Inhaled Xenon Washout as a Biomarker of Alzheimer’s Disease

Biomarkers have the potential to aid in the study of Alzheimer’s disease (AD); unfortunately, AD biomarker values often have a high degree of overlap between healthy and AD individuals. This study investigates the potential utility of a series of novel AD biomarkers, the sixty second 129Xe retention time, and the xenon washout parameter, based on the washout of hyperpolarized 129Xe from the brain of AD participants following inhalation. The xenon washout parameter is influenced by cerebral perfusion, T1 relaxation of xenon, and the xenon partition coefficient, all factors influenced by AD. Participants with AD (n = 4) and healthy volunteers (n = 4) were imaged using hyperpolarized 129Xe magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS) to determine the amount of retained xenon in the brain. At 60 s after the breath hold, AD patients retained significantly higher amounts of 129Xe compared to healthy controls. Data was fit to a pharmacokinetic model and the xenon washout parameter was extracted. Xenon washout in white and grey matter occurs at a slower rate in Alzheimer’s participants (129Xe half-life time of 42 s and 43 s, respectively) relative to controls (20 s and 16 s, respectively). Following larger scale clinical trials for validation, the xenon washout parameter has the potential to become a useful biomarker for the support of AD diagnosis.

Cerebro-vascular factors have been demonstrated to both influence and to be influenced by AD pathology [26][27][28]. These vascular effects include an increased risk of stroke in AD patients [29]. Paradoxically, a history of strokes excludes a diagnosis of AD in favor of vascular dementia [30]. This only muddies the waters in differentiating between these two causes of dementia and raises discordance: If an individual suffers a stroke and then presents with dementia symptoms, the dementia is likely to be differentiated as vascular dementia, yet if an AD patient suffers a stroke following diagnosis, the stroke is likely to be attributed to the AD pathology.
Impaired cerebral perfusion is one of the many pathological hallmarks of AD [49] and its use as a biomarker for AD was established even before the amyloid cascade hypothesis was first posited [50] in the early 1990s. Numerous research groups have observed a reduction in cerebral perfusion in AD patients using a variety of techniques [51][52][53]. In one such example, Perani, et al. showed a reduction in global cerebral perfusion compared to healthy controls using technetium-99m hexamethylpropyleneamine oxime ([99mTc]HM-PAO) Single Positron Emission Tomography (SPECT) [54]. Parkes and colleagues employed arterial spin labelling (ASL), an MRI-based technique, to demonstrate a decrease in the grey matter to white matter (GM:WM) perfusion ratio, which was attributed to the reduction in GM perfusion [55]. Work by Du et al. and Johnson et al. localized areas of hypoperfusion to the parietal cortex [56,57]. In addition to localizing impaired cerebral perfusion to the parietal cortices, Schuff, and colleagues localized impaired perfusion to the frontal cortices [56]. ASL and phase contrast MRI have been used to detect changes in cerebral blood flow (CBF) in order to stage AD disease severity [58,59].
In addition to detecting endogenous nuclei in the body (i.e., protons), MRI can also be used to detect exogeneous nuclei. A variety of techniques have been developed to increase the sensitivity of MRI to these exogeneous nuclei. One of these techniques involves hyperpolarizing the nuclei of a variety of elements such as 3 He [60], 129 Xe [61][62][63] and 13 C [64], to align their nuclear spin angular momentum, providing an increase in the signal-to-noise (SNR) ratio of approximately five orders of magnitude [65]. Because these nuclei are not ubiquitous to the body under ordinary physiological conditions, they can be detected and tracked throughout the body using MRI and magnetic resonance spectroscopy (MRS) [66][67][68]. In particular, a hyperpolarized (HP) gas, 129 Xe, dissolves into the blood following inhalation [69] and then travels throughout the vasculature and accumulates in highly perfused tissues such as the brain [70,71]. This process is referred to as xenon wash-in. Xenon then washes out of the brain tissue, dissolving in the blood and is then exhaled following gas exchange in the lungs.
The diagnosis of AD is a diagnosis of exclusion and is primarily based upon clinical presentation [30]. Biomarkers obtained from the cerebrospinal fluid (CSF) or from imaging may be used as confirmatory tools [72]. One of the difficulties in using AD-associated biomarkers for AD diagnosis is the large overlap in values between AD patients and healthy individuals; while the mean values between the two populations may be statistically significant, many AD patients may have biomarker values close to the "normal" range, and vice versa [73]. Abnormal Aβ42 levels are typically set at <550 ng/L [74]. This overlap has motivated the search for AD biomarkers with higher diagnostic accuracy. Recently, Nakamura, et al. demonstrated convincing plasma AD biomarkers at accuracies exceeding 90% using highly sensitive immunoprecipitation and mass spectrometry [75]. At present, biomarkers are ideally suited to quantitatively monitor disease progression over time, comparing current values to baseline values.
In the present study, we employed HP 129 Xe, MRI and MRS to probe the xenon gas exchange characteristics in the brain of AD participants [63]. We demonstrate the preliminary results of two potential biomarkers of AD based on the washout of HP 129 Xe from the brain of AD participants, which was detected using 129 Xe MRS. The washout of the 129 Xe from the brain can be used as an indirect measure of brain perfusion. In contrast to the high degree of overlap of many existing AD biomarkers, our preliminary data demonstrate a five standard deviation difference of 129 Xe-GM signal retention at 60 s following the breath hold and a nearly two standard deviation difference in the xenon washout parameter between AD participants and healthy controls. The xenon washout parameter is calculated by fitting the xenon signal curve to a pharmacokinetic equation (see Equation (1) in results). The difference in xenon retention values between AD participants and healthy controls makes these potential biomarkers candidates for larger scale future studies.

Ethical Approval and Consent to Participate
This research study was approved by the research ethics boards (REB) of Lakehead University (LU) and the Thunder Bay Regional Health Sciences Centre (TBRHSC) (Reference number RP-307) and was conducted in accordance with the Tri-Council Policy Statement-2 (TCPS-2). All participants consented to their data being used for publication.

Participant Recruitment
Four participants diagnosed with mild to moderate AD were recruited from the community for participation in this study. AD patients were diagnosed using clinical criteria by a qualified neurologist or gerontologist. Additionally, four age-matched healthy volunteers were recruited to serve as controls. Age-matched control participants were all cognitively normal. Informed consent was obtained from all human participants.

129 Xe Magnetic Resonance Spectroscopy
Enriched 129 Xe was polarized to~35% using a Xemed xenon (Xemed LLC, Durham, NH, USA) gas polarizer and dispensed into a 500 mL Tedlar bag. The participants inhaled the Xe gas and held their breath for 20 s. Sixty dynamic spectra were acquired every 2 s beginning with Xe inhalation. Xe MRS parameters were as follows: 60 dynamic scans, bandwidth 32 kHz, sample number: 4096, TR/TE = 2 s/0.17 ms, FA = 10 • . The signal was a single voxel encompassing the entirety of the brain region. We used a low flip angle to maintain polarization of the 129 Xe gas throughout all dynamic scans. Both Xe-GM and Xe-WM peaks were plotted as a function of time. Signal intensity was calculated by measuring the peak divided by the standard deviation of the noise.

129 Xe Image Processing
All images were processed using a custom Matlab script that converted the raw data in k-space into an MR image using a Fast Fourier Transform (FFT) algorithm. SNR maps were created by dividing each pixel by the standard deviation of the noise. Complete details on how the SNR maps were created are contained within the supplementary information.

Xenon Washout Parameter Maps Image Processing
The three dynamic 129 Xe MRI were processed as described above. A custom Matlab script was used to calculate the xenon washout parameter of each pixel as described above to create a "xenon washout parameter map". Xenon washout parameter maps from all individuals were averaged to create a mean xenon washout parameter map for all AD participants and healthy age-matched controls. A mask was created to remove noise from outside the brain region for image clarity. The xenon washout parameter maps were overlaid on T2W anatomical MRI using GIMP v.2.8 image processing software.

Statistical Analysis
Data for all participants was aggregated and the means and standard deviations were calculated. 16 data points for healthy controls and 13 data points for AD were used. (Each participant had xenon washout measured 3 or 4 times). A Welsh's t-test (2-tail, unpaired) was conducted to establish statistical significance. For all comparisons, p < 0.01. However, due to the preliminary nature of this study and small sample size, p values were not stated.

Magnetic Resonance Spectroscopy
In this work, we analyzed the 129 Xe magnetic resonance (MR) spectra from the brains of AD participants (n = 4) compared to healthy controls as a function of time. We began our study by ensuring that the 129 Xe spectra we acquired were consistent with previously reported spectra. We observed five 129 Xe MRS peaks measured at +189 ppm, +193 ppm, +196 ppm, +199 ppm, +219 ppm ( Figure 1A). These peaks are consistent with previously reported spectra that assigned these peaks to 129 Xe interacting with muscle, WM, GM, CSF and red blood cells (RBC), respectively [71,76]. The chemical shift of the 129 Xe MRS acquired from AD participants did not differ appreciably from the MRS acquired from healthy controls. However, the signal intensity of brain matter peaks differed in AD participants versus healthy controls ( Figure 1A). The 129 Xe-GM signal was 43% lower in AD participants than in healthy controls. Whereas the 129 Xe-WM was not statistically different between healthy controls and AD participants ( Figure 1B). We also quantified the ratio of GM to WM. We calculated that the 129 Xe-GM/ 129 Xe-WM was 32% lower in AD participants than in healthy controls.

MRS as a Function of Time
Next, we tracked the SNR of the 129 Xe-WM and 129 Xe-GM to probe the effect of AD on the washout time of 129 Xe from the brain (Figure 2). We observed a considerable difference in 129 Xe washout between AD participants and healthy controls. Healthy controls had a 129 Xe washout half-life of 20 s and 16 s in WM and GM, respectively; while AD participants had a 129 Xe washout half-life time of 42 s and 43 s in WM and GM, respectively. AD participants had a slower washout compared to healthy controls indicating increased 129 Xe retention in the WM and GM of the brain. Sixty seconds (60 s) following 129 Xe inhalation, we observed a five standard deviation difference in the 129 Xe-GM between healthy controls and AD participants ( Table 1, Figures 2D and 3A). In fact, there was no overlap between healthy controls and AD participants. to 129 Xe interacting with muscle, WM, GM, CSF and red blood cells (RBC), respectively [71,76]. The chemical shift of the 129 Xe MRS acquired from AD participants did not differ appreciably from the MRS acquired from healthy controls. However, the signal intensity of brain matter peaks differed in AD participants versus healthy controls ( Figure 1A). The 129 Xe-GM signal was 43% lower in AD participants than in healthy controls. Whereas the 129 Xe-WM was not statistically different between healthy controls and AD participants ( Figure 1B). We also quantified the ratio of GM to WM. We calculated that the 129 Xe-GM/ 129 Xe-WM was 32% lower in AD participants than in healthy controls.

MRS as a Function of Time
Next, we tracked the SNR of the 129 Xe-WM and 129 Xe-GM to probe the effect of AD on the washout time of 129 Xe from the brain (Figure 2). We observed a considerable difference in 129 Xe washout between AD participants and healthy controls. Healthy controls had a 129 Xe washout halflife of 20 s and 16 s in WM and GM, respectively; while AD participants had a 129 Xe washout half-life time of 42 s and 43 s in WM and GM, respectively. AD participants had a slower washout compared to healthy controls indicating increased 129 Xe retention in the WM and GM of the brain. Sixty seconds (60 s) following 129 Xe inhalation, we observed a five standard deviation difference in the 129 Xe-GM between healthy controls and AD participants ( Table 1, Figures 2D and 3A). In fact, there was no overlap between healthy controls and AD participants.

MRS as a Function of Time
Next, we tracked the SNR of the 129 Xe-WM and 129 Xe-GM to probe the effect of AD on the washout time of 129 Xe from the brain (Figure 2). We observed a considerable difference in 129 Xe washout between AD participants and healthy controls. Healthy controls had a 129 Xe washout halflife of 20 s and 16 s in WM and GM, respectively; while AD participants had a 129 Xe washout half-life time of 42 s and 43 s in WM and GM, respectively. AD participants had a slower washout compared to healthy controls indicating increased 129 Xe retention in the WM and GM of the brain. Sixty seconds (60 s) following 129 Xe inhalation, we observed a five standard deviation difference in the 129 Xe-GM between healthy controls and AD participants ( Table 1, Figures 2D and 3A). In fact, there was no overlap between healthy controls and AD participants.  129 Xe NMR spectra for healthy controls (blue) and AD patients (red). (B) Topographic "streak" plot of (A) depicting the NMR dynamic spectra from the top with SNR in "hotter" colors. Notice a higher SNR in AD patients for a longer time than that of the healthy controls. SNR of 129 Xe-WM (C) and 129 Xe-GM (D) as a function of time for healthy controls (blue) and AD participants (red). The participants inhaled 500 mL of HP 129 Xe and held their breath for 20 s. 129 Xe MRS from the brain region was acquired every 2 s. Notice an increase in 129 Xe signal after approximately 10 s as the 129 Xe reached the brain. At 20 s, the participant exhaled and the 129 Xe signal began to decrease at different rates for AD participants vs. healthy controls for WM and GM.  129 Xe NMR spectra for healthy controls (blue) and AD patients (red). (B) Topographic "streak" plot of (A) depicting the NMR dynamic spectra from the top with SNR in "hotter" colors. Notice a higher SNR in AD patients for a longer time than that of the healthy controls. SNR of 129 Xe-WM (C) and 129 Xe-GM (D) as a function of time for healthy controls (blue) and AD participants (red). The participants inhaled 500 mL of HP 129 Xe and held their breath for 20 s. 129 Xe MRS from the brain region was acquired every 2 s. Notice an increase in 129 Xe signal after approximately 10 s as the 129 Xe reached the brain. At 20 s, the participant exhaled and the 129 Xe signal began to decrease at different rates for AD participants vs. healthy controls for WM and GM. β = + 1 (2) where S is the SNR as a function of time since exhalation, S(T) is the maximum xenon SNR at time T, f is the cerebral perfusion, p is the xenon partition coefficient, T1 is the 129 Xe spin-lattice relaxation time in tissue, and α is the MR flip angle. By fitting the data to this model, we were able to extract the xenon washout parameter for each subject ( Figure 3B). We calculated a 41% and 31% reduction in the GM and WM xenon washout parameters, respectively, in AD patients compared to healthy controls (Table 1).

129 Xe MRI
Lastly, we acquired 129 Xe MRI from the brain of healthy controls and AD participants (Figure 4). We were able to obtain significantly higher SNR images from healthy controls than from AD participants.
By fitting the SNR of each voxel in the three dynamic images, we were able to calculate a xenon washout parameter in each pixel to create a xenon washout parameter map ( Figure 5). Similar to our spectroscopic results, we qualitatively observe a higher xenon washout parameter in age-matched healthy controls ( Figure 5, top) than in AD patients ( Figure 5, bottom). Additionally, the xenon washout parameter remains similar in caudal brain regions in AD participants versus healthy controls, whereas there is a reduction in the localized xenon washout parameter in the prefrontal regions of AD participants. We then quantified this washout time by applying the following pharmacokinetic model to the 129 Xe washout signal [77]: where β is the xenon washout parameter and is defined as, where S is the SNR as a function of time since exhalation, S(T) is the maximum xenon SNR at time T, f is the cerebral perfusion, p is the xenon partition coefficient, T1 is the 129 Xe spin-lattice relaxation time in tissue, and α is the MR flip angle. By fitting the data to this model, we were able to extract the xenon washout parameter for each subject ( Figure 3B). We calculated a 41% and 31% reduction in the GM and WM xenon washout parameters, respectively, in AD patients compared to healthy controls (Table 1).

129 Xe MRI
Lastly, we acquired 129 Xe MRI from the brain of healthy controls and AD participants (Figure 4). We were able to obtain significantly higher SNR images from healthy controls than from AD participants.   By fitting the SNR of each voxel in the three dynamic images, we were able to calculate a xenon washout parameter in each pixel to create a xenon washout parameter map ( Figure 5). Similar to our spectroscopic results, we qualitatively observe a higher xenon washout parameter in age-matched healthy controls ( Figure 5, top) than in AD patients ( Figure 5, bottom). Additionally, the xenon washout parameter remains similar in caudal brain regions in AD participants versus healthy controls, whereas there is a reduction in the localized xenon washout parameter in the prefrontal regions of AD participants.

Discussion
In this work, we made two significant findings. Firstly, AD participants have significantly lower 129 Xe in the GM than healthy controls. Secondly, we found that AD subjects retain 129 Xe within both the GM and WM of the brain significantly longer than healthy controls. Our results support the hypothesis that cerebral perfusion may be affected by AD pathology.
AD is considered primarily a disease of grey matter; however, white matter has been implicated as well in the AD pathology [78]. Our observed decrease in 129 Xe-GM signal between AD participants

Discussion
In this work, we made two significant findings. Firstly, AD participants have significantly lower 129 Xe in the GM than healthy controls. Secondly, we found that AD subjects retain 129 Xe within both the GM and WM of the brain significantly longer than healthy controls. Our results support the hypothesis that cerebral perfusion may be affected by AD pathology.
AD is considered primarily a disease of grey matter; however, white matter has been implicated as well in the AD pathology [78]. Our observed decrease in 129 Xe-GM signal between AD participants and healthy controls suggests a reduction in GM volume or a decrease in 129 Xe uptake in the GM. In contrast, the 129 Xe-WM signal was not different between AD participants and healthy controls.
While there is considerable overlap in many AD biomarkers between healthy controls and AD patients, we calculated a five standard deviation difference in the 129 Xe retention in the GM between AD patients compared to healthy controls (Table 1, Figures 2D and 3A). Furthermore, no AD patients had 129 Xe retention values below that of any healthy control and no healthy control had 129 Xe retention values higher than the lowest AD 129 Xe retention values.
Additionally, we introduce a potential AD biomarker that we denote as the xenon washout parameter from Equation (1). The measurement of xenon signal as a function of time (i.e., Xe washout) using 129 Xe MRS was fit to a pharmacokinetic model (Equation (1)) and the xenon washout parameter was calculated. While other pharmacokinetic models [76] have been developed and fit our data, we utilized the model developed by Martin,et al. [77] because it was a better fit for our data as it is only applied to the washout phase and therefore relied on fewer assumptions than the model developed by Kilian et al. [76]. We calculated that the xenon washout parameter is nearly 2 standard deviations lower in AD participants than in healthy controls for grey matter.
In addition to our spectroscopic data, we were able to localize the xenon washout parameter to different brain regions. Our analysis of the localized xenon washout parameter from the xenon washout parameter maps indicates a lower xenon washout parameter in AD participants than in healthy controls ( Figure 5). While this observation is consistent with our spectroscopic results, the xenon washout parameter obtained from imaging is less accurate than that obtained from spectroscopy because the model was fit using only three data points (three dynamic 129 Xe images) for the imaging data compared to 60 data points for the spectroscopic data. From our qualitative observations of the sagittal β-parameter maps, we observed that the xenon washout parameter is higher in the caudal regions than it is in the frontal lobes ( Figure 5) for both healthy controls and AD participants. In AD participants, while the mean xenon washout parameter decreases throughout the brain, the xenon washout parameter remains higher in the posterior regions of the brain. This observation could possibly indicate lower perfusion in the frontal lobes than in the caudal brain regions. This observation is consistent with previous reports indicating that AD pathology begins near the rostral regions of the brain and slowly migrates towards the caudal regions of the brain [79].
The cerebral perfusion, T1 of 129 Xe in the brain, and the xenon partition coefficient all influence the xenon washout parameter as expressed in Equation (2). Cerebral perfusion has long been demonstrated to be reduced in AD patients [50][51][52][53][54][55][56][57][58]. Work by Binnewijzend et al. reported a difference of cerebral perfusion of approximately 27% which was one standard deviation lower in AD patients compared to healthy controls [58]. In contrast, we observed a difference in the xenon washout parameter of 42% or nearly two standard deviations between AD subjects and healthy controls. Since the xenon washout parameter is a function of cerebral perfusion, our results raise the question of why the xenon washout parameter demonstrates a greater difference between AD participants and healthy controls. While ASL provides a direct measure of cerebral perfusion, our technique incorporates additional factors, such as T1 and the xenon lipid diffusion coefficient, that may be affected in AD patients.
It is possible that the xenon washout parameter is affected by a difference in T1 relaxation time values between healthy controls and AD participants. It is interesting to speculate that small changes in T1 may be caused by the abundance of trace metals in the brain of AD participants [10,[80][81][82][83].
An alternative hypothesis is that the partition coefficient of xenon in the brain tissues is different in AD participants compared to healthy controls. It is well established that Aβ affects the membrane properties of the brain [84][85][86][87][88][89]. We speculate that because of the increased membrane permeability in AD brain tissue, the partition coefficient of xenon in the brain of AD participants is increased, creating a reservoir of xenon that causes a slower wash out. Additionally, AD is known to breakdown the blood brain barrier (BBB) [90]. This breakdown may affect the clearance of xenon from the brain tissue into the blood. However, it is more likely that this phenomenon would cause increased xenon clearance (and hence faster washout in AD patients) rather than our observations of greater retained xenon in AD brains. This biomarker may offer some advantages over existing biomarkers, especially those relying on CSF. Our proposed biomarker does not require an invasive spinal tap and, so far in our preliminary results, shows a greater difference between AD participants and healthy controls compared to CSF Aβ values.
Like all techniques, this proposed biomarker has a number of limitations. First, it requires an expensive and highly specialized 129 Xe polarizer to polarize the 129 Xe gas. Secondly, this technique relies on the patient holding their breath for 20 s, a task that some individuals with dementia could have difficulty with.
Moreover, this study has notable limitations. Firstly, it relied on a small number of subjects precluding accurate determination of the sensitivity and specificity of this technique. Second, because of the small sample size, we were unable to infer the predictive power of this study or to correlate disease severity with the xenon washout parameter value. Lastly, we were unable to differentiate whether our proposed biomarkers are indicative of just AD or all dementias, or even, any neurologic disease. Regardless of these limitations, the proposed 129 Xe xenon washout parameter biomarker has the potential for validation with a larger sample size study to determine both its accuracy and predictive power of impending AD. Additionally, future experiments could attempt to correlate the Montreal Cognitive Assessment (MOCA) scores to 129 Xe retention and the xenon washout parameter; these biomarkers could be tested as a potential correlate to future AD risk.
In conclusion, we demonstrate a difference in the 129 Xe retention between AD participants and healthy controls. Additionally, we introduce the termed xenon washout parameter which accounts for changes in cerebral perfusion, and differences in 129 Xe T1 relaxation and lipid partition coefficients associated with AD pathology. The xenon washout parameter is considerably different in healthy controls and AD participants with little overlap between the two groups.