Anal Bioanal Chem (2007) 389:1577–1583 DOI 10.1007/s00216-007-1541-3
ORIGINAL PAPER
Quality control of cosmetic mixtures by NIR spectroscopy Marcelo Blanco & Manel Alcalá & Joaquim Planells & Rufí Mulero
Received: 30 May 2007 / Revised: 25 July 2007 / Accepted: 1 August 2007 / Published online: 3 September 2007 # Springer-Verlag 2007
Abstract Cosmetic preparations typically consist of mixtures of various compounds of natural origin or their derivatives. Their analysis is made rather difficult by their usually high complexity and is utterly impossible with a single analytical method; also, there is usually little to be gained by determining every individual component of the mixture. Rather, analyses are aimed at ensuring a proper balance between the contents of each component and thus require the use of methods capable of delivering global information. The combined use of near-infrared (NIR) spectroscopy and multivariate spectral processing chemometric techniques has enabled the development of effective methods for establishing the composition of complex samples with acceptable levels of analytical properties, such as accuracy, precision and throughput. In this work, we developed partial least squares calibration models for the determination of each component in a cosmetic mixture, and global indices (viz. the hydroxyl value), simply from the NIR spectrum of the sample. The models thus obtained are accurate enough for use in quality control analyses of cosmetic preparations and provide an effective alternative to existing conventional global methods. M. Blanco (*) : M. Alcalá : J. Planells Departament de Química, Unitat Analítica, Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain e-mail:
[email protected] M. Alcalá e-mail:
[email protected] R. Mulero Development and Quality Assurance Department, Croda Spain, S.A. (Mevisa Site), Ctra Hostalric a Tossa km 1.8, 08495 Fogars de la Selva, Spain
Keywords Cosmetics . Near-infrared spectroscopy . Quality control . Multivariate calibration
Introduction Cosmetic preparations are commonplace consumer products, the regular use of which helps correct flaws and preserve and improve the health status of the human skin. Cosmetics are available in various physical forms, including compact, creamy, milky, oily, alcoholic, aqueous and powdery formulations consisting of a variety of substance types such as bases or excipients, active substances, correctors, preservatives, stabilizers, and colouring and flavouring agents. Worthy of special note among such substances on account of their widespread use are excipients forming emulsions of variable consistency. An emulsion consists of small drops of a liquid homogeneously dispersed, but not dissolved, in another. Therefore, an emulsion consists of two immiscible liquids that are dispersed with the aid of a third component known as an “emulsifier”. There are two types of emulsions: they consist of a dispersion of (1) an aqueous phase in a oily medium (water-in-oil emulsions) or (2) an oil, fat or wax in water (oil-in-water emulsions). Emulsifiers typically consists of surfactants, glycerin esters and oxyethylene fatty acids and their derivatives [1]. The emulsion of choice for each application is dictated by the intended use of the cosmetic concerned. Many components of cosmetic emulsions are natural products consisting of complex mixtures of chemically very similar substances (isomers, oligomers). This makes their analysis rather complicated and often requires the determination of global indices for the mixture rather than individual values for specific components.
1578
Anal Bioanal Chem (2007) 389:1577–1583
Many industrial cosmetic production processes are controlled by using classic analytical methods such as those involving the volumetric determination of the iodine, acid or hydroxyl values [2–4]. These methods are labourintensive and time-consuming, use large amounts of reagents and produce large amounts of waste. Product quality is additionally assessed in terms of sensory properties such as colour or odour, which provide little information about the overall quality of the product concerned and none about its components. Using a chromatographic method to analyse such complex samples is not always feasible or useful, either because the analytes are poorly resolved or because the information derived is of little significance. Spectroscopic methods provide a highly effective choice for the straightforward, expeditious analysis of many industrial products. Near-infrared (NIR) spectroscopy is especially useful in this context as it allows spectra to be recorded in a noninvasive, nondestructive manner at any point in the industrial production process without the need for any reagents. Also, the use of multivariate chemometric techniques facilitates the development of analytical methods for the simultaneous determination of both physical and chemical parameters from the NIR spectrum of the sample. In recent years, NIR spectroscopy has been used in a large number of instances for quality control [5–9] and process control [10–15] purposes. In this work, we used NIR spectroscopy for quality control analyses of a cosmetic base consisting of a mixture of paraffins, various natural and modified triglycerides, and sorbitol esters. Every component of the mixture was successfully determined from the NIR spectrum for the sample by using various partial least squares (PLS) calibration models.
obtained by reacting sorbitol with isostearic acid and is used as an emulsifier in cosmetic creams and lotions. Finally, poly (oxyethylene glycol) hydrogenated castor oil is a poly (ethylene glycol) derivative with an average content of 2 mol ethylene oxide per mole of castor oil; it is classified as both an alkoxylated alcohol and a glyceryl ester and derivative, and is typically used as a fragrance ingredient, and also as an emulsifying, solubilizing surfactant [16].
Experimental
Reference method for the hydroxyl value
Mixture components
The hydroxyl value was determined by titration with ethanolic KOH, using the method officially endorsed by the European Pharmacopoeia [17].
The cosmetic mixture studied consisted of hydrogenated castor oil (15% m/m, OIL_HY), poly(oxyethylene glycol) hydrogenated castor oil (30% m/m), paraffin wax (25% m/m) and sorbitan isostearate (30% m/m). Hydrogenated castor oil is a mixture of glycerides of ricinoleic and isoricinoleic acids plus small amounts of the glycerides of stearic and dihydrostearic acids; the oil is colourless to pale yellow, has a characteristic odour and is sparingly soluble in water. Paraffin wax is a mixture of saturated hydrocarbons of variable chain length (CnH2n+2, n=20–40) occurring as a waxy solid insoluble in water that is commonly used as a base for creams and masks. Sorbitan isostearate is a waxy solid
Production samples The product studied was an internal base for oil-in-water emulsions consisting of a mixture with the composition described in the previous section. The samples studied were portions of finished product collected from 11 different production processes. Laboratory samples A laboratory sample set was constructed by using mixtures of the previously described components at concentrations of ±5% m/m around their respective nominal values. The concentrations of the four components in the mixtures were selected in such a way as to minimize correlation between analyte pairs; all correlation coefficients thus obtained were less than 0.6. The previous concentrations were calculated by hand, using appropriate formulae in a spreadsheet. A total of 32 samples were prepared by weighing each ingredient on an analytical balance and subsequently mixing all four ingredients. In order to ensure homogeneous blending, each mixture was melted in a stove at 105 °C and held at that temperature for 30 min with stirring by hand several times prior to cooling. Once each mixture had been checked to be homogeneous, the melt was transferred to a vial for storage at room temperature.
NIR spectra NIR spectra were recorded in the transmittance mode using a model 5000 spectrophotometer from FOSS NIRSystems (Silver Springs, MD, USA) equipped with a vial heater module. The temperature was set at 100 °C. Sample spectra were obtained at 2-nm intervals over the wavelength range 1,100–2,500 nm. The vials used to hold the samples in order to record the NIR spectra were of 8-mm outer diameter and 6.5-mm path length.
Anal Bioanal Chem (2007) 389:1577–1583
1579
Fig. 1 Near-infrared (NIR) spectra from the pure components of the cosmetic product’s constituents. OIL_HY hydrogenated castor oil (15% m/m), PAR paraffin wax (25% m/m), OIL_OX hydrogenated castor oil (30% m/m), SOR and sorbitan isostearate (30% m/m)
Software The NIR instrument was controlled, and spectra were acquired and processed, by using the software Vision version 2.51, also from FOSS NIRSystems. The chemometric package used for data processing and model construction was Unscrambler version 9.2, from Camo Process (Trondheim, Norway). Data processing Partial least squares (PLS) models were constructed and principal component analysis (PCA) was performed by using both NIR spectra and reference values. To this end, we used various spectral pretreatments, including the standard normal variate (SNV) and first and second spectral derivatives, which were obtained by using the Savitzky–
Golay algorithm with an 11-point moving window and adjustment for a second-order polynomial. Different wavelength ranges were employed depending on the positions of the main bands for each analyte. PLS 1 calibration models were constructed by cross-validation, using the leave-one-out procedure. The number of PLS factors (latent variables) selected for each model was established from the minimum percentage relative standard error of calibration from the cross-validation values of the training set, and finally the model was refined applying the minimum percentage relative standard error of prediction of an external sample set, not used during the development of the model. The quality of the calibration models and their predictive ability were assessed in terms of the relative standard errors.
Results and discussion Table 1 Correlation coefficients between the spectra of the cosmetic product’s constituents
OIL_HY PAR OIL_OX SOR
OIL_HY
PAR
OIL_OX
SOR
1.0000 0.9890 0.9937 0.9945
1.0000 0.9862 0.9801
1.0000 0.9975
1.0000
OIL_HY hydrogenated castor oil (15% m/m), PAR paraffin wax (25% m/m), OIL_OX hydrogenated castor oil (30% m/m), SOR and sorbitan isostearate (30% m/m)
The four components of the product studied were paraffins and esters of long-chain aliphatic acids with very similar NIR spectra (Fig. 1) exhibiting no significant differences between their absorption bands. In fact, the NIR spectra for the four components exhibited two strong bands in the regions 1,700–1,800 nm (first overtone for the C–H bond) and 2,300–2,500 nm (combination band for the C–H bond). The greatest differences between analytes occurred in the 2,100-nm region (combination band for the O–H bond), where the paraffins exhibited no absorption, whereas
1580
Anal Bioanal Chem (2007) 389:1577–1583
Table 2 Correlation coefficients between the concentrations of the constituents in the laboratory samples
OIL_HY PAR OIL_OX SOR
OIL_HY
PAR
OIL_OX
SOR
1.00 –0.59 –0.06 –0.37
1.00 –0.40 0.00
1.00 –0.59
1.00
sorbitan isostearate absorbed maximally—because of its large number of O–H bonds—and the two castor oil derivatives exhibited bands of intermediate strength differing only slightly in shape and height. In any case, the NIR spectra were very similar, as reflected in the correlation coefficients between pairs (Table 1), which exceeded 0.98 in all instances. The reference method used for routine analytical control of the mixture studied is based on the determination of the hydroxyl value. No other effective method for quantifying the components of the mixture has been developed to date as chromatographic techniques lack the resolution level required to resolve the large number of signals obtained. This precludes the use of production samples to construct NIR calibration models as no effective method for obtaining the reference values needed to develop a multivariate calibration model based on NIR spectra exists as yet. Rather, the calibration set must be constructed from laboratory-made samples spanning the required analyte concentration range. The high spectral correlation between the components of the mixture precludes the calibration
Fig. 2 Principal component analysis score plot obtained from NIR spectra of laboratorymade samples. PC principal component
models with an acceptable predictive ability being obtained and requires minimizing correlation between laboratorymade samples—otherwise PLS models cannot discriminate spectral variations due to specific analytes. The proposed approach seeks to reduce correlation between the components of the mixture in order to obtain robust models with an acceptable predictive ability. We used a total of 32 laboratory samples with correlation coefficients between analyte pairs all less that 0.60 (Table 2). After their NIR spectra had been recorded, the samples were split into a calibration set and a prediction set in such a way as to encompass as wide a spectral variability range as possible (based on a PCA scores plot) while spanning the required concentration range for each analyte. Fig. 2 shows a scores plot used to select the samples for inclusion in the calibration and prediction sets for the OIL_HY calibration model. As can be seen, both sets spanned the variability of the first two principal components (PC1, 88% X-variance; PC2, 10% variance). One calibration set per component was selected and the remaining samples were used to construct the prediction sets spanning the required concentration range for each component. The selection of those samples sets was done by visual inspection of the scores plot of a PCA and the concentration ranges for each analyte. PLS calibration models for each analyte were tested in the absorbance, SNV and Savitzky–Golay first-derivative and second-derivative modes, both with the whole spectrum (1,100–2,500 nm) and with the specific regions where each analyte exhibited substantial bands; however, the strong
Anal Bioanal Chem (2007) 389:1577–1583
1581
Table 3 Figures of merit for the four partial least squares (PLS) calibration models
Calibration
NIR (%) =A + B Ref (%)
Prediction NIR (%) = A + B Ref (%)
Factor
OIL_HY
PAR
OIL_OX
SOR
Concentration range (%) Spectral pretreatment Wavelength range (nm)
9.6–19.8 2nd derivative 1,600–2,200
19.2–31.1 1st derivative 1,100–2,200
25.1–37.1 2nd derivative 1,600–2,200
24.1–35.1 2nd derivative 1,800–2,200
Number of samples PLS factors Y–explained variance (%) RSEC (%) Correlation (r) Intersection (A) Slope (B)
20 4 98.71 0.7 0.993 0.16±0.96 0.99±0.06
20 2 99.51 1.2 0.997 0.11±0.92 1.00±0.04
20 4 97.94 1.2 0.990 0.71±2.33 0.98±0.07
20 4 99.68 0.8 0.998 0.13±0.86 1.00±0.13
Number of samples RSEP (%) Correlation (r) Intersection (A) Slope (B)
12 1.0 0.989 1.23±1.35 0.93±0.10
12 0.9 0.997 1.21±1.27 0.95±0.05
12 1.3 0.976 –2.04±5.06 1.08±0.17
12 1.0 0.997 0.05±1.86 0.99±0.06
NIR near infrared, RSEC relative standard error of calibration, RESP relative standard error of prediction
spectral similarities between the four components precluded an accurate selection of the specific bands to be used with each. Table 3 shows the figures of merit provided by the calibration models with the highest predictive ability. We used the first-derivative or the second-derivative spectral treatment in all cases as such treatments provided better results than absorbance-treated and SNV-treated data. The 2,200–2,500-nm region was avoided as it provided poorer calibration results or failed to significantly improve the predictive ability of the models. The optimum number Fig. 3 Process control chart for the sum of each percentage component determination by the NIR method. SD standard deviation
of PLS factors was 2 or 4, depending on the particular analyte, and the calibration and prediction errors were always less than 1.2 and 1.3%, respectively. The plots of NIR concentration against reference concentration values were all straight lines of unity slope and zero intercept (95% significance). Therefore, the four PLS calibration models used were subject to no systematic errors and their results exhibited no significant differences from the reference values, which were obtained by weighing using an analytical balance.
1582
Anal Bioanal Chem (2007) 389:1577–1583
Table 4 Figures of merit for the hydroxyl value (OHV) calibration model
Calibration
NIR (%) =A + B Ref (%)
Prediction
NIR (%) =A + B Ref (%)
Factor
OHV
OH range (mg KOH/g) Spectral pretreatment Wavelength range (nm)
105–123
Number of samples PLS factors Y-explained variance (%) RSEC (%) Correlation (r) Intersection (A) Slope (B)
20
The hydroxyl value, which is the sole parameter used at present for quality control purposes with the product studied, is only useful for determining the overall quality of the mixture. The quality specifications for release of each batch include a hydroxyl value for the finished product ranging from 113 to 130 mg KOH/g sample. We developed a NIR method to determine the hydroxyl value as a replacement for the wet reference method. We determined the hydroxyl value for each laboratory sample by using the reference method and various PLS calibration models. The approach was similar to that used in determining the four components: available samples were used to construct two sets—one for calibration and the other for prediction—that were employed with various spectral ranges and treatments. Table 4 shows the results obtained with the model exhibiting the highest predictive ability. Very similarly to the previous models, using first-derivative spectra over the 1,100–2,200-nm range, we could construct a PLS model with four factors that featured calibration and prediction errors less than 1%. The results provided by the NIR method exhibited no significant differences from those obtained with the reference method. The method was used to control the previous production batches and was found to provide results meeting the quality specifications for the product. Table 5 summarizes the determinations performed on the different production batches studied. As can be seen, the NIR results were never significantly different from the product specifications.
1st derivative 1,100– 2,200
Number of samples RSEP (%) Correlation (r) Intersection (A) Slope (B)
3 98.29 0.5 0.992 2.06±7.28 0.98±0.06 12 0.6 0.995 0.69±8.57 1.00±0.08
The proposed NIR method, which affords the simultaneous quantification of the four components of the mixture from the NIR spectrum of the sample, was used for quality control analyses of finished product samples from 11 industrial processes. Fig. 3 shows the combined analyte concentrations predicted by the NIR method against those obtained with the batch number. As can be seen, the combined concentrations never exceeded the accepted limits for the process (95–105%); therefore, the proposed method possesses an acceptable predictive ability.
Conclusions The combined use of NIR spectroscopy and multivariate calibration affords the simultaneous determination of the components of a cosmetic mixture and its hydroxyl value.
Table 5 Results obtained by the NIR method Batch number
1 2 3 4 5 6 7 8 9 10 11
NIR prediction OIL_HY (15%)
PAR (25%)
OIL_OX (30%)
SOR (30%)
Sum of concentration (100%)
OHV (113–130 mg KOH/g)
13.7 15.3 14.4 14.8 15.1 14.8 15.2 16.1 16.0 16.0 15.5
25.3 24.3 24.1 25.0 24.3 23.6 23.8 25.0 23.6 23.7 25.3
29.6 29.5 30.6 29.3 29.6 29.6 28.3 27.4 28.3 28.3 28.1
31.1 30.3 30.6 30.7 29.4 30.4 30.5 31.4 30.7 30.7 30.9
99.7 99.4 99.7 99.8 98.4 98.4 97.8 99.9 98.6 98.7 99.8
114.5 115.2 116.3 114.5 115.0 116.8 117.0 115.3 116.5 116.5 114.9
The values in parentheses are the quality assurance specifications.
Anal Bioanal Chem (2007) 389:1577–1583
Unlike its classic, wet counterpart, the proposed method requires no sample pretreatment and can provide results in real time simply by using the NIR spectrum of the sample. Acknowledgement The authors are grateful to Spain’s Ministry of Education and Science for funding this research within the framework of project CTQ 2006–12923.
References 1. Neubauer M (1945) Formulario práctico de la cosmética moderna. Albatros, Buenos Aires 2. Orster TF, Waldmann M, Both W, Jassoy C (1999) Int J Cosmet Sci 21:253 3. Ogunniyi DS (2006) Bioresource Technol 97:1086 4. Blanco M, Villaescusa V (2007) Talanta 71:1333 5. Büning-Pfaue H, Kehraus S (2001) Eur J Lipid Sci Technol 103:793
1583 6. Flávio SG, Lima M, Araújo AS, Borges L (2004) J Near Infrared Spectrosc 12:159 7. Blagoi G, Bleotu A, Puica M, Vasilescu M (1998) J Near Infrared Spectroc 6:285 8. Grunewald H, Kurowski C, Timm D, Grummisch D, Meyhack U (1998) J Near Infrared Spectrosc 6:215 9. McClure F (2003) J Near Infrared Spectros 11:487 10. Cozzolino D, Parker M, Dambergs RG, Herderich M, Gishen M (2006) Biotechnol Bioeng 95:1101 11. Finn B, Harvey LM, McNeil B (2006) Yeast 23:507 12. Ionita-Manzatu M, Scarlat I, Vasilescu M, Puica M, Blagoi G (1998) J Near Infrared Spectrosc 6:211 13. Santos AF, Silva FM, Lenzi MK, Pinto JC (2005) Polym Plast Technol Eng 44:1 14. Blanco M, Castillo M, Peinado A, Beneyto R (2006) App Spectrosc 60:641 15. Houmøller LP, Kristensen D, Rosager H (2007) Talanta 71:868 16. Greensberg LA (1954) Handbook of cosmetic materials. Interscience, New York 17. European Directorate for the Quality of Medicines (2002) In: European pharmacopoeia, 4th edn. European Directorate for the Quality of Medicines, Strasbourg
