Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Jun 1.
Published in final edited form as: Ann Neurol. 2020 Apr 20;87(6):907–920. doi: 10.1002/ana.25725

A prospective study of acute blood-based biomarkers for sport-related concussion

Timothy B Meier 1,2,3, Daniel L Huber 1, Luisa Bohorquez-Montoya 1, Morgan E Nitta 1,4, Jonathan Savitz 5,6, T Kent Teague 7,8,9,10, Jeffrey J Bazarian 11, Ronald L Hayes 12, Lindsay D Nelson 1,13, Michael A McCrea 1,13
PMCID: PMC7477798  NIHMSID: NIHMS1619597  PMID: 32215965

Abstract

Objective

Prospectively characterize changes in serum proteins following sport-related concussion and determine if candidate biomarkers discriminate concussed athletes from controls and are associated with duration of symptoms following concussion.

Methods

High school and collegiate athletes were enrolled between 2015 and 2018. Blood was collected at pre-injury baseline and within 6 hours (early-acute) and at 24–48 hours (late-acute) following concussion in football players (n=106), matched uninjured football players (n=84) and non-contact sport athletes (n=50). Glial fibrillary acidic protein, ubiquitin c-terminal hydrolase-L1, S100 calcium binding protein B, alpha-II-spectrin breakdown product 150, interleukin-6, interleukin-1 receptor antagonist and c-reactive protein were measured in serum. Linear models assessed changes in protein concentrations over time. Receiver operating curves quantified the discrimination of concussed athletes from controls. A Cox proportional hazard model determined if proteins were associated with symptom recovery.

Results

All proteins except glial fibrillary acidic protein and c-reactive protein were significantly elevated at the early-acute phase post-injury relative to baseline and both control groups and discriminated concussed athletes from controls with areas under the curve of 0.68–0.84. The candidate biomarkers also significantly improved the discrimination of concussed athletes from non-contact controls compared to symptom severity alone. Glial fibrillary acidic protein was elevated post-injury relative to baseline in concussed athletes with a loss of consciousness or amnesia. Finally, early-acute levels of interleukin-1 receptor antagonist were associated with the number of days to symptom recovery.

Interpretation

Brain injury and inflammatory proteins show promise as objective diagnostic biomarkers for sport-related concussion, while inflammatory markers may provide prognostic value.

INTRODUCTION

Sport-related concussion (SRC) is a major public health issue estimated to affect millions of individuals in the U.S. each year.1 Currently, clinical decisions regarding the diagnosis of SRC and the determination of when athletes can return-to-play (RTP) are based on clinical judgement that is largely informed by the patient self-report of symptoms along with other clinical testing (e.g., balance and neurocognitive testing). Thus, there is great interest in identifying objective biomarkers to assist in diagnosis of SRC, identification of players at risk of delayed recovery, and determination of an athlete’s readiness to RTP after concussion.

Blood-based biomarkers have been a focus recently because of their relative cost-effectiveness and their potential as point-of-care markers. Biomarkers that putatively capture various aspects of the neurometabolic cascade of concussion have been investigated following SRC, including S100 calcium binding protein B (S100B), glial fibrillary acidic protein (GFAP), ubiquitin c-terminal hydrolase-L1 (UCH-L1), alpha-II-spectrin derivatives, and general inflammatory markers.2–11 As highlighted in a recent systematic review, however, much of the prior blood biomarker work in SRC has been limited to relatively small sample sizes and/or did not include preseason (i.e., preinjury) assessments.12 Moreover, recent findings that some of the proposed biomarkers for SRC are sensitive to exercise and/or exposure to minor head impacts during sport participation without frank concussion highlight the importance of appropriate control comparisons.13–16

The current study assessed the effects of acute SRC in a large, prospective cohort of high school and collegiate football players on serum markers of brain injury and inflammation including GFAP, UCH-L1, S100B, alpha-II-spectrin breakdown product 150 (SBDP150), interleukin (IL) 6, IL-1 receptor antagonist (IL-1RA) and c-reactive protein (CRP). To address limitations in prior work, blood was collected at a preinjury baseline visit and at two acute post-injury time points. Matched uninjured football players and matched non-contact sport athletes were enrolled as separate control groups to account for the potential effects of football exposure without concussion. Based on prior work,2–11 we hypothesized that serum levels of the selected markers would be elevated at the early-acute stage following SRC relative to preinjury levels and relative to both control groups. Furthermore, we hypothesized that the panel of markers would improve the discrimination of concussed from controls athletes compared to concussion symptoms alone (diagnostic biomarker). Finally, we hypothesized that acute elevations in these serum biomarkers would be associated with the duration of symptoms following SRC (prognostic biomarker).

METHODS

Participants

High school and collegiate athletes were enrolled as part of a prospective study of the natural history of concussion between August 2015 and June 2018 (Project Head to Head-II). The institutional review board at the Medical College of Wisconsin approved the study. Adult participants and parents of minors provided written informed consent while minors completed written assent.

A total of 1,136 football players were enrolled prior to the season and completed a baseline visit that included a detailed clinical battery and collection of blood specimen. Exclusion criteria for participation in follow-up visits included: injury that would precluded participation in the study, current psychotic disorder or narcotic use, history or suspicion of conditions known to cause cognitive dysfunction (e.g., epilepsy, moderate-to-severe TBI), any contraindication to study procedures, prior concussion in the last 6 months (contact controls), and prior concussion or football experience in high school or college (non-contact controls). Concussions were identified and diagnosed by certified athletic trainers or team physicians at each institution and subsequently screened and triaged by study investigators to confirm they met the study definition and requirements. The study definition of concussion was based on the Centers for Disease Control and Prevention HEADS up educational initiative, as previously described.5 A total of 106 football players that enrolled at baseline sustained a concussion during the study period and met the criteria for the current study. Concussed athletes completed follow-up visits with blood collection within 6 hours (early-acute) and at 24–48 hours (late-acute), 8 days, 15 days, and 45 days post-injury.

A total of 84 contact control athletes (CC) were selected from uninjured football players that enrolled at baseline to match injured athletes using a closest possible approach based on the following matching criteria: level of play (i.e., college or high school), institution, team, age, estimated intellectual functioning (word reading performance at baseline), race, handedness, concussion history, and position. In addition, a total of 50 non-contact control athletes (basketball, n=16; track and field, n=8; baseball, n=26; NCC) without current or high school football exposure or prior concussion served as additional controls matched using a closest possible approach based on level (high school or college), age, estimated premorbid intelligence, and race. CC and NCC athletes completed the same study protocol at similar follow-up visits to injured athletes. The first follow-up visits for control participants occurred as soon as possible following identification of the participant based on study resources and participant availability. The focus of the current study is on acute biomarkers of concussion; thus, only data from the baseline, early-acute, and late-acute visits are included. Sample characteristics are reported in Table 1. Football athletes were allowed to participate as a control and then subsequently in the injury protocol, while football athletes could also complete the injury protocol for multiple concussions during the study period. A total of 5 football players initially participated as controls and subsequently sustained a concussion, and 4 football players participated in the injury protocol for multiple concussions. These data points are treated as independent in analyses. Sensitivity analyses demonstrated that the inclusion of these data did not affect results (see Results). UCH-L1 and S100B levels for 32 concussed athletes and 29 CC have been previously reported.5 Samples from the original report were reanalyzed using an updated GFAP assay; thus, GFAP data have not been previously reported. IL-6, IL-1RA, and CRP have been previously reported in 41 concussed and 43 CC athletes.2

Table 1:

Sample characteristics

SRC CC NCC Statistic p-value
N 106 84 50
Age 18.00 (1.52) 18.37 (1.68) 18.74 (1.85) F=3.59 0.029
Body mass index 29.57 (6.13) 28.81 (5.26) 24.23 (3.21) F=17.85 <0.001
Years of sport Participation 8.02 (3.37) 8.15 (2.65) 11.44 (4.49) F=19.16 <0.001
Median prior concussions [IQR] 1 [0–2] 0 [0–1] 0 [0–0] H=37.16 <0.001
WTAR Standard Score 99.39 (14.17) 99.46 (12.39) 104.66 (9.19) F=3.39 0.035
Race, N (%) FET 0.17
White 81 (76.4%) 64 (76.2%) 43 (86.0%)
Black 23 (21.7%) 19 (22.6%) 5 (10.0%)
American Indian or Alaska Native 0 (0.0%) 1 (1.2%) 1 (2.0%)
Unknown/Not Reported 2 (1.9%) 0 (0.0%) 1 (2.0%)
Ethnicity FET 0.76
Non-Hispanic 91 (85.8%) 74 (89.2%) 46 (92.0%)
Hispanic 7 (6.6%) 5 (6.0%) 3 (6.0%)
Unknown/Not Reported 8 (7.5%) 4 (4.8%) 1 (2.0%)
Competition Level, N (%) College 87 (82.1%) 68 (81.0%) 37 (74.0%) X2=1.46 0.48
ADHD, N (%) 9 (8.5%) 7 (8.3%) 5 (10.0%) FET 0.91
Learning Disorder, N (%) 2 (1.9%) 0 (0.0%) 0 (0.0%) FET 0.69
Open in a new tab

Note: Mean (standard deviation) are shown unless otherwise indicated. SRC = sport-related concussion; CC = contact controls; NCC = non-contact controls; N = number; IQR = interquartile range; WTAR = Wechsler Test of Adult Reading; ADHD = attention-deficit/hyperactivity disorder; FET = Fisher’s exact test.

Clinical battery

Demographic and health history information were collected at baseline along with the Wechsler Test of Adult Reading (WTAR) as an estimate of intellectual function. The Sport Concussion Assessment Tool – 3rd Edition symptom checklist was collected at each visit to assess concussion symptom severity. Detailed concussion information was collected at follow-up visits (e.g., injury characteristics, the number of days that athletes were symptomatic).

Biomarker data

Venous blood was collected using Red Top BD Vacutainer tubes, left to clot at room temperature for 30 minutes, and centrifuged at 1,500 RCF for 15 minutes. Serum concentrations of UCH-L1, GFAP, and SBDP150 were determined using enzyme-linked immunosorbent assays (Banyan Biomarkers, Inc., Alachua, FL) and serum concentrations of S100B were determined using an electrochemiluminescence immunoassay designed for diagnostic testing (Roche, Cobas 6000). Serum concentrations of IL-6, IL-1RA, and CRP were measured using a Meso Scale Discovery (MSD) QuickPlex SQ 120 instrument and MSD V-PLEX assays following manufacturer’s instructions. All assays were run blinded to diagnosis. The lower limit of quantification (LLOQ) and the upper limit of quantification (ULOQ) of each biomarker are noted in Table 2.

Table 2:

Symptom severity and biomarker levels

Measures Baseline M (SD), N Early-acute M (SD), N Late-acute M (SD), N Group Effect p-value Visit Effect p-value Group-by-Visit Effect p-value
Symptom Severity Score SRC 3.76 (6.96), 106 29.19 (21.98), 100 23.54 (20.86), 103 <0.001* <0.001* <0.001*
CC 2.13 (4.22), 83 1.29 (1.98), 49 1.73 (3.31), 84
NCC 2.48 (3.81), 50 3.59 (6.53), 44 3.32 (6.36), 50
GFAP SRC 3.00 (0.85), 98 3.12 (0.86), 59 3.15 (0.93), 98 0.09 <0.001* 0.06
CC 2.88 (0.94), 78 2.97 (0.93), 80 2.94 (0.90), 79
NCC 2.79 (0.87), 46 2.79 (0.85), 46 2.87 (0.88), 43
UCH-L1 SRC 4.84 (0.58), 102 5.24 (0.61), 58 4.59 (0.55), 99 0.001* <0.001* <0.001*
CC 4.82 (0.60), 80 4.74 (0.65), 68 4.63 (0.65), 73
NCC 4.60 (0.58), 44 4.57 (0.62), 46 4.53 (0.58), 40
S100B SRC 4.06 (0.57), 105 4.29 (0.56), 60 3.87 (0.76), 103 <0.001* <0.001* <0.001*
CC 4.10 (0.54), 84 3.92 (0.61), 84 3.80 (0.56), 82
NCC 3.76 (0.52), 50 3.70 (0.50), 50 3.56 (0.50), 49
SBDP150 SRC 5.27 (0.34), 104 5.38 (0.26), 60 5.15 (0.32), 103 0.001* <0.001* 0.001*
CC 5.21 (0.27), 84 5.12 (0.19), 83 5.07 (0.27), 81
NCC 5.25 (0.40), 49 5.20 (0.31), 50 5.13 (0.36), 49
IL-6 SRC −0.61(0.83), 91 0.03 (0.90), 54 −0.61 (0.78), 90 <0.001* <0.001* <0.001*
CC −0.76 (0.72), 67 −0.78 (0.57), 71 −0.73 (0.60), 68
NCC −0.80 (0.85), 38 −0.69 (0.71), 37 −0.85 (0.71), 39
IL-1RA SRC 5.63 (0.52), 96 6.00 (0.59), 56 5.69 (0.63), 94 <0.001* 0.001* <0.001*
CC 5.56 (0.51), 80 5.49 (0.44), 80 5.49 (0.43), 77
NCC 5.45 (0.38), 41 5.36 (0.39), 41 5.40 (0.41), 40
CRP SRC 13.31 (1.41), 96 13.48 (1.35), 56 13.56 (1.38), 93 <0.001* 0.53 0.14
CC 13.36 (1.35), 80 13.31 (1.49), 80 13.22 (1.56), 77
NCC 12.33 (1.31), 41 12.48 (1.37), 41 12.42 (1.41), 40
Open in a new tab

Note: Natural log-transformed levels of biomarkers in pg/mL are shown. The lower and upper levels of quantification for markers in pg/mL are as follows: GFAP, 5 and 448; UCH-L1, 48 and 3,584; SBDP150, 19 and 1,200; S100B, NA and 39,000; IL-1RA, 0.17–5.14 and 732–1,150; IL-6, 0.006–0.26 and 721–764; CRP, 1.11–4.62 and 202,000. GFAP = glial fibrillary acidic protein; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist; CRP = c-reactive protein; SRC = sport-related concussion; CC = contact controls; NCC = non-contact controls; M (SD) = mean (standard deviation); N = number of usable samples.

*

Indicates effects significant at False Discovery Rate q<0.05.

Statistical analysis

Statistical analyses were performed using IBM SPSS Statistics version 24 (Armonk, NY). One-way analyses of variance (ANOVA), Kruskal-Wallis tests, chi-square tests, or Fisher’s exact tests were used to compare demographic variables between groups. Biomarker levels were natural log-transformed to better approximate a normal distribution. Biomarker levels below the LLOQ or above the ULOQ were excluded from analyses. For inflammatory markers, samples with coefficient of variation between replicates of greater than 25% were also excluded. The coefficient of variation for non-inflammatory markers were all less than 15%. The total final N for each biomarker is listed in Table 2. Generalized linear mixed models were used to assess changes in variables of interest within athletes over time as a function of group, with visit modeled as a repeated factor across participants (i.e., baseline, early-acute, late-acute), group (i.e., SRC, CC, NCC), and the group-by-visit interaction. Continuous data were modeled with normal distributions; symptom severity scores were modeled with a negative binomial distribution. Finally, exploratory linear mixed models were performed to determine if biomarker levels in SRC athletes were associated with acute injury characteristics (i.e., loss of consciousness [LOC], post-traumatic amnesia [PTA], and/or retrograde amnesia [RGA]). Changes in biomarkers across visits in SRC athletes were assessed with visit modeled as a repeated factor across participants, SRC group (i.e., with or without either LOC, PTA, or RGA), and the group-by-visit interaction. Mixed model analyses were conducting under the assumption that missing data were missing at random.

Area under the receiver operating curves (AUC) quantified the ability of markers at the early-acute visit to discriminate SRC from either CC or NCC. Penalized logistic regression analyses were conducted using the least absolute shrinkage and selection operator (LASSO) to determine if the inclusion of blood biomarkers significantly improved the discrimination of SRC from CC or NCC relative to symptom severity alone at the early-acute visit. Predicted probabilities resulting from the logistic regression models were used to calculate AUCs for combinations of markers. The Delong Test was used to compare AUCs from dependent models. Finally, a Cox proportional hazard model, implemented using the coxph function in R with the Efron method, was used to determine if early-acute biomarker levels were associated with time to symptom recovery (in days) in concussed athletes following verification that the proportional hazards assumption was met. In absence of an independent validation cohort, repeated (n=10) 5-fold cross-validation was conducted using the caret (AUCs and logistic regression models) and RMS (cox regression) R packages. An alpha of 0.05 (2-tailed) was considered for all analyses. Benjamini-Hochberg False Discovery Rate (q<0.05) correction was applied to analyses aligned with the study’s primary hypotheses (i.e., whole sample mixed models, logistic regressions, and proportional hazard models). Post-hoc tests were conducted with Bonferroni correction at p<0.05 when indicated for mixed models.

RESULTS

Demographic data, injury characteristics, and symptom severity scores

There were no differences in baseline demographic variables between SRC and CC (Table 1). NCC were significantly older at baseline than SRC (mean difference(MD)=0.74, 95% Confidence Interval(CI)[0.06, 1.42], p=0.03), had significantly higher WTAR standard scores than SRC (MD=5.27, 95%CI[0.03, 10.51], p=0.048), had significantly lower BMI than both SRC (MD=−5.34, 95%CI[−7.54, −3.13], p<0.001) and CC (MD=−4.58, 95%CI[−6.88, −2.28], p<0.001), and had more years of participation in their sport relative to both SRC (MD=3.42, 95%CI[2.01, 4.84], p<0.001) and CC (MD=3.29, 95%CI[1.81, 4.76], p<0.001). By design, NCC also had significantly fewer concussions than SRC and CC (ps<0.001).

Three SRC athletes reported loss of consciousness, 17 reported post-traumatic amnesia, and 4 reported retrograde amnesia. The median number of days that SRC athletes reported symptoms was 7 [IQR 5–11], while the median number of days before unrestricted return-to-play was 11 [IQR 8.5–14.5].

There was a significant group-by-visit interaction for symptom severity scores (Figure 1; Table 2), with SRC athletes having significantly elevated symptoms at the early-acute and late-acute visits relative to both control groups and relative to their own baseline (ps<0.001; Table 3).

Figure 1: Longitudinal changes in symptom severity and biomarker levels.

Figure 1:

Open in a new tab

Shown are the mean and standard error of symptom severity scores and biomarker levels at the pre-injury baseline, early-acute, and late-acute visits in athletes with sport-related concussion (SRC), contact control athletes (CC), and non-contact control athletes (NCC). Natural log-transformed levels of biomarkers in pg/mL are shown. GFAP = glial fibrillary acidic protein; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist; CRP = c-reactive protein.

Table 3:

Mean differences and 95% confidence intervals for post-hoc comparisons for the generalized linear mixed models

Group Contrast Symp. Sev. UCH-L1 S100B SBDP150 IL-6 IL-1RA
SRC EA-BL 24.84, [15.48, 34.19] 0.42, [0.28, 0.56] 0.21, [0.08, 0.34] 0.11, [0.02, 0.20] 0.58, [0.38, 0.78] 0.39, [0.29, 0.50]
EA-LA 5.07, [−4.29, 14.43] 0.67, [0.53, 0.82] 0.40, [0.28, 0.54] 0.22, [0.13, 0.32] 0.56, [0.35, 0.76] 0.33, [0.22, 0.44]
LA-BL 19.76, [12.22, 27.30] −0.25, [−0.37, −0.14] −0.19, [−0.30, −0.09] −0.11, [−0.19, −0.04] 0.02, [−0.14, 0.19] 0.06, [−0.03, 0.15]
CC EA-BL NS −0.12, [−0.26, 0.02] −0.18, [−0.30, −0.07] −0.09, [−0.18, −0.01] NS NS
EA-LA NS 0.07, [−0.07, 0.21] 0.12, [0.00, 0.24] 0.05, [−0.04, 0.13] NS NS
LA-BL NS −0.19, [−0.33, −0.06] −0.30, [−0.42, −0.18] −0.14, [−0.23, −0.06] NS NS
NCC EA-BL NS NS −0.05, [−0.20, 0.10] −0.05, [−0.16, 0.06] NS NS
EA-LA NS NS 0.14, [−0.02, 0.29] 0.07, [−0.04, 0.18] NS NS
LA-BL NS NS −0.19, [−0.34, −0.04] −0.12, −0.23, −0.004] NS NS
Visit Contrast Symp. Sev. UCH-L1 S100B SBDP150 IL-6 IL-1RA
BL SRC-CC NS NS −0.03, [−0.24, 0.18] NS NS NS
SRC-NCC NS NS 0.32, [0.07, 0.56] NS NS NS
CC-NCC NS NS 0.35, [0.09, 0.60] NS NS NS
EA SRC-CC 27.19, [17.24, 37.15] 0.56, [0.33, 0.79] 0.36, [0.14, 0.59] 0.26, [0.14, 0.38] 0.73, [0.43, 1.04] 0.53, [0.34, 0.72]
SRC-NCC 25.07, [14.96, 35.19] 0.69, [0.42, 0.95] 0.58, [0.32, 0.84] 0.18, [0.05, 0.32] 0.70, [0.34, 1.07] 0.66, [0.43, 0.90]
CC-NCC −2.12, [−4.23, −0.01] 0.13, [−0.14, 0.40] 0.22, [−0.04, 0.47] −0.08, [−0.21, 0.05] −0.03, [−0.39, 0.33] 0.13, [−0.10, 0.36]
LA SRC-CC 21.80, [13.66, 29.93] NS 0.08, [−0.13, 0.29] NS NS 0.20, [0.01, 0.38]
SRC-NCC 20.20, [11.91, 28.50] NS 0.31, [0.06, 0.56] NS NS 0.30, [0.07, 0.52]
CC-NCC −1.59, [−3.53, 0.34] NS 0.23, [−0.02, 0.49] NS NS 0.10, [−0.13, 0.33]
Open in a new tab

Note: The group-by-visit interaction was not significant in the primary analysis for c-reactive protein and glial fibrillary acidic protein. NS = simple main effect not significant; SRC = sport-related concussion; CC = contact controls; NCC = non-contact controls; EA = early-acute visit; LA = late-acute visit; BL = baseline visit; Symp. Sev. = symptom severity score; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist.

Brain injury biomarker data

The median time from injury to blood specimen collection was 3.8 hours [IQR 2.2–5.2] at the early-acute visit and 33 hours [IQR 21.1–46.5] at the late-acute visit. There was a significant group-by-visit interaction for UCH-L1, S100B, and SBDP150 (Figure 1, Table 2). Mean differences and confidence intervals for all post-hoc comparisons are presented in Table 3. Follow-up analyses demonstrated UCH-L1 levels were significantly elevated in SRC at the early-acute visit relative to both CC and NCC as well as relative to their own baseline and late-acute visits (p<0.001). Compared to baseline values, 79% of SRC athletes had increased UCH-L1 at the early-acute visit. UCH-L1 levels at the late-acute visit were significantly lower than baseline levels for both SRC and CC (p<0.005).

S100B levels were also significantly elevated at the early-acute visit in SRC relative to both NCC and CC as well as to their own baseline and late-acute visits (ps<0.001). Compared to baseline values, 70% of SRC athletes had elevated S100B at the early-acute visit. S100B levels were still elevated at the late-acute visit in SRC relative to NCC (p<0.01). In addition, baseline levels in all groups were significantly higher than levels at the late-acute visit (ps<0.01), while baseline levels in CC were also higher than the early-acute visit (ps<0.001). Finally, NCC had lower baseline levels of S100B relative to both CC and SRC (ps<0.01).

SRC had elevated SBDP150 at the early-acute visit relative to both NCC and CC, and relative to their own baseline and late-acute visits (ps<0.05). A total of 72% of SRC athletes had elevated SBDP150 levels at the early-acute visit relative to baseline. Baseline levels were significantly elevated relative to the late-acute visit in all groups (ps<0.05), while baseline levels were also elevated relative to the early-acute visit in CC (p<0.05).

Inflammatory biomarker data

There was a significant group-by-visit interaction for IL-1RA and IL-6 (Figure 1; Table 2; Table 3). SRC also had elevated levels of IL-1RA and IL-6 at the early-acute visit relative to CC and NCC and relative to other visits (ps<0.001). Relative to baseline levels, higher IL-6 and IL-1RA levels were evident at the early-acute visit in 75% and 80% of SRC athletes, respectively. IL-1RA levels were also elevated in SRC relative to CC and NCC at the late-acute visit (ps<0.05). Finally, there was a main effect of group on CRP levels, with higher CRP levels in both SRC (MD=1.04, 95%CI[0.47, 1.61], p<0.001) and CC athletes relative to NCC (MD=0.90, 95%CI[0.31, 1.48], p<0.001).

Biomarker data - sensitivity analyses

Sensitivity analyses accounting for age, BMI, and WTAR scores did not affect the significant findings in the primary analysis, with the exception of the following: group differences in CRP levels were no longer significant (p=0.35; i.e., were driven by group differences in BMI); and there was a significant effect of group on GFAP levels (p=0.047), with SRC athletes having elevated levels relative to NCC (MD=0.40, 95%CI[0.01, 0.80], p=0.04). Sensitivity analyses excluding data from athletes with repeat SRC showed equivalent results as the primary analyses. Finally, additional sensitivity analyses with biomarker marker values below or above quantification levels replaced by the LLOQ and ULOQ, respectively, showed consistent results relative to the primary analyses for all markers, though the group-by-visit interaction for GFAP went from a non-significant trend in the primary analysis to significant (p=0.02). SRC athletes had elevated GFAP levels at the early-acute (MD=0.25, 95%CI[0.12, 0.37], p<0.001) and late-acute visits (MD=0.20, 95%CI[0.09, 0.30], p<0.001) relative to the baseline visit and relative to NCC (early-acute: MD=0.48, 95%CI[0.08, 0.87], p=0.01; late-acute: MD=0.41, 95%CI[0.03, 0.80], p=0.03).

Discrimination of concussed athletes from controls

Area under the receiver operating curves (ROC) for the whole sample and the repeated 5-fold cross-validation for markers at the early-acute visit are presented in Table 4. Individual biomarkers had area under the curves (AUC) ranging from 0.54 to 0.81 for discriminating SRC from CC and AUCs ranging from 0.63 to 0.84 for discriminating SRC from NCC. The combination of all biomarkers at the early-acute visit discriminated SRC from CC and NCC with AUCs of 0.90 for both comparisons. As seen in Table 4, results for the cross-validation were similar to those in the whole sample.

Table 4:

Area under the receiver operating characteristic curves for tested biomarkers at the early-acute phase

SRC vs CC AUC [95% CI] AUCcv (SD) SENScv (SD) SPECcv (SD)
GFAP 0.57 [0.47, 0.66] 0.56 (0.09) 0.91 (0.11) 0.06 (0.08)
UCH-L1 0.74 [0.65, 0.83] 0.74 (0.09) 0.76 (0.11) 0.56 (0.15)
S100B 0.68 [0.60, 0.77] 0.69 (0.10) 0.82 (0.09) 0.39 (0.12)
SBDP150 0.81 [0.74, 0.89] 0.81 (0.07) 0.86 (0.09) 0.65 (0.13)
IL-6 0.78 [0.70, 0.87] 0.78 (0.09) 0.80 (0.11) 0.69 (0.14)
IL-1RA 0.78 [0.70, 0.85] 0.77 (0.07) 0.85 (0.08) 0.47 (0.14)
CRP 0.54 [0.44, 0.64] 0.48 (0.11) 0.97 (0.05) 0.02 (0.05)
All biomarkers 0.90 [0.84, 0.96] 0.85 (0.08) 0.83 (0.11) 0.74 (0.14)
Symp. Sev.* 0.95 [0.91, 1.00] 0.95 (0.05) 1.00 (0.00) 0.85 (0.11)
Symp. Sev + biomarkers 0.99 [0.97, 1.00] 0.97 (0.04) 0.93 (0.11) 0.90 (0.08)
Symp. Sev.* (subsample) 0.88 [0.75, 1.00] 0.87 (0.13) 1.00 (0.00) 0.62 (0.25)
Symp. Sev + biomarkers (subsample) 0.96 [0.91, 1.00] 0.93 (0.08) 0.98 (0.05) 0.72 (0.24)
SRC vs NCC AUC [95% CI] AUCcv (SD) SENScv (SD) SPECcv (SD)
GFAP 0.63 [0.52, 0.74] 0.62 (0.11) 0.35 (0.17) 0.80 (0.14)
UCH-L1 0.79 [0.70, 0.88] 0.80 (0.09) 0.71 (0.15) 0.76 (0.13)
S100B 0.79 [0.70, 0.88] 0.79 (0.10) 0.67 (0.13) 0.73 (0.13)
SBDP150 0.73 [0.63, 0.82] 0.73 (0.10) 0.58 (0.17) 0.78 (0.12)
IL-6 0.75 [0.64, 0.85] 0.74 (0.10) 0.52 (0.17) 0.81 (0.11)
IL-1RA 0.84 [0.75, 0.92] 0.83 (0.10) 0.72 (0.15) 0.81 (0.11)
CRP 0.71 [0.60, 0.81] 0.70 (0.12) 0.54 (0.15) 0.78 (0.14)
All biomarkers 0.90 [0.82, 0.97] 0.88 (0.08) 0.63 (0.15) 0.92 (0.08)
Symp. Sev.* 0.94 [0.89, 0.99] 0.94 (0.05) 0.89 (0.11) 0.91 (0.09)
Symp. Sev + biomarkers 0.98 [0.96, 1.00] 0.96 (0.05) 0.86 (0.16) 0.92 (0.09)
Symp. Sev.*(subsample) 0.84 [0.71, 0.98] 0.84 (0.15) 0.96 (0.09) 0.64 (0.24)
Symp. Sev + biomarkers (subsample) 0.95 [0.89, 1.00] 0.92 (0.09) 0.91 (0.12) 0.68 (0.26)
Open in a new tab

Note: AUC = area under the receiver operating characteristic curve; SENS = sensitivity; SPEC = specificity, SD = standard deviation, CV= results from the repeated 5-fold cross validation

*

SCAT for those with biomarkers at 6 hours for comparison. SRC = sport-related concussion; CC = contact controls; NCC = non-contact controls; Symp. Sev. = symptom severity score; glial fibrillary acidic protein = GFAP; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist; CRP = c-reactive protein. Subsample = analyses limited to SRC with lower tercile of symptoms.

Logistic regression analyses showed that the inclusion of blood biomarkers significantly improved the discrimination of SRC from NCC at the early-acute visit relative to symptom severity scores alone (X2(3)=12.27, p=0.006, q<0.05), improving the AUC to 0.98 from 0.94. Despite the significant improvement in the logistic regression model fit, DeLong’s test showed that the improvement in AUC was not statistically significant (p=0.12). The inclusion of blood biomarkers also improved the AUC for discriminating SRC from CC at the early-acute visit compared to the use of symptoms alone (0.99 versus 0.95), although this improvement was not significant based on the logistic regression analysis (X2(5)=10.94, p=0.053) or Delong’s test (p=0.10).

Because biomarkers might have utility in clinical scenarios where self-reported symptoms are lower or potentially unreliable, exploratory analyses compared AUCs of symptom severity combined with biomarkers at the early-acute visit to symptom severity alone in concussed athletes in the bottom tercile of self-report symptoms at the early-acute visit (n=19). Relative to symptoms alone, AUCs were higher for differentiating SRC from CC (AUC=0.96) and NCC (AUC=0.95) compared to the use of symptoms alone (AUCs = 0.88 and 0.84, respectively), though these differences were not significant based on DeLong’s test (ps=0.07). However, the addition of biomarkers did significantly improve the logistic regression model fit based on symptoms alone for both the CC (X2(3)=8.24, p=0.04) and NCC comparisons (X2(2)=9.19, p=0.01).

Association with symptom duration

Relevant statistics for tests of associations between early-acute biomarkers and symptom duration in SRC are reported in Table 5. Higher levels of IL-1RA at the early-acute visit were significantly associated with greater symptom duration following concussion (p=0.03, q<0.05), with a one unit increase in natural log-transformed IL-1RA associated with 40% lower hazards of recovery (Figure 2). Higher levels of IL-6 were also associated with greater symptom duration following concussion, though this relationship was not statistically significant (p=0.08). There was also a non-significant association between higher S100B and shorter symptom duration (p=0.09).

Table 5:

Cox regression results for concussed athletes at the early-acute visit

HR 95% CI p-value C-Index C-Indexcv
GFAP 1.01 0.74, 1.38 0.94 0.46 0.35
UCH-L1 1.13 0.73, 1.74 0.58 0.50 0.44
S100B 1.59 0.93, 2.71 0.09 0.55 0.55
SBDP150 0.54 0.23, 1.29 0.16 0.54 0.53
IL-6 0.75 0.55, 1.03 0.08 0.58 0.57
IL-1RA 0.60 0.38, 0.95 0.03* 0.56 0.57
CRP 1.03 0.85, 1.25 0.75 0.50 0.42
Open in a new tab

Note: HR = hazard ratio, CI = confidence interval, C-Index = concordance index, CV= results from the repeated 5-fold cross validation, glial fibrillary acidic protein = GFAP; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist; CRP = c-reactive protein.

*

Indicates effects significant at False Discovery Rate q<0.05.

Figure 2: Association of IL-1RA and IL-6 with symptom duration.

Figure 2:

Open in a new tab

Shown is Kaplan-Meier plot for recovery in concussed athletes for interleukin-1 receptor antagonist (IL-1RA) and interleukin-6 (IL-6) at the early-acute post-concussion visit. Line colors denote different terciles for each marker for illustrative purposes.

Association with acute injury characteristics

Exploratory generalized linear models compared changes in biomarker levels over time in SRC athletes with loss of consciousness, post-traumatic amnesia, and/or retrograde amnesia (SRC+, n=21) and concussed athletes with none of these signs of acute altered mental status (SRC-, n=85). There was a significant group (SRC+ vs. SRC-)-by-visit interaction for GFAP (p=0.021; Figure 3). Follow-up tests demonstrated that GFAP was significantly elevated in SRC+, but not SRC-, at the early-acute (MD=0.49, 95%CI [0.19, 0.80], p<0.001) and late-acute visits (MD=0.39, 95%CI [0.13, 0.66], p<0.001) relative to baseline. No group (SRC+ vs. SRC-) or group-by-visit effects were observed for any other marker (ps>0.10).

Figure 3: Effect of acute concussion characteristics on biomarker levels.

Figure 3:

Open in a new tab

Shown are the mean and standard error of natural log-transformed biomarker levels, in pg/mL, at the pre-injury baseline, early-acute, and late-acute visits in concussed athletes with loss of consciousness, post-traumatic amnesia, and/or retrograde amnesia either (SRC+) and in concussed athletes without loss of consciousness, post-traumatic amnesia, or retrograde amnesia (SRC-). GFAP = glial fibrillary acidic protein; S100B = S100 calcium binding protein B; SBDP150 = alpha-II-spectrin breakdown product 150; UCH-L1 = ubiquitin c-terminal hydrolase-L1; IL-6 = interleukin 6; IL-1RA = interleukin 1 receptor antagonist; CRP = c-reactive protein.

DISCUSSION

Results from this large-scale prospective study showed that UCH-L1, S100B, SBDP150, IL-6, and IL-1RA were significantly elevated at the early-acute phase (i.e., within 6 hours post-concussion) following SRC relative to both pre-injury baseline levels and to two groups of non-concussed athlete controls. Individually, the studied markers showed varying ability to discriminate concussed athletes from uninjured contact and non-contact athletes at the early-acute window (AUCs up to 0.84). When used in combination, however, blood biomarkers showed good-to-excellent discrimination of concussed athletes from contact (combined AUC=0.90) and non-contact controls (AUC=0.90). Moreover, across the whole sample, the addition of these biomarkers improved the discrimination of concussed athletes from non-contact athletes at the early-acute phase above and beyond a common concussion symptom checklist alone. Exploratory logistic regression analyses focusing on concussed athletes with lower symptom burden at the early-acute phase showed that the inclusion of biomarkers improved the discrimination relative to both contact and non-contact controls. In addition, early-acute levels of the inflammatory marker IL1-RA were associated with the duration of symptoms in concussed athletes. Finally, although GFAP levels were not significantly elevated acutely in the SRC group as a whole, early-acute and late-acute levels of GFAP were sensitive to the presence of acute signs of altered mental status (i.e., loss of consciousness, post-traumatic and/or retrograde amnesia). These results highlight the potential diagnostic and prognostic utility of early-acute levels of these markers for the clinical management of SRC.

Blood-based biomarkers for concussion diagnosis

The early-acute elevations of brain injury biomarkers observed in the current work are consistent with prior studies of SRC or other forms of mild traumatic brain injury (mTBI). For example, elevated S100B and UCH-L1 have be reported in multiple samples of athletes with SRC3,5,6,8 and in emergency department (ED) patients with mTBI.17 Current results also extend prior work demonstrating the potential utility of alpha-II-spectrin breakdown products for the acute assessment of brain injury.18–20 Consistent with some prior reports,5,6 we did not observe acute elevations in GFAP across the entire SRC group. GFAP levels were, however, elevated at both acute visits relative to baseline in concussed athletes with either loss of consciousness, post-traumatic amnesia, and/or retrograde amnesia. These results suggest that GFAP may be sensitive specifically to more severe gradients of concussion, consistent with observations of elevated GFAP in ED patients with mTBI.17,21 In addition, if validated, markers sensitive to acute injury characteristics, such as loss of consciousness, may ultimately have clinical relevance in cases where injuries are not witnessed, and thus, circumvent the reliance on patient self-report for loss of consciousness or amnesia status.

In addition to the acute elevation in more traditional blood markers of brain injury, we also report an early-acute elevation in general markers of inflammation (i.e., IL-6 and IL-1RA). Several studies have reported acute elevations in serum or plasma levels of IL-6 and increased IL-1 activity following moderate to severe TBI, 22–26 while elevated IL-6 has also been observed following moderate blast exposure without mTBI diagnosis.27 Growing evidence supports the hypothesis that SRC leads to changes in peripheral, or systemic, inflammation,7,28,29 and we have reported acute elevations in IL-6 and IL-1RA in an overlapping sample.2 Certainly, serum levels of inflammatory markers are sensitive to a wide variety of conditions and injuries and are thus not specific to SRC or brain injury. Therefore, their isolated use as diagnostic markers is not recommended. It is noteworthy, however, that IL-6 and IL-1RA levels in both control groups were more stable across visits than most of the other studied markers, with no significant variations in either control group over time. Furthermore, the AUCs for IL-6 and IL-1RA at the early-acute visit were equivalent to or even surpassed the AUCs of the other studied markers that are putatively more specific to brain injury.

As noted above, most of the studied biomarkers showed fair-to-good discrimination of concussed athletes from both control groups at the early-acute time point (i.e., within 6 hours), with AUCs reaching 0.84. Combined, these markers showed good-to-excellent discrimination (AUCs 0.90). These AUCs compare favorably to those reported for non-symptom reporting measures that are currently used clinically, such as neurocognitive tests and balance measures. For example, a recent large-scale study in collegiate athletes (N>500) reported that objective measures within 6 hours post-concussion, which included scores on a common cognitive screen (Standardized Assessment of Concussion), a common balance measure (Balance Error Scoring System), and age, discriminated concussed athletes from controls with an AUC of 0.73.30

While self-reported symptoms remain the primary basis for rendering clinical diagnoses of concussion, there is growing recognition of the limitations of this approach to approximate the pathophysiological processes that underlie concussion and the potential benefit of objective biomarkers to augment clinical diagnosis of SRC. In that context, the addition of blood-based biomarkers to self-reported symptom severity significantly improved this discrimination of concussed athletes from non-contact controls compared to symptom severity alone at the early-acute visit. When analyses were limited to concussed athletes with fewer acute symptoms, arguably those athletes who present the greatest diagnostic challenge to clinicians, the addition of biomarkers improved the discrimination of concussed athletes from both control groups. Clinically, it is not advised or recommended that blood biomarkers replace current clinical measures used for assessment and diagnosis of concussion. However, the ultimate utility of biomarkers may be in providing objective markers of brain injury that can be used in addition to clinical examination, symptom assessment, neurosensory testing, neurocognitive testing, and other tools that characterize the clinical effects of injury. The use of a blood biomarker panel in this scenario is particularly appealing due to the insensitivity of these markers to symptom reporting biases (i.e., over or underreporting) or practice effects (e.g., neurocognitive testing). For example, biomarkers might prove particularly useful in clinical scenarios where athletes might be motivated to not report or otherwise minimize concussion-related symptoms in order to prevent removal from competition. Moreover, in contrast to self-reported symptoms or neurocognitive testing, blood biomarkers also provide a measure of the physiological, rather than clinical, effects of concussion. Thus, an additional utility of blood biomarkers may be to help identify brain injury in scenarios where symptoms do not fully capture the underlying physiological effects of concussion and complicate clinical diagnosis. Additional large-scale studies are needed, however, before the promise of blood-based biomarkers can be realized.

Blood-based biomarkers for prognosis

In addition to decisions regarding the diagnosis of concussion, there is also great interest in identifying objective prognostic biomarkers for the identification of individuals at risk for prolonged recovery following SRC. Interestingly, the acute increases in the putative brain injury markers (i.e., UCH-L1, S100B, and SBDP150) were not associated with symptom duration. Rather, at the early-acute visit, IL-1RA, a non-specific inflammatory marker, was significantly associated with the number of days that concussed athletes reported symptoms. Prior studies have identified relationships between inflammatory markers in blood and outcome following mTBI. ED admission levels of CRP predicted the presence of post-concussion symptoms at 3-months post-injury in a cohort of adult patients with mTBI.31 Inflammatory markers contributed to a multivariate predictor model for recovery at 3 and 6 months post-injury in an ED sample of mostly mTBI patients.32 Tumor necrosis factor levels at 1–4 days post-concussion were elevated in pediatric mTBI patients that developed persistent symptoms at one-month post-injury.33 Finally, sub-acute levels of two chemokines were associated with recovery in a sample of athletes with SRC.7 In addition, we recently reported that elevated IL-6 within 6-hours post-SRC was associated with the number of days that athletes were symptomatic.2 Although not statistically significant, we observed a similar trend for IL-6 in the current study (p<0.10). Nevertheless, current results add to the emerging data that suggest that peripheral markers of inflammation have potential as prognostic markers for SRC and brain injury in general.

Limitations

The current study has several strengths, including the large sample size, the use of two matched control groups, and the prospective study design that included collection of pre-injury blood samples. There are, however, limitations that should be considered. First, as in all longitudinal studies, some participants did not participate at every time point. The number of participants in each group with available data for each marker is presented in Table 2. Sensitivity analyses demonstrated that symptoms, acute injury characteristics, and duration of symptoms did not significantly differ in injured athletes with and without missing data, demonstrating that missingness did not bias current results. Second, it is unclear the extent to which the current results in male high school and collegiate football players generalize to other sports, female athletes, athletes of different ages, or non-athletes. Third, we conducted several exploratory analyses without correction for multiple comparisons (i.e., the additional logistic regression analyses in the lower tercile of symptoms and the analyses of biomarkers based on acute injury characteristics); we cannot rule out the possibility of false positives for these analyses. In addition, we have previously reported results from an inflammatory panel in a limited subsample of the current cohort. 2 Several markers that were not significant in those analyses were not assayed in the entire, final cohort; thus, there is some selection bias specifically for the inflammatory markers reported here. Future studies are needed to replicate current findings for the inflammatory markers in an independent sample. A post-hoc FDR was applied to all tests in the current manuscript as well as those in those published work on inflammatory markers. All significant associations based on the primary FDR correction remained significant, with the exception of the association between IL-1RA and symptom recovery, which became a non-significant trend (q=0.054) It should be noted that several aspects of the current work are distinct from the prior work, including the inclusion of a non-contact control group, hypotheses relating to the added value of biomarkers in addition to self-report symptoms, the effects of acute injury characteristics on biomarker levels, and the extensive cross-validation. Therefore, primary FDR correction applied in this manuscript are limited to the specific hypotheses of the current manuscript. Finally, markers selected for this study were primarily elevated only at the early-acute post-SRC window, defined as within 6 hours of injury. The clinical utility of specific biomarkers will depend on the kinetics of the biomarker in question and the clinical setting (e.g., sideline versus emergency department versus concussion clinic). Additional studies are needed to determine if the markers in the current study are also sensitive at more immediate post-injury visits. Conversely, the use of additional markers with more delayed or prolonged signals, such as neurofilament light,10 could also increase the clinical utility of any potential blood biomarker panel.

Conclusion and future directions

A combination of brain injury and inflammatory biomarkers in serum show promise as objective diagnostic markers for SRC, while inflammatory markers may also provide prognostic value. Future work is needed to determine the possible clinical cut-off values of best performing biomarkers, the optimal combination of biomarkers, and the validation of potential point-of-care assays for biomarker quantification in order to move these markers from research to clinical tools.

ACKNOWLEDGEMENTS

The authors thank Ashlee Taylor and Brenda Davis from the University of Oklahoma Integrative Immunology Center for analysis of serum inflammatory markers; Ashley LaRoche and Alexa Wild from the Department of Neurosurgery at the Medical College of Wisconsin for study coordination; Carisa Bergner from the Comprehensive Injury Center at the Medical College of Wisconsin and Dr. Aniko Szabo from the Department of Biostatistics at the Medical College of Wisconsin for statistical consultation, and Arthur Weber from Banyan Biomarkers, Inc., for analysis of blood markers.

This work was supported by the Defense Health Program under the Department of Defense Broad Agency Announcement for Extramural Medical Research through Award No. W81XWH-14-1-0561. Opinions, interpretations, conclusions and recommendations are those of the authors and are not necessarily endorsed by the Department of Defense. Support for this work was also provided by the National Institute of Neurological Disorders And Stroke of the National Institutes of Health under Award Number R21NS099789. TM acknowledges additional support from the National Institute of Neurological Disorders And Stroke (R01NS102225) and through a project funded through the Research and Education Program, a component of the Advancing a Healthier Wisconsin endowment at the Medical College of Wisconsin. JS acknowledges support from the National Institute of General Medical Sciences (P20GM121312) and the National Institute of Mental Health (R21MH113871). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The REDCap electronic database and the Adult Translational Research Unit used for this project were supported by the National Center for Advancing Translational Sciences, National Institutes of Health, Award Number UL1TR001436.

POTENTIAL CONFLICTS OF INTEREST

Dr. Bazarian has received research support from Banyan Biomarkers Inc, which develops diagnostic blood tests for traumatic brain injury. Dr. Hayes received salary and stock compensation from Banyan Biomarkers Inc., where he is Founder and Chief Science Officer. All other authors have nothing to report.

REFERENCES

RESOURCES