Matthews and Wells [ 183] have presented an interesting variation on the normal selection protocol, designated 'substrate phage'. They displayed human growth hormone (hGH) on phage linked to pIII with a random linker. This library is bound to immobilized hGH-receptor. Elution was carried out by treatment with a protease which normally did not cleave the phage or hybrid protein. The subset of clones enriched by cleavage in the linker thus presents a consensus for the preferential protease substrates. This method has also been used to characterize the preferential cleavage site of furin, a mammalian enzyme involved in the cleavage of many constitutively expressed protein precursors .
Vectors constructed with such proteolytic linkers can thus be released from their target during panning with a mild buffer containing the appropriate protease, without having to guess at optimal elution conditions (see Ref. 85; thrombin cleavage for elution).
The following examples use competition during panning to select for enhanced specificity for a particular target. Cunningham et al.  selected an A-type-ANP receptor-specific agonist (atrial natriuretic peptide analogue or variant) using competition panning with soluble receptor B while selecting on receptor A. Dennis and Lazarus [I85] selected APPI (Alzheimer's P-amyloid precursor protein inhibitor) variants specific for blood-clotting factor VIIa which bound more weakly, or not at all, to FXa. They used competitive selection with FXIa, plasmin and kallikrein in the panning buffer to prevent enrichment for those with specificity for these proteases. An overall consensus inhibitor (TF7I-C), previously isolated, was a 1.9 nM inhibitor for FVII, but a 0.8 nM inhibitor of FXIa, 1.2 nM for kal-likrein and 40 nM for plasmin. The competitive enrichment amongst variants of this latter inhibitor from a library in which nine positions were very partially randomized (ca. 360 000 variants) yielded inhibitors with a single change. Clear effects ofcounterselection were seen only in one of four banks. The final inhibitor constant, although all worse than TF7I-C for FVIIa, was found to be a 100-fold more selective relative to FXIa and kallikrein.
An alternative to negative selection has been presented by Stausbol-Gran et al. , who demonstrated that the addition of control proteins during the adsorption and washing steps during panning can guide the specificity of the selection process towards a target which is absent in the control mixture. It is an interesting possibility that such a protocol would help specific enrichment in selection of disease-specific phagotopes as described above, whereby control patient sera would be added as competitor during panning. It is, of course, questionable as to whether or not the concentrations of the individual antibodies in the serum are high enough to produce a significant effect.
SH3-ligands and mirror-imaging with D-peptide ligands
Peptide ligands which bind Src-homology type 3 domains (SH3) have been investigated by a number of groups using both synthetic and phage-display systems, subsequent to Baltimore's group  inferring that a proline-rich sequence was involved in the interaction. Screening on SH3 has shown nonconstrained proline-rich core sequences [49,135-137], where the specificity for different SH3-domains lies mainly in adjacent sequences . This latter paper is, in fact, a major 'tour de force', in which the consensus specific for a very wide range of SH3-domains is defined by phage display with a very large dedicated peptide library presenting (X)ePXXP(X)e. Binding affinities of peptides, based on the motifs identified, confirm results with phage. Clones of type II (bind in the opposite orientation ), which showed a broader binding specificity amongst SH3-domains, were used to screen mouse and human cDNA expression libraries, selecting 18 genes, of which 10 were novel .
One can compare these phage-display experiments with the synthetic peptide and rational design approach for Abl-SH3 binding clones. Pisabarro and Serrano  varied the decapeptide APTYPPPLPP substituting single positions, one at a time with two to four different amino acids (a total of 29 peptides). Binding varied between twofold better and 20 times worse, e.g. the latter particularly when the prolines were exchanged. Combining favourable substitutions yielded a 0.4 pM binder, APTYSPPPPP. In the model, the long Pro-run is favourable due to reduction in entropic freedom. Although the consensus and/or optimal sequences arrived at with the two techniques are very similar, it seems to me that the relevant biological information obtained, relative to the effort involved, greatly favours the empirical approach.
The potential therapeutic value of targetting the signal transduction pathway has recently been demonstrated with specific Jak2-kinase inhibitors in an animal leukaemia model . Since these pathways require that SH3 and SH2 interact with their appropriate ligands, this binding is also of interest as a target for therapeutic design. The development of ligands with clinical potential obviates the direct use of short peptides, which are easily degraded in vivo, either in the serum or in the cell cytoplasm. In a dramatic demonstration of a novel strategy, Schumacher et al.  have developed D-amino acid poly-peptide ligands, which should be resistant to proteolysis, but which, because of their small size, easily diffuse to and into target organs/tumour tissue. They used synthetic enantio-meric Src-SH3 domain, composed entirely of D-amino acids, as the target during panning, With a type 3 phage library presenting a disulphide constrained C(X)UC loop they identified binding epitopes as L-amino acid ligands. Interestingly, the synthetic D-aa enantiomer ligands bind to the natural L-protein. The D-forms of CLSGLRLGLVPC, CLMGLRLGL-LPC, CAYGFKLGLIKC, with a consensus containing a long hydrophobic region interrupted by an arginine or lysine, bind normal Src-SH3. The binding is dependent on the constrained structure, i.e. on disulphide bond formation, with the best binding clones having a moderate affinity (Kd of 63 pM). NMR was used to investigate the binding site of the D-peptide on SH3. It fits into the shallow groove used for binding the natural ligand, which is in agreement with the competitive binding with natural ligands. Modelling implied that the conserved arginine/lysine residue binds in pocket-A, as does the conserved arginine in the normal substrate, but the portion of the cleft binding the proline residues is not involved in strong interactions with the D-peptide. The authors consider that D-enantiomer synthesis is feasible by chemical synthesis for domains of up to 100 aa, and as such could be applicable for screening with any of the numerous small domains which occur in natural proteins (e.g. WW, PHD, SH2, etc.; Ref. 192). This approach is, of course, not confined to screening with phage-display libraries.
A knowledge of disease-related antigens in autoimmune disease is often lacking. A resolution of this problem would simplify the search for the early immune response antigens correlated with the initiation of the disease, or at the worst provide a diagnostic tool to assess the disease status. In terms of vaccination and the development of an immune competent status subsequent to either infection or vaccination, there is also the need for diagnosis of the immune response against protective antigens. The aim in both cases is to select a collection of antigens, which can be used to recognize the presence of antibodies against specific epitopes in the serum (or cerebrospinal fluid), where, initially, the antigens are unknown. Although faced with the apparently insurmountable problem ofchallenging 108 to 109 phage variants with perhaps a million different antibodies present in the serum, a number of groups have attempted to carry out just such a selection.
One of the first studies, Dybwad et al.  used panning of a type 8 nonapeptide library, alternating positive selection on pooled sera from a number of rheumatoid arthritis (RA) patients, and negative selection (see reservations about the effectivity of such an approach, above) on sera from a control group. The experiment was moderately successful, in that the enriched phage population contained five unrelated clones which could react with 44% of sera from RA patients, whereas they reacted with only 13% of sera from the control group.
Cortese's group [7,60,61,115,194,195] have developed a different screening approach. In view of the concentration problems involved during the initial panning step, the high avidity of multivalent type 8, or type 8+8 vectors (see Refs. 114 and 196) is considered essential to the success of the enterprise, as in the previous example given. Pools of positive sera are immobilized on beads for the panning, and pools of both positive and negative sera are used for screening gridded blots of up to 100000 clones selected during the panning. The method has been applied to prediabetic and high risk insulin dependent diabetes mellitus patient sera (Mennuni et al., in preparation, quoted in Ref. 195; preliminary results show significantly higher reactivity of some clones for patient compared to control sera), cold, insoluble IgM-complex fractions from type II cryoglobulinemia patients (CryoII; Macchini et al., in preparation, cited in Ref. 195) and the cerebrospinal fluid of multiple sclerosis patients . In all three examples, phage (termed phagotopes) were isolated which could interact specifically with antibodies from patient sera or, in the latter case, with high titre to patient cerebrospinal fluid (CSF). In the case of the CSF-specific clones, three consensus motifs were found. However, the clones isolated bound to antibodies from both the multiple sclerosis and the control groups. The association with the disease state, although not implying an etiological relationship, was indicated by the finding that the antibodies binding phagotopes were present at a much higher titre in the CSF of a few individuals from the MS patient group. In the case of the CryoII study, phagotopes showed a motif which implied that the dominant antigen recognized by the disease-specific IgMs was the T-lymphocyte and natural killer cell surface antigen LAG-3. This was confirmed by further experiments.
The same serum panning and selection strategies have been used to isolate phagotopes recognized by antibodies specifically present in hepatitis B virus- (HBV; Ref. 115) and hepatitis C virus- (HCV; Ref. 195) immune sera. In the first study two different clones were found, one of which was similar to the natural viral antigen sequence. At least one of the two antigens were recognized by 80% of sera from immunized individuals and by 40% of sera from HBV-infected patients. In the case of the HCV study a whole panel of (at least 40) phagotope clones were selected (panned and positively and negatively screened, with HCV-infected patient sera and control sera) which reacted only with pooled patient sera. Although any one clone reacted with serum antibodies in only half of the patient samples, the total collection gave a 'fingerprint' of the immune status of the individuals, with close to 100% identification of positive sera and zero cross-reaction with negative control sera. This is a very encouraging result which should be extremely useful in the development of diagnostic kits.
The isolation of disease-related antibodies specific for either melanoma [ 180], the autoimmune thyroid peroxidase antigen [ 174], or neutrophil antigen characteristic of ulcerative colitis  from nonnaive patient repertoire phage libraries has been reviewed above in the 'Panning on cells' section. The isolation of anti-viral antibodies with diagnostic or therapeutic potential from antibody repertoire libraries has been described in earlier papers: e.g. against HIV , hepatitis B virus  and human respiratory syncytial virus .
The use of phage-display libraries to characterize a number of epitopes recognized by polyclonal autoimmune antibodies has also been reported for affinity purified anti-TNFa autoantibodies . A number of papers have also pointed out that multivalent type 8 phage are extremely good immunogens [24,201,202]. As such they have been proposed as potential vaccines.
Selecting human antibodies directed to a particular epitope
Jespers et al.  present a protocol whereby, using phage display and starting from a mouse monoclonal antibody to a particular epitope, a completely human antibody can be derived to the same epitope. The mouse light and heavy chains are cloned from the hybridoma. A phage library presenting human light chains is infected into an E. coli host expressing the mouse monoclonal heavy chain. Panning on the original antigen selects hybrid Fab directed to the same antigenic epitope. A host strain, expressing the human light chain, is infected with a phage library of human heavy chains. Panning of the phage produced, which are now displaying completely human Fab fragments, on the original antigen allowed selection of human antibodies directed to the same epitope that is recognized by the original mouse antibody. In this manner human anti-amino-terminal TNFa antibodies were produced with 15 nM dissociation constants to human TNFa This differs from previous protocols [204-207] to 'humanize' antibodies in that (i) it is a straightforward approach, not requiring a knowledge of the structure of the antibody; and (ii) the final antibody is of completely human origin.
Was this article helpful?