Structural Organization of the Human Insulin Receptor Ectodomain”

To provide an experimental system amenable to a detailed biochemical and structural investigation of the extracellular (ligand binding) domain of the insulin receptor, we developed a mammalian heterologous cell expression system from which tens of milligrams of the soluble secreted ectodomain (the IR92 1 protein) can be routinely purified using methods that do not require harsh elution conditions. The purified IR921 protein has a Stokes radius of 6.8 nm and a sedimentation coefficient of 9.8 S, from which we calculate a hydro- dynamic mass of 281 kDa. Electron microscopic images, using both rotary shadowing and negative stain- ing techniques, demonstrate a characteristic substructure for the IR921 protein consisting of two elongated arms, with a globular domain at each end, connected to each other at a point somewhat off-center to form a Y structure. Analysis using circular dichroism and fluorescence spectroscopy illustrate that insulin bind- ing results in conformational changes in the ectodomain. Furthermore, fluorescence anisotropy decay data reveal segmental mobility within the IR921 protein that is successively frozen as a result of insulin binding, in contrast to results obtained in a previous study of the epidermal growth factor receptor ectodo- main. This result suggests a divergence in hormone-induced signaling mechanisms

autonomous function ( i e . the secreted ectodomain is a highaffinity insulin binding protein, and the cytoplasmic domain is an active protein-tyrosine kinase; Ellis et al., 1987Ellis et al., , 1988aEllis et al., , 1988bHerrera et al., 1988;Johnson et al., 1988 Whittaker andOkamoto, 1988). Despite the wealth of information available concerning the structure and sequences of vertebrate insulins, the details of how the ligand interacts with its receptor are poorly understood. In addition, little is known about the structural organization of the ligand binding domain. To provide an experimental system amenable to a detailed biochemical and structural investigation of the receptor ectodomain, we have developed a mammalian heterologous cell expression system from which tens of milligrams of the soluble secreted ectodomain (the IR921 protein; see Schaefer et al., 1990) can be routinely harvested. As part of our efforts to understand structure/function relationships intrinsic to the organization of this complex domain of the receptor, we present the results of hydrodynamic, electron microscopic and spectroscopic (circular dichroism and fluorescence) studies of this molecule.

MATERIALS AND METHODS
Selection of Overproducing Cell Strains-Chinese hamster ovary cells (CHO-K1, obtained from ATCC, Rockville, MD) were maintained at 37 "C in an atmosphere of 5% CO, and 95% air in Ham's F-12 medium with 5% fetal bovine serum, 100 units/ml of penicillin G, 100 pg/ml streptomycin, and 10 mM Hepes (pH 7.4). CHO-K1 cells were co-transfected with a recombinant plasmid (pehIRO1) containing a 2983-base pair cDNA fragment that encodes all but 8 amino acids of the human IR (hIR) extracellular domain (Ellis et al., 1988b) and a recombinant plasmid that confers resistance to the neomycin analog G418 (Southern and Berg, 1982). IR residues are numbered as described (Ebina et al., 1985). A solid-phase insulin binding assay was used to screen for high expressing clones that maximally secrete the soluble ectodomain. The binding of lZ5I-insulin (monoiodinated porcine insulin (80-120 mCi/wg) was from Du Pont-New England Nuclear; unlabeled porcine insulin was a gift from Eli Lilly) to receptors immobilized by anti-IR monoclonal antibodies (Soos et al., 1986; provided as a gift from Dr. Ken Siddle (University of Cambridge, United Kingdom)) on microtiter plates was performed as described (Morgan and Roth, 1986;Ellis et al., 1988b;Sissom and Ellis, 1989;Schaefer et al., 1990).
Purification of the IR921 Protein-The highest expressing cell line was expanded, seeded into 900-cm2 roller bottles (Costar, Cambridge, MA), and adapted to growth in low serum-containing medium (1:l mixture of Ham's F-12 and Dulbecco's modified Eagle's medium with 1% fetal bovine serum, 100 units/ml of penicillin G, 100 pg/ml streptomycin, 800 pg/ml of G418, and 10 mM Hepes (pH 7.4)). Conditioned medium (100-300 ml/bottle) was harvested every 24-48 h for 2-3 weeks and was centrifuged at 1000 X g for 30 min in a adjusted to 50 mM Tris (pH 7.4) and 0.02% sodium azide and was Sorval GS-3 rotor to remove any cells or debris. The supernatant was stored at 4 "C. The pooled conditioned medium then was sterilefiltered (0.22 p m ) and concentrated -200-fold (i.e. from -20 liters to -100 ml) using a Minisette ultrafiltration system (Filtron Technology Corp., Clinton, MA). The use of a 100-kDa cut-off membrane also provided an initial purification step, by removing proteins of lower molecular mass. Protease inhibitors were added to a final concentration of 1.5 p~ pepstatin A, 75 PM antipain hydrochloride, 2 p~ 23394 Insulin Receptor Ectodomain leupeptin, 6.4 p~ benzamidine, 130 p~ hestatin, 5 mM EDTA, and 2 mM phenylmethylsulfonyl fluoride. The concentrated medium then was adjusted to 37% ammonium sulfate (pH 7.4, added dropwise overnight at 4 "C while stirring), the resulting precipitate was pelleted by centrifugation at 10,000 X g for 30 min, and the supernatant was adjusted to 55% ammonium sulfate (added dropwise overnight while stirring). The precipitated protein was pelleted by centrifugation at 10,000 X g for 30 min, and the resulting pellet was resuspended in 20 mM Tris (pH 7.4) and dialyzed exhaustively at 4 "C against 20 mM Tris (pH 7.8). This material was fractionated by anion-exchange chromatography with an FPLC Mono Q 16/10 column (Pharmacia LKB Biotechnology Inc.; buffer A was 20 mM Tris (pH 7.8); buffer B was buffer A plus 0.5 M NaCl). Fractions containing insulin binding activity were pooled and concentrated down to 2 ml using Centriprep-100 and Centricon-100 concentration devices (Amicon, Beverly, MA). This material was fractionated further by gel filtration chromatography (with an FPLC . Purified ectodomain was analyzed by SDS-PAGE (Laemmli, 1970) with or without reducing agent (P-mercaptoethanol) on 5-15% gradient gels (1.0 mm) with an acry1amide:bisacrylamide ratio of 37.5:l. Molecular masses of standard proteins (Bio-Rad) are myosin (200 kDa), @-galactosidase (116 kDa), phosphorylase b (97 kDa), bovine serum albumin (66 kDa), and ovalbumin (43 kDa).
Stokes Radius-To estimate the Stokes radius (Rs) of the IR921 protein, purified material (500 pg) was rechromatographed on the gel filtration column and its R.9 value determined from a standard curve of Rs uersus (-log KaV)l" (Laurent and Killander, 1964;Florke et al., 1990). The distribution coefficient (&) was calculated as K., = ( V , -V,J/(Vt -Vo). Blue dextran (2000 kDa) and glycyltyrosine (238 Da) were used to estimate the void ( Vo) and small solute ( V,) volumes, respectively. V, is the elution volume. Standard proteins (Pharmacia or Boehringer Mannheim) and their Stokes radii in nanometers are as follows: thyroglobulin (8.5), P-galactosidase (6.91), ferritin (6.1), catalase (5.22), aldolase (4.81), bovine serum albumin (3.55), and ovalbumin (3.05). To assess the Stokes radius of the ectodomaininsulin complex, ectodomain (500 pg, 20 p M ) was incubated with a 3fold molar excess of unlabeled insulin and 250,000 cpm of tracer '"Iinsulin at 4 "C for 4.5 h, and the mixture was fractionated as described above. The absorbance at 280 nm and the radioactive signal were determined for each fraction.
Hydrodynamic Properties-Zone sedimentation through glycerol gradients was used to determine the sedimentation coefficient and as an additional purification step prior to electron microscopy. Material for electron microscopy studies was generated by layering 200 pl of IR921 protein (at either 670 or 134 pg/ml for rotary shadowing and negative staining techniques, respectively) or equivalent amounts of reference proteins, onto 5-ml gradients of 15-40% (v/v) glycerol in 0.2 M ammonium bicarbonate (pH 8). The gradients were centrifuged in an SW 50.1 rotor at 38,000 rpm at 20 "C for 15 h, eluted in 20 fractions (collected from the bottom), and the fractions were analyzed by SDS-PAGE to identify protein peaks. Standard proteins used for sedimentation calibration were thyroglobulin (19.2 S), @-galactosidase (15.9 S), catalase (11.3 S), aldolase (7.2 S), and bovine serum albumin (4.6 S).
Based on the Rs and s values determined for the IR921 protein, its hydrodynamic molecular mass was calculated as described by Siege1 and Monte (1966), where N is Avogadro's number, v is the viscosity of the solvent (0.01 g/cm s for water at 20 "C), s is the sedimentation coefficient (in Svedberg), u2 is the partial specific volume (0.73 cm'/g for proteins), and p is the density of the solvent (1.0 g/cm" for water). The frictional ratio ( f/fmln) for the IR921 protein was estimated from the ratio of sedimentation coefficients, S&S, where smax is the sedimentation coefficient of an unhydrated sphere of radius R,,, that is sufficient to contain the mass of the protein.
Electron Microscopy-For rotary shadowing, a 10-ml sample of the peak glycerol gradient fraction (estimatedprotein concentration -100 pg/ml) was sprayed on mica and shadowed in a Balzars model BAEl2O vacuum evaporator, as described (Fowler and Erickson, 1979). For negative staining, the protein was diluted 5-fold, and one drop was applied to a freshly glow-discharged carbon film, withdrawn, and the grid stained with three drops of 2% uranyl acetate (unadjusted pH).
Sample Preparation for Spectroscopic Studies-All buffers were prepared with Milli-Q System (Millipore S.A., Molsheim, France) water. Ectodomain (2.5 mg) was ultrafiltered using a Centricon-100 device to exchange the solvent into 10 mM Tris buffer (pH 7.5), and the protein solution was adjusted to 7 p~. An extinction coefficient at 280 nm of 272,700 M" cm" was calculated for the IR921 protein as described (Gill and von Hippel, 1989). The concentration was confirmed by absorption spectroscopy using a Pye-Unicam PU 8800 UV/VIS spectrophotometer (Philips, Kassel, Germany) to assure that the protein concentration would be optimum for CD measurements in optical cells of appropriate path lengths for the different wavelength ranges (see below). A stock solution of human insulin, a generous gift from Eli Lilly, in 10 mM Tris buffer (pH 7.5), was adjusted photometrically to 14 pM (using the absorption coefficient A (1 cm; 1 g/liter; 276 nm) = 1.065). The following solutions then were prepared ED = 3.5 pM IR921 protein; I1 and I2 = 3.5 and 7.0 p~ insulin, respectively; and the corresponding mixtures: ED11 = 3.5 p~ IR921 protein and 3.5 pM insulin; ED12 = 3.5 p~ IR921 protein and 7.0 p~ insulin. The 1 ED:2 insulins stoichiometry was included to test the recent proposal of Markussen and colleagues (Markussen et al., 1991) that both insulin binding sites on the ectodomain become occupied with hormone. The relevant questions are thus: ED11 -I1 = ED? and ED12 -I2 = ED? After completion of the study, an aliquot of the IR921 protein solution was submitted to quantitative amino acid analysis (Stein and Moore), which gave a concentration of 3.3 p~. Thus, the actual combinations were just slightly over stoichiometric with respect to insulin, i.e. ectodomain to insulin ratios of 1:1.06 and 1:2.12, respectively.
AVIV 62 DS (Lakewood, NJ) or a Jasco 5-600 (Japan Spectroscopic Circular dichroism measurements were carried out on either an Co., Ltd., Tokyo) CD spectrometer. The instruments were calibrated with a 0.1% aqueous solution of d-10-camphorsulfonic acid at 290.5 nm according to Chen and Yang (1977). The ratio of the d-10camphorsulfonic acid hands at 192.5 and 290.5 nm was -2.0. Further details were as described (Renscheidt et al., 1984). The spectral bandwidth was 1.5 nm. The path lengths of the optical cells used were: 180-205 nm, 0.1 mm; 205-250 nm, 0.5 mm; 250-320 nm, 10.0 mm. The same three cells were used for all measurements. Measurements were carried out at 26 "C, but otherwise all solutions were kept at 4 "C.
Steady-state fluorescence spectra were recorded on a Spex Fluorolog 211 photon counting spectrofluorimeter (Spex Industries, NY), with bandwidths of 2.7 and 2.2 nm for excitation and emission monochromators, respectively. They were corrected for changes in the lamp intensity and for spectral sensitivity of the emission monochromator-photomultiplier system. Excitation wavelengths were 295 or 300 nm in order to excite tryptophan residues directly and to avoid energy transfer and inner filter effects. measured in the single photon counting mode with an Edinburgh Fluorescence lifetimes and fluorescence anisotropy decay were Instruments Ltd. (United Kingdom) model 199 lifetime spectrometer. The thyratron-gated hydrogen flashlamp was operated at 40 kHz, and it provided a lamp pulse of 1.4 ns FWHM. The excitation wavelength was 300 nm with a maximum bandwidth of 12 nm. Residual first and second order stray light was suppressed by adequate black and cut-off glass filters (Schott, Mainz, Germany). At least 80,000 counts were accumulated in the peak channel of the total fluorescence intensity, S(t). The lamp pulse, U t ) , was recorded with a suspension of Ludox (Du Pont) using incident light at 336 nm, which is the wavelength where the fluorescence emission of the IR921 protein is maximum. Data handling and iterative nonlinear least squares fit of the decays were accomplished by a program supplied by Edinburgh Instruments Ltd. In the equations provided below, upperand lowercase letters indicate experimental and fitted decays of the total intensity, difference, and anisotropy, respectively.
The rotation-free total intensity decay, s(t), was fitted to a sum of the exponentials (Lakowicz, 1983), where b,, represents the background caused by the dark counting noise of the instrument, and b, and T~ are the amplitude and lifetime of the ith excited state, respectively. The contribution of thejth exponential term to the total fluorescence decay was calculated as follows.  Table I. The fractional intensity permits calculation of the mean lifetime, <T> = ZB,T,.
The experimental anisotropy, R(t), is defined as (Lakowicz, 19831, where D(t) is the decay of the difference between the polarized fluorescence intensities, and g is the sensitivity correction parameter. Two rotational modes are probably the limit which can be resolved with the current instrumentation and analytical procedures (Bucci and Steiner, 1988). thus, r ( t ) = r;e"'/+i) + r,, i = 1, 2 r, = r, + r2 + r, (Eq. 6) where r; is the anisotropy, 6; is the rotational correlation time, r, is the zero point and ris the long-time limit of the emission anisotropy: r( t + 0) = r,, r( t -P 00) = r-. Calculation of the anisotropy parameters using D(t) and s(t) was as described (Barkley et al., 1981). In the simplest case of a fluorophore fixed to a spherical molecule the rotational correlation time, 4 (in nanoseconds), is calculated as, where q is the viscosity of the solution (0.94 centipoise for water at 23 "C), V is the hydrated volume of the molecule (in cm'), k is Roltzmann's constant, and T i s the absolute temperature. According to Bucci and Steiner (1988), there are two ways to estimate the hydrated volume (in cm"). The first approach is dependent on the molecular mass, M (in g/mol), where u2 is the partial specific volume (0.73 cm'/g for proteins), h is the hydration level, which we assume to be 0.2 cm" water/g of protein and N is Avogadro's number. On the other hand, a nonspherical molecule can be represented by an equivalent spherical rotor whose effective radius is referred to as the Stokes radius (Rs). Based on the H s of the IR921 protein determined by gel filtration, the effective volume can be calculated and hence its predicted d, from Equation 7. When the rotational diffusion of the protein is superimposed by the motion of a flexible segment, the time-resolved anisotropy is (Lakowicz, 1986), where $,, is the overall correlation time of the protein that is calculated via Equation 7, is the effective rotational correlation time of the segment, and a is the fraction of the total depolarization which is lost by the segmental mobility. A detailed analysis of this problem was given by Lipari and Szabo (1980). This formula holds true only if the segmental motion occurs independently of the overall rotational motion. A single is apparent only if either the emission or the absorption dipole of the fluorophore is pointing along the major axis of the segment or if the segmental motion is a "wobbling cone" motion rather than spinning about the axis. In other cases is split into two components.

RESULTS AND DISCUSSION
Expression Strategy for Purification of a Soluble IR Ectodomain-An experimental prerequisite for the application of biophysical methods to study structure/function relationships of the IR ectodomain is the availability of a heterologous cell expression system from which tens of milligrams of the protein of interest can be readily purified. Our initial studies of a secreted soluble derivative of the ectodomain utilized a stably transfected Chinese hamster ovary cell line (the CHO.hIRO1 cells of Ellis et al., 1988b). The expression plasmid designed for these studies, in which the receptor was truncated following residue 921, encodes all of the a-subunit and all but 8 of the extracellular residues of the P-subunit (this protein is now designated IR921; see Schaefer et al., 1990). While these cells efficiently synthesized, processed, and secreted the heterotetrameric (apo)z truncated receptor that bound insulin with wild-type affinity (i.e. -1 nM), the overall yield of protein was too low for large-scale protein purification to be feasible (-50 pg/liter of conditioned culture medium). Subsequent attempts to produce the ectodomain in insect cells using a recombinant baculovirus demonstrated that secretion of the properly processed IR921 protein is much less efficient in this system; a significant fraction of the protein either accumulates intracellularly as a noncleaved precursor with high-mannose oligosaccharide chains, or it is secreted as a noncleaved precursor whose oligosaccharide side chains have been partially trimmed Ellis, 1989, 1991). Therefore, we revisited the mammalian CHO cell expression system and used a solid-phase insulin binding assay to screen -300 cotransfected G418-resistant clones in order to isolate high-producing cell lines that maximally secrete the IR921 protein. Using this expression system and the purification scheme described under "Materials and Meth-od~,'' we can now reproducibly purify -30 mg of ectodomain from 20 liters of conditioned medium. Moreover, this purification scheme does not utilize harsh elution conditions (i.e. 3 M urea and pH 5; Markussen et al., 1991).
The final step of the purification is illustrated in Fig. L4.
As assessed by '251-insulin binding, the ectodomain elutes coincident with the &galactosidase standard ( M I 540,000; RS = 6.9 nm; Fig. lA). The receptor protein so purified is visualized by Coomassie Blue staining as a single species on both reducing (left lane, the a-and truncated @-subunits) and nonreducing (right lane, the (a/30)n receptor complex) SDSpolyacrylamide gels (Fig.  123). Note that three bands are visible for the reduced &subunit. Treatment with endoglycosidase F demonstrates that this heterogeneity is due to differences in glycosylation, which are observed in cell lines selected for high level secretion of the ectodomain (data not shown). The protein purified by these methods exhibits the same degree of interaction with a panel of conformationsensitive receptor-specific monoclonal antibodies as the protein that is assayed in the conditioned medium prior to purification (data not shown). Furthermore, the apparent dissociation constant observed with radiolabeled insulin is identical before and after purification, i.e. -1 nM (data not shown).

The IR921 Protein in Solution Has an Elongated Shape-
The hydrodynamic properties of the soluble ectodomain were assessed in two ways. First, the IR921 protein, purified as shown in Fig. 1, was rechromatographed on the gel filtration column to determine its elution behavior. Parallel runs of standard proteins were also performed, and the resulting data are presented in Fig. 2A. From such plots, we estimate the Stokes radius of the IR921 protein to be 6.8 nm. The same estimate was also obtained when the ectodomain was preincubated with insulin (data not shown), illustrating the lack of detectable changes in shape induced by binding of the hormone (however, see further below). Similar results were obtained by . In contrast, a shift in the Stokes radius from 9.5 to 7.9 nm is observed upon insulin binding to the intact detergent-solubilized receptor (F16rke et al., 1990).
To characterize further the hydrodynamic nature of the ectodomain, purified material was subjected to glycerol density gradient centrifugation (Fig. 2B). In glycerol gradients the IR921 protein sediments one or two fractions behind catalase, which migrates at 11.3 S (Fig. 2B, fractions 10 vers'sus 8 or 9, respectively). Based on six gradients with three different preparations, our best estimate for the sedimentation coefficient of the IR921 protein is 9.8 S. The combined RS and S values give a mass of 281 kDa for this soluble ectodomain protein (see Equation 1 under "Materials and Methods"), slightly larger than the -210 kDa calculated from the sequence that is deduced from the cDNA. This difference in molecular mass presumably results from extensive posttranslational glycosylation (there are 17 potential N-linked sites (Ebina et al., 1985;Ullrich et al., 1985) as well as 0-linked oligosaccharides (Collier and Gorden, 1991). The estimated mass of the IR921 protein is close to that of catalase (232 kDa), which was used as a standard in both sedimentation and gel filtration analyses. Catalase is a compact globular protein with dimensions of 10.3 x 7 x 7 nm (as determined by x-ray crystallography (Murthy et al., 1981)) or approximately 10 nm in diameter (as determined by electron microscopy, see below). The fact that the IR921 protein appears larger than catalase when assessed by gel filtration (Rs of 6.8 uersus 5.2 nm, respectively) and smaller than catalase when assessed by sedimentation (9.8 uers'sus 11.3 S, respectively), suggests that the ectodomain has a more extended shape in solution. This is reflected in the calculated frictional ratio ( f / f m i n = smax/s, see Equation 2) of 1.64 for the IR921 protein, where a value of 1.3-1.4 is typical of globular proteins and a value of 1.6-1.9 is typical of moderately elongated proteins (Erickson, 1982). Thus the ectodomain in solution has an elongated shape. A similar discrepancy between Rs and the apparent. size, as assessed by sedimentation analysis or electron microscopy, is observed for the intact detergent solubilized receptor (Baron et al., , F16rke et al., 1990al., , Christiansen et al., 1991.

Electron Microscopic Analysts of the Ectodomain Reveals a -20-nm Extended Particle with Globular Domains at Either
End-The shape of the IR921 protein was studied further by electron microscopy, using both rotary shadowing and negative staining techniques. The most vivid substructural detail is seen in the rotary shadowed specimens (Fig. 3). Although many molecules appear globular or show no interpretable substructure, two or three characteristic views of more extended molecules are consistently identified. These images, Magnification is X 250,000, and the bar is 50 nm. B, a field of molecules at lower magnification (X 150,000; the bar is 100 nm). In addition to the X and Y structures, several less distinctive globular particles are seen. C, a cartoon summary for the observed images of the ectodomain. The X and Y structures are the middle and right-hand diagrams, respectively. The diagrams depict two elongated arms with globular domains on each end. The kit-hand diagram depicts a single arm, which corresponds to one CUB half-ectodomain. which are summarized in Fig. 3C, consist of two elongated arms, with globular domains at each end, connected to each other at a point slightly off-center to form an " X or a "Y" structure. The X structure is demonstrated by images in the first two rows in Fig. 3A (note that some of these, e.g. the two left-hand images of row 1, might be described as an "H," but for simplicity we group these all together). The Y structure is demonstrated by the third row of images in Fig. 3A. The average length of the arms measured across the X structure is 20 nm (after correcting for the estimated -1-nm shell of metal).
In contrast, it was more difficult to identify the substructure of the negatively stained IR921 protein. To provide a standard for comparison, a sample of catalase was also examined by negative staining. As illustrated in Fig. 4C, catalase appears as compact globular molecules, with an average diameter of 10.3 nm. With the negatively stained ectodomain, most of the particles that could be identified as single molecules appear as irregular globular or variably elongated structures, on the order of 10 X 15 nm. These globular structures are generally larger more asymmetric and more variable in stain density than catalase. Upon closer examination, we also identified a reasonable number of images that convincingly resemble the Y structure (see Fig. 4, A and B, for selected images). The size of these negatively stained Y-shaped molecules is very similar to that determined by rotary shadowing; each arm of the Y is about 10-12 nm long and the shaft is about 6-10 nm long. Remarkably, the Y-shaped molecules are often found in close proximity to one another. Thus most of the small panels in Fig. 4A show more than one Y structure (we count 2,2,2, 3, and 1 Y structure, respectively, in the five cut-outs). We

FIG. 4. Electron micrographs of negatively stained IR ec-
todomain and catalase. A , selected images of the ectodomain, which correspond to the Y structures seen by rotary shadowing. H , a larger field of molecules, in which more than a dozen images have an identifiable Y structure. C, images of catalase, which appears as compact globular molecules with an average diameter of 10.3 nm (magnification is X 250,000 for all panels, and the bar is 50 nm). Note that catalase molecules are both smaller and more dense than the IR921 molecules. also found a larger field of molecules in which about half of the particles appear as Y structures. This field, shown in Fig.  4B, is not presented as being representative, it is the best field we could find. An average field this size typically shows only the globular particles. The occasional cluster of two or more well preserved Y-shaped molecules suggests that there are rare and usually small "good" areas of the specimen, where the molecules are better preserved and/or better stained.
Thus both rotary shadowing and negative staining techniques demonstrate a characteristic substructure, with two elongated arms connected to form a Y. The closely related but somewhat more expanded X structure was rarely seen by negative staining, so we refer to the Y as the consensus structure. In this respect, these images of the ectodomain are similar to those of immunoglobulin G molecules, observed by negative staining (Valentine and Green, 1967). IgG exists as a dimer comprised of two light chains and two heavy chains held together by disulfide and noncovalent bonds. The Fab regions, which each possess an epitope binding site, form the arms of the Y, whereas the Fc region forms the stalk. Similarly, each arm of the IR ectodomain contains an insulin binding site. Thus certain structural and functional features appear to be shared between these two distinct classes of proteins. The Y structure of the IR ectodomain may indeed be the true shape of the molecule in solution. Alternatively, the solution conformation might be somewhat more compact (corresponding to the globular particles that constitute the majority of the images), and the Y structures would represent an expanded or somewhat unfolded conformation. In either case, these images demonstrate the fundamental organization of the heterotetrameric ectodomain, i e . each a/3 half-molecule appears capable of extending to a length of -20 nm, possesses a globular domain at each end and is connected to its partner (presumably via disulfide bonds, as well as potential proteinprotein interactions) at a point somewhat off-center.
The images we present are consistent with two previous studies of the IR. Johnson et al. (1988) presented images of negatively stained IR ectodomain that show somewhat elongated globular particles. Although they did not present quantitative measurements and their images are at rather low magnification, we measured dimensions in the range of 12-19 nm from their Fig. 7B. This is reasonably close to the dimensions of the globular particles in our own images. They, however, observed no substructure. In a more recent and detailed study (Christiansen et al., 1991), intact insulin receptors were isolated by detergent extraction of placenta and they were examined by negative staining. This study demonstrated a structure resembling a "T," with a height of -24 nm and a width across the T of -18 nm. It seems clear that our Y structure corresponds to the upper half of this T. Indeed, in several of their images the two arms of the T are bent upward to form an angle very similar to that in our Y-shaped molecules, suggesting that a certain degree of flexibility may exist in this region. The dimensions they report are also quite similar to ours, the major difference being that transmembrane and cytoplasmic domains of the @-subunits apparently extend the length of the stalk of the Y to that observed for the T.
The electron microscope images of the IR ectodomain allow further insight with .respect to the results of our earlier carboxyl-terminal deletion analysis of this protein . One particularly striking result is that even the smallest deletions in this series (i.e. the IR883 and IR794 proteins, which terminate within the carboxyl-terminal portion of the @,,-subunit; the numbering scheme refers to the number of IR amino acid residues in each protein) dramatically compromise the secretion of the truncated proteins, as well as the stability of the IR794 protein (see Fig. 2 Receptor Ectodomain et al., 1990). It is only when the sites of truncation are placed much further amino-proximal (beginning with IR486) that efficient secretion is restored. At the time, we imagined that there might well be an element of heretofore unrecognized structure in this expanse of the receptor sequence, which our restriction site-based deletions perturbed, but we had only a minimal understanding of the nature and extent of folding in this region. The electron microscope images presented in the current study illustrate that the carboxyl terminus is indeed folded so as to form a globular domain at the end of each a@ half-receptor.

Insulin
An important related observation stems from the recent identification of two copies of the fibronectin type I11 repeat in the primary sequence of the ectodomains of several ligandactivated tyrosine kinase receptors including those for insulin and insulin-like growth factor I (O'Bryan et al., 1991;Pasquale, 1991). As illustrated in Fig. 5, alignment of the carboxyl terminus of the human IR ectodomain with that of the human growth hormone receptor and domain 3 of tenascin, for which crystal structures have recently been reported (De Vos et al., 1992;Leahy et al., 1992), reveals conservation of characteristic proline, tryptophan, tyrosine and leucine (as well as conserved acidic and other hydrophobic) residues. Interestingly, although the second type I11 repeat of the IR is continuous, the first repeat is interrupted by an insert that contains the a-@ cleavage site and an alternatively spliced exon (number 11) that is present in the Ebina (Ebina et al., 1985) sequence but not that of Ullrich (Ullrich et al., 1985) (see Fig. 9). The type I11 repeat is a seven stranded @-structure (Baron et al., 1992;Leahy et al., 1992), whose strand topology is analogous to that of the D2 domain observed in the crystal structure of human CD4 (Ryu et al., 1990;Wang et al., 1990). In the alignment of Fig. 5, the insert within the first type I11 repeat of the IR resides within the loop between @ strands E and F, suggesting that it is an independently folded domain. The large globular domain at the carboxyl terminus of each a@ half-receptor observed by electron microscopy (Fig. 3) is consistent with our prediction that the insert region extends toward the second type I11 repeat and in fact suggests that these two regions are intimately associated. As the interaction between strands within each repeat is not in the linear order of the protein primary sequence, it is to be expected in retrospect that a random truncation (i.e. with respect to protein sequence, since our earlier carboxyl-terminal deletion analysis was based on the use of DNA restriction sites; Schaefer et al., 1990) within such a structure would play havoc with the structural integrity of this region. Although the role of the type I11 repeats in insulin receptor structure/function has yet t o be tested experimentally, they may serve as a spacer to extend the insulin binding site away from the plasma mem-brane (cf. the discussion of Fig. 4c in Leahy et al., 1992) and/ or facilitate transmembrane signaling or additional receptorprotein interaction(s).
In addition, the electron microscope images predict that the amino-terminal portions of each aP0 half-ectodomain (the arms of the Y) are independently folded. Consistent with these images, we demonstrated that truncations within the amino-terminal -35-50% of the ectodomain result in monomeric proteins that are secreted as efficiently as the intact heterotetrameric ectodomain (e.g. the IR486 protein of Schaefer et al. (1990)). Recent work using photoactivatable derivatives of insulin illustrate that the a-subunit appears to fold such that the amino terminus and the beginning of the carboxyl-terminal half come into close contact with the hormone (Wedekind et al., 1989;Fabry et al., 1992;see Fig. 10).
Surprisingly, however, carboxyl-terminal truncations of the ectodomain, which contain the determined insulin contact sites, do not bind the hormone , suggesting that additional elements are required to constitute the hormone binding site. The insulin contact sites are apparently all within the cy-subunit since a truncated ectodomain containing all but 7 carboxyl-terminal residues of the a-subunit binds insulin with near wild-type affinity (the IR728 protein of Schaefer et al. (1990)). It is quite striking that different regions of the closely related insulin and insulin-like growth factor I receptor ectodomains confer specificity for their respective ligands despite very similar overall organization of their ligand binding domains (Gustafson and Rutter, 1990;Kjeldsen et al., 1991;Schumacher et al., 1991;Zhang and Roth, 1991a). The involvement of noncontiguous regions in ligand binding by a related receptor, that for epidermal growth factor, has also been reported (Wu et al., 1990;Lax et al., 1990Lax et al., , 1991Woltjer et al., 1992). Thus, this may represent an important aspect of transmembrane signaling by this class of hormone receptors.
Spectroscopic Analysis of the IR921 Protein-To explore further the organization of the ectodomain and to assess the extent of conformational changes that occur upon insulin binding, we undertook an analysis using CD and fluorescence spectroscopy. Both techniques are nondestructive, and they are sensitive indicators of changes in protein conformation (Lakowicz, 1986;Johnson, , 1990, although a specific structural interpretation cannot always be made. Additional insight into the spectroscopic properties of the IR921 protein can be gleaned by comparing results obtained in a similar analysis on a soluble form of the epidermal growth factor receptor (EGFR) ectodomain (Greenfield et al., 1989). Indeed, significant primary sequence similarity exists between these two ectodomains; based on conserved amino acid residues and secondary structure predictions, a model for the  and tenascin (TNfn3). Amino acid sequences were aligned by hand. The consensus sequence indicates residue identity among at least four of the five sequences. Dashes indicate any residue. An asterisk indicates conserved hydrophobic residues (I, L, or V), and 9 indicates conserved acidic residues. Gaps introduced to optimize the alignment are indicated by periods. ZRI, ZR2, GHRl, and GHR2 refer to the first and second type 111 repeats in each protein. IR1 includes residues N594-P654 and E794-A810 (IR residues are numbered according to Ebina et al., 1985). The vertical arrow between strands E and F indicates the insert (S655-1793) within IR1. IR2 includes residues A821-Y920. Overlined residues in IR1 and IR2 indicate amino acid identity between the human IR, the insulin-like growth factor I receptor, and the insulin receptor-related receptor (see Abbott et al., 1992). Assignment of @-strands A-F in TNfn3 and their alignment with those in the GHR (De Vos et al., 1992) is from Leahy et al. (1992). The alignment of IR and GHR sequences optimizes the conservation of cysteine residues. In GHR1, disulfide bonds are formed between the cysteine residues in strands A and B, C' and E, and F and G. In contrast, cysteine residues are absent in GHR2. By analogy with GHR1, the alignment suggests that the cysteine residues in strands F and G of IR1 and C' and E of IR2 are also disulfidelinked. extracellular domains of the insulin and the epidermal growth factor receptors has been proposed (Bajaj et al., 1987). This model predicts two independently folded subdomains, designated L1 (residues 1-119) and L2 (residues 311-428), on either side of the cysteine-rich domain (residues 155-312), which comprise -42% of the IR ectodomain (see Figs. 9 and 10). Organization of the remaining ectodomain sequence diverges considerably for these two receptors. In the EGFR the L2 domain is followed by a second copy of the cysteine-rich region, which terminates just prior to the transmembrane region. In contrast, the L2 domain of the IR is followed by a region (residues 429-593, comprising -18% of the ectodomain) that contains residues that are proximal to the insulin binding site (Fabry et al., 1992) and the epitope(s) for several monoclonal antibodies that inhibit insulin binding (Zhang and Roth, 1991b). This region is followed by the two fibronectin type I11 repeats (residues 594-920, comprising -35% of the ectodomain). In addition, the EGFR is a monomeric protein, whereas the IR is a heterotetramer (formed as a dimer of two cup precursor molecules). The potential role of these divergent organizations in transmembrane signaling by these two receptors is discussed further below. Fig. 6, A and B, illustrate the CD spectra (including noise) of the isolated ectodomain and insulin at the same concentrations used in the mixtures of these proteins (see below). The disparity in amplitude is a reflection of the large difference in their molecular masses. When replotted in terms of mean residue weight ellipticity, the spectrum of the IR921 protein (Fig. 7, A and B ) is smaller than that of insulin (data not shown). There is fairly high overall similarity between the ectodomain spectra of the IR and the EGFR in the far-UV ( Fig. 6A and Greenfield et al., 1989). The former is, however, slightly more intense, particularly in the positive band a t
Protein concentrations are as described in the legend to Fig. 6. approximately 190 nm. This difference may be partially due to differences in glycosylation, since oligosaccharide side chains on a protein contribute to negative ellipticity below 210 nm (Bush et al., 1980(Bush et al., ,1982. In the near-UV, the patterns observed are also very similar. However, the altogether negative ellipticity is again more pronounced for the IR than for the EGFR ectodomain, with more tryptophyl fine structure resolved around 295 nm (Fig. 6 B and Greenfield et al., 1989).
To explore whether formation of the insulin and ectodomain complex is accompanied by a detectable conformational change, we compared CD spectra of the IR921 protein in the absence (ED) and presence (ED11 and EDI2) of insulin. The corresponding spectra are overlaid for the 1:l and the 1:2 stoichiometry in Fig. 7A (far-UV) and Fig. 7B (near-UV). The consistent effect observed is that at wavelengths above 260 nm the absolute ellipticity of the IR921 protein in the complex is greater and below 260 nm it is smaller than that of the ectodomain alone. Although these spectral changes are relatively small, they suggest that conformational changes do occur upon binding of insulin to the IR ectodomain. Similar observations were reported for the EGFR ectodomain (Greenfield et al., 1989). The fact that changes in the spectrum are observed within the fine structure around 295 nm (Fig. 7 B ) , which is indicative of relatively well defined tryptophans (there are 30 Trp residues per ectodomain and fortuitously none in insulin), indicates that conformational changes occur in the ectodomain. Furthermore, were the observed spectral changes to be attributed solely to insulin (in contrast to Fig.   7, A and B, where they have been attributed solely to the ectodomain), an insulin spectrum results that is incompatible with our knowledge of the structural stability of this hormone (i.e. beyond the flexibility required for induced fit) (data not shown).

Steady-state Fluorescence Properties of the Ectodomain-
The fluorescence spectrum of the IR921 protein (using an excitation wavelength of 295 nm, to excite only tryptophan residues) peaks at 335 nm and has a FWHM of 58 nm (data not shown). According to the classification of Burstein et al. (1973), the 30 Trp residues in the IR921 protein are on the borderline between being buried in nonpolar regions (class I) and having a limited contact with water but being immobilized (class 11). The FWHM of 58 nm, instead of the expected 49-53 nm, indicates a nonhomogeneous environment of the Trp residues and thus probably that there are tryptophans belonging to class I and class 11. In contrast, Trp residues in the EGFR ectodomain all appear to be buried within the protein (Greenfield et al., 1989). Upon binding of the insulins there are slight changes in the fluorescence intensity, but neither a wavelength shift nor a change in the FWHM occur (data not shown). This is not surprising, since presumably only a small portion of the fluorophores are affected by insulin binding, especially because many of them are buried within the protein matrix.
We also examined the dynamics of fluorescence of Tyr residues (72 in the ectodomain and 4 in insulin) and Phe residues (76 in the ectodomain and 2 in insulin). To assess the existence of intra-and intermolecular energy transfer from Tyr or Phe to Trp, the fluorescence excitation spectra (using an emission wavelength of 369 nm) of the ectodomain in the presence and absence of insulin were divided by the absorption spectra of the corresponding samples as suggested by Saito et al. (1981). We observed that the transfer efficiency increased slightly upon addition of insulin (data not shown).
It cannot be determined whether this effect is due to the aromatic side chains of insulin coming close to Trp residues within the ectodomain or whether it is due to a reduction of donor-acceptor separation within the ectodomain as a consequence of insulin binding.
Fluorescence Lifetimes and Fluorescence Anisotropy Decay Measurements Reveal That Insulin Binding Increases the Rigidity of the Ectodomain-Extension of the steady-state experiments to time-resolved fluorescence can yield more information about the Trp environment and protein mobility. For these measurements an excitation wavelength of 300 nm was used to provide both a high limiting anisotropy ( r J and to avoid excitation of Tyr residues. The latter point is of special interest here because insulin does not contain any Trp residues, and hence the changes observed have to be ascribed to the ectodomain. Two representative examples of the timeresolved fluorescence measurements are given to illustrate the decay of the total fluorescence intensity (Fig. 8 A ) and the decay of the difference between the polarized fluorescence intensities (Fig. 8 B ) of the IR921 protein. All lifetime data are listed in Table I. The goodness of fit of the data is reflected in a uniform distribution of the residuals (see bottom of Fig.   8, A and B ) and a reduced Chi squared (xz) value near unity (Table I).
In all cases a sum of three exponentials was necessary to achieve a good fit with the total fluorescence intensity decay. However, the large number of tryptophans in the IR921 protein precludes the resolution of the fluorescence of individual Trp residues. The fluorescence decay of the free ectodomain is described by three lifetimes: 0.73, 3.2, and 7.0, with a mean lifetime, <T>, of 4.48 ns. This value for <T> further supports the notion that Trp residues within the ectodomain are immobilized (Kouyama et al., 1989). Upon addition of insulin, the two shorter lifetimes consistently decrease. In parallel, the amplitudes B1 and B2 (relating to these shorter lifetimes) tend to increase slightly at the expense of B3 (relating to the longer lifetime). Both of these effects reduce the mean fluorescence lifetime (<T>) and may indicate that   (Kouyama et al., 1989).
Another useful measure of fluorescence, which derives from the anisotropy decay data, is the rotational correlation time, 4 (for review, see Lakowicz, 1986). Rotational diffusion during the lifetime of the excited state displaces the emission dipole of the fluorophore, which results in a reduction in the measured anisotropy. 4 is a measure of this diffusion event. The IR921 protein is described by a 4 of 11.6 ns ( Table I); an attempt to analyze the data based on two rotational correlation times did not improve the x ' value. For proteins that are not simply spherical, Bucci and Steiner (1988) suggest that + be calculated on the basis of the Stokes radius (which we estimated by gel filtration; see above). With an RS of 6.8 nm, + is calculated to be 304 ns (see Equation 7). Since our measured value of + of 11.6 ns is much shorter, but longer than that of a freely mobile side chain (typically on the order of fractions of a nanosecond), it is indicative of segmental mobility within the IR921 protein. This hypothesis is further supported by the fact that only a small part ( r l ) of the total amplitude ( ro) follows the time course of the anisotropy decay ( i e . most of the tryptophans are fixed within the protein matrix according to class I and I1 tryptophans). As discussed earlier, the IR ectodomain and immunoglobulins appear to share certain structural and functional characteristics. Interestingly, Hanson et al. (1981) demonstrated that IgG is described by two rotational correlation times, both of which derive from flexible motions within the Fab region ( i e . both values are much shorter than + predicted for the overall tumbling of the molecule in solution). Moreover, this segmental flexibility appears to play a critical role in the biological function of the immunoglobulin molecule (Valentine and Green, 1967;Dudich et al., 1978;Hanson et al., 1981).
According to Lakowicz (1986), the anisotropy decay is described by Equation 9 when segmental mobility is present. Comparison of our estimated overall rotational correlation time for the IR921 protein (+* = 304 ns) with its mean fluorescence lifetime (<T> = 4.48 ns) clearly shows that the tumbling of the ectodomain as a whole can be neglected and, hence, +* + 00 in Equation 9. The fraction of the total depolarization which is lost by the segmental mobility can be determined as a = rJr0 (Lakowicz, 1986). However, attempts t o accurately estimate the segmental mass are complicated due to the large number of Trp residues in the ectodomain and the fact that they reside in somewhat different environments ( i e . classes I and 11). Nonetheless, it is of interest to estimate the segmental mass of the ectodomain in the ideal case ( i e . assuming that the segment can wobble freely and that all 30 Trp side chains have equal mobility and fluorescence intensity). Under these conditions, the wobbling segment comprises 15% of the total ectodomain. According to the calculations of Wegener et al. (1980), flexible attachment of a small mass to a larger one reduces its diffusion coefficient by a factor of three. Therefore, + of the segment has to be divided by three to estimate its mass with Equations 7 and 8. The resulting value is 11 kDa, which corresponds to only 4% of the ectodomain. The two different estimates of the segmental mass (i.e. 11 uers'sus 42 kDa) could be explained if the motion originates from -four nearly equivalent segments. In this regard, it is possible that the mobile segments are located within the L1 and L2 domains (Bajaj et al., 1987), or alternatively, within the fibronectin type I11 repeats (see Figs. 9 and 10).
We next examined the effect of insulin binding on the observed segmental flexibility of the ectodomain. Quite surprisingly, we found that + for the IR921 protein increases dramatically (from 11.6 to 170 ns) when the first insulin is bound and it exceeds the time range of the measurement upon binding of the second insulin (Table I). This indicates that the segmental mobility is successively frozen as a result of insulin binding, since aggregation of the ectodomain in the presence or absence of insulin, as assessed by gel filtration and NMR spectroscopy,' is not observed. Thus in contrast to the relatively small changes observed with CD, the changes upon insulin binding observed by fluorescence anisotropy decay measurements are quite large. Furthermore, since insulin does not contain any Trp residues, the observed confor-  Bajaj et al. (1987); Cys, cysteine-rich region; III, fibronectin type I11 repeats. Note that the first type I11 repeat is interrupted by an insert that contains the a-@ cleavage site and an alternatively spliced exon (number 11). The numbers below the uertical hatchmarks indicate exons (Seino et al. (1989). The numbers at the bottom of the diagram indicate amino acid residues in the primary sequence of the a@ proreceptor. a, asubunit; @, @-subunit; &,, extracellular portion of the @-subunit.
Insulin FIG. 10. Cartoon of the ectodomain and its interaction with insulin. The dotted pattern corresponds to the cysteine-rich region. See the legend to Fig. 9 for further description. Note that insulin contact sites have been mapped to both L1 and L2 (for discussion see Fabry et al., 1992). mational changes must occur within the ectodomain. The fact that these changes in fluorescence become more pronounced upon addition of the second insulin suggests that both binding sites on the ectodomain become occupied with ligand. This conclusion supports the recent stoichiometry measurements of Markussen et al. (1991). Photoaffinity labeling studies with insulin (Wedekind et al., 1989;Fabry et al., 1992) demonstrate that the hormone binds proximal to the L1 and L2 domains and thus suggest that the arms of the Y may well contribute to the segmental flexibility that is lost upon ligand binding (see Figs. 3 and 10). Analysis of smaller truncated derivatives of the ectodomain should help to localize the regions that contribute to the segmental flexibility.
It is interesting that the IR ectodomain differs from the EGFR ectodomain in that the latter is much more flexible (as apparent from r, and rl/ro). Moreover, the binding of EGF to its ectodomain causes a reduction in + (from 6.2 to 2.5 ns) for the complex (Greenfield et al., 1989), suggesting that there is an increase in mobility or energy transfer (Leenders et al., 1990). The diverging properties of the IR and the EGFR ectodomains (i.e. heterotetramer uersus monomer, respectively, and the organization of the carboxyl-terminal region) presumably contribute to the opposing nature of the conformational changes (measured by fluorescence spectroscopy) that occur upon ligand binding. The different behavior of the ectodomains upon ligand binding may reflect the use of different mechanisms to facilitate transmembrane signaling.

SUMMARY
Given that the IR921 protein is stable, is efficiently processed and secreted and binds insulin with wild-type affinity, we have developed a heterologous cell expression system by which this protein can be obtained in milligram amounts. The availability of milligrams of purified ectodomain has facilitated both biochemical and biophysical analyses, including the identification of insulin contact sites (via photoaffinity labeling techniques ;Fabry et al., 1992) and the elucidation of the overall shape and organization of this complex polypeptide (via rotary shadowing and negative staining electron microscopy). In addition, the inherent mobility of the ectodomain and the nature and extent of conformational changes induced by insulin binding have been characterized using spectroscopic methods.
A number of valuable landmarks provided by motifs deduced within the primary sequence of the receptor protein (i.e. L1, cysteine-rich, L2, fibronectin type I11 repeats), together with the exon-intron boundaries of the IR gene (Seino et al., 1989;see Fig. 9 and the cartoon of Fig. lo), provide a framework for a new generation of molecular genetic, biochemical, and biophysical studies designed to explore further the modular (or subdomain) organization of the ectodomain and to delineate the minimum sequence required for highaffinity insulin binding. Given the size and complexity of the ectodomain protein, the identification of a minimal ligandreceptor complex is also expected to facilitate the generation of crystals suitable for the elucidation at high resolution of the three-dimensional structure of the ectodomain-hormone complex by x-ray crystallography.