Metabolomics-oriented Bioinformatics at the MPI for Molecular Plant Physiology.

Transcription

1 Metabolomics-oriented Bioinformatics at the MPI for Molecular Plant Physiology. Dirk Walther Head Bioinforamtics group Max Planck Institute for Molecular Plant Physiology Potsdam-Golm

2 Overview GC/MS Data Metabolomics databases: The Golm Metabolome Database (GMD) Automated GC/MS-spectra interpretation Metabolites Metabolite-transcript causal relationships from time course data Pathways Metabolomics-assisted genome annotation in Chlamydomonas reinhardtii Genomes

3 From GC/MS Data to Metabolites GC/MS Data Metabolomics databases: The Golm Metabolome Database (GMD) Automated GC/MS-spectra interpretation Metabolites Metabolite-transcript causal relationships from time course data Pathways Metabolomics-assisted genome annotation in Chlamydomonas reinhardtii Genomes

4 GMD Golm Metabolome Database Spectra NIST Wiley GMD Chemicals Beilstein (8.Mio.) Profiles PubChem (3.Mio.) CAS (25 Mio.) KEGG (0.02 Mio.) KNApSAcK Publications BRENDA Pathways EcoCyc METLIN AraCyc Kopka et al., (2005) The Golm metabolome database. Bioinformatics 21: Hummel et al. (2007) The Golm Metabolome Database: a Database for GC-MS based Metabolite Profiling. In: Hohmann, S. (ed) Topics in Current Genetics: Nielsen, J., Jewett, M. (eds) Metabolomics. Springer-Verlag, Berlin Heidelberg New York,

5 GMD Golm Metabolome Database Spectra GMD in numbers Chemicals NIST Wiley Beilstein (8.Mio.) GMD Profiles Reference substances: 3,056 Metabolites: 830 Analytes (derivatized compound): 3,175 Number of spectra: 24,383 PubChem (3.Mio.) CAS (25 Mio.) KEGG (0.02 Mio.) KNApSAcK Profiles (samples): 668 Publications BRENDA Pathways EcoCyc METLIN AraCyc Kopka et al., (2005) The Golm metabolome database. Bioinformatics 21: Hummel et al. (2007) The Golm Metabolome Database: a Database for GC-MS based Metabolite Profiling. In: Hohmann, S. (ed) Topics in Current Genetics: Nielsen, J., Jewett, M. (eds) Metabolomics. Springer-Verlag, Berlin Heidelberg New York,

6 GMD Inventory, Reference substances acid 52% sugar 21% polyol 4% terpenoid 4% alcohol 3% amine 3% alkane 2% conjugate 2% indole 1% others less than 1%: aldehyde, alkaloid, amide, calystegine, chalcone, flavonoid, imide, lactam, nucleotide, purine, stilbene nucleoside 1% pyrimidine 1%

7 Annotation of GC/MS-spectra Recorded GC/MS spectrum Library of GC/MS spectra for reference substances Identified Compound

8 relative frequency Annotation via GC/MS-spectra comparisons 0,025 0,02 0,015 sugars amino acids vs sugars 0,01 0, ,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 Similar spectra Similarity score Dissimilar spectra Without RI information

9 relative frequency Annotation via GC/MS-spectra comparisons 0,025 0,02 0,015 sugars amino acids amino acids vs sugars 0,01 0, ,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 Similar spectra Similarity score Dissimilar spectra Without RI information

10 Knowledge-based annotation of GC/MS-spectra Same metabolite, but different derivatives 3TMS-GABA 2TMS-GABA Identification via characteristic mass peaks Spectra interpretation Scsibrany & Varmuza (1992) J Anal Chem Varmuza & Werther (1996). Chem. Inf. Comput. Sci.

11 Annotation of GC/MS-spectra using decision trees O Si m/z = 103 Si O O Si m/z = 217 Fraction Amino Acids Fraction Sugars Predicted\True amino acid sugar amino acid sugar % correct predictions! O O TMS m/z = 117

12 recall Compound class predictions based on fragments s p P m / A l e intensity lg 1 P H y P l r A C m C a a l P H h e A l r C o P h A c 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Classification tree models for 20 functional groups 1 2 Diol Acetal Alcohol Aldehyde Alkene Alpha Aminoacid Amine Aromatic Carbonyl Carboxylic Acid Carboxylic Acid Deriv Carboxylic Acid Ester Hemiacetal Heterocycle Hydroxy Phenol Phosphoric Acid Deriv Prim Alcohol Prim Aliph Amine Prim Amine Sec Alcohol F-measure = 0.65 Hemiacetal Heterocycle Carboxylic Acid Ester 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% precision r S alcohol like amine like carbonyl like diverse r A ac C aa x x x x x x x x x x x x x x x - x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x - x - x x x x x x x x x x x x x x x x x x x x x x - x x - - x x - x - x x x x x x x - x x x - - x x x x x x x x x *1) *2) RI VAR m.-diff ratio lg Hummel et al., Submitted

13 Annotation of user-submitted GC/MS spectra at gmd.mpimp-golm.mpg.de

14 Annotation of user-submitted GC/MS spectra at gmd.mpimp-golm.mpg.de

15 From Metabolites to Pathways GC/MS Data Metabolomics databases: The Golm Metabolome Database (GMD) Automated GC/MS-spectra interpretation Metabolites Metabolite-transcript causal relationships from time course data Pathways Metabolomics-assisted genome annotation in Chlamydomonas reinhardtii Genomes

16 Can we infer causal and, thus, pathway relationships from metabolomics data? Which molecules function as cause, which as effect?

17 From Correlation Networks to Pathways X,Y. gene expression level, metabolite level etc.? X,Y are connected (edge) if r>r c Is there a straightforward connection between the underlying system and the observed correlations? Can we deduce novel pathways based on the observed correlation matrix? Steuer et al. (2003) Bioinformatics 19: 1019

18 Yeast Transcripts & Metabolite Profiles in Response to Temperature Stress: Time series data Temperature shifts: 28 C 37 C heat 28 C 10 C cold 28 C 28 C control Yeast Time series: 8 time points 0, 0.25, 0.5, 1, 2, 4, 8 and 24 h Transciptome Analysis Metabolome Analysis Affymetrix Gene-chips GC-MS Katrin Straßburg, Joachim Kopka

19 Metabolite Profiles GC/MS with in vivo stable isotope labeling for accurated quantification Heat Stress 42 metabolites 11 unknowns Cold Stress 44 metabolites 13 unknowns Katrin Straßburg, Joachim Kopka

20 Expression Ratio Expression Ratio Transcript Profiles Affymetrix whole yeast genome microarray 2.0 Heat Stress 3.0 Cold Stress (37 C 28 C) (10 C 28 C) Cluster genes [7%] Cluster genes [26%] Cluster genes [24%] Cluster genes [33%] Cluster genes [9%] Time (min) Time (min) Cluster genes [10%] Cluster genes [19%] Cluster genes [44%] Cluster genes [16%] Cluster genes [11%] Katrin Straßburg, Joachim Kopka

21 Deducing pathways from time-course data? Omics-viewer from YeastCyc Patrick May

22 Metabolite correlations are more informative than transcript correlations YeastCyc & KEGG yeast pathway maps Katrin Straßburg, Joachim Kopka

23 Causal relationships via time-delayed correlations Time metabolites leading concurrent metabolites trailing Metabolites Transcripts metabolites leading metabolites trailing Cognate Met-Transcript pair X Random pair Stronger correlations when metabolites are considered leading

24 level Causal relationships using the concept of Granger causality T variable 1 (transcript) M variable 2 (metabolite) A parameters E residual error Monotonic Signals: time High Pearson Correlation for all three relative time shifts, r=1 Low Granger causality, p-value=1 data not informative, Katrin Straßburg, Joachim Kopka

25 level Causal relationships using the concept of Granger causality T variable 1 (transcript) M variable 2 (metabolite) A parameters E residual error Artificial data: s2(t)=f[s1(t-1)] s1=>s2, p=0 s2=>s1, p=1 time Directionality deducable! Katrin Straßburg, Joachim Kopka

26 Causal relationships using the concept of Granger causality Cause Effect p-value Heat BH-corrected p-value Phenylalanine -> YPR047W, _at 3.38E E-02 Adenosine-5-monophosphate -> YBR115C, _at 1.46E E-01 Serine -> YGL026C, _at 2.37E E-01 YJR109C, _at -> Glutamic 3.41E E-01 Cold YKL106W, _at -> Phenylalanine 1.99E E-01 Fumaric -> YKL148C, _at 2.42E E-01 YER073W, _at -> Glyceric 2.75E E-01 YBR299W, _s_at -> Fructose 3.23E E-01 Arginine -> YDR341C, _at 3.38E E-01 YFL022C, _at -> Phenylalanine 3.95E E-01 Multiple Testing Correction problem: ~450 tests (metabolites-cognate transcript) Solution: Take only direction of greater significance (cuts number of tests in half) Katrin Straßburg, Joachim Kopka

27 Cause/Effect Metabolites: Examples Serine= Cause metabolite Phenylalanine= Effect metabolite Walther, Strassburg, Kopka, submitted

28 Identification of central Cause-Metabolites Glutamic acid and serine identified as central cause metabolites Metabolite is Cause is Effect p-value BH-corrected p-value Glutamic acid E E-04 is Cause Serine E E-04 is Effect Fructose-6-phosphate E E-02 Adenosine-5-monophosphate E E-02 Threonine E E-01 Glucose-6-phosphate E E-01 Glucose E E-01 Aspartic acid E E-01 Lysine E E-01 Valine E E-01 Proline E E-01 Isoleucine E E-01 Leucine E E-01 Alanine E E-01 Phenylalanine E E-01 Glyceric acid-3-phosphate E E-01 Glutamine E E-01 Citric acid E E-01 Glyceric acid E E-01 Glycerol E E-01 Homoserine E E-01 Trehalose E E-01 Asparagine E E-01 Malic acid E E-01 Fumaric acid E E-01 Ribose-5-phosphate E E-01 Tyrosine E E-01 Glycine E E+00

29 From Metabolites and Pathways to Genomes GC/MS Data Metabolomics databases: The Golm Metabolome Database (GMD) Automated GC/MS-spectra interpretation Tools Metabolites Metabolite-transcript causal relationships from time course data Pathways Metabolomics-assisted genome annotation in Chlamydomonas reinhardtii Genomes

30 Metabolomics & Proteomics-assisted genome annotation in Chlamydomonas reinhardtii Central Metabolism 159 metabolites 3,483 proteins identified by MS proteins with no peptide support MapMan view of Central Metabolism May et al. Genetics 2008 Björn Usadel, Patrick May, Stefanie Wienkoop, Stefan Kempa

31 Metabolic Expansion Algorithm: From metabolites to genes in medium detected by GC/MS Nils Christian, Oliver Ebenhöh (MPIMP, U Potsdam)

32 Metabolic Expansion Algorithm: From metabolites to genes in medium detected by GC/MS from KEGG : annotated in genome Nils Christian, Oliver Ebenhöh (MPIMP, U Potsdam)

33 Metabolic Expansion Algorithm: From metabolites to genes?? in medium detected by GC/MS from KEGG : annotated in genome Nils Christian, Oliver Ebenhöh (MPIMP, U Potsdam)

34 Metabolic Expansion Algorithm: From metabolites to genes in medium detected by GC/MS from KEGG : annotated in genome : inferred from KEGG Nils Christian, Oliver Ebenhöh (MPIMP, U Potsdam)

35 Metabolic Expansion Algorithm: From metabolites to genes in medium detected by GC/MS from KEGG : annotation : inferred from KEGG : presence predicted = minimal expansion Nils Christian, Oliver Ebenhöh (MPIMP, U Potsdam)

36 Omics-assisted metabolic pathway analysis in Chlamydomonas reinhardtii Building a Chlamydomonas draft network from KEGG using reciprocal best hit analysis of JGI 3.1 against KEGG 3365 nc, cp, mt proteins mapped on 198 KEGG pathways (JGI: 114) 7330 KEGG reactions 713 EC classifications (JGI: 552) network expansion used to determine which of the 159 metabolites are producible by the draft network 127 metabolites are represented in KEGG 70 are in the scope 57 are NOT in the scope AGs Ebenhöh, Weckwerth, Walther (MPIMP), May et al. (2008) Genetics

37 Chlamydomonas genome sequence Metabolic Expansion Algorithm: From metabolites to genes Sucrose detected in Chlamydomonas Sucrose pahtway genes including SPS (Sucrose Phosphate Synthase) must be encoded in the genome, but were not annotated in the draft genome Bioinformatic analysis of the Chlamydomonas genome identified candidate SPS gene aligned regions N s in genomic sequence sucrose-6-phosphate phosphohydrolase domain Arabidopsis SPS sequence

38 Summary GC/MS Data Classification tree based automated GC/MSspectra interpretation is possible Metabolites Pathways Metabolite-transcript causal relationships from time course data by using Granger causality metabolites appear to be leading and are more pathway-informative Metabolomics measurements help complete genome annotations Genomes

39 Acknowledgements AG Walther (MPIMP) Jan Hummel Patrick May Pawel Durek AG Kopka (MPIMP) Katrin Straßburg Nadine Strehmel Joachim Kopka AG Weckwerth (GoFORSYS) Stefanie Wienkoop Stefan Kempa Julia Weiss Luis Recuenco-Munoz Wolfram Weckwerth (now U Vienna) AG Ebenhöh (GoFORSYS) Nils Christian Oliver Ebenhöh (now U Aberdeeen) AG Usadel/Stitt (MPIMP) Björn Usadel AG Selbig (MPIMP, U Potsdam) Joachim Selbig Zoran Nikoloski