Structure-Function Analysis of FLT3 Ligand-FLT3 Receptor Interactions Using a Rapid Functional Screen*

FLT3 ligand (FLT3L) stimulates primitive hematopoietic cells by binding to and activating the FLT3 receptor (FLT3R). We carried out a structure-activity study of human FLT3L in order to define the residues involved in receptor binding. We developed a rapid method to screen randomly mutagenized FLT3L using a FLT3R-Fc fusion protein to probe the relative binding activities of mutated ligand. Approximately 60,000 potential mutants were screened, and the DNA from 59 clones was sequenced. Thirty-one single amino acid substitutions at 24 positions of FLT3L either enhanced or reduced activity in receptor binding and cell proliferation assays. Eleven representative proteins were purified and analyzed for receptor affinity, specific activity, and physical properties. Receptor affinity and bioactivity were highly correlated. FLT3L affinity for receptor improved when four individual mutations that enhance FLT3L receptor affinity were combined in a single molecule. A model of FLT3L three-dimensional structure was generated based on sequence alignment and x-ray structure of macrophage colony-stimulating factor. Most residues implicated in receptor binding are widely dispersed in the primary structure of FLT3L, yet they localize to a surface patch in the tertiary model. A mutation that maps to and is predicted to disrupt the proposed dimerization interface between FLT3L monomers exhibits a Stokes radius that is concentration-dependent, suggesting that this mutation disrupts the FLT3L dimer.

FLT3 ligand (FLT3L) 1 binds to and activates the cell surface tyrosine kinase receptor, FLT3R, initiating signaling events that regulate the proliferation and differentiation of multiple lineages of cells of the hematopoietic system (1)(2)(3). Alone, FLT3L stimulates little growth of highly purified bone marrow progenitor cells in vitro, yet when FLT3L is combined with other growth factors it has potent synergistic proliferative effects on these hematopoietic precursors (1,4). The human FLT3R has been cloned and shown to belong to a subfamily of structurally related tyrosine kinase receptors that contain five extracellular Ig-like domains and an intracellular tyrosine kinase domain (5). Whereas FLT3R is expressed in a limited number of tissues, including human bone marrow, thymus, spleen, liver, and lymph nodes, FLT3L is widely expressed in human tissue (6,7).
The human FLT3L gene encodes a 235-amino acid type I transmembrane protein consisting of four domains: 1) an Nterminal 26-residue signal peptide, 2) a 156-residue extracellular domain, 3) a 23-amino acid transmembrane domain, and 4) a 30-residue cytoplasmic domain (1,2,7). Soluble FLT3L is thought to be released into circulation from the cell membrane by protease cleavage (8). Soluble FLT3L is a noncovalently linked dimer and contains six cysteine residues that apparently form intramolecular disulfides. Karplus and co-workers sorted the three-dimensional coordinates of nine-helical bundle cytokines into three structural subfamilies: long, intermediate, and short chain (9). Although the three-dimensional structure of human FLT3L is unknown, sequence comparisons with other members of the four-helix bundle protein family whose threedimensional structures are known, imply that it is a member of the left-handed short chain antiparallel four-helix bundle cytokine structural subfamily (1). The three-dimensional structure of five members of the short chain helical cytokines have been solved, including interleukin-4, granulocyte-macrophage colony-stimulating factor, interleukin-2, M-CSF, and interleukin-5 (9). Structural features common to these proteins include an up-up-down-down antiparallel topology for helices A-D and two short strands of ␤-sheet contained in the long cross-over segments between helices A-B and C-D. The unique up-updown-down topology for helices A-D, similar genetic organization of exons and introns, distant relatedness observed between some members, and similar biological function indicate that helical bundle cytokines arose from divergent evolution of a common ancestral protein. It is widely believed that helical cytokines initiate signal transduction by ligand-induced receptor oligomerization (10).
The current methods for site-directed mutagenesis are powerful tools for the study of structure and function of proteins and are highly efficient and effective when some idea of where to make the specific mutation is available. However, in the case of FLT3L, the protein region of interest is not well defined, so it is difficult to predict which mutation to make in order to bring about a desired change. The cross-reactivity of murine and human FLT3L for FLT3R (7) precludes the potential to identify residues of interest by swapping interspecies segments of polypeptide between these ligands. Comprehensive mutational studies of other four-helix bundle proteins may not be applicable to FLT3L because a number of these species are monomeric and bind class I hematopoietic receptors, whereas FLT3L is dimeric and binds to and activates a class III tyrosine kinase receptor. Therefore, the method employed in this study of FLT3L structure-function relationship is random mutagenesis of the DNA encoding FLT3L coupled with an activity-screening system.

EXPERIMENTAL PROCEDURES
Plasmid Manipulation and DNA Sequencing-The multicopy expression vector paADH2 was described previously (11). This plasmid has the 2 origin of replication, the ADH2 promoter to drive expression of foreign genes, and the ␣-factor leader to direct secretion of heterologous proteins. Vector PIXY456 is derived from paADH2 by the addition of a BamHI site adjacent to the Asp718 site. Primers JM37 (AATTAGGT-ACCTTTGGATAAAAGACTCAGTGGGACCCAGGAC) and P11673 (ATATGGATCCCTACGGGGCTGTGGCCTCCAGGGGCCG) were used to amplify human FLT3L from FLT3L-clone 9 cDNA (7). JM37 primer contains an Asp718 site 20 bases from the 3Ј-end of the ␣-factor and fuses the ␣-factor leader to FLT3L. P11673 contains a BamHI site for cloning into the PIXY456 expression vector. The amplified fragment was digested with Asp718 and BamHI and ligated to PIXY456 to construct PIXY771. The Asn-Phe-Ser glycosylation site at positions 123-125 was mutated to Asn-Phe-Ala using primer JM40(CCTCCTG-CAGGAGACCTCCGAGCAGCTGGTGGCGCTGAAGCCCTGGATCAC-TCGCCAGAACTTCGCCCGGTGCCTG). Primers JM40 and P11673 were used to amplify the 3Ј-end of the human FLT3L cDNA. This amplified fragment was digested with BsaI and BamHI and used to replace the BsaI-BamHI fragment in PIXY771, resulting in plasmid PIXY797. Plasmid DNA was rescued from yeast clones (12) and was sequenced with synthetic primers that would hybridize to the coding and noncoding strands of the vector DNA 5Ј and 3Ј to the FLT3L gene.
Multiple amino acid substitutions occurred as single independent isolates in 13 of the 59 FLT3L DNA sequences. Single mutations were isolated by subcloning gene fragments. In this manner, we were able to assign function to individual residues for the multiply substituted proteins (Table I).
We combined several of the FLT3L ϩϩ mutations into a single molecule. A mutant combining W118R and Q122R was constructed by PCR mutagenesis but was poorly expressed and not characterized further. A mutant combining K84E and Q122R was constructed by subcloning gene fragments. A multiple mutant containing LϪ3H, H8Y, K84E, and Q122R was constructed by PCR mutagenesis. The 5Ј oligonucleotide JM116.46 (TGGATAAAAGAcacAGTGGGACCCAGGACTGCTCCTTC-CAATAcag) encoding the LϪ3H and H8Y (mutagenized codons in lowercase type) was used to amplify the FLT3L using the the double mutant K84E/Q122R as a template and a 3Ј vector primer. A second PCR reaction was used to extend the 5Ј-end using the oligonucleotide JM117.42 (TGGATAAAAGACACAGTGGGACCCAGGACTGCTCCTT-CCAATACAG) and 3Ј vector primer. This was introduced into PIXY456 by recombination as described below for the mutagenized PCR fragments. All constructs were confirmed by DNA sequence analysis.
A soluble form of the human FLT3R was generated and purified as described for murine FLT3R (2). PCR was used to clone the human FLT3R cDNA described by Rosnet et al. (13). The entire extracellular domain of the human FLT3R (amino acids 1-541) was fused to the Fc portion of human IgG.
Mutagenesis and Transformation-Mutagenic PCR was performed as described previously (14). Mutagenized FLT3L cDNA was introduced into the PIXY456 expression vector by homologous recombination (15). The PCR fragment has a 30-base pair segment 5Ј to and a 60-base pair segment 3Ј to the FLT3L gene that are homologous to the regions immediately adjacent to the gap of vector DNA. Transformants were selected on synthetic complete minus tryptophan medium (16).
Screening-Transformed colonies, 0.5-2 mm in diameter, were lifted onto cellulose acetate membranes. A second "capture" membrane (nitrocellulose) was coated with 0.1 mg/ml monoclonal antibody M5␣-FLT3L in 50 mM phosphate, pH 7.4, 150 mM NaCl (PBS) for 16 h at 4°C. The membrane was subsequently blocked for 4 h at 4°C in PBS containing 3% (w/v) bovine serum albumin, 5% (w/v) nonfat dry milk (blocking solution) and then washed twice in PBS. The wet capture nitrocellulose membrane was placed on a YEPD plate (16), care taken to avoid air pockets, and covered in turn with the cellulose acetate membrane with yeast colonies uppermost, avoiding air pockets. The membrane sandwich was punctured with a needle for later orientation, and the assembly was incubated at 30°C for 18 -22 h. The cellulose acetate membrane with colonies was removed and placed on a fresh YEPD plate and stored at 4°C for later recovery of the yeast.
The capture membrane was then used to detect receptor binding to immobilized FLT3L. This protocol was carried out at room temperature, and all reagents and washes were in PBS containing 1% (w/v) bovine serum albumin unless otherwise stipulated. The capture nitrocellulose membrane was blocked for 1 h in blocking solution and then washed three times. The membrane was then probed for 1 h with 0.5 g/ml human FLT3R-Fc, washed three times over 10 min, probed 1 h with a 0.53 g/ml horseradish peroxidase-goat anti-human IgG, Fc␥ conjugate (Jackson ImmunoResearch Laboratories), washed three times over 30 min, washed in PBS, and developed with 4-chloro-1-napthol (Bio-Rad) in PBS plus 18% methanol and 0.002% hydrogen peroxide.
Single yeast colonies whose receptor binding signal differed from the wild type background were isolated and plated on selective medium. Two-ml YEPD cultures of these chosen colonies were grown as described (11), and medium was used immediately for screens.
ELISA and Cell Proliferation Assay-The ELISA used in this study was described previously (17). Murine WWF7 cells were assessed in [ 3 H]thymidine incorporation assays, as described earlier (6).
BAF/B03 cells were engineered to express the full-length human FLT3R using the same procedure previously described (2) for producing BAF/B03 cells expressing the murine FLT3 receptor. Expression of human FLT3R by the cells was confirmed by examining the capacity of the cells to proliferate in response to soluble FLT3L and by flow cytometric analysis using biotinylated FLT3L.
Recombinant Human FLT3L Preparation-Wild type and mutant FLT3L proteins were purified to greater than 90% homogeneity as follows. 1.2 liters of yeast medium were filtered through a 0.22-m membrane, the pH was adjusted to 4.0 by the addition of glacial acetic acid with rapid mixing, and the mixture was filtered through a 0.22-m membrane again (conductivity 2-6 milliohms). This filtrate was applied to a 30-ml Fractogel EMD SO 3 -650 (M) (EM Separations) equilibrated with 25 mM NaCH 3 COO, 50 mM NaCl, pH 4.0, at 20 ml/min. Protein was eluted with 25 mM MES, 200 mM NaCl, pH 6.0. The pH was adjusted to 7.5-8 by the addition of 1 ⁄20 volume of 1 M Tris, pH 9.0, and the solution was then filtered through a YM100 membrane (Amicon). FLT3L was affinity-purified by passing the filtrate over a monoclonal antibody (M5␣-FLT3L; Ref. 8; CNBr-activated Sepharose 4B, Amersham Pharmacia Biotech) column equilibrated in 50 mM NaHPO 4 , 300 mM NaCl, pH 7.4. Protein was eluted with 25 mM NaHPO 4 , pH 11.3, and the eluant was neutralized by the addition of 1 ⁄100 volume of 1 M monobasic phosphate. The eluant was concentrated and applied to a Superdex 200 column (Amersham Pharmacia Biotech) equilibrated in PBS at 2.5 ml/min. The dimer peak was collected, concentrated, filtered through a 0.22-m membrane, and stored at 4°C. The concentration of purified FLT3L protein was determined by quantitative amino acid analysis after acid hydrolysis. The level of contamination of purified protein was assessed by running 2-4 g of various FLT3L proteins on 16% SDS-polyacrylamide gels and visualized with colloidal Coomassie stain (Fig. 3). In all cases, the proteins were a single species (Ͼ90%) with a M r of 21,000.
In this study, the amino acid sequence of FLT3L is numbered according to the mature fully processed soluble protein, whose amino-terminal residues are Thr-Gln-Asp (positions 1, 2, and 3, respectively (7)). The gene construct we utilized for protein expression in yeast encodes an additional 3 amino acids (Leu-Ser-Gly) amino terminus of the mature protein. Amino-terminal sequence analysis of purified yeast-derived FLT3L shows that 80% of the N-terminal sequence is Leu-Ser-Gly-Thr-Gln-Asp, while 20% begins at the penultimate residue, Ser. The specific activity and K ai for receptor binding of this yeast-derived protein (K ai ϭ 2.2 ϫ 10 10 M Ϫ1 ) used in this study is similar to FLT3L expressed in mammalian cells that has native mature sequence (K ai ϭ 2.5 ϫ 10 10 M Ϫ1 ), and will be referred to as wild type FLT3L. K ai is defined as the affinity constant of unlabeled FLT3L for FLT3R as determined by inhibition of binding of 125 I-labeled wild type FLT3L.
Physical Characterization-Gel filtration chromatography of purified proteins was performed on a 300 ϫ 7.8-mm Bio-Sil 125-5 column (Bio-Rad) in PBS at a flow rate of 1 ml/min. Circular dichroism spectra of proteins were obtained using a Jasco 500 C spectropolarimeter interfaced with an IBM AT computer. All spectra were measured in a 0.1-mm cell in PBS at 22°C, between 260 and 195 nm using a 1-nm bandwidth and a 1-s time constant. The percentage of helical content was estimated as described previously by Sreerama and Woody (18).
Radiolabeling, Binding Assays, and Data Analysis-Purified recombinant FLT3L was labeled with 125 I using a solid phase chloramine T analog (Pierce) to a specific radioactivity of 4 ϫ 10 14 cpm/mmol with no detectable loss of specific binding activity as assessed by inhibition assays with unlabeled FLT3L. Binding assays and data analysis were performed as described previously (19). Briefly, BAF-BO3 cells transfected with human FLT3R cDNA (0.5-1 ϫ 10 6 ) were incubated with serial dilutions of 125 I-FLT3L in binding media (RPMI 1640, 2.5% bovine serum albumin, 0.2% NaN 3 , 20 mM Hepes, pH 7.2) in 96-well microtiter plates maintained on a mini-orbital shaker (Bellco) at 37°C for 90 min. Inhibition binding assays were carried out by holding the radiolabeled FLT3L concentration constant at 0.3 nM, while unlabeled competitor proteins were ranged from 150 to 0.001 nM.
Modeling of FLT3L-We generated the quaternary FLT3L structure using FOLDER, a distance geometry-based method for homology modeling (20). FOLDER uses a sequence alignment between a template and model protein to identify residues in topologically equivalent positions. For topologically nonequivalent atoms such as variable loops and some side chains, chemical constraints, standard geometrical parameters, and chemical information like disulfide cross-links are used to compute a set of distances between these atoms, which is appended to the set of distances for topologically equivalent atoms. Distance geometry methods are a powerful technique for structure elucidation when modeling proteins with low sequence identity to the template. The x-ray crystallographic coordinates for M-CSF (21) 2 served as the structural template. M-CSF is the most closely related protein to FLT3L, where a crystal structure has been solved. The amino acid sequence alignment between M-CSF and FLT3L described previously is based on genetic and structural constraints and is shown in Fig. 1 (1). FLT3L is a noncovalently linked dimer, and we modeled FLT3L with the assumption that the dimer interface would be the same in M-CSF and FLT3L. We used the computer graphics program InsightII to generate the images in Fig. 5.

RESULTS
Isolation of FLT3L Mutants-The screening protocol used in this study is described in Fig. 2 and in experimental procedures. We lifted yeast colonies transformed with mutagenized FLT3L cDNA onto a cellulose acetate membrane, which was then overlaid onto a nitrocellulose capture membrane previously coated with a non-neutralizing antibody to FLT3L (M5␣FLT3L). Protein secreted by yeast passed through the cellulose acetate membrane and was captured by the M5␣FLT3L immobilized on the nitrocellulose membrane. The capture membrane was probed with human FLT3R-Fc followed by a reporter reagent, horseradish peroxidase-goat anti-human IgG. The receptor binding properties of the secreted FLT3L were assessed visually by noting the variation in intensity of the stained spots after the enzyme-linked reaction. Most colonies secreted FLT3L, which stained similarly to protein secreted from yeast expressing wild type FLT3L. About 1% of the colonies gave rise to spots devoid of stain ( Fig. 2A). We designated these clear white spots as receptor binding deficient (FLT3L Ϫ ) and isolated respective individual yeast colonies by relating to their positions on the cellulose acetate membrane.
The appearance of stained spots whose intensity was greater than that of wild type was a far rarer event (Fig. 2B). We isolated these colonies and designated their phenotypes FLT3L ϩϩ , species that exhibit increased receptor binding properties. In this manner, we screened approximately 60,000 colonies for receptor binding, with up to 300 colonies assayed per standard size Petri dish (82-mm diameter). We isolated 214 FLT3L Ϫ colonies, representing only a portion of the FLT3L Ϫ species, and 114 FLT3L ϩϩ species, representing all of the de- Top, mutagenized DNA encoding FLT3L was transformed into yeast, and colonies were lifted onto a cellulose acetate membrane. The cellulose acetate membrane was overlaid with a nitrocellulose membrane coated with M5␣FLT3L. After expression of FLT3L secreted from yeast and captured on the coated nitrocellulose membrane, the capture membrane was probed with FLT3R-Fc followed by goat anti-human Fc-horseradish peroxidase. A, enzyme-linked signal on nitrocellulose after development from FLT3L Ϫ clones. B, enzymelinked signal on nitrocellulose after development from FLT3L ϩϩ clones. tectable FLT3L ϩϩ colonies.
In order to eliminate mutants that expressed poorly or had grossly altered structures, we used Western blots to analyze the growth medium from all 328 FLT3L Ϫ and FLT3L ϩϩ clones. We discarded mutants because little or no protein was secreted, because only high molecular weight smears were observed (possibly due to hyperglycosylation), and several because of lower than expected molecular weight. We determined the specific activity of the remaining proteins by subjecting the yeast medium to an ELISA and a WWF7 cell proliferation assay, which measure the levels of expression and biological activity, respectively. We discarded additional species with specific activities near that of wild type. We subjected a final 30 FLT3L Ϫ species and 29 FLT3L ϩϩ species to DNA sequence analysis.
The 59 FLT3L mutants isolated and sequenced in this study are presented in Table I along with their specific activity. We identified 31 amino acid substitutions at 24 sites that alter FLT3L activity. When we plot our collection of 24 positions along the primary structure of FLT3L, presented in Fig. 3, three linear clusters of high frequency amino acid substitution, or "hot spots," appear at positions 8 -15, 81-87, and 116 -124. Each of the three hot spots contain amino acid substitutions that improve the binding and biological activity of FLT3L, positions 8, 84, 118, and 122.
Mapping of these hot spots onto our FLT3L model indicate that mutations on the putative solvent-exposed surface of the N terminus of helix A or disruption of the packing of helix A appears to be deleterious to the function of FLT3L as indicated by the large number of mutations isolated between positions 8 and 15. Large regions of relative mutational quiescence, positions 22-75, 95-113, and 125-153, separate these hot spots. This last segment is unlikely to be involved in receptor binding, because progressive C-terminal deletions have indicated that residues C-terminal to cysteine 131 are dispensable to normal activity but are required for binding of the non-neutralizing monoclonal antibody M5␣FLT3L. 3 Examination of the data in Table I shows that the same residue is substituted by different amino acids. The addition of the single hydroxyl group from the tyrosine side chain in mutein F15Y is sufficient to reduce activity by 80%, while the other conservative substitution, F15L, completely abolishes detectable WWF7 activity. We postulate, based on our structural model of FLT3L, that this phenylalanine residue is tightly packed in the hydrophobic core of the protein. Similarly, substitutions observed for phenylalanine at positions 81 (F81L and F81S) and 124 (F124L and F124S), both conserved in murine FLT3L, suggest a packing phenomenon. Substitution of serine at position 13 by either proline or phenylalanine substantially reduces activity. The equivalent site in murine FLT3L is serine, suggesting that position 13 cannot tolerate a large side chain. We obtained 14 FLT3L ϩϩ independent isolates of the K84E substitution and one FLT3L ϩϩ K84T substitution (Table I).
In two cases, we isolated clones with a FLT3L ϩϩ phenotype with an amino acid substitution at position Ϫ3 (LϪ3H, ( Table  I). The FLT3L produced in this study contains an additional three residues, derived from the signal sequence, amino terminal to the threonine of FLT3L protein expressed in mammalian cells. Amino-terminal sequence analysis of wild type protein expressed in yeast confirms this conclusion (see "Experimental Procedures"). This result indicates that a substitution outside the mature protein is able to increase FLT3L activity.
We isolated L26F, L27P, and V34L in the primary membrane screen as species with a FLT3L ϩϩ phenotype. Later, when subjected to secondary screens, these clones became FLT3L Ϫ . The activities of L26F and V34L are 69 and 58% of wild type, respectively; thus, their designation as FLT3L ϩϩ from the primary membrane screen may be due to subjective error in discerning a enhanced reporter signal from wild type background. By contrast, we isolated L27P independently two times as a FLT3L ϩϩ species, yet its activity is between 7 and 20% of wild type (Table I and Fig. 4). The level of expression, since L27P is expressed at almost 3 times the level of wild type protein, probably accounts for its FLT3L ϩϩ signal we observed in the primary membrane screen. L27P is of interest, since it occurs 3 S. D. Lyman, unpublished results. a The first letter is the one-letter code for the wild type amino acid; the number is the amino acid position of mature FLT3L; the second letter is the one-letter code for the replacing amino acid. The table is divided into two sections, those species that were designated as deficient in receptor binding (FLT3L Ϫ ) and those species that exhibit greater than wild-type receptor binding (FLT3L ϩϩ ), according to the results from the initial filter binding assay.
b The level of FLT3L expression as determined by ELISA of yeast supernatants is presented in this column as g/ml.
c Direct assay of yeast supernatants yields the ratio of cell proliferation activity in WWF7 cells (units/ml) to concentration of FLT3L protein as determined by ELISA (ng/ml), and this value is referred to as specific activity of mutant (Sp. Act. MUT) or wild-type protein (Sp. Act. WT). The ratio of specific activity of mutant to wild-type FLT3L in a given assay is averaged over one or more independent assays and is presented in this column. A value of zero indicates that the proliferation activity of the mutant protein is below the limit of detection of our assay. d The number of independent assays, n, is indicated in this column. e Independent isolate, n, refers to the number of times a particular mutation was isolated.
f The specific activity of wild type protein averaged over 11 independent assays equals 0.95 Ϯ 0.34 units/ng. g Not applicable. h These mutants were subcloned from species that had multiple amino acid substitutions.
at the putative C terminus of helix A, the proposed dimerization interface in our model for FLT3L quaternary structure.
Three substitutions introduce cysteine residues, R20C, R55C, and R95C. The potential for disruption of the disulfide bond network due to these mutations causes us to view these cysteine substitutions with less interest. Substitutions that replace or introduce prolines, P10S, S13P, and P90S, were not pursued due to the probable large structural perturbations these replacements would introduce.
Purification and Analysis of FLT3L Muteins: Receptor Binding and Cell Proliferation Activity-We chose to purify all FLT3L muteins that exhibited greater than wild type specific activity because these are likely to be directly involved in receptor binding (H8Y, K84E, K84T, W118R, and Q122R). Proteins with specific activities lower than wild type were selected for further study due to their relative proximity to the mutations resulting in a FLT3L ϩϩ phenotype and their near wild type levels of expression (H8R, I11Y, F81S, K116E). We also selected L27P for purification for the reasons stated above.
The data in Fig. 4 clearly denote a strict correlation between relative biological activity and receptor binding affinity for this subset of purified FLT3L muteins. It is unknown if this strong correlation extends to other muteins listed in Table I. The K a and the K ai for the wild type protein used in this study are 1.5 Ϯ 0.7 ϫ 10 10 and 2.2 Ϯ 0.7 ϫ 10 10 M Ϫ1 , respectively.
The specific activity reported in Fig. 4 must be distinguished from that presented Table I, because the FLT3L proteins have been purified, their concentration has been determined by amino acid composition, and the BAF/hFLT3R cell proliferation assay uses cells transfected with human FLT3R. Note that there is a close correlation between the specific activities of the murine WWF7 assay using yeast medium and the human BAF/ hFLT3R assay using purified proteins (Table I and Fig. 4). The specific activities for the FLT3L ϩϩ species H8Y, K84E, K84T, W118R, and Q122R are in good agreement. However, when comparing FLT3L Ϫ species, we note that whereas the human BAF/hFLT3R proliferation assay exhibits some activity in all cases, the murine WWF7 assay does not. For example, H8R activity is 20% of wild type in the BAF/FLT3R assay, but it is not active in the WWF7 assay. This is also true for substitutions I11Y and K116E. It appears that the BAF/FLT3R assay is a more sensitive assay than the WWF7 assay. This effect is pronounced for mutein F81S. We isolated F81S as a FLT3L Ϫ species in the primary membrane screen using human FLT3R-Fc. The lack of activity in the murine WWF7 assay and near wild type activity in the BAF/FLT3R assay observed for F81S (see Table I and Fig. 4) may reflect a difference in the emphasis between murine and human receptors for the side chain of position 81 of the ligand.
We constructed, expressed, and purified two multiple FLT3L ϩϩ muteins, K84E/Q122R and LϪ3H/H8Y/K84E/ Q122R. The results form BAF/hFLT3R binding assays of purified LϪ3H/H8Y/K84E/Q122R mutein show that this molecule is 8-fold more active than wild type protein (Table II). Similarly, the findings from the BAF/hFLT3R cell proliferation assays indicate that biological activity of the quadruple mutant is also enhanced, albeit only 3-fold.
Physical Characterization of FLT3L Muteins-The results we present in Table III indicate an absence, with the exception of W118R, of gross structural perturbations for the subset of purified proteins. Monomeric unglycosylated FLT3L should have a molecular mass of 17,686 daltons. When analyzed by SDS-gel electrophoresis, yeast-produced FLT3L migrates at molecular mass of approximately 21 kDa due to the presence of core glycosylation at a single N-linked site (Fig. 4B). The Stokes radius as determined by size exclusion chromatography (M r ϭ 44,000) indicates that wild type and mutant proteins are dimeric. The helical content of wild type and mutant proteins as determined by circular dichroism are similar to that obtained for wild type protein. These results are expected, given that the near wild type levels of expression was an important criterion for choosing this subset of proteins. Previous studies have shown that misfolded proteins are degraded in the endoplasmic reticulum (22), suggesting that well expressed proteins secreted from yeast are in a stable fully folded form.
The helical content for amino acid substitution W118R is 7% less than that of wild type protein. A hydrophobic residue occupies the equivalent site for murine and human SCF (phenylalanine) and for murine and human M-CSF (leucine) as determined by amino acid sequence alignment (1). Utilizing the sequence alignment between FLT3L and M-CSF and the threedimensional structure of M-CSF, residue 118 of FLT3L is positioned 4 residues from the C terminus of helix D and is mostly buried in the hydrophobic core of the protein. It appears that the W118R substitution, a FLT3L ϩϩ species, disrupts the helical content of FLT3L, yet this mutation increases FLT3L activity.
A three-dimensional homology model FLT3L is presented in Fig. 5. This model is based on M-CSF and predicts a dimeric form of FLT3L. The L27P mutation maps to the proposed dimerization interface, and a mutation at the dimerization interface may destabilize FLT3L dimers, resulting in monomeric FLT3L. Analysis of L27P by size exclusion chromatography indicated that it was dimeric at 0.1 mg/ml (Table III). To test whether monomeric FLT3L species would be observed at lower concentrations, L27P or wild type FLT3L was diluted and analyzed by size exclusion chromatography. Fig. 6A shows that wild type FLT3L eluted at 8.1 min at 0.28 or 0.017 mg/ml. At 0.28 mg/ml, the elution time of the L27P protein was nearly   FIG. 3. Linear representation of FLT3L amino acid substitutions. Top, representation of FLT3L, including the intramolecular disulfide linkage, placement of ␤-sheet segments, and placement of ␣-helices A-D based on sequence alignment with M-CSF (1). Bottom, primary structure representation of FLT3L with a superimposition of the relative specific activity profile of muteins with single amino acid substitutions (see Table  I for details).
identical to the wild type dimeric FLT3L. However, as the concentration of the L27P protein is reduced to 0.017 mg/ml, the peak at 8.1 min is reduced in size, and a second peak is observed at 8.6 min (Fig. 6B). The change in the elution time of the L27P protein when diluted from 0.28 mg/ml to 0.017 mg/ml corresponds to a shift in the observed molecular weight from 44 to 28 kDa. In contrast, the wild type FLT3L protein remains at 44 kDa for both concentrations. These data indicate that the L27P mutation disrupts the FLT3L dimerization interface, resulting in monomeric FLT3L species at reduced protein concentrations. DISCUSSION We have generated a map of the FLT3L binding surface for its receptor using random mutagenesis and a functional screen. The data in Table III indicate that there is minimal gross structural perturbation, with the exception of W118R, in our collection of purified FLT3L muteins. These data suggest that most of the residues concerned may be directly involved in the energetics of receptor binding. Interpretation of the mutation pattern observed in this study (Fig. 3) is greatly facilitated by the three-dimensional homology model of FLT3L (Fig. 5).
The solved crystal structure of M-CSF was chosen as a template for modeling FLT3L, since several properties of M-CSF are most similar to FLT3L. M-CSF, and by analogy FLT3L and SCF, belong to the short chain helical bundle structural subfamily. Other members of this family share little identity, yet have remarkable structural similarity (9). FLT3L (1), SCF (23), and M-CSF (21) are dimeric, but unlike M-CSF, FLT3L, and SCF they are non-disulfide-linked dimers. The M-CSF dimerization interface was used to model the dimerization interface of FLT3L.
In addition to the sequence similarity between FLT3L and M-CSF, the FLT3L receptor-ligand system appears most sim-ilar to the M-CSF and SCF receptor-ligand systems. The receptors for FLT3L, SCF, and M-CSF are all type III tyrosine kinase receptors and share considerable sequence identity. The binding stoichiometry for polypeptide chains of SCF ligand with subunits of Kit receptor is 2:2 (24). Koths and colleagues (25) assert that M-CSF binds only one receptor chain per ligand polypeptide. These observations suggest that for SCF and M-CSF, and by analogy FLT3L, there is only one receptor binding site per ligand subunit and an overall 2:2 stoichiometry in the ligand-receptor complex.
The mutagenesis data and the three-dimensional model of FLT3L corroborate each other. The three hot spot regions (positions 8 -15, 81-87, and 116 -124), widely separated in the primary structure, cluster together in a small surface patch of the tertiary structure. The clustering of mutations in a small surface patch is consistent with a single receptor binding site per monomer as suggested by studies of similar receptor-ligand systems. In addition, the model indicates that some mutations map to the proposed dimer interface. Mutations that disrupt the dimerization interface would be predicted to shift the equilibrium from dimeric to monomeric FLT3L. The L27P mutation maps to this proposed dimerization interface and reduces the biological activity of the mutant protein. As predicted, the L27P mutant protein exhibits a concentration-dependent change in the apparent molecular weight consistent with an alteration in a monomer-dimer equilibrium (Fig. 6). In addition to L27P, we also identified another mutation that maps to the dimer interface, A64T. The A64T mutant protein was not purified or analyzed for its size exclusion properties. However, in the protein sequence alignment proposed by Hannum   a The specific activity of purified mutant (Sp. Act. MUT) or wild type protein (Sp. Act. WT) was determined by taking the ratio of proliferation activity in BAF/hFLT3R cells (units/ml) to concentration of FLT3L protein as determined by amino acid composition (ng/ml). The ratio of specific activity of mutant to wild type FLT3L is averaged over three independent assays, and the S.D. is reported.
b The ratio of K ai of mutant to wild type FLT3L, averaged over at two independent assays, and the S.D. is reported. a Stokes radius was determined by size exclusion chromatography monitored at an absorbance wavelength of 280 nm, with a 45-g loading at a concentration of 0.1 mg/ml. The value presented represents at least two measurements, and the S.D. was less than 5% of the reported value.
b Percentage of helix was determined by circular dichroism measurements (see "Results" for details).
c Not determined.

FIG. 4.
A, the relative values of FLT3 receptor binding and BAF/ hFLT3R cell proliferation activity for purified FLT3L proteins. The ordinate is the log 10 of the relative value, and the abscissa is the FLT3L protein designation. The values of K ai of mutant/K ai of wild type (filled bars) and specific activity of mutant/specific activity of wild type (open bars) are shown. The ratio of K ai of mutant to wild type FLT3L, determined in at least duplicate, in a given assay is averaged over at least three independent assays, and the S.E. is reported. The specific activity of purified mutant or wild type protein was determined as described for the WWF7 assay in Table I. The ratio of specific activity of mutant to wild type FLT3L, determined in duplicate, in a given assay is averaged over at least three independent assays, and the S.E. is reported. ND, not determined. B, Coomassie-stained SDS-polyacrylamide gel electrophoretic profiles of the corresponding purified FLT3L proteins are presented.
Phe 67 in SCF altered monomer-dimer equilibrium toward monomer and reduced the biological activity of SCF (23). The observation that the L27P mutation mapped to the dimerization interface and appeared to alter monomer-dimer equilibrium helped validate M-CSF as a template for modeling FLT3L structure.
As noted above, mutations that map to three hot spots scattered throughout the primary sequence cluster together in a residues are blue. Position 8 is colored blue, although a FLT3L Ϫ mutein also occurs at this site. FLT3L muteins listed in Table I whose activity is greater than wild type or reduced more than 75% of wild type are represented (except cysteine substitutions R20C and R55C). Position labels represent those proteins that were purified. B, space filling model of FLT3L. Orientation, coloration (except helices, which are uniform gold), representation, and numbering are as in A. FLT3L muteins listed in Table I whose activity is greater than wild type or reduced more than 75% of wild type are represented and labeled. patch on the FLT3L three-dimensional model (Fig. 5). His 8 maps to the center of this patch. We isolated two substitutions for histidine at position 8, H8R, which is FLT3L Ϫ , and H8Y, which is FLT3L ϩϩ . This histidine is conserved in murine FLT3L and M-CSF (Fig. 1). Koths and co-workers (25) report that the activity of M-CSF is reduced when the equivalent histidine, His 9 , is substituted by alanine. High resolution three-dimensional analysis of M-CSF mutant H9A/H15A shows no significant structural perturbations. These data suggest that His 9 of M-CSF and His 8 of FLT3L are directly involved in the binding energetics with their respective receptors. Lys 84 is another basic residue of particular interest. Removal of positively charged Lys 84 , the penultimate residue of the C terminus of helix C, in muteins K84E and K84T results in the FLT3L ϩϩ phenotype. The threonine substitution of K84T is conservative with the serines found in the equivalent site in M-CSF and murine FLT3L. This suggests that Lys 84 acts to diminish activity in native FLT3L by destabilizing the interaction with FLT3 receptor.
Inspection of Fig. 5B reveals that K84E, Q122R, and H8Y form a triangle of residues that enhance binding of FLT3L to its receptor. Residues that map within this triangle include Asp 3 , Cys 4 , Ser 5 , and Gln 7 . We identified the mutation D3G by subcloning a double mutant, but this mutation had little influence on FLT3L biological activity (Table I). A previous study showed that the C4S mutation reduced biological activity 10-fold (26). Other residues that map within the triangle include Ser 5 and Gln 7 , which are Tyr 5 and Ser 7 in murine FLT3L, respectively (Fig. 1). Since murine FLT3L stimulates the human receptor, these residues are not critical for receptor binding. These data suggest that residues that can be changed to increase receptor affinity may include residues that may not be part of a continuous patch; another explanation is that this section of the homology model is inaccurate. Consistent with the former interpretation is the observation that W118R and LϪ3H increase binding affinity for the FLT3 receptor and introduce an extra positive charge to the molecule. The W118R mutation disrupts some of the helical content of the molecule (Table III), and Leu Ϫ3 is a residue that is not normally found in the mammalian expressed molecule, so it is unlikely that either residue is part of receptor binding site. These results are consistent with a less direct effect of charge rather than alterations in an area of the molecule critical for receptor binding.
Double mutein K84E/Q122R maintains the close correlation between receptor affinity and bioactivity that we observe for the individual FLT3L ϩϩ mutations (Table II). This correlation breaks down for quadruple FLT3L ϩϩ mutein, LϪ3H/H8Y/ K84E/Q122R. Whereas receptor affinity for the quadruple mutant is over 8-fold greater than wild type, cell proliferation activity is only 3-fold greater (Table II). Two recent reports also indicate that the correlation between receptor affinity and bioactivity diminishes as high affinity mutations are combined. Multiply substituted muteins of interleukin-6 (27) and ciliary neurotrophic factor (28) exhibit high affinity for their respective receptors but do not evince the same high levels of bioactivity. Two explanations potentially account for the apparent limit to bioactivity. First, Toniatti et al. (27) propose that receptor clustering on the surface of cytokine-activated cells acts to increase local receptor concentration to levels above the K d of the IL-6 muteins, thus limiting maximum bioactivity of the high affinity mutants. Second, Reddy et al. (29) demonstrate that an epidermal growth factor mutein with reduced receptor affinity exhibits greater bioactivity than wild type ligand by diminishing two attenuation mechanisms, endocytic internalization and degradation of epidermal growth factor and receptor down-regulation. Therefore, the maximum bioactivity of high affinity FLT3L muteins is potentially reduced by a rapid rate of ligand-mediated receptor endocytosis.
In summary, we have developed a novel method for screening a randomly mutagenized ligand in order to identify residues involved in receptor binding. This method should be applicable to ligand-receptor systems where appropriate receptor binding and biological assays are available. In addition, the ability to simply identify species with "improved" receptor affinity and bioactivity points toward a significant utility for this technology. The strength of this method is further enhanced when mutations can be correlated to a homology-modeled structure. The corroboration among primary, tertiary, and quaternary structural information provides a powerful indication that we have correctly mapped the functional epitopes of FLT3L.