Single Proline Substitutions in Predicted a-Helices of Murine Granulocyte-Macrophage Colony-stimulating Factor Result in a Loss in Bioactivity and Altered Glycosylation*

Contributions of a-helices to biological activity in murine granulocyte-macrophage colony-stimulating factor were analyzed using site-directed mutagenesis and protein expression in COS- 1 cells. A series of single proline substitutions were made for residues within the four predicted a-helices as a means of disrupting local helical secondary structure. Mutations in three of the four helices resulted in marked reductions in bioactivity. Five mutants E21P, L56P, E60P, L63P, and L107P showed 102-104-fold reduction in bioactivity as well as hyperglycosylation. The same Pro substitu- tions made on non-N-glycosylated molecules had a similar loss in bioactivity implying that a Pro-induced structural change and not hyperglycosylation was re-sponsible for the major decrease in bioactivity. Addi- tional amino acid substitutions at these residues which conserved charge or hydrophobicity, or replaced the original residue with an Ala, verified that conformational changes in the protein structure were specifi- cally due to steric constraints imposed by the Pro residue rather than loss of important side chain functions. (GM-CSF)’

Contributions of a-helices to biological activity in murine granulocyte-macrophage colony-stimulating factor were analyzed using site-directed mutagenesis and protein expression in COS-1 cells. A series of single proline substitutions were made for residues within the four predicted a-helices as a means of disrupting local helical secondary structure. Mutations in three of the four helices resulted in marked reductions in bioactivity. Five mutants E21P, L56P, E60P, L63P, and L107P showed 102-104-fold reduction in bioactivity as well as hyperglycosylation. The same Pro substitutions made on non-N-glycosylated molecules had a similar loss in bioactivity implying that a Pro-induced structural change and not hyperglycosylation was responsible for the major decrease in bioactivity. Additional amino acid substitutions at these residues which conserved charge or hydrophobicity, or replaced the original residue with an Ala, verified that conformational changes in the protein structure were specifically due to steric constraints imposed by the Pro residue rather than loss of important side chain functions.
Granulocyte-macrophage colony-stimulating factor (GM-CSF)' is a member of a family of glycoproteins essential for regulating growth and differentiation of hematopoietic progenitor cells (1)(2)(3) as well as stimulating functional activation of mature cell populations (4, 5 ) . Besides having similar and overlapping biological activity the colony-stimulating factors show synergistic effects when present together (6,7). Taken with the fact that multiple colony-stimulating factors are often produced by a single source (8,9) and that each protein is active at picomolar concentrations, studies using purified native molecules were difficult and often misleading. With the advent of cDNA expression systems, preparations of pure recombinant factors became available for extensive characterization of activity in vitro and in vivo and have allowed structural analysis to ensue.
The protein sequences of GM-CSF from human (lo), gibbon ( l l ) , murine (12,13), and bovine (14) sources have been deduced from cloned cDNA sequences (Fig. 1). Despite considerable sequence homology among the proteins, including location of cysteine residues, some species specificity of biological activity is still maintained. Human and murine proteins have the least homology (54% identical amino acids) and fail to cross-react (15). Detailed biophysical studies using purified proteins demonstrated a high similarity between the two molecules. The proteins exhibited similar peptide backbone conformations, physical characteristics, and conformational stability (16).
In the absence of structural information Parry et al. (17) using a series of predictive algorithms proposed the four ahelical bundle motif as a structural model for a family of related cytokines including hGM-CSF. A number of structure-function studies by other investigators fit the model suggested by Parry. Synthetic peptide analogues synthesized by Clark-Lewis et al. (18) identified minimal sequence necessary for detectable activity. They showed that neither the amino-terminal 15 residues nor the six carboxyl-terminal amino acids were critical for activity. Removal of additional amino acids from either end which include residues predicted to be in helix 1 or helix 4, significantly lowered activity. Shanafelt and Kastelein (19) used scanning deletion analysis to map four regions critical to activity in the murine protein. These regions aligned with the same residues proposed by Parry et al. as helical (Fig. 1). More recently, Kaushansky et al. (20) identified two distinct regions of hGM-CSF, residues Glu-21 to Asp-31 and Thr-78 to Thr-94, that are important for the GM-CSF bioactivity by assaying various hybrid molecules for species-specific activity. Additional receptor competition assays using these chimeric molecules suggest that the two regions comprise the active receptor-binding domain (20). These results were then extended further. Brown et al. (21) used the same chimeras to map the binding epitopes of two monoclonal antibodies that neutralize hGM-CSF activity. Their results confirmed that the two regions identified by Kaushansky et al. (20) were near each other in the folded conformation. They also identified specific residues critical to antibody binding which may also be involved with receptor interactions (21).
In this report we examined the contribution that each of the four predicted helices has on mGM-CSF activity. By making a series of Pro substitutions for residues located in the proposed helices our aim was to cause local disruptions of an individual helix. Mutant proteins were transiently ex-5333 pressed in COS-1 cells. Bioactivity of each mutant protein was assayed and compared to that of native mGM-CSF.

MATERIALS AND METHODS
Bacterial Strains and Vectors-The Escherichia coli strain XL1-Blue (Stratagene, La Jolla, CA) was used for the propagation and maintenance of all plasmid DNA. The VCS-M13 helper phage was used for helper phage rescue of single strand DNA used as template DNA in mutagenesis reactions.
A partial length cDNA containing the coding region for mGM-CSF was subcloned into the H i n d site of pUC18. A HindIII-XbaI fragment from the pUC18 subclone was ligated into pRJBlOB (22) which had been digested with HindIII-XbaI yielding pRJB-GM. pRJB-GM is a eukaryotic expression vector utilizing the Rous sarcoma virus promoter and bovine growth hormone polyadenylation signal. When coupled with a SV40 origin placed upstream of the Rous promoter, this expression cassette is highly efficient in COS-1 cells. In addition, pRJB-GM has an M13 phage F1 origin that permits rapid production of single-stranded phagemid DNA (23) which can be used for single strand sequencing and site-directed mutagenesis.
Mutagenesis-Site-directed mutagenesis was performed using the protocol provided with the oligonucleotide-directed in vitro mutagenesis system V.2 (Amersham RPN.1523, Arlington Heights, IL) which is based on the method of Eckstein (24)(25)(26). Oligonucleotides, 22 nucleotides long, corresponding to mGM-CSF sequences incorporating the desired amino acid substitutions were made complementary to the template DNA and used as primers in the mutagenesis reactions. Bacteria were transformed with DNA from the final step of the mutagenesis reactions and cultured on agar plates. Plasmid DNA was isolated from three to five colonies, digested with diagnostic restricted enzymes to check vector integrity, and then sequenced using Sequenase (United States Biochemical, Cleveland, OH) protocol (27,28). At least 200 nucleotides of GM-CSF-coding sequence was sequenced for all mutants and entire coding sequences were often determined. Confirmed mutants were prepared and plasmic DNA was twice banded on CsCl gradients for transfection.
Transfections and Metabolic Labeling-COS-1 cells were transfected by the DEAE-dextran method (29) with modifications (30-33). Twenty h before transfection, cells were plated on 60-mm culture dishes (Corning, Corning, NY) at 4 X lo5 cells/dish in 4 ml of complete Dulbecco's minimal Eagle's medium (DMEM) high glucose (GIBCO) supplemented with 10% fetal calf serum (FCS) (Hyclone, Logan, UT). After 20 h, the medium was aspirated and the cells washed once with 5 ml of Dulbecco's phosphate-buffered saline. One ml of complete DMEM supplemented with 10% Nu-serum (Collaborative Research, Lexington, MA) was added to the plate. Ten pg of plasmid DNA in 0.2 ml of Tris-buffered saline (34), pH 7.4, was mixed with 0.8 ml of serum-free DMEM containing 1 mg/ml DEAE-dextran and 250 pM chloroquine, the mixture was added to the dish and incubated at 37°C for 4 h. The transfection solution was aspirated and replaced by 2 ml of 10% dimethyl sulfoxide (Sigma) in Dulbecco's phosphate-buffered saline and incubated for 5 min. After removing the dimethyl sulfoxide, cells were washed once with 5 ml of phosphate-buffered saline and incubated in 5 ml of complete DMEM with 10% FCS for 24 h. The medium was then replaced with 2.5 ml of harvest media (methioninefree DMEM, 10% FCS, supplemented with methionine (Irvine Scientific, Santa Ana, CA) to a final concentration of 0.3 pg/ml with or without 100 pCi/ml [35S]Met (ICN Biomedicals Trad5S-Label, 51006, Irvine, CA)) and incubated for 30 h. Supernatants were harvested and tested for the presence of GM-CSF by bioassay and, if the cells had been metabolically labeled with [%]Met, GM-CSF was immunoprecipitated from the supernatant and analyzed by polyacrylamide gel electrophoresis (SDS-PAGE).
Proliferation Assay-The mGM-CSF-responsive murine leukemia cell line DA-3 (35) was used to determine the biological activity of the mutant proteins. DA-3 cells were maintained in RPMI-1640 medium (GIBCO) supplemented with 10% FCS and 2% conditioned medium from a cloned Chinese hamster ovary cell line producing mGM-CSF. For proliferation assays, the cells were washed twice in Dulbecco's phosphate-buffered saline adjusted to lo5 cells/ml in complete DMEM with 10% FCS. Culture supernatants from the transfected COS-1 cells were harvested and titered in triplicate using serial 3-fold dilutions in 96-well microtiter plates (Costar, Cambridge, MA) such that the volume in each well was 100 pl; 100 p1 of the DA-3 suspension culture was added yielding a final volume of 200 pl. Proliferation was assessed by measuring the incorporation of [methyl-"Hlthymidine (Amersham TRA.310). A 12-h pulse of 2 pCi/well was given at 48 h of culture. Cells were harvested on glass fiber filters (Whatman, Clifton, NJ) with a multi-well cell harvester (Cambridge Technology, Cambridge, MA) washed, and air dried. [3H]Thymidine incorporation was measured using an LKB 1209 Rack-beta (Pharmacia LKB Biotechnology Inc.) scintillation counter. Values reported as counts/minute are the mean of the triplicate for each dilution. Assays were performed on at least two separate transfection experiments.
native or mutant GM-CSF metabolically labeled with [35S]Met were Immunoprecipitation-COS transfection supernatants containing immunoprecipitated using either polyclonal rabbit antisera or a purified rat monoclonal antibody mAb A2' as previously described (36, 37) with modifications. COS supernatants (0.5 ml) were clarified by centrifugation for 10 min then transferred to a fresh Eppendorf tube. Samples were incubated for two 1-h cycles with 100 pl of heatinactivated formalin fixed Staph A protein (Igsorb, Enzyme Center, Malden, MA) suspended in wash buffer (50 mM Tris, pH 7.8,150 mM NaCI, 5 mM EDTA, 0.1% NaN', 0.5% Nonidet P-40,0.5% deoxycholic acid, 10% glycerol) to adsorb nonspecific proteins. Purified rat monoclonal antibody (mAb A2) (3.5 pg) was added and mixed 10 h followed by a second 10-h incubation with 10 pg of rabbit anti-rat IgG (Vector Laboratories, Burlington, CA). Alternatively, 5 pl of concentrated rabbit serum containing polyclonal antibody made against mGM-CSF was added to the precleared COS-1 supernatant and mixed for 10 h. Samples were incubated with 200 pl of Igsorb for 3 h and then pelleted. The Igsorb pellet was washed three times with wash buffer, suspended in 100 pl of denaturing buffer, and boiled for 3 min. After centrifugation, the proteins were separated by SDS-PAGE (15%) (38) and either exposed to XAR-5 film (Kodak, Rochester, NY) or a phosphorimager screen (Molecular Dynamics, Sunnyvale, CA).
Determination of Relatiue Specific Actiuity-The concentration (%) of native mGM-CSF COS supernatant that gave 50% maximal stimulation in the DA-3 proliferation assay was determined from the 16point serial 3-fold dilution curve. Values for a and b were calculated for the equation Y = a x b using a least squares algorithm to fit the curve to the data from the GM-CSF COS supernatant titration. Y,i, was assigned from the background count as determined by the average value (cpm) for all dilutions of the mock transfection titration curve.
YmaX was the average count for all the dilutions of the GM-CSF COS transfection outside the linear region that define the upper limits of the titration curve. The 50% stimulation index (50% S.I.) equals % (V,,, -YmiJ and the bioactivity equals 100 times the reciprocal of the (%) COS supernatant at 50% S.I. The least squares algorithm was then used to fit the curve to the mutant titration values and utilizing the 50% S.I. from the native GM-CSF curve the mutant protein bioactivity was calculated.
The amount of immunoreactive protein in the COS transfection supernatants are quantified by integrating the area x density of the signal from the autoradiogram using a scanning densitometer (Molecular Dynamics, Sunnyvale, CA) or scanning an exposed screen using a phosphorimager (Molecular Dynamics). Protein concentration (volume) was reported after subtracting the background protein concentration of the mock transfection.
Specific activity equals bioactivity/protein concentration. Relative specific activity is defined as specific activity mutant/specific activity GM-CSF and allows for comparison of mutant bioactivity from several COS transfection experiments.
It should be noted that the sensitivity of this quantification method of GM-CSF may not be as reliable as other immunological techniques (19,39) and permits merely an estimate of specific activity. Since mAb A2 only recognizes the folded form of GM-CSF Western immunoblotting was not possible.' Our method of immunoprecipitation as the means of quantitating protein concentration has some endogenous inaccuracies, but complete control experiments were done to insure that all of the labeled immunoreactive protein was measured (data not shown). Differences of 3-4-fold were occasionally seen, consequently we considered differences in relative specific activity of 10-fold or greater to be significant.  tional consideration was also given to the relative a-helical propensity of the residue as determined by a number of predictions based on observed frequencies in a-helices (40,41). Residues with high helical propensity were chosen for substitution so the steric effects of the Pro residue would be most dramatic. Residues selected for Pro replacement are identified by circles located in the murine sequence of Fig. 1.

Introduction of Proline Substitutions and Expression of Mu-
The eukaryotic expression vector pMB-GM, which uses the Rous sarcoma virus promoter, was designed to generate mutants and to evaluate expression in COS-1 cells rapidly precluding the need to shuttle mutant constructs between vectors (Fig. 2). pMB-GM contained a modified mGM-CSF cDNA inserted betweon the Rouse sarcoma virus promoter and the bovine growth hormone polyadenylation signal. The mGM-CSF cDNA fragment (37) which includes only 17 amino acids of the native secretory signal sequence starting from the second in-frame Met has been shown to be secreted efficiently (42). GM-CSF or the desired mutation was expressed transiently in COS-1 cells, and the supernatants containing the secreted nascent proteins were collected and analyzed as described under "Materials and Methods." Proline Substitutions Show a Differential Effect on mGM-CSF Bioactiuity-To examine the effects that various Pro substitutions have on protein function, proliferation assays using the mGM-CSF-responsive cell line DA-3 were compared (Fig. 3). Four separate graphs are shown corresponding to the four regions predicted to be a-helices. Each graph includes titrations of native mGM-CSF and mock transfections as references. Table I includes calculations of relative specific activity for the first series of Pro substitutions. The two mutations located in region A had distinct differences in bioactivity. Mutant E17P showed similar activity to that of native GM-CSF while the activity of mutant E21P was about 850-fold less active. Region B contained the three mutations E35P, E38P, and E43P, all of which had activity levels resembling that of the native molecule. The three mutations, L56P, E60P, and L63P, located in region C exhibited various degrees of reduced bioactivity ranging from 25-to 450-fold less activity than that of the native protein. Of the three mutants, L63P had the most profound effect on activity (450-fold) followed by E60P (50-fold) and L56P (25-fold). The two mutants in region D resembled those in region A in that one functioned like the native protein and the other had the lowest bioactivity of all the mutations. The titration curve of mutant AlOlP paralleled that of the native molecule while mutant L107P (5,500-fold) was unable to reach maximal stimulation even at the highest concentration tested.
Assessing the contribution of a single helix to bioactivity is most evident for regions B and C. In both cases all three mutants in each helix had similar effects on bioactivity and glycosylation. Additional studies using combinations of 2 Pro replacements supported the original findings. Double Pro mutations in region B showed only a minimal reduction (-10fold) in bioactivity and no changes in glycosylation (data not shown). In addition, it appears that Pro substitution mutagenesis can be used either to define the boundaries of an ahelix or to determine the critical residues within a single helix. For example, two of the least active mutants E21P and L107P were located in regions A and D indicating the importance of these regions. While mutants E17P and AlOlP, located in the same regions, had no measurable effect on bioactivity suggesting that these residues are not critical for helix formation.
Proline Substitutions Cause Variations in Glycosylntion Patterns-As shown in Fig. 4, no immunoreactive protein was found in the mock transfection, while native GM-CSF had three major bands ranging in molecular mass from 14. glycosylation. It has also been suggested that 0-linked glycosylation occurs, and this may be the reason for the difference between the molecular mass of 14.3 kDa predicted from gene structures and the larger apparent molecular mass encountered from recombinant protein made in mammalian expression systems (45). Proline substitutions that decreased activity of the protein (E21P, L56P, E60P, L63P, L107P) also showed hyperglycosylation when compared to the native molecule ( Fig. 4 lanes 4, 8-10 and 12). It is possible that these Pro substitutions cause a conformational change which results in both decreased bioactivity and increased glycosylation. Alternatively, the mutations may cause increased glycosylation which in turn results in reduced bioactivity.
Hyperglycosylution Reflects a Change in Protein Conformation and Is Not the Major Cause of Decreased Bioactivity-A series of mutants was made to determine if loss in bioactivity was caused by a change in protein structure or steric hindrance due to additional carbohydrate groups (Fig. 5). The generation of these mutants involved three sequential amino acid substitutions. The first two replacements eliminated both N-glycosylation sites by altering the tripeptide consensus sequence. The existing Asn residues were changed into the amino acid found in the other known GM-CSF molecules ( Fig. 1). In the case of Asn-66 a Ser was inserted and for Asn-75 a Thr was used. The resulting double mutant referred to as 2N was confirmed as a single species of GM-CSF in COS transfection supernatants (Fig. 6, lune 3). The third substitution was either E21P, L56P, E60P, L63P or L107P made on the 2N mutant template. The activity of the mutants are shown in Fig. 5 and compared in Table 11. As anticipated from previous studies (43-45), the activity of mutant 2N was at least as active as the native glycosylated form. Similarly the Pro substitutions made on the non-N-glycosylated form of the protein were more active than the glycosylated Pro substitution mutants. The non-N-glycosylated Pro substitutions, however, had the same relative effect on activity as they did in the glycosylated molecule, indicating that increased glycosylation is not the major cause for decreased bioactivity. Proline Substitutions Are the Major Cause for Decreased Bioactivity-Proline substitution at 5 residues measurably decreased the activity of the protein. Our concern was whether the differences in activity caused by the Pro substitutions were due to constraints imposed by the Pro residue itself or the inability of the protein to tolerate any amino acid changes at these sites. To test this, two additional amino acid substitutions were made for those 5 residues exhibiting loss of from the immunoprecipitation using polyclonal antisera. precipitation. .

FIG. 4. SDS-PAGE analysis of metabolically radiolabeled COS transfected supernatants. COS cells were transfected with either plasmid RJB-GM or various mutants and labeled with [:"S]
Met. Equal volumes of transfection supernatants were immunoprecipitated with rat anti mGM-CSF monoclonal antibody mAb A2 as described under "Materials and Methods." Precipitated samples were denatured in denaturing buffer, separated on a 15% SDS-PAGE gel and exposed to x-ray film. bioactivity and hyperglycosylation. For each residue a functionally similar amino acid, either in charge or hydrophobicity, was substituted. Native Glu residues were mutated to Asp residues and Leu residues were changed to Val residues. The final substitution for the five less active mutants was Ala as a means to determine if the removal of an important side chain function would yield the observed changes in bioactivity.
DA-3 proliferation assays of all three substitutions made for a single residue are compared to the native molecule in each of the 5 graphs in Fig. 7. In all cases the functional analogue had no effect on bioactivity demonstrating that similar side chains conserve function. Alanine substitutions alsu had IIU Imasurable effecr. UII b l~a c~l v l~y excep~ posslWy mutant L63A which showed a consistent 5-10-fold decrease. With the present methods, differences of less than 10-fold are not considered significant; therefore L63A is considered to be similar to native bioactivity although more rigorous exami-  nation may verify some difference. The conclusion based on this series of mutations confirmed that diminished bioactivity of mutants E21P, L56P, E60P, L63P, and L107P is due to conformational changes induced by the Pro substitution.

GM-CSF BIOACTIVITY
Other Amino Acid Substitutions Can Alter Glycosylation and Not Change Significantly the Bioactivity of GM-CSF-In all cases the functional analogue substitution had no effects on N-linked glycosylation (Fig. 8). In contrast two Ala substitution mutants caused hyperglycosylation. Unexpectedly, both L83A anu Llu-IA were nyperglycosylateu but St111 malntalneu full or near full bioactivity. The fact that two mutants exhibited hyperglycosylation but retained levels of bioactivity like the native protein was additional proof that hyperglycosylation was not the major cause for the decreased bioactivity of the mutants which have both hyperglycosylation and reduced bioactivity.

DISCUSSION
Protein crystallographic studies have revealed common structural features among proteins that neither resemble each other in sequence or function. One such recurring structural topology some proteins with high a-helical content have is the arrangement of helices to form a four a-helical bundle (46). In a study of 30 cytokines, Parry et al. (17) applying a series of protein prediction algorithms proposed that interleukins and colony-stimulating factors belong to a family of proteins that fit the four a-helical bundle motif. Specifically, a structural model was discussed for hGM-CSF that identified the four helices that comprise 54% of the molecule and the possible relative orientation of the helices based on disulfide bonds and connecting loop constraints. Of particular interest was the orientation of helices C and D due to the disulfide bond formed between Cys-51 and Cys-93. The GM-CSF molecule was the only cytokine predicted to have a parallel arrangement of two neighboring helices on the linear protein:' In good agreement, Wingfield et al.

14.3.
FIG. 6. SDS-PAGE analysis of metabolically radiolabeled COS transfection supernatants. COS cells were transfected with either plasmid RJB-GM or various mutants and labeled with ["SI Met. Equal volumes of transfection supernatants were immunoprecipitated using the rat monoclonal antibody mAb A2 as described under "Materials and Methods." Precipitated samples were boiled in denaturing buffer, separated on a 15% SDS-PAGE gel, and quantitated using a phosphorimager.
' D. A. D. Parry, unpublished observations. circular dichroic spectra. From the same experiment comparable CD spectra support the evidence from sequence alignment data that the two proteins have similar backbone conformation. Shanafelt and Kastelein (19) examined structurefunction relationships by generating a large panel of three amino acid deletion mutants that extended along the entire length of the mGM-CSF. Their results identified four regions of the protein defined as critical to GM-CSF bioactivity. Our analysis of the Parry et al. predictions and the scanning deletion studies (see Fig. 1) indicated that the critical regions aligned with homologous sequences in the human counterpart, predicted to be a-helical, suggesting that helices play an important role in maintaining protein integrity and bioactivity. The fact that deletions were three amino acids, which corresponds to nearly a full turn of an a-helix, implied that merely shortening any one of the four helices had significant consequences on protein bioactivity. Our results dispute those of Shanafelt and Kastelein regarding helix 2. Our analysis suggests that helix 2 has a noncritical role in maintaining mGM-CSF bioactivity. This contradiction may simply reflect the differences between the two methods of probing mGM-CSF structure. The use of scanning deletion mutagenesis is likely to result in greater conformational changes in the molecule which are necessary to compensate for the gap left by the deletions.
The aim of our work was to investigate more closely the role of a-helices and compare the differential contributions that helices play in preserving mGM-CSF bioactivity. The approach used single amino acid substitutions as a means of minimizing gross structural perturbations caused by inserting or removing peptide fragments. Our strategy involved substituting Pro for residues localized to a particular predicted helix with the intention of disrupting that helix. Proline was selected as the replacement residue because it is regarded as sterically incompatible with helical secondary structure. Constraints due to the unusual Pro side chain characteristics limit peptide bond rotation and prevent H-bonding, both of which are important to a-helical conformations. Clearly, a single Pro insertion may not be capable of overcoming the cooperative interactions of the neighboring residues or stabilizing effects of adjacent helices. This has been seen in the interleukin 2 molecule where helix B accommodated a Pro by introducing a kink (47). Therefore, it is important to remember that all structural changes due to the Pro residue are theoretical. A true understanding of the alterations imposed by the  Specific activity (SA) equals the bioactivity/protein concentration. Relative specific activity (RSA) equals SA mutant/SA mCM-CSF and was calculated from the data presented RSA was calculated from a separate COS transfection experiment using mAb A2 for the immunoprecipitation.
' RSA was calculated from a separate COS transfection experiment using mAb A2 for the immunoprecipitation.
in Figs. 5 and 6 (using mAb A2). Relative specific activity (RSA) equals SA mutant/SA GM-CSF and was calculated from the data presented e RSA was calculated from the same COS transfection as in RSAd except protein concentration was determined ' RSA was calculated from a separate COS transfection experiment using mAb A2 for the immunoprecipitation.
(area X density) from the autoradiogram and subtracting the volume of the mock lane.  substitutions can only be addressed by crystallographic data (48).
Since alterations in primary structure can lead to secondary structural changes which may affect bioactivity (e.g. glycosylation), it is important to determine the contribution of each structural change to bioactivity. Earlier experiments using hGM-CSF with mutated N-linked glycosylation sites (44,45) or proteins which had the carbohydrates removed enzymatically (43) measured increased activity with decreasing Nlinked glycosylation. Elegant biochemical studies using purified preparations of hGM-CSF from activated T-lymphocytes recently demonstrated that increased N-linked glycosylation decreased the specific activity of native protein and confirmed that previous observations were not artifacts of mammalian expression systems. Cebon et al. (49) was also able to attribute the lower activity to reduced receptor affinity. One of the most interesting findings from our experiments is the strong correlation between mutations that decreased bioactivity and changed N-linked glycosylation processing. Previous studies in this laboratory demonstrated the same phenomenon with a carboxyl-terminal deletion mutant (37). LaBranche et al. generated a deletion mutant that had over 10'-fold reduced bioactivity and showed that the hyperglycosylation was Nlinked. The major species of mGM-CSF has only 1 Asn residue glycosylated (Fig. 4, lane 2). Since the deletion mutant removed Cys-118 and therefore eliminated the second disulfide bond a more open protein conformation may have existed that was able to accommodate additional carbohydrate groups. Two mutants L63A and L107A support this idea. Both mutations removed hydrophobic side chains that, in the Parry model, are important for helix packing. The loss of hydrophobic interactions in the interior of the protein might result in a more relaxed conformation and allow greater exposure of both Asn residues permitting occupancy of both sites. Similar effects were seen with some Pro substitutions that disrupted critical helices and possibly helix packing. While hyperglycosylation does result in a minor (6-fold) decrease in bioactivity (49), other structural changes are necessary for a major loss in bioactivity. Assays of Pro substitutions made on non N-glycosylated molecules (Table 11) showed that hyperglycosylation was not the major cause of bioactivity differences a d iuclicakxl L~I~( L hyperglycusylatlun represents a cnange In protein conformation. Thus, it is likely that specific Pro substitutions in mGM-CSF result in conformational changes which are reflected by both decreased bioactivity and hyperglycosylation. The present results with mGM-CSF and prior investigations of hGM-CSF suggest a close proximity of N-linked carbohydrate attachment sites to the active site. This is an important consideration because in the two species the carbohydrates are located in different regions of the molecule. The N-linked carbohydrates are attached a t Asn-66 and Asn-75 in mGM-CSF and Asn-27 and Asn-37 in the human molecule (Fig. 1). Protein modeling by Kaushansky et al. (50), based on the predictions of Parry et al. and bond energy minimization analysis, placed these two regions next to each other in the folded form. Independent confirmation of this prediction comes from two different studies. Initially, Kaushansky et al. (20) using human-mouse chimeric molecules demonstrated the importance of the regions Glu-21 to Asp131 and Thr-78 to Thr-94 to the bioactivity of hGM-CSF. Both of these regions overlap with a t least one of the N-linked glycosylation sites in either the human Asn-27 or the murine Asn-75 protein (Fig. 1). Using the same series of chimeras Brown et ~l . (21) mapped the binding epitopes of two mAbs that neutralize hGM-CSF activity. They showed that mAb 221 recognized residue Arg-24 and region His-83 to Thr-94 and that mAb 213 bound to the region Leu-77 to His-83. Together these experiments established the proximity of these two regions to each other and present a strong case for their involvement with receptor binding.
Methods for predicting secondary structure are constantly being refined. Improvements in accuracy can be attributed to both the increasing numbers of protein structures solved by x-ray crystallographers and the use of high powered computers to analyze large amounts of available structural information. These predictions can provide testable models for studies designed to identify secondary structural elements and examine their molecular interactions or contribution to function. Our study shows that proline insertion mutagenesis is a sensitive means to evaluate helix contribution to protein function. Beyond comparing different helices, our data suggest that this method may be gentle enough to probe subtle differences that can occur along the length of a single helix. This approach should be suited both for molecules whose helices have been identified through crystallographic data and for molecules where helices are predicted.