Diversity and censoring of landscape phage libraries

Abstract

Libraries of random peptides displayed on the surface of filamentous phages are a valuable source for biospecific ligands. However, their successful use can be hindered by a disproportionate representation of different phage clones and fluctuation of their composition that arises during phage reproduction, which have potential to affect efficiency of selection of clones with an optimal binding. Therefore, there is a need to develop phage display libraries with extended and varied repertoires of displayed peptides. In this work, we compared the complexity, evolution and representation of two phage display libraries displaying foreign octamers and nonamers in 4000 copies as the N-terminal part of the major coat protein pVIII of phage fd–tet (landscape libraries). They were obtained by replacement of amino acids 2–4 and 2–5 of pVIII with random octa- and nonamers, respectively. Statistical analysis of the libraries revealed their dramatic censoring and evolution during amplification. Further, a survey of both libraries for clones that bind common selectors revealed the presence of different non-overlapping families of target-specific clones in each library justifying the concept that different landscape libraries cover different areas of a sequence space.

Introduction

Ff class of filamentous phage includes three strains: f1 (Loeb, 1960), M13 (Hofschneider, 1963) and fd (Marvin and Hoffman-Berling, 1963). The wild-type phages are thread-like particles of ∼1 µm long and 7 nm in diameter. The bulk of their tubular capsid consists of 2700 identical, largely α-helical subunits of the 50-residue major coat protein pVIII arranged in an array with 5-fold rotational and 2-fold screw symmetries. About half of the pVIII protein’s amino acids are exposed to the solvent, the other half being buried in the capsid (Marvin, 1998; Marvin et al., 2006). One tip of the phage outer tube is capped with five copies each of minor coat proteins pVII and pIX and another tip with minor coat proteins pIII and pVI. Viral DNA of varying sizes, including recombinant genomes with foreign DNA inserts, can be accommodated in the filamentous capsid whose length is altered to match the size of the enclosed DNA by adding proportionally fewer or more pVIII subunits during phage assembly (Hunter et al., 1987).

In phage display constructions (here and after, the terms requiring a definition are italicized) foreign coding sequences are spliced in-frame into one of the five phage coat protein genes, so that the ‘guest’ peptide, encoded by that sequence, is fused to the coat protein and thereby displayed on the surface of the virion (reviewed in Smith and Petrenko, 1997). A phage display library is a collection of such fusion phage clones, each harboring a different foreign coding sequence, and therefore displaying a different guest peptide on their surface. When a foreign coding sequence is spliced into the major coat protein’s gene, gpVIII, the guest peptide is displayed on every pVIII subunit (Ilyichev et al., 1989; Felici et al., 1991; Greenwood et al., 1991) increasing the virion’s total mass by up to 20% (Iannolo et al., 1995). Such particles were eventually given the name ‘landscape phage’ to emphasize the dramatic change in surface architecture caused by arraying thousands of copies of the foreign peptide in a dense, repeating pattern around the viral capsid. A landscape library is a large population of such phages, encompassing billions of clones with different surface structures and biophysical properties (Petrenko et al., 1996; Iannolo et al., 1997; Legendre and Fastrez, 2002; Petrenko et al., 2002).

Different applications of phage display strategy in biomedicine and nanotechnology have promoted a fast growing interest in landscape phages as a new selectable nanomaterial. Depending upon the particular foreign peptide sequence inserted, the resulting landscape phage can bind organic and inorganic compounds, proteins and antibodies, induce specific immune responses in animals or resist stress factors such as chloroform or high temperature. Landscape phages serve as substitutes for antibodies against cell-displayed antigens and receptors, diagnostic probes for bacteria and spores, gene- and drug-delivery systems and biospecific adsorbents (reviewed in Petrenko, 2008). Phage-derived probes inherit the extreme robustness of wild-type phage and allow fabrication of bioselective materials by self-assembly of phage or its composites on metal, mineral or plastic surfaces (Flynn et al., 2003; Reiss et al., 2004).

Although landscape phages retain their ability to infect Escherichia coli and form phage progeny, it is not surprising that these chimeras, with the alien peptides composing up to one-fifth of the phage mass, are defective to some degree when compared with parental wild-type strains. Some recombinant pre-coat proteins cannot be processed normally at the inner bacterial membrane (Malik et al., 1998) and it is likely that other types of defects are operative as well (Li et al., 2003). Most guest peptides are not tolerated when displayed on every pVIII subunit in high-copy-number vectors, and it was not until the introduction of vectors based on the replication-defective phage fd–tet that multibillion-clone landscape libraries could be constructed (Petrenko et al., 1996).

It was observed, however, that even in highly diverse libraries different guest peptides are presented unequally (Petrenko et al., 2002). The disproportionate representation of different phage clones in the original libraries and fluctuation of their composition that arises during phage construction and reproduction may affect efficiency of selection of clones with an optimal fitness. Therefore, there is a need to develop phage display libraries with extended repertoires of displayed sequences. In this work we compared the complexity, evolution and representation of two landscape libraries displaying foreign octamers and nonamers in the same register on the surface of the phage. They were obtained by replacement of amino acids 2–4 and 2–5 of pVIII with random octa- and nonamers, respectively. It was found that although these libraries have similar size and complexity, they represent phages that may belong to different families of homologous clones that were identified by parallel affinity selection with common selectors: β-lactamase, streptavidin and bacteria Salmonella typhimurium. These results justify development of separate landscape libraries covering different areas of a ‘sequence space’.

Materials and methods

Bacterial strains and general procedures

Escherichia coli strains MC1061 (F^- araD139 Δ (ara-leu)7696 galE15 galK16 Δ(lac)X74 rpsL (Str^r) hsdR2 (r_k⁻m_k+) mcrA mcrB1) (Meissner et al., 1987) and K91BlueKan (Kan^r Hfr-C thi lacZΔM15 lacY::mkh lacI^Q) (Smith and Scott, 1993) were obtained from George Smith (University of Missouri, Columbia, MO, USA). Salmonella typhimurium, reference strain ATCC 13 311, was obtained from the American Type Culture Collection (Rockville, MD, USA). All general procedures employed for construction of the libraries, production and analysis of recombinant phages, DNA sequencing, media and buffers, selection procedures and analysis of phage binding by enzyme linked immunosorbent assay (ELISA) are detailed in recently published protocols (Brigati et al., 2008) and previous publications (Sorokulova et al., 2005).

Statistical analysis of phage populations

Phage populations in original landscape libraries and selected phage clones were analyzed using bioinformatics programs RELIC (Mandava et al., 2004). A population diversity of libraries was estimated using the program POPDIV. Positional amino acid diversity in each randomized position was evaluated with program DIVAA and compared with parental primary libraries. Program INFO was used to calculate information content of selected peptides using the population of analyzed clones in primary libraries as a reference for calculation of amino acid occurrence probabilities. The AAFREQ program was used to analyze the frequency of amino acid occurrence in the random peptide inserts including 73 clones from primary and 73 clones from amplified f8/8 libraries, 49 phage-producing clones from the pre-f8/9 library (clones obtained as colonies of transformed bacteria after their electroporation with mutagenized phage DNA) and 61 clones from primary f8/9 libraries. Since the observed frequency of amino acids at different positions of the insert does not implicitly follow the theoretical profile of frequency, we obtained an intuitive portrayal of this discrepancy using an adaptation of the one-sample tests for proportions (Rosner, 2006). On the basis of our assumptions, the observed value is expected to follow a binomial distribution based on a fixed number of clones analyzed (N) and expected frequency (E). The result of this analysis was designated as the ‘frequency disparity index’ and was calculated for each amino acid for each position in the insert as follows:

where, the formula is an algebraic equivalent to that found in (Rosner, 2006) and FDI gives the measure of how much the expected frequency differs from the observed frequency, E the expected frequency of an amino acid at a given position in the phage insert; O the observed frequency of the amino acid at the same position in the phage insert as noted for ‘E’ above and N is the number of clones used to calculate the observed and expected frequencies.

The FDI scores thus obtained have an expectation of being zero, negative or positive depending on whether the amino acid being considered occurred at the expected level, at a lower frequency than expected or at a higher frequency than expected, respectively. Thus, a graphical representation of such scores provides a picture of over-representation or under-representation of different amino acids at different positions of the phage inserts. Furthermore, since we analyzed a relatively large number of clones, we expect the FDI scores to approximately follow a normal distribution that allows using two-tailed P-values obtained from a standard Z-table (Rosner, 2006) to prove statistical significance of the difference seen between the observed and expected frequencies of an amino acid at a particular position.

Results

Complexities and diversities of phage libraries

Previous studies indicated that randomization of the N-terminal region of each copy of the phage major coat protein pVIII may be subject to some limitations in size and structure (Ilyichev et al., 1989; Iannolo et al., 1997; Rodi et al., 2005). It has been demonstrated that phage M13 can tolerate a 6-mer peptide insertion at the beginning of the pVIII protein, but when the length of the peptide increases, the percentage of clones able to form infective phage particles drops significantly. Only 40% of clones with inserts corresponding to 8-mer peptides formed infective phage particles and this value decreased to 20% for 10-mer peptide inserts and to 1% for 16-mer peptide inserts (Iannolo et al., 1997). Attempts to construct an 8-mer representative library in each copy of the pVIII protein based on the multi-copy vector M13 failed (Petrenko and Smith, unpublished). The first comprehensive multibillion-clone landscape library was successfully constructed only when the very low-copy number vector f8-1 (AF218734), a derivative of the phage fd–tet with a defective origin of replication, was employed (Petrenko et al., 1996). The replication defectiveness of the vector becomes an advantage as it averts a phenomenon called ‘cell killing’. Also, toxic effects of fusion proteins may be better tolerated when carried by replication-compromised vectors such as fd–tet, which produce a very low number of replicative form (RF) DNA in infected bacteria (Smith, 1988). In this library a random octamer replaces amino acids 2–4 (EGE) at the beginning of the pVIII protein of the f8-1 vector extending the total length of the fusion pVIII protein by five amino acids (Figure 1).

Fig. 1.

Vectors and libraries. In the nucleotide sequences corresponding to the part of recombinant gene gpVIII encoding the N-terminal part of the major coat protein, randomized structures are designated as nnk, where n = A, T, G or C, and k = G or T. Restriction sites for PstI and BamHI are underlined. N-terminal amino acid structures of mature recombinant pVIII proteins in libraries indicated by capital single letters according to amino acid abbreviations. Randomized amino acids are designated in small letters (a–h in the f8/8 library and a–i in the f8/9 library). Amino acids are numbered as in vector phage f8–5 (Petrenko et al., 2002).

Here we have demonstrated that another amino acid (D5 at the beginning of the pVIII protein) could also be randomized. In the f8/9 library, four amino acids (EGED) were replaced by random nonamers bringing a total length of the fusion pVIII protein to 55 amino acids, the same length as in the previously constructed f8/8 library (Petrenko et al., 1996). For construction of the f8/9 library we used vector f8–6, derivative of f8–5 (AF464138) (Petrenko and Smith, 2005) (Fig. 1). A characteristic feature in f8–6—two amber TAG codons at the beginning of the gene gpVIII—ensures the absence of residual wild-type phage in the library. Such wild-type phage was shown to overgrow their recombinant counterparts during successive amplification that may hinder the results of affinity selection. Double-stranded randomized DNA fragments were ligated into cleaved vector DNA followed by electroporation of resulting recombinant DNA molecules into MC1061 E.coli cells. A portion of transformed bacterial clones (named the ‘pre-library’) was grown on an indicator plate with tetracycline, and the major part of transformed bacteria was cultured in liquid media with tetracycline. The size of the pre-library (2 × 10⁹ clones) was estimated in proportion to the number of clones that grew on the indicator plate. The phage library isolated from the liquid culture, containing ∼4.4 × 10¹³ phage virions, was named the ‘primary library’. It was observed that about half of the transformed bacteria in the pre-library produced phage particles, in agreement with our previous observations (Petrenko et al., 1996). It was determined by growing individual pre-library clones in liquid cultures, separating bulk cells by centrifugation and titering the phage remaining in the supernatants. The complexity (the size) of f8/9 library (1.2 × 10⁹ clones)—the number of primary clones able to produce phage—was determined as a portion of the primary clones producing phage particles (61%) compared with the total number of primary clones (2 × 10⁹ clones). The rest of the pre-library clones (39%) did not produce phage in the host bacteria because they contained stop codon TAG in different positions of the gene-encoding fusion coat proteins that did not allow expression of the fusion pVIII protein and assemblage of the phage in the non-suppressor strain MC1061. That was demonstrated by PCR amplification of the corresponding segments of double-stranded viral DNA from tetracycline-resistant bacterial clones and their sequencing. Thus, we found that complexities of f8/8 and f8/9 libraries were similar to each other but both differ from their theoretical complexities [1.28 × 10¹¹ for f8/9 library encoded by DNA’s Gnk(nnk)₈ where n = G, A, T, or C and k=G or T; and 4.16 × 10⁹ for f8/8 library encoded by DNA’s Gnk(nnk)₆nnG].

The quality of a phage display library can be characterized by its completeness or diversity—the proportion of all possible sequences actually present in the library. Population diversity of the libraries was estimated from the sequences of a limited number of the members of the libraries (73 for f8/8 and 61 for f8/9) using the statistical program RELIC POPDIV (Mandava et al., 2004). An algorithm of the program has been previously shown to provide reasonable estimates of population diversities even when based on as few as 50 peptides even though a high standard deviation is observed due to the low number of sequences analyzed (Makowski and Soares, 2003). Since both theoretical and observed diversities were calculated using the same algorithm, using at least 50 peptides in each case, the estimates derived can serve as an acceptable measure of the population diversities of our libraries. Calculated population diversities of f8/8 and f8/9 primary libraries (0.0091 ± 0.0045 and 0.0029 ± 0.0016, respectively, as per POPDIV) were 37 and 3 times lower than the observed diversities of these libraries (1.4 × 10⁹/4.16 × 10⁹ = 0.34 and 1.2 × 10⁹/1.28 × 10¹¹ = 0.0094, respectively) probably because of unequal presentation of different clones in the libraries and uneven positional distribution of amino acids in foreign peptides. It is interesting to note that population diversity of phage-producing clones in the pre-library f8/9 (0.0116 ± 0.0074) exceeds the diversity of the primary library f8/9 (0.0029 ± 0.0016) four times, indicating that occurrence of unique clones in the library change during its growth in the liquid medium because the efficiency of phage assembly and export is different for different clones. An even more dramatic change in population diversity was observed when a portion of the primary library f8/8 was used for amplification in the presence of fresh E.coli K91BlueKan cells (0.0091 ± 0.0045 and 0.0012 ± 0.0006 for the primary and amplified libraries, respectively). The decrease in population diversity of the library more than seven times during its amplification can be attributed to differences in infectivity of individual clones in the library toward the host bacterium and differences of biosynthesis of distinct phage particles inside the bacterial cell during phage assembly and export.

Positional diversity is a statistical measure of the proportion of the 20 possible amino acids that are observed at any given position (Mandava et al., 2004). If a position in the peptide is populated by equal proportions of all 20 amino acids, then the diversity is equal to 1 (20/20). If a position in the peptide is populated by only one amino acid in all sequences, then the diversity is 0.05 (1/20), as is the case for the 9th position in the f8/8 library which is always occupied by aspartic acid. We estimated the positional diversity of both libraries using the statistical program RELIC DIVAA. The positional diversities of both libraries exhibited similar patterns (Figure 2) with the diversity in position one for f8/8 and f8/9 libraries and position eight for the f8/8 library being significantly lower than in the other positions because they have codons Gnk and nnG, encoding 5 and 13 amino acids, respectively.

Fig. 2.

Positional diversities of libraries used in the study as determined by the program DIVAA of the RELIC suite. Diversity of f8/8 and f8/9 primary libraries; x-axis denotes amino acid position, y-axis denotes diversity measured. Solid line corresponds to f8/8 library, dashed line corresponds to f8/9 library.

Biological censoring of the phage libraries

We assumed that the diversities of f8/9 and f8/8 libraries were affected by biological censoring of phage-producing clones during phage amplification. To check this hypothesis, amino acid profiles of the theoretically randomized 8-mer and 9-mer libraries were compared with amino acid profiles of the obtained libraries (Figure 3) using the RELIC program AAFREQ (Mandava et al., 2004). Further, the one-sample test for proportions was used to obtain a statistical measure (FDI) of the divergence from the theoretical diversities:

where, the formula is an algebraic equivalent to that found in Rosner (Rosner, 2006) and FDI gives a measure of how much the expected frequency differs from the observed frequency, E the expected frequency of an amino acid at a given position in the phage insert, O the observed frequency of the amino acid at the same position in the phage insert as noted for ‘E’ above and N is the number of clones used to calculate the observed and expected frequencies.

Fig. 3.

Frequencies of amino acids at different positions in peptide inserts. The expected frequency of amino acids and the observed frequency calculated using the AAFREQ program of the RELIC suite (Mandava et al., 2004) were used in an adaptation of the one-sample tests for proportion (Rosner, 2006). The results of this analysis were expressed as a frequency disparity index (FDI) which is a measure of how much the observed frequency deviates from the expected. Amino acids at various positions of the peptide inserts are arrayed along the x-axis whereas the y-axis represents the FDI. FDIs for each amino acid are presented by clusters of columns, in which positions of each column correspond to the amino acid’s position in the peptides (a–h for 8-mer libraries, and a–i for 9-mer libraries, see Fig. 1 and insert lower panel). The dashed line represents statistical significance of a frequency disparity at the α-level of 0.05. (A) f8/8 amplified library; (B) f8/8 primary library; (C) f8/9 primary library; (D) f8/9 pre-library.

We found that the observed frequency of amino acids at different positions of the randomized area does not implicitly follow the theoretical profile of frequency. On the basis of the formula applied in our calculations, an FDI of 0 would imply that the occurrence of the amino acid at a given position follows the expected frequency. A positive FDI score (with the bar above the X-axis) would imply that the amino acid is over-represented at a given position whereas a negative FDI score (with the bar being below the X-axis) would imply an under-representation of the amino acid at a given position. Furthermore, at an α-level of 0.05, an FDI score of 1.96 would represent significant deviation of amino acid frequency from the expected value with the amino acid being either overrepresented or underrepresented depending on the sign of the FDI score. A critical issue is that even though the graphical representation shows the amino acids to be grouped together for ease of comparison, the occurrence of each amino acid at each individual position in our approximation is an independent event and cannot be correlated to either neighboring positions or to other amino acids. This means that the degrees of over- and under-representation are unique for each amino acid at each position. Also, no adjustments were made for multiple comparisons as our focus was on individual amino acids at individual positions. A special case to be considered is the amino acid cysteine that is conspicuously absent in all clones of the library due to structural restraints. This absence is accurately depicted as under-representation in our analysis supporting the prediction of independence of each amino acid and position in our analysis. Our observations based on the above technique can be summarized subsequently.

Evolution of the pre-library f8/9 during its amplification

Distribution of FDI scores in the f8/9 pre-library and f8/9 primary library are very similar (Figure 3D and C, respectively). In both libraries, the acidic amino acids aspartic acid (D) and glutamic acid (E) are significantly overrepresented while the positively charged amino acids lysine (K) and arginine (R) are relatively underrepresented. Glycine (G), asparagine (N) and serine (S) are overrepresented in several positions while leucine (L) and tryptophan (W) are relatively underrepresented. Only few statistically significant changes in amino acids distributions were observed during amplification of the pre-library f8/9. For example, aspartic acid (D) in position ‘i’ is much less overrepresented in the primary library; isoleucine (I) is underrepresented in most positions of the primary library, while similar amino acids leucine (L) and valine (V) are less underrepresented. Thus, noted above the difference in diversities of the pre-library and primary f8/9 libraries can be attributed to the changes in their amino acid distribution, probably influenced by biological mechanisms of phage assembly and secretion.

Evolution of the primary library f8/8 during its amplification

Distribution patterns of amino acids for the f8/8 primary library and amplified f8/8 library are very similar (Fig. 3B and A, respectively). In both libraries aspartic acid (D) and asparagine (N) are overrepresented in the N-terminus of the peptide inserts whereas glutamine (Q) is overrepresented at the C-terminus. Proline (P) and threonine (T) are overrepresented at various positions except the first. Serine (S) is overrepresented in the first few positions. Arginine (R) is underrepresented in the first and last positions. Glycine (G), leucine (L), arginine (R), valine (V) and tryptophan (W) are underrepresented at several positions. The amplification of the primary library leads to ‘contrasting’ of the biased distribution of amino acids in different positions changing overall library’s diversity, as noted above.

Comparison of the primary libraries f8/8 and f8/9

Frequencies with which the amino acids occurred in both libraries are quite different with a distinct preference toward negatively charged amino acids D and E in the f8/9 library and very different patterns of glycine (G), proline (P) and threonine (T) distribution in both libraries (Fig. 3B and C, respectively, for the primary library f8/8 and primary library f8/9). Comparison of the amino acid profiles for the f8/8 and f8/9 libraries against theoretically randomized libraries provides evidence of censoring of the real libraries. Randomization of an additional amino acid (D5) in phages belonging to f8/9 library also affected amino acid frequency profiles and library evolution with a preference toward better surviving clones during amplification of the library.

Selection of phages from primary libraries f8/8 and f8/9 that bind model targets

We hypothesized that using different landscape phage libraries may be advantageous in the search for specific ligands by affinity selection against different targets because the range of clones belonging to the same sequence ‘family’ in each library may be different. Selection of both libraries with the same target should yield a greater variety of structurally non-overlapping clones. To prove this hypothesis, we surveyed both primary libraries in selection of binding phage against the same selectors: monomeric TEM-1 β-lactamase, tetrameric streptavidin and whole S.typhimurium cells.

Tem-1 β-lactamase-binding phages

The simplest target we chose was TEM-1 β-lactamase, a monomeric bacterial protein which has been used to select phage-binding clones also by Huang et al. (Huang et al., 2003). Phage libraries were depleted against plastic and bovine serum albumin to eliminate phages binding to these entities during the selection process. Three rounds of affinity selection were performed to obtain phage clones with a putative propensity to bind β-lactamase. Fifty clones, randomly picked from the last round of selection, were sequenced for each library. Most of the sequences of the selected peptides from both libraries (Table I) demonstrate the clustering into families, with a few ‘orphans’ that did not belong to any identifiable family structure. Positional diversity of amino acids in affinity selected guest peptides was distinct from diversity of the guest peptides in the original library and showed a predictable decrease in all positions except for position 7 in the octamer library (Figure 4) and position 6 in the nonamer library (data not shown).

Table I.

Phage displayed peptides selected from landscape phage libraries

	β-lactamase	Streptavidin	S. typhimurium
			Acid	DOC
Library f8/8	DPKPTAAA4	VPEGAFSS7	VTPPQSSS	DPKGPHSM
	DPRPESAP	VPEGAFTS3	VTPPTSPQ	DPKSPLHT
	EPKPTPAA2	VPEGAFGS	VTPSSPHS	DPKSPQQT
	DPPKRPDV5	VPEGAFST	VTPQGSHP	DPRSPASL2
	DPSSRQTP	VPEGAFSQ	VSTQSTHP	DPRPAQHT
	EHPQPPTP	VPESAFAQ	TPGQPSHP	DPHKAGGL
	ERPAPQLP	VPDGAFST	VPPPSPQS2	EPRLAHGA
	DRVQPAMQ3	VPDSAFNT	VPPPSPHS3	EPHRAASV
	SDTSSPGQ	VPDGAFSQ2	VPPPSASS	DPSKRTQP
	VNTSSPGQ6		VPPPSQSQ2	EPNKHSQS
	VSPPSHST		VPPPSNPS	DRPSPNTV
	VTGSPPST2		VPPPGQHQ	VTPPQQGS
	VTPSPTPQ		VPPSSSSP	DNKMTSHS
	VPPQSNSM		VPQQDKAQ
	APVHQESS
	DNASAPRS
Library f8/9	ARSVAMSDS	VPVGAYSDT21	ELPLAFGND2	VHGETSNQD2
	ASSVAMSDS	AALGHPAMD22	ELPLDPGLD	VHSEGSVNT
	DFGYAKEDT	ETHLDPNRT4	GSYSDMVDN2	VHSDHSISD
	DQRGDRDDT	EYVLHGSED	GVYSDISGD	DKNSGGGES4
	DHLNVASSD		VNYDDMTST2	DKASPGSSD2
	ERTQDGSSD		VPYADMSES2	DKHEGSNTD
	DQSGAVGMG		AGMTYDLPD2	DDYNFYGVN
	ENTGTSIPE		DAFSQSATD2
	VLSSDHNED		VAEPVDLPA
	VSSSDHNED2
	VPSGDVSME
	VQGYGPSMD
	VSMEVAPDA
	VSSGTGPDG
	GMGPEYGGD
	VTAPSTAED
	AIETTVGDD
	EPQTLYGTQ
	GHTGGLEED
	VHNGNLRLD
	GDSGTGDSH
	EMDTGKDGN
	ATFSVPEAD
	ASSPGIGSE

Common motifs in families are indicated by bold letters. Superscripted arabic numbers show how many identical clones were identified in the group of the sequenced phages.

Fig. 4.

Positional diversities of amino acids in selected β-lactamase-binding peptides. The x-axis denotes amino acid position, y-axis denotes diversity measure. Diversity of primary library f8/8 (solid line) and selected β-lactamase binding clones (dashed line).

An initial screening phage capture ELISA was performed to identify the target-avid clones and based on its results, we selected two clones (VSPPSHST and DNASAPRS) from the 8-mer library and three clones (GHTGGLEED, DHLNVASSD and DQRGDRDDT) from the 9-mer library for further assays. The binding profile of selected phages was analyzed by micropanning. This format was similar to the affinity selection procedure with the binding activity of the clone being assessed as the number of infective phage particles obtained after elution. The highest levels of recovery was demonstrated by phage VSPPSHST selected from the f8/8 library (22 times higher than the control vector phage) and phages GHTGGLEED and DHLNVASSD from the f8/9 library (52 and 42 times higher, respectively, than that of the control phage). Phage f8–5 (AF464138) derivative of f8–1 (AF218734), which has the same pVIII amino acid structure but differences in three codons (codons 16, 17 and 24), was used as a control for all phages analyzed. Clones selected for their binding efficiency were then analyzed by target capture ELISA in which the format of the binding reaction was reversed: a phage clone was immobilized and allowed to bind the dissolved target TEM-1 β-lactamase. Phage VSPPSHST selected from the library f8/8 in this test bound at a level which was twice that of the control phage f8–5. Clones GHTGGLEED and DHLNVASSD selected from the library f8/9 bound at levels which were approximately twice that of the vector phage, while phage DQRGDRDDT bound at a level three times higher than the control. Numerical data from these assays can be found online as supplementary material (Supplementary data are available at PEDS online). These data clearly showed that the best binders of β-lactamase recruited from libraries f8/8 and f8/9 belong to different families. The use of two separate libraries allowed widening the spectrum of binding clones.

Streptavidin-binding phages

Streptavidin was chosen as a more complex target possessing a tetrameric structure with four potential binding sites. It has commonly been used as a model for analysis of random peptide libraries and several consensus motifs such as HPQ (Devlin et al., 1990; Kay et al., 1993; McLafferty et al., 1993; Giebel et al., 1995), GDF/WXF, PWXWL (Roberts et al., 1993), EPDWF/Y (Caparon et al., 1996) and DVEAWL/I (Lamla and Erdmann, 2003) that bind streptavidin have been identified.

Streptavidin-binding phages were selected in the same set up as described above for TEM-1 β-lactamase (Table I). Fifty clones that were isolated and sequenced contained seven unique sequences: VPVGAYSDT, AALGHPAMD, DQFSLQSQD, GDDYANKES, ETHLDPNRT, EYVLHGSED and VGGFGHPDD with a predominance of two clones, VPVGAYSDT (21/50) and AALGHPAMD (22/50). In the phage capture ELISA with the fixed concentration of the phages, phage VPVGAYSDT bound streptavidin 129 times stronger, phage GDDYANKES—107 times stronger, phage EYVLHGSED—24 times stronger, phage AALGHPAMD—10 times stronger and phage ETHLDPNRT—8 times stronger than negative control (vector f8–5). Phage VPVGAYSDT resembles the structures of clones with the motif (VPxGA Y/F S/Txx) isolated previously from the f8/8 library (Table I) (Petrenko and Smith, 2000). The best binder of this family, phage VPEGAFSS, demonstrated a 135 times higher binding efficiency than the control phage in the same experiment with the rest of the binding clones selected from f8/9 library.

Affinities of two representative phages from different libraries, VPEGAFSS from the f8/8 library and VPVGAYSDT from the f8/9 library, were compared in a competition ELISA, in which the previously characterized streptavidin-binding phage VPEGAFSS was immobilized and used as a detector for streptavidin. The surveyed phages were pre-incubated with AP-SA in solution and applied to the wells containing immobilized detector phage (Fig. 5). Phage VPVGAYSDT bound to streptavidin (AP-SA) with about 50 times higher affinity than phage VPEGAFSS (estimated as the proportion of phage concentration when 50% of AP-SA is bound to the immobilized detector phage). Numerical data from these assays can be found online as supplementary material (Supplementary data are available at PEDS online). These data confirmed the assumption that the use of separate phage libraries can be beneficial for selection of phages with higher level of binding to a corresponding selector.

Fig. 5.

Competition ELISA of streptavidin-binding phages. Streptavidin-conjugated alkaline phosphatase (AP-SA, final concentration in the mixture—0.67 µg/ml),) was mixed with gradually increasing concentrations (from 9.7 × 10¹³ to 0.55 × 10⁹ vir/ml) of phage VPVGAYSDT (dashed line) and phage VPEGAFSS (solid line) and tested on ELISA plates with immobilized detector phage VPEGAFSS. ELISA signals of characterized phages were presented as a percentage of ELISA signals of the samples without phage, which were considered as 100%. (Details for the procedure described in (Petrenko and Smith, 2000).).

Salmonella binding phages selected from f8/8 and f8/9 libraries

The affinity selection regimen for our library analysis culminated with the most complex target, S.typhimurium. As described above, phage libraries were depleted against plastic and bovine serum albumin to eliminate phages binding to these entities during the selection process. Two elution strategies were employed to recover the phages—using acid and deoxycholate buffers (the last allows isolation of membrane-bound or cell-penetrating phages). Most clones randomly chosen after the fourth round of affinity selection from the f8/8 library congregated into three families (Table I). Almost all clones recovered bound S. typhimurium cells more strongly than control vector f8–5 in cell capture ELISA (Sorokulova et al., 2005). Clones from Acid and DOC fractions encoded no apparent overlapping structures of the displayed peptides except for clone VTPPQQGS, originating from the f8/8 library, which was found in the DOC fraction and shared the common motif VTPPxxxS with a family found in the Acid fraction. Most clones selected from the f8/9 library (12/16) could be clustered into four families with structures distinct from those identified in the f8/8 library.

Several phage clones from the f8/8 and f8/9 libraries which demonstrated high ELISA signals were evaluated in co-precipitation assay (Table II). In this test, selected phage particles were incubated with S.typhimurium cells in solution, unbound phage was removed by centrifugation followed by washing, and bound phage was eluted with deoxycholate buffer and titered. Yields of recovered phages were compared with the yield of negative control phage f8–5. Six ELISA positive clones from the f8/8 library and two clones from the f8/9 library demonstrated binding with S.typhimurium cells in solution (Table II). Numerical data from these assays can be found online as supplementary material (Supplementary data are available at PEDS online). As in two preceding examples, the use of two separate libraries for selection of Salmonella binders permitted increasing diversity of bacterial binders.

Table II.

Co-precipitation assay of S.typhimurium-binding phages

Library	Peptide insert	Yield, recovery/input in %
f8/8	DRPSPNTV	8.0 × 10−1
	VPQQDKAQ	6.0 × 10−2
	VTPPQSSS	5.0 × 10−2
	DPKSPQQT	1.9 × 10−3
	VTPQGSHP	8.2 × 10−4
	VSTQSTHP	2.5 × 10−4
f8/9	VNYDDMTST	7.0 × 10−2
	AGMTYDLPD	2.3 × 10−4
	ELPLDPGLD	<1.2 × 10−4
	VHSEGSVNT	<1.2 × 10−4
Control	EGE (f8-5 vector)	1.2 × 10−4

Phage suspensions were mixed with suspension of bacterial cells, incubated and phage–bacteria complexes precipitated. Phages were recovered by elution buffer. Phage inputs and recoveries were determined by biological tittering (Brigati and Petrenko, 2005; Sorokulova et al, 2005).

Analysis of a possible correlation of information content of the selected peptides with their binding ability

Bioinformatics has enabled the analysis of information relating to the physico-chemical properties of biologically relevant molecules in a relatively short span of time. These tools, though extremely useful, provide only a theoretical picture, the merit of which could be considerably improved if it could be translated to a practical application estimating predictability of peptide binding affinity to a specific target. Toward such a goal, we analyzed data generated from the INFO program of the RELIC suite for affinity selected peptides and attempted to correlate the results from theoretical predictions with our experimental observations.

The information value of a given peptide arises as a functional extension of the theory of information, the basic precept of which is that information is a decrease in uncertainty (Shannon, 1948). Accordingly, information calculated by the program INFO is a measure of the probability of encountering a particular peptide in a pool of affinity selected peptides due to random chance (nonspecific binding or good growth characteristics) as opposed to ligand affinity. It is equal to the negative logarithm of the probability of its natural occurrence. INFO uses the program AAFREQ to calculate position-specific frequencies of each amino acid at each position of the insert and uses these data to generate the probability of random observation of the insert as a whole by multiplying the probability of each amino acid occurring at each position within the peptide. The lower the probability of occurrence is, the higher the information value will be as its presence in the affinity selected sub-library is likely due to its ligand affinity rather than good growth characteristics or non-specific binding. The higher the probability of occurrence, the higher is the chance that the peptide will be carried over to the subsequent rounds due to favorable viral growth rather than ligand affinity and thus lower is the information value. Similarly, peptides with high information content are rare in the parental library and therefore their increased presence in the affinity-selected sub-libraries should be due to affinity to their target and vice versa. We estimated information content of the selected clones from affinity-selected phage populations (shown in Table I) and compared them to the information content of the clones from the original landscape libraries. The analysis revealed that in each case information content profiles of the clones from the selected population were different from that of the original libraries. We postulated that the increased occurrence of a clone with high information (as calculated by the INFO program) is a function of its binding affinity as demonstrated by ELISA. It was predicted that a set of clones with high information content would emerge during selection and that the best binder would be among these selected phages. This expectation was fulfilled in the selection against streptavidin (Fig. 6; the best binder is starred) and partially in the selection against Salmonella (data not shown), although in selection against β-lactamase the best binder demonstrated relatively low information content (data not shown). Though these results indicate a possibility of a direct relationship between the theoretical information that a peptide contains and the practical rendering of this information into a binding signal, it may not be universally true. Within their own realms, both information content of a peptide and the cognate binding signal are complex factors that are dependent on multiple sub-factors. Thus, a more complicated model may be better able to define the relationship between these two entities and enable the use of information content of a peptide as a predictor of the binding signal that it will eventually produce in an assay.

Fig. 6.

Shift in the Information profile of the f8/8 library following affinity selection against streptavidin. The information content is a measure of the probability of encountering a particular peptide in a pool of affinity-selected peptides due to random chance (non-specific binding or good growth characteristics) as opposed to ligand affinity. Thus, the profile of information is supposed to change as a result of affinity selection that is demonstrated in the above figure with shift to the right. Also we predicted that increased occurrence of a clone with high information (as calculated by the INFO program) is a function of its binding affinity as demonstrated by ELISA. The information profile of the parent f8/8 library is shown as dashed line; information content profile of affinity selected clones clones is shown as solid line. The x-axis denotes information content of peptides, y-axis denotes the occurrence of peptides in populations of the primary library f8/8 and selected clones. The star represents the best binder identified by the binding assays (VPEGAFSS).

Discussion

The ultimate goal of most phage display selection projects is identification of peptide ligands that bind strongly and specifically to a target receptor. It may be assumed a priori that in every population of random peptides there are peptides with ideal fitness and highest affinity to the target, so called leading peptides. A collection of structural homologues to a leading peptide would therefore form a structural family within a library and would be expected to possess lower fitness and thereby lower affinity to the target. A model of such relationships, suggested previously (Smith and Petrenko, 1997), can be envisioned as an affinity cone with the lead peptide sequence occupying the peak and different members of the family occupying different and descending levels in the cone based on a hierarchical gradient of affinity to the target (Fig. 7).

Fig. 7.

Hypothetical three dimensional representation of peptides population in a sequence space. Following a previously suggested model (Smith and Petrenko, 1997), a highly simplified two-dimensional tableau of all possible combinations of amino acids for a given number of randomized positions can be represented as a grid with each position in the grid standing for a unique combination of amino acids (e.g. 4.16 × 10⁹ or 5 × 20⁶ × 13 for f8/8 library encoded by DNA’s Gnk(nnk)₆nnG). For convenience, we assume that points that are close in our two-dimensional representation indicate similar amino acid sequences. To make the model more illustrative, we can represent the possible sequences for a hypothetically 4-mer library in two non-overlapping subsets of the randomized positions along the axes of the plane, e.g. subset of 400 sequences or (20 × 20) (positions 1–2, along the x-axis, and subset of 400 or (20×20) (positions 3–4, along the y-axis). In phage libraries displaying longer peptides, the sequence space would be considerably larger (e.g. 4.16 × 10⁹ for the 8-mer library and 1.28 × 10¹¹ for the 9-mer library). Real phage libraries, which hardly reach the theoretical diversities, would be seen as a collection of random points in sequence space and be expected to form overlapping domains within such a grid as depicted (A and B). An additional dimension is superimposed onto this grid when phage libraries are considered in the context of their affinity toward a target. Thus, we would expect to find structurally related peptide clusters stratified on the basis of their target affinity so as to form ‘affinity cones’ (1, 2 and 3) with affinity increasing from the base to the peak of each cone. Hence, the peptide with the highest affinity to the target (‘leading peptide’) would form the peak of the cone.

In an ideal situation, the leading peptide should be revealed by affinity selection (biopanning) as we have demonstrated. However, in practice, the leading peptide may be absent from the selected pool of phage clones because the theoretical complexity (total number of all possible peptide structures) of the library exceeds the actual observed complexity. For example, there are 1.28 × 10¹¹ possible peptide structures for the 9-mer library Gnk(nnk)₈, which is about 60 times higher than the actual number of clones we have observed in this library. Furthermore, since the peptide is genetically coupled to a biological entity, the phage, the availability of this peptide for selection is contingent upon the growth capacities of its carrier phage. For example, a phage clone may be a moderate binder but yet be selected and propagated by virtue of its better multiplicative potential. Thus, more abundant phage clones in the lower echelons of the affinity cone would be more likely to be selected.

To address this problem, we hypothesized that the chances of revealing the leading ligands may be considerably increased by using ‘separate libraries’. These are libraries enriched with clones that belong to different families. To explore this hypothesis, we obtained the landscape library f8/9 by randomizing position 5 in the major coat protein pVIII normally occupied by aspartic acid (D). This change represented an increase in the number of randomized amino acid residues compared with our previous landscape library, f8/8, in which positions 2, 3 and 4 of the major coat protein had been randomized. Our primary observation was that, in addition to enhancing the diversity of the random peptides displayed on the phages, the type of amino acid in position 5 also played a functional role in determining the viability of the phage. A direct evaluation of the diversity of peptides generated by the randomization of amino acids at position 5 would require statistical analysis of thousands of clones. Instead, we chose to analyze the common shift in the peptide diversity as a result of the randomization of position 5 in the major coat protein. Despite some bias in amino acid selection at this position, such as a complete absence of cysteine, an under-representation in hydrophobic and basic amino acids, and an increased representation in acidic acids, the novel library differs considerably from the former f8/8 library. Distinguishing characteristics of the new 9-mer library included an even greater bias against positively charged amino acids (K and R), an increased number and distinct dispersion of negatively charged amino acids (D and E), a change in frequency and dispersion of valine, a decreased number of alanines and prolines and an increase in the number of glycines. Overall, we demonstrate here that both libraries encode unique amino acid sequence profiles suggesting the prevalence of different families of phage clones represented in each library.

To further characterize the sequence repertoires encoded and the behavior of the libraries during affinity selection, we screened the libraries against three selectors with a gradient of structural complexity: TEM 1 β-lactamase, streptavidin and S.typhimurium. Analysis of the unique clones obtained during affinity selection against TEM 1 β-lactamase revealed non-overlapping and dissimilar families of peptides and sometimes even individual phages within each library. A significant observation was that the diversity of the clones selected from the f8/9 library was higher than the diversity of clones selected from the f8/8 library. Thus, the use of both libraries in selection protocols against the same target is justified as this would provide with a greater chance of finding a peptide with high affinity and specificity. Furthermore, the selection with ‘separate libraries’ such as the one used in our experiments is warranted based on the increased diversity that is generated in the selected clones.

Affinity selection against streptavidin has already been done in our laboratory using the f8/8 library (Petrenko and Smith, 2000). Here, we used the novel f8/9 library and contrasted the results obtained with those previously obtained for the f8/8 library. A predominant family of clones with the motif VPxGAY/FS/T had been previously identified in the f8/8 library as being specific for streptavidin. Among the highly diverse affinity-selected clones from the 9-mer library, only one clone showed homology with this family. This clone displayed an affinity to streptavidin that was 50 times higher than the best binder from the f8/8 library. Thus, separate libraries create greater diversity in the repertoire of the clones represented in the library, provide a wider range of alternative clones and allow phage clones with maximal affinity to a particular target to be selected.

Phage display is rapidly being adapted as a high throughput screening system for affinity ligands against biologically relevant targets like tumor receptors, surface components of bacteria and others. With these applications in mind, we decided to test our hypothesis by screening both the f8/8 and f8/9 libraries against a complex target, S.typhimurium cells, exposing numerous binding receptors on their surface. As with our earlier targets, we identified non-overlapping and highly divergent sets of clone families in each library and isolated unique clones with a variety of binding capacities.

In summary, we have demonstrated that the use of separate landscape libraries enhances the likelihood that novel ligands can be isolated including some with very high target affinities and simultaneously increase the diversity of the target-specific ligands that can be selected. It has previously been demonstrated that landscape libraries with high diversity are a rich source of specific binders with applications in cancer cell recognition, gene- and drug-delivery systems, diagnostic probes against pathogenic bacteria and spores, etc. Such a wide range of potential applications for this technology has generated a considerable demand for many new target-specific ligands. Techniques that enhance the repertoire of peptide ligands during affinity selection would help quench this demand and create new novel adaptations of phage display technology.

Funding

This work was supported by Army Research Office ARO/DARPA grant no. DAAD 19-01-10454 (to V.A.P.), National Institutes of Health grants NIH-1 R21 AI05564501 and 1 R01 CA125063-01 (to V.A.P.). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute of NIH.