Structural Guided Scaffold Phage Display Libraries as a Source of Bio-Therapeutics

We have developed a structurally-guided scaffold phage display strategy for identification of ligand mimetic bio-therapeutics. As a proof of concept we used the ligand of integrin αvβ6, a tumour cell surface receptor and a major new target for imaging and therapy of many types of solid cancer. NMR structure analysis showed that RGD-helix structures are optimal for αvβ6 ligand-interaction, so we designed novel algorithms to generate human single chain fragment variable (scFv) libraries with synthetic VH-CDR3 encoding RGD-helix hairpins with helices of differing pitch, length and amino acid composition. Study of the lead scFv clones D25scFv and D34scFv and their corresponding VH-CDR3 derived peptides, D25p and D34p, demonstrated: specific binding to recombinant and cellular αvβ6; inhibition of αvβ6-dependent cell and ligand adhesion, αvβ6-dependent cell internalisation; and selective retention by αvβ6-expressing, but not αvβ6-negative, human xenografts. NMR analysis established that both the D25p and D34p retained RGD-helix structures confirming the success of the algorithm. In conclusion, scFv libraries can be engineered based on ligand structural motifs to increase the likelihood of developing powerful bio-therapeutics.


Introduction
The use of combinatorial phage display scFv libraries for generation of therapeutic antibodies is well established and has resulted in clinically valuable reagents [1,2]. ScFv libraries are commonly made from immune or naïve B cells or as synthetic libraries where antibody variable heavy (V H ) and variable light (V L ) gene segments are rearranged in vitro with synthetic complementarity determining regions (CDRs) coding for random sequences of varying lengths [3][4][5]. The use of the phage display library has been used to develop antibodies for therapeutic intervention using the above combinatorial libraries. We reasoned that the use of antibody engineering in combination with ligand structural studies will result in robust libraries that can lead to isolation of potent ligand-mimetic bio-therapeutic antibody candidates.
Since receptor:ligand interactions must be considered as interacting topographical maps we wondered if it were possible to generate a target-selective library by incorporating a panel of specific three-dimensional shapes into the CDR3 of the variable heavy (V H -CDR3). If such a library used stereochemical shapes that corresponded to a ligand-binding interface then it would more likely generate scFv(s) that will block the ligand:receptor interaction than would a conventional random library. To test this hypothesis we considered a therapeutically relevant target, the integrin avb6, which represents a novel and important tumourselective target that is expressed on the surface of cancer cells. We, and others, have shown that avb6 promotes cancer cell migration, invasion and growth in vivo [6][7][8][9][10]. Moreover strong expression of avb6 correlates with poor prognosis in multiple cancers [11][12][13] and thus human therapeutic antibodies to this integrin are likely to have a significant therapeutic value.
In previous studies we identified the 20mer peptide A20FMDV2 (N 1 A 2 V 3 P 4 N 5 L 6 R 7 G 8 D 9 L 10 Q 11 V 12 L 13 A 14 Q 15 K 16 -1 A 2 V 3 P 4 N 5 L 6 R 7 G 8 D 9 L 10 Q 11 V 12 L 13 A 14 Q 15 K 16 V 17 A 18 R 19 T 20 ), derived from the foot and mouth disease virus VP1 coat protein, as a highly specific and potent ligand for avb6 [14]. The peptide included the avb6-binding motif RGDLXXL identified by Kraft et al (1999) [15]. Using NMR we determined the threedimensional structure of A20FMDV2 in 30% TFE: a hairpin structure with RGD at the tip of the turn followed by a C-terminal helix [14]. This structural motif has also been observed in additional avb6-binding peptides (Wagstaff et al, 2012, [14]). In addition, we have previously probed the role of the helix in A20FMDV2 through the use of specific D-amino acids to disrupt helix formation, and found the resulting helix-attenuated peptide had greatly reduced binding to cells expressing avb6 [14].
In this study we have used our NMR data to design algorithms to build phage display human scFv libraries that would retain the key structural residues that would encode a library of RGD-helix hairpin structural motifs where the helices could vary in length, pitch and sequence composition. Our data show that it is possible to design scFv libraries that include structural motifs within the V H -CDR3 to provide potent ligand antagonising antibody candidates that can be developed for cancer therapy.

Library design
Our rationale was to introduce a structural selectivity to a phage display library where the three-dimensional (3D) avb6 ligand recognition motif (RGD-helix) was genetically encoded into the antibody binding pocket at the V H -CDR3. Figure 1A shows the amino-acid sequence identity and mean NMR solution structure of A20FMDV2. From these data plus the STD-NMR data [14] we designed two algorithms to develop V H -CDR3 libraries encoding a hairpin containing at its turn, an RGD motif, followed by a Cterminal a-helix ( Figure 1B) or a 3 10 -helix ( Figure 1C); the helical wheel map for each library is also shown. The template synthetic V H -CDR3 was based on the a-helix donor sequence L 8 A 9 R 10 L 11 K 12 R 13 E 14 F 15 N 16 E 17 , which is helix 1 from the Drosophila engrailed homeodomain ((EN-HD) [16]). However in the library the L 8 was changed to A 8 in order to prevent formation of an LXXLL motif. The a-helix template V H -CDR3 algorithm was: E 1 P 2 R 3 G 4 D 5 L 6 X 7 X 8 L 9 A 10 A 11 R 12 Z 13 K 14 R 15 Z 16 F 17 N 18 E 19 Z 20 L 21 A 22 Z 23 L 24 Q 25 E 26 K 27 G 28 I 29 where Z and X were random amino-acid residues introduced into the same quadrant as the leucine residues of the RGDLXXL motif: at position 13, 16, 20 and 23 in the a-helix based on the helical wheel (3.6 residues per turn of helix). The EP and the LQEKGI motifs were N-and Cterminal helix-capping regions, respectively (based on the standard Schellmann C-cap: Leu-Gln-Glu-Lys-Gly-Ile (LQEKGI) [17] ( Figure 1B). To extend the helix length further, the EN-HD donor sequence was repeated from positions 21/22 in the library. To build the 3 10 helix library we used the E 1 P 2 R 3 G 4 D 5 L 6 X 7 X 8 L 9 A 10 A 11 Z 12 L 13 K 14 Z 15 E 16 F 17 Z 18 E 19 N 20 Z 21 L 22 A 23 Z 24 L 25 Q 26 E 27 K 28 G 29 I 30 template inserting a random amino acid residue in every third position following the RGDLXXL motif in the 3 10 helix (3.0 residues per turn of 3 10 helix, Figure 1C). Hence, the random residues automatically provided helices of different lengths because they included helix stabilising residues, such as alanine, lysine or arginine or, alternatively, helix destroying residues such as proline [18] as well as including residues covering all side chain properties to test the effect of charge, hydrophobicity and steric interactions.

Library selection and screening for lead candidates
After 3 rounds of alternate panning on immobilised recombinant avb6 and cells expressing avb6, greater than two thirds of clones bound to recombinant avb6 in ELISA ( Figure 2A). Binders revealed at least 24 unique V H -CDR3 sequences corresponding to both an a-helix and 3 10 helix libraries. In flow cytometry experiments none of the scFv bound to the avb6-negative cells and many scFv bound well to the avb6-expressing cells. Since we used pair of cell lines that are isogenic and differ only by avb6 but endogenously express four other RGD directed integrins a5b1, avb3, avb5 and avb8 [19,20] these data suggest strongly that the scFvs exhibited avb6-specific binding (Fig. 2B). However, based upon a combination of scFv protein expression yields, biochemical stability (size-exclusion chromatography) and strength of binding to cellular avb6 (flow cytometry) we focussed our study on 2 different scFvs clones: D25scFv (D25) and D34scFv (D34) ( Figure 2C). V H -CDR3 sequences of both D25 and D34 encoded a-helix as follows: EPRGDLRTLAAREKRNFNETLARL-QEKGI and QPRGDLRELAARSEAQLQEKGI, for D25 and D34, respectively. Clone D34 had mutation from C to G that resulted in replacement of E to Q in the N-terminal helix-capping regions. The size-exclusion chromatography profile of the purified D34scFv showed a main peak eluting at 70 ml, corresponding to the 30 kDa scFv. For D25scFv we also observed a lower peaks eluting at 45-60 ml, corresponding to scFv dimers and tetramers, occasionally seen with some scFv [5] ( Figure 2C).

avb6-specific internalization
D25scFv and D34scFv exhibited cellular internalisation in avb6-expressing cells but not in avb6-negative cells. At 0 minutes, the scFvs were localised at the cell surface ( Figure 4A and 4D for D25scFv and D34scFv, respectively) but after incubation at 37uC for 45 minutes, they were localised within the cell cytoplasm and nucleus ( Figure 4B and 4E for D25scFv and D34scFv, respectively). Omitting the scFv primary layer and labelling only with antimyc and the fluorochrome-labelled secondary antibody showed very little nuclear staining suggesting the nuclear localisation of the scFvs was real ( Figure 4C and 4F for D25scFv and D34scFv, respectively). Similarly, biotinylated-D25p and biotinylated-D34p also underwent cellular internalisation in avb6-expressing cells ( Figure 4H and 4K for D25p and D34p, respectively) but not in avb6-negative cells ( Figure 4I, and 4L for D25p and D34p, respectively). Efficient internalisation was observed at 30-45 minutes ( Figure 4H, K for D25p and D34p, respectively) but no nuclear localisation was observed.
Biotinylated-D25 peptide localises to avb6-expressing tumours in vivo When 12.5MBq of [In111]-DTPA-Streptavidin decorated with biotinylated-D25p was intravenously injected into three mice bearing both a subcutaneous avb6-positive A375Pb6 tumour and an avb6-negative A375Ppuro tumour, on opposite shoulders, we observed a 23% injected dose per gram uptake in the avb6positive tumour compared with only 3% for the avb6-negative tumour at 1 hour post-injection, a ratio of almost 8:1 ( Figure 5). This translated into a very clear discrimination of the avb6positive tumour by single-photon emission computed tomography (SPECT) imaging ( Figure 5).

Structural determination of D25 and D34 peptides by NMR
We used NMR to solve the solution structures for peptides D34p and D25p. Figure 6 shows the 3D-rendering model closest to the mean calculated for each peptide from an ensemble of 20 NMR structures; the associated structural statistical data from CNS for both peptides is available in Table S1 and NOE and structural contact information is available in Figures S2, S3, S4. Both peptides exhibited the RGD-helix motif. D34 which has 22 amino-acids has a shorter helix than peptide D25, which has 29 amino-acids. Helices for both peptides were defined as standard ahelix with the D34 a-helix running from Leu6-Leu17 and D25 ahelix running from Leu6-Gln25.

Discussion
CDR grafting, the replacing of complementarity-determining regions in a mouse antibody with those from a human antibody, has been used for more than two decades for humanization of mouse monoclonal antibodies and development of immunotherapy [20]. CDR grafting of ligands into CDR loops to bind to a specific receptor was originally reported by Barbas et al in 1993 where they produced semisynthetic human antibodies library that included RGD motifs followed by random sequences to select for antibody fragments specific to the integrins avb5, avb3, and a I IIbb3 [21]. Kogelberg et al inserted 17 residues from A20FMDV2 into the CDR3 region into an anti-CEA scFv thereby creating an antibody with avb6-specificity [22]. More recently, a peptide sequence that bound to an inorganic material surface, was grafted into the CDR of a camel-type single domain antibody rearranged with a library of random sequences in additional CDR. Authors noted a synergistic effect from the grafted and selected random CDR loops that drastically increased the affinity for the inorganic target [23]. In this study we have taken a different approach, namely, grafting a 3-dimensional geometry based library designed from a ligand:receptor binding stereochemical interface. To test the model we chose a therapeutically valuable target, the integrin avb6 that we, and others, have Figure 1. a-helix and 3 10 -helix Library design. A) Amino acid sequence and NMR solution structure of A20FMDV2 with a hairpin structure with RGD at the tip of the turn followed by a C-terminal helix is shown. B) a-helix algorithm used to design the V H -CDR3 encoding a hairpin containing at its turn an RGD motif, followed by a C-terminal a-helix. C) 3 10 -helix algorithm used to designed the V H -CDR3 encoding a hairpin containing at its turn an RGD motif, followed by a C-terminal 3 10 -helix. Amino acid positions that are available for randomisation are highlighted (X and Z shown in blue and red, respectively). doi:10.1371/journal.pone.0070452.g001 reported is associated with poor survival from cancer, presumed to be because this integrin promotes carcinoma invasion and survival [6][7][8][9]. We had identified previously avb6-binding peptides from high affinity ligands for avb6 and shown that interrogation of the peptide structures by various NMR techniques revealed 1) all three ligands (A20FMDV1, LAP, A20FMDV2) were hairpin-shaped peptides with RGD at the turn followed by an helix and 2) the Asp+1 and Asp+4 residues were exposed on the same face of the helix and appeared to form a hydrophobic binding interface with the integrin and 3) potency of avb6 inhibition appeared to correlate with the length of the helix [14,24]. Thus we designed two algorithms to retain these structural elements while allowing for variation in amino-acid composition and helix length.
We have used our NMR data to design algorithms that would retain the key structural residues that would encode a library of RGD-helix-hairpin structural motifs where the helices would be of varying lengths and sequence composition. We used the algorithms to create two structurally-guided scFv libraries that incorporate either an aor a 3 10 -helix C-terminal to the RGDLXXL motif within V H -CDR3.
Screening of 96 clones isolated following three rounds of biopanning with the combined a-helix and 3 10 -helix libraries revealed H-CDR3 sequences of both a-type and 3 10 -type, indicating that both library designs are capable of producing avb6-binding scFv. The scFv and the V H -CDR3 derived peptides from the two lead clones, D25scFv, D34scFv, D25p and D34p: 1) bound only to avb6-expressing cells (A375Pb6) but not to cells that expressed avb3, avb5, avb8 and a5b1 (A375Ppuro) 2) exhibited dose-dependent inhibition of the avb6-specific ligand A20FMDV2 binding to cellular avb6 3) inhibited carcinoma cell avb6dependent adhesion to fibronectin 4) and were internalised into cells in an avb6-dependent manner. These characteristics make these two lead clones excellent candidates for development as therapeutic antibodies. In advance of this we tested whether the D25p possessed similar in vivo targeting capabilities that we had previously shown for A20FMDV2. Data showed that the radiolabelled D25p selectively located to avb6-expressing tumours in vivo showing that the structurally designed library could generate unique compounds with in vivo efficacy and specificity. Our attempts to radiolabel the D25scFv and D34scFv resulted in loss of antibody activity so in vivo experiments were not attempted. This will need to be addressed in future studies where radiolabeling protocols will need to be modified. Hence antibody derivatives such as diabody and intact antibody derivatives of D25 and D34 could be studied further in an in vivo setting.
Our data also show that RGD-helix structures with very long helices retain excellent avb6-specificity and function-blocking activity. Thus D25p has an a-helix almost 3-fold longer than the parental peptide A20FMDV2. Interestingly, the D25p was a more efficient inhibitor of avb6-cellular function than D34p which has a shorter a-helix, whereas there was no difference in binding of D34scFv versus D25scFv to cellular avb6 ( Figure 3E). Thus, while we have not established whether there is an optimal length of ahelix for scFv specificity for avb6, we have established that The concept introduced here can be adopted for other therapeutic targets. Current studies have identified tumour associated receptor tyrosine kinases that are being considered as potential therapeutic targets [25][26][27] and it is likely that key chemokine receptors [28] also will be similarly targeted. A similar strategy to that described here, using the 3-dimensional shape of the ligand-binding interface for these receptors, can generate receptor-selective structurally-guided scFv, peptide or any other protein scaffold libraries that would herald a new method for creating valuable therapeutic antibodies. This approach has the potential to replace the traditional approach whereby random libraries are used to developed ligand antagonists and may result in more powerful therapeutic antibodies.  Figure S1) showed any binding. To avoid repetition in the figure, the bar indicated as puro represent the binding of D25 scFv, D34 scFv which gave similar MFI when the primary scFv or peptide were omitted as the binding to puro control cells (data not shown). B) Dose-dependent inhibition of the avb6 specific binding of A20FMDV2 (A20) was demonstrated with both D25 (D25scFv) and D34 (D34scFv). A20FMDV2 binding is expressed here as a percentage of the MFI value detected in the absence of scFv. C) Dose-dependent inhibition of the avb6 specific binding of A20FMDV2 (A20) was demonstrated with D25p and D34p peptides. A20FMDV2 binding is expressed here as a percentage of the MFI value detected in the absence of peptide inhibitors. D) avb6dependent adhesion to fibronectin was similarly inhibited by D25scFv and D34scFv (p.0.05). E) avb6-dependent adhesion to fibronectin was inhibited by D25p and D34p. Significantly better inhibition of avb6-dependent adhesion to fibronectin was seen by D25p in comparison to D34p (p = 0.0156). doi:10.1371/journal.pone.0070452.g003 Scaffold Libraries for Development Bio-Therapeutic PLOS ONE | www.plosone.org

Library construction and selection
Library construction is shown in Figure 7. A pool of 50 human V H genes cloned into the pHEN1 vector was used as a PCR template [5]. The library initially was amplified with LMB3 primer (Table 1), which anneals to pHEN1 vector sequences 59 to the cloned V H gene and primer 1or 2, which anneals to 39 end of the V H gene which was composed of the frame work 3 (FR3) region, the structurally guided motifs encoded by the aor 3 10helix algorithms and finally the JH4 sequences (Table 1, Fig. 7A). In a second PCR step, XhoI restriction site was introduced 39 to the JH4 sequence after amplification with primer 1 or 2 and LMB3 ( Figure 7B). The V H gene amplicons containing inserts encoding the algorithms were then digested with XhoI and NcoI restriction enzymes and inserted into the NcoI and XhoI treated pIT2 vector containing the VL repertoire (Fig. 7C, [3]).
The a-helix and 3 10 libraries were each rescued separately KM13 helper phage as described previously [29]. For selection, we used in parallel each library separately as well as mixed alpha helix and 3 10 libraries. The initial phage selection was performed using immobilised recombinant avb6 protein as described previously [5,29] and http://www.lifesciences.sourcebioscience.com/ media/143421/tomlinsonij.pdf. After two rounds of selection on immobilised avb6 we then selected the libraries on cell-expressed avb6 using A375Pb6 and A375PPuro cell lines, for avb6-positive and avb6-negative selection, respectively [14,22]. First line screening were done by ELISA using 5 mg/ml immobilised recombinant avb6 as described [29].

Cell lines
The adherent melanoma cell lines A375Pb6 and A375Ppuro, described previously [22] were cultured in Dulbecco's Modified Eagles' Medium (DMEM) supplemented with 10% Foetal Calf Serum (FCS). The human oral squamous cell carcinoma VB6, which was engineered to express high levels of avb6, was grown in Keratinocyte Growth Medium as described [7].

Screening for integrin specificity by Flow Cytometry
Purified scFv were tested for specificity for avb6 by flow cytometry using the isogenic pairs of cell lines, A375Pb6 and A375Ppuro, or DX3b6 and DX3puro. Each pair expresses similar levels of four RGD-binding integrins (avb3, avb5, avb8, a5b1) but only A375Pb6 and DX3b6 express avb6 [14,22,30]. Briefly, cells were detached with trypsin/EDTA, rinsed twice in ice-cold wash buffer (0.1% BSA/DMEM/0.1% NaN 3 ) and approximately 2610 5 cells re-suspended in 50 ml of wash buffer per sample. In between all incubation steps, all samples were washed twice with wash buffer unless otherwise stated. Cells were incubated on ice for 1 hr with 0.1 or 1 uM purified scFv of peptide, washed twice. For scFv, cells were then incubated for 1 hr on ice with mouse anti-Myc antibody (clone 9E10, Santa Cruz) at 1:100. After washing, bound antibody was detected with anti-mouse IgG-Alexaflour 488 (1:250 for 1 h; Molecular Probes), on ice. Bound V H -CDR3 derived peptide were detected by rabbit anti-biotin IgG (1:200), followed by anti-rabbit IgG-Alexafluor 488 (1:250; Molecular Probes). Cells were analysed using the FACSCalibur (Beckton Dickinson). Cell expression of avb6 integrin was detected with mouse monoclonal antibody (clone 10D5, 10 mg/ml; Millipore) and non-specific (control) binding with non-immune class matched IgG.

Inhibition of avb6 ligand binding
The scFv lead candidates propensity to inhibit binding of the avb6-specific biotinylated-A20FMDV2 [19] to avb6-expressing cells was determined by pre-incubation with purified scFv or peptide (0.1 and 1.0 mM) for 10 mins, immediately followed by 40 mins incubation with the biotinylated A20FMDV2 (10 nM). Figure 7. Cloning strategy used to build the scFv phage display libraries with synthetic structurally shaped V H -CDR3 containing the aand 3 10 -helix. A. pHEN1 plasmid backbone containing scFv was used as the PCR template gene to be amplified with the primers LMB3 and the Primer 1 or 2 that includes the FR3 sequences (yellow), flanked by the aor 3 10 -helix algorithms (red) and followed by JH4 sequences (brown). B. After PCR amplification the resulted V H gene segment was rearranged with aor 3 10 -helix V H -CDR3 (red). To incorporate a unique Xho1 restriction site, the PCR products derived from step A were amplified with LMB3 and Primer 3. C. The resultant PCR products from step B were digested with Nco1 and Xho1 for insertion into the pIT2 plasmid backbone that contain light chain repertoire. doi:10.1371/journal.pone.0070452.g007

Fast protein liquid chromatography (FPLC)
The structural stability of scFv proteins was assessed by gel filtration chromatography using a HiPrep 16/60 sephacryl S-200 (GE healthcare, Amersham, UK) connected to an AKTA FPLC (GE healthcare, Amersham, UK). Fractionations were performed in PBS at a flow rate of 0.3 ml/min with detection at 280 nm. A Calibration Kit (GE healthcare, Amersham, UK) containing protein markers (aprotinin, ribonuclease A, carbonic anhydrase, ovalbumin and conalbumin) was used to generate a calibration curve. Equal amounts of each of the protein markers were prepared to 500 ml in PBS prior to loading onto the column. Kav values derived from the equation: Kav = (V e -V 0 )/(V c -V 0 ); where Ve = elution volume, Vc = geometric column volume, and V0 = column void volume were plotted against log molecular weight to generate a protein calibration curve. Purified scFv (50 ug/500 ul) was loaded onto the gel filtration column and molecular weight of the different peaks was determined using the calibration curve.

Internalisation Assay
In 24-well culture plates, 2610 4 cells avb6-positive andnegative cells were seeded onto 13 mm diameter glass coverslips and allowed to incubate overnight at 37uC in growth medium. Cells were washed twice in serum free medium (SFM) and then scFv (4 uM) or biotinylated-CDR3-peptide (100 nM) diluted in SFM added on ice for 10 mins. After two ice-cold washes and a further 109 on ice with mouse anti-biotin antibody (10 ug/ml) prewarmed media containing 10% FCS was added to the cells and incubated at 37uC. At 10 min intervals between 0-60 mins, cells were fixed in 2% HCHO in PBS and permeabilised with 0.1% TritonX-100 (PBS) for 3 mins. scFv was detected with mouse anti c-myc (1:100 dilution, clone 9E10; 309, ambient temperature) while peptides were detected by rabbit anti-biotin IgG (1:200). Both peptide and scFv detected with anti-mouse Alexa488 for 309. Nuclei were counter-labelled with 49,6-diamidino-2-phenylindole (DAPI), coverslips mounted with Mowiol and samples examined by confocal microscopy (Zeiss LSM510, Welwyn Garden City).

Picogreen adhesion assay
All adhesion assays were performed in quadruplicate and repeated at least 2-3 times. In a 96-well plate format, test wells were coated with Fibronectin (10 ug/ml/PBS) and negative control wells were coated with 0.1% BSA (bovine serum albumin)/PBS. After incubation at 37uC for 1 hour, plates were washed in PBS twice and blocked with 0.1% BSA/PBS for 30 mins at 37uC. Following a PBS rinse, cells were seeded into the wells, the plate resting on ice to avoid dehydration and to ensure an initial uniform temperature for the experiment. Purified scFv or peptide was added (25 ul) to the wells at the desired concentrations, before adding 1.5610 4 cells (25 ul) per well. In some experiments, cells (1.5610 4 ) were pre-treated with b1 blocking antibody AIIB2 at 10 ug/ml before seeding into test wells to block the b1-dependent fibronectin adhesion. To determine percentage adhesion, standard curves were generated by plating 0-2.5610 4 cells in separate wells. After incubating plates at 37uC for 309 plates were washed twice in 1 mM CaCl 2 /0.5 mM MgCl 2 /PBS and transferred into 280uC for 15 mins. Adherent cells were quantified using a Picogreen kit (Invitrogen) and analysed on a fluorescence reader (FLUOstar Optima, BMG Labtech Ltd, Bucks, UK).

In vivo localization
All animal procedures followed strict Home Office (UK) guidelines under license number PPL 80/2279. Biotinylatedpeptide was radiolabelled with Indium-111. To 10 ug of DOTA-Biotin (Macrocyclics.com #C-100) buffered in 1M Ammonium Acetate (pH 5.5) was added Indium [ 111 In]-acetate. The mixture was heated to 80uC for 30 mins and cooled at RT for 10 mins. The labelled DOTA-Biotin was added to streptavidin at a 1:1 Molar Ratio (1 mg of streptavidin per 17.6 ug of DOTA-Biotin). To this mixture, 88 ug of botinylated-D25p was added and 10 ul of the mixture analysed by size exclusion-HPLC to verify the stability of radiolabelled products pre-and post-labelling (data not shown). A total of 50MBq was used to label 25 ug of peptide; the sample was divided into four such that each mouse received 12.5 MBq (6.25 ug) of 111 In-labelled D25 peptide.
Female CD1 nu/nu athymic nude mice (Charles River) were subcutaneously injected with 100 ml (2610 6 cells) of A373Pb6 into the right shoulder and A375puro in the left shoulder. Tumours were allowed to develop for 20 days and 200 ml of freshly radiolabelled 111 In-labelled D25 peptide was administered intravenously. Tumours were imaged by NanoSPECT/CT (Bioscan, Inc) at 1 hr, 4 hr and 24 hr-time points as follows: Mice were placed onto the imaging bed and initial low resolution CT scans were collected at 45KVp, 180 projections per rotation, 500 ms per projection. Subsequently SPECT images were acquired (45 minutes acquisition time) and data reconstructed using on-board HiSPECT-NG software (Bioscan). To measure radioactivity associated with tumours the NanoSPECT/CT machine was calibrated (before the experiment) by imaging a phantom with an Indium-111 standard solution. Subsequently Invivoscope software (Invicro) was used to generate three-dimensional regions of Interest (ROI) around the tumours and the ROIs converted to megabequerels. Structural determination by NMR study Solution NMR structures of peptides D25p and D34p on 0.1%TFE were solved as previously (13). All NMR data for peptides A22 (D25p) and A29 (D34p) were obtained at 283 K from a 14.1 T (600 MHz 1 H) Bruker Avance III NMR spectrometer equipped with a 5 mm QCI-F cryoprobe. All NMR samples were 350 mL within a Shigemi NMR tube and contained 1 mM peptide in 25 mM sodium phosphate buffer at pH 6.5 also containing 50 mM sodium chloride, 4% (v/v) dimethylsulfoxide (DMSO), and 30% (v/v) trifluroethanol-d3 (TFE). NMR data processing was completed using TopSpin 3.1 (Bruker), assignments were completed using CCPN Analysis. 1 H chemical shifts and through-space structural assignments were obtained from two-dimensional TOCSY and NOESY NMR experiments with mixing times of 20 ms/60 ms for TOCSY and 250 ms for NOESY. The observed NOE contacts support the presence of an a-helical conformation along the length of the peptide with NOEs observed between Ha and HN (i-i+3) as well as Ha and Hb (i-i+3). Structural ensembles were calculated using CNS [31] and including dihedral angles confirmed by DANGLE analysis [31] and predicted hydrogen-bond donor acceptor pairs. The final ensemble was water-minimised using YASARA Structure software (available from http://www.yasara.org) and Ramachandran analysis of each peptide ensemble was completed using PROCHECK-NMR [32].

Statistical analysis
The Wilcoxon matched-pairs signed rank test was used to compare activity. Statistical analysis was performed using the GraphPad Prism software package (GraphPad Software, San Diego, CA).