Journal of Chromatography A, 1341 (2014) 73–78

Contents lists available at ScienceDirect

Journal of Chromatography A journal homepage: www.elsevier.com/locate/chroma

Selective separation of ferric and non-ferric forms of human transferrin by capillary micellar electrokinetic chromatography b ´ Paweł Nowak a , Klaudyna Spiewak , Julia Nowak a , Małgorzata Brindell b , a,∗ ´ ˙ Michał Wozniakiewicz , Grazyna Stochel b , Paweł Ko´scielniak a a b

Jagiellonian University in Kraków, Faculty of Chemistry, Department of Analytical Chemistry, Kraków, Poland Jagiellonian University in Kraków, Faculty of Chemistry, Department of Inorganic Chemistry, Kraków, Poland

a r t i c l e

i n f o

Article history: Received 5 February 2014 Received in revised form 9 March 2014 Accepted 11 March 2014 Available online 18 March 2014 Keywords: Electrophoretically mediated microanalysis Iron saturation Micellar electrokinetic chromatography On-line reaction Short-end injection Transferrin

a b s t r a c t The previously published method allowing the separation of non-ferric (iron-free) and ferric (ironsaturated) forms of human serum transferrin via capillary electrophoresis has been further developed. Using a surface response methodology and a three-factorial Doehlert design we have established a new optimized running buffer composition: 50 mM Tris–HCl, pH 8.5, 22.5% (v/v) methanol, 17.5 mM SDS. As a result, two previously unobserved monoferric forms of protein have been separated and identified, moreover, the loss of ferric ions from transferrin during electrophoretic separation has been considerably reduced by methanol, and the method selectivity has been yet increased resulting in a total separation of proteins exerting only subtle or none difference in mass-to-charge ratio. The new method has allowed us to monitor the gradual iron saturation of transferrin by mixing the iron-free form of protein with the buffers with different concentrations of ferric ions. It revealed continuously changing contribution of monoferric forms, characterized by different affinities of two existing iron binding sites on N- and C-lobes of protein, respectively. Afterwards, the similar experiment has been conducted on-line, i.e. inside the capillary, comparing the effectiveness of two possible modes of the reactant zones mixing: diffusion mediated and electrophoretically mediated ones. Finally, the total time of separation has been decreased down to 4 min, taking the advantage from a short-end injection strategy and maintaining excellent selectivity. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Human serum transferrin (Tf) is the iron-binding transport protein essential for delivering of iron ions into the cells. It binds two trivalent iron (ferric) ions to the N- and C-binding sites, respectively, resulting in the entirely iron-saturated form of protein (holo, h-Tf). Alternatively only one ferric ion can be bound to the N- or Cbinding site, hence two possible monoferric forms of Tf may exist: FeN –Tf and FeC –Tf [1,2]. The efficient separation of iron-free (apo, a-Tf), monoferric, and holo forms was a problematic task, because they exhibit the structural and chemical similarity [3]. Both a capillary zone electrophoresis and a micellar electrokinetic chromatography (MEKC) have been reported to be suitable techniques for the analysis of proteins and peptides [4–6]. There are numerous parameters which may be adjusted to improve the

∗ Corresponding author at: Jagiellonian University in Kraków, Ingardena 3, 30-060 Kraków, Poland. Tel.: +48 12 663 20 84; fax: +48 12 663 20 84. E-mail addresses: [email protected], ´ [email protected] (M. Wozniakiewicz). http://dx.doi.org/10.1016/j.chroma.2014.03.037 0021-9673/© 2014 Elsevier B.V. All rights reserved.

separation, however, the fast method optimization is concurrently impeded. Therefore, the use of experimental design approach is advisable, especially in the case when the results seem to be dependent on several factors with no determined correlation. MEKC, due to the surfactant presence, minimizes the adsorption of proteins to the capillary inner surface, thus making the results more precise and reproducible [7–10]. Tf was a subject of several CE-based analysis conducted usually with the use of MEKC. These studies were mainly focused on its determination in the biological material, including determination of a carbohydrate-deficient transferrin, a common marker of alcohol abuse [11–15]. Recently, we have also published a method allowing the separation of a-Tf and h-Tf using MEKC technique [16]. Our protocol was based on the addition of 7 mM SDS to running buffer consisting of 50 mM Tris–HCl, pH 8.5. Unfortunately, the presence of the two monoferric forms of Tf (FeN –Tf and FeC –Tf) could not be confirmed by using described method. Our present work was aimed at the further method development to enable the detection of all four forms of Tf including the differentiation between FeN –Tf and FeC –Tf, prevention from iron release during electrophoretic separation, and reduction in

74

P. Nowak et al. / J. Chromatogr. A 1341 (2014) 73–78

separation time. A particular effort has been made to provide the evidence that the peaks observed on the electropherograms are the assumed Tf forms, and to investigate the contribution of each ferric form at different saturation levels. Distinct analytical purpose was to attempt two modes of on-line protein saturation conducted directly inside the capillary, and to use the experimental design and the surface response methodology for method optimization. 2. Materials and methods Human a-Tf, human h-Tf (powder, BioReagent, suitable for cell culture, ≥98%), and human serum albumin (HSA) (powder, fatty acids free, globulin free, ≥99%), iron(III) nitrate nonahydrate, and nitrilotriacetic acid (NTA) were obtained from Sigma–Aldrich (Germany). Chemicals of analytical reagent grade: sodium hydroxide and ethanol were supplied by POCH S.A. (Gliwice, Poland), while LC–MS grade solvents: methanol, isopropanol an acetonitrile were purchased from Sigma–Aldrich (Germany). All solutions were prepared in MiliQ quality water and filtered through 45 ␮m regenerated cellulose membrane and degassed by centrifugation. All separation buffers were stored at +4 ◦ C, while NaOH and HCl solutions at ambient temperature. The samples of proteins were prepared in 50 mM Tris buffer, pH 8.5 with neither surfactant nor organic solvent addition, at the final concentration ranging from 0.10 to 0.50 mg/mL. The accurate value of buffer pH was obtained by using concentrated HCl (Tris–HCl). Before each analysis, samples were centrifuged for 5 min, 5000 × g. The minimal volume of the sample was 50 ␮L. The measurements were performed using a P/ACE Capillary Electrophoresis System (Beckman-Coulter, USA) equipped with a diode array detector. During all experiments the whole spectra within the range of 200–600 nm were collected, however, the results obtained at 200 nm were used in further processing. The uncoated fused-silica capillaries (laser burned detection window) were of 60 cm × 50 ␮m i.d. or 30 cm × 50 ␮m (Beckman-Coulter), with a 50 or 20 cm distance to the detector, respectively. Temperature of the sample tray and capillary were set up to 22 ◦ C. The capillary was being rinsed between runs as follows: 0.138 MPa (20 psi) of MiliQ water for 2 min, 0.138 MPa of 0.1 M NaOH for 3 min; and 0.138 MPa of running buffer for 3 min. Before the first run at a working day the rinsing protocol involved 0.138 MPa of methanol for 6 min; 0.138 MPa of 0.1 M HCl for 4 min; 0.138 MPa of MiliQ water for 3 min; 0.138 MPa of 0.1 M NaOH for 10 min; and 0.138 MPa of running buffer for 10 min, whereas in case of the first use of the capillary after mounting in cartridge the same procedure has been used, but the duration of all steps was doubled. Sample injection was performed using a forward pressure at anodic side, applying: 3.4 kPa (0.5 psi) for 5 s, unless stated otherwise. Forward voltage ranging from 15 kV to 30 kV was being applied. Each analysis has been repeated at least three times. The instrumental noise produced during detection has been smoothed out using Origin 8.0 software (OriginLab Corporation, USA). During chemometric calculations Statistica 10 (StatSoft Inc., Tulsa, OK, USA) software was used. Conductivity measurements were performed using a microcomputer conductivity meter (Elmetron CC-551) with a conductivity sensor (CD-2 type, 0.51 cm−1 sensor constant). 3. Results and discussion 3.1. Addition of organic solvent In the beginning, the effect of four different organic solvents as the buffer additives has been tested: methanol, ethanol,

propan-2-ol, and acetonitrile, each one in final 20% (v/v) concentration. The solvents were added to the 50 mM Tris buffer, pH 8.5, containing 10 mM SDS. According to the previously reported data, in this buffer SDS is able to form the micelles, probably a crucial factor enabling the separation of holo and apo forms of Tf. A reference buffer was the buffer without the addition of any organic modifier. The results have been generally compared by separating the sample containing mixed h-Tf, a-Tf and HSA. HSA was used as an internal standard to compare the obtained relative migration times standing for the particular transferrin forms and for calculation of relative peak areas. Among the four tested organic solvents, each one resulted in altered electropherogram, however only in the case of methanol the appearance of the two minor, totally separated peaks has been reported, localized in a gap between the peaks corresponding to hTf and a-Tf. Another effect which has been observed for methanol was the change in peak intensity, i.e. the peak standing for h-Tf had similar intensity as that for a-Tf, contrary to the buffer without methanol where the peak of h-Tf was considerably diminished. It has supported us to conclude, that the loss of iron by h-Tf can be minimized in presence of the methanol (see Supplementary Material for more detailed investigations of the role played by methanol). Apart from the organic solvents, also the addition of urea in final concentration of 6 M, 1 M, and 1 M with combination with 20% methanol, has been investigated. Urea was reported to improve separation of Tf forms according to the iron saturation in gel electrophoresis [17]. In our case it has turn out, that the changes in electropherograms have been more extensive than those caused by the solvents, however the peaks were incompletely separated and vastly diminished, i.e. the sensitivity of the method was considerably weaker after addition of urea. In the end, we have concluded that methanol without urea is the most promising buffer additive. Then, the effect of increasing concentration of methanol in the running buffer has been investigated, as it has been shown in Fig. 1. Final concentration of Tris 50 mM, SDS 10 mM and pH 8.5 were kept for each buffer. The methanol-free buffer and the buffers containing: 10.0%, 17.5% and 25.0% (v/v) methanol were compared. It has revealed the strong and increasing with concentration effect of this additive on the peak area ratio between holo and apo forms, and the appearance of two novel peaks. We have hypothesized, that these peaks may originate from monoferric forms of protein, previously unobserved. The analysis performed for the pure h-Tf and a-Tf resulted also in the appearance of these peaks, but their intensity was very low. 3.2. Experimental design Taking into account that finding of the optimal conditions for a method involving addition of the two different but crucial buffer components is not a trivial task, we have decided to use experimental design approach. For that purpose, set of experiments based on a three-factorial Doehlert uniform shell design and surface response methodology was conducted [18]. The optimized factors were: concentration of methanol cMeOH , concentration of SDS cSDS and pH. Preliminary experiments performed initially helped us to point out the range of optimized factors (see Supplementary Material for more details) [19]. Three system responses have been chosen: time of analysis defined as time of the end of HSA peak, sum of FWHM (Full Width at Half Maximum) for all five peaks of interest, and inverse resolution between h-Tf and HSA peaks. Such defined responses reflect the improvement in separation if their values are decreasing. According to the surface response methodology, a quadratic polynomial response model for the sample was

P. Nowak et al. / J. Chromatogr. A 1341 (2014) 73–78

75

Fig. 1. Electropherograms presenting the separation of the mixed sample containing h-Tf, a-Tf, and HSA, with reference to the increasing concentration of methanol, as well as the samples containing only h-Tf or a-Tf with HSA in the highest methanol concentration. Each Tf form was prepared at final concentration 0.33 mg/mL, while HSA at 0.33 mg/mL for mixed sample and 0.16 mg/mL for Tf standards sample. The value of voltage was: 15 kV for buffer without methanol, 20 kV for 10% methanol, 25 kV for 17.5% methanol, and 30 kV for 25% methanol, to ensure the similar electrophoretic mobilities. Effective capillary length was 50 cm.

estimated, according to formula (1).

3.3. Gradual Tf saturation

2 2 yk = ˇ0 + ˇ1 cMeoH + ˇ2 cSDS + ˇ3 pH + ˇ11 cMeoH + ˇ33 cSDS

In the next step, an experiment assuming the preparation of samples characterized by various iron saturation of a-Tf by ferric ions has been carried out. To this end, 6 mixtures with different a-Tf and ferric ions ratios (added as the nitrilotriacetic-Fe(III) complex, Fe-NTA) have been prepared [22], and incubated 20 min at ambient temperature in vials placed directly on a sample tray in CE instrument. Hence, it was an example of an off-line process conducted in an at-inlet format. After incubation, the samples were subjected to the CE analysis in optimized conditions and using a truncated capillary to 20 cm of effective length and 15 kV voltage. Electropherograms obtained for all of the 6 samples have been shown in Fig. 3. As it can be seen, this experiment allowed us to monitor the gradually increasing saturation of a-Tf by ferric ions, revealing that intensities of the two novel peaks are considerably changing with increasing concentration of iron in sample. Moreover, different behavior of these peaks has been reported. One of them is rapidly declining, while the second one is gradually growing up, reaching the intensity similar as that of h-Tf in a fourth sample. For the last sample, a complete saturation of a-Tf to h-Tf was observed. These results prove, that the peaks localized between h-Tf and a-Tf are the two predicted monoferric forms of protein, completely separated by this method and sensitive on the concentration of iron in sample. After calculation of the relative peak areas (Fig. S-3) we have concluded, that obtained outcomes are consistent with the data reporting on different binding affinities of the two iron biding sites and an existing cooperativity between them (see Supplementary Material for more details). As a consequence, monoferric forms have been identified, i.e. FeC –Tf is a faster migrating form than

+ ˇ33 pH2 + ˇ12 cMeOH cSDS + ˇ13 cMeOH pH + ˇ23 cSDS pH

(1)

where yk is the given response of the system (time of analysis, sum of FWHM or inverse resolution). The coefficients ˇi and ˇij were calculated by constructing the model by means of general linear regression, with the application of effective hypothesis test [20]. Statistical significance of coefficients ˇi and ˇij was evaluated with the use of the ANOVA test at the level ˛ = 0.05. According to the developed models, all three factors (cMeOH , cSDS , and pH) have significant influence on analysis time, both methanol and SDS concentration and SDS are significant with regard to the FWHM values, inverse resolution of chosen peaks is influenced by none of these factors. This may be caused by relatively good resolution in most of the experiments. The correlations between factors were found insignificant and the calculated model reveals mostly linear relations between factors and analyzed responses. In order to estimate conditions optimal for all of the responses of the system, a Derringer and Suich desirability function (D function) [21], has been applied to the calculated model. The desirability values for the analyzed factors were between 0 for the lowest response (yk ) values and 1 for the highest. The overall function D has been calculated as the geometric mean of the individual desirability values. Function D surfaces diagrams are presented in Fig. 2. The optimum conditions correspond to a local maximum value of function D, and they are as follows: cMeOH = 22.5% (v/v), cSDS = 17.5 mM (w/v), pH 8.5. Since that time, these values have been kept for running buffer during the all next experiments.

76

P. Nowak et al. / J. Chromatogr. A 1341 (2014) 73–78

Fig. 3. Electropherograms obtained for the samples containing equal amount of aTf (0.16 mg/mL) and HSA (0.10 mg/mL) in buffer composed of 50 mM Tris, 100 mM NaCl, 25 mM NaHCO3 , pH 7.4, and different concentration of iron added as Fe-NTA, incubated at-inlet for 20 min. Molar ratio of a-Tf to iron amounting to 1:2 theoretically enables the complete saturation of protein.

Fig. 2. Diagrams of function D surfaces: cSDS –cMeOH , cSDS –pH, cMeOH –pH, respectively. It is worth highlighting, that searching for a maximum value of function D beyond the considered ranges of parameters has been excluded on the basis of the first preliminary set of experiments. In particular, cMeOH lower than 22.5% (v/v) initiated undesirable iron release, and pH higher than 8.5 strongly deteriorated the peak shape of h-Tf (see Supplementary Material).

FeN –Tf. To our knowledge, this is the first demonstration of a gradual saturation of a-Tf by iron using the chromatography-related technique able to distinguish between all possible forms of protein. Subsequently, the potential of the CE instrument for performing on-line reactions has been attempted. In this case, the binding between protein and metal ions were assumed to be possible

directly inside the capillary, as the effect of interactions between the distinct zones injected consecutively into capillary. For that purpose two different methodologies are possible. Firstly, the reaction may be triggered by diffusion-mediated mixing, i.e. an idle process occurring when the zones get in a direct contact inside capillary. Secondly, the efficiency of mixing may be improved via electrophoretically-mediated mixing, by applying the voltage directly after injections to mix properly the reactants. This methodology was broadly used to investigate the activity of different types of enzymes, and is known as an electrophoretically mediated microanalysis (EMMA) [23–27]. In both cases, an additional incubation step may be introduced without the voltage application, when the reaction might take place. Thus, during incubation period the capillary played a role of a specific microreactor. The largest advantage of on-line approach is that the volumes of reactants needed to be injected into capillary are even three orders of magnitude lower than in off-line approach. Both mixing modes have been applied to study iron binding to a-Tf. The obtained electropherograms showing the similar level of protein saturation by iron are depicted in Fig. 4(A). In addition, the schematic illustration of these methodologies is depicted there as well. The molar ratio between iron and protein at the point of contact between zones was 10:1, which reflects the concentration of iron and protein in given zones. Electrophoretically-mediated mixing combined with a step of an idle incubation of the sample in capillary without the voltage application was more efficient than the diffusion-mediated mode combined with the same step of incubation. To reach similar iron saturation of Tf, the step of incubation required in the case of electrophoretically-mediated mixing was five times shorter than that in the case of diffusion-mediated approach. More details concerning these experiments and additional electropherograms obtained for different times of incubation are available in Supplementary Material. 3.4. Short-end injection At the end, we have decided to apply another variant of CE allowing for the large reduction in time of analysis, i.e. a

P. Nowak et al. / J. Chromatogr. A 1341 (2014) 73–78

77

Fig. 4. The outcomes obtained after the application of on-line reaction and short-end injection approaches. (A) – iron saturation of a-Tf performed inside capillary applying diffusion-mediated (top) and electrophoretically mediated (bottom) modes. In each case the step of incubation has been introduced directly after the injections: 5 min for diffusion-mediated mode and 1 min for electrophoretically-mediated mode. The protein zones contained: a-Tf (0.5 mg/mL), HSA (0.3 mg/mL), 50 mM Tris, 100 mM NaCl, 25 mM NaHCO3 , pH 7.4, while iron was injected as Fe-NTA solution in the same buffer (10 times higher molar concentration than a-Tf). Effective capillary length was 20 cm, while voltage value 15 kV. (B) – separation of the mixture of h-Tf, a-Tf and HSA (each one 0.33 mg/mL) performed by long-end (top) and short-end (bottom) injections, applying 20 cm effective length – 30 kV and 10 cm effective length – 15 kV voltage, respectively. On the electropherograms, the schematic illustration of these modes is depicted.

short-end injection strategy [28]. Contrary to the conventional long-end injection, it uses the shorter outlet part of capillary, which normally is about 10 cm long. The sample vial is placed at the outlet, and then negative pressure is applied for injection and the reverse polarity for separation. Therefore, this is one of the simplest way for reduction of separation time, unless the efficient separation in 10 cm section of capillary cannot be obtained. In our case, we have compared short-end injection performed using 10 cm capillary effective length and applying 15 kV voltage with longend injection using 20 cm capillary length and 30 kV (the same capillary). Theoretically, both methods should give similar times of migration. Indeed, the results shown in Fig. 4(B) confirm our assumption. However, in the case of long-end injection the peak standing for h-Tf was decreased, while one of the monoferric Tf and a-Tf peaks increased. After performing additional experiments we have concluded, that the high electric field (1000 V/cm) generated in the relatively short capillary could be a direct or indirect reason for partial iron release (see Supplementary Material to see more results). Interaction with SDS micelles depends on proteins structure, charge and hydrophobic amino acids residues exposure [29]. Release of iron from h-Tf is accompanied by conformational changes resulting in an “open” structure formation, characterized by an increased exposure of aromatic residues [30,31]. Due to that fact, a-Tf can be more amenable to interactions with amphiphilic SDS micelles. Two monoferric forms in turn, might be able to exert an intermediate behavior, but slightly differing between each other. Methanol was possibly an additional enhancement of this phenomenon, and caused the preservation of ferric ions bounded to the particular lobes of protein.

4. Conclusions To conclude, short end injection mode enabled the fully efficient separation of all protein forms within the time less than 4 min. 10 cm was a sufficient effective length to separate all forms of proteins with the migration times associated rather with separation of largely smaller molecules than of proteins. It is a clear-cut demonstration of the method selectivity. It is worth highlighting, that the relatively large protein molecules exerting only a subtle difference in mass caused by the binding of only one or two additional atoms could be fully separated by this method. Moreover ferric ions do not alter the charge of molecule, since they replace H+ ions. Consequently, two monoferric forms were of the same mass-tocharge ratio, and they have been also separated to the baseline. It proves that our method could be denoted as an ultraselective one. Such a good outcome can derive from the interactions between the micelles formed by SDS and protein molecules. The faster hTf spends probably most of time being dissolved in liquid phase, while slower a-Tf is localized in micellar phase characterized by a reverse vector of electrophoretic mobility due to the negatively charged SDS molecules. The results concerning gradual Tf saturation turn out to be consistent with the theoretical assumptions. On-line methodology has proven to be good alternative for conventional off-line strategy, using significantly lower amounts of reactants. The method seems to be easily transferrable to a on-chip format, where the separation time could be further shortened. Owing to the significant automation, sensitivity, and speed of analysis, we are convinced that MEKC-based method is the most effective from the all reported to date Tf iron saturation assays. Such a good outcome

78

P. Nowak et al. / J. Chromatogr. A 1341 (2014) 73–78

may encourage to the searching for the novel protein–ligand assays among CE-related techniques. Additionally it proves, that MEKC may give the great separation efficiency irrespective of difference in mass-to-charge ratio of analytes.

[8] [9] [10] [11] [12]

Acknowledgements Author Paweł Nowak has received the financial support from ´ within the Krakowskie Konsorcjum “Materia-Energia-Przyszło´sc” subsidy KNOW. The research was carried out with equipment purchased with financial support from the European Regional Development Fund within the framework of the Polish Innovation Economy Operational Program (contract no. POIG.0 2.01.00-12-0 23/08). Appendix A. Supplementary data

[13] [14] [15] [16] [17] [18] [19] [20] [21]

Supplementary material related to this article can be found, in the online version, at http://dx.doi.org/10.1016/j.chroma.2014. 03.037. References [1] [2] [3] [4] [5] [6] [7]

H. Sun, H. Li, P.J. Sadler, Chem. Rev. 99 (1999) 2817. M.E. Brandsma, A.M. Jevnikar, S. Ma, Biotechnol. Adv. 29 (2011) 230. P.T. Gomme, K.B. McCann, Drug Discov. Today 10 (2005) 267. S. Terabe, Annu. Rev. Anal. Chem. 2 (2009) 99. M. Silva, Electrophoresis 34 (2013) 141. S.E. Deeb, H.A. Dawwas, R. Gust, Electrophoresis 34 (2013) 1295. Z. El Rassi, Electrophoresis 31 (2010) 174.

[22] [23] [24] [25] [26] [27] [28] [29] [30] [31]

P.G. Righetti, G. Candiano, J. Chromatogr. A 1218 (2011) 8727. P.G. Righetti, R. Sebastiano, A. Citterio, Proteomics 13 (2013) 325. V. Kaˇsiˇcka, Electrophoresis 35 (2014) 69. Y.W. Wu, J.F. Liu, T.X. Xiao, D.Y. Han, H.L. Zhang, J.C. Pan, Electrophoresis 30 (2009) 668. J. Caslavska, J. Joneli, U. Wanzenried, J. Schiess, W. Thormann, J. Sep. Sci. 35 (2012) 3521. J. Joneli, U. Wanzenried, J. Schiess, C. Lanz, J. Caslavska, W. Thormann, Electrophoresis 34 (2013) 1563. Y. Kuroda, R. Hamaguchi, K. Moriyama, T. Tanimoto, J. Haginaka, J. Pharm. Biomed. Anal. 76 (2013) 81. F. Bortolotti, M.T. Trevisan, R. Micciolo, L. Canal, A. Vandoros, T.M. Palmbach, F. Tagliaro, Clin. Chim. Acta 416 (2013) 1. ´ ´ P. Nowak, K. Spiewak, M. Brindell, M. Wozniakiewicz, G. Stochel, P. Ko´scielniak, J. Chromatogr. A 1321 (2013) 127. J. Williams, K. Moreton, Biochem. J. 185 (1980) 483. S.L.C. Ferreira, W.N.L. Dos Santos, C.M. Quintella, B.B. Neto, J.M. Bosque-Sendra, Talanta 63 (2004) 1061. S.L. Byrne, A.B. Mason, J. Biol. Inorg. Chem. 14 (2009) 771. R.R. Hocking, Methods and Applications of Linear Models: Regression and the Analysis of Variance, Wiley, Hoboken, NJ, 2003. D. Vojnovic, M. Moneghini, F. Rubessa, A. Zanchetta, Drug Dev. Ind. Pharm. 19 (1993) 1479. ´ G. Majka, K. Spiewak, K. Kurpiewska, P. Heczko, G. Stochel, M. Strus, M. Brindell, Anal. Bioanal. Chem. 405 (2013) 5191. X. Wang, K. Li, E. Adams, A.V. Schepdael, Electrophoresis 35 (2014) 119. ´ P. Nowak, M. Michalik, L. Fiedor, M. Wozniakiewicz, P. Ko´scielniak, Electrophoresis 34 (2013) 3341. ´ P. Nowak, M. Wozniakiewicz, P. Ko´scielniak, Electrophoresis 34 (2013) 2604. J. Bao, F.E. Regnier, J. Chromatogr. 608 (1992) 217. ˇ R. Remínek, M. Zeisbergerová, M. Langmajerová, Z. Glatz, Electrophoresis 34 (2013) 2705. Z. Glatz, Electrophoresis 34 (2013) 631. M. Hadjmohammadi, M. Salary, J. Chromatogr. B 912 (2013) 50. P.G. Thakurta, D. Choudhury, R. Dasgupta, J.K. Dattagupta, Biochem. Biophys. Res. Commun. 316 (2004) 1124. E.N. Baker, H.M. Baker, R.D. Kidd, Biochem. Cell Biol. 80 (2002) 27.