Preliminary Report: Localized Cerebral Blood Flow Mediates the Relationship between Progesterone and Perceived Stress Symptoms among Female Collegiate Club Athletes after Mild Traumatic Brain Injury

Abstract

Female athletes are under-studied in the field of concussion research, despite evidence of higher injury prevalence and longer recovery time. Hormonal fluctuations caused by the natural menstrual cycle (MC) or hormonal contraceptive (HC) use impact both post-injury symptoms and neuroimaging findings, but the relationships among hormone, symptoms, and brain-based measures have not been jointly considered in concussion studies. In this preliminary study, we compared cerebral blood flow (CBF) measured with arterial spin labeling between concussed female club athletes 3–10 days after mild traumatic brain injury (mTBI) and demographic, HC/MC matched controls (CON). We tested whether CBF statistically mediates the relationship between progesterone serum levels and post-injury symptoms, which may support a hypothesis for progesterone's role in neuroprotection. We found a significant three-way relationship among progesterone, CBF, and perceived stress score (PSS) in the left middle temporal gyrus for the mTBI group. Higher progesterone was associated with lower (more normative) PSS, as well as higher (more normative) CBF. CBF mediates 100% of the relationship between progesterone and PSS (Sobel p value = 0.017). These findings support a hypothesis for progesterone having a neuroprotective role after concussion and highlight the importance of controlling for the effects of sex hormones in future concussion studies.

Introduction

Concussion is a form of mild traumatic brain injury (mTBI) that accounts for the majority of the 300,000 sports-related brain injuries among high school and college athletes in the United States each year.¹ Repetitive head impacts, whether resulting in a clinically diagnosable concussion or not, may put athletes at higher risk for cumulative functional deficits later in life.^2–4

Neuroimaging has consistently detected changes in the brains of concussed subjects both in the acute post-injury phase and during recovery.^5–8 In fact, brain changes persist after post-concussion symptoms have resolved,⁹ suggesting that neuroimaging may offer increased sensitivity to study injury recovery. Cerebral blood flow (CBF) is emerging as a useful metric for studying concussive injury and recovery, mainly because of its tight coupling to neuronal activity and involvement in inflammatory processes.^10–15 In the acute phase of concussion, CBF is often reduced, and remains low during the recovery process.^5,9,16,17 CBF is significantly correlated with symptom severity and cognitive performance,^7,17 providing an objective and quantitative brain-based complement to clinical assessment.

Collegiate athletes can be generally divided into those participating in varsity and those participating in club teams. According to the National Collegiate Athletic Association, the estimated number of varsity athletes in the United States is 460,000,¹⁸ whereas >2,000,000 athletes participate in collegiate clubs.¹⁹ Despite their much larger number, club athletes tend to be less studied than their varsity counterparts, as clubs receive fewer resources and medical supervision, as well as less education about sports-related concussion.²⁰

Independent of the varsity versus club divide, athletes can also be divided by sex. As of 2004–2005, ∼ 41% of collegiate females participated in sports,²¹ which translates into roughly 200,000 female athletes. However, the majority of concussion studies to date have focused on male athletes.¹⁵ Those studies that have included females have either grouped them together with males or have not accounted for sex-specific factors such as hormonal fluctuations, which may lead to inconsistent findings.^22–24 Therefore, a significant need exists for studies involving female athletes, especially because female athletes appear to have a higher injury prevalence and longer recovery times.^25–28

A physiological factor that might account for gender differences is hormones. Females experience significant fluctuations in the levels of estrogen and progesterone during the menstrual cycle (MC). These gonadal hormones have widespread non-reproductive functions in the central nervous system (CNS)²⁹ and have neuroprotective effects in various disorders including ischemia, TBI, and spinal cord injury in both animal models and humans.^30–33 These putative neuroprotective effects emphasize the importance of considering MC phase when mTBI occurs.

Further complicating the picture is the widespread use of hormonal contraceptives (HC), which typically use synthetic forms of estrogen and/or progesterone to inhibit the natural cycling of their levels to prevent pregnancy. It is estimated that four out of five sexually active women use oral hormonal contraceptives, although use of other hormonal contraceptive methods are on the rise.³⁴ HC users have lower post-concussive symptom severity than non-HC users.^28,35 Although HC use does not affect length of recovery,^28,35 it does appear to be associated with better cognitive performance after injury.³⁶

MC-based hormonal fluctuations are reported to influence neuroimaging outcomes, which complicates data interpretation. For example, patterns of brain activations differed across the MC for emotional processing, verbal memory and visual-spatial tasks.^37–43 Structural and connectivity differences across the MC have also been reported.^44,45 Similarly, neurometabolite concentrations are reported to fluctuate across the MC and may contribute to changes after exposure to head acceleration events (HAEs).⁴⁶ Our previous work encouraged controlling for MC phase in studies of reward/aversion processing^47–49 or comparison between clinical groups with structural brain imaging;⁵⁰ however, this practice has not yet been applied to studies of concussion in female athletes.

The current research sought to move past studies that controlled for MC effects^47–50 to investigate whether levels of progesterone, a gonadal hormone that has demonstrated neuroprotective effects in brain injury,^51–53 influences post-concussive symptom severity in women during the acute phase of mTBI (i.e., 3–10 days post-injury). We incorporated a rigorous MC tracking and HC user/non-HC user matching technique for this project. Experimental procedures involved arterial spin labeling (ASL), a non-invasive neuroimaging technique for measuring CBF, which was used to test its potential mediation of the relationship between serum progesterone level and post-concussive symptom severity. Evidence of strong statistical mediation by CBF of a negative relationship between progesterone and post-concussive symptom severity would support the hypothesis that progesterone can have a neuroprotective role in concussion. We used a stepwise approach to reveal brain regions that have a three-way relationship among progesterone, symptom score, and CBF. Mediation analysis was then applied to determine whether there is a directed pathway (CBF as mediator and progesterone as independent variable [IV]) for the interactions, using a negative control where these roles were switched.

Methods

Subjects

At the beginning of each sports season, Northwestern University female club athletes were approached and asked to fill out an online screening form, which collects demographic information and self-report of MC start dates for the previous three cycles. Subsequently, injured athletes were identified by the university's athletics department and referred to the study team for study enrollment. Athletes provided written informed consent in compliance with guidelines of the university's internal review board. Study visits were divided into 2 days, when study assessments included: Perceived Stress Scale (PSS),⁵⁴ Post-Concussive Symptom Scale (PCSS),⁵⁵ and the Beck Depression Inventory II (BDI-II).⁵⁶ We further assessed caffeine consumption as well as alcohol and cannabis use for the past 30 days using the timeline followback (TLFB) method.⁵⁷ Briefly, the TLFB method is an online, calendar-guided questionnaire that collects self-reported substance use for the past 30 days. It has been shown to be highly reliable.^58–60 For this study, we further adapted it to the evaluation of menstrual cycle, as has been done by others for neuroimaging studies.^47–49,61 These assessments were collected on Day 1, and blood sampling and magnetic resonance imaging (MRI) were completed on Day 2. Study visits were scheduled within 3–10 days post-injury. Sixteen injured athletes were identified and recruited for this study, but only 15 of them completed the MRI data collection because one subject declined MRI.

Sixteen control subjects were recruited from non-collision sports teams and matched based on age, ethnicity, handedness, and contraceptive use/type. For MC matching purposes,⁶² all subjects were divided into three groups: (1) non-HC users with regular MCs, (2) non-HC users with irregular MCs and (3) HC users. For non-HC users, days 1–7 of MC were estimated as the follicular phase, and days 20 or higher were estimated as the luteal phase. Based on the 3-month TLFB (MC tracking self-report), we scheduled the matched control subjects to be studied during the same MC phase as the mTBI athlete was scanned. For example, if the mTBI athlete was scanned during the follicular phase, the matched control was scheduled during days 1–7 of her MC. Users of HC such as oral contraceptives or NuvaRing do not have normal MC, as their hormone levels are suppressed; therefore, their cycles were divided into active and inactive phases, depending on whether they were on the active hormone pills or with NuvaRing inserted, or on placebo pills or with NuvaRing not inserted. Control subjects were scheduled within 2 days of the matched mTBI athlete's pill pack or NuvaRing day. All athletes on contraceptives were on synthetic hormones, as this is the only form of contraceptive that is approved by the Food and Drug Administration (FDA) for use within the United States.

MRI acquisition

Imaging data were acquired on a 3.0T whole body Siemens Prisma scanner (Erlangen, Germany), using a 64-channel head/neck receive-only coil. High resolution, T1-weighted anatomical images were collected using three-dimensional magnetization-prepared rapid acquisition with gradient echo (3D-MPRAGE) with the following parameters: repetition time (TR)/echo time (TE)/inversion time (TI)/flip angle (FA) = 2300 ms/2.94 ms/900 ms/9 degrees; 176 sagittal slices; 1 mm isotropic resolution; parallel aquisistion techniques (iPAT) acceleration factor = 2; iPAT reference lines = 38 (total duration = 5 min 37 sec). ASL data were collected using a two-dimensional echo-planar imaging (2D-EPI) acquisition, using pseudo-continuous labeling.⁶³ Other parameters include: TR/TE = 4500 ms/12 ms, label duration (τ) = 1.5 sec, post-labeling delay (PLD) = 1.8 sec, labeling plane offset = 90 mm from center of imaging slices, resolution 3.4 × 3.4 × 6 mm³, 24 slices with 1.5 mm gap acquired in ascending order, iPAT acceleration factor = 2, 35 pairs of interleaved control, and tag images (total duration = 5 min 29 sec).

CBF mapping

Imaging data were processed using in-house scripts written in Matlab R2016a (Mathworks, Natick, MA) with Statistical Parametric Mapping SPM8 (Wellcome Department of Imaging Neuroscience, London, UK). All ASL data were motion corrected with the first image of the series as the reference and then co-registered to the high resolution anatomical image. Perfusion weighted images were generated by pairwise subtraction between control and tag images and averaged over the entire time series. Images were converted to quantitative CBF units in mL/100 g/min using the single-blood-compartment model:⁶⁴

$$f=\frac{\lambda \cdot \Delta M\cdot e^{PLD∕T_{1b}}}{2\alpha \cdot M_0\cdot T_{1b}\cdot \left(1-e^{-\frac{\tau }{T_{1b}}}\right)}$$

where f represents CBF in quantitative units, ΔM is the perfusion weighted signal, λ is the blood/water tissue partition coefficient (assumed to be 0.9 g/mL⁶⁵), α is the inversion efficiency assumed to be 0.85, ⁶³ M₀ is the equilibrium magnetization, estimated from the mean of all the control images, and T_1b is blood T1 assumed to be 1664 ms.⁶⁶ The quantitative CBF maps were then transformed to Montreal Neurological Institute (MNI) template space and up-sampled to 1.5 mm isotropic resolution based on the transformation matrix calculated from the high-resolution anatomical image using VBM8.⁶⁷

Given the poor spatial resolution of the ASL acquisition, partial volume correction (PVC) is necessary to minimize contamination of ASL signal from different tissue types. Tissue probabilities from segmented gray and white matter (WM) maps of the high-resolution anatomical image were used to calculate true gray matter (GM) CBF based on the following equation, where the GM flow was assumed to be 2.5 times that of WM flow, and P_GM and P_WM refer to GM and WM probabilities respectively.⁶⁸

$$f_{corr}=\frac{f_{uncorr}}{\left(P_{GM}+0.4\ast P_{WM}\right)}$$

The equation above results in artifactual hyperperfusion in areas where both P_GM and P_WM are low (such as near the cortical–skull interface); therefore, only voxels with at least 30% GM were corrected.^68,69 Voxels with <30% GM were set to zero. This threshold also limits the final analysis to areas with >30% GM, which is more robust because the single PLD ASL method used here is mainly optimized for healthy GM.⁶⁴ PVC maps were smoothed with 8 mm full width at half maximum (FWHM) kernel before entering into statistical analysis. All statistical analysis involving CBF maps used each subject's global GM CBF (gCBF), a scalar value calculated from the mean of all GM voxels, for proportional scaling each subject's CBF map to a global value of 50 mL/100 g/min. This step is necessary to minimize the contribution of inter-subject variability in gCBF.⁷⁰ Subsequent analyses were performed on regional CBF (rCBF), the average CBF across all voxels in the significant clusters detected in the voxelwise statistical analyses, adjusted by the proportional scaling factor based on gCBF.

Blood draw

For assessment of progesterone level, 1 mL of blood was collected by trained nurses of the Northwestern Memorial Hospital outpatient Clinical Research Unit and analyzed at the Northwestern Memorial Hospital Clinical Lab. All except four subjects had blood draws between 2 p.m. and 6 p.m. The remaining four subjects had blood draws before noon. This was because of scheduling challenges, as the blood draws were performed by different nurses in the Clinical Research Unit, and their availability was limited.

Statistical analysis

mTBI versus controls (CON)

Given the pilot sample size and multiple symptom scores, we sought to increase our statistical power by computing a symptom composite score from symptom scores that were significantly different between groups, namely the total PCSS and PSS. This composite score was computed in two steps. First, for each of these scores, a z-score was computed by subtracting the group mean of the CON from each subject's score and dividing by the standard deviation of the CON. Second, the composite score was calculated as the average of these z-scores. Standard two-sampled t tests were then used to determine group differences in the individual symptom and composite scores.

For analysis of brain imaging data, a stepwise process was used within the Statistical Parametric Mapping (SPM8, Wellcome Institute) software suite. We first identified regions in the brain where the mTBI and CON groups differed in CBF using a voxelwise, two-sampled t test. Given the exploratory nature of this analysis on a small sample, we used a liberal threshold of voxelwise p = 0.05, a cluster of 10 adjacent voxels to identify potential clusters for subsequent processing.

General framework

We deployed a multi-step statistical process with three steps: (1) identifying two-way associations that met p value thresholds <0.05 to reduce the number of potential variable combinations tested; (2) assessing if the conjunction p value from three-way associations was less than the Bonferroni correction for all tests done before them (including prospective mediation/moderation tests); and (3) mediation testing with mediation effects >50% and Sobel p value <0.05. Because of the substantial number of component prerequisite effects in a traditional mediation approach, even when the percent mediated effect is very large, some of the component prerequisite effects are not as large; and it is too easy for one of these to miss even though the study is more than adequately powered for detecting large mediation effects (Step 5 in Mediation section). Two caveats must be noted: (1) this multi-step process was adapted given the small sample size and potential for large indirect effects; and (2) all results must therefore be considered exploratory and requiring replication for both confirmation and more precise effect size estimation.

Two-way associations

As the main focus of this study was to determine if there were three-way relationships among regional CBF, progesterone levels, and symptom scores, which could then go into mediation testing, we first computed Pearson's correlation coefficients between symptom scores and progesterone levels to determine the appropriate symptom score for imaging-based statistics. Number of days before return to play and days to symptom resolution were regressed against symptom scores and progesterone levels to determine whether these variables should be included as covariates. Because neither of the relationships approached significance (data not shown), results without these two variables are reported subsequently.

Within areas with salient group differences in CBF, we then used voxelwise multiple regression to identify regions where there was a significant relationship between symptom score and CBF for the mTBI subjects. Each subject's global CBF level and the number of days from injury to imaging were included as covariates. Clusters with at least 20 adjacent voxels or more and voxelwise p < 0.005, having positive or negative relationship with symptom scores, were considered for mediation analysis, which was corrected for multiple comparisons.

Mediation analysis

Clusters where rCBF significantly correlated with both symptom score and progesterone levels (i.e., three-way association) were entered into a directed mediation analysis, where progesterone was designated as the IV, cluster rCBF as the mediator (M), and symptom score as the dependent variable (DV). The following steps, as applied and tested with other imaging studies,⁷¹ were used to test for significance at p < 0.05 for each step.

• Step 1 (Path A): M = β₀ + β_1A (IV) + ϵ

• Step 2 (Path B): DV = β₀ + β_1B (M) + ϵ

• Step 3 (Path C, model 1)): DV = β₀ + β_1,1C (IV) + ϵ

• Step 4 (Path C, model 2): DV = β₀ + β_1,2C (IV) + β_2,2C (M) + ϵ

• Step 5: Use Sobel's test to determine if β_1,2C is significantly lower than β_1,1C. If p_Sobel < 0.05, then the mediation was considered significant.

Mediation analyses were conducted in SPSS Version 26, utilizing the mediation testing from Hayes.⁷² In this study, mediation used three-way associations where a set of common variables (progesterone levels, rCBF, and symptom scores) demonstrated negative or positive associations to each other, with each association meeting a significance threshold of p < 0.05. All corrections for multiple comparisons were performed on these three-way associations up front in the following manner. A conjunction analysis was computed for p_a x p_b x p_c, where p_subscript represents the p value for each pathway (negative or positive relationship between two variables) in the three-way association. For each three-way association submitted to mediation analysis, the conjunction p value was compared with a threshold set as the Bonferroni-corrected p value (p[Bonf]), based on p < (0.05)³ divided by the number of contrasts tested. This p(Bonf) was therefore the following: there were three relationships between CBF and progesterone (Path A), four relationships between rCBF and PSS_Z (Path B), and three symptom score-progesterone level relationships (Path C, model 1) tested. This resulted in 3 × 4 × 3 = 36 relationships being tested. This value was then multiplied by 2 for testing directed mediation versus control mediation, resulting in 72 tests run. Our p(Bonf) was therefore set as (0.05 × 0.05 × 0.05)/72 = 0.000125 /72 = 1.74 × 10⁻⁶. As noted, directed mediation tested progesterone as the IV and rCBF as the mediator, and control mediation tested rCBF as the IV and progesterone as the mediator. This control analysis was done to ensure that the mediation was no longer significant when the IV and M were switched (i.e., progesterone does not mediate the relationship between rCBF and symptom score).

Results

mTBI versus CON

Descriptive statistics of subject demographics, symptom scores, and progesterone levels are shown in Table 1, along with testing of differences across groups. Of the original sample of 15 mTBI and 15 CON athletes who completed MRI scans, one mTBI athlete did not have usable ASL data and was excluded with her matched control. Another pair of mTBI/CON athletes was assessed during the luteal phase of their MC, when their progesterone levels were more than 10 times higher than the rest of our cohort. Given the small sample size of this study, this pair of athletes was excluded from the analysis to increase the homogeneity of the sample in terms of progesterone levels, resulting in a final sample of 26 club female athletes (mTBI n = 13, CON n = 13). Both total PCSS and PSS raw scores, and the symptom composite Z-score, were significantly higher in the mTBI group.

Table 1.

Summary of Demographics, Symptoms and Outcomes, and Differences between Groups

	mTBI (n = 13)	CON (n = 13)	p value
Age (M ± SD)	20.2 ± 1.5	20.1 ± 1.1	n.s.
No. hormonal contraceptive users	7	7	n.s.
Total PCSS (raw, range 0–49)	21 ± 15	4 ± 4	0.003
PSS (raw, range 0–26)	15 ± 6	8 ± 5	0.002
BDI (raw, range 0–22)	4 ± 6	1 ± 1	n.s.
Symptom composite (Z-score)	2.59 ± 2.02	0 ± 0.9	0.008
Return to play (days)	22 ± 14	-	-
Days to symptom baseline	1 ± 5	-	-
Days from injury to scan	7 ± 2	-	-
Progesterone (ng/mL)	0.48 ± 0.27	0.43 ± 0.32	n.s.

mTBI, mild traumatic brain injury; CON, control; M ± SD, mean ± standard deviation; n.s., group comparisons are not significant at p < 0.05 level; PCSS, Post-Concussive Symptom Scale; PSS, Perceived Stress Scale; BDI, Beck Depression Inventory.

Association of progesterone level with symptom scores

For each group, we assessed the relationship of progesterone level to standardized symptom scores. Table 2 shows the Pearson's correlation coefficients (p values) for groupwise correlations between individual symptom Z scores and the composite Z score with progesterone levels. Only the PSS_Z score was significantly correlated with progesterone levels in the mTBI group (r = -0.586, p = 0.035); the negative association indicated that as progesterone levels were higher, perceived stress symptoms were lower or less severe. Based on this finding, all subsequent imaging and mediation results are reported in reference to PSS_Z.

Table 2.

Pearson's Correlation Coefficients (p Values) between Standardized Symptom Scores (Z Score) and Progesterone Levels

Symptom Score	mTBI	CON
PCSS_Z	-0.294 (0.329)	-0.007 (0.983)
PSS_Z	-0.586 (0.035)	-0.3 (0.319)
Composite_Z	-0.397 (0.179)	-0.1 (0.745)

Symptom scores were converted into Z scores based on the mean symptom scores of the control (CON) group. Correlation between PSS_Z of the mild traumatic brain injury (mTBI) group and progesterone level was the only one significant at p < 0.05, highlighted in bold.

PCSS, Post-Concussive Symptom Scale; PSS, Perceived Stress Scale.

Two-way associations with neuroimaging

Figure 1 shows clusters with significant correlation between voxelwise CBF and PSS_Z, overlaid onto a single subject's high resolution anatomical image in MNI template space. All images are in neurological convention, with left side of the brain on the left side of the image. Red clusters represent positive correlations; that is, higher symptoms associated with higher rCBF, and blue clusters represent negative correlations. This analysis was limited to voxels where there were significant differences in CBF between the mTBI and CON groups (clusters shown in Supplementary Figure S1, summarized in Supplementary Table S1). A summary of the clusters with significant correlation to PSS_Z, including number of voxels for each cluster, MNI coordinates of the voxel with the highest T value, anatomical labels based on the Harvard–Oxford cortical (48 regions) and subcortical (21 regions) atlases,⁷³ and a probabilistic cerebellar atlas with 28 anatomical regions⁷⁴ is shown in Table 3.

FIG. 1.

Clusters where cerebral blood flow (CBF) in the mild traumatic brain injury (mTBI) group correlated with perceived stress score (PSS)_Z score, overlaid onto a single subject's T1 in Montreal Neurological Institute (MNI) space. Red represents positive correlation; that is, higher PSS_Z score associated with higher CBF, and blue represents negative correlation; that is, higher PSS_Z score associated with lower CBF. Clusters were generated by thresholding the results at p < 0.005, at least 20 voxels for each contrast (positive and negative correlations with PSS_Z score), and only in regions where a significant group difference in CBF was detected. Color image is available online.

Table 3.

Summary of Clusters with Significant Correlation between CBF and the PSS_Z Score in the mTBI Group, Masked by Areas with Significantly Different CBF between mTBI and CON Groups

+PSS, days to scan, proportional
Nvoxels	T	equivZ	p(unc)	x,y,z (mm)	Anatomical label	mTBI vs. Prog	CON vs. Prog
21	5.534	3.663	0.0001	-36, -51, 46	Left SPL (59%), post. Left supramarginal (26%)	-0.562 (0.046)	0.028 (0.929)

-PSS, days to scan, proportional
Nvoxels	T	equivZ	p(unc)	x,y,z (mm)	Anatomical label	mTBI vs. Prog	CON vs. Prog
42	4.812	3.385	0.0004	-54, -48, 1	Left MTG, temporoocc.(80%), OUTSIDE (18%), post. Left MTG (2%)	0.713 (0.006)	0.023 (0.94)
27	4.179	3.107	0.0009	-58, -3, -11	Sup. Left STG (67%), ant. Left MTG (26%)	0.654 (0.015)	0.228 (0.455)
32	3.767	2.905	0.0018	-8, -48, 42	Left precuneus (59%), Right precuneus (27%), post. Left cingulate (12%), post. Right cingulate (2%)	0.506 (0.077)	-0.286 (0.343)

Clusters were generated using a p value threshold of 0.005 and a cluster threshold of 20 voxels. The two rightmost columns show Pearson's correlation coefficients and p values in parentheses for correlations between rCBF and progesterone levels for each group. Correlations with p value <0.05 are shown in bold.

CBF, cerebral blood flow; PSS, Perceived Stress Scale; mTBI, mild traumatic brain injury; CON, controls; prog, progesterone; Nvoxels, number of voxels in cluster; cereb., cerebellum; post., posterior; temporoocc, temporooccipital; MTG, middle temporal gyrus; STG, superior temporal gyrus; SPL, superior parietal lobule; ant., anterior; sup., superior; SMA, supplementary motor cortex. OUTSIDE means the cluster was not in a labeled region of the atlas.

Four clusters had significant correlations between rCBF and progesterone for the mTBI group only: left superior parietal lobule (L SPL), left medial temporal gyrus (L MTG), superior left superior temporal gyrus (L STG), and left precuneus. Whereas the L SPL cluster had a positive correlation between rCBF and PSS_Z (i.e., higher rCBF associated with higher or more severe PSS_Z), the remaining three clusters had negative correlations between rCBF and PSS_Z (lower rCBF associated with higher or more severe PSS_Z). Pearson's correlation coefficients and p values between rCBF extracted from these clusters and progesterone levels for the mTBI and CON groups are listed in the two rightmost columns in Table 3, with significant results shown in bold. No significant correlations to progesterone were found for the CON group. Three of these clusters had significant correlations between rCBF and progesterone levels in the mTBI group.

Three-way associations with neuroimaging

Three clusters showed significant three-way relationships among rCBF, PSS_Z, and progesterone. The locations and the corresponding two-way scatterplots for these clusters are shown in Figure 2, with the mTBI group in red and the CON group in black. In the L SPL cluster, which had a positive correlation between rCBF and PSS_Z, the correlation between rCBF and progesterone was negative. Similarly, in the clusters with a negative correlation between rCBF and PSS_Z, the correlation between rCBF and progesterone was positive. None of the correlations were significant for the CON group. In all three clusters, mean CBF for the mTBI group was lower than that of the CON group (p < 0.05).

FIG. 2.

Clusters where there were significant correlations between cerebral blood flow (CBF) and progesterone levels for the mild traumatic brain injury (mTBI) group. The corresponding CBF-perceived stress score (PSS)_Z and CBF-progesterone plots are shown in the middle two columns, where red represents the mTBI group and black represents the control (CON) group. The box plots in the rightmost column show the distribution of CBF values extracted from this cluster for the two groups. The red line represents the group mean CBF, and outliers are denoted by a red cross. Color image is available online.

Mediation testing

Mediation results are shown in Table 4, where odd-numbered rows represent the directed mediations, with progesterone as IV and rCBF as M, and even-numbered rows represent the control mediations, with rCBF as IV and progesterone as M. In all the directed mediation cases, paths A, B, and C without M (model 1) were statistically significant, but after including the effects of M, path C (model 2) was no longer significant, indicating that rCBF carries (i.e., mediates) the relationship between progesterone and PSS_Z. In all three clusters, the % effect of progesterone on PSS_Z mediated by rCBF were greater than the % effect of rCBF on PSS_Z mediated by progesterone, suggesting that only the directed mediation was valid. The L MTG cluster, denoted by*, was the only cluster with a Sobel's test p value <0.05 and a mediation effect of >50% (namely, the mediation effect was 100%); it should be noted that the beta slopes of these relationship suggest that progesterone level has a neuroprotective effect on PSS_Z. Correcting for multiple comparisons, each conjunction analysis p value of paths A, B, and C (p_a*p_b*p_c) was less than the Bonferroni correction p = (0.05 × 0.05 × 0.05)/72 = 1.74 × 10⁻⁶ (rightmost column of Table 4). The location of this cluster, as well as scatterplots for paths A, B, and C model 1, are shown in Figure 3.

Table 4.

Mediation Results for mTBI Group, Using CBF Values Extracted from the Three ROIs where a Significant Correlation between CBF and Progesterone Were Detected in Table 3

	Path: Model & predictor(s):		Path A: IV predicting M		Path B: M predicting DV		Path C model 1: IV predicting DV no M		Path C model 2: IV predicting DV with M		% Effect nediated	Sobel test p value	pa*pb*pcp(Bonf) = 1.74 × 10−6
IV	M	DV	Std β	p value	Std β	p value	Std β	p value	Std β	p value
Progesterone	L SPL	PSS_Z	-0.566	0.044	0.748	0.003	-0.585	0.036	-0.238	0.35	59	0.091	4.752 × 10−6
L SPL	Progesterone	PSS_Z	-0.566	0.044	-0.585	0.036	0.748	0.003	0.613	0.03	18	0.368	4.752 × 10−6
Progesterone	L MTG	PSS_Z	0.728	0.005	-0.826	0.001	-0.585	0.036	0.034	0.898	100	0.017	1.80 × 10−7
L MTG	Progesterone	PSS_Z	0.728	0.005	-0.585	0.036	-0.826	0.001	-0.851	0.008	0	0.896	1.80 × 10−7
Progesterone	L STG	PSS_Z	0.664	0.013	-0.772	0.002	-0.585	0.036	-0.13	0.636	78	0.052	9.36 × 10−7
L STG	Progesterone	PSS_Z	0.664	0.013	-0.585	0.036	-0.772	0.002	-0.685	0.028	11	0.630	9.36 × 10−7

In all three regions of interest (ROIs), cerebral blood flow (CBF) mediates a higher percentage of the effect of the independent variable (IV) (progesterone) on the dependent variable (DV) (PSS_Z), compared with progesterone as a mediator for the relationship between CBF and PSS_Z. Significant effects were noted by mediation effects >50% and Sobel's test p value of <0.05, and denoted with bold text. Although both L MTG and L STG had a conjunction analyses p value (p_a^*p_b^*p_c) less than the Bonferroni correction p(Bonf) = 1.74 × 10⁻⁶, only the L MTG mediation had a Sobel's test p value <0.05.

mTBI, mild traumatic brain injury; SPL, superior parietal lobule; MTG, middle temporal gyrus; STG, superior temporal gyrus; PSS, PSS, Perceived Stress Scale.

FIG. 3.

Correlation plots of the three-way relationship among progesterone, regional cerebral blood flow (rCBF) and perceived stress score (PSS)_Z of the left medial temporal gyrus (L MTG) cluster, which had the strongest mediation effect. Color image is available online.

Discussion

In this pilot study, we investigated whether rCBF mediated the relationship between progesterone and post-concussive symptoms among female collegiate club athletes assessed 3–10 days after concussion. We did not expect this relationship to be observed in demographic-, HC-, and MC-matched CON athletes participating in non-contact sports. If rCBF carried (i.e., mediated) a negative association between progesterone level and post-concussive symptoms measured in mTBI subjects but not in CON, this would support a hypothesis that progesterone might be neuroprotective during concussion. Our results support this expectancy. In particular, both whereas total PCSS and PSS were elevated in the mTBI group, only PSS_Z was significantly correlated with progesterone levels in the mTBI group in a negative relationship. Namely, higher progesterone levels were associated with lower PSS scores. Using voxelwise statistical analysis, we found three brain regions, all localized in the left hemisphere, with a three-way relationship among progesterone, CBF, and PSS_ Z: left SPL, left MTG, and left STG. Mediation analysis revealed that only the left MTG cluster was statistically significant, based on mediation effects being >50% and a Sobel's test p value <0.05 (specifically, p = 0.017). In this case, rCBF for this cluster mediated 100% of the relationship between progesterone and PSS_ Z.

Progesterone is a gonadal hormone that is synthesized by the ovaries in females, and by the testes and the adrenal cortex in males.³³ Sex hormones have been shown to affect sports performance,⁷⁵ and are attributed to the differences in sports injury risk between male and female athletes. For example, females in the luteal phase of the MC are less prone to anterior cruciate ligament (ACL) injury,^76,77 likely because of the ability of progesterone to increase fibroblast proliferation and collagen synthesis in ACL cell cultures.⁷⁶ Progesterone also plays an important role in the CNS via an array of progesterone receptors that are widely distributed in the brain.⁷⁸ Multiple animal models of brain injury have consistently suggested that progesterone may have neuroprotective effects. For example, in rodents, progesterone administered before middle cerebral artery occlusion resulted in smaller areas of infarction and improved outcome.⁷⁹ Exogenous progesterone also improved functional measures when administered after stroke.^{80, 81} Progesterone also appears to have neuroprotective effects in a rodent model of TBI, where levels of lipid peroxidation, cerebral edema, and inflammatory proteins associated with brain damage were all reduced after progesterone administration.^32,82 Clinical trials in adult TBI patients have also reported positive outcomes associated with progesterone treatment, including reduced death rate and improved functional outcomes.^51–53

In our cohort, we found that higher progesterone levels were associated with lower PSS score; that is, more normative levels of stress measure. Concurrently, higher progesterone was associated with higher rCBF values in the left MTG. Our between-group data also showed that rCBF levels were lower for the mTBI athletes relative to CON athletes. Our findings are supported by the literature, as higher rCBF has traditionally been associated with better tissue health and cognitive performance.^83–85 Therefore, our findings suggest that higher progesterone levels may be associated with more normative tissue responses in injured athletes. Lastly, mediation analysis showed that rCBF carries the relationship between progesterone and our stress measure (PSS_ Z), which can be viewed as evidence that progesterone has a neuroprotective role and that this in turn results in the reduction of post-concussion stress symptoms.

The locus of this mediation relationship is a small cluster within the posterior part of the left MTG. The MTG is a known network hub that facilitates communication between parallel and distributed brain networks.^86,87 The posterior part of the MTG is an important center for semantic processing, which is slightly lateralized to the left hemisphere based on a meta-analysis of 120 studies on semantic processing.⁸⁸ This is a heteromodal brain region that is activated by both visual and auditory stimuli and is thought to integrate various types of stimuli and to control semantic retrieval.^88,89 Although it is not immediately obvious how the left MTG is associated with stress, there is evidence that this region is also involved in emotional face processing.⁹⁰ The left MTG has also been implicated in social anxiety disorder, in which functional connectivity of this network hub is altered and proportional to both clinical and patient-reported symptom severity.⁹¹

The current finding warrants further investigation, as the relationship between progesterone and CBF has not been widely studied. Although CBF in females fluctuates over the course of the MC, during which the balance between estrogen and progesterone varies greatly, the exact mechanism is unclear.⁹² Different types of progestin for hormone replacement therapy in menopausal women also seem to play a role in modulating CBF.⁹³ A natural extension of this study would be to include measures of estrogen, as estrogen and progesterone can work in an antagonized manner.^31,33,94 Estrogen also has widely studied effects on CBF, primarily through the interaction between estrogen receptors and endothelial nitric oxide synthase.⁹⁴ A recent study aimed at studying sex differences in athletes with a history of concussion (HOC) found that females with HOC had a higher variability in temporal lobe CBF.⁹⁵ As this study did not account for the MC phase or hormonal levels, the authors could only speculate that hormonal variations may have contributed to this finding. Future studies that include hormonal level assessments might improve interpretation of such findings.

Among multiple limitations of the current study was the small sample size, which limited our analysis to subjects in the follicular phase of the MC. Given the current findings, further work is warranted to also study subjects in the luteal phase when progesterone levels are typically higher. A larger sample size would provide adequate statistical power to study both progesterone and estrogen to improve understanding of how the balance of these hormones might support a neuroprotection hypothesis. Further, a larger sample size including more HC users would also help elucidate whether HC use is also related to CBF and post-injury outcomes. Another future direction is inclusion of resting state connectivity analysis, which can identify networks of brain regions that may be involved in mediating the potential neuroprotective effects of hormones on post-injury outcomes. A potential weakness in the current study is related to the nonspecific nature of the PSS score. PSS assesses a person's ability to handle stress, which can vary even in healthy subjects without concussion. Future studies may benefit from other complementary measures of stress such as serum cortisol levels or emotion regulation tasks, to enable better understanding of the relationship between stress and CBF. Additionally, future studies may benefit from monitoring and accounting for sub-symptom threshold exercise during the recovery period, as it has been shown recently to normalize compromised CBF autoregulation and improve physiological perturbations observed post-injury.^96,97 Finally, although our matching protocol for MC/HC use helps to minimize measurement bias from the effect of MC phases on neuroimaging measures, it does not allow us to determine whether hormone levels at the time of injury also play a role in affecting symptoms and recovery. Inclusion of this information in future studies would help improve our understanding of the multidimensional relationship among hormones, neuroimaging measures, and post-injury symptoms.

Conclusion

In conclusion, this preliminary study examined three-way associations across three distinct measures of hormonal level, regional brain perfusion, and stress, in order to determine if regional brain perfusion mediated the relationship between progesterone and perceived stress. Our findings support the hypothesis that progesterone levels might reflect neuroprotection in a group of collegiate club female athletes after concussion. We found high progesterone levels associated with lower or more normative stress symptoms, as well as higher rCBF in the left MTG, reflecting potential neuroprotective effects of progesterone. rCBF in the left MTG mediated 100% of progesterone's relationship with PSS, which could be interpreted as the underlying mechanism. These findings warrant further investigation in a large-scale study to better understand the role of multiple hormones, both natural and synthetic, in influencing post-concussion symptoms by their actions in the brain.

Funding Information

Funding was provided in part by the Eleanor Wood-Prince Grant Initiative: A Project of the Woman's Board of Northwestern Memorial Hospital, the National Institute of Neurological Disorders and Stroke of the National Institutes of Health (NIH) under Award Number F31NS106840, and data collection via REDCap was supported in part by a Clinical and Translational Science Award (CTSA) grant from the NIH under Award Number UL1TR001422. Funding for scanning was supplied by the Warren Wright Adolescent Center (WWAC), Northwestern Memorial Hospital, and the WWAC provided space and/or salary for V.G., S.B., N.L.V., J.L.R., and H.C.B. A.A.H. was supported by VA RR&D Career Development Award RX000949. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.

Author Disclosure Statement

No competing financial interests exist.

Supplementary Material

Supplementary Figure S1

Supplementary Table S1