Cumulative

HEMATOPOIESIS1,2,3,4 SCF is a glycosylated, small-protein growth factor found in both membranebound and soluble forms. The membrane-bound form is expressed on the surface of bone marrow stromal cells and acts synergistically with other circulating hematopoietic growth factors (M-CSF, GM-CSF, G-CSF, IL-3, erythropoietin, etc.) to stimulate the proliferation of hematopoietic stem cells (HSCs) which lead to a wide variety of cell lineages including the erythrocytic, granulocytic and megakaryocytic lineages. Thus, SCF is a crucial component of the hematopoietic microenvironment. On the stromal cell surface, SCF also functions as a chemoattractant and adhesion structure for early hematopoietic stem cells. The function of soluble SCF is unknown.

MAST CELL BIOLOGY5,6,7 & OTHER FUNCTIONS8,9 SCF also plays an important role in mast cell biology and thus also goes by the name of mast-cell growth factor (MGF). Similar to its hematopoietic functions, SCF acts to stimulate the proliferation and maturation of mast cells in the peripheral tissues. In addition, SCF has been shown to stimulate mast cell degranulation to release mediators such as serotonin and histamine. SCF also plays a role in early germ cell development in which it both enhances the proliferation and maintains the survival of primordial germ cells (PGCs). SCF is also involved in melanocyte biology (possibly as a chemoattractant) and in this way influences skin color.

DISCOVERY of SCF1,2 SCF was discovered through a classic combination of genetics and biochemistry. It had been known for some time that two distinct genetic loci (SI and W) were responsible for a similar set of mutant phenotypes in mice. Both of these mutant phenotypes involved a typical triad of hematopoietic defects, coat color alterations and defective gonadal development. Since the W gene encodes the c-kit tyrosine kinase receptor, the search was on for the ligand for this receptor. A protein that bound to the c-kit product was isolated and purified which was also shown to map to the SI locus. Thus the fact that SI encoded SCF and W encodes the c-kit product provided a cogent biochemical explanation for the genetics.

CLINICAL ASPECTS10,11 Pathologically, SCF acts as an autocrine factor in some leukemias and some small-cell lung cancers. Some anemias arising from bone-marrow failure may also involve defects in the SCF/c-kit system. Therapeutically, SCF antagonists may be useful agents in combating cancers that involve SCF as an autocrine factor. SCF also has potential use in ameliorating various acquired anemias (such as those resulting from chemotherapy) as well as facilitating stem-cell mediated gene transfer & gene therapy.

REFERENCES 1 2 3 4 5 6 7

Martin FH, et al., (1990), Cell, 63:203. Zsebo KB, et al., (1990), Cell, 63:195. Migliaccio G, et al., (1991), J Cell Physiol., 148:503. Andrews RG, et al., (1991), Blood, 78:1975. Anderson DM, et al., (1990), Cell, 63:235. Meininger CJ, et al., (1992), Blood, 79:958. Tsuji K, Zsebo M, Ogawa M, (1991), J Cell Physio., 148:362.

8 9 10 11

Dolci S, et al., (1991), Nature, 352:809. Matsui Y, et al., (1991), Nature, 353:750. Cassel A, et al., (1993), Exp Hematol., 21:585. Rygaard K, et al., (1993), Br J Cancer, 67:37.

HEMATOPOIESIS

MAST CELL BIOLOGY

DISCOVERY of SCF

SEQUENCE ANALYSIS1,12 Full-length SCF is composed of 248 amino acids and includes extracellular, transmembrane and cytoplasmic domains. Soluble SCF is 164/165 amino acids long and shares ~16% identity (~32% similarity) with another hematopoietic growth factor M-CSF. M-CSF, in turn, has been found to be a member of a growing list of growth factors (such as IL-4, GM-CSF, Growth Hormone, etc.) which share a 4 -helix bundle folding motif.

GLYCOSYLATION13,14 SCF undergoes rather extensive and heterogeneous N-linked & O-linked glycosylation. However, the absence of glycosylation at normally glycosylated sites does not significantly affect receptor binding or biological activity. It is interesting to note that in human SCF one of the glycosylation consensus sequences (at Asn120) is always glycosylated while another such consensus sequence (at Asn72) is never glycosylated. O-linked glycosylation occurs at residues 142,143 & 155 which are believed to be in the flexible spacer domain just proximal to the transmembrane domain.

CONSERVED CYSTEINES & DISULFIDE BONDS14 Among the various species variants of SCF, four cysteines are absolutely conserved. These have been shown to form two disulfide bonds at Cys 4-Cys89 and Cys43-Cys138.

SECONDARY STRUCTURE13 Circular dichroism studies show significant -helical secondary structure ranging between 38% - 47% depending on the type of theoretical analysis used to interpret the spectroscopic data. According to the CD results, about 33% -sheet structure is also present.

DIMERIZATION13,15,16 Sedimentation equilibrium and gel filtration analyses indicate that SCF forms non-covalently linked homo-dimers. Activation of the c-kit receptor has been shown to involve receptor dimerization. Dimerization of SCF (the c-kit ligand) may facilitate this signal transduction process.

REFERENCES 1 12 13 14 15 16

Martin FH, et al., (1990), Cell, 63:203. Bazan, JF, (1991), Cell, 65:9. Arakawa, T, et al., (1991), J Biol Chem., 266:18942. Lu, HS, et al., (1991), J Biol Chem., 266:8102. Arakawa T, et al., (1992), Anal Biochem., 203:53. Lev S., et al., (1992), J Biol Chem., 267: 15970.

SEQUENCE ANALYSIS

N-Linked Glycosylation Asn120

Glycosylated

Asn65

Some

glycosylation

(accounts for heterogeneity)

Asn72

SECONDARY STRUCTURE13 l

Not glycosylated

RESTRICTED CONDITIONS Initial crystallization trials of the 165 amino acid-long soluble form of SCF were unsuccessful. The high solubility of this protein as well as the postulated flexibility of the spacer domain proximal to the transmembrane domain may account for these difficulties with crystallization. We undertook, therefore, crystallization efforts with a 141 amino acid-long truncated form of SCF which still retains receptor binding capability and biological activity. In this case, crystallization trials were successful but the conditions were highly restricted with even small changes in precipitant concentration, pH and temperature having adverse effects. For example, crystals grew in our warm room at about 22-23C but did not grow in the 20C incubators. The optimum conditions for crystallization are shown in the panel on the right. Until all of these conditions could be precisely determined, reproducibility was a frustrating problem.

MULTIPLE ORTHORHOMBIC FORMS Several closely related orthorhombic forms were grown sometimes within the same droplet. One of these (form B) had relatively small unit cell dimensions and diffracted very well. Unfortunately, this crystal was impossible to reproduce with A form (with much poorer characteristics) dominating. Stabilization protocols (see below) made it possible to observe suitable diffraction from some of these A form crystals.

CALCIUM REQUIREMENT Crystal growth was absolutely dependent on calcium (0.25M). Interestingly, crystals did not grow with other divalent cations such as Mg, Zn or various lanthanides; use of these cations resulted in heavy precipitation. However, crystals treated with 10 mM EDTA were not disrupted suggesting that calcium was not essential for crystal packing.

STABILIZATION Stabilization of the crystals proved to be essential in improving diffraction as well crystal lifetime. Unstabilized crystals diffracted rarely beyond 3.8Å spacings and typically had an x-ray lifetime of less than 18 hours. Stabilization with slightly higher precipitant concentrations and a lower pH (pH 7.2 as opposed to pH 7.5 during crystal growth) greatly improved diffraction to beyond 3.0Å and increased the crystal lifetime to beyond 100 hours.

MISCELLANEOUS ISSUES Addition of the crystallization conditions with -octyl glucoside was unproductive. In addition crystallization with MOPS or TRIS at the equivalent or slightly different pH ranges resulted in poorly formed crystals. HEPES was the only effective buffer at the crystallization pH of

7.5. The crystals were moderately biaxially birefringent. purposes, the best crystals measured 0.35  0.25  0.25 mm.

For diffraction

DIFFRACTION PARAMETERS 0-level precession photographs were used to determine unit cell dimensions as well as Laue group symmetry. As mentioned before, several orthorhombic crystal forms were obtained many of which had very similar unit cell dimensions and varying diffraction quality. Inspection of the morphology did not lend any clues as to which form a particular crystal belonged to. This of course complicated our initial analysis of the diffraction. The best crystals had unit cell dimensions of a=88.1Å, b=83.4Å, c=72.7Å. Based on these results, four molecules per asymetric unit gave the most plausible solvent content (46%). These results were also confirmed using a biochemical analysis based on amino acid composition (see the panel on PACKING MODEL). Initial diffraction patterns showed 90C reciprocal lattice angles, mm symmetry each of the 0-level nets, and systematic absences of odd reflections along all three axes h, k, and l. The space group of these protein crystals is thus P212121.

SCF hk TRANSFORM17,18,13 Unusual diffraction patterns can sometimes indicate large-scale structural features in the crystal structure. This fact was used by Watson & Crick to lead them to the double-helical structure of DNA. A similar situation may be possible with SCF although a closer analogy would be the crystallographic studies of bacteriorhodopsin or more recently a leucine zipper. In the case of bacteriorhodopsin strong reflections are observed at 10Å and 5Å spacings. These patterns, termed the 10Å band & 5Å band are mutually perpendicular to each other and reflect (no pun intended) the interhelical spacings and helical-rise spacings respectively of this predominantely -helical membrane protein. The SCF hk transform also shows a stronger ring of diffraction at about 9.25Å spacings along the k-axis. The presence of this pattern may indicate an approximately side-by-side packing of the -helices in SCF along the k(c) axis. This data is consistent with the largely -helical content of SCF as determined by CD & sequence analysis.

SCF hl TRANSFORM Both a 10Å band (10.5Å) along the l-axis and a 5Å band (5.2Å) along the haxis are observed in the SCF hl 0-level net. This is suggestive of the helices lining up along the h(a) axis and packing side-by-side along the l axis. Of course, for this to be possible all four molecules in the asymetric unit must be oriented in approximately the same direction. Such a packing model is consistent with Patterson self-rotation searches we have computed with native data (see the SELF-ROTATION panels).

REFERENCES

13 Arakawa, T, et al., (1991), J Biol Chem., 266:18942. 17 Blaurock, A, J Mol Biol. 18 Rasmussen, R., et al., (1991), Proc Natl Acad Sci USA, 88:561.

DIFFRACTION PARAMETERS

Unit cell dimensions

a=88.1Å, b=83.4Å, c=72.7Å.

Za

4

Solvent content

46%

Space Group

P212121

BACTERIORHODOPSIN TRANSFORM SCF hk TRANSFORM

HELIX PARAMETERS SCF hl TRANSFORM

lDATA

COLLECTION STRATEGY

The most important factor in data collection was the stabilization and careful handling of the SCF crystals. Even with such care, data collected was often poor. For example, data collection from one crystal gave an Rsym of 0.087 out to 3.8Å but rapidly decayed in quality further out. Another crystal gave reliable and complete data out to 3.0Å and some data out to 2.8Å. Data was collected on a Hamlin area detector using a home x-ray source at 50kV and 100mA. Our best native data set was collected from a single crystal for a total irradiation time of 42 hours. Three orientations were used to meet the competing demands of crystal decay, data completeness and adquate overlap for merging statistics. Attempts to collect data beyond 2.8Å never succeeded although there is hope that using a synchrontron source may substantially improve the observable diffraction.

DATA REDUCTION The data was reduced using the CCP4 crystallographic computing package (ROTAVATA & AGROVATA). The data totaled 21,028 measurements with 5934 of these independent. The cumulative Rsym was 0.039 out to 2.8Å although by this point the data was rapidly falling off in quality since the last shell (2.87-2.8Å) had an Rsym of 0.377. The fact that the data looks good at high intensities points favorably to the possibility that data collection at a synchrontron may prove useful. Note also that the average intensity tends to rise between 5.5Å and 4.5Å which is a relatively common phenomenon but in this case may also be indicative of the diffraction arising from the helical rise repeats (see DIFFRACTION panel).

PSEUDO-CENTERING As a possible confirmation of potential packing models, a rough search for the presence of pseudo-centering was conducted. Ratio's of even to odd reflections (k+l,h+l,h+k) were computed and nothing indicative of psuedocentering was observed either at low or high resolution.

AGVROATA RESULTS

SELF-ROTATION FUNCTIONS We know (see BIOCHEMISTRY panel) that SCF naturally forms homo-dimers. In addition, since apparently there are four molecules per asymmetric unit we would like to characterize the non-crystallographic symmetry of these crystals. Patterson self-rotation searches can be used to find such noncrystallographic symmetry elements. The basic idea is to compute a native Patterson map, systematically rotate this map by a specified angle and then compare it to the original, unrotated map. A high correlation between the rotated and unrotated map is indicative (although not necessarily so) of crystallographic or non-crystallographic symmetry. In the latter case may be masked by noise or other such uncertainties. In any case, the search for a dimer axis should reveal strong peaks when searching at  = 180. Several programs are used to calculate these self-rotation functions and the three we have used are MERLOT, GLRF and XPLOR. These results are described below.

MERLOT The MERLOT plot ( = 180) is shown on the right side of this panel. Data used for the calculation range from 35 to 10Å and a 30Å Patterson radius cutoff. Many such plots were produced with varying data limits and cutoff radii. The most common peaks were at =90 and =12, =78; the two peaks are symmetrically related. The peaks look curiously elongated in the direction of the a*-axis. This suggests that each of these peaks is actually a summation of two unresolved peaks. An analysis of peak heights (from the GLRF program discussed below) shows that the height of these fused peaks (being about 1/2 as large as teh origin peak) is to large for it to be a single peak. In addition, since there are actually two dimers in the asymmetric unit, we should likewise see two dimer axes. Given the fact that the two SCF dimers are equivalent to each other, it would be highly improbable that only one of these axes would be apparent. Our conclusion is that the two dimer axes are on either side of the fused peak. The two dimer axes are approximately: =81

=78

=99

=78

The other possibility is that one axis is along the c* axis and the other two-fold is just flanking it.

GLRF18 Prior knowledge about non-crystallographic symmetry can be used in GLRF to enhance self-rotation function peak heights. In our calculations we presupposed the existence of a two-fold axis so that the GLRF program would rotate the Patterson by 180 for each calculated comparison. A plot as well as quantitative data is shown on the right side of this panel. Note the fused peaks (1 & 2) and the flanking peaks (3 & 4).

REFERENCES 18 Tong L, and Rossmann M, (1990), Acta Cryst., A46:783.

MERLOT

GLRF

THE "DIMER-DIMER" AXIS After locating the two-folds relating each of the two dimers we need now to find the non-crystallographic symmetry element relating the two dimers to each other. We refer to this as the "dimer-dimer" axis. Because this axis relates all four molecules to each other we would expect the corresponding Patterson self-rotation function peak height to be rather large. Because the search is throughout the complete angle space, we have used XPLOR to calculate the self-rotation functions. The results are given below.

XPLOR SELF-ROTATION FUNCTION CALCULATIONS There are two possibilities for this "dimer-dimer" axis. As can be seen on the right side of this panel, XPLOR shows similar results under a wide variety of data limits & Patterson cutoff parameters. Two large peaks invariably show up.

The =0, =7.5, =180 peak This first peak is also visible in the MERLOT plot as an elongation of the origin peak along the a*-axis where the axis points out approximately in the same direction as the b*-axis. Unfortunately, because it is so close to the origin peak it is difficult to properly evaluate; this peak, for example, is not picked up by the GLRF package. In any case, if we consider the M-CSF dimer structure as prototypical, a "dimer-dimer" axis in the b*-axis direction would suggest that the helices are oriented approximately in the baxis direction. According to our general analysis of the diffraction (see the DIFFRACTION panel) the helices are oriented in the a-axis direction. Because of its proximity to the origin and its inconsistency with the other diffraction results the next peak may be a more plausible candidate for the "dimer-dimer" axis.

The =0, =90, =15 peak Since in this case   180, this peak does not correspond to a perfect twofold. The axis in this case, is approximately oriented in the direction of the a*-axis so that all the helices would now point in the direction of the a-axis. This is consistent with the model discussed in the DIFFRACTION panel. The strength and consistency of this peak using the XPLOR package is also impressive and we regard this peak as the most plausible one corresponding to the "dimer-dimer" axis.

XPLOR RESULTS: Rmax=8Å,Rmin=5Å,Patrad=20Å

XPLOR RESULTS: Rmax=35Å,Rmin=4Å,Patrad=15Å

Za calculation19 In order to more definitively determine the number of molecules in teh asymmetric unit we used a quantitative amino acid analysis protocol. This protocol involves carefully measuring the macroscopic volume of a specific crystal and then submitting this crystal to an amino acid analysis in order to determine the number of molecules in the crystal and thus the number of molecules in the asymmetric unit. The results of this study are summarized in the chart on the right side of this panel.

PACKING MODEL Based on the previously described self-rotation functions, a packing model for the asymmetric unit can be proposed. All four monomors have their ahelices approximately disposed in the direction of the a-axis. The two dimer two-folds do not overlap since the axis relating the two dimers involves a rotation of 15°.

REFERENCES 19 Kwong PD, et al., (1990), Proc Natl Acad Sci USA, 87:6423.

PHILOSOPHY OF MOLECULAR REPLACEMENT SCF shares a common 4 -helix folding motif along with several other growth factors and hormones such as GM-CSF, G-CSF, M-CSF, IL-4, Growth hormone, IL2. More such examples will no doubt be discovered in the near future. In an attempt to begin obtaining phasing information we have tried to generate conventional heavy atom as well lanthanide derivatives but to no avail. Because of the structural similarity of these proteins we have attempted a molecular replacement solution to the phase problem. In addition, with the ubiquity of this folding motif, this study will also establish the feasibility of using molecular replacement as a possible strategy for quickly solving growth factor structures in the future.

TRIAL STRUCTURE20 and INITIAL RESULTS Our first trial structure is the hematopoietic growth factor GM-CSF which is a prototypical 4 -helix bundle but nevertheless has little sequence similarity with SCF. In order to use our knowledge of the dimer two-fold, we have used the GLRF package with a locked two-fold (whose orientation is determined by the self-rotation functions). Patterson cross-rotation searches (in a manner analogous to the self-rotation procedures) were performed using the GM-CSF monomer as a trial structure. Results with and without the locked two-fold are shown on the right side of this panel. Note that there are four to five relatively high peaks which could potentially correspond to solutions for the four SCF monomers in the asymmetric unit. The peak heights for these "solutions" are enhanced when the GLRF calculation is done with the locked two-fold. In addition, noise peaks have become less prominent.

FURTHER STUDIES The next step is to refine the dimer axes and re-do the cross-rotation functions in the hope of enhancing the peaks of interest. The strategy then would be to apply a translation function to search for the orientation and position of one of the GM-CSF monomers. The refined non-crystallographic symmetry elements would then be used to generate the second GM-CSF molecule and a translation search would then be applied to this molecules. The third and fourth molecules would be similarly generated and positioned. Variations on this theme could include the use of the GM-CSF fold with a polyalanine sequence or an averaging of several 4 -helix bundle proteins. Ideally, we would like to use the M-CSF molecule since this is both more closely related to SCF as well as shares its dimerization properties.

REFERENCES 20 Diederichs K, Boone T, and Karplus PA, (1991), Science, 254:1779.

GM-CSF TRIAL STRUCTURE

INITIAL GLRF RESULTS

THE M-CSF STRUCTURE21 The three-dimensional x-ray crystallographic structure of M-CSF has been recently determined. As expected, it has a 4 -helix bundle folding motif characteristic of other such growth factors. It shares about 16% sequence identity with SCF and as with SCF it is a dimeric structure. However, unlike SCF which is a non-covalently bound dimer, the M-CSF dimer is covalently bound via a single intramolecular disulfide bond.

FEATURES OF THE SCF MODEL Based on sequence comparisons and helical hydrophobicity moments a model structure of SCF superimposed on a schematic rendition of M-CSF structure has been proposed that corroborates some of the biochemical data. The conserved cysteines, for example, easily fit into positions that allow the formation of the two intramolecular disulfides. The first disulfide bond attaches the Cterminus to the top of helix B while the second disulfide bond attaches the N-terminus to the top of helix C. It has also been observed that not all of the consensus sites for N-linked glycosylation are, in fact, glycosylated in the native protein. In particular, Asn120 is always glycosylated, Asn69 sometimes glycosylated and Asn72 is never glycosylated. According the model structure Asn120 is on the outer face of helix D while Asn72 is situated towards the core of the molecule at the confluence of helices A and B. Obviously glycosylation at this site would seriously disrupt the SCF structure. Asn69 is situated on the outer edge of helix C. Other important features of this model include the acidic surfaces of the helices, and the conserved dimer interface.

ACIDIC RIDGES The helices in SCF have rather marked hydrophobic moments: the interior facing portions of the helices are quite hydrophobic while the exterior surfaces are often charged. Helices B, C and D, in particular have a large preponderance of negatively charged acidic aspartates and glutamates situated on their surfaces. Helix B for example has an uninterrupted, continuous ridge of five acidic residues on its surface. It is possible that the role of divalent cations in both the precipitation and and crystallization of SCF may in part be derived from the neutralization of these "acidic ridges" by these cations.

CONSERVED DIMER INTERFACE The region of the molecule distal to the C-terminus turns out to be completely and strictly conserved among the SCFs from different species. This region is also analogous to the dimer interface for M-CSF so it is very likely that it too is the dimer interface for SCF. Knowledge of the dimer site may help in designing SCF antagonists that may prove useful in cancer chemotherapy, etc.

PROPOSED RECEPTOR BINDING SITE Because the C-terminus leads towards the cell surface and the dimer interface interacts with the other SCF monomer, it seems plausible to propose that the receptor binding site is composed of the surfaces of helices A & C. A more detailed knowledge of these surfaces in conjunction with mutant studies would assist in the development of SCF agonists --potentially useful agents in treating various clinical anemic states.

REFERENCES 21 Pandit J, et al., (1992), Science, 258:1358.

M-CSF STRUCTURE

SCF MODEL