INTRODUCTIONFood supplements are concentrated sources ofnutrients (i.e. minerals and vitamins) or other substanceswith a nutritional or physiological effect that aremarketed in “dose” form (e.g. pills, tablets, capsules,or liquids in measured doses) [1]. In the EU, foodsupplements are regulated as foods. Therefore, it is theresponsibility of the manufacturer, importer, supplieror distributor to ensure the safety of food supplementsplaced on the market.The use of dietary supplements is mainly widespreadin sport. People are continually searching forsupplements to help them lose weight, boost energy, andbuild muscles. There are some supplements which arecommonly used to achieve these goals [2].One of them is a whey protein supplement, themost important nutrient to boost athletic performance.Whey protein is popular among athletes, bodybuilders,fitness models, as well as people seeking to improvetheir performance in the gym. Numerous studies showthat it can help increase strength, gain muscle, and losesignificant amounts of body fat [3, 4]. Some specifictypes of protein are made for certain scenarios, suchas casein protein for a slow-release protein and wheyprotein for a faster release. The main types of wheyprotein are concentrates (WPC), isolates (WPI), andhydrolysates (WPH).The branched-chain amino acids (BCAAs) –leucine, iso-leucine, and valine – are among the nineessential amino acids for humans that account for 35%of essential amino acids in muscle proteins. They areunique as the only amino acids used directly by musclesas energy during exercise [5].178Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185The next commonly used supplement in sport iscreatine which increases lean body mass, skeletal musclestrength, as well as muscle power and endurance [6].Creatine supplementation appears to raise the creatinelevel in muscle cells and cause weight gain throughan increase in lean body mass with no effect on fatmass [7, 8].Pre-workout supplements are multi-ingredientdietary formulas designed to boost energy and athleticperformance. While some pre-workout supplements havecarbohydrates, most are carbohydrate- and calorie-free.Others contain caffeine, beet juice, or amino acids, suchas arginine, citrulline, and ornithine, to increase bloodflow to the muscles.Mass gainers are products mostly directed formen who find it difficult to build lean muscle mass.They contain high amounts of calories, as well ascarbohydrates and protein, making them a perfect mealreplacement for people with quick metabolism.Due to increased consumption of sports supplementsand EU regulations, there is a need for a quick andprecise method to evaluate their quality. The defectsof traditional measurements create a possibility ofadulteration. For example, the commonly used theKjeldahl method, which determines protein content insamples, is time-consuming and unable to distinguishthe protein nitrogen from the non-protein nitrogen [9].Nowadays, adding inexpensive amino acids and aminoacid derivatives to protein supplements to modify theircontent has become a common adulteration methodwhich is hard to detect [9]. Moreover, dishonestproducers provide incorrect information on thepackaging regarding the amounts of ingredients.Some methods have been proposed to ensure thequality of sports supplements. Jiao et al. used the Ramanspectroscopy combined with multivariate analysis forrapid detection of adulterants in whey protein [8]. Highvalues of R2 and low errors of prediction for partialleast squares (PLS) analysis prove that it could beused to detect adulterants in WPC. Champagne andEmmel demonstrated the Fourier transform infrared(FTIR) with attenuated total reflectance (ATR) asa tool for detecting adulteration in raw materials ofdietary supplements [10]. The researchers proved thatvibrational spectroscopy could be used to identify thepresence of known adulterants intentionally spiked intodietary ingredients, including erectile dysfunction drugs,steroids, weight loss drugs, and Melamine.Pereira et al. proposed using fluorescencespectroscopy to detect and characterize adulteratedwhey protein supplements [11]. The adulteration wasperformed by adding creatine, caffeine, and lactoseto WPC samples at different levels (10%, 20%, and30% w/w). The time-resolved fluorescence analysisshowed increased mean intensity lifetime in alladulterated samples, compared to pure WPC. This studyproved that fluorescence spectroscopy was able to evinceadulteration in WPC powders.Another use of the fluorescence technique wasreported by Pulgarin et al. [12]. The authors used theemission spectroscopy to characterize several wheysamples subjected to different treatments and conditions.Their results indicated that the fluorescent amino acids,tyrosine and tryptophan, were responsible for theintrinsic fluorescence of whey. Martin et al. predictedthe protein content in single wheat kernels usinghyperspectral imaging, while Ingle at al. applied NIRspectroscopy to determine the protein content in powdermix products [13, 14].High-performance liquid chromatography (HPLC)is one of the most common techniques used todetermine the concentration of ingredients. The HPLCtechnique was applied by several authors to measure theconcentration of taurine, caffeine or vitamins in energydrinks [15–17]. These studies exemplify a growingdemand for new, more efficient techniques to assess thequality of food products and their ingredients. Comparedto conventional techniques or chromatography analysis,infrared spectroscopy allows measuring the sample’seco-friendliness – without sample preparation or the useof chemical reagents. In addition, FTIR spectroscopycan be successfully used in the analysis of amino acidprofiles, as confirmed by [18, 19].In this study, we applied FTIR spectroscopy coupledwith chemometrics to evaluate the quality of sportssupplements. This method is very efficient as thespectral profile in one measurement can provide variousinformation about the product that could not be given byany conventional technique in common use.Our main objectives were to create a regressionmodel using PLS analysis to determine the totalamount of protein in the product and to distinguishvarious ingredients by the FTIR spectra and principalcomponent analysis (PCA).STUDY OBJECTS AND METHODSSamples. Our study objects included forty-eightsamples of sports supplements from different producers:whey (15 samples), BCAAs (12 samples), creatine(3 samples), mass gainers (6 samples), and pre-workouts(12 samples). The samples were in the form of powdersor liquids.Protein determination. The protein content in wheyprotein samples was assessed by the Kjeldahl method,using a conversion factor of total nitrogen to protein(6.38 for milk products, 6.25 for meat products, and 5.70for vegetables) [20]. Three parallel trials were performedfor each sample. The percentage of protein in a sample(X) was calculated according to the formula [20]:X= (a ∙ n ∙ 1.4 ∙ f )/mwhere a is the amount of the standard solution ofhydrochloric acid used for titration of ammonia in aspecific sample, cm3; n is the molar concentration ofhydrochloric acid used for titration; m is the sample179Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185mass, g; f is the conversion factor of total nitrogen toprotein (6.38 for milk products, 6.25 for meat products,and 5.70 for vegetables); 1.4 is the amount of nitrogencorresponding to 1 cm3 of 0.1 molar solution ofhydrochloric acid, mg.FTIR measurements. Medium infrared spectrawere performed on a 4700 FTIR spectrometer (Jasco,Japan). Single beam spectra of the sample were collectedand rationed against the background of air. For eachsample, MIR spectra were recorded from 4000 to600 cm–1 by co-adding 16 interferograms at a resolutionof 4 cm–1. The measurements were performed intriplicate.Data analysis. Principal component analysis(PCA). Principal component analysis was performedon the FTIR spectra of whey protein supplements todistinguish the samples. PCA is a multivariate techniquethat linearly transforms an original set of variables intoa substantially smaller set of uncorrelated variables thatrepresents most of the information in the original dataset. Data for PCA are arranged in a two-way matrix,in which column vectors represent variables and rowvectors represent the “objects” whose variables aremeasured [21]. The PCA analysis was carried out usingUnscrambler X (CAMO, Oslo, Norway) software.Partial least squares (PLS). The partial leastsquares (PLS) regression method was used to determinethe relation between the samples’ spectra and the contentof protein in whey supplements. We selected regions ofspectra and data preprocessing options to optimize themodel. In total, 45 spectra were measured (15 samplesin triplicate). The set of independent variables X was theFTIR spectra and the set of dependent variables Y wasthe protein content. Full cross-validation was applied tothe regression model.The regression models were evaluated using theadjusted R2 and the root mean-square error of crossvalidation(RMSECV), as the term indicating theprediction error of the model. The quality models wereevaluated by the ratio of the standard deviation ofreference data for the validation samples to the RMSEP(RPD). The predicted values were compared to thereference values. The PLS analysis was carried out usingUnscrambler X (CAMO, Oslo, Norway) software.RESULTS AND DISCUSSIONProtein determination. The protein contentsin whey protein samples (measured by the Kjeldahlmethod) are given in Table 1.The results show that the producers declaredsimilar values to those marked. For most producers,the differences from the declared values did not exceed5 g/100 g of protein, which is considered as acceptable.The highest difference between the value declaredand that determined by the Kjeldahl method wasobserved for three samples. They were from Producer 9(84.7 g/100 g vs. 73.87 g/100 g), Producer 8 (82 g/100 gvs. 75.92 g/100 g), and Producer 15 (85 g/100 g vs.76.73 g/100 g). However, most producers declared thecorrect protein value on the package of their products.Sports supplements spectra in medium infraredrange. The medium infrared absorption spectra of thesports supplements measured against air are presented inFig. 1.According to the data reported in [23], the absorptionspectra of whey products had two prominent features,Amide I (about 1650 cm–1) and Amide II (about1540 cm–1) bands. The former arose primarily from theC=O stretching vibration and the latter was attributedto the N-H bending and C-N stretching vibrations ofthe peptide backbone. The band with the maximumabsorption at about 3268 cm–1 was assigned to Amide A.The band at 3000–2825 cm–1 corresponded to the C-Hstretching vibration, while the low intensity bands atTable 1 Protein content in whey protein samples measured by the Kjeldahl method [22]Producer Type of protein Declaration, g Determined, g Standard deviation, gProducer 1 WPC+WPH+WPI 72.72 72.72 1.98Producer 2 (vegetarian) – 56.7 54.72 0.29Producer 3 WPI 85 86.30 4.39Producer 4 WPI+WPC 78.5 81.49 0.28Producer 5 WPC+WPI+WPH 71 74.18 1.25Producer 6 WPC+WPI 71 73.04 1.08Producer 7 WPC+WPI+WPH 63 65.31 1.67Producer 8 WPC+WPI 82 75.92 2.48Producer 9 WPI 84.7 73.87 0.91Producer 10 WPI+WPC 79.2 78.02 1.45Producer 11 WPC+WPI 71 68.69 0.31Producer 12 WPC 70 71.08 0.36Producer 13 WPC+WPI 80 78.21 0.52Producer 14 WPI 88 85.32 0.81Producer 15 WPI 85 76.73 0.87WPI – whey protein isolate, WPC – whey protein concentrate, WPH – whey protein hydrolysate180Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185about 1241 cm–1 and 1100 cm–1 were assigned to the P-Ostretching vibrations [23].Zhu et al. [24] found that the absorption spectraof branched chain amino acids (BCAA) showed theconcentration of effective wavelengths of amino acids(e.g. valine, leucine, isoleucine, and glycine) mainlyin the fingerprint region (500–1700 cm–1). Based onthe literature, we can describe the main bands in theseproducts. The band with two maximum absorptionpeaks at about 1575 cm–1 and 1509 cm–1 could beassigned to isoleucine [24]. The band with the maximumat about 1400 cm–1 corresponded to glycine that does notcontain asymmetric carbon atoms [24]. The valine bandswere also observed in the fingerprint region. The band at665 cm–1 was assigned as a bending mode of CO2. Thebands at about 753 cm–1 and 776 cm–1 were assigned aswagging and bending of CO2 group, while vibrationsbetween 900 and 965 cm−1 as mainly due to the C–Cstretching vibration. The medium intensive band at2817–3000 cm–1 was assigned to the C-H hydroxylgroup [24].The creatine MIR absorption spectra have not beenwidely reported in literature. Based on the chemicalcomposition of creatine (which is also an organic acid),we can infer that the creatine spectrum should be similarto the BCAA spectrum. The differences in the intensityand shape of some bands are probably due to a highconcentration of aminoacetic acid and guanidine in thecreatine sample.The pre-workout absorption spectra have not beenwidely described in literature either. According tothe studies of caffeine determination in Singh et al. orAbdalla, pre-workouts containing caffeine have sometypical bands for that component [25, 26]. Thus, it couldbe used to confirm the presence of this ingredient in theproduct.Principal component analysis (PCA). The PCAwas used to distinguish the medium infrared spectraobtained from different types of sports supplements.The PCA data were plotted on a graph of first principalcomponent (PC1) vs. second principal component(PC2), as shown in Fig. 2. The PCA was conducted forall the products and for groups of products. The resultswere diversified into (1) all supplements, (2) proteinsupplements, (3) creatine supplements, (4) BCAAs, (5)mass gainers, and (6) pre-workout supplements (Fig. 2).Sports supplements are products to which producersadd various mixes of ingredients depending on marketneeds and prevailing trends. These ingredients mayinclude vitamins, minerals, herbs, and amino acids.In our study, we applied the PCA analysis to thespectra acquired from forty-eight samples whichwere measured in triplicate and then averaged. For thewhole spectrum (4000–400 cm–1), the first and secondprincipal components (PC1 and PC2) described 78% oftotal variation (61% and 17%, respectively), as shown inFig. 2a.Based on the data in Fig. 2a, we identified threemain groups of products. The first one included productscharacterized by positive values of PC1 and negativevalues of PC2 (BCAAs and pre-workouts). Theseproducts differed from the others in their physical state(they were liquids). The second group was productswhich primarily contained proteins and amino acids.They included mostly proteins, mass gainers, andcreatine. The third group (mostly with a positive PC1)was composed of BCAAs and pre-workouts. The mainingredient in these products was branched amino acids.Figure 1 Absorption spectra of sports supplements in medium infrared region (4000–600 cm–1) [22]181Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185The PCA results for protein supplements arepresented in Fig. 2b. For the whole spectrum, the firstand second principal components described 93% oftotal variation (53% and 40%, respectively). Based onthe distribution of the samples, we distinguished threegroups of protein products. The first group includedsupplements with high amounts of whey proteinisolate (WPI) and negative values of PC1. The secondgroup contained supplements made from whey proteinconcentrate and characterized by positive PC1 and PC2.Finally, the third group included products based ongreen protein (plant proteins for vegans) with positivePC1 and negative PC2.Fig. 2c presents the PCA results for creatine samples.According to the data, creatine with the addition ofcaffeine was characterized by positive values of PC1and PC2. Pure creatine and creatine with additives werein the opposite sites (negative values of PC1) and closetogether.The PCA results for BCAA supplements are shownin Fig. 2d. The first and second principal componentsdescribed 96% of total variation (90% and 6%,respectively). Based on the distribution of samples, weidentified three main groups of BCAA products. FluidBCAAs were characterized by negative values of PC1and positive values of PC2. Solid samples had a positiveFigure 2 PCA results for medium infrared spectra of sports supplements. Scores plot for two significant principal components:PC1 vs PC2. (a) all samples and full spectra at 4000–600 cm–1; (b) proteins; (c) creatines; (d) BCAA; (e) gainers;(f) pre-workouts [22](c) (d)(а) (b)(e) (f)182Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185PC1. Pure glutamine was in the quarter which had apositive PC1 and a negative PC2. We also found someBCAA samples with additives in that quarter. Accordingto the information on the packaging, these samplescontained glutamine. The rest of the BCAA supplements(with positive values of PC1 and PC2) were sampleswithout glutamine.Fig. 2d features the PCA results for mass gainersupplements. We found that the first and secondprincipal components described 83% of total variation(64% and 19%, respectively). Based on the data, wedistinguished two groups of gainers. The first groupwas characterized by negative values of PC1 while thesecond, by positive values of PC1. Due to insufficientinformation on the packaging, it was hard to determinethe differences between them. It is worth emphasizingthat group A contained supplements from variousproducers, while group B had only two products of thesame producer. In addition, the products in group Bcould be found on a low-price shelf on the market.The last group of sports supplements exposed toPCA included pre-workout products (Fig. 2e). The firstand second principal components described 93% of totalvariation (70% and 23%, respectively). We identifiedtwo groups of pre-workout supplements. The first groupincluded liquid samples with negative values of PC1,while the second contained solids with positive valuesof PC1. The samples in the second group differed fromeach other in the amount of caffeine. Those with loweramounts of caffeine had negative values of PC2, whilethose with higher amounts of caffeine had positivevalues of PC2.Partial least squares regression (PLS). PLS wasused to quantitatively evaluate the concentration ofprotein in whey protein supplements based on theirspectral characteristics. Different types of mathematicalpre-processing were applied to the spectra beforebuilding the model. First, we analyzed complete spectrain all the spectral regions. Next, we chose specific subregions,relying on the regression coefficients for thecomplete spectra and the chemical information in thespecific sub-regions (Fig. 3).The PLS regression results for the full spectrum(4000–400 cm–1) without any pretreatment revealeda correlation between the spectra and the proteincomposition. R2 for the calibration and validation modelsamounted to 0.76 and 0.62, respectively. It meant thatthe models had a medium-good capability to predict theprotein content in whey supplements. The RMSE for thecalibration and validation models was also low (3.5%and 4.7 %, respectively), confirming their medium-goodquality (Fig. 3a). The regression results were improvedwhen specific spectral regions were used instead of thecomplete spectra. R2 for the calibration and validationmodels reached 0.85 and 0.76, respectively. It meantthat the models had a good capability to predict theprotein content in whey supplements. The RMSE for thecalibration and validation models was also low (2.7%Figure 3 Predicted versus actual concentration of protein in whey supplements obtained by PLS calibration. (a) full FTIRspectrum; (b) sub-region: 1467–1600 cm–1; (c) full spectra after SNV normalization; (d) sub-region: 1160–2205 cm–1 after SNVnormalization [22](c) (d)(а) (b)183Wójcicki K. Foods and Raw Materials, 2020, vol. 8, no. 1, pp. 177–185and 3.7%, respectively), confirming the good quality ofthe models (Fig. 3b).Next, we performed the mathematical preprocessingof the spectra (using SNV normalization).R2 for the calibration (full spectrum) and validationmodels equaled 0.73 and 0.54, respectively. The RMSEwas also low (3.7% and 5.2%, respectively), whichconfirmed that the quality of the models was mediumgood(Fig. 3c). The regression results were improvedwhen specific spectral regions were used instead of thecomplete spectra. R2 for the calibration and validationmodels amounted to 0.91 and 0.75, respectively. Thissuggested a good capability of the models to predict theprotein content in whey supplements. The RMSE for thecalibration and validation models was also low (2.1%and 3.8%, respectively), which confirmed their goodquality (Fig. 3d).Based on the results, we found that the rapidFTIR method had an accuracy comparable to theKjeldahl method. The difference between the valuesdetermined by the Kjeldahl method and those predictedby FTIR was about 1.2 g (Table 2). In addition, ourcomplementary method offered several advantages:it is simple, fast (less than a minute) and requires nochemicals or reagents, compared to traditional methods.CONCLUSIONOur study aimed to investigate the potential ofmedium infrared (FTIR) radiation in combination witha multiway analysis in monitoring the quality of sportssupplements. The spectra of selected sports supplementshad a different shape and intensity, depending on thechemical composition. Based on the characteristicspectra, the FTIR could be used to confirm the presenceor absence of a given ingredient in the sample.The results of the PCA analysis (sample distribution)showed that the FTIR spectra coupled with PCA offereda promising tool for distinguishing sports supplementsbased on their ingredients.The regression analysis (PLS) indicated that FTIRspectroscopy could replace the time-consuming Kjeldahlmethod as a much faster technique to predict theconcentration of protein in whey supplements that doesnot require any reagents.Thus, we found FTIR spectroscopy to be a promisingapproach to quality evaluation of sports supplements.CONFLICT OF INTERESTThe author declares that there is no conflict ofinterest.