Inflammatory Mediators Related to Vascular Dysfunction are Linked to ICP,PRx,and CPP Following Human Severe Traumatic Brain Injury

Abstract

Disturbed cerebral autoregulation (represented by a positive pressure reactivity index [PRx]), elevated intracranial pressure (ICP), and decreased cerebral perfusion pressure (CPP) are key treatment targets following severe traumatic brain injury (sTBI). This study investigated neuroinflammation as a potential mechanism underlying these intracranial disturbances. Plasma samples from 11 sTBI patients (from a prior Phase II drug trial) were analyzed for 174 proteins using an antibody-based suspension bead array, with intervention effects accounted for where possible. Dimensionality reduction techniques, including principal component analysis (PCA) and supervised methods, were applied to protein data, informed by physiological variables (ICP, CPP, and PRx). PCA revealed distinct protein clustering patterns related to ICP >20 mmHg and PRx > 0, with PC1 linked to patient ID, time from injury, and intervention, and PC2/PC3 significantly associated with PRx dose (p < 0.001). Markers relating to inflammation of the vascular system comprised 20% of the top 50 proteins influencing PC2, implicating complement inflammation in these processes. Notably, MASP-2 (p = 0.027) and complement factor I (p = 0.039) were significantly associated with PRx dose in a mixed-effects model. These findings suggest that vascular inflammation, particularly complement activation, may contribute to intracranial physiological disturbances in sTBI, highlighting the complement pathway as a potential target for further investigation.

Keywords

cerebral autoregulation cerebrovascular pressure reactivity complement inflammation inflammation neurocritical care

Introduction

In the neurocritical care unit, treatment of severe traumatic brain injury (sTBI) is aimed at avoiding or managing secondary brain injury. After the primary damage caused by the traumatic insult, the brain is vulnerable to secondary injury via dynamic pathophysiological changes, which leads to poor outcomes and death after TBI.^1,2 Secondary brain injury includes the endogenous cascade of injury within the brain, and the additional effects of extra-cerebral damage.³

Intracranial hypertension is a secondary injurious mechanism following sTBI, which is commonly monitored,⁴ and used as a goal-directed target. Both the severity and duration of raised intracranial pressure (ICP) are associated with poor outcome.⁵ Elevated ICP leads to a drop in cerebral perfusion pressure (CPP) and puts the patient at risk of ischemia. Cerebral autoregulation (CA) aims to maintain a constant cerebral blood flow (CBF) despite changes in CPP, however, this mechanism is impaired following TBI.^6,7 Hence, changes in CPP may lead to passive fluctuations in the CBF, and secondary damage via hypoxia or hyperemia.⁸ Cerebrovascular pressure reactivity, the principal mechanism of CA, can be monitored via a surrogate correlation between ICP and arterial blood pressure, referred to as pressure reactivity index (PRx).⁹ A positive PRx indicates disturbed cerebrovascular reactivity and has been associated with worse patient outcomes after sTBI.¹⁰ Monitoring intracranial physiology dynamics (specifically ICP, CPP, and PRx, henceforth referred to as neuromonitoring-derived parameters) after a brain injury is proposed to ameliorate secondary injury as a key component in multimodality monitoring.¹¹

CA has been shown to function via myogenic, metabolic, and neurogenic mechanisms,¹² but the molecular and proteomic pathways involved are largely unknown. Neuroinflammation entails the sequential involvement of various inflammatory cells and signaling molecules across the central nervous system. In addition, a systemic inflammatory response is ubiquitous following TBI.^13,14 Persistent inflammation triggered by the initial trauma and further exacerbated by secondary injury has been linked to increased oedema, vascular permeability via blood–brain barrier (BBB) breakdown, disrupted cell signaling, and cell death.¹⁵ The initiation of neuroinflammation following TBI is well recognized but the contribution to intracranial physiological disturbances is unknown.

In this study, we aimed to explore inflammation as a biological mechanism of intracranial physiological derangements (as measured with ICP, CPP, and PRx) post-sTBI using state-of-the-art proteomic and computational techniques in a clinical cohort of 11 sTBI patients who had both blood plasma samples and high-frequency physiological monitoring for the first week post-injury.

Methods

Study participants and trial design

This was a retrospective analysis of samples and data collected as part of a prospective, interventional phase II (open-label) randomized-controlled trial of recombinant human IL1ra (rhIL1ra, anakinra) in sTBI.¹⁶ The original phase II trial was approved by the local research ethics committee (06/Q0108/64), and the next of kin provided full assent in line with prospective ethical approvals. All patients received sedation (with/without neuromuscular blockade), endotracheal intubation, mechanical ventilation, and multimodality monitoring as part of standard clinical practice. In total, n = 20 patients were randomized to drug or no drug groups (1:1), where rhIL1ra was given at a dose of 100 mg subcutaneously once a day for 5 days. As part of this trial, arterial and jugular blood samples were collected twice daily for the first week post-injury. The trial protocol included sampling 1 h before and after drug administration (either hypothetical for the control group, or rhIL1a for the treatment group). Samples were centrifuged for 15 min at 4000 g at 4°C, and the plasma was stored at −80°C. Eleven patients from the original trial had matched high-frequency ICM+ physiological data and were included in this analysis.

Accounting for the effect of intervention

In the current study of 11 patients (who had blood sampling with matched physiological data), we have two sub-cohorts of patients: those that received the IL1ra drug (intervention group, n = 7) and those that did not (control group, n = 4). To understand the effect of the IL1ra drug on the brain physics parameters, the mean dose of physiological variables was compared for the intervention and control groups across the total monitoring time (Mann–Whitney test) and per day (mixed model). Furthermore, we portrayed a subset of proteins (details on the choice of proteins specified below) temporally across treatment groups. In summary, the intervention and control sub-cohorts were not significantly different for physiological parameters; however, trends were acknowledged. Henceforth, subsequent analyses adjust for intervention where possible.

Proteomic analysis

In total, 177 proteins were analyzed, targeted by 219 antibodies (Supplementary Table S1). Of these, n = 6 antibodies failed initial quality control and were removed, yielding 174 proteins targeted by 213 distinct antibodies available for final analysis. Two antibodies targeting different epitopes of the same protein were included for 39 proteins.¹⁷ The protein panel was chosen based on CNS enrichment¹⁸ or previous clinical, experimental, or proteomic neuroinflammatory studies.^13,19–24 Antibodies, selected based on the available antibody set in the Human Protein Atlas, were immobilized onto color-coded magnetic beads, as previously described.^20,25 The protein content of the samples was directly labelled with biotin and incubated with the bead array. Target proteins in the sample then bind to their specific antibodies, and read-out was facilitated using a streptavidin-conjugated fluorophore in a FlexMap 3D instrument (Luminex Corporation). The relative protein abundance in each plasma sample is reported as median fluorescent intensity (MFI) levels.

Physiological data acquisition and processing

ICP monitoring was performed in all patients using an intraparenchymal pressure monitor (Codman, Raynham, MA), and ABP was measured using a radial/femoral arterial line (Baxter Healthcare, Deerfield, IL) zeroed at the level of the right atrium. Real-time high-resolution (240 Hz) physiological monitoring was performed at the bedside using ICM+ software (Cambridge Enterprise, Cambridge, UK). Clinicians had full access to bedside monitoring data and treated patients as per the Brain Trauma Foundation Guidelines.²⁶

Curation and processing of physiological data were done in ICM+. Curation included manual inspection for gross artefacts pertaining to non-physiological waveform disturbances. Then, automated removal based on non-physiological minima/maxima and lack of pulsatility was applied. Cleaned data were averaged per minute for each physiological signal. CPP was calculated as ABP-ICP, and PRx was calculated as previously described.²⁷ Finally, daily averages of ICP, CPP, and PRx were calculated.

Statistical analysis

All analyses were performed in R Studio (version 4.4.1: https://www.r-project.org/), with statistical significance set at the p = 0.05 level. See a comprehensive list of packages used in Supplementary Data S2.

Dose calculations of deranged brain physiology

To investigate the effect of deranged intracerebral physiology, pathological thresholds of ICP, PRx, and CPP known to associate with worse outcome following sTBI were selected: ICP > 20 mmHg,⁵ CPP <60 mmHg,²⁶ and PRx > 0.⁹ The chosen CPP threshold is based on zeroing of the mean arterial pressure transducer at the level of the right atrium, in line with current recommendations from the Brain Trauma Foundation Guidelines.²⁶ For each parameter, the deviation from this threshold value over time was calculated as the area under the curve, as previously described.^28,29 A visual depiction of this calculation can be found in Supplementary Data S3. To account for variations in monitoring duration as well as duration of deranged brain physiological variable, the daily “dose” of pathological ICP, PRx, or CPP is reported and adjusted for the monitoring length (computational details provided in Supplementary Data S3). For robustness, we also report the percentage of monitoring time spent with a “pathological” ICP, PRx, and CPP (i.e., number of minutes each metric was above/below threshold, divided by the total number of monitoring minutes).

Daily values of all metrics (mean, dose, and percentage time) were calculated and matched with individual blood samples or their daily averages for each analyte as deemed appropriate per the analysis undertaken. Where necessary, per patient values for each metric were used. As analyses required time-linked neuromonitoring and proteomic data, the limiting data was the blood sampling frequency, and hence, daily values for the first week postinjury were used.

Dimensionality reduction and multivariable analysis of daily physiological burden

Principal component analysis (PCA) was undertaken with the full panel of 174 proteins targeted by 213 antibodies. We compared MFI for all analytes (arterial and jugular, sampled every 12 h) with the daily dose of the brain physiological variables. This means that one brain physiological data point (daily dose) will be linked to several protein measurements (arterial and jugular, possibly several time points). As biological sampling of this sort varies based on clinical interventions and partially missing monitoring periods/blood sampling intervals, this was deemed pragmatic in cohorts of this kind. The PCA data was zero-centered and scaled to unity. Univariable regression analysis was used to assess the contribution of each PC to the daily burden of deranged intracranial physiology. Linear mixed modeling was done with the individual sample ID (unique for arterial/jugular blood, and per sampling time point) as a random effect. Independent variables demonstrating either a trend toward, or significance itself (p < 0.05), in univariable analysis were incorporated into a multivariable model, where variables were excluded in a step-down fashion. In addition, PCA loadings were investigated to distinguish individual protein contributions to each PC.

Supervised PCA: partial least square regression and partial least square discriminant analysis

Following an unbiased PCA analysis, we sought to find the optimal subset of proteins to describe physiological derangement. Therefore, supervised dimensionality reduction via partial least square regression (PLS) and partial least square discriminant analysis (PLS-DA) were undertaken. For PLS-DA, log10-transformed brain physiological data were split at the median value, thereby creating a binary variable to which analyte levels were assigned. Model fine tuning was done with cross-fold validation (Mfold, n = 10 for PLS and n = 5 for PLS-DA), repeated 5 and 10 times, respectively. For PLS, proteins from the suggested subset were considered for stability if variable importance in projection (VIP) coefficients were close to 1.0. For PLS-DA, the number of components was minimized using multiple distance metrics as reported by the R package (max, centroid, mahalanobis³⁰) with variables driving PC1 reported if VIP ≥ 0.8.

Further analyses in the vascular protein subset

We have previously identified a panel of inflammatory proteins likely to be involved in vascular dysfunction,¹² and hence created a subset of 17 proteins targeted by 21 antibodies (Supplementary Data S4). Note that these are proteins selected for their general involvement (in any way) in vascular inflammation, and include both those representing inflammation affecting blood vessels, and as well as inflammation of the vessels themselves. Henceforth, for the purpose of this article, we use the term “vascular inflammation” to mean any inflammation relating to the vasculature.

Investigations into this vascular subset included arterial samples only. Associations between inflammatory protein concentrations and neuromonitoring-derived parameters were done using correlation and regression analyses. Spearman’s correlations were done for the entire cohort and by intervention group. Linear mixed effect modeling was done: Scaled protein levels were modelled as predictors of physiological variables, patient and time were used as random effects, and intervention group as a fixed effect. The interaction between protein levels and intervention was modelled and reported if significant.

Results

In the current study of 11 patients (who had blood sampling with matched physiological data), we have two sub-cohorts of patients: those that received the IL1ra drug (intervention group, n = 7) and those that did not (control group, n = 4). Clinical and baseline characteristics of the patients participating in the study are described in Table 1, with further details on each patient in Supplementary Table S3. Eleven patients had brain physiological monitoring and protein samples quantified. The cohort had a median age of 43 years at injury, with a post-resuscitation total Glasgow Coma Scale (GCS) of 3. The predominant cause of injury was a road traffic accident (n = 7, 64%). All patients had major/severe injury as defined by ISS > 15, with a median ISS of 43.

Table 1.

Cohort Demographics

	Cohort	IL1ra treatment group
Characteristic	N = 11^a	Control, N = 4^a	Intervention, N = 7^a	p-Value^b
Age (years)	43 (33, 49)	48 (40, 55)	41 (33, 45)	0.3
Sex				0.2
Female	5 (45%)	3 (75%)	2 (29%)
Male	6 (55%)	1 (25%)	5 (71%)
GCS total	3 (3, 7)	3 (3, 4)	5 (3, 8)	0.4
GCS motor	1 (1, 5)	1 (1, 2)	3 (1, 5)	0.5
Injury severity score	43 (37, 45)	42 (38, 45)	43 (35, 48)	>0.9
Physiological variables
Mean ABP (mmHg)	91 (87, 99)	86 (83, 87)	95 (92, 103)	0.006
Mean ICP (mmHg)	13.4 (10.6, 16.8)	11.6 (9.4, 13.5)	16.5 (12.3, 18.5)	0.3
Mean PRx	−0.02 (−0.10, 0.07)	0.04 (−0.03, 0.18)	−0.07 (−0.13, 0.03)	0.2
Mean CPP (mmHg)	77 (74, 84)	74 (72, 75)	82 (78, 88)	0.042
Adjusted dose of ICP	23 (22, 37)	25 (23, 31)	23 (21, 37)	>0.9
Adjusted dose of PRx	0.34 (0.26, 0.41)	0.30 (0.26, 0.35)	0.38 (0.27, 0.50)	0.5
Adjusted dose of CPP	56 (54, 84)	55 (54, 69)	56 (27, 84)	>0.9
Ptime ICP >20 mmHg (%)	34 (28, 53)	48 (28, 67)	34 (28, 41)	0.5
Ptime PRx >0 (%)	46 (34, 58)	53 (41, 69)	46 (29, 53)	0.3
Ptime CPP <60 mmHg (%)	10 (5, 14)	5 (5, 6)	13 (11, 15)	0.042

^aMedian (IQR); n (%).

^bWilcoxon rank sum exact test; Fisher’s exact test; Wilcoxon rank sum test.

IL1ra, interleukin 1 receptor antagonist; n, number of patients; GCS, Glasgow Coma Scale; ABP, arterial blood pressure; ICP, intracranial pressure; PRx, pressure reactivity index; CPP, cerebral perfusion pressure; Ptime, percentage time.

In total, 174 proteins were analyzed in up to 362 plasma samples (arterial and/or jugular) from 11 sTBI patients within their first week from injury. Physiological data consisted of 162.3 h (6.8 days) of data per patient, originally sampled at 250 Hz.

Accounting for the effect of intervention

Based on the nature of retrospective analyses and the fact that we investigate our hypotheses in the context of the IL1ra trial, we are aware of a possible effect of intervention on our analyses. Across the total monitoring time with physiological monitoring, there was no difference between total dose of deranged brain physiological parameter between the treatment and control groups (Fig. 1A). Throughout the first week post-trauma, when patients had dual monitoring (physiological and blood samples) treatment group (control or rh-ILra treatment) did not affect ICP, or CPP burden, or PRx dose burden per day (Fig. 1B). However, dose of PRx > 0 seemed to be slightly lower in the IL1ra group as compared with control patients (β = −0.10, p = 0.06). The protein levels were plotted over time by treatment groups, with an overlapping 95% confidence intervals throughout the 5 days of sampling. Figure 1C shows a subset of the proteins (chosen for visualization based on their later significance). We have considered the effect of IL1ra where possible in our analysis.

FIG. 1.

Treatment effect on dose of derangement of neuromonitoring-derived parameters during entire monitoring period (A) and over the dual-monitoring time (B) as well as vascular protein subsets longitudinally (C). Utilizing accumulated doses of ICP, PRx, and CPP, the total burden of deranged brain physics (adjusted for monitoring duration) could be calculated. (A) shows no obvious difference between total dose (per study subject) of deranged ICP, PRx, or CPP between the control and rh-IL1ra groups (p values at the top of the graph) across the full time, all patients had neuromonitoring equipment inserted. (B) plots the dose over time throughout the monitoring time with concomitant blood samples and neuromonitoring (i.e., the first week after trauma). Graphical trends are seen indicating a treatment effect (more periods of deranged ICP in the control group (top panel), and a lower burden of PRx values in the treatment group (middle panel). Similarly to the total dose burden, however, treatment group (control or rh-ILra treatment) did not affect brain physics dose burden per day, but there was a trend for PRx (mixed model, p values not shown). (C) plots the vascular subset of proteins over time for the intervention and control groups, showing overlapping 95% confidence intervals (indicated with colour ribbons), indicating similar levels and temporal patterns for the two patient sub-cohorts. CPP, cerebral perfusion pressure; ICP, intracranial pressure; PRx, pressure reactivity; rhIL1ra, recombinant human interleukin 1 receptor antagonist.

Dimensionality reduction

Dimensionality reduction was conducted on all proteomic data (arterial and/or jugular samples) from the 11 patients. The accompanying PCA scree plot (Supplementary Data S5) indicated that a maximum of three to four principal components (PCs) described most of the variance in the data. The first and second PCs were mostly explained by patient identity and time from trauma. Notably, there was an influence of rh-IL1ra treatment on PC1 and 2 (Supplementary Data S6).

Neuromonitoring-derived parameters, as assessed via time-adjusted dose-per-day of ICP, PRx, or CPP, visually seem to be related to PC2 and 3 (Fig. 2A). PC1 was not associated with daily deranged dose of ICP, PRx, or CPP upon univariable linear mixed regression model assessment (Supplementary Table S4). Following these univariable results, we investigated the association between our neuromonitoring parameters and the PCs, while controlling for the blood sample and time (via unique sample IDs) as a random effect. Further, interaction effects between the PCs and day from trauma or intervention are investigated, with all results shown in Table 2. In these models, variables were excluded in a step-down fashion. PC2 (t = −3.8, p < 0.001), and PC3 (t = −5.5, p < 0.001) were associated with the daily dose of deranged ICP. PC2 and 3 were significantly associated with PRx dose in a multivariable regression (t = −6.4, p < 0.001 and t = −5.5, p < 0.001, respectively). In this model, treatment and time from trauma were modelled as interaction effects with PC2 and were significant (p < 0.001 for both interactions). In a step-down multivariable regression, PC2 was the final variable retained with a significant association to CPP (t = −2.0, p = 0.045, Table 2). The proteins within each PC driving these regression results are illustrated in Figure 2B. Here, there is an over-representation (10 proteins, 20%) of proteins relating to vascular inflammation in PC2. This was investigated further using supervised dimensionality reduction techniques.

FIG. 2.

Principal component analysis of proteins, demonstrating clustering based on deranged neuromonitoring-derived metrics. PC results where daily pathological doses of ICP, PRx, and CPP are demonstrated using the colour gradient (A). Notably, protein samples cluster in accordance with the extent of physiological derangement. Investigations into the loadings driving the PCs (B) show an over-representation of proteins involved in vascular inflammation (dark shading = vascular inflammatory protein). CPP, cerebral perfusion pressure; ICP, intracranial pressure; PC, principal component; PRx, pressure reactivity.

Table 2.

Multivariable Analysis (PC Regressions, Linear Mixed Effect Model, Backwards Step-Wise Multivariable Regression)

Predictor tested	Outcome variable	t-value	p-Value
ICP model
PC2	Dose of deranged ICP per day	−3.8	<0.001
PC3	Dose of deranged ICP per day	−5.5	<0.001
Day from trauma	Dose of deranged ICP per day	1.26	0.21
PC2*Day from trauma		1.54	0.13
Intervention	Dose of deranged ICP per day	0.88	0.38
PC2*Intervention		−0.95	0.34
PRx model
PC2	Dose of deranged PRx per day	−6.4	<0.001
PC3	Dose of deranged PRx per day	−5.5	<0.001
Day from trauma	Dose of deranged PRx per day	−1.6	0.11
PC2*Day from trauma		4.0	<0.001
Intervention	Dose of deranged PRx per day	−0.19	0.85
PC2*Intervention		3.4	<0.001
CPP model
PC2	Dose of deranged CPP per day	−2.0	0.045
Day from trauma	Dose of deranged CPP per day	1.14	0.26
PC2*Day from trauma		0.55	0.58

All regressions were undertaken using a linear mixed model framework, where individual sample ID (unique for arterial/jugular blood, and per sampling time point) was included as a random effect. In this multivariable regression, variables were excluded in a step-down fashion. Significant interaction effects are written as variable*variable. The dose of “deranged” neuromonitoring metric is defined as follows: ICP > 20 mmHg, PRx > 0, CPP < 60 mmHg. Significant p-values (<0.05) are in bold highlight. For the ICP and PRx models, the starting model included PC2, PC3, day from trauma, and intervention while allowing for interaction effects between PC2 and day from trauma as well as PC2 and intervention. For the CPP model, the starting model included PC2 and day from trauma, while allowing for interaction between the two predictor variables.

CPP, cerebral perfusion pressure; ICP, intracranial pressure; PC, principal component; PRx, pressure reactivity.

Partial least square regression and partial least square discriminant analysis

The protein profile contributing to deranged neuromonitoring-derived metrics was further examined using PLS and PLS-DA. PLS confirmed that there was no benefit in adding four or more principal components (Supplementary Data S7). Following model tuning, PLS regression suggested that n = 2 dimensions should be kept and that a smaller subset of proteins was sufficient to describe the deranged neuromonitoring-derived variables CPP, ICP, and PRx (Fig. 2A). The exact number of proteins returned varies because of parallel processing through cross-fold validation in the PLS model. It is more useful to examine stability (VIP ≥ 0.8) across model iterations, thereby yielding the most stable proteins across all models: Mannan binding lectin serine peptidase 2 (MASP2), MCHR2, complement factor 1QA (C1QA), C9, Ficolin-3 (FCN3), DSCAM, GRM1, BTBD17, STX1A, OPALIN, GRIA2A, and CFI. Specifically, the complement proteins were positively correlated with the brain physiological metrics as shown using a correlation and network analysis (orange/red connecting lines, Fig. 3A–B), where higher complement levels correspond to higher doses of deranged ICP or PRx. Correlations were stronger between proteins and ICP/PRx burden, compared to that of CPP burden (Fig. 3C), and hence only ICP and PRx burden were pursued in PLS-DA (Supplementary Data S7).

FIG. 3.

PLS regression using two components identified a subset of proteins that describe derangements in the neuromonitoring-derived parameters ICP, PRx, and CPP. The correlation circle plot (A) shows the association between the variables in the PLS model: a set of proteins (blue) and deranged physiological parameters (orange), and the model components 1 and 2. This visualisation aids in understanding how the original variables (blue and orange) contribute to the 2 PCs of the final PLS model. Only variables with VIP ≥ 0.8 were included. Text has been displaced minimally to enable reading of the proteins. The network plot of correlations (B) shows that proteins positively (orange/red lines) associated with deranged neuromonitoring indices are proteins largely related to vascular inflammation. The relationship was especially strong for ICP and PRx burden (C). AUC, area under the curve; CPP, cerebral perfusion pressure; ICP, intracranial pressure; PC, principal component; PLS, partial least square regression; PRx, pressure reactivity index; sTBI, severe traumatic brain injury.

For PLS-DA supervised by PRx derangement, the primary proteins driving PC1 were C1QA, MASP2, and MCHR2—these same proteins also drove PC1 in a PLS-DA model supervised by ICP derangement. In the ICP model, additional complement proteins (C1QB, C9, and FCN3) also contributed but with lower stability values (VIP = 0.40, 0.26, and 0.08, respectively).

Further analysis in the “vascular inflammation” protein subset

Correlation matrices were created for the entire cohort, and by treatment group (Fig. 4). Clinical parameters were also examined, with age having the strongest correlation with the vascular inflammatory proteins. The GCS (both total and motor scores), as well as the injury severity scale, had relatively weak correlations with the vascular inflammatory proteins (all coefficients weaker than ±0.5 for entire cohort, Fig. 4A). When all patients were considered, complement proteins C5 and C9 showed strong negative correlations with mean ICP (−0.76 and −0.70 respectively). This relationship remained consistent across the intervention (−0.57 for both, Fig. 4B) and control (−0.82 and −0.89, Fig. 4C) subgroups. When subgrouping our small patient cohort, statistics are limited. Hence, we simply demonstrate that the intervention and control subsets showed different correlations between some other proteins and neuromonitoring-derived parameters, with different directions of correlation in some cases (e.g., C5 has a strong positive correlation with the adjusted dose of PRx > 0 with a coefficient of 0.8 in the control group, compared to −0.64 in the intervention group).

FIG. 4.

Spearman’s correlation between various clinical and neuromonitoring-derived variables and the vascular antibodies. Correlations done for the entire cohort (A), the intervention group (B), and the control group (C). Correlation coefficients are shown as values, where the shading colour represents the strength and direction of correlation, with yellow indicating a positive correlation and blue a negative correlation. n, number of patients in each cohort; GCS, Glasgow Coma Scale; ISS, injury severity scale; ICP, intracranial pressure; ABP, arterial blood pressure; CPP, cerebral perfusion pressure; PRx, pressure reactivity index; % Time, percentage time. Abbreviations of the vascular proteins are available in Supplementary Data S4.

Linear mixed-effect modeling further highlighted the complement cascade. Both MASP2 and complement factor I (CFI) were associated with the dose of deranged pressure reactivity (β coefficient = 0.13 [95% CI 0.02–0.25] and 0.08 [0.0–0.16] respectively), even when patient, monitoring time, and intervention were accounted for (Table 3). Interestingly, ficolin 1 (FCN1) was significantly associated with the percentage time spent with an ICP >20 mmHg (p value = 0.005) but also showed a significant interaction (p value = 0.006) based on treatment group. An interaction plot shows the difference in FCN1 trends by control/intervention in Supplementary Data S8.

Table 3.

Multivariable Regression Analyses Were Done with All Antibodies and Physiological Variables, Where Patient, Time, and Intervention Were Taken into Consideration

Antibody	Predictor:Gene/protein	Outcome:physiological variable	β coefficient:	p-Value
HPA029313	MASP2_1	Adjusted dose PRx >0	0.13 (0.02–0.25)	0.027
HPA029313	MASP2_1	Intervention	−0.01 (−0.16–0.14)	0.9
HPA001143	CFI	Adjusted dose PRx >0	0.08 (0.00–0.16)	0.039
HPA001143	CFI	Intervention	−0.05 (−0.19–0.08)	0.4
HPA001295	FCN1_1	% Time with ICP >20	18 (6.3–30)	0.005
		Intervention	−9.4 (−29–11)	0.3
		Interaction effect with Intervention	−24 (−39–−7.8)	0.006
HPA001295	FCN1_1	% Time with PRx >0	−12 (−24–0.78)	0.065
HPA001295	FCN1_1	Intervention	−16 (−39–5.9)	0.13
HPA071173	FCN3	Adjusted dose PRx >0	−0.15 (−0.30–0.00)	0.05
HPA071173	FCN3	Intervention	−0.04 (−0.18–0.10)	0.5
HPA004138	VEGFC_2	Adjusted dose PRx >0	−0.09 (−0.18–0.01)	0.067
HPA004138	VEGFC_2	Intervention	−0.06 (−0.20–0.08)	0.4

A linear mixed effect model was done, where patient and time were included as random effects, with intervention group included as a fixed effect. An interaction between antibody level and intervention was allowed, with values reported if significant. Only significant associations are shown.

Only significant associations (p-value < 0.05), or those approaching significance (<0.07), are reported. Proteins may be represented by multiple antibodies, antibody specified by ProteinName_#. HPA, human protein atlas (ID number); FCN1, ficolin-1; FCN3, ficolin-3; VEGFC, vascular adhesion molecular epitope 3; MASP2, membrane associated serine protease 2; CFI, complement factor I; ICP, intracranial pressure; PRx, pressure reactivity index.

Discussion

We explored inflammation as a biological mechanism of intracranial physiological derangement, as measured with doses of ICP, CPP, and PRx above thresholds, post-sTBI using advanced statistical techniques. Unsupervised dimensionality reduction identified two PCs that were significantly associated with the daily dose of deranged ICP and PRx. At the protein level, these PCs seemed to be driven by vascular inflammatory proteins. Using supervised dimensionality reduction, we corroborated these findings and identified key vascular inflammatory proteins, several of which were related to the complement cascade. Further investigation using this vascular panel showed notable associations with neuromonitoring-metric derangements, and importantly, significance of FCN1, MASP2, and CFI complement proteins even when patient, time, and intervention are considered. Considering the positive correlations between both proteins MASP2/CFI and the dose of disturbed PRx (coefficient of 0.8 for both relationships, control patients only, Fig. 4C), our results suggest that higher complement levels are associated with a greater dose of disturbed pressure reactivity. Taken together, our work has identified a significant relationship between inflammation relating to the vasculature, particularly of complement origin, and deranged ICP and PRx following sTBI.

Neuroinflammation is triggered immediately following trauma and entails a sequential cellular and humoral response to injury.^31,32 Several of the proteins identified in the post-injury neuroinflammatory cascade have been linked to long-term outcome following sTBI,²⁰ but clinical intervention has yet to show therapeutic benefit.¹⁵ Previously, we have reviewed existing literature and suggested candidate vascular pathways involved in CBF control in acute brain injury. Interestingly, using a broad screen of proteins broadly involved in TBI, we can corroborate our previous systematic review and demonstrate a relationship between specific vascular inflammatory proteins and intracerebral physiology. Systemic inflammation and endothelial dysfunction lead to microglia activation, vascular permeability, BBB disruption, reduced vasodilation, and hence reduced CBF.¹² Complement is activated and persistent in sTBI and has been found intrathecally in human sTBI,^20,33 while its role in TBI pathophysiology is vastly complex, there has been evidence to suggest that complement proteins specifically localize to cerebral blood vessels and brain contusions, supporting a direct link between complement activation and CBF control after sTBI.³⁴ In this study, we have used several overlapping statistical methods to demonstrate the close association between complement and deranged ICP and PRx, although it must be noted that we cannot comment on inflammation specifically derived from the brain versus the influence of systemic inflammation. Considering the prevalence of BBB disruption and that TBI patients often suffer from polytrauma, it is likely that the intracranial inflammatory environment influences, and is influenced by, systemic inflammation. This will be of pivotal importance in future work.

As our analyses were done in the context of the IL1ra drug trial, we have taken treatment into account where possible. Notably, PC1 and 2 of our unsupervised PCA demonstrated a strong effect of treatment, similar to earlier work analyzing cerebral microdialysate from the same cohort.¹⁶ Interestingly, correlation analyses between vascular inflammatory proteins and intracranial metrics demonstrated different associations in the intervention versus control groups (Fig. 4), tentatively sparking interest in the potential effect of targeted treatment: when the immune-modulator IL1ra is administered, associations between vascular inflammatory proteins and neuromonitoring-derived parameters change. This is further supported by ficolin1’s significant interaction effect with treatment in the linear mixed effect modeling. These results flag the (1) potential benefit of pursuing a possible mechanistic link between these two secondary injury cascades, and (2) successful target engagement by the study drug.

Interventional clinical trials of complement modulation are currently underway for acute brain injury. Koopman et al. have recently described safety and pharmacodynamic efficacy of a C5 inhibitor following aneurysmal subarachnoid haemorrhage,³⁵ and a C1 inhibitor is currently being investigated for the same endpoints following sTBI.³⁶ While results are still pending for sTBI, C5 inhibition following aneurysmal subarachnoid hemorrhage has been shown to be acceptably safe under prophylactic anti-infectious treatment. Complement inhibition also seems to be efficient. Across both pathologies, it will be critically important to investigate the relationship between brain physiological derangements and complement modulation, as our analyses suggest that this relationship could be mechanistic in nature and have therapeutic potential.

Limitations

As our analysis required matched physiological and protein data, we have a small cohort of 11 intensively monitored patients. This limited us in our statistical analysis, and we recognize that we could not account for various factors that could confound our results, such as surgical intervention or injury severity (as surgery, and/or a more injured brain, is likely to elicit a greater inflammatory response and greater pressure disturbances). Furthermore, as this study was hypothesis-generating, we did not correct for multiple testing. While we accounted for the effect of intervention as best as possible, we are aware that anakinra is an immunomodulatory drug and is likely to influence our results. Hence, interpretation is done with this in mind.

While we used dose and percentage time metrics to account for both the extent and time of deviation from thresholds, these, as well as summary ICP, CPP, and PRx values, are representative measures of intracranial dynamics, and careful interpretation is required. The protein panel, while selected for CNS enrichment and results from previous literature, contains an over-representation of complement, thereby yielding a possible selection bias. Further, we quantified the proteins in blood sera, and therefore we are unable to distinguish between systemic and brain-derived inflammation.

Conclusion

Inflammation relating to the vasculature, particularly the complement cascade, was shown to be a possible biological mechanism of deranged intracranial dynamics, as characterized by intracranial hypertension and dysfunctional pressure reactivity post-sTBI.

Transparency, Rigor, and Reproducibility Summary

This was a retrospective analysis of samples and data collected as part of a prospective, interventional phase II (open-label) randomized-controlled trial of recombinant human IL1ra (rhIL1ra, anakinra) in sTBI, and hence, as a retrospective analysis, the study itself and the analysis plan were not formally registered. The original study included a total of 20 patients (randomized 1:1), however, a sample size of 11 patients was used based on data availability, we were limited to time-linked proteomic and neuromonitoring data. Patients were randomized as part of the original trial, and we accounted for the effect of intervention where possible, as detailed in the methodology. Key inclusion criteria and neuromonitoring outcomes were based on established standards in the field and data availability. The statistical tests were based on non-parametric data, with missing/non-matched data excluded. Confidence intervals have been reported where relevant, and statistical analysis was performed by both C.A.S. and C.L., with review from all authors. As this was a hypothesis-generating analysis, multiple comparisons were not corrected for. External validation studies are ongoing in an independent cohort. Reasonable requests for data and analytic code used to conduct the analyses presented will be carefully considered by the corresponding author (cas244@cam.ac.uk).

Authors’ Contributions

A.H., E.T., and P.S. conceptualized and designed the study. A.H. ran the original drug trial and subsequently collected patient biological samples and data. S.B. and P.N. did the biological proteomic analysis and quality control. C.A.S., C.L., E.B., and P.S. did the data cleaning and preprocessing. C.L. and C.A.S. did the analysis and visualization, with verification and interpretation by A.H., E.T., and E.N. C.L. and C.A.S. wrote the article, which was subsequently edited and approved by all authors. All authors had full access to the data. C.L. and C.A.S. contributed equally at all stages.

Footnotes

Acknowledgment

The authors thank the Human Protein Atlas project for contributing antibodies to the project.

Author Disclosure Statement

P.S. receives part of the licensing fees for ICM+ software, licensed by Cambridge Enterprise Ltd, University of Cambridge, Cambridge, UK.

Funding Information

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Medical Research Council (Grant no. G1002277 ID98489), National Institute for Health and Care Research Biomedical Research Centre, Cambridge (Neuroscience Theme; Brain Injury and Repair Theme) and the Addenbrooke’s Charitable Trust (Grant no. 900404).

Authors’ support: C.A.S.—Patrick & Margaret Flanagan Skye Cambridge Trust Scholarship; C.L.—the Uppsala County (Region Uppsala), Uppsala University, and Uppsala University Hospital Research Residency Grant; the Uppsala University MedFak Grant, the Swedish Society of Medicine, the Wenner-Gren Foundations; E.B.—the Medical Research Council (grant no.: MR N013433-1) and by the Gates Cambridge Scholarship; E.N.—Brain Research UK; P.S.—nothing to disclose; A.H.—Medical Research Council/Royal College of Surgeons of England Clinical Research Training Fellowship (Grant no. G0802251), MRC Grant (MR/X021882/1) the NIHR Biomedical Research Centre and the NIHR Brain Injury MedTech Co-operative. The views expressed are those of the authors and are not necessarily those of the NIHR, the Department of Health and Social Care or of any of the other funding bodies; E.T.—Karolinska Institutet Research Grants (#2022-01576), The Swedish Society of Medicine (#SLS-985504), The Swedish Brain Foundation (Hjärnfonden, #FO2023-0124) and Region Stockholm Clinical Research Appointment (#FoUI-981490).

Supplemental Material

References

Lazaridis

, Rusin

, Robertson

. Secondary brain injury: Predicting and preventing insults. Neuropharmacology, 2019; 145(Pt B):145–152; doi: 10.1016/j.neuropharm.2018.06.005

Bonds

, Yang

, Hu

, et al. Predicting secondary insults after severe traumatic brain injury. Journal of Trauma and Acute Care Surgery, 2015; 79(1):85–90; doi: 10.1097/TA.0000000000000698

Kochanek

, Clark

RSB

, Ruppel

, et al. Biochemical, cellular, and molecular mechanisms in the evolution of secondary damage after severe traumatic brain injury in infants and children: Lessons learned from the bedside. Pediatr Crit Care Med, 2000; 1(1):4–19; doi: 10.1097/00130478-200007000-00003

Marmarou

, Anderson

, Ward

, et al. Impact of ICP instability and hypotension on outcome in patients with severe head trauma. J Neurosurg, 1991; 75(Supplement):S59–S66; doi: 10.3171/sup.1991.75.1s.0s59

Stocchetti

, Colombo

, Ortolano

, et al. Time course of intracranial hypertension after traumatic brain injury. J Neurotrauma, 2007; 24(8):1339–1346; doi: 10.1089/neu.2007.0300

Rangel-Castilla

, Gasco

, Nauta

HJW

, et al. Cerebral pressure autoregulation in traumatic brain injury. Neurosurg Focus, 2008; 25(4):E7; doi: 10.3171/FOC.2008.25.10.E7

Miller

, Stanek

, Langfitt

. Concepts of cerebral perfusion pressure and vascular compression during intracranial hypertension. Prog Brain Res, 1972; 35:411–432; doi: 10.1016/S0079-6123(08)60102-8

Bouma

, Muizelaar

, Choi

, et al. Cerebral circulation and metabolism after severe traumatic brain injury: The elusive role of ischemia. J Neurosurg, 1991; 75(5):685–693; doi: 10.3171/jns.1991.75.5.0685

Czosnyka

, Smielewski

, Kirkpatrick

, et al. Continuous assessment of the cerebral vasomotor reactivity in head injury. Neurosurgery, 1997; 41(1):11; doi: 10.1097/00006123-199707000-00005

10.

Sorrentino

, Diedler

, Kasprowicz

, et al. Critical thresholds for cerebrovascular reactivity after traumatic brain injury. Neurocrit Care, 2012; 16(2):258–266; doi: 10.1007/s12028-011-9630-8

11.

Le Roux

, Menon

, Citerio

, et al. The international multidisciplinary consensus conference on multimodality monitoring in neurocritical care: A list of recommendations and additional conclusions: A statement for healthcare professionals from the neurocritical care society and the European society of intensive care medicine. Neurocrit Care, 2014; 21(Suppl 2):S282–S296; doi: 10.1007/s12028-014-0077-6

12.

Smith

, Carpenter

, Hutchinson

, et al. Candidate neuroinflammatory markers of cerebral autoregulation dysfunction in human acute brain injury. J Cereb Blood Flow Metab, 2023; 43(8):1237–1253; doi: 10.1177/0271678X231171991

13.

Helmy

, Carpenter

, Menon

, et al. The cytokine response to human traumatic brain injury: Temporal profiles and evidence for cerebral parenchymal production. J Cereb Blood Flow Metab, 2011; 31(2):658–670; doi: 10.1038/jcbfm.2010.142

14.

Samanta

, Chiollaz

A-C

, Needham

, et al.; CENTER-TBI investigators and participants. Parsimonious immune-response endotypes and global outcome in patients with traumatic brain injury. EBioMedicine, 2024; 108:105310; doi: 10.1016/j.ebiom.2024.105310

15.

Simon

, McGeachy

, Bayır

, et al. The far-reaching scope of neuroinflammation after traumatic brain injury. Nat Rev Neurol, 2017; 13(9):572; doi: 10.1038/nrneurol.2017.13

16.

Helmy

, Guilfoyle

, Carpenter

KLH

, et al. Recombinant human interleukin-1 receptor antagonist in severe traumatic brain injury: A phase II randomized control trial. J Cereb Blood Flow Metab, 2014; 34(5):845–851; doi: 10.1038/jcbfm.2014.23

17.

Sjöstedt

, Fagerberg

, Hallström

, et al. Defining the human brain proteome using transcriptomics and antibody-based profiling with a focus on the cerebral cortex. PLoS One, 2015; 10(6):e0130028; doi: 10.1371/journal.pone.0130028

18.

Sjöstedt

, Zhong

, Fagerberg

, et al. An atlas of the protein-coding genes in the human, pig, and mouse brain. Science, 2020; 367(6482):eaay5947.

19.

Zeiler

, Thelin

, Czosnyka

, et al. Cerebrospinal fluid and microdialysis cytokines in severe traumatic brain injury: A scoping systematic review. Front Neurol, 2017; 8(JUL):331; doi: 10.3389/fneur.2017.00331

20.

Lindblad

, Pin

, Just

, et al. Fluid proteomics of CSF and serum reveal important neuroinflammatory proteins in blood–brain barrier disruption and outcome prediction following severe traumatic brain injury: A prospective, observational study. Crit Care, 2021; 25(1):103; doi: 10.1186/s13054-021-03503-x

21.

Bellander

, Singhrao

, Ohlsson

, et al. Complement activation in the human brain after traumatic head injury. J Neurotrauma, 2001; 18(12):1295–1311; doi: 10.1089/08977150152725605

22.

Hanrieder

, Wetterhall

, Enblad

, et al. Temporally resolved differential proteomic analysis of human ventricular CSF for monitoring traumatic brain injury biomarker candidates. J Neurosci Methods, 2009; 177(2):469–478; doi: 10.1016/j.jneumeth.2008.10.038

23.

Harish

, Mahadevan

, Pruthi

, et al. Characterization of traumatic brain injury in human brains reveals distinct cellular and molecular changes in contusion and pericontusion. J Neurochem, 2015; 134(1):156–172; doi: 10.1111/jnc.13082

24.

Salim

, Hadjizacharia

, Brown

, et al. Significance of troponin elevation after severe traumatic brain injury. J Trauma, 2008; 64(1):46–52; doi: 10.1097/TA.0b013e31815eb15a

25.

Schwenk

, Gry

, Rimini

, et al. Antibody suspension bead arrays within serum proteomics. J Proteome Res, 2008; 7(8):3168–3179; doi: 10.1021/pr700890b

26.

Carney

, Totten

, O’Reilly

, et al. Guidelines for the management of severe traumatic brain injury, fourth edition. Neurosurgery, 2017; 80(1):6–15; doi: 10.1227/NEU.0000000000001432

27.

Czosnyka

, Smielewski

, Kirkpatrick

, et al. Continuous monitoring of cerebrovascular pressure-reactivity in head injury. Acta Neurochir (Suppl 1998), 1998; 71:74–77; doi: 10.1007/978-3-7091-6475-4_23

28.

Sheth

, Stein

, Aarabi

, et al. Intracranial pressure dose and outcome in traumatic brain injury. Neurocrit Care, 2013; 18(1):26–32; doi: 10.1007/s12028-012-9780-3

29.

Kahraman

, Dutton

, Hu

, et al. Automated measurement of “pressure times time dose” of intracranial hypertension best predicts outcome after severe traumatic brain injury. Journal of Trauma - Injury, Infection and Critical Care, 2010; 69(1); doi: 10.1097/TA.0b013e3181c99853

30.

Rohart

, Gautier

, Singh

, et al. mixOmics: An R package for ‘omics feature selection and multiple data integration. PLoS Comput Biol, 2017; 13(11):e1005752; doi: 10.1371/journal.pcbi.1005752

31.

Gadani

, Walsh

, Smirnov

, et al. The glia-derived alarmin IL-33 orchestrates the immune response and promotes recovery following CNS injury. Neuron, 2015; 85(4):703–709; doi: 10.1016/j.neuron.2015.01.013

32.

Alam

, Hana

, Jin

, et al. Surgery, neuroinflammation and cognitive impairment. EBioMedicine, 2018; 37:547–556; doi: 10.1016/j.ebiom.2018.10.021

33.

Stahel

, Trentz

, Kossmann

, et al. Intrathecal levels of complement-derived soluble membrane attack complex (sc5b-9) correlate with blood-brain barrier dysfunction in patients with traumatic brain injury. J Neurotrauma, 2001; 18(8):773–781; doi: 10.1089/089771501316919139

34.

De Blasio

, Fumagalli

, Orsini

, et al. Human brain trauma severity is associated with lectin complement pathway activation. J Cereb Blood Flow Metab, 2019; 39(5):794–807; doi: 10.1177/0271678X18758881

35.

Koopman

, Tack

RWP

, Wunderink

, et al. Safety and pharmacodynamic efficacy of eculizumab in aneurysmal subarachnoid hemorrhage (CLASH): A phase 2a randomized clinical trial. Eur Stroke J, 2023; 8(4):1097–1106; doi: 10.1177/23969873231194123

36.

van Erp

IAM

, van Essen

, Fluiter

, et al. Safety and efficacy of C1-inhibitor in traumatic brain injury (CIAO@TBI): study protocol for a randomized, placebo-controlled, multi-center trial. Trials, 2021; 22(1):874; doi: 10.1186/s13063-021-05833-1

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.94 MB