746 PREPARATIVE SEPARATIONS 1:10 10:1 0.005 g A 0.05 g B 0.05 g A 0.005 g B A B 23 4 2 3 4 2 34 2 3 4 2 3 4 23 4 0.5 g A 5 g B 5 g A 0.5 g B T-P 2 g A 20 g B 20 g A 2 g B (a) (b) (c) Figure 15.9 Isocratic separation of a two-component sample as a function of sample size. Computer simulations based on the Langmuir isotherm; Conditions: 250 × 50-mm column (7-μm), 210 mL/min flow rate, N = 800; k = 1 and 1.5, respectively. Sample weights indi- cated in figure. Adapted from [9]. (Section 9.3.1). If there is any doubt as to the identity of the product peak in these initial separations, this can be confirmed by a separate injection of pure product. Step 2 of Figure 15.8. Following the adjustment of %B in step 1, separation conditions are varied for the best possible separation of the product peak from adjacent impurity peaks. Usually the product peak should be placed midway between the adjacent impurities on each side. Previous chapters provide a detailed discussion of how selectivity α can be optimized, depending on the kind of sample and whether NPC or RPC is used (see Table 2.2 for conditions that affect α). Because of the importance of maximizing α in prep-LC (Eq. 15.5), more work on step 2 may be warranted than for analogous analytical separations. Unlike the case of analytical separation, in prep-LC it is important—if possible—to avoid separation conditions that result in > 50% ionization of the product molecule (see Section 15.3.2.1 and the discussion of Fig. 15.6). Large changes in α (without ionizing the product) are most likely to be achieved by a change in B-solvent or the column. 15.3 ISOCRATIC ELUTION 747 The same conditions used for this optimized separation can be used to assay fractions collected during prep-LC (but with the initial small-scale column). If the resolution of the product peak is R s 2 (desirable for prep-LC), the assay separations can be carried out with a shorter column and increased flow rate to speed up fraction analysis. Step 3 of Figure 15.8. An initial estimate of the weight of injected sample is possible, based on (1) Equation (15.5), (2) a value of α, (3) the column capacity w s , and (4) a rough estimate of sample purity (%-product in the sample). For a 150 × 4.6-mm column with 10-nm pores, and a product that does not ionize in the mobile phase, the column capacity can be estimated as w s ≈ 150 mg (≈100 mg/g of column packing), from which the weight of sample for T-P separation is w s ≈ (1/6) × (150) × ([%-product]/100) × ([α − 1]/α) 2 ; if the product is partly or completely ionized, the allowed sample weight can be much lower than the latter estimate. Following a separation with this estimated sample weight, sample weight can be increased or decreased by trial and error to achieve T-P separation. Alternatively, the use of fully automated equipment allows a number of trial separations where sample size is varied; from such experiments the correct sample weight can be quickly determined. Once a promising separation is identified in this way, the product peak should be collected and assayed, in order to confirm ≈100% recovery and purity. It may also be worthwhile at this point to see if an increase in flow rate can maintain the latter separation, but with a reduced run time. The object of prep-LC is usually maximum production of purified product in minimum time, which favors short run times. Step 4 of Figure 15.8. The final step in Figure 15.8 (scale-up) completes method development. The desired scale-up factor can be calculated from the results of step 3 (see Section 15.1.2.1), taking into account the availability of (1) columns of different i.d. and (2) equipment that can provide the required flow rate. A final separation with this larger column can then be carried out, allowing verification of the product recovery and purity obtained with the previous (smaller) column. Scale-up should result in essentially the same purity and recovery of product as found for the small-scale separation. 15.3.2.5 Fraction Collection The usual goal of fraction collection—whether carried out manually or with an automated system—is to obtain a maximum yield of adequately pure product, with as little effort as possible. The initial step is to collect a number of fractions across the product peak, followed by their analysis for content and purity. These results can be used to determine the time during which the product peak should be collected (best ‘‘cut points’’) in the final separation(s), so as to achieve the purification goals (Section 15.4.1). Prior to finalizing the prep-LC procedure, a trial separation can be carried out to confirm the latter cut points. A few small fractions around each cut point can be collected for this purpose. Once the separation procedure and cut points are finalized, only a single product fraction need be collected. However, additional fractions can provide insurance against unanticipated changes in the separation. 748 PREPARATIVE SEPARATIONS 15.4 SEVERELY OVERLOADED SEPARATION A detailed study of severe column overload (i.e., sample sizes larger than those that correspond to T-P separation) is beyond the scope of the present book; however, it is useful to consider certain aspects of such separations. Such severely overloaded separations can result in a greater production of purified product per hour with reduced consumption of the mobile phase, as well as requiring smaller columns and smaller scale equipment—all of which can be of great practical importance. The disadvantage of such separations is that more effort is required for method development, and individual separations usually require the collection and analysis of several product fractions so that only adequately pure material is obtained. It may also be necessary to re-process product fractions that are insufficiently pure. The interested reader is referred to several texts [4, 9, 10] for further study. 15.4.1 Recovery versus Purity As sample weight increases to the point of severe overload, the prediction of individual peak shapes becomes more uncertain. This is illustrated in Figure 15.9 for small-sample (Fig. 15.9a), T-P (Fig. 15.9b), and severely-overloaded (Fig. 15.9c) separations of a sample where the relative concentrations of the two components A and B vary from 1:10 to 10:1. Thus we can see what happens to a minor peak that elutes either before or after the (larger) product peak. For severe overload (Fig. 15.9c), when the impurity peak precedes the product peak, it is displaced and compressed so that peak height increases. There is also some overlap of the two peaks. When the impurity peak follows the product peak, it is dragged into the product peak (so-called tag-along effect). The relative importance of these two effects can be difficult to predict, so the optimum sample weight must be determined experimentally. This optimum weight will also vary with the relative concentrations of product and impurities. 0 10 20 30 40 50 60 70 80 90 100 88 90 92 94 96 98 100 Purit y( % ) Recovery (%) bovine insulin porcine insulin Figure 15.10 Plot of recovery against product purity for a given sample load (5 mg) for a 1:1 mixture of bovine and porcine insulins. Bovine insulin (solid line); porcine insulin (dashed line). Conditions: 250 × 4.6-mm (20-μm) C8 column; 10–29% acetonitrile–0.1% aqueous TFA in 10 minutes. Adapted from [13]. 15.4 SEVERELY OVERLOADED SEPARATION 749 (a) (d )(e)(f ) (c)(b) Isocratic elution Gradient elution 2.5 mg A 2.5 mg A 2.5 mg A 2.5 mg B 10 mg B 25 mg B A, A’ B, B’ A A’ B, B’ A A’ B, B’ 3 10 11 10 10 11 1112 12 12 (min) (min) 13 13 (min) 13 3 5 (min)6 (min)64 35 4 (min)654 2.5 mg A 2.5 mg A 2.5 mg A 2.5 mg B 10 mg B 25 mg B A, A’ B, B’ A A’ B, B’ A A’ B, B’ Figure 15.11 Similar effects of column overload in corresponding separations: (a)isocratic and (b) gradient elution. Separation of two xanthines (β-hydroxyethyltheophylline [A] and 7β-hydroxypropyltheophylline [B]) with k (isocratic) equal k ∗ (gradient). Sample weights shown in figure. Peaks labeled A and B are for the injection of samples of pure of A or B; peaks labeled A and B are for the separation of mixtures of A and B. Adapted from [2]. 15.4.2 Method Development The selection of an optimum sample size for severely overloaded prep-LC can start with an optimized T-P separation (as in Fig. 15.2b), followed by injecting successively larger sample weights. For each separation (or sample weight), a number of fractions that surround and include the product peak are collected and assayed, and the results are collected within a spreadsheet. Based on pooling the purest fractions, product recovery (or yield) can then be plotted against product purity, as in the examples of Figure 15.10 for two different products (bovine and porcine insulin). Similar plots will result for different sample weights, allowing selection of the most attractive sample weight. To a first approximation, the maximum weight of purified product with some predetermined purity (e.g., 98%) can be established in this way. For the 750 PREPARATIVE SEPARATIONS examples of Figure 15.10 the recovery of 98%-pure material would be 92 and 42%, respectively, for bovine and porcine insulin. Similar plots for different sample sizes might result in a better compromise between the weight of purified product and its recovery. The separation of Figure 15.10 was carried out with RPC using gradient elution. However, the same approach would be used for NPC or isocratic elu- tion. The principle of estimating sample size is the same regardless of whether isocratic or gradient elution chromatography is being used. Optimum separation conditions—other than sample weight—may not be the same for T-P as compared with severely overloaded separation. In both cases a very considerable experimental effort can be required in order to simultaneously optimize both sample size and separation conditions. 15.4.2.1 Column Efficiency As the (small-sample) separation factor α 0 increases for T-P separation, and a larger sample weight becomes possible, the effect of the small-sample column plate number N 0 on product resolution decreases (Eq. 15.4a). Smaller values of N 0 are therefore required, with little effect on the recovery or purity of the product. This is no longer the case for severely overloaded separations. Displacement effects as in Figure 15.9c, for a small peak that precedes a large peak, can improve separation. This is better shown in the isocratic separations of Figure 15.11c,whereinthe absence of sample displacement peak B would completely overlap peak A. Because of displacement, there is some separation of the two peaks (compare the similar situation of Fig. 15.9c). Sample displacement is highly advantageous in severely overloaded separation, but unlike the case of T-P separation, it appears to be favored by higher values of N 0 [11]. As a result large-scale separations are generally carried out with moderately efficient columns that use particle diameters of 10 to 15 μm. 15.4.2.2 ‘‘Crossing Isotherms’’ This unusual behavior can arise for solutes with different saturation capacities. An example is seen in the separation of alcohols from phenols [13], where alcohols can have significantly higher saturation capacities than phenols. Figure 15.12a shows the RPC separation of benzyl alcohol and phenol for a small sample (10 μg), where benzyl alcohol elutes last. Figure 15.12c shows the same separation for a larger sample (1-mg phenol, 3-mg benzyl alcohol); the two peaks are almost completely separated. When the order of elution of a phenol and alcohol are reversed, while the weights of early- and late-eluting compound are held the same (as in Fig. 15.12a,c), a very different result is obtained; see Figure 15.12b,d for the separation of phenethyl alcohol and p-cresol. In the overloaded separation of Figure 15.12d, peak overlap is almost complete—contrasting strongly with the analogous separation of Figure 15.12c. The reason for the contrasting separations of Figure 15.12 is somewhat complicated, but can be pictured in terms of ‘‘crossing isotherms’’—as illustrated in Figures 15.12e (phenol and benzyl alcohol) and Figure 15.12f (phenethyl alcohol and p-cresol). For the separation of phenol and benzyl alcohol, the isotherms do not cross (Fig. 15.12e) because phenol is always more retained than benzyl alcohol, and 15.5 GRADIENT ELUTION 751 8 (min) 108 6 10 8 10 12(min) (min) 8 10 12 (min) OH CH 2 OH CH 2 OH CH 2 OH C s C s C m C m CH 3 CH 3 CH 3 CH 2 CH 2 OH CH 2 CH 2 OH CH 2 CH 2 OH OH OH OH (a)(b) (c)(d ) (e)(f ) OH OH Figure 15.12 Example of crossing-isotherm behavior, with decrease in allowed sample weight for touching-peak separation. Conditions: 150 × 4.6-mm (5-μm) C 18 column; methanol-water mobile phases; 1.0 mL/min. (a)3μg phenol and 7 μg benzyl alcohol (BA); (b)3μg 2-phenylethanol (PE) and 7 g p-cresol; (c)sameas(a), except 4-mg sample weight; (d) same as (b), except 4-mg sample weight; (e, f ) hypothetical isotherms corresponding to separa- tions of phenol-benzyl alcohol and 2-phenylethanol/p-cresol, respectively. Adapted from [14]. the two compounds are well separated. For the separation of phenethyl alcohol and p-cresol (Fig. 15.12f), the greater retention of p-cresol for a small sample, combined with its smaller column capacity, leads to crossing of the isotherms for a sufficiently large sample. For the latter sample weight, the two compounds are equally retained, with no separation—as observed in the separation of Figure 15.12d. The latter explanation is intentionally oversimplified. 15.5 GRADIENT ELUTION While gradient elution is often used for analytical separations and small-scale prep-LC, its use for large-scale separations can be less convenient and more costly. 752 PREPARATIVE SEPARATIONS An exception to this generalization occurs for the separation of large biomolecules because their isocratic retention can vary greatly for small changes in %B (Section 13.4.1.4), making isocratic elution impractical or impossible. An example of the industrial-scale purification of biosynthetic human insulin by gradient elution is discussed in Section 13.9.2. Method development for gradient separations closely parallels that for isocratic separation, as discussed in Chapter 9. Thus, when the gradient retention factor k ∗ is the same as k for isocratic elution, and other conditions are the same (‘‘corresponding’’ separation; Section 9.13), the separation of a product peak from its impurities will be the same for both isocratic and gradient elution. Similarly any change in conditions that can improve isocratic selectivity can be used in the same way to improve gradient separation. Consequently virtually everything that applies for isocratic prep-LC in Section 15.3 applies equally for gradient elution. This will simplify our remaining discussion of gradient prep-LC in this section. For additional information about gradient prep-LC, see [2]. 15.5.1 Isocratic and Gradient Prep-LC Compared Figure 15.11a–c was used previously to compare the effect of sample size on an isocratic separation, where only the weights of two compounds in the sample are varied. A similar series of separations is shown in Figure 15.11d–f for the gradient separation of the same sample (compounds A and B) with the same conditions (except that gradient steepness replaces %B). In each case chromatograms are overlaid for (1) the separate injection of each compound (A and B ), and (2) the injection of the mixture (A plus B); see the related discussion of Section 2.6.2 and Fig. 2.24. The isocratic and gradient chromatograms for separations of equal sample weights (e.g., Fig. 15.11a vs. d, b vs. e, c vs. f ) are seen to be virtually identical, with the exception of the more rounded (‘‘shark-fin’’ shaped) peaks for overloaded gradient elution in Figure 15.11d –f. This similarity of isocratic and gradient separations under comparable con- ditions was discussed in Section 9.1.3. For equivalent results as in Figure 15.11 for ‘‘corresponding’’ isocratic and gradient separations, the retention factor for each peak in the isocratic (k) and gradient (k ∗ ) separations must be approximately equal, and all other separation conditions (column, A- and B-solvents, flow rate, temperature) must be the same. In the gradient separations of Figure 15.11d–f , separation conditions were adjusted so that (small-sample) values of k ∗ were equal to isocratic values of k in Figure 15.11a–c. As discussed in Section 9.2, values of k ∗ are determined by gradient conditions: k ∗ = 0.87t G F V m ΔφS (9.5) Here t G is the gradient time, F is flow rate, Δφ is the change in φ ≡ 0.01 × (%B) during the gradient, S is related to the change in k for a given change in φ or %B (equal to d[log k]/dφ), and V m is the column dead-volume (mL)—which can be determined from an experimental value of t 0 and the flow rate F (Section 2.3.1; V m = t 0 F). Changes in isocratic separation as a result of a change in %B can be replicated in gradient elution, by a change in gradient time t G (Eq. 9.5), so that the new values of both k and k ∗ are the same. 15.5 GRADIENT ELUTION 753 A B f* f* f* log k* log k* log k* t G = 30 t G = 30 t G = 30 S A = S B (Parallel) S A < S B (Divergent) S A > S B (Convergent) A B A B A B (a)(b) A B A B (c) Figure 15.13 Effect of unequal values of S on the overload separation of two peaks by gradient elution. Adapted from [2]. 15.5.2 Method Development for Gradient Prep-LC The general plan of Figure 15.8 for isocratic method development can be followed for gradient elution also, with a few changes. The selection of initial conditions will be virtually the same, except than the initial separation will be carried out with a 0–100%B gradient, followed by narrowing the gradient range in most cases so as to shorten run time (Section 9.3.4). The optimization of the initial gradient separation for improved selectivity (step 2) can be carried out in the same way as for isocratic elution (Section 9.3.3), except that the goal is a maximum resolution for the product peak, rather than acceptable resolution of all peaks in the chromatogram. Aside from the choice of conditions for maximum α, the gradient program can be further modified so as to minimize separation time, while maintaining the resolution of the product peak from its adjacent impurity peaks (see Sections 9.3.4, 9.3.5). Maximizing sample weight (step 3) and scale-up (step 4) then proceed in exactly the same way as for isocratic prep-LC. For further details, see [2]. A complication not found for isocratic prep-LC is observed occasionally in gradient prep-LC. When two adjacent peaks have different values of S, this can affect the sample weight for T-P separation, as illustrated conceptually in Figure 15.13. Figure 15.13a gives the T-P gradient separation for the case of equal S-values for two compounds (‘‘parallel’’ case), which is often a close approximation for most samples. At the top of Figure 15.13a is a plot of log k ∗ versus φ* for each peak, where φ* is the value of φ (≡0.01 × %B) when the peak is at the column mid-point 754 PREPARATIVE SEPARATIONS (see Eq. 9.5a); values of φ* in Figure 15.13 track the time during the separations shown at the bottom of Figure 15.13a–c . The dotted lines connect the log k ∗ –φ ∗ plot for compounds A and B to (small-sample) peaks in the chromatogram below. The values of k ∗ and φ* for each peak are determined by gradient conditions, with an assumed gradient time of 30 minutes for each separation in Figure 15.13. Now assume that a large enough sample weight has been injected to allow peak B to cover the space between the two small-sample peaks (T-P separation), giving the wide cross-hatched peaks in the chromatogram of Figure 15.13a.Wesee that the vertical separation of the two log k ∗ versus φ plots is constant and equal to log α for each value of φ. Thus, at the beginning of elution of overloaded peak B (at a lower value of φ*, corresponding to the elution of a small sample of A), α isthesame(= α o ) as at the end of elution; that is, the separation factor is not a function of sample weight. (Note that Eq. 15.5, which relates sample weight for T-P separation to values of α o , assumes approximately equal values of S for the two adjacent peaks.) Figure 15.13b is similar to that of 15.13a (same weight of injected sample for T-PseparationinFig.15.13a), except that now the plots of log k ∗ versus φ*are no longer parallel but diverge for lower values of φ* (‘‘divergent’’ case); that is, the value of S for compound B is greater than for compound A. For higher loading of the column (at lower values of φ*), the vertical separation of the two log k ∗ –φ* curves increases, corresponding to an increase in α with increasing sample weight. A larger value of α means a larger sample weight for T-P separation (Eq. 15.5), so the same injected weight of sample as in Figure 15.13a is no longer sufficient to cause the peaks to touch. That is, the divergent case allows a larger weight of injected sample (other factors equal), compared to the equal-S case of Equation (15.5) and Figure 15.13a. Figure 15.13c illustrates the third possibility: log k ∗ –φ* plots that converge for smaller φ* (‘‘convergent’’ case); namely S for compound B is less than for compound A. Now α decreases with increasing sample weight, and injection of the same weight of sample as in Figure 15.13a for T-P separation leads to a more rapid column overload with overlap of the two peaks. When convergent behavior is suspected (because of lower than expected sample weights for T-P separation), further changes in separation conditions should be considered—with the goal of reversing the elution order of the two peaks (product and nearest impurity bands). A similar approach can also be used to minimize the problem of crossing isotherms (Section 15.4.2.2). For a further discussion of the consequences of unequal S-values in gradient prep-LC, see [2]. 15.6 PRODUCTION-SCALE SEPARATION Production-scale separations are well beyond the scope of this book, but the simple theory and practice outlined here is still pertinent. Separations of this kind are usually highly optimized, so as to result in the highest possible production rate for the desired product, with the required purity and recovery. At this scale, process economics are of primary importance; the goal is a combination of purity, recovery, and production rate that yield the lowest cost per kg of the desired product, including the cost of removing mobile phase from the purified product. Separation conditions REFERENCES 755 are usually developed empirically, using the approach of Figure 15.10, for samples much larger than correspond to T-P separation. For an example, see Section 13.9 for the production-scale separation of rh-insulin. For production-scale separations, the use of simulated moving bed (SMB) techniques are increasingly important. This is a binary separation technique that relies on the simulation of a countercurrent separation system by the use of multiple columns and switching valves. The reader is referred to specialized texts on this topic [14]. Although this approach has been used for many decades in the petroleum industry (the Molex process for isolation of p-xylene) and the food industry (the Sorbex process for fructose-rich syrups), it has only been used by the pharmaceutical industry since the 1990s. The use of countercurrent separation with a continuous sample input and product output gives a more effective use of the chromatographic bed than the traditional procedures discussed in this chapter. SMB thus reduces both the size of the columns and the amount of solvent used (and therefore costs); it is widely used. The column efficiency required for countercurrent separations is relatively low, which allows the use of very short ‘‘pancake’’ columns. For example, successful enantiomer separations are carried out under SMB conditions with columns of 800- or 1000-mm i.d., but only 100 mm in length (e.g., packed with 20-μm particles). REFERENCES 1. M. S Villeneuve and L. A. Miller in Preparative Enantioselective Chromatography,G. B. Cox, ed., Wiley-Blackwell, 2005. 2. L. R. Snyder and J. W. Dolan, High Performance Gradient Elution, Wiley, Hoboken, NJ, 2007, ch. 7. 3. Azeotropes: http://en.wikipedia.org/wiki/Azeotrope (data). 4. G. Guiochon, A. Felinger, D. G. Shirazi, and A. M. Katti, Fundamentals of Preparative and Non-linear Chromatography, 2nd ed., Academic Press, Boston, 2006. 5. J. H. Knox and H. M. Pyper, J. Chromatogr., 363 (1986) 1. 6. L.R.Snyder,G.B.Cox,andP.E.Antle,Chromatographia, 24 (1987) 82. 7. U. D. Neue, C. B. Mazza, J. Y. Cavanaugh, Z. Lu, and T. E. Wheat, Chromatographia Suppl., 57 (2003) S-121. 8. U. D. Neue, T. E. Wheat, J. R. Mazzeo, C. B. Mazza, J. Y. Cavanaugh, F. Xia, and D. M. Diehl, J. Chromatogr. A, 1030 (2004) 123. 9. L. R. Snyder, J. J. Kirkland, and J. L. Glajch, Practical HPLC Method Development, 2nd ed., Wiley-Interscience, New York, 1997. 10. H. Schmidt-Traub, Preparative Chromatography of Fine Chemicals and Pharmaceutical Agents, Wiley-VCH, New York, 2005. 11. G. Guiochon and S. Ghodbane, J. Phys. Chem., 92 (1988) 3682. 12. G. B. Cox and L. R. Snyder, J. Chromatogr., 590 (1992) 17. 13. G. B. Cox and L. R. Snyder, J. Chromatogr., 483 (1989) 95. 14. O. Dapremont in G. B. Cox, Preparative Enantioselective Chromatography, Wiley-Blackwell, New York, 2005. . 15.13a is no longer sufficient to cause the peaks to touch. That is, the divergent case allows a larger weight of injected sample (other factors equal), compared to the equal-S case of Equation. Preparative and Non-linear Chromatography, 2nd ed., Academic Press, Boston, 2006. 5. J. H. Knox and H. M. Pyper, J. Chromatogr., 363 (1986) 1. 6. L.R.Snyder,G.B.Cox,andP.E.Antle,Chromatographia, 24 (1987). columns of 800 - or 1000-mm i.d., but only 100 mm in length (e.g., packed with 20-μm particles). REFERENCES 1. M. S Villeneuve and L. A. Miller in Preparative Enantioselective Chromatography,G. B.