Canadian Bioinformatics Workshops
2Module #: Title of Module
Module 3 Metabolite Identification and Annotation – Part II
Learning Objectives Become aware/familiar with of other NMR spectral deconvolution tools Learn about how spectral deconvolution is applied to GC-MS data Become aware of NIST and AMDIS Learn about various MS database searches and MS databases Learn about molecular formula generation Learn about other techniques for unknown compound ID by MS
Goal of Metabolite Annotation ppm
Metabolite ID by Spectral Deconvolution (NMR) Mixture Compound A Compound B Compound C
Alternatives to Chenomx AMIX (Bruker) AutoFit (automated fitting) MetaboMiner (2D NMR) HMDB (NMR spectral match) PRIMe Spin Assgn (NMR spectral matching server) rNMR and BRMB Peaks Server BATMAN
AutoFit - Automated NMR Profiling
Performance of Autofit Synthetic Real P. Mercier et al. J Biomol NMR Apr;49(3-4):307-23
NMR Compound ID from Mixtures - MetaboMiner Raw TOCSY Spectrum ID’d Compounds
MetaboMiner Software Design Standard reference libraries –225 TOCSY spectra –488 HSQC spectra –Specialized sub-libraries for CSF, plasma and urine Algorithms for automatic processing & compound identification –“Minimal signature peaks” –1D 1H peak list as sanity check –Extra dimensional information for identification Support for direct spectral annotation
MetaboMiner Performance
NMR Compound ID - HMDB Phenyllactate Phenylpyruvate Phenylacetic acid Tropic acid Benzyl alcohol … NMR spectrum of mixture Peak list to HMDB High scoring matches
PRIMe Spin Assign
rNMR
BMRB Peaks Server
BATMAN
Metabolite ID by GC-MS GC -MS total Ion chromatogram
EI Breaks up Molecules in Predictable Ways Molecular ion Recall EI MS Generates Multiple Peaks
GC-MS Spectrum
Recall GC-MS Analytes are Derivatized Methoxime
Metabolite ID by GC-MS GC-MS is often best for identification of amino acids, organic acids, sugars, fatty acids and molecules with MW<500 GC has higher resolution and reproducibility than LC EI-MS is more standardized than soft ionization methods, so EI spectra are more comparable Most common route is to use AMDIS + NIST database
NIST 11 MS Database 243,893 EI spectra of 212,961 cmpds 9934 ion trap MS for 4649 cmpds 91,557 Qtof & QqQ spectra for 3774 compounds 224,038 RI values for 21,847 cmpds
NIST MS Search Software
AMDIS (Automated Mass Spectral Deconvolution and Identification System) Noise analysis –Determines background noise level Component perception –Identifies peaks by comparing to noise Spectral deconvolution –Generates a “clean” or model spectrum Compound identification –Identifies compounds via a library search using a match factor
Match Factor (MF) Measures the similarity of the MS spectrum of the query to the MS spectrum in the reference database Defined as the normalized dot product of the query and the reference spectra I ref corresponds to the intensities of the reference spectra, I qry corresponds the intensities of the query spectra, M corresponds to the masses (m/z) w is a weighting term to penalize uncertain peaks
GC-MS Protocol Prepare a set of external n-alkane standards (8-9 n-alkanes spanning octane to hexadecane) and run as an external calibration standard Run a “blank sample” containing just the solvent and derivatization agents Run the sample of interest (under the same conditions as the blank)
GC-MS Protocol External n-alkane standard used for RI calculation
GC-MS Protocol Create a calibration file using the n-alkane mixture (sets retention indices [RI’s] to the standard values) Analyze the sample data file against the CAL(calibration)-file for the alkane mixture (sets and recalculates RI's using the n- alkanes) Search the NIST database for matches and displaying the results of the search Get rid of “false” positives by comparing the “blank” against the sample spectrum
Step 1- Create Calibration File AMDIS
Step 2 – Calibrate Sample Spectrum Using CAL-file
AMDIS GC Peak List EI-MS Spectrum For Step 3 – Search NIST Database for Matches
Match factor 60% (if in doubt compare “blank” and your signal) Step 3 – Search NIST Database for Matches (Zero in) Reference Spectrum Peak Spectrum MF = 84% Match To Valine 73 & 144 are 2 most abund. m/z
Other GC-MS Options Alternatives to AMDIS –AnalyzerPro (SpectralWorks) –ChromaTOF (Leco) –Evaluated in TrAC Trends in Analytical Chemistry Volume 27, Issue 3, March 2008, Pages Alternatives to NIST08 or NISTII –Golm Database (Open access) –FiehnLib (Leco, Agilent) –HMDB???
The Golm Database GC-MS (Quad and TOF) database Contains MSRI (MS + retention index) or MST data for 1450 identified metabolites Includes 10,336 spectra linked to analytes Downloadable libraries compatible with NIST08 and AMDIS software Primary focus on plant metabolites Supports compound name and MS queries MS submissions via NIST08 or AMDIS format
Golm Database
Golm Database
The FiehnLib GC-MS Database 2212 EI MS and RI data for quadrupole &TOF GC-MS Over 1000 primary metabolites below 550 Da Covers lipids, amino acids, fatty acids, amines, alcohols, sugars, amino- sugars, sugar alcohols, sugar acids, sterolsphosphates, hydroxyl acids, purines
Metabolite ID by LC-MS LC -MS total Ion chromatogram
Levels of Metabolite Identification in MS 4 levels of metabolite identification Positively identified compounds –Confirmed by match to known standard Putatively identified compounds –Match to MS + RT or MS/MS + RT Compounds putatively identified in a compound class Unknown compounds
Metabolite ID by LC-MS LC-MS is often best for identification of lipids, bases, amino acids, organic acids, fatty acids and other somewhat hydrophobic molecules Metabolite ID typically requires both MS and MS/MS data (along with retention time information) and internal standards Compound ID can be done by high accuracy mass matching and/or by MS/MS matching to spectral databases
Simple MW Search DBs ChEBI ( PubChem ( ChemSpider ( HMDB (
PubChem MW Search Available Under “Advanced Search”
PubChem Results
ChEBI MW Search
Advanced MS Search DBs NIST/AMDIS ( Metlin ( HMDB ( MassBank (
Advanced MS Search DBs These databases support not only MW or MW range searches, but also support parent ion searches (positive, negative, neutral), peak list searches (from MS or MS/MS data) as well as MS/MS spectral matching These DBs are intended more for MS- based metabolomics and compound ID than the simple MW search tools
MS Compound ID - HMDB Phenyllactate Phenylpyruvate Atrolactic acid Homovanillin Coumaric acd LC-MS Spectrum Peak list to HMDB High scoring matches
MS Compound ID - HMDB Database of ~400,000 predicted masses from ~40,000 known metabolites Includes adduct mass calculations for 30+ possible or expected metabolite adducts Allows selection of different databases (DrugBank, HMDB, FooDB, T3DB), mass tolerance and ionization mode Designed for mixture deconvolution (i.e. identification of multiple compounds at a time)
MS/MS Compound ID - HMDB Database of 1000 experimental MS/MS spectra (low, medium and high collision energies) collected on QqQ - but largely valid for ion trap instruments as well Allows selection of different instruments (QqQ, ion trap, FT-MS qTOF), collision energies, ionization modes, parent ion mass tolerance and fragment ion mass tolerance Designed for identification of a single compound at a time
Metlin MS Search Step 1: Enter Mass Step 2: Select Charge Step 3: Select “all” Step 4: “Find Metabolites”
Metlin Results
Metlin MS/MS Search mzXML mzML mzData
Metabolite ID - Complications LC-ESI-MS often leads to the production of salt adducts, neutral loss species and multiply charged species Up to 50% of LC-MS signals arise from these “noise” sources Key challenge is to distinguish adducts or multiply charged species from parent ions or to group adducts or multiply charged species with parent ions
Adduct Formation Effect on ESI Mass Spectrum Sample Na Adducts
Common Adducts in DI-MS
Fiehn Lab Adduct Table
MZedDB – Adduct Calculator
MZedDB – Results for C6H12O6
Neutral Loss Fragments
Handling MS Complications MZedDB, Metlin and HMDB are able to handle or predict adducts Metlin and MZedDB are able to handle or predict ion pairs or multiply charged species Metlin can potentially handle or predict neutral loss species Searching by MS or MS ranges can lead to lots of hits (high FP rate)
Exploiting High Mass Accuracy to ID Compounds ppm Linear IonTrap ppm Triple Quad ppm Q-TOF ppm TOF-MS ppm Magnetic Sector ppm Orbitrap ppm FT-ICR-MS Mass AccuracyType (10 ppm in Ultra-Zoom)
Molecular Formula Generators Formula generators are used to create molecular formulae from very accurate masses obtained by FT-MS or OrbiTrap Assist in compound ID by LC-MS (formula is more restrictive than MW) Input typically requires: –Accurate isotopic mass (with or without adduct) –Error in ppm or mDa (milliDaltons)
Molecular Formula Generators (MWTWIN) Accurate mass Mass error
Molecular Formula Generators (HighChem)
Molecular Formula Generator Server (MZedDB)
Finding Compounds By Molecular Formula - PubChem
Finding Compounds By Molecular Formula - ChEBI
Formula Filters Use additional MS information (isotopic abundance) as well as chemical bonding restrictions (Lewis & Senior rules), known or presumed atomic compositional data and matches to known or hypothesized structures to reduce the possible # of structures/formulas that are generated
Fiehn’s 7 Golden Rules (7GR) Formula Filter
7GR Software
Molecular Formula Space of Small Molecules
Frequency Distribution of Molecular Formulas
Impact of Mass Accuracy on Formula Numbers
Mass + Isotope Abundance Example: ESI-MS (+) of Solanine on a LTQ Resolving Power: 1700 Mass Accuracy: 46 ppm Isotopic Abundance Error: ±1.46% C45H73NO15 MW = [M+H] +
Mass Isomers Are Hard To Distinguish by MS Alone Use Retention Time or Isomer Generators to Distinguish
Molecular Isomer Generators Example: MOLGEN DEMO (Bayreuth)MOLGEN DEMO Creates all possible structural isomers from a given molecular formula
Size of Molecular Isomer Space is Unknown Accurate massFormulaNumber Isomersin Beilstein DB CH2O CH6N2O C2H6O C4H2N C5H2O C6H C7H6N2O2100,082, C7H10N466,583, C8H6O36,717, C8H10N2O76,307, C9H10O26,843, C9H14N29,459, C10H2N265,563, C10H14O1,548, C11H2O9,414, C11H1884, C12H634,030,90512
Some Points of Caution Many databases (PubChem, ChEBI, Metlin, FiehnLib, NIST) mix non-metabolites with metabolites or plant metabolites with animal and/or microbial metabolites or drugs/buffer reagents with metabolites This leads to many “silly” hits If you know the source organism use this information to limit the search or use organism-specific metabolome databases (HMDB, FooDB, DrugBank, KnapSack, etc.)
Alternatives to Mass Filtering and Mass Matching Use chemoselective labeling (similar to proteomics) to simplify the identification of “true” metabolites, reduce number of signals and eliminate false positives Use MS-based kits (Biocrates) Use concepts in Computer-Aided Structure Elucidation (CASE) to assist in compound ID
Quantitative MS Metabolomics With Chemoselective Labeling LC-MS Analysis Mix Pooled AnalysisIndividual Analysis
Quantitative MS Metabolomics With Chemoselective Labeling
Quantitative MS Metabolomics in Human Urine 2.51mM 30 nM 672 peaks by amino labeling 120 standards spiked 92 peaks identified/quantified 30 nM mM 820 peaks by carboxy labeling Still assessing Guo K. & Li L. Anal Chem May 15;81(10):
Advantages to Derivitization Tags can convert non-UV active compounds into UV or fluorescently detectable cmpds Tags improve ionization efficiency and lower limit of detection Tags permit affinity purification and concentration Tags make polar molecules hydrophobic, leading to better LC separations Tags permit isotope based quantification Tags greatly increase # compounds detected Tags allow independent confirmation of “real” peaks Best route to automated ID & quantification by LC-MS
BioCrates IDQ Kit 40 acylcarnitines, 13 amino acids, 15 LysoPCs, 77 PCs, 15 SMs = 160
Multiple Reaction Monitoring Q1 Q3 CH 3 CD 3
Sample Urine Metabolite List Concentration range from 10 nM to 7.2 mM (1,000,000 X concentration) Arginine 38.7 uM Tyrosine uM C14:2 Carn 0.03 uM C4:1 Carn uM C8 Carnitine 1.05 uM PC(36:5) aa uM LysoPC-20: uM SM(22:3) uM Glutamine uM Valiine 37.0 uM C14:2-OH 0.02 uM C5 Carnit 4.39 uM C9 Carnitine 1.37 uM PC(38:5) aa uM LysoPC-6: uM SM(24:0) uM Glycine uM Leu/Ile uM C16 Carn uM C6-OH Carn uM PC(28:1) aa uM PC(42:4) aa uM SM(OH)16: uM SM(24:1) uM Histidine uM Carnitiine 73.2 uM C16-OH Cr uM C5-M-DC uM PC(30:2) aa uM PC(38:3) ae uM SM(OH)22: uM SM(26:0) uM Methionine 15.6 uM C10 Carn uM C16:1-OH uM C5-OH Carn 1.46 uM PC(34:1) aa uM PC(38:4) ae uM SM(OH)22: uM SM(26:1) uM Phenylalanin 52.7 uM C10:1 Carn 1.83 uM C2 Carnitine 45.2 uM C5:1 Carn 1.84 uM PC(34:2) aa uM PC(38:5) ae uM SM(OH)24: uM Glucose 2264 uM Proline 42.9 uM C10:2 Carn uM C3 Carnitine 2.12 uM C5:1-OH uM PC(34:4) aa uM PC(38:6) ae uM SM(16:0) uM Creatinine 7222 uM Serine uM C12 Carn uM C3-OH Carn uM C6 Carnitine uM PC(36:1) aa uM PC(40:5) ae uM SM(16:1) uM Threonine uM C14 Carn uM C4 Carnint 11.0 uM C6:1 Carnt uM PC(36:3) aa uM PC(42:3) ae uM SM(18:1) uM Tryptophan 15.0 uM C14:1-OH uM C4-OH Carn uM C8-OH Carn uM PC(36:4) aa uM PC(44:3) ae uM SM(20:2) uM
CASE – Computer-Aided Structure Elucidation Two approaches – Bottom Up and Top Down Top-Down uses known metabolites and generates variants (via metabolic transformation or other bio-informed methods). Properties/spectra/MW are predicted and then compares them to observed spectra/properties of unknown Bottom-Up uses known fragments of molecules, assembles the fragments into logical structure, predicts the properties/spectra and compares to observed spectra/properties of unknown
Top - Down CASE Methods Known metabolites (20,000) Predicted biotransformations (20,000 --> 200,000) Predicted MS, MS/MS, NMR, GC-MS Spectra Match observed spectra to predicted specta to ID
MyCompoundID Computationally metabolizes 8000 compounds in HMDB to create a dataset of 400,000 possible theoretical metabolites
MyCompoundID
Bottom-Up (Traditional) CASE Known metabolite substructures or metabolite EI or CID fragments Match observed spectra to predicted specta to ID Predicted (or kown) MS, MS/MS, NMR, GC-MS fragment spectra Neural Network or GA driven fragment assembly +