Variation in Caffeine Concentration in Single Coffee Beans

Subscriber access provided by MICHIGAN STATE UNIVERSITY | MSU LIBRARIES

Journal of Agricultural and Food Chemistry is published by the American ChemicalSociety. 1155 Sixteenth Street N.W., Washington, DC 20036Published by American Chemical Society. Copyright © American Chemical Society.However, no copyright claim is made to original U.S. Government works, or worksproduced by employees of any Commonwealth realm Crown government in the courseof their duties.

Article

Variation in caffeine concentration in single coffee beansGlen p. Fox, Alex Wu, Yiran Liang, and Lesleigh Force

J. Agric. Food Chem., Just Accepted Manuscript • Publication Date (Web): 27 Sep 2013

Downloaded from http://pubs.acs.org on September 29, 2013

Just Accepted

“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are postedonline prior to technical editing, formatting for publication and author proofing. The American ChemicalSociety provides “Just Accepted” as a free service to the research community to expedite thedissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscriptsappear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have beenfully peer reviewed, but should not be considered the official version of record. They are accessible to allreaders and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offeredto authors. Therefore, the “Just Accepted” Web site may not include all articles that will be publishedin the journal. After a manuscript is technically edited and formatted, it will be removed from the “JustAccepted” Web site and published as an ASAP article. Note that technical editing may introduce minorchanges to the manuscript text and/or graphics which could affect content, and all legal disclaimersand ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errorsor consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Variation in caffeine concentration in single coffee beans 1

2

Glen P Foxa*, Alex Wu

b, Liang Yiran

b and Lesleigh Force

b 3

4

aThe University of Queensland, Queensland Alliance for Agricultural & Food 5

Innovation, Toowoomba, Qld 4350 Australia. 6

bSchool of Agriculture and Food Sciences, The University of Queensland, St Lucia 7

Qld 4072, Australia. 8

9

10

*Corresponding Author: Dr Glen Fox 11

Tel: +61 746398830 12

Fax: +61 746398800 13

Email [email protected] 14

15

16

Page 1 of 28

ACS Paragon Plus Environment

Journal of Agricultural and Food Chemistry

ABSTRACT 17

Twenty eight coffee samples from around the world were tested for caffeine levels to 18

develop near infrared reflectance spectroscopy (NIRS) calibrations for whole and 19

ground coffee. Twenty-five individual beans from five of those coffees were used to 20

develop a NIRS calibration for caffeine concentration in single beans. We used an 21

international standard high performance liquid chromatography method to analyse for 22

caffeine content. Coffee is a legal stimulant and possesses a number of heath 23

properties. However, there is variation in the level of caffeine in brewed coffee and 24

other caffeinated beverages. Being able to sort beans based on caffeine concentration 25

will improve quality control in the level of caffeine in those beverages. There was a 26

range in caffeine concentration was from 0.01 mg/g (decaffeinated coffee) up to 19.9 27

mg/g (Italian coffee). The majority of coffees were around 10.0 mg/g to 12.0 mg/g. 28

The NIRS results showed r2 values for bulk unground and ground coffees were greater 29

than 0.90 with standard errors less than 2 mg/g. For the single bean calibration the r2 30

were between 0.85 and 0.93 with standard errors of cross validation of 0.8 to 1.6 mg/g 31

depending upon calibration. The results showed it was possible to develop NIRS 32

calibrations to estimate the caffeine concentration of individual coffee beans. One 33

application of this calibration could be sorting beans on caffeine concentration to 34

provide greater quality control for high-end markets. Further, bean sorting may open 35

new markets for novel coffee products. 36

37

Page 2 of 28



INTRODUCTION 38

Coffee is one of the most-consumed beverages in the world with aroma and taste 39

being the main sensory characteristics. The dominant commercial species of coffee 40

are Arabica (Coffea arabica L.) and Robusta (C. canephora Pierre) and most 41

commercial coffee beverages are produced from these roasted beans or blends of 42

these two. However, there are differences in quality and sensory profiles between 43

these species, and suitable analytical methods and trained taste panels are required to 44

quantify these differences. 45

In regards to quantifying chemical components, such as lipids, protein or 46

caffeine concentration, expensive and time-consuming wet chemistry methods are 47

required. However, near infrared reflectance spectroscopy (NIRS) has been used as a 48

high through-put and cheap surrogate for quantification of quality (including caffeine 49

concentration) 1-7 and sensory parameters

8. The differences detected in the chemical 50

composition within and between Arabica and Robusta varieties can be useful for 51

quality classification purposes and NIRS has been used successfully in this context 1. 52

In addition, NIRS has also been used for quantification of green bean quality 9, 10

and 53

to understand genetic, environmental and processing effects on quality 11. 54

In most biological studies, a bulk sample is tested to obtain a representative 55

mean value, although there is little information on the variation within the sample and 56

the impact of the variation on the result. Several studies in cereal crops investigated 57

the evaluation of single grains to ascertain the variation within a sample for a range of 58

attributes using NIRS 12-24

. To date, there has been no report of assessment of single 59

coffee beans by NIRS. However, single coffee beans have been tested for specific 60

purposes including optimising a wet chemistry method for caffeine concentration 25 61

and identifying breeding lines for reducing caffeine concentration 2, 26

. 62

Page 3 of 28



The aim of our study was to develop NIRS calibrations for caffeine 63

concentration in bulk unground and ground coffees as well as single coffee beans 64

using a high performance liquid chromatography (HPLC) method as the reference 65

method to quantify caffeine levels. The NIRS calibration in bulk unground and 66

ground coffees would provide a rapid, predictive method of caffeine concentration. A 67

NIRS calibration on single beans would provide the opportunity to understand the 68

potential variability in coffee samples, e.g. blends, or sort beans based on caffeine 69

concentration to further quantify quality and potentially develop new products (e.g. 70

naturally low caffeine coffee). 71

72

MATERIALS AND METHODS 73

Twenty-eight roasted Arabica and blended whole bean coffees, purchased from retail 74

outlets, from a number of countries were used. Most of the coffees were Arabica type 75

as per the information on the packet. Unfortunately, some of packets lacked specific 76

or complete information on the origin or blend of the coffee. Table 1 shows the types 77

and origins of coffees. 78

79

Ground coffee 80

A 20 g sub-sample from each coffee was ground in a standard electric coffee grinder 81

for NIR scanning and HPLC analysis. The grinder was cleaned with a fine brush 82

between coffees. 83

84

FT-NIR 85

Three forms of coffees (bulk unground, ground and single beans) were scanned in a 86

Fourier Transform (FT) NIR multipurpose analyzer instrument (Bruker, Germany) 87

Page 4 of 28



from 4000 to 12000 cm at 8 cm resolution. For bulk unground and ground coffees, 88

the 97 mm cup on the rotating sphere was used, where 64 scans of the rotating cup 89

were averaged into a single spectrum. Coffees were scanned in duplicate. 90

91

Five coffees were selected for single bean analysis (coffees numbered 5, 7, 11, 15, 92

and 24 in Table 1). Four samples were blends which would provide a broad range 93

within each coffee. A single Arabica was also included. Twenty-five individual beans 94

were randomly selected from each of these five coffees and scanned in a 19 mm glass 95

vial using the microsphere attachment. A chrome insert was placed on top of each 96

bean to prevent light escaping, as per Bruker recommendations. The wavenumber 97

range and resolution were the same as for the bulk unground and ground coffees. 98

After scanning, each bean was sealed in a plastic bag and stored in the freezer prior to 99

grinding for caffeine analysis. 100

101

Opus (V 7.0) was used for calibration development with all pre-treatments selected in 102

the Optimise process to find optimal calibration math treatment and factors. Savistky-103

Golay, with 1st and 2

nd derviatives were used with 4, 4 gap and smoothing. Individual 104

spectra files were exported in data point table format (Opus software), loaded into 105

Excel, then imported into Unscrambler (V 9.8, [Camo, Sweden]) for calibration 106

development. The calibrations developed from these two multivariate software 107

programs were compared using the Fearn 27 calculation which tests for significant 108

differences (P <0.05) between the standard deviation of prediction errors of predicted 109

values (from cross validation) using both packages. 110

111

Caffeine analysis 112

Page 5 of 28



For the ground coffees, 2.000 g of coffee was weighed out on a three decimal place 113

analytical balance. For single beans, each bean was ground in the coffee grinder and 114

recovered. The recovered coffee was weighed and used for caffeine extraction. 115

116

The weighed coffee was dissolved in boiling water, mixed well and kept heated for 10 117

minutes with constant stirring. The coffee solution was quickly cooled in iced water 118

then equilibrated at room temperature. The coffee solution was made up to 100 ml in 119

a volumetric flask, mixed well and then filtered through a No.42 filter paper. Five ml 120

of the filtrate was further filtered through a 0.45 µm mesh nylon membrane, and 121

diluted 1 in 100 ml in a volumetric flask for HPLC analysis. 122

123

A Waters HPLC system was used for caffeine quantification. Caffeine was separated 124

in an Alltech Prevail C18, 150 mm × 4.6 mm, 5 µm column with a guard column. 125

Column temperature was 30oC and injection volume 10 µl. The mobile phase was 126

40% methanol and 0.5% phosphoric acid. The final pressure in the instrument was 127

119 - 122 bar with the detector set at 254 nm. Caffeine was identified by matching 128

peak retention time against a caffeine standard and quantified by plotting against the 129

standard curve of prepared caffeine standards at 1, 5, 10 and 20 mg/ml. 130

131

RESULTS 132

Range in caffeine between coffees 133

Of the 28 coffees collected in this study, one (No. 15) was a decaffeinated coffee 134

while all others were caffeinated. The caffeine concentration of the decaffeinated 135

coffee was 0.01 mg/g (Table 1). The range in caffeine concentration for the other 27 136

samples was from 10.6 mg/g and 19.9 mg/g (Table 1). Most coffees were between 137

Page 6 of 28



10.6 mg/g and 12.8 mg/g whereas coffees 1 and 5 were much higher (17.7 mg/g and 138

19.9 mg/g, respectively). These two high caffeine coffees were obtained from Italy 139

and, as per the information on the packet, were identified as blends. 140

141

Range in caffeine within selected coffees for single bean analysis 142

Five of the 28 samples were selected to investigate variation in caffeine concentration 143

within each sample. Table 2 shows the HPLC caffeine concentrations for the 25 144

beans from the five coffees. For all coffees used in the single bean analysis, the 145

calculated average of the 25 individual beans was slightly less than the result for the 146

coffee when tested as a bulk ground coffee (Table 2). However, all of these five 147

coffees had individual kernels well above the bulk ground average. For the 148

decaffeinated coffee, all beans contained trace amounts of caffeine. While the high 149

caffeine coffee had eight individual beans above 20 mg/g with one kernel at 25.3 150

mg/g. The distribution plots of caffeine concentration of each of the individual five 151

coffees is shown in Figure 2. 152

153

Figure 3 shows the distribution plots for weight of beans used for calibration 154

development. A normal distribution was seen in bean weights for caffeinated and 155

decaffeinated beans. Figure 4 a and b show the correlation plots between caffeinated 156

beans and decaffeinated beans, respectively, when compared to bean weights. There 157

was a positive in caffeine between caffeinated beans and bean weights (R2 = 0.31). 158

There was a similar positive relationship between been weight and caffeine 159

concentration in decaffeinated beans (R2 = 0.34). These relationships indicated 160

around 30% of bean weight is associated to caffeine concentration which would 161

suggest around 70% of the caffeine concentration is not explained by weight. Hence, 162

Page 7 of 28



selecting for weight would not assist in calibration development if trying to select for 163

heavier or lighter beans. 164

165

NIR calibrations for bulk unground and ground coffees 166

For all calibrations developed, either a standard normal variation (SNV) first 167

derivative or multiplicative scatter correction (MSC) first derivative was selected as 168

best. All calibrations for bulk unground and ground beans had greater than eight 169

factors. The Opus calibration for bulk unground and ground coffees resulted in high 170

coefficients of determination of 0.93 and 0.95, respectively (Table 3). While in 171

Unscrambler the r2 values were 0.92 and 0.95 respectively. The r

2 values for internal 172

cross-validation were equal to or greater than 0.92 for bulk unground or ground 173

coffees. The NIRS correlation plots for the whole bulk unground and ground coffees 174

are shown in Figures 4a and 5b respectively. 175

176

The inclusion of a single decaffeinated coffee (coffee 15), with a zero level of caffeine 177

well below the next coffee at 10.5 mg/g, could have a major influence on the overall 178

regression. Hence, the calibrations were repeated with the decaffeinated coffee 179

excluded. The cross validation r2 values were again greater than 0.90 for unground 180

and ground coffees using both multivariate analysis programs. 181

182

All standard error of calibration (SEC) results for whole bulk unground and ground 183

coffees were less than 1.2 mg/g and 1.3 mg/g respectively by both modelling 184

packages (Table 3). The SECs for the calibrations excluding the decaffeinated coffee 185

showed a slight improvement. 186

187

Page 8 of 28



Both multivariate analysis packages allowed for internal cross-validation. The 188

internal standard error of cross-validation (SECV) for both unground and ground 189

coffee was less than 1.3 mg/g (Table 3). The SECV decreased slightly when the 190

decaffeinated coffee was excluded from the calibration set. 191

192

A ratio of standard error prediction to standard deviation (RPD) is used to indicate the 193

potential application value of the calibration 28. For RPD values greater than 2.5, the 194

calibrations would be suitable for screening lines in breeding programs. For values 195

greater than 5, the calibration would be useful in quality control. For the unground 196

and ground calibrations using both multivariate programs, the RPDs were calculated 197

at greater than 4.0 for unground and ground coffees (Table 3), suggesting reasonably 198

good calibrations for both product types. This was also the case when the 199

decaffeinated coffee was excluded. 200

201

The bias for each calibration was also less than 0.1 which is acceptable based on the 202

suggestion of Shenk, ie bias less than 0.6 x SEC. 203

204

Fearn’s statistic 205

Fearn proposed a calculation to compare if standard deviations of prediction errors 206

and bias’ between two different calibrations were significantly different when 207

compared to reference values at P <0.10, P <0.05 or P <0.01 levels . Using the Fearn 208

calculation, updated by Reid and Guthrie , there was no significant difference (P 209

<0.05) between the standard deviations of prediction errors and bias’ for the 210

calibration developed using the two multivariate packages for unground and ground 211

coffee calibrations when the predicted values were compared to the reference values. 212

Page 9 of 28



213

NIR calibrations for single beans 214

The calibrations for the single beans produced high r2 greater than 0.94 with both 215

multivariate packages with the decaffeinated coffee included. When the decaffeinated 216

coffee was excluded the Opus program gave a r2 greater 0.9304 while the 217

Unscrambler package gave a r2 = 0.9275. The r

2 values for cross-validation were 218

greater than 0.91 from both programs when using all five coffees. However, when the 219

decaffeinated coffee was excluded, the r2 were considerably lower (Table3). 220

221

For single beans, the SECs and SECVs were similar to those for the unground and 222

ground coffee calibrations. The RPD(CV) were also similar to the unground and 223

ground coffee RPDs with the calibrations with all five coffees included being higher 224

than the RPDs for calibrations with the decaffeinated coffee excluded. 225

226

The Fearn statistic showed no significant difference (P <0.05) for standard deviation 227

of prediction errors and bias’ for predicted values from both multivariate packages 228

when compared to the reference values. 229

230

Loading plots/wavelengths associations 231

The raw spectra and 1d SNV spectra for the pure caffeine are shown in Figure 6a and 232

6b, respectively. The structure of caffeine is shown in Fig.1, with obvious C=C, C=N, 233

O=C-NR and CH3 bonds. These combination bonds are strongly associated with 234

regions beyond 5000 cm (2200 nm). The Opus loading plots from the calibrations for 235

bulk unground, ground coffees and single beans are shown in Figs 7a, 7b and 7c, 236

respectively. These loading plots show a number of strong associations in regions 237

Page 10 of 28



associated with proteins and oils. In particular, there are a number of regions 238

associated with NH amino acids groups and urea, and C=O associated with oils 239

groups. 240

241

Discussion 242

The development of a NIRS calibration to predict caffeine content is dried coffee has 243

been reported previously and using more samples than in this study 11, 29

. However, 244

those studies used a relatively small number of samples but expanded the calibration 245

set by either having samples from multiple locations or by collecting samples from 246

different processing techniques, respectively. Our calibration used only 28 samples 247

but included two samples from Australia which is the first time caffeine content has 248

been reported for Australian grown coffee. These samples were within the average 249

range of caffeine when compared to international samples. 250

251

The novel aspect of our research was the development of a single bean NIRS 252

calibration to predict caffeine concentration. We believe this to be the first report for 253

a NIR calibration to predict caffeine concentration in single coffee beans. Even with 254

the five different coffees used for the NIRS calibration, there was considerable 255

variation in caffeine concentration. The variation in caffeine concentration was not 256

dependent upon bean weight as only around 30% of caffeine concentration was 257

correlated to bean weight which indicated selecting for bean weight would not 258

enhance selecting for caffeine concentration. 259

260

Our results for unground bulk and ground bulk coffee for caffeine concentration were 261

similar to other researchers, who reported r2 values greater than 0.9, with SECV or 262

Page 11 of 28



SEP around 1 to 2 mg/g. The potential application value of unground and ground 263

bulk calibrations were similar based on the RPD values despite the low number of 264

samples in our calibrations and only using a cross-validation approach. The ground 265

coffee calibrations were the best models from both multivariate packages based on the 266

reduced and more uniform particle size. However, in the practical sense, the potential 267

uses of NIR for any coffee trait would be of more value when predicting and sorting 268

unground beans. As part of a quality control program at packaging facilities where 269

beans are ground and undergo vacuum sealed packaging, then calibrations for ground 270

coffee could be used. 271

272

The calibrations developed used a number of spectral regions associated with the 273

major bonds in caffeine (C=O, C=C, O=C-NR). These molecular bonds are also in 274

protein and oil, which make up around 30% in coffee so it could be expected to see 275

these regions used in the calibration. When observing the spectra for pure caffeine, 276

these regions also appeared as major peaks on a 1d spectra. There are a number of 277

regions identified which were associated with urea O=C-((NH2)2). The C=N bonds 278

have also been identified in a study of hydrogen cyanide (HCN) in forage sorghum 30 279

In our study, we compared two multivariate programs, one being the software of the 280

FT-NIR instrument and the other being a commercial package which can use any type 281

of multivariate data. There were some differences in the final calibrations developed 282

using each package. However, as a real world application using the software 283

dedicated to operating the instrument is a better option. Exporting data into dedicated 284

multivariate packages provides the scope to understand how good a calibration could 285

be. 286

287

Page 12 of 28



The development of a NIR calibration to predict caffeine concentration in single 288

coffee beans has shown the potential to understand the variation in pure Arabica 289

coffees or blended coffees. In addition, to produce a decaffeinated coffee, the 290

caffeine is removed from beans usually by chemical extraction. Using a non-291

destructive technique such as NIRS could assist processing in knowing the in-situ 292

concentration of caffeine in individual beans which could then in turn help optimise 293

the decaffeination process. Alternatively, this tool could be used to identify and sort 294

naturally low (or high) caffeine beans to develop new coffee products and open new 295

markets. 296

297

While caffeine may not contribute to the major flavour (bitterness) of coffee, 298

consumers purchase coffee with some anticipation of a stimulating effect from 299

caffeine. Based on our results, it would be possible to build a calibration to sort beans 300

on caffeine concentration. However, current commercial NIR sorting technology is 301

quite limited in applications and costly and further research and development is 302

required. 303

304

Acknowledgements 305

All coffee were purchased except one sample donated by Green Cauldron Coffee, 306

Federal, NSW, Australia. The authors wish to thanks Dr Heather Smyth for review of 307

the manuscript. 308

309

LITERATURE CITED 310

1. Downey, G.; Boussion, J. Authentication of coffee bean variety by near-311

infrared reflectance spectroscopy of dried extract. J Sci. Food Agric 1996,71, 312

41-49. 313

Page 13 of 28



2. Gous, P.;, Martin, A; Lawson, W.; Kelly, A.; Fox, G.; Sutherland M. QTL 314

associated with barley (Hordeum vulgare) feed quality traits measured through 315

in situ digestion. Euphytica 2012, 185, 37-45. 316

3. Casal, S.; Oliveira, M.; Alves, M. Ferreira MA. Discriminate analysis of 317

roasted coffee varieties for trigonelline, nicotinic acid, and caffeine content. J 318

Agric. Food Chem. 2000, 48, 3420-3424. 319

4. Tripathi, S.; Mishra, H. A rapid FT-NIR method for estimation of aflatoxin 320

B(1) in red chili powder. Food Control 2009, 20, 840-46. 321

5. Tallada, J.; Wicklow, D.; Pearson, T. Armstrong PR. Detection of fungus-322

indected corn kernels using near-infrared reflectance spectroscopy and colour 323

imaging. Trans. ASABE 2011, 54, 1151-1158. 324

6. McGoverin, C.; Engelbrecht, P.; Geladi, P.; Manley, M. Characterisation of 325

non-viable whole barley, wheat and sorghum grains using near-infrared 326

hyperspectral data and chemometrics. Anal. Bioanal. Chem. 2011, 401, 2283-327

2289. 328

7. Salgo, A.; Gergely, S.; Juhasz, R. Characterizing the maturation and 329

germination processes in wheat by NIR methods. Using cereal science and 330

technology for the benefit of consumers. Proceedings of the 12th International 331

ICC Cereal and Bread Congress, Harrogate, UK, 23-26th May 2004. 332

8. Ribeiro, J.; Ferreira, M.; Salva, T. Chemometric models for the quantitative 333

descriptive sensory analysis of Arabica coffee beverages using near infrared 334

spectroscopy. Talanta 2011, 83, 1352-58. 335

9. Bertrand, B.; Etienne, H.; Lashermes, P.; Guyot, B.; Davrieux, F. Can near-336

infrared reflectance of green coffee be used to detect introgression in Coffea 337

arabica cultivars? J Sci. Food Agric 2005, 85, 955-62. 338

10. Santos, J.; Sarraguca, M.; Rangel, A.; Lopes, J. Evaluation of green coffee 339

beans quality using near infrared spectroscopy: a quantitative approach. Food 340

Chem. 2012, 135, 1828-35. 341

11. Posada, H.; Ferrand, M.; Davrieux, F.; Lashermes, P.; Bertrand, B. Near 342

infrared spectral signature and their stability across environments. 22nd 343

International Conference on Coffee Science, ASIC 2008, Campinas, SP, 344

Brazil, 14-19 September, 2008 . 345

12. Delwiche, S. Measurement of single-kernel wheat hardness using near-346

Infrared transmittance. Trans. ASAE 1993, 36, 1431-37. 347

13. Delwiche, S. Single wheat kernel analysis by near-infrared transmittance 348

protein content. Cereal Chem. 1995, 72, 11-16. 349

14. Angelino, S.; VanLaarhoven, H.; VanWesterop, J.; Broekhuijse, B.; Mocking, 350

H. Total nitrogen content in single kernel malting barley samples. J. Inst. 351

Brew. 1997, 103, 41-46. 352

15. Home, S.; Wilhelmson, A.;, Tammisola, J.; Husman, J. Natural variation 353

among barley kernels. J.Am. Soc. Brew. Chem. 1997, 55, 47-51. 354

16. Dowell, F. Automated color classification of single wheat kernels using visible 355

and near-infrared reflectance. Cereal Chem. 1998, 75, 142-144. 356

17. Delwiche, S.; Hruschka, W. Protein content of bulk wheat from near-infrared 357

reflectance of individual kernels. Cereal Chem. 2000, 77, 86-88. 358

18. Jiang, H.; Zhu, Y.; Wei, L.; Dai, J.; Song, T.; Yan, Y.;Chen; S. Analysis of 359

protein, starch and oil content of single intact kernels by near infrared 360

reflectance spectroscopy (NIRS) in maize (Zea mays L.). Plant Breed. 2007, 361

126, 492-497. 362

Page 14 of 28



19. Pearson, T.; Brabec, D.; Dogan, H. Improved discrimination of soft and hard 363

white wheat using SKCS and imaging parameters. Sensing Instrument. Food 364

Quality Safety 2009, 89-99. 365

20. Tallada, J.; Palacios-Rojas, N.; Armstrong, P. Prediction of maize seed 366

attributes using a rapid single kernel near infrared instrument. J. Cereal Sci. 367

2009, 50, 381-87. 368

21. Fox, G.; Sweeney, N.; Hocroft, D. Development of a single kernel NIR barley 369

protein calibration and assessment of variation in protein on grain quality. J. 370

Inst. Brew. 2011, 117, 582-586. 371

22. Peiris, K.; Dowell, F. Determining weight and moisture properties of sound 372

and Fusarium-damaged single wheat kernels by near-infrared spectroscopy. 373

Cereal Chem. 2011, 88, 45-50. 374

23. Williams, P.; Geladi, P.; Fox, G. Manley M. Maize kernel hardness 375

classification by near infrared (NIR) hyperspectral imaging and multivariate 376

data analysis. Anal. Chim Acta 2009, 653, 121-30. 377

24. Williams, P.; Manley, M.; Fox, G.; Geladi P. Indirect detection of Fusarium 378

verticillioides in maize (Zea mays L.) kernels by near infrared hyperspectral 379

imaging. J. Near Infrared Spec. 2010, 18, 49-58. 380

25. Bonnlander B, Lonzarich V, Liverani FS. Caffeine determination in single 381

beans by LC-MS/MS DAD; 21st International Conference on Coffee Science, 382

Montpellier, France, 11-15 September, 2006. 383

26. Baumann, T.; Sondahl, M.; Waldhauser, S.; Kretschmar, J. Non-destructive 384

analysis of natural variability in bean caffeine content of Laurina coffee. 385

Phytochem. 1998, 49, 1569-1573. 386

27. Reid, D.; Guthrie, J. Comparing two competing calibrations for a given data 387

set using Fearn's Criteria. Version 2.0. NIR Spectrum 2004, 2, 7. 388

28. Williams, P.; Sobering, D. Comparison of commercial near infrared 389

transmittance and reflectance instruments for analysis of whole grains and 390

seeds. J. Near Infrared Spec. 1993, 1, 25-32. 391

29. Esteban-Diez, I.; Gonzalez-Saiz, J.; Pizarro, C. Prediction of roasting colour 392

and other quality parameters of roasted coffee samples by near infrared 393

spectroscopy. A feasibility study. J. Near Infrared Spec. 2004, 12, 287-297. 394

30. Fox GP, O'Donnell NH, Stewart PN, Gleadow RM. Estimating Hydrogen 395

Cyanide in Forage Sorghum (Sorghum bicolor) by Near-Infrared 396

Spectroscopy. J Agric. Food Chem. 2012, 60, 6183-87. 397

398

399

Page 15 of 28



Table 1. Caffeine concentration of samples collected. 400

Coffee

No.

Brand1 Type

1 Origin

1 Caffeine

(mg/g)

1

Brasilia Coffee Italian

Espresso

Blend Italy

17.7

2

Brasilia Coffee Supa

Cream

Blend Italy

12.7

3

Oxfam Aus World Blend

Organic Espresso

Arabica Africa / Latin American

/ Asia Pacific 12.5

4 Vittoria Espresso Coffee Blend Italy 12.2

5

Caffe Aurora Medaglia

Doro

Blend Italy

19.9

6 Monjava Blend Central America 12.4

7 Dormans Breakfast (MR) Blend Kenya 11.7

8 Dormans Espresso (DR) Blend Kenya 11.8

9

Dormans AA Blue

mountain (MR)

Arabica Kenya

11.4

10 Masai (MR) Arabica Kenya 10.8

11

CTM Lush Blend Colombia/Nicarigua

/Timor/New Guinea 10.7

12

CTM House Blend Brazil / Colombia

/Sumatra / India 12.3

13

CTM Revive Blend Colombia / Costa Rica /

Indonesia / New Guinea

/ Brazil 11.6

14

CTM Jamocha Blend South America / Central

America / Africa / East

India 12.7

15 CTM Mimic (decaf) Blend Colombia / New Guinea 0.01

16 CTM Single Origin Blend Rwanda 10.6

17

Woolworths Select

Espresso

Arabica

12.2

18

Gloria Jean’s Coffee

smooth Classic Blend

Arabica Central and South

America 12.1

19 North Queensland Gold Arabica Australia 11.6

20 PNG Hand Roast Arabica Papua New Guinea 12.2

21 Brazil Hand Roast Arabica Brazil 12.8

22 Colombia Hand Roast Arabica Colombia 12.8

23 Doisaket Coffee Brand Arabica Thailand 11.7

24 Native Organic Arabica Brazil 10.5

25 Green Cauldron Arabica Australia 10.7

26 Café one Italian No.5 Arabica Italy 11.8

27 Spar Espresso (SRR) Arabica South Africa 11.8

28

Café de chamarel L’lle

Maurice

Europe

11.7 1As per information on sample packet 401

MR - medium roast 402

DR - Dark roast 403

SR – Strong rich roast 404

405

Page 16 of 28



Table 2. Variation in caffeine concentration (mg/g) in 25 single beans from five 406

selected coffees. 407

Sample

Bean 5 7 11 15 24

1 10.1 11.2 9.9 1.2 10.4

2 16.8 10.1 11.3 0.7 7.4

3 16.5 9.3 8.4 0.7 9.5

4 15.8 13.6 8.9 0.9 10.1

5 21.7 13.2 10.0 0.7 12.0

6 14.7 14.1 13.0 0.6 11.8

7 22.7 12.4 10.3 1.0 8.6

8 17.9 13.9 9.0 0.5 10.1

9 19.4 17 11.0 1.2 8.3

10 20.2 10.1 10.1 1.2 10.1

11 18.9 8.6 8.9 0.6 9.7

12 15.9 10.7 7.7 0.6 9.7

13 20.2 12.3 9.9 0.6 9.4

14 18.2 10.8 8.5 0.3 9.5

15 23.7 11.2 10.5 0.5 11.0

16 20.1 14.4 8.3 0.8 10.5

17 15.4 12 11.2 0.9 12.2

18 25.3 11 10.9 0.6 9.3

19 18 11.4 10.5 0.9 9.0

20 20.1 10.6 9.6 0.8 11.0

21 16.1 10 7.9 1.1 11.0

22 15.3 10 9.6 0.6 11.0

23 16.7 11.8 8.9 0.6 9.8

24 15.4 6.7 7.9 0.6 7.2

25 18.2 10.5 9.1 0.6 11.2

Calculated

mean 18.1 11.5 9.7 0.8 10.0

Median 18.0 11.2 9.6 0.7 10.1

SD 3.2 2.8 1.2 0.3 1.3

Max 25.3 14.4 13.0 1.2 12.2

Min 14.7 6.7 7.7 0.6 7.2

Ground

coffee1 19.9 11.7 10.7 0.0 10.5

1Caffeine concentration of bulk ground coffee for each of these coffees 408

409

Page 17 of 28



Table 3. Statistics for NIR calibrations for unground bulk, ground and single bean coffees. 410

Sample type No. R2 SEC R

2CV SECV RPDCV Factors Bias Treatment

(smoothing)

Opus calibrations

Bulk 28 0.9325 0.9 0.9255 1.0 3.65 8 0.003 2dSNV (9)

Ground 28 0.9504 0.8 0.9333 0.9 4.10 5 0.001 2dSNV (9)

Single bean 124 0.9510 1.4 0.9197 1.7 3.53 10 0.005 1dSNV (13)

Bulk less decaff coffee 27 0.9311 0.9 0.9178 1.0 3.55 8 0.003 2dSNV (9)

Ground less decaff coffee 27 0.9434 0.8 0.9199 0.9 4.25 5 0.001 2dSNV (9)

Single bean less decaff coffee 99 0.9304 1.1 0.8749 1.5 2.83 9 0.02 1dMSC (13)

Unscrambler calibrations

Bulk 28 0.9225 0.9 0.9255 1.0 3.65 8 0.003 2dSNV

Ground 28 0.9511 0.8 0.9343 0.9 4.10 5 0.001 2dSNV

Single bean 122 0.9491 1.3 0.9228 1.6 3.55 8 -0.013 1dSNV

Bulk less decaff coffee 27 0.9301 0.9 0.9098 1.0 3.52 8 0.003 2dSNV

Ground less decaff coffee 27 0.9407 0.8 0.9124 0.9 4.21 5 0.001 2dSNV

Single bean less decaff coffee 97 0.9275 1.2 0.8599 1.6 2.75 8 -0.02 1dSNV

No. - Number of samples used in the calibration 411

R2 - Coefficient of determination of calibration set 412

SEC - Standard error of calibration 413

R2CV - Coefficient of determination of cross validation 414

SECV - Standard error of cross validation 415

RPDCV - Ratio of SD to standard error of prediction using the cross validation 416

417

418

Page 18 of 28



419

420

Figure 1. Structure of caffeine. 421

422

423

424

Page 19 of 28



a

0

1

2

3

4

5

6

7

10 12 14 16 18 20 22 24 26

Caffeine Conc. (mg/g) 425

b

0

2

4

6

8

10

12

6 8 10 12 14 16 18 20 22


c

0246

8101214

6 8 10 12 14 16

Caffeine Conc. (mg/g)

427

d

0

1

2

3

4

5

6

7

8

9

0 0.2 0.4 0.6 0.8 1 1.2 1.4


Page 20 of 28



e 429

0

2

4

6

8

10

12

6 8 10 12 14


Figure 2. Distribution of caffeine between single beans of five samples: 5 (a), 7 (b), 431

11 (c), 15 (d) and 24 (e). 432

433

Page 21 of 28



0

24

6

8

1012

14

110

120

130

140

150

160

170

180

190

200

210

220

230

240

250

260

Bean Weight (mg)

No.

434

Figure 3. Distribution of bean weights used in NIRS calibration development. 435

436

Page 22 of 28



(a)

y = 0.073x + 0.12

R2 = 0.3188

0.000

5.000

10.000

15.000

20.000

25.000

30.000

100 150 200 250 300

Bean Weight (mg)

Caffeine content (m

g/g)

437

438

(b)

y = 0.0051x - 0.0771

R2 = 0.3421

0.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

100 150 200 250 300

Bean Weight (mg)

Caffeine content (m

g/g)

439

Figure 4. Correlation plot between bean weight and caffeine concentration for used 440

in NIR calibration development (a) caffeinated coffees and (b) decaffeinated coffee 441

beans. 442

443

Page 23 of 28



(a) 444

445

(b) 446

447

(c) 448

Figure 5. Correlation plots for whole (a) ground (b) and (c) single bean coffees 449

(actual (X axis) vs predicted (Y axis)). 450

451

Page 24 of 28



a 452

b 453

Figure 6. Pure caffeine standard (a) raw spectra and (b) 1st derivative SNV treatment. 454

455

Wavenumber (cm)

Page 25 of 28



a 456

b 457

4500500055006000650070007500

-10000

-5000

0

5000

Absorbance Units

Page 26 of 28



c

500055006000650070007500

-0.1

5-0

.10

-0.0

50.0

00.0

50.1

00

.15

0.2

0

Abso

rban

ce U

nits

458

Figure 7. Loading plots for (a) bulk unground, (b) ground and (c) single beans.. 459

460

Page 27 of 28



TOC Graphic 461

462

463

Page 28 of 28



Variation in Caffeine Concentration in Single Coffee Beans

Documents

Transcript of Variation in Caffeine Concentration in Single Coffee Beans