Investigation of the Confinement Odour 

Problem in Exported Lamb using NMR-

based Metabolomics 

 

 

A thesis presented in partial fulfilment of the requirements for the 

degree of 

 

 

Master of Science 

in 

Chemistry 

 

 

at Massey University, Manawatū, New Zealand. 

 

 

 

Natalia Olivecrona 

2015



Copyright is owned by the Author of the thesis.  Permission is given for 
a copy to be downloaded by an individual for the purpose of research and 
private study only.  The thesis may not be reproduced elsewhere without 
the permission of the Author. 
 



   
 

i 
 

 

 

 

Dedicated to the Memory of Moira Lynette Fay Olivecrona 



   
 

ii 
 

Abstract 

Recent changes to the supply chain practices of meat exporters has increased the potential for 

consumers to be exposed to the phenomenon of confinement odour, the smell produced by 

vacuum or modified atmosphere packaged meat which has been chilled and stored for 

extended periods. This harmless odour, which does not indicate meat spoilage, can lead to the 

rejection of the product by consumers. This is a problem for NZ lamb meat producers as they 

form the largest group of exporters of lamb meat in the world, and their largest market is the 

UK and other EU countries.  

The processes behind confinement odour development are poorly understood. In this thesis, 

NMR spectra were acquired of meat, and drip extracts of meat from two different processing 

plants stored under different temperatures for 11-13 weeks to simulate conditions of exported 

meat during overseas shipment, transport to warehouse and retail display. The spectra were 

analysed by multivariate data analysis to find metabolic differences between meat which 

produces confinement odour and meat which produces either spoilage odour or no odour.  

Optimisation of extraction of metabolites from meat and drip samples was also carried out. 

The best sample preparation method for meat and drip included homogenisation by bead 

beating (meat samples only), protein precipitation using an acetonitrile, methanol and acetone 

solvent mixture, and removal of solvent by vacuum centrifugation. 

Multivariate data analysis demonstrated the ability to discriminate drip samples with 

confinement odour from spoiled samples and the former showed increased lactate 

concentration with low levels of leucine indicating the presence of Lactic Acid bacteria. The 

spoiled samples had increased butyrate levels which is indicative of the presence of 

Clostridium spp. Both bacterial populations were in a late stage of growth. This is consistent 

with confinement odour as an early indicator of spoilage. This result indicates the potential for 

drip to be utilised more widely for the analysis of meat metabolites. 

Additionally, samples could be discriminated by processing plant of origin using multivariate 

data analysis. Increased levels of pyruvate and decreased levels of glucose in samples from 

Plant 2 indicated their bacterial populations had progressed to a later stage of growth than the 

bacterial populations in samples from Plant 1.  

  



   
 

iii 
 

Acknowledgements 

First I would like to thank my supervisor Dr Patrick Edwards who encouraged me to pursue this 

project and introduced me to the field of metabolomics. His wealth of knowledge of NMR 

spectroscopy and valuable time was freely shared and hugely appreciated. I would also like to 

thank my supervisor Dr Linda Samuelsson, who provided endless encouragement, proof-

reading of my thesis and expert knowledge of so many of the techniques and methods used in 

this research. Many thanks to Dr Mariza Reis and my supervisor Dr Marlon Reis for providing 

the framework and samples for this work, and their guidance throughout the entirety of my 

research. Thanks to the C. Alma Baker Trust for the provision of a scholarship, and AgResearch 

for providing a stipend without which I would not have been able to complete these studies.  

Particular thanks go to Assoc. Prof. Kathryn Stowell for her assistance and guidance in helping 

me obtain a suspension to my studies during a particularly distressing time. I would like to 

thank my mum Moira Olivecrona for her constant support and encouragement; she was my 

greatest rock and my strongest advocate. Sadly she passed away before this could be 

completed but she never doubted for a moment that it would be. Thanks to my dad Stenfinn 

and my sister Courtney who contributed with their love, support and in so many other 

innumerable ways. Without them this would never have happened. Finally I would like to 

thank Jim for sharing every happy, sad and frustrating moment and supporting me every step 

of the way. 

  



   
 

iv 
 

Contents 

Abstract .......................................................................................................................................... ii 

Acknowledgements .................................................................................................................. iii 

List of Figures ............................................................................................................................... vii 

Glossary of Abbreviations .............................................................................................................. x 

NMR Experiments ...................................................................................................................... x 

Statistical Analysis ...................................................................................................................... x 

Meat Science .............................................................................................................................. x 

Chapter One Introduction ............................................................................................................. 1 

 Export of NZ Meat .............................................................................................................. 1 1.1.

1.1.1. Meat Spoilage ............................................................................................................. 2 

1.1.2. Acceptability Characteristics in Meat ......................................................................... 3 

1.1.3. Confinement Odour and Sensory Evaluation .............................................................. 3 

1.1.4. Packaging and Storage for Improved Meat Quality .................................................... 4 

 Metabolomics .................................................................................................................... 5 1.2.

1.2.1. Analysis Methods ........................................................................................................ 7 

1.2.2. Meat Metabolomics .................................................................................................... 9 

1.2.3. Previous Comparison of Drip and Meat .................................................................... 11 

1.2.4. Experimental Design Considerations ........................................................................ 11 

 NMR ................................................................................................................................. 12 1.3.

1.3.1. Underlying Principle .................................................................................................. 12 

1.3.2. Experimental Variations ............................................................................................ 13 

1.3.3. Benefits for Use with Metabolomics ........................................................................ 14 

1.3.4. Experimental Considerations .................................................................................... 14 

 Multivariate Statistical Analysis ....................................................................................... 15 1.4.

1.4.1. Data Pre-Processing for NMR and Multivariate Statistical Analysis ......................... 16 

1.4.2. Principal Components Analysis ................................................................................. 21 

1.4.3. Outliers ...................................................................................................................... 23 

1.4.4. Partial Least Squares – Discriminant Analysis ........................................................... 24 

1.4.5. Orthogonal Partial Least Squares – Discriminant Analysis ....................................... 25 

1.4.6. Limitations and Common Pitfalls .............................................................................. 26 

1.4.7. Validation .................................................................................................................. 26 

1.4.8. Univariate Statistical Analysis ................................................................................... 28 

 Aims and Objectives ......................................................................................................... 28 1.5.



   
 

v 
 

Chapter Two Materials and Methods ......................................................................................... 29 

 Evaluation of sample preparation methods .................................................................... 29 2.1.

2.1.1. Chemicals .................................................................................................................. 29 

2.1.2. Methods for preparation of drip samples ................................................................ 29 

2.1.3. Methods for preparation of meat samples ............................................................... 30 

 Preparation of confinement odour samples .................................................................... 33 2.2.

2.2.1. Preparation of drip samples ...................................................................................... 34 

2.2.2. Preparation of meat samples .................................................................................... 34 

 NMR Analysis ................................................................................................................... 34 2.3.

2.3.1. Spectral Acquisition .................................................................................................. 34 

2.3.2. Spectral Processing ................................................................................................... 35 

 Statistical Analysis ............................................................................................................ 35 2.4.

2.4.1. Processing for Metabolomics .................................................................................... 35 

2.4.2. Statistical Analysis Methods ..................................................................................... 37 

2.4.3. Metabolite Identification .......................................................................................... 38 

Chapter Three Results and Discussion ........................................................................................ 43 

 Evaluation of Sample Preparation Methods .................................................................... 43 3.1.

3.1.1. Methods for Preparation of Drip Samples ................................................................ 44 

3.1.2. Methods for Preparation of Meat Samples .............................................................. 49 

3.1.3. Chosen Sample Preparation Methods ...................................................................... 52 

 Confinement Odour Study ............................................................................................... 54 3.2.

3.2.1. Confinement Odour and Spoilage Odour Metabolite Differences ........................... 54 

3.2.2. Metabolite Differences between Processing Plants ................................................. 62 

3.2.3. Metabolite Differences between Drip and Meat ...................................................... 73 

Chapter Four Conclusion and Future Work ................................................................................ 80 

 Conclusions ...................................................................................................................... 80 4.1.

4.1.1. Optimal Preparation Protocol for NMR Analysis of Lamb Drip and Meat ................ 80 

4.1.2. Confinement Odour Metabolites in Lamb Drip and Meat ........................................ 80 

4.1.3. Metabolites Differentiating Processing Plant of Lamb Drip and Meat ..................... 81 

4.1.4. Metabolites Differentiating Lamb Drip and Lamb Meat ........................................... 81 

 Future Work ..................................................................................................................... 81 4.2.

4.2.1. Sampling of Lamb Drip and Meat at Various Time Points for Time Series Analysis of 

Confinement Odour Metabolites ........................................................................................ 81 



   
 

vi 
 

4.2.2. Mass Spectrometry Analysis of Meat and Drip for Confinement Odour Metabolites

 ............................................................................................................................................ 81 

4.2.3. Treatment Changes to Increase Control Samples ..................................................... 82 

Appendix A PLS-DA Analysis and Validation ............................................................................... 87 

Analysis of Drip for Confinement Odour Metabolites ........................................................ 87 

Analysis of Drip for Processing Plant Metabolites .............................................................. 87 

Analysis of Meat for Confinement Odour Metabolites ...................................................... 89 

Analysis of Meat for Processing Plant Metabolites ............................................................ 89 

Analysis of Drip and Meat for Confinement Odour Metabolites ........................................ 91 

Analysis of Drip and Meat for Partitioning of Metabolites ................................................. 92 

Analysis of Drip and Meat for Processing Plant Metabolites ............................................. 93 

 

  



   
 

vii 
 

List of Figures 

Figure 1 Interaction of the Metabolome (Griffin & Shockcor, 2004) ........................................... 6 

Figure 2 Splitting of Nuclei Spin States in an External Magnetic Field ("Nuclear Magnetic 

Resonance,") ............................................................................................................................... 12 

Figure 3 1H NMR Spectrum of a Mixture of α-glucose and β-glucose in D2O Acquired at 

400MHz ....................................................................................................................................... 13 

Figure 4 Overlay of 1H NMR Spectra of Filtered Plasma and Unfiltered Plasma ........................ 15 

Figure 5 Section of Overlaid Plasma 1H NMR Spectra from 7.71 - 7.95 ppm ............................. 17 

Figure 6 Section of Overlaid Plasma 1H NMR Spectra from 7.71 - 7.95 ppm with Local Alignment

 .................................................................................................................................................... 18 

Figure 7 Section of Full Resolution Plasma 1H NMR Spectrum Showing 0.8 - 2.75 ppm ............ 19 

Figure 8 Section of Plasma 1H NMR Spectrum Showing 0.8 - 2.75 ppm Binned to 0.04 ppm .... 19 

Figure 9 Example of Principal Components Scores Plot, Outliers are Shown Circled in Red ..... 22 

Figure 10 Example of Principal Component Analysis Loadings Plot ........................................... 23 

Figure 11 PCA Scores Plot of Plasma Coloured by Year of Collection ......................................... 24 

Figure 12 PLS-DA Scores Plot of Plasma Coloured by Year of Collection .................................... 25 

Figure 13 Permutation Test for a Valid PLS-DA Model ............................................................... 28 

Figure 14 1D Loadings Plot of an Example OPLS-DA Model with Loadings Coloured by VIP Value

 .................................................................................................................................................... 39 

Figure 15 1D Loadings Plot of an Example OPLS-DA Model with Loadings Coloured by VIP 

Value, Loadings with VIP≥1.5 Shown and a Loading with Error Greater than its Value Circled in 

Red .............................................................................................................................................. 40 

Figure 16 S-Plot From an Example OPLS-DA Model with Loadings Coloured by VIP Value, 

Loadings Combining High Correlation and High Covariance Shown Circled in Red ................... 41 

Figure 17 Overlaid 1H NMR Spectra of Drip with Protein Removed by Centrifugation (Dark Red) 

or Ultrafiltration (Light Blue) ....................................................................................................... 43 

Figure 18 Overlaid 1H CPMG (Dark Red) and NOESY (Light Blue) NMR Spectra of a Drip Sample 

with Protein Removed by Centrifugation ................................................................................... 45 

Figure 19 Overlaid Spectra of Three Experimental Replicates of Drip with Protein Removed by 

Centrifugation (Enlargement of Section 8.14 - 8.42 ppm) .......................................................... 46 

Figure 20 Overlaid Spectra of three Experimental Replicates of Drip with Protein Removed by 

Ultrafiltration (Enlargement of Section 8.13 - 8.40 ppm) ........................................................... 47 

Figure 21 Overlaid Spectra of Three Experimental Replicates of Meat with Protein Removed by 

Perchloric Acid Solvent Precipitation (Enlargement of Section 8.06 - 8.58 ppm) ...................... 51 

Figure 22 PCA Scores Plot of Drip and Meat Samples Coloured According to the Procedures 

Outlined in the Sample Preparation Evaluation ......................................................................... 53 

Figure 23 PCA Scores Plot of Drip Coloured by Odour Status ..................................................... 54 

Figure 24 PCA Loadings Plot of Drip Coloured by Odour Status, Bins with Largest Variation 

across Each Component Circled in Red ....................................................................................... 55 

Figure 25 Box Plots of Significant Bins from T-Test of Drip Samples Classed by Odour Status 

(Red Indicates Samples with Confinement Odour, Green Indicates Samples which are Spoiled)

 .................................................................................................................................................... 56 

Figure 26 PCA Scores Plot of Meat Coloured by Odour Status ................................................... 57 



   
 

viii 
 

Figure 27 PCA Loadings Plot of Meat Coloured by Odour Status, Bins with Largest Variation 

across Each Component Circled in Red ....................................................................................... 57 

Figure 28 PCA Scores Plot of Drip and Meat Samples Coloured by Odour Status ...................... 58 

Figure 29 PCA Loadings Plot of Drip and Meat, Bins with Largest Variation across Each 

Component Circled in Red .......................................................................................................... 59 

Figure 30 PCA Scores Plot of Drip Coloured by Processing Plant of Origin................................. 62 

Figure 31 OPLS-DA Scores Plot of Drip Samples Classified by Processing Plant of Origin, Samples 

which are More Similar to Samples from the Other Plant than from Their Own Shown Circled in 

Red .............................................................................................................................................. 63 

Figure 32 OPLS-DA 1D Full Loadings Plot of Drip Samples Classified by Processing Plant of 

Origin, Loadings Coloured by VIP Value ...................................................................................... 64 

Figure 33 OPLS-DA 1D Loadings Plot of Drip Samples Classified by Processing Plant of Origin, 

Loadings with VIP Value Greater than 1.5 Shown ...................................................................... 65 

Figure 34 S-Plot of Drip Samples Classified by Processing Plant of Origin, Coloured by VIP value, 

Loadings which combined High Covariance with High Correlation are Shown Circled in Red ... 65 

Figure 35 PCA Scores Plot of Meat Coloured by Processing Plant of Origin ............................... 66 

Figure 36 OPLS-DA Scores Plot of Meat Samples Classified by Processing Plant of Origin ........ 67 

Figure 37 OPLS-DA 1D Full Loadings Plot of Meat Samples Classified by Processing Plant of 

Origin, Loadings Coloured by VIP Value ...................................................................................... 67 

Figure 38 OPLS-DA 1D Loadings Plot of Meat Samples Classified by Processing Plant of Origin, 

Loadings Coloured by VIP Value, Loadings with VIP values greater than 1.5 shown ................. 68 

Figure 39 S-Plot of Meat Samples Classified by Processing Plant of Origin, Coloured by VIP 

value, Loadings Combining High Covariance with High Correlation Shown Circled in Red ........ 69 

Figure 40 PCA Scores Plot of Drip and Meat Coloured by Processing Plant Origin .................... 69 

Figure 41 OPLS-DA Scores Plot of Drip and Meat Classified by Processing Plant of Origin ........ 70 

Figure 42 OPLS-DA 1D Full Loadings Plot of Drip and Meat Classified by Processing Plant of 

Origin, Loadings Coloured by VIP Value ...................................................................................... 71 

Figure 43 OPLS-DA 1D Loadings Plot of Drip and Meat Classified by Processing Plant of Origin, 

Loadings Coloured by VIP Value, Loadings with VIP values greater than 1.5 shown ................. 71 

Figure 44 S-Plot of Drip and Meat Samples Classified by Processing Plant of Origin, Coloured by 

VIP value, Loadings Combining High Covariance with High Correlation Shown Circled in Red .. 72 

Figure 45 PCA Scores Plot of Drip and Meat Coloured by Sample Type ..................................... 73 

Figure 46 OPLS-DA Scores Plot of Drip and Meat Classified by Sample Type ............................. 74 

Figure 47 OPLS-DA 1D Full Loadings Plot of Drip and Meat Classified by Sample Type, Loadings 

Coloured by VIP Value................................................................................................................. 74 

Figure 48 OPLS-DA 1D Loadings Plot of Drip and Meat Classified by Sample Type, Loadings 

Coloured by VIP Value, Loadings with VIP Value Greater than 1.5 Shown ................................ 75 

Figure 49 S-Plot of Drip and Meat Samples Classified by Sample Type, Coloured by VIP value, 

Loadings Combining High Covariance with High Correlation Shown Circled in Red .................. 76 

Figure 50 PLS-DA Scores Plot of Drip Classified by Odour Status ............................................... 87 

Figure 51 PLS-DA Scores Plot of Drip Samples Classified by Processing Plant ............................ 88 

Figure 52 Permutation Test Plot Showing 999 Permutations for PLS-DA model of Drip Samples 

Classified by Processing Plant ..................................................................................................... 88 

Figure 53 PLS-DA Loadings Plot of Drip Samples Classified by Processing Plant ........................ 89 

Figure 54 PLS-DA Loadings Plot of Meat Samples Classified by Processing Plant of Origin ....... 90 



   
 

ix 
 

Figure 55 Permutation Test Plot Showing 999 Permutations for PLS-DA model of Meat Samples 

Classified by Processing Plant ..................................................................................................... 90 

Figure 56 PLS-DA Loadings Plot of Meat Samples Classified by Processing Plant ...................... 91 

Figure 57 PLS-DA Scores Plot of Drip and Meat Classified by Odour Status ............................... 91 

Figure 58 PLS-DA Scores Plot of Drip and Meat Classified by Sample Type ............................... 92 

Figure 59 Permutation Test Plot Showing 999 Permutations for PLS-DA model of Drip and Meat 

Samples Classified by Sample Type............................................................................................. 93 

Figure 60 PLS-DA Loadings Plot of Drip and Meat Classified by Sample Type ............................ 93 

Figure 61 PLS-DA Scores Plot of Drip and Meat Classified by Processing Plant of Origin ........... 94 

Figure 62 Permutation Test Plot Showing 999 Permutations for PLS-DA model of Drip and Meat 

Samples Classified by Processing Plant ....................................................................................... 94 

Figure 63 PLS-DA Loadings Plot of Drip and Meat Classified by Processing Plant of Origin ....... 95 

 

Table 1 Method Test Treatment Conditions ............................................................................... 30 

Table 2 Storage Conditions of Meat Samples and Subsequent Odour Types ............................ 33 

Table 3 Summary of the Result of T-Test Analysis on Drip Samples Classed by Odour Status ... 55 

Table 4 Summary of Metabolites Characterised from 1H and 13C NMR Signals for Reference .. 77 

  



   
 

x 
 

Glossary of Abbreviations 

NMR Experiments 

HSQC ........................................................................... Heteronuclear Single Quantum Coherence 

NOESY  ............................................................................. Nuclear Overhauser Effect Spectroscopy 

TOCSY ............................................................................................. Total Correlation Spectroscopy 

CPMG ..................................................................................................... Carr Purcell Meiboom Gill 

 

Statistical Analysis 

PCA ................................................................................................. Principal Components Analysis 

PLS-DA ...................... Partial Least Squares Projection to Latent Structures-Discriminant Analysis 

OPLS-DA .................................................. Orthogonal Partial Least Squares-Discriminant Analysis 

UV ................................................................................................................ Unit Variance (Scaling) 

VIP ............................................................................................. Variable Importance in Projection 

 

Meat Science 

MAP ............................................................................................. Modified Atmosphere Packaging 

LAB ................................................................................................................... Lactic Acid Bacteria 

GN ........................................................................................................................... Gram Negative 

GP .............................................................................................................................. Gram Positive 



   
 

1 
 

Chapter One Introduction 

 Export of NZ Meat 1.1.
Meat is New Zealand’s second highest value export after dairy products (NZ Trade and 

Enterprise), and New Zealand is the world’s largest exporter of mutton and lamb meat (Meat 

and Livestock Australia). With more than 340 million kilograms of lamb meat shipped to 

around 110 countries every year (The Lamb Company), and ensuring the quality of meat when 

it reaches its export destination is incredibly important. By volume the European Union is the 

largest importer of New Zealand meat (Beef and Lamb NZ, 2015), and so most exported meat 

travels for about forty to fifty days to reach market (Richie, 2014). Taking into account the time 

required for the product to be processed before shipment, move through customs and be 

distributed the remaining shelf life is around two weeks and any delays reduce this time (The 

Meat Industry Association, 2011).  

Recently, the time taken to deliver meat from producer to market has increased as fuel prices 

and environmental pressures result in exporters reducing the speed of the ships transporting 

the product by up to 20% (Mills, Donnison, & Brightwell, 2014), a measure termed “slow 

steaming” (Psaraftis & Kontovas, 2013). There is a non-linear relationship between ship speed 

and emissions; not only is it more environmentally friendly to reduce ship speed but also 

financially worthwhile (Psaraftis & Kontovas, 2013). Therefore the life of chilled meats is being 

pushed beyond the recognized, and achievable, time of sixty to seventy days (Mills et al., 

2014). Meat is usually stored at temperatures of -1.5-0°C (Johnson, 1991) during export, with 

the minimum value set to stop the meat from freezing.  

A positive side effect of longer storage times for exported red meat is that the prolonged aging 

period improves consumer reported scores for attributes such as tenderness and flavour 

(Graham et al., 2010), both of which increase with aging. This has been linked to the 

proteolysis of myofibrillar proteins in the meat (Graham et al., 2010). Higher levels of amino 

acids, nucleotides and sugars are found in aged meat and these help to impart flavour in the 

final cooked product. 

Another adjustment to the procedures surrounding the export of NZ meat relates to the 

changing practices of retailers. Previously it was common for retailers to repackage 

transported primal meat cuts into retail-ready portions at in-house butcheries, however many 

now choose to do away with the additional space and labour this requires in favour of the 



   
 

2 
 

more convenient and economical option of purchasing retail-ready cuts directly from the 

supplier (Jeremiah & Gibson, 2001). 

1.1.1. Meat Spoilage 

Meat spoilage is linked to the growth of certain classes of bacteria on the surface of meat. 

These bacteria can contaminate the meat at any stage from the farm to the final packaging. 

Tearing of packaging can also allow bacterial contamination. Inhibiting the growth of these 

bacteria is the goal of storage temperature and packaging systems for exported meat. Removal 

of oxygen has one of the biggest effects on the rate of bacterial growth (Gill, 1989); therefore 

packaging is designed to achieve this. Meat will also show discolouration due to oxidation of 

muscle pigment in the presence of oxygen (Gill, 1989), and this is another characteristic that 

results in rejection of product by the consumer. 

1.1.1.1. Meat Spoilage Characteristics 

Certain groups of bacteria are associated with the production of off odours and spoilage in 

meat upon reaching sufficient concentration. Precursors for the metabolites implicated in 

spoilage are glucose, lactic acid, water soluble proteins and amino acids. As the quantity of 

these precursors increases the rate and extent of spoilage also increases (Ercolini et al., 2011). 

1.1.1.2. Spoilage Bacteria 

Meat which has been processed and stored in adherence with industry standard best practice 

(Devine, 2014) will nonetheless contain a wide range of bacterial species. Which of these 

species is able to predominate is dependent on the meat environment; this is in turn 

determined by factors such as temperature, atmosphere and pH (Davies, Board, & Board, 

1998). Strong off-odours are largely caused by gram negative (GN) bacteria: these bacteria are 

fast growing and proliferate in aerobic environments. Gram positive (GP) bacteria, such as 

many strains of lactic acid bacteria (LAB) and Micrococcaceae, are slower growing than GN 

bacteria but are able to grow in both aerobic and anaerobic conditions, although they 

generally form only a small proportion of the overall meat bacterial ecosystem in the presence 

of oxygen. LAB species are also tolerant of low temperatures (Davies et al., 1998). Therefore 

refrigeration (around -1.5-1 degree C) and vacuum or modified atmosphere packaging (MAP) 

of meat will cause LAB to thrive relative to GN bacteria. However, LAB will still grow slowly 

under these conditions.  

LAB utilise carbohydrate fermentation for energy production to produce a range of organic 

acids which produce sour or acid smells (Davies et al., 1998). The acid production lowers the 

pH of the meat surface and, in conjunction with other ‘antimicrobial’ metabolites produced by 



   
 

3 
 

LAB; this has the effect of creating an inhospitable environment for pathogenic, putrefactive 

and toxinogenic bacteria(Davies et al., 1998). LAB also competes for resources with these 

undesirable bacteria. 

Both types of bacteria produce malodorous substances, for example LAB produce small 

amounts of dimethlysulphide and methanethiol in addition to organic acids which contribute 

to the distinctive smell that accompanies these colonies(Davies et al., 1998). As part of their 

metabolism pathogenic GN bacteria produce sulphurous compounds which have an offensive 

smell; therefore the odours of meats with greater proportions of LAB present will be less 

offensive than meat where GN bacteria proliferate(Davies et al., 1998). This means that meat 

on which LAB is the dominant bacterial species will have a longer shelf life than other meat, 

therefore meat environments are generally manipulated to provide conditions in which GP 

bacteria will grow in preference to GN bacteria(Davies et al., 1998). 

However, once bacterial counts reach around 107 to 108 cells/cm2 on the meat surface glucose 

is generally exhausted meaning that bacteria turn to amino acids for metabolism, breaking 

these down in to foul smelling sulphides, esters and amines(Davies et al., 1998). 

1.1.2. Acceptability Characteristics in Meat 

For the majority of consumers the acceptability of meat is mostly based on visual and olfactory 

cues, since information on its age or quality is normally not provided. The appearance of 

'freshness' is very important and can be suggested by colour (deep red for beef and lamb) or 

state (frozen is suggestive of long transport times and old meat). Meat is expected to have 

little smell, as a strong odour is most often associated with spoilage.  

From a manufacturing perspective freezing meat for transport is preferable to chilling due to 

the increased shelf life of frozen meat. As a consequence, frozen meat can be shipped slower 

which reduces costs. For chilled meat some storage time is preferable as it increases the 

flavour of the meat and results in a more tender product (Graham et al., 2010). Chilling also 

increases the water holding content of the meat compared with freezing/thawing, meaning 

that the same quantity of meat will weigh more and can therefore be sold for a higher price.  

1.1.3. Confinement Odour and Sensory Evaluation 

Confinement odour (CO) is a phenomenon that has always been an issue with vacuum and 

modified atmosphere packaged meat. Meat that has been stored in these ways will sometimes 

produce a possibly unpleasant smell immediately upon the packaged being opened; however 

this doesn’t indicate that the meat is unfit for consumption as other odours might. As it is a 



   
 

4 
 

well-known phenomenon in the meat industry its detection does not result in meat being 

rejected by a retailer. However, changes to supply chain practices have resulted in the 

consumer being exposed to this phenomenon in place of the retailer. As consumer have not 

been educated about CO they are most likely to equate the smell to spoilage.  

Odour assessment of meat is often carried out by selected panels using a hedonic scale, in 

which the odour is categorised over a point system ranging from acceptable to unacceptable 

(Gill & Penney, 1986). In the assessment of CO, it is necessary to evaluate the meat 

immediately upon opening its packaging and again after a short wait period. Meat exhibiting 

CO can then be differentiated from spoiled meat as it will produce unacceptable odours only 

upon first opening the package, which will dissipate over the wait period and not be detectable 

at the second assessment. It may also be possible to distinguish CO from spoilage odour by the 

notes each produces. CO has been described as cheesy, dairy or milky with additional sour or 

acid notes (Johnson, 1991). In comparison spoilage odours are generally referred to as putrid, 

sweet or sulfurish (Gill & Penney, 1986; Seman, Drew, Clarken, & Littlejohn, 1988). 

Hydrogen sulfide is the cause of sulfur notes associated with spoilage odours; it can be 

produced by particular strains of Lactobacillus sakei (a species of LAB) (Mills et al., 2014). 

Sulfur notes are also linked to the depletion of glucose levels in meat(McMullen & Stiles, 

1994). 

It has been suggested that LAB proliferation in meat is linked to confinement odour and that 

the specific odour produced is dependent on the particular type of LAB present (Mills et al., 

2014).  

1.1.4. Packaging and Storage for Improved Meat Quality 

Reduction of bacterial growth in meat packaging is often achieved in one of two ways – 

vacuum packaging and MAP. Vacuum packaging involves the complete removal of air from the 

product and effective sealing of the package. MAP involves the complete removal of air from 

the product followed by replacement with an alternative gas, or mixture of gases. A common 

MAP technique uses the addition of carbon dioxide (Gill, 1989), a gas which inhibits microbial 

growth (Bill B.A.; Small, Sikes, & Doral, 2008) through slowed respiration and increased lag 

phase/generational times. For both types of packaging it is important that the packaging film is 

strong, to reduce the chance of breakage, and has a very low permeability, to reduce the 

transmission of oxygen back into the meat during storage (Gill, 1989) as well as reducing the 

transmission of carbon dioxide out of the film. Trace amounts of gases remain in vacuum 

packaged products; however tissue and microbial respiration will reduce any oxygen to less 



   
 

5 
 

than 1% and increase carbon dioxide to around 20% (Davies et al., 1998). The rest of the 

modified atmosphere consists of nitrogen gas. 

Storing meat at temperatures low enough to freeze (<-1.5°C) significantly extends the 

acceptable storage time when compared to meat that is only chilled (-1.5°C-4°C) (Small et al., 

2008). However, chilled meat displays a number of beneficial characteristics that are less 

evident in meat which has been frozen. For example the water holding capacity (WHC) is 

higher in chilled meat, which results in less water loss during storage and increases the 

tenderness of the final product (Trout, 1988). Tender meat is more desirable to the consumer. 

A high WHC also means that, for the same pre-storage weight, chilled meat can be sold for a 

higher price than frozen and thawed meat as the water that stays in the muscle tissue is 

included in the weight of the final product sold to the consumer. This makes chilled meat a 

more desirable product for the producer. Spoilage bacteria are still able to grow at 

temperatures slightly below -1.5°C; therefore chilling meat can only extend the shelf life of 

meat (Davies et al., 1998). 

 Metabolomics 1.2.
Metabolomics is concerned with the investigation of all the low molecular weight compounds 

present within a cell or organism as a result of metabolism or ingestion (Dixon et al., 2006), 

and more particularly with the changes in metabolite concentrations that occur due to a 

perturbation to the genetic or environmental conditions of the organism. These compounds 

are called metabolites and are the starting materials, intermediates and products of cellular 

metabolism. They include a number of classes of compounds such as amino acids, sugars, 

lipids and organic acids. Metabolites are affected not just by gene expression but also by the 

interaction of the environment with the genome, transcriptome and proteome. Figure 1 also 

shows that while signalling is generally thought of as occurring from genome to transcriptome 

to proteome it is possible for it to go in the opposite direction and from any of these to any 

other. The entire collection of an organism’s metabolites is called its metabolome. 



   
 

6 
 

 

Figure 1 Interaction of the Metabolome (Griffin & Shockcor, 2004) 

The goal of metabolomics can generally be considered to be the investigation of the 

metabolome to determine the metabolic response to various forms of stimulus (for example 

disease, diet intervention or drug exposure) (Lindon & Nicholson, 2008). Traditional statistical 

approaches follow the changes to a single metabolite concentration when the system is 

perturbed, for example monitoring the level of lactate in human sweat before and after an 

exercise intervention. The metabolomics approach is to follow the changes that occur to the 

entire metabolome when the system is perturbed. This means that there is no need to narrow 

down the field of possible effects on the system in order to perform an analysis, therefore 

unexpected changes can be discovered. All of the metabolite concentration information which 

a particular analysis technique can provide is able to be analysed together, which means that a 

holistic view of the metabolic response is obtained. 



   
 

7 
 

In animal systems the metabolome is made more complex due to the presence of symbiotic 

microorganisms or parasites which produce their own metabolites and interact with the host’s 

metabolism (van der Greef & Smilde, 2005). 

There are two main types of metabolomic analysis. The first is metabolic profiling, in which the 

biological sample is analysed and as many of the metabolite signals as possible identified. In 

targeted metabolic profiling these metabolites are quantified using internal standards at 

known concentrations (Savorani, Rasmussen, Mikkelsen, & Engelsen, 2013). The second of 

these techniques is metabolic fingerprinting, in which the analysis of the metabolome is 

coupled with multivariate analysis techniques in order to compare the metabolic patterns of 

different classes of samples. The metabolites are not initially identified; instead the signals 

they produce due to the particular analytical technique used are all used to provide a 

‘fingerprint’ of the samples’ metabolomes. The mean metabolome is computed for each class 

and these means are compared to produce a list of metabolites which describe the 

perturbation produced by the stimulus (Broadhurst & Kell, 2006). The metabolites are used to 

discover the biological mechanisms activated by the stimulus (Savorani et al., 2013). 

1.2.1. Analysis Methods 

Two analysis platforms are used for the majority of metabolomics studies – Nuclear Magnetic 

Resonance (NMR) spectroscopy and Mass Spectrometry (MS). Both platforms offer a large 

number of protocols which have their own strengths and weaknesses. NMR can be applied 

both to solid and liquid state samples, many different elements can be probed to produce 

numerous one-dimensional and two-dimensional experiments. MS is coupled to many 

separation and infusion methods. These combinations include gas chromatography MS (GC-

MS), liquid chromatography MS (LC-MS), two dimensional GC time-of-flight MS (2D-GC-TOF-

MS), high performance LC-MS (HPLC-MS), direct infusion electrospray ionization TOF-MS (DI-

ESITOF-MS), capillary LC-ESI-TOF-MS, ESI-MS, capillary electrophoresis ESI-MS and Fourier 

transform ion cyclotron resonance MS (FTI-CR-MS) (Rochfort, 2005Rochfort, 2005).  

Less commonly utilised platforms are Fourier transform infrared spectroscopy (FTIR), high 

performance liquid chromatography-ultraviolet (HPLC-UV) and fluorescence microscopy 

(Rochfort, 2005Rochfort, 2005).  

NMR and MS differ in a number of fundamental ways (Lindon & Nicholson, 2008Lindon & 

Nicholson, 2008) but are both useful for metabolomics applications as they are considered 

complementary techniques (Rochfort, 2005metabolome, and therefore many platforms are 

necessary for application to metabolomic studies.). The strengths of one platform are related 



   
 

8 
 

to the weaknesses of the other, and vice-versa.  There is currently no technique which is able 

to measure the entire metabolome, and therefore many platforms are necessary for 

application to metabolomic studies. 

Because of the techniques’ specific limitations and strengths it is necessary for the analysis 

method to be chosen based on the established needs of the study. A one size fits all 

recommendation should not be made, with some papers (Ercolini et al., 2011; Rochfort, 2005) 

reporting the use of both analysis methods on the same sample sets to take advantage of the 

benefits of both platforms.(Ercolini et al., 2011; Rochfort, 2005) reporting the use of both 

analysis methods on the same sample sets to take advantage of the benefits of both platforms. 

NMR is generally only sensitive enough to pick up metabolites at micro-molar quantities, 

although improvements to the detection limit down to nano-molar can be made if the 

spectrometer is fitted with a cryoprobe. The techniques generally require a sample size of a 

few hundred microlitres or milligrams (with the use of a microprobe only tens of microlitres 

are required) which for certain studies can be very difficult or time consuming to obtain 

(Young, Berdagué, Viallon, Rousset-Akrim, & Theriez, 1997Young, Berdagué, Viallon, Rousset-

Akrim, & Theriez, 1997). 

However, NMR does have a number of characteristics which are advantageous when applied 

to metabolomics. It is a non-destructive technique (Rochfort, 2005Rochfort, 2005)(Worley, 

2013). This can be important for metabolomics experiments where previous biological 

knowledge of the samples can lead to false (but seemingly reliable) interpretations of the data 

(Worley, 2013). The technique provides high analytical reproducibility, which removes the 

need for machine replicates (Rochfort, 2005). This robustness also means that comparison of 

sample data taken in different labs using different machines is possible (Rochfort, 2005). NMR 

also allows for fairly easy identification of metabolites through the use of online databases 

containing spectra of pure metabolites, in addition to 2D experiments as described elsewhere.  

which means the same sample can be reanalysed at a later date to investigate stability. 

Minimal sample preparation is necessary for NMR analysis, and many different classes of 

compounds can be analysed together without the need for separation techniques. It allows 

many biofluids, cell lysates and tissues (whether intact or not) to be analysed without bias for 

particular classes of compounds. This can be important for metabolomics experiments where 

previous biological knowledge of the samples can lead to false (but seemingly reliable) 

interpretations of the data. The technique provides high analytical reproducibility, which 

removes the need for machine replicates.. This robustness also means that comparison of 



   
 

9 
 

sample data taken in different labs using different machines is possible. NMR also allows for 

fairly easy identification of metabolites through the use of online databases containing spectra 

of pure metabolites, in addition to 2D experiments as described elsewhere.  

It should be noted that analysis can be performed using many types of animal tissues and 

fluids including muscle (Graham et al., 2010; Mannina et al., 2008; Savorani et al., 2010), 

organs, adipose tissue (Atherton et al., 2006), hemolymph, glands (Viant, Rosenblum, & 

Tjeerdema, 2003), cerebrospinal fluid (CSF), blood, urine (Osorio, Moloney, Brennan, & 

Monahan, 2012) and freeze exudate (Straadt, Aaslyng, & Bertram, 2011).(Graham et al., 2010; 

Mannina et al., 2008; Savorani et al., 2010), organs, adipose tissue , hemolymph, glands , 

cerebrospinal fluid (CSF), blood, urine  and freeze exudate. 

 

1.2.2. Meat Metabolomics 

Metabolomic meat analysis has been carried out in a diverse range of animal species for an 

equally diverse range of applications. 

Beef and pork have been frequently investigated, providing the ability to authenticate breed 

(Straadt et al., 2011), production system (Osorio et al., 2012) and storage length (Graham et 

al., 2010); this being a common source of research for food and nutrition science on products 

of both animal and plant origin.  Food authenticity is a hugely important issue for many 

countries globally as they seek to maintain premiums on food and drink produced in certain 

geographical locations, using specific techniques, or made with particular constituents.  

Discrimination of meat based on the breed of the animal which it came from is not only useful 

for food authenticity analysis but also investigating the effect of breed on meat quality 

characteristics. This was the objective of a study by (Straadt et al., 2011) in which the NMR and 

metabolomics were used to probe biophysical and biochemical factors of pork quality. NMR T2 

relaxation was used to discover the water holding content (WHC) of five pig cross breeds, in 

order to determine which breed produced the most tender meat, and this result was 

combined with scores from sensory analysis. 1H NMR spectroscopy was also used to discover 

the metabolic profile differences between the breeds; it was found that a number of amino 

acids, choline-containing compounds, lactic acid, inosine, inosine-monophosphate (IMP) and 

glycerol differed significantly between the breeds.  

Metabolomics has been utilised for investigating the possibility of discriminating meat samples 

based on production system and type of feed for a number of different meat types. Both of 



   
 

10 
 

these factors can influence meat quality and as such consumers are willing to pay different 

prices based on the treatment of the animal before slaughter. For example, meat from animals 

which have been raised outdoors and fed on pasture commands a high price as consumers 

become more concerned about animal welfare and are made aware of its nutritional 

superiority over more intensively produced meat.  NMR metabolomics has been used to 

discriminate between beef from pasture-fed cows which had been reared outdoors and barley 

concentrate fed cows which had been confined indoors (Osorio et al., 2012). The ability to 

discriminate meat based on the type of feed used in its production not only prevents 

producers from being able to mislead consumers but can also detect illegal practices. One such 

practice is adding chicken bone meal to chicken feed for a cheap source of nutrients; it is illegal 

in many cases as an animal cannot have feed adulterated with processed animal protein from 

its own species. Direct Analysis in Real Time Ionization-Mass Spectrometry (DART-MS) used in 

conjunction with chemometric methods is able to differentiate between the meat of chickens 

which have been fed with chicken bone meal and the meat of those which have been fed the 

same diet, under the same conditions, without the addition of bone meal (Cajka, Danhelova, 

Zachariasova, Riddellova, & Hajslova, 2013). The analysis showed that while the polar meat 

metabolites differed corresponding to the feed type the non-polar meat metabolites were 

more influenced by seasonal differences. 

A number of food products have a Protected Designation of Origin (PDO) to indicate they have 

been produced from a certain geographical location with a reputation for quality. Proton 

transfer reaction time-of-flight mass spectrometry (PTR-ToF-MS) has been used in conjunction 

with multivariate analysis techniques in order to discriminate the headspace gas from dry-

cured ham from PDOs from dry-cured hams from other regions (del Pulgar et al., 2011). In 

order for producers in these regions to profit from their PDOs it is important that imposters 

are easily discriminated and the illegal practice punished.  

A number of studies have focussed on the discrimination of meat based on post-slaughter 

treatment. For example, NMR metabolomics has been applied to ground beef to determine 

whether the meat has been irradiated (Zanardi et al., 2015), a treatment which extends the 

shelf-life of the meat but is not legal in a number of countries. Principal Components Analysis 

(PCA) and Classification Trees were applied to the data and revealed that glycerol, lactic acid 

esters and tyramine (or a p-substituted phenolic compound) were discriminating metabolites 

between irradiated and non-irradiated beef. Another example shows that the chilling process 

applied to lamb meat immediately after slaughter can affect the tenderness of the meat by an 



   
 

11 
 

altered energy metabolism (Warner et al., 2015). This change can be assessed by a number of 

NMR and MS-based methods which reveal the differences between the meat metabolites 

based on the chilling process. 

It has been shown that the age of duck meat can be ascertained from inspecting the 

metabolite profiles of equivalent reference samples (Liu, Pan, Ye, & Cao, 2013). 

It is possible to monitor the microbial populations in meat over its storage time. This has been 

performed by Ercolini et al. who compared this information with metabolomic analysis in 

order to assess the effects of various forms of packaging conditions on beef (Ercolini et al., 

2011). They were able to follow the changes in the microbial population as the storage 

progressed and showed that the storage conditions had a great effect on which species 

predominated and the metabolites which were produced. 

1.2.3. Previous Comparison of Drip and Meat 

A previous study sought to determine the relative merits of analysing meat or drip 1H NMR 

spectra to differentiate the metabolite profiles of various pork producing pig breeds. Based on 

multivariate statistical analysis it appears that analysis of the drip was able to yield a better 

grouping of samples based on breed. The authors suggest that this may be the result of many 

metabolites important to the discrimination being lost from the meat in the form of drip 

(Straadt et al., 2011). 

1.2.4. Experimental Design Considerations 

For a robust analytical method (such as NMR spectroscopy) generally only one analytical 

replicate per sample is necessary (Savorani et al., 2013).  

It is necessary to minimise variation introduced to the samples during preparation and 

analysis. Any variation unrelated to wanted classification can confound the efforts of the 

statistical analysis to discriminate the samples into groups. Examples of unwanted sample 

variation include differences in collection, storage, temperature, pH, post-collection 

metabolism and the presence of precipitate (Savorani et al., 2013). Sample preparation and 

analysis is an important area to consider in minimising variation. For instance, if different 

sample classes are prepared separately it would be impossible to discriminate whether any 

variation found was due to inherent differences between the classes or if unwanted variation 

had been introduced during preparation. Problems of this kind can be mitigated by 

standardising all aspects of preparation and analysis, for example the order/time of collection, 

preparation, transport and spectral acquisition (Liland, 2011). In addition, samples are 



   
 

12 
 

normally randomised during preparation and analysis to avoid run order and/or batch effects. 

This allows the analysis to be robust, reproducible and easily interpretable (Trygg, Holmes, & 

Lundstedt, 2006). 

 NMR 1.3.

1.3.1. Underlying Principle 

Certain isotopes, specifically those with an odd number of protons and/or neutrons, have a 

non-zero spin (an intrinsic quantum property of nucleons) which allows for their spin systems 

to be perturbed by an external magnetic field, these are said to be magnetically susceptible. 

Examples of such isotopes include 1H, 2H, 13C, 19F and 31P.  This perturbation comes in the form 

of splitting of the spin energy levels, proportional to the size of the applied field, within the 

nucleus of the atoms (Figure 2). This splitting behaviour creates a small magnetic field. The 

nuclei can absorb and re-emit applied radio frequency pulses which match the frequency of 

the energy transition between the two spin states. The frequency of the energy transition 

depends on the size of the effective magnetic field at the nucleus; electron density shields the 

nucleus from the external magnetic field thereby lowering the size of the effective field. If the 

effective magnetic field at the nucleus is lowered then the spin system will be split less, 

meaning that the matching radio frequency will be less. The difference between the amount of 

splitting that is expected from the applied magnetic field, and the amount of splitting that is 

observed through the emitted radio frequency is termed the chemical shift. The chemical shift 

is highly dependent on the chemical environment of the nucleus and is therefore useful in 

molecular structure elucidation (Savorani et al., 2013). Nuclei in many common compounds 

have known chemical shift values and this information can be found in many online databases. 

A large number of metabolites are catalogued this way, and therefore analysis of a sample by 

NMR can show the metabolite components.  

 

Figure 2 Splitting of Nuclei Spin States in an External Magnetic Field ("Nuclear Magnetic Resonance,") 



   
 

13 
 

Additional structural information is contained in the NMR spectrum, for example the 

multiplicity of a peak in a 1D spectrum indicates the number of nuclei neighbouring the 

nucleus of interest, while the integrals of all the individual peaks for a given metabolite 

indicate the proportions of nuclei corresponding to those signals. In Figure 3 the spectrum 

corresponding to a mixture of α-glucose and β-glucose shows the various peaks which are 

produced by the molecules and their different multiplicities can clearly be observed. The 

various anomeric signals have been annotated. 

 

Figure 3 1H NMR Spectrum of a Mixture of α-glucose and β-glucose in D2O Acquired at 400MHz 

Many types of biological samples can be analysed by NMR, for example cell lysates, whole 

tissue, tissue extracts and biofluids such as blood, urine and cerebrospinal fluid. The most 

common technique for metabolomics is 1D 1H NMR spectroscopy which investigates liquid 

samples. However, other elements (such as 13C and 31P) can also be targeted, and these can be 

combined with 1H in order to enable 2D exploration (Rochfort, 2005). These elements are 

generally less abundant than 1H and their ability to produce quantitative information is limited. 

2D 1H-13C experiments can be employed to improve metabolite identification as the two 

dimensions reduce the risk of peaks from different metabolites overlapping (Rochfort, 2005). 

Intact tissues and cells can also be used if high resolution magic angle spinning (HRMAS) NMR 

is available (Straadt et al., 2011).  

1.3.2. Experimental Variations 

High resolution magic angle spinning (HR-MAS) NMR spectroscopy is a technique which allows 

for the analysis of solid samples such as intact muscle tissue (Lindon & Nicholson, 2008). The 

sample must be rapidly spun (around four to six kHz) at an angle of 54.7 degrees in relation to 

the applied magnetic field, to mitigate the line broadening effect otherwise seen in solid 

samples. This effect is not seen in liquids as, unlike in solids, molecules are able to tumble 

freely in solution which averages out anisotropic NMR parameters.  



   
 

14 
 

Recent experiments using HR-MAS NMR spectroscopy (Savorani et al., 2013) have shown that 

solid tissue samples can be directly analysed with only the addition of D2O to the sample. This 

incredibly simple preparation is ideal for metabolomics as the sample is analysed in as natural 

a state as possible. Many major signals can be identified; these include lactic acid, creatine, 

fatty acids, amino acids, organic acids and nucleosides. The information obtained from this 

analysis has been used, in conjunction with stable ratio isotope analysis and multivariate 

analysis, to discriminate the geographic origin of meat with a 96% predictive ability (Sacco, 

Brescia, Buccolieri, & Caputi Jambrenghi, 2005). The robustness and reliability of this method 

suggest that HR-MAS NMR is a very useful technique for analysis of meat metabolite 

characteristics. However, logistically it is difficult to analyse the large number of samples that 

is required for metabolomic analysis without a sample changer.  

1.3.3. Benefits for Use with Metabolomics 

Simple sample preparation is obviously very important for metabolomics studies involving 

multivariate analysis as these require high numbers of samples for statistical robustness. In 

order to reduce preparation induced variability it is important to complete preparation quickly 

and efficiently. There is also very little variation introduced by NMR analysis due to its 

reproducibility and uniformity (Dixon et al., 2006). 

1.3.4. Experimental Considerations 

Proteins are generally not included in metabolomic analysis and there are a number of reasons 

why physical protein removal is done before NMR analysis. For instance many low molecular 

weight metabolites will remain bound to the proteins present and will therefore not be 

detected in the spectra; this hinders the ability to quantify the concentrations of these species. 

High molecular weight molecules can also distort the spectral baseline, protein-bound 

metabolites can exchange with free metabolites over the period of acquisition resulting in 

broadened peaks, and a high concentration of protein will lower transverse relaxation (T2) 

times, all phenomena leading to a reduction in quantitative accuracy. Protein can be removed 

from a biological sample in a number of ways each with varying levels of efficacy and 

appropriateness depending on the sample type, for example a chemical method involves 

precipitating proteins out of solution using a solvent, while a physical method for removal 

involves filtration of the sample.  

In Figure 4 a spectrum of drip which includes protein (blue) overlays a spectrum of the same 

drip sample where the protein has been physically removed by ultrafiltration (red). The sharp 

low molecular weight metabolite peaks can be seen to be partially or completely obscured by 



   
 

15 
 

the broad resonances of the protein. Not only is quantification of the metabolites impossible 

from the protein-containing spectrum, but some metabolites are so obscured as to not be 

identifiable. For analysis which would seek to monitor the concentration differences of these 

metabolites based on NMR spectroscopy the protein-containing samples would be useless. 

 

Figure 4 Overlay of 
1
H NMR Spectra of Filtered Plasma and Unfiltered Plasma 

In addition to the above methods the Carr Purcell Meiboom Gill (CPMG) pulse sequence is 

often used for analysis of samples containing protein where it is undesirable to remove the 

protein prior to spectral analysis. For metabolomics this approach is often not appropriate due 

to its varied effects on both quantitative and qualitative investigation (Savorani et al., 2013).  

 Multivariate Statistical Analysis 1.4.

Multivariate statistical analysis describes the analysis of more than one statistical variable at a 

time (Liland, 2011) to obtain information about the differences between groups (classes) of 

samples, which means that it is particularly useful for applying to metabolomics research. 

Metabolomics experiments produce a very large number of variables (in the case of NMR, bin 

integrals or metabolite concentrations) compared to the small number of samples. Each of 

these variables can be considered as representing a dimension in the metabolic space, and 

therefore an entire spectrum can be thought of as a point in a multidimensional metabolic 



   
 

16 
 

hyperspace (Lindon & Nicholson, 2008). A lot of these variables do not relate to the class 

differences in the experiment and can interfere with the analysis. This produces the need for 

methods of dimension reduction in order to expose the information most relevant to the 

experiment. 

Dimensionality reduction can be achieved by combining the variables from input data space 

(i.e. the chemical shift bins) with a linear transformation to lower dimensional output space 

(i.e. scores/principal components) (Worley, 2013).  

There are two broad categories of model-based methods. Unsupervised methods (such as 

principal components analysis, hierarchical cluster analysis and nearest-neighbour clustering) 

require no previous knowledge of the sample characteristics and therefore don't incorporate a 

hypothesis bias. Supervised methods (for example partial least squares and orthogonal partial 

least squares analysis) require prior knowledge of expected sample group variation (i.e. class 

membership) (Worley, 2013). 

A commonly used statistical analysis program for metabolomics is SIMCA 13 (Umetrics; Umeå, 

Sweden). 

1.4.1. Data Pre-Processing for NMR and Multivariate Statistical Analysis 

The following techniques are used to transform the raw data before statistical analysis in order 

to create fewer data points, reduce the impact of experimental variation and make the spectra 

more comparable. This vastly improves the ease of model interpretation. 

1.4.1.1. Exclusions 

It may not be suitable to use the entire NMR spectrum for analysis. There are certain regions 

which may have to be excluded prior to the steps described below. One region with is often 

excluded from all spectra is the region around the water peak. This peak varies in intensity 

primarily due to acquisition parameters, not the sample properties. Unless accounted for 

through spectroscopic techniques, proteins will appear as large broad peaks. Often these 

molecules are not specifically investigated, as is the case in this work, and so are excluded so 

as not to have a confounding effect on the normalisation (described in 1.4.1.4). Because 

protein signals are especially broad it can be difficult to exclude them from spectra without 

also excluding areas which contain other metabolites, therefore it is best in this situation to 

physically remove proteins from samples before spectroscopy is performed. Additional 

exclusions may be made considering the properties of the samples and the needs of the study.  



   
 

17 
 

1.4.1.2. Alignment 

Peak positions can vary, by small or large amounts depending on the metabolite, due to a 

number of factors that can affect chemical environments. These factors include pH and 

temperature (the effect of pH differences can be mitigated somewhat with the addition of 

buffer to the samples before NMR analysis). Because of this, comparing unprocessed spectra 

against one another can result in position differences obscuring the intensity differences that 

are the object of metabolomic analysis. These position shifts must therefore be either 

compensated for (see 1.4.1.3 below) or reduced. Position shift reduction can be achieved by 

certain programs which incorporate peak alignment features (e.g. MestReNova Suite NMR 

Plugin v. 10.0.1). Alignment is often applied one of two ways; globally, whereby the entire 

spectrum is shifted by a set amount in order to optimise overall peak overlap, or locally, by 

which individual peaks are selected and shifted relative to the rest of the spectrum. Figure 6 

shows the effect of applying local alignment on a peak in overlaid NMR spectra; Figure 5 shows 

the same peak pre-alignment. 

 

Figure 5 Section of Overlaid Plasma 
1
H NMR Spectra from 7.71 - 7.95 ppm 



   
 

18 
 

 

Figure 6 Section of Overlaid Plasma 
1
H NMR Spectra from 7.71 - 7.95 ppm with Local Alignment 

1.4.1.3. Binning 

The procedure of binning, alternatively called bucketing, is used to reduce the dimensionality 

of spectra while also compensating for pH-induced chemical shift differences. Spectra are 

divided by chemical shift into sections, or “bins”, generally of equal width (although some 

sophisticated software programs can provide variable bin widths), the sum of the intensities of 

all the individual plot points within that section are then used to give a single variable for each 

bin. In this way a spectrum containing many thousands of variables (spectral points) may now 

be represented using only a few hundred (bins). Figure 8 shows the effect binning to 0.04 ppm 

has on the spectra in Figure 7. It can be seen that while the fine detail of the peak multiplicity 

has been lost the number of variables in the same section of a spectrum has decreased 

dramatically.  



   
 

19 
 

 

Figure 7 Section of Full Resolution Plasma 
1
H NMR Spectrum Showing 0.8 - 2.75 ppm 

 

Figure 8 Section of Plasma 
1
H NMR Spectrum Showing 0.8 - 2.75 ppm Binned to 0.04 ppm 

A bin width of 0.04 ppm is predominantly used for metabolomics studies. This value is a good 

compromise between too narrow bins, where the bin width is insufficient to contain 

misaligned peaks, and too wide bins, where many peaks may be contained within the same bin 

and therefore individual peak changes are masked. However, it is important to choose the 

optimal bin width with consideration to the particular needs of the study. If peaks are very well 



   
 

20 
 

aligned and not very wide then narrower bins would be appropriate, while spectra with broad 

and/or badly aligned peaks would require wider bins. 

1.4.1.4. Normalization 

Samples for metabolomics studies can, and often do, have differences in their total 

concentration. More concentrated samples will show higher peak intensities, possibly for all 

metabolites, than less concentrated samples. Relative changes in metabolites will then be 

obscured. It is therefore necessary to normalise the spectra in order to apply direct 

comparison. Normalisation is a process whereby every point in a spectrum is either multiplied 

or divided by a constant. The most common normalisation methods for metabolomics involve 

either dividing the intensity of each point by the total sum of intensities over the whole 

spectrum, by the intensity of the largest peak, or by the area from an internal standard. The 

third requires a reference chemical of known concentration, while the first assumes that the 

total concentration of metabolites is approximately constant. Normalisation to total spectral 

intensity is suitable for most data however very large peaks with highly variable intensities can 

confound this approach. 

1.4.1.5. Scaling 

The goal of multivariate analysis for metabolomics is, in part, to find metabolites with 

systematic quantity differences between two, or more, groups. These differences can be very 

large, and obvious to the eye, or very small, and difficult to find without the assistance of a 

statistical program. Even with a sophisticated analysis package it is possible for small, but 

significant, variation to go unnoticed amidst the larger variation of high abundance 

metabolites. Often this large variation is unrelated to the biological difference of interest. To 

mitigate this, a technique known as scaling is employed; this is a process whereby each 

variable (bin) is divided by a scaling factor which is different for each bin. Initially it is common 

to subtract the mean of the value of the bin across all samples from the corresponding bin in 

each spectrum which produces mean-centred data. A variety of scaling factors can then be 

applied; which factor is chosen depends on the data and the objectives of the study. The two 

most common scaling methods are described below. 

Unit variance (UV) scaling, also known as autoscaling, uses the standard deviation of each bin 

as the scaling factor; this allows metabolites with small abundance changes to be detected 

with equal weight as those which vary by a large amount. The principal disadvantage of this 

method is that it amplifies the influence of noise. 



   
 

21 
 

Pareto scaling is similar to UV scaling, but it uses the square root of the standard deviation of 

each bin as the scaling factor. This leaves the data closer to its original form than UV; it still 

increases the influence of low abundance metabolites without inflating noise.  

It is important to note that the type of data and multivariate analysis method used must be 

taken into consideration when selecting a scaling method. At times it can be necessary to 

employ no scaling method. 

1.4.2. Principal Components Analysis 

PCA is generally the first multivariate analysis method to be performed as it provides an 

unbiased overview of groupings, trends and outliers within the given data (Trygg et al., 2006).  

The process seeks to find the most important dimensions of variability in the data and present 

this information in plots which allow for simple interpretation. This is achieved by creating a 

multidimensional space where each variable represents a dimension and each spectrum is 

represented by a point in this space. The origin of the coordinate system is moved to the 

average data point and a new variable is created by finding the direction in the data space 

which describes the greatest variance. This variable is called the first principal component (PC). 

This step is then repeated, with the condition that each PC is orthogonal to each other PC, until 

the desired number of components are produced (or the desired level of variability is 

explained). The requirement for orthogonality means that the components are uncorrelated. 

Each component may be described as explaining a certain percentage of the variability. For the 

most part the correlation between the variables is strong enough that only a few components 

are needed to be calculated to account for the vast majority of the variance. 

Two plots are produced to represent the output of PCA.  

The scores plot (an example of which is shown in Figure 9) shows the projection of the data 

onto the new coordinate system, the axes represent the PCs and each spectrum is represented 

on the plot as a single point. This shows the relationships between the samples (sometimes 

called objects), if the explained variance is sufficiently high then samples which are close 

together in this plot share similar properties, and those which are far apart have very different 

properties. 



   
 

22 
 

 

Figure 9 Example of Principal Components Scores Plot, Outliers are Shown Circled in Red 

Like the scores plot, the loadings plot (an example of which is shown in Figure 10) has axes 

representing the PCs however here each point represents a single variable. It therefore shows 

the relationships between the original variables and the PCs. Each point (variable) has a 

loading value which shows how much the original variable contributed to the component. A 

high loading value indicates that the PC is strongly related to the variable. If the model explains 

a sufficiently large amount of the variance then variables which are close together in the 

loadings plot have a high correlation, and variables which fall on opposite sides of the plot 

have a negative correlation. 

The scores and loadings plots are used in conjunction, samples which are positioned on a 

particular side of the scores plot have high values of variables which are located on the same 

side of the loadings plot. The combination of the scores, loadings and residuals values gives all 

the variation in the original data set. 

 



   
 

23 
 

 

Figure 10 Example of Principal Component Analysis Loadings Plot 

While PCA is a very useful analysis tool, it is not frequently used in isolation from the other 

methods described below. There are a few disadvantages associated with this method, for 

example the loadings plots are often not easy to interpret and local patterns in the data can be 

obscured by global patterns (van der Greef & Smilde, 2005). 

1.4.3. Outliers 

Samples which are positioned far from the other samples in the PCA scores plot are likely 

candidates to be outliers (Næs, Brockhoff, & Tomic, 2011), especially those which fall outside 

of the Hotelling’s T2 region (an ellipse on the scores plot which shows the 95% confidence 

interval of the modelled variation) (Trygg et al., 2006). Figure 9 shows an example of a scores 

plot with potential outliers (circled in red). However, it is not enough to say that a sample is an 

outlier based only on this condition. It is necessary to investigate the original spectrum to find 

reasons which would qualify the sample to be excluded. These could include contaminant 

compounds, solvent residues, noise overwhelming the metabolite signals, poor water 

suppression and poor spectral quality. To narrow down the portions of the spectrum which 

must be examined to find the way in which the candidate outlier differs from the rest of the 

samples the loadings plot can be used.  



   
 

24 
 

1.4.4. Partial Least Squares – Discriminant Analysis 

Partial Least Squares or Projections to Latent Structures (Worley, 2013) (PLS) is an extension of 

PCA in which the principal components are calculated by maximising the covariance of the 

linear functions of X (e.g. the values of the bins) and the response Y (i.e. the classes). As with 

PCA the subsequent latent variables (similar to principal components from PCA) are found 

from the residuals of the previously calculated component with the condition that each 

component is independent of all other components (Næs et al., 2011).  

PLS-DA is the combination of PLS with discriminant analysis, a technique which results in the 

discriminant surface being placed in the best position to allow for the greatest separation of 

the classes (Lindon & Nicholson, 2008). This is achieved by first making the Y variables (classes) 

into dummy variables (for the most common case of two classes the labels will be set as 

vectors of length zero and one respectively (Næs et al., 2011)) and then the latent variables are 

calculated from the relation between the vectors of X and the dummy variables (maximising 

the between-group covariance matrix (Liland, 2011)). 

 

Figure 11 PCA Scores Plot of Plasma Coloured by Year of Collection 

Figure 11 is the scores plot of a PCA model of sheep plasma that was collected in two separate 

years, this same data was used to construct the PLS-DA model whose scores plot is shown in 

Figure 12. It is clear to see that the PLS-DA model has a scores plot in which the samples are 

separated by their year of collection, while the PCA model shows some mixing of the samples 



   
 

25 
 

and therefore an incomplete separation. Therefore the PLS-DA model is more likely to be able 

to give accurate information on the metabolites which cause this separation. 

 

 

Figure 12 PLS-DA Scores Plot of Plasma Coloured by Year of Collection 

1.4.5. Orthogonal Partial Least Squares – Discriminant Analysis 

While PLS-DA gives better definition of the classes than PCA, allowing for improved 

classification of samples, it can be difficult to interpret the scores and loadings plots. This is 

because the latent variables combine variation correlated with the Y-response (class) with 

variation which is uncorrelated to Y (Trygg et al., 2006). A technique called OPLS-DA combines 

Orthogonal Signal Correction (OSC) with PLS-DA (Worley, 2013). This results in all Y-correlated 

variation being contained in the first component (called the predictive component) while 

systematic variation uncorrelated to Y is confined to the higher orthogonal principal 

components (Trygg et al., 2006). This process can be shown to amount to a rotation of the PLS-

DA solution (Liland, 2011), which makes it clear that while OPLS-DA increases interpretability it 

does not improve classification (Savorani et al., 2013). If there is no Y-uncorrelated variation in 

the data-set then OPLS-DA will produce models identical to PLS-DA (Worley, 2013). 



   
 

26 
 

1.4.6. Limitations and Common Pitfalls 

Although multivariate statistical methods can be very useful in investigating metabolomic 

datasets there are a number of issues that must be kept in mind in order to prevent spurious 

deductions.  

As previously mentioned, PLS-DA and OPLS-DA are very powerful analysis tools that are prone 

to over-fitting models to data. Validation is an absolutely necessary component of any analysis 

involving these methods. These validation steps are most robust when used on two-class 

problems (for example diseased and control populations).  

Because variation correlated directly with the class differences is the subject of metabolomic 

analysis it is very important to minimise all other sources of variation and be aware of the 

effects of those which are still present.  

Once metabolites have been identified from these analysis methods it is important to consider 

them only as candidates for biomarkers until they have been validated against correlations 

from other sources. Statistical analysis is unable to distinguish between causal effects and 

indirect correlations. 

1.4.7. Validation 

A well-known downside to supervised analysis techniques (i.e. PLS-DA and OPLS-DA) is their 

potential to over-fit data to produce unreliable models. How well a model performs is 

generally based on two parameters – the goodness of fit statistic R2 (an indication of how 

much of the variance in the data is explained by the model) and the prediction statistic Q2 

(how well the model will be able to correctly predict the class assignment of new samples).  It 

is possible however, that models with high R2 and Q2 values can be formed from samples with 

randomly assigned group labels. There is no absolute value above which a Q2 value can be 

deemed to indicate a statistically valid model (Westerhuis et al., 2008). A sign of possible over-

fitting is when a model’s R2 and Q2 values differ by a large amount (i.e. more than 0.2) (Worley, 

2013) However, this in itself is not conclusive. It is therefore necessary to validate the 

reliability of models generated from supervised analysis.  Described below is a common 

validation method. 

1.4.7.1. Cross Validation 

If all possible latent variables are included in PLS-DA model building, there is a high chance that 

they will be over-fit. In order to compensate for this SIMCA 13 chooses an appropriate number 

of significant components for each model through cross validation. There are a number of 



   
 

27 
 

cross validation methods with SIMCA 13 utilising 7-fold cross validation. This method consists 

of initially splitting the dataset into 7 equal subsets, removing one of the subsets and building 

a model with a set number of components using the other 6 subsets (calibration set). The Y 

values for all members of the excluded subset are predicted using this model. This procedure is 

repeated for each subset. The differences between the observed and predicted Y values of the 

data are used to calculate the Q2 value for this particular number of components. The number 

of significant components is chosen by repeating this for an increasing number of components 

until the Q2 increase is below a certain threshold set by a particular set of rules. 

1.4.7.2. Permutation Testing 

Permutation testing can only be applied to PLS-DA models; however the inference of validation 

can be extended to OPLS-DA models with entirely identical parameters/data. It is for this 

reason that PLS-DA models were built for each of the OPLS-DA models analysed in this study. 

A permutation test consists of a specified number of iterations of creating models where the 

class labels have been randomly assigned. Often two hundred permutations is considered 

sufficient, however the more permutations used, the clearer the result can be (Westerhuis et 

al., 2008). The R2 and Q2 values for these models are then plotted against those values 

calculated from the original model. The class assignments in the permuted models are 

assumed to have no biological significance, and therefore no inherent differentiation. The y-

axis of the graph represents the values of the parameters R2 and Q2 while the x-axis represents 

the degree of similarity of the permuted models to the original model. Regression lines are 

drawn from the original values as a line of best fit through the permuted values.  An example 

of a permutation plot for a valid model is shown in Figure 13. 

The information presented in this plot must be interpreted to discover whether the model can 

be considered statistically valid. As mentioned above, model over-fit is indicated if the original 

model’s R2 and Q2 values differ by more than 0.2. This is because adding latent variables to the 

model will optimise its parameters and R2 and Q2 will increase toward 1, however at a certain 

point the two values will diverge with Q2 trending toward 0 as the model becomes over-fitted 

(Broadhurst & Kell, 2006). Therefore it can be assumed that where the values begin to diverge 

indicates the optimal parameters for prediction ability. Secondly, all permuted Q2 values must 

be lower than the original model’s Q2 value. However for R2 it is only necessary that the 

majority of the permuted values are lower than the original value. Additional requirements 

have been proposed – the y intercept of the Q2 regression line must be lower than -0.2 and the 



   
 

28 
 

R2 intercept must be lower than 0.3 (Worley, 2013). The example in Figure 13 shows that this 

model is valid. 

 

Figure 13 Permutation Test for a Valid PLS-DA Model 

1.4.8. Univariate Statistical Analysis 

A common analytical technique for determining the variance between two classes of samples 

is the t-Test. This test individually compares the means of the variables of the two classes and 

establishes the significance of the differences between them. Type I errors (rejection of a true 

null hypothesis) are a hazard of metabolomic studies where the number of variables (e.g. 

spectral bins) is far greater than the number of samples. The term False Discovery Rate (FDR) is 

used to describe the number of Type I errors. To counter the FDR it is necessary to apply a 

suitable correction to the p-value (for example a Bonferroni correction which decreases the p-

value relative to the number of variables) (Broadhurst & Kell, 2006). 

 

 Aims and Objectives 1.5.
The primary goal of this research is to use multivariate data analysis to find metabolic 

differences between meat which produces confinement odour and meat which produces 

either spoilage odour or no odour. An additional goal is to optimise the method of extraction 

of metabolites from meat and drip. 

  



   
 

29 
 

Chapter Two Materials and Methods 

 Evaluation of sample preparation methods 2.1.

A trial of the available methods for preparation of meat and drip (or drip-like substances) for 

NMR analysis from the literature was conducted in order to establish the quickest, safest and 

most convenient polar metabolite extraction protocol. Procedures for homogenising the meat 

and removing protein and solvent from both sample types were tested. 

Meat and drip (from a leg of lamb) were obtained from one of the commercial meat 

processing plants providing samples for the confinement odour trial. The leg was vacuum-

packaged and stored at -1.5°C for 16 weeks before the packaging was opened and the meat 

and drip sampled. The samples were then stored at -80°C until analysis. 

2.1.1. Chemicals 

Ultrapure water was obtained from a Milli-Q® system (Millipore, Bedford, MA). Acetonitrile 

and methanol were purchased from Thermo Fisher Scientific (Auckland, New Zealand) and 

were of Optimal LC-MS grade. Acetone and chloroform were purchased from BDH Laboratory 

Supplies (Poole, England). 3-(Trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS) and 

imidazole were purchased from Sigma. Deuterium oxide (D2O, 99.5 atom%) was purchased 

from Cambridge Isotope Laboratories (Tewksbury, MA). Perchloric acid (70%) was purchased 

from Univar (Auburn, Australia). 

2.1.2. Methods for preparation of drip samples 

As there is very little in the literature concerning preparation of drip for NMR analysis (Straadt 

et al., 2011), the methods for the drip methods test were adapted from methods for protein 

removal from plasma. 

2.1.2.1. Centrifugation 

400 µL of drip was centrifuged at 20,000 x g for 15 minutes at 4°C, 300 µL of supernatant was 

then added to 240 µL of phosphate buffer (187.5 mM, pH 7.4) and 60 µL of DSS (5 mM) in D2O 

(with 100 µM imidazole). 

2.1.2.2. Ultrafiltration 

Traces of glycerol were removed from 3,000 MWt cut-off centrifugal filters (Nanosep 3K 

Omega Microcentrifuge filters; Pall, USA) by soaking in MilliQ water overnight (Chenomx, 

2001). The next morning the filters were replaced in their tubes and were centrifuged at 

20,000 x g for 5 min to remove any excess water. 400 µL of drip was immediately added to the 



   
 

30 
 

filters and centrifuged at 20,000 x g for 18 hours at 4°C.  300 µL of supernatant was then 

added to 240 µL of phosphate buffer (187.5 mM, pH 7.4) and 60 µL of DSS (5 mM) in D2O (with 

100 µM imidazole). 

2.1.2.3. Solvent 

Four solvents or solvent mixtures were used to precipitate protein from the drip samples. The 

first solvent was acetonitrile, the second methanol, the third a mixture of acetonitrile, 

methanol and acetone, and the fourth a mixture of methanol and chloroform. 

400 µL of drip was added to one of the four ice cold solvents or solvent mixtures in either 1:2 

or 1:3 ratios. The drip-solvent mixtures were vortexed for 10 seconds before being left to 

partition on ice for 10 minutes. The mixtures were then centrifuged at 14,000 rcf for 10 

minutes at 4°C. The supernatant was collected and either freeze-dried or speed-vacuumed 

overnight as described below in 3.1.2.3. The remaining pellet was resuspended in 240 µL of 

phosphate buffer (187.5 mM, pH 7.4), 60 µL of DSS (5 mM) in D2O (with 100 µM imidazole) and 

300 µL of MilliQ water. 

2.1.3. Methods for preparation of meat samples 

The protein removal methods adapted for the meat preparation test were not only taken from 

established procedures on meat but also other forms of tissue (liver, heart and adipose 

tissue(Atherton et al., 2006)). 

Table 1 summarises the different methods applied to the meat samples. A total of sixteen 

methods were applied with combinations of each of the described components. 

Table 1 Method Test Treatment Conditions 

 

Method Liquid Nitrogen Bead Beater Methanol/Chloroform/Water Methanol/Water Perchloric Acid Acetonitrile/Methanol/Acetone Vacuum Centrifugation Freeze Drying

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

Homogenisation DryingSolvent Combination



   
 

31 
 

2.1.3.1. Homogenisation 

Homogenisation of meat samples was achieved through either liquid nitrogen grinding or bead 

beating. Liquid nitrogen grinding was conducted at AgResearch Ruakura Research Centre 

Hamilton. 

Liquid nitrogen homogenisation 

Samples were first frozen and then ground under liquid nitrogen using a SPEX Sample Prep 

Freezer/Mill 6970 (SPEX; Stanmore, United Kingdom). Once ground the samples were stored at 

-80°C until weighing. 200 mg of the ground tissue was weighed directly into 2 mL microtubes 

with screw caps and frozen in liquid nitrogen. The samples were kept frozen at -80°C until 

solvent extraction at which point they were thawed on ice. 

Bead beater homogenisation 

For the second homogenisation method 200 mg of tissue was weighed directly into 2 mL 

microtubes with screw caps (Sarstedt; Nümbrecht, Germany) containing three high density 

zirconium oxide beads of 1.4-1.7 mm in diameter (Glen Mills Inc.; Clifton NJ, USA) and frozen in 

liquid nitrogen. The sample was kept frozen at -80°C until solvent extraction. The tissue was 

thawed on ice prior to extraction and then a solvent mixture was added to the tissue which 

was homogenised in a Mini-Beadbeater-96 (Biospec Products, Inc.; Bartlesville OK, USA) at 

2100 oscillations per minute for four minutes. 

2.1.3.2. Solvent Extraction 

Methanol/Chloroform/Water 

The method of Mannina et al. (2008) was followed with slight alterations. 400 µL of ice cold 

methanol and 85 µL of ice cold MilliQ water were added to 200 mg of either liquid nitrogen 

ground tissue or unground tissue. The unground tissue was then homogenised in a bead 

beater as described above. 400 µL ice cold chloroform and an additional 200 µL of ice cold 

MilliQ water were then added to the homogenised tissue, which was vortexed for 20 seconds. 

The samples were left on ice for 10 minutes to partition then centrifuged at 16,000 x g  for 10 

minutes at 4°C. The supernatant was transferred to an eppendorf tube and kept on ice until 

solvent evaporation.  

Methanol/Water 

The protein precipitation method of Straadt et al. (2011) was used for the methanol/water 

extraction. 1.2 ml of aqueous methanol (v/v, 2:1 methanol to water) was added to 200 mg of 

liquid nitrogen ground tissue or unground tissue. The unground tissue was homogenised in a 



   
 

32 
 

bead beater as described above. The samples were left on ice to partition for 10 minutes and 

then centrifuged at 16,000 x g for 10 minutes at 4°C. The supernatant was transferred to an 

Eppendorf tube and kept on ice until solvent evaporation. 

Perchloric Acid 

400 µL of perchloric acid was added to 200 mg of liquid nitrogen ground tissue or unground 

tissue. The unground tissue was homogenised in a bead beater as described above. The 

samples were centrifuged at 16,000 x g for 10 minutes at 4°C, and then the supernatant was 

removed to an Eppendorf tube and neutralized to pH 7.8 with KOH (6 M). The samples were 

centrifuged at 16,000 x g for a further 10 minutes at 4°C, and then the supernatant was 

transferred to an eppendorf tube and kept on ice until solvent evaporation.  

Acetonitrile/Methanol/Acetone 

200 μL each of ice cold acetonitrile, methanol and acetone were added to 200 mg of tissue. 

Unground tissue mixtures were homogenised in the bead beater. All mixtures were then 

vortexed for 10 seconds and left to partition on ice for 10 minutes. The mixtures were then 

centrifuged at 14,000 rcf for 10 minutes at 4°C. The supernatant was transferred to an 

Eppendorf tube and kept on ice until solvent evaporation. 

2.1.3.3. Solvent Evaporation 

Two solvent evaporation methods were trialled. 

Vacuum centrifugation 

Samples were centrifuged in a Speed Vac Plus (SC110A) Concentrator (GMI; Minnesota, USA) 

speed vacuum with a SavantTM Refrigerated Vapor Trap (RVT400) Vapornet (VN100) and Digital 

Vacuum Gauge (DVG50) (Thermo Scientific) with a vacuum supplied by a VacuuBrand RZ-6 

vacuum pump (VacuuBrand; Wertheim, Germany); tube lids were left open and samples were 

dried for a minimum of 16 hours. The remaining residue was resuspended in 240 µL of 

phosphate buffer (187.5 mM, pH 7.4), 60 µL of DSS (5 mM) in D2O (with 100 µM imidazole) and 

300 µL of MilliQ water. 

Freeze drying 

Samples were first frozen in liquid nitrogen for ten minutes and then freeze dried with their 

tube lids open in a Dura-Dry MP (600-4413) (FTS Systems; Warminster PA, USA) freeze drier for 

a minimum of 16 hours. The remaining residue was resuspended in 240 µL of phosphate buffer 

(187.5 mM, pH 7.4), 60 µL of DSS (5 mM) in D2O (with 100 µM imidazole) and 300 µL of MilliQ 

water. 



   
 

33 
 

 Preparation of confinement odour samples 2.2.

Ninety meat samples were obtained from two commercial meat processing plants and were 

stored at various chilled temperatures to simulate the conditions of exported meat during 

shipment overseas, transport to warehouse, and retail display. The various conditions are 

shown in Table 2. The number of samples and conditions were chosen to produce meat with a 

variety of odour types. At the end of the storage time meat and drip samples were collected 

just after opening the packs and immediately frozen in liquid nitrogen. Some samples did not 

produce enough drip for analysis. Immediately upon opening the packages the meat was 

assessed for odour and again ten minutes later, those with persistent odour were classified as 

having spoilage-related odour, those with an odour upon opening which dissipated before the 

second assessment were classified as having confinement odour and those which had no 

odour at either assessment were classified as having no odour. Samples with no odour made 

up 10% of the samples, while those with confinement odour or spoilage odour made up 22% 

and 68% respectively. 

Table 2 Storage Conditions of Meat Samples and Subsequent Odour Types 

 Treatment No of samples 

from Plant 1 

No of samples 

from Plant 2 

No of Odour Type 

None CO Spoil. 

Control 1 11 weeks -1.5°C 5 5 3 1 6 

Control 2 12 weeks -1.5°C 5 5 1 4 5 

Control 3 13 weeks -1.5°C 5 5 1 0 9 

Treatment 1 7 weeks -1.5°C 

4 weeks +2.0°C 

5 5 2 3 5 

Treatment 2 7 weeks -1.5°C 

2 weeks +2.0°C 

2 weeks +4.0°C 

5 5 0 5 5 

Treatment 3 7 weeks -1.5°C 

5 weeks +2.0°C 

5 5 0 5 5 

Treatment 4 7 weeks -1.5°C 

2 weeks +2.0°C 

3 weeks +4.0°C 

5 5 2 2 6 

Treatment 5 7 weeks -1.5°C 

6 weeks +2.0°C 

5 5 0 0 10 



   
 

34 
 

Treatment 6 7 weeks -1.5°C 

2 weeks +2.0°C 

4 weeks +4.0°C 

5 5 0 0 10 

 

Before preparation all drip and meat samples were separately randomised into six groups 

using a random number generator (random.org). The order of preparing the groups was also 

randomized; this was to prevent any possible systematic variation being introduced to the 

samples which could confound the statistical analysis. 

2.2.1. Preparation of drip samples 

400 µL of drip was added to 800 µL of a mixture of acetonitrile, methanol and acetone in equal 

parts. The drip-solvent mixtures were vortexed for 10 seconds before being left to partition on 

ice for 10 minutes. The mixtures were then centrifuged at 14,000 rcf for 10 minutes at 4°C. The 

supernatant was collected and kept on ice until solvent evaporation. Samples were vacuum 

centrifuged, with their tube lids open, for a minimum of 16 hours. The remaining powder was 

resuspended in 240 µL of phosphate buffer (187.5 mM, pH 7.4), 60 µL of DSS (5 mM) in D2O 

(with 100 µM imidazole) and 300 µL of MilliQ water. 

2.2.2. Preparation of meat samples 

For the sample preparation itself 200 mg of tissue was weighed directly into 2 mL microtubes 

with screw caps (Sarstedt) containing three beads and frozen in liquid nitrogen. The samples 

were kept frozen at -80°C overnight. The following day the samples were thawed on ice prior 

to extraction and 200 μL each ice cold acetonitrile, methanol and acetone were added to each. 

The tissue was homogenised in the bead beater f