Blood-Based Protein Biomarkers in the Chronic Phase of Traumatic Brain Injury: A Systematic Review

Aims

Assessment of risk of bias, data extraction, and data synthesis

This systematic review is reported according to the Preferred Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. ⁴⁸ Prior to the commencement of this review, a protocol was published to the PROSPERO database (CRD42023435535). There were six deviations from the protocol (Supplementary Data S1).

This systematic review was conducted in parallel with an additional systematic review that examined blood-based protein biomarkers of neurovascular injury at any time after TBI. The search strategy was optimized for both reviews, and the study selection was completed in tandem at both the title and abstract and full-text review stage (i.e., the reviewers assessed each article simultaneously for eligibility in both reviews). After full-text review, the studies were allocated to their topic area (chronic phase TBI or neurovascular injury).

Information sources and search strategy

We searched English literature from Embase and MEDLINE via Ovid SP and Science Citation Index–Expanded via Web of Science on July 24, 2023. There were no date/time, document type, or publication status restrictions. The search strategy was developed based on elements of the PICO (population: brain injury; outcome: blood biomarkers) and study type and included a range of controlled vocabulary (Medical Subject Headings = MeSH and Excerpta Medica Tree = Emtree) and keywords linked by Boolean operators. Keywords were collected through literature review, authors’ expert opinions, and review of the primary search results. The search strategies were developed with the assistance of a medical information scientist and underwent seven rounds of revisions and testing. The MEDLINE search strategy was peer-reviewed by the medical information scientist using the Peer-Review of Search Strategies Checklist (PRESS), ⁴⁹ before translating the strategy to other databases (Supplementary Data S2). The reference lists of previous systematic reviews and included studies were checked for additional studies that may have been missed by the database search. Google Scholar article alerts were used to monitor new publications from July 2023 to review submissions (June 2024; no new publications were identified). This systematic review was last assessed as up-to-date in July 2023.

Study selection

Throughout the review selection process and risk of bias assessment, reviewers were not blinded to journal titles, study authors, or their institutions. All study screening was completed independently and in duplicate by two pairs of reviewers (A.H., L.B., H.C., R.P., A.Y., B.Y.). Disagreements were resolved through consensus, and if required, a third team member adjudicated (A.H., L.B., H.C., R.P., A.Y., B.Y.).

Citations identified by database search were first uploaded to EndNote for deduplication. Unique citations were uploaded to Covidence for screening. Titles and abstracts were screened against inclusion criteria. Studies that potentially met inclusion criteria were retrieved in full and assessed against the inclusion criteria. The full length of potentially eligible records was retrieved using institutional subscriptions, loan requests, or open sources. Full-text studies that did not meet inclusion criteria were excluded.

Eligibility criteria

Search results

Database searching identified 12,523 records (Fig. 1). Search results were deduplicated in EndNote X9. Following title and abstract screening, 11,845 were deemed irrelevant and removed. Full texts were obtained for 633 articles. There were 440 articles excluded, most commonly because the participants were not in the chronic post-injury period (≥12 months post-injury; n = 198). No studies were excluded due to publication in a language other than English. We considered 40 articles excluded to be “near miss” articles: Studies that appear to meet the inclusion criteria but which were subsequently excluded. ⁴⁸ The most common reason for this was that participants were not restricted to only greater than 12 months post-injury, and data collected from those greater than 12 months post-injury were not reported separately. For several studies, the lower bound for time since injury could not be discerned, and the authors could not provide this information (n = 26 authors contacted). Supplementary Data S3 provides the citation and rationale for the exclusion of each “near miss” article. Searching the citation lists of previous reviews (n = 5) and included studies identified a further 126 records to screen of which 2 were included. Overall, the database search and other sources identified 30 studies included in this review (Supplementary Data S6 has the full list of citations).

OPEN IN VIEWER

FIG. 1.

PRISMA flow diagram showing each stage of the screening process for study inclusion. Note. We have separated the studies identified from other sources by using “+.” Supplementary Data S3 provides the citations for the three studies that could not be retrieved. “Chronic phase TBI” is defined as ≥12 months post-injury. Reasons for “no useable outcome data” included intervention studies with no pretreatment analyses, reporting of protein levels only with no analysis relevant to our aims. The full-text review identified (n = 193) studies that were eligible for inclusion in one of the two systematic reviews (screening was completed concurrently for two systematic reviews). Thirty studies were then allocated to the current systematic review. ABI, acquired brain injury; PRISMA, Preferred Items for Systematic Reviews and Meta-Analyses; RHI, repetitive head impacts; TBI, traumatic brain injury.

There were two studies by Shahim and colleagues published in 2020 ^53,54 that used the same sample and methodology. Shahim 2020a ⁵³ reports on four proteins (NfL, tau, UCHL-1, and GFAP), and Shahim 2020b ⁵⁴ reports NfL data only. Per the study authors’ statements in Shahim 2020a, the NfL data in Shahim 2020b are reproduced in Shahim 2020a. As such, we have combined these studies in Table 1 (Study Characteristics). We have used Shahim 2020a as our reference document for the evidence tables and narrative synthesis. We have provided a separate risk of bias analysis rationale for each study (the overall risk of bias score for each study was the same).

There were a further n = 7 studies that analyzed data from the same parent studies (Trondheim Mild TBI Study ^57,58 ; Chronic Effects of Neurotrauma Consortium Longitudinal Study ^80,82 ; 15 Year Longitudinal TBI Study ^64,70,71 ). Although there was likely some sample overlap, all studies used either different proteins or conducted different analyses and were considered separate studies for our review.

General characteristics of the included studies

There were 30 studies included in the review. However, two studies were deemed to have used the same data (see above), giving a final sample of n = 29 studies (Table 1; extended version of Table 1 is in Supplementary Data S7). Studies were published between 2010 and 2023, with the majority published after 2020 (n = 23, 79%). Half of the studies were published in the USA (n = 14), with the remaining half published in Europe (n = 14) and Australia (n = 1).

Most studies employed at least one control group for their analyses (n = 23, 79%), with two studies using two control groups (non-TBI civilian group + non-TBI veteran group; ⁶¹ non-TBI healthy group + non-TBI injured group ⁷¹ ). Most cohorts comprised civilian populations (n = 19, 65%), with n = 8 (28%) from the military and only n = 2 (7%) studies using sports athletes (rugby union ⁷⁵ and flat-track jockeys ⁷² ). Population sizes for participants with TBI who completed a blood draw ranged from n = 8 to n = 298 (median [Md]: 39; interquartile range [IQR]: 55.5), with the majority having fewer than n = 100 participants with TBI (n = 23, 79%). The majority of TBI participants in each study were male, with four studies having only male participants. ^59,61,75,77 Most studies did not report participant race (n = 19), nine studies had mostly White participants and one study had only White participants ⁸³ (Supplementary Data S7). One-third of studies used samples of mTBI only (n = 10), and a further third focused only on moderate-severe (n = 5) or severe TBI (n = 5). The remaining studies included the full spectrum from mild to severe TBI (n = 8), with one study broadly looking at “TBI with loss of consciousness” without further severity specification. Only 12 studies reported the number of lifetime exposures to TBI.

We extracted cross-sectional data from 28 studies. In 11 (39%) of these studies, the time post-injury was exactly 12 months. The remaining 17 studies provided data from greater than 12 months post-injury: Three studies provided a range with no upper bound (≥12 months, n = 2; >12years, n = 1); one study provided cross-sectional data at 1, 2, 3, 4 and 5 years post-injury; three studies provided range (R) of time since injury only (R: 2–5 years; R: 6–13 years; R: 7–20 years); for the remaining 10 studies the average time post-injury ranged from 3.9 to 32 years (Md: 7.3; IQR: 4.4).

We extracted longitudinal data from seven studies (24%). There were only two studies that collected longitudinal data beyond 12 months post-injury; up to 2 ⁷⁸ and 5 years. ⁵³ For the remaining studies the longitudinal data points were before 12 months post-injury (n = 5).

Regarding the blood compartment in which the studies were performed, 22 were done in serum, 7 in plasma, and 2 used EVs (n = 4 studies conducted analyses in several blood compartments).

A total of 36 proteins were examined across studies. Protein variants and ratios of proteins were also examined in several studies (Table 1). The most data were provided for GFAP, NfL, total tau, and UCHL-1. Across the n = 29 studies, data were provided for all five aims.

Risk of bias assessment for the included studies

Although 16 studies (53%) were scored an overall low risk of bias, 13 were rated an intermediate risk of bias (43%), and one study was rated a high risk of bias (Table 1; justification for scoring Supplementary Data S8). Across studies, there were several common issues worth noting. First, the TBI and control group participants were matched in only seven studies. However, this was usually done for the baseline sample (i.e., participants were not rematched at the chronic stage (≥12 months post-injury), and it was not clear if the participants who provided data at ≥12 months post-injury remained matched. Likewise, although most studies did compare their groups on some key demographic details, it was unclear if the subsamples who provided data at ≥12 months post-injury remained demographically similar. Second, covariates were often not controlled for in analyses, including group comparisons of TBI vs control cohorts. Third, the sample sizes for some analyses, especially those using subsamples stratified by injury severity, were very small (n ≤15)—suggesting these analyses were likely underpowered. Fourth, for seven studies, control data were only collected at baseline and used as a comparator for all time points. This approach does not optimize the use of a control group to account for natural variation over time. Finally, the eligibility criteria and participant characteristics were often lacking in detail which reduces generalizability and comparison with other studies.

Evidence synthesis

Tables 2 and 3 provide the evidence for Aims 1 and 2, respectively. The evidence tables are organized in three hierarchical levels: pathophysiological class (neurodegenerative, astroglial damage, inflammatory, vascular, other), time since injury (earliest to latest), and injury severity (mild to severe). The assignment of each protein to a pathophysiological class was completed by the research team, and it is acknowledged that some proteins may be considered under more than one class. The evidence tables for Aims 3–5 are provided in Supplementary Data S5.

Within the narrative synthesis below, we use the subheadings “Key Findings” and “Additional Findings,” with “Key Findings” being those from studies with low risk of bias and large samples (n > 100 in TBI group with blood draw at 12 months or more post-injury). Proteins included in “Key Findings” are italicized for emphasis. The key findings are summarized in Table 4.

Structural MRI

For GFAP, NfL, total tau, and UCHL-1, 28 analyses were conducted for each protein in one study with intermediate risk of bias in mTBI+moderate-severe TBI (1, 2, 3, 4, 5 years postinjury ⁵³ ). There were no significant findings from these analyses for GFAP, UCHL-1, or total tau, and three significant findings for NfL. NfL measured at 12 months was associated with volume loss in the central and mid-anterior regions of the corpus callosum at 1–2 years postinjury ⁵³ ; NfL measured at 3 years was associated with volume loss in the central region of the corpus callosum at 3–4 years post-injury. ⁵³ One study with a small moderate-severe TBI sample (n = 8) contributed significant findings for p-tau and p-tau/t-tau ratio: both were associated with change in brainstem shape from 17.53 to 23.95 months. ⁷⁸

Diffusion MRI

Two studies of mTBI+moderate-severe TBI provided diffusion MRI data (n = 48 analyses of GFAP, NfL, total tau, UCHL-1 ⁵³ ; n = 135 analyses of GFAP, NFL ⁷⁴ ). From these analyses, there were 6 significant findings (GFAP n = 1; NfL n = 3; total tau n = 2), with no significant findings for UCHL-1. GFAP measured at 3 years was associated with a decrease in fractional anisotropy in the genu region of the corpus callosum from 3 to 4 years. ⁵³ NfL measured at 3 years was associated with a decrease in fractional anisotropy and an increase in radial diffusivity in the genu of the corpus callosum. ⁵³ At M: 8.3 years, NfL was associated with increased mean diffusivity for whole brain white matter. ⁷⁴ Tau levels at 12 months were associated with an increase in mean diffusivity at 1–2 years post-injury, and tau levels at 4 years were associated with a decrease in fractional anisotropy at 4–5 years post-injury. ⁵³

PET

One study with a small mTBI sample (n = 10, R: 7–20 years post-injury) contributed significant findings: higher NfL in those who were tau PET ([¹⁸F]AV1451) positive. ⁵⁹

EEG

One study with a small moderate-severe TBI sample (n = 8) contributed nonsignificant findings for p-tau and p-tau/t-tau ratio: both were not associated with change in resting state EEG from 17.53 to 23.95 months. ⁷⁸

Demographic variables

The demographic variables examined were age, sex, education, genotype, medical health, and medication. There are no “Key Findings” for demographics as contributing studies with low risk of bias did not have large sample sizes (n <100 for all TBI analysis groups).

Medical health

The association between protein level and medical health was examined in a single study. This low risk of bias study (moderate-severe TBI; M: 3.9 years post-injury) reported IGF-1 (insulin-like growth factor 1) levels were not associated with growth hormone deficiency status. ⁷⁶ Small sample sizes (n = 9–11) in subgroups may have precluded detection of associations due to low statistical power.

Medication

In the single intermediate risk of bias study (moderate-severe TBI; 12 months postinjury ⁸³ ) BDNF was not associated with antidepressant use. Small sample sizes for subgroups (no antidepressant use, n = 10) may have impacted results.

Injury variables

The injury variables examined were injury severity, time since injury, age at injury, and number of TBIs. Key Findings are presented for all injury variables with the exception of age at injury.

Injury severity

Injury severity was the most studied injury characteristic, with data from seven studies covering BDNF, NSE, GFAP, total tau, and UCHL-1.

Key Findings: There were significant findings for GFAP (higher GFAP in moderate-severe vs mild TBI), and nonsignificant findings for NfL, total tau, and UCHL-1 at M: 7.1–7.5 years postinjury ⁷¹ .

Additional Findings: For GFAP, three additional studies contributed 22 analyses across 12 months to Md: 5 years post-injury. ^53,60,64 There was only one additional significant finding: higher GFAP in severe TBI compared to mTBI (12 months post-injury; low risk of bias ⁶⁰ ). There were no other significant findings for NfL (mTBI + moderate-severe TBI 12 months-M: 7.1–7.5 years; low risk of bias, ^60,71 and intermediate risk of bias ⁵³ ), total tau (mTBI + moderate-severe TBI 12 months-M: 7.1–7.5 years; low risk of bias ⁷¹ and intermediate risk of bias ^53,69 ), and UCHL-1 (mTBI + moderate-severe TBI 12 months-M: 7.1–7.5 years; low risk of bias ⁷¹ and intermediate risk of bias ⁵³ ). There were also no significant associations between injury severity and BDNF (moderate-severe TBI; 12 months post-injury; intermediate risk of bias ⁸³ ) and NSE (severe TBI; 12 months+; low risk of bias but small samples ⁵⁶ ).

Time since injury

Four studies examined the association of time since injury with GFAP, IL-6, NfL, NSE, S100B, and total tau.

Key Findings: There was a significant increase in NfL with time post-injury but not total tau (mTBI, M: 8.3 years R: 2–15 postinjury ⁸⁰ ).

For NfL, 2 additional studies contributed 4 analyses across <1 month to M: 9.8 years post-injury. The one low risk of bias study found no significant association with time since injury (mTBI, Md: 5.5 years post-injury, low risk of bias ⁶⁴ ; cmTBI+moderate-severe TBI; M: 9.8 years postinjury ⁶⁴ ). A significant finding was reported when comparing 2–12 months vs 12 months+ (mTBI, intermediate risk of bias and small sample ⁷⁷ ).

There were no significant associations between time since injury and GFAP (mTBI, 5.5 years post-injury, low risk of bias ⁶⁴ ; complicated mTBI + moderate-severe TBI; M: 9.8 years postinjury ⁶⁴ ), IL-6 (mTBI, <1 month to 12 months+, intermediate risk of bias and small sample ⁷⁷ ), NSE (mTBI, <1 month to 12 months+, intermediate risk of bias and small sample ⁷⁷ ; severe TBI, <12 months to 12 months+, low risk of bias ⁵⁶ ), or S100B (mTBI, <1 month to 12 months+, intermediate risk of bias and small sample ⁷⁷ ).

Age at injury

There was no association of age at injury with GFAP or NfL at 12 months post-moderate-severe TBI (low risk of bias ⁶³ ).

Number of TBI

Key Findings: Greater number of TBIs was associated with higher NfL, but not total tau (mTBI, M: 8.3 years R: 2–15 post-injury ⁸⁰ ).

Two additional low risk of bias studies examined the association of number of TBI with IL-6, IL-10, NfL, and TNF-α: all reporting nonsignificant findings (NfL: mTBI, Md 5.5 years ⁶⁴ ; IL-6, IL-10, TFN-α:mTBI, M: 9 years). ⁸²

Aim 5. To determine whether there is an association of blood-based protein biomarker concentration with any clinical outcomes

Fourteen studies provided data for Aim 5 examining cognitive, neurobehavioral and emotional, functional, sleep, and quality of life outcomes; 7 with low risk of bias ^{60,64,68,71,76,80,82} ; and 7 with intermediate risk of bias. ^{53,59,62,66,74,78,83} Eighteen proteins were examined for Aim 5: amyloid-β₄₂, BDNF, bd-tau, GFAP, IGF-I, IL-1β (interleukin-1 beta), IL-4 (interleukin-4), IL-5 (interleukin-5), IL-6, IL-7 (interleukin-7), IL-8, IL-10, IL-12, NfL, p-tau, TNF-α, total tau, and UCHL-1. Key Findings are only presented for sleep and cognitive outcomes. For Aim 5 data summarized by protein, see Supplementary Data S5.

Cognitive outcomes

Five studies examined the association of cognitive outcomes with BDNF, IGF-I, GFAP, NfL, total tau, UCHL-1, and p-tau.

Key Findings: For GFAP and UCHL-1, there were mixed findings dependent on TBI severity and the cognitive measure: no associations for mTBI (M: 7.5 years post-injury; low risk of bias ⁷¹ ) but in the moderate-severe TBI group increased GFAP was associated with worse perceptual reasoning scores, and increased UCHL-1 was associated with worse immediate and delayed memory (M: 7.1 years post-injury, low risk of bias ⁷¹ ). There were no associations with NfL or total tau from this study.

Additional Findings: The nonsignificant findings for NfL were supported by another study with intermediate risk of bias (mTBI, R: 7–20 years, small sample). ⁵⁹ Nonsignificant findings were also reported for cognitive outcomes and BDNF (moderate-severe TBI, 12 months post-injury, intermediate risk of bias ⁸³ ) and IGF-I (moderate-severe TBI, M: 3.9 years post-injury, low risk of bias). ⁷⁶ For p-tau there were mixed findings dependent on the cognitive measure: increases in p-tau were counterintuitively associated with improvement in RBANS (Repeatable Battery for the Assessment of Neuropsychology Status) Digit Span (M: 17.52–23.95 months) but not two other RBANS subtests or the MoCA (Montreal Cognitive Assessment; severe TBI, intermediate risk of bias, small sample ⁷⁸ ). In contrast, when using the ratio of p-tau/t-tau, there were no significant associations with cognition (severe TBI, M: 17.52–23.95 months, intermediate risk of bias, small sample ⁷⁸ ).

Emotional and neurobehavioral outcomes

Six studies examined the association of neurobehavioral outcomes with BDNF, GFAP, NfL, TNF-α, and p-tau.

Neurobehavioral function

There were no significant associations with neurobehavioral function and GFAP, NfL, or TNF-α. There were no associations of neurobehavioral function and GFAP or NfL (mTBI+moderate-severe TBI, 12 months; 12 months to 2–5 years, low risk of bias ⁶⁰ ; mTBI, 5.5 years post-injury, low risk of bias ⁶⁴ ; complicated mTBI + moderate-severe TBI; M: 9.8 years post-injury, low risk of bias ⁶⁴ ). There were no associations between disinhibition and TNF-α (severe TBI, 12 months, intermediate risk of bias, small sample ⁶⁶ ).

Depression

Depression was examined with NfL, p-tau, and BDNF. Higher p-tau was associated with less severe depression 6 months later (severe TBI, intermediate risk of bias, small sample ⁷⁸ ). There were mixed findings for the association of depression with BDNF depending on analysis type (moderate-severe TBI, 12 months post-injury, intermediate risk of bias ⁸³ ). BDNF level was positively correlated with depression score on the PHQ-9, but there were no significant differences in BDNF levels for those categorized as post-traumatic depression (PTD) vs those without PTD (PTD was defined as a positive endorsement of =>5 symptoms on the PHQ-9 with ≥1 cardinal symptom—depression or anhedonia). There were no associations of depression with NfL (mTBI, R: 7–20 years, intermediate risk of bias and small sample ⁵⁹ ) and p-tau/t-tau ratio (M: 23.95 months; severe TBI, intermediate risk of bias, small sample ⁷⁸ ).

Suicidality

There were no associations between suicidality and TNF-α (severe TBI, 12 months, intermediate risk of bias, small sample ⁶⁶ ).

PTSD

There were no associations of PTSD with GFAP or NfL (mTBI, 5.5 years post-injury, low risk of bias ⁶⁴ ; complicated mTBI + moderate-severe TBI, M: 9.8 years post-injury, low risk of bias ⁶⁴ ).

Functional outcomes

Six studies examined the association of functional outcomes with bd-tau, GFAP, IL-1β, IL-4, IL-5, IL-6, IL-8, IL-10, IL-12, NfL, p-tau-231 TNF-α, total tau, and UCHL-1. All studies used the GOS/GOSE (Glasgow Outcome Scale/Glasgow Outcome Scale–Extended), except for one using Disability Rating Scale. ⁷⁸ There were significant associations for IL-6 and p-tau, and significant but mixed findings for GFAP and NfL. Those with unfavorable GOS scores had higher IL-6 levels (severe TBI, 12 months, low risk of bias ⁶⁸ ). Higher p-tau and p-tau/total tau ratio was associated with improvement in functional outcomes (M: 17.52–23.95 months; severe TBI, intermediate risk of bias, small sample ⁷⁸ ).

For GFAP, one low risk of bias study found higher GFAP at 12 months was associated with lower GOSE at 12 months and improvement in GOSE from 12 months to 2–5 years (mTBI + moderate-severe TBI ⁶⁰ ). This was in contrast to an intermediate risk of bias study that found no association between GFAP at 12 months and changes in GOSE at 12 months to 2 years (mTBI + moderate-severe TBI ⁵³ ). After 12 months post-injury, there were no further significant associations for GFAP and change in GOSE for mTBI + moderate-severe TBI from 2 years to M: 8.3 years post-injury (intermediate risk of bias ^53,74 ).

For NfL, one low risk of bias study found higher NfL at 12 months was associated with lower GOSE at 12 months but no improvement in GOSE from 12 months to 2–5 years (mTBI+moderate-severe TBI ⁶⁰ ). In addition, an intermediate risk of bias study found that higher NfL at M: 8.3 years was associated with worsening GOSE from approximately 8 months to M: 8.3 years post-injury (mTBI + moderate-severe TBI ⁷⁴ ). All other analyses were nonsignificant both at 12 months post-injury (mTBI + moderate-severe TBI, intermediate risk of bias ⁵³ ; severe TBI, intermediate risk of bias and small sample ⁶² ) and up to 5 years post-injury (mTBI + moderate-severe TBI, intermediate risk of bias ⁵³ ).

There were no associations between functional outcomes and bd-tau; p-tau-231 (severe TBI, 12 months post-injury, intermediate risk of bias and small sample ⁶² ); IL-1β; IL-4; IL-5; IL-8; IL-10; IL-12; TNF-α; IL-1β/IL-10 ratio; IL-6/IL-10 ratio; IL-8/IL-10 (severe TBI, 12 months, low risk of bias ⁶⁸ ); total tau (severe TBI, 12 months post-injury, intermediate risk of bias and small sample ⁶² ; 12 months to 5 years post-injury, mTBI + moderate-severe TBI, intermediate risk of bias ⁵³ ); or UCHL-1 (12 months to 5 years post-injury, mTBI + moderate-severe TBI, intermediate risk of bias ⁵³ ).

Sleep outcomes

Key Findings: Two low risk of bias studies of mTBI M: 8.3–9 years post-injury contributed several analyses examining sleep outcomes with amyloid-β42, IL-6, IL-10, TNF-α, and total tau. ^80,82 There were no associations with sleep outcomes and amyloid-β₄₂. ⁸⁰ There were significant but mixed findings for all other proteins.

Quality of life outcomes

There were no associations between quality of life and GFAP or NfL (mTBI + moderate-severe TBI, 12 months; 12 months to 2–5 years, low risk of bias ⁶⁰ ).

Protein level in TBI as compared to non-TBI control groups

For mild TBI, the most robust findings were for elevation in inflammatory markers (eotaxin-1, IFN-y, IL-8, IL-9, IL-17A, MCP-1, MIP-1β, FGF-basic, and TNF-α) at 12 months post-injury, ⁵⁷ with some evidence of long-term elevations in IL-6 at an average of 9 years post-injury. ⁸² Less robust mTBI studies also suggest that CRP and MME may be promising markers. ⁶¹ For moderate-severe TBI, the most robust significant findings were for elevations in GFAP at 12 months post-injury, ^53,60,63 with some evidence of more chronic elevations provided at an average of 7.1 years post-injury. ⁷¹ Studies with low risk of bias but smaller samples confirmed elevations in GFAP 12 months post moderate-severe TBI ⁶³ and also suggest elevations can be found in some mild TBI samples. ^53,60 Elevated GFAP is considered an indicator of astrocytic cell impairment following TBI, ^{84 –86} and may indicate disruptions to the blood–brain barrier (BBB). ⁶³ Other promising markers from these smaller studies included NSE in severe TBI, ⁵⁶ and several protein variants for α-syn (10H, D5), amyloid-β (A4, C6T), TDP-43 (AD-TDP 1;2;3;9;11), and tau (D11C) in a very long-term cohort (M: 29.4 years) that included mTBI and moderate-severe TBI. ⁸¹ NfL, one of the most studied proteins, also had some notable significant findings albeit from less robust studies. For moderate-severe TBI at 12 months postinjury, there was inconclusive evidence for NfL mostly drawn from smaller studies or those with intermediate risk of bias. ^53,60,63,79 Elevated NfL is thought to reflect axonal injury and degeneration. Alterations in NfL levels in the chronic post-injury period may be facilitated by ongoing release from degenerating axons, a process that is known to persist for years after injury. ^84,85

Neuroinflammation is a key component of TBI-related pathology; however, it remains poorly characterized in the chronic period. Our data synthesis shows there are several inflammatory markers elevated at 12 months post-injury when compared to controls (eotaxin-1, IFN-y, IL-8, IL-9, IL-17A, MCP-1, MIP-1B, FGF-basic, and TNF-,α ⁵⁷ ) with elevations in some proteins persisting to 5 years (CRP ⁶¹ ) and 9 years post-injury (IL-6 ⁸² ). Evidence of ongoing neuroinflammation aligns with evidence from PET ^86,87 and postmortem ^84,85,88 studies of TBI survivors in the chronic post-injury period. Chronic neuroinflammation may be the result of self-perpetuating acute inflammatory processes ⁸⁹ and is thought to be a key factor linking TBI with neurodegenerative disease. ⁹⁰

The group differences in NSE identified at 12 months were counterintuitive: levels were higher in non-TBI controls. ⁵⁶ This finding is surprising, as NSE is released when neurons are injured, and NSE levels are higher in TBI survivors in the acute post-injury period. ^{91 –94} This raises the intriguing possibility that later reductions in NSE may reflect a distinct chronic phase TBI-related neuropathological process that emerges after the acute period. ⁵⁶ This may be seeded by early NSE activity causing a cascade of other biological processes that ultimately feed forward to a reduction in NSE. Alternatively, novel pathological processes could be established when post-TBI sequelae such as sleep dysfunction and comorbid medical conditions influence protein dysregulation.

The weight of higher quality evidence indicated nonsignificant group differences in UCHL-1, amyloid-β₄₂, and total tau. However, when select neurodegenerative-disease protein variants of amyloid-β₄₂ and total tau, as well as α-syn, and TDP-43, were examined on average almost 30 years post-injury, there were significant group differences. ⁸¹ This suggests neurodegenerative-disease protein variant assays may be more sensitive to chronic phase TBI-related pathologies. Replication of findings is required as samples were small and not matched (TBI n = 25, non-TBI control n = 18), and analyses were not adjusted for age.

There may be value in stepping away from measurements of total tau in favor of bd-tau and p-tau. Total tau quantifies tau of both CNS and peripheral origin, and recent work suggests that peripheral origin makes up approximately 80% of the total tau signal in the bloodstream. ⁹⁵ One study in our review did include both total tau and bd-tau, and although no statistical comparisons were made, the authors reported that bd-tau was a more sensitive marker and CNS-tau signal could be masked when total tau was used. ⁶²

It is possible that group-level analyses are being obscured by the heterogeneity we would expect in long-term cohorts. Growing evidence suggests that only a subgroup of TBI survivors have ongoing long-term neuropathology. ^{85,96 –99} This idea was explored in one study comparing a non-TBI control group with a cohort of moderate-severe TBI survivors 6–13 years post-injury. ⁷³ A Mann–Whitney U test comparing group-level data did not identify significant group differences; however, significant findings were identified when comparing the percentage of participants above the normal range. ⁷³ Even within this subgroup of survivors with long-term neuropathology, evidence suggests this is likely to be a mixed polypathology, ¹⁰⁰ in which constellations of pathological processes may differ across people. In support of this, one study examining 11 neurodegenerative-disease-associated protein variants reported that the pattern of protein elevations differed between participants. ⁸¹

Trajectories of protein levels over time

The current evidence provides little insight into trajectories of protein levels beyond 12 months post-injury. Within the first 12 months post-injury, the overall evidence for change in NfL, total tau, GFAP, bd-tau, and p-tau231 suggests a trajectory characterized by a decrease over time, ^58,62 whereas increases were reported in FGF-basic and IL-17A. ⁵⁷ Both proteins increased from admission to 3 months post-injury, and levels remained stable at 12 months post-injury. IL-17A has a role in stimulating and controlling the recruitment of neutrophils, ¹⁰¹ and FGF-basic has been shown to increase neurogenesis, preserve BBB integrity, and enhance blood vessel proliferation, while suppressing autophagy. ¹⁰² Chronic elevations in inflammatory proteins may be a positive phenomenon depending on the inflammatory marker and time since injury. It is generally accepted that persistent neuroinflammation may serve as a restorative mechanism; however, a prolonged inflammatory reaction can become destructive and ultimately neurotoxic over time through mechanisms such as oxidative stress, apoptosis, and excitotoxicity. ^{103 –106} Where this balance lies for each inflammatory marker or combination of markers within each individual is yet to be elucidated.

The relationship between protein level and biological markers

Establishing an association between blood-based protein biomarkers and brain pathology—as characterized by other clinically accessible biomarkers—is critical to establish whether changes in proteins can be mapped to changes in the brain. The overall evidence base addressing this question in the chronic TBI period is limited and diminished by small samples in studies with intermediate risk of bias. For both structural and diffusion MRI, a large number of analyses were conducted in multiple grey and white matter regions for several proteins in a very small sample, with either no or very few significant findings reported.

It is unclear what meaningful conclusions can be drawn from a small number of significant findings in the context of large numbers of nonsignificant results. It is possible that studies were underpowered and as such only identified areas with a stronger association. It also seems likely that heterogeneity in protein levels across individuals, and potentially differences in how protein levels are associated with brain pathology, could contribute to inconclusive findings. These findings do, however, show preliminary evidence that some biomarker elevations in the chronic period do reflect ongoing neural damage.

It is worth considering whether we should expect contemporaneous or closely paired (within 12 months) associations between blood-based protein levels and direct metrics of brain dysfunction, such as MRI and PET, within the chronic TBI period. Pathophysiological changes in the brain may be a slowly evolving process for which associations with protein biomarkers are only detectable after several years. ⁷⁴ In this context, we might expect to see detectable changes in different types of biomarkers (i.e., blood, CSF, MRI, PET) at different time points, akin to what is widely accepted in Alzheimer’s disease biomarker staging. ¹⁰⁷ Furthermore, pathophysiological changes in the brain such as volume loss are likely to be the downstream consequence of multiple pathological processes reflected by several blood-based biomarkers, thereby limiting the strength of any single protein-pathology association.

The relationship between protein levels with demographic, clinical, and injury factors

To understand how TBI may influence protein levels in the chronic period, it is important that we identify factors that may also affect protein levels, including key demographic and genetic factors. This will allow for precision prognostic modeling in the chronic period, wherein relevant contributors to biomarker levels can be included in statistical models. The most robust evidence identified was for the effect of APOE genotype: total tau was higher for APOE e4+ carriers compared to APOE e2+ carriers. ⁷⁰ In comparison, evidence for age, sex, and education was drawn from small studies with intermediate risk of bias.

It is critical to understand how biomarkers change with age to ensure normal age-related variations are disentangled from post-TBI pathological elevations. Age may have both direct effects on protein levels and synergistic effects through interactions with TBI. There is substantial evidence outside of TBI that age can directly impact the levels of some proteins, with older age associated with increased levels of p-tau, ¹⁰⁸ total tau, ¹⁰⁹ NfL, ¹¹⁰ and inflammatory markers. ¹¹¹ Age is also associated with differences in biological responses to injuries, as well as brain repair and recovery processes which then impact protein levels. ^112,113 Evidence from the acute period does indeed show that the association between TBI and the levels of some proteins differs by age. ^{57,58,114 –116} Generally, blood-based biomarkers show less specificity in older adults with TBI, as their uninjured peers also have elevated levels of several proteins. ^114,115,117 Protein levels may also differ by sex. There is preliminary work that has considered the interaction between TBI and sex on protein levels—mostly inflammatory markers ^{57,58,68,118 –120} —however, even in acute studies there has been little attempt to specifically examine the influence of sex. ³⁵

TBI is now recognized as a chronic and dynamic health condition, ¹²¹ with often includes polytrauma and impacts on multiple body systems. ¹⁴ Comorbidity of medical ^{122 –124} and psychiatric ^{125 –127} conditions are common. Several conditions that are commonly comorbid with TBI ¹²⁸ have been independently associated with increased levels of blood-based proteins including obstructive sleep apnea, ¹²⁹ insomnia, ¹³⁰ osteoarthritis, ¹³¹ depression, ¹³² and PTSD. ¹³³ Consideration also needs to be paid to conditions impacting renal and liver function, given their role in the elimination of proteins in the blood. ¹³⁴ Notably, only two studies controlled for or statistically explored important medical health conditions in their analyses. ^65,81 This is a major limitation potentially confounding the findings of previous work.

Although there is evidence to suggest injury factors may impact blood-based protein levels in the acute post-injury period, ¹³⁵ this was inconclusive in our review of the chronic period. Our “Key Findings” did report associations for GFAP with time since injury and for NfL with both time since injury and number of TBIs. These findings were, however, directly contrasted with nonsignificant findings from low risk of bias studies with smaller samples. Replication of analyses in larger samples is required to draw strong conclusions.

The prognostic role of blood-based proteins in predicting clinical outcomes

Our review provides preliminary evidence that some blood-based biomarkers are associated with cognitive and sleep outcomes, with less robust albeit promising findings for functional outcomes. In the few studies reviewed, there were no such associations with neurobehavioral outcome, PTSD, or quality of life. It is possible that earlier in the chronic period, proteins are accumulating, but not yet associated with an impact on outcomes. This has been found in other clinical cohorts where elevations in GFAP, NfL, and tau in older patients predated the development of cognitive decline, MCI, and AD with a latency of 8 years. ¹³⁶

Sleep dysfunction was associated with greater levels of IL-6, IL-10, NfL, TNF-α, and total tau. ^80,82 Sleep dysfunction may directly impact the production of some proteins (e.g., BDNF, CRP ¹³⁰ ). Preliminary research examining both TBI and poor sleep has found an interaction in their effect on enlarged perivascular spaces—an MRI marker used to infer glymphatic system disruption. ^137,138 The glymphatic system promotes the clearance of proteins and interstitial waste solutes out of the brain to maintain homeostasis ^30,139 and is thought to be twice as efficient during sleep ¹⁴⁰ (although this remains controversial ¹⁴¹ ). Sleep dysfunction ¹⁴² and TBI ^143,144 are both associated with glymphatic system disruption, which would lead to a build-up of proteins within the brain. It is not clear how glymphatic system disruption in the context of TBI and poor sleep may be reflected in protein levels in the blood. Indeed, TBI and sleep dysfunction are thought to directly increase most protein levels; however, protein levels would be reduced by glymphatic system disruption sequestering proteins in the parenchyma. ⁴⁰ Interestingly, there is evidence from animal studies that mice with TBI and a suppressed glymphatic system (through various mechanisms including sleep deprivation) had lower protein levels than those with TBI alone, ¹⁴⁵ suggesting that TBI and sleep dysfunction may actually lead to lower protein levels.

Review limitations

Our review has several limitations that should be considered. First, we applied strict eligibility criteria requiring that analyses of participants 12 months post-injury were presented separately. This resulted in the exclusion of several articles, as we could not confirm the lower bound of time since injury. Though unfortunate, any co-inclusion of acute samples would have obscured patterns specific to the chronic TBI period and constrained conclusions specific to chronic phase biomarker activity. Secondly, although we did reach out to a large number of authors to clarify methodological questions, we were not able to seek or receive answers to every item we might have hoped to clarify. There are also technical limitations when measuring blood-based biomarkers, and differences in preanalytical and analytical techniques make direct comparisons in protein levels difficult. ¹⁴⁶ There have been attempts made to define common data elements for the collection, preparation, and storage of biofluids for biomarkers. ¹⁴⁷ However, existing studies across the spectrum from acute to chronic TBI rarely make reference to these common data elements (CDEs). It is also important to consider that the proteins studied to date likely represent those for which established platforms are most readily available and may therefore be over-represented in currently published work. As such, the results of this review reflect the utility of blood biomarkers that are most accessible for inclusion in research.

There are also inherent limitations in the use of blood-based biomarkers as measures of CNS-derived processes. This is particularly problematic for proteins with non-CNS expression, in which the contribution from the CNS will potentially be obscured or even drowned out by protein expression in peripheral tissue. ⁴⁰ The role of the blood–brain barrier (BBB) may also complicate the interpretation of blood biomarker levels. The integrity of the BBB is commonly disrupted by TBI, ^{148 –151} and this can persist for decades post-injury. ¹⁵² The degree of BBB permeability is thought to influence the levels of biomarker proteins in the blood. As such, changes in blood biomarkers over time may reflect true changes in the levels of that protein in the brain, changes in BBB permeability, or a combination of these factors. ¹⁴⁶ Likewise, and as discussed above, the role of the glymphatic system may also influence protein levels in the blood.

Future research directions

At this preliminary stage of building the evidence base, future work should be broad in scope and endeavor to both replicate findings summarized here as well as consider other proteins, covariates, and outcomes. The selection of proteins should not be limited to those substantiated in the acute period. In the current review, the most studied proteins—GFAP, NfL, total tau, and UCHL-1—mirrored that of the acute period with the exception of S100B. ^35,153 This is an important piece of the puzzle as acute pathology may seed future pathogenesis—but it is unlikely that these biomarkers of acute injury capture the full story in the chronic phase of TBI pathology. Indeed, the clinical questions we ask of acute blood-based biomarkers, such as determining the need for CT scan, are substantially different from the clinical questions we seek to address in the chronic period. Foremost of which is who may be developing neurodegenerative pathology—requiring the examination of neurodegenerative-disease-related proteins. Taken together, we recommend further examination of acute TBI blood-based biomarkers along with exploration of novel markers—with efforts guided by promising markers in neurodegenerative diseases. Ideally, such examinations would look at proteins independently and use panels to examine polypathology. Finally, as the field continues to advance it will be important to identify biomarkers that are sensitive to treatment effects in the chronic stages of TBI that would provide mechanistically relevant outcome measures for clinical trials, and ultimately, tools for clinical monitoring of disease progression.

No single biomarker is likely to capture the full pathophysiology of all TBIs. Preliminary work in the acute ¹⁵⁴ and chronic period ^58,63,68,81 has shown the utility of using panels, but the majority of work across time since injury focuses on single proteins. ³⁵ A blood-based biomarker panel may be optimized by including proteins sampled at different time points (i.e., acute, subacute, chronic) when they are known to be most sensitive to outcome prognostication. ^58,63 Panel accuracy and efficacy may also be improved by including multiple measurements of protein levels over time, ³⁶ demographic characteristics, ⁵⁸ and advanced neuroimaging markers. ¹⁵⁵ These panels could then be used to identify subgroups with distinct biomarker profiles and trajectories. Deep phenotyping of these subgroups may create opportunities for precision-medicine approaches to address TBI-related pathophysiology in the chronic post-injury period.

Direct review of methodological issues and limitations in the summarized studies also provides clear areas for strengthening future work: matching TBI and non-TBI groups on key demographic variables at chronic time points; controlling for key confounding factors including age and sex; increasing sample sizes particularly when conducting stratified sub-analyses; more diverse samples (sex, race, ethnicity); consideration of number of TBI; and clear reporting of eligibility criteria and participant demographics including the range of time since injury. Longitudinal prospective studies must be designed that extend beyond 12 months post-injury if we are to identify markers that portend clinically meaningful post-recovery decline. Evolving technologies in biomarker discovery that permit ultrasensitive detection of circulating biomarkers may be of particular value for elucidating clinically relevant long-term elevations in biomarkers that are, on average, in lower abundance in the chronic stages relative to acute (e.g., circulating structural proteins such as UCH-L1 and S100B). For markers of chronic processes that we might hypothesize to progress over time with age (e.g., neuroinflammation, neurodegeneration), the ability to detect subtle but clinically important elevations in biomarker progression relative to appropriately matched uninjured controls will require similarly precise detection sensitivity.

Clinical implications

Based on our critical review of the current evidence base, it is our opinion that there remains insufficient evidence to substantiate any marker being used in clinical practice.