Anal Bioanal Chem (2010) 397:3575–3579 DOI 10.1007/s00216-010-3890-6

ORIGINAL PAPER

Study of ε-caprolactone polymerization by NIR spectroscopy Marcelo Blanco & M. Jesús Sánchez & Manel Alcalà

Received: 26 March 2010 / Revised: 5 May 2010 / Accepted: 29 May 2010 / Published online: 30 June 2010 # Springer-Verlag 2010

Abstract Near-infrared (NIR) spectroscopy is proposed for the in-line quantitative and kinetic study of the polymerization of ε-caprolactone and eventually to facilitate realtime control of the manufacturing process. Spectra were acquired with a fibre-optic probe operating in transflectance mode immersed in the reactor. The NIR data acquired were processed using a multivariate curve resolution alternating least squares (MCR-ALS) algorithm. The proposed method allows calculation of the concentration and spectral profiles of the species involved in the reaction. The key point of this method is the lack of reference concentrations needed to perform the MCR-ALS method. The use of an extended spectral matrix using both process and pure analyte spectra solves the rank deficiency. The concentration profiles obtained were used to calculate a kinetic fitting of the reaction, but the method was improved by applying kinetic constraints (hard modelling). The rate constants of batches at different temperatures and the energy of activation for this reaction were calculated. Whenever possible, the hard modelling combined with the MCR-ALS method improves the fit of the experimental data: the results show good correlation between the NIR and reference data and allow the collection of high-quality kinetic information on the reaction (rate constants and energy of activation). Keywords Near-infrared spectroscopy . Process analysis . Chemometrics . Multivariate curve resolution . Kinetics

M. Blanco : M. J. Sánchez : M. Alcalà (*) Grup de Quimiometria Aplicada, Departament de Química (Unitat Analítica), Facultat de Ciències, Universitat Autònoma de Barcelona, 01893 Bellaterra, Barcelona, Spain e-mail: [email protected]

Introduction Biodegradable polymers possess interesting properties which enable their use in a variety of fields and make them effective alternatives to similar products by virtue of their fairly rapid degradation and the lack of toxicity of their degradation products. Their degradation rate depends on various factors, including sample thickness, moisture, temperature and oxygen content. Caprolactone polymers are an example of biodegradable polymers. They are used as gypsum substitutes in orthopaedic applications. The mixture of these polymers with starch is especially useful in the manufacturing of films, injected goods and thermoformed products, among others [1]. Even more interestingly, polycaprolactone is used as a matrix for controlled drug delivery applications [2]. Local delivery of drugs by biodegradable polymers allows the reduction of systematic side effects, ensures high local levels of the target drugs, avoids the need for a second surgery to remove the carrier and also avoids the need for recurrent intravenous injections or intravenous maintenance. Moreover, polycaprolactone has been extensively studied on account of its good hydrolysability and biocompatibility [3]. The ε-caprolactone polymerization reaction is performed in the presence of a carboxylic acid, such as lauric acid, that modifies the properties of the polymer [1, 3] and in the presence of a catalyst that improves the yield of the reaction. Many industrial production processes are chemically monitored by using volumetric or chromatographic methods. These methods are labour-intensive and timeconsuming, use large amounts of reagents and also produce large amounts of waste. Product quality is generally dependent on the time of the end point, but most of these analytical methods cannot provide information in real time. Spectroscopic methods provide a highly effective choice for

3576

the straightforward, expeditious analysis of many industrial products [4]. Moreover, the use of both spectroscopy techniques and effective analysis of the spectral data obtained provides the opportunity to improve process knowledge/understanding and provides a better industrial approach to obtain final products with higher quality [5]. Near-infrared (NIR) spectroscopy can be efficiently used for the in-line real-time monitoring of the ε-caprolactone polymerization reaction as it requires no sample pretreatment. The use of multivariate data analysis techniques allows the extraction of the relevant chemical information of the process from the NIR spectra. Quantitative measurements can be performed by the development of multivariate calibration models [6]. Partial least squares regression is the most widely used chemometric algorithm for this purpose, but previous to its calculation, it is necessary to obtain reference values using the most appropriate analytical technique [5]. There is a family of chemometric techniques known as self-modelling curve resolution (SMCR) that can be used to monitor the chemical composition during the course of a reaction. This method is especially attractive because it does not require reference analytical information to determine spectral changes in a process. The application of the SMCR techniques provides two types of information for each individual component of a mixture: the concentration profiles with time and the pure spectrum [7]. The alternating least squares (ALS) algorithm is the algorithm most widely used in combination with the SMCR method. Even though the ALS algorithm does not require reference information, it can profit from available knowledge about the target product [5]. In this work, NIR spectroscopy was used to monitor the polymerization of ε-caprolactone in the presence of lauric acid to obtain polycaprolactone. The application of the multivariate curve resolution (MCR)–ALS method provides a means for obtaining the concentration profiles, pure NIR spectra for each component, kinetic rate constants and energy of activation for the polymerization reaction. Two different strategies of MCR-ALS soft and hard modelling were tested. The relevant information from the NIR spectra must provide an improvement of the process knowledge/ understanding.

Experimental Pilot laboratory experiments Figure 1 shows the components of the polymerization reaction. Mixtures of ε-caprolactone and lauric acid (12:1 molar ratio) were prepared. Titanium(IV) butoxide was used as a catalyst. The reactions were conducted in a 1-L LabMax reactor (Mettler Toledo, Columbus, OH, USA). This appara-

M. Blanco et al.

tus allows the control of different parameters, including temperature, agitation and distillation. Four batches were used for polymerization at different temperatures (170, 180, 190 and 210°C). The ε-caprolactone was introduced into the reactor and heated to the selected temperature before lauric acid and the catalyst were added. The experimental conditions were selected to only evaluate the kinetic changes of the reaction rather than to optimize the yield or reduce the amount of by-products or impurities. Gas chromatography reference method A Hewlett-Packard HP5890 series II gas chromatograph (Agilent Technologies, Santa Clara, CA, USA) equipped with a flame ionization detector and a SUPELCOWAX-10 fused-silica capillary column of 30-m length × 0.32-mm diameter × 0.25-µm inner film thickness (Sigma-Aldrich, St. Louis, MO, USA) was used. The method consisted of the injection of 1 µL of diluted sample (injector temperature 250°C), using methyl palmitate as an internal standard and ethanol–xylene (1:1) as the solvent. The oven temperature was programmed at 180°C for 3 min, then a temperature ramp of 5°C min-1 was applied from 180 to 250°C and finally the temperature was maintained at 250°C for the last 1 min. The ε-caprolactone content was calculated by interpolation of a calibration regression line of analyte to internal standard peak area versus analyte concentration. NIR spectroscopy NIR spectra for the reaction mixture were recorded by using an immersed fibre-optic probe in transflectance mode. The spectra were acquired with a NIR spectrophotometer (Foss NIRSystems 5000, Silver Spring, MD, USA) equipped with a transflectance optical probe of 2.5-m length (fibre and probe) and 1-mm path length. The instrument was controlled via Vision version 2.51. Spectra were recorded at regular 3-min time intervals and each spectrum was the average of 32 scans over the wavelength range 1,100–2,500 nm at 2-nm resolution. Data analysis Experimental data were processed in MATLAB version 7.0 (The MathWorks, Natick, MA, USA). Soft and hard modelling methods were combined by using software from Tauler and de Juan [8]. Several reports have been published giving a theoretical background on the MCR-ALS algorithm and the application to NIR spectral data [9–12]. Multiplicative scatter correction was applied to the spectra to minimize scattering effects of the radiation [13]. Baseline offset correction was also applied to avoid wavelength ranges with negative absorbance.

Study of ε-caprolactone polymerization by NIR spectroscopy

3577

Fig. 1 Polymerization reaction of ε-caprolactone in the presence of lauric acid. cat catalyst

Figure 2 shows the NIR spectra of the monomer εcaprolactone, lauric acid and the polymer polycaprolactone. Polycaprolactone as a reference standard is not commercially available. However, the last spectrum obtained at the end of the polymerization reaction was considered as the reference spectrum for the polymer. In all cases, the maximum monomer concentration at the end of every polymerization reaction was lower than 1% w/w. As can be seen, the major spectral differences between the monomer and the polymer can be explained in three ranges: 1,650–1,800, 2,000–2,100 and 2,325–2,500 nm. In the range 1,650–1,800 nm, where the first overtone for the C–H bond shows absorption, the monomer presented two major bands, but the intensity of the first band (1,715 nm) decreased in the spectrum of the polymer. The range 2,000–2,100 nm, where there is a combination band for the O–H bond, presented interesting behaviour, where the maximum at 2,120 nm suffered a decrease in intensity and a displacement to higher wavelengths (2,140 nm). In the range 2,325–2,500 nm, the monomer presented a minimum and a maximum of intensity, whereas the polymer spectrum did not present relevant peaks. This latter range corresponds to the combination bands for the C–H bond. The spectrum of lauric acid is similar to the spectra of the monomer and polymer, with the typical NIR wide and overlapped bands. However, there are a number of significant spectral differences between both reagents and the product, as those observed in Fig. 3 (ranges 1,650–1,800, 2,000–2,100 and 2,325–2,500 nm) plus a small band at 1,440 nm.

Figure 3 shows the NIR spectra of a batch prepared at 170°C. The spectral changes marked with arrows are consistent with those explained for Fig. 2. The most relevant changes were observed in the ranges 1,650– 1,800, 2,000–2,100 and 2,325–2,500 nm. As the reaction evolved, the intensities of the bands related to the monomer decreased, whereas the intensities of the bands of the polymer increased. The initial spectra of all the batches were not considered for analysis because of the inhomogeneity of the mixture, which contributed to non-reproducible spectral acquisition. The initial mixture of lauric acid and εcaprolactone is a liquid solution with two phases, the top phase of monomer and the bottom phase of acid. After the first minutes of the reaction, both phases merge into a turbid dispersion, as drops of acid into a mixture of monomer/polymer. Finally, the solution is totally clear and the reaction continues. After 5 min of stirring, the mixture was homogenous and the spectra acquired showed adequate reproducibility. Two strategies of soft and hard modelling were tested to obtain in all cases the concentration profiles of the reagents and product for individual batches (170, 180, 190 or 210°C), the NIR spectra of each compound, the kinetic rate constants at each temperature and the energy of activation of the reaction. The two strategies can be summarized as follows: (1) MCR-ALS soft modelling for each temperature and (2) MCRALS hard modelling using kinetic constraints for each temperature. The application of singular value decomposition (SVD) allows the mathematical rank of the spectral matrix to be used

Fig. 2 Near-infrared spectra of the reagents and the product: a 1,650– 1,800 nm, b 2,100–2,000 nm, c 2,325–2,500 nm

Fig. 3 Spectra of the process during a polymerization (example 5 h of reaction): a 1,650–1,800 nm, b 2,100–2,000 nm, c 2,325–2,500 nm

Results and discussion

3578

for the MCR-ALS experiments to be checked. In this case, the NIR spectral matrix corresponding to the polymerization performed at 170°C was used for this purpose, and the SVD provided a dimensionality of 2, which means the presence of only two components during the reaction. However, the reaction consisted of three compounds: monomer, lauric acid and polymer. Therefore, the spectral matrix of a single batch was rank-deficient, which can be ascribed to the high correlation between the spectra of the reactants and products. To accurately solve the rank deficiency, every spectral matrix for each temperature (170, 180, 190 or 210°C) was extended with the spectra of the three components to resolve rotational ambiguity. This strategy was confirmed by SVD, and in all cases a dimensionality of 3 was obtained. The first MCR-ALS strategy of soft modelling was performed by applying spectral and concentration nonnegativity restrictions, and a concentration closure constraint (sum of concentrations equal to 1) for a system of three components. Figure 4 shows the calculated NIR spectra of the monomer, lauric acid and polymer for the polymerization performed at 170°C. The variance in this calculation was 99.9997%. The comparison between the spectra in Fig. 3 (reference) and those in Fig. 4 (MCR-ALS) reveals a high match.The correlation coefficient between the reference and the MCR-ALS spectra was higher than 0.98 in all cases. Figure 5 shows the calculated concentration profiles for the three components. The monomer concentrations calculated by the gas chromatography (GC) reference method are also plotted in the same figure. The correlation between the NIR and reference profiles is remarkable (R2 of 0.996 and a relative error of 2.02%). The reaction ceased to evolve appreciably after 250 min, so it was assumed to finish within that time. The soft modelling calculation was repeated for the

M. Blanco et al.

Fig. 5 Concentration profiles of the three components calculated for the process at 170°C by the application of the MCR-ALS method. Values for monomer determination by the reference method are shown. GC gas chromatography

rest of the batches (180, 190 and 210°C). In all cases, the correlation coefficient between the calculated and the reference spectra was higher than 0.98. The polymerization of ε-caprolactone is a first-order reaction in terms of the monomer [14]. A first-order kinetic model can be calculated using the monomer concentrations during the reaction. A first-order kinetic fitting should obey Eq. 1: ln ½"
caprolactone ¼
kt þ ln ½"
caprolactone0 ;

ð1Þ

where [ε-caprolactone]0 is the initial monomer concentration, [ε-caprolactone] is the monomer concentration at time t and k is the reaction rate constant. This kinetic fitting was calculated using the GC reference concentrations of the monomer for the polymerization performed at 170°C. The calculated rate constant was 0.0119 min-1, with a correlation R2 of 0.9943. The rate constant calculated with the NIR concentrations of the monomer determined by the GC reference method was 0.0121 min-1. Both reference and NIR kinetic fittings gave very similar results. The rate constants calculated using the NIR

Table 1 Kinetic constants and R2 obtained for the different models

Fig. 4 Spectral profiles of the three components obtained after the application of the multivariate curve resolution alternating least squares (MCR-ALS) method to the data of the process at 170°C: a 1,650–1,800 nm, b 2,100–2,000 nm, c 2,325–2,500 nm

T (°C)

Soft modelling

170 180 190 210

k (min-1) 0.0121 0.0122 0.0262 0.1519

Hard modelling R2 0.993 0.987 0.991 0.986

k (min-1) 0.0126 0.0141 0.0376 0.1809

R2 0.995 0.997 0.999 0.995

Study of ε-caprolactone polymerization by NIR spectroscopy

3579

concentration profiles for all the temperatures are summarized in Table 1. In all cases, the correlation of the leastsquares curve fitting to calculate the rate constants was higher than 0.99. The second strategy involved the combined use of the MCR-ALS algorithm with a specific kinetic behaviour of the process as a constraint. This method allows concentration profiles and reaction kinetics to be calculated. In addition, forcing the system to follow a first-order kinetic model helps minimize the typical rotational ambiguity of the MCR-ALS algorithm, thereby improving resolution and providing the rate constants for the process as additional output information. This computational method was applied to all batches (170, 180, 190 and 210°C). In this case, an extended matrix of batch spectra plus the spectra of the pure components was also used to improve the initial rank deficiency. The correlation between the calculated and the reference spectra for the three components was high (R2 > 0.98) but not significantly better than that obtained by the soft modelling strategy. Also, successful results were obtained for the calculation of the NIR concentration profiles. The correlation R2 between the NIR and reference profiles was higher than 0.997, which is slightly higher than that obtained with the first MCR-ALS strategy. Table 1 shows the rate constants for this strategy. The rate constants are slightly higher than those calculated with the soft modelling strategy, but the correlation for each kinetic fitting was better (R2 >0.995) with the hard modelling calculations. The rate constants obtained at different temperatures can be used to calculate the energy of activation. The Arrhenius equation (Eq. 2) relates the rate constant of a reaction to temperature: k ¼ A:e
Ea=RT ;

ð2Þ

where A is a constant, Ea the energy of activation and R the gas constant (8.314472 JK-1·mol-1). Equation 2 can be expressed in logarithmic form as ln k ¼ ln A
ðEa =RT Þ:

ð3Þ

The constant (A) and the energy of activation (Ea) were calculated by using least-squares fitting of ln k versus 1/RT. The calculated energies of activation were -119.1 kJ mol-1 (R2 =0.915) and -125.6 kJ mol-1 (R2 =0.949) for the first and second MCR-ALS strategies, respectively. The two strategies gave similar energies of activation. However, the MCR-ALS hard modelling calculation allowed a better fit between the rate constants and the temperature to be obtained.

Conclusions The polymerization reaction between ε-caprolactone and lauric acid was monitored by NIR spectroscopy and the MCR-ALS method was used to analyse the spectral data obtained. Soft and hard modelling were tested for each temperature used. In all cases, concentration profiles and NIR spectra of the three components were obtained and no significant differences were obtained between MCR-ALS soft and hard modelling for the calculation of concentration profiles and NIR spectra. However, hard modelling is preferred over soft modelling because of four aspects: (1) slightly better spectral matching between reference and MCR spectra, (2) slightly better quantitative matching between reference GC values and MCR results, (3) better kinetic fitting for each temperature and (4) better fitting for the calculation of the energy of activation. It has been concluded that the combination of NIR spectroscopy and the MCR-ALS method is an adequate choice for monitoring this type of reaction with a minimum of supportive experiments. The method allows improvement of process knowledge and understanding through the study of kinetic parameters. Acknowledgement The authors are grateful to Spain's MCyT for funding this research within the framework of project CTQ2009-08312.

References 1. Centro de Información Técnica (2007) Degradación de los materiales plásticos. Boletín técnico informatico no 21. http:// www.plastivida.com.ar/pdf/21.pdf 2. Gonsalves KE, Chen X, Cameron JA (1992) Macromolecules 25:3309–3312 3. Giavaresi G (2004) Biomed Pharmacother 58:411–417 4. Blanco M, Castillo M, Peinado A, Beneyto R (2006) Appl Spectrosc 60:641–647 5. Blanco M, Castillo M, Beneyto R (2007) Talanta 72(2):519–525 6. De Beer TRM, Alleso M, Goethals F, Coppens A, Heyden Y, De Diego H, Rantanen J, Verpoort F, Vervaet C, Remon JP, Baeyens WRG (2007) Anal Chem 79(21):7992–8003 7. Pedge NI, Walmsley AD (2007) Appl Spectrosc 61(9):940–949 8. Jaumot J (2010) Multivariate curve resolution homepage. http:// www.mcrals.info/ 9. Tauler R, Kowalski B, Fleming S (1993) Anal Chem 65 (15):2040–2047 10. Tauler R, Smilde A, Kowalski B (1995) J Chemom 9:31–58 11. de Juan A, Maeder M, Martinez M, Tauler R (2000) Chemom Intell Lab Syst 54:123–141 12. Jaumot J, Gargallo R, de Juan A, Tauler R (2005) Chemom Intell Lab Syst 76:101–110 13. Geladi P, MacDougall D, Martens H (1985) Appl Spectrosc 39:491–500 14. Cayuela J, Bounor-Legare V, Cassagnau P, Michel A (2006) Macromolecules 39:1338–1346