Direct Determination of Ni2+-Capacity of IMAC Materials Using Near-Infrared Spectroscopy

The present paper reports a new method for the quantification of the Ni2+-capacity of an immobilized metal affinity chromatography (IMAC) material using near-infrared spectroscopy (NIRS). Conventional analyses using UV absorption spectroscopy or atomic absorption spectrometry (AAS) need to dissolve the silica-based metal chelate sorbent as sample pretreatment. In the first step, those methods were validated on the basis of an ideal homogenous NiSO4-solution and unveiled that UV with an intermediate precision of 2.6% relative standard deviation (RSD) had an advantage over AAS with an intermediate precision of 6.5% RSD. Therefore, UV analysis was chosen as reference method for the newly established NIRS model which has the advantage of being able to measure the material directly in diffuse reflection mode. Partial least squares regression (PLSR) analysis was used as multivariate data analysis tool for quantification. The best PLSR result obtained was: coefficient of determination (R2) = 0.88, factor = 2, root mean square error of prediction (RMSEP) = 22 µmol/g (test-set validation) or 7.5% RSDPLSR. Validation of the Ni2+-capacity using UV absorption spectroscopy resulted in an intermediate precision of ±18 µmol/g or 5.0% RSD. Therefore, NIRS provides a fast alternative analysis method without the need of sample preparation.


Introduction
The affinity of proteins to transition elements such as zinc, copper, nickel, etc. in their ionic state bonded to a stationary phase was described in 1975 by Porath et al. for the first time [1]. Based on this knowledge further investigations were carried out on purification and analysis of proteins using this technique, which is nowadays well known as immobilized metal affinity chromatography (IMAC) [2][3][4][5][6]. Ni 2+ on an IMAC material based on a nitrilotriacetic acid adsorbent is found to bind specific to proteins which contain neighboring histidine residues [7]. This approach is the fundament for the purification of proteins due to their histidine content [8]. Therefore, six attached histidine residues (HisTag) are common as affinity tag and result in a high purification of the proteins in high yield [9,10].

IMAC Materials
The repeated syntheses of the IMAC materials showed that the resulting Ni 2+ -capacity was not reproducible. The minimum difference by using the same silica gels as starting materials was 50 µmol/g and went up to a maximum difference of 138 µmol/g of the Ni 2+ -capacity. Therefore, 32 IMAC materials differing in Ni 2+ -capacity were synthesized by reapplying 12 silica gels. The usage of different pore and particle size ensured that a variation of these physical parameters which strongly affect the NIR spectra was taken into account. Correlation analysis of the Ni 2+ -capacity and pore size gave a Pearson correlation coefficient [19] of −0.6 and therefore showed a moderate negative correlation. No correlation analysis was done for the particle size as just a range was specified by the producers. The Ni 2+ -capacity of all the IMAC materials was in the range from 156 to 444 µmol/g. Details of all 32 IMAC materials are given in Table S1 in the Supplementary Materials.

UV and AAS Method Validation
Two techniques are typically used for the quantification of the Ni 2+ -capacity of IMAC materials: UV absorption spectroscopy and AAS [11,12]. The first goal of this study was to identify the quantification method which is better suited as reference analysis for the establishment of a PLSR model on the basis of NIRS. As the precision of the reference values limits the possible performance of the PLSR model, the repeatability and the intermediate precision were used as decision criteria to determine the better suited reference analysis. In the first step, the IMAC materials were not used for this determination as they need an additional sample preparation step which takes an additional error into account and therefore was determined after (Section 2.3). Three solutions containing low, middle and high nickel concentration referred to the range of the calibration were prepared and used for the determination of the repeatability and intermediate precision. For the intermediate precision calibration, solutions were prepared freshly for each of the five measurement days but were used four times within a day for the repeatability. The coefficients of determination (R 2 ), limits of detection (LOD) and limits of quantification (LOQ) for the calibrations were 0.9949 ≤ R 2 ≤ 0.9998, 0.00028 ≤ LOD ≤ 0.0014 mol/L and 0.0010 ≤ LOQ ≤ 0.0047 mol/L for the UV method and 0.9821 ≤ R 2 ≤ 0.9972, 5.6 × 10 −6 ≤ LOD ≤ 1.4 × 10 −5 mol/L and 1.9 × 10 −5 ≤ LOQ ≤ 4.7 × 10 −5 mol/L for the AAS method, respectively. UV measurements resulted in a repeatability of 1.9% relative standard deviation (RSD) and intermediate precision of 2.6% RSD. AAS measurements resulted in a repeatability of 6.2% RSD and intermediate precision of 6.5% RSD. The results unveiled that the UV absorption spectroscopy method performed better than the AAS one and therefore was better suited for the determination of the Ni 2+ -capacity of the IMAC materials. The raw data and regression parameters for all nickel solution measurements and calibrations are given in Tables S2-S17 in the Supplementary Materials.

Reference Analysis Method Validation
In the second step, the performance of the Ni 2+ -capacity reference analysis for creating the PLSR model was evaluated. Therefore, six IMAC materials were synthesized for the determination of the repeatability and the intermediate precision. The R 2 , LOD and LOQ of all calibrations were 0.9977 ≤ R 2 ≤ 0.9998, 0.00028 ≤ LOD ≤ 0.0010 mol/L and 0.00092 ≤ LOQ ≤ 0.0032 mol/L, respectively. The Ni 2+ -capacity reference analysis gave a repeatability of 4.2% RSD or ±15 µmol/g and an intermediate precision of 5.0% RSD or ±18 µmol/g. In contrast to the UV method validation of the standard solutions which had an intermediate precision of 2.6% RSD (Section 2.2), the reference analysis resulted in a worse precision. This was caused by the additional sample preparation step in which the nickel had to be released from the IMAC material. Furthermore, handling of solid materials takes possible inhomogeneity into account, which also contributes to the lower precision of the analysis. The raw data and regression parameters for all precision measurements of the reference analysis are given in Tables S18-S23 in the Supplementary Materials.

Multivariate Data Analysis of NIR Spectra
Raw NIR spectra of all 32 IMAC materials used for MVA are shown in Figure 1. The band assignment was done by using the raw NIR spectra ( Figure 1) and the second derivative ( Figure 2). The latter one enhances spectral features and therefore resolves small bands. This means that a single peak in a zeroth order spectrum appears as negative band with its minimum at the same wavenumber as the maximum in the absorption band [20]. The broad band with its maximum at 9500 cm −1 appears after nickel ion immobilization. This signal can be assigned to the electronic transition of the nickel complex but is not due to overtone or combination vibration [21,22]. The bands at 7316 cm −1 and 7300 cm −1 observed through the analysis of the second derivative arise from the first overtone of the OH stretching vibrations of the free silanol groups (literature: 7316 cm −1 ) [20,[23][24][25]. The band at 7236 cm −1 can be assigned to silanol groups hydrogen bonded to water molecules (literature: 7225 cm −1 ) [20]. The band at 7116 cm −1 is due to water molecules hydrogen bonded to silanol groups and water molecules hydrogen bonded to other water molecules (literature: 7121 cm −1 ) [20]. The broad band at 6860 cm −1 (literature: ca. 6900 cm −1 [26] and 6861 cm −1 [20]) is due to combination involving the symmetric and asymmetric stretching modes of the water molecule [26]. The bands at 5900, 5828, 5724 and 5660 cm −1 in region from 6000 to 5600 cm −1 can be assigned to overtones of CH stretching vibrations [26]. Analysis of the second derivative ( Figure 2) reveals a band at 5316 which is due the water molecules adsorbed onto the free silanol groups on the surface (literature: 5314 cm −1 ) and a band at 5268 cm −1 due to the water molecules adsorbed onto hydrogen bonded silanol groups (literature: 5270 cm −1 ) [20,23]. The broad band, which appears at 5150 cm −1 , can be assigned to the combination of the asymmetric stretch and bending of the water molecule (literature 5150 cm −1 ) [26]. The bands at about 4500 cm −1 in the raw spectra or seen at 4568 and 4524 cm −1 in the second derivative represent the combination band of OH stretch with one of the SiO 2 fundamentals (literature: 4450 and 4520 cm −1 [26] or 4579, 4518 and 4422 cm −1 [20]). The bands in the region below 4400 cm −1 can be assigned to the first combination region of CH [26]. The best PLSR model was obtained by applying the following spectral pre-treatment: second derivative (Savitzky-Golay, polynomial order: 2, number of smoothing points: 5). Wavenumber regions used for the model were: 7500-6700 cm −1 and 6120-4008 cm −1 . Figure 2 illustrates the pretreated spectra and the removed regions (indicated by the crossed-out areas). The region from 10,000 to 7504 cm −1 was kept out since the very broad band that can be found in this region in the raw spectra ( Figure 1) gives no signal in the second derivative spectra ( Figure 2). This means that this band is unsuitable for the calculation of a multivariate PLSR model. The region 6696-6124 cm −1 only contains the foothills of broad bands. Therefore, no defined signals can be observed in the second derivative spectra. The results of the established PLSR model are depicted in Table 1. The regression coefficients for factor 2 which summarize the relationship between the wavenumbers and the Ni 2+ -capacity are plotted in Figure 3. By comparing the second derivative spectra in Figure 2 to the regression coefficients in Figure 3, it can be concluded that the observed regression coefficients correspond to the observed molecular vibrations described above. As the root mean square error of prediction (RMSEP) shows the best estimation for the future performance of the PLSR model [13], this quality parameter should be used for the direct comparison of the reference analysis and the introduced PLSR model. Due to the recording of the NIR spectra and the measurement of the reference values at different days, the RMSEP of 22 µmol/g (7.5% RSD PLSR ) regarding the Ni 2+ -capacity represents the intermediate precision of the NIRS method (including the PLSR model, NIRS measurements and MVA). The UV absorption spectroscopy method used as Ni 2+ -capacity reference analysis achieved an intermediate precision of ±18 µmol/g (5.0% RSD). Therefore, the NIRS model performed comparable. Raw spectra, reference data and the established PLSR models are provided in a Microsoft Excel File and The Unscrambler X File in the Supplementary Materials.  Table 1. The regression coefficients for factor 2 which summarize the relationship between the wavenumbers and the Ni 2+capacity are plotted in Figure 3. By comparing the second derivative spectra in Figure 2 to the regression coefficients in Figure 3, it can be concluded that the observed regression coefficients correspond to the observed molecular vibrations described above. As the root mean square error of prediction (RMSEP) shows the best estimation for the future performance of the PLSR model [13], this quality parameter should be used for the direct comparison of the reference analysis and the introduced PLSR model.     Table 1. The regression coefficients for factor 2 which summarize the relationship between the wavenumbers and the Ni 2+capacity are plotted in Figure 3. By comparing the second derivative spectra in Figure 2 to the regression coefficients in Figure 3, it can be concluded that the observed regression coefficients correspond to the observed molecular vibrations described above. As the root mean square error of prediction (RMSEP) shows the best estimation for the future performance of the PLSR model [13], this quality parameter should be used for the direct comparison of the reference analysis and the introduced PLSR model.      Abbreviations: full cross validation (CV), test-set validation (TSV), root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV) for CV, root mean square error of prediction (RMSEP) for TSV, coefficient of determination (R 2 ), relative standard deviation (RSD).
The following 12 silica gels differing in pore and particle size were gained: Carl Roth (Karlsruhe, Germany   Abbreviations: full cross validation (CV), test-set validation (TSV), root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV) for CV, root mean square error of prediction (RMSEP) for TSV, coefficient of determination (R 2 ), relative standard deviation (RSD).
The following 12 silica gels differing in pore and particle size were gained: Carl Roth (Karlsruhe, Germany

Activation of Silica Gel
First, 0.56 g (14 mmol) NaOH were dissolved into 28 mL H 2 O. Then, 3 g silica gel were added to the solution and stirred for 30 min at room temperature. Next, 2.3 mL 6 M HCl (14 mmol) were used to neutralize the solution. Afterwards, the silica gel was filtered and washed several times with H 2 O. The particles were dried at 45 • C under vacuum (300 mbar).

Preparation of GLYMO-IDA
A solution of 29.4 g NaOH and 19.6 g (144 mmol) iminodiacetic acid (IDA) in 210 mL H 2 O was prepared at 0 • C. Thirty-five grams (144 mmol) of 3-glycidyloxypropyl)trimethoxysilane (GLYMO) were slowly added during 30 min. The mixture was allowed to warm up to room temperature and was stirred for 4 h. Afterwards, the temperature was raised to 65 • C and the reaction mixture was stirred for 18 h [18].

Immobilization of GLYMO-IDA on the Silica Gel
Nine milliliters of 6 M HCl were added to 24 mL of the GLYMO-IDA solution to adjust the pH-value to 3.5. The solution was diluted with 50 mL H 2 O and 2 g of activated silica gel were added. The suspension was refluxed at 100 • C for 3 h. The mixture was allowed to cool down to room temperature. The particles were filtered, washed several times with H 2 O and dried at 45 • C under vacuum (300 mbar) [18].

Nickel Ion Immobilization
Forty milliliters of 0.1 M nickelsulfate solution were added to 1 g of the silica-GLYMO-IDA material. The suspension was stirred for 2 h at room temperature. To remove unbound nickel ions, the particles were filtered and washed with H 2 O, 0.5 M NaCl solution and H 2 O successively. The obtained turquoise IMAC material was dried at 45 • C under vacuum (300 mbar). The structure [6] is shown in Figure 4.

Activation of Silica Gel
First, 0.56 g (14 mmol) NaOH were dissolved into 28 mL H2O. Then, 3 g silica gel were added to the solution and stirred for 30 min at room temperature. Next, 2.3 mL 6 M HCl (14 mmol) were used to neutralize the solution. Afterwards, the silica gel was filtered and washed several times with H2O. The particles were dried at 45 °C under vacuum (300 mbar).

Preparation of GLYMO-IDA
A solution of 29.4 g NaOH and 19.6 g (144 mmol) iminodiacetic acid (IDA) in 210 mL H2O was prepared at 0 °C. Thirty-five grams (144 mmol) of 3-glycidyloxypropyl)trimethoxysilane (GLYMO) were slowly added during 30 min. The mixture was allowed to warm up to room temperature and was stirred for 4 h. Afterwards, the temperature was raised to 65 °C and the reaction mixture was stirred for 18 h [18].

Immobilization of GLYMO-IDA on the Silica Gel
Nine milliliters of 6 M HCl were added to 24 mL of the GLYMO-IDA solution to adjust the pHvalue to 3.5. The solution was diluted with 50 mL H2O and 2 g of activated silica gel were added. The suspension was refluxed at 100 °C for 3 h. The mixture was allowed to cool down to room temperature. The particles were filtered, washed several times with H2O and dried at 45 °C under vacuum (300 mbar) [18].

Nickel Ion Immobilization
Forty milliliters of 0.1 M nickelsulfate solution were added to 1 g of the silica-GLYMO-IDA material. The suspension was stirred for 2 h at room temperature. To remove unbound nickel ions, the particles were filtered and washed with H2O, 0.5 M NaCl solution and H2O successively. The obtained turquoise IMAC material was dried at 45 °C under vacuum (300 mbar). The structure [6] is shown in Figure 4.  [6]. X, available orbitals for electron donors (water, imidazole, etc.).

UV Absorption Spectroscopy
First, 1.5 mL 0.05 M EDTA solution were added to 100 mg of the IMAC material and mixed for 10 min at 1600 rpm using a Thermomixer C (Eppendorf, Hamburg, Germany). After centrifugation for 5 min at 13,200 rpm or 15,000 rcf using a Centrifuge 5418 (Eppendorf, Hamburg, Germany), the supernatant was removed and the procedure was repeated. The supernatants were combined and filled up to 5 mL using a volumetric flask. The solution was transferred into a cuvette and the absorption was measured in triplicate at λ = 380 nm using a Jenway Genova Plus Life Science Spectrophotometer (Cole-Parmer, Stone, UK). External calibration was done by dissolving nickel sulfate hexahydrate in 0.05 M EDTA solution.

UV Absorption Spectroscopy
First, 1.5 mL 0.05 M EDTA solution were added to 100 mg of the IMAC material and mixed for 10 min at 1600 rpm using a Thermomixer C (Eppendorf, Hamburg, Germany). After centrifugation for 5 min at 13,200 rpm or 15,000 rcf using a Centrifuge 5418 (Eppendorf, Hamburg, Germany), the supernatant was removed and the procedure was repeated. The supernatants were combined and filled up to 5 mL using a volumetric flask. The solution was transferred into a cuvette and the absorption was measured in triplicate at λ = 380 nm using a Jenway Genova Plus Life Science Spectrophotometer (Cole-Parmer, Stone, UK). External calibration was done by dissolving nickel sulfate hexahydrate in 0.05 M EDTA solution.  The solutions used for UV absorption spectroscopy were diluted with 0.05 M EDTA to a nickel concentration of approximately 0.14 mM or 8 µg/mL using a 50 mL volumetric flask to be in the range of the calibration (2-15 µg/mL). Measurements were done in triplicate. External calibration was done by dissolving nickel sulfate hexahydrate in 0.05 M EDTA solution.

Near-Infrared Spectroscopy
Measurements of the NIR spectra of all 32 samples were done using the NIRFlex N-500 FT-NIR spectrometer (Büchi, Flawil, Switzerland) equipped with the Solids Cell N500-001 and the Vial Add-on N510-002, which sequentially scans 6 samples. The operating software was NIR Ware 1.4.3010 (Büchi, Flawil, Switzerland). Spectra were recorded in diffuse reflection in the wavenumber region from 10,000 to 4000 cm −1 accumulating 64 scans with a spectral resolution of 8 cm −1 and a digital resolution of 4 cm −1 . Sample measurements were done in triplicate and averaged to one representative spectrum for each sample.

UV and AAS Method Validation
The precision of the UV absorption spectroscopy and AAS methods were evaluated by determination of the repeatability (also termed intra-assay precision) and intermediate precision referred to ICH guidelines [27,28]. Therefore, three solutions with low, mid and high nickel concentrations (11,23 and 34 mM for UV absorption spectroscopy and diluted to 4, 8, and 11 µg/mL for AAS, respectively) were prepared using nickel sulfate hexahydrate in 0.05 M EDTA. The measurements were carried out four times per day for five days. External calibration was done for each measurement. Six Calibration standards for UV absorption spectroscopy (5,13,21,29,37 and 45 mM) and AAS (2,5,7,9,12 and 15 µg/mL) were prepared freshly each day. Calculations of LOD and LOQ were executed according to DIN32645: 2008-11 [29]. Experimental data for all measurements are given in Tables S2-S17 in the Supplementary Materials.

Reference Analysis Method Validation
Method validation of the precision of the IMAC material Ni 2+ -capacity determination was done by using UV absorption spectroscopy which showed a better performance than AAS (Sections 2.2 and 2.3). Therefore, six IMAC materials were synthesized. Repeatability and intermediate precision were determined. The measurements were carried out three times per day for three days. Calibration solutions were prepared freshly on each of the three measurement days but were used three times within a day. External calibration was done as described in Section 3.6. Calculations of LOD and LOQ were executed according to DIN32645: 2008-11 [29]. Experimental data are given in Tables S18-S23 in the Supplementary Materials.

Multivariate Data Analysis
MVA was carried out by using the software The Unscrambler X Version: 10.5 (CAMO Software, Oslo, Norway). Since the measured spectra were recorded in reflectance, all spectra had to be transformed to absorbance units to ensure linear relation between spectral and reference data. Furthermore, spectra were pretreated by multiplicative scatter correction (MSC) [30] or standard normal variate (SNV) transformation [31] and first or second derivative. All spectral pretreatments were tested alone and in combination. Derivatives were carried out by Savitzky-Golay differentiation [32] with a quadratic polynomial and optimization of the applied smoothing points (SP). PLSR models were calculated with the NIPALS algorithm [33] and UV absorption spectroscopy data as reference analysis. Spectral regions without relevant information were kept out. Validation was carried out by test-set validation (TSV) and full cross validation (CV) also known as leave one out cross validation (LOOCV) [33]. TSV was done as follows: Two thirds of the IMAC materials (21 samples, distributed representatively over the nickel concentration range and including extreme values) were used for calibration. The remaining third (11 samples) were used as test-set. Evaluation of the calculated PLSR models was carried out by observation of following parameters: root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV) for CV, root mean square error of prediction (RMSEP) for TSV, coefficient of determination (R 2 ), regression coefficients, the number of factors and the relative standard deviation (RSD) calculated as follows: % RSD PLSR = (RMSECV or RMSEP/Mean value of the calibration range) × 100%.

Conclusions
UV absorption spectroscopy was unveiled to be superior to AAS for the quantification of the Ni 2+ -capacity of the synthetized IMAC materials. This method was selected as reference analysis for the establishment of a PLSR model using NIRS in hyphenation with MVA. The NIRS model performed almost as well as the UV reference analysis and better than the AAS method. Therefore, the analysis using NIRS represents an attractive alternative for the determination of the Ni 2+ -capacity of IMAC materials as it operates noninvasively, fast and needs no sample preparation.
Supplementary Materials: The following are available online, Table S1: List of synthesized IMAC materials, Table S2: Calibration data for method validation of UV absorption spectroscopy Day 1, Table S3: Calibration data for method validation of UV absorption spectroscopy Day 2, Table S4: Calibration data for method validation of UV absorption spectroscopy Day 3, Table S5: Calibration data for method validation of UV absorption spectroscopy Day 4, Table S6: Calibration data for method validation of UV absorption spectroscopy Day 5, Table S7: Measured sample concentrations of the method validation of UV absorption spectroscopy Day 1-2, Table S8: Measured sample concentrations of the method validation of UV absorption spectroscopy Day 3-4, Table S9: Measured sample concentrations of the method validation of UV absorption spectroscopy Day 5, Table S10: Calibration data for method validation of AAS Day 1, Table S11: Calibration data for method validation of AAS Day 2, Table S12: Calibration data for method validation of AAS Day 3, Table S13: Calibration data for method validation of AAS Day 4, Table S14: Calibration data for method validation of AAS Day 5, Table S15: Measured sample concentrations of the method validation of AAS Day 1-2, Table S16: Measured sample concentrations of the method validation of AAS Day 3-4, Table S17: Measured sample concentrations of the method validation of AAS Day 5, Table S18: Calibration data for method validation of the Ni 2+ -capacity determination by UV absorption spectroscopy Day 1, Table S19: Calibration data for method validation of the Ni 2+ -capacity determination by UV absorption spectroscopy Day 2, Table S20: Calibration data for method validation of the Ni 2+ -capacity determination by UV absorption spectroscopy Day 3, Table S21: Measured Ni 2+ -capacity of the method validation by UV absorption spectroscopy Day 1, Table S22: Measured Ni 2+ -capacity of the method validation by UV absorption spectroscopy Day 2, Table S23