Detection of Vitamin C

The detection and quantification of vitamin C is essential and helpful for research, especially the non-invasive methods. However, the invasive methods are commonly used due to their higher sensibility to possibly determine the metabolites of low concentrations even though it is not applicable to all aspects to collect the data from the living subjects. For example, plasma ascorbate concentration is often determined by directly collecting blood samples and then using high-performance liquid chromatography. Biopsy is used to measure its concentration of some specific organs, but not available to collect the data from living brains.

The development of MRS provides one of the non-invasive methods to measure chemicals within the body. Because of the limitation of MRS, the chemicals would be detectable at least the concentration larger than hundreds of micro-molar (Table 1).

However, the concentration of vitamin C in human brain is within the range of several millimolar [44], it is hence high enough to be detectable via proton MRS. Dr. Terpstra and her colleagues [23, 37, 45] have proposed to use a spectral editing method to detect and quantify the concentration of ascorbate and other antioxidants, and their results are comparable to those estimated by using other methods.

When blood ascorbate levels are below 70 μM, the kidney’s sodium-dependent vitamin C transporters would reabsorb ascorbate to prevent the loss by renal clearance [65]. However, when the ascorbate concentration is higher than 70 μM, the body would excrete ascorbate more rapidly [66]. The plasma half-life of ascorbate is dependent on the intake. During the period of deficient intake, the plasma half-life of ascorbate is between 8 to 40 days [67], and ascorbate has a half-life time of about 30 minutes when the higher intake levels lead to more rapid excretion [68]. Some early clinical studies of vitamin C reported the benefits to use high-dose treatment on some diseases suffering from oxidative stress, such as cancer. However, studies about its pharmacokinetics of high-dose treatment are still rare in the recent decade. How much vitamin C needs to be administrated via the intravenous or oral route to achieve the desired concentration and how long it takes are questions that remain to be answered.

Dr. Padayatty and his colleagues [43] discussed the plasma vitamin C concentration varying with the route of administration, oral and intravenous use, in healthy volunteers. Vitamin C concentrations are tightly controlled when using oral dose and only intravenous administration can produce high plasma vitamin C concentrations (Fig. 7). Besides, the plasma concentrations reach the maximal values

immediately after the intravenous administration, while they arrive at the peaks with about 3 hours delay after oral supplements. The pharmacokinetic data at high intravenous dose of vitamin C in cancer patients are still sparse, and such research about brains is even rare. Fig. 8 shows the preliminary results from two healthy volunteers [69]. The concentrations of vitamin C in human brains reach the maximum in about 24 hours after intravenous bolus, with only 30% to 40% of increment, which implies that the concentration of vitamin C is also strictly controlled in human brains.

It is essential to collect more in vivo spectra data from healthy volunteers and patients for further investigations.

Figure 7. Plasma vitamin C concentrations in healthy subjects after intravenous or oral administration of vitamin C. The larger subfigure shows the change of plasma vitamin C concentration after the 1.25-g oral or intravenous dose administrated at steady state (N = 12). The inset subfigure shows peak plasma vitamin C concentrations as a function of dose after oral or intravenous administration of vitamin C [43].

Figure 8. It shows the normalized concentration change of ascorbic acid and the sum of glutamine, glutamate and glucose after an intravenous bolus of 3-g ascorbic acid.

The spectral data were acquired from occipital lobe. The ascorbate concentration reaches a maximal value in about 24 hours after the intravenous bolus [69].

Table 1. Ranges of some steadily MR-detectable metabolite concentrations reported for normal adult human brains and biopsy tissues [70].

Metabolites Concentration range (mmol/kgww)

NAA 7.9-16.6 (average 10.3)

NAAG 0.6-2.7 Choline (total) 0.9-2.5

Creatine 5.1-10.6 Glutamate 6.0-12.5 Glutamine 3.0-5.8 Myo-inositol 3.8-8.1 Phosphocreatine 3.2-5.5 Aspartate 1.0-1.4 GABA 1.3-1.9

Chapter 3

Detectability and Reliability of Vitamin C Using MRS

3.1 Motives

The importance of vitamin C has shown in many previous studies. It relates to the storage of iron and stimulation of the immune system. It also has the connection with some brain diseases, such as AD and MS as well as its correlation to age. Furthermore, it is the most concentrated non-enzymatic antioxidant in the central nervous system.

Since vitamin C plays a significant role in brain functioning [44, 71], a non-invasive and quantitative detection method of vitamin C in the human brain would be highly desirable.

The concentration of ascorbate in brains is about proportional to neuron density, and its concentration in human brain is roughly 1.0 mM [44]. This level is sufficiently high to be detectable with MRS. However, in traditional clinical routines of MRS, the concentration of ascorbate was not discussed and even its component was not

included during MR spectral analysis. Table 2 shows the chemical shifts of ascorbate and other metabolites, which possess the chemical shifts close to it [70, 72, 73]. Due to the multiple overlaps with the spectra of other metabolites, such as Gln (C2H group at 3.75 ppm), Glu (C2H group at 3.74 ppm) and myo-inositol (mI), therefore, it is not trivial to quantify the concentration of ascorbate (Fig. 9). To evaluate the detectability of ascorbate and the possible impacts on other evaluated metabolite concentrations, we focused on Gln, Glu, and mI as they all own resonance peaks close to the most prominent ascorbate signal at 3.73 ppm.

The detection of ascorbate was recently reported in human subjects at 4T and in rat brain at 9.4T by MEGA-PRESS edited spectroscopy [23, 45]. Although the technique is based on the typical 3D localization spectral acquisition method, PRESS, it is more time-consuming for several reasons. Such a spectral editing method needs to be acquired twice, one time for with and the other for without frequency selective RF pulse. Thus, the scan time at least becomes double (Fig. 4(a)). Besides, the longer TE acquisition of spectra sacrifices the SNR, so it is essential to increase the number of average to compensate the loss of the signal.

In order to avoid these disadvantages and apply MR spectroscopy techniques to

vitamin C quantification, we use traditional PRESS sequence on a clinical 3T MR scanner to acquire in vivo spectra. The goal of this thesis is to evaluate the possibility and to verify the reliability of detecting vitamin C in human brains by using clinical standard 3T MR spectroscopy methods.

Figure 9. Basis spectra of NAA, mI, Glu, Gln, and Asc. These spectra were used in LCModel analysis and the resonance peaks of Glu, Gln, and Asc are close to each other at about 3.75 ppm.

Table 2. ¹H chemical shifts of ascorbic acid and other metabolites. The metabolites that possess resonance peaks close to ascorbic acid are listed. DD, double-doublet; M, multiplet; Q, quartet; T, triplet [70, 72, 73]

Metabolites Group Chemical shift Multiplicity

Asc C6H2 3.73 M

3.2 Material and Methods

3.2.1 In Vivo Spectra Collection and Analyses

From our data archives we collected 76 in vivo single voxel spectra (SVS) of 49 subjects from different brain regions. These included cerebellum (25 cases, average age = 27.50 ± 6.35 years), frontal lobe (29 cases, average age = 28.56 ± 6.77 years), occipital lobe (7 cases, average age = 29.43 ± 16.28 years), parietal lobe (10 cases, average age = 18.60 ± 16.09 years), and others (5 cases, average age = 26.40 ± 15.61 years) (Fig. 10). These spectra were acquired on a 3T Siemens Magnetom Trio system (Erlangen, Germany) with a common PRESS sequence, voxel size = 8 cm³, TE = 30 ms, TR = 3000 ms, NEX = 128 (with water-suppression), 1024 complex data points, a receiver bandwidth = 1200 Hz (centered on the water resonant frequency) with a standard ¹H quadrature single-channel coil.

In order to observe the impact on other metabolite concentrations if Asc was included in the LCModel (v. 6.1-4A) estimation [74], these data were fitted twice within the spectral range of 0.2-4.2 ppm in the frequency domain. We included 14 metabolites in the standard basis-set: aspartate (Asp), Cre, GABA, glucose (Glc), Gln,

Glu, glycerophosphocholine (GPC), guanidoacetate (Gua), phosphocholine (PCh), mI, NAA, N-acetylaspartylglutamate (NAAG), scyllo-inositol (Scyllo), and taurine (Tau).

The analyses were operated once using the basis-set with an Asc basis spectrum (15 metabolites included in total) and once without the Asc basis spectrum (14 metabolites). To simplify the discussion of this approach we will use the description

“with” and “without” Asc basis throughout the following.

Metabolite concentrations were evaluated as creatine ratio (1/Cre). The differences of the estimated concentrations using a “with” Asc and a “without” Asc basis set analyses were calculated for each metabolite. This was done by deducting the “with”

Asc results from the “without” Asc results. The inter-individual standard deviations (SDs) of these estimated metabolite concentrations were calculated to depict variations of these metabolites in various regions. Furthermore, we evaluated the consistency of the estimated concentrations of different metabolites through comparing these SDs.

In order to verify the reliability of LCModel analysis of Asc concentration we simulated a “virtual titration”, where a modified Asc spectrum was added to a human

in vivo spectrum from the collected data were chosen, with different FWHM values

(0.076, 0.043 and 0.033 ppm) representing the linewidth range of the evaluated in vivo spectra. The FWHM values of in vivo spectra are actually the spectral line-width of NAA evaluated by LCModel.

The modifications adapted the ideal Asc basis spectrum according to the line shape, linewidth, and phase characteristics of the in vivo spectra before they were added. The necessary parameters for this purpose were derived from the corresponding in vivo spectra by LCModel. These parameters were related to some features of the in vivo spectrum and were applied to the ideal Asc basis using eq. [1]

in Ref. [74] to obtain the modified Asc spectrum. Six different Asc concentration levels were included: 0.1 mM, 0.2 mM, 0.5 mM, 1.0 mM, 1.6 mM, and 2.0 mM. For further details of this process, refer to the paragraph Virtual Titration section below.

In addition, the impact of SNR was evaluated. For this purpose the Asc spectrum SNR was altered to values of 20, 10 and 5 before adding it to the in vivo spectra.

Random noise signals were added with different amplitudes in order to change the SNR of the Asc signal. Here, the SNR is defined as the maximum amplitude of the Asc signal divided by the mean amplitude of the generated random noise in the frequency domain.

Figure 10. The acquired spectra from human brains are mostly from (a) parietal lobe, (b) occipital lobe, (c) cerebellum, and (d) frontal lobe. The large plot of each subfigure shows the sagittal view of the human brain. The upper right one and the lower right one show the corresponding coronal and transverse views, respectively.

All the voxel sizes of these spectra are of 2x2x2 cm³.

3.2.2 Virtual Titration

The virtual titration experiment aims to modify the Asc basis and then add it to the in

vivo spectra to observe whether the estimated concentration of Asc would be

proportional to the artificial added Asc concentration. Modifications were done to explore various Asc basis spectra, reflecting the possible range of the in vivo spectra.

The formulas for the modification are from the appendix of Ref. [74], which describe the parameters taken into consideration and their constraints during curve fitting.

LCModel uses the concepts of linear combination to describe the analyzed spectra as the superposition of each spectrum of the corresponding metabolites with different weightings. In other words, parameters C_l in eq. (7) are the concentrations of the NM

metabolites. Due to the magnetic field inhomogeneity, the lineshape coefficients Sn in eq. (7) are responsible for peak distortions. The NM metabolite spectra of the in vitro basis set can be written as M_l(ν;0,0). In the meanwhile, the parameters ε_l account for small errors in referencing the spectra, and the parameters ν_l relate to shorter T2

of in vivo spectra. The zero- and first-order phase corrections are φ₀ and φ , and the ₁ baseline is represented by the summation of β_jB_j(ν_k). Therefore, the fitted spectrum can be modeled as eq. (7) starting from the in vitro spectra and the difference )

ˆ( Y νk

of it and the real in vivo spectrum should be minimized. Equation (8) represents the relation between FID signals of metabolites (ml) and their spectra, where F indicates the discrete Fourier transform.

First, an in vivo spectrum was analyzed in LCModel to estimate the concentrations of different metabolites. The evaluated parameters for baseline correction, phase adjustment, line shape, spectral shifting, and line broadening are logged in addition to the actual analysis result. In the next step, these parameters, which represent the features of this in vivo spectrum, are applied to modify the Asc basis spectrum to create an artificial Asc spectrum and change the concentration of the artificial signal by changing its amplitude. Additionally, a randomly generated noise signal is added to this modified Asc spectrum. This is done to achieve a certain SNR in the resulting dataset. Then this modified Asc spectrum (including noise) and the in vivo spectrum are combined to create a new spectrum. This final spectrum is analyzed in LCModel to observe the effects on concentration estimation before and after adding the modified Asc signal. For each case with a certain added concentration, we generated the random signal 10 times to create 10 modified in vivo spectra with the same SNR

metabolite concentrations and their SDs were calculated and used for comparison to determine the impact of SNR.

Before applying virtual titration to general in vivo spectra, we extracted spectral information of each metabolite from the analysis basis-set and created some artificial spectra with different weighting combinations to simulate the in vitro spectra acquired from uniform phantoms. In order to approach the reasonable in vivo metabolic spectra in human brains, the simulated concentrations of metabolites are chosen according to some previous review articles [44, 70]: NAA 10 mM, Cre 7.5 mM, PCh 0.6 mM, Lac 0.4 mM, Asp 1.2 mM, mI 6.0 mM, Glu 8.0 mM, Gln 4.5 mM, and Glc 1.0 mM.

Because the resonance peaks of last four metabolites are close to each other, we added them gradually into the simulated in vitro spectrum instead of at once. The combined spectrum was analyzed in LCModel to check if the quantification is consistent with the initial concentration settings. It would be helpful to clarify the overlapping effects between different metabolites during spectral curve fitting when proceeding virtual titration. Ascorbate is the target metabolite of virtual titration. Therefore, we changed the concentration of ascorbate to 0.2 mM, 0.5 mM, 0.8 mM, 1.0 mM, 1.5 mM, 2.0 mM, and 5.0 mM, and added it to the simulated in vitro spectrum. These spectra were analyzed in LCModel to observe the relationship between estimated concentrations

and real concentrations of ascorbate.

In this dissertation, we first showed the analysis results from in vivo spectra and compared the quantification difference between the results from the analysis basis-set without and with ascorbate basis. Besides, the Cramér-Rao Lower Bounds (CR-SD) of glutamine and glutamate in two analyses were compared to make the impact on quantification using different basis-sets more clearly. In order to verify the feasibility of virtual titration, we demonstrated the quantification results using simulated in vitro spectra. Furthermore, virtual titration was applied to in vivo spectra and discussed the quantification effects resulting from different spectral SNR values and linewidths to verify whether vitamin C in human brains could be stably detected and quantified via traditional PRESS sequence and LCModel analysis.

3.3 Results

3.3.1 In Vivo Spectra Analysis

The in vivo spectra analysis results derived from LCModel were classified into five regions and are shown in Figs. 11 and 12. In three cerebellum and two frontal lobe spectra the Asc concentrations could not successfully be evaluated by LCModel.

These datasets are excluded from further analysis since their CR-SD of the derived Asc concentrations are larger than 25%. Fig. 11 depicts the concentration ratios of five commonly evaluated metabolites including Asc within the five investigated regions.

Estimated concentrations of total NAA (NAA+NAAG) are located between 1.1 (1/Cre) and 1.4 (1/Cre), and the SDs range from 0.061 (1/Cre) to 0.143 (1/Cre). Hence, the relative SD, which is calculated as the SD divided by the corresponding estimated concentration, is about 12%. Similarly, the relative SDs of mI and choline (GPC+PCh) are about 13% and 15%, respectively. CR-SD values of total NAA, mI and choline range from 3% to 4%. In the five brain regions investigated, we found average values for vitamin C concentration ranged from 0.442 (1/Cre) to 0.535 (1/Cre).

Corresponding SDs were between 0.04 (1/Cre) and 0.13 (1/Cre), resulting in a relative SD of about 24%. This is twice the value that was found for NAA. The relative SD of

Glu+Gln with about 25% is similar to that of Asc. CR-SD values of these two metabolites are 7% and 11%, respectively.

Including Asc into the LCModel basis set may alter the estimated concentrations of other metabolites. Figure 12 displays this impact in relation to the evaluated Asc concentration after inclusion to the basis set. It shows the concentration differences of four different metabolites in all 71 datasets. From Figs 12(b) to 12(e) show the concentration differences of metabolites according to spectral data from different brain regions and the concentration differences of mI in four regions are obviously independent of the ascorbate concentration (p > 0.5). Besides, the analysis of choline shows resembling outcomes of the associated p-value larger than 0.05 in these four subfigures. From Fig. 12(a) the R² of mI is equal to 0.005 of associated p-value larger than 0.5, and the R² of GPC+PCh is 0.052 with p-value larger than 0.05. The results from individual regions and all subjects are consistent to confirm that no change is detectable as a function of Asc concentration for these two metabolites plotted. The statistic result of Glu+Gln in cerebellum is of p-value less than 0.3, and from other three regions their p-values are less than 0.1. Comparing to the results from Fig. 12(a), the concentration differences of Glu+Gln ranging from 0.067 (1/Cre) to 0.416 (1/Cre) shows an increasing trend with higher ascorbate concentrations (R² = 0.189, p <

0.001). Although the NAA concentration tends to go down slightly from -0.006 (1/Cre) to -0.248 (1/Cre) as a whole and the NAA concentration difference shows dependency on Asc concentration in fontal lobe (p < 0.01), the analysis results from 71 cases does not show such a phenomenon (R² = 0.019, p > 0.2). Figures 12(b) and 12(e) display results from parietal and occipital lobe, respectively. Their concentration distributions of ascorbic acid, ranging from 0.4 (1/Cre) to 0.6 (1/Cre), are much narrower compared with data from all cases, as shown in Fig. 12(a).

It is observed that CR-SD values of Asc and Glx are larger than that of other metabolites. Therefore, we focused on those two metabolites to plot the relation between their CR-SD values and the corresponding Asc concentration in Fig. 13. It depicts the trend that the CR-SD values decrease with a higher Asc concentration. The CR-SD values are above 20% for low Asc concentrations of about 0.25 (1/Cre) and decrease to 7% for Asc concentrations above 0.8 (1/Cre). The CR-SD changes of Glx do not express an obvious dependency on Asc concentration. There is a clear difference, however, between Glx CR-SD values from analysis “with” and “without”

Asc in the basis dataset. The latter are in most cases smaller.

Figure 11. The estimated relative concentrations of five metabolites, Asc, mI, choline (GPC+PCh), total NAA (NAA+NAAG), and the sum of Glu and Gln (Glu+Gln) in five different brain regions are depicted. The error bars show the SD of the relative concentration of each metabolite in different regions. The metabolites with larger concentration display a larger SD. Total NAA and Glu+Gln are examples of this. The relative SD of Glu+Gln (about 25%) is higher than that of total NAA (about 15%).

Concentrations of Asc and mI are at a similar level, and their relative SDs are also about 25% and 15%, respectively. The concentration of choline is the lowest among these metabolites, and its relative SD is similar to total NAA and mI.