![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
3
*
Dementia Research Service, Burke Medical Research Institute, White Plains, NY 10605;
ESA, Incorporated, Chelmsford, MA 01824;
**
Antioxidants Research Laboratory, Jean Mayer U.S. Department of Agriculture Human Nutrition Research Center on Aging at Tufts University, Boston, MA 02111; and
Departments of Biochemistry and Neuroscience, Cornell University Medical College, New York, NY 10021
3To whom correspondence should be addressed. E-mail: Bkristal{at}burke.org.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
KEY WORDS: • dietary restriction • HPLC • serum metabolite • multivariate • biomarker • rats
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
–11
). DR alters many basic physiologic processes, including metabolism, hormonal balance and the generation of, detoxification of and resistance to reactive oxygen species (12
). These data are consistent with multiple epidemiologic studies in humans showing the detrimental effects of obesity on health (13
). The beneficial effects of DR on animal health reveal that the physiologic effect of DR in mammals is large, and encourage us to hypothesize that a serum profile that can distinguish long-term patterns of energy intake exists and can be identified using the DR rat as a model system.
To establish serotypes that accurately reflect substantial changes in food and/or energy intake such as that which occurs in animals subject to DR (1
–10
,14
–17
), we are investigating low-molecular-weight molecules that are chemically redox active using HPLC coupled with coulometric electrochemical array detectors. Initial studies showed that HPLC separations coupled with coulometric array detectors (18
–27
) could detect 
1200 compounds in rat sera, of which 300 were analytically reliable and 240/290 were biologically reliable in young female/male rats [(14
) and companion paper(28
)]). Among them, 101 (female) and 112 (male) metabolites differed between AL and DR rats by automated analysis. As described in the companion paper (28
), attention to analytical issues reduced the number of metabolites under study to 63 in female rats and 52 in males. Metabolite subsets enabled both HCA and PCA to group the initial cohorts of both male and female rats with 100% accuracy.
These previously obtained data demonstrate feasibility, that is, that quantitative analysis of selected sera metabolites can yield sufficient information by which to classify the dietary intake of a group of rats and set the stage for validation of these metabolic serotypes in independent cohorts. In this report, we utilize independent cohorts of male and female rats to address the next stage of this problem, i.e., whether specific metabolites exist that robustly (e.g., across different cohorts of rats, and eventually across species) enable determination of dietary group of origin.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
Details of the rats and husbandry conditions used in this study were reported previously (14
). Briefly, male and female Fischer 344 x Brown Norway F1 rats were obtained monthly from the National Institute on Aging colony at Harlan (Indianapolis, IN). The basic animal husbandry for the animals used followed NIA/NIH guidelines as implemented by Harlan. Rats were fed NIH-31 (AL rats only) or vitamin/mineral–fortified NIH-31 (DR rats only); see (14
) for detailed diet compositions. All rats were housed individually, and DR feeding regimens (40% less food than eaten by AL rats) were implemented at 6 wk of age. Rats were imported from Harlan at 5 mo of age and killed 1 mo later. As noted in the previous report (14
), we found that rats in our colony that consumed food AL ate slightly less food than at Harlan, and restriction was therefore slightly modified. Specifically, once adjusted, food restriction was imposed at a level of 35% less than ad libitum consumption (14
). Sera were collected after killing by decapitation to avoid known differential effects of anesthesia on variables of interest. Collected blood was allowed to clot on ice for 30 min before centrifugation (1000 x g, 10 min). Female cohorts 1 and 2 had 8 AL and 8 DR rats; female cohort 3 had 6 AL and 5 DR rats; male cohort 1 had 5 AL and 8 DR rats; male cohort 2 had 7 AL and 8 DR rats. All animal experiments were performed under institutionally approved protocols and complied with the Guide for the Care and Use of Laboratory Animals (29
).
HPLC separations and coulometric array detection was conducted essentially as described previously, using an ESA CoulArray system (ESA, Chelmsford, MA) (14
,18
,25
,26
).
Data analysis is described in the text. Data were analyzed using the programs CEAS 504 (ESA), Statview 5.0.1 (SAS Institute, Cary, NC) and Pirouette 2.7/3.0 (Infometrix, Woodinville, WA). Descriptions of the applications of the techniques of hierarchical cluster analysis (HCA) and principal component analysis (PCA) are presented in the companion paper (28
); mathematical formulas are available in the Pirouette manual. Other aspects of the data analysis are described in the text.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
)]. In this report, we present data showing that these metabolites previously identified as informative in one cohort can also be used to distinguish dietary groups in independent cohorts. The first part of the Results and Discussion section shows that the markers of female rats are sufficient to allow discrimination of the groups by HCA and PCA. The second portion shows that initially identified male serotype is also capable of distinguishing the groups with HCA and PCA. Subsequent analyses identified additional metabolites that could be used to establish more powerful data set for future studies.
Validation of female serotype.
The companion paper (28
) identified 63 variables that were biologically (across cohorts) and analytically robust and that were sufficient to discern dietary group in cohort 1 using classification algorithms. The previous study (on cohort 1) was carried out using two basic scaling techniques, range scaling and autoscaling, and three grouping algorithms, single, complete and centroid. Each of the six combinations of preprocessing and grouping algorithms yielded 100% accuracy with cohort 1. As noted in that report, however, autoscale preprocessing involves mean-centering and variance-scaling the data, and is a generally robust technique. In contrast, in range scale preprocessing, which is commonly used for graphing, the highest value in the dataset is assigned a value of 1 and the lowest a value of 0. Remaining values are determined by proportion. This method is highly sensitive to outliers and has limited utility in subsequent cohorts because the overall range in a subsequent cohort may change. For this reason, we will not pursue range scale preprocessing further, and will focus on autoscale-based analysis. Each of the three grouping methods (single, centroid, and complete) correctly identified the dietary group of origin of 100% of the cohort 1 samples, but the complete grouping algorithm gave the best separation in cohort 1. Therefore, our initial validation test (in cohort 2) will use the complete grouping algorithm.
HCA with the 63 previously identified metabolites distinguished AL and DR rats with 94% accuracy.
HCA was used to determine whether the metabolites previously shown to identify the dietary group of origin of a given rat in one cohort retained this ability in an independent cohort. Using autoscale preprocessing and the complete grouping algorithm, HCA was 94% accurate at distinguishing the diet group of the 16 rats in cohort 2. Analyses using single and centroid grouping algorithms are also shown for comparison (Fig. 1
).
|
show that there were no outliers in the dataset from cohort 2. The two-dimensional projection of a plot of the first three principal components (Fig. 2
B) shows that a planar separation of the two dietary groups existed. Figure 2
C shows that the principal components identified were, as expected, weaker than in the initial study [>30% in Principal Component 1 vs. >50% initially (28
)], but nearly 60% of the variability was still captured within the first two principal components. We further addressed the ability of PCA to capture the analyte by analyte variability present by looking at modeling power. Modeling power analysis revealed that 10 components captured nearly 60% of the variation of nearly 60% of the analytes (Mean modeling power ± SD = 59 ± 15%; Median modeling power 59%).
|
and 2
reveal that it was possible to use all 63 metabolites to distinguish dietary groups using predefined algorithms in an independent cohort. We then sought to further improve our dataset.
As described in the companion paper (28
), our initial strategy was chosen to reduce Type II statistical errors (believing that the serum concentration of a given metabolite does not differ between AL and DR when in reality it does) at the expense of making an increased number of Type I statistical errors (believing that the serum concentration of a given analyte does differ between AL and DR when in reality it does not). We expect that many of the metabolites initially identified represent statistical noise (artifacts). Because removing noninformative data can help classification algorithms work more efficiently, we again pared the dataset by keeping only those metabolites that reached a significance of P 
0.2. Overall, t test analysis determined that 37 of the 63 metabolites tested differed between AL and DR rats at P 0.2. Note that only 1 in 25 metabolites would be expected to be significant in both P 0.2 tests by chance alone, suggesting that the majority of metabolites in this 37 metabolite dataset displayed a real statistical difference (i.e., P < 0.05) between AL and DR rats.
To determine whether the dataset remaining (37 metabolites) retained the properties of interest after this winnowing, we repeated the multivariate analyses described above on the same cohort used above. HCA showed that discrimination between dietary groups was retained in the smaller dataset, and, indeed, slightly improved (compare "single" in Figs. 1
, 3
and see below). In addition, empirical analysis suggested that the misclassification seen in the complete analysis resulted from a single metabolite because upon the removal of this analyte, 100% accuracy was obtained (Fig. 3
, Panel B). PCA analysis showed that there were no outliers in the 36-metabolite dataset (Fig. 4A
). The two-dimensional projection of a plot of the first three principal components shows that a planar separation of the two dietary groups existed (Fig. 4
B). Figure 2
C shows that the primary principal component was better able to capture analyte variability after we reduced the metabolite number (45% in Principal Component 1 vs. <30% initially). Modeling power was also enhanced (Mean modeling power ± SD = 70 ± 12%; Median modeling power 72%). These data suggest that we can clean the dataset of biologically or statistically "noisy" metabolites without losing any of the discriminating power of the metabolome.
|
|
yielded a separation of 0.38 (0.244 + 0.136 = 0.380) with 94% accuracy. Complete analysis of the 37-peak dataset (Fig. 3
A) yielded a separation of 0.55 (0.330 + 0.218 = 0.548) with 94% accuracy. Complete analysis of the 36-peak dataset (Fig. 3
B) yielded a separation of 0.54 (0.198 + 0.337 = 0.535) with 100% accuracy. These data suggest that cleaning the dataset of variables that appear noisy within that dataset may indeed contribute to an apparent increase in separation between the groups. Tests of the effect of removing these variables on the ability to discern diet in independent cohorts will be examined below.
In addition, the ability of the 26 metabolites removed to distinguish AL and DR groups was tested. As shown in Figure 5
, clusters built on these 26 metabolites had only 60–63% accuracy in distinguishing AL from DR (note that 50% is the floor for accuracy given the nature of the analysis). We also further pursued these analyses using PCA. PCA analysis was also unable to distinguish AL and DR (data not shown). The inability of PCA to distinguish AL and DR can be appreciated by examining the series of 2-D plots shown in Figure 6
, which show the two-dimensional representations of components 1–5 against each other. Together these data indicate that the metabolites removed from this analysis did not contribute significant signal to the discrimination of cohort 2 using the 63-metabolite dataset.
|
|
, the 63-metabolite profile was able to distinguish dietary group using the previously defined approach in a third independent cohort (cohort 3) with 91% accuracy using HCA (Autoscale, complete). In contrast to the efficacy of HCA observed with the 63 metabolite dataset, HCA (Autoscale, complete) was unable to distinguish AL and DR rats using the 37-metabolite dataset (63% accuracy, Fig. 7
C). These data suggest two points: 1) there does exist at least one peak in the 26 cut during the previously described analysis that does contribute to distinguishing AL and DR sera under some conditions. We therefore elected to continue using 63 peaks for the immediate future. 2) HCA analysis, at least that attainable using the complete grouping method, lacks sufficient versatility/robustness to accomplish our aims.
|
). These data suggest the possibility that HCA-based approaches are too weak overall to perform the necessary grouping. Another possibility, that the problem might lie in the grouping algorithm used for our HCA validation test (complete), was the problem. As shown in Figure 9
, single- and centroid-based analyses were sufficient to give 100% accuracy in those samples grouped, but were unable to assign groups to 18% (2 of 11) of the samples. Incremental analysis was able to group the samples with 91% accuracy. Thus, alternative grouping algorithms were better than the complete algorithm, but HCA still did not reach the accuracy and power offered by PCA. These exploratory approaches are consistent with the interpretation that PCA-based analyses are more powerful in addressing the grouping problem with our datasets. We are currently extending this approach to n-dimensional analysis based on pattern recognition driven categorization algorithms and supervised expert systems.
|
|
A 52-metabolite data set was best in identifying the dietary groups in the first male cohort, but we analyzed all 76 metabolites in the second male cohort to avoid losing potentially valuable variables.5
After preliminary automated analysis of the sera from male cohort 2, each of the 76 peaks (metabolites) in the HPLC chromatograms of the sera from this cohort was inspected manually. Although the initial study had addressed analytical and biological validation including cohort-cohort instability [companion paper (28
)], the closer examination now conducted determined that 20 of the 76 peaks were not sufficiently analytically and biologically robust to be included in the study. The levels of these peaks displayed inconsistencies in intragroup DR and AL samples. The remaining 56 metabolites contained 42 of the 52 metabolites defined previously and 14 of the 56 metabolites that were not included in the previous set of 52 that best distinguished the sera in cohort 1. The levels of the remaining 56 metabolites were confirmed manually.
HCA and PCA analyses with the 56 metabolite dataset distinguished AL and DR samples in male cohort 2 with >80% accuracy.
PCA and HCA analyses were carried out to determine whether the 56-metabolite data set retained the ability to identify dietary group of origin. Incremental analysis with Autoscale preprocessing was chosen as the test case. The results show that HCA with the 56-metabolite data set separated the dietary groups with 87% accuracy, and the remaining three grouping algorithms (shown for comparison) also separated at >80% accuracy (Fig. 10)
. With the 56-metabolite data set, PCA showed that there were no outliers in the 15 samples analyzed (Fig. 11A
). Despite relatively weak modeling power (see below), the first three components were able to distinguish AL and DR sera with 87% efficacy (Fig. 11
B). The first 2–3 mathematical factors (components) captured 50% of the total variability present, and 6–7 components captured 80% of the variability present (Fig. 11
C). Modeling power analysis revealed that even 10 components captured only about one third of the variation present in the dataset (Mean modeling power ± SD = 33 ± 18%; Median modeling power 34%, Fig. 11D
). A two-dimensional multiplot showed that there were no two components that could efficiently separate the groups (Fig. 12
). Excluding metabolite #671 (which was found to have a very high X-residual value) did not improve the separations (HCA and PCA analyses, data not shown).
|
|
|
). Using this 20-metabolite dataset, HCA distinguished the samples at 66.7, 86.7, 80 and 80% efficacy with grouping algorithms, single, centroid, complete and incremental, respectively (Fig. 13
). Thus, this shorter dataset had minimal effect on the efficacy of HCA.
|
|
). Analysis again showed that there were no outliers in the 15 samples analyzed (Fig. 14
A). Figure 14
B shows that 7 AL and 8 DR rats were distinguished by 3 components in PCA at 100% accuracy Two mathematical factors (components) captured >50% of the total variability present (compared with 2–3 components in the 56-metabolite data set), and 5–6 components captured 80% of the variability present, compared with 6–7 components in the 56-metabolite data set (Fig. 14
C). Modeling power analysis revealed substantial improvement (Mean modeling power ± SD = 72 ± 16%; Median modeling power 75%, Fig. 14
D). Figure 15
shows a two-dimensional multiplot of five components. Principal Component 1 could be used to isolate AL samples from DR samples at > 90% accuracy in conjunction with components 2, 3, 4 or 5. This separation differed from that obtained using the 56-metabolite dataset (Fig. 12)
, in which DR and AL samples could not be distinguished using any two components. The above results indicate that a subgroup of metabolites in the 76 metabolite data set identified in the primary study retained sufficient information to distinguish dietary group of origin by classification algorithms.
|
|
). We reevaluated the other 221 metabolites and identified 9 that were able to help differentiate the dietary groups in the male cohort 2. These 9 metabolites had P 0.2. In studies utilizing HCA, the inclusion of these 9 metabolites improved the ability of the 20-metabolite dataset to discern diet group of origin. The combined 29-metabolite dataset enabled HCA to distinguish the dietary group of origin with 100% accuracy using the incremental grouping algorithm (data not shown). PCA with these 29 metabolites separated the cohort at 100% accuracy (data not shown).
Back-testing analyses were then carried out to examine whether these 9 additional metabolites affected the separations previously obtained in the other cohorts. We quantified the 9 metabolites in the samples of the first male cohort and carried out analyses with the data set combining the 9 metabolites and the original data set. When the 9 metabolites were added to the subset of 52 metabolites in the first cohort, HCA grouped samples at 100% accuracy with grouping algorithm incremental similar to the original 52 metabolite subset. However, the 9 metabolites affected the grouping accuracy of the 52 metabolites with grouping algorithms single, centroid, and complete. X-residuals from PCA analysis based on the 61 (9 + 52) metabolites were less than those based on the original 52 metabolites (i.e., metabolite variability was better explained by the same number of components). The PCA based on the 61 (9 + 52) metabolites distinguished the two dietary groups of the first cohort at 100% accuracy as did the original 52 metabolites. This indicates that use of the 9 additional metabolites did not weaken the power of the data set to discriminate provided more powerful classification algorithms were used. We will therefore continue to evaluate these 9 new metabolites in our next studies, because more powerful classification algorithms will be utilized.
As noted at several points in this and the companion paper (28
), PCA consistently outperformed HCA in enabling us to distinguish categories. At first consideration, this appears surprising because HCA is specifically designed to distinguish classes, whereas PCA is designed to simplify datasets by collapsing variables into synthetic mathematical factors termed components. One possibility, which we currently consider the most likely, is that the data reflected in the primary principal components had an increased signal:noise ratio relative to that observed in the raw data. Alternatively stated, this implies that larger numbered (i.e., later) principal components would be comprised primarily of noise. Another possibility is that the third "representational" dimension offered by rotation gives us additional viewing flexibility that increases separation. Two points argue against this being the major reason why PCA seems more effective. The first is that HCA works in multidimensional space, and therefore should not be hindered by the limitation of our visualization. A second is that two-dimensional visualizations of the PCA data also appeared better than the HCA data (e.g., compare Figs. 13
and 15
). In addition, the possibility that the increased efficacy offered by PCA is "artifactual" would suggest that supervised cluster-based analysis (e.g., KNN or K-nearest neighbor) would be as efficacious as supervised PCA-based analysis (e.g., SIMCA, Soft Independent Modeling of Class Analogy) in distinguishing classes. Our preliminary evidence suggests that this is not true (unpublished data). Thus, overall, our evidence suggests that PCA-based data simplification is eliminating an aspect of biological noise in the metabolome that we have not yet recognized. We expect to continue to examine this issue as we move to larger datasets.
In conclusion, the major finding of this paper is that previously identified components of the serum metabolome robustly retain sufficient information to distinguish AL and DR female and male rats in independent cohorts. We utilized additional cohorts to further improve the serotypes. For example, after this validation experiment, data from the second male cohort was then used to further improve the profile by eliminating noninformative metabolites and identifying additional potentially informative metabolites. Although the 63-metabolite female serotype does retain some statistically noisy metabolites, paring the dataset to 37 appears to discard potentially useful information, and we therefore choose to retain these analytes in our current serotype model at this time. Our study also demonstrated the utility and arguably, the necessity, for the application of progressively more powerful grouping algorithms to establish categorical separations. As seen to a lesser extent in the female dataset, PCA is significantly more powerful than HCA in utilizing the data set to distinguish dietary groups of origin. In the analyses currently in progress, we are developing expert system– based approaches to further aid in categorical separations.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
2 Supported by National Institutes of Health National Institute on Aging R01-AG15354 (B.S.K.), ESA, Incorporated and the Winifred Masterson Burke Relief Foundation.
4 Abbreviations used: AL, ad libitum; DR, dietary restricted; HCA, hierarchical cluster analysis; PCA, principal component analysis.
5 For reasons discussed above, we will use only autoscale preprocessed data during the validation portions of this study. Our initial validation test in HCA will use the incremental grouping algorithm, which gave the best separations in the proof of principle study (28
). Data using the single, centroid, and complete grouping algorithms on cohort 2 will be presented for comparison.
Manuscript received 9 June 2001. Initial review completed 25 July 2001. Revision accepted 11 February 2002.
|
LITERATURE CITED |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
1. Kristal, B. S. & Yu, B. P. (1994) Aging and its modulation by dietary restriction. Yu, B. P. eds. Modulation of Aging Processes by Dietary Restriction 1994:1-36 CRC Press Boca Raton, FL. .
2. Weindruch, R. & Walford, R. (1988) The Retardation of Aging and Disease by Dietary Restriction 1988 Charles C. Thomas St. Louis, MO. .
3. McCay, C. M. (1935) Cellulose in the diets of rats and mice. J. Nutr. :435-447.
4. Maeda, H., Gleister, C. A., Masoro, E. J., Murata, I., McMahan, C. A. & Yu, B. P. (1985) Nutritional influences on aging of Fischer 344 rats: II Pathology. J. Gerontol. 40:671-688.
5. Carlson, A. J. & Hoelzel, F. (1946) Apparent prolongation of the life span of rats by intermittent fasting. J. Nutr. 31:363-375.
6. Goodrick, C. L., Ingram, D. K., Reynolds, M. A., Freeman, J. R. & Cider, N. (1990) Effects of intermittent feeding upon body weight and lifespan in inbred mice: interaction of genotype and age. Mech. Age. Dev. :55.
7. Ross, M. H. & Bras, G. (1970) Food preference and length of life. Science (Washington, DC) 190:165-167.
8. Weindruch, R. H. & Walford, R. L. (2000) Dietary restriction in mice beginning at one year of age: Effects on life-span and spontaneous cancer incidence. Science (Washington, DC) 215:1415-1418.
9. Nolen, G. A. (1972) Effect of various restricted dietary regimens on the growth, health, and longevity of albino rats. J. Nutr. 102:1477-1494.
10.
Tannenbaum, A. (1945) The dependence of tumour formation on the composition of the calorie-restricted diet as well as on the degree of restriction. Cancer Res. 5:616-625.
11. Stucklikova, E., Juricova-Horakova, E. M. & Deyl, Z. (1975) New aspects of the dietary effect of life prolongation in rodents: what is the role of obesity in aging. Exp. Gerontol. 10:141-144.[Medline]
12. Yu, B. P. (1996) Aging and oxidative stress: modulation by dietary restriction. Free Radic. Biol. Med. 21:651-668.[Medline]
13.
Willett, W. C., Dietz, W. H. & Colditz, G. A. (1999) Guidelines for healthy weight. N. Engl. J. Med. 341:427-434.
14.
Vigneau-Callahan, K. E., Shestopalov, A. I., Milbury, P. E., Matson, W. R. & Kristal, B. S. (2001) Characterization of Diet-Dependent Metabolic Serotypes: Analytical and Biological Variability Issues in Rats. J. Nutr. 131:924S-932S.
15. Iwasaki, K., Gleister, C. A., Masoro, E. J., McMahan, C. A. & Yu, B. P. (1988) The influence of dietary protein source on longevity and age-related disease processes of Fischer rats. J. Gerontol. 43:B5-B12.
16. Iwasaki, K., Gleister, C. A., Masoro, E. J., McMahan, C. A., Seo, E.-J. & Yu, B. P. (1988) Influence of the restriction of individual dietary components on longevity and age-related disease of Fischer rats: the fat component and the mineral component. J. Gerontol. 43:B13-B21.
17. McCay, C. M., Crowell, M. F. & Maynard, L. A. (1935) The effect of retarded growth upon the length of lifespan and upon the ultimate body size. J. Nutr. 10:63-79.
18. Kristal, B. S., Vigneau-Callahan, K. E. & Matson, W. R. (1998) Simultaneous analysis of the majority of low-molecular-weight, redox-active compounds from mitochondria. Anal. Biochem. 263:18-25.[Medline]
19. Beal, M. F., Matson, W. R., Swartz, K. J., Gamache, P. H. & Bird, E. D. (1990) Kynurenine pathway measurements in Huntington’s disease striatum; evidence for reduced formation of kynurenic acid. J. Neurochem. 55:1327-1339.[Medline]
20. Matson, W. R., Gamache, P. H., Beal, M. F. & Bird, E. D. (1987) EC array sensor concepts and data. Life Sci. 41:905-908.[Medline]
21. Matson, W. R., Bouckoms, A., Svendson, C., Beal, M. F. & Bird, E. D. (1990) Generating and controlling multiparameter databases for biochemical correlates of disorders. Basic, Clinical and Therapeutic Aspects of Alzheimer’s and Parkinson’s Diseases II:513-516 Plenum New York, NY. .
22.
LeWitt, P. A., Galloway, M. P., Matson, W. R., Milbury, P. M., McDermott, M. & Srivastava, D. K. (1992) Markers of dopamine metabolism in Parkinson’s disease: The Parkinson’s Study Group. Neurology 42:2111-2117.
23.
Ogawa, T., Matson, W. R., Beal, M. F., Myers, R. H., Bird, E. D., Milbury, P. E. & Saso, S. (1992) Kynurenine pathway abnormalities in Parkinson’s disease. Neurology 42:1702-1706.
24. Beal, M. F., Matson, W. R., Storey, E., Milbury, P. E., Ryan, E. A., Ogawa, T. & Bird, E. D. (1992) Kynurenic acid concentrations are reduced in Huntington’s disease cerebral cortex. J. Neurol. Sci 108:80-87.[Medline]
25.
Matson, W. R., Langials, P., Volicer, L., Gamache, P. H., Bird, E. D. & Mark, K. A. (1984) n-Electrode three dimensional liquid chromatography with electrochemical detection for determination of neurotransmitters. Clin. Chem. 30:1477-1488.
26. Milbury, P. E. (1997) CEAS generation of large multiparameter databases for determining categorical process involvement of biomolecules. Coulometric Array Detectors for HPLC 1997:125-141 VSP International Science Publication Zeist, The Netherlands. .
27. Acworth, I. N. & Gamache, P. H. (1996) The coulometric electrode array for use in HPLC analysis, part 1: theory. Am. Lab. 5:33-37.
28.
Shi, H., Vigneau-Callahan, K. E., Shestopalov, A. I., Milbury, P. E., Matson, W. R. & Kristal, B. S. (2002) Characterization of diet-dependent metabolic serotypes: II. Proof of principle in female and male rats. J. Nutr. 132:1031-1038.
29. National Research Council (1985) Guide to the Care and Use of Laboratory Animals. Publication no. 85–23 (rev.) 1985 National Institutes of Health Bethesda, MD. .
This article has been cited by other articles:
|
|
 |
|
  V. Shulaev Metabolomics technology and bioinformatics Brief Bioinform, June 1, 2006; 7(2): 128 - 139. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. L. Prentice, W. C. Willett, P. Greenwald, D. Alberts, L. Bernstein, N. F. Boyd, T. Byers, S. K. Clinton, G. Fraser, L. Freedman, et al. Nutrition and Physical Activity and Chronic Disease Prevention: Research Strategies and Recommendations J Natl Cancer Inst, September 1, 2004; 96(17): 1276 - 1287. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. S. Kristal and U. Paolucci Caloric Restriction in trans Sci. Aging Knowl. Environ., July 9, 2003; 2003(27): pe19 - 19. [Abstract] [Full Text]
|
|
|
|
 |
|
 Y. Noguchi, R. Sakai, and T. Kimura Metabolomics and its Potential for Assessment of Adequacy and Safety of Amino Acid Intake J. Nutr., June 1, 2003; 133(6): 2097S - 2100. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. Shi, K. E. Vigneau-Callahan, A. I. Shestopalov, P. E. Milbury, W. R. Matson, and B. S. Kristal Characterization of Diet-Dependent Metabolic Serotypes: Proof of Principle in Female and Male Rats J. Nutr., May 1, 2002; 132(5): 1031 - 1038. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
