MAPPING THE CONSEQUENSES OF PHYSICAL EXERCISE AND NUTRITION ON HUMAN HEALTH
A PREDICTIVE METABOLOMICS APPROACH
Elin Chorell
Abstract
Contents
List of papers 7
Other papers by the author not appended to the thesis 8
Abbreviations 9
Notation 10
AIM 11
BACKGROUND 13
Exercise, nutrition and health 13
Metabolomics 16
Multivariate data processing and analysis 18
Predictive metabolomics 23
RESULTS 25
Papers I and II 25
Paper III 29
Paper IV 33
Paper V 40
CONCLUSION 45
ACKNOWLEDGEMENTS 48
POPULÄRVETENSKAPLIG SAMMANFATTNING 50
REFERENCES 52
List of papers
Other papers by the author not appended to the thesis
Abbreviations
Notation
Aim
Background
Exercise, nutrition and health
Figure 1
Figure 1. Left Swedish national health recommendations on daily dietary intake displayed as a food circle
†. Right Guidance from the U.S. department of Health and Human Services (HHS) on how to track calories, nutrients and ingredients.
Reprinted with kind permission from the Swedish National Food Administration (Livsmedelsverket).
Figure 2
Figure 2. A schematic picture of the building blocks in systems biology and their inherent hierarchy:
Genomics, mapping the entire genome (DNA), Transcriptomics, measurements of overall transcript expression (RNA), Proteomics, identifying, sequencing and characterizing the functional protein network (proteins), Metabolomics, the comprehensive analysis of the metabolome under specified conditions (metabolites).
Metabolomics
Figure 3 Genome
DNA
Transcriptome
RNA
Proteome
Protein
Metabolome
Biochemicals Metabolites
Building blocks of systems biology
Phenotype
Function
Genomics Transcriptomics Proteomics Metabolomics
Figure 3. A few examples of the physio-chemically diverse set of metabolites targeted by metabolomics
analysis.
Metabolomics in human nutrition and physical exercise
Multivariate data processing and analysis
Hierarchical multivariate curve resolution
Figure 4
Figure 4. Schematic picture of the HMCR method including the predictive feature. A) A representative training set (or all samples) are subjected to HMCR and simultaneously aligned. B) The aligned sample chromatograms are divided into time windows with borders set at local minima. C) Each time window is deconvoluted separately resulting in deconvolution of quantitative profiles (i.e. metabolite concentrations) and mass spectral profiles (i.e. metabolite identities), collected in two separate data tables. D) The obtained HMCR parameters can be used to deconvolute new samples, resulting in resolving the same compounds (if possible) as in the original data.
Chemometrics
Retention time
Intensity
A) Alignment of samples
Retention time
Intensity
D) Prediction of new samples - using obtained HMCR settings
THE HMCR PROCEDURE
B) Division of data into time windows
Retention time
Intensity
Resolved compounds (Quantity)
Mass spectra (Identity)
Retention time
Intensity
1 4
2 5
3 6
m/z
m/z m/z
m/z m/z
m/z 1
4 2
5 3
+
6C) Resolving each time window separately
PCA
PLS and OPLS
Predictive metabolomics
Figure 5
Figure 5. An overview of the predictive metabolomics strategy. SED is used iteratively throughout the whole process from creating the original study design to the selection of sample batches for sample preparation and analysis as well as for selection of training and test sets. All sample handling protocols are standardized and designed to maximize a diverse metabolite output. The HMCR method is used to resolve overlapping compounds that were not separated by the chromatography and also allows for independent sample prediction by resolving the same metabolites as obtained in the original HMCR processing in independent sample sets. Finally, multivariate methods are used for sample comparison modeling, biomarker extraction, interpretation and validation.
STATISTICAL EXPERIMENTAL DESIGN
Maximize information output
STANDARDIZED PROTOCOLS Sample collection, preparation
and chemical characterization
HIERARCHICAL MULTIVARIATE CURVE RESOLUTION Resolving overlapping compound
MULTIVARIATE ANALYSIS Sample comparison analysis
PREDICTIVE METABOLOMICS
BIOLOGICAL INTERPRETATION
Results
Papers I and II
Metabolomics analysis of physical exercise induced stress responses
and macronutrient modulation
figure 6.
Figure 6. A schematic overview of the study design used in papers I and II. All subjects performed four matched exercise sessions on individualized workloads and were either given water, low-carb+protein, low-carb or high-carb beverage immediately after completed exercise. Blood samples were collected pre and post exercise (0min) as well as in the early recovery phase (15-90min).
Results
1)
WaterPre +15
min+30
min+60
min+90
minBeverage intake 0
minFO U R D I FFE R E NT SE SSI O NS*
2)
Low-carb + protein3)
Low-carb4)
High-carb90min of strenous exercise on individualized workloads Plasma sample collection
All subjects performed all four sessions (1-4) with one week apart according to a randomized scheme
*
Figure 7A
Figure 7. A) Cross validated OPLS-DA scores (tcv[1-3]P) revealing clustering of subjects in relation to which macronutrient composition that were ingested in the early recovery period following physical exercise. B) OPLS-DA covariance loadings plot (w[1-2]) of the resulting metabolic pattern explaining the clustering in the two first components of the model scores.
Figure 7B
3-methylhistidine Pseudouridine
w[1]
w[2]
Fatty acids
Amino acids Carbohydrates
Low Carb+Protein
Water
High Carb Low Carb tcv[3]P
tcv[2]P tcv[1]P
4
0
10
5 10
5
A B
Figure 8
Figure 8. Left: Cross validated OPLS-DA scores describing the separation of subjects with high and low fitness status (i.e. VO
2max) in samples taken in the early recovery period after exercise. Subjects with top five VO
2max values are plotted as blue dots while the five subjects with bottom VO
2max values are plotted as yellow dots. All model samples were collected after ingestion of water only. Corresponding low-fit samples, collected after ingesting low-carb+protein beverage, were predicted into the existing model and are displayed as yellow circles in the plot. The same subject/time point is connected with a dashed line. Right: Insulin concentration of the deviating subject from the OPLS-DA predictions when ingesting the low-carb+protein beverage (red line) compared to the average (grey line).
Summary and conclusion
HighFit
Top 5 individuals
LowFit
Bottom 5 individuals
tPScv[1]
Num (Subject)
Model samples (subject ingesting water)
Predicted samples (subjects ingesting low carb+protein)
Average*
95% confidence interval Deviating subject Insulin concentration when ingesting low carb+protein
Pre 0 15 30 60 90
120
80
40 0
µU/mL
Time point (min after ingestion/exercise)
Paper III
Investigation of metabolite profiles associated with cardiorespiratory
fitness in healthy subjects
Results
Figure 9
Figure 9. A schematic overview of the collected blood samples included in paper III. 1) All subjects were given a nutritional load during resting. Blood samples were collected pre and every following 15min for a total of 90min. 2) An exercise session with individualized workloads was performed where subjects were given either water or a nutritional load during the recovery period. Blood samples were collected pre and immediately after completed exercise as well as every 15min for a total of 60min. 3) All subjects performed a matched exercise session for validation purposes. Blood samples were collected pre and immediately after completed exercise as well as every 15min for the following 45min.
Figure 10
1 )
Resting + nutritional load All subjectsT H R E E D I FFE R E NT SE SSI O NS *
2)
Exercise + nutritional load/water Water: N=14 (7 high fit + 7 low-norm) Nutritional load: N=13 (6 high fit + 7 low-norm)3)
Exercise + nutritional load All subjects, validation setPre Nutritional +15min+30min+45min
load
Pre Nutritional +15min+30min+45min+60min
load/water
Pre +30min+45min +60min+75min
Nutritional load
+15min +90min
65min of ergometer cycling on individualized workloads
Plasma sample collection All subjects performed session 1 in randomized order.
Session 2 and 3 were performed on the same day with 6h of rest and a standardized meal in-between.
*
High fit: VO2max > 60 ml VO2/min/kg Low-norm: VO2max 38-48 ml VO2/min/kg
Figure 10. A PCA model based on anthropometric measurements from which the sample preparation and GC-TOF/MS batches were designed. Left: Cross validated PCA scores (tcv[1-2]) describing the inter subject variation where high fit subject are displayed as dots and low-norm as circles. Batches are in different colors and matched subjects are connected by a line. Right: PCA loadings (p[1-2]) describing the variation among the anthropometric variables from which the PCs were calculated.
Figure 11A B
Fat in tissue (%) BMI
Body weight (kg)
Hemoglobin (g/L)
Age
Length (cm)
Effect on VO2max
VO2max
VO2max ml/kg/min Effect on VO2max/kg Effect/kg VO2max
Batch 1 Batch 2 Batch 3 Batch 4 Batch 5 Batch 6 Batch 7
PCA scores PCA loadings
High fit Low-norm
Figure 11. OPLS-DA model comparing plasma metabolite profiles from high fit to low-norm subjects following exercise and nutritional load. A) Cross-validated OPLS-DA scores, where high fit subjects are displayed as dark green dots and low-norm as light green dots. The independent sample predictions (from a replicate session) are displayed as circles. B) OPLS-DA model loadings describing the metabolite pattern related to fitness level. Metabolites in the upper part of the plot are positively correlated, i.e.
increased, in high fit subjects while metabolites in the bottom part are negatively correlated, i.e.
decreased in high fit subjects. Identified metabolites are displayed in black, where squares denotes metabolites classified as amino acids, crosses denotes metabolites classified as fatty acids and dots denotes non-classified metabolites.
Summary and conclusion
2.5
-2.5 0
15 30 45 60
Time after finished exercise + nutritional load (Min)
tPScv[1]
Num (resolved compound)
0 50 100 150 200
DHA (22:6w3)*
RIBITOL ASPARTIC ACID PH. ACID
ILE
18:1
G-TOCOPHEROL*
A-TOCOPHEROL
18:2w6 16:1
16:0 METHYL-CYS
b-ALA 18:0
INCREASE IN HIGH FIT
DECREASE IN HIGH FIT
Fatty acid (Identified)
Amino acid (Identified) Non identified compound Identified compound 0.2
-0.2 0 0.1
-0.1
p[1]
High fit subject (exercise session 2) Predicted high fit subject (exercise session 3) Low-norm subject (exercise session 2) Predicted low-normsubject (exercise session 3)
A. OPLS-DA SCORES B. OPLS-DA LOADINGS
Paper IV
The impact of long and short term diet interventions on the
metabolome in overweight and obese postmenopausal women
Results
Figure 12
Figure 12. A schematic overview of the collected blood samples in paper IV. 1) A short term dietary intervention study including ten overweight and obese postmenopausal women following a high protein and MUFA diet. Blood samples were collected pre intervention, after five weeks on the restricted diet as well as after an additional three months on a non restricted diet. 2) A long term dietary intervention study comparing two different diets, low versus a high protein and MUFA diet, in 66 overweight and obese postmenopausal women. Blood samples were collected pre intervention as well as after six and 24 months on the restricted diets.
Short term dietary restriction (study1)
Figure 13A
1) Short term effect of a high protein and MUFA diet Five weeks, 10 subjects
T W O D I FFE R E NT I NT E R V E NT I O NS *
2)Long term effect of two different diets Two years 34 subjects on a high protein and MUFA diet 32 subjects on a low protein and MUFA diet
Pre +3 months
Restricted diet +5 weeks
Blood sample collection The dietary composition in the short term study
were comparable to the high protein and MUFA diet used in the two year intervention study
*
Non restricted diet
Pre +24 months
Restricted diet +6months
Restricted diet
High protein and MUFA diet: 30E% carbohydrates, 30E% proteins and 40E% fats (mainly MUFA and PUFA)
Low protein and MUFA diet (i.e. Nordic nutrition recomendation from 2004): 55E% carbohydrates, 15E% proteins and 30E% fats
Figure 13B
Figure 13. An OPLS-DA model comparing plasma samples from ten overweight and obese postmenopausal women collected pre and post five weeks on a high protein and MUFA diet. A) Cross validated model scores (tPScv[1]) describing a clear separation between samples collected pre (white dots) to those after five weeks on the restricted diet (blue dots). Analytical replicates (circles) were predicted into the existing OPLS-DA model. B) Additional samples collected three months on a non- restricted diet (red dots) and their analytical replicates (red circles) were predicted into the existing OPLS-DA model, as described in A.
Figure 14
Figure 14
3 month follow-up (model sample)
3 month follow-up (predicted analytical replicate) Post 5 weeks of diet (model sample)
Post 5 weeks of diet (predicted analytical replicate) Pre diet (model sample)
Pre diet (predicted analytical replicate)
A B
tPScv[1]
Subject (Num) Subject (Num)
Figure 14. OPLS-DA loading (p[1]) describing the metabolic pattern responsible for the observed separation between plasma samples collected pre and post five weeks on a high protein and MUFA diet. Entities displayed in the upper part of the plot increased in concentration after the five week intervention while entities in the lower part decreased.
Long term dietary restriction (study 2)
Figures 15A B
170. AMP 85. QA
118. Trp 97. Tyr
179. Mal 173.ɑT
174. Chol 64. LauA
67. Asp 63. Phe
12. Pro 10. PhA
31. Thr 35. MeCys
81. CA105. myo-Inositol
121. OA (C18:1) 149. DHA (C22:6ω3)
17. PA 103. PalA (C16:1)
104. PA (C16:0)
86 107 122 134
4 8. EA 25
51
48 120
148 145
167
95 119
Increases after 5w of high protein and MUFA diet
Decreases after 5w of high protein and MUFA diet
Identified metabolite Amino acid Amine Alcohol/polyol Carbohydrate Fatty acid Carboxylic acid Hydroxy acid Unidentified metabolite
p[1]
Compound (Num)
Figure 15. OPLS-DA models comparing plasma metabolite profiles in overweight and obese postmenopausal women in plasma samples collected pre to samples collected after 24 months on a restricted diet A) High protein and MUFA diet. Cross-validated OPLS-DA scores (tcv[1]) revealing a clear separation between plasma samples collected pre (white) to those after 24 months (red) on the high protein and MUFA diet. B) Cross-validated OPLS-DA scores (tcv[1]) revealing a clear separation of plasma sample collected pre (white) to those after 24 months on low protein and MUFA diet (grey). A and B) The highest responder within each diet is denoted by an arrow, low responders by a pound sign and reverse responder by a dashed line.
Figure 16
Figure 16
Figure 16
#
#
A B
tcv[1]
Subject (Num) Subject (Num)
G Pre diet
24 months on a high protein and MUFA diet 24 months on a low protein and MUFA diet
High responder Low responder Reverse respoder
#
High protein and MUFA Low protein and MUFA
Figure 16. OPLS-DA correlation loadings (p[1]) from two different OPLS-DA models plotted against each other in a shared and unique structure (SUS)-plot. Y-axis: Low protein and MUFA diet; p[1] for the OPLS-DA model discriminating between samples collected pre to those after 24 months of restricted diet. X-axis: High protein and MUFA diet; p[1] for the OPLS-DA model discriminating between samples collected pre to those after 24 months of restricted diet. The joint and opposite metabolite changes for both diets are displayed along the diagonals (arrows), while diet specific changes are seen along the x- (high protein and MUFA) and y-axis (low protein and MUFA).
Figure 17
102
100. AA (C20:4ω6) 56. LauA
(C12:0)
101. DGLA (C20:3ω6) 71 84. PA (C16:0)
96. SA (C18:0) 21. EA
22
30. Ser
86. myo-Inositol 116. Chol
34. Thr
16. GHB 87. NSAID 107. Inosine
98. Cys
93. LA (C18:2ω6) 8 106. DHA(C22:3ω3) 94
64. Orn 58 66 72
54 12. Ala 74. Tyr
69. CA 102 85
5
p[1] High protein and MUFA diet
Identified metabolite Amino acid Amine Alcohol/polyol Carbohydrate Fatty acid Carboxylic acid Hydroxy acid Unidentified metabolite
p[1] Low protein and MUFA diet
Figure 17. Cross-validated OPLS-DA scores (tcv[1]) displaying a clear separation between samples collected pre (white) to sample collected after six months (grey) on a high protein and MUFA diet. The individual metabolic response of each subject is highlighted by a dashed line. A solid line divides the plot into segments corresponding to group assignments. Diet group one started their diet in October 2007, diet group two in November 2007 and diet group three in March 2008.
Short and long term response to a high protein and MUFA diet
Summary and conclusion
Figure 17
tcv[1]
Subject (Num)
Diet group one Diet group two Diet group three
Six months on a high protein and MUFA diet Pre diet
G
Paper V
The impact of feeding Lactobacillus F19 during weaning:
A study of the plasma metabolome
figure 18
Figure 18. A schematic overview of the study design and blood sample collection in paper V. A double- blind study of Lactobacillus F19 (LF19) treated weaning infants compared to placebos. Blood samples were collected at 5½ months of age and at 13 months of age after a daily intake of LF19 treated cereals or placebo.
Results
Figure 19
LF19 19 infants
P R O B I O T I C S I N I NFANT S
5½ months of age 13 months of age
Blood sample collection
LF19: At least one daily serving of cereal supplemented with probiotic Lactobacillus paracasei ssp. paracasei strain F19 Placebo: At least one daily serving of non-supplemented cereals
Placebos 18 infants
LF19 treatment
5½ months of age 13 months of age Placebo treatment
Figure 19. Separate OPLS-DA models for the LF19 and placebo groups, comparing samples at 5½ and 13 months of age. A) Cross-validated scores (tcv[1]) describing a clear response to age for all samples in the LF19 treated infants. Samples collected at 5½ months of age are displayed as white dots and samples collected at 13 months as orange dots; each subject’s corresponding samples are connected with a line. B) Cross-validated scores (tcv[1]) describing a clear separation of samples related to age in the placebo group. Samples collected at 5½ months of age are displayed as white dots and samples collected at 13 months as green dots; each subject’s corresponding samples are connected with a line.
Figure 20
Figure 20
Figure 20
tcv[1]
Subject (Num) Subject (Num)
13 months of age (LF19) 13 months of age (Placebo) 5½ month of age (pre LF19)
A. LF19 B. Placebo
Figure 20. Loadings (p[1]) from two different OPLS-DA models discriminating between samples collected at 5½ months and 13 months of age plotted against each other in a shared and unique structure (SUS)-plot. Y-axis: LF19 (probiotics) group. X-axis: placebo group. In the plot the general (joint) and opposite metabolite changes for both groups are displayed along the diagonals, while group specific (unique) changes are seen solely along the x- (placebo) and y-axis (LF19).
Summary and conclusion
Identified metabolite Sterol
Amino acid Amine Alcohol/polyol Carbohydrate Fatty acid Carboxylic acid Hydroxy acid Unidentified metabolite 59. PA
112.Phe 135. AzA
58. Glycerate 204
171. LA
148. DGL 25. BuA
172 143 173
159
92. Erythrose 149.
116
153. myo-inositol 126
97. OH-Pro
74. Glutarate
54. Gly 95. AspA 39. Val 44
52 48 50 9
157
117
137. CA 81. MeCys 127 83
