J. Mol. Biol. (1997) 268, 107±117

Stabilization of a Recombinant Fv Fragment byBase-loop Interconnection and VH-VL Permutation

Ulrich Brinkmann, Angelina Di Carlo, George VasmatzisNatalya Kurochkina, Richard Beers, Byungkook Lee and Ira Pastan*

Laboratory of Molecular BiologyNational Cancer Institute,National Institutes of HealthBuilding 37, Room 4E16 37Convent Dr Msc 4255Bethesda, MD20892-4255, USA

Abbreviations used: IPTG, isoprothiogalactopyranoside; scFv, singledisul®de stabilized Fv; VH and VL,heavy or light chain; pFv, base-looppermutated Fv fragment; PCR, polyreaction; GuCl, guanidine chloride;dithioerythritol; IB, inclusion body.

0022±2836/97/160107±11 $25.00/0/mb9

We have developed a novel method to stabilize a recombinant antibodyFv fragment. The VH and VL domains of this Fv fragment, called pFv(permutated Fv), are covalently interconnected to each other at the two``base-loops'' that normally connect VH beta strand 3 to 3b and VL betastrand 3 to 3b. To produce the base-loop stabilized Fv fragment, we con-nected the N-terminal half of the VL domain (VL 1-40) of murine anti-body anti-Tac to the C-terminal half of VH (VH 42-115). We also fused theC terminus of VH by a (Gly4Ser)3 linker to the N-terminal half of VH

(VH 1-40, thereby generating a permutated VH domain). Finally we con-nected the base loop of VH (N-terminal half) to the C-terminal half of VL

(VH 42-115). The anti-Tac pFv fragment was fused to a truncated form ofPseudomonas exotoxin to generate a pFv-immunotoxin. Fvs with the cor-rect structure were produced by refolding of recombinant inclusion bodyprotein using a renaturation protocol that was originally developed forFab and scFv fragments. Due to the arti®cially connected and permutatedprimary sequence, the folding pathway for the pFv structure may poss-ibly be different from the conventional folding of antibody domains.Analysis of antigen binding of anti-Tac pFv, and of the speci®c cytotox-icity of pFv-immunotoxin towards antigen expressing cancer cells demon-strated that the anti-Tac pFv retained most of its af®nity and fullspeci®city when compared to anti-Tac scFv. Also anti-Tac pFv was rela-tively stable, retaining 25% of its binding activity after a 24 hour incu-bation in human serum at 37�C. This indicates that connection of baseloops can be a useful alternative to linker or disul®de stabilization of Fvfragments.

Keywords: antibody engineering; cancer therapy; anti-Tac; scFv; dsFv

*Corresponding author

Introduction

Fv fragments of antibodies are heterodimers ofantibody VH and VL domains. They are the smal-lest antibody fragments that contain all structuralinformation necessary for speci®c antigen binding.Fv fragments are useful for diagnostic and thera-peutic applications such as imaging of tumors ortargeted cancer therapy. Because of their smallsize, Fv fragments are particularly useful in appli-cations that require good tissue or tumor pen-

pyl-b-D-chain Fv; dsFv,variable region of

connectedmerase chainDTE,

60850

etration, because small molecules penetrate tumorsmuch faster than large molecules (Yokota et al.,1992). Although native unstabilized Fv heterodi-mers have been produced from unusual antibodies(Skerra & Pluckthun, 1988; Webber et al., 1995),generally Fv fragments by themselves are unstablebecause the VH and VL domains of the heterodimercan dissociate (Glockshuber et al., 1990). This re-sults in drastically reduced binding af®nity and isoften accompanied by aggregation.

A solution to the stabilization problem resultedfrom a combination of genetic engineering and re-combinant protein expression techniques. The mostcommon method of stabilizing Fvs is the covalentconnection of VH and VL by a ¯exible peptide lin-ker which results in single-chain (sc)Fv molecules(Bird et al., 1988; Huston et al., 1988). scFvs aremore stable than Fvs alone, and several scFvs and

Figure 1. Design of anti-Tac pFv by base-loop connection and VH permutation. a, Structural model of anti-Tac scFv.CDRs are on the top of the molecule, the base loops on the bottom. b, Model (for demonstration only) in which theVHC-VHN linker of the permutated VH has been introduced and the base loops have been opened. c, Final permu-tated Fv that is stabilized by base-loop connection between VH and VL.

108 Permutated Fv Fragment

scFv-fusion proteins (e.g. recombinant immunotox-ins) have been made that may be useful for tumorimaging and the therapy of tumors (Yokota et al.,1992; Webber et al., 1995; Chaudhary et al., 1989;Brinkmann & Pastan, 1994; Pastan et al., 1995;Brinkmann et al., 1991). On the other hand, thereare many examples of unstable scFvs or scFvs withsuboptimal binding properties, because the peptidelinker is not always suf®cient for VH-VL stabiliz-ation, and it may interfere with antigen binding(Brinkmann et al., 1993; Reiter et al., 1994a,b).Another way to generate stable recombinant Fvs isto connect the VH and VL by an interdomain disul-®de bond instead of a linker peptide; this results indisul®de stabilized (ds)Fvs. dsFvs, when they canbe successfully produced, solve most problemsthat can be associated with scFvs; they are verystable, often show full antigen binding activity,and sometimes have better af®nity than scFvs(Webber et al., 1995; Brinkmann et al., 1993; Reiteret al., 1994a,b,c).

Many, and perhaps most, antibody Fv fragmentscan be successfully stabilized by one of thesemethods and some Fvs can be stabilized by both

methods. However, there are also Fvs that cannotbe stabilized in an optimal manner by either meth-od. For example, Fvs can be unstable in forms ofscFvs and may have reduced binding as dsFv(Brinkmann et al., 1993; Reiter et al., 1996).

Here we describe the development of an alterna-tive stabilization method for Fv fragments inwhich the VH and VL domains are covalentlylinked to each other at the ``base loops'' that con-nect the beta strands 3 to 3b of VH and of VL. Inprinciple this method might circumvent some ofthe dif®culties encountered using other approachesto stabilize Fvs.

Results

Design of anti-Tac pFv: base loop connectionand VH permutation

pFv molecules are recombinant Fvs in which theVH and VL are connected to each other at the loopbetween beta strand 3 and 3b of VL and the loopbetween strands 3 and 3b of VH. A structuralmodel of such a pFv derived from anti-Tac, a mur-

Figure 2. Composition and foldingof scFv and pFv. A, Linear compo-sition of the anti-Tac scFv and pFvchains. B, Representation of anti-Tac scFv (left), correctly folded pFvand hypothetical incorrectly foldedmixed scFv structure that mightalso form from the pFv primarysequence. The (incorrect) mixedscFv would contain structuralincompatibilities that prevent theformation of this structure or makeit very unstable (Table 1). The dia-grams on the right indicate thestrand connections in the barrel(beta sandwich). Upward trianglesrepresent strands that come up,inverted triangles represent strandsthat go down. Thin lines are loopsat the bottom, thick lines are loopsat the top, i.e. CDRs.

Permutated Fv Fragment 109

ine antibody that recognizes the p55 subunit of thehuman IL2 receptor (Webber et al., 1995;Chaudhary et al., 1989) is shown in Figure 1. Theloops that were used to connect VH and VL are ex-posed on the bottom of the Fv (Figure 1a), and we,therefore, refer to these two loops as base loops.The linear composition of the gene that encodesthe pFv protein is shown in Figure 2A and the se-quence in Figure 3. We fused the amino terminusof VL (positions 1 to 40) to the C terminus of VH

(positions 42 to 115) using a GSSAGG linking pep-tide. This fusion connects the ®rst half of the VL

base loop with the second half of the VH base loop.The ¯exible loop connection contains GSSAGG tofacilitate spanning the distance between VH and VL

(12 to 14 AÊ ) in the ®nal structure (see Figure 3 forsequence details). The C terminus of VH (VH(42-115)) is also fused by a (Gly4Ser)3 linker to the N-terminal 40 amino acids of VH, thus forming a per-mutated form of VH that ends with the VH baseloop. This is connected with a short connector,SSAGG (Figure 3A), to the second half of the VL

base loop followed by VL 41-106(end) of VL. Theproline 41 of VH was replaced by serine and formsthe ®rst amino acid of the SSAGG connector(Figure 3A). For ease of production and puri®-cation and for analysis of speci®c binding of theFv, the pFv molecule was then fused to a truncatedform of Pseudomonas exotoxin, PE38, to generate apFv-PE38 immunotoxin (Brinkmann & Pastan,1994, 1995; Pastan et al., 1995).

In order for the designed pFv sequence to attain acorrectly folded functional protein as shown inFigure 1, the arti®cially separated and reconnectedamino acid chain has to fold into a permutated VH

domain (second half of VH! linker! ®rst half ofVH) and the VL domain has to fold from two frag-ments that are separated (in sequence) by the per-mutated VH. Despite this complex primarysequence, we thought that the acquisition of a cor-rect Fv structure should be possible because bothbase loop connections are long enough to span thedistance between VH and VL (Figure 1c) and thelinker of the permutated VH is long and ¯exible en-

Figure 3. Construction of a base-loop connected permutated pFv fragment. A, Sequence of anti-Tac(pFv). B, PCR-cloning strategy and construction of the plasmid for expression of anti-Tac(pFv)-PE38 immunotoxin.

110 Permutated Fv Fragment

ough to connect the VH fragments without causingstructural disturbances (Figure 1b and c).

One potential problem in our design is that, be-cause of the similarity of VL and VH structures andthe symmetrical design of the permutated sequence(Figure 2), a ``classical'' scFv of mixed VH/VL do-mains could form instead of the pFv. Because weconsidered the formation of mixed scFv a poten-tially signi®cant side reaction to the formation ofpFv, we built a model of the mixed scFv structureand analyzed whether it could exist as a properlyfolded Fv (Table 1). We found that several pos-itions in the mixed scFv are structurally incompati-ble and it is unlikely that these incompatibilitiescan be compensated by chain rearrangements and¯exibility. Therefore, the predominant stable pro-duct resulting from the permutated Fv gene shouldbe the base-loop stabilized functional pFv shownin Figure 1c.

Construction of the anti-Tac pFv gene

A gene that codes the base-loop connected andpermutated Fv fragment of anti-Tac was con-structed by overlap extension polymerase chain re-action (PCR) using the single chain Fv gene of anti-Tac (Chaudhary et al., 1989) as PCR template. Theassembly of the pFv gene from four separate PCR

fragments encoding the N-terminal half of VL

(VLN), the C-terminal half of VH (VHC), the C-terminal half of VL (VLC) and the C-terminal halfof VH (VHC) is described in detail in Materials andMethods and in Figure 3B. The pFv gene wasfused in-frame to the gene encoding PE38 andcloned into a vector for recombinant expression inEscherichia coli (Brinkmann & Pastan, 1995).

Expression in inclusion bodies, refolding andpurification of a recombinant anti-Tac(pFv)-immunotoxin

E. coli BL21/lDE3 cultures containing the plasmidpADC1 for expression of the anti-Tac pFv-immu-notoxin were grown and induced with IPTG. ThepFv fusion protein accumulated in insoluble intra-cellular inclusion bodies (IBs). These IBs containalmost pure recombinant protein, but in an insolu-ble, denatured and aggregated conformation. Togenerate protein with a proper conformation, wesolubilized and reduced the IBs in GuCl and DTE,and refolded the protein by dilution in a buffercontaining arginine as described in Materials andMethods (Brinkmann & Pastan, 1995; Buchner et al.,1992). Refolded, soluble monomeric protein wasthen puri®ed from other bacterial and improperlyfolded proteins by ion exchange (Q-Sepharose,

Table 1. Design of a base-loop connected permutated Fv fragment

Position 1 Position 2 Design/problem Effect on structure

pFv VL 40 Pro VH 42 Gly Connected (base loop) None, stabilizingVH 40 Arg VL 41 Gly Connected (base loop) None, stabilizingVH 115 Ser VH 1 Gln Connected (G4S)3 None, stabilizing

(VH permutation)

Mixed VH 115 Ser VH 1 Gln Connected (G4S)3 None, stabilizingscFv (classical scFv)

VH 29 Phe VL 70 Tyr Partial overlapImpossible, requires majorrearrangement

VH 29 Phe VL 66-68 backbone Partial overlapImpossible, requires majorrearrangement

VH 45 Leu VL 35 Phe Partial overlapImpossible, requires majorrearrangement

VL 34 Trp VH 6 Gln Partial overlapImpossible, requires majorrearrangement

VL 34 Trp VH 95 backbone Partial overlapImpossible, requires majorrearrangement

VH 38 Lys VL 44 Lys Close proximity Destabilizing, electrostatic repulsionVH 36 Gln VL 46 Glu Salt bridge removed Destabilizing

The designed linkers and connections in the pFv are long enough and ¯exible to not interfere with the ®nal Fv structure. Thehypothetical mixed Fv that might form from the pFv sequence in an alternative folding reaction contains structural incompatibilitiesthat make its formation highly unlikely and unstable. The most predominant ``¯aws'' in the mixed scFv structure are shown here.Position 1 and position 2 refer to pairs of amino acids which are either arti®cially connected (linkers, connectors), or which are inter-fering with a stable structure. More details are available upon request (G.V., VASM@helix.nih.gov).

Permutated Fv Fragment 111

Mono Q) and size exclusion chromatography tonear homogeneity. Figure 4 shows the recombinantpFv-immunotoxins present in the total cell pelletfrom induced cells, puri®ed inclusion bodies, andthe active refolded product. A comparison of thepuri®cation pro®les of scFv and pFv immunotoxinsshows that both proteins behave in a very similarmanner. The pFv-immunotoxin appears as charac-teristic chromatography peaks in the same pos-itions as scFv immunotoxins (data not shown).Although the amount of recombinant protein in in-duced cells and in inclusion bodies was the sameas with anti-Tac(pFv)-PE38 (Figure 4), there was alower yield of properly folded active pFv-immuno-toxin compared to the scFv immunotoxin. Using a

Figure 4. Production of anti-Tac pFv-PE38 immunotoxin.Reducing SDS-12% PAGE of total cell extracts of E. coliexpressing anti-Tac(pFv)-PE38, inclusion bodies contain-ing anti-Tac(pFv)-PE38, and pure anti-Tac(pFv)-PE38after refolding and puri®cation by ion exchange andsize exclusion chromatography. The molecular mass ofthe standard proteins are 94, 67, 43, 30, and 20 kDa.

standard refolding protocol (for antibodies and im-munotoxins) only �1% of the input protein wasobtained as monomeric pFv-toxin, compared to ayield of 4 to 6% with the anti-Tac scFv immunotox-in. We observed that this low yield correlates withthe formation of increased amounts of aggregatesduring refolding of the pFv compared to the scFv.The reason for increased aggregation may be thatthe folding of the pFv is less ef®cient than that ofthe scFv. On the other hand, the puri®ed properlyfolded pFv does not have an increased tendency toaggregate compared to the scFv (see the compari-son of stability below). It is possible that the refold-ing yield may be signi®cantly increased uponoptimization of the refolding conditions (Rudolph& Lilie, 1996).

Antigen binding of anti-Tac scFv and pFv

The binding af®nities of anti-Tac scFv and of thenewly constructed anti-Tac pFv to the IL2 receptoralpha subunit (p55) were determined by surfaceplasmon resonance (BIAcore). The recombinantscFv or pFv toxin fusions were applied to p55coated sensor chips. The sensorgrams of bindingand dissociation of anti-Tac scFv and pFv to thep55 coated chip are shown in Figure 5; both curvesappear very similar. Figure 5 (bottom) shows thatthe binding to the sensor chip is antigen speci®cbecause it can be competed by preincubation of theFv with soluble p55 antigen. A determination ofassociation and dissociation rates from several sen-sorgrams (see Materials and Methods) reveals akass of 1.4 � 105 to 2 � 105 for the scFv and 6 � 104

to 8 � 104 for the pFv. The small differences in thekass are within the range of experimental scatter.The dissociation rates of the pFv and scFv were

Figure 5. Speci®c binding of anti-Tac scFv and pFv to immobilized p55. BIAcore sensor chips coated with p55 (Tac-antigen) were incubated with 25 mg/ml anti-Tac scFv or pFv-immunotoxin at a ¯ow rate of 10 ml/min Hepes bufferedsaline (HBS) until saturation of binding was reached (®rst half of the curve, on-reaction). Then, the system was¯ushed with HBS to observe the dissociation from the chip (off-reaction, second part of the pro®le). For competitionof binding, anti-Tac(pFv) was preincubated for 15 minutes with a ®vefold excess (molar ratio) of p55 before appli-cation to the chip.

112 Permutated Fv Fragment

also very similar with a kdiss of 2.2 � 10ÿ4 to5.2 � 10ÿ4 for the scFv and 2.5 � 10ÿ4 to 3.6 � 10ÿ4

for the pFv. The KD at binding equilibrium, calcu-lated as KD � kdiss/kass is 1 to 4 nM for the scFvand 3 to 6 nM for the pFv (the slightly higher cal-

culated KD of the pFv results from the higher kass).These data indicate that the base-loop stabilizedpermutated form of anti-Tac Fv possesses a verysimilar antigen binding af®nity as does the singlechain Fv.

Figure 6. Speci®c cytotoxicity of anti-Tac(pFv)-PE38immunotoxin. A, Activity of anti-Tac pFv-PE38 towardsvarious cell lines. HUT-102 and MT-1 and ATAC-4 cellsexpress the TAC antigen; A-431, U-266 and CA-46 cellsare antigen negative (see Table 2). B, Activity of the cor-responding anti-Tac scFv-PE38 immunotoxin. C, Compe-tition of the cytotoxicity of anti-Tac pFv and scFv-PE38towards ATAC-4 cells by addition of excess anti-TacIgG demonstrates that the toxicity is mediated by the Fvmoiety of the immunotoxin.

Permutated Fv Fragment 113

Specific cytotoxicity of anti-Tac scFv andpFv-immunotoxins

PE38 is a truncated but enzymatically fully activeform of Pseudomonas exotoxin (Brinkmann &Pastan, 1994; Pastan et al., 1995). Fusion proteins ofantibody fragments with PE38 are cytotoxic to cellsthat bind and internalize the fusion protein but arenot cytotoxic to antigen negative cells. Thus, cyto-toxicity re¯ects speci®c antigen binding. Because

the degree of cytotoxicity to a given antigen (withde®ned antigen density) depends on binding af®-nity, cytotoxicity assays of Fv-immunotoxin fusionproteins can serve as an accurate test system forspeci®city as well as relative af®nity of engineeredantibody fragments (Brinkmann & Pastan, 1995).To test whether the pFv of anti-Tac retains notonly the af®nity but also speci®city, we analyzedand compared the activity of anti-Tac scFv- andpFv-immunotoxins towards different antigen ex-pressing and negative cell lines. The antigen (p55)expressing cell lines that we tested were MT-1 andHUT-102 leukemia cells. Antigen negative controlswere CA-46 (leukemia) and U-266 (myeloma) cells.To allow a direct comparison of antigen positiveand negative cells in an otherwise identical ``back-ground'', we also compared A-431 cells that do notcontain the IL2Ra with an A-431 cell line whichcontains an IL2 receptor expression plasmid andmakes IL2Ra (Kreitman et al., 1994). Figure 6 andTable 2 show that both anti-Tac immunotoxins arevery cytotoxic to antigen expressing cells while notaffecting antigen negative cells. Competition exper-iments in which excess anti-Tac IgG blocked theircytotoxicity con®rmed that their activity is antigenspeci®c (Figure 6C). The IC50 values of the scFv-im-munotoxin to the antigen expressing MT-1, HUT-102 and ATAC-4 cells are 0.25 ng/ml, 0.65 ng/mland 0.02 ng/ml, respectively. The IC50 values ofthe corresponding pFv-immunotoxins to MT-1,HUT-102 and ATAC-4 cells were 0.25 ng/ml,0.9 ng/ml and 0.06 ng/ml, respectively. Neithertype of anti-Tac Fv immunotoxin affected antigennegative cells (CA-46, U-266 and A-431) even at1000 ng/ml (Figure 6, Table 2). These results de-monstrate that anti-Tac scFv and pFv immunotox-ins are cytotoxic to p55 expressing cells with avery similar activity and with the same speci®city.We noticed that in some experiments the activityof the pFv was slightly decreased compared to thescFv-immunotoxins, but these differences were notlarge enough to be signi®cant (Table 2, and Discus-sion). We conclude that not only the binding af®-nity to P55 but also the speci®city of the anti-TacpFv is well preserved.

Comparison of the stability of anti-Tac pFv,scFv and dsFv

One important application of Fvs is the diagnosisand therapy of cancer, e.g. in the form of radio-labeled Fvs for tumor imaging or Fv-toxin fusionproteins for therapy (immunotoxins). For such ap-plications, it is important that Fvs are stable at37�C in human serum so they will retain activityfor some time after injection into patients. To ana-lyze the stability of our novel pFv and to compareit with Fvs that are stabilized by peptide linkers(scFv) or disul®des (dsFv), we assayed the stabilityof anti-Tac scFv, dsFv, or pFv-immunotoxins inPBS and human serum at 37�C (see Materials andMethods). Since the toxin moiety of PE-derivedimmunotoxin is very stable, any instability can be

Table 2. Af®nity and speci®c cytotoxicity of anti-Tac scFv and pFv-immunotoxins

Af®nity: Anti-Tac scFv Anti-Tac pFv

kass 1.4±2.0 � 105 6.0±8.1 � 104

kdiss 2.2±5.2 � 10ÿ4 2.5±3.6 � 10ÿ4

KD (nM) 1±4 3±6

Cytotoxicity: IC50, ng/ml anti-Tac(Fv)-PE38Cell line Tac-antigen scFv pFv

HUT-102 � 0.25 0.25MT-1 � 0.65 0.9CA-46 ÿ >1000 >1000U-266 ÿ >1000 >1000A-431 ÿ >1000 >1000ATAC-4a � 0.02 0.06

The af®nity to p55 antigen (IL2 receptor alpha subunit) was determined by surface plasmon reso-nance (BIAcore, Figure 5). kass, kdiss and KD (KD � kdiss/kass; binding at equilibrium) were calculatedfrom the sensorgrams using the BIAevaluation software package. Cytotoxicity was assayed in 96-well plates (see Figure 6 and Materials and Methods). IC50 is the concentration that causes 50%inhibition of protein synthesis after 20 hours incubation with immunotoxin.

a ATAC-4 is a recombinant cell line derived from A-431 that is stably expressing p55 (Kreitmanet al., 1994).

114 Permutated Fv Fragment

attributed to the binding moiety of the immuno-toxin (Brinkmann et al., 1993; Reiter et al., 1994b,c).Our stability analyses showed that for anti-Tac thedsFv fragment is the most stable of the three differ-ent Fv variants, with no loss of activity even afterthree days at 37�C (Reiter et al., 1994c). Analyses ofscFv and pFv-immunotoxins showed that both Fvslost �50% activity after 12 hours and 75% activityafter 24 hours. Thus, scFv and pFv-fragments ofanti-Tac have about the same stability.

Discussion

We have constructed a novel type of recombinantantibody Fv fragment (pFv for permutated Fv), inwhich the VH/VL heterodimer is covalently linkedand thereby stabilized at the two base loops thatconnect the VH beta strands 3 and 3b, and the VL

strands 3 and 3b. A pFv of the anti-Tac antibodyshowed very similar antigen binding af®nity as thesingle-chain Fv (VH and VL connected by a peptidelinker). A pFv-containing immunotoxin, anti-TacpFv-PE38, had the same cytotoxic speci®city toantigen expressing cancer cells as the correspond-ing scFv immunotoxin.

Design of pFv

To design the pFv connections, we used a structureof anti-Tac that was generated by homology mod-eling (Figure 1, Materials and Methods). The con-nections were visually inserted and the process didnot involve a computerized optimization such asenergy minimization of the resulting designedstructures. It is possible that pFv molecules can befurther improved using more sophisticated mol-ecular design techniques. The data presented herewith the anti-Tac Fv show that in principle it ispossible to obtain active stable pFv molecules. Be-cause the framework structures of all Fvs are very

similar, and because the designed pFv connectionsare in the (¯exible) base loops of VH and VL awayfrom the antigen binding region, it should be poss-ible to apply this method of stabilization to otherFvs. Since stable Fvs with good antigen bindingproperties cannot be prepared from all antibodiesusing current methods like single-chain or disul-®de-stabilized Fvs (Webber et al., 1995;Glockshuber et al., 1990; Bird et al., 1988; Hustonet al., 1988; Chaudhary et al., 1989; Brinkmann et al.,1993; Reiter et al., 1994a,b,c; Reiter et al., 1996),pFvs could be useful for antibody Fvs that cannotsuccessfully be generated with these othermethods. So far, we have only made and analyzeda pFv fragment of anti-Tac, which is quite stable asa scFv and dsFv (Webber et al., 1995; Reiter et al.,1994c). Further analyses with other antibodies areneeded to show whether the pFv design is appli-cable to other Fvs. The pFv approach could be par-ticularly useful for Fvs with an induced ®t bindingmode because such Fvs may have reduced bindingas dsFvs (U. B., unpublished results) and are rela-tively unstable as scFv (low interface energy; Riniet al., 1992; Stan®eld & Wilson, 1994).

Folding of pFv

An additional interesting aspect of pFvs may bethe mechanism by which these molecules fold. Ourexperimental data show that the ®nal structure ofthe original Fv and scFv and of the novel pFv arethe same. However, the primary sequence of thepFv gene is permutated and the chains are arti®-cially separated and reconnected. It is known thatother proteins can be permutated and correctlyfolded from a permutated sequence (Yang &Schachman, 1993; Goldenberg & Creighton, 1983;Goldenberg, 1989; Luger et al., 1989). But this is the®rst example demonstrating that permutation isalso feasible for antibody domains. We produce

Permutated Fv Fragment 115

pFv protein by refolding of the molecule startingwith bacterial inclusion bodies. In the process theprotein is totally unfolded, reduced and then re-folded. Unstabilized Fvs and linker and disul®de-stabilized Fv fragments assume their structure bythe relatively independent folding of their singledomains (VH and VL) which then assemble into theFv heterodimer. It is possible that pFvs fold in adifferent manner, by simultaneous folding and as-sembly of the domains. Such a folding pathwaycould allow the strands of VH and VL (halves ofthe beta sandwiches), which are totally separatedin the primary sequence, to come together to formthe barrel-like Fv structure more easily. The lowyield of properly refolded product of pFv com-pared to scFv and the formation of large amountsof aggregates during the refolding reaction also in-dicates that the refolding reaction of pFvs is differ-ent or at least not as ef®cient as the folding andassembly reaction of scFvs. Furthermore, the sym-metrical structure of antibodies, VL and VH, mayallow unproductive side reactions such as the for-mation of a mixed scFv (see Figure 1c, Table 1).We have no indication that this mixed scFv is pre-sent in our ®nal Fv preparation but it is likely thatthis competing folding reaction (leading to aggre-gation) reduces the yield of properly folded pFv.

The reduced refolding yields is a disadvantage forthe large scale production of pFvs from bacterialinclusion bodies, but optimization has not been ex-plored to increase the production yields. However,it is interesting that pFvs, once properly foldedand puri®ed, are relatively stable. To fully dis-sociate VH and VL in a pFv, it is probably necess-ary to break the barrel-like sandwich structures.This would require a much greater energy than thedissociation of VH and VL in a scFv. Stability analy-ses with anti-Tac (which is already stable as anscFv; Webber et al., 1995; Reiter et al., 1994b,c)show that the pFv is as stable as the scFv. Furtheranalyses are required to analyze the stability andfolding and unfolding patterns of pFvs in more de-tail, and compare them with scFv and dsFvs.

Materials and Methods

Computer modeling and analyses

A model of the Fv fragment of the murine anti-Tac anti-body (anti IL2 receptor alpha chain; Webber et al., 1995;Chaudhary et al., 1989; Uchiyama et al., 1981) was ob-tained by ``homology-modeling'' the known anti-Tacprotein sequence into building blocks of known antibodycrystal structures as previously described (Vasmatziset al., 1995). The ribbon ®gures of the modeled structureswere created by the program ribbons (Carson, 1987). Theamino acids were numbered according to the Kabatscheme (Kabat et al., 1991). The peptide linkers and con-nectors in the scFv and pFv structures were manuallyadded without further energy minimization. It is knownthat such linkers, at least the (Gly4Ser)3 linker that weused, are ¯exible without a preferred structure (Freundet al., 1994). The ``mixed'' scFv structure was generatedby matching overlay of the VH and VL C structures (VH

on VL and VL on VH), deletion of the appropriate N- orC-terminal ``halves'' of the domains from the overlaystructure followed by fusion of the remaining halves intothe mixed scFv (see Figure 3B).

Plasmid constructions

The plasmid pRK79 encodes the scFv-immunotoxin,anti-Tac(scFv)-PE38 (Chaudhary et al., 1989; Kreitmanet al., 1994). Anti-Tac scFv is derived from the murinemAb anti-Tac, directed against IL2-receptor alpha (p55)subunit. PE38 is a truncated but enzymatically fully ac-tive form of Pseudomonas exotoxin (Brinkmann & Pastan,1994; Pastan et al., 1995; Hwang et al., 1987). Proteins inwhich the antibody fragments are fused to PE38 are cy-totoxic to cells that bind the antibody, but not to antigennegative cells (Brinkmann & Pastan, 1994; Pastan et al.,1995). The plasmid pADC1 that encodes the anti-TacpFv fused to PE38 was generated by overlap extensionPCR using the anti-Tac scFv plasmid as the PCR tem-plate as described in detail in Figure 3. The primerPL5FOR, 50GGCGGCCATATGCAAATTGTTCTCACCC-AGTCTCCAGCA30 was used to provide the 50 end andATG start codon of VL (NdeI site underlined). PL5REV,50GCCACCGGCAGAAGAGCCTGGCTTCTGCTGGAA-CCA30 and PH3FOR, 50CAGAAGCCAGGCTCTTCT-GCCGGTGGCGGACAGGGTCTGGAATGGATTGGA30were used to fuse the N-terminal half of VL to the C-terminal half of VH. The C-terminal half of VH waslinked to the N-terminal half of VH via a (Gly4Ser)3 linkerby the primers PH3REV, 50CAGATGGACCTGAGA-GCCGCCGCCACCCGAGCC30 and PH5FOR, 50TCGGG-TGGCGGCGGCTCTCAGGTCCATCTGCAGCAGTCTG-GGCCT30. PH5REV, 5 0TCCACCAGCAGAAGACCT-CTGTTTTACCCAGTGCATCCTGTA30 and PL3FOR,50AACAGAGGTCTTCTGCTGGTGGAGGCACTTCTCC-CAAACTCTGGATTT30 were used to fuse the N-terminalhalf of VH to the C-terminal half of VL. The complemen-tary overlaps of the primer pairs that enable overlap ex-tension are underlined. The primers PL5FOR andPL3REV, 50TCCGGAAGCTTTGAGCTCCAGCTTGGTC30were used to amplify the complete overlap extended pFvsequence. To construct the pFv gene, separate PCR reac-tions (20 cycles 94 to 55 to 72�C, one minute each) wereperformed with primer pairs PL5FOR/REV, PH3FOR/REV, PH5FOR/REV and PL3FOR/REV. The resultingfragments VL5, VH3, VH5, and VL3, respectively, wereseparated on a low melting point agarose gel and recov-ered as gel pieces. The pFv gene was assembled by twosubsequent steps of overlap extension PCR; ®rst VLNwas joined to the VHC and VHN to VLC, then the re-sulting VLN-VHC and VHN-VLC fragments were fusedand ampli®ed. The strategy for assembly of the ®nal pFvfrom the PCR fragments is shown in detail in Figure 3B.Two separate 50 ml PCR reactions were performed for®ve cycles without any PCR primer; one contained 1 mlmolten gel with fragment VL5 and fragment VH3, theother fragments VH5 and VL3. After ®ve cycles, both re-actions, now containing joined VL5-H3 and VH5-L3 frag-ments, were combined and subjected to ®ve furthercycles without primers to generate the VL5-H3-H5-L3fragment (see Figure 3B). Then primer PL5FOR andPL3REV were added to the 100 ml ®nal reaction mix, andampli®ed for 20 cycles to obtain the ®nal pFv gene. High®delity polymerase mix (Boehringer Mannheim) wasused to avoid PCR errors. The pFv gene was cloned asan NdeI/HindIII fragment into pULI7 (Brinkmann &Pastan, 1994), a vector for fusion of antibody fragmentsto PE38, a truncated form of Pseudomonas exotoxin

116 Permutated Fv Fragment

(Brinkmann & Pastan, 1995). The NdeI-site provides thetranslation start site and the HindIII site fuses the scFvgene in frame to the truncated toxin (Hwang et al., 1987).The vector contains the phage T7 promoter for ex-pression in Studier's E. coli BL21/lDE3 expression sys-tem (Studier & Moffatt, 1986). The pFv gene was provento be correct by DNA sequencing on an ABI 373A se-quencer using the dideoxy chain terminator sequencingkit.

Production of recombinant proteins

Anti-Tac scFv- and pFv immunotoxins were expressed inE. coli BL21/lDE3, and accumulated in inclusion bodies(IBs) as previously described for other recombinant im-munotoxins (Brinkmann & Pastan, 1995). IBs were solu-bilized in GuCl, reduced with DTE and refolded bydilution in a refolding buffer containing arginine to pre-vent aggregation and glutathione to facilitate redox shuf-¯ing (Buchner et al., 1992). Active protein was puri®edfrom the refolding solution by ion exchange and size ex-clusion chromatography to near homogeneity as de-scribed (Buchner et al., 1992). Protein concentrationswere determined by Bradford assay (BioRad CoomassiePlus).

Binding assays

The af®nity of anti-Tac scFv and pFv was assayed andcompared by surface plasmon resonance (BIAcore,Pharmacia Biosensor). Recombinant Tac antigen (p55,Il-2 receptor alpha; Sauve et al., 1991), was coupled toBIAcore sensor-chips (CM5, research grade, PharmaciaBiosensor) according to the manufacturer's speci®ca-tions. scFv or pFv immunotoxin was applied to thechips and binding and dissociation (kass and kdiss) deter-mined from binding and release curves of the sensor-grams with the BIAevaluation software package(Pharmacia Biosensor). KD at equilibrium was calcu-lated as KD � kdiss/kass.

Cytotoxicity assays

The cytotoxicity of the anti-Tac(Fv)-immunotoxins wasassessed by protein synthesis inhibition assays (inhi-bition of incorporation of tritium labeled leucine into cel-lular protein) in 96-well plates as previously described(Brinkmann et al., 1991). IC50 is the toxin concentrationthat reduces incorporation of radioactivity by 50% com-pared to cells that were not treated with toxin. For com-petition assays, 10 mg/ml (®nal concentration) of mAbanti-Tac was added 15 minutes prior to the addition ofimmunotoxin.

Stability assays

The stability of the anti-Tac(Fv) immunotoxins was de-termined by incubating them at 10 mg/ml at 37�C inhuman serum (Reiter et al., 1994c). Active immunotoxinremaining after incubation was determined by cytotox-icity assays on HUT-102 cells.

Acknowledgements

We thank R. Kreitman for the anti-Tac scFv-PE38 ex-pression plasmid pRK79, J. Hakimi for providing recom-

binant IL2-receptor alpha p55 protein and anti-Tac IgG,V. Fogg and I. Margulies for cell culture, Dr M. Carsonfor the use of the program ribbons, Joe Cammisa forhelp with computer graphics and Althea Jackson, JennieEvans and Robb Mann for editorial assistance. A.D.C.is partially supported by the Leonardo Di CapuaAssociation, Italy.

References

Bird, R. E., Hardman, K. D., Jacobson, J. W., Johnson, S.,Kaufman, B. M., Lee, S. M., Lee, T., Pope, S. H.,Riordan, G. S. & Whitlow, M. (1988). Single-chainantigen-binding proteins. Science, 242, 423±426.

Brinkmann, U. & Pastan, I. (1994). Immunotoxinsagainst cancer. Biochim. Biophys. Acta, 1198, 27±45.

Brinkmann, U. & Pastan, I. (1995). Recombinant immu-notoxins: from basic research to cancer therapy.Methods, 8, 143±156.

Brinkmann, U., Pai, L. H., FitzGerald, D. J., Willingham,M. & Pastan, I. (1991). B3(Fv)-PE38KDEL, a single-chain immunotoxin that causes complete regressionof a human carcinoma in mice. Proc. Natl Acad. Sci.USA, 88, 8616±8620.

Brinkmann, U., Reiter, Y., Jung, S. H., Lee, B. & Pastan,I. (1993). A recombinant immunotoxin containing adisul®de-stabilized Fv fragment. Proc. Natl Acad.Sci. USA, 90, 7538±7542.

Buchner, J., Pastan, I. & Brinkmann, U. (1992). Amethod to increase the yield of properly foldedrecombinant fusion proteins: single-chain immuno-toxins from renaturation of bacterial inclusionbodies. Anal. Biochem. 205, 263±270.

Carson, M. (1987). Ribbon models of macromolecules.J. Mol. Graph. 5, 103±106.

Chaudhary, V. K., Queen, C., Junghans, R. P.,Waldmann, T. A., FitzGerald, D. J. & Pastan, I.(1989). A recombinant immunotoxin consisting oftwo antibody variable domains fused to Pseudomo-nas exotoxin. Nature, 339, 394±397.

Freund, C., Ross, A., Pluckthun, A. & Holak, T. A.(1994). Structural and dynamic properties of the Fvfragment and the single-chain Fv fragment of anantibody in solution investigated by heteronuclearthree-dimensional NMR spectroscopy. Biochemistry,33, 3296±3303.

Glockshuber, R., Malia, M., P®tzinger, I. & Pluckthun,A. (1990). A comparison of strategies to stabilizeimmunoglobulin Fv-fragments. Biochemistry, 29,1362±1367.

Goldenberg, D. P. (1989). Circularly permuted proteins.Protein Eng. 2, 493±495.

Goldenberg, D. P. & Creighton, T. E. (1983). Circularand circularly permuted forms of bovine pancreatictrypsin inhibitor. J. Mol. Biol. 165, 407±413.

Huston, J. S., Levinson, D., Mudgett-Hunter, M., Tai,M. S., Novotny, J., Margolies, M. N., Ridge, R. J.,Bruccoleri, R. E., Haber, E., Crea, R. & Oppermann,H. (1988). Protein engineering of antibody bindingsites: recovery of speci®c activity in an anti-digoxinsingle-chain Fv analogue produced in Escherichiacoli. Proc. Natl Acad. Sci. USA, 16, 5879±5883.

Hwang, J., FitzGerald, D. J., Adhya, S. & Pastan, I.(1987). Functional domains of Pseudomonas exotoxinidenti®ed by deletion analysis of the gene expressedin E. coli. Cell, 48, 129±136.

Kabat, E. S., Wut, T. T., Perry, H. M., Gottesman, K. S. &Foeller, C. (1991). Sequences of Proteins of Immuno-

Permutated Fv Fragment 117

logical Interest, 5th edit., NIH Publication, NationalInstitutes of Health, Bethesda, MD, pp. 91±3242.U.S. Dept Health Human Services.

Kreitman, R. J., Bailon, P., Chaudhary, V. K., FitzGerald,D. J. & Pastan, I. (1994). Recombinant immunotox-ins containing anti-Tac(Fv) and derivatives of Pseu-domonas exotoxin produce complete regression inmice of an interleukin-2 receptor-expressing humancarcinoma. Blood, 83, 426±434.

Luger, K., Hommel, U., Herold, M., Hofsteenge, J. &Kirschner, K. (1989). Correct folding of circularlypermuted variants of a beta alpha barrel enzymein vivo. Science, 243, 206±210.

Pastan, I., Pai, L. H., Brinkmann, U. & FitzGerald, D. J.(1995). Recombinant toxins: new therapeutic agentsfor cancer. Ann. NY Acad. Sci. 758, 345±354.

Reiter, Y., Brinkmann, U., Jung, S.-H., Lee, B., Kasprzyk,P. G., King, C. R. & Pastan, I. (1994a). Improvedbinding and anti tumor activity of a recombinantanti-erbB2 immunotoxin by disul®de-stabilization ofthe Fv fragment. J. Biol. Chem. 269, 18327±18331.

Reiter, Y., Brinkmann, U., Jung, S.-H., Lee, B. & Pastan,I. (1994b). Engineering disul®de bonds into con-served framework regions of Fv fragments: recom-binant immunotoxins containing disul®de-stabilizedFv with improved biochemical characteristics. Pro-tein Eng. 7, 697±704.

Reiter, Y., Kreitman, R. J., Brinkmann, U. & Pastan, I.(1994c). Cytotoxic and antitumor activity of arecombinant immunotoxin composed of disul®de-stabilized anti-Tac Fv fragment and truncated Pseu-domonas exotoxin. Int. J. Cancer, 58, 142±149.

Reiter, Y., Brinkmann, U., Lee, B. & Pastan, I. (1996). En-gineering antibody Fv fragments for cancer detec-tion and therapy: disul®de-stabilized Fv fragments.Nature/Biotechnology, 14, 1239±1245.

Rini, J. M., Schulze-Gahmen, U. & Wilson, I. A. (1992).Structural evidence for induced ®t as a mechanismfor antibody-antigen recognition. Science, 255, 959±965.

Rudolph, R. & Lilie, H. (1996). In vitro folding of in-clusion body proteins. FASEB J. 10, 49±56.

Sauve, K., Nachman, M., Spence, C., Bailon, P.,Campbell, E., Tsien, W. H., Kondas, J. A., Hakimi,J. & Ju, G. (1991). Localization in human interleukin2 of the binding site to the alpha chain (p55) of theinterleukin 2 receptor. Proc. Natl Acad. Sci. USA, 88,4636±4640.

Skerra, A. & Pluckthun, A. (1988). Assembly of a func-tional immunoglobulin Fv fragment in Escherichiacoli. Science, 240, 1038±1041.

Stan®eld, R. L. & Wilson, I. A. (1994). Antigen-inducedconformational changes in antibodies: a problem forstructural prediction and design. Trends Biotech. 12,275±279.

Studier, F. W. & Moffatt, B. A. (1986). Use of bacterio-phage T7 RNA polymerase to direct selective high-level expression of cloned genes. J. Mol. Biol. 189,113±130.

Uchiyama, T., Broder, S. & Waldmann, T. A. (1981). Amonoclonal antibody (anti-Tac) reactive with acti-vated and functionally mature human T cells. I.Production of anti-Tac monoclonal antibody anddistribution of Tac (�) cells. J. Immunol. 4, 1393±1397.

Vasmatzis, G., Cornette, J., Sezerman, U. & DeLisi, C.(1995). TCR recognition of the MHC-peptide dimer:structural properties of a ternary complex. J. Mol.Biol. 261, 72±89.

Webber, K. O., Reiter, Y., Brinkmann, U., Kreitman,R. J. & Pastan, I. (1995). Preparation and character-ization of a disul®de-stabilized Fv fragment of theanti-Tac antibody: comparison with its single-chainanalog. Mol. Immunol. 4, 249±258.

Yang, Y. R. & Schachman, H. K. (1993). Aspartate trans-carbamoylase containing circularly permuted cataly-tic polypeptide chains. Proc. Natl Acad. Sci. USA, 90,11980±11984.

Yokota, T., Milenic, D. E., Whitlow, M. & Schlom, J.(1992). Rapid tumor penetration of a single-chainFv and comparison with other immunoglobulinforms. Cancer Res. 52, 3402±3408.

Edited by P. E. Wright

(Received 9 September 1996; received in revised form 12 December 1996; accepted 13 December 1996)