Content uploaded by Gerard Downey
Author content
All content in this area was uploaded by Gerard Downey on Apr 25, 2016
Content may be subject to copyright.
Food Authentication Using
Vibrational Spectroscopic
Techniques
Professor Gerard Downey
Teagasc Food Research Centre, Ashtown
Dublin 15, Ireland
1
Talk Outline
•Food authentication
–analytical challenge, expected outcome, possible
implementation…….
•Modern food assembly
–supply chains, ingredients, analytical support needs….
•Vibrational spectroscopy
–principles, practice, chemometric tools, reported
applications, needs, how to use analytical outputs,
limitations…….
•Conclusions
2
Food authentication
•Extremely complex problem
–Range of potential adulteration issues
•geographic origin, species, process, substitution
–Complexity and variability of sample matrices
•pastes, liquids, solids, powders
–Each detected incident is unique to a food and adulterant;
adulterant-specific assays can only be applied after the
event
–Sensitivity of the issue means pressure for quick and
definite response
•can we identify possible issues in advance and prepare?
•can we develop reliable untargeted analytical methods?
–Linkage to food safety issues increases pressure to monitor
3
Modern Food Assembly Operations
•Modern food assembly operations have
complex linkages often with no or limited
face-to-face interaction
•Companies need pharmaceutical industry
approach to control:
–control incoming raw material and
–confirm outgoing product
4
Modern Food Assembly Operations
5
Transfer Point 1
Transfer Point 2
Transfer Point 3
Transfer Point 4
Transfer Point 5
Traceability
Paper
Trail
Confirmatory Analysis
For Authenticity?
Analytical Requirements – Processor
•Should be able to confirm conformance to
specification
•Should be capable of operation at factory intake;
that means simple to operate by relatively
unskilled workers
•Should be reliable enough to give confidence in
decisions to prevent entry of material into plant
•Should be easy to adapt to the involvement of
new ingredients or suppliers
•Cost of analysis should be low
6
Can Vibrational Spectroscopy Help?
•Sensors based on vibrational spectroscopy
have particular advantages:
–non-destructive analysis
–rapid, so real-time measurements possible
–robust so withstand even harsh processing
environments
–relatively inexpensive; low cost per
analysis
•How can we deploy them?
7
Basic Principles of Vibrational
Spectroscopy
•Spectroscopy deals with the interaction between
photons of radiation and molecules
•Interactions at specific wavelengths or
frequencies arise because of molecular vibrations
•Responses depend on the molecular structure of
compounds - chemical bonds
•Measurement of the response pattern can thus
give information about molecular structure;
what & how much is present?
8
Types of Data Collected
9
FT-IR
NIR Reflectance
Raman spectra of milk powder
10
Spectroscopic Sensor Characteristics
•4000-400 cm-1
•Fundamental
absorption bands
•High structural
info
•Little or no sample
preparation
required
•Small sample size
only
•Limited fibre optic
use
•Some sensitivity
to water content
of sample
•Weak absorptions
•12500-4000 cm-1
•Overtones &
combination bands
•Low structural info
•Little or no sample
preparation required
•Large sample sizes
•Considerable fibre
optic use
•Very sensitive to
water in sample
•Very sensitive to
particle size of
sample
• 400-4000 cm-1
Raman shift
• Fundamental
absorption bands
•High structural info
• Little or no sample
preparation required
•Very small sample
sizes
• Limited fibre optic
use
•Not affected by
water in sample
Mid IR NIR Raman
Equipment
•Laboratory and industrial equipment available
for all spectroscopic methods
•Handheld units now a reality.
11
How Do We Use This Data?
•There are two questions we can ask
–Is this sample the same as some specified
material?
•Classification
–To which of a limited number of possible sample
types does this sample belong?
•Discrimination
12
13
The Discriminant Process
Classification
techniques build a
boundary between the
classes,
They always make
an assignment for
unknown objects
Class 1 objects Class 2 objects
Assignments seem
definitive:
YES/NO
One class or the other
14
…so how do we interpret this?
15
An alternative: class-modelling
Class-modelling
techniques build a
model for each class
by defining
┼
┼acentroid
and a dispersion
vector
16
Class-modelling: 2
For this reason, an
object may be assigned
to
1 class
>1 class, or
no class
17
SIMCA UNEQ
POTFUN
Sophisticated
algorithms can
handle complex
sample
distributions
Example 1 – Beef adulteration by offal
•Beef is an important component of the typical European diet
–in 2012, EU beef consumption was 6.94 million tonnes
•In recent years, there has been a shift to more
economical, minced meat products such as beefburgers
•Such products may be particularly susceptible to fraud given
–the long supply chains involved in their manufacture
–the pressure to achieve financial returns, especially from lower-
cost items
18
Background Issues - 2
•Critically, all physical meat identification clues are lost to the
consumer on mincing
•Reliance is instead placed on the brand, retailer reputation or
label
•Blending of beef offal (heart, kidney, liver or lung) is a
potential adulteration issue
•In a designed experiment, could a vibrational spectroscopic
method detect offal-adulteration of beefburgers?
19
The work - 1
•Authentic beefburgers were prepared at two quality levels
–High quality (lean beef + beef fat only)
•lean meat content varied between 80% w/w and 100% w/w of beefburger in 2.5%
increments; fat accounted for the remainder
–Lower quality ( lean beef, beef fat, rusk (5% w/w)+ water (20% w/w)
•beef at 45–62.5% w/w in 2.5% increments
•beef fat at 22.5–10% w/w in 2.5% increments
•Each group of beefburgers was produced on two separate occasions; therefore, a
total of 36 (18 higher quality and 18 lower quality) authentic beefburgers was
prepared
•A number of individual beefburgers was prepared for each recipe; 1 was chosen
randomly and stored overnight at 4 °C prior to analysis while the others were
stored at -20 °C for 2-3 weeks 20
The work - 2
•Adulterated beefburgers were formulated with lean beef, beef fat, water,
rusk and offal (liver, lung, kidney and heart).
•Formulations were produced using a D-optimal experimental design
•A total of 46 different beefburger formulations was generated
•These beefburgers were produced in random order over a period of
several days; because of time constraints, each formulation was produced
once. They were analysed fresh ( overnight at 4 °C ) and after storage at -
20 °C for 2-3 weeks
23
The work - 3
•Samples were homogenised prior to spectral collection
(Robot Coupe R301 ultra) for 30 s
•NIR: Spectra recorded from 400-2498 nm at 2 nm intervals.
Samples analysed in duplicate with re-packing
•FT-IR: Spectra recorded from 800–4000 cm-1 at a
nominal resolution of 4 cm-1; 64 sample scans averaged
and corrected using an air blank reference
•Raman: Dispersive spectra recorded from 250-3300cmˉ¹
using 780nm laser.
•All measurements made at ambient temperature (~20 °C);
samples were scanned in random order
22
The work - 4
•Discriminant analysis was by PLS; class-modelling was by
SIMCA
•Models were developed and evaluated on separate
calibration and evaluation sample sets
•Models were developed for fresh samples, frozen-then-
thawed samples and both together
•Raman results available for frozen-then-thawed only
23
The spectra-NIR
NIR reflectance [log (1/R)] spectra of fresh and frozen-then-thawed beefburgers (850–1098 nm). (a) Fresh authentic;
(b) frozen-then-thawed authentic; (c) fresh adulterated; (d) frozen-then-thawed adulterated.
Peak centred
around 970 nm
likely to involve
water: 2nd
overtone –OH
stretch
24
The spectra-FTIR Plots of (a) all beefburger spectra and mean
spectra of (b) authentic and (c) adulterated
samples
~1640 cm-1: -OH or amide I
~1550 cm-1: amide II
~950–1200 cm-1: CHO, maybe glycogen
~1740 cm-1: lipid C=O absorptions
25
The spectra - Raman
26
Spectra of (a) authentic
beefburgers; (b) adulterated
beefburgers; (c) average spectra
of authentic and adulterated
beefburgers.
The principal component scores - NIR
Score plots (a) fresh, (b)
frozen-then-thawed; PC1
and PC2 loading plots of all
beefburger samples
(c) fresh, (d) frozen-then-
thawed
1: authentic sample
0: adulterated sample
27
ab
c
The principal component scores - FTIR
(a) Scores of fresh beefburgers; (b) PC6 loading for fresh beefburger samples; (c) Scores of frozen- then-
thawed beefburgers; (d) PC3 loading of frozen-then-thawed beefburgers.
1 = authentic sample
2 = adulterated sample
28
The
principal
component
scores -
Raman
29
PCA score plots of all beefburger samples (a)
PC1 vs. PC2, (b) PC1 vs. PC3; loading plots of
all beefburger samples; (c) PC1, (d) PC2 and
(e) PC3. (# = number of adulterated beefburger
formulation in Table 1)
The results – PLS discriminant analysis
PLS-DA % Correct Classification of
Validation Samples
Sample Type Pre-treatment # loadings Authentic Adulterated
NIR
Fresh 2der17 8 100 95.5
Frozen-then-thawed MSC 3 100 91.3
Both types MSC 3 100 88.9
FTIR
Fresh Any or none 6-10 100 100
Frozen-then-thawed MSC 13 100 100
2der11 8 100 100
Both types MSC 13 100 100
Raman
Frozen-then-thawed Unit vector
normalisation
3 100 90
30
The results – SIMCA class-modelling
Sample Type Pre-treatment # PCs Sensitivity Specificity Efficiency
NIR
Fresh MSC 2 1 1 1
Frozen-then-thawed MSC 2 0.88 1 0.94
Both types MSC 2 0.91 0.99 0.95
(0.97*) (0.96*) (0.96*)
FTIR
Fresh None 4 0.94 0.80 0.87
Frozen-then-thawed MSC 5 0.94 0.87 0.90
Both types MSC 5 1.0 0.91 0.95
Raman
Frozen-then-thawed Unit vector
normalisation
51 0.89 0.94
2der.17 plus u.v.nor 2 (0.94*) (1*) (0.97*)
*Using PLS scores
31
The outcome
•On this well-defined but not unrepresentative sample set,
–FTIR is the technique of choice for discrimination (PLS-DA) of
any sample type
–NIR is the technique of choice for class-modelling (SIMCA) of
fresh samples; Raman is the best choice for frozen-then-thawed
samples
•A spectroscopic approach is appropriate for use in a production
facility with a defined recipe or recipes to detect samples which are
not what they should be
•Any such applications will require a research input for model
development involving e.g. sampling over longer time-periods ,
including the use of different ingredients such as onions or onion
powder etc.
32
Implications
•These and many other reports reveal the
undoubted potential (practical and scientific)
of vibrational spectroscopic methods for
specific authentication problems.
•Are there any problems preventing their use?
•How can we apply them in industry?
–check conformance to specification
33
Issues - 1
•A significant issue is the development, storage and maintenance of
spectral collections. All models depend on these
•Many models will be for a specific problem but enough global issues
exist to make generic models worthwhile e.g. skim milk powder
•Who will do this?
–Food Industry?
–Private R&D company?
•Most likely model may be licencing of spectral databases and
qualitative models by tech companies just as many quantitative
calibrations are licenced currently
•Development of models will depend on industry demand
34
Issues - 2
•Chemical basis for model performance not easily
understood
–critical importance of library construction and data analysis
•Ability to transfer models between instruments needs
to be demonstrated
•Results of screening analyses need to be seen as
supports to business decisions:
–what risks are associated with the acceptance of incorrect
product?
–what risks are associated with the rejection of correct
product?
35
Conclusions
•Vibrational spectroscopic tools have demonstrated a
capability to provide useful answers to food
authentication problems
•They are easy and economical to deploy
•They have been used for monitoring conformance to
specification in the pharmaceutical industry for many
years (PAT systems)
•The food industry should capitalise on this expertise and
improve its security
36
Thank You For Your Attention!
37