WO2017153783A1 - Nmr method of detecting a marker in a coffee product - Google Patents

Abstract

A method for detecting a marker in a sample is provided. The method comprises adding solvent to the coffee to prepare a sample; scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint, and; analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker, wherein the marker is caffeine or 16-OMC.

Description

NMR METHOD OF DETECTING A MARKER IN A COFFEE PRODUCT

This invention relates to improvements in or relating to a method of detecting a marker in a sample. In particular, the invention relates to a method of detecting a marker in a sample, such as coffee beans, using low-field NMR spectroscopy.

Background

Coffee beans are one of the most widely traded commodities in the world, and as such, are vulnerable to fraud within the supply chain. The two main species that are traded are Coffea arabica L. (~ 70% of the market) and Coffea canephora Pierre ex A. Froehner (variety robusta) (Belitz et al., 2009). Arabica beans are the highest value coffee beans, prized for their smooth, rounded flavour, whilst the more disease resistant robusta beans are much cheaper, and as their name suggest, produce a rougher brewed drink. There is potential, therefore, for unscrupulous traders to make economic gain by partially or wholly substituting "arabica" with "robusta" products, deceiving other parties in the supply chain and, ultimately, the consumer. Objective methods are needed for the reliable identification of both species, and for the estimation of their contents in coffee products. Whole coffee beans may be distinguished by inspection, but chemical analysis is required to confirm the identity of ground roasted products, and for example to detect the adulteration of arabica by amounts of robusta.

Ground roast coffee is a complicated mixture of hundreds of different organic compounds, present in concentrations ranging from trace amounts up to tens of % by weight. Amongst components found in the lipophilic fraction of the coffee bean are diterpenes of the kaurene family, in amounts of up to 20% of the total lipid content. These include cafestol, found in both bean types, and kahweol, found in arabica beans and in small amounts in some, but not all, robusta beans. However, a further diterpene, 16-0-methylcafestol (16-OMC) is found exclusively in robusta beans according to the prior art, and has thus been proposed as a marker for distinguishing between the two bean types (Speer & Mischnick, 1989). The stability of 16-OMC with respect to the roasting process means that it can also be used to detect the presence of robusta in processed coffee products (Speer & Kolling-Speer, 2006). An official method exists for the determination of 16-OMC in roasted coffee by HPLC, but it requires a time-consuming sample preparation (DIN 10779, 1999).

High-field ¹H NMR spectroscopy has been previously reported for the analysis of coffee. The majority of studies have examined aqueous extracts of coffee in a variety of applications (Cagliani et al., 2013, Charlton et al., 2002, Consonni et al., 2012, Schievano et al., 2014, Wei et al., 201 1 and Wei et al., 2012). In contrast, the recent study by Monakhova et al (2015) was focussed on the analysis of lipophilic extracts from coffee beans and products, and their potential for addressing issues of authenticity in arabica and robusta coffees. It showed that many minor components, including kahweol and 16-OMC, produce clearly identifiable signals in 400MHz ¹H NMR spectra, and further, that the integration of the 16-OMC signals can be used to estimate the amount of robusta in coffee blends with an approximate limit of detection of 1 -3%. The authors concluded that high-field NMR spectroscopy has the potential as a screening tool for identifying and detecting coffee species before applying the more time-consuming official method, for example HPLC analysis.

However, the maintenance of high-field NMR instruments can be costly and difficult. Typically, most high-field NMR spectroscopy instruments require cryogens such as liquid nitrogen and helium, which may present additional risks to the user.

A recent development in NMR technology, low-field ("benchtop") spectroscopy, can be used to address some of the issues associated with high-field NMR spectroscopy. However, low-field NMR instruments lack the sensitivity and spectral resolution that higher-field NMR spectrometers are able to offer. Moreover, a typical doublet splitting of 6Hz translates into a peak separation of 0.01 ppm when collected at 600MHz, but becomes a larger separation of 0.1 ppm when working at 60MHz. This can make it difficult for a user to detect markers within a sample in the low-field NMR spectra, even though these markers may be readily identifiable in the high-field NMR spectra.

It is against this background that the present invention has arisen.

Summary of the invention

The recent development in NMR technology, low-field ("benchtop") spectroscopy, can be used to address issues of coffee authentication and quality control. Compared with high-field instruments, low-field spectrometers are smaller and more robust. Capital and maintenance costs are lower, as these instruments utilise permanent rather than electro-magnets and thus do not need any cryogens. Consequently, low-field spectroscopy can be a rapid and cost-effective means for increasing the uptake of authenticity testing of products. According to an aspect of the invention, there is provided a method for detecting a marker in a sample, the method comprising (i) scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint and; (ii) analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker. Preferably, the marker is a strain-specific marker in the sample. In some embodiments of any aspect of the invention, the marker may be a naturally-occurring marker in the sample.

According to the present invention there is provided, a method for detecting a marker in coffee, the method comprising

(i) adding solvent to the coffee to prepare a sample; (ii) scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint, and;

(iii) analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker, wherein the marker is caffeine or 16-OMC. The coffee may be, but is not limited to, whole, ground, bean, powder, extract or concentrated coffee. The coffee may also be freeze dried or spray dried instant coffee. The beans may be fresh or dried, roasted or unroasted coffee beans.

In some embodiments, the method may further comprise the steps of; (a) extracting a lipophilic composition of the coffee;

(b) concentrating the extract to produce a test sample;

(c) scanning the test sample using low-field NMR spectroscopy to generate a spectral fingerprint; and

(d) comparing the spectral fingerprint of the test sample to the spectral fingerprint of the known authentic sample to detect the presence of the marker, wherein the marker is caffeine or 16-OMC.

In some embodiments, the method may further comprise obtaining a spectral fingerprint of a known authentic coffee sample, which is indicative of a low amount of the marker such as 16-OMC or caffeine, wherein the low amount can be below an effective detection limit.

In some embodiments, the method may further comprise the step of comparing the spectral fingerprint of the test coffee sample to the spectral fingerprint of the known authentic sample, wherein the detection of the marker above the effective detection limit in the test coffee sample can indicate the presence of the marker. This step may be particularly useful for analysing a high concentrated sample.

In a further embodiment, the method may further comprise setting an effective detection limit, wherein the effective detection limit can be expressed as a ratio between two coffee beans (w/w), such as robusta to arabica.

The sample can be a coffee composition. In some embodiments, the sample may be a robusta extract. In some embodiments, the sample may be a mixture of arabica and robusta extracts. In some embodiments, the sample may be an extract from a composition claiming to be an arabica extract. The test coffee sample may be a robusta extract, an arabica extract, or it may be a mixture of arabica and robusta extracts. The known authentic coffee sample may be, but not limited to an arabica extract or a robusta extract. Preferably, the known authentic coffee sample is arabica. In some embodiments, the methods of the present invention may further comprise preparing the sample to provide a lipophilic composition prior to the scanning step (ii). Lipophilic compositions can be advantageous as it may provide a clearer NMR spectrum. Furthermore, extracting lipophilic composition from a sample is a relatively straightforward process. In a further embodiment, a solvent may be added to the sample prior to the scanning step (ii). Preferably, the solvent is chloroform. Most preferably, the solvent is deuterated chloroform. By using a simple solvent based process, lipophilic compositions from coffee samples can be readily extracted. Furthermore, a protonated solvent may give rise to a large solvent peak in the NMR spectrum, which may obscure the peaks from the sample in the NMR spectra. By using a deuterated solvent, the dominant peaks of the solvent can be avoided in the NMR spectra.

The method of the present invention may further comprise concentrating the extract prior to the scanning step to produce a test sample. This may be useful for increasing the signal to noise ratio in the NMR spectra, which can result in larger peaks appearing in the spectra. Therefore, concentrating the test sample prior to the scanning step may increase the sensitivity of detecting minor compounds or markers in test coffee samples. Preferably, the extract may be suitably concentrated prior to the scanning step for detecting low levels of 16-OMC or caffeine in test samples.

Alternatively, increasing the number of scans during NMR acquisition may increase the signal to noise ratio in the NMR spectra. Preferably, the signal to noise ratio increases as the square root of the number of scans. The enhanced signal to noise ratio in the NMR spectra, obtained by increasing the number of scans, may be useful for detecting low amounts of the marker such as 16-OMC or caffeine in test coffee samples.

Optionally, the authentic arabica sample may be extracted and then concentrated using the methods of the present invention. Low levels of 16-OMC in authentic arabica samples have not previously been reported in the prior art. By concentrating the authentic arabica sample as disclosed in the present invention, it may increase the signal to noise ratio in the NMR spectra, which may enable the discovery or detection of low amounts of 16-OMC or caffeine in authentic arabica.

Suitably the sample is placed in a NMR sample holder of the low-field NMR spectrometer in such a way as to ensure that the sample experiences a substantially uniform magnetic field strength with a common direction.

Typically, when the NMR spectrometry is in use, the scanning step (ii) comprises applying a radio-frequency pulse to the sample in the sample holder and detecting a free induction decay (FID) signal, which can then be suitably Fourier Transformed (FT) into peaks that are easily identifiable in the NMR spectrum. The spectral fingerprint of the sample may be a 1 D spectrum. Most preferably, the spectral fingerprint is a proton 1 D NMR spectrum.

Alternatively, the NMR spectrum may be a 2D NMR spectrum. The spectral fingerprint of the sample using 2D-NMR may be useful for detecting markers. Furthermore, 2D-NMR may reduce signal overlap and therefore provide a higher resolution in the spectra.

The low-field NMR instrument may have a magnetic field strength of 40 to 100 MHz or it may exceed 45, 50, 60 and 75 MHz. In some embodiments, the magnetic field strength of the NMR instrument may be less than 100, 75, 60, 50 or 45 MHz. Preferably, the low-field NMR instrument has a magnetic field strength of 60 MHz.

In some embodiments, the marker is 16-OMC. Typically, 16-OMC may be present in robusta extracts or the mixture comprising robusta extracts thereof. In some embodiments, a low level amount of 16-OMC marker may be present in arabica extracts or the mixture comprising arabica extracts thereof. For example, the low level amounts of 16-OMC marker in arabica sample or in a mixture comprising arabica extracts thereof can be 1 % w/w.

In some embodiments, the method may further comprise analysing the spectral fingerprint unique to the marker at a chemical shift region of between 0.8 ppm to 6.20 ppm. In some embodiments, the detection of a peak at the chemical shift region of 3.16 ppm may indicate the presence of 16-OMC.

The 16-OMC has a plurality of peaks at a chemical shift region of 0.8 ppm to 6.20 ppm in the 1 D spectrum. In addition, the peaks of 16-OMC may also occur at the chemical shift regions of 0.8 to 2.6 ppm, 3.16 ppm and 6.20 ppm in the 1 D spectrum. The locations of peak centres on the chemical shift scale are invariant to field strength, peak splittings are not. For instance a typical doublet splitting of 6Hz translates into a peak separation of 0.01 ppm when collected at 600MHz, but becomes a larger separation of 0.1 ppm when working at 60MHz. Thus, spectra that contain many resonances will often exhibit substantially different profiles at low and high field strengths; consequently, it is not obvious that an analysis based upon interpretation of high-field data will translate readily to low-field measurements.

This issue may be overcome by detecting the presence of 16-OMC within samples using low-field NMR spectroscopy, since 16-OMC has a distinctive chemical shift fingerprint in the low-field NMR spectrum.

The peaks positions at 0.8 to 2.6 ppm, 3.16 ppm and 6.20 ppm in the 1 D spectrum are a unique fingerprint to the 16-OMC. Furthermore, these peaks are relatively isolated from the main peaks in the NMR spectrum. Therefore, these peaks act as good markers for quickly identifying and detecting the presence of 16-OMC in the NMR spectra.

The peak that occurs at 3.16 ppm in the NMR spectra can be immune from the effects of the external magnetic field strength. This is particularly advantageous because the peak can occur in exactly the same chemical shift position i.e. at 3.16 ppm in the NMR spectra at low or high field strengths. Therefore, the peak at 3.16 ppm is particularly useful as marker for detecting the 16-OMC compound in samples when using different field strengths. In some embodiments, the peak at 3.16 ppm may be a singlet peak in the NMR spectra.

In some embodiments, the peak positions at 0.8 to 2.6 ppm, 3.16 ppm and 6.20 ppm can be obtained from the NMR spectra at 60 MHz. In a further aspect of the present invention there is provided, a method of preparing a test coffee sample for authentication, the method comprising

(a) extracting a lipophilic composition of the coffee comprising a marker;

(b) concentrating the extract to produce a test sample; wherein the test sample is capable of being authenticated by detecting the marker in a spectrum.

The extraction process may be a solvent-based extraction for example; it may be a chloroform-based extraction process of lipophilic compositions comprising a marker. In some embodiments, the chloroform may be deuterated. Optionally, the solvent may be hexane, ethanol or methanol. Using hexane, ethanol or methanol can be advantageous because it may be a cheaper solvent to use than chloroform during the extraction of lipophilic compositions. In addition, using hexane or methanol may also permit the extraction process to be carried outside of a fume cupboard.

Alternatively, the solvent may be an aqueous solution. A solvent such as water may be used to extract an aqueous composition of the coffee comprising the marker. In certain examples the solvent may be a mixture of water and a water miscible liquid such as ethanol. In some embodiments, the water or water mixture may be deuterated. As an example, an aqueous-based extraction process can be useful to apply to coffee containing little amounts of lipid content such as instant coffee.

According to another aspect of the invention, there is provided a method of enabling verification of the authenticity of a product, the method comprising scanning the product using low-field NMR spectroscopy and; identifying a marker of the product in a spectrum.

In a further aspect of the invention, there is provided a method of enabling verification of the authenticity of a coffee product, the method comprising scanning the product using low-field NMR spectroscopy and identifying a marker of the product in a spectrum, wherein the marker is caffeine or 16-OMC.

Preferably, the marker is 16-OMC. In one embodiment, the marker is naturally- occurring.

The method may further comprise scanning the product or an example of the product using low-field NMR spectroscopy to generate spectroscopic data. This may act as a signature of the product, thereby enabling the authenticity of a product to be determined by comparing the signature of the product with a signature of a known authentic product.

The product may be a food or beverage product. Preferably, the product is a beverage product such as coffee. The product may be a mixture of arabica and robusta extract, or it may be a robusta extract, or it may be an arabica extract. In some embodiments, the known authentic product is an arabica extract.

The method may further comprise analysing the spectral fingerprint unique to the marker using Monte Carlo Simulation. In one embodiment, the marker may be a naturally-occurring marker. Analysing the NMR spectra using Monte Carlo Simulation can be an advantage because it provides a statistical analysis of the spectral fingerprint data.

The Monte Carlo Simulation of arabica extract may have a statistical distribution range of between 0.5 to 99.5 percentiles. Moreover, the Monte Carlo Simulation of robusta coffee may have a statistical distribution range of above 99.5 percentile. Monte Carlo Simulation can provide a means to distribute data, which may enable a user to easily visualise the data and distinguish between arabica and robusta products.

The effective detection limit of the robusta to arabica in the product may be 10% w/w. In some embodiments, the concentration step prior to scanning step (ii) may increase the effective detection limit. For example, the effective detection limit of robusta to arabica in the product may be 0.5% to 10% w/w, 0.5% to 5%, or 1 % w/w to 3% w/w. In some embodiments, the effective detection limit of robusta to arabica in the product may be 1 % w/w. In a preferred embodiment, the detection limit of the robusta comprising 16-OMC marker to arabica in the product is 10% w/w. In some embodiments, the effective detection limit of robusta comprising the 16-OMC marker to arabica in the product may be 0.5% to 10%, 0.5 to 5% w/w, 1 % w/w to 3% w/w or 1 % w/w.

The product may comprise 10-40% of robusta coffee. In some embodiments, the effective detection limit for robusta coffee to arabica coffee is 10% w/w. In some embodiments, the effective detection limit for robusta coffee to arabica coffee is 0.5 to 10%, 0.5% to 5% w/w, 1 to 3% w/w or 1 % w/w.

In another aspect of the invention, there is provided a method for detecting a 16- OMC marker, the method comprising scanning a coffee composition having the 16-OMC using low-field NMR spectroscopy, whereby the scanning of the coffee composition results in a NMR spectral fingerprint unique to 16-OMC.

In a further aspect of the invention, there is provided a database of reference spectral data generated according to the method of any one of the previous aspects. Preferably, the spectral data is NMR data. Further aspects and embodiments of the invention are described in more detail below.

Description of Figures

The invention will now be further and more particularly described, by way of example only, and with reference to the accompanying drawings, in which:

Figure 1 shows an NMR spectrum of lipophilic extract from arabica coffee beans according to an aspect of the invention; Figure 2 shows a spectral fingerprint of 16-OMC as shown in Figure 1 ;

Figure 3 illustrates the spectral fingerprints of arabica and robusta extracts;

Figures 4a, 4b, 4c and 4d provide data showing varying intensity levels of robusta in a mixture; Figures 5a, 5b and 5c show the analysis of robusta and arabica samples using Monte Carlo Simulation;

Figures 6a, 6b and 6c show NMR spectra of lipophilic extract from robusta coffee sample;

Figure 7a provides a graph showing the concentration of robusta in a sample and Figure 7b shows the relevant spectral region of the robusta content;

Figure 8 provides data showing the estimated robusta content in 60 coffee samples;

Figures 9a and 9b show a graph of theoretically calculated probability function for a study of 60 coffee samples; Figures 10a and 10b show the 1 D NMR spectra of various samples at a chemical shift range of between 3.1 ppm to 3.7 ppm;

Figures 1 1 a show the integrated 3.16 ppm peak areas in 60MHz spectra,

Figure 1 1 b shows a normal probability plot for the data according to Figure 1 1 a,

Figure 1 1 c shows the empirical and fitted cumulative distribution functions for typical arabica coffees;

Figure 12a shows a calibration chart, at the low concentration region, to estimate the concentration of adulterant (robusta or other non-arabica) present in samples,

Figure 12b provides a calibration graph according to Figure 12a, which shows the integrated 3.16 ppm peak areas for 60 samples; Figure 13a shows an integrated area of 3.16 ppm and Figure 13b shows the 1 D NMR spectra at chemical shift range of 3.0 ppm to 3.7ppm; Figures 14a, 14b and 14c show plots of the 3.16 ppm peak area in 60 MHz versus 600 MHz;

Figure 15a shows a calibration chart for arabica and non-arabica coffee samples,

Figure 15b provides a table of estimated confidence intervals for predicted concentrations of the samples as illustrated in Figure 15a; and

Figure 16a provides a graph showing the number of samples as a function of the fraud prevalence,

Figure 16b shows the relative chance of obtaining cases of fraud at a range of prevalence rates according to Figure 16a; and Figure 16c shows the cumulative distribution functions for data according to Figures 16a and 16b.

Detailed description

The present invention relates to a method of detecting a marker, such as a naturally-occurring marker in a sample, the method comprising scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint and; analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker. In one embodiment, the method may relate to detecting one or more marker in a sample.

The present invention also relates to a method for detecting a marker in a sample, the method comprising (i) scanning the test sample using low-field NMR spectroscopy to generate a spectral fingerprint of the marker and; (ii) analysing the spectral fingerprint of the test coffee sample to detect the marker unique to the sample. The method may further comprise the step of obtaining a spectral fingerprint of a known authentic coffee sample, which is indicative of a low amount of the marker prior to step (ii), wherein the low amount can be below an effective detection limit. Furthermore, the method may comprise the step of comparing the spectral fingerprint of the test coffee sample to the spectral fingerprint of the known authentic sample, wherein the detection of the marker above the effective detection limit in the test coffee sample can indicate the presence of the marker.

Prior to detecting and analysis of the marker in the sample, the samples can be prepared using a lipophilic extraction procedure. Lipophilic extraction for samples may be achieved by straightforward mixture of the sample, e.g. coffee grounds, with an organic solvent such as chloroform followed by filtering. This may make the experimental protocol as straightforward as possible, with the ultimate goal of developing a high-throughput, low-cost screening approach for detecting undeclared addition of compounds to a sample, for example, the undeclared addition of ground roast robusta beans to products sold as 100% arabica.

Typically, the sample may be placed in a tube, such as a 1 mm, 5mm or 10 mm NMR tube or a shigemi tube. Suitably, the tube may be chosen to improve sample read-out. The sample may then be placed in an NMR sample holder of the low- field NMR spectrometer. The sample holder may comprise a spinner.

The NMR spectrometer used in this invention is preferably a low-field NMR instrument, which is a self-contained unit that can be placed on a bench or surface and moved as necessary (unlike superconducting NMR spectrometers). When the NMR spectrometry is in use, a radio-frequency pulse, such as a 90 or 180 degree pulse, may be applied to the sample in the sample holder and detect a free induction decay (FID) signal according to standard protocol. The FID signals related to the sample may then be suitably Fourier Transformed (FT) into peaks that are easily identifiable in the NMR spectrum. Optionally, the spectral fingerprint may be a 1 D, 2D, 3D, 4D or 5D NMR spectrum. Typically, the spectral fingerprint is a 1 D NMR spectrum. Preferably, the spectral fingerprint is a proton 1 D NMR spectrum.

A 60MHz spectrum of an extract prepared from one of the whole bean arabica sample is shown in Figure 1 , along with a 600MHz spectrum collected from the same extract. Alternatively, the high-field NMR spectrum may be a 400 MHz, 600 MHz, 800 MHz or 1 GHz NMR spectrum. For low-field NMR spectrum, a spectrum may be 40 MHz, 60 MHz, 80 MHz or 100 MHz spectrum.

As indicated in Figure 1 , the spectra have been independently scaled and offset to facilitate comparison. At both field strengths (60 MHz and 600 MHz), the spectral profile may be dominated by resonances attributable to the triglyceride component of the extract. Lipids (mostly triglycerides, but also di- and mono-glycerides as well as free fatty acids) are present in roast coffee in concentrations up to 14% w/w, thus triglycerides are the major constituent of the lipophilic extract.

Although the resonances in the 60MHz spectrum are broader and more overlapped than in the 600MHz spectrum, the spectra nevertheless contain analogous information, as shown in Figure 1 . Furthermore, where there is no overlap from the triglyceride signals, small resonances arising from the more minor constituents of the extract can be discerned. This may assist in annotating the features as they appear in the low-field spectrum, in which peaks at 3.42, 3.59 and 3.99ppm can be attributed to caffeine, and the somewhat more overlapped features at 5.95, 6.19, 6.30ppm to the main diterpenes (kahweol, cafestol) found in arabica coffees.

As illustrated Figure 1 , there is shown the spectra of the sample, which was prepared from authentic arabica beans. The samples were not concentrated prior to the scanning step. Therefore, there are no features of any signals arising from another diterpene, 16-OMC shown in Figure 1 , which can act as a recognized marker compound in robusta beans. However, in some embodiments, isolated resonances from 16-OMC of robusta extracts could be seen in high-field NMR spectra, such as the 600MHz spectra. Referring to Figure 2, there is shown the spectra of the 16-OMC in chloroform collected at 60MHz, and for comparison purposes, at 600MHz. The majority of the 16-OMC resonances are found in the 0.8 - 2.6ppm range, which is also where the most prominent triglyceride resonances occur. However, peaks corresponding to 16-OMC are centred at 3.16, 3.75 and 6.20ppm (respectively labelled [i] - [iii] in Figure 2) in the NMR spectrum are comparatively more isolated. A peak [iii] at 6.20 is a doublet arising from the H-|₈ proton. The splitting can be seen in the 600 MHz spectrum, but cannot be resolved at 60MHz. In terms of using this peak for discrimination, the issue is that other diterpenes found in coffees exhibit resonances at similar chemical shifts that also arise from the H-|₇ and Hi s protons (H-|₈ doublet at 6.21 ppm in cafestol and dehydrocafestol; H-|₈ singlet at 6.30ppm in kahweol and 16-O-methylkahweol and at 6.31 ppm in dehydrokahweol; H-|₇ singlet at 3.78ppm in 16-O-methylkahweol; H-i _7a doublet at 3.83ppm in kahweol; H-i _7b doublet at 3.72ppm in cafestol and at 3.70ppm in kahweol). In addition, these resonances are resolved into discrete signals at high- field strengths, but at 60MHz one must expect considerable overlap, as indeed is seen in low-field spectrum.

Referring to Figure 2, a peak [ii] at 3.75ppm is attributed to the two H-|₇ protons. Note that in the esterified form of the 16-OMC molecule, as is the case in coffee, these protons signals appear instead as two doublets at 4.28 and 4.45ppm, which are overlapped with triglyceride signals in the low-field NMR spectra.

Referring to Figure 2, a peak [i] at 3.16ppm is a singlet arising from H₂i protons in the methyl functional group that distinguishes 16-OMC from cafestol. As shown in Figure 2 there seems to be no visible or small resonances seen in this region of the 60MHz spectrum of the arabica extract, making this peak the obvious candidate to be used as a marker signal for the presence of robusta coffee using low-field NMR spectroscopy.

As shown in Figure 3, the low-field spectra from the arabica extract and two further extracts prepared from decaffeinated arabica and robusta beans are plotted for the region between 3 and 6.5ppm, using a greatly expanded and offset y-scale for clarity. Typically, the presence of caffeine resonances appear in the spectra of both the arabica and robusta extracts but not in that from the decaffeinated beans, and the presence of kahweol in the arabica extracts but not the robusta.

In Figure 3, the plotted data also includes the 16-OMC 1 D spectrum: the three peaks [i] - [iii] occur in this region of the chemical shift scale. From Figure 3, one or more prominent peaks corresponding to 16-OMC can be seen in robusta coffee bean extracts. Peak [i] at 3.16ppm is clearly visible as an isolated feature. Peak [ii] can just be discerned between the caffeine bands. Peak [iii] is also apparently isolated, but is coincident with the kahweol features seen in the arabica extracts (kahweol is sometimes present in small amounts in robusta) and also with cafestol resonances which are known from high-field assignments to occur at around this chemical shift.

Preferably, low-field NMR spectroscopy may be suitably used as a high- throughput screening tool for detecting the undeclared addition of robusta to ground arabica coffees. In this context, peak [i] at 3.16ppm is an obvious candidate marker signal, as it is not overlapped by any other resonances. Furthermore, resonance [i] was examined quantitatively in sets of spectra obtained from two mixture series along with the authentic arabica and robusta samples used to prepare the mixtures in each case. In each spectrum, the peaks were locally baseline corrected using a polynomial fit, and then normalized through division by the integrated glyceride peak area, approximately 3.8 - 4.6ppm, as indicated in Figure 3. A normalization step is often useful in NMR when there is unavoidable variation in sample concentration, which in the present case arises from variable extraction efficiency at the sample preparation stage. The glyceride resonances are present in all triglyceride spectra and thus provide a useful internal reference signal.

Referring to Figure 4a, there is shown an expansions of peaks at around 3.0 - 3.3ppm ppm following baseline correction and normalization, from a sample mixture. The progression between 0% robusta (in which the peak is absent) and 100% is shown in Figure 4a. According to Scharnhop et al, the resonance at 3.16ppm as observed in the spectrum of pure 16-OMC may be a singlet originating from three protons. However, it may also have been noted that chemical breakdown of 16-OMC, for example due to exposure to light, may cause the information in this peak to be shifted to lower ppm values, although the nature of the decomposition products is not known. Typically, this resonance at 3.16ppm can be seen in the 60MHz spectra. Furthermore, the shape of the feature suggests that it may comprise multiple overlapped resonances, with signal-to- noise and field strength limitations preventing these from being fully resolved.

Referring to Figure 4b, the data shows the spectra of varying robusta content collected on the 600 MHz spectrometer. The data collected on the 600 MHz spectrometer are comparable to the data collected on the 60 MHz spectrometer, as shown in Figure 4a.

To represent the amount of 16-OMC present in the sample, the 3.16ppm resonance as well as the breakdown product peaks were integrated in the spectra from both series collected at both field strengths. These are plotted versus the concentration of robusta in Figures 4c and 4d.

In addition, calibrations serve to illustrate that quantitation, for example in the context of quality control, is a feasible prospect using low-field NMR. The consistency of the results from the two or more mixture series implies that an analytical precision of +/-6%w/w on the compositional values can be achievable by 60MHz NMR. It compares well with +/-4%w/w obtained by 600MHz spectroscopy, considering the difference in cost and complexity between the two techniques. In the context of adulteration detection, however, quantitation may not be essential.

According to one aspect of the invention, by scanning the product or an example of the product such as coffee using low-field NMR spectroscopy, this can generate spectroscopic data which acts as a signature of the product, thereby enabling the authenticity of a product to be determined by comparing the signature of the product with a signature of a known authentic product.

For example, to identify whether there are any presence of robusta extracts within a mixture, one approach is to analyse the NMR spectrum and identify or detect the peaks that correspond to 16-OMC. This may require the use of a reference NMR spectrum where only pure 16-OMC peaks are present in the NMR spectrum. Since the 16-OMC can be absent or occur only at low level amounts in arabica coffees, the presence of a peak corresponding to 16-OMC in the NMR spectrum would indicate the presence of robusta in the mixture. For data analysis, one such method is matched filtering, in which a "template" signal is cross-correlated with a measured signal in order to detect the presence of the template in the measured signal. This approach was developed for application to baseline-corrected sections of the low-field spectra in the region 3.05 - 3.30ppm.

As shown in Figure 5, a statistical analysis, preferably Monte Carlo simulation was used to determine the distribution of the maximum normalised cross-correlation under conditions of the null hypothesis (HO, no matching signal is present) using a Lorentzian line shape of a suitable width and location as the template. This statistic is normally distributed, and the parameters of its distribution can be used directly to establish threshold values for the rejection of HO at the desired probability level. In doing so, the assumption is made that for arabica extracts, this spectral region amounts to nothing other than Gaussian white noise. This approach offers the obvious advantage that a database of reference samples is not required, since the boundaries of the arabica group are estimated by simulation only.

Referring to Figure 5a, the figure shows the values obtained from the authentic samples, such as arabica extracts prepared from whole beans. As shown in Figures 5a and 5b, also marked on the plots are 0.5 and 99.5 percentiles for the statistic's distribution obtained from the simulation. Typically, the values for the authentic arabica samples lie between these percentiles. Furthermore, the robusta samples may lie far above the 99.5% percentile, meaning that for these samples, HO can be confidently rejected. This may indicate that in each case, a peak consistent with a 16-OMC signal has been found in the data. This may show beyond doubt that the low-field NMR approach is capable of distinguishing entirely reliably between unadulterated arabica and robusta ground roast coffees.

Referring to Figure 5b, there is shown the values of the test statistic obtained from the mixture samples. As shown in Figure 5b, the exception of one 10% w/w and one 40% w/w robusta, all of the mixture spectra are detected as containing a 16- OMC peak and thus some robusta coffee. Determining a precise detection limit may not be possible, since it would require prior knowledge of the concentration of 16-OMC present in the adulterant, and this may be quite variable between different robusta beans. However, the results shown in Figure 5b suggest that mixtures containing 20% w/w robusta are likely to be detected by this method present in the invention, and there is a suggestion that at least some samples containing 10% w/w robusta can also be identified.

Referring to Figure 5c, all the samples in a surveillance study of retail purchased coffees are found to lie within the 0.5 and 99.5 percentiles, with the exception of two extracts which fall just above the 99.5% boundary.

Suitably, a library of reference spectra is generated. The reference spectra may be NMR spectra of samples such as arabica and/or robustra extract or a mixture of arabica and robusta extracts thereof.

Other aspects

Aspects and embodiments of the invention are also set out in the following clauses:

1 . A method for detecting a marker in a sample, the method comprising

(i) scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint and;

(ii) analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker.

2. The method according to any one of the preceding clauses, wherein the sample is a coffee composition.

3. The method according to any one of the preceding clauses, wherein the sample is a robusta extract. 4. The method according to any one of the preceding clauses, wherein the sample is a mixture of arabica and robusta extracts. 5. The method according to any one of the preceding clauses, further comprising preparing the sample to provide a lipophilic composition prior to the scanning step (i).

6. The method according to any one of the preceding clauses, further comprising adding a solvent to the sample prior to the scanning step (i).

7. The method according to clause 6, wherein the solvent is chloroform.

8. The method according to any one of the preceding clauses, wherein the scanning step (i) comprises applying a radio-frequency pulse to the sample and detecting a free induction decay using the NMR spectrometer. 9. The method according to clause 8, wherein the free induction decay is Fourier transformed.

10. The method according to any one of the preceding clauses, wherein the spectral fingerprint is a 1 D spectrum.

1 1 . The method according to clause 10, wherein the spectral fingerprint is a proton 1 D spectrum.

12. The method according to any one of the preceding clauses, wherein the low field NMR spectroscopy has a magnetic strength of 40 to 100 MHz.

13. The method according to any one of the preceding clauses, wherein the magnetic field strength is 60 MHz. 14. The method according to any one of the preceding clauses, wherein the marker is caffeine.

15. The method according to any one of the preceding clauses, wherein the marker is 16-OMC.

16. The method according to clause 15, wherein the 16-OMC is present only in robusta extracts or a mixture comprising robusta extracts thereof. 17. The method according to any one of clauses 15 or 16, wherein the 16-OMC has a plurality of peaks at a chemical shift region of between 0.8 ppm to 6.20 ppm in the 1 D spectrum.

18. The method according to any one of clauses 15 to 17, wherein the peaks of 16-OMC occur at the chemical shift regions of 0.8-2.6 ppm, 3.16 ppm, 3. 75, 10 ppm and 6.20 ppm in the 1 D spectra.

19. A method of enabling verification of the authenticity of a product, the method comprising scanning the product using low-field NMR spectroscopy and identifying a marker of the product in a spectrum. 20. The method according to clause 19, further comprising scanning the product or an example of the product using low-field NMR spectroscopy to generate spectroscopic data which acts as a signature of the product, thereby enabling the authenticity of a product to be determined by comparing the signature of the product with a signature of a known authentic product. 21 . The method according to clauses 19 to 20, wherein the marker is 16-OMC.

22. The method according to clauses 19 to 21 , wherein the product is a beverage product.

23. The method according to clauses 19 to 22, wherein the product is coffee.

24. The method according to clauses 19 to 23, wherein the known authentic product is an arabica extract.

25. The method according to clauses 19 to 24, further comprising analysing the spectral fingerprint unique to the marker using Monte Carlo simulation.

26. The method according to clauses 19 to 25, wherein the Monte Carlo simulation of arabica extract has a statistical distribution range of between 0.5 to 99.5 percentiles.

27. The method according to clauses 19 to 25, wherein the Monte Carlo simulation of robusta coffee has a statistical distribution range above 99.5 percentile. 28. The method according to clauses 19 to 27, wherein an effective detection limit of the marker in the product is 10% w/w.

29. The method according to clauses 19 to 28, wherein the effective detection limit for 16-OMC in the product is 10% w/w. 30. The method according to clauses 19 to 29, wherein the product comprises 10- 40% of robusta coffee.

31 . The method according to clauses 19 to 30, wherein the effective detection limit for robusta coffee is 10% w/w.

32. A method for detecting a 16-OMC marker, the method comprising scanning a coffee composition having the 16-OMC using low-field NMR spectroscopy, whereby the scanning of the coffee composition results in a NMR spectral fingerprint unique to 16-OMC.

33. A database of reference spectral data generated according to the method of any one of the preceding clauses. 34. The database according to clause 33, wherein the spectral data is NMR data.

Various further aspects and embodiments of the present invention will be apparent to those skilled in the art in view of the present disclosure.

All documents mentioned in this specification are incorporated herein by reference in their entirety.

"and/or" where used herein is to be taken as specific disclosure of each of the two specified features or components with or without the other. For example "A and/or B" is to be taken as specific disclosure of each of (i) A, (ii) B and (iii) A and B, just as if each is set out individually herein. Unless context dictates otherwise, the descriptions and definitions of the features set out above are not limited to any particular aspect or embodiment of the invention and apply equally to all aspects and embodiments which are described. Certain aspects and embodiments of the invention will now be illustrated by way of example and with reference to the figures described above and tables described below. Example 1

Materials and methods

Around 17 samples of roast coffee beans were obtained from UK retailers and from the British Coffee Association, as detailed in table 1 a. The authenticity of these intact bean samples was confirmed by inspection. Combinations of these samples may be used to produce a set of 42 mixtures, as detailed in table 1 b. Finally, 27 samples of ground roast coffees, all of which displayed the labelling claim "100% arabica" on their packaging (and two of which were also labelled decaffeinated), were purchased from UK retailers (table 1 c).

Extraction of the lipophilic fraction Whole bean samples can be ground with pestle and mortar, to produce a particulate sample physically comparable to purchased ground coffees. The mixture samples were prepared from these grounds as detailed in table 1 b.

To extract the lipophilic fraction, 1 g of ground sample (authentic, mixture, or surveillance) may be mixed with 3.0ml deuterated chloroform and agitated in shaker bath for 5 minutes. For a subset of samples only (whole bean retail samples, as well as 5 of the surveillance samples), the extraction procedure can be carried out twice to allow investigation of the technical repeatability.

Table la Whole bean samples:

Number of

Number of

extracts Comments samples

prepared

Purchased from UK retailers.

Includes one decaffeinated

7 14

sample. Two extracts arabica

prepared per sample.

Supplied by British Coffee

4 4

Association

Purchased from UK retailers.

3 6 Two extracts prepared per robusta sample.

Supplied by British Coffee

3 3

Association

Total: 27

Table lb Mixtures prepared from whole bean samples:

Number of

%w/w %w/w Number of

extracts

arabica robusta samples

prepared

Series prepared from a

90, 80, 70,... 10, 20,

9 9 randomly selected pair of 10 30,... 90

whole bean samples

Series prepared from a

90, 80, 70,... 10, 20,

9 9 randomly selected pair of 10 30,... 90

whole bean samples

Prepared from various

80 20 12 12 pairwise combinations of whole bean samples

Prepared from various

60 40 12 12 pairwise combinations of whole bean samples

Total: 42

Table lc Surveillance samples (ground):

Number of

Number of

extracts Comments samples

prepared

Purchased from UK retailers. Includes two decaffeinated labelled arabica 27 32 sample. Two extracts prepared for 5 of the samples. 60 MHz H spectra

60 MHz ¹H NMR spectra may be acquired on a Pulsar low-field spectrometer (Oxford Instruments, Tubney Woods, Abingdon, Oxford, UK) running SpinFlow software (v1 , Oxford Instruments). The sample temperature can be 37 °C, and the 90 ° pulse length may be ~7.2 s as determined by the machine's internal calibration cycle. 256 FIDs can be collected from each extraction with a fixed RD of 2 s, resulting in an acquisition time of ~40 min per extract. These parameters represent an acceptable compromise between speed and spectral quality. The linewidth (chloroform FWHM) can be maintained between 0.5-0.9Hz by daily checking and shimming as and when necessary.

In all cases, the FIDs can be Fourier-transformed, co-added and phase-corrected using SpinFlow and MNova (Mestrelab Research, Santiago de Compostela, Spain) software packages to present a single frequency-domain spectrum from each extract. Where spectra were examined qualitatively, apodisation was additionally applied to the FIDs. The chemical shift scale in all spectra can be referenced to the residual chloroform peak at 7.26 ppm.

The spectral acquisition conditions may vary and can be dependent on the sample composition and/or the external magnetic field. For example, 60 MHz 1 H NMR spectra may be acquired on a Pulsar low-field spectrometer (Oxford Instruments, Tubney Woods, Abingdon, Oxford, UK) running SpinFlow software (v1 , Oxford Instruments). The sample temperature can be 37 °C, and the 90° pulse length may be 7.2 s as determined by the machine's internal calibration cycle. For each sample, 256 free induction decays (FIDs) may be collected using a filter width of 5000 Hz and recycle delay of 2 s, resulting in an acquisition time of approximately 40 min per extract. These parameters represent an acceptable compromise between speed and spectral quality. FIDs can be zero-filled to give spectra of 65536 points. The linewidth may be maintained between 0.5-0.9 Hz by daily checking of the chloroform FWHM and shimming as and when necessary. 600 MHz Ή spectra

600 MHz ¹H NMR spectrum may be collected from selected extracts only using a Bruker Avance III HD spectrometer running TopSpin 3.2 software and equipped with a 5 mm TCI cryoprobe. The probe temperature can be regulated at 27 °C. The spectra may be referenced to chloroform at 7.26 ppm.

The NMR spectra of the samples may be acquired on a high-field NMR spectrometer. For example, 600 MHz 1 H NMR spectra were collected from selected extracts using a Bruker Avance III HD spectrometer running TopSpin 3.2 software and equipped with a 5 mm TCI cryoprobe. The probe temperature may be regulated at 27 °C. For each spectrum 64 scans may be collected using 30° pulses with a spectral width of 20.5 ppm, acquisition time of 2.67 s and recycle delay of 3 s. FIDs can be zero-filled and transformed using exponential line broadening (0.3 Hz) to give spectra of 65536 points. The spectra can be referenced to the residual chloroform peak at 7.26 ppm. Data analysis

All data visualization and processing of the frequency-domain spectra may be carried out in Matlab® (The Mathworks, Cambridge, UK) installed along with the "Statistics and Machine Learning" and "Signal Processing" toolboxes.

Example 2 Concentration step of coffee samples to improve sensitivity

The present invention described the use of 3.16 ppm peak area as a proxy for the amount of robusta coffee present in the sample. The estimated detection limit may be typically at 10% w/w robusta in (impure) arabica samples. The sensitivity of the present method may be increase further, primarily through the development of a new sample preparation procedure such as concentrating the sample prior to scanning.

A solvent like chloroform, which may be deuterated, can be used to extract the lipophilic phase from ground roast coffee samples followed by a concentration step in which the chloroform may be evaporated using a vortex evaporator, and the residue dissolved in a much smaller amount of solvent for the NMR analysis. For the concentration step, a large amount of coffee (~10g) and solvent (~30ml) sample may be required for the extraction step.

The second solvent may be chloroform, which is suitable for low field NMR, or relatively high-cost deuterated chloroform, which is a preferred solvent for high field NMR. The procedure of extracting, evaporating and dissolving the sample as described throughout the specification, and in conjunction with the ability to use a different solvent at the dissolution stage, prevents the method of the present invention becoming prohibitively expensive. This can be an important consideration for a putative screening technique.

In an example, the lipophilic fraction can be extracted by taking 10 g of ground coffee and stirring (600 rpm) with 30 ml of chloroform for 5 min. The extract may be filtered through filter paper (Whatman No. 1 ) then through an empty SPE cartridge (Bond Elut) into sovirel tubes. The extract can be dried using a vortex evaporator with heating at 30°C and a pressure of 30 in Hg for 30 min. The vortex evaporator can be used to evaporate the chloroform leaving a dried extract. The dried extract may be redissolved in 800 μΙ of chloroform, which may be deuterated and filtered through cotton wool directly into NMR tubes.

In another example, a solvent like hexane or methanol, can be used to extract the lipophilic phase from coffee samples, which may be optionally followed by a concentration step in which the hexane or methanol can be evaporated using a vortex evaporator to leave a dried extract. The dried extract may be redissolved in chloroform, which may be deuterated and filtered through directly into NMR tubes.

Surveillance studies

Table 2 - Arabica and "non-arabicas" of assured origin

Sample Species/hybrid Category Cultivar (where Country of Code applicable) origin

1 arabica wild type Ethiopia

2 arabica wild type Ethiopia

3 arabica wild type Ethiopia

4 arabica wild type Yirgacheffe Ethiopia

5 arabica wild type Limu Ethiopia

6 arabica wild type Ethiopia

7 arabica wild type Ethiopia

8 arabica wild type Ethiopia

9 arabica wild type Ethiopia

10 arabica wild type Lekemte Ethiopia

11 arabica wild type Ethiopia

12 arabica wild type Ethiopia

13 arabica wild type Ethiopia

14 arabica wild type Ethiopia

15 arabica wild type Ethiopia

16 arabica wild type Ethiopia

17 arabica wild type Ethiopia

18 arabica wild Geisha Colombia type/cultivar

19 arabica cultivar Typica Colombia

20 arabica cultivar Bourbon Colombia

21 arabica cultivar Mundo Nuovo Brazil

22 arabica cultivar Castillo Colombia

23 arabica cultivar Bourbon Pointu La Reunion

24 arabica cultivar Laurina Brazil

25 arabica cultivar Uganda

26 arabica cultivar Caturra Colombia

27 arabica cultivar Caturra Costa Rica

28 arabica cultivar SL 28/SL 34 Kenya

29 arabica x cultivar Batian Kenya robusta < (hybrid)

arabica

30 arabica x cultivar uiru 11 Kenya robusta < (hybrid)

arabica

31 arabica x cultivar Arabusta Kenya robusta (hybrid)

32 canephora S. 274 India

33 canephora Indonesia canephora Vietnam canephora Brazil canephora Rwanda canephora India congesis x hybrid India robusta

liberica Uganda liberica Uganda

Table 3 - Survey of retail coffees sourced worldwide

Sample Code Country of Purchase Country of Coffee Origin as stated on label

CSOl Poland Not specified

CS02 Poland Not specified

CS03 Poland Brazil

CS11 France Not specified

CS12 France Not specified

CS13 France Not specified

CS14 France Ethiopia

CS18 The Netherlands Not specified

CS19 UK Java

CS20 UK Kenya

CS21 UK Colombia

CS22 UK Colombia

CS23 UK Kenya

CS25 UK Africa, Indonesia, Latin America

CS26 UK Not specified

CS27 UK Colombia

CS28 UK Colombia

CS29 UK Kenya

CS30 UK Not specified

CS31 UK Guatemala

CS32 UK Not specified

CS33 UK Not specified

CS34 UK Not specified CS36 Italy Not specified

CS37 France Not specified

CS39 France Not specified

CS40 Germany Not specified

CS41 France Not specified

CS42 Italy Not specified

CS43 Spain Colombia

CS44 Italy Nicaragua

CS45 USA Not specified

CS46 USA Not specified

CS47 USA Not specified

CS48 USA Not specified

CS49 USA Not specified

CS50 The Netherlands Not specified

CS51 The Netherlands Not specified

CS53 USA Not specified

CS54 USA Mexico

CS55 USA Mexico

CS56 USA Not specified

CS57 Australia Not specified

CS58 Italy Not specified

CS59 Italy Not specified

CS60 Estonia Not specified

CS61 UK Mexico

CS62 UK Costa Rica CS63 UK Brazil

CS64 UK Brazil

CS65 UK Ethiopia

CS66 Italy Costa Rica

CS67 UK Brazil, Nicaragua, Honduras

CS68 UK Brazil, Nicaragua, Honduras

DH1 UK Ethiopia

DH2 UK Brazil

DH3 UK Guatemala

DH4 UK Colombia

DH5 UK Kenya

DH6 UK Not specified

Table 4 - Surveillance study showing P-value, estimated concentration of robusta and estimated confidence intervals of samples.

Arabica and "non-arabicas" of assured origin

40 samples of coffee beans were obtained from an assured source. 30 of these were arabica coffees (18 Ethiopian wildtypes, and 12 cultivars from a range of commercially important coffee-producing countries). Complete details of these samples are given in Table 2. Survey of retail coffees sourced worldwide

60 samples of ground roast coffees were purchased by IFR staff, students and collaborators from a range of outlets in 1 1 different countries. All displayed the labelling claim "100% arabica" (or equivalent in the relevant local language). The geographic origins of the coffees as stated on the labels covered 1 1 different coffee-growing countries and represented all producing continents. Details of each sample are given in Table 3. All samples were supplied to IFR's Analytical Sciences Unit in original unopened packaging.

Referring to Figures 6a, 6b and 6c, the Figures show 60 MHz NMR spectra obtained from two lipophilic extracts prepared from a sample of robusta coffee beans. Figure 6a shows two NMR spectra; spectrum A and spectrum B. Spectrum A is an extract as prepared using the method of the present invention, which involves a concentration step. For comparative purposes, spectrum B is a 1 D- NMR spectrum of an extract without the concentration step. The figure shows that the concentrated extract produces an NMR spectrum with larger peaks and greater signal-to-noise from a given set of spectral acquisition conditions.

In Figures 6b and 6c, there is shown a region of the NMR spectra at around the 16-OMC (-3.16 ppm) and caffeine peaks (-3.38, ~3.58ppm). Figure 6b shows the spectrum of the concentrated sample and Figure 6c shows the NMR spectrum of the un-concentrated sample, respectively. These are plotted internally normalized to the integrated glyceride peaks in each case; note the same y-scale for both Figures 6b and 6c. The results show the concentrated sample provides clearer NMR spectra i.e. less signal to noise ratio than the spectra of the un-concentrated sample.

Referring to Figure 7a, there is provided a graph showing the calibration relating to the concentration of robusta in a sample to the integrated peak area of 3.16 ppm peak. In Figure 7b, there is shown a 1 D-NMR spectral region of between 3.1 ppm to 3.7 ppm. The NMR spectra of the measured samples are stacked in increasing order of robusta content. The first three NMR traces (i.e. bottom three traces) in Figure 7b comprise 0% w/w robusta.

By concentrating the sample prior to the scanning step, the detection limit can change from 10% w/w to around 1 % w/w robusta in a mixture of arabica and robusta samples. This is evidenced by the ability to calibrate for the amount of robusta in prepared arabica/robusta mixtures using the 3.16 ppm peak area as shown in Figures 7a and 7b. The calibration is well-behaved and linear, with an R² value of 0.99 and a prediction root-mean-square error (RMSE) of +/- 0.6% w/w.

The effective detection limit of 1 % level may be an important milestone because it may generally be accepted to mean intentional substitution or adulteration of food products for economic gain, rather than simply adventitious contamination during normal processing. Even substituting, modifying or adding robusta in arabica samples at a rate of a few percent could yield substantial economic advantage, considering the price differential between the two species and the amount of coffee traded.

It has been proposed that for many products, 1 % w/w of adulterant is a reasonable cut-off for making this distinction. With regards to the undeclared presence of robusta in arabica samples, the 1 % w/w detection level represents a reasonable threshold. A range of test mixtures were prepared by the inventors to validate the calibration, from different pairwise combinations of original coffee samples. For test samples containing 18% w/w robusta or less (i.e. the range of the calibration), errors in prediction were no greater than 3% w/w. This is greater than the RMSE for the calibration series, but this may be expected, since different coffees were involved rather than a single pair used to prepare all the calibration samples.

Previous convention in the prior art is to regard 16-OMC as absent from arabica coffees. However, the inventors have shown the improvements in the sensitivity of the disclosed method i.e. by concentrating the sample prior to the scanning step, also led to an unexpected result: the discovery of low levels of 16-OMC in coffee samples such as arabica samples. The evidence for this is hinted at in the nonzero intercept of the regression line in Figure 7a, and in the small feature discerned in the lowest three traces of Figure 7b, which are spectra prepared from a 100% arabica coffee from a trusted source.

Referring to Figure 8, there is provided a calibration used to estimate the robusta content of 60 ground roast coffee samples sourced from a range of retailers around the world. Of the 60 samples, 52 samples were predicted as having robusta contents of 1 % w/w or less, and a further 3 samples as having robusta contents in the range 1 % to 3% w/w. Given the accuracy and precision of the calibration, there appears to be no evidence that these samples contain any robusta coffee.

However, the remaining five samples were predicted to have robusta contents of 5%, 17%, 19%, 21 % and 40% w/w. These results can be confirmed by applying high-field NMR spectroscopy and/or by low-field NMR spectroscopy on the samples. The estimated percentages are far greater than could reasonably be found by normal contamination during processing. In conclusion, the five coffee samples seem to be subject to fraudulent partial substitution with robusta at some point in their production.

Referring to Figures 9a and 9b, there is shown a theoretically calculated probability functions for a surveillance study of 60 roasted coffee samples. The surveillance exercise provides an estimate of the prevalence of fraud in the sector overall. The chance of obtaining exactly five instances of fraud from a study size of 60 samples is calculated from the binomial theorem and plotted as a function of prevalence in Figure 9a. The cumulative distribution is given in Figure 9b, from which the estimated prevalence of fraud across the coffee sector to be between 4- 17% (90% confidence interval). In Figures 16a to 16c, the estimated prevalence of fraud across the coffee sector is between 5 - 20% (95% confidence interval).

Referring to Figures 10a and 10b, there is shown a region of interest in the 60Mz (Figure 10a) and 600MHz (Figure 10b) spectral collections. Figures 10a and 10b show the region around the 3.16 ppm peak in the spectra acquired from 30 arabica coffees of assured origin at 60 MHz and 600 MHz, respectively. In both Figures the spectra have been internally normalized to the glyceride peaks to facilitate side-by-side comparison on the same vertical scale. The spectra were collected from all 30 arabica coffees of assured origin as detailed in Table 2. As shown in Figures 10a and 10b, the presence of the 3.16 ppm 16-OMC peak was confirmed in every sample, irrespective of genetic background and provenance. Of particular note are four of the wildtype samples (codes 1 , 13, 14 and 16) which appear to show much larger 16-OMC contents than the other samples. The extraction procedure and spectral acquisition at 60 MHz and 600 MHz field strengths can be repeated for these samples; these technical replicates are indicated by brackets on Figures 10a and 10b. These samples are all from marginal coffee-growing areas, with 13, 14 and 16 in particular from much drier locations than other samples in the collection. The rainfall pattern may also be bimodal (as it is for some of the other sample locations, e.g. Yirgacheffe, although these are much wetter overall). The obtained data presented in Figures 10a and 10b may be from low-producing areas and rarely, if ever, do the coffees grown here get utilised in commercial, commodity-type coffee.

Referring to Figure 1 1 a, there is shown an integrated 3.16 ppm peak area in 60MHz spectra from the assured source arabicas. Samples 1 , 13, 14 and 16 originate from atypical coffee-growing locations. Replicate measurements made on repeat extractions from these samples are indicated by joined points.

Figure 1 1 b provides a normal probability plot for the data in Figure 1 1 a (excluding the atypical samples). As illustrated in Figure 1 1 b, the 26 arabica coffees from typical coffee-growing regions form a well-behaved normal population with regards to the 3.16ppm peak. This may enable the user to choose an upper threshold for the peak area, at a desired probability level, for use in a test to verify the authenticity of arabica coffees.

Figure 1 1 c illustrates the empirical and fitted cumulative distribution functions for typical arabica coffees. The value of the integrated peak area corresponding to the 95th percentile is marked on Figure 1 1 c, and also on Figure 1 1 a. Referring to Figures 12a, there is shown the low concentration region in a calibration chart developed to estimate the concentration of adulterant (robusta or other non-arabica) present in samples that fail to be accepted as authentic arabicas. As shown in Figure 12a, the open and closed markers indicate the actual and predicted concentrations for the mixture series, with the error indicated by vertical lines. The threshold value shown in Figure 12a is marked on the horizontal axis. For samples with a peak area below this value, the null hypothesis (that the sample is an authentic arabica) may be accepted. For peak areas above this value, the sample may be considered suspicious, and the calibration can be used to estimate its non-arabica content.

Referring to Figure 12b, there is provided an integrated 3.16 ppm peak areas for the 60 surveillance samples. The right-hand vertical axis is an equivalent concentration scale obtained from the calibration line in Figure 12a. As shown in Figure 12b, 8 samples have 3.16 ppm peak areas above the threshold value and can be rejected as authentic arabicas (p < 0.05). Of these, 2 are only slightly above the threshold value. The other 6, however, have estimated non-arabica contents ranging from 3 - 33% w/w. This can be seen by reading values from the right-hand vertical axis in Figure 12b, which indicates an equivalent concentration scale obtained using the established calibration line.

These values are considered to be in excess of what could reasonably be claimed as adventitious contamination. Thus, the authors suggest this is highly indicative of fraudulent substitution, or at the very least, unacceptably poor quality control. For three of these samples, sufficient beans were available to carry out repeat extractions. The outcomes for these are also shown on the plot and detailed in Table 4, again indicating the very high reproducibility of the methods disclosed in the present invention.

Referring to Figure 13a and 13b, there is provided a plot showing the concentration of robusta in the mixture series versus integrated area of the 3.16 ppm in the 600 MHz, and the associated simple linear regression line. Figure 13b show the spectra used to obtain the calibration and are shown as a stacked plot for clarity purposes.

Both Figures 13a and 13b illustrate the samples contain a small but consistent feature at 3.16ppm on the 600 MHz spectrometer. The clear non-zero intercept of the regression line shown in Figures 13a appear to be in agreement with the results obtained on a 60 MHz spectrometer, as shown in Figures 7a and 7b.

Figures 14a, 14b and 14c show plots of the 3.16 ppm peak area in 60 MHz spectra versus 600 MHz spectra for 18 mixture series extracts (Figure 14a); extracts from the arabica of assured origins (Figure 14b) and extracts from the 60 survey samples (Figure 14c). The correlation or pattern of results between the 60 MHz and 600 MHz peak area values in analogous experiments were found to be similar. Thus, for the analysis of a single, isolated resonance such as the 3.16 ppm peak, there appears to be little benefit in carrying out NMR spectroscopy at the higher field strength. The precision and accuracy of the spectral measurements can be substantively the same, whereas the 60 MHz approach offers considerable advantages in terms of ease-of-use, as well as lower capital and maintenance costs.

Figures 15a and 15b provide a calibration chart developed from 26 different arabica coffees and 10 different "non-arabicas" (robusta and other coffee species). The calibration line indicates the median of the regression lines obtained by simple linear regression onto all possible pair-wise combinations of arabica and non- arabicas. Various percentiles are also indicated in Figures 15a and 15b, which can be used to estimate confidence intervals for predicted concentrations. This approach may exploit the excellent linearity of NMR peak areas as a function of concentration, as demonstrated in Figures 7a and 7b.

In addition, the known concentrations versus peak areas for 28 samples (the mixture series, and other assorted pairs of samples) are also marked in Figure 15a. The predicted values for these samples are shown in Figure 15b, along with the error in prediction and estimated confidence interval. Referring to Figure 16a, there is shown a planning survey graph. The graph in Figure 16a shows the number of samples that need to be examined as a function of the (a priori unknown) fraud prevalence, in order to achieve 99%, 95% or 90% chance of seeing at least one instance of fraud. The probability of finding some instances of fraud in a surveillance study depends on the number of coffees examined and the prevalence of fraud in the sector. As shown in Figure 16a, finding some instances of fraud in a surveillance study can be calculated using the binomial sampling theorem for different survey sizes and prevalences. In cases of fraud which have been discovered by the surveillance study as shown in Figure 16a, an estimate can be made a posterior of the likely fraud prevalence. As illustrated in Figure 16b, the functions show the relative chance of obtaining exactly Q cases of fraud at a range of prevalence rates. Figure 16c shows the cumulative distribution functions for cases Q=6 and Q=8, for which the estimated 95% confidence intervals for the fraud prevalence are 5-20% (Q=6) and 7-23% (Q=8).

References

Belitz, H. D., Grosch, W., & Schieberle, P. (2009). Food chemistry {4 ed.). Berlin Heidelberg: Spinger.

Cagliani, L. R., Pellegrino, G., Giugno, G., & Consonni, R. (2013). Quantification of Coffea arabica and Coffea canephora var. robusta in roasted and ground coffee blends. Talanta, 106, 169-173.

Charlton, A. J., Farrington, W. H. H., & Brereton, P. (2002). Application of H-1 NMR and multivariate statistics for screening complex mixtures: Quality control and authenticity of instant coffee. Journal of Agricultural and Food Chemistry, 50(1 1 ), 3098-3103.

Consonni, R., Cagliani, L. R., & Cogliati, C. (2012). NMR based geographical characterization of roasted coffee. Talanta, 88, 420-426.

D'Amelio, N., De Angelis, E., Navarini, L, Schievano, E., & Mammi, S. (2013).

Green coffee oil analysis by high-resolution nuclear magnetic resonance spectroscopy. Talanta, 110, 1 18-127.

DIN 10779 (201 1 ) Analysis of coffee and coffee products - Determination of 16-0- methyl cafestol content of roasted coffee - HPLC-method): German Institute for Standardization (DIN). Guillen, M. D., & Ruiz, A. (2001 ). High resolution H-1 nuclear magnetic resonance in the study of edible oils and fats. Trends in Food Science & Technology, 12(9), 328-338.

Monakhova, Y. B., Ruge, W., Kuballa, T., Use, M., Winkelmann, O., Diehl, B., Thomas, F., & Lachenmeier, D. W. (2015). Rapid approach to identify the presence of Arabica and Robusta species in coffee using H-1 NMR spectroscopy. Food Chemistry, 182, 178-184.

Parker, T., Limer, E., Watson, A. D., Defernez, M., Williamson, D., & Kemsley, E.

K. (2014). 60 MHz H NMR spectroscopy for the analysis of edible oils. Trends Analyt Chem, 57(100), 147-158.

Scharnhop, H., & Winterhalter, P. (2009). Isolation of coffee diterpenes by means of high-speed countercurrent chromatography. Journal of Food Composition and Analysis, 22(3), 233-237.

Schievano, E., Finotello, C, De Angelis, E., Mammi, S., & Navarini, L. (2014).

Rapid Authentication of Coffee Blends and Quantification of 16-O-

Methylcafestol in Roasted Coffee Beans by Nuclear Magnetic Resonance. Journal of Agricultural and Food Chemistry, 62(51 ), 12309- 12314.

Speer, K., & Koelling-Speer, I. (2006). The lipid fraction of the coffee bean. Brazilian Journal of Plant Physiology, 78(1 ), 201 -216.

Speer, K., & Mischnick, P. (1989). 16-O-METHYLCAFESTOL - A NEW

DITERPENE IN COFFEE - DISCOVERY AND

IDENTIFICATION. Zeitschrift Fur Lebensmittel-Untersuchung Und- Forschung, 189(3), 219-222. Wei, F., Furihata, K., Koda, M., Hu, F., Miyakawa, T., & Tanokura, M. (2012).

Roasting Process of Coffee Beans as Studied by Nuclear Magnetic Resonance: Time Course of Changes in Composition. Journal of Agricultural and Food Chemistry, 60(4), 1005-1012. Wei, F. F., Furihata, K., Hu, F. Y., Miyakawa, T., & Tanokura, M. (201 1 ). Two- Dimensional H-1 -C-13 Nuclear Magnetic Resonance (NMR)-Based Comprehensive Analysis of Roasted Coffee Bean Extract. Journal of Agricultural and Food Chemistry, 59(17), 9065-9073.

Claims

1 . A method for detecting a marker in coffee, the method comprising

(i) adding solvent to the coffee to prepare a sample;

(ii) scanning the sample using low-field NMR spectroscopy to generate a spectral fingerprint, and;

(iii) analysing the spectral fingerprint, wherein the presence of a fingerprint unique to the marker indicates the presence of the marker, wherein the marker is caffeine or 16-OMC.

2. A method according to claim 1 comprising

(a) extracting a lipophilic composition of the coffee;

(b) concentrating the extract to produce a test sample;

(c) scanning the test sample using low-field NMR spectroscopy to generate a spectral fingerprint; and

(d) comparing the spectral fingerprint of the test sample to the spectral fingerprint of the known authentic sample to detect the presence of the marker, wherein the marker is caffeine or 16-OMC.

3. The method according to any one of the preceding claims, wherein the coffee is a mixture of arabica and robusta extracts.

4. The method according to any one of the preceding claims, wherein the solvent is chloroform.

5. The method according to any one of the preceding claims, wherein the scanning step (ii) comprises applying a radio-frequency pulse to the sample and detecting a free induction decay using an NMR spectrometer.

6. The method according to claim 5, wherein the free induction decay is Fourier transformed.

7. The method according to any one of the preceding claims, wherein the spectral fingerprint is a 1 D spectrum.

8. The method according to claim 7, wherein the spectral fingerprint is a proton 1 D spectrum.

9. The method according to any one of the preceding claims, wherein the low- field NMR spectroscopy has a magnetic strength of 40 to 100 MHz.

10. The method according to any one of the preceding claims, wherein the low- field NMR spectroscopy has a magnetic field strength is 60 MHz.

1 1 . The method according to any one of the preceding claims, wherein the 16- OMC is present in robusta extracts or a mixture comprising robusta extracts thereof.

12. A method of enabling verification of the authenticity of a coffee product, the method comprising scanning the product using low-field NMR spectroscopy and identifying a marker of the product in a spectrum wherein the marker is caffeine or 16-OMC.

13. The method according to claim 12, further comprising scanning the product or an example of the product using low-field NMR spectroscopy to generate spectroscopic data which acts as a signature of the product, thereby enabling the authenticity of a product to be determined by comparing the signature of the product with a signature of a known authentic product.

14. The method according to claim 12 or claim 13, further comprising analysing the spectral fingerprint unique to the marker using Monte Carlo simulation.

15. The method according to claim 14, wherein the Monte Carlo simulation of arabica extract has a statistical distribution range of between 0.5 to 99.5 percentiles.

16. The method according to claim 15, wherein the Monte Carlo simulation of robusta coffee has a statistical distribution range above 99.5 percentile.

17. The method according to any one of claims 12 to 16, wherein an effective detection limit of robusta to arabica in the product is 10% w/w.

18. The method according to any one of claims 12 to 16, wherein an effective detection limit of robusta to arabica in the product is 1 % w/w.