An MRspec database query and visualization engine ...

    An MRspec database query and visualization engine with applications as a clinical diagnostic and research tool Filip Miscevic, Justin Foong, Benjamin Schmitt, Susan Blaser, Michael Brudno, Andreas Schulze PII: DOI: Reference:

S1096-7192(16)30284-0 doi: 10.1016/j.ymgme.2016.11.003 YMGME 6118

To appear in:

Molecular Genetics and Metabolism

Received date: Revised date: Accepted date:

28 September 2016 7 November 2016 8 November 2016

Please cite this article as: Miscevic, F., Foong, J., Schmitt, B., Blaser, S., Brudno, M. & Schulze, A., An MRspec database query and visualization engine with applications as a clinical diagnostic and research tool, Molecular Genetics and Metabolism (2016), doi: 10.1016/j.ymgme.2016.11.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

An MRspec Database Query and Visualization Engine with

PT

Applications as a Clinical Diagnostic and Research Tool Filip Miscevic1,2, Justin Foong1, Benjamin Schmitt3,4,5, Susan Blaser6, Michael Brudno1,2,5,# and Andreas

RI

Schulze3,5,7,#,* 1

SC

Centre for Computational Medicine, Peter Gilgan Center for Research and Learning, The Hospital for

Sick Children, Toronto, Canada; 2Department of Computer Science, University of Toronto, Toronto,

NU

Canada; 3Department of Paediatrics, University of Toronto, Toronto, Canada; 4Siemens Healthcare, Sydney, Australia; 5Genetics and Genome Biology, Peter Gilgan Center for Research and Learning, The Hospital for Sick Children, Toronto, Canada; 6Department of Diagnostic Imaging, The Hospital for Sick

MA

Children, Toronto, Canada, 7Division of Clinical and Metabolic Genetics, The Hospital for Sick Children,

D

Toronto, Canada.

AC CE P

TE

# Authors contributed equally to the manuscript.

* To whom correspondence should be addressed: The Hospital for Sick Children,

Peter Gilgan Centre for Research and Learning 686 Bay Street,

Toronto ON M5G 0A4 Canada.

Tel: (416) 813-7654 ext. 304828 E-mail: [email protected]

ACCEPTED MANUSCRIPT Abstract Purpose

RI

PT

Proton magnetic resonance spectroscopy (MRspec), one of the very few techniques for in vivo assessment of neuro-metabolic profiles, is often complicated by lack of standard population norms and paucity of computational tools.

SC

Methods

MA

NU

7,035 scans and clinical information from 4,430 pediatric patients were collected from 2008 to 2014. Scans were conducted using a 1.5 T (n = 3,664) or 3 T scanner (n = 3,371), and with either a long (144 ms, n = 5,559) or short echo time (35 ms, n = 1,476). 3,055 of these scans were localized in the basal ganglia (BG), 1,211 in parieto-occipital white matter (WM). 34 metabolites were quantified using LCModel. A web application using MySQL, Python and Flask was developed to facilitate the exploration of the data set.

Results

Conclusions

AC CE P

TE

D

Already piloting the application revealed numerous insights. (1), N-acetylaspartate (NAA) increased throughout all ages. During early infancy, total choline was highly varied and myo-inositol demonstrated a downward trend. (2), Total creatine (tCr) and creatine increased throughout childhood and adolescence, though phosphocreatine (PCr) remained constant beyond 200 days. (3), tCr was higher in BG than WM. (4), No obvious gender-related differences were observed. (5), Field strength affects quantification using LCModel for some metabolites, most prominently for tCr and total NAA. (6), Outlier analysis identified patients treated with vigabatrin through elevated γ-aminobutyrate, and patients with Klippel-Feil syndrome, Leigh disease and L2-hydroxyglutaric aciduria through low choline in BG.

We have established the largest MRSpec database and developed a robust and flexible computational tool for facilitating the exploration of vast metabolite datasets that proved its value for discovering neurochemical trends for clinical diagnosis, treatment monitoring, and research. Open access will lead to its widespread use, improving the diagnostic yield and contributing to better understanding of metabolic processes and conditions in the brain.

Keywords MRSpec; magnetic resonance spectroscopy; inborn errors of metabolism; clinical dataset; computational analysis; neurochemical profiling

ACCEPTED MANUSCRIPT 1 Introduction

SC

RI

PT

Proton magnetic resonance spectroscopy, also known as MRSpec or 1H-MRS, is increasingly used in conjunction with diagnostic brain magnetic resonance imaging (MRI) to obtain neurometabolic profiles. Changes in these profiles are clinically significant, as they are associated with a variety of morbidities, such as inborn errors of metabolism [1], tumorigenesis [2, 3] and neurodegenerative diseases [4]. Furthermore, MRSpec is non-invasive, rapid, and easily integrated as part of an MRI workflow. Here we describe findings from an exploratory analysis of a comprehensive database of 7,035 MRSpec brain scans, as well as the computational tool we have developed to enable the exploration of MRSpec databases such as this one.

MA

NU

While a rich source of information, analyzing quantitative MRSpec data is not without its challenges. As our data have been obtained in a clinical setting, standard procedures (though existing) were not always applied, there was no control cohort, and results may be confounded by the presence of underlying conditions at the time of data acquisition. Furthermore, there are many partitions of the data with possibly different trends, such as age, sex, or scan localization.

AC CE P

2 Methods

TE

D

We sought to address these issues by developing a computational tool for exploring MRSpec data for clinical and research purposes. The tool enables the rapid investigation and validation of trends or research hypotheses concerning patient populations captured in this database. By facilitating the visualization of trends and individual patient data against the whole, the tool is also a promising clinical and diagnostic aide.

2.1 Obtaining MRSpec Data

Patients were examined using either a 1.5 T or a 3 T Philips Achieva clinical MRI system (Philips Healthcare, Andover, MA). Point-resolved spectroscopy (PRESS) acquisitions (NEX = 128, repetition time (TR) = 2,000 ms) were obtained at two different echo times (TE) (short TE = 35 ms, long TE = 144 ms). Two different locations in the left hemisphere were used: basal ganglia (BG), followed by parietooccipital white matter (OCC WM). Choice of volume was according to size of the brain area with focus on tissue homogeneity (minimal edge length of a voxel is 1 cm, minimal voxel volume of 1 cm3). Depending on the clinical indication either a short, 12-min protocol consisting of two scans was used (BG at 144 ms followed by OCC WM at 35 ms), or a ‘metabolic’, 24-min protocol with four scans (BG at 144 and 35 ms followed by OCC WM at 144 and 35 ms). Raw spectrogram data was quantified using LCModel Version 6.2-1L (LCMODEL Inc., Oakville, ON) [5] without partial volume correction or signal normalization against a standard of reference, i.e. in relative terms using LCModel’s inbuilt quantitation routine. Metabolite signals were normalized against the unsuppressed bulk water signal intensity. For measurements taken with the 1.5 T scanner, the metabolites available for analysis were guanidinoacetate (Gua)1, total N-acetylaspartate (tNAA), aceto-acetate (AcAc), lactate (Lac), glutamine (Gln), taurine (Tau), glucose (Glc), scyllo-inositol (Scyllo), acetone (Acn), total choline (tCho), myoinositol (Ins), glutamate (Glu), aspartate (Asp), L-alanine (Ala), N-acetylaspartylglutamate (NAAG), total 1

More precisely, it is a simulated singlet to account for an occasional significant signal at 3.78ppm.

ACCEPTED MANUSCRIPT creatine (tCr), γ-aminobutyrate (GABA) as well as the lipid resonances (Lip) at 2.0 ppm (Lip20), two species at 1.3 ppm (Lip13a/b), one at 0.9 ppm (Lip09). Macromolecule resonances (MM), MM09, MM12, MM14, MM17 and MM20, were also recorded [5].

PT

For measurements taken with the 3 T scanner, metabolites analyzed included those above as well as glycerophosphocholine (GPC), phosphocholine (PCh), phosphocreatine (PCr), and creatine (Cr).

RI

tNAA, tCho, tCr and Glx were calculated, where applicable, by taking the sum of PCr and Cr for tCr; NAA and NAAG for tNAA; Cho, GPC and PCh for tCho; and Gln and Glu for Glx.

MA

NU

SC

The 7,035 scans of 4,430 unique patients were dated February 2008 to October 2014. 3,664 scans were conducted using 1.5 T scanner, while 3,371 scans were conducted using 3 T. 5,559 of the scans were conducted using a 144 ms echo time; 1,476 scans were conducted using a short echo time. 3,055 of these scans were localized in the basal ganglia (BG), 1,211 in parieto-occipital white matter (OCC WM), 55 in the cerebellum and 26 in fronto-temporal white matter (FT WM). Location data was unavailable for 958 of the scans. Research ethics board approval was obtained for retrospective chart review to record patient histories, medication, and diagnoses, which was transcribed manually and deidentified prior to being added to the database. For 6,555 of these scans, the indication for the scan and/or a working or final diagnosis was available.

TE

D

Mean patient age was 4.6 years, (0-18 years, median 3.3 years). The male-to-female ratio was 1.4:1. 4,931 scans were known to have been conducted under some form of sedation, with 4,468 of these conducted under general anesthesia. Average age of anesthetization was 4.4 years (median 3.2 years).

2.2 The MRSpec Database Query and Visualization Engine

AC CE P

Patient health information was deidentified, assigned a study ID, and imported into a MySQL 5.6 database. The database query and visualization tool was implemented in Python 2.7 and a lightweight Flask 0.10 web application. The full source code is available at https://github.com/compbioUofT/mrspec.

2.3 Data Preprocessing

Because the data are noisy, we excluded certain results from analysis with user-definable thresholds for the confidence of measurement determined by LCModel when fitting data to the raw spectrograph curve (see [5]). The thresholds used are available on the project page at https://github.com/compbioUofT/mrspec/blob/master/config/metabolite_thresholds.txt. We assumed that although each patient underwent 1H-MRS for a particular indication, the severity of the condition and the likelihood that it would systematically affect metabolite measurements was distributed normally. This would mean that while the data would potentially have greater variance, it would still be most dense around the population norm, and so it would be analytically useful to compare individual patients to overall trends in the data.

2.4 Age-adjusted Standard Deviation Score Given that metabolite concentrations can vary with age, we found it useful to compare a given data point with a moving ‘average’ of an age-matched cohort rather than the population average. For every

ACCEPTED MANUSCRIPT

PT

metabolite for each patient, we calculated the expected metabolite value at that age using a linear regression of the 50 closest patients by age on either side of the current patient being considered, for a total of 100 patients. The standard deviation of this patient’s metabolite with respect to the regression was then calculated, hereafter referred to as the standard deviation score. Since standard deviation is a unitless measure that indicates where the measurement falls in the spread of the data, it is appropriate for comparing differences in concentration between metabolites.

RI

2.5 Examining Trends using Linear Regression

SC

Partitions of the database were examined for trends using the multiple linear regression model

D

MA

NU

Where is metabolite concentration, is age, and is a binary indicator variable for whether a particular measurement belongs to one partition or the other (i.e., 1 for male, 0 for female). The statistical significance of coefficients of regression and was examined to determine whether this partition revealed trends explainable above and beyond chance, using a Bonferroni correction at an level of significance, where and refers to the number of hypotheses that were tested (i.e., number of partitions that were examined). Since 3 partitions were examined, trends were only considered significant if p < 0.0167. Refer to the tables in the supplementary material for the p values of these coefficients for all metabolites.

AC CE P

TE

Certain metabolites followed a clear sublinear trend, for which we used a zero-shifted logarithm of age, , in place of the age if the adjusted R2 was better. Residual plots were examined in determining the appropriateness of the log-linear model over the linear model. In comparing trends that were not different partitions of the same data (i.e. trends between different metabolites), we calculated the statistic for the coefficients of regression [6], comparing it to the critical value for a two-sided test at :

Where are the slopes of the regressions being compared and SE is the standard error of the respective coefficients.

3 Results and Discussion 3.1 A Tool to Visualize and Analyze MRSpec data We developed a robust tool for exploring and visualizing this highly partitioned database of MRSpec measurements and to mitigate issues of noise with this kind of clinical data. The key functionality of our tool is summarized in Figure 1, with additional information provided in the sub-sections below.

ACCEPTED MANUSCRIPT

2

PT

1

NU

SC

RI

3

D

MA

4

7

AC CE P

6

TE

5

Figure 1: Some of the features of the MRSpec Query and Visualization engine. 1) Options to select the metabolites to display, gender, field strength of localization of the scan and to export the data as a spreadsheet or image. 2) Graphs of the queries. The graph pictured shows total N-acetylaspartate measurements in the basal ganglia, stratified in intervals of 0.5 standard deviation scores versus the age of the patient. This visualization is useful for identifying outliers in the data (see Methods for calculations). 3) Selecting a data point opens up a side bar with information about the scan such as indication and diagnosis. 4) Information about metabolites that are elevated or lowered are also displayed. Clicking ‘More Options’ reveals a popup window with additional query options such as 5) searching for particular patients by ID and 6) adding a patient’s results manually to compare to the data. 7) When two or more scans are selected, a second tab can be clicked which displays useful information about the group such as pooled standard deviation scores.

We opted for a clean, uncluttered interface with a comprehensive set of options for refining queries. Queries can be filtered on a wide range of scan parameters (localization, field strength, echo time) and patient information (sex, age, reason for scan). Ranges of values can also be queried – for instance, a particular age range can be displayed, or results themselves can be queried based on whether they fall outside of a ‘normal’ range (see Figure 1). Different queries can be overlaid on top of each other – for instance, to compare measurements from males and females. Individuals or groups of patients can also be singled out – by manual entry of results or by directly accessing the record in the database – in order to compare them to the remainder of the cohort. Exporting any of this data for further analysis – either as an image of the graphs generated or of the raw query data in a spreadsheet – is likewise straightforward.

ACCEPTED MANUSCRIPT

MA

NU

SC

RI

PT

Additionally, we have augmented the search capabilities of the tool to include medical records apart from just the MRI requisition, and so we can query working- or final diagnoses, comorbidities, concomitant medications, and patient histories. This is implemented using a range of sophisticated keyword search options, including terms to include, terms to exclude, wildcards, and whether to search for these terms in the MRI requisition form or the patient records. Altogether, this allows clinically relevant information to be evaluated in relation to the MRSpec data on the fly. As we shall discuss, this greatly simplifies the process of mining the data for clinically significant trends. For instance, outliers can be rapidly identified and investigated whether they share the same underlying pathology. It also becomes easy to evaluate how a particular pathology, such as a cancer, affects the MRSpec metabolite profile, if at all. If there is an established association between a treatment and the metabolite profile, its progress can be evaluated using MRSpec – and in fact, our tool is capable of investigating whether such an association can be measured. All of these possibilities become reality thanks to a sophisticated visualization tool. To the best of our knowledge, no paper to date has reported on a body of MRSpec data of this size (n = 7,035). The richness of available data and flexibility of our tool gives us the unique opportunity to characterize and report on trends, and compare our findings to those in the literature. The tool is available for public use at https://mrspec.ccm.sickkids.ca/.

3.2 Characterizing the Data and Showcasing the MRSpec Database Query and Visualization Engine

AC CE P

TE

D

The comprehensive options for refining and visualizing queries using the tool afford the ability to gain powerful insight into trends in the data, which has potential clinical significance diagnostically as well as in shaping research questions and hypothesis. Several examples underlining the research and clinical utility of the tool are described below. Note that for consistency, we use the 3 T scans; the results of the 1.5 T scans are consistent with reported trends but not shown.

3.2.1 Age-related Metabolite Changes in the Brain Several studies have examined age-related changes in metabolite concentrations in normally developing children [7-14]. Previously reported upward trends in the ratios of NAA:Cr and NAA:tCho and a downward trend in the ratio of tCho:Cr in both BG and WM are consistent with our present findings [7]. That these changes appear to stabilize in later life is also consistent with previous findings [8]. The most comprehensive of those studies also examined location-related developmental changes which are discussed in detail below [9, 12]. Consistent with previous work, NAA increased with age in both BG and WM, with the greatest increase in the first 2.5 years of life, as illustrated in Figure 2A [9, 12]. While the role of NAA in the brain is debated, it appears to be associated with the functional viability of neurons, and so its rapid increase in the first few years of life is thought to reflect the massive increase in neural arborisation and synaptogenesis [15]. It is also associated with myelination, which increases substantially over the first few years of life [15-17]. It is important to note that the absolute NAA concentrations reported here may differ from those in the literature. For example, the basal ganglia NAA concentration is approximately 20 mM on average, which is somewhat high compared to literature values that normally range from about 9 mM to 13 mM [18, 19]. This is because the data acquired by LCModel is not absolutely quantitative, and depend on the acquisition protocol (this is discussed in the conclusion below).

ACCEPTED MANUSCRIPT

50

PT

40 30

RI

20 10 0 0

1000

2000

SC

NAA [mM per kg wet wgt.]

A. Age vs. NAA

3000

4000

5000

6000

5000

6000

NU

Age [days]

MA

12 10 8

D

6 4 2 0

1000

AC CE P

0

TE

tCho [mM per kg wet wgt.]

B. Age vs. tCho

2000

3000

4000

Age [days]

C. Log of Age vs. Ins

Ins [mM per kg wet wgt.]

12 10 8 6 4 2 0

1

10

100

1000

Age [logarithm of days] Figure 2: Age versus N-acetylaspartate (NAA), inositol (Ins) and total choline (tCho) at 3 T in the basal ganglia (BG).

tCho remains at a constant concentration beyond the first 100 days of life in both BG and WM, although there is great variance in concentration in the neonatal period, possibly reflecting the large turnover and

ACCEPTED MANUSCRIPT myelination events occurring early in development. Interestingly, while previous studies have reported a decrease of tCho from birth onwards [9, 12], we observed high variance in tCho, which later stabilized by 100 days (Figure 2B). The reason for this variance warrants further investigation.

PT

myo-Inositol, a major brain osmolyte and astrocyte marker, revealed a downward trend during the first 100 days of life and remained constant afterwards in both BG and WM (Figure 2C). GABA, Glu and Gln were constant at all ages, consistent with previously reported findings [9, 12].

SC

RI

3.2.2 Increase of Creatine but stable Phosphocreatine over time

18

NU

16 14 12

MA

10 8 6

Cr PCr

D

4

tCr

0 0

TE

2 50

100

150

200

Age [days]

AC CE P

Concentration [mM per kg wet wgt.]

A. Age vs. Cr, PCr and tCr

B. Age vs. Cr, PCr and tCr

Concentration [mM per kg wet wgt.]

25

20

15

tCr Cr

10

PCr 5

0 0

1000

2000

3000

4000

5000

6000

Age [days] Figure 3: Age versus creatine (Cr), phosphocreatine (PCr) and total creatine (tCr) in the basal ganglia (BG) at 3 T: A. in the first 200 days of life, B. at all ages.

ACCEPTED MANUSCRIPT

SC

3.2.3 Differences Related to Scan Localization

RI

PT

We examined the relationship between Cr and PCr at 3 T in both BG and WM, as only at 3 T can PCr signals be resolved from Cr. Both metabolites increase rapidly in the first 200 days (6.6 months) of life (Figure 3A), with Cr and subsequently tCr continuing to rise gradually throughout childhood and adolescence (Figure 3B). PCr and Cr in the first 200 days do not appear to differ significantly in their rate of increase (p > 0.1), however PCr remains constant after 200 days (Figure 3B). This is at odds with previous findings, which report stabilization in the Cr to PCr ratio after the first year of life [9]. However, the Cr measurements at 3 T are heteroscedastic: variance increases over time. Interestingly, this heteroscedasticity is not observed in any other metabolites; another intriguing research question generated by visualizing the data.

NU

3.2.3.1 Basal Ganglia and White Matter Consistent with previous work, it was found that tCr was higher in BG at 1.5 T, irrespective of age, compared with WM (p < 0.01) [9]. No significant difference in Ins was detected (p > 0.1).

MA

3.2.3.2 Cerebellum and Basal Ganglia tCr was not found to be significantly different in the cerebellum (p > 0.2) compared with BG, which is inconsistent with previous findings, although many of these measurements were taken to rule out cerebellar anomalies, and in some cases have confirmed diagnoses of cerebellar atrophy.

TE

D

tCho appeared to be elevated in the cerebellum compared to BG at 3 T (p < 0.0123) but not at 1.5 T (p > 0.4), possibly for the same reason.

AC CE P

Analyses of other metabolites revealed no significant differences, or were not possible due to the paucity of cerebellar measurements following data preprocessing.

3.2.4 Sex-related Differences

There were no statistically significant differences observed with regards to sex-related differences in any of the metabolites (p > 0.045) over all ages. Ozturk et al. [20] report a minor but statistically significant increased ratio of NAA:Cr in female versus male children, by 0.2%, in right frontal white matter – but nowhere else in the brain. We are unable to comment directly on this finding as our scans were obtained exclusively in the left hemisphere, but otherwise, their results are consistent with the data examined.

3.2.5 Explaining Trends in the Data: Effect of Field Strength on Measurements For some of the metabolites, such as tCr and tNAA, a strong bimodality was observed in the data, as in Figure 4. The ability to rapidly compare different partitions in the data allowed us to quickly identify that this bimodality was completely explained by the field strength, and this was confirmed by multiple linear regression (p < 0.01) [21]. This demonstrates that differences in scan parameters can lead to significant quantitative differences, and moreover, that such differences can be found quickly using this tool.

ACCEPTED MANUSCRIPT A. Age vs. tCr

PT RI

15

SC

10

5

NU

tCr [mM per kg wet wgt.]

20

0

1000

MA

0

2000

3000

4000

5000

6000

D

Age [days]

TE

B. Age vs. tCr

tCr [mM per kg wet wgt.]

AC CE P

20

15

1.5 T

10

3T

5

0 0

1000

2000

3000

4000

5000

6000

Age [days] Figure 4: A. Total creatine (tCr) concentrations plotted against age – notice the strong bimodality. B. tCr concentrations plotted against age, with 1.5 T in blue and 3 T in orange. The field strength completely explains the bimodality (p < 0.01).

Statistically significant differences in field strength on metabolite concentration (p < 0.0167 for at least one of b2 and b3, refer to methods) were observed for AcAc, Cr, CrCH2, Gln, Glu, Glx, MM20, NAA, Tau, tCho, tCr, tNAA.

ACCEPTED MANUSCRIPT

PT

The observed differences in the quantitative values between field strengths can be explained by metabolite T1 relaxation times that consequently lead to metabolite-specific differences in steady state signal during averaging with a short TR of 2 s. While this is uncompensated for in the LCModel quantification, a difference will exist between the quantitative values from the two field strengths, but this can be assumed systematic across all scans, and particularly strong for metabolites with long T1 times, e.g. NAA [22, 23].

RI

3.2.6 Outlier analysis: GABA Elevated in Patients Treated with Vigabatrin

MA

NU

SC

All patients with an elevated GABA level (n = 6) were observed to suffer from infantile spasm or a generalized seizure disorder. These patients frequently have low GABA levels, hence the cause of the seizures, so it was hypothesized that these patients were on vigabatrin therapy, which inhibits GABA transamination and would therefore explain the elevated GABA levels [24]. Follow-up with this cohort confirmed that all of these patients were on vigabatrin therapy at the time of the scan, as seen in Figure 5. Importantly, unlike Cr or NAA, GABA has no distinguishable peaks on a raw spectrogram. Thus, this trend is only discernible by a quantitative analysis of MRSpec data. Furthermore, this information may be of clinical value; for instance, the responsiveness of GABA levels to vigabatrin therapy may be predictive of relapse or outcome, a research question that warrants further investigation.

D

GABA and Vigabatrin Treatment

TE

6 5

AC CE P

GABA [mM per kg wet wgt.]

7

4 3 2 1 0

0

500

1000

1500

2000

2500

3000

Age [days] All patients

Infantile spasms or generalized seizure

On vigabatrin treatment

Figure 5: Elevated γ-aminobutyrate (GABA) levels in patients suffering from infantile spasm or generalized seizure disorders.

3.2.7 Outlier analysis: Choline decreased in patients with Klippel-Feil syndrome, Leigh disease and L2-Hydroxyglutaric aciduria

ACCEPTED MANUSCRIPT

RI

PT

Additionally, the tool is useful in generating hypotheses from the data. For example, cherry-picking several outliers of decreased tCho in basal ganglia at 3 T (Figure 6) revealed patients with Klippel-Feil syndrome, Leigh disease, and L2-Hydroxyglutaric aciduria. Decreased tCho has not previously been reported in these conditions. In the lattermost condition, one toddler with an elevated Cho to Cr ratio has been reported [25]. Taken together, one could hypothesize that over the course of L2Hydroxyglutaric aciduria, Cho first increases before it decreases as a reflection of high membrane turnover in early stages and membrane loss in later stages of the disease.

SC

tCho in L2HGA, Leigh's disease and Klippel-Feil Syndrome

NU

10

MA

8 6

All Patients L2HGA Leigh's

D

4 2 0 1000

2000

AC CE P

0

TE

tCho [mM per kg wet wgt.]

12

3000

Klippel-Feil

4000

5000

6000

Age [days]

Figure 6: Decreased total choline (tCho) in patients with Klippel-Feil syndrome, Leigh’s disease, and L2-Hydroxyglutaric aciduria (L2HGA) measured from scans taken at 3 T in the basal ganglia (BG).

4 Conclusion

MRspec is one of the very few techniques that can be used for the neurochemical profiling, or more specifically, in vivo assessment of metabolic processes and conditions in the brain. It provides a noninvasive monitor of brain metabolism allowing early and specific ascertainment of disease processes and treatment monitoring. We present a comprehensive database of MRSpec data and a robust, flexible tool for discovering trends for clinical diagnostic and research purposes. Cursory analysis of the database reveals age and location-related trends, some of them consistent with the literature [9, 12] and others previously unreported. While we present the ‘raw’ data from the whole group of 4,430 children, Bluml at al. [12] selected 309 out of the cohort of approximately 2,500 children to study metabolic brain maturation. It is interesting to see that our approach using ‘unfiltered’ data acquired in a routine clinical setting confirms the observation made in this previous study. Furthermore, we demonstrate that changes in MRI parameters, such as field strength, affect the outcome of findings so that caution must be taken when comparing MRSpec results obtained using different MRI parameters. This is due to the fact that the concentrations, i.e. those provided by

ACCEPTED MANUSCRIPT

PT

LCModel, are not absolutely quantitative as there are biases due to the acquisition protocol. The quantification assumes fully-relaxed spectra, which would require a TR of 6s at 3T and 3.5s at 1.5T, which is not clinically feasible. Concentrations are weighted by this and other cascaded effects in a nonlinear relationship. That explains the difference in absolute metabolite values obtained at 3 T and 1.5 T, as well as those reported in the literature. Hence, one has to stress the relative nature of these protocols.

NU

SC

RI

With regards to our tool, our ability to report these new findings is a testament to its utility. As we have demonstrated, outlier analysis is easily performed applying our tool and reveals observations that may lead to the generation of new scientific hypotheses. Provided that the same acquisition parameters are used, our tool can be used to compare data between institutions. Furthermore, many of the technical challenges addressed using the tool would also be useful in other kinds of multi-dimensional patient data apart from MRSpec data. We hope to release the tool and database for outside analysis, and for use with other MRSpec databases, as well as other kinds of highly partitioned patient data.

MA

Acknowledgements

[2]

[3]

[4] [5] [6] [7]

[8]

[9]

[10]

Prust, M. J.; Gropman, A. L.; Hauser, N. New frontiers in neuroimaging applications to inborn errors of metabolism. Mol. Genet. Metab., 2011, 104(3), 195-205. Law, M.; Yang, S.; Wang, H.; Babb, J. S.; Johnson, G.; Cha, S.; Knopp, E. A.; Zagzag, D. Glioma grading: sensitivity, specificity, and predictive values of perfusion MR imaging and proton MR spectroscopic imaging compared with conventional MR imaging. Am. J. Neuroradiol., 2003, 24(10), 1989-1998. Stadlbauer, A.; Moser, E.; Gruber, S.; Buslei, R.; Nimsky, C.; Fahlbusch, R.; Ganslandt, O. Improved delineation of brain tumors: an automated method for segmentation based on pathologic changes of 1HMRSI metabolites in gliomas. Neuroimage, 2004, 23(2), 454-461. Martin, W. R. MR spectroscopy in neurodegenerative disease. Mol. Imag. Biol., 2007, 9(4), 196-203. Provencher, S. W. Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn. Reson. Med., 1993, 30(6), 672-679. Clogg, C. C.; Petkova, E.; Haritou, A. Statistical methods for comparing regression coefficients between models. American Journal of Sociology, 1995, 1261-1293. Hashimoto, T.; Tayama, M.; Miyazaki, M.; Fujii, E.; Harada, M.; Miyoshi, H.; Tanouchi, M.; Kuroda, Y. Developmental brain changes investigated with proton magnetic resonance spectroscopy. Dev. Med. Child Neurol., 1995, 37(5), 398-405. Brooks, J. C.; Roberts, N.; Kemp, G. J.; Gosney, M. A.; Lye, M.; Whitehouse, G. H. A proton magnetic resonance spectroscopy study of age-related changes in frontal lobe metabolite concentrations. Cereb. Cortex, 2001, 11(7), 598-605. Pouwels, P. J.; Brockmann, K.; Kruse, B.; Wilken, B.; Wick, M.; Hanefeld, F.; Frahm, J. Regional age dependence of human brain metabolites from infancy to adulthood as detected by quantitative localized proton MRS. Pediatr. Res., 1999, 46(4), 474-485. Cady, E. B.; Penrice, J.; Amess, P. N.; Lorek, A.; Wylezinska, M.; Aldridge, R. F.; Franconi, F.; Wyatt, J. S.; Reynolds, E. O. Lactate, N-acetylaspartate, choline and creatine concentrations, and spin-spin relaxation in

AC CE P

[1]

TE

References

D

We wish to thank students Elizabeth Mason, Tommy Tam, Moeen Imran, Andrew Zasowski, and Jacalyn Kelly for their great work in chart reviews and entering data. Students were funded by the Starbucks Summer Studentship program. FM was supported by a Natural Sciences and Engineering Research Council of Canada Undergraduate Student Research Award.

ACCEPTED MANUSCRIPT

[15] [16]

[17]

[20] [21] [22] [23] [24] [25]

PT

AC CE P

[19]

TE

D

[18]

RI

[14]

SC

[13]

NU

[12]

MA

[11]

thalamic and occipito-parietal regions of developing human brain. Magn Reson Med, 1996, 36(6), 878886. Toft, P. B.; Leth, H.; Lou, H. C.; Pryds, O.; Henriksen, O. Metabolite concentrations in the developing brain estimated with proton MR spectroscopy. J Magn Reson Imaging, 1994, 4(5), 674-680. Bluml, S.; Wisnowski, J. L.; Nelson, M. D., Jr.; Paquette, L.; Gilles, F. H.; Kinney, H. C.; Panigrahy, A. Metabolic maturation of the human brain from birth through adolescence: insights from in vivo magnetic resonance spectroscopy. Cereb.Cortex, 2013, 23(12), 2944-2955. Krishna, S. H.; McKinney, A. M.; Lucato, L. T. Congenital genetic inborn errors of metabolism presenting as an adult or persisting into adulthood: neuroimaging in the more common or recognizable disorders. Semin. Ultrasound. CT MR, 2014, 35(2), 160-191. Kreis, R.; Ernst, T.; Ross, B. D. Development of the human brain: in vivo quantification of metabolite and water content with proton magnetic resonance spectroscopy. Magn. Reson. Med., 1993, 30(4), 424-437. Moffett, J. R.; Ross, B.; Arun, P.; Madhavarao, C. N.; Namboodiri, M. A. A. N-Acetylaspartate in the CNS: From Neurodiagnostics to Neurobiology. Prog. Neurobiol., 2007, 81(2), 89-131. Madhavarao, C. N.; Arun, P.; Moffett, J. R.; Szucs, S.; Surendran, S.; Matalon, R.; Garbern, J.; Hristova, D.; Johnson, A.; Jiang, W.; Namboodiri, M. A. Defective N-acetylaspartate catabolism reduces brain acetate levels and myelin lipid synthesis in Canavan's disease. Proc. Natl. Acad. Sci. U. S. A., 2005, 102(14), 52215226. Chakraborty, G.; Mekala, P.; Yahya, D.; Wu, G.; Ledeen, R. W. Intraneuronal N-acetylaspartate supplies acetyl groups for myelin lipid synthesis: evidence for myelin-associated aspartoacylase. J. Neurochem., 2001, 78(4), 736-745. Wu, W. E.; Gass, A.; Glodzik, L.; Babb, J. S.; Hirsch, J.; Sollberger, M.; Achtnichts, L.; Amann, M.; Monsch, A. U.; Gonen, O. Whole brain N-acetylaspartate concentration is conserved throughout normal aging. Neurobiol. Aging, 2012, 33(10), 2440-2447. Jessen, F.; Fingerhut, N.; Sprinkart, A. M.; Kuhn, K. U.; Petrovsky, N.; Maier, W.; Schild, H. H.; Block, W.; Wagner, M.; Traber, F. N-Acetylaspartylglutamate (NAAG) and N-Acetylaspartate (NAA) in Patients With Schizophrenia. Schizophr. Bull., 2013, 39(1), 197-205. Ozturk, A.; Degaonkar, M.; Matson, M. A.; Wells, C. T.; Mahone, E. M.; Horska, A. Proton MR spectroscopy correlates of frontal lobe function in healthy children. Am. J. Neuroradiol., 2009, 30(7), 1308-1314. Drost, D. J.; Riddle, W. R.; Clarke, G. D.; Group, A. M. T. Proton magnetic resonance spectroscopy in the brain: report of AAPM MR Task Group #9. Med. Phys., 2002, 29(9), 2177-2197. Mlynárik, V.; Gruber, S.; Moser, E. Proton T1 and T2 relaxation times of human brain metabolites at 3 Tesla. NMR Biomed., 2001, 14(5), 325-331. Li, Y.; Srinivasan, R.; Ratiney, H.; Lu, Y.; Chang, S. M.; Nelson, S. J. Comparison of T(1) and T(2) metabolite relaxation times in glioma and normal brain at 3T. J. Magn. Reson. Imaging, 2008, 28(2), 342-350. Chudomelova, L.; Scantlebury, M. H.; Raffo, E.; Coppola, A.; Betancourth, D.; Galanopoulou, A. S. Modeling new therapies for infantile spasms. Epilepsia, 2010, 51 Suppl 3, 27-33. Read, M. H.; Bonamy, C.; Laloum, D.; Belloy, F.; Constans, J. M.; Guillois, B.; Kottler, M. L.; Verhoeven, N. M.; Jakobs, C. Clinical, biochemical, magnetic resonance imaging (MRI) and proton magnetic resonance spectroscopy (1H MRS) findings in a fourth case of combined D- and L-2 hydroxyglutaric aciduria. J. Inherit. Metab. Dis., 2005, 28(6), 1149-1150.