Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 3

AFFINITY SEPARATION

K. Jones, Affinity Ltd, Freeport, Separation and puriRcation methods for biological Ballsalla, Isle of Man, UK macromolecules vary from the very simple to the esoteric. The type of technique adopted is basically Copyright ^ 2000 Academic Press a function of source, the fragility of the molecule and the purity required. Traditionally, high purity Introduction pharmaceuticals have used multistage processing, but this is very inefRcient as measured by the well- Of the collection of separation technologies known documented fact that 50}80% of total production as ‘afRnity’, afRnity chromatography is by far the costs are incurred at the separation/puriRcation stage. most widely used variant. AfRnity chromatography is In contrast, the highly selective indigenous properties becoming increasingly important as the speed of the of the afRnity method offer the alternative of revolution taking place in biotechnology processing very elegant single-step puriRcation strategies. The increases. The concept of an ‘afRnity’ separation re- inherent simplicity and universality of the method has sults from a naturally occurring phenomenon existing already generated a wide range of separation tech- within all biological macromolecules. Each biological nologies, mostly based upon immobilized naturally macromolecule contains a unique set of intermolecu- occurring proteinaceous ligands. By comparing the lar binding forces, existing throughout its internal ‘old’ technologies of ‘natural’ ligands or multistage and external structure. When alignment occurs be- processing with the ‘new’, exempliRed by synthetic tween a speciRc site of these forces in one molecule designed ligands, the most recent advances in af- with the site of a set of forces existing in another Rnity processing can be described. (different) molecule, an interaction can take place R between them. This recognition is highly speci cto Biological Recognition the pair of molecules involved. The interactive mech- anism can be converted into a universal mutual bind- As nature evolved, life forms had to develop a protec- ing system, where one of the binding pair is attached tive mechanism against invading microorganisms if to an inert matrix, packed into a column and used they were to survive. Thus there is a constant battle exclusively to capture the other matching molecule. between the cell’s defence mechanism and the attack- When used in this (afRnity) mode, the technique is ing microorganisms, a battle resolved by the cells probably the simplest of all chromatographic generating (the immunoglobulins) able to methods. It is, however, restricted almost exclusively recognize the protein coat of attacking microorgan- to the separation and puriRcation of biological mac- isms and signal killer cells to destroy the invaders romolecules, and is unsuitable for small molecules. before they cause harm to the host. Equally, if micro- AfRnity chromatography or bioselective adsorp- organisms were to survive, they had continually to tion chromatography was Rrst used in 1910, but it was mutate and change their protein coats to avoid detec- only in the 1960s that afRnity chromatography as tion by existing antibodies. The ‘attack and destroy’ practised today was developed as a puriRcation tech- process is a function of changes in the molecular nique. By the late 1970s the emergence of recombinant structure in a speciRc part of the protein, with only DNA technology for the manufacture of protein phar- the most minute of changes occurring at the surface of maceuticalsprovidedanewimpetusforthishighly the protein. Evolution has thus designed a system speciRc chromatographic method, implemented by the where every protein has a very precise structure, but demand for ever-increasing product purity implicit in one which will always be recognized by another. One regulatory frameworks devised by (amongst others) element of the interacting pair can be covalently the USA’s Food and Drug Administration (FDA). Fi- bonded onto an inert matrix. The resulting chromato- nally, the need to reduce the cost of drugs is under graphic medium can then be packed into a column, constant scrutiny by many Governments, particularly and used to separate exclusively its matching partner those with controlled health schemes funded by rev- from an impure mixture when added as a solution to enue raised by taxation. These mutually incompatible the top of the column. This fact can be stated as pressures indicate the need for more efRcient sep- follows } for every protein separation problem there aration systems; the afRnity technique provides the is always an afTnity solution. The process of promise of meeting all necessary requirements. producing a satisfactory medium is quite difRcult. 4 I / AFFINITY SEPARATION / Derivatization

The matching pair must be identiRed, and one afRnity method is not restricted to protein separ- of them isolated in a pure form. Covalent bond- ations; nucleic acids and whole cells can also be ing onto an inert matrix in a stable manner must separated. always allow the ‘docking’ surface of the protein to The simplicity of the chromatographic process is be positioned to make it available to the target pro- shown in Figure 1. The of interest, covalently tein. The whole also has to be achieved at an accept- bonded onto the inert matrix, is contained in the able cost. column, and a solution containing the target (the This technique has resulted in many successful ap- ligate) is passed through the bed. The ligand recog- plications, often using antibodies as the afRnity nizes the ligate to the exclusion of all other molecules, medium (immunoafRnity chromatography), but with the unwanted materials passing through the col- large scale separations using these ‘natural’ ligands umn packing while the ligate is retained. Once the are largely restricted by cost and regulatory reasons. bed is saturated with the target molecule (as mea- Although immunoafRnity chromatography is still sured by the breakthrough point), contaminating spe- widely practised, in recent years the evolution of cies are washed through, followed by collection of the design technologies has provided powerful new ap- target molecule as a very pure fraction using an proaches to mimic protein structures, resulting in the eluting buffer solution. Finally, the column is development of synthetic ligands able to work in cleansed from any strongly adsorbed trace materials, harsh operational environments and at low cost. usually by regeneration with a strong alkali or acid, making it available for many more repeat runs. An R The Af\nity Process outstanding advantage of the af nity process is an ability to concentrate very dilute solutions while The afRnity method is critically dependent upon stabilizing the captured protein once adsorbed onto the ‘biological recognition’ existing between species. the column. Many of the in-demand manu- By permanently bonding onto an inert matrix a mol- factured by genetically engineered microorganisms ecule (the ligand) that speciRcally recognizes the mol- are labile, allowing only minute quantities to be pres- ecule of interest, the target molecule (the ligate) can ent in the fermentation mix before they begin to be separated. The technique can be applied to any deteriorate. An ability to capture these very small biological entity capable of forming a dissociable quantities while stabilizing them in the adsorbant complex with another species. The dissociation con- phase results in maximization of yield, making mass- S stant (Kd) for the interaction re ects the comp- ive savings in total production costs. lementarity between ligand and ligate. The optimal Although the technical processing advantages are R R range of Kd for af nity chromatography lies be- clear there is a major dif culty in the appli- tween 10\4 and 10\8 mol L\1. Most biological cation of afRnity chromatography as understood ligands can be used for afRnity purposes provid- by most practitioners today. Most ligands described ing they can be immobilized, and once immobilized in Table 1 suffer from two primary disadvan- continue to interact successfully with their respective tages: a lack of stability during use; and high ligates. The ligand can be naturally occurring, an cost. Fortunately these problems have now been over- engineered macromolecule or a synthetic molecule. come, and afRnity chromatography is now accep- Table 1 provides some examples of immobili- ted as the major separations technology for zed ligands used to purify classiRed proteins. The proteins.

Table 1 Affinity ligands and purified proteins

Immobilized ligand Purified protein

Divalent and trivalent metal ion Proteins with an abudance of his, tryp and cys residues , cells Carbohydrates Lectins Reactive dyes Most proteins, particularly nucleotide-binding proteins Nucleic acids Exo and endonucleases, polymerases, other nucleic acid-binding proteins Amino acids (e.g. lys, arg) Nucleotides, cofactors substrates and inhibitors Proteins A and G Immunoglobulins Hormones, drugs Receptors Antibodies Antigens Antibodies Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 5

The beaded have captured over 85% of the total market for biological macromolecule separ- ations, and are regarded as the industry standard to which all other supports are compared. They have achieved this position by providing many of the desir- able characteristics needed, and are also relatively inexpensive. Beaded agaroses do have one severe lim- itation } poor mechanical stability. For analytical applications speed and sensitivity are essential, de- manding mechanically strong, very small particles. Beaded agaroses are thus of limited use analytically, Figure 1 Schematic diagram of affinity chromatography. a gap Rlled by high performance liquid chromatogra- phy (HPLC) using silica matrices. For preparative and Matrices large scale operations other factors are more impor- tant than speed and sensitivity. For example, mass By deRnition matrices must be inert and play no part transfer between stationary phase and mobile phase is in the separation. In practice most play a (usually) much less important when compared to the contri- negative role in the . To minimize bution from the chemical kinetics of the binding these disadvantages matrices have to be selected with reaction between stationary phase and protein. great care. There is a theoretically perfect matrix, Band spreading is also not a serious problem. When deRned as consisting of monodispersed perfectly combined with the highly selective nature of the shaped spheres ranging from 5 to 500 m in dia- afRnity mechanism, these factors favour the com- meter, of high mechanical strength, zero nonspeciRc mon use of large sized, low mechanical strength par- and with a range of selectable pore sizes ticles. from 10}500 nm, a very narrow pore size distribution In recent years synthetic polymeric matrices have and low cost. This idealized matrix would then pro- been marketed as alternatives. Although nonbiodeg- vide the most efRcient separation under all ex- radable, physically and chemically stable, with good perimental conditions. As always, a compromise has permeabilities up to molecular weights greater than to be reached, the usual approach being to accentuate 107 Da, the advantages provided are generally off- the most attractive characteristics while minimizing set by other quite serious disadvantages, exempliRed the limitations, usually by manipulating the experi- by high nonspeciRc adsorption. Inorganic matrices mental conditions most likely to provide the optimum have also been used for large scale protein separ- result. ations, notably reversed-phase silica for large scale The relative molecular masses of proteins vary recombinant human insulin manufacture (molecular from the low thousands to tens of millions, making weight approximately 6000 Da), but are generally pore size the most important single characteristic of not preferred for larger molecular weight pro- the selected matrix. Very large molecules need very ducts. A very slow adoption of synthetic matrices is open and highly porous networks to allow rapid and easy penetration into the core of the particle. Struc- tures of this type must therefore have very large pores, Table 2 Support matrices but this in turn indicates low surface areas per unit volume, suggesting relatively low numbers of surface Support matrix Operational pH range groups to which ligands can be covalently attached. The matrix must also be biologically and chemically 2}14 inert. A special characteristic demanded from biolo- Cellulose 1}14 gical macromolecular separations media is an ability Dextran 2}14 ( to be sanitized on a routine basis without damage. Silica 8 Glass (8 This requires resistance to attack by cleansing re- Polyacrylamides 3}10 agents such as molar concentrations of strong alkali, Polyhydroxymethacrylates 2}12 acids and chaotropes. In contrast to analytical separ- Oxirane}acrylic copolymers 0}12 ations, where silica-based supports are inevitably Styrene}divinylbenzene copolymers 1}13 } used, silica cannot meet these requirements and is Polyvinyl alcohols 1 14 N-Acryloyl-2-amino-2-hydroxy-1, 2-propane 1}11 generally not favoured for protein separations. PTFE Unaffected Table 2 contains examples of support matrices used in afRnity separations. PTFE, polytetrafluoroethylene. 6 I / AFFINITY SEPARATION / Derivatization indicated as improvements are made to current ma- coupled directly to the matrix, but most require coup- terials and the prices of synthetics begin to approach ling via a previously activated matrix. The afRn- those of agarose beads. Other factors resist any signif- ity matrix selected must have an adequate number of icant movement towards synthetic matrices. Most appropriate surface groups onto which the ligand can installed processing units are designed for low perfor- be bonded. The most common surface group is hy- mance applications. Higher performance matrices droxyl. The majority of coupling methods involve the would need reinstallation of new, much higher cost activation of this group by reacting with entities con- high performance plant; plant operators would need taining halogens, epoxy or carbonyl functional retraining; operating manuals would need rewriting; groups. These surface residues are then coupled to and plant and factory would need reregistration with ligands through primary amines, hydroxyls or the FDA. In combination, the implication is that groups, listed in Table 3. penetration of high performance systems for large Polysaccharides, represented by agarose, have scale applications will be slow, and agarose beads will a high density of surface hydroxyl groups. Tradition continue to dominate the market for protein separ- still dictates that this surface is activated by cyanogen ations. bromide, but it is well established that this reagent forms pH-unstable iso-urea linkages, resulting in Covalent Bonding a poorly performing product. Furthermore CNBr- activated agarose needs harsh coupling conditions A basic requirement of all chromatographic media is if high yields of Rnal media are to be obtained, the need for absolute stability under all operational suggesting high wastage of often expensive ligands. conditions through many cycles of use. Consequently This factor is particularly evident with fragile entities all ligands must be covalently bonded onto the such as the very-expensive-to-produce antibodies, matrix, and various chemistries are available to and yet many workers simply read previous literature achieve this. and make no attempt to examine alternative far A number of factors are involved: superior coupling methods. The advantages of mild coupling regimes are demonstrated in Figure 2, where 1. The performance of both ligand and matrix are the use of a triazine-activated agarose is compared to not impaired as a result of the coupling process. CNBr-activated agarose. Yield is signiRcantly in- 2. Most of the coupled ligand is easily accessible to creased, largely by coupling under acidic rather than the ligate. alkaline conditions. 3. Charged or hydrophobic groups are not generated R on the matrix, so reducing nonspeci c adsorption. Intermolecular Binding Forces 4. The immobilized ligand concentration is optimal for ligate bonding. Almost all chromatographic separations rely upon 5. There is no leakage of immobilized ligand from the interaction of the target molecule with either the matrix. a liquid phase or a covalently bonded molecule on the solid phase, the exceptions being those relying upon Some ligands are intrinsically reactive (or can be molecular size, e.g. molecular sieves and gel Rltra- designed to be so) and contain groups that can be tion. In afRnity separations ligates are inevitably

Table 3 Activation materials

Activating reagent Bonding group on ligand

Cyanogen bromide Primary amines Tresyl chloride Primary amines, Tosyl chloride Primary amines, thiols Epichlorohydrin Primary amines, hydroxyls, thiols 1,4-Butanediol diglycidyl ether Primary amines, hydroxyls, thiols 1,1-Carbonyldiimidazole Primary amines, hydroxyls Cyanuric chloride Primary amines, hydroxyls Divinylsulfone Primary amines, hydroxyls 2-Fluro-1-methylpyriinium-toluene-4-sulfonate Primary amines, thiols Sodium periodate Primary amines Glutaraldehyde Primary amines Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 7

Figure 2 Triazine coupling. (A) Coupling of human serum albumin (HSA) to ready-acitivated supports as a funciton of pH. (B) Time course of coupling of human IgG to ready-activated supports at 43C. , CNBr-activated agarose 4XL; , triazine-activated agarose 4XL. complex biological macromolecules or assemblies, Fortunately there are a vast number of biological mostly or exclusively consisting of amino acids enti- ligands that can interact with more than one macro- ties linked together in a speciRc manner. This com- molecule and consequently group-speciRc ligands are plexity of structure provides many opportunities to commonplace. Since group-speciRc adsorbents retain exploit the physicochemical differences between a range of ligates with similar binding requirements, the target molecule and the ligand to be used. Each a single adsorbent may be used to purify a number of structure contains the four basic intermolecular bind- ligates. Group-speciRc ligates can be used when the ing forces } electrostatic, hydrogen bonding, hydro- desired ligate is present in high concentration, but this phobic and van der Waals interactions } spread implies that some preprocessing has taken place and throughout the structure in an exactly deRned spatial a concentration step interposed. The use of non/ manner. The degree of accessibility and spatial pre- group-speciRc adsorbents can only offer partial sentation within the pore of the medium, and the separations. This results in having to apply several strength of each force relative to each other, dictate stages in series, each only capable of removing a pro- whether these forces are utilized to effect the portion of the impurities. In contrast a highly selec- separation. The biological recognition between spe- tive ligand can exclusively remove the target in one cies is a reSection of the sum of the various mo- step, but often the resulting complexes are very tight- lecular interactions existing between them, and ly bound, have low binding capacities, are easily this summation is Rxed for the ligate. However, denatured and are expensive to produce. Until re- various ligands may be found that emulate some cently these adsorbents were restricted to technically or all of the available binding forces to various difRcult isolations. Today the use of computer- degrees. assisted molecular modelling systems provides oppor- AfRnity adsorbents are therefore assigned to tunities to investigate relationships between designed one of three broad ligand categories: nonspeciRc, ligands and relevant protein structures. For the Rrst group speciRc or highly speciRc. NonspeciRc ligands time logical design approaches can be applied and have only a superRcial likeness to biological ligands consequently stable inexpensive ligands have now and binding is usually effected by just one of the become available. four binding processes described above. Although ion exchange materials can be used in a similar manner to Analytical Scale-Up afRnity adsorbents, they only exhibit the single force of electrostatic binding. They are thus limited to Modern biotechnology uses two different types relatively indiscriminate binding. In this case the only of chromatography. Analytical separations require criterion for binding is that of an overall charge. that run time is minimized, while resolution and 8 I / AFFINITY SEPARATION / Derivatization sensitivity are maximized. In contrast, for preparative used to determine degradation of the target protein and process applications, the objective is to maximize (for example deamidated and oxidized elements); to purity, yield and economy. These techniques have identify previously unidentiRed components; to estab- developed separately, simply because biological mac- lish the chromatographic identity between recom- romolecules pose unique difRculties, making binant and natural materials; to develop orthogonal them unsuitable for ‘standardized’ analysis. A major methods for the identiRcation of unresolved impu- inSuence on this division has been that scale-up usu- rities; and for many other demanding analytical ally occurs very much earlier in the development of approaches. a process, causing biochemists to turn to the tradi- tional low performance methods of ion exchange Af\nity versus Traditional Media (IE), hydrophobic interaction (HI) and gel per- meation chromatography (GPC). The highly efR- When projects are transferred from research to devel- cient afRnity chromatography method was gener- opment two sets of chromatographic techniques are ally ignored, primarily because of the difRculty of carried forward: analysis, usually based on RP- having to develop a unique ligand for each separation HPLC; and larger scale serialized separation steps, rather than having ‘off-the shelf’ column pack- often incorporating traditional methods of ion ex- ings immediately available from external suppliers. change, hydrophobic interaction and gel permeation For analytical purposes high performance afRnity chromatography. Major decisions have to be taken at liquid chromatography (HPALC) is a rarity, a func- this juncture } to scale up the separation processes tion of the limited availability of suitable matrices developed during the research phase or to investigate (Table 2) and the afRnity process itself. alternatives. Regulatory demand and shortened pat- Where quantiRable high speed chromatography is ent lifetimes compel managements to ‘fast track’ new required, reversed-phase HPLC (RP-HPLC) has no products. Commercial pressure is at a maximum. equal. Unfortunately there is no general purpose Being Rrst to market has the highest priority in terms method for biomolecules to parallel the inherent of technical and commercial reward. Very little time power of RP-HPLC. The success of RP-HPLC for is left to explore other separation strategies. It is analysis can be judged from the large number of known that serial application of IE, HI and GPC published applications developed for the ‘Rrst wave’ inevitably leads to very high manufacturing costs, of protein pharmaceuticals manufactured by geneti- but which comes Rrst? Most often the decision is cally engineered microorganisms. These extracellular taken to begin manufacture using unoptimized separ- (relatively) low molecular weight proteins include ations as deRned in research reports. It is only in human insulin, human growth hormone and the in- retrospect that very high production costs become terferons. However, as molecular size and fragility apparent. By then it is too late } regulatory systems increase, so difRculties in using HPLC increase, are Rrmly in place. a primary reason why much analysis is still conducted There is an alternative. If researchers were more on low performance systems. Low performance sys- aware of process economics and the consequences of tems are easily scaled up; RP-HPLC is not. regulatory demand, selection of superior separation Biological separation systems must be aseptically processes could then result. Although most resear- clean throughout the process. The mixtures are inevi- chers are fully aware of the advantages of single-step tably complex and usually contain many contamina- afRnity methods, paradoxically the high selectiv- ting similarly structured species. Such species can ity advantage of afRnity chromatography is also a adsorb very strongly onto the medium, demanding weakness. Suitable off-the-shelf afRnity adsor- post-use washing with very powerful reagents to ster- bents are often unavailable, in which case an adsor- ilize and simultaneously clean the column. Silica- bent has to be custom synthesized. Since the majority based matrices cannot survive this type of treatment, of biochemists have no desire (or time) to undertake hence scale-up of analytical procedures is generally elaborate chemical synthesis, -based adsor- precluded. The Rrst wave of commercial protein bents are commonly used. However, raising suitable pharmaceuticals have generally proved to be relative- antibodies and purifying them before immobilization ly stable under high stress conditions. On the other onto a preactivated support matrix is an extremely hand intracellular proteins, often of high molecular laborious procedure. In addition, proteins are so of- weight, are unstable. Analysis by high performance ten tightly bound to the antibody that subsequent RP-HPLC methods then becomes problematic. De- involves some degree of denaturation and/or mand for fast, high resolution, analytical methods loss of acitivity. Ideal media require the incorporation will continue to increase for on-line monitoring and of elements of both nonselective and selective adsor- process validation. Such techniques have already been bents to provide adsorbents with a general applicabil- Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 9 ity. If stable, highly selective and inexpensive af- a simple synthesized entity or a high molecular weight Rnity ligands were available, then opportunities would protein. The afRnity technique is theoretically of exist for researchers to develop efRcient high yield universal application and any protein can be separ- separations even in the earliest phases of investigation. ated whatever its structure and origin. As always, These systems could then be passed forward to pro- there are major limitations. The most effective duction with the knowledge that optimally efR- afRnity ligands are other proteins. Unfortunately cient separations are immediately achievable. such proteins are difRcult to Rnd, identify, isolate Production costs of any pure material reSect the and purify. This results in high costs. An even greater absolute purity level required and the difRculty of deterrent is that most proteins are chemically, cata- achieving it. Therapeutic proteins have high purity lytically and enzymically unstable, a particularly un- requirements and the larger the administered dose, attractive feature if they are to be used for the manu- the purer it has to be. Since many protein pharma- facture of therapeutic substances; and regulatory ceuticals will be used at high dose levels, purities need authorities generally reject applications using pro- to exceed 99%, occasionally up to 99.999%. That teinaceous ligands. these purities can be met by traditional methods is In anticipating that one day stable inexpensive af- possible, but it is widely documented that the applica- Rnity media would be in demand, a team led by C.R. tion of such methods massively increases production Lowe began an investigation into which synthetic costs. Between 50 and 80% of total production costs ligand structures offered the greatest possibility of therapeutic proteins are incurred at the puriRcation of developing inexpensive stable ligands. It was con- stage. The manufacturing cost of a product is directly cluded that structures that could be manipulated into related to its concentration in the mother liquor; speciRc spatial geometries and to which intermolecu- the more dilute it is, the higher the cost of recov- lar binding forces could easily be added offered ery. Since traditional puriRcation processes on their the highest chance of success. Model compounds own cannot selectively concentrate a target protein to were already available; the textile dyes. the exclusion of all others, they have to be used in series. The number of stages required can vary be- Synthetic Ligands tween four and 15. Each step represents a yield loss, and incurs a processing cost. Yields of less than 20% Textile dyes had already proved to be suitable ligands are not uncommon. Figure 3 shows an puriR- for protein separations. Blood proteins, dehydrogen- ed in multiple stages and by a one-step afRnity ases, kinases, oxidases, proteases, nucleases, transfer- process. ases and ligases can be puriRed by a wide variety of It was these limitations that caused biochemists to dyes. However, they did not prove to be the break- examine highly selective ligands. Almost any com- through so eagerly awaited. An essential feature of all pound can be used as an afRnity ligand provided chromatographic processes is exact repeatability it can be chemically bonded onto a support matrix from column to column, year after year. Textile dyes and, once immobilized, it retains its ability to interact are bulk chemicals, most of which contain many with the protein to be puriRed. The ligand can be by-products, co-produced at every stage of the dye

Figure 3 Comparison of multistep versus affinity separation. 10 I / AFFINITY SEPARATION / Derivatization manufacturing process. This fact alone makes repro- ducibility problematic. Furthermore, the bonding process between dye and matrix was poorly re- searched. This resulted in extensive leakage. All com- mercially available textile dye products leak exten- sively, especially under depyrogenating conditions (Figure 4). Despite these limitations, it was recog- nized that dye-like structures had a powerful underly- ing ability to separate a very diverse range of proteins. Their relatively complex chemical structures allow spatial manipulation of their basic skeletons into an inRnite variety of shapes and conRgurations. Pro- teins are complex three-dimensional (3-D) structures and folds are present throughout all protein struc- tures. An effective ligand needs to be shaped in Figure 5 Schematic representtion of ligand}protein interaction. such a manner that it allows deep insertion into W, electrostatic interaction; X, hydrogen bonding; Y, van der a suitable surface Rssure existing within the 3-D Waals interaction; Z, hydrophobic interaction. ***, original structure (Figure 5). In contrast, if the ligand only backbone; - - -, new structure added; 2, original backbone interacts with groups existing on external surfaces, move; , fields of interaction. then nonspeciRc binding results and proteins other than the target are also adsorbed. A much more selective approach is to attempt to insert a ligand into ing; Y, van der Waals; Z, hydrophobic) align with the an appropriate fold of the protein, and add binding binding areas in the protein fold, idealized afRn- groups to correspond with those present in a fold of ity reagents result. The use of spacer arms minimizes the protein. If all four of the basic intermolecular steric hindrance between the carrying matrix and forces (Figure 5: W, electrostatic; X, hydrogen bond- protein.

Figure 4 Leakage of blue dye from various commercial products. , 0.1 mol L\1 NaOH; , 0.25 mol L\1 NaOH; , 1 mol L\1 NaOH. Key: A, Mimetic Blue 1 A6XL (affinity chromatography); B, Affi-Gel Blue (Bio-Rad); C, Blue Trisacryl-M (IBF); D, Fractogel TSK AF-Blue (Merck); E, C.I. Reactive Blue 2 polyvinyl alcohol-coated perfluropolymer support; F, Blue Sepharose CL-6B (Pharmacia); G, immobilized Cibacron Blue F3G-A (Pierce); H, Cibacron Blue F3G-A"Si500 (Serva); I, Reactive Blue 2-Sepharose CL-6B (Sigma). Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 11

The Rnal step is to design appropriate bonding purities, substantial increases in column lifetime, and technologies to minimize potential leakage. Until re- improvements in batch-to-batch reproducibility. cently this type of modelling was a purely theoretical exercise. It was only the introduction of computer- assisted molecular modelling techniques that allowed Rational Design of Af\nity Ligands the theory to be tested. Before the arrival of logical Modi\cation of Existing Structures modellling the discovery of selective ligands was en- tirely based upon empirical observation, later fol- The Rrst example of a rational design of new bio- lowed by a combination of observation, experience mimetic dyes used the interaction between horse and limited assistance from early computer generated liver alcohol dehydrogenase (ADH) and analogues of models. Although several novel structures evolved the textile dye Cibacron Blue F3G-A (Figure 7). It during this period, a general approach to the design of had been established that the parent dye binds in the new structures remained elusive. At this time only NAD#-binding site of the enzyme, with the an- very few 3-D protein structures were available, again thraquinone, diaminobenzene sulfonate and triazine greatly restricting application of rational design ap- rings (rings A, B and C, respectively, in Figure 7) proaches. As more sophisticated programmes, simu- apparently adopting similar positions to those of the lation techniques, protein fragment data and many adenine adenosine ribose and pyrophosphate groups more protein structures were released, logical design of NAD\. The anthraquinone ring (A) binds in methods were revolutionized. However, many mil- a wide apolar fold that constitutes, at one end, the lions of proteins are involved in life processes, and it adenine bridging site, while the bridging ring (B) is is clear that many years will elapse before the major- positioned such that its sulfonate group interacts with ity of these will be fully described by accurate models. the guanidinium side chain of Arg271 (Figure 8). Consequently intuition and experience will continue Ring C binds close to where the pyrophosphate to play a major role in the design of suitable ligands. bridge of the coenzyme binds with the reactive Of available rationally designed synthetic molecules, triazinyl chlorine adjacent to the nicotinamide ribose- the Mimetic2+ range can currently separate over 50% binding site. The terminal ring (D) appears to be of a randomly selected range of proteins. Stability bound in a fold between the catalytic and coenzyme under depyrogenating conditions has been demon- binding domains, with a possible interaction of the strated for these products (Figure 6). This results in sulfonate with the side chain of Arg369. The binding minimal contamination from ligand and matrix im- of dye to horse liver ADH resembles ADP binding but differs signiRcantly at the nicotinamide end of the molecule with the mid-point position of ring D dis- placed from the mid-point position of the nicotinamide ring of NAD# by about 1 nm. Consequently a number of terminal-ring analogues of the dye were synthesized and characterized in an attempt to improve the speciR- city of dye binding to the enzyme. Table 4 lists some of the analogues made by substituting }RintheDring (Figure 7), together with their dissociation constants. These data show that small substituents bind more tightly than bulkier groups, especially if substituted in the o-orm-positions with a neutral or anionic group. Further inspection of the computer model given as Figure 8 showed that the dye analogues were too short andrigidtobindtohorseliverADHinanidentical manner to the natural coenzyme, NAD#.Conse- quently analogues of the parent dye were designed and synthesized with central spacer functionalities to increase the length and Sexibility of the molecule (Figure 9). This product proved to be some 10 times superior to any previously synthesized compound. This work provided the Rrst proof that rationally Figure 6 Comparison of ligand leakage from mimetic ligand designed molecules could be converted into stable, affinity adsorbent A6XL ( ) and conventional textile dye agarose inexpensive, chromatographic media, while providing ( ). the most remarkable separations. 12 I / AFFINITY SEPARATION / Derivatization

Figure 7 Principal structural elements of the anthraquinone dye, Cibacron Blue F3G-A.

De Novo Design mimic the binding of naturally occurring anionic het- erocycles such as NAD#, NAPD#, ATP, coenzyme Most early efforts in proving the rational design A, folate, pyridoxal phosphate, oligonucleotides technology was based upon dye structures. To date and polynucleotides. However, some proteins, all dyes considered have been anionic, presumably particularly proteolytic enzymes, interact with cationic because the charged chromophores of these ligands substrates. The trypsin-like family of enzymes forms

Figure 8 Putative binding pocket for the terminal-ring analogue (m-COO\) of Cibacron Blue F3G-A in the coenzyme binding site of horse liver alcohol dehydrogenase (ADH). The site lies lateral to the main coenzyme binding site and comprises the side chains of two juxtaposed cationic residues Arg47 and His51. Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 13

Table 4 Apparent affinities of terminal-ring analogues of an- by insertion of a hexamethylene spacer arm between thraquinone dyes for horse liver alcohol dehydrogenase (ADH) the designed ligand and the matrix. After synthesis of this medium it was demonstrated that puriRed pan- creatic kallikrein was strongly bound, with over 90% of activity being recovered on elution with 4- aminobenzamidine, whereas trypsin appeared largely in the void of the column. This medium was able to purify kallikrein 110-fold from a crude pancreatic acetone powder in a single step. There is an alternative to the rational design ap- proach } the use of combinatorial libraries.

1 R Apparent Kd (mol L\ ) Combinatorial Libraries m-COO\ 0.06 The driving force behind the development of combi- H0.2 natorial libraries has been the many failed attempts o-COO\ 0.2 to design therapeutic substances using theoretical \ o-SO3 0.4 knowledge allied to rational design; very few such \ m-SO3 1.6 approaches succeeded. In contrast, combinatorial li- m-CH OH 4.5 2 brary design is now thought by some to provide the m-CONH2 5.7 p-COO\ 5.9 best opportunity of discovering new novel peptides p-SO3\ 9.3 and small molecule structures for pharmaceutical ap- 2\ p-PO3 10.5 plication. A quite natural extension of the concept # p-N (CH3)3 172.0 is to use combinatorial libraries to discover ligands capable of achieving highly efRcient protein sep- arations. When directed at drug discovery the earliest one of the largest groups of enzymes requiring workers built libraries from peptides. For ligands cationic substrates and includes enzymes involved in libraries will generally utilize simple chemical mol- digestion (trypsin); blood clotting (kallikrein, throm- ecules and occasionally smaller peptides. To distin- bin, Factor Xa); Rbrinolysis (urokinase, tissue plas- guish this subsection from the earlier methods a minogen activator) and complement Rxation. These convenient designation is the term Chemical Combi- enzymes possess similar catalytic mechanisms and natorial Library (CCL)2+. bind the side chains of lysine or arginine in a primary Although procedures for rationally designed pocket proximal to the reactive serine (Ser195), with ligands are well established, the newness of CCL speciRcity being determined partly by the side chain suggests that CCL design of afRnity ligands of Asp189 lying at the bottom of the pocket, and should be regarded as embryonic rather than immedi- partly by the ability of the individual enzymes to form ately available for commercial application. There are secondary interactions with the side chains of other thus two diametrically opposed systems } the rational nearnearby substrate amino acids. For example, tis- design process, based on logic, experience and know- sue kallikrein differs from pancreatic trypsin in ledge; and CCL which is illogical and completely that it displays a marked preference for phenylalanine random. One description of CCL is ‘a method of in the secondary site, probably because the phenyl increasing the size of the haystack in which to Rnd ring on the phenylalanine residue neatly slips into your needle’. A very recent approach is to combine a hydrophobic wedge-shaped pocket between the both rational and CCL techniques, a process termed aromatic side chains of residues Trp215 and Tyr99 ‘intelligent’ combinational design. At the time of (Figure 10). SpeciRcity for the secondary amino acid writing there are no published examples of ligands residue is less stringent in trypsin since Tyr99 is re- derived from CCL, although patents have been Rled placed by Ala99 and the hydrophobic pocket cannot in this area. be formed. By designing a mimic for the Ph}Arg R dipeptide should result in a speci city for kallikrein. Regulations and Drug Master Files Figure 11 uses p-aminobenzamidine and phene- thylamine functions substituted on a monochloro- For researchers the relevance of regulations often triazine moiety. However, the active site of pancreatic seems remote, and yet the decisions taken in even the kallikrein lies in a depression in the surface of the earliest stages of research can take on a great signiR- enzyme. The expected steric hindrance is eliminated cance if the target product becomes a commercial 14 I / AFFINITY SEPARATION / Derivatization

Figure 9 Horse liver alcohol dehydrogenase separation. Using the modified Cibacron Blue F3G-A (Figure 7) structure given above, the selectivity is greatly enhanced making it possible to separate the isoenzymes.

reality. It is regrettable that many researchers slavish- deRnition and suppliers, stability data for every step ly follow previously published data on a given separ- of the process, formulation methods, packaging, ation problem without giving thought to the longer- labelling, toxicity data and so on. The documentation term implications of their decisions. Sections above has to be revised annually and any changes notiRed. describe the adverse economic effects of multi- Furthermore plant and process is open to inspection stage processing, but an equally important factor is at all times for full audit of procedure. There is one regulatory issues. The most widely used regulations large anomaly within the regulations. The largest are those deRned by the USA’s Food and Drug Ad- volume of material in contact with a drug during ministration (FDA). Any company wishing to import manufacture is water, solvents and salts, all of which products relevant to regulations existing in the USA are exactly deRned in terms of their physicochemical must conform exactly to FDA requirements; drugs in characteristics. The next largest is chromatographic particular are very strictly controlled. Detailed de- media; ambiguously media do not have to be de- scriptions of any plant and process used in drug scribed in the same detail. manufacture have to be lodged in documented form The outstanding stability of the synthetic with the FDA, wherein every aspect of process de- chromatography media provides an excellent oppor- scription is given. This must include raw material tunity to develop and register Drug Master Files Sepsci*1*TSK*Venkatachala=BG I / AFFINITY SEPARATION 15

Figure 10 Model of the Phe}Arg dipeptidyl substrate bound in the acitve site of porcine pancreatic kallikrein. The illustration shows Asp189 at the bottom of the primary binding pocket as well as the side chains of Tyr99 and Try215, which form the secondary binding pocket, with the phenyl ring of the Phe residue sandwiched between the hydrophobic side chains of these residues.

(DMFs) with the FDA. DMFs allows companies to standards and Good Manufacturing Practice (GMP) use synthesized ligands for very high purity protein is an integral part of a DMF. Few researchers select- pharmaceuticals with total conRdence. New Drug ing a speciRc medium consider the long-term implica- Applications (NDA) and Investigational New Drugs tions of stability under depyrogenating conditions, (IND) documents incorporating such stable afRn- the number of cycles that can be achieved (lifetime in ity media can now be submitted to the FDA, safe in use), its availability in bulk, whether it is manufac- the knowledge that all appropriate information is on tured under aseptic conditions and the price when Rle. The effective guarantee of minimum quality supplied in bulk. If the researcher makes a good initial selection, the research data produced can be utilized in development phases with conRdence. ‘Fast tracking’ is facilitated, with minimum aggravation, maximum efRciency and minimum puriRcation costs.

Alternative Af\nity Approaches Aqueous two-phase systems have been extensively applied to bimolecular puriRcations, by attaching af- Rnity ligands to one of a pair of phase-forming poly- mers, a method known as afTnity partitioning. Although a substantial body of research literature is available, few systems appear to have been adopted for commercial purposes. Reactive dyes, with their simple and well-deRned coupling chemistries, have Figure 11 Comparison of the structures of (A) the Phe}Arg generally been favoured as the active ligand. The dipeptide, and (B) the ‘biomimetic’ ligand designed to bind at the advantage of afRnity partitioning is that the pro- active site of the porcine pancreatic kallikrein. cess is less diffusion-controlled, binding capacities 16 I / AFFINITY SEPARATION / Derivatization are high and the recovery of bound proteins is Af\nity Membranes easier, created by the process operating with fewer theoretical plates than those generated by chromatog- UltraRltration membranes are commonly employed raphy columns. This technique has also been com- as a ‘polishing’ stage of multistage separation pro- bined with afTnity precipitation, where a homo- cesses for several commercially important proteins. bifunctional ligand composed of two ligand entities Consequently attaching standard afRnity ligands connected by a spacer (for example a bis-dye) is used. to create afRnity membranes has become an ac- However, even in combination this approach suf- tively researched area. The most obvious advantage fers from considerable nonspeciRc binding and rela- of a membrane structure is the high rate of transport tively low puriRcation factors. A review of this com- of the medium through the porous structure by Rltra- bination suggests that it is more suited to large scale, tion, thus minimizing the normally encountered dif- low purity products. In contrast, perSuorocarbon fusion limitations of mass transfer. High adsorption emulsion chemistry utilizing mixer-settlers may of- rates are achieved, especially if long distance electro- fer more promise. By using a series of mixer-settlers static interactions are involved in the binding mech- connected in a loop a continuous process has been anism. However, in contrast to ion exchange developed. A ligand (usually a reactive dye) is membranes, similar high transport effects are not covalently bonded to a high density perSuorocarbon observed when used in the afRnity mode, elimin- emulsion and contacted with the crude protein solu- ating much of the initial attraction of this form of tion. After settling in the Rrst tank the emulsion is device. Other theoretically attractive features in- pumped to a second settler and washed before passing cluded: an inherent ability to control pore size across to the third settler for elution. The emulsion is regen- a very wide range, offering an opportunity to erated in the fourth settler. The supernatants from increase capacity of a given system; and ability to each settler, still containing some unbound target operate in either batch mode or Rltration mode. In protein, are normally discarded. Although reasonable both cases experimental data have not conRrmed these recoveries and yields are obtained, signiRcant devel- assumed advantages; a 10-fold change in the pore size opment is needed for this type of system to become resulted in only a two-fold capacity increase and when competitive with conventional chromatography col- in Rltration mode, although adsorption is fast, severe umn methods. peak broadening on elution is experienced. Another favoured research approach to improving The chemistry relevant to particulate media is iden- efRciency is to use expanded beds. Various tech- tical to that required for membranes, in effect niques have been tried, with the primary objective of making the systems compatible and allowing an easy eliminating the ‘solid bed’ effect, where the bed technology interchange. The covalent bonding of af- acts as a Rlter, trapping insolubles and creating signif- Rnity ligands to the surface of a membrane follows icant back-pressure. By partially removing the normal exactly the same chemistry as that applied to partic- constraints of upper and lower retaining frits, which ulate media, and the same adsorption/desorption pack the particles tightly in the bed, the particles can principles apply to both. Consequently the only dif- expand, thereby releasing trapped solid impurities. ference between membrane systems and those of con- Consequently longer operational cycles and higher ventional chromatography is the exploitation of the Sows result. One limitation of the expanded bed characteristics of the membrane matrix compared system is that adsorption can only be carried out in with a particulate bed. Although the high mechanical one stage, resulting in a less efRcient process. strength of membranes is one major advantage, plus Expanded beds are only an intermediate stage to- the scale-up is claimed to be very easy by stacking wards Uuidized beds. Several variations of Suidized membranes (although scaling afRnity columns is bed technology have been adopted to evaluate them also very straightforward), it has been discovered that for afRnity processing. One example is the use of if the pressure drop across the membrane is too high perSuorocarbon emulsions in a countercurrent con- sealing problems occur; the mobile phase then Sows tactor. The afRnity perSuorocarbon emulsion is beyond the edges and past the membranes. Further- loaded with crude source material into the base of more afRnity membranes should be capable of a column in a similar manner to that of an expanded use with unclariRed extracts, but is has been generally bed. The adsorbent is then removed from the base of observed that membrane capacity and lifetime are the bed and carried forward through four identical progressively reduced with time of use. Even with contactors where washing, elution and regeneration clariRed broths, membrane fouling regularly occurs. are carried out successively. This process is claimed to This is almost certainly the reason why afRnity improve signiRcantly removal of target proteins com- membranes have not found favour in large scale pared to an expanded bed system. processing. Sepsci*11*TSK*Venkatachala=BG I / CENTRIFUGATION 17

Conclusion icals. A mandatory part of any new protein pharma- ceutical process is the acceptance by regulatory Protein separations can be achieved by a variety of authorities of the separation process involved. That R af nity techniques, but separations in the synthesized afRnity ligand separation processes chromatography mode are by far the most widely have now been fully accepted by the foremost regula- R used. Nature de ned an appropriate pathway to highly tory authority, the USA’s Food and Drug Administra- R } ef cient separation utilization of the phenom- tion, conRrms a worldwide acceptance of the power enon of the automatic recognition mechanism existing of ligand design technologies. between a given protein and at least one other. By covalently bonding one of the pair onto an inert matrix See Colour Plate 1. a theoretically simple separation process can be de- vised. Although these immunoafRnity separations Further Reading are widely practised today, severe limitations exist, not least of which are cost and instability of the afRn- Briefs K-G and Kula M-R (1900) Fast protein chromatogra- R ity medium when in use. As modern design aids have phy on analytical and preparative scale using modi ed micro-porous membranes. Chemical Engineering become commonplace, in conjunction with newer Science 47: 141}149. techniques such as the development of combinatorial Burton SJ, Stead CV and Lowe CR (1988) Design and library arrays, it has proved possible to mimic nature applications of biomimetic anthraquinone dyes. Journal and replace immunoafRnity matrices by speciR- of Chromatography 455: 201}216. cally designed synthetic ligands. These new ligands not Chase HA (1994) PuriRcation of proteins using expanded only accurately emulate the exquisite precision of the beds. Trends in Biotechnology 12: 296}305. natural protein}protein interaction mechanisms, but Dean PDG, Johnson WS and Middle FA (1985) AfTnity also provide the opportunity to manipulate the ligand Chromatography: A Practical Approach. Oxford: IRL structures, thus offering far more efRcient Press. R separations than any previously achieved. For a given Jones K (1990) A review of af nity chromatography. } protein, from whatever source and at any dilution, it is Chromatographia 32: 469 480. Kenny A and Fowell S (1990) Methods in Molecular Biol- now possible virtually to guarantee that a highly cost- R ogy: Practical Protein Chemistry. New York: Humana effective and highly ef cient separation pro- Press. cess can be developed for eventual commercial use. Kopperschlager G (1994) AfRnity extraction with dye- Designed ligand processes have already been ad- ligands. Methods in Enzymology 228: 121}129. opted for several very large biotechnology projects Walker JM and Gaastra W (1987) Techniques in Molecular scheduled to manufacture bulk protein pharmaceut- Biology. London: Croom Helm.

CENTRIFUGATION

D. N. Taulbee and M. Mercedes Maroto-Valer, has been employed for well over 100 years. Applica- University of Kentucky-Center for Applied Energy tions that range from the mundane, industrial-scale Research, Lexington, KY, USA dewatering of coal Rnes to the provision of an invalu- Copyright ^ 2000 Academic Press able tool for biomedical research. The Rrst scientiRc studies conducted by Knight in Introduction 1806 reported the differences in orientation of roots and stems of seedlings when placed in a rotating Centrifugation is a mechanical process that utilizes an wheel. However, it was not until some 60 years later applied centrifugal force Reld to separate the compo- that centrifuges were Rrst used in industrial applica- nents of a mixture according to density and/or tions. The Rrst continuous centrifuge, designed in particle size. The principles that govern particle be- 1878 by the Swedish inventor De Laval to separate haviour during centrifugation are intuitively compre- cream from milk, opened the door to a broad range of hensible. This may, in part, explain why centrifu- industrial applications. About this same time, the Rrst gation is seldom a part of post-secondary science centrifuges containing small test tubes appeared. curricula despite the broad range of scientiRc, medical These were modest, hand-operated units that attained and industrial applications in which this technique speeds up to 3000 rpm. The Rrst electrically driven