Oscillating hypercapnia induces neural abundant protein efflux and potential depletion in health and chronic traumatic brain injury

Introduction

Inefficient clearance of parenchymal waste represents a hallmark of chronic traumatic brain injury (TBI), atypical aging and various other neurodegenerative disorders.^1–3 Metabolic waste, macromolecules and cellular debris are typically cleared through glymphatic/lymphatic pathways (primary) and through the blood–brain barrier (secondary) via transcytosis. ⁴ The reciprocal relationship between intracranial cerebral blood volume (CBV) and cerebral spinal fluid (CSF) volumes with fixed parenchymal mass was first described over two centuries ago (Monro–Kellie doctrine, Figure 1(a)). ⁵ More recent preclinical work^6–8 has emphasized the importance of various vascular factors (Figure 1(b)) in the control of CSF flow. CSF flow represents a key and potentially necessary component for glymphatic/lymphatic clearance, as it facilitates subsequent mixing with interstitial solutes, efflux along perivenous, cranial and spinal nerve spaces, and drainage through lymphatic meningeal vessels.^4,6,9 Other studies in humans indicate that low frequency hemodynamic oscillations (LFHO) are the primary driver of bulk CSF flow in various brainstem compartments,^10–12 suggesting that exogenous agents which enhance bulk CSF flow can potentially be harnessed for therapeutic purposes. ¹³

OPEN IN VIEWER

Figure 1.

Theoretical and dynamic relationships between PaCO₂, CH, CSF flow and NAP efflux. CSF occupies ~13%–17% (mustard trapezoid) of intracranial volume, and is inversely proportional (black arrows) to CBV (pink trapezoid; ~3%–7%) assuming constant parenchymal volume (~80%) and normal brain compliance (Monro–Kellie doctrine; a). Global cerebrovascular caliber and tone (b) are primarily determined by CVR, vasomotion and physiological (cardiac and respiratory) fluctuations (box size denotes relative contribution), operating at different frequencies. Typical cerebrovascular caliber range (b, top row) is decreased following TBI (b, bottom row), as depicted by differences between actual (red) and potential (black) vessel diameter. Dynamic changes in PaCO₂ (c) drives CH changes (pink arrows) in a non-linear fashion during mild, moderate and severe hypercapnia (maximum ∆CBV = ~30%). Changes in CH subsequently drive bulk CSF flow (c, mustard yellow arrows) at currently unknown scale (depicted as linear relative to CH changes) and maximum value (denoted by X). Changes in CH (pink arrows)/bulk CSF flow (mustard arrows) are hypothesized to stimulate increased NAP clearance in an unknown dose-dependent fashion (denoted by size of blue arrows (d)) from the parenchyma to blood. Empirical PSC data from a single subject demonstrates coupling between GM blood–oxygen level dependent response (e, pink trace) and bulk CSF flow (e, mustard trace; note difference in scale) as a function of 2.5, 5 and 10 mmHG changes in ETCO₂.

Waste clearance is greatest during sleep,^14,15 when large amplitude LFHO occur secondary to global neuronal entrainment ¹⁰ and/or cyclical changes in norepinephrine levels resulting in alterations in vasomotion. ⁶ In contrast, during wakeful states, changes in CBV are primarily mediated by local neuronal activity within distributed neural networks, with anti-correlated activity occurring across competing networks (e.g. default-mode and executive network). ¹⁶ This diffuse neuronal activity results in small-amplitude changes in global CBV and subsequent small-amplitude changes in vascular-enhanced bulk CSF flow.^11,12 However, cerebrovascular reactivity (CVR; a primary determinant of cerebrovascular caliber; Figure 1(b)) can be parametrically manipulated by the potent vasodilator carbon dioxide (CO₂) to mimic or exceed the large amplitude changes in CBV that naturally occurs during slow wave sleep (Figure 1(c) and 1(e)). CO₂ acts directly on smooth muscle cells via H⁺ ions dissolved in CSF from arterial blood (primary mechanism) and secondarily through the release of other vasoactive agents.^17,18 CVR has also been shown to be impaired in both acute and chronic TBI^3,19–21 (Figure 1(b)). Collectively, these data suggest CO₂-induced LFHO may be sufficient to enhance waste clearance (Figure 1(d)) through advective and diffusion mechanisms associated with increased CSF flow and interstitial solute exchange, ¹³ and that this process may be impacted by cerebrovascular disease and atrophy following TBI.

The clearance of amyloid proteins from the brain represented a foundational finding establishing the existence of the glymphatic system. ²² Elevated blood protein concentrations have been traditionally equated with protein upregulation and/or cellular breakdown, with each protein linked to individual pathological substrates such as neuronal cell body injury (ubiquitin C-terminal hydrolase L1 (UCH-L1)), axonal injury (neurofilament light chain (NfL)), amyloid aggregation ((Aβ) 40 and Aβ42), tauopathies (brain-derived tau and phosphorylated tau species), or gliosis (glial fibrillary acidic protein (GFAP)).^23–26 Importantly, the concentration gradient for several of these proteins has been established to be 9.3%–130.8% higher in apical tissues and fluids (i.e. CSF and interstitial) relative to blood.^27–30 Several of these neural abundant proteins (NAP) are also increased in the bloodstream following general trauma, anesthetic exposure, cardiac surgery, renal dysfunction and in atypical aging.^31–34 Most recently, blood-based concentration levels for Aβ40, Aβ42, NfL, and ptau have been shown to be highest post-sleep, ³⁵ suggesting that NAP may also serve as a surrogate marker for general waste clearance. Collectively these findings suggest that there are multiple, non-exclusive pathways for brain-to-blood efflux, as well as ultimate systemic clearance.^36,37

The primary objective of the current study was therefore to investigate whether ~30 min of CO₂-induced LFHO would be sufficient to promote both NAP efflux and subsequent depletion in patients with chronic TBI (N = 22) and healthy controls (HC; N = 22). Our a priori hypotheses were that CO₂-induced LFHO would increase protein efflux, and that this relationship would be moderated by both global tissue volume and the magnitude of CVR. Our second objective was to examine whether ~30 min of CO₂-induced LFHO would be safe and well-tolerated in both healthy individuals and those with chronic TBI. The final a priori hypotheses were that the majority of participants would be able to safely tolerate the prescribed CO₂ exposures, ¹⁸ but that ~30 min of CO₂-induced LFHO would not be sufficient to affect performance on complex cognitive control tasks.

Methods

A prospective cohort design examined effects of hypercapnia on protein efflux (primary endpoints), complex cognition and common symptoms over an ~4-h session (Figure 2(a)). Participants with chronic TBI were recruited from the local community and clinics. Age- and sex-matched HC were recruited from the local community based on word of mouth, online databases and posted fliers. Inclusion criteria for the TBI group were 18–68 years old, age-of-injury >18 years, >6 months post-injury, and actively symptomatic. Exclusion criteria included history of neurological, intellectual, or major psychiatric disorders prior to the TBI, active substance abuse/dependence other than cannabis use (determined by urine screen), contraindications for MRI, or non-fluency in English. Exclusion criteria for HC were history of previous TBI with >5-min loss of consciousness, any neurological diagnosis, psychiatric disorders other than adjustment disorder, autism spectrum disorder, intellectual disability, active substance abuse/dependence, contraindications for MRI or non-fluency in English.

OPEN IN VIEWER

Figure 2.

Study design and protein efflux results. Study design (a) examining effects of CO₂ administration on Sx, cognitive control (depicted by computer icon) and protein efflux. CO₂ challenges were repeatedly administered (first cycle start = Time 0) during blood–oxygen level dependent imaging (depicted with brain icon). Blood samples were acquired at B, pre, 45 m, 90 m, and 2.5 h post-hypercapnia. GFAP (b), NfL (c), BD tau (d), and ptau217 (e) concentrations (log of pg/mL) are presented for participants with chronic TBI (light blue) and HC (coral) at baseline in black scale. Change from B (∆log) are depicted in gray scale for all points right of dashed line. Significant post-hypercapnia effects (denoted by brackets with asterisk) associated with both protein efflux (45 m; GFAP, NfL, and BD tau) and subsequent depletion (90 m; all proteins) were observed, followed by return to pre-hypercapnia levels at 2.5 h (all proteins).

Participants with chronic TBI were classified as experiencing “mild” (N = 9) or “moderate–severe” (N = 13) injury using a semi-structured interview (Ohio method). However, primary analyses compared all chronic TBI participants to HC in agreement with recent expert panel recommendations. ³⁸

Symptom rating and cognitive control task description

All participants completed symptom rating (i.e. symptoms of headache, nausea, fogginess, and dizziness; 11-point Likert scale) and a cognitive control task (see Supplemental Materials) before and after hypercapnia procedures. Reactive and proactive cognitive control tasks were selected given their critical role for evaluating executive functioning in both health and following TBI.^39–41 Briefly, the task consisted of multisensory (auditory and visual) cues simultaneously presented on either a black (reactive cognitive control task; Figure 3(a)) or gray (proactive control task; Figure 3(d)) background. A cue prompted participants to attend to either the auditory or visual modality. Cues were followed by simultaneously presented congruent or incongruent multisensory numeric stimuli (targets of “one,” “two,” or “three”). During reactive cognitive control, participants pressed a button corresponding to the target number presented in the attended modality while ignoring information in the unattended modality. Cognitive interference was quantified via subtraction of congruent from incongruent trials (i.e. large difference = poor reactive control). During proactive cognitive control (1-back trials), participants were required to hold the attended stimuli in working memory for 1.5 s while ignoring the unattended modality, and to only respond to the target upon the subsequent trial. Proactive control was quantified by subtracting congruent 0-back trials from congruent 1-back trials.

OPEN IN VIEWER

Figure 3.

Cognitive control task methods and results. Reactive (a, black background) and proactive (d, gray background) cognitive control tasks with simultaneously presented auditory and visual numbers were administered pre- and post-hypercapnia to participants with chronic TBI (blue colors) and HC (coral colors). The difference between reaction times (∆ = incongruent 0-back − congruent 0-back trials) during the reactive cognitive control task significantly decreased indicating less interference for both groups on AA trials (b, bottom row) post-hypercapnia (hypercapnia effects denoted by brackets with carat symbol), and for controls on AV trials (c, bottom row). In contrast, there were no pre/post-hypercapnia effects during proactive cognitive control (Δ = 1-back congruent − 0-back congruent trials; (e) and (f), bottom rows). As expected, chronic TBI patients exhibited significantly lower accuracy across both AA and AV trials relative to HC during both reactive and proactive cognitive control (significant group effects denoted by black brackets with asterisk).

MRI hypercapnia task description and ETCO₂ analyses

All imaging data were acquired on a Siemens 3T scanner with a 32-channel head coil. High resolution T₁-weighted (voxel size = 1.0 mm³) images were obtained first followed by six runs of single-shot, gradient-echo echoplanar pulse sequence (voxel size = 3.0 mm³) to quantify bulk CSF flow dynamics during a hypercapnia challenge (Figure 4(a)). Each run included four separate blocks of increased CO₂ exposure (35 s duration) interspersed with gas concentrations more typical of atmospheric air (~30 s duration), for a total of 24 hypercapnia cycles administered across ~30 min. As in previous studies, ¹⁹ research staff manually switched between room air (inhalation duration 30 ± 5 s) and the predetermined gas mixture (5% CO₂, 21% O₂, balance N₂) using a Douglas bag, with exhaled CO₂ data sampled at 1000 Hz. Following pre-processing (see Supplemental Methods), the ETCO₂ regressor was min–max normalized to approximate an amplitude of 1 across the entire timeseries, and shifted thirty-five times to capture individual differences in physiology.

OPEN IN VIEWER

Figure 4.

Hypercapnia methods and imaging results: CVR (a–c) and VE-bCSF (b, d–f) flow were quantified using a variety of metrics. Unfiltered global GM (gray trace in (c)) as well as Bp filtered data (GM_Bp; pink trace in (d)) and its derivative (dGM_Bp/dt; blue trace in (d)) were derived from a subject specific mask (dark blue, b). The bCSF flow signal (mustard trace in (f)) was derived from a ROI immediately below the fourth ventricle (yellow tracing, b). The ETCO₂, filtered global GM signal, and derivative vectors were temporally shifted at various lengths (c, e) to account for physiological lags (denoted by blue arrow and bracket in (c)). A series of regressions identified the maximal fit for both CVR (GM ROI–ETCO₂ (c)) and VE-bCSF flow (CSF ROI–GM_Bp and CSF ROI–dGM_Bp/dt (f)) following CO₂ administration. Participants with chronic TBI (light blue) exhibited a significantly (group effects denoted by asterisk) decreased fit (Fisher’s Z) and Amp (AU) for CVR metrics (g) relative to matched HC (coral). TBI participants exhibited decreased VE-bCSF flow Amp (h) at a trend level (denoted by a “T”), with similar results observed for the derivative ((i) dVE-bCSF flow/dt).

BOLD data during the hypercapnia task were processed using individual CSF and gray matter (GM)-specific pipelines to characterize global CVR ¹⁹ as well as CO₂-induced bulk CSF flow (Figure 4(b)–(f); Supplemental Materials and Supplemental Figure S1).^10,12 For CVR quantification, a global GM mask was first created from the T₁-weighted image (SPM12) using a 60% probability threshold, spatially transformed to the single-band reference image (i.e. native space), and multiplied by the fully preprocessed BOLD data. The array of participant-specific, time-lagged ETCO₂ vectors was then regressed against the global gray matter signal. The unstandardized beta (i.e. amplitude) from regressor with the maximal fit (i.e. highest absolute Pearson correlation between ETCO₂ and global GM signal) was selected, followed by a Fisher’s Z-transformation of the Pearson correlation coefficient, and calculation of latency to maximal fit.

CSF that has not been exposed to radiofrequency pulses (i.e. flowing into the most inferior slice) has a higher T2* signal, while radiofrequency exposed CSF has a lower signal. ¹⁰ The CSF-specific voxel-wise preprocessing stream included only time-slice correction, followed by bandpass filtering (0.0008–0.10 Hz) tailored to CVR cycle length, detrending, mean reversion and truncation to remove edge artifacts. A CSF region of interest within the cervical central canal inferior to the medulla oblongata was identified via a semi-automated process. To assess CO₂-induced bulk CSF flow, the fully preprocessed voxel-wise data were similarly bandpass filtered, detrended and truncated, and multiplied by the subject specific GM mask to form a single vector. The bandpassed GM vector was time-shifted 83 times to form an array of regressors (0–40.32 s). The single vector was also numerically differentiated to create a GM derivative to specifically quantify the relationship between changes in hemodynamics and bulk CSF flow, as well as for quantifiable quality assurance metrics (see Supplemental Methods and Supplemental Figure S1). The derivative was then time-shifted six times to create a regressor array spanning from 0 to 3.36 s. Both the bandpassed GM and derivative arrays were subsequently regressed on the CSF flow vector to quantify CO₂-induced bulk CSF flow using identical metrics as the CVR analyses (i.e. normalized amplitude, Fisher’s Z and latency of maximal fit).

Total GM, WM, and CSF volumes were obtained from the T₁-weighted image (FreeSurfer pipeline) and normalized by total intracranial volume as primary structural metrics. Brain age (Supplemental Figure S2) was estimated using a deep learning network (DeepBrainNet) architecture. ⁴² The predicted brain age difference (PAD) was calculated by subtracting actual subject age from the median value of all slice-wise predictions, and subsequently served as a secondary measure of overall brain structure health.

Results

Demographics and symptom characterization

A total of 22 individuals with chronic TBI (nine “mild” and 13 “moderate–severe”) and 22 HC completed all hypercapnia exposures (24 cycles across six MRI runs) for a completion rate of 95.7%. Two participants (4.3% of sample) were unable to tolerate the hypercapnia equipment (mask inside of scanner) due to claustrophobia rather than the CO₂ exposure itself. Participants who completed the hypercapnia task did not differ on biological sex or age between the TBI and HC groups (all p’s > 0.05; Table 1). The total change in ETCO₂ summed across the six individual runs did not significantly differ as a function of Group (p = 0.124; Cohen’s d = −0.47), sex or age.

OPEN IN VIEWER

Table 1.

Participant demographics and symptom summary scores (i.e. Likert ratings of headache, nausea, fogginess, and dizziness) across the entire hypercapnia challenge.

	TBI (N = 22)	HC (N = 22)	p value
Age	42.41 ± 14.44	42.73 ± 11.26	0.934
Sex (% male)	14 (63.6%)	13 (59.1%)	0.757
Injury characteristics
Severity (“mTBI”/“msTBI”)	9 (40.9%)/13 (59.1%)	NA
YPI	3.7 (1.6–6.1)	NA
Symptoms
B	0 (0–3.25)	0 (0–1)	<0.001
∆CO22	0 (−1.25 to 3)	0 (−0.25 to 0)	0.884
∆CO24	0 (−1.25 to 2.25)	0 (−0.25 to 0)	0.432
∆CO26	1 (0–4)	0 (−0.25 to 0.25)	0.052
∆Bl4	0 (−1 to 0)	0 (−0.25 to 0)	0.851
∆Bl5	0 (−2 to 0)	0 (−1 to 0)	0.170

“mTBI”: “mild” TBI; “msTBI”: “moderate-to-severe” TBI; NA: not applicable; B: baseline; ∆: change score between the timepoint provided and baseline; ∆CO₂2: symptom rating after second block of hypercapnia challenge; ∆CO₂4: symptom rating after fourth block of hypercapnia challenge; ∆CO₂6: symptom rating after sixth and final block of hypercapnia challenge; ∆Bl₄: symptom rating after fourth blood draw; ∆Bl₅: symptom rating after fifth and final blood draw; YPI: years post-injury.

All data are reported in median and IQR (denoted by parentheses) or mean plus standard deviation (denoted by ±).

Structural imaging results

A series of LME models comparing tissue volumes (Figure 5(a)–(c)) across Group with sex and age as additional covariates indicated that normalized CSF volume was significantly increased for participants with chronic TBI relative to HC (F = 7.31; p = 0.010; Cohen’s d = −0.81). In contrast, normalized WM (p = 0.180; Cohen’s d = 0.41) and GM (p = 0.169; Cohen’s d = 0.42) volumes did not show a significant main effect of Group at current sample sizes (p > 0.10), although medium-to-large effect sizes were present. A similar LME model indicated that predicted brain age (Supplemental Figure S2) was significantly higher for participants with chronic TBI relative to HC (F = 4.84; p = 0.034; Cohen’s d = −0.65).

OPEN IN VIEWER

Figure 5.

Structural imaging results. Participants with chronic TBI (light blue) exhibited significantly (group effects denoted by asterisk) increased normalized CSF (a) Vol relative to HC (coral), whereas normalized GM (b) and WM (c) volumes were not significantly different.

Relationships between imaging results and protein efflux

Separate multiple regression analyses investigated associations between fit/amplitude from CVR and CO₂-induced bulk CSF flow (independent variables) with protein efflux (change scores 45 min post-hypercapnia), as well as between structural volume metrics (independent variables: normalized CSF, GM, and WM) and protein efflux. Omnibus results from the vascular/bulk CSF flow metric models were significant only for ptau217 (Figure 6(a); p = 0.010; R² change = 30.9%), with significant positive partial correlations observed between ptau efflux and CVR fit (ρXY⋅Z = 0.34) and CO₂-induced bulk CSF flow amplitude (ρXY⋅Z = 0.35).

OPEN IN VIEWER

Figure 6.

Relationship between imaging metrics and protein efflux. Significant and unique associations (denoted by partial correlation = ρXY⋅Z; all data graphed in rsd form respective to all other variables) were observed between CVR fit (Fisher’s Z; pink circles) and ptau217 (a) efflux (log of pg/mL) at 45 min post-hypercapnia, as well as between CO₂-induced bulk CSF Amp (AU; mustard circles) with ptau217 efflux. Associations between structural volumes (right of dashed line) and protein efflux were significant for GFAP (b) and BD tau (c), with both CSF volume (orange circles) and white matter volume (white circles) each uniquely associated with protein efflux. Similar structural relationships existed for ptau217 (d), although the overall model was not significant (denoted by †).

In contrast, omnibus results for tissue volume models (Figure 6(b)–(d)) were significant for GFAP (p = 0.024; R² change = 26.3%) and brain-derived tau (p = 0.039; R² change = 23.9%) efflux, with significant positive partial correlations observed between normalized CSF (GFAP ρXY⋅Z = 0.33; brain-derived tau ρXY⋅Z = 0.34) and white matter (GFAP ρXY⋅Z = 0.42; brain-derived tau ρXY⋅Z = 0.35) volumes with the magnitude of protein efflux. Although the overall model for ptau217 was not significant (p > 0.05), a similar magnitude of partial correlations were observed for CSF (ρXY⋅Z = 0.38) and white matter (ρXY⋅Z = 0.35) volumes.

Discussion

The dynamic, inverse relationship that exists between global CBV and CSF volumes within the fixed cranial vault was originally posited over 200 years ago, ⁵ but the importance of vascular-enhanced CSF flow in glymphatic functioning has only recently emerged.^6–8 Consistent with a priori hypotheses, CO₂-induced LFHO were effective for increasing the efflux of neural abundant protein concentrations in the blood, which may represent a surrogate marker of brain waste clearance. ³⁶ This temporary increase was subsequently followed by a reduction in blood protein concentrations below pre-hypercapnia levels, providing preliminary evidence of a putative reduction in parenchymal protein concentrations following ~30 min of oscillating CO₂ exposure. Current imaging results indicated that individuals with chronic TBI exhibited reduced CVR and increased atrophy relative to matched controls, but no group differences existed for CO₂-induced bulk CSF flow at our limited sample sizes. The magnitude of protein efflux was more strongly associated with CSF and white matter tissue volumes, although ptau217 was also associated with both vascular functioning and bulk CSF flow coupling. Collectively, current and previous ⁴⁴ results suggest that CO₂-induced LFHO enhances waste clearance across multiple neurodegenerative conditions, and may therefore have therapeutic indications.

In normal tissue, healthy epithelia (e.g. blood–brain and blood–CSF barriers) typically restrict the exchange of proteins between apical and basolateral spaces through continuous non-fenestrated endothelial cells and tight junctions, with protein exchange occurring through highly regulated transcytosis.^45,46 The glymphatic/lymphatic systems therefore represent the primary pathway for clearance of waste products into the bloodstream through both advective and diffusion processes.^4,6,9 Glymphatic/lymphatic clearance is dependent on both macroscopic and microscopic mechanisms which vary across gray and white matter tissue, but CSF influx is recognized as a necessary component as it facilitates mixing of interstitial solutes and subsequent waste removal. Human^10–12 and rodent^6–8 data suggest that LFHO respectively govern both global (e.g. laminar flow in the fourth ventricle) and local (e.g. at the level of neurovascular unit) CSF flow dynamics. Thus, increases in both global (e.g. cerebrovascular reactivity) and local (e.g. following neurovascular coupling) cerebral blood flow/volume not only provide needed metabolic substrates, but also play a critical role in indirectly promoting waste clearance via increased CSF flow.^7,8

Prescribed CO₂ can be used to mimic the oscillating levels of global parenchymal engorgement that occur naturally during slow wave sleep secondary to neuronal entrainment ¹⁰ or vasomotion, ⁶ and have been shown to drive glymphatic clearance in rodent studies. Importantly, the majority (90.9%) of individual hypercapnia sessions (i.e. six runs per individual; 264 runs in total) resulted in strong anti-correlation between the derivative of CO₂-induced LFHO and bulk CSF flow, indicative of the fundamental inverse relationship (i.e. increased blood volume = decreased CSF volume) that exists for global fluid dynamics within the fixed cranial vault. ⁵ The tight coupling that occurs between CO₂-induced LFHO and bulk CSF flow did not differ between HC and individuals with chronic TBI, although differences in vascular-induced CSF flow amplitude changes existed at trend level. Consistent with a priori hypotheses, GFAP, NfL, and brain derived tau concentrations were increased in the blood stream ~45 min after the hypercapnia challenge across both groups of participants, followed by evidence of potential parenchymal depletion (i.e. reduced plasma protein levels) at ~90 min post-hypercapnia for all measured proteins. The kinetics of protein efflux and depletion were temporally uniform regardless of protein molecular weight at the current temporal sampling regimen, a finding more commonly associated with advection during glymphatic clearance. ⁴ Although UCH-L1 was quantified, results indicated poor data quality, similar to previous studies with this assay for “mild” TBI in more acute settings.^47,48

TBI alters endothelial and smooth muscle cell function, changes capillary distensibility, reduces capillary recruitment and cerebral blood flow, and causes pericyte loss.^3,49,50 Post-traumatic vascular pathology has been histologically confirmed even after controlling for vascular risk factors and vascular disease burden. ⁵¹ The traumatically-compromised cerebrovasculature has been linked to reduced waste clearance,^1,3 which further disrupts the neurovascular unit through positive feedback loops (“double-hit” hypothesis) and increases the likelihood of prion-like spreading. Hypoperfusion has traditionally been proposed as the primary link between cerebrovascular dysfunction and reduced waste clearance, but previous^3,20,21 and current findings suggest CVR is also reduced post-TBI, and therefore represents a second potential mechanism. Specifically, a limited vessel caliber range may result in delayed or reduced cerebral blood flow, as well as restrict the magnitude of subsequent bulk CSF flow (Figure 1(b) and (c)), both of which may negatively impact waste clearance in the long-term. However, contrary to our a priori predictions, there was only a significant relationship between CVR and vascular-induced bulk CSF flow with ptau217 rather than all proteins.

Multiple studies^52–55 have indicated progressive tissue loss (i.e. neurodegeneration) following TBI, which is greater in white (+5.97 years) relative to gray (+4.66 years) matter and is associated with blood protein concentration levels. CSF volumes and predicted brain age were significantly increased in the chronic TBI cohort relative to the control sample, and global CSF and white matter volumes accounted for unique variance for GFAP, brain-derived tau, and ptau217 efflux. These findings preliminarily suggest that CO₂-induced protein efflux may be more dependent on global parenchymal tissue to fluid ratios (Monro–Kellie doctrine) rather than fluid dynamics at the neurovascular unit level (i.e. local interactions between changes in vessel caliber and CSF flow).^6–8 However, these findings require replication in larger sample sizes and other neurodegenerative conditions. Moreover, current results are likely only pertinent to acute alterations in fluid dynamics resulting from CO₂ administration rather than the long-term relationships that exist between cerebrovascular dysfunction and abnormal protein aggregation.^1,3 Therefore, future preclinical and clinical studies are required to examine the role of these individual biological mechanisms on waste clearance in more detail.

For example, systemic clearance through renal and hepatic pathways represents a frequently overlooked factor for understanding protein concentration levels in blood. ³⁶ Clearance via renal filtration occurs exponentially faster for proteins with lower molecular weights when other molecular factors (charge and size) are held constant, whereas >60 kDa proteins are typically catabolized hepatically.^56,57 Moreover, although CO₂ is a potent vasodilator in most tissue beds, it acts as a vasoconstrictor in the kidneys and results in both decreased blood flow and increased tissue resistance in healthy individuals.^58,59 Similarly, recent evidence suggests that there may be at least two separate pathways for glymphatic clearance from parenchymal tissue versus the CSF that are at least partially dependent on molecular weight, with larger proteins potentially captured and cleared through lymph nodes. ³⁷ Thus, future preclinical studies are also required to determine how vascular-enhanced manipulations of bulk CSF flow affect different glymphatic clearance pathways and compounds of various molecular weights.

Previous preclinical studies have demonstrated that glymphatic impairment decreases the release of waste products into the blood stream during the acute phases of TBI, theoretically increasing cellular exposure to these toxins. ⁶⁰ Hypercapnia may therefore represent a generalized therapy that can be prescribed to enhance waste clearance early post-injury, when treatments are most likely to be more effective. Similarly, the therapeutic administration of hypercapnia could also reduce post-traumatic toxic protein concentrations in the parenchyma during the chronic injury phase, as such proteins can remain elevated months-to-years post-injury and are associated with increased atrophy.^52–54 Finally, regularly prescribed hypercapnia sessions may also augment homeostatic waste clearance given that sleep is frequently disrupted following TBI or in atypical aging. ⁶¹

Although individuals with TBI and HC experienced a temporary increase in symptoms during the hypercapnia session, total symptom burden quickly returned to baseline levels, suggesting that the hypercapnia procedures were safe and well tolerated. No other adverse events were observed during the study. Contrary to a priori hypotheses of a null effect, performance on a reactive cognitive control task improved after ~30 min of oscillating hypercapnia, with evidence of reduced interference from conflicting stimuli in both individuals with chronic TBI and HC. In contrast, there were no significant post-hypercapnia changes in proactive cognitive control for either group. Individuals with TBI exhibited the expected impairments in accuracy across both reactive and proactive tasks. ⁴¹ Thus, current results preliminarily suggest that hypercapnia may potentially improve cognitive performance during tasks which require the parsing of conflicting information (i.e. stimulus-driven), but may not improve performance on tasks that require the planned/anticipatory allocation of cognitive resources.^39,43 Importantly, the reduction in interference during reactive cognitive control requires both replication in a controlled clinical design with placebo CO₂ administration to rule-out practice effects, as well as testing in other cognitive domains such as vigilance and processing speed to investigate whether findings generalize to other domains.

Limitations of the current study include moderate sample sizes within each group that may have reduced power to detect more subtle differences in protein efflux and/or depletion as a function of TBI (small to medium effect sizes), as well as potential group differences in bulk CSF flow metrics (e.g. amplitude a statistical trend). Similarly, the current study followed recent recommendations to analyze TBI as a spectrum rather than based on traditional labels of “mild” or “moderate-to-severe” TBI.^38,62 However, we were underpowered to examine for known imaging differences (e.g. greater atrophy and cerebrovascular impairment in more severe injuries) and possible differences in protein efflux based on injury severity, representing a critical future direction. As a result, means and standard deviations are provided in Supplemental Results for all proteomic measurements based on sub-group membership.

Second, the measurement of bulk CSF flow via BOLD imaging in humans is both a surrogate and indirect measurement of glymphatic functioning. Like most time-dependent imaging techniques, it is affected by multiple other physiological confounds (motion, respiratory patterns, etc.). However, multiple rodent studies suggest that bulk CSF flow represents a proxy measure for the penetration of CSF into parenchymal spaces,^6–8 suggesting that it may represent a reasonable surrogate of glymphatic function. Moreover, the methodologies used in the current study to manipulate cerebral blood flow (hypercapnia), proxies for CSF flow (T2* effects within the most inferior slice) as well as surrogates for waste clearance (commercially available proteomics) provide immediate clinical translation relative to the more complex methods/markers that have proven to be feasible only in animal models to date.^4,6,7 Finally, additional preclinical studies are required to determine if the decrease in neural abundant protein concentrations in the blood truly reflects a reduction in parenchymal protein levels, or is instead reflective of a reduced systemic clearance or temporary change in cerebrovascular tone post-hypercapnia.

In summary, there are currently no approved therapies for TBI outside of neurosurgery and acute symptom management. Current results suggest that CO₂-induced LFHO are sufficient for increasing protein efflux, and may result in a brief period of parenchymal protein depletion before returning to pre-hypercapnia levels. The putative underlying mechanism of action for CO₂ therapy (i.e. enhanced advective clearance through vascular-enhanced bulk CSF flow) harnesses normal brain physiology, and may therefore have widespread clinical applications for treating multiple neurodegenerative disease states relative to drugs that specifically target abnormal tau, amyloid, or alpha-synuclein aggregation.

Supplemental Material