project final report - cordis...meta-predict final report 3 4.1c - a description of the main s&t...
Post on 21-Feb-2021
2 Views
Preview:
TRANSCRIPT
Meta-predict Final Report 1
PROJECT FINAL REPORT
Grant Agreement number: 277936
Project acronym: META-PREDICT
Project title: Developing predictors of the health benefits of exercise for individuals
Funding Scheme: [HEALTH.2011.2.4.3-3 HEALTH.2011.2.4.2-2 HEALTH] Molecular and physiological effects of lifestyle factors on diabetes/obesity. Evaluation and validation studies of clinically useful biomarkers in prevention and management of cardiovascular disease Health.
Period covered: from 01/12/2011 to 01/07/2016 (incl. extension 6 months)
Name of the scientific representative of the project's co-ordinator1,
Olav Rooyackers
Title and Organisation: Professor Karolinska Institutet, Stockholm, Sweden
Tel: +46 8 58586182
Fax: +46 8 7795424
E-mail: olav.rooyackers@ki.se
Project website address: www.metapredict.eu
1 Usually the contact person of the coordinator as specified in Art. 8.1. of the Grant Agreement.
Meta-predict Final Report 2
4.1 Final publishable summary report
THIS FILE CONTAINS THE FIGURES AND TABLES ONLY. These belong to the text submitted online 4.1a - Executive Summary 4.1b - A summary description of project context and objectives
Figure 1. A high-level overview of the META-PREDICT. We have independent exercise life-style modification programs (A-D) with common clinical end-points that feed into a 'predictor' pipeline where we generate diagnostic screens from RNA, mDNA and metabolites. This will lead to personalised diagnostics, greater understanding of the underlying biology, and biomarkers for validating a novel drug screening in vivo model. Studies A and part of D have existing RNA gene-chip profiles; Study C has miRNA and mRNA deposited in the public domain (GEO). Study B will be the largest study of time-efficient cycle-sprint training in at-risk (metabolic disease) subjects studied using HTA 2.0 RNA, miRNA and mDNA chips.
Meta-predict Final Report 3
4.1c - A description of the main S&T results/foreground
Figure 2. Metabolomics analyses in a smaller cohort of Twins with significant difference in level of physical activity (left) which was confirmed in larger cohorts of matched individuals with high and low levels of physical activity. From original publication (16).
Table 1. Interval-training induced concentration differences between HRT and LRT female rats. The
concentrations for the metabolites are given as mean±SD in µM.TRP:tryptophane.
Meta-predict Final Report 4
Metabolite(µM) FemaleHRT FemaleLRT p-value FDR
PRE POST PRE POST
Lauroylcarnitine 0.04±0.02 0.02±0.01 0.03±0.02 0.02±0.01 3.97E-09 1.71E-07
Stearoylcarnitine 0.11±0.10 0.05±0.04 0.07±0.04 0.04±0.02 5.30E-08 1.14E-06
Octanoylcarnitine 0.02±0.01 0.02±0.01 0.02±0.01 0.03±0.02 1.23E-07 1.77E-06
Myristoylcarnitine 0.12±0.09 0.04±0.03 0.10±0.07 0.04±0.02 7.69E-07 8.26E-06
Hexanoylcarnitine 0.05±0.02 0.05±0.02 0.05±0.03 0.09±0.06 1.27E-05 0.00010945
TRP 107.4±9.7 98.8±9.5 102.5±15.1 98.4±13.1 0.0016521 0.01184
Meta-predict Final Report 5
Figure 3: Human Studies - Presentation of the linear relationship between changes in plasma insulin and plasma C-peptide concentrations following 6 weeks of ‘5-by-1’ HIT protocol. Oral glucose tolerance test (OGTT). Value is based on Pearson correlation coefficients (coefficient of determination).
Table 2. Subject characteristics for the HIT study
Figure 4: Comparison of the efficacy of 6 weeks of the ‘5-by-1’ HIT high intensity cycle based exercise training protocol (black circles) with the less effective ‘7-by-1’ HIT (black diamonds) protocol in obese or over-weight adults. The values are presented as mean ± 95% CI for the main training responses. AUC: area under the curve during oral glucose tolerance testing (180 min); SBP: Supine systolic blood pressure; DBP: Supine diastolic blood pressure; MAP: Supine mean arterial pressure; RHR: resting heart rate.
Non-exercise
(n=13) ‘7-by-1’ HIT
(n=40) ‘5-by-1’ HIT
(n=137)
Gender (men/women) 4 / 9 20 / 20 63 / 74
Age (y) 25 (20-51) 38 (20-53) 38 (18-51)
Height (m) 1.66 (1.52-1.81) 1.71 (1.53-1.94) 1.72 (1.51-2.01)
BMI (kg·m-2
) 32.7 (27.3-41.4) 29.8 (27.0-45.5) 31.2 (26.5-48.1)
Fasting glucose (mmol·L-1
) 4.6 (4.0-5.2) 4.5 (3.8-5.4) 4.6 (3.3-5.6)
2hr OGTT glucose (mmol·L-1
) 6.3 (5.2-9.4) 6.5 (4.2-10.5) 6.8 (3.7-10.1)
IPAQ Score 248
(160-1188)
382
(73-594)
330
(0-597)
Meta-predict Final Report 6
Figure 5: Heat-map representation of the individual responses to exercise training for 5 variables following ‘5-by-1’ HIT (V̇O2max, MAP, Oral glucose tolerance test (OGTT) Insulin AUC180 and HOMA-IR). It can be observed that each subject demonstrates a differential response to the 5 clinical parameters. Some subjects improve all five, most improve 2 or 3 while some do not improve any of these health biomarkers.
Meta-predict Final Report 7
Figure 6. META-PREDICT HIT study design. Six weeks of supervised HIT was preceded by two study visits (baseline measurements, one week apart) and followed by another two study visits, post-training (~3 days after final session) and post-training monitoring (3 weeks after the final session). After an overnight fast, a DXA scan was followed by two fasted blood samples and a muscle biopsy, a standard 75-g OGTT, a standardised meal and a V̇O2max-test. The DXA scan and muscle biopsies were not performed during Study visit 2.
Meta-predict Final Report 8
Figure 7. Differences in waist gain (cm, mean and 95%CI) during follow-up (decreased activity: changed from upper third to a lower one; increased activity: changed from a lower third to upper one). (A) Sex-specific differences among individuals taking into account clustered observation of twin pairs. (B) Pairwise difference among leisure-time physical activity discordant same-sex twin pairs. (C) Pairwise differences among leisure-time physical activity discordant monozygotic twin pairs (from (28)).
Meta-predict Final Report 9
Figure 8. Content of selected signaling proteins [mean (SD)] in skeletal muscle (m.gastrocnemius) of high and low responder rats following acute high intensity exercise. * = statistically significant difference between the rat lines.
Figure 9. The impact WP5 D5.3 experiment on Rat miRNA PCA (Affymetrix)
Meta-predict Final Report 10
Figure 10. The impact WP5 D5.3 experiment on Rat miRNA PCA (Exiqon)
Figure 11 - Affymetrix miRNA array 2.0 PCA plot – Heritage Samples
Meta-predict Final Report 11
A
B
C
Meta-predict Final Report 12
Figure 12. NUSE plots for muscle samples from A) the TWINN study, B) STRIDDE III, C) LVL Maastricht study and D) HIT study.
Training set A (select 150 genes using KNN and High vs Low
muscle profiles
Apply 150 genes using gene-ranking method & regression to
2 muscle data sets B &C
Apply 150 genes selected and validated using muscle data to
Cohort D - blood
R2=0.806p-value=1.01E-06
R2=0.483p-value=1.11E-08
Baseline Blood Profilein subjects that did AT.
R2=0.461 pvalue=2.65E-05
Figure 13. Using new (WP8) DUKE blood chip data (n=124 baseline from 289) we took the muscle based aerobic predictor signature genes and attempt to reproduce this classifier from the blood RNA profile.
D
Meta-predict Final Report 13
Figure 14. Demonstrating the large peak of background probe signal seen in the ‘raw’ HTA 2.0 chip data in MP muscle samples.
Figure 15. Demonstrating the impact of re-alignment of the original HTA 2.0 probes to the latest genome sequence - ~2.4Million probes are removed.
Meta-predict Final Report 14
Figure 16. Demonstrating the impact of processing the filtered HTA 2.0 probes to reflect the probes expressed in a tissue specific manner - ~2.3Million probes are removed.
Meta-predict Final Report 15
Figure 17. A muscle specific HTA 2.0 ‘probe’ map that is then assembled into probe-sets to detect individual exons. Only exons (probesets) with n=3 probes or more, aligned to 1 place in genome and containing probes above background signal are included.
Figure 18. An example of a proto-type blood RNA versus HOMA linear model (Using STRRIDE II blood data). This is a simple Knn + Ranking model was generated prior to custom CDF methodology and will be further developed to replicate across cohorts using the new linear modeling code and CDFS.
Spearman rank correlation
r=0.6951; p<2.2e-16
Meta-predict Final Report 16
Figure 19. XRGenomics computational strategy for linear code
Figure 20. RNA sequencing not able to produce a strongly linear signal for lower expressed human tissue genes (especially muscle and blood).
Features Expression Matrix
TRAINcor Features Expression Matrix
TRAINcorphenotype data TRAINsel
phenotype data
TRAINsel Features Expression Matrix
OPTION 1:Log transform
clinical variable
OPTION 2:Log transform
expression matrices
OPTION 3:Scale and center expression data
Leave one sample out
OUTPUT: Annotated list of all the
features in the study
OUTPUT: Annotated list of selected features
Data processing for Linear classifier
Linear Classifier
Calculate correlation coefficients between clinical variable and
expression values for each exon
Adjustment for covariates(chosen number of covariates
can be one or multi)
OPTION 4 Flexibility to subset the features
used for TRAINsel
Selection 1: After every loop, one more feature is added to calculate sample
score. Effect of addition of each feature is saved.
OPTION 6select the maximal number of
features you want to test
OPTION 5Order TRAINsel
features
OPTION 9 , 10, 11Subsetting features for
final selection
Test the model by running cor.test between clinical
variable and sample score for selected features
OPTION 7Select how sample
scores are calculated
Selection 2: From selected features, selection process
is repeated with final model feature selection
OPTION 8Set threshold for slope
of the models
Adjustment for covariates(chosen number of covariates
can be one or multi)
CCvalue,p-value,adjustedpvalue,slopeofbestfitline
Pearson orSpearman
FigureS8:RelationbetweensamplessizeandgenesexpressedatalowvalueasdetectedusingRNA-sequencing(RNA-seq)technology.Therewasavery
limitedlinearrelationshipbetweensequencingdepthoflowexpressedgenesinbloodandmuscletissueswhiletherewasamorerobustlinearrelationship
inadiposetissue.ThismeansthatcompletecoverageofthetranscriptomeinmuscleandbloodwouldbeverychallengingandthusRNA-seqwouldnot
necessarilybepreferentialtechnologyfordetectingalternativeexonusageevents.
2.0e+07 4.0e+07 6.0e+07 8.0e+07 1.0e+08 1.2e+08
05
10
15
20
Corelation between median number of counts in 141 lowest expr genes with increasing number of total counts in Mele three tissues
Total counts
Me
dia
n r
aw
cou
nts
fo
r g
en
es
Blood
Muscle
Adipose
R^2 =0.79
R^2 =0.56
R^2 =0.38
Total&counts&
Median&raw
&counts&for&ge
ne&
20000000& 40000000& 60000000& 80000000& 120000000&100000000&
0&
5&
10&
15&
20&
Meta-predict Final Report 17
Figure 21. Full-Transcript, 3’UTR-transcript and 5’UTR-transcript regulation in response to 6 weeks of 5by1 HIT (WP3) in the skeletal muscle of subjects that demonstrate a gain in aerobic capacity (FDR 10%, FC>1.2)
Meta-predict Final Report 18
Figure 22. Global correlation between lncRNA and CIS protein coding genes in human skeletal muscle (and HIT exercise regulated genes in red).
Figure 23. Up-stream regulators associated with the protein-coding genes that are co-regulated with the lncRNA exercise responsive genes in subjects that demonstrate a robust improvement in aerobic capacity
Meta-predict Final Report 19
Figure 24. Probe-sets for TCF7L2 in muscle and adipose (WP5).
Meta-predict Final Report 20
Figure 25. iGEMS analysis pipeline
No#
*#Use#Exon+#level#CDF#
SD#filter#to#carry#forward#only#the#expressed#genes#
Raw#Gene#Chip#Data
*#Use#Gene+level#CDF#
Gene+level#analysis*
Significant#GENE#level#
MUF#score#
e.g.##FDR<1%#
Exon+level#analysis*
Locate'candidate'EXONS'using'splicing'
Index'(SI)'
Specific#AEU#per#gene#
e.g.#top##10%#SI##scores#
Step 1 Step 2 Step 3
optional
e.g.##SD#>#5#in#at#least#two#exons#of#a#candidate#gene#
Select&top&EXON&SI&scores&from&Significant&
GENE&list&
29step*&nega; ve&selec; on:&false&
posi; ve?
Meta-predict Final Report 21
Figure 26. Plot of DECR1 in people with high versus low insulin sensitivity. The blue box indicates alternative splicing for several Exons.
Figure 27. Application of iGEMS to muscle tissue related to high responders in VO2max
X
X
X
X
X
X
X
X
X
X
X
X
X
X– higherintensityaftertrainingX– higherintensitybeforetraining
X
T11,exon3
X
T2,exon11
T3,exon10
X
X
X
X
X
X
X
X
X
X
X
Exonfrom2,3,6,8,13,14,or15
X
T15,exon9
X
Meta-predict Final Report 22
Figure 28. The wCCS for up and down regulated miRNAs WP5 D5.3 experiment.
Meta-predict Final Report 23
top related