Synopsis

Keywords: Data Analysis, Spectroscopy

Numerical Monte Carlo analysis was performed to quantify metabolite-metabolite correlations of spectral origin in ³¹P MRS spectra acquired from human brain at both 3 and 7 Tesla without any confounding biological correlations. Significant correlations were found for many ³¹P-containing metabolite pairs. In particular, the 3 and 7 Tesla NAD⁺-NADH correlations are significantly different because of the field strength difference. These results demonstrate that it is necessary to incorporate metabolite-metabolite correlations originating from spectral overlap into statistical models that correlate MRS measurements with clinical parameters when overlapping ³¹P signals are of clinical interest.

Introduction

In vivo ³¹P MRS has been widely used to study brain disorders. However, there is severe spectral overlap in ³¹P MRS spectra even at 7 Tesla (for example, spectral overlap among α-ATP, NAD⁺, NADH, and UDPG). Since spectral overlap per se can lead to significant correlations between metabolites¹ it is necessary to incorporate correlations of spectral origin into statistical models for correlating overlapping ³¹P MRS signals with clinical parameters. In this work, Monte Carlo analysis was performed to systematically quantify metabolite-metabolite correlations of spectral origin in ³¹P MRS spectra in the absence of any confounding correlations of biological origins. The effects of field strengths and background spectral baseline on metabolite-metabolite correlations were also characterized in detail.

Methods

In vivo ³¹P MRS data In vivo data of five healthy participants for each field strength were analyzed. The in vivo ³¹P MRS data were acquired using Siemens Skyra 3 Tesla and Magnetom 7 Tesla scanners (Siemens Healthcare, Erlangen, Germany) and home-built coil assemblies using a circular ³¹P coil with 7.0 cm diameter as described previously². Briefly, in vivo ³¹P MRS data acquisition used a pulse-acquire sequence. Relevant parameters at 3 Tesla were: TR = 2 seconds; spectral width = 5 kHz; number of acquisitions = 128; number of data points = 1024. Identical parameters were used at 7 Tesla except that TR = 3 seconds.
Data processing and quantification The same data processing procedure was applied for both field strengths. The first two data points in FID were set to zero to suppress the baseline in ³¹P MRS spectrum³. Subsequently, zero-filling, 1-Hz exponential line broadening, Fourier transform, zero- and first-order phase corrections, and chemical shift referencing (PCr was set at 0 ppm) were performed. All spectra were quantified using an in-house developed fitting program⁴ implementing the linear combination model fitting algorithm⁵. Spectral regions covering -20 to 1 ppm at 3 Tesla and -20 to 10 ppm at 7 Tesla, respectively, were quantified. The downfield region at 3 Tesla was excluded². All basis data were numerically calculated with chemical shifts and coupling constants taken from the literature^6,7. UDPG was excluded in the basis data of 3 Tesla due to its low sensitivity. The background spectral baseline was modeled as a polynomial. For calculating absolute concentrations, total ATP (sum of α-, β-, and γ-ATP) were used as an internal reference, assumed to be 9 mM⁸.
Monte Carlo analysis of in vivo spectra Individually fitted metabolites including spectral baseline were derived from linear combination model fitting. Subsequently, two different datasets comprising metabolites only and metabolites + baseline were generated. Lastly, noise with its level derived from the corresponding in vivo spectrum was added. For each spectrum 2000 different noise realizations were used. Each spectrum was then fitted using the linear combination model fitting program described in Data processing and quantification.
Correlation analysis Using concentrations of individual metabolites from Monte Carlo analysis, Pearson’s correlation coefficients were calculated to investigate the effect of field strength and spectral baseline on metabolite-metabolite correlations.

Results

Figure 1 shows representative 3 Tesla (left panel) and 7 Tesla (right panel) in vivo ³¹P MRS spectra. Means and standard deviations of metabolite concentrations obtained at both field strengths are listed in Table 1. Tables 2 and 3 show the mean Pearson’s correlation coefficient matrices derived from Monte Carlo simulations performed using in vivo ³¹P MRS data at 3 Tesla (n = 5) and 7 Tesla (n = 5), respectively. Due to the severe spectral overlap between NAD⁺ and NADH, the Pearson’s correlation coefficient of the NAD⁺-NADH pair at 3 Tesla is quite large (r = -0.73 ~ -0.75) with and without the baseline. In comparison, the Pearson’s correlation coefficient for the NAD⁺-NADH pair is markedly reduced at 7 Tesla (r = -0.56). While UDPG is detectable at 7 Tesla it has significant correlations with both NAD⁺ and NADH. The background spectral baseline was found to have significant influence on certain metabolite-metabolite correlations. For example, the Pearson’s correlation coefficient of the PC-PE pair changed from -0.11 without the baseline to +0.15 with the baseline at 7 Tesla.

Discussion

Compared to the crowded short echo time proton MRS spectra there is generally a wide chemical shift dispersion in ³¹P MRS spectra. To the best of our knowledge correlations originating from spectral overlap in ³¹P MRS spectra have not been taken into account when correlating overlapping ³¹P MRS signals with clinical parameters. However, at magnetic field strength such as 3 Tesla there is significant spectral overlap in the downfield region involving phosphoesters, Pi, and the prominent membrane phosphate signal that is not well understood². Even at 7 Tesla there is severe spectral overlap involving α-ATP, NAD⁺, NADH, and UDPG. The results of this study showed that spectral correlations can be significant for ³¹P MRS data and are affected by both field strength and the background spectral baseline. Therefore, correlations in ³¹P MRS data due to spectral overlap need to be built into statistical models for clinical analysis.

Acknowledgements

No acknowledgement found.