A melanoma-specific VH antibody cloned from a fusion phage ...

Proc. Natl. Acad. Sci. USAVol. 93, pp. 6280-6285, June 1996Medical Sciences

A melanoma-specific VH antibody cloned from a fusion phagelibrary of a vaccinated melanoma patientXIAOHONG CAI*t AND ALAN GAREN**Department of Molecular Biophysics and Biochemistry, Yale University, 266 Whitney Avenue, New Haven, CI 06520-8114; and tCancer Research Institute,Hunan Medical University, Changsha, Hunan 410078, People's Republic of China

Contributed by Alan Garen, February 28, 1996

ABSTRACT The human antimelanoma antibody V86 wascloned from a single-chain Fv molecule (scFv) fusion phagelibrary displaying the heavy chain variable domain (VH) andlight chain variable domain (VL) repertoire of a melanomapatient immunized with genetically-modified autologous tu-mor cells. Previous ELISA tests for binding of the V86 fusionphage to a panel of human metastatic melanoma and carci-noma cell lines and primary cultures of normal melanocytes,endothelial, and fibroblast cells showed that measurablebinding occurred only to the melanoma cells. In this commu-nication, the strict specificity of V86 for melanoma cells wasconfirmed by immunohistochemical staining tests with cul-tured cells and frozen tissue sections. The V86 fusion phagestained melanoma cell lines but did not stain carcinoma celllines or cultured normal cells; V86 also stained specifically themelanoma cells in sections of metastatic tissue but did notstain any of the cells in sections from normal skin, lung, andkidney or from metastatic colon and ovarian carcinomas anda benign nevus. An unexpected finding is that V86 contains acomplete VH domain but only a short segment of a VL domain,which terminates before the CDR1 region. This VL deletionresulted from the occurrence in the VL cDNA of a restrictionsite, which was cleaved during construction of the scFvlibrary. Thus V86 is essentially a VH antibody. The effect ofadding a VL domain to V86 was examined by constructing scFvfusion phage libraries in which V86 was coupled to VA or VKdomains from the original scFv library of the melanomapatient and then panning the libraries against melanoma cellsto enrich for the highest affinity antibody clones. None of theV86-VA clones showed significant binding to melanoma cellsin ELISA tests; although binding occurred with most of theV86-V,K clones, it was generally weaker than the binding ofV86.These results indicate that most of the VL domains in theoriginal scFv library reduce or eliminate the affinity ofV86 formelanoma cells. Accordingly, VH libraries could provide ac-cess to anti-tumor antibodies that might not be detected inscFv or Fab libraries because of the incompatibility of mostrandomly paired VH and VL domains.

In an earlier communication (1) we described a generalprocedure for cloning anti-tumor antibodies from single-chainFv molecule (scFv) fusion phage libraries derived from pe-ripheral blood lymphocytes of cancer patients who werevaccinated with genetically modified tumor cells. The antibod-ies are selected by panning the libraries against the tumor cells.The vast repertoire of human antibodies that could bescreened in this way constitutes a potentially rich source ofanti-tumor antibodies for immunodiagnosis and immunother-apy and for isolating tumor-specific antigens. This procedurewas used to clone from two vaccinated melanoma patientsantibodies that react specifically with human melanoma cellsor melanocytic-lineage cells in ELISA tests (1). In this com-

munication, the melanoma-specific antibody V86 is furthercharacterized by immunohistochemistry with cultured cellsand tissue sections. V86 is unusual because it is essentially aheavy chain variable domain (VH) antibody rather than a scFvantibody, most of its light chain variable domain (VL) havingbeen deleted during construction of the scFv library. Todetermine whether V86 functions better as a VH or scFvanti-melanoma antibody, we analyzed the effect of attachingVL domains from the original scFv library on the bindingspecificity and affinity of V86 for melanoma cells. The resultsof this analysis reveal problems associated with scFv librariesthat might be circumvented with VH libraries.

MATERIALS AND METHODSHuman Cells. Endothelial cells were isolated from umbilical

cords and cultured by the Yale Endothelial Cell CultureFacility; the cells were used after 2-6 transfers. Melanocytesand fibroblast cells were cultured from foreskin samples by theCell Culture Core Service of the Yale Skin Disease ResearchCenter; the cells were used after 1-4 transfers. The melanomalines DM343 and DM414 were provided by Dr. H. Seigler'sLaboratory at Duke University Medical Center. The mela-noma lines SIT1 and ZAZ6 were provided by the Yale SkinDisease Research Center; TF-2 is a derivative of SIT1 trans-fected with the human tissue factor gene, which was providedby Dr. M. E. Bromberg. The melanoma line A2058, the breastcarcinoma line BT-20, the gastric carcinoma line MS, and therenal carcinoma line Caki-1 were obtained from the AmericanType Culture Collection. The culture medium for the mela-noma and carcinoma lines was DMEM/10% fetal calf serum(FCS). Frozen tissue sections were provided by the YaleCritical Technologies Facility.Immunoperoxidase Staining of Cells and Tissue Sections.

The cultured cells were grown in 16-well culture chambers(Lab-Tek Chamber Slide) until about 50% confluent, washedwith PBS, and fixed with 0.24% glutaraldehyde for 10 min atroom temperature. The fixed cells were washed with PBS, andthe wells were filled with 2% FCS in PBS and kept for 1 hr atroom temperature. The wells were emptied, and 200 ,ul of astock of cloned V86 fusion phage or fUSE5 control phage (2),diluted 1:1 with 2% FCS in PBS, was added to each well andkept for 2 hr at room temperature, and the wells were washed3 times with PBS for 5 min each. The bound phage weredetected by reacting first with a peroxidase-conjugated anti-M13 polyclonal antibody (Pharmacia) diluted 1:200 and thenstaining with a diaminobenzidine/H202 substrate (VectorLaboratories). Tissue sections cut from frozen samples ofmelanoma tumors or normal skin were attached to glass slides

Abbreviations: scFv, single-chain Fv molecule; VH and VL, variabledomain of a heavy chain and light chain, respectively; VK and VA,variable domains of K and A light chains, respectively; CDR, compli-mentarity-determining regions of the variable domain; FCS, fetal calfserum.Data deposition: The sequence reported in this paper has beendeposited in the GenBank data base (accession no. U58144).

6280

The publication costs of this article were defrayed in part by page chargepayment. This article must therefore be hereby marked "advertisement" inaccordance with 18 U.S.C. §1734 solely to indicate this fact.

Proc. Natl. Acad. Sci. USA 93 (1996) 6281

and fixed in 0.24% glutaraldehyde at room temperature for 10min. The fixed sections were used for immunohistochemistryas described for cultured cells.DNA Sequencing. A single bacterium colony carrying a

phage clone was inoculated into 50 ml of 2 x TY tet broth andgrown overnight; the bacteria were removed by centrifugation,and the supernatant medium was filtered through a 0.45-,mmembrane. The phage in the medium were precipitated byadding 0.2 volume of a 20% polyethylene glycol 8000/2.5 MNaCl stock solution, and the phage were pelleted and resus-pended in PBS. The precipitation procedure was repeatedonce, and the phage pellet was resuspended in 100 ,ul of water.The phage DNA was extracted twice with phenol/chloroformand once with chloroform, and the DNA was precipitated withethanol/NaAc and resuspended in 50 ,ul of water. The primerfor sequencing the VL domain was TGATTTTCTGTAT-GAGG, which hybridizes 90 bases away from the 5'-end ofgene-3 of the phage; the primer for sequencing the VH domainwas ACCCGACCCACCACCGCCCGA, which hybridizes tothe linker of the scFv molecule. A full description of thissequencing procedure has been published (2).

Construction of V86-VA and V86-V, Libraries. The cDNAencoding the VH domain of V86 was synthesized by PCR fromthe V86 fusion phage clone (1). The sequence of the 5'-primerwas ATTATTATTCGCAATTCCTTTAG, which hybridizesto the third gene of the phage, and the sequence of the3'-primer was GCCACCCGACCCACCACCGCCCGAGC-CACCTGARGAGACGGTGACCRKKGTYCC, which in-cludes part of the JH and linker sequences (1). The cDNAsencoding the VK and VA domains were synthesized fromperipheral blood lymphocytes of melanoma patient DM414 asdescribed previously (1). To construct the V86-VA and V86-VKscFv cDNAs, the complementary linker sequences were hy-bridized, and the cDNAs were synthesized by PCR as described(2). The procedure involved adding 10 ng of V86 cDNA and 10ng of VK or VA cDNAs to 100 ,ul of PCR reagents and cyclingseven times at 94°C for 1 min, 60°C for 1 min, and 72°C for 2min; 10 pM of a mixture of VH back primers and 10 pM of JKor JA forward primers (1) were then added, and the PCRsynthesis was continued for 25 cycles. The PCR products wereprecipitated with ethanol and purified by electrophoresis in1% LMT-agarose gel and extraction with 03-agarose. To con-struct the fusion phage libraries; the scFv cDNAs and thereplicative form of the fUSE5 DNA (2) were digested with SfiIand ligated in 100 ,ul of reaction mixture containing 1.5 ,ug ofscFv cDNA, 8 ,ug of vector DNA, and 2000 units of T4 DNAligase for 16 hr at 16°C. The ligation products were purified byextraction with phenol/chloroform and precipitation withethanol, and the purified DNA was used to transform DH1OBElectroMax cells (GIBCO/BRL). The cells were plated on 2x TY tet agar. The total number of transformed clones was 4x 107 for the V86-VK library and 1 x 108 for the V86-VAlibrary. The fusion phage stocks for each library were preparedfrom the transformed cells as described (1). The fraction ofphage with scFv inserts, as determined by PCR synthesis usingprimers complementary to phage sequences flanking theinsert, was 75% (15/20) for the V86-VK library and 83%(19/23) for the V86-VA library.Panning the V86-VA and V86-V,, Libraries. The melanoma

cell line A2058 was grown as an attached monolayer in 24-cm2flasks until almost confluent, washed with PBS, and fixed with0.24% glutaraldehyde for 10 min at room temperature. Thefixed cells were washed with PBS and blocked with DMEM/10% FCS for 1 hr at room temperature. The phage from theV86-VA and V86-VK libraries were precipitated in 4% PEG/0.5 M NaCl and resuspended in water, and about 1011 phagein 2 ml of DMEM/10% FCS were added to the melanomacells. The culture flask was shaken gently for 2 hr at roomtemperature, the medium was removed, and the cells werewashed rapidly 10 times with PBS at room temperature. The

attached phage were eluted from the cells in 2 ml of E-bufferfor 10 min at room temperature and immediately neutralizedwith 0.375 ml N-buffer (2). The eluted phage were mixed with15 ml of Escherichia coli K91 Kan cells, and after 30 min atroom temperature, the cells were plated on 2 x TY tet agar (2).The colonies were collected in 50 ml of 2 x TY tet mediumand incubated for 1 hr. The bacteria were pelleted, and thesupernatant medium was filtered through a 0.45-,um mem-brane. For each subsequent panning step, the amplified phagefrom the previous panning step were precipitated in 4%PEG/0.5 M NaCl and resuspended in water, and about 1011transforming units of the phage were used for panning againstmelanoma cells A2058 as described for the first panning step.

Preparation of Fusion Phage Clones. The phage were mixedwith E. coli K91 Kan cells at low phage to cell ratios, and thecells were plated on 2 x TY tet agar (2). Individual colonieswere inoculated into 2 ml of 2 x TY tet medium and grownovernight. The cells were pelleted, and the medium containingthe cloned phage was used for the initial ELISA assays. For theassays reported in Figs. 1 and 2, the medium was filteredthrough a 0.45-,um membrane and precipitated in 4% PEG/0.5M NaCl twice. The phage titer in each cloned stock wasdetermined as described (2) and adjusted to contain 1010 phagetransforming units per ml.ELISA Assays. The cells were grown in 96-well microtiter

plates until almost confluent, washed with PBS, and fixed with0.24% glutaraldehyde for 30 min at room temperature. Thefixed cells were washed with PBS, and the wells were filled withDMEM/10% FCS and kept for 1 hr at room temperature. Thewells were emptied, and 100 ,ul of a stock of cloned phagediluted 1:1 with DMEM/10% FCS was added to each well.After 2 hr at room temperature, the wells were washed rapidly10 times with PBS, and the amount ofbound phage was assayedusing a peroxidase-conjugated anti-M13 polyclonal antibodyand O-phenylenediamine as the peroxidase substrate (Phar-macia Detection Module). All assays were done in triplicate.

Restriction Analysis of V86-VA and V86-VK cDNAs. ThecDNA insert in a phage clone was synthesized by PCR usingthe 5'-primer ATTATTATTCGCAATTCCTTTAG and the3'-primer GAATTTTCTGTATGAGGTTTTGCT, which hy-bridize to phage sequences flanking the insert. The purifiedcDNA products were digested separately with Sau3AI andHinfl, and the digests were analyzed by electrophoresis in 2%agarose gel.

Relative Binding Affinities of the Fusion Phage Antibodiesto Melanoma Cells. The melanoma cell line A2058 washarvested from culture flasks in PBS/10 mM EDTA, and thecells were pelleted and resuspended in DMEM/10% FCS. Aphage stock was serially diluted in DMEM, and each dilutedsample was added to 5 x 105 melanoma cells. After incubationwith gentle rotation at room temperature for 2 hr, the cellswere washed 3 times by centrifugation at 3000 rpm in amicrofuge for 3 min and resuspended in PBS. The phageremaining bound to the cells after the third centrifugation wereeluted with 400 ,ul of E-buffer for 10 min at room temperatureand were neutralized immediately with 75 ,lI N-buffer (2). Thetotal phage and eluted phage titers were determined byinfecting a suspension of E. coli K91 Kan cells from a frozenstock for 1 hr at room temperature and plating the cells on 2x TY tet agar (2). The relative affinities of the fusion phageantibodies were calculated as described for monoclonal anti-bodies (3).

RESULTSThe antibody V86was selected from a scFv fusion phage libraryof a melanoma patient by panning against the autologoustumor cells. Previous ELISA tests with a panel of melanomalines, carcinoma lines, and primary cultures of normal celltypes including melanocytes showed that V86 fusion phage

Medical Sciences: Cai and Garen

6282 Medical Sciences: Cai and Garen

^ .* .Fw t; s r * * A ++ ~* - * Pa 0* |

z t 4 41,V .~~~~~~~~~~~~~~~~~~~~I.

ts* fr Uot 54~~~~~~~~~~~~~~~~~~~~~~~~#~~~ ~ ~ ~~~~.....4,

* is ;t *..A

i.

I e

a.

'p

a% A

p.-.

-S..

W .

w e p..

*& I *. :..

.^ k11

a. ..,

C.AMt

.Ikb

* a .9** a

_4'* hr,AA

.4.** ,."aD-b

A

*. H

I

..

I

E*gI

E

j

e.

,'1

f

% 04. * . ..a

0,.

_~0

A.

i

N

G

FIG. 1. Immunohistochemical staining of cultured cells with V86fusion phage. The cells were grown on slides, fixed with gluteralde-

AIqr hyde, and reacted with V86 fusion phage or fUSE5 control phage. Thesecond antibody was a polyclonal antiphage conjugated to horseradishperoxidase, and the staining reagent was diaminobenzidine/H202,

* i which produces a brown product; hematoxylin was used as a counter-, *- *.r stain. Panel A, melanoma DM414 with V86; panel B, melanoma

t DM414 with fUSE5; panel C, melanoma A2058 with V86; panel D,normal melanocytes with V86; panel E, normal endothelial cells with

* * V86; panel F, normal fibroblasts with V86; panel G, renal carcinomah Caki-1 with V86. Additional immunohistochemical tests of V86 fusion

phage with melanoma lines DM 341, DM343, ZAZ6, and TF2 showed

a strong staining reaction similar to panels A and C; there was no

staining reaction of V86 with the gastric carcinoma line MS and thebreast carcinoma line BT-20 similar to panel G.

Proc. Natl. Acad. Sci. USA 93 (1996)

a

*:... .4-m.

.I.....::d-:"

1. .%

a:,i,a.

fJ._* A

a.6.

ft.

W.

V

let,

N.-i...> 0Ni..

B

AP'

.-I

I

B,., .0

i..'aUp

40SI:

4CW

-..

.

I

41.--

lbo:-ft.

.ft

Im:..w

N.. .

"'. IC%%1%ik I* -'-%

I V4k VA -.-

4k 1. 4wi

45P0 a

o

..f ..

".......i:

4.

Proc. Natl. Acad. Sci. USA 93 (1996) 6283

A B

C

could bind to the melanoma cells but not to the carcinoma cellsor normal cells (1). The melanoma-specific binding of V86 wasfurther tested by immunohistochemistry with several of thetumor cell lines and normal cells used for the ELISA tests (Fig.1). V86 showed a staining reaction with the melanoma cells butnot with either the carcinoma cells or normal cells, confirmingthe ELISA results. Because the reaction of V86 with melanomacells might involve an antigen expressed by the cultured cellsbut not by melanoma tumors in vivo, additional immunohis-tochemistry was done with sections of frozen human meta-static melanoma tissue and normal skin tissue. V86 showed a

staining reaction with the melanoma tissue but not with theconnective tissue in the melanoma sections or with any of thetissues in the section of normal skin (Fig. 2); V86 also failed tostain any of the tissues in sections of frozen normal lung andkidney or of metastatic colon and ovarian carcinomas and a

benign nevus (data not shown). These results demonstrate thatV86 can bind specifically to melanoma cells in a metastatictumor as well as to cultured melanoma cells. The staining ofthe melanoma sections appears to concentrate along the

FIG. 2. Immunohistochemical assay for specific binding 'of V86fusion phage to a frozen section of a human metastatic melanoma. Theprocedure for reacting the sections with the phage and antiphagesecond antibody was as described for Fig. 1; the sections were stainedwith diaminobenzidine/H202 and counterstained with hematoxylin.Panel A, melanoma section reacted with V86; panel B, melanomasection reacted with fUSE5; panel C, normal skin section reacted withV86. Additional immunohistochemical tests of V86 with frozen tissuesections of normal lung and kidney, metastatic colon, and ovariancarcinomas and benign nevus showed no staining reaction similar topanel C.

borders of the melanoma cells, suggesting that the antibodybinds at the cell surface.The sequence of the V86 antibody, as deduced from the

sequence of the cDNA insert in the fusion phage, shows acomplete VH domain followed by the linker, which couples theVH and VL domains in a scFv molecule; however, the VLdomain terminates before the CDR1 region (Table 1). Thereason for the premature termination of the VL domain is thepresence of an Sfil restriction site that was cut in the processof cloning the V86 scFv cDNA into the fusion phage, resultingin the deletion of the distal VL segment.The next series of experiments was designed to test the effect

of adding a VL domain to V86 on the specificity and affinity ofthe antibody for melanoma cells. Two scFv fusion phagelibraries were constructed for this purpose, one expressingV86-VK antibodies and the other V86-VA antibodies. The VAand VK domains for these libraries were derived from therepertoire of VL domains in the original scFv library used forthe isolation of V86. Individual clones were isolated from eachlibrary, and those containing different VA or VK domains, as

Table 1. Amino acid sequence of clone V86.

FR1 CDR1 FR2 CDR2QVQLVQSGGGLVQPGGSLRLSCAASGFTFS SYAMS WVRQAPGKGLEWVA AISGSGGSTYYADSVKG

FR3 CDR3 FR4RFTISRDNSKNTLYLQMNSLRAEDTAVYYCAR GWGLRGEEGDYYMDV WGKGTMVTVSS

LINKER FR1GGGGSGGGGSGGGGS SYELTQEPRGGGTQLTVLGGAAGAThe amino acid sequence was derived from the nucleotide sequence of the cDNA insert in the V86 fusion phage.

The completeVH domain is followed by the linker and the truncated VL domain.

Medical Sciences: Cai and Garen

6284 Medical Sciences: Cai and Garen

Table 2. Effect of different VA and VK domains on the binding ofV86 to melanoma cells.

ELISA absorbance

Clones + + + + + -

V86-VA 0 0 0 50V86-VK 2 20 28 18

The V86-VA and V86-VK clones are random isolates from the twofusion phage libraries, each expressing the V86-VH domain linkedeither to VA or VK domains from the original scFv library. Themelanoma cell line for the ELISA tests was A2058. The ELISAabsorbance scales are as follows: + + +, >0.8; + +, 0.8-0.5; +, 0.5-0.1;-, <0.1. The absorbance for V86 without a VL domain is 1.0.

determined by restriction mapping the cDNAs in the fusionphage, were tested by ELISA for binding to the melanoma cellline A2058 (Table 2). None of the V86-VA clones showeddetectable binding, indicating that VA domains are incompat-ible partners for V86; most of the V86-V,K clones showedsignificant binding, but it was generally weaker than thebinding of V86. Thus most of the VL partners for V86 reduceor eliminate its capacity to bind to melanoma cells. TheV86-VA and V86-VK libraries were then panned twice againstmelanoma cells to enrich for any higher affinity antibodies thatmight be present, and clones of the panned phage containingdifferent VA or VK. domains were tested by ELISA for bindingto the melanoma line A2058. As before, binding occurred withthe V86-V, clones and not with the V86-VA clones. Therelative binding affinities for A2058 cells of 15 panned V86-V"clones were determined, using a Scatchard plot assay (3). Theresults showed that the affinity of V86 was the highest (Fig. 3),suggesting that V86 functions more effectively as a VH than asa scFv anti-melanoma antibody. The panned V86-VK cloneswere also tested by ELISA for specific binding to melanomacells, using a panel of cells consisting of A2058, four carcinomalines and primary cultures of melanocyte, endothelial, andfibroblast cells. All of the V86-VK clones bound only to themelanoma cells (data not shown), indicating that addition of aVK domain to V86 affects only its affinity but not its specificityfor melanoma cells.

DISCUSSIONThe strict specificity of the human antibody V86 for melanomacells was first demonstrated by ELISA tests with a panel ofhuman melanoma and carcinoma cell lines and primary cul-tures of normal cells including melanocytes (1). The specificityof V86 has now been confirmed by immunohistochemistry withthe tumor lines and cultured normal cells and also with frozensections of metastatic melanoma and carcinoma tissues, abenign nevus and three normal tissues including skin: V86reacted specifically with melanoma cells in the melanomatissue and cell lines and did not react with any of the normalcells or other tumor cells or the benign nevus. V86 appears tobind to a cell surface antigen expressed by all of the metastaticmelanomas tested and not by normal cells or other tumors.Further immunohistochemistry with staged melanoma tissuesare needed to determine whether expression of the V86antigen also occurs at earlier stages of the disease.Although V86 was isolated from a fusion phage library

designed to display the antibody repertoire of a cancer patientas scFv molecules and therefore should have contained both aVH and VL domain, most of the VL domain is missing becausean extraneous cloning site located near the 5'-end of the VLcDNA was cleaved during construction of the library, resultingin a deletion of the distal VL region. Despite the absence of aVL partner, V86 appears to be one of the most specificanti-melanoma antibodies isolated from the library (1). It wasshown in other studies that an isolated heavy chain of amonoclonal antibody can retain the specificity of the intact

10

0

x1 5 * '

0

0 1 2 3 4 5

0~~~~~~

Fig. 3. Scatchard plot for binding of V86 fusion phage to melanomacells A2058. The experiments were done as described in Materials andMethods. r is the number of phage bound per cell, A-x is the numberof unbound phage and r/A-x is the ratio expressed as M-1 (3). Theaffinity constant K can be estimated from these measurements, butbecause of uncertainty in the phage titer, which is based on infectivity,only a relative affinity value can be used. Assigning a relative affinityof LK0 to V86, the affinity of 15 other fusion phage clones that expressV86 as a scFv molecule conjugated to various VL domains, as describedin the text, ranged from 0.6 to 0.02.

antibody for its cognate determinant, although the affinity isusually reduced by one or two orders of magnitude (4, 5). Asimilar finding was reported for VH antibodies synthesizedfrom spleen DNA of mice immunized with lysozyme. Thespecificity for lysozyme was retained, but the affinity was about10-fold weaker as compared with a complete monoclonalanti-lysozyme antibody (6). To determine the effect of linkinga VL domain to V86 on its specificity and affinity for melanomacells, we constructed two scFv fusion phage libraries expressingthe V86 domain in random pairwise combinations with eitherVA or VK, domains derived from the original scFv library usedfor the isolation of V86. All of the different VA, domains testedas partners for V86 prevented binding to melanoma cells;binding occurred with most of the different V,, domains tested,but it was generally weaker than the binding of V86. Thus theV86 single-domain VH antibody binds more strongly to mela-noma cells than do most of the V86 scFv antibodies containingrandomly paired VL domains. A similar finding was reportedfor the VH domain of a mouse lupus IgG monoclonal antibody,which was shown to bind to native DNA at least as well aloneas in combination with its natural VL partner in a scFvconstruct (9). Although the affinity of V86 for melanoma cellsis usually reduced by addition of a VK, domain, the strictspecificity for melanoma cells is not affected, consistent withother evidence that the VH domain alone can determineantibody specificity (4-8).

T'he anti-tumor antibodies expressed by the mature B cellsin the peripheral blood lymphocytes of vaccinated cancerpatients, such as the patient from whom V86 was derived (1),have probably been subjected to affinity maturation, whichoptimizes the specificity and affinity of an antibody for itscognate determinant. However, because construction of a scFvor Fab library involves random pairings between highly com-plex populations of VH and VL CDNAS, there is virtually nochance that the original combination ofVH and VL partners forany antibody will be included in a library of the usual size.Pairing a VH domain with a new VL partner usually results ina less active or inactive antibody, as shown here for V86.

Proc. Natl. Acad. Sci. USA 93 (1996)

Medical Sciences: Cai and Garen

Consequently VH domains that could function as anti-tumorVH antibodies might remain undetected in single scFv or Fablibraries because of an incompatible VL partner. This is a majorlimitation for random combinatorial antibody libraries, whichmight be circumvented by using VH libraries to pan foranti-tumor antibodies. To test this strategy, we constructed a

VH fusion phage library containing the same population ofVHgenes as the scFv library used for the isolation of V86.Preliminary results indicate that additional melanoma-specificantibodies can be cloned from the VH library (unpublishedresults).A natural repertoire of functional heavy chain antibodies has

been reported in the camel (10). The antibodies are composedof two identical disulfide-linked heavy chains each containinga VH, CH2, and CH3 domain. The camel also has a minorpopulation of antibodies with the expected composition of twocomplete heavy chains and two complete light chains. Thisremarkable discovery raises intriguing questions concerningthe evolutionary and functional significance of heavy chainantibodies. It also provides additional evidence that a func-tional antibody need not contain a VL domain.

Proc. Natl. Acad. Sci. USA 93 (1996) 6285

We thank Dr. David Leffell for supplying a frozen section of benignnevus tissue. This study was supported by a Special Use Fundadministered by Dr. William H. Konigsberg and by the Joseph J.Konigsberg Melanoma Memorial Fund.

1. Cai, X. & Garen, A. (1995) Proc. Natl. Acad. Sci. USA 92,6537-6541.

2. Smith, P. G. & Scott, J. K. (1993) Methods Enzymol. 217, 228-257.

3. Trucco, M. & de Petris, S. (1981) Immunol. Methods 2, 1-26.4. Haber, E. & Richards, F. F. (1963) Proc. R. Soc. Ser. B 166,

176-187.5. Jaton, J. C., Klinman, N. R., Givol, D. & Sela, M. (1968) Bio-

chemistry 7, 4185-4195.6. Ward, E. S., Gussow, D., Griffiths, A. D., Jones, P. T. & Winter,

G. (1989) Nature (London) 341, 544-546.7. Fleischman, J. B., Porter, R. R. & Press, E. M. (1963) Biochem.

J. 88, 220-228.8. Utsumic, S. & Karush, F. (1964) Biochemistry 3, 1329-1338.9. Jang, Y.-Y., Lecerf, J.-M. & Stollar, B. D. (1996) Molec. Immu-

nol. 33, 197-210.10. Hammers-Casterman, C., Atarhouch, T., Muyldermans, S., Rob-

inson, G., Hamers, C., Songa, E. B., Bendahman, N. & Hamers,R. (1993) Nature (London) 363, 446-448.