PTC Orbe “We innovate for our Consumers

beyond their Expectations”

Learning from other industries –

Insights from coffee on

advanced sensory-analytical

correlations

Imre Blank et al.

15th Australian Wine Industry Technical Conference

Sydney, July 16, 2013

2

Why to learn from coffee research ?

There are so many differences …

15th AWITC / I. Blank / 16.07.2013 2

Raw materials:

Beans vs

grapes

Technology:

Roasting vs

fermentation

Consumption:

Hot vs

cold/ambient

3

Why to learn from coffee research ?

There are quite a few similarities as well …

15th AWITC / I. Blank / 16.07.2013 3

Quality aspects

Freshness, off-

flavours Freshness Off-

flavours

Composition:

Aroma, different

structures

Taste,

polyphenols

Primary aromas

R4

R3

R2

R1

COOH R3

R2

R1

COOH

HO

R

C

O

O CH

COOH

CH

COOH

OH

O

O

OH

Sotolon

White wine

Wine Flavour Ageing aromas

Perception threshold: 2 µg/L:

Cork taint

(Silva Ferreira 2003; Pons et al 2008;

Tanner et al. 1981)

15th AWITC / I. Blank / 16.07.2013 4

2-Furanmethanethiol

Perception threshold: 0.4 ng/L

(Tominaga, 2003)

Wine Flavour – Sulphur compounds

(Dubourdieu & Tominaga, 2009)

15th AWITC / I. Blank / 16.07.2013 5

6

Why to learn from coffee research ?

There are similar challenges

15th AWITC / I. Blank / 16.07.2013 6

2. Sensory /

analytical correlation

Predictive model

qualitative quantitative sensory-analytical

correlation

>1000 volatiles 50 important

aroma cpds

23-26 key

odorants

Detection threshold: 0.8 ng/L

1. Analysis of trace

compounds

Chemical structure

Low thresholds

Formation of key

flavour compounds

Analytical challenge:

Flavour formation upon coffee roasting

15th AWITC / I. Blank / 16.07.2013 7

1

Green coffee beans (GC)

Incorporation of labelled precursors

EB + BR + labeled

precursors

Roasting

Spiked green beans

Spiking with precursors

Roasted beans spiked with precursors

Roasting

Precursor omission

EB + BR omitted in sugars/AA

Roasting

The Approach: New experimental set-up

Exhausted beans (EB)

Water extraction at 95°C for 2 h

Biomimetic recombinate

(BR)

EB + BR Roasting

+

Biomimetic in-bean study

Omission

Labelling Spiking

8 15th AWITC / I. Blank / 16.07.2013 8

Poisson et al. (2009) J. Agric. Food Chem., 57, 9923-9931

Composition of the biomimetic recombinate (BR)

15th AWITC / I. Blank / 16.07.2013 9

10

Formation of 2-furfurylthiol (FFT):

Model studies

Model System

• Tressl et al. (1993): FFT is formed from arabinose/cysteine via 3-deoxypentosone

(3-DP) and furfural while maintaining the intact carbon chain

• Grosch (1999): Arabinogalactans suggested as precursor of FFT by isolating the

polysaccharide from green coffee and roasting it in the presence of cysteine

Figure: Hypothetical formation of FFT in coffee from arabinogalactans or

arabinose (R=H) and cysteine (protein-bound)

O

OHOH

OH

OR

NH2

O

SH OH

OH

O

OSH

SH2

+

Arabinose Cysteine

3-DP

2-Furdurylthiol (FFT)

2-Furaldehyde

15th AWITC / I. Blank / 16.07.2013 10

11

Formation of 2-Furfurylthiol (FFT):

Omission and spiking experiments

• Omission of sugars favored the generation of FFT, whereas furfural content was highly suppressed

In-bean experiment

0%

25%

50%

75%

100%

125%

150%

175%

EB + BR EB + BR -Omission AA

EB + BR -Omission sugars

Re

lative

Co

nc. [

%]

2-furaldehyde 2-furfurylthiol

• Spiking with sucrose increased furfural amounts but considerably decreased concentrations of FFT

• Spiking experiment with cysteine resulted in enhanced FFT amounts, thus indicating cysteine as a suitable sulphur source

0%

25%

50%

75%

100%

125%

150%

175%

Green Coffee (GC)

GC + Sucrose

GC + Arabinose

GC + Cysteine

Re

lative

Co

nc. [

%]

2-furaldehyde 2-furfurylthiol

O

SH

15th AWITC / I. Blank / 16.07.2013 11

Poisson et al. (2009) J. Agric. Food Chem., 57, 9923-9931

12

Formation of 2-furfurylthiol (FFT):

Labelling experiments

2-furfurylthiol

88%

2.7% 4.2% 5.1%0.4% 0.9%

0%

20%

40%

60%

80%

100%

M

114

M+1

115

M+2

116

M+3

117

M+4

118

M+5

119

rela

tive In

ten

sit

y (

%)

Incorporation of D-[U-13C5]-arabinose did not yield fully labelled furfural nor

FFT, but partially labelled FFT with 13C1, 13C2 and 13C3-moieties

furfural

88%

8.5%1.4% 0.0% 1.2% 1.0%

0%

20%

40%

60%

80%

100%

M

96

M+1

97

M+2

98

M+3

99

M+4

100

M+5

101

rela

tive In

ten

sit

y (

%)

2-Furfurylthiol (FFT) is most likely not generated via the

furfural pathway

In-bean experiment

! 15th AWITC / I. Blank / 16.07.2013 12

Poisson et al. (2009) J. Agric. Food Chem., 57, 9923-9931

13

Analytical challenge:

Formation of wine favour

15th AWITC / I. Blank / 16.07.2013

Grape compound

•Nutrients

•Flavour precursors

•Non-precursor flavour-active compounds

Metabolism

•Catabolic/anabolic pathways

•Biotransformation

•Metabolism

Metabolites

•Fermentation bouquet

•Varietal compounds

•Phenolic adducts and polymers

Esters, higher alcohols, acids,

carbonyls, polysaccharides, volatile

sulphur compounds

Aging

•Oxygen / T effect

•Chemical reactions

•Controlled conditions

•Final wine flavour

Product storage

•Temperature effect

•Uncontrolled conditions

•Modif. composition

•Off-flavour

→ Advanced analytics

→ Molecular understanding

→ Labelling experiments

→ Sensory dimension

→ Targeted vs. holistic

13

1

14

Sensory/Analytical challenge:

Correlation & Predictive model

15th AWITC / I. Blank / 16.07.2013

P14 1st International Congress on Cocoa Coffee and Tea

THE CHALLENGE

SENSORY – ANALYTICAL CORRELATION

DIFFERENT NATURE OF DATA

THE APPROACH

ADVANCED ANALYTICS

MONADIC SENSORY PROFILING

ADVANCED STATISTICS

THE SOLUTION

RELIABLE PREDICTIVE TOOL

14

2

Can we predict sensory profiles by analytical data ?

15

The challenge: Understanding the coffee ‘melodie‘

15th AWITC / I. Blank / 16.07.2013 15

The flavour of coffee can be compared to a symphony played by an orchestra

Fundamentally different nature of sensory & analytical data

16

Relationship between “signal” intensity

and aroma concentration

Aroma concentration

Peak i

nte

nsit

y

Linear relationship between peak

intensity and aroma concentration

Major problem to overcome: The fundamentally different nature of

analytical and sensory data

Sigmoid relationship between perceived

aroma intensity and aroma concentration (Fechner, 1877)

Perception ~ k log (conc.)

15th AWITC / I. Blank / 16.07.2013 16

17

The Approach: Systematic study

15th AWITC / I. Blank / 16.07.2013

Develop a mathematical model based on quantitative analysis of

flavour compounds to predict coffee sensory profiles

Identify well correlated marker compounds for sensory

descriptors

17

12 coffee blends

(25 mL, 40 mL, 110

mL)

Sensory analysis

(12 panelists, 9

sensory descriptors)

Profiling

Quantitative

analysis

(42 odorants, 12

taste compounds)

Targeted approach Predictive analytical-

sensory correlation

model

18

Monadic sensory profiling with an expert panel

(n=12)

15th AWITC / I. Blank / 16.07.2013 18

19

Concentrations of 54 aroma and taste

compounds were determined

15th AWITC / I. Blank / 16.07.2013 19

20

Absolute quantification was carried out with

different state-of-the-art methods

Quantitative analysis of

• 42 aroma compounds

(quantification with isotope

dilution assay)

a. SPME-GC-MS

b. SPME-GCxGC-TOF MS

c. SPE-GC-MS

• 12 taste compounds

(external quantification)

a. HPLC-DAD

b. LC-MS/MS

15th AWITC / I. Blank / 16.07.2013 20

21

Comprehensive GCxGC-TOF/MS for quantification

of high impact trace coffee components

RT: 0.00 - 33.66

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32

Time (min)

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

11000

12000

13000

14000

15000

16000

17000

18000

19000

20000

21000

22000

23000

Co

un

ts

RT: 13.08

AA: 41504RT: 11.44

AA: 40188

RT: 5.60

AA: 20990

RT: 1.35

AA: 10422

RT: 5.67

AA: 4261

RT: 8.61

AA: 4626

NL:

2.32E4

FID

Analog

ICIS

42GC6-

101

?

Analysis of

methional

Methional peak

hidden behind

other peaks

Resolved by

deconvolution and

2-dimensional

techniques

15th AWITC / I. Blank / 16.07.2013 21

22

Pre-processing of analytical & sensory data is

key to perform multivariate statistics

Fechner‘s law:

perception ~ k log (concentration)

Sensory data

Normalize

Instrumental Data

Take logarithm

Normalize

Subtract intensity

Subtract (instrumental)

intensity

Correlation of the two datasets

X‘‘ = Y‘‘ + P

15th AWITC / I. Blank / 16.07.2013 22

23

Normalisation & transformation of analytical data

15th AWITC / I. Blank / 16.07.2013 23

24

The solution: Predictive model

15th AWITC / I. Blank / 16.07.2013 24

Coffees are widely distributed over sensory space

25

Combination of sensory & analytical spaces

using PCA

15th AWITC / I. Blank / 16.07.2013 25

26

Quality markers – 30 compounds exhibit strong

correlation to the sensory descriptors

15th AWITC / I. Blank / 16.07.2013 26

27

The robust statistical model allows a reliable

prediction of the sensory profile

15th AWITC / I. Blank / 16.07.2013 27

Principle components regression: 101 out of 106 data are below LSD

28

Holistic analytical approach – Also suitable to build

predictive sensory models from head-space data

15th AWITC / I. Blank / 16.07.2013 28

Lindinger et al. (2008)

29

Received for review October 24, 2007

Accepted November 23, 2007

Major source of data: Key literature

(Proceedings of the ASIC Symposium, 2010)

15th AWITC / I. Blank / 16.07.2013 29

30

Conclusions & Enjoy your coffee !

A mathematical model has been developed which

allows predicting the sensory profiles of coffee

Deeper understanding of link between sensory

descriptors and aroma markers

Useful tool to support product development of

coffee blends with new taste experiences

30

J. Baggenstoss,

T. Davidek,

A. Glabasnia,

J. Kerler,

Ch. Lindinger,

F. Mestdagh,

L. Poisson,

Ph. Pollien,

A. Rytz,

E. Thomas,

Ch. Yeretzian

15th AWITC / I. Blank / 16.07.2013