NBI patients or the E1-NBI, E2-NBI, E1-TBI subcohorts (Fig. S1A–E, Supplemental Digital Content 1, https://links.lww.com/SLA/E51).

F2
FIGURE 2:
Top circulating proteins (identified by linear regression + mediation analysis) that mediate the protective effect for reducing early mortality only in E2-TBI subgroup. A, Scheme of endotyping trauma patients in PAMPer Trial by the level of the systemic response to injury. B, Levels of representative features that associate with the two trauma patient Endotypes (patient numbers: E1=80, E2=69). C, Forest plot showed the total effect, average causal mediation effects and proportion of mediation effects with 95% confidence interval for 4 representative proteins in trauma patients grouped by presence or absence of brain injury. The outcome was set as 3 day mortality. CI indicates confidence interval; OR, odds ratio; Prop, proportion.

Linear regression followed by mediation analysis for the overall cohorts as well as the subgroups was performed on the proteomic dataset. In this analysis, proportion of mediated effect for each variable is calculated as the averaged mediated effect (the effect of TP in improving outcome via circulatory factors) divided by the total treatment effect (the total effect of TP in improving outcome). This identified 64 proteins with a proportion of mediated effect over 30%, the threshold set for possible significance, for the outcome of 3-day mortality (Table S1, Supplemental Digital Content 2, https://links.lww.com/SLA/E52). Of note, none of the 64 proteins were significant for a mediated effect of TP for 30 days mortality in E2-TBI subgroup (proportion <30% and P<0.05) suggesting that TP exerted most of its survival benefit early after injury. To prioritize the candidate proteins, we applied criterion for classifying the 64 proteins into 3 groups or tiers (Table S1, Supplemental Digital Content 2, https://links.lww.com/SLA/E52). Tier 1 consisted of the 9 proteins [Serine protease inhibitor A5 (SERPINA5), coiled coil-helix coiled coil-helix domain 7 (CHCHD7), Apoprotein E3 (APOE3), plasminogen (PLG), insulin-like growth factor binding protein 4 (IGFBP4), Vitamin K-dependent protein C (PROC), neuropeptide S (NPS), carboxypeptidase B2 (CPB2), and coagulation factor XI (F11)] with the proportion of mediated effect in the total population, TBI group and E2-TBI subgroup of over 50%. Tier2 consists of the 7 proteins [LPS binding protein (LBP), soluble phospholipase A2-Group XIIA (sPLA2-XII), myoglobin (MB), acetyl-Coenzyme A acyltransferase-1 (ACAA1), cathepsin F (CTSF), complement factor 3 (C3), and ubiquitin conjugating enzyme E2 S (UBE2S)] with the proportion of mediated effect of over 50% only in the E2-TBI subgroup. Tier3 consists of 48 other proteins with a proportion of mediated effect in E2-TBI subgroup between 30% and 50%. Forrest plots across all of the patient cohorts for the 4 proteins with the highest proportion of mediation is shown in Figure 2C and includes SERPINA5, PLG, PROC, and APOE3. The top 10 biologic pathways represented by the 64 proteins is shown in Figure S2, Supplemental Digital Content 1 (https://links.lww.com/SLA/E51). The most significant processes include fibrin clot formation, Insulin-like growth factor transport and uptake, mitochondrial protein import, and lipoprotein assembly, remodeling, and clearance.

The same mediation analysis was carried out on untargeted metabolomic (n=898 variables) and targeted lipidomic (n=997 variables) datasets. This identified 41 lipids with a mediated proportion effect over 50% only for the E2-TBI subgroup in the reduction of 3d mortality associated with TP administration. The top 10 features for each dataset are shown in Figure S3 and Figure S4 (Supplemental Digital Content 1, https://links.lww.com/SLA/E51). The lipids were comprised mostly of glycerophospholipids with the top 3 as follows: PE (18:0/22:5) (mediated proportion:85%), TAG52:4-FA20:4 (mediated proportion: 69%), LPC (16:1) (mediated proportion: 68%). None of the nonlipid metabolites achieved significance for 3-day mortality.

ER Identifies Circulating Causal Latent Factors Involved in Reducing Early Mortality in E2-TBI Patients

ER is a novel latent-factor-regression-based interpretable machine learning approach that overcomes the limitations of high dimensionality and multicollinearity often encountered in multiomic analysis.15 ER can integrate high-dimensional multiomic datasets, without making any assumptions regarding data-generating mechanisms, and identify significant causal latent factors, beyond predictive biomarkers, that drive an outcome(s) of interest. ER identified 70 latent clusters of features (designated by the letter Z), of which 4 were significantly involved in positively mediating (proportion >30%, Z8, Z27, Z33, Z24) the effect between TP and 3-day mortality in the E2-TBI subgroup (Fig. 3A). A secondary ER analysis using the Dantzig estimator18 (see the Methods section on essential regression) honed in on 2 of these 4 latent factors (Z8, Z24) that had already been identified as significant based on both ER (using the least squares estimator), and the mediation analyses. Together, our analyses suggest that these 4 latent factors capture higher-order relationships between proteins and lipids that contribute to differences in mortality, with the most significant effects being explained by Z8 and Z24. We also explored the mediation effect of E2-TBI specific latent factors and identified Z16 as the only with proportion over 30% (Fig. 3A). Interestingly, Z8 consists of proteins involved in coagulation cascades and overlaps with several proteins identified earlier in Table S1 (Supplemental Digital Content 2, https://links.lww.com/SLA/E52) (based on the univariate mediation analyses). Z24 captured a sPLA(2)GXIIA signature that was also seen in the univariate mediation analyses. Z27 and Z33 captured glycolipid [DAG (16:1/22:6), DAG (18:0/22:6)] and phospholipid [PE (P-16:0/16:0), PE (P-16:0/20:5), PE (P-18:0/20:5)] signatures, respectively. E2-TBI specific factor Z16 captured the apolipoprotein E signatures along with the mitochondrial integrity signature protein, CHCHD7. Importantly, our ER analyses agree well with the univariate mediation analyses and converge on a consistent set of 6 proteins, including, SERPINA5, CHCHD7, APOE3, PLG, protein C and PLA2GXIIA, and multiple lipids that explain differences in 3-day mortality. The results of the mediation analysis for 3 and 30 days survival adjusted for age and sex for these proteins is shown in Table S2 (Supplemental Digital Content 2, https://links.lww.com/SLA/E52).

F3
FIGURE 3:
Latent factors comprised of proteomic and lipidomic features aligned with treatment effect of thawed plasma in reducing early mortality. A, Forest plot shows average causal mediation effects and proportion of mediation effects with 95% confidence interval for 5 latent clusters with mediated proportion over 30% identified by essential regression from a multiomic dataset in E2-TBI patients (n=44). Outcome (3 day mortality) related latent clusters (Z8, Z27, Z33, Z24) were marked as red. A subgroup (E2-TBI vs. E2-NBI)-related latent cluster (Z16) is marked as blue. aIdentified by the least squares method. bIdentified by the Dantzig method. B, Loading scores for selected pure (identified to a specific latent cluster) features in 5 latent clusters identified in (A). CI indicates confidence interval; OR, odds ratio; Prop, proportion.

E2 Patients With TBI process TP Differently than Other Subgroups

To further understand why effective mediators were only identified in E2-TBI patients, we explored the relationship (shown in Fig. 1) of pathway B (TP with molecules/ factors measured in patients). We centered this on the tier 1 and tier 2 proteins (n=16) causally linked to survival in E2-TBI patients that received TP, which included the 6 proteins identified in both the linear and ER analyses. Relative levels of the proteins, represented as z-scores normalized across the patient groups, were projected onto a heatmap. A cohort of low or no-injury trauma patients transported by air medical transport and sampled upon arrival (n=2912) is included in the heat map as minimally injured controls. Two distinct patterns emerged, defined as module 1 and module 2. Module 1 consisted of proteins (mainly related with latent factor Z24) that were low at baseline, such as sPLA(2)GXIIA, and were only significantly higher in E2-TBI patients that received TP (Fig. 4A). Module 2 consisted of proteins (mainly related with latent factor Z8 and Z16), such as PLG and APOE3, that were high in controls but much lower after trauma. Notably, these proteins were higher in E2-TBI and E1-NBI patients that received TP, but not in E1-TBI or E2-NBI patients that received TP (Fig. 4A). Levels for 4 of the proteins (based on florescence activity) are shown in Figure 4B. The module 1 pattern with reconstitution of baseline levels is seen for SERPINA5, APOE3, and protein C and the module 2 pattern with an increase above baseline levels after TP in E2-TBI is evident for sPLA(2)GXIIA.

F4
FiGURE 4:
Polytrauma patients with TBI process thawed plasma differently than polytrauma patients without TBI. A, Heatmap shows the expression pattern of 16 tier 1 and tier 2 proteins for reducing 3 day mortality in TBI-E2 patients upon admission. Patients were grouped by TBI, endotypes and treatment arms. Z scores for each protein were normalized across all patient groups and relative levels shown. Low ISS is a group of low injury trauma patients that serve as control (n=29). Proteins (rows) were clustered by hierarchical clustering. *Indicate P<0.05 for Wilcox sum test comparison between the 2 treatment arms for each patient subgroup. B, Comparing expression levels of 4 representative proteins (plasma serine protease inhibitor: SEPRINA5, protein C, soluble phospholipase A(2)-XII, and Apoprotein E3) differently processed by TBI-E2 patients. P value was calculated by Wilcox sum test.

The protein factors causally linked to improved survival in E2-TBI patients are part of larger protein networks, most of which were well-represented in the Somascan platform. To visualize the differences between patient endotypes with or without TBI across the coagulation and fibrinolysis proteins, relative levels of the proteins from these pathways available from proteomic analysis are depicted in a heat map (Fig. S5, Supplemental Digital Content 1, https://links.lww.com/SLA/E51). Most of the proteins involved in coagulation and fibrinolysis were higher in the controls. In the absence of TP, protein levels dropped dramatically after injury in E1-NBI patients, and to an even greater degree in the E2-NBI and E2-TBI groups. Prehospital TP prevented this drop in the E1-NBI group and partially reversed the drop in the E2-TBI group, but had minimal impact on the protein levels in the E2-NBI cohort. Many of the end effectors of clotting and fibrinolysis were low at baseline and in the E1 groups. These increased dramatically in the E2 cohorts with the greatest increases in tissue factor, procoagulant proteases (eg, thrombin) and fibrinolytic enzymes (eg, plasmin) seen in the E2-TBI group. These changes were not influenced by TP. These findings confirm that TP is metabolized differentially depending on the injury severity and pattern. Within the most severely injured groups (E2), patients with TBI responded to TP with a partial reconstitution of coagulation factors while NBI patients did not.

DISCUSSION

In this analysis, we confirm that prehospital TP dramatically improves survival in severely injured trauma patients that have TBI and a heightened systemic response to trauma. Our mediation analysis of a multiomic dataset identified circulating proteins and lipids that are likely contribute to the selective survival benefit afforded by TP in E2-TBI patients. Based on the mediation and causal latent factor analyses, we conclude that improved blood clotting resulting from partial restoration of depleted clotting factors (eg, factor XI) and increases in proteases known to inhibit activated protein C and block fibrinolysis (SERPINA5 and CPB2) partially accounts for the improved survival. The mediation analysis also suggest that molecules involved in lipid transport and metabolism (APOE3 and sPLA(2)GXIIA), cell growth and survival (IGFBP4), and mitochondrial integrity (CHCHD7) may also contribute to the survival benefit of early TP administration. These findings also yield new insights into the nature of coagulopathy in severely injured patients with or without TBI.

The inclusion of even moderate TBI in patients with hemorrhagic shock is known to significantly increase trauma-induced coagulopathy (TIC) as measured by thromboelastography and PTT, which in turn, is associated with increased mortality.19–21 Tissue trauma with shock drives coagulopathy through consumption of clotting factors, endothelial injury, and inflammation.22 The addition of TBI is thought to amplify TIC through damage to the endothelium that comprises the blood–brain barrier with release of procoagulant factors from the injured brain.23 While a role for increased fibrinolysis has been suggested in isolated TBI, no increase in fibrinolysis was found in studies examining TBI + hemorrhagic shock.24 Research using a swine model of TBI has shown that early fresh frozen plasma improves brain function25 while the PAMPer study established that early TP increases survival only in patients with the combination of TBI with polytrauma associated with heightened systemic inflammatory response.6,7 It is notable from our work that early TP normalized the INR in both TBI and NBI patients groups. However, our proteomic analysis shows that TP restored coagulation factor levels only in patients with moderate nonbrain injury (E1-NBI) and in the E2-TBI patients. Our mediation analysis points to an important role for SERPINA5 and UPB2, proteases known to block protein C and fibrinolysis,26 in the survival benefit of TP. The increase in levels of inactivated protein C observed in our proteomic analysis could result from the increase in SERPINA5, which was initially described as an inhibitor of protein C and is known to block protein C activation by inhibiting thrombin-thrombomodulin.26 Interestingly, SERPINA5 also blocks urokinase and could contribute to an increase in PLG by preventing its conversion to plasmin. Future work charactering SERPINA5-centric protein network modules in E2-TBI could provide deeper insights into underlying molecular mechanisms. Taken together, our results suggest that improved coagulation afforded by TP in the E2-TBI subset could prevent early progression of the brain lesion by preventing ongoing intracranial bleeding where even small volumes of blood can be lethal.

It has been shown that lipid transport and metabolism play an important roles during brain development, hemostasias and injury.27–29 Interestingly, experimental studies showed that the lipid transport protein APOE3, but not APOE4 secreted from astrocytes can protect the integrity of the brain–blood barrier and ameliorate brain injury.30,31 Products of lipolysis and peroxidation by phospholipid enzymes was shown to mediate brain energy metabolism, mitochondrial damage and neuron ferroptosis.32–34 Prehospital TP prevented the drop in lipids that follows severe injury.11 Several glycolipids and glycerophospholipids were associated with the survival benefit of TP. These findings along with the highly significant mediation effect of APOE proteins (including APOE3) and the circulating lipid metabolizing enzyme, sPLA(2)GXIIA hint at the possibility that lipid transport into the injured brain or an increase in the production of bioactive lipids could also contribute to the salutary effects of TP in TBI patients. It is intriguing that proteins involved in growth factor transport and mitochondrial integrity strongly associate with the survival benefit of TP. Insulin-like growth factor is known to have neuroprotective actions.35 Whether, these processes, or other factors identified in the mediation analysis, such as the brain-enriched proteins, NPS36 or CTSF,37 contribute to the brain protective actions of TP will require studies in experimental models.

There are several important limitations to our study. Our observations from the PAMPer trial may not be representative of other populations. For instance, the single center COMBAT trial did not find a survival benefit in early administration of TP in patients transported by ground ambulance.38 However, post hoc analyses comparing the trials suggest this discrepancy might have been caused by the substantial difference between transport times; the shorter times in the COMBAT meant controls could have received additional hemostatic care sooner, which in turn might have masked the beneficial effect of TP evident in longer transport times of PAMPer.39 In addition, the COMBAT trial only consisted of few TBI patients (n=28), the subset found to be most sensitive to TP in PAMPer trial.6 We propose that discrepancies between trials might be better understood if future trial design included deeper analysis of circulating biomarkers. Despite the precision afforded by our complimentary strategies to identify causal features, perturbation experiments in model organisms are necessary to prove causality. In addition, our analysis is based entirely on the levels of circulating biomolecules, most of which have critical biologic activities. We do not assess the activities of these proteins and lipids, or identify regulatory networks driving corresponding activities. Finally, our targeted analysis of proteins, lipids and other metabolites does not represent the entire spectrum of circulating biomolecules and therefore, we cannot conclude that we have assessed all of the circulating factors that are likely to associate with the beneficial effect of early TP.

ACKNOWLEDGMENTS

The authors acknowledged the contribution of collaborators involved in PAMPer study for the clinical data collection.

DISCUSSANT

Dr. Leopoldo C. Cancio (San Antonio, TX)

Hi, Lee Cancio from San Antonio, Texas. Dr. Billiar and colleagues have greatly extended our understanding of the effect of resuscitation strategy on the response to severe injury. Over the last 20 years, we've witnessed a revolution in resuscitation. These changes began in Iraq with a retrospective study in 2007 that showed that combat casualties who underwent massive transfusions had decreased death from hemorrhage when they received plasma and red-blood-cell transfusions in equal ratios. This was followed by the PROMMT study, a prospective observational study in civilians. This study confirmed that those who received plasma and red cells in equal ratios as well as platelets and red cells in equal ratios had decreased 6-hour mortality.

The PROPPR study was an RCT, also conducted in civilians, comparing a 1-to-1-to-1 ratio of plasma, platelets, and red blood cells to a 1-to-1-to-2 ratio. There was no difference in mortality at 24 hours or at 30 days, but fewer patients in the 1-to-1-to-1 group died of hemorrhage at 24 hours.

Next came 2 studies of prehospital plasma versus crystalloid. The COMBAT study in Denver, a ground-ambulance study, was interrupted for futility. The PAMPer study in Pittsburgh was an aeromedical study, which found a decrease in mortality at 24 hours and at 30 days and a decrease in the INR in the plasma group. Pusateri reanalyzed these data, concluding that the difference in results in the two studies could be explained by a longer prehospital time in the latter study. That is, plasma became effective if the prehospital time was >20 minutes.

Subsequent subgroup analyses isolated the therapeutic effect in PAMPer to patients with TBI as found on computed tomography scan. They then further honed in on TBI patients with a specific E2.

The present study further examines the effects of prehospital plasma in TBI patients with E2. They found that plasma prevented the decrease in lipids after injury and improved coagulation, but had no effect on markers of endotheliopathy or on systemic inflammation.

My main questions are as follows. Number 1, given that previous in-hospital PROMMT and PROPPR studies identified decreased hemorrhage as a main benefit of balanced-ratio treatment, would the authors not agree that it would be premature to limit the use of plasma in the prehospital setting to only those with TBI?

Two, is it possible that a benefit of plasma in non-TBI patients was missed in the PAMPer study because these patients were on average less likely to receive a massive transfusion than those in the study from Iraq?

Three, do the authors envision any way to rapidly identify endotype in order to personalize therapy?

And finally, extensive burns are another condition with systemic inflammation and endothelial injury. Would plasma exert similar effects there as well, or is this pathophysiology unique to TBI?

And I thank the Association for the privilege of discussing this outstanding paper.

Response from Timothy Billiar

Thank you, Dr Cancio, for your insightful questions. The issue of the timing of plasma and possible other benefits of plasma are important, and of course all trauma patients are not the same. Trials enroll patients with different characteristics. I think some of the question has to do with the mechanism of action of this early prehospital plasma in these patients in which transport took a little bit longer than in the COMBAT trial. The need to have a significant brain injury to realize the benefit aligns with the mechanism of action where improving clotting to just a small degree and preventing even low-volume bleeding in the brain likely accounts for some of the benefit.

As we also showed, prehospital plasma improved the INR in patients with evidence of early coagulopathy, and we are not reporting here any of the other possible benefits of pre-hospital plasma beyond the effect on mortality, so I think based on the demonstration that coagulopathy was improved by prehospital plasma and strong evidence that using the one-to-one plasma administration within the hospital, there is no change in the recommendation that plasma should be used, especially in the setting of hemorrhagic shock.

Your question about rapidly identifying, especially in the prehospital setting, patients likely to benefit from the plasma is an important one. In the previous study that I mentioned that we published, we did look at biomarkers measured at admission. However, we found that a well-known marker of brain injury, UCHL1, as highly predictive of which patients would fall into the E2 TBI group. AUCs were over 0.9. Now again that was measured at admission. Now we need to measure that at point of care. If we can measure that at point of care, we may have a biomarker that would allow us to accurately predict which patients at the scene would benefit from prehospital plasma.

Using plasma in burns, I think it needs to be tested. It could have therapeutic benefits beyond the coagulation effects mediated through other pathways that cause mortality, but until we test it, and your work obviously is leading the way in this area, we won't know.

Dr. Rachael Callcut (Sacramento, CA)

Rachel Callcut from Sacramento. A really important study, and I think you're right that part of the differential outcome that our patients have experienced in some of these trials probably comes down to some element of proteomic expression as well as metabolomic. I'm wondering if you have enough data within the TBI cohort to show if there's a differential expression of these depending on the severity of brain injury? Thanks.

Response from Timothy Billiar

So we know that the E2 patients have higher proinflammatory mediators. They have endotheliopathy markers as well as many other markers showing that there's a heightened systemic response. They also have typical brain injury markers including GFAP and UCHL1. UCHL1 is a little better than GFAP for early detection. Since we know there is development of point-of-care instruments that will be able to measure these biomarkers, I believe that we'll be able to identify early on patients more likely to benefit from early plasma. It's the progression of the brain lesion most likely that occurs early that we're able to impact in patients that have a high degree of coagulopathy. What I haven't been able to show you (but will be in the paper) is the remarkable changes across the whole family of coagulation and fibrinolysis proteins that are evident within the proteomic dataset. Levels of these proteins are changing dramatically, and what we see when we administer the prehospital plasma to the TBI E2 group is a partial restoration of many of the coagulation factors.

The patients without TBI that have polytrauma in this E2 group show very little reversal with the prehospital plasma. It may not be enough plasma for the coagulopathy that is seen in the polytrauma patients without TBI. We're putting the heat map in the paper.

Dr. David Soybel (White River Junction, VT)

Yes, thanks, Tim. It’s been awesome to watch you for 40 years work these pathways out. I wonder if I could push you a little bit more on the difference between the E1 and the E2 groups. It must be genetics plus environment, but I’m wondering if you have any thoughts about why some people are the responders and some people are protectors?

Response from Timothy Billiar

That’s a good question. The endotypes were selected by a clustering strategy, and what came out were some of the features that I showed in my presenaton. But, I did not show that E1 patients were not injured quite as badly, so their ISS is more in the 19 to 20 range where the E2’s are closer to 30. That doesn’t capture all the difference in the injury characteristics, and the TBI patients within E1, many of them were isolated TBI with minor injury. So, I think part of the difference is the injury pattern, but not everything, and there are of course people with moderate to minor injury that have a big systemic response. Then there are patients with a high injury burden but a limited systemic response. So, the endotyping makes sure that all patients end up in the right group. I think that is why we can correctly identify the 20% to 30% of patients that actually respond to the prehospital plasma for survival benefit.

There is definitely evidence of endotheliopathy measured by several endotheliopathy markers in E2 patients, along with differences across the multiomics. I think this is driven, in part, by the injury severity, and by other factors, such as genetics, and then transfer times.

Dr. Orlando Soybel (Abington, PA)

Great study, Dr Billiar. The question that I have is you mentioned about the noncoagulation protein and that they also are impacted. Is there a sensitivity issue here in terms of is there a balance that may be related to the severity of injury or the time the plasma was elevated? It seems that that could also turn off the whole effect that you’re trying to propose. Thank you.

Response from Timothy Billar

Yeah, thank you for raising that question. So if you look across all the coagulation proteins, most are present at baseline. Even with the minor injury that the E1 NBI patients experience, there was a rapid drop. By the time they arrived to the hospital, at least using our proteomic analysis, almost all of those constitutively expressed coagulation proteins have dropped several fold. The TBI E1 patients have a small drop, and that's consistent with the failure to see an elevation in INR. The E2 patients, whether they are NBI or they have TBI, display precipitous drop in all coagulations proteins, and this is all happening within one to two hours. The administration of prehospital plasma in the NBI patients, did not reverse the drop in coagulation proteins. These patients are typically in some degree of shock and have extensive peripheral tissue trauma.

For some reason, the patients with TBI and polytrauma, exhibited partial restoration. The serine protease that I talked about may be a big factor based on the causal mediation analysis. Levels of SERPINA5 were partially restored in E2-TBI patients that received pre-hospital plasma. There were other proteins that were low at baseline and became elevated selectively in the TBI E2 group, soluble phospholipase A2’s, for example.

Causality will have to be established by appropriate experiments, but this gives us a little bit of a roadmap where to go, and better yet, as I said, I think it gives us a signature. We should look for these signatures in these other trials, and I think biobanks should become mandatory when you're doing an interventional trial, so that we can compare between trials why we see different results in seemingly similar trails. There's going to be design issues that explain the differences as well, but we could normalize this across a set of biomarkers by using the same omics and plasma stored the same way. It seems to me otherwise we're wasting money. We’re getting different results in different trials, and we don’t know why.

This study was intended to help us understand why plasma was having this effect in about 20% to 25% of the patients. The other 75%, there's no mortality effect. That’s why, so you’ve got to find the right patient subset and then the right circumstance and the right transfusion. I think that’s where precision medicine needs to go in trauma. Thank you.

REFERENCE

1. Moore FA, McKinley BA, Moore EE. The next generation in shock resuscitation. Lancet. 2004;363:1988–1996.
2. Harris T, Davenport R, Mak M, et al. The evolving science of trauma resuscitation. Emerg Med Clin North Am. 2018;36:85–106.
3. Cap AP, Pidcoke HF, Spinella P, et al. Damage control resuscitation. Mil Med. 2018;183(suppl_2):36–43.
4. Jansen JO, Thomas R, Loudon MA, et al. Damage control resuscitation for patients with major trauma. BMJ. 2009;338:b1778–b1778.
5. Sperry JL, Guyette FX, Brown JB, et al. Prehospital plasma during air medical transport in trauma patients at risk for hemorrhagic shock. N Engl J Med. 2018;379:315–326.
6. Gruen DS, Guyette FX, Brown JB, et al. Association of prehospital plasma with survival in patients with traumatic brain injury: a secondary analysis of the pamper cluster randomized clinical trial. JAMA Netw Open. 2020;3:e2016869.
7. Wu J, Vodovotz Y, Abdelhamid S, et al. Multi-omic analysis in injured humans: patterns align with outcomes and treatment responses. Cell Rep Med. 2021;2:100478.
8. Lutton EM, Farney SK, Andrews AM, et al. Endothelial targeted strategies to combat oxidative stress: improving outcomes in traumatic brain injury. Front Neurol. 2019;10:582.
9. Barelli S, Alberio L. The role of plasma transfusion in massive bleeding: protecting the endothelial glycocalyx? Front Med (Lausanne). 2018;5:91.
10. Vigiola Cruz M, Carney BC, Luker JN, et al. Plasma ameliorates endothelial dysfunction in burn injury. J Surg Res. 2019;233:459–466.
11. Wu J, Cyr A, Gruen D, et al. Lipidomic signatures align with inflammatory patterns and outcomes in critical illness. Res Sq. Preprint published January 8, 2021. doi:10.21203/rs.3.rs-106579/v1.
12. Guyette FX, Brown JB, Zenati MS, et al. Tranexamic acid during prehospital transport in patients at risk for hemorrhage after injury: a double-blind, placebo-controlled, randomized clinical trial. JAMA Surg. 2020;15:11–20.
13. Davies DR, Gelinas AD, Zhang C, et al. Unique motifs and hydrophobic interactions shape the binding of modified DNA ligands to protein targets. Proc Natl Acad Sci USA. 2012;109:19971–19976.
14. Tingley D, Yamamoto T, Hirose K, et al. mediation:R package for causal mediation analysis. J Stat Softw. 2014;59:1–38.
15. Bing X, Lovelace T, Bunea F, et al. Essential Regression: a generalizable framework for inferring causal latent factors from multi-omic datasets. Patterns (N Y). 2022;3:100473.
16. Yu G, Wang L-G, Han Y, et al. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS. 2012;16:284–287.
17. Imai K, Keele L, Yamamoto T. Identification, inference and sensitivity analysis for causal mediation effects. Stat Sci. 2010;25:51–71.
18. Dantzig GB Nash SG. Origins of the simplex method. A History of Scientific Computing. New York, NY: ACM; 1990:141–151.
19. Cannon JW, Dias JD, Kumar MA, et al. Use of thromboelastography in the evaluation and management of patients with traumatic brain injury: a systematic review and meta-analysis. Crit Care Explor. 2021;3:e0526.
20. Kunio NR, Differding JA, Watson KM, et al. Thrombelastography-identified coagulopathy is associated with increased morbidity and mortality after traumatic brain injury. Am J Surg. 2012;203:584–588.
21. van Gent JAN, van Essen TA, Bos MHA, et al. Coagulopathy after hemorrhagic traumatic brain injury, an observational study of the incidence and prognosis. Acta Neurochir (Wien). 2020;162:329–336.
22. Moore EE, Moore HB, Kornblith LZ, et al. Trauma-induced coagulopathy. Nat Rev Dis Primers. 2021;7:30.
23. Zhang J, Zhang F, Dong J-F. Coagulopathy induced by traumatic brain injury: systemic manifestation of a localized injury. Blood. 2018;131:2001–2006.
24. Samuels JM, Moore EE, Silliman CC, et al. Severe traumatic brain injury is associated with a unique coagulopathy phenotype. J Trauma Acute Care Surg. 2019;86:686–693.
25. Imam A, Jin G, Sillesen M, et al. Fresh frozen plasma resuscitation provides neuroprotection compared to normal saline in a large animal model of traumatic brain injury and polytrauma. J Neurotrauma. 2015;32:307–313.
26. Meijers JCM, Herwald H. Protein C inhibitor. Semin Thromb Hemost. 2011;37:349–354.
27. Betsholtz C. Lipid transport and human brain development. Nat Genet. 2015;47:699–701.
28. Lai J-Q, Shi Y-C, Lin S, et al. Metabolic disorders on cognitive dysfunction after traumatic brain injury. Trends Endocrinol Metab. 2022;33:451–462.
29. Raulin A-C, Martens YA, Bu G. Lipoproteins in the central nervous system: from biology to pathobiology. Annu Rev Biochem. 2022;91:731–759.
30. Teng Z, Guo Z, Zhong J, et al. ApoE influences the blood-brain barrier through the NF-κB/MMP-9 pathway after traumatic brain injury. Sci Rep. 2017;7:6649.
31. Bell RD, Winkler EA, Singh I, et al. Apolipoprotein E controls cerebrovascular integrity via cyclophilin A. Nature. 2012;485:512–516.
32. Knopp RC, Lee SH, Hollas M, et al. Interaction of oxidative stress and neurotrauma in ALDH2-/- mice causes significant and persistent behavioral and pro-inflammatory effects in a tractable model of mild traumatic brain injury. Redox Biol. 2020;32:101486.
33. Chitturi J, Li Y, Santhakumar V, et al. Consolidated biochemical profile of subacute stage traumatic brain injury in early development. Front Neurosci. 2019;13:431.
34. Lamade AM, Anthonymuthu TS, Hier ZE, et al. Mitochondrial damage & lipid signaling in traumatic brain injury. Exp Neurol. 2020;329:113307.
35. Zheng WH, Kar S, Doré S, et al. Insulin-like growth factor-1 (IGF-1): a neuroprotective trophic factor acting via the Akt kinase pathway. J Neural Transm Suppl. 2000:261–272.
36. Pape H-C, Jüngling K, Seidenbecher T, et al. Neuropeptide S: a transmitter system in the brain regulating fear and anxiety. Neuropharmacology. 2010;58:29–34.
37. Wang B, Shi GP, Yao PM, et al. Human cathepsin F. Molecular cloning, functional expression, tissue localization, and enzymatic characterization. J Biol Chem. 1998;273:32000–32008.
38. Moore HB, Moore EE, Chapman MP, et al. Plasma-first resuscitation to treat haemorrhagic shock during emergency ground transportation in an urban area: a randomised trial. Lancet. 2018;392:283–291.
39. Pusateri AE, Moore EE, Moore HB, et al. Association of prehospital plasma transfusion with survival in trauma patients with hemorrhagic shock when transport times are longer than 20 minutes: a post hoc analysis of the pamper and COMBAT clinical trials. JAMA Surg. 2020;155:e195085.
Keywords:

causal mediation analysis; multiomics; PAMPer; polytrauma; thawed plasma; traumatic brain injury

Supplemental Digital Content