Systematic investigations of peak distortions due to additives in supercritical fluid chromatography

The impact of eluent components added to improve separation performance in supercritical fluid chromatography was systematically, and fundamentally, investigated. The model system comprised basic pharmaceuticals as solutes and eluents containing an amine (i.e., triethylamine, diethylamine, or isopropylamine) as additive with MeOH as the co-solvent.

Journal of Chromatography A 1621 (2020) 461048 Contents lists available at ScienceDirect Journal of Chromatography A journal homepage: www.elsevier.com/locate/chroma Systematic investigations of peak distortions due to additives in supercritical fluid chromatography Emelie Glenne a, Jörgen Samuelsson a,∗, Hanna Leek b, Patrik Forssén a, Magnus Klarqvist c, Torgny Fornstedt a,∗ a b c Department of Engineering and Chemical Sciences, Karlstad University, SE-651 88 Karlstad, Sweden Early Chemical Development, Pharmaceutical Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden Early Product Development, Pharmaceutical Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden a r t i c l e i n f o Article history: Received 10 January 2020 Revised 12 March 2020 Accepted 14 March 2020 Available online 16 March 2020 Keywords: Supercritical fluid chromatography Peak performance Peak distortions Additives Basic components Overloaded peaks a b s t r a c t The impact of eluent components added to improve separation performance in supercritical fluid chromatography was systematically, and fundamentally, investigated The model system comprised basic pharmaceuticals as solutes and eluents containing an amine (i.e., triethylamine, diethylamine, or isopropylamine) as additive with MeOH as the co-solvent First, an analytical-scale study was performed, systematically investigating the impact of the additives/co-solvent on solute peak shapes and retentions, using a design of experiments approach; here, the total additive concentration in the eluent ranged between 0.021 and 0.105 % (v/v) and the MeOH fraction in the eluent between 16 and 26 % (v/v) The co-solvent fraction was found to be the most efficient tool for adjusting retentions, whereas the additive fraction was the prime tool for improving column efficiency and peak analytical performance Next, the impacts of the amine additives on the shapes of the so-called overloaded solute elution profiles were investigated Two principal types of preparative peak deformations appeared and were investigated in depth, analyzed using computer simulation with mechanistic modeling The first type of deformation was due to the solute eluting too close to the additive perturbation peak, resulting in severe peak deformation caused by coelution The second type of deformation was also due to additive–solute interactions, but here the amine additives acted as kosmotropic agents, promoting the multilayer adsorption to the stationary phase of solutes with bulkier aryl groups © 2020 The Authors Published by Elsevier B.V This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/) Introduction Preparative supercritical fluid chromatography (SFC) is an important technique in the pharmaceutical industry and of increasing interest in the scientific community In SFC, the mobile phase usually contains highly pressurized carbon dioxide (CO2 ) as the main weak solvent, with a polar co-solvent added to control retention Many active pharmaceutical ingredients contain amine functional groups [1]; therefore, an amine additive must be added to the eluent to obtain acceptable separation performance, reproducibility, and productivity [2–4] The addition of an amine additive will generally decrease the retention and improve the peak shapes for basic solutes in SFC [5,6], as was demonstrated in the early 1980s in reversed-phase liquid chromatography (RPLC) [7] In SFC it is ∗ Corresponding authors E-mail addresses: Jorgen.Samuelsson@kau.se Torgny.Fornstedt@kau.se (T Fornstedt) (J Samuelsson), common to use 0.1–1% amine additive in the co-solvent [8] The selection of the amine additive is based on several considerations, such as the effect on the chromatographic performance, the detector used in controlling the fractioning, post-purification removal, and reactivity In preparative industrial applications, diethylamine (DEA) is often used when the fraction collection is based on ultraviolet (UV) detection and ammonia when the fractions are analyzed by mass spectrometry detection Recently, the component ammonium hydroxide has been introduced as a water rich additive for improved chromatographic separation and purification of highly polar pharmaceuticals and peptides [9] Both modifiers and additives are used to modulate the retentions and shapes of the eluted peaks of the injected solutes, but they operate in different ways A modifier is defined as an added compound that operates by modifying the solvent strength of the mobile phase, whereas an additive is defined as a compound that operates by competing with the solutes for the limited adsorption sites on the stationary phase surface [10,11] https://doi.org/10.1016/j.chroma.2020.461048 0021-9673/© 2020 The Authors Published by Elsevier B.V This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/) E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 When modeled, the effect of a modifier on retention should therefore be described using linear solvent strength theory [12]; here the logarithm of the retention factor (k) versus fraction cosolvent in the eluent should have a linear relationship [11,12], whereas the effect of an additive is better described by a competitive adsorption isotherm [10] If we have a competitive mechanism, the logarithm of the retention factor (k) versus fraction cosolvent will not show a non-linear relationship, instead the relationship will be non-linear [,13] In a recent study we showed that, in SFC using a diol silica adsorbent, the organic co-solvent MeOH, considered a modifier, acts as an additive at low concentrations in the eluent and as a modifier at higher MeOH concentrations [13] Here, the MeOH fraction 13% (v/v) was the turning point since, since the excess isotherm reached its maximum using a diol column at a backpressure of 120 bar and a column temperature of 40 °C [13]; this value might differ for other polar phase systems in SFC since the methanol adsorption to the stationary phase depend strongly on the column chemistry as well as the temperature and pressure In later studies, we demonstrated that in the regions where MeOH is acting as an additive, the shape of the so-called overloaded elution profiles of the solutes changed in a most unusual way [14,16], depending on the level of the co-solvent fractions, such as observed before in RPLC systems having strongly adsorbed additives [15] A more complete census and understanding of all interactions between the solute and the surface of the stationary phase in analytical separation systems requires analyzing the shapes of the overloaded solute bands under so-called nonlinear conditions [10,11] The reason why nonlinear studies of solute adsorption are also important in analytical chromatography is that the different adsorption sites operate to different degrees in different concentration ranges since they have different saturation capacities, so one can distinguish the sites from one another by using a broad concentration range from low to high In addition, the shapes of the overloaded elution zone are related to the curvature of the adsorption isotherm, classified as of five main types [10] The most common Type I adsorption isotherm is the Langmuir isotherm, which is shaped convex upwards and reaches a limiting surface capacity The corresponding overloaded elution profile has a sharp front and diffuse rear—the classical overloaded Langmuirian peak shape [10] However, adsorption studies were we combine analytical-scale and overloaded, non-linear, experiments for obtaining a deeper understanding of solutes underlying retention mechanism, requires welldesigned selections of model compounds accounting for many aspects among others purity, availability and solubility, due to the large amounts used [10] Systematic investigations of the remarkable peak deformations in SFC have revealed that the overloaded solute bands have the common Langmuirian shape at low co-solvent concentrations, whereas at higher co-solvent concentrations (i.e., levels at which the co-solvent acts as an additive), the overloaded solute band profiles have unusual anti-Langmuirian shapes (i.e., with diffuse fronts and sharp rears) [14,16] Between these characteristic shapes, the bands appear generally deformed and almost round, with a hump at the front or rear These remarkable perturbation-peak–generated deformations, were explained on a firm theoretical basis by Fornstedt and Guiochon for LC [11,15] and occur in unusual LC/RPLC systems especially designed to produce the deformation effects It was recently demonstrated that perturbation-peak–generated deformations appear in most general SFC systems under certain conditions, far more seriously than in RPLC, where the deformations appear in more unusual systems and/or in systems especially designed to generate the effects by, for example, adding a more hydrophobic ion-pair reagent in RPLC than is necessary The requirements for the effects to take place in SFC due to the co-solvent are summarized in the following rule of thumb [14,16] First, the co-solvent in the SFC system should be operated using co-solvent levels at which it acts as an additive (see above) Second, the cosolvent should adsorb more strongly (or at an equal strength) to the stationary phase than does the solute in eluent without cosolvent Third, the co-solvent perturbation peak should elute earlier than the solute in the actual eluent co-solvent concentration plateau As discussed above, peak deformations can occur in most common SFC systems due to strong co-solvent adsorption to the stationary phase, i.e., under conditions in which the co-solvent acts as an additive [14,16] West et al recently showed that ammonium acetate used as an additive adsorbs even more strongly than MeOH to a hybrid silica phase (BEH; Waters Corporation, Milford, MA, USA) [8] Therefore, it can be assumed that amine additives should generate deformations similar to the way polar organic co-solvents [14] The aim of this study is to systematically investigate the effects that amine additives have on analytical and preparative retentions and band shapes, as well as to better understand the underlying mechanisms by which the additive affects the band shapes Three common basic pharmaceutical compounds were selected as model solutes, i.e., alprenolol, metoprolol, and propranolol, fulfilling the combined requirements for analytical-scale and overloaded adsorption studies As model additives three amines were selected: triethylamine (TEA), diethylamine (DEA), and isopropylamine (iPrNH2 ) To understand how the co-solvents and additives affect retention and efficiency under analytical conditions, design of experiments (DoE) modeling was used The underlying thermodynamics of the separation systems were further investigated by making overloaded injections of the solutes and then analyzing the corresponding preparative elution zones using mechanistic numerical modeling We also intended to use the deeper understanding thus attained to offer guidelines for designing these more complicated SFC systems intended for separating basic solute components, to give high performance and minimum risk of peak distortion Theory An adsorption isotherm describes the equilibrium distribution of a solute between the stationary and mobile phases at a specific and constant temperature When the mobile phase contains an additive, which competes with the solute for the available stationary phase, a multi-component adsorption isotherm is required [10,16] In this study, we will use the multi-component competitive bi-Langmuir adsorption isotherm that, for the ith component in the mixture, can be written [10]: qi (C ) = aI,iCi 1+ n j=1 bI,jC j + 1+ aII,iCi , n j=1 bII,jC j (1) where a and b are the adsorption isotherm parameters and C and q are the concentrations in the mobile and stationary phases, respectively Here, we have two different adsorption sites, denoted I and II in Eq (1) To describe the multilayer adsorption, the extended liquid–solid bi-BET adsorption isotherm can be used [17]: q(C ) = aI C (1 − bL,IC )(1 − bL,IC + bIC ) + aIIC (1 − bL,IIC )(1 − bL,IIC + bIIC ) (2) where a, b, and bL are positive adsorption isotherm parameters The parameter b is related to the adsorption to the stationary phase surface and the parameter bL to adsorption to a layer of previously adsorbed solute [10] The inverse method [18] was used to estimate the adsorption isotherm parameters by fitting calculated chromatograms using the , E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 equilibrium dispersive column model [10] to experimental chromatograms; see Section 3.4 for more details Experimental section 3.1 Chemicals and material Carbon dioxide (CO2 , 99.99%) was obtained from AGA Gas AB (Lidingö, Sweden), while HPLC-grade methanol (MeOH) (>99.9%), diethylamine (DEA) (≥99.5%), triethylamine (TEA) (99%), and isopropylamine (iPrNH2 ) (≥99.5%) were obtained from Sigma-Aldrich (St Louis, MO, USA) As solutes, alprenolol HCL (≥99%), metoprolol tartrate (≥98%), and propranolol HCL (≥99%) were used after transforming delivered form into free-base form, see Section 3.2 All solutes were obtained from Sigma-Aldrich A Kromasil Diol (150 × 4.6 mm) column with 5-μm particles and a pore size of 60 A˚ (Nouryon, Bohus, Sweden) was used Nitrous oxide (99.998%; Sigma-Aldrich) was used to determine the ˚ “dead volume” of the columns to be 1.69 mL, according to Asberg et al [19] The SFC system was an ACQUITY UPC2 (Waters Corporation, Milford, MA, USA) in its default configuration with a diode-array detector and a column oven A 10-μL loop was used for all injections The mass flows of both the co-solvent and total eluent were measured simultaneously using two mini CORI-FLOW M12 Coriolis mass flow meters (Bronkhorst High-Tech B.V., Ruurlo, Netherlands), as described by Glenne et al [16] The pressure was measured at the inlet and outlet of the column using two EJX530A absolute pressure transmitters (Yokogawa Electric Corporation, Tokyo, Japan) [16] The temperature was measured at the middle of the column using a PT-100 four-wire resistance temperature detector with an accuracy of ±0.2°C (Pentronic AB, Gunnebo, Sweden) permanently attached to the column surface using a silver-based epoxy adhesive (Arctic Silver, CA, USA) 3.2 Procedures In this study, the free-base forms of the solutes were always used The solutes in salt form were transformed into freebase form by shaking a 1% solution of the solute salt dissolved in 100 mM NaOH aqueous solution with an equal volume of dichloromethane The dichloromethane phase was then collected and evaporated In all experiments, the backpressure regulator was adjusted to give an average pressure over the column of 150 bar and the column oven was set to 40 °C The flow rate was set to mL min–1 in all experiments The solutes were detected at 220 nm in the analytical analysis; in the overloaded study, the solute zones were recorded at 248 nm for alprenolol and metoprolol, at 325 nm for propranolol, and at 205 nm for the additive zones All samples were filtered through a 0.45-μm PTFE filter prior to injection The retention volumes were calculated from the peak apex of the elution profiles and the column efficiencies were based on measuring the peak width at half height In the procedure described in Section 4.1, “Analytical-scale analysis”, 1-μL samples of solute were injected; the concentrations were 0.77 mM for propranolol, 0.75 mM for metoprolol, and 0.80 mM for alprenolol, all dissolved in neat MeOH The additive sample contained 0.1 M additive in MeOH, and 2-μL samples of DEA or TEA or 5-μL samples of iPrNH2 were injected The retention times of the additives were estimated using the first statistical moment due to the severe degree of peak tailing For more information about the DoE study, see Section 3.3 In Section 4.2, “Overloaded analysis using DEA additive”, 2–8μL volumes of 99.90 mM alprenolol or metoprolol or of 99.95 mM propranolol were injected using MeOH with mM DEA as the sample diluent The MeOH content was calculated as the actual MeOH fraction of 16, 21, or 26% (v/v), according to Section 3.4 DEA was added to the co-solvent to achieve the overall (total) eluent concentrations of 0.021, 0.063, and 0.105% (v/v) corresponding to total eluent concentrations of 2.1, 6.3, and 10.2 mM, respectively The experiments described in Section 4.3, “Overloaded analysis using different amine additives,” also including the additives TEA and iPrNH2 , were conducted in a similar fashion as for DEA However, in these experiments, the MeOH content in the mobile phase was not varied, but was instead fixed at 16% (v/v), and MeOH was used as the diluent with the different additive amines: either mM TEA or mM iPrNH2 , depending on the additive used For comparison purposes, the concentrations of iPrNH2 in the eluent were 2.5, 7.5, and 12.4 mM and of TEA were 1.5, 4.6, and 7.6 mM, corresponding to 0.021, 0.063, and 0.105% (v/v) additive in the eluent, respectively The additive concentration is often expressed as the additive fraction added to the co-solvent This practice, however, results in a varied amount of additive in the total eluent depending on the co-solvent fraction To avoid this variation, we instead express the additive concentration as the fraction of the total eluent The additive concentrations used in this study correspond to additive fractions of 0.1, 0.3, and 0.5% (v/v) added to a MeOH fraction of 21% (v/v) in the eluent In this context, it should be noted that when we selected the ranges of co-solvent (MeOH) and additive concentrations, we had in mind that, as well as covering interesting and practically useful experimental concentration ranges, the analytes selected should have sufficient solubility 3.3 Design of experiments A three-level, two-factor, full-factorial design with three center points was used to study the variation in the retention factor and efficiency (number of theoretical plates, N) with the two operational parameters: MeOH fraction and DEA concentration The actual MeOH fractions were 16, 21, and 26% (v/v) and the DEA concentrations in the eluent were 0.021, 0.063, and 0.105% (v/v) Models were fitted using multiple linear regression implemented in MODDE 11 (Umetrics AB, Sweden) Statistically insignificant (95% confidence level) parameters were removed to refine the models All regression models had excellent R2 and Q2 values (see Table S.1 in the Supplementary Material) In this study a full factorial three level design with two factors (CMeOH and CDEA ) was used For the regression model, constructed after removing insignificant coefficients at a 95% confidence level, give: 2 k = c0 + cCMeOH C MeOH + cCDEA C DEA + cC2 MeOH CMeOH + cC2 DEA CDEA , (3) where ci are regression coefficients The column efficiency was estimated using the half-height method and the retention factor was calculated from the retention volume and column void volume; the latter was determined using N2 O as non-retained marker [19] 3.4 Calculations The measured mass flows (total and co-solvent), pressures, and temperatures were used to estimate the actual average volumetric flow rates and the actual average volumetric co-solvent fractions, as described by Glenne et al [13] The partial molar volume was calculated according to Kato et al [20] and the density was estimated using the Kunz and Wagner [21] equation of state implemented in REFPROP v 9.1 from the National Institute of Standards and Technologies (NIST) [22] The inverse method [18] was used to estimate the adsorption isotherm parameters in this study (see Section 2) For the alprenolol using DEA as the additive (Fig 4) and metoprolol with E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 iPrNH2 as the additive (Fig 9), simulated elution profiles were calculated using orthogonal collocation on finite elements [23], and the bi-Langmuir model was used as the adsorption model, see Eq (1) The column efficiency was set to 80 0 and the total porosity to 0.678 (calculated from the column dimensions and the ‘dead time’ t0 measured using an unretained substance); the flow rate was 1.08 mL min–1 when using DEA and 1.09 mL min–1 when using iPrNH2 as the additive The isotherm parameters for Fig were estimated by fitting simultaneously to the three different additive (iPrNH2 ) concentration levels in the eluent which makes the fitting task somewhat more challenging than that in Fig 4, comprising only one level of the additive (DEA) For more experimental information regarding the inverse calculations, see Figs and 9; for the isotherm parameters, see Table S.3 in the Supplementary Material For the propranolol case shown in Fig 5, the simulated elution profiles were calculated using a finite volume algorithm [24] and the bi-BET adsorption model, see Eq (2), where the total porosity was 0.69 (calculated from the column dimensions and the ‘dead time’ t0 measured using an unretained substance) and the flow rate 1.094 mL min–1 For the experimental data used for the inverse calculations, see Fig In addition to estimating the adsorption isotherms in Eq (2), the inverse method was also used to estimate the number of theoretical plates, N, and time adjustment factors, t, for the elution profiles, i.e., the elution time is adjusted by adding the factor t (here we set t = for the largest injection volume) This was done to account for possible delays between injections, flow rate variations inside the column, etc For more experimental information regarding the inverse calculations, see Fig 5; for the isotherm parameters, see Table S.4 in the Supplementary Material Results and discussion Previously, we have shown how different co-solvents can generate peak deformations when they adsorb more strongly to the stationary phase than does the solute [14,16] Here, we will systematically investigate the impact on retention and peak/band shapes of both co-solvent and amine additives in the more complicated separation systems required for analysis of basic components In a previous study, we learned that with large co-solvent (i.e., MeOH) fractions using diol silica as adsorbent, beyond the maximum point of co-solvent excess, MeOH acts more as a modifier than as an additive (see Fig in that study, where that point is 13% (v/v) [13]) To distinguish the effects of the amine as an additive from the co-solvent additive effects, in this study we selected co-solvent levels of MeOH in the eluent at which the MeOH level is at least 16% (v/v), well inside the region where the co-solvent function of MeOH dominates [13,16] In Section 4.1, “Analytical-scale analysis,” we investigate how the retentions and peak shapes (efficiency) of analytical-size injections of the three model solutes depend on varying both the co-solvent fraction and additive composition of the eluent additive diethylamine (DEA), respectively In Section 4.2, we investigate how the shapes of the overloaded eluted solute bands depend on varying the amount of the DEA additive in the eluent as well as varying the co-solvent (i.e., MeOH) fraction Adsorption studies over a broad range of solute concentrations give a complete census of all interactions between the solute and stationary phases and are required for a deeper understanding of the underlying retention mechanisms of separation systems [10] The results were interesting and unexpected, and it was relevant to expand the study with additional overloaded experiments confirming the generality of the results Thus, in Section 4.3, we investigate overloaded elution profiles also using the additives triethylamine (TEA) and isopropylamine (iPrNH2 ), keeping the co- Fig Retention factors of the three basic model solutes used in the study, i.e., propranolol, metoprolol, and alprenolol, and the three additives used, i.e., diethylamine (DEA), triethylamine (TEA), and isopropylamine (iPrNH2 ), versus the MeOH fraction (v/v) in eluent lacking an amine additive solvent MeOH fraction in the eluent constant at 16% (v/v) We used numerical tools for adsorption isotherm determinations and mechanistic modeling to analyze the overloaded data generated in Sections 4.2 and 4.3 4.1 Analytical-scale analysis First, we investigated the relative retention of the three basic β receptor antagonist model solutes and the DEA additive with different fractions of MeOH in the eluent without having any DEA additive in the eluent The model solutes are β -receptor antagonists with similar structures but different hydrophilicities The experimental log KD values for metoprolol, alprenolol, and propranolol are 1.88, 3.10, and 3.56, respectively [25] Thus, the bulky propranolol containing a naphthyl group, instead of a benzyl group, is the most hydrophobic of the model solutes (see solute structures in Fig S.1 in the Supplementary Material) Fig shows the resulting retention factors of all solutes and amine additives, injected using eluent lacking additive, versus the MeOH fraction in the eluent The retention factors of the β receptor antagonists decrease with increasing MeOH content (cf Fig 1); see Table S.2 in the Supplementary Material for the corresponding numerical values The most hydrophobic solute, propranolol, has much higher retention factors than the less hydrophobic metoprolol and alprenolol We can also see that all the solutes are more retained than is DEA and that the retentions of the solutes relative to the additive decrease with increasing MeOH fraction; this observation regarding the relative retentions is in line with West et al [8] With the high MeOH content of 26% (v/v), the two less retained solutes alprenolol and metoprolol have almost a combined elution with DEA To better investigate the impacts of the co-solvent and additive on retention and efficiency, we employed a DoE approach (see Section 3.3 for details) In the DoE, the variation of all factors are simultaneously evaluated in a minimal number of experiments [26] In this study a full factorial three-level design with two factors (CMeOH and CDEA ) was used, see Section 3.3 for details Fig 2a presents the centered and normalized coefficients of the model fit for the retention factors of the solutes, showing that the retentions decrease with both increasing MeOH fraction and increasing DEA concentration in the eluent, and that the MeOH fraction has a fivetimes-larger impact on the retention than does the DEA fraction E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig Centered and normalized coefficients from the DoE model fit for (a) the retention factor and (b) the column efficiency of the analytical-size peaks resulting from injections of the three β -receptor antagonists using eluents with varying MeOH fractions and DEA concentrations The error bars represent the 95% confidence intervals of the coefficients CMeOH and CDEA are the eluent concentrations of MeOH and DEA, respectively, C2 MeOH and C2 DEA are the corresponding quadratic terms Fig S.2 and Table S.1 in the Supplementary Material contain the corresponding response surfaces and regression coefficients, respectively Therefore, it is much more effective to use the co-solvent to adjust the solute retention than it is to modulate the additive fraction level The effect is largest for propranolol, followed by metoprolol and alprenolol; note that this is the same order as the relative retention order shown in Fig The model contains significant quadratic MeOH and DEA terms visualized as curves in the response surfaces (see Fig S.2 in the Supplementary Material) Starting from zero additive concentration in the eluent and going to the lowest additive level, the solute retention factors decrease strongly with the added additive (see Table S.2 in the Supplementary Material) As soon as normal operative DEA levels have been reached in the eluent contents, the solute retention factors decrease only slightly with further increased DEA levels (cf Table S.2), in line with SFC studies using ammonium acetate as an additive [4,8,27] Fig 2b shows that the impacts of the additive and co-solvent on column efficiency are comparable in size and that eluents having low MeOH and high DEA contents produce the highest possible column efficiencies This model contains significant quadratic DEA factors, visualized as some skewness in the response surfaces in Fig S.2 (see also Table S.1 in the Supplementary Material, containing the regression coefficients for the data used in Fig 2a and b) To summarize, the co-solvent fraction is the most important factor controlling the retention of the basic solutes, whereas both the co-solvent and additive have an impact on the solutes’ column efficiencies For the highest possible column efficiencies, separations using eluents with a combined small MeOH fraction and high DEA content are recommended Here, one also need to consider the solutes solubility in the actual eluent composition More particularly for this system, the lowest allowed MeOH fraction in the eluent should be selected so that the solutes have good enough solubility to avoid precipitations in te separation system 4.2 Overloaded analysis using DEA additive The analytical investigation showed that adding the co-solvent MeOH and the additive DEA to the eluent clearly affects both retention and efficiency For a deeper investigation aiming at understanding the physico-chemical mechanisms, we have to make a complete census of all possible sources of interactions between the solutes and the stationary phase, and thus study the interactions over broad concentration ranges of the solutes—the nonlinear operational conditions described by Guiochon et al [10] We complemented our analytical results with studies of overloaded elution profiles resulting from the injections of high-concentration samples of the model solutes In the analytical section, we can see that alprenolol has the lowest retention factors of the three model solutes (cf Fig 1) Fig shows the overloaded elution profiles resulting from injections of four different injection volumes (2–8 μL) of 100 mM alprenolol In Fig we see, in agreement with the experimental design described in Section 4.1, shorter retentions of alprenolol with increasing MeOH fractions, going from top to bottom, and a sharpened overloaded elution profile with increasing DEA concentrations, going from left to right In Fig 3a we see some sign of peak deformation at low DEA concentrations, becoming more severe with increasing MeOH fractions in the eluent, going first to 21% (Fig 3d) and then to 26% (Fig 3g) This trend to more severe peak deformation is because, as mentioned in the introduction (see Section 1), the second of three requirements for perturbationpeak–generated deformation to appear [14,16] is better approached with a higher MeOH content in the eluent The first requirement for perturbation-peak–generated deformation is fulfilled in all cases in this study, since we use amine additives, not a co-solvent that can have a dual function, as in the earlier SFC studies of this deformation in which we defined the requirements [14,16] The second requirement was that the additive should adsorb more strongly (or E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig Overloaded alprenolol profiles resulting after 2-, 4-, 6-, and 8-μL injections of 100 mM alprenolol eluted with a mobile phase containing increasing DEA contents (left to right, 0.021, 0.063, and 0.105%) and increasing MeOH fractions (top to bottom, 16, 21, and 26%) The arrows mark the positions of the DEA perturbation peaks at an equal strength) to the stationary phase than does the solute in eluent without co-solvent From Fig we can see that it is only in Fig 3d and g that the second requirement is approached, using combined low DEA/high MeOH eluent contents, conditions in which the peak alprenolol and DEA retentions approach, in eluent lacking DEA (cf Fig 1) The third requirement was that the additive perturbation peak should elute earlier than the solute in the actual eluent co-solvent concentration plateau, which is fulfilled for all chromatographic conditions in Fig 3, as indicated by the arrows that mark the perturbation peak positions in the actual chromatographic runs Thus, and in line with the rule of thumb for co-solvent acting as the additive in SFC [14,16], the deformations become more pronounced as the MeOH content increases, while keeping the DEA content low (cf Fig 3d and g) However, some degree of the perturbation-peak–generated type of deformation should also be expected for metoprolol at this MeOH level, which can be seen in the overloaded metoprolol experiments with the highest eluent MeOH 26% (v/v) content (see Fig S.3g, in the Supplementary Material) In the analytical section, we could see that the metoprolol has the next shortest retention of the three model solutes and also elutes close to DEA at the highest MeOH content, i.e., 26% (v/v), in eluents lacking DEA (cf Fig 1) However, the deformations are less pronounced than those for alprenolol, appearing for metoprolol only at the highest MeOH/lowest DEA contents (cf Fig 3g), because the second requirement is not fulfilled at MeOH contents below 26%, unlike for alprenolol and DEA (cf Fig 1) The perturbation-peak–generated deformations are due to complex competitive interactions between the additive perturbation zone and the solute zones as they resolve while traveling along the column, forming internal gradients for each other, as previously observed for 1-phenyl-1-propanol when the co-solvent MeOH was acting as the additive [16] If we can simulate the characteristic deformations using an established column model, such as the equilibrium dispersive column model [10], combined with a mechanistic competitive adsorption model, such as Eq (1) in Section 2, this will provide valid confirmation that a competitive mechanism is the underlying reason for the observed deformations described above The fitting of valid mechanistic models to experimental data is important, and one always starts with the simplest model, thereafter, if it is needed, select a more complex model Therefore, the simple one-site Langmuir competitive model was first tested for the adsorption of alprenolol and DEA, however, this model failed to fit the data well Next, the bi-Langmuir competitive adsorption isotherm equation, Eq (1), was used, this two-site model fitted the data very well (for model coefficients, see Table S.3 in the Supplementary Material) Fig 4a shows the agreement between experimental (solid lines) and simulated (dotted lines) overloaded alprenolol elution profiles using the competitive bi-Langmuir model, and Fig 4b shows the corresponding simulated perturbation signal of the DEA concentration plateau We can see that the fits are very good, especially bearing in mind that these deformations have very unusual and characteristic shapes Interestingly, the sharp rear of the negative perturbation zone of DEA has a combined elution with the rear of the alprenolol zone Since the mechanistic simulations fit the experimental profiles so well, we can conclude that the sharp rear part of the eluted alprenolol band is the result of a complex competitive interplay with the rear of the DEA zone as the zones travel along the column (cf Fig 4a and b) Propranolol is the most hydrophobic and bulky of the three model solutes (cf Fig S.1); it has much higher analytical retention factors that are always well resolved from the DEA peak, as seen in the analytical section (cf Fig and Table S.2) The important second requirement for perturbation-peak–generated deformations is far from being fulfilled Therefore, when overloaded injections of propranolol are made, we cannot expect a similar tendency for deformations of a competitive nature as discussed above for alprenolol and metoprolol The overloaded experimental propranolol profiles can be seen as the solid lines in the subplots in Fig with varying MeOH and DEA contents in the eluent Surprisingly, there are also deformations for propranolol, but these deformations in- E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig (a) Comparison between simulated (dotted lines) and experimental (solid lines) overloaded solute elution profiles resulting after 2-, 4-, 6-, and 8-μL injections of 100 mM alprenolol (b) The corresponding simulated DEA concentration plateau perturbations (dashed lines) The mobile phase contained 21% (v/v) MeOH and 0.021% DEA The simulations are based on the binary competitive bi-Langmuir adsorption isotherm model, i.e., Eq (1) in Section 2; for the best bi-Langmuir model coefficients, see Table S.3 in the Supplementary Material stead become more pronounced with decreasing MeOH fractions in the eluent (bottom to top in Fig 5) and with increasing amine additive contents in the eluent (left to right in Fig 5), finally resulting in clearly overloaded anti-Langmuirian profiles (cf Fig 5c) Thus, the deformations of the overloaded propranolol bands display a completely opposite pattern to those of the earliereluting model solutes alprenolol and metoprolol For these two latter solutes, which fulfill the requirements for perturbation-peak– generated deformations, the overloaded deformations are most pronounced at high MeOH/low DEA contents in the eluent (for alprenolol see Fig 3g and for metoprolol see Fig S.3g) For propranolol, however, under these conditions the deformations not appear at all; in contrast, at the highest MeOH/lowest DEA contents, the overloaded propranolol displays a “normal” overloaded Langmuirian band shape (see Fig 5g) If we now, starting from this position, decrease the MeOH content in the eluent while keeping the DEA content constant, i.e., moving upwards in the left column from Fig 5g to 5a using 26% MeOH, we can see how the Langmuir profiles become successively deformed, developing a hump at the front at the lowest MeOH content (16%), as shown in Fig 5a Now, moving from this position and increasing the DEA content while keeping the lowest MeOH content of 16%, i.e., moving from Fig 5a to c, we see how the overloaded propranolol is transformed, in a fascinating way, from being a deformed band with a hump in front (Fig 5a) to a nice classical anti-Langmuirian profile (Fig 5c) This pattern is completely opposite to that displayed by the more hydrophilic solutes alprenolol (cf Fig 3) and metoprolol (cf Fig S.3), strongly indicating a completely different underlying mechanism for the deformation of the bulkier, hydrophobic model solute, propranolol We more closely inspected the overloaded propranolol elution profiles for the largest load at 16% (v/v) MeOH and the lowest DEA concentration in the eluent in Fig The propranolol elution profile initially has a relatively sharp front at the lower concentra- Fig Overloaded experimental (solid lines) and simulated (dotted lines) elution profiles for 2-, 4-, 6-, and 8-μL injections of 100 mM propranolol eluted with increasing DEA contents in the eluent (left to right, 0.021, 0.063, and 0.105%) and with increasing MeOH contents (top to bottom, 16, 21, and 26%) The arrows mark the position of the DEA perturbation peaks The simulations (dotted lines) are based on the single-component bi-BET adsorption isotherm model, i.e., Eq (2) in Section 2; for model coefficients and model agreement, see Table S.4 in the Supplementary Material 8 E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig Adsorption isotherms of propranolol at different eluent compositions (a) Constant MeoH fraction of 16% (v/v) and varying DEA contents of 0.021, 0.063, and 0.105% (v/v) (b) Constant DEA content of 0.063% (v/v) and varying MeOH fractions of 16, 21, and 26% (v/v) The adsorption isotherm of propranolol eluted with 16% (v/v) MeOH and 0.021% (v/v) DEA has an inflection point, indicated by the green circle in (a) tions, which becomes dispersed at higher concentrations, whereas the tail is sharp at higher concentrations and becomes dispersed at lower concentrations (cf Fig 5a) Thus, the overloaded band shape cannot be described by a Type I isotherm (Langmuirian), which has a convex upwards curvature that flattens out The profiles in Fig 5a instead indicate that the adsorption isotherm is of Type II [28,29], with an initial convex upwards curvature (as in Type I) that, instead of flattening out, becomes a concave (upwards) curvature With higher additive concentrations (Fig 5b to c), the overloaded solute profile instead becomes anti-Langmuirian, indicating that the adsorption isotherm is of Type III [28,29], with only a concave (upwards) curvature, i.e., the isotherm does not flatten out Both Type II and III adsorption models are multilayer models [10] and could be described using the BET adsorption model The adsorption isotherm parameters for the elution profiles for propranolol were estimated using the inverse method (see Section 3.4) Several adsorption isotherm models that could describe multilayer or solute–solute interactions (i.e., the Moreau, bi-Moreau, and BET models) were evaluated, but only the bi-BET model agreed well with all the experimental data In Fig 5, the dotted lines show the simulated overloaded elution profiles of propranolol using the estimated bi-BET adsorption isotherm We can see how perfectly the bi-BET model (dotted lines) bands agree with the corresponding experimental (solid lines) bands For model coefficients and model agreement terms, see Table S.4 in the Supplementary Material Fig shows the propranolol adsorption isotherms for (a) varying DEA contents but a constant MeOH fraction of 16% (v/v) and for (b) varying MeOH fractions but a constant DEA content of 0.063% The adsorption isotherm for propranolol eluted with 16% (v/v) MeOH and 0.021% DEA is of Type II, with an inflection point marked by the circle in Fig 6a (green line) With increasing DEA concentrations but a constant MeOH fraction, the adsorption isotherm becomes concave (upwards), indicating transformation from a Type II to Type III isotherm (red line in Fig 6a) in which adsorbed solute–solute interactions become dominant With a constant DEA concentration but increasing MeOH fractions in the eluent, the isotherm is transformed from Type III to Type I (Fig 6b) There is an enhanced tendency for multilayer formation with increasing DEA concentrations and decreasing MeOH fractions with propranolol as the solute, as indicated by the overloaded elution shapes in Fig 5b and c A possible underlying explanation for the layer formation is the kosmotrope effect, in which hydrophobic interactions are favored Kosmotropic ions usually have a large charge density and Fig Retention factors of the β -receptor antagonists and the additive perturbation peaks investigated in this study with 16% (v/v) MeOH in the eluent with different amine additives: (a) iPrNH2 , (b) DEA, and (c) TEA interact strongly with water, tending to increase the order of the water structure [30] These so-called structure makers are commonly used for “salting-out” proteins by aggregating them The ions compete for the water molecules originally associated with the protein surface, and in that way promote hydrophobic interactions between proteins that result in protein precipitation The order of the kosmotropic effect can be related to the Hofmeister series, in which ions are arranged according to their ability to precipitate proteins [31] In the Hofmeister series, the ammonium salts are classified as kosmotropic agents [32] Consequently, amine additives may promote hydrophobic interactions that enhance the solute–solute interactions, especially in the hydrophobic parts of a molecule, for example, the naphthyl group in propranolol E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig Overloaded metoprolol elution profiles for 2-, 4-, 6-, and 8-μL injections of 100 mM metoprolol eluted with 16% (v/v) MeOH and increasing additive contents (left to right, 0.021, 0.063, and 0.105%) in the eluent, with different additives in the eluent: (a–c) iPrNH2 , (d–f) DEA, and (g–i) TEA The arrows mark the position of the additive perturbation peak 4.3 Overloaded analysis using different amine additives To confirm the generality of our conclusions, we expanded the model separation system by using two other common SFC amine additives: isopropylamine (iPrNH2 ) and triethylamine (TEA) These additives were selected to cover the difference between primary, secondary, and tertiary amines as well as because they are commonly used for separating amines in SFC [3,33,34] First, we investigated the relative retentions of the solutes and these additives with different fractions of MeOH in the eluent without having the additive in the eluent In Fig we can see the retention factors of these additives together with the analytical size retention factors of the three β -receptor antagonists and of DEA, versus the MeOH fraction in eluent lacking additive As a reminder, in this plot the amine additives are also injected as if they were solutes Fig shows that all model solutes are more retained than are the amine additives used in this study and that iPrNH2 adsorbs somewhat more strongly to the stationary phase than does DEA (already studied above), whereas TEA adsorbs somewhat more weakly than does DEA (cf Fig and Table S.2) The adsorption strength of the amines to the stationary phase is in line with their order concerning increasing capability for forming hydrogen bonding: primary amine > secondary amine > tertiary amine In Fig 7, the retention factors of all solutes and the perturbation peaks of all amine additives are plotted versus different amounts of additive in the eluent using 16% (v/v) MeOH in the eluent As in the case with DEA as the additive, adding even small amounts of additive to the eluent, compared with no additive, drastically affects the retention (cf Table S.2 in the Supplementary Material) The amine additive iPrNH2 is more potent at reducing the retention factors of the solutes, followed by DEA and finally by TEA (cf Fig 7); this is the same order as the additives’ relative adsorption to the stationary phase when injected into eluent lacking additive (cf Fig 1) Inspecting the retentions of the perturbation peaks using 16% (v/v) MeOH in the eluent, one can observe that the iPrNH2 perturbation peak has higher retention factors than does alprenolol but co-elutes more or less with metoprolol (cf Fig 7a and Table S.2) The DEA perturbation peak (Fig 7b) always elutes earlier than the solutes, but metoprolol and especially alprenolol elute very close to the perturbation peak, especially with high MeOH contents, which is why it displays the perturbation-peak–generated competitive type of deformation (cf Fig 3d and g; and Fig S.3g) The TEA perturbation peak always elutes earlier than the solutes even with high MeOH contents (Fig 7c), so we should not expect any competitive deformations of any of the three model solutes when using TEA as the additive in this separation system Now let us investigate what happens when we perform overloaded studies using the amine additives iPrNH2 and TEA, respectively, and compare this with what we have already seen regarding how DEA affected the solute peak shapes Fig shows the overloaded solute elution profiles resulting from to 8-μL injections of 100 mM metoprolol with a constant 16% (v/v) MeOH content in the eluent and increasing concentrations of different additives Inspecting the elution profiles in the top row using iPrNH2 (Fig 8a–c), severely deformed elution profiles can be observed, and the metoprolol band deformation changes strongly with the increasing amount of iPrNH2 in the eluent, from left (Fig 8a) to right (Fig 8c) On the other hand, no peak deformations appear under the same operational conditions in the middle row using the DEA additive (see Fig 8d–f) or in the bottom row using the TEA additive (see Fig 8g–i) This observation is in line with Fig 7, that one could expect solute peak deformation due to co-elution or elution close to the perturbation peak to be more severe for metoprolol using iPrNH2 than using DEA or TEA By visually comparing the overloaded metoprolol elution profiles formed when using the less retained TEA in the eluent (Fig 8g–i), it is evident that TEA as the additive does not sharpen and concentrate the preparative peaks as efficiently as does using DEA as the additive (Fig 8d–f) This was confirmed using alprenolol as the solute (see Fig S.4 in the Supplementary Material) The last obser- 10 E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 Fig (a–c) Overloaded experimental (solid lines) and simulated (dotted lines) elution profiles for 8-μL injections of 100 mM metoprolol with different amounts of iPrNH2 in the eluent (d–e) Corresponding simulated concentration plateau perturbations of iPrNH2 (dashed lines); here d corresponds to a, e to b, and f to c The eluent contained 16% (v/v) MeOH with 0.021% (a, d), 0.063% (b, e), and 0.105% (c, f) iPrNH2 The simulations are based on the binary competitive bi-Langmuir adsorption isotherm model, i.e., Eq (1) in Section 2; for the model coefficients, see Table S.3 in the Supplementary Material Fig 10 Overloaded elution profiles for 2-, 4-, 6-, and 8-μL injections of 100 mM propranolol eluted with 16% (v/v) MeOH and increasing additive contents (left to right, 0.021, 0.063, and 0.105%) in the eluent, and different additives in the eluent: (a–c) isopropylamine (iPrNH2 ), (d–f) diethylamine (DEA), and (g–i) trietylamine (TEA) The arrows mark the amine additive perturbation peak position vation is very interesting because the purpose of using an amine additive is to improve the separation performance The transformation of the deformed elution profiles when overloaded metoprolol is eluted with increasing iPrNH2 levels in the eluent, as described above (cf Fig 8a–c), should be most interesting to understand mechanistically Therefore, adsorption parameters assuming a competitive bi-Langmuir model describing the ad- sorption competition between metoprolol and iPrNH2 were estimated using the inverse method, and the parameters were used to simulate the elution profiles In Fig 9a–c, the most overloaded experimental elution profiles (following 8-μL injections) are compared with the corresponding simulated profiles using different amounts of iPrNH2 in the eluent The model fits very nicely to the characteristic and unusual solute band shapes at medium and high E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 additive levels (cf Fig 9b and c) but less accurately to the more withdrawn solute band shape at the low additive case (cf Fig 9a); the reason for the latter can be that the isotherm parameter estimation for the data in the figure were estimated by fitting simultaneously to the three different additive (iPrNH2 ) concentration levels resulting in lower relative weight on the low additive concentration data of Fig 9a In the bottom row of the Figure (Fig 9d–f), the corresponding simulated concentration plateaus of the additive are shown; here it is remarkable how well these simulated additive perturbations reflects the unusual experimental top row solute shapes, i.e the experimental and simulated profiles in Fig 9a–c, in this case even the weird profile region of Fig 9a was well captured Thus, acceptable model agreement was found using the same adsorption parameters for all iPrNH2 concentrations This confirms that the cause of these deformations is competition between the additive and the solute, as also observed for alprenolol using DEA as the additive (cf Fig 4) We can also confirm the other type of deformation mechanism observed for propranolol when the elution profile moves from a Langmuirian towards an anti-Langmuirian shape when using DEA as the additive (cf Fig 5) Fig 10 presents the overloaded shapes of propranolol with use of the three different additives In line with our hypothesis, the effect is clearly more pronounced with iPrNH2 (Fig 10a–c) in the eluent, and the profile turns towards an antiLangmuirian shape with a lower additive content than when using DEA as the additive (Fig 10d–f) This suggests that the multilayer tendency is stronger for propranolol separated using iPrNH2 rather than DEA in the eluent TEA seems to be the weakest additive, and the elution profile for propranolol eluted with a small fraction of TEA is Langmuirian, and multilayer characteristics are first observed with higher amounts of additive in the eluent Many primary, secondary, tertiary, and quaternary amines have a positive Jones−Dole viscosity B coefficient [35] and therefore are kosmotropic agents [36] This result indicates that if a certain volume fraction of amine additive is added to the co-solvent, iPrNH2 will promote multilayer formation the strongest, followed by DEA and then TEA Conclusions and practical implications For compounds containing amine functional groups there is often a need, except having a co-solvent in the eluent, to also add an additive to the eluent to achieve acceptable peak shapes for both analytical and preparative purposes To understand the impact that different compounds in the mobile phase have on retention and peak shape, we undertook an analytical and overloaded investigation combined with design of experiments modeling and fundamental numerical modeling, respectively The model system comprised three β -receptor antagonists as model solutes, a Diol column as the stationary phase, and CO2 with varied MeOH fractions and additive concentrations in the eluent Using design of experiments, the co-solvent fraction and the additive concentration were investigated It was shown that (i) the co-solvent fraction is the most important factor controlling the retention and that (ii) the highest analytical peak efficiency is obtained by using eluents containing small fractions of MeOH and high amine additive contents In the model system used in this study the solubility of the model solutes was good enough, but if that should not be the case, it is not recommended using such small MeOH fractions In the overloaded analysis, it could be concluded that more compact elution zones are generally observed when using a larger fraction of amine additives in the eluent Also, two different types of peak deformations were observed The first type of deformation probably results from the tag-along effect as the solute elutes close to the additive perturbation peak For the second type of deformation concerning a later-eluting solute, the adsorption study re- 11 vealed that the deformation was caused by multilayer adsorption In this case, the amine additive probably acts as a kosmotropic agent promoting the multilayer adsorption of solutes to the stationary phase Multilayer formation is strongly dependent on the type of amine additive used, with the additive iPrNH2 promoting multilayer formation the strongest, followed by the additives DEA and then TEA A related topic is “stacked injections” [37], an approach often used to increase the productivity of preparative separation In this approach, the next injection is made before the complete elution of the current cycle The cycle time of an injection is defined as the difference between the time when the first-eluting component exceeds a certain threshold concentration and when the elution profile of the last-eluting component drops below this concentration The goal when conducting stacked injections is to have these cycles as close together as possible: if they are too far apart productivity will decrease, and if they are too close together, the new cycle might be affected by the previous injection, possibly deforming the elution profiles It should be noted that here we have an “invisible” additive, and it might be necessary to take this into account when setting the cycle time In previous LC studies we have shown that in certain cases the additive must be taken into account [38], while in others it can be ignored [39] The conditions when the additive must be taken into account, and when it can safely be ignored, remain to be systematically investigated In the meantime, we recommend studying the peak shapes when using the stacked injection mode; if deformations occur that degrade separation performance, the most practical action is to increase the cycle time Declaration of Competing Interest The authors declare that they have no conflict of interest CRediT authorship contribution statement Emelie Glenne: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Visualization Jörgen Samuelsson: Conceptualization, Methodology, Software, Validation, Formal analysis, Data curation, Writing - original draft, Writing - review & editing, Supervision Hanna Leek: Conceptualization, Resources, Writing - original draft, Writing - review & editing Patrik Forssén: Software, Formal analysis, Writing - original draft, Visualization Magnus Klarqvist: Conceptualization, Resources, Writing - original draft, Writing - review & editing Torgny Fornstedt: Conceptualization, Methodology, Data curation, Writing - original draft, Writing - review & editing, Supervision, Project administration, Funding acquisition Acknowledgements This work was supported by the Swedish Knowledge Foundation via the KKS SYNERGY project “BIO-QC: Quality Control and Purification for New Biological Drugs” (grant number 20170059) and by the Swedish Research Council (VR) via the project “Fundamental Studies on Molecular Interactions aimed at Preparative Separations and Biospecific Measurements” (grant number 2015–04627) Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.chroma.2020.461048 References [1] P.S Charifson, W.P Walters, Acidic and basic drugs in medicinal chemistry: a perspective, J Med Chem 57 (2014) 9701–9717, doi:10.1021/jm5010 0a 12 E Glenne, J Samuelsson and H Leek et al / Journal of Chromatography A 1621 (2020) 461048 [2] L Miller, Preparative enantioseparations using supercritical fluid chromatography, J Chromatogr A 1250 (2012) 250–255, doi:10.1016/j.chroma.2012.05.025 [3] K.W Phinney, L.C Sander, Additive concentration effects on enantioselective separations in supercritical fluid chromatography, Chirality 15 (2003) 287–294, doi:10.1002/chir.10196 [4] E Lemasson, S Bertin, P Hennig, H Boiteux, E Lesellier, C West, Development of an achiral supercritical fluid chromatography method with ultraviolet absorbance and mass spectrometric detection for impurity profiling of drug candidates Part I: Optimization of mobile phase composition, J Chromatogr A 1408 (2015) 217–226, doi:10.1016/j.chroma.2015.07.037 [5] O Gyllenhaal, A Karlsson, Evaluation conditions for SFC of metoprolol and related amino alcohols on hypercarb (porous graphitic carbon) with respect to structure–selectivity relations, Chromatographia 71 (2010) 7–13, doi:10.1365/ s10337- 009- 1406- [6] T.A Berger, J.F Deye, Effect of basic additives on peak shapes of strong bases separated by packed-column supercritical fluid chromatography, J Chromatogr Sci 29 (1991) 310–317 [7] A Sokolowski, K.-G Wahlund, Peak tailing and retention behaviour of tricyclic antidepressant amines and related hydrophobic ammonium compounds in reversed-phase ion-pair liquid chromatography on alkyl-bonded phases, J Chromatogr A 189 (1980) 299–316, doi:10.1016/S0 021-9673(0 0)80312-X [8] C West, E Lemasson, Unravelling the effects of mobile phase additives in supercritical fluid chromatography—Part II: adsorption on the stationary phase, J Chromatogr A 1593 (2019) 135–146, doi:10.1016/j.chroma.2019.02.002 [9] J Liu, A.A Makarov, R Bennett, I.A Haidar Ahmad, J DaSilva, M Reibarkh, I Mangion, B.F Mann, E.L Regalado, Chaotropic effects in sub/supercritical fluid chromatography via ammonium hydroxide in water-rich modifiers: enabling separation of peptides and highly polar pharmaceuticals at the preparative scale, Anal Chem 91 (2019) 13907–13915, doi:10.1021/acs.analchem 9b03408 [10] G Guiochon, D.G Shirazi, A Felinger, A.M Katti, Fundamentals of Preparative and Nonlinear Chromatography, 2nd ed., Academic Press, Boston, MA, 2006 [11] Torgny Fornstedt, Georges Guiochon, Theoretical study of high-concentration elution profiles and large system peaks in nonlinear chromatography, Anal Chem 66 (1994) 2116–2128, doi:10.1021/ac0 085a030 [12] L.R Snyder, J.W Dolan, High-Performance Gradient Elution: The Practical Application of the Linear-Solvent-Strength Model, Wiley, 2007 [13] E Glenne, K Öhlén, H Leek, M Klarqvist, J Samuelsson, T Fornstedt, A closer study of methanol adsorption and its impact on solute retentions in supercritical fluid chromatography, J Chromatogr A 1442 (2016) 129–139, doi:10.1016/ j.chroma.2016.03.006 [14] E Glenne, H Leek, M Klarqvist, J Samuelsson, T Fornstedt, Systematic investigations of peak deformations due to co-solvent adsorption in preparative supercritical fluid chromatography, J Chromatogr A 1496 (2017) 141–149, doi:10.1016/j.chroma.2017.03.053 [15] T Fornstedt, P Forssén, D Westerlund, System peaks and their impact in liquid chromatography, TrAC Trends Anal Chem (2016), doi:10.1016/j.trac.2016 01.008 [16] E Glenne, H Leek, M Klarqvist, J Samuelsson, T Fornstedt, Peak deformations in preparative supercritical fluid chromatography due to co-solvent adsorption, J Chromatogr A 1468 (2016) 200–208, doi:10.1016/j.chroma.2016.09.019 [17] P Vajda, G Guiochon, Modifier adsorption in supercritical fluid chromatography onto silica surface, J Chromatogr A 1305 (2013) 293–299, doi:10.1016/j chroma.2013.06.075 [18] A Felinger, A Cavazzini, G Guiochon, Numerical determination of the competitive isotherm of enantiomers, J Chromatogr A 986 (2003) 207–225, doi:10 1016/S0021-9673(02)01919-2 ˚ [19] D Asberg, M Enmark, J Samuelsson, T Fornstedt, Evaluation of co-solvent fraction, pressure and temperature effects in analytical and preparative supercritical fluid chromatography, J Chromatogr A 1374 (2014) 254–260, doi:10 1016/j.chroma.2014.11.045 [20] K Kato, M Kokubo, K Ohashi, A Sato, D Kodama, Correlation of High Pressure Density Behaviors for Fluid Mixtures made of Carbon Dioxide with Solvent at 313.15 K, Open Thermodyn J (2009) 1–6 [21] O Kunz, R Klimeck, W Wagner, M Jaeschke, The GERG-2004 wide-range equation of state for natural gases and other mixtures, Fortschr.-Ber VDI, VDI-Verlag, Düsseldorf (2007) [22] E.W Lemmon, I.H Bell, M.L Huber, M.O McLinden, NIST Standard Reference Database 23: Reference Fluid Thermodynamic and Transport Properties-REFPROP, National Institute of Standards and Technology, Standard Reference Data Program, Gaithersburg, 2018 Version 10.0 [23] K Kaczmarski, M Mazzotti, G Storti, M Mobidelli, Modeling fixed-bed adsorption columns through orthogonal collocations on moving finite elements, Comput Chem Eng 21 (1997) 641–660, doi:10.1016/S0 098-1354(96)0 030 0-6 [24] S Javeed, S Qamar, A Seidel-Morgenstern, G Warnecke, Efficient and accurate numerical simulation of nonlinear chromatographic processes, Comput Chem Eng 35 (2011) 2294–2305, doi:10.1016/j.compchemeng.2010.10.002 [25] G Götmar, T Fornstedt, G Guiochon, Retention mechanism of β -blockers on an immobilized cellulase relative importance of the hydrophobic and ionic contributions to their enantioselective and nonselective interactions, Anal Chem 72 (20 0) 3908–3915, doi:10.1021/ac9914824 [26] L.S Riter, O Vitek, K.M Gooding, B.D Hodge, R.K Julian, Statistical design of experiments as a tool in mass spectrometry, J Mass Spectrom 40 (2005) 565– 579, doi:10.1002/jms.871 [27] A Cazenave-Gassiot, R Boughtflower, J Caldwell, L Hitzel, C Holyoak, S Lane, P Oakley, F Pullen, S Richardson, G.J Langley, Effect of increasing concentration of ammonium acetate as an additive in supercritical fluid chromatography using CO2 –methanol mobile phase, J Chromatogr A 1216 (2009) 6441–6450, doi:10.1016/j.chroma.2009.07.022 [28] W Zhang, Y Shan, A Seidel-Morgenstern, Breakthrough curves and elution profiles of single solutes in case of adsorption isotherms with two inflection points, J Chromatogr A 1107 (2006) 216–225, doi:10.1016/j.chroma.2005.12 094 [29] J Samuelsson, T Undin, T Fornstedt, Expanding the elution by characteristic point method for determination of various types of adsorption isotherms, J Chromatogr A 1218 (2011) 3737–3742, doi:10.1016/j.chroma.2011.04.035 [30] Z Yang, Hofmeister effects: an explanation for the impact of ionic liquids on biocatalysis, J Biotechnol 144 (2009) 12–22, doi:10.1016/j.jbiotec.2009.04.011 [31] Y Marcus, Effect of Ions on the Structure of Water: Structure Making and Breaking, Chem Rev 109 (2009) 1346–1370, doi:10.1021/cr8003828 [32] M.G Cacace, E.M Landau, J.J Ramsden, The Hofmeister series: salt and solvent effects on interfacial phenomena, Q Rev Biophys 30 (1997) 241–277, doi:10 1017/S0033583597003363 [33] V Desfontaine, D Guillarme, E Francotte, L Nováková, Supercritical fluid chromatography in pharmaceutical analysis, J Pharm Biomed Anal 113 (2015) 56– 71, doi:10.1016/j.jpba.2015.03.007 [34] D Speybrouck, C Doublet, P Cardinael, C Fiol-Petit, D Corens, The effect of high concentration additive on chiral separations in supercritical fluid chromatography, J Chromatogr A 1510 (2017) 89–99, doi:10.1016/j.chroma.2017.06 049 [35] H.D.B Jenkins, Yizhak Marcus, Viscosity B-Coefficients of Ions in Solution, Chem Rev 95 (1995) 2695–2724, doi:10.1021/cr0 040a0 04 [36] A.P dos Santos, A Diehl, Y Levin, Surface tensions, surface potentials, and the Hofmeister series of electrolyte solutions, Langmuir 26 (2010) 10778–10783, doi:10.1021/la100604k [37] L Miller, D Honda, R Fronek, K Howe, Examples of preparative chiral chromatography on an amylose-based chiral stationary phase in support of pharmaceutical research, J Chromatogr A 658 (1994) 429–435, doi:10.1016/ 021-9673(94)80 033-2 [38] P Forssén, L Edström, M Lämmerhofer, J Samuelsson, A Karlsson, W Lindner, T Fornstedt, Optimization strategies accounting for the additive in preparative chiral liquid chromatography, J Chromatogr A 1269 (2012) 279–286, doi:10 1016/j.chroma.2012.05.052 [39] P Forssén, R Arnell, M Kaspereit, A Seidel-Morgenstern, T Fornstedt, Effects of a strongly adsorbed additive on process performance in chiral preparative chromatography, J Chromatogr A 1212 (2008) 89–97, doi:10.1016/j.chroma 2008.10.040  ... pronounced with decreasing MeOH fractions in the eluent (bottom to top in Fig 5) and with increasing amine additive contents in the eluent (left to right in Fig 5), finally resulting in clearly overloaded... with increasing MeOH fractions, going from top to bottom, and a sharpened overloaded elution profile with increasing DEA concentrations, going from left to right In Fig 3a we see some sign of peak. .. stationary phase is in line with their order concerning increasing capability for forming hydrogen bonding: primary amine > secondary amine > tertiary amine In Fig 7, the retention factors of all solutes