1RESEARCH HIGHLIGHT

2
3  to determine calibration models using NIR-chemometric technique for analysis caffeine
4 content in commercial tea product.
5  IR spectra data of tea product were correlated with the caffeine concentration using
6 chemometric methods
 7 Title: IR SPECTROSCOPY COUPLED WITH CHEMOMETRIC AS A SIMPLE AND
8 RAPID METHOD FOR DETERMINATION OF CAFFEINE CONTENT OF TEA
9 PRODUCT
10
11
12 Lestyo Wulandari*, Koko Dwi Pratoko and Diana Hanifiyah Sutipno
13
14
15Faculty of Pharmacy, University of Jember
16Jl. Kalimantan 37, Jember 68121, Indonesia
17
18
19
20Email adresses:
21Dwi Koko Pratoko: dwikoko.farmasi@unej.ac.id
22Diana Hanifiyah Sutipno: dhinnalova@gmail.com
23
24
25*Corresponding author: Lestyo Wulandari
26Tel.: 0331-324736
27Email: lestyowulandari@unej.ac.id
28
29 ABSTRACT
30

31 Tea is a popular beverage that comes from the Camellia sinensis. There are four types of
32tea in general, namely black tea, oolong tea, green tea, and white tea. The four types of tea are
33distinguished based on the presence or absence of the fermentation process in the processing.
34One of the compounds that plays a role in providing freshness in tea is caffeine. The purpose of
35this study was to determine caffeine content in tea samples on the market using the NIR-
36Chemometric method, while the comparison method used was TLC-Densitometry. Infrared (IR)
37spectroscopy combined with chemometrics has been developed for simple analysis of caffeine in
38the tea sample. IR spectra of tea sample were correlated with caffeine content using
39chemometrics. In this study, the PLS model of NIR model that showed the best calibration with
40R-Square and RMSEC value were 0.9579185 and 0.0698975 respectively. PLS calibration model
41of NIR models was further used to predict unknown caffeine content in commercial samples. The
42significance of caffeine content that has been measured by NIR and TLC-Densitometry was
43evaluated with Two Paired Samples t-Test. The caffeine content that has been measured with
44both methods gave no significant difference.

45
46Keywords: tea (Camellia sinensis), black tea, oolong tea, green tea, white tea, caffeine content;
47NIR, FTIR, chemometrics
48
491. INTRODUCTION
50 Tea is a plant whose parts of leaves and shoots are used to make drinks. There are four
51types of tea. The four types of tea are distinguished based on the fermentation process. Black
52tea through fermentation, green tea without going through the fermentation process, oolong tea
53through a partial fermentation process (semifermentation), while white tea is taken from the
54youngest leaves of tea which are immediately evaporated and dried without the fermentation
55process first [7].
56 Consumption of tea can provide general benefits, which reduce fatigue, increase physical
57endurance and mental alertness, and play a role in the body's recovery process. People often
58consume tea for health and beauty purposes. One type of tea is green tea which acts as a weight
59loss and inhibitor of premature aging [16]. One of the main compounds contained in tea is
60caffeine. Caffeine is one of the compounds of the methylated xanthine derivative alkaloid. The
61safe limit of caffeine consumption according to BPOM is 150 mg / day which is divided into at
62least three doses [9]. If caffeine is consumed in the right amount, the benefits will be obtained by
63the body, but conversely if excessive it can trigger acceleration of heart rate, feeling nervous,
64anxious and even insomnia, so special attention is needed if consumed by people whose body
65tolerance to caffeine is still lacking, such as children and adolescents and pregnant women [2].
66 Based on side effects if consumed in excess, it is important to determine the caffeine
67content of tea products in the market. There are several methods of determining caffeine content
68that have been used, namely UV-Vis Spectrophotometry [2] [5] [11], High Performance Liquid
69Chromatography (HPLC) [1] [3] [8], and TLC-Densitometry [12] [4] [18] [17]. The methods that
70have been used have several disadvantages, namely requiring certain solvents and reagents, so
71that researchers are currently faced with the challenge of being able to process data and analyze
72the results using fast and reliable analytical methods. One method that has the potential to be
73used is NIR Spectroscopy, because sample preparation from the NIR Spectroscopy method is
74easier, ie without using additional solvents or other reagents [13] [15].
75 The spectrum produced by NIR spectroscopy is quite complex so it is difficult to interpret.
76Based on these problems, then to overcome them a method called chemometrics is needed.
77Chemometric is a method based on statistical and mathematical approaches to find the
78relationship between spectra data and chemical parameters of a substance that is difficult to
79measure directly.
80This study aimed to create calibration model using NIR-chemometric technique to determine
81caffeine content from commercial tea product.
82
832. MATERIALS AND METHODS
842.1 Materials and instrumentation
 85 The material samples were commercial tea product (black tea, oolong tea, green tea, and
86white tea), caffeine standard, methanol, chloroform, thin layer chromatography (TLC) plates. The
87instruments used were sieve No. 60, chamber, ultrasonicator, micro pipette capillaries, CAMAG
88densitometer, NIR device (Brimrose Luminar 3070), and The Unscrambler X 10.2 software.
89
902.2 Sample simulation preparation
91 Tea sample simulation was divided into training set and test set were collected from various
92shops, shopping centers, and traditional markets in Jember, East Java, Indonesia. Twenty tea
93product simulation were prepared as a training set consisted of three type of tea (black tea, green
94tea, and oolong tea). Four tea sample prepared as test set that consist about black tea, oolong
95tea, and white tea, and three sample used as real sample. The sample used in the study can be
96seen in table 1.
97
982.3 Determination of IR spectral data
99 All samples were scanned on both infrared spectroscopy instruments (NIR and FTIR). In
100all samples were scanned five replication use NIR spectroscopy (Brimrose Corporation Luminar
1013070) with five shots for each sample and FTIR spectroscopy (Bruker Alpha) with three
102replications for each sample, so that the NIR and FTIR spectra data are obtained through
103Acquire Brimrose (NIR) and OPUS (FTIR) software, each spectra data was code named.
104

1052.4 Preparation of caffeine standard

106 The caffeine standard is weighed as much as 50 mg and 30 mg, then diluted in a 25 ml
107methanol. The standard caffeine mother solution is obtained with a concentration of 2000 µg / ml
108and 1200 µg / ml respectively. The mother liquor 2000 µg / ml was diluted to a concentration of
109500 µg / ml, 800 µg / ml, and 1000 µg / ml with methanol. Meanwhile, the mother liquor 1200 µg /
110ml became 300 µg / ml and 600 µg / ml.
111
1122.5 Determination of caffeine content by TLC-Densitometry method
113 The tea samples were weighed 400 mg each which was replicated three times. The tea
114sample was put into a 10 ml methanol. The sample is inserted in an ultrasonicator and run for 10
115minutes. The sample solution is then left to stand for 24 hours in the refrigerator to optimize
116caffeine which is being extracted and then filtered into the vial using filter paper.
117 All standard caffeine with sample solutions were each 2 µl at the TLC plate using a
118capillary micro pipette. After the result of the spots was dried, the TLC plate are inserted into the
119chamber which has been saturated by the chloroform:methanol (9.5: 0.5). eluent, if the eluent
120has reached the eluate limit, the plate is lifted and dried. The stain from eluation was scanned
121and the purity of the spectra was produced using CAMAG densitometers. The caffeine content in
122tea samples is calculated based on the scanning data.
123
1242.6 Determination of calibration models
125 The NIR and FTIR spectra data of the training set samples were analyzed quantitatively
126using PLS, PCR and SVR regression chemometrics through The Unscrambler X 10.2 software. Y
127variable (predictor) is absorbance value of the spectrum data was used as an combined with the
128% w/w caffeine value from a previously determined using the TLC-densitometry method used as
129an X variable (reference). R-square value > 0.91 or greater and the smaller RMSE error value of
130the model indicates that the model was chosen has the best predictive ability [10]. The best
131calibration model was then tested by Leave One Out Cross Validation (LOOCV) and 2-Fold
132Cross Validation (2-FCV) techniques. The model was validated by LOOCV with removing a set of
133sample data from the training set, and the rest data was used to create the new model. The 2-
134FCV by using test set as an independent sample.
135
1362.7 Application in commercial sample
137 The spectra of commercial tea samples were determined by IR spectroscopy. The
138selected and validated model is then applied to the real sample. The caffeine content of the TLC-
139Densitometry method results compared to the IR spectroscopy prediction of caffeine sample,
140results of both then input into to the SPSS Trial version 23.0 program for further analysis.
141
1423. RESULTS AND DISCUSSION
143 Samples were given an identity using a code that was adjusted to the brand in each tea
144sample. Determination of caffeine content is carried out two or three times from the replication
145results because there are several replication samples whose concentration of analyte cannot
146appear on the output densitometer (out of permitted & no peak detected or peak deleted). The
147results of NIR and FTIR spectra data are shown in Figure 1, the spectral data of all samples were
148used as predictors on the calibration model.
149
1503.1 Determination of caffeine content by TLC-Densitometry
151 The wavelength used to scan the spots on the TLC plate using desitometer is 277 nm.
152Caffeine content from each sample are expressed with % w/w. The level of % w/w of each tea
153sample is determined by converting the concentration value of the analyte that appears in the
154form of a nanogram unit on the densitometer scan results to the initial weighing and dilution.
155Caffeine content was obtained as in Table 2.
156
1573.2 Determination and validation of the calibration model
158 320 spectra data from NIR and 58 spectra from FTIR were analyzed quantitatively using
159chemometrics on the The Unscrambler X 10.2 software. The spectra analyzed using PLS, PCR
160and SVR technique. The spectra of the training set sample was correlated with the caffeine
161concentration for determining the calibration model. Spectra data as the variable y (response)
162and the value of caffeine content of the results of TLC-Densitometry expressed by % w/w as the
163variable x is used to predict the variable y (predictor). The best calibration model was PLS,
164because it has R2 and RMSEC value were 0.9579185 and 0.0698975 respectively (Table 3). This
165value show a good R value above 0.91 [10] and good RMSEC because it is an error or deviation
166between the concentration of predicted results with actual concentration, so that if the value gets
167smaller then it will also produce a better calibration model because it shows predictive results the
168same or near the actual concentration [18]. The slope parameter shows the average increase or
169decrease in the Y variable for an increase in one variable X (the slope size of a line). If the slope
170is positive then the shape of the line will increase to the right as in the results of this study. This
171value indicates that the PLS model of NIR has been formed as good regression linearity, which
172was the actual value, and predictive value have a close correlation [10].
173 The results of the LOOCV PLS model of NIR showed that the R-square value > 0.91 and
174RMSE value were small (±0.06), while the results of 2-FCV validation through the prediction of
175the test set sample showed that the R-square and RMSE values obtained were 0.9250847 and
1760.0909219. The esults of 2-FCV can be seen in Figure 2. Based on the results of LOOCV and 2-
177FCV validation, it can be concluded that the reliability or consistency of the prediction ability of
178the PLS calibration model NIR was well-formed so that it can be implemented in the actual
179sample.
180
1813.2 Application of Commercial Sample
182 The PLS as the best model calibration that has been validated is then applied to the real
183sample. Real samples obtained were scanned using an NIR spectrophotometer, then the caffeine
184samples were determined using TLC-Densitometry. The caffeine content of TLC-Densitometry
185results compared to the NIR scan caffeine levels by inputting both levels of data into the SPSS
186trial version 23.0 program for further analysis. The mean value of % w/w caffeine results from the
187NIR spectroscopy method was compared with the results of the TLC-Densitometry method
188(Table 4).
189 The analysis used is in the form of an analysis of Two Paired t-Test samples to obtain
190information about whether there is a significant difference between the levels of real samples of
191NIR scan results with TLC-Densitometry. Based on the results of the analysis of the T-Test two
192paired samples showed that the significance value produced was 0.122 (> 0.05), so that Ho was
193accepted which means there was no significant difference between the caffeine levels of real
194samples obtained from NIR and TLC-Densitometry.
195
1964. CONCLUSION
197 Infrared spectroscopy method in the PLS calibration model of NIR spectroscopy-coupled
198with chemometrics methods can be used to determine the caffeine content. The FTIR calibration
199model doesn’t have a good R-Square validation value. The results of determining the caffeine
200content in both NIR Spectroscopy and TLC-Densitometry methods were the same or there are no
201significant differences. IR spectroscopy method was rapid, precise, accurate and eco-friendly.
202
2035. CONFLICT OF INTEREST
204The authors declare that there is no conflict of interest.
205
2066. REFERENCES
207[1] Bae, I. K., H. M. Ham, M. H. Jeong, D. H. Kim, dan H. J. Kim. 2015. Simultaneous
208 determination of 15 phenolic compounds and caffeine in teas and mate using RP-HPLC/UV
209 detection: method development and optimization of extraction process. Food Chemistry.
210 172:469–475.
211[2] Belay, A., K. Ture, M. Redi, dan A. Asfaw. 2008. Measurement of caffeine in coffee beans
212 with uv/vis spectrometer. Food Chemistry. 108(1):310–315.
213[3] Cunha, R. R., S. C. Chaves, M. M. A. C. Ribeiro, L. M. F. C. Torres, R. A. A. Muñoz, W. T.
214 P. D. Santos, dan E. M. Richter. 2015. Simultaneous determination of caffeine,
215 paracetamol, and ibuprofen in pharmaceutical formulations by high-performance liquid
216 chromatography with uv detection and by capillary electrophoresis with conductivity
217 detection. Journal of Separation Science. 38(10):1657–1662.
218[4] Ford, M. J., M. A. Deibel, B. A. Tomkins, dan G. J. Van Berkel. 2005. Quantitative thin-layer
219 chromatography/mass spectrometry analysis of caffeine using a surface sampling probe
220 electrospray ionization tandem mass spectrometry system. Analytical Chemistry.
221 77(14):4385–4389.
222[5] Hasanah, U., R. K. Wardani, dan M. A. H. F. Fernanda. 2016. Analisis kadar kafein dari
223 serbuk teh hitam , teh hijau dan teh putih ( camellia sinensis l .). Journal of Pharmacy and
224 Science. 1(1):21–23.
225[6] Heckman, M. A., J. Weil, dan E. G. de Mejia. 2010. Caffeine (1, 3, 7-trimethylxanthine) in
226 foods: a comprehensive review on consumption, functionality, safety, and regulatory
227 matters. Journal of Food Science. 75(3):77–87.
228[7] Hirthe, K. 2007. Identification and comparison of phenolic compounds in the preparation of
229 oolong tea manufactured by semifermentation and drying processes. Agricultural and Food
230 Chemistry. 7462–7468.
231[8] Jiang, H., U. H. Engelhardt, C. Thräne, B. Maiwald, dan J. Stark. 2015. Determination of
232 flavonol glycosides in green tea, oolong tea and black tea by uhplc compared to hplc. Food
233 Chemistry. 183:30–35.
234[9] Keputusan Kepala Badan Pengawasan Obat dan Makanan Republik Indonesia Nomor.
235 HK.00.05.23.3644 Tahun 2003. Ketentuan Pokok Pengawasan Suplemen Makanan. 9
236 agustus 2004. Jakarta.
237[10] Lengkey, L. C. E. C., I. W. Budiastra, K. B. Seminar, dan B. S. Purwoko. 2013.
238 Determination of chemical properties in jatropha curcas l . seed ip-3p by partial least-
239 squares regression and near-infrared reflectance spectroscopy. International Journal of
240 Agriculture Innovations and Research. 2(1)
241[11] Navarra, G., M. Moschetti, V. Guarrasi, M. R. Mangione, V. Militello, dan M. Leone. 2017.
242 Simultaneous determination of caffeine and chlorogenic acids in green coffee by uv/vis
243 spectroscopy. Journal of Chemistry. 2017:8.
244[12] Pavlik, J. W. 1973. TLC detection of caffeine in commercial products. Journal of Chemical
245 Education. 50(2):134.
246[13] Rohman, A., Sismindari, Y. Erwanto, dan Y. B. Che Man. 2011. Analysis of pork
247 adulteration in beef meatball using fourier transform infrared (ftir) spectroscopy. Meat
248 Science. 88(1):91–95.
249[14] S. Pindyck, R. dan D. Rubinfeld. 1976. Econometric Models and Economic Forecasts.
250 XF2006305658.
251[15] Shafirany, M. Z., Y. Susilawati, dan I. Musfiroh. 2018. Aplikasi kemometrik dalam
252 penentuan mutu tumbuhan obat. Majalah Farmasi, Sains, Dan Kesehatan. 4(2):6–14.X[16]
253 Sudaryat, Y., M. Kusmiyati, C. R. Pelangi, dan A. Rustamsyah. 2015. Aktivitas antioksidan
254 seduhan sepuluh jenis mutu teh hitam ( camellia sinensis ( l .) o . kuntze ) indonesia
255 antioxidant activity of ten grades of indonesia black tea. Jurnal Penelitian Teh Dan Kina.
256 18(2):95–100.
257[17] Torres Y Torres, J. L., S. L. Hiley, S. P. Lorimor, J. S. Rhoad, B. D. Caldwell, G. L.
258 Zweerink, dan M. Ducey. 2015. Separation of caffeine from beverages and analysis using
259 thin-layer chromatography and gas chromatography - mass spectrometry. Journal of
260 Chemical Education. 92(5):900–902.
261[18] Trianto, G. O. 2009. Penetapan kadar kafein dalam teh celup yang beredar di supermarket
262 di wilayah kabupaten jember secara klt-densitometri. Skripsi. Jember: Fakultas Farmasi
263 Universitas Jember.
264
265
266
267
268
269
270
271
272
273
274
275
2767. TABLE AND FIGURE CAPTIONS
277
278Table 1. The sample used in the study
279
280Table 2. Results determining caffeine content in each sample
281
282Table 3. Calibration results of training set samples
283
284Table 4. Results of caffeine content of real samples
285
286Figure 1. Result of NIR and FTIR spectra
287
288Figure 2. Result of 2-Fold Cross Validation;
289
290Table 1. Wulandari
291
No. Sample group Code name Tea Brand
1 Training set PC Poci
2 SR Sariwangi
3 SS Sosro
4 DD Dandang
5 KJ Kepala Djenggot
6 GD Gardoe
7 NG Naga
8 BD Bandulan
9 BT Cap Botol
10 PR Prendjak
11 SM Sarimurni
12 SB Cap Sepeda Balap
13 CT Tjatoet
14 IDG Indomaret
15 BTG Cap Botol
16 SSG Sosro
17 JWG Jawa
18 2TG 2 Tang
 19 XNO Xiamen
20 GLG Galan 999
21 Test set TJ Tong Tji
22 ZTW Zet White Tea
23 GG Galan 999
24 GP Gopek
25 Real sample DD2 Dandang
26 SS2 Sosro
27 XOO Xiamen
292
293Table 2. Wulandari
294
Sample group Name code Caffeine content (%b/b) ± SD
Training set PC 1.26±0.03
SR 1.43±0.03
SS 1.43±0.04
DD 1.27±0.01
KJ 1.27±0.00
GD 1.31±0.07
NG 1.41±0.03
BD 1.81±0.01
BT 1.95±0.02
PR 1.28±0.03
SM 1.56±0.02
GLG 2.5±0.05
SB 1.30±0.02
TJT 1.30±0,00
IDG 1.64±0.01
BTG 1.35±0.03
SSG 2.08±0.01
JWG 1.96±0.01
2TG 1.92±0.00
XNO 1.27±0.03
Test set TJ 1.67±0.03
ZTW 1.75±0.01
GG 2.35±0.05
GP 1.49±0.03
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322Table 3. Wulandari
323
324 No Model Slope RMSE R-Square
325 .
326
327 1 PLS Calibration 0.958199 0.0698975 0.9579185
328 Validation 0.958211 0.0698972 0.9579188
329 5
330 2 PCR Calibration 0.621883 0.2107912 0.6218834
331 2
332 Validation 0.613763 0.2151758 0.6078126
333 3
334 3 SVR Calibration - 0.2300169 0.5597217
335 Validation - 0.2384573 0.5242738
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375Table 4. Wulandari
376
Sample Caffeine content (%b/b)
NIR ± SD TLC-Densitometry ± SD
DD2 1.28±0.03 1.33±0.01
SS2 1.52±0.02 1.53±0.03
XOO 1.33±0.09 1.36±0.03
377
378
379
380
381Figure 1. Wulandari
382
1,9

383 1,8
1*
384 1,7

1,6

1,5

1,4

1,3

1,2

1,1

0,9

0,8

0,7

0,6

0,5

0,4
850 872 896 920 944 968 992 1018 1048 1078 1108 1138 1168 1198 1228 1258 1288 1318 1348 1378 1408 1438 1468 1498 1528 1558 1588 1618 1648 1678 1708 1738 1768 1798 1828 1858 1888 1918 1948 1978

0,01

2*
0

-0,01

-0,02
3998,419 3843,98 3696,625 3545,02 3409 3299,9 3163,88 3023,61 2876,255 2728,9 2592,88 2445,525 2288,252 2139,48 1999,21 1851,855 1704,5 1572,73 1436,71 1296,44 1160,42 1024,4 886,9631 729,6902

*Spectra data 1. NIR, 2. FTIR; (A) DD, (B) KJ, (C) PC, (D) PR, (E) XNO
385
386Figure 2. Wulandari
387
Predicted vs. Reference
388 2,3
Elements: 32
389 2,2 Slope: 0,8126091
Offset: 0,278228
2,1 Correlation: 0,9936818
R2(Pearson): 0,9874035
2
Predicted Y (C1, Factor-4)

R-Square: 0,9250847
1,9
RMSEP: 0,0909219
SEP: 0,0704236
1,8 Bias: -0,0588412

1,7

1,6

1,5

Y Reference
1,5 1,6 1,7 1,8 1,9 2 2,1 2,2 2,3 2,4
390
391Cover Letter

392
393
394
395
396
397 22nd July 2019
398
399
400Dear editor,
401
402I am pleased to send you an original research article file entitled “IR Spectroscopy Coupled
403with Chemometric as a Simple and Rapid Method for Determination of Caffeine
404Content of Tea Product” by Wulandari et al., that we would like you to consider for
405publication in the Food Chemistry journal.
406
407
408We are looking forward to hear your reply.
409
410
411Sincerely,
412Lestyo Wulandari
413
414