Proteomic Discovery of Noninvasive Biomarkers Associated With Sport-Related Concussions

Abstract

Background and Objectives

Sport-related concussions affect millions of individuals across the United States each year, and current techniques to diagnose and monitor them rely largely on subjective measures. Our goal was to discover and validate objective, quantifiable noninvasive biomarkers with the potential to be used in sport-related concussion diagnosis.

Methods

Urine samples from a convenience series of healthy control collegiate athletes who had not sustained a concussion and athletes who sustained a concussion as diagnosed by a sports medicine physician within 7 days were collected prospectively and studied. Participants also completed an instrumented single-task gait analysis as a functional measure. Participants were recruited from a single collegiate athletic program and were ≥18 years of age and were excluded if they had a concomitant injury, active psychiatric conditions, or preexisting neurologic disorders. Using Tandem Mass Tags (TMT) mass spectroscopy and ELISA, we identified and validated urinary biomarkers of concussion.

Results

Forty-eight control and 47 age- and sex-matched athletes with concussion were included in the study (51.6% female, 48.4% male, average age 19.6 years). Participants represented both contact and noncontact sports. All but 1 of the postconcussion participants reported experiencing symptoms at the time of data collection. Insulin-like growth factor 1 (IGF-1) and IGF binding protein 5 (IGFBP5) were downregulated in the urine of athletes with concussions compared to healthy controls. Multivariable risk algorithms developed to predict the probability of sport-related concussion showed that IGF-1 multiplexed with single-task gait velocity predicts concussion risk across a range of postinjury time points (area under the curve [AUC] 0.786, 95% confidence interval [CI] 0.690–0.884). When IGF-1 and IGFBP5 are multiplexed with single-task gait velocity, they accurately distinguish between healthy controls and individuals with concussion at acute time points (AUC 0.835, 95% CI 0.701–0.968, p < 0.001).

Discussion

These noninvasive biomarkers, discovered in an objective and validated manner, may be useful in diagnosing and monitoring sport-related concussions in both acute phases of injury and several days after injury.

Trial Registration Information

ClinicalTrials.gov Identifier: NCT02354469 (submitted February 2015, first patient enrolled August 2015).

Classification of Evidence

This study provides Class III evidence that urinary IGF-1 and IGFBP5 multiplexed with single-task gait velocity may be useful in diagnosing sport-related concussion.

The Centers for Disease Control and Prevention estimates that up to 3.8 million people sustain sport-related traumatic brain injuries (TBIs) each year, 80% to 90% of which are concussions.¹ Due to the challenges that clinicians face in determining concussion status,^2,3 a substantial need exists for accurate and unbiased methods by which to diagnose and manage the injury. To date, most commonly used assessments lack appropriate clinical reliability.⁴ Therefore, diagnosis typically relies on assessing a combination of subjective measures such as patient-reported symptoms, balance, and neurocognitive function.^5,6

Recovery monitoring is also important to determine the optimal time to initiate appropriate rehabilitation or return-to-activity protocols.^2,7 The methods commonly used to monitor recovery, however, have demonstrated limited validity, specificity, and test/retest reliability.⁸ Recent studies have attempted to address these limitations by using more objective approaches such as gait analysis.^8-12 Regardless, a substantial clinical need remains for an objective and quantifiable test to diagnose concussion. Noninvasive biomarkers have the potential to meet these needs in an economical way² and are especially attractive in populations in which routine or emergent blood tests can be difficult to obtain for both procedural and compliance reasons.

We have used both a global proteomics and a biologically driven approach to identify and validate potential noninvasive biomarkers of sport-related concussion. Proteomic analysis by Tandem Mass Tags (TMT) mass spectroscopy identified 71 proteins differentially expressed in the urine of individuals with concussions. On the basis of the availability of commercial ELISA kits and the relevance of each identified species to the pathophysiology of concussion, 10 proteins were validated by ELISA. Ultimately, insulin-like growth factor 1 (IGF-1) and IGF binding protein (IGFBP) 5 were determined to be the most promising noninvasive urinary biomarkers of sport-related concussion.

IGF-1 is a peptide hormone produced primarily in the liver in response to growth hormone stimulation but one that can also be made in the brain.¹³ IGF-1 functions in both endocrine and paracrine manners to promote anabolism by binding IGF-1 receptors and initiating phosphoinositide 3-kinase or mitogen-activated protein kinase signaling.¹³ IGF-1 receptors are abundant in the brain,¹⁴ and this protein can modulate neuronal growth, differentiation, proliferation, and survival.^15,16 IGF-1 also plays an important role in the maintenance of myelination and the repair process after brain damage.^17,18

The majority of circulating IGF-1 is bound to IGFBPs, most frequently IGBFP5 and IGFBP3, which promote its stability and half-life in circulation.¹⁷ IGFBP5 is particularly abundant in the brain¹⁷ and can exert both stimulatory and inhibitory effects on IGF-1 signaling.^19,20 Recently, it has been shown that IGFBP5 also has actions that are independent of its interactions with IGF-1. These include a role in cell survival, differentiation, and tissue remodeling through its ability to bind the extracellular matrix.¹⁹

These noninvasive biomarkers provide an objective and quantifiable method for diagnosing sport-related concussion and may have utility in monitoring symptom changes. They may also provide insight into the pathophysiology of sport-related concussion, which may help better inform clinical management.

Methods

Experimental Design

The purpose of this study was to determine whether urinary IGF-1 and IGFBP5 multiplexed with single-task gait velocity may be useful in objectively diagnosing sport-related concussion using Class III evidence. Baseline urine samples from healthy controls were collected prospectively from a convenience series of collegiate athletes during preseason evaluation between August 17, 2015, and August 27, 2017. During the evaluation, all athletes were approached and invited to provide a sample. Additional urine samples were collected from athletes within 1 to 7 days after sustaining a sport-related concussion between November 1, 2015, and November 15, 2017. All participants were recruited from a single collegiate athletic program through the institution's athletic training staff and team physician. Inclusion criteria consisted of being ≥18 years of age, participating on a varsity athletic team, having medical clearance to participate in a sport as determined by medical staff, and having an age- and sex-matched counterpart in the comparison group. Exclusion criteria included having sustained a concussion in the 6 months before the initial examination (excluding the current injury for the concussion group); a history of epilepsy, prior seizure, or any previous abnormal EEG finding or abnormal brain imaging (CT or MRI) finding, as reported by the patient; any unstable medical condition; any preexisting neurologic disorder, including but not limited to stroke, intracranial hemorrhage, any movement disorder, and cerebral palsy; and any currently active psychiatric condition, including specifically major depression, bipolar disorder, or schizophrenia. A history of a psychiatric condition but no ongoing psychiatric episode (e.g., not currently undergoing treatment for a major depressive episode) was not an exclusion. Among the healthy baseline group, 321 athletes were approached for participation in the study, and 21 did not enroll due to not meeting inclusion/exclusion criteria or not electing to participate. Fourteen baseline individuals did not provide a urine sample, and 238 healthy baseline individuals were excluded due to not meeting age- and sex-matching criteria of concussed individuals (Figure 1). Single-task gait testing was also performed at the time of urine collection, as described below. All concussions were diagnosed by a sports medicine physician in the days after a single concussion according to the criteria determined by the International Conference on Concussion in Sport consensus statement.² Physicians were blinded to all biomarker and gait results. No participants were diagnosed with multiple or repeated concussions between the index concussion and the urine and gait assessments.

Figure 1. Flow of Participants Through the Study.

In brief, inclusion criteria consisted of being an active participant on a collegiate varsity sports team, being ≥18 years of age, having a concussion diagnosis within the prior 7 days (concussion group only), and having no previous concussion sustained in the prior year (control group only). We excluded potential participants from the study if they had a concomitant injury, a currently active psychiatric condition, or a preexisting neurologic disorder.

Urine Collection and Storage

Clean-catch urine was collected midstream as we previously reported using alcohol-free Triad Medical-Benzalkonium chloride antiseptic towelettes (Allegro Medical, Mesa, AZ).^21-26 Briefly, after collection, urine samples were immediately placed on ice and frozen at −20°C until analyzed. Samples were thawed and tested for the presence of blood and leukocytes with Multistix 9 strips (Siemens Healthcare Diagnostics Inc, Tarrytown, NY), as we first described.²² Among the 95 samples studied, only 3 control samples and 1 concussion sample were positive for blood. Samples were then divided into aliquots and stored at −20°C for future analysis. All samples were coded and anonymized. Clinical information was accessed only to age- and sex-match healthy control and concussion samples for analysis.

Gait Testing

To obtain a measure of functional ability, participants completed a steady-state gait assessment consistent with protocols described previously.^27,28 In brief, participants walked at a self-selected pace toward a target 8 m in front of them, walked around it, and returned to the original starting position. Participants wore a set of 3 inertial measurement units (Opal Sensor; APDM Inc, Portland, OR) to calculate the average stride velocity of each trial.^27,28

TMT Identification of Differentially Expressed Proteins

Proteins were isolated from human urine samples by a well-established spin-filter cleanup protocol.²⁹ Briefly, urine samples were centrifuged, and the pH was adjusted to neutral. They were then spun in Vivaspin 6 spin filters with anti-human serum albumin to concentrate, purify, and albumin-deplete the proteins. Proteins were harvested from the filter and quantified by the BCA assay (Thermo Fisher Scientific, Waltham, MA). An aliquot of 20 μg was taken from each sample and each internal standard and digested overnight with mass spectrometry (MS)–grade trypsin (Promega, Madison, WI) with a 1:30 ratio of trypsin:total protein in 100 mM triethylamonium bicarbonate. Peptides were labeled with 6-plex TMT (Thermo) according to the manufacturer's instructions.

Mass Spectrometry

Labeled peptides were pooled and cleaned with Oasis SPE (Waters Corp, Milford, MA). Each TMT label group had 1 pooled internal standard sample and 4 experimental standards to enable cross-group normalization and quantification. TMT groups were reconstituted in MS loading buffer composed of 5% formic acid, 5% acetonitrile at a concentration that corresponded to 0.8 μg starting material per 1 μL MS loading buffer. Samples were loaded with an Ekspert nanoLC 400 autosampler (Eksigent, Redwood City, CA) and Ekspert nano LC 415 system (Eksigent) with separate loading and enrichment pumps coupled to a Thermo Q Exactive MS. A picochip system (New Objective, Littleton, MA) was used to trap and elute peptides. The stationary phase was BEH C18, 3 μm and 10 cm. A 60-minute linear gradient from 5% buffer B (0.1% formic acid in acetonitrile) to 30% B was used. Buffer A was 0.1% FA in water. MS settings were as follows: MS1 settings, range m/z 300 to 1,400 Da; resolution 70,000; automatic gain control target 3e6; 50-millisecond maximum ion accumulation time; 1 scan range (single scan); and profile data were obtained. MS2 scans were obtained as follows: a top 10 method was used with a fixed first mass of 100 m/z with dynamic higher mass range, isolation window of 1.8 m/z, automatic gain control target of 2e5, 130-millisecond maximum accumulation time, collision energy of 30 normalized collision energy, and an underfill ratio of 2.0%. Dynamic exclusion was set to 15 seconds, and precursor charge states 2 to 5 were permitted. All samples were acquired in duplicate.

Database Analyses

The UniProt homo sapiens (March 2018) database and Proteome Discoverer (version 2.2) were used for all searches. An initial spectrum files recalibration node search was performed using a precursor mass tolerance of 100 ppm and product ion mass tolerance of 0.2 Da. Recalibrated spectra were searched with Byonic (version 2.2.0.388). Settings were as follows: trypsin (semi), 1 missed cleavage, precursor mass tolerance 10 ppm, and fragment mass tolerance 20 ppm. Modifications were limited to 2 common per peptide. Modifications were TMT6plex n-terminal and lysine, fixed; deamidated asparagine and glutamine, common 1; and oxidation of methionine, common 1. Carbamidomethylation of cysteine residues was fixed. Peptide level score cutoff was set to automatic. Protein false discovery rate was set to 2% (or 50 reverse counts) per raw file. Reporter ion quantification was based on intensity, and a threshold of 50% coisolation was used. TMT channels were normalized according to total peptide amount.

Pathways and Interactive Biological Network Analysis

Ingenuity pathway analysis (Qiagen, Venlo, the Netherlands) was used to analyze the proteins determined to be differentially expressed in the sport-related concussion urine samples by TMT. Using the expression analysis tool, we identified the key canonical pathways enriched in concussion urine samples and built a network based on the overlap of these functions and key players of the pathways. The null hypothesis was that molecules in our dataset did not overlap with a particular pathway. Right-tailed Fisher exact test was used to determine the significance of differential expression, with significance set at p < 0.05.

Enzyme-Linked Immunosorbent Assay

On the basis of the accessibility of commercially available monospecific ELISA kits, 9 of the differentially expressed proteins were validated for their presence in the urine samples of all study participants, including S100A9 (Aviva Systems Biology, San Diego, CA), APP (Abcam, Cambridge, MA), TFF3 (R&D Systems, Inc, Minneapolis, MN), TNC (Abcam), THBS1 (R&D Systems, Inc), MMP7 (R&D Systems, Inc), IGFBP5 (Aviva Systems Biology), VTN (Abcam), and IGF2 (R&D Systems, Inc). All urine samples were also analyzed for IGF-1 (R&D Systems, Inc). ELISAs were performed according to the manufacturer’s instructions. All analyses were conducted in duplicate, and values with coefficients of variation >25 were excluded according to the manufacturer’s recommendations.

Statistical Analysis

Univariate comparisons of urinary biomarkers between athletes 1 to 2 days after concussion and age-matched controls were performed with the Mann-Whitney U test. Receiver operating characteristic curve analysis was used to determine diagnostic performance with area under the curve (AUC) for combinations of predictors using multivariable logistic regression. IGFBP-5 was excluded from the first multivariable logistic regression model because it was not statistically found to be a strong predictor. We next created a multivariable risk algorithm using 3 predictor variables (IGFBP5, IGF-1, single-task gait velocity) for which cutoff values were determined by the Youden J index to maximize the combination of sensitivity and specificity. A decision tree was developed to provide a clinical framework for 8 different probabilities of sport-related assessing concussion risk according to combinations of 3 multivariable predictors (2 urinary biomarkers and gait velocity). Bootstrapping model validation of the multivariable risk model for predicting sport-related concussion using 1,000 bootstrap resamples was conducted with R (R Foundation for Statistical Computing, Vienna, Austria). Statistical analysis was performed with Stata 16.0 (StataCorp LLC, College Station, TX). A 2-tailed Bonferroni-adjusted threshold of p < 0.017 was considered statistically significant to protect against false-positive results due to multiple-group comparisons.³⁰

Classification of Evidence

The purpose of this study was to identify objective urinary biomarkers for the diagnosis of sport-related concussion in collegiate athletes. This prospective study provides Class III diagnostic evidence that urinary IGF-1 and IGFBP5 multiplexed with single-task gait velocity may be helpful in the objective, quantifiable, and noninvasive diagnosis of sport-related concussion.

Standard Protocol Approvals, Registrations, and Patient Consents

All data were collected prospectively with written informed consent according to the institutional bioethical guidelines of the Institutional Review Board at Boston Children's Hospital (IRB-P00016469). The study was registered at ClinicalTrials.gov (identifier: NCT02354469).

Data Availability

Anonymized data not published within the article will be shared by request from any qualified investigator.

Results

A total of 95 urine samples were collected and analyzed, including 47 samples from athletes after concussion and 48 from healthy control athletes who had not experienced a concussion in the year before study enrollment. The participating athletes represented both contact and noncontact sports (Table 1). Urine was collected from the majority of athletes within 3 days of injury. All but 1 of the postconcussion participants reported experiencing symptoms at the time of urine collection. Further clinical information is summarized in Table 2. No adverse events during sample collection were observed.

Table 1.

Characteristics of the Study Groups

Table 2.

Clinical Characteristics of the Postconcussion Study Group

Proteomic Analysis by TMT

TMT, a quantitative MS technology, was used to identify and quantify the urinary proteome of healthy control athletes and those with a sport-related concussion. Seventy-one proteins were significantly differentially expressed (p < 0.05) in the urine of control vs athletes with concussion after the use of a standard 1.5-fold cutoff, and results remained significant after accounting for multiple comparisons (Figure 2A). We used ingenuity pathway analysis to investigate the canonical signaling pathways that were differentially expressed between groups on the basis of the 71 proteins identified by TMT (Figure 2B). These analyses suggest that these differentially expressed proteins are implicated in a variety of biological pathways and functions, including brain development and plasticity, inflammation and the immune response, extracellular matrix remodeling, regulation of lipid metabolism, cellular transport, bone development and remodeling, and coagulation (Figure 2B).

Figure 2. Proteometric Analysis.

(A) Relative quantification of the urinary proteome of athletes with concussions vs healthy age- and sex-matched controls by Tandem Mass Tags. Waterfall plot shows the differentially expressed proteins between athletes after concussion and healthy controls. See eAppendix 1, links.lww.com/WNL/B630 for the key of abbreviations for panel A. (B) Ingenuity pathway analysis of overlapping canonical signaling pathways differentially expressed between control and concussion urine samples. Connected pathways share at least 1 common gene, and the intensity of the node color is proportional to the p value calculated with a right-tailed Fisher exact test, with a darker color indicating a more significant differential expression. Significance was set to p < 0.05. Only the 25 most significant pathways are shown.

Downregulation of Urinary IGFBP5 and IGF-1 in Athletes After Concussion

To validate the findings from the TMT, urine samples from all 95 participants were tested for the presence of S100A9, APP, TFF3, TNC, THBS1, MMP7, IGFBP5, VTN, and IGF2 by ELISA (control n = 48, postconcussion n = 47). IGFBP5, which was downregulated 2.411-fold in the urine of athletes with sport-related concussion by TMT, was found by ELISA to be decreased in athletes after concussion (p = 0.026) (Figure 3A). The remaining 8 proteins analyzed did not show any significant differences between individuals with concussions and healthy controls (data not shown). In addition to using TMT, an unbiased and global proteomics approach to biomarker discovery, we used a biologically driven approach in which we identified candidate biomarkers on the basis of their biological function and relevance to concussion pathophysiology, as well as their biological relationship to candidates discovered via TMT. Therefore, we investigated free IGF-1 as a candidate marker given that its function is modulated in part by IGFBP5 binding. Univariate analysis demonstrated that free IGF-1 was also downregulated in the urine of athletes with concussion compared to controls (p = 0.048) (Figure 3B). No significant sex-specific differences in urinary IGFBP5 or free IGF-1 concentrations were observed. Control levels of IGFBP-5 were similar for males and females (male median 1,002 pg/mL, interquartile range [IQR] 1,660 pg/mL; female median 1,477 pg/mL, IQR 1,471 pg/mL), as were concussion levels (male median 859 pg/mL, IQR 1,257 pg/mL; female median 974 pg/mL, IQR 1,407 pg/mL). Similarly, control levels of IGF-1 did not differ between males and females (male median 0.132 ng/mL, IQR 0.242 ng/mL; female median 0.084 ng/mL, IQR 0.251 ng/mL), nor did concussion levels (male median 0.078 ng/mL, IQR 0.105 ng/mL; female median 0.028 ng/mL, IQR 0.113 ng/mL).

Figure 3. IGFBP5 and IGF-1 Decrease After Concussion and Predict Concussion Risk When Multiplexed With Single-Task Gait Velocity.

(A and B) Quantification of urinary insulin-like growth factor binding protein (IGFBP) and insulin-like growth factor 1 (IGF-1) by ELISA in healthy control athletes (n = 48) vs athletes after concussion (n = 47). Plots show medians and interquartile ranges. (C and D) Multivariable risk algorithm for concussion when single-task gait velocity is multiplexed with urinary free IGF-1 levels. Values presented are probabilities of concussion, determined by multivariable logistic regression analysis. The multiplexed risk algorithm demonstrates excellent predictive ability for concussion (area under the curve 0.786, 95% confidence interval 0.690–0.884).

Multiplexing of IGF-1 With Single-Task Gait Velocity Predicts Concussion Risk

We developed a multivariable risk algorithm to predict the probability of sport-related concussion in the cohort of athletes from 1 to 7 days after injury. We used multivariable logistic regression that considered both urinary free IGF-1 concentrations and single-task gait velocities (Figure 3C). For IGF-1 levels <0.18 ng/mL, the odds ratio [OR] was 7.72 (95% confidence interval [CI] 1.75–88.90, p = 0.008). For a gait velocity of 0 to 1 m/s compared to a reference of >1.15 m/s, the OR was 19.02 (95% CI 2.18–174.87, p = 0.009); for a gait velocity of 1.01 to 1.15 m/s compared to a reference of >1.15 m/s, the OR was 17.29 (95% CI 2.54–119.29, p = 0.004). Of the 95 urine samples collected and analyzed by ELISA, 28 controls (15 female, 13 male) and 37 with concussion (20 female, 17 male) had available data on single-task gait velocity. No significant differences in either biomarker levels or demographics (age, sex, weight, height, and concussion history) were noted between the individuals with and without gait velocity data. IGFBP5 provided no additional predictive information and thus was not included in this model. The model confirmed that low levels of urinary free IGF-1 were associated with a higher risk of sport-related concussion. Furthermore, the postconcussion group walked significantly more slowly than the controls (mean 1.03 m/s, SD 0.11 m/s vs mean 1.17 m/s, SD 0.16 m/s, p = 0.0001). Thus, the risk of concussion was higher for individuals with low to moderate single-task gait velocity, while those with higher gait velocity had much lower risk (Figure 3). The multiplexed risk algorithm demonstrates excellent predictive ability for sport-related concussion (AUC 0.786, 95% CI 0.690–0.884) and provides a range of concussion likelihood that is almost 150-fold (Figure 3).

Urinary IGFBP5 Levels After Injury Are Time Dependent

Because concussion biomarkers can peak or decline at various postinjury time points, we assessed whether IGFBP5 and IGF-1 could detect differences between groups within the first 2 days of injury. Of the 47 concussion samples, 21 were collected within the first 1 to 2 days. In these samples, there was a steep decrease in IGFBP5 concentration compared to controls (p = 0.001) (Figure 4A). Comparatively, for IGFBP5 levels in concussion samples that were collected 3 to 7 days after injury (n = 26, median 1,256 pg/mL, IQR 1,444 pg/mL), there was no significant difference compared to the healthy controls (n = 48, median 1,325 pg/mL, IQR 1,569 pg/mL, p = 0.715). A similar temporal trend was not observed with IGF-1 (Figure 4B).

Figure 4. Multiplexing Biomarkers From Athletes 1 to 2 Days After Concussion Provides Enhanced Capacity to Predict Probability of Concussion.

Urinary levels of (A) insulin-like growth factor binding protein 5 (IGFBP5) and (B) free insulin-like growth factor 1 (IGF-1) as determined by ELISA were compared between healthy control athletes (n = 48) and athletes after concussion whose urine sample was collected during the acute phase of injury (days 1 or 2 after injury, n = 21). Plots show medians an interquartile ranges. (C) Multiplex of markers from patients 1 to 2 days after concussion using logistic regression. For analyses including only free IGF-1 or IGFBP5, concussion n = 21 and control n = 48. For all other analyses, including gait velocity, concussion n = 16 and control n = 29 because gait velocity data were not available for all participants. (D) Receiver operating characteristic curve for multiplex of IGFBP5, free IGF-1, and single-task gait velocity for days 1 and 2 after concussion (area under the curve 0.835, 95% confidence inerval 0.701–0.968, p < 0.001).

Multiplexing of Biomarkers Predicts Concussion Risk

Receiver operating characteristic curves were created to evaluate the discriminatory power of IGFB5 and IGF-1 as acute biomarkers of sport-related concussion in urine samples collected 1 to 2 days after concussion. Alone, IGFBP5 demonstrated excellent discriminatory capability between groups (AUC 0.764, p = 0.001), whereas IGF-1 had minimal discriminatory capability (AUC 0.581, p = 0.280) (Figure 4C). When multiplexed together, however, the sensitivity and specificity of the markers in detecting concussion were enhanced (AUC 0.783, p < 0.001). We next multiplexed these objective markers with gait velocity. Single-task gait velocity data were available from 29 of the 48 control patients and 16 of the 21 patients with concussion whose urine was collected within 2 days of injury. Compared to single-task gait velocity alone (AUC 0.753, p = 0.006), inclusion of these 2 objective biomarkers increased the discriminatory capability in concussion diagnosis (AUC 0.835, p < 0.001) (Figure 4, C and D).

A multivariable risk algorithm for the prediction of sport-related concussion risk in the first 2 days after injury was developed by creating cut points in the levels of IGFBP5, IGF-1, and single-task gait velocity. This was used to develop a clinical decision tree to predict the likelihood of sport-related concussion based on the 3 measures (Figure 5). Patients with lower levels of urinary IGFBP5 and IGF-1 and slower single-task gait velocity within 1 to 2 days after injury had a higher probability of concussion. Furthermore, by multiplexing these 3 variables, our model predicted the likelihood of concussion over an almost 40-fold range of probabilities (Figure 5). To determine the validity of our model, internal bootstrapping of the multivariable risk model to predict concussion was performed. This analysis determined large values for the bias-corrected Somers D rank correlation of 0.61 and Nagelkerke R² of 0.27. The AUC for our risk score is 0.835 (95% CI 0.700, 0.973). The bias-corrected intercept and slope of the logistic calibration equation were −0.04 and 0.77, respectively. The bias-corrected Brier score was 0.17. Therefore, the bootstrap resampling results indicate good internal validity of the full multivariable predictive model.

Figure 5. Clinical Decision Tree to Predict Probability of Concussion.

Clinically relevant cutoffs for insulin-like growth factor binding protein 5 (IGFBP5), insulin-like growth factor 1 (IGF-1), and gait velocity were determined with receiver operating characteristic curve analysis based on the J index to maximize the combination of sensitivity and specificity. The decision tree created by multiplexing these 3 measures can be used to predict the probability of concussion in athletes 1 to 2 days after injury. As above, analysis includes data only from patients with concussion 1 or 2 days after injury and for whom data are available for all 3 variables (concussion n = 16; control n = 29).

Discussion

Concussions affect millions of individuals each year, yet the ability to precisely and accurately diagnose this injury has been a longstanding challenge. The goal of this study was to identify and validate a set of objective and quantifiable noninvasive urinary biomarkers to be used in sport-related concussion diagnosis.

The discovery of these markers was grounded in an objective, global proteomics approach using state-of-the-art TMT complemented by a biologically driven discovery approach, as described previously.³¹ As a quantitative method of MS, TMT permits both the identification and quantification of the proteins present in our urine samples. This approach to identify differentially expressed proteins in biofluids has been validated as a reliable, reproducible method for biomarker discovery.^21,32,33

Using monospecific ELISAs, we demonstrated that levels of both free IGF-1 and IGFBP5 were decreased in the urine of athletes after concussion compared to healthy age- and sex-matched controls. Disease biomarkers can be upregulated or downregulated as a function of their physiologic context. One might speculate that this difference could be due to the possibility that these proteins may be involved in brain repair after concussion and are excreted at relatively lower levels at the time of injury. These results are consistent with the literature, which has shown that circulating serum IGF-1 levels in both human and rodent models are often decreased after TBI.^14,17,34 Moreover, pituitary dysfunction and disruption of the growth hormone/IGF-1 axis after concussion have been reported in instances of blast-induced concussion.³⁵ However, we now report the presence of these alterations in free IGF-1 in urine. It is important to note that the use of IGF-1 supplementation has been investigated in the treatment of both TBI and ischemic brain injury with the goal of augmenting the neuroprotective effects of IGF-1,^15,16 suggesting that IGF-1 may play a critical role in the ability of the brain to heal after concussion and may represent a means by which to manage this injury.

IGFBP5 has been associated in the recovery of the brain from severe hypoxic-ischemic injury in rats, suggesting that it may play a role in brain injury repair.³⁶ Furthermore, its expression in the brain during development is regulated by IGF-1,³⁷ suggesting cross-talk between the 2 molecules, because IGFBP5 plays a large role in modulating IGF-1 function. The current study supports a potential important connection between IGBFP5 and concussion pathophysiology in humans.

We also investigated the predictive value of urinary biomarkers for sport-related concussion diagnosis alongside average walking speed (i.e., gait velocity). Gait velocity is measured with inertial sensors placed on the body, and as an objective measurement of gross motor control, it demonstrates high test-retest reliability^38,39 and has been used to distinguish between participants with concussion and controls immediately after concussion.^27,40 While not widely used after sport-related concussion, a recent clinical practice guideline suggests that gait evaluations may assist clinical decision-making,⁴¹ potentially using sensors embedded within everyday devices such as smartphones.¹² When multiplexed with gait velocity, urinary IGF-1 discriminated between healthy controls and athletes after concussion with excellent sensitivity and specificity across postinjury time points. Our multivariable risk model provided an almost 150-fold range of probabilities in the likelihood of concussion and may provide a noninvasive tool that could be broadly applied to concussion diagnosis at a variety of time points post-injury.

Our work also revealed that urinary IGFBP5 levels decrease most significantly 1 to 2 days after injury, suggesting that urinary IGFBP5 may be a marker of acute injury and could represent a method to noninvasively diagnose sport-related concussions. The ability to make early diagnoses is crucial to injury management.² Alone and when multiplexed with IGF-1 in urine samples collected 1 and 2 days after injury, urinary IGFBP5 demonstrated significant ability to discriminate between patients with concussion and healthy controls at acute time points. By multiplexing these objective and noninvasive biomarkers with single-task gait velocity, we have developed a test that is substantially more sensitive and specific in diagnosing sport-related concussion than currently available options such as balance and cognitive measures (AUC 0.58 and 0.62, respectively).⁴² While clinical symptoms alone provide essential diagnostic information in diagnosing sport-related concussion, they do not take into consideration a number of confounding variables. For example, female athletes, younger athletes, and those who have a history of attention-deficit/hyperactivity disorder report more frequent and severe concussion symptoms at baseline, which can confound the results of a test that relies solely on patient-reported symptoms.⁴² The unbiased and quantitatively objective nature of our test can allay such concerns. Furthermore, our multivariable risk algorithm provides a clear and simple clinical decision tree that has the ability to predict the likelihood of concussion over a 40-fold range of probabilities using objective data, which has the potential to improve the accuracy of sport-related concussion diagnoses.

The utility of an objective, noninvasive, fluid-based biomarker for concussion diagnosis has been widely acknowledged, and various candidate markers have previously been identified in the CSF and blood.⁴³ Many of these investigated candidate biomarkers are products of axonal and glial injury such as neurofilaments, tau, and glial fibrillary acidic protein, as well as markers of inflammation. None of these biomarkers have been clinically approved to predict the occurrence of a concussion. However, 1 blood test has been approved for clinical use as a means to determine whether a CT scan should be conducted when evaluating a head injury.⁴⁴ It is important to acknowledge, however, that, unlike urine, both CSF and serum require invasive methods of sample collection.

To date, few published studies have been reported using urine as a liquid biopsy for concussion. One study demonstrated the presence of elevated marinobufagenin in the urine of football players with concussion compared to healthy controls, but because very few postconcussion samples were included in the study (n = 6), this finding merits further investigation.⁴⁵ A panel of dysregulated metabolites in the urine of patients after concussion using nuclear magnetic resonance spectroscopy has been reported.⁴⁶ However, to the best of our knowledge, these data have not yet been validated.

The current study represents the first demonstration of an objective and validated opportunity to diagnose and monitor sport-related concussion. To date, most commonly used diagnostic assessments rely on a combination of subjective measures such as patient-reported symptoms, balance, or neurocognitive function, which have been shown to lack significant clinical reliability.^4,47 Therefore, the need for an objective diagnostic is critical. Here, we compared healthy control collegiate athletes to separate collegiate athletes who had sustained a concussion. Further validation studies should be performed to determine how the levels of urinary biomarkers might change after concussion within the same individual. Given that the current study focused specifically on sport-related concussions in college-aged athletes, larger validation studies in a broader population are necessary for these results to be generalized to other patient populations. Given that standard neuroimaging sequences are not typically used for the care of acute concussion because they are not useful in identifying postconcussion changes,^48,49 no imaging results were available in this study. In addition, information about substance or alcohol use was not collected in the study from which these samples were derived. The noninvasive nature of our markers will allow frequent, painless, economical monitoring, along with the ability to test biosamples in real time, features that would be beneficial in injury monitoring and return-to-activity decision-making.

Glossary

AUC

area under the curve

CI

confidence interval

IGF-1

insulin-like growth factor 1

IGFBP

IGF binding protein

IQR

interquartile range

MS

mass spectrometry

OR

odds ratio

TBI

traumatic brain injury

TMT

Tandem Mass Tags

Appendix. Authors

Study Funding

This study was supported by the Eastern Athletic Trainers Association Foundation, the Eunice Kennedy Shriver National Institute of Child Health & Human Development (R03HD094560), the National Institute of Neurologic Disorders and Stroke (R01NS100952, R03HD094560, and R43NS108823), and the MINDSOURCE Brain Injury Network. This study was registered at ClinicalTrials.gov (NCT02354469).

Disclosure

C.C. Daisy, S. Varinos, K. Kaplan, F. Wang, B. Berkstresser, R.S. Lee, J. Froehlich, D. Zurakowski, and M.A. Moses report no disclosures relevant to the manuscript. D.R. Howell reports grants from Eunice Kennedy Shriver National Institute of Child Health & Human Development, National Institute of Neurologic Disorders And Stroke, MINDSOURCE Brain Injury Network, and The Tai Foundation. R. Mannix reports funding by the Department of Defense (DOD W81XWH1920011) and the NIH (U01NS096835, 1R01NS115942) and philanthropic support from the National Hockey League Alumni Association through the Corey C. Griffin Pro-Am Tournament and a grant from the National Football League. W.P. Meehan reports royalties from ABC-Clio publishing for the sale of Kids, Sports, and Concussion: A Guide for Coaches and Parents and Concussions; from Springer International for Head and Neck Injuries in Young Athlete; and from Wolters Kluwer for working as an author for UpToDate. His research is funded in part by philanthropic support from the National Hockey League Alumni Association through the Corey C. Griffin Pro-Am Tournament and a grant from a grant from the National Football League. Go to Neurology.org/N for full disclosures.