01/06/2017 v1.0.0

NMRANNOTATION

RémiServienPatrickTardivel

MarieTremblay-FrancoCécileCanlet

WHYDOWENEEDANAUTOMATEDNMRANNOTATIONTOOL

2

Metabolomic workflow

QuickTime™ et undécompresseur

sont requis pour visionner cette image.

QuickTime™ et undécompresseur

sont requis pour visionner cette image.

QuickTime™ et undécompresseur

sont requis pour visionner cette image.

QuickTime™ et undécompresseur

sont requis pour visionner cette image.

-0,25

-0,20

-0,15

-0,10

-0,05

-0,00

0,05

0,10

0,15

0,20

0,25

-0,30 -0,25 -0,20 -0,15 -0,10 -0,05 0,00 0,05 0,10 0,15 0,20 0,25 0,30

t[2]

t[1]SIMCA-P+ 12.0.1 - 2011-11-02 16:45:27 (UTC+1)

Urine, plasma, tissues…

Analytical analysis(NMR, MS…)

NMR and MS spectra

Data ReductionMultivariate

Statistical Analysis

GlucoseTaurineGlycine

Glutamate LactateLysineLeucine

Metabolites Identification

Control mice

Exposedmice

1HNMRspectrum ofreference compound

Glycine

9 8 7 6 5 4 3 2 1 0Chemical Shift (ppm)

3.57

2 1OH

NH23

O

§ Each group of protons gives one signal§ A singlet for Glycine

9.0 8.5 8.0 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0Chemical Shift (ppm)

4.21

4.21

4.22

4.23

4.23

4.26

4.27

4.37

4.38

4.39

4.44

4.49

4.52

4.53

4.55

4.55

4.56

4.77

4.79

4.79

6.03

6.04

6.106.118.15

8.21

8.22

8.41

8.84

8.84

8.85

8.85

9.15

9.16

9.35

NAD

10

11

O

1213

14a,14b

O OHOHP

15

O

OH

O

O- O

OP16

17a,17b

1819

O20

21

N+

OH

OH

26

25

22

24

23

27

NH2

O

N9

5

4

6

N3

N12

8

N7

NH2

• 1 signal is observed for each type of proton• 18 different signals for this compound• Complex spectrum

1HNMRspectrum ofreference compound

1HNMRspectrum ofbiological matrices

5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0Chemical Shift (ppm)

• No separation of compounds beforeNMR analysis

• A metabolite can have many signalsat different chemical shifts

• Many signals overlap

Identification of metabolites in mixtures is really complex

Metabolite identification(manually)

• Based on1Hchemical shift,coupling patternandcouplingconstants

• Comparison with annotated spectra ofsimilar matricesinliterature• Spiking :areference compoundis added inthesample• Comparison with NMRspectra ofreference compounds

– database in-house– Chenomx (commercial)– HMDB(freeware)

• Multi-dimensional NMR

2DNMRexperiments mostly used

• 1H-1HCOSY(Correlation SpectroscopY)experiment :correlations 1H-1Hvia3bonds

• 1H-1HTOCSY(TOtal Correlation SpectroscopY)experiment :correlations between allprotonswithinagiven spinsystem

• 1H-13CHSQC(Heteronuclear SingleQuantumCoherence)experiment :correlations 1H-13Cvia1bond

• 1H-13CHMBC(Heteronuclear MultipleBondCoherence)experiment :correlations 1H-13Cvia2,3or4bonds

C C

H H

C H

C C

H

C

C C

H H

C

H

Informationis splitted intwo dimensions

Informationonhydrocarbon skeletal

Example :1HNMRspectrum ofaqueous pup extract (PND2)ofrat

2D1H-1HCOSYNMRspectrum ofaqueous pup extract (PND2)

CH3-CH(OH)COOH

1,33 ppm

4,11 ppm lactatethreonine

alanine

glutamate

glutamineglutathion

glutathion

leucineIle valine

valine

lysine lysine

citrate

choline

taurine

b-glucose

glucose

3.78 ppm

1.33 ppm

4.11 ppm

TMSP

taurine

taurine

glucose

glucose

glycinecitrate

leucine

arginine

lysine

choline

lactate

threonine

uridine

inosine

lactatemethionine

betaine

1,33 ppm4,11 ppm 3,25 ppm

22,35 ppm

71,25 ppm

2D1H-13CHSQCNMRspectrum ofaqueous pup extract (PND2)ofrat

www.hmdb.ca/

Comparison with NMRspectra ofreference compounds:Human Metabolome database

Many responses (299) : the expert must check all the spectra to find the right metabolite

Comparison with NMRspectra ofreference compounds:Human Metabolome database

Comparison with NMRspectra ofreference compounds:Human Metabolome database

Comparison with NMRspectra ofreference compounds:Human Metabolome database

13 results

Comparison with NMRspectra ofreference compounds:Database in-house

Aqueous pupextract

Lactate

CreatineTaurine

Alanine

Valine

Glutamate

• Time-consuming• Complicated• Some resonances arenotidentified• Anautomated NMRannotationtool will be helpful :

– PackageRBatman:very complex andnotuser-friendly– BAYESIL(http://bayesil.ca/):webinterface(limited toserum,plasma,andCSFbiofluids)

– Chenomx :commercialsoftwareHao J,Liebeke M,Astle W,DeIorio M,Bundy JG,Ebbels TMD.Bayesian deconvolution andquantificationofmetabolites incomplex1DNMRspectra using BATMAN,2014,NatureProtocols,9(6):1416-1427.Hao,J.,Astle W,DeIorio M,Ebbels TMD.BATMANanRpackagefortheautomated quantificationofmetabolites from NMRusing aBayesian Model.Bioinformatics,2012,28(15):2088-2090.Ravanbakhsh S,LiuP,Bjordahl TC,Mandal R,GrantJR,WilsonM,EisnerR,Sinelnikov I,HuX,Luchinat C,greiner R,Wishart DS.(2015)Accurate,Fully-Automated NMRSpectralProfiling forMetabolomics.PLoS ONE 10(5):e0124219.

Metabolite identification(manually)

We developed a new tool for the annotation of NMR spectra : ASICS (Automatic Statistical Identification in Complex Spectra)

Objectives

Ø Automated identification of compounds in NMR spectra of complex mixtures

Ø Estimation of the proportion of the metabolites in the mixture

Modeling

Mixture Y spectrum Library Z1, …., Zn

Problems : noise and peaks warping

NoiseModeling

Several NMRspectra ofglucoseallow tomodelthenoise

Warping problems

Ø Metabolite concentration is different in pure compound spectrum and in mixture spectrum : peaks do not have necessary the same shape

Ø Experimental conditions are different between pure compound and mixture (pH, ionic strength, analyticalvariability…) : peaks can be shifted

Warping modeling

Two NMRspectra ofasame mixtureobtained indifferentconditions

Warping modeling

Before modeling After modeling

NMRspectra ofcholinepureorinthemixture

We used awarping function :F

Warping

Ø Shift depend on pH : shift is not the same for all compounds

Ø Deformations are localized

Ø Peaks can be shifted, but also expanded or contracted

Ø Maximum deformation given by the expert

Ø NMR spectra are recorded at pH 7 to limit the shifts

Metabolite spectrum modeling

• f1(t) :NMRspectrum ofpurecompound• NMRspectrum observed with noise

• Library{Z1,….,Z200}

Mixturespectrum modeling

Ø Amixturegis modeled asapositivefunction g(t):

Ø Signalis noised andwarped :

Ø a is therelativeproportionofmetabolite inthemixturespectrum :areaunderthecurve ofthemetabolite divided bythetotalareaofthespectrum

Ø Determine non-zero proportionsØ Estimate theproportions:quantification

Method

Ø Afalsedetection is ametabolite identified butthis metabolite isnotpresent inthemixture

Ø Thedetection threshold is theproportionfrom which we aresurethat themetabolite is identified

Ø Compromisebetween low probability offalsedetection andlowthreshold

Ø Inthis method :Ø Controltheprobability toobtain oneorseveral falsedetectionsØ Give thedetection threshold ofeach metaboliteØ Minimize thedetection threshold

Realdataset

A mixture of 6 metabolites :

Metabolites ProportionsD-Glucose 5.8 %Creatinine 21.0 %

L-Phenylalanine 3.3 %L-Proline 6.7 %

Benzoic acid 8.4 %Choline chloride 54.6 %

Library of 36 metabolites

Automatic cleaning

Black:mixturespectrum;Red :hydroxybenzoic spectrum

Eliminationof18metabolites

Identification

Metabolites Identification DetectionThreshold

Benzoic acid Yes 1.6 %

Choline Yes 2.3 %

Creatinine Yes 1.1 %

Glucose Yes 3.1 %

Phenylalanine Yes 2.1 %

Proline Yes 2.6 %

Acetic acid No 0.9 %

Galactose No 2.5 %

Mannitol No 1.6 %

Mannose No 2.4 %

Autres No < 2.5 %

Quantification

Metabolites Estimatedproportions

Real proportions

Benzoic acid 5.7 % 8.4 %Choline 55.8 % 54.6 %

Creatinine 13.3 % 21 %Glucose 5.5 % 5.8 %

Phenylalanine 3.9 % 3.3 %Proline 5.7 % 5.9 %

Detection thresholds are too high

Improvement ofdetection thresholds

Metabolites Identification New thresholds Previousthresholds

Acetic acid No 0.16 % 0.9 %Galactose No 0.19 % 2.5 %Mannitol No 0.24 % 1.6 %Mannose No 0.17 % 2.4 %Others No < 0.68% < 2.5 %

Synthesis Urine

• Mixin500mlofwater:- 3,8gofpotassiumchloride- 8,5gofsodiumchloride- 24,5gofurea- 1,03gofcitric acid- 0,34gofascorbic acid- 1,18gofpotassiumdihydrogenophosphate- 1,4gofcreatinine- 0,64gofsodiumhydroxide- 0,47gofsodiumbicarbonate- 0,28mlofphosphoric acid

Synthesis Urine

• Additionofmetabolites:- Hippuric acid :8.2mM- TMAO:7mM- acetic acid :0.8mM- Alanine:1mM- Betaine :1mM- Carnitine :0.7mM- Dimethylamine :1mM- Ethanolamine :1.5mM- Formate:0.4mM- Glucose:1mM- Glutamine:2mM- Glycine:0.6mM- Guanidinoacetate :0.8mM- Lactate:0.6mM- Lysine:1mM- Malonate :1.7mM- Trigonelline :0.6mM- Tyrosine:0.3mM

21metabolites detectables inNMR

1HNMRspectrum

Annotation:comparison with other tools

True Positive

False Positive

False Negative

True Negative

Accuracy (%)

Compounds in library

Computing Time

ASICS 17 10 4 145 92 176 2 mn 38s

MetaboHunter 4 51 17 795 92 867 < 1 mn

Batman 21 125 0 1 18 147 74 hours

Bayesil 12 17 7 53 73 89 10 mn 48s

Chenomx 15 48 6 269 54 338 < 1 mn

ASICS detects 17 metabolites among 21, but there are 10 false positive. The metabolites must be checked by the NMR expert.

The accuracy is acceptable compared to other tools

Quantification:comparison with other toolsCompound Real proportion ASICS’s

proportionBayesil’sproportion

Chenomx’sproportion(automatic fit)

BATMAN’sproportion

Creatinine 1 1 1 1 1Citric acid 0.434 0.693 12.38 Not identified 0.089Hippuric acid 0.338 0.344 Not identified 0.312 0.072Trimethylamine-N-Oxide

0.286 0.439 Not in library 0.311 1.311

Ascorbic acid 0.156 Not identified Not identified Not identified 0.568Malonic acid 0.073 Not identified Not identified 0.015 0.058Ethanolamine 0.062 0.044 Not identified Not identified 0.259L-Lysine 0.044 0.076 0.512 Not identified 0.214Dimethylamine 0.047 0.054 0.079 Not identified 0.025Betaine 0.042 0.053 0.246 0.055 0.754L-Alanine 0.042 0.056 0.219 0.045 0.230D-Glucose 0.041 0.059 0.705 0.046 0.023Guanidinoacetic acid 0.033 0.052 Not identified 0.022 0.226

L-Carnitine 0.033 0.045 0.159 0.023 0.029L-Glutamine 0.032 Not identified 4.100 Not identified 0.342Acetic acid 0.032 0.031 Not identified 0.035 0.129Glycine 0.031 0.036 0.089 0.032 0.304Lactic acid 0.028 0.027 0.307 0.018 0.025Trigonelline 0.026 0.011 Not in library 0.018 0.044Formic acid 0.017 Not identified 0.006 0.029 0.007L-Tyrosine 0.012 0.024 Not identified 0.014 0.571

The quantifications provided by ASICS or by Chenomx both fit quite well the order of magnitude of the real proportion of the different metabolites, but these quantifications are not accurate

Conclusion

• The program is powerful for detecting the presence of compounds in mixture

• The program is not very powerful for estimating the trueproportion, but we are working on this problem

• At the moment, pH of compounds in the library is 7, but wecan import spectra at different pH

• At the moment, we have 175 compounds in the library but wewill record more compounds (MetaboHub)

• More tests are needed (other mixtures)• This tool can be used to help the expert to annotate the

metabolites in a complex mixture • Publication has been submitted

HOWTODOWITHGALAXY?

39

FORM

Name of the zip file

Exclusion zone

Maximum shiftExecute

MATRIXRESULT

Metabolites names

Relative concentrations

GRAPHRESULT

GRAPHRESULT

GRAPHRESULT

Example :Synthesis Urine

• Upload the1HNMRspectrum ofsynthesis urine• Region toexclude :6.5-4.5ppmforurine• Determine themetabolites present inthis sample• Displaythespectra• Comparetheresults with thelist ofmetabolitespresent inthemixture

Example :Synthesis Urine• Listofmetabolites present inthesample :

– Creatinine (24.8mM;0.378)– Citric acid (10.7mM;0.163)– Hippuric acid (8.2mM;0.125)– TMAO(7mM;0.107)– Glutamine(2mM;0.030)– Malonate (1.7mM;0.026)– Ethanolamine (1.5mM;0.023)– Alanine(1mM;0.015)– Betaine (1mM;0.015)– Dimethylamine (1mM;0.015)– D-Glucose(1mM;0.015)– Lysine(1mM;0.015)– Acetic acid (0.8mM;0.012)– Guanidinoacetate (0.8mM;0.012)– Carnitine (0.7mM;0.011)– Glycine(0.6mM;0.009)– Lactate(0.6mM;0.009)– Trigonelline (0.6mM;0.009)– Formate(0.4mM;0.006)– Tyrosine(0.3mM;0.004)