1961

ClinicalChemisty 42:121961-1969 (1996) Enzyid1

Epitope mapping of prostate-specific antigen withmonoclonal antibodies

DIANE C. JETTE,’ FERNANDO T. KREUTZ,1 BRUCE A. MALCOLM,’- DAVID S. WISHART,2

ANTOINE A. NOuJAIM,3 and MAVANUR R. SURESHI*

Prostate-specific antigen (PSA) is a widely used marker forscreening and monitoring prostate cancer. We identified

and characterized the epitopes of two anti-PSA monoclonal

antibodies (niAbs) designated B80 and B87. The epitopeswere initially mapped as nonoverlapping by developing a

sandwich inimunoassay to measure PSA with the two anti-

PSA mAbs. The two antibodies do not cross-react withhomologous pancreatic kallikrein, but recognize epitopes

unique to PSA. B80 and B87 can recognize both free andcomplexed PSA and hence measure total PSA. Epitopescanning and bacteriophage peptide library affinity selec-tion procedures were used to identify and locate an epitopeon PSA. A possible epitope for B80 was identified as being

located on or near PSA amino acid residues 50-58 (-GRH-

SLFHP-). The epitope for B87 was likely on an exposednonlinear conformational determinant, unique to PSA, and

not masked by the binding of B80 or a1-antichymotrypsin.

INDEXING TERMS: immunoassay #{149}prostate cancer . bacterio-phage peptide library

Prostate-specific antigen (PSA) is a 33-kDa single-chain glyco-

protein produced by prostatic epithelium.4 The primary struc-ture of PSA was first described by Watt et al. [1] and shows a

high degree of homology with serine proteases of the kallikrein

family [2, 3]. PSA is a major protein in seminal plasma and shows

chymotrypsin-like substrate specificity [4-6/. It is normally

Faculty of Phannacy and Pharmaceutical Sciences, and 2 1)eparrment of

Biochemisny, Faculty of Medicine, University of Alberta, Edmonton, AB, Canada

T6G 2N8.Biomira Research Inc., Edmonton, AB, Canada.

Author for correspondence. Fax 403-492-8241; e-mail rnsuresh@gpu.srv.

ualberta.ca.

Nonstandard abbreviations: PSA, prostate-specific antigen; BPII, benign

prostadc hyperphasia; inAb, inonoclonal andbody; hK2, human glandular kal-

hikrein; hKh, human pancreatic kallikrein; AMG, a2-macroglohulin; ACT, o,-

antichymotrypsin; I-IRPO, horseradish peroxidase; hsmAb, bispecific niAb; BTS

2,2-azino-bis(3-ethylbcnz-thiazoline-6-suhfonic acid); SDS, sodium dodecyl sul-

fate; PAGE, polyacrylamide gel electrophoresis; TMB, tetramethylbenzidine;

PBS, phosphate-buffered saline; BSA, bovine serum albumin; pKl, porcine

pancreatic kalhikrein; and TBS, Tris-buffered saline.

Received January 16, 1996; revised June 25, 1996; accepted June 25, 1996.

found in low concentrations (<2.5 jigfL) in male blood plasmabut can be increased in conjunction with prostate cancer, benignprostatic hyperplasia (BPH), and surgicaltrauma to the prostate

[7-9].Monoclonal antibody (niAb)-based RIAs and ELISAs are

used to measure PSA in serum. The results are used to screen for

prostate cancer and monitor patients during treatment[8, 10, 11].

The PSA gene is a member of the human tissue kallikrein

gene family referred to as serine proteases because the mecha-nism of proteolytic cleavage involves a serine residue at the

active site [1, 12, 13]. PSA is a 6-kb gene product of chromo-

some 19 in the region of q 13.2-q 13.4 with 4 introns and 5 exons[14]. It has a >84% nucleotide sequence homology with humanglandular kallikrein (hK2), and 73% homology with human

pancreatic kallikrein (hKl), indicating a common ancestral gene[3, 15-18]. Human serum kallikrein is not a member of this gene

family and does not have a high degree of sequence homology to

PSA. The amino acid homology with porcine pancreatic kal-

likrein(pKl) is59.6%. Previously all three gene products were

renamed [15, 17], as shown in Table 1.

In serum, PSA can form stable complexes with two majorserum protease inhibitors, a,-macroglobulin (AMG) and a1-

antichymotrypsin (ACT) [20-22]. When PSA iscomplexed to

AMG it is thought to be completely encapsulated with noepitopes accessible for immunodetection [23]. However, when

PSA is complexed to ACT, some epitopes proximal to the activesite are masked while others are available for binding to

anti-PSA antibodies [21]. Immunoassays involving antibodies

directed against PSA epitopes that are masked by ACT will failto detect this complex. The proportion of serum PSA corn-plexed to ACT has heen found to be significantly higher withprostate cancer than with BPH. The ratio of free to complexed

PSA, therefore, may be useful in distinguishing between pros-tatic cancer and BPH [21-23]. It is important, therefore, to

determine if antibodies used to measure PSA in serum arerecognizing free PSA, PSA-ACT, or both. In addition, several

authors have recently discussed the importance of standardizingPSA assays, because many of the commercially available assays

yield different PSA values [24-30]. In this paper, we attempt toidentify and characterize the epitopes that are recognized by two

anti-PSA mAbs.

1962 Jetteet al.:PSA epitopemapping

Table 1. Nomenclature of kallikrein family proteins.aFormal name Commonname Description

hKl PRK Human pancreatic/renal kallikrein,hKLK1 gene product

hK2 hGK-1 Human glandular kallikrein, hKLK2gene product

hK3 PSA PSA, hKLK3 gene product

pKl Porcine pancreatic kallikrein

Modified from Mccormack et al. [19].

Materials and MethodsMaterials. Anti-PSA mouse IgG mAbs B80 and B87 were kindly

provided by Biomira, Edmonton, AB, Canada, and the anti-PSA X anti-peroxidase (HRPO) bispecific mouse IgG mAb

(bsmAb B8OXHRPO) was developed in our laboratory [31].Briefly, a quadroma or hybrid hybridoma was generated by

fusing the B80-secreting hybridoma with an anti-HRPO-secret-ing hybridoma. The bsmAb was purified by DE-52 gradient

chromatography. This bsmAb has one Fab arm with anti-PSAspecificity and another Fab arm with anti-HRPO specificityengineered in one molecule. Standard media used for cellculture, RPMI-1o40 supplemented with 2 mmnol/L L-glutamine,50 kilounits/L penicillin, 50 mg/L streptomycin, and 50 mL/L

fetal bovine serum, were obtained from Gihco BRL, Gaithers-burg, MD. Anti-mouse and anti-rabbit IgG-peroxidase conju-

gate, cyanogen bromide-activated Sepharose 4B, pKl, kallikreinfrom human serum, and 2,2’-azino-bis(3-ethylbenz-thiazoline-

6-sulfonic acid) (ABTS) were obtained from Sigma, St. Louis,MO. Maxisorp” 96-well immunoplates were obtained fromNunc, Roskilde, Denmark. PSA calibrator was obtained from

Scripps Labs., San Diego, CA. ACT from human plasma andrabbit anti-ACT IgG were obtained from Calbiochem, La Jolla,CA. Molecular-mass markers for sodium dodecyl sulfate-poly-acrylamide gel electrophoresis (SDS-PAGE), precast 4% to15% gradient polyacrylamide gels, SDS buffer cartridges, andthe Phast electrophoresis system were obtained from Pharma-

cia, Uppsala, Sweden. The Mimotope kit, used to synthesize theiinniobilized hexapeptides, was obtained from Cambridge Re-

search Biochemicals, Cheshire, UK. A Vmax microplate reader

and MAX Line plate washer from Molecular Devices, MenloPark, CA, was used. Escherichia coli strain K 91 (Hfi-C thi) was

used in conjunction with the bacteriophage (phage) decapeptide

display library [32J. Dideoxmucleotide sequencing was per-formed with the Sequenase 2.0 system obtained from UnitedStates Biochemicals, St. Louis, MO. One Step TMB (tetram-ethylbenzidine) substrate was obtained from Pierce Chemical

Co., Rockford IL.

Purification of PSA. A human prostate cancer cell line (LNcap)was used as a source of PSA. The cells were grown (standardmedia, 275 mm2 tissue culture flask) and the supernatant was

collected and pooled every 3 to 5 days. The concentration ofPSA in the unpurified supernatant was determined to be 0.9-1.5irmg/L by ELISA. PSA was purified from I L of cell supernatant

by using an affinity column with immobilized anti-PSA antibod-

ies, B80 and B87. The column was made by first washing

Sepharose 4B (1 g in 3.5 mL, cyanogen bromide activated) with2 volumes of buffer (0.5 mol/L phosphate, pH 6.8). A mixture of

B80 and B87 [10 mg in 3 mL of phosphate-buffered saline(PBS), pH 7.21 was dialyzed overnight against three changes of

buffer. The antibodies were added to the Sepharose gel and

gently agitated at room temperature. The amount of antibodyremaining in solution was determined by monitoring the reac-

tion mixture absorbance at 280 nm. The reaction was stoppedwhen -90% of the antibody was bound. The column was then

washed twice with PBS (10 mL), and ethanolamine (10 mL of100 ninol/L in distilled water) was added to block any unbound

reaction sites on the Sepharose (2 h at room temperature).Before use, the mAb affinity column was washed twice (10

mL of 10 mmollL phosphate buffer, pH 6.4). Cell supernatant

containing 1.5 mg/L PSA (1 L) was cycled for 24 h through thecolumn with a peristaltic pump (1.7 mL/min) so that the totalvolume of supernatant passed through the column three times.

After washing the column with buffer, the PSA was eluted with

glycine (100 mmol/L, pH 2.5). Column fractions (0.5 mL) were

collected into tubes containing buffer (50 .tL of 1 mol/L

phosphate, pH 8.0) to neutralize the eluates. The proteinconcentration of each fraction was determined by measuring theabsorbance at 280 nni and the appropriate fractions werepooled. After pooling, an ELISA was performed as describedbelow to determine the concentration of purified PSA. The

column, which can be reused, was washed (1 g/L sodium azide

in 10 mmol/L phosphate) and stored at 4 #{176}Cuntil needed. PSA

eluted from the monoclonal affinity column in fractions 3 to 5and a total of 1.37 mg of PSA was recovered from 1.5 L of

supernatant. No remaining PSA was detected in the supernatantafter affinity purification.

Preparation and analysis of PSA -ACT complex. The complex was

prepared by using the method described by Lilja et al. [22].Purified PSA (7 g) was added to purified ACT (50 g) in

reaction buffer (50 iL of 50 mmol/L Tris-HC1, pH 7.8, with

0.1 molJL NaC1) and incubated for 6 h at 37 #{176}C.The degree of

complex formation was analyzed by SDS-PAGE. A Phast elec-trophoresis system was used to analyze free and complexed PSA.

A precast aciylamide gel (4% to 15% gradient) was run under

standard conditions (10 mA, 250 V in SDS-barbital buffer) as

per the manufacturer’s instructions. Samples containing 4 [IL of

purified PSA, complexed PSA, unbound ACT, and low-rangemolecular-mass markers were analyzed. SDS-PAGE could ef-

fectively resolve PSA at -30 Wa, ACT at -60 Wa, andPSA-ACT complex at -90 Wa. A small amount of free PSAwas seen on the electrophoresis gel after complex formation.The PSA-ACT complex immunoassay used in this study will not

detect free PSA or ACT.

MEASUREMENT OF PSA AND PSA-ACT BY ELISA

PSA sandwich ELISA. We previously developed two mouse

mAbs, B80 and B87, using human PSA isolated from seminal

plasma. These two antibodies were selected for their ability toform an efficient sandwich assay for PSA in serum. The

antibodies have high affinity for PSA (Kd -2 X o o mol/L forboth) and the assay exhibited a good correlation (r = 0.98) with

50 100 150 200 250

Antibody concentration (pg/L)

Clinical Chemistry 42, No. 12, 1996 1963

the Hybritech PSA Tandem RIA with 154 prostate cancer

serum samples (Krantz MJ and Suresh MR, manuscript inpreparation). A PSA sandwich ELISA was developed with mAbB87, as the capture antibody, immobilized on plastic in a

microtiter plate. The bsmAb B8OXHRPO was used as the tracer

antibody [31].Microtiter wells were coated with mAb B87 (1 tg

in PBS, 16 h, 4 #{176}C).The plate was then blocked [10 g/L bovine

serum albumin in PBS (BSA-PBS), 1 h, 37 #{176}C}.Samples orcalibrators (50 [IL) were added to each well along with bsmAb

B8OXHRPO (10 ng) and HRPO (300 ng in 50 [IL of PBS). Theplate was incubated with shaking (30 mm, room temperature).

After washing [3 x 200 [IL of PBS containing 0.5 mL/L Tween20 (PBS-Tween)], ABTS substrate (100 [IL of 0.5 g/L ABTS,

0.5 [IL of 30% H,O,, in 0.1 mol/L citrate with 0.1 mollLphosphate, pH 4.0) was added to each well and the amount of

PSA present in the saIiple was determined by measuring theabsorbance at 405 nm with a microplate reader. The assay wascalibrated for PSA concentration with 99% pure PSA obtained

commercially. The calibrators were diluted in BSA-PBS (andnot serum). This assay was used to determine the PSA concen-

tration in cell supernatant and PSA affinity-purified eluates. Thedose-response curve, shown in Fig. 1, is linear from 0 to 75[Lg/L PSA and is not saturated even at 250 [Ig/L PSA. Each

point is an average of duplicate measurements and the SEM isindicated. Further optimization and the clinical utility of this

assay are currently being evaluated in our laboratory [31 and

work in progress]. Another experiment directly compared thequantitative recovery of fixed samples of PSA ranging from 3 to

100 [Ig/L with and without ACT in molar ratios of 1:0, 1:0.5,1:1, and 1:5. The absorbance values obtained at 100 [Lg/L were

1.286, 1.210, 1.296, and 1.206 and at 6 [Ig/L were 0.107, 0.086,

0.112, and 0.097 for the respective ratios. The absorbance values

similarly were within experimental error for 3, 12, 25, and 50

[Ig/L PSA with and without ACT in the above ratios.

2.5

2.0

1.5

10

0.5

0 50 100 150 200 250

PSA (pg/L)

300

Fig. 1. Dose-response curve for a PSA sandwich ELISA.mAb B87, immobilized on the plate, was used as the capture antibody andbsmAb B8OXHRPO was used as the tracer antibody. Pure PSA, diluted in 10 g/LBSA-PBS, was used to calibrate the assay for PSA concentration. The assay wasperformed as described in Materials and Methods.

PSA -ACT ELISA. Microtiter wells were coated with either B80

or B87 (1 g in PBS, overnight, 4 #{176}C)and blocked as above.

PSA-ACT complex was titrated from 100 ng/well (100 [IL

BSA-PBS) in twofold serial dilutions. The plate was incubated

(shaking, 1.5 h, room temperature) and then washed. Anti-ACT

rabbit IgG polyclonal antisera was diluted (100 [IL, 1:1000

dilution in BSA-PBS) and added to each well. The plate was

again washed. Anti-rabbit IgC-HRPO was diluted as above and

added to each well. The plate was incubated for I h and washed.

ABTS substrate was added and the amount of immobilized

PSA-ACT complex present in the sample was determined by

measuring the absorbance at 405 nm.

Cross-reactivity analysis. Human serum kallikrein and pKl were

immobilized on plastic microtiter wells. B80 or B87 was incu-

bated with the immobilized proteins to evaluate their cross-

reactivity. The binding of these antibodies to immobilized PSA,shown in Fig. 2, reaches saturation in this assay at -12 [Lg/L for

B80 and 50 [Ig/L for B87. No binding to either of the kallikreins

was observed for B80 or B87 up to a concentration of 1 mg/L.

Purified PSA was used as a positive control and BSA was used as

a negative control. Microtiter wells were coated with either

purified PSA, hK2, or pKl (0.5 g, PBS, overnight, 4 #{176}C).Theplates were blocked, and B80 or B87 was titrated from 1 [Ig/well

in BSA-PBS in twofold dilutions. The plate was incubated for

3 h, then washed. Anti-mouse IgG-HRP() was diluted as above

and added to each well. The plate was incubated for I h and

washed. ABTS substrate was added and the amount of antibody

binding to kallikreins or PSA was determined by measuring the

absorbance at 405 nm.

EC

0

0

Fig. 2. Comparison of the binding of mAbs B80 and 687 to PSA.glandular (or human serum) kallikrein, or pKl in an immunoassay.PSA or kallikrein was immobilized on the plate. After incubating with B80 or B87.the plate was washed to remove unbound material, and anti-mouse lgG-HRPOconjugate was used as the secondary antibody. Neither B80 nor B87 cross-reacted with either kallikrein. The assay was performed as described in Materialsand Methods.

1964 Jette et al.: PSA epitope mapping

SYNTHESIS AND SCREENING OF PSA HEXAPEPTIDES

The 240 overlapping hexapeptides representing the entire PSA

sequence were synthesized on pins with the Mimotope kit.

Standard solid-phase peptide synthesis with Fmoc-amino acid

active esters was used to build each hexapeptide on amino-functionalized pins. The peptides were acetylated at the N

terminus. The pins with the immobilized synthetic peptides can

be regenerated and retested several times.A standard ELISA was used to screen the PSA-derived

peptides for binding to B87 and bsmAb B8OXHRPO. The pins

were blocked in a 96-well microtiter plate. For screening with

B87, antibody (100 ng/well, 200 [IL of BSA-PBS) was added to

each well of a microtiter plate and the pins incubated in thissolution for 3 h and washed. Anti-mouse IgG-HRPO was

diluted and added to each well of a microtiter plate. The pinswere again incubated and washed. ABTS substrate was added

and the amount of antibody binding to immobilized peptide was

determined by measuring the absorbance at 405 nm. For

screening with B80, bsmAb B8OXHRPO (10 ng) and HRPO(300 ng in 200 [IL of PBS) were added to each well. The pins

were incubated for 1 h and subsequently washed. ABTS sub-strate was added and the absorbance at 405 nm was measured.

AFFINITY SELECTION OF B80 AND B87 SPECIFIC PEPTIDES

FROM A RANDOM PHAGE PEPTIDE DISPLAY LIBRARY

An aliquot of a bacteriophage peptide library displaying random

decapeptides [32] [4.4 X 10’’ colony-forming units in 40 [IL of

BSAJTris-buffered saline, pH 7.2 (BSA-TBS)I was added to 40

g of immobilized B80 and B87 (10 [IL of a 50% slurry of

antibody-linked Sepharose 4B, described above). The sample

was gently mixed (room temperature, 1.5 h) and the beads were

collected by ccntrifugation (30 s, 4000g). The supernatant was

removed and the beads resuspended, washed with 500 [IL ofBSA-TBS (5 nun), and recovered by centrifugation. This wash

was repeated two times with BSA-TBS, three times with 0.5

mL/L Tween 20 in TBS (TBS-Tween), and once with BSA-TBS. The washed beads were then incubated with PSA (1 h,

room temperature with shaking, 3 g in 50 [IL of BSA-TBS).

The supernatant, designated as ligand eluate, was collected and

the beads washed as before with 500 [IL of 10 mmollL citrate

buffer, pH 2.0. After 5 mm, the supernatant, designated as the

acid eluate, was collected. Both the ligand and acid eluates wereamplified by infecting E. co/i cells with isolated phage, following

procedures described in Christian et al. [32]. A second round ofselection was performed as above.

IMMUNOLOGICAL SCREENING OF AFFINITY-SELECTED PHAGE

Affinity-selected phage particles were screened in an in situ

immunological assay with phage-infected colonies [32]. Phage

clones from the eluates were used to infect fresh cells. The cellswere then grown in selective media (15 X 150 mm plates, LB

medium with 20 mg/L tetracycline) at a density of 200 to 500

colonies/plate and transferred to nitrocellulose membrane fil-

ters. For screening with B80, the filters were washed (30 mm,TBS-Tween), blocked (30 mm, 30 mL/L BSA-PBS), and thenincubated with bsmAb B8OXHRPO (20 g, with 1 mg of

HRPO). The filters were washed again and incubated with

TMB substrate for I h. The filters were then washed with

distilled water and allowed to air dry overnight. Alternatively,

filters with colonies from a second set of plates were incubated

with B87 (1 mg in 100 mL of BSA-PBS) for 2 h at room

temperature. After washing, the B87 filters were incubated with

anti-mouse-HRPO (1:1000 dilution in BSA-PBS, I h, room

temperature). The filters were washed three times and incubated

with TMB substrate. After 1 h the filters were washed with

distilled water and allowed to air dry overnight. Positive clones

were picked from the plates and amplified for sequencing.

Dideoxynucleotide sequencing was performed by using the

Sequenase 2.0 system according to the manufacturer’ s instruc-

tions.

MODELING OF PSA

A homology model of PSA was generated with INSIGHT II

molecular modeling package (Biosym Technologies, San Diego,

CA). The model was prepared by initially aligning the human

PSA sequence against those of both porcine kallikrein (59%

sequence identity) and rat tonin (54% sequence identity) by

using structural alignment routines in the SEQSEE software

[33] package (Biotools, Edmonton, AB, Canada). On the basis ofthese two alignments and on the quality of the available

structural data, the x-ray structure of rat tonin (Brookhaven

Protein Datahank accession code: ITON) was selected as the

three-dimensional template for homology modeling the PSA

structure. By using the sequence alignment produced by SEQ-

SEE, the amino acid sequence of PSA was then substituted into

that of tonin by using the Biopolymer module of INSIGHT II.

The model was subjected to a brief period of conjugate gradient

energy minimization to reduce unfavorable van der Waals

contacts and torsional strain.

Results

ANALYSIS OF PSA-ACT COMPLEX

An immunoassay, shown in Fig. 3, that is specific for the

PSA-ACT complex was developed by using mnAbs B80 or B87

imnmobilized on microtiter plates and rabbit anti-ACT poly-

clonal antibody. Goat anti-rabbit antibody conjugated with

HRPO was used as the signal-generating antibody. Free PSA or

ACT was not detected in this assay format. Goat anti-rabbit-

HRPO conjugate and rabbit anti-ACT antibody did not bind toimmobilized mAbs B80 or B87 (data not shown). Twofold

dilutions of PSA-ACT from 500 [Ig/L PSA were made (the

concentration of analyte is expressed in [Lg/L coinpiexed PSA).The dose-response curves were linear up to 50 [Ig/L and

saturated at -500 [Ig/L PSA-ACT complex. As little as 20 [Ig/L

of complexed PSA could be detected in this unoptimized assay.

Because both antibodies recognize the complex, we can con-

clude that both epitopes are not on or near the region that binds

to ACT. Another experiment directly compared a fixed sample

of PSA with and without ACT in ratios of 1:0.5, 1:1, and 1:5.

Here again no differences were observed in the PSA values

obtained, further confirming the PSA specificity of the mAbs.

0 100 200 300 400 500 600

Antigen (pg/L)

0CO CO CCC’)) CD C’) 0 F.. CO CO CCC’)))’) C, OF.. .- CO CO CCC’)) CD C’) OF.. COO CCC’))

CCC CCC C’) ‘ CO CO CON F- CO 0)0)0 CCC CCC C’) CO CO CD F.. COO) 0)0,- CCC C’) C’)CCC CCC CCC CCC CC) C’)

Peptlde number

Clinical Chemist-ty 42, No. 12, 1996 1965

Fig. 3. Detection of PSA-ACTcomplex by B80 and 687 in a sandwichformat immunoassay.In this assay, mAb B80 or B87 was used to capture PSA or PSA-ACTcomplex.Rabbit anti-ACT serum was used to detect any immobilized PSA-ACTcomplex.Both B80 and B87 were able to bind to PSA when complexed to ACT. Free PSAor ACT will not be detected in this assay. The difference in sensitivity of thisassay to PSA and PSA-ACTcomplex is due to differences in reagents and assayformat. The PSA assay involves two mAbs. whereas this assay involved rabbitanti-ACT polysera and anti-rabbit polysera to detect the complex. The concentra-tion of complex is expressed as g/L of PSA. The assay was performed asdescribed in Materials and Methods.

EPITOPE SCANNING WITH PSA-DERIVED HEXAPEPTIDES

To further identify the PSA epitopes bound by the two mAbs,

we synthesized all the overlapping hexapeptides derived fromthe linear amino acid sequence on PSA. Fig. 4 shows the results

of the screening of PSA sequence-derived hexapeptides with

bsmAb B8OXHRPO. The PSA peptide number represents theposition of the first amino acid in the hexapeptide relative to the

PSA amino acid sequence. The data shown was not background

subtracted and illustrates the low background signal from thehexapeptides, which do not bind to the bsmAb. In contrast, themonospecific B80 antibody binding measured by a goat anti-

mnouse-HRPO or biotinyl B80 along with streptavidin-HRPOgave high backgrounds and numerous peptides identified as

(false) positives due to the several complex secondary polyclonalreagents used. Four potential epitopes, exhibiting strong bsmAb

binding, were identified by this method: peptide 2 1-27

(RGRAVC), peptide 30-36 (\TLVHPQ), peptide 53-59 (RH-SLFH), and peptide 195-20 1 (PLVCNG). A few other peptidesshow moderate reactivity. No significant binding to PSA-

derived hexapeptides was seen for B87 in this assay.

AFFINITY SELECTION OF B80- AND B87-SPECIFIC PEPTIDES

FROM A PHAGE PEPTIDE DISPLAY LIBRARY

A phage decapeptide display library [32] was used (4.4 X 10”particles in round 1) to identify peptides that would specifically

bind to B80 or B87 and mnimic PSA epitopes. The Sepharose

beads with immobilized antibody were used for the affinityselection, and the elution conditions were the same as those used

for the affinity purification of PSA. This column, with 133 pmol

of immobilized antibody, was able to bind 40 pmol of PSAwithout saturating (molar ratio of antibody to PSA of 3.3 to 1).

The molar ratio of antibody to phage was -70 to 1. The phagewas eluted by incubating resin-bound phage with PSA and acid

and separated by centrifugation, providing a yield of 3.1 X i#{248}

and 5 x l0 particles, respectively. The molar ratio of the

immobilized antibody to the PSA used to elute the phage was-2.7 to 1. Functionally, this ratio would be much closer,since

not all antibody combining sites would be available for PSAbinding because CNBr-based coupling is a random linkage of

macromolecules to the solid phase. In addition, the affinity of

the mAbs for PSA is high (i.e., Kd -2 X 10m0 molJL). Themolar ratio of PSA used to elute phage was 25 to 1. The ligand

eluate was then amplified and reselected (2 X l0 particles as

1.2

O.8C’)0

0.6C

20.4

0.2

Fig. 4. Epitope scanning: binding of bsmAb B8OxHRPO to PSA amino acid sequence-derived hexapeptides.Hexapeptides representing the entire PSA amino acid sequence were synthesized on derivatized pins. This histogram shows several PSA.derived peptides that boundB80 antibody.The assaywas performedas described in Materials and Methods.

1966 Jette et al.: PSA epitope mapping

‘)Frequency of appearance of each sequence from among those sequenced.

Underlined sequences are consensus sequences; these indicate motifs that the selected sequences have in common.

input in round 2) with ligand and acid to generate 2.3 X iO and7 X 102 particles, respectively. We chose to elute the phage

particles from immobilized antibody with ligand (PSA) and acidto selectfor PSA-specific sequences [32]. After the secondround, the yield increased by 140-fold for the ligand elution,

whereas a threefold increase was seen with the acid elution,

suggesting that antibody-binding phage had been selected. The

ligand eluates gave the highest enrichment, indicating that acideluate was less specific. Enrichment was calculated as the total

yield of phage from the affinityselection elution from round 2divided by total yield of phage from the affinity selection elution

from round I.

To selectsequences that were specificto B80 or B87, an

immunoblotting assay was performed on nitrocellulose lifts of

colonies of cells infected with affinity-selected phage. The

fraction obtained by eluting with acid in both rounds ofselection did not yield any antibody-specific phage clones.B80-positive clones were seen for both eluants with ligand in the

first round. B87-positive clones were only found in the eluant,which was eluted with acid in the first round and ligand in the

second round.

IMMUNOSCREENING OF PHAGE DISPLAYING B80- AND B87-

SPECIFIC PEPTIDES IDENTIFIED BY AFFINITY SELECTION

To identify the phage specifically binding to B80 and to B87, anin situ immunoassay was performned. Phage eluted from the

affinity beads were grown in E. co/i as individual colonies on agar

plates. A nitrocellulose filter was used to lift the phage-infected

colonies onto a solid support [32]. After washing, phage particlesbound to the membrane were screened for binding to the twomAbs in a standard immunoblot assay. Two sets of plates and

filters were prepared; one set was screened with B80 and the

other with B87. Table 2 shows the number of positive clones foreach eluate. Colonies were selected for DNA sequencing on thebasis of the intensity of the immunoblotting results. Screeningwith B87 revealed 24% positive colonies with the phage that

were eluted first with ligand and second with acid. The other

phage eluants did not bind to B87.

The resulting peptide insert sequences are shown in Table 3along with the frequency of appearance of each sequence from

among those sequenced. Although it is tempting to speculate

that higher frequency of a particular sequence is indicative ofpreferential binding relative to other sequences, such a conclu-

sion may be premature and beset with problems of genetically

identical siblings picked randomly in the screening step. In all,66 clones were sequenced. Many gave the identical DNA

sequence, indicating that they were copies of the same clone.

Five unique sequences were identified for B80 and eight for B87.

Consensus sequences in the Table are underlined and indicate

motifs that the selected sequences have in common and requirefurther study that they are equivalent to binding motifs. The

most interesting finding was that one consensus sequence wasfound to be homologous to a linear sequence on PSA, peptide

51-55 (-LGRHS-). This sequence closely matches the B80-

specific peptide sequence 53-59 (-GRHSLFH-) identified by

epitope scanning (Fig. 4). None of the B87-specific sequences

identified by this method is homologous to linear sequences

found in PSA. These sequences represent mimotopes, peptidesequences that mimic the natural epitope and are able to bind

specifically to the antibody paratope.

The only criteria used to assess homology was sequencealignment performed both manually and by the SEQSEE

Eluate

Table 2. Number of positive clones ideSelection procedure

ntlfled by lmmunoblottlng of phage-infeB87.posltive colonies

cted colonies.B80.positlve colonies

Round 1 Round 2 no./totai % no./total %

1 Ligand Ligand 0/1500 0 49/1500 3

2 Ligand Acid 0/1500 0 28/1508 23 Acid Ligand 238/983 24 7/209 3

4 Acid Acid 0/1500 0 0 0

‘)Four eluat es are generated after two rounds of panning.

Table 3. B80- and B87-specific peptide Insert sequences Identified by affinity selection of a bacteriophage peptide displaylibrary.

B80 sequences

WLGRPSRSGG (3)8

TLGRRSRPGG(2)TLGRPSGLVG(1)ANHTRGRPLV (3)

WGFDFGFGSS(2)SCLFYNTVCV (3)

NEWSDEENGQ(2)

B87 sequencesSVWGA.ASSPW(3)SRWVRRSSGW(4)SRWSRSRNPR (4)SQWVRRSSGT (4)

RPWPRRSSGW(2)SQWRRSSPW (5)SQWVRRSSG(5)

RFRQRATIAF (1)

Cross-reactive sequences

AFH1TGRIAA (3)

DVRAFHWWGR (5)

ADEAFHTVGR (4)

PRRRFHTTGA (3)

DVPSFHWWGR(3)LVHAFHUGA (3)

SVWGMSSPW (1)

PSA sequence45-60

NKSVILLGRHSLFHPE

ClinicalChernistiy42, No. 12, 1996 1967

software[33].Any homologous pattern less than three amino

acids was considered insignificant. Multiple alignments wereperformed with the inputsequences by usingSEQSEE with the

rbo.wtmatrixwith gap penalties of 10 and extension penalties of2. In the caseof the B80 bindingpeptides,those sequences that

clusteredtogether as being >40% identical(i.e.,the top four

sequences in Table 3) were aligned separatelywith SEQSEE.

Those residuesdisplaying>70% sequence conservationfrom

thissetwere used to form a “consensus”sequence.Thus forthe

B80 set,theconsensussequence would be TLGRPS***G (where

* isany amino acid).In the B87 set,a 70% sequence identity

thresholdwas alsoused.These criteriaidentifiedthe consensus

sequence S*W*RRSS. For the sake of simplification, only

conserved contiguous residues were underlined in Table 3. Aseries of alignments was performed with the eight sequences in

the B87 column (Table 3) against the PSA sequence (PIRaccession A26757) with SEQSEE. The same default scoring

matrix and gap penalties were used. The only sequence that

yielded any reasonable score was at the N terminus(WECEKHSQPW) of PSA. This particular sequence still re-

quired the introduction of two gaps and only two of the most

highly conserved residues were matched. By any criteria, this is

a poor match. Typically over a stretch of 10 residues, one shouldexpect to find better than 60% sequence identity (with no gaps

or breaks) if the match is significant. The nonspecific or

cross-reactive sequences (Table 3) appear to be reactive to both

mouse antibodies B80 and B87 and may be motifs capable of

binding the constant domains of IgG 1. This, however, requires

further study.

COMPUTER MODEL OF PSA

A structural model of the molecule, if not available from x-ray

crystallography, can be generated by using proteins of knownstructure that have a high degree of homology with the antigen.

The model can assist in locating highly exposed surface regions

and illustrate the spatial relations between various binding or

catalytic sites and antibody epitopes.Figure 5 shows the computer-generated homology model of

PSA. The catalytic triad for this serine protease-like molecule

(His-41, Asp-96, and Ser-71) is shown in white and the locationof the ACT binding site, based on what is known about the

binding of ACT to chymouypsin, is indicated by the shadedresidues. The B80 epitope identified by hexapeptide scanningand phage affinity selection, residues 53-60, is indicated inyellow. It is seen to be on an exposed region of the molecule wellaway from the ACT binding site. An epitope for B87 was notidentified in this study and is likely in an exposed region that is

not masked by the binding of ACT or B80 to PSA.

Computer models of PSA have been described by Vijinen

[34] and Bridon and Dowell [35]. The three-dimensional struc-ture of PSA and glandular kallikrein were modeled on the basis

of pKl. Loops unique to PSA were modeled by searching froma database and the structure was refined by energy minimization

and molecular dynamics. The resulting model closely resemblesthe PSA model generated in this study (Fig. 5). From the

location of the binding site of ACT it is apparent that most ofthe molecule is still exposed and available for binding to

Fig. 5.MolecularmodelingofPSA: The catalytic residues are shown inwhite and theproposedB80 epitope in yellow.The modelwas created as described in Materials and Methods.

antibodies when ACT is complexed to PSA. The B80 epitope(shown in yellow in Fig. 5), located in this study, is on a highly

exposed region near the N terminus, well away from the ACTbinding site. The exact location of the B87 epitope was not

determined but the possible epitopes could be limited to ex-posed regions that are unique to PSA and that do not overlap

with the B80 epitope or the ACT binding site.

Discuss#{226}onIn this study we attempt to addressthe issueof immunospeci-

ficity of anti-PSA antibodies at the molecular level by charac-terizing the epitopes recognized by two new mAbs. At least fivePSA isomers have been identified with pls ranging from 6.8 to

7.5. This heterogeneity is believed to be caused by differing

sialic acid content of the carbohydrate side chains and not by

changes in the amino acid sequence [9]. All of this variabilitymeans that antibodies directed against specific epitopes may

recognize some forms of PSA and not others. There may beepitopes on PSA that are especially sensitive to the conforma-

tional changes that occur when the enzyme is converted fromprecursor to zymogen to enzyme to inactive enzyme. Antibodies

directed against conformationally sensitive epitopes may only

recognize one of these forms of PSA.Two anti-PSA mAbs designated B80 and B87 were used in

this study. The B80 hybridoma cell line was fused in ourlaboratory with an anti-HRPO cell line to produce a hybrid

hybridoma or quadroma secreting a bispecific mAb with one

binding site directed against PSA and the other against HRPO[31]. This useful reagent eliminates the need for biotinylated or

secondary antibody-based detection methods and provides uswith a probe potentially having the highest specific activity ofHRPO-generated signal. Further, this bispecific mAb probeeliminates the high background often created by the secondary

polyclonal antibody- enzyme conjugates used in conventionalimmunoassays.

1968 Jette et al.: PSA epitope mapping

B80 and B87 were used to measure PSA in a sandwichimmunoassay, indicating that they recognize two distinct and

nonoverlapping epitopes. The PSA used to calibrate this assay

was obtained commercially as 99% pure protein in BSA buffer

to ensure that PSA-ACT complex was not present in this

calibrator. The ability of these antibodies to bind to PSAsimultaneously indicates that the epitopes are nonoverlapping

and far enough apart to avoid steric hindrance. Both of these

antibodies fail to form a homosandwich, indicating that each

antibody recognizes one unique and nonrepeating epitope. This

result is not unexpected considering the relatively small size of

PSA. Another recent study [36] also demonstrates by Western

blot analysis that the mAb B80 binds most consistently (com-

pared with other commercially available monoclonals studied)

not only to freePSA and PSA-ACT but also to PSA complexed

with AMG.

The specificity of the antibodies was determined by measur-ing the degree of cross-reactivity with pKl, which shows a

59.6% amino acid sequence homology with PSA. The fact thatneither B80 nor B87 cross-reacts with this related protease

shows that both antibodies recognize sequences on the molecule

that are not common to pKl and PSA. Since we can also assume

that the epitope resides on an exposed surface structure and is

not hidden in the molecule, the number of possible epitopes is

greatly reduced. Further studies on cross-reactivity with hK2

(84% homology to PSA) is needed to establish the unique

specificity of these mAbs.Sequential or linear epitopes are continuous strands of

residues present on the surface of a protein. Discontinuous or

nonlinear epitopes are sequences brought into proximity on the

surface of the molecule. Antibodies to discontinuous epitopesmay react weakly to subregions of the epitope made up of a few

residues in a linear sequence. Linear epitopes can be identified

and located by epitope scanning, a technique first introduced by

Geyson et al. [37]. It is based on the principle that linear

epitopes can be mimicked by peptides with as few as six residues.

Some of these peptides represent actual sequences found on the

antigen and indicate the location of the epitope on the protein

molecule. Others mimic the conformation and binding charac-

teristics of the epitope and can bind specifically to the antibody.

These peptides are referred to as mimotopes. Hexapeptides

representing all possible overlapping amino acid sequences of

PSA were synthesized on a solid-phase pin and tested for

binding to B80 and B87 in an ELISA format. The hexapeptidescan identified four possible locations of the B80 epitope. Two

of these peptides at residues 30-36 and 195-201 are found in the

interior of the molecule (on the basis of a computer-generated

model of PSA [34]), and are hence not likely candidates for the

epitope. The remaining two sequences at 21-27 and 5 3-59 are

adjacent to one another on the surface of the molecule. This

technique of epitope scanning, which can identify linear

epitopes that can be defined by a short linear peptide, failed tolocate the B87 epitope. It is likely that B87 recognizes a

conformational epitope that cannot be mimicked by a short

peptide based on the linear sequence of PSA. It is also possiblethat the epitope for B87 may be a linear epitope that is >6

residues. A short peptide representing a part of this epitope may

not be able to bind to the antibody paratope. Further, differ-

ences in conformation caused by interactions with neighboring

residues that are not directly involved in binding may be critical

to binding. A short peptide will not be able to mimic this

conformation and thus may not bind to the antibody.

Bacteriophage peptide display libraries can be used to affin-

ity-select peptides that specifically bind to a given antigen or

antibody [38-40]. The major advantage of using recombinant

peptide libraries is that a huge diversity of conformational

structures can be created and screened within a single library. By

this approach, a bacteriophage display library [32] was used to

identify sequences that bind specifically to B80 or B87.Among the phage sequences identified (Table 3), one con-

sensus sequence is homologous to a linear sequence seen on PSA

and identified as a possible epitope by epitope scanning. The

fact that two different techniques independently identified the

same region on PSA as the B80 epitope is strong evidence that

the epitope involves residues 51-54 (GRHS). The other anti-

body-specific sequences (Table 3) may represent mimotopes,peptidesthatmimic the conformation of theepitope.The other

remaining sequences listedin the third column of Table 3 are

peptides that were cross-reactivewith B80 or B87. This con-sensus sequence has been observed in other unrelated affinity

selection experiments and may represent cross-reactive binding

to constant regions of antibodies. The other potential PSA

mimotopes identified by this method are currently being eval-

uated.

In conclusion, the epitopes of two high-affinity anti-PSA mAbs,B80 and B87, were studied and well characterized. They bind

both free PSA and PSA complexed with ACT, and do not

cross-react with pKl or hK2. To locate the epitopes, overlap-

ping hexapeptides representing the entire PSA amino acid

sequence were synthesized and screened in an immunoassay for

binding with the two mAbs. B80 showed specific binding to

several peptides. B87 did not show binding to a specific peptide,

suggesting that the B87 epitope may be nonlinear. A bacterio-

phage peptide display library was used to select for peptide

sequences that bind specifically to B80 and B87. Several con-

sensus sequences were identified with this technique. One

peptide sequence was homologous to a linear sequence on PSA,

which was also identified as a possible B80 epitope by PSA

hexapeptide scanning.

MRS. acknowledges Biomira Inc. and the Medical Research

Council of Canada (MRC) for support of a University-IndustryChair, and the Natural Sciences and Engineering Research

Council of Canada (NSERC) for a strategic grant and ACB for

a group grant.

References1. Watt KWK, Lee P, Timkulu TM, Chan W-P, Loor R. Human prostate-

specific antigen: structural and functional similarity with serineproteases. Proc NatI Acad Sci U S A 1986;83:3166-70.

2. Schallar J, Akujama K, Tsuda R, Hara M, Marti T, Rickli B.Isolation, characterization and amino-acid sequence of y-semino-

Clinical Cheinistiy 42, No. 12, 1996 1969

protein, a glycoprotein from human seminal plasma. Eur J Bio-chem 1987:170:111-20.

3. Lundwall A, Lilja H. Molecular cloning of human prostate-specificantigen cDNA. FEBS Left 1987;214:317-22.

4. Lilja H, Abrahamsson PA, Lundwall A. Seminogelin, the predomi-nant protein in human seminal semen: primary structure andidentification of closely related proteins in the male accessory sexglands and on the spermatozoa. J Biol Chem 1989;264:1894-900.

5. Akijama K, Nakamura T, Iwanaga S, Hara M. The chymotrypsin-likeactivity of human prostate-specific antigen, ysemino-protein. FEBSLeft1987;225:168-72.

6. Christensson A, Laurell C-B, Lilja H. Enzymatic activity of theprostate-specific antigen and its reactions with extracellular serineproteinase inhibitors. Eur J Biochem 1990;194:755-63.

7. Drago JR, Badalament RA, Wientjes MG, Smith ii, Nesbit JA, YorkJP, et al. Relative value of prostate-specific antigen and prostateacid phosphatase in diagnosis and management of adenocarci-noma of the prostate: Ohio State University experience. Urol1989;34:187-92.

8. Catalona WJ, Smith DS, RaltiefTL, Dodds KM, Coplen DE, Yuan Ji,et al. Measurement of prostate-specific antigen in serum as ascreening test for prostate cancer. New EngI J Med 1991;324:1156-61.

9. SchellhammerPF,WrightGL. Biomolecular and clinical character-istics of PSA and other candidate prostate tumor markers. UrolClin North Am 1993;2O:597-6O6.

10. Crawford ED, DeAntoni EP. PSA as a screening test for prostatecancer. Urol Clin North Am 1993;20:637-46.

11. Andriole GL, Catalona Wi. Using PSA to screen for prostatecancer: the Washington University experience. Urol Clin North Am1993:20:647-52.

12. Lilja H. A kallikrein-like serine protease in prostatic fluid cleavesthe seminal vesicle protein. J Clin Invest 1985:76:1899-903.

13. Schaller J, Akiyama K, Tsuda R, Hara M, Marti T, Rickli EE.Isolation, characterization, and amino-acid sequence of gamma-seminoprotein, a glycoprotein from human seminal plasma. EurJ Biochem 1987;170:111-20.

14. Reigman PH, Vlietstra Ri, Suurmeifer L, Cleutjens CB, Trapman J.Characterization of the human kallikrein locus. Genomics 1992;14:6-11.

15. Berg T, Bradshaw R, Carretero 0, Chao J, Chao L, Clements J. Acommon nomenclature for members of the tissue (glandular)kallikrein gene families. Recent progress on kinins. Agents Ac-tions 1992;38(Suppl 1):19-25.

16. Carbini I, Scicli A, Carretero 0. The molecular biology of thekallikrein-kinin system; Ill. The human gene family and kallikreinsubstrate. J Hypertens 1993:11:893-8.

17. Chung DW, FujikawaK, McMullen BA, Davie EW. Human plasmaprekallikrein, a zymogen to a serine protease that contains fourtandom repeats. Biochemistry 1986;25:2410-7.

18. Clements J. The kallikrein gene family: a diversity of expressionand function. Mol Cell Endocrinol 1994;99:C1-C6.

19. McCormack RT, Rittenhouse HG, Finlay JA, Sokoloff RL, Wang Ti,Wolfert RL, et al. Molecular forms of prostate-specific antigen andthe human kallikrein gene family: a new era. Urology 1995;45:729-44.

20. Jensen-Scottrup L. a-Macroglobulins: structure, shape, and mech-

anism of proteinase complex formation. J Biol Chem 1989;264:11539-42.

21. Stenman U-H, Leinonen J, Alfthan H, Rannikko S, Tuhkanen K,Alfthan 0. A complex between prostate-specific antigen andal-antichymotrypsin is the major form of prostate-specific antigenin serum of patients with prostate cancer: assay of the complex

improves clinical sensitivity for cancer. Cancer Res 1991;51:222-6.

22. Lilja H, Christensson A, Dahl#{233}nU, Matikainen M-T, Nilsson 0,Pettersson K, Lovgren T. Prostate-specific antigen in serumoccurs predominantly in complex with a1-antichymotrypsin. ClinChem 1991:37:1618-25.

23. Christensson A, Bjork T, Nilsson 0, Dahlen U, Matakainen M-T,Cockett ATK, et al. Serum prostate-specific antigen complexed toal-antichymotrypsin as an indicator of prostate cancer. J Urol1993:150:100-5.

24. Graves HCB. Standardization of immunoassays for prostate-specific antigen. Cancer 1993;72:3141-4.

25. Zhou AM, Tewari PC, Bluestein BI, Caldwell GW, Larson Fl.Multiple forms of prostate-specific antigen in serum: differencesin immunorecognition by monoclonal and polyclonal assays. Clin

Chem 1993:39:2483-91.26. Jurince-Winkler C, von der Kammer H, Horlbeck R, Klippel K-F.

Pixberg HU, Scheit K-H. Clinical evaluation of a new prostate-specific antigen sandwich ELISA which employs four monoclonalantibodies directed at different epitopes of prostate-specific anti-gen. Eur Urol 1993:24:487-91.

27. Vessella RL, Lange PH. Issues in the assessment of PSA immu-noassays. Urol Clin North Am 1993;20:607-19.

28. Vessella RL. Trends in immunoassays of prostate-specific anti-gen: serum complexes and ultrasensitivity lEditoriall. Clin Chem1993;39:2035-9.

29. Jolley NL, Bacarese-Hamilton 1. SR1’ assay of a-fetoprotein,carcino-embryonic antigen, and prostate-specific antigen com-pared with corresponding established commercial assays. ClinChem 1994;40:895-9.

30. Cattini R, Robinson CR, Jolley GO, Bacarese-Hamilton T. Measure-ment of prostate-specific antigen in serum using four differentimmunoassays. Eur J Clin Chem Clin Biochem 1994;32:181-5.

31. Kreutz Fr, Suresh MR. Bispecific monoclonal antibodies in immu-nodiagnostics of prostate cancer. Fourth International Conferenceon bispecific monoclonal antibodies and targeted cellular toxicityProceedings, 1995. Hawk’s Key, FL:17.

32. Christian RB, Zuckermann RN, Kerr JM, Wang L, Malcolm BA.Simplified methods for constwction, assessment and rapidscreening of peptide libraries in bacteriophage. i Mol Biol 1992;227:711-8.

33. Wishart DS, Boyko RE, Willard L, Richards EM, Sykes BD. SEQSEE:a comprehensive program suite for protein sequence analysis.Comput AppI Biosci 1994;10:121-32.

34. Vijinen M. Modelling of prostate specific antigen and humanglandular kallikrein structures. Biochem Biophys Res Commun1994;204:1251-6.

35. Bridon D, Dowell B. Structural comparison of prostate-specificantigen and human glandular kallikrein using molecular modeling.Urology 1995;45:8O1-6.

36. Wu iT, Zhang P, Wang T, Wilson L, Astill M. Evaluation of free PSAisoforms, PSA complex formation and specificity of anti-PSAantibodies by HPLC and PAGE-immunoblotting techniques. i ClinLab Anal 1995:9:1-14.

37. Geyson HM, Meloen RH, Barteling Si. Use of peptide synthesis toprobe viral antigens for epitopes to a resolution of a single aminoacid. Proc NatI Acad Sci U S A 1984;81:3998-4002.

38. Scott 1K, Smith GP. Searching for peptide ligands with an epitopelibrary. Science 1990:249:386-90.

39. Cwirla SE, Peters EA, Barrette RW, Dower Ui. Peptides on phage:a vast library of peptides for identifying ligands. Proc NatI Acad SciU S A 1990:87:6378-82.

40. Devlin ii, Panganiban LC, Devlin PE. Random peptide libraries: asource of specific protein binding molecules. Science 1990;249:404-6.