Assembly of the gigantic hemoglobin of the earthworm Lumbricus terrestris. Roles of subunit equilibria, non-globin linker chains, and valence of the heme iron.

The extracellular hemoglobin of the earthworm Lumbricus terrestris has four major kinds of O2-binding chains: a, b, and c (forming a disulfide-linked trimer), and chain d. Non-heme, non-globin structural chains, “linkers,” are also present. Light-scattering techniques have been used to show that the ferrous CO-saturated abc trimer and chain d form an (abcd)4 complex of 285 kDa at neutral pH. Formation of the full-sized 4-MDa molecule requires the addition of linker chains in the proportion of two linkers per (abcd)4 and occurs much more rapidly in the presence of 10 mM calcium. This stoichiometry is supported not only by direct quantitative analysis of the intact hemoglobin but also by the fact that the addition of 50% of the proposed stoichiometric quantity of linkers results in the conversion of 50% of the (abcd)4 to full-sized molecules. Isolated CO-saturated abc trimers self-associate to (abc)2 and higher aggregates up to an apparent limit of (abc)10 ∼550 kDa. The CO-saturated chain d forms dimers, (d)2, and tetramers, (d)4. Oxidation of the (abcd)4 complex with ferricyanide causes complete dissociation of chain d from the abc trimer, but addition of CN− maintains the (abcd)4 complex. Valence hybrids have also been studied. The ferrous CO-saturated abc trimer and met (ferric) chain d also associate to form (abcd)4, but the met abc trimer and ferrous CO-saturated chain d do not. Oxidation of the abc trimer and chain d to the ferric form causes the formation of a characteristic hemichrome spectrum with a maximum at 565 nm and a shoulder near 530 nm. These results show that interactions between the abc trimer and chain d are strongly dependent on the ligand and valence state of the heme iron. Light-scattering measurements reveal that oxidation of the intact Hb produces a significant drop in molecular mass from 4.1 to 3.6 MDa. Inclusion of CN− prevents this drop. These experiments indicate that oxidation causes the Hb to shed subunits. The observations provide an explanation for the wide variations in the molecular mass of L. terrestris Hb that have been observed previously.

The extracellular hemoglobin of the earthworm Lumbricus terrestris has four major kinds of O 2 -binding chains: a, b, and c (forming a disulfide-linked trimer), and chain d. Non-heme, non-globin structural chains, "linkers," are also present. Light-scattering techniques have been used to show that the ferrous CO-saturated abc trimer and chain d form an (abcd) 4 4 and occurs much more rapidly in the presence of 10 mM calcium. This stoichiometry is supported not only by direct quantitative analysis of the intact hemoglobin but also by the fact that the addition of 50% of the proposed stoichiometric quantity of linkers results in the conversion of 50% of the (abcd) 4 to full-sized molecules. Isolated COsaturated abc trimers self-associate to (abc) 2 and higher aggregates up to an apparent limit of (abc) 10 ϳ550 kDa. The CO-saturated chain d forms dimers, (d) 2 , and tetramers, (d) 4 . Oxidation of the (abcd) 4 complex with ferricyanide causes complete dissociation of chain d from the abc trimer, but addition of CN ؊ maintains the (abcd) 4 complex. Valence hybrids have also been studied. The ferrous CO-saturated abc trimer and met (ferric) chain d also associate to form (abcd) 4 , but the met abc trimer and ferrous CO-saturated chain d do not. Oxidation of the abc trimer and chain d to the ferric form causes the formation of a characteristic hemichrome spectrum with a maximum at 565 nm and a shoulder near 530 nm. These results show that interactions between the abc trimer and chain d are strongly dependent on the ligand and valence state of the heme iron. Light-scattering measurements reveal that oxidation of the intact Hb produces a significant drop in molecular mass from 4.1 to 3.6 MDa. Inclusion of CN ؊ prevents this drop. These experiments indicate that oxidation causes the Hb to shed subunits. The observations provide an explanation for the wide variations in the molecular mass of L. terrestris Hb that have been observed previously.

complex of 285 kDa at neutral pH. Formation of the full-sized 4-MDa molecule requires the addition of linker chains in the proportion of two linkers per (abcd)
The extracellular hemoglobins of annelids were first shown by Svedberg and Eriksson (1) to be gigantic molecules of at least 3 MDa in molecular mass. Although several subsequent studies gave molecular masses close to 4 MDa or higher for the Hb of Lumbricus terrestris and related species (2)(3)(4), recent studies by scanning transmission electron microscopy (5) have suggested masses of 3.5-3.6 MDa, and early light-scattering measurements suggested even lower values (6,7). A possible reason for this great variation in reported molecular masses is oxidation which has been shown to cause extensive dissociation of many annelid Hbs (8 -10). Goss et al. (9) used light scattering to show that oxidation of L. terrestris Hb at neutral pH caused a large drop in molecular weight. Similarly, Ascoli et al. (10) found that oxidation of the similar Hb from Octolasium complanatum 1 caused hemichrome formation and dissociation outside of a narrow range near pH 7.
The stoichiometry has also been uncertain. The Hb of L. terrestris has four major kinds of O 2 -binding globin chains: a, b, and c (forming the disulfide-linked abc trimer) and chain d, together with non-heme structural chains, "linkers," designated L. Kapp et al. (11) and Vinogradov et al. (12) have reported that linkers comprise 33-36% of the total mass, but this conclusion was based largely on the staining of SDS-gels with Coomassie Blue. This dye, however, binds to different proteins to quite different extents (13,14). Ownby et al. (15) redetermined the stoichiometry by reverse-phase high performance liquid chromatography (HPLC). 2 The weight proportions of linkers were found to be approximately 16.4% by two independent procedures: the integrated absorption of the HPLC peaks and amino acid analysis of these peaks. These results led to the conclusion that the overall stoichiometry is (abcd) 2 L. This stoichiometry is further supported by the results of SDS-capillary gel electrophoresis monitored at 214 nm (see our companion study (16)).
The goals of the present experiments are to redetermine the molecular mass by light scattering, to examine the equilibria between subunits, to investigate the effects of oxidation, and to examine the role of the linker chains in assembly. Vinogradov et al. (12) were the first to conclude that these chains are required for assembly. This conclusion was based on the finding that preparations of the abc trimer and chain d subunits phase HPLC (15). The molar ratios of trimer to linker and chain d to linker were found to be Ͼ52 and ϳ30, respectively, in terms of 220 nm absorbance for the trimer and chain d preparations. The linker chain fraction contained 0.16 trimers and 0.72 chains d per linker. The maximal amount of full-sized molecules that could be generated from these constituents of the linker chain fraction is 0.08 moles of (abcd) 2 L/mole of linker. These quantities were allowed for in the reassembly experiments (Fig. 13, a and b). The absorption spectrum of the CO-saturated abc trimer in this preparation was indistinguishable from that of freshly prepared HbCO, indicating that no measurable oxidation had occurred during storage in (NH 4 ) 2 SO 4 .
Oxidation-The HbCO (preparation from (NH 4 ) 2 SO 4 ) was first converted to HbO 2 by exposure to a strong light ("SunGun," Sylvania) at 0°C under an atmosphere of O 2 for 30 s, repeated 10 times. This procedure was also used to convert the abc trimer and chain d to the oxy derivatives. Oxidized Hb (or subunits) was made by adding various amounts of K 3 Fe(CN) 6 according to the experiment. Two sets of experiments were done: (i) 100% oxidized Hb was produced by adding K 3 Fe(CN) 6 (25-or 125-fold molar excess over the heme) in the dark 3 on ice overnight against N 2 -saturated 25 mM bis-tris propane, 100 mM NaCl, and 1 or 5 mM EGTA, pH 7.0. The K 3 Fe(CN) 6 was added in the same buffer. Light-scattering experiments were performed on these solutions with the size-exclusion column equilibrated with the same buffer. (ii) In the second set of experiments, HbO 2 (freshly prepared, unfrozen) with and without calcium was prepared with Ca 2ϩ , 25 mM bis-tris propane, 100 mM NaCl, 10 mM CaCl 2 , pH 7.0, or without Ca 2ϩ , in 5 mM sodium phosphate, 100 mM NaCl, 1 mM EDTA, pH 7.0. The latter preparation without calcium is the same solution used for the dry weight determinations (16). Aliquots of each of these solutions were oxidized in the dark by adding K 3 Fe(CN) 6 as 5, 10, 25, 50, 75, 90, or 100% of the molar quantity of heme-iron. The actual percent oxidation was estimated from the change in absorbance. The end point of the absorbance change was calculated approximately from the ratio between the extinction coefficients for human HbO 2 and metHb at 576 nm because slow transformation of the L. terrestris metHb to the hemichrome precluded direct determination of the end point. Free ferro-and ferricyanide were removed by dialysis against the same buffer with the NaCl increased to 300 mM. Although this procedure should also remove bound ferrocyanide if the binding is similar to that in human Hb (21), this was not tested.
Light Scattering-A DAWN model F multiangle light-scattering photometer (Wyatt Technology, Santa Barbara, CA) with a 5-milliwatt He-Ne vertically polarized laser of wavelength 632.8 nm was used in all experiments. The construction and principles of operation of this instrument together with the associated interferometric refractometer have been described in detail (22). The photometer is equipped with a K5 flow cell (Wyatt) that has a 67-l total fluid volume in which the scattering volume is Ϸ 0.5 l. All measurements were made in the in-line flow mode. The laser light beam is collimated through the center of the flow path, and the fluid flow is in the same direction as the laser. The windows of each end of the flow cell are recessed, and the edges of the flow cell are away from the Ϸ 0.5-l scattering volume so that stray light from the air/glass/solvent interfaces is not seen by the detectors. High gain photodiodes are placed at 18 refracted angles. A Beckman Altex model 110A HPLC pump with a Beckman model 210 injection valve was used to deliver samples through a switching valve (catalog no. V240, Upchurch) either directly to the photometer or first through a Bio-Gel SEC 40XL column (300 ϫ 7.8 mm, Bio-Rad catalog no. 125-0604). Experiments shown in Fig. 13, a and b, used a Waters "Protein-Pak" 300 SW column (300 ϫ 7.8 mm, catalog no. 800B). Injection loops of 20 l, 180 l, and 1 ml were used, depending on the experiment. Connections were made with Upchurch "PEEK" 0.01-inch internal diameter tubing. The flow rate for all experiments was 0.3 ml/min. A model LP-21 "Low-Pulse" pulse damper (Scientific Systems, Inc., State College, PA) was inserted after the pump. An Optilab-903 interferometric refractometer (Wyatt Technology, Santa Barbara, CA) was connected in-line following the light-scattering photometer. The P2 sample cell used in the refractometer has a path length of 0.2 mm. A Hitachi model 100-10 uv-visible spectrophotometer with flow cell in the fluid line follows the refractometer.
Photometer Calibration-Calibration was performed as described in the Wyatt manual. The relationship between molecular weight and light-scattering is given (23,24) by the equation, (n 0 is the refractive index of the solvent, dn/dc, is the refractive index increment, N is Avogadro's number, and is the wavelength in vacuo) and c is the weight concentration of protein. B is the second virial coefficient. R is the angle-dependent Rayleigh ratio, given by R ϭ I r 2 /I 0 V where I is the intensity of scattered light at angle ; I 0 , the incident light intensity; V, the volume of scattering solution; and r, the distance from this volume to the detector. P() Ϫ1 is given by the Zimm equation (23,24).
This is a limiting expression, strictly valid only as approaches zero (24). R G is the radius of gyration. Here is the wavelength in vacuo and n 0 is the refractive index of the solvent. All light-scattering data are linearly extrapolated to ϭ 0; an example is shown in Fig. 1. Calibration of the photometer is achieved by using toluene with known Rayleigh ratio of 1.406 ϫ 10 Ϫ5 cm Ϫ1 at the wavelength of the laser to relate the R 90 value (R at 90°) to the detector voltage.
The configuration-specific calibration constant, A CSCC , is solvent-specific. An instrument constant, A inst , dependent only on geometrical factors, is derived from A CSCC , as described in the Wyatt manual (Astra Program, Version 2.11). Once A inst is known, values of A CSCC in other solvents can be calculated from known refractive indices of solvent and glass. Calibration of the other detectors was achieved by the use of an isotropic scatterer that scatters light equally in all directions. Although the initial normalization was performed with methanol, later measurements utilized the averaged results from bovine serum albumin and carbonic anhydrase. No significant difference in values for molecular weights was found.
Refractometer Calibration and dn/dc Determination-The Wyatt Optilab model 903 interferometric refractometer measures the difference between the refractive indices of sample and reference cells. The refractometer has the following features (22). Collimated white light polarized at 45°is split by a Wollaston prism into vertically polarized sample and horizontally polarized reference beams. Recombination of these beams by a second Wollaston prism followed by a quarter-wave plate and analyzer filter gives a plane polarized beam rotated with respect to the initial 45°. The rotational phase difference is proportional to the refractive index difference between solvent and solution. An interference filter with maximal transmission at 630 nm is placed in front of the detector. A three-way valve makes possible fluid flow through both sample and reference cells (flush) or only through the sample cell. The cells are thoroughly flushed at a flow rate of 0.3 ml/min with deionized "nanapure" HPLC-grade water (resistance ϭ 17.9 megohm) filtered through a 0.2-m nylon filter (Schleicher & Schuell, catalog no. 00040). The valve is then turned so that fluid flows only through the sample cell. The temperature is maintained at 25°C for all experiments with water from an external bath. Calibration provides the difference in refractive index between sample and reference cells per volt from the signal detector. Both NaCl and sucrose were used as calibrants. NaCl (Aldrich, catalog no. 20 -4439), 99.999% pure, was found to give satisfactory results with the value of the refractive index increment, dn/dc, taken to be 0.1741 ml/g at 25°C and 589 nm (25). This value was used in all calculations. Interpolation of the data of Kruis (26) yields a slightly higher value, 0.1744 ml/g at 632.8 nm and 25°C. The NaCl (Aldrich) was dried at 180°C overnight, weighed, and dissolved in a 1-liter volumetric flask with nanapure water passed through a 0.2-m filter. All dilutions were made with great care using a single calibrated Pipetman. Calibration and dn/dc measurements were made either "offline" directly into the refractometer by siphon or through the HPLC system but bypassing the column. For the latter procedure, a 1-ml injection loop was used with a flow rate of 0.3 ml/min. Sucrose (Baker, analytical reagent, catalog no. 4072-01) with a dn/dc value of 0.1422 ml/g (25) was used as a second standard. The calibration constants obtained from NaCl and from sucrose agreed to within 0.3%. The calibrating solutions were not filtered because filtration was found to give poor, non-reproducible results. Recalibration with sucrose one year later gave an identical result.
Determination of the molecular mass by light scattering is critically dependent on the accurate determination of dn/dc. This was measured for L. terrestris Hb as follows. The last dialysate used for the dry weight and heme determination (see our companion study (16)) was also used to obtain dn/dc. Dilutions were made by weight from the stock solution (0.1642 mM heme). The dry weight determination (16) resulted in a dn/dc value of 0.185 ml/g. An identical value was obtained by assuming the (abcd) 2 L stoichiometry previously determined (15,16) together with the measured heme concentration of 0.1642 mM (16). The weight concentration c i in each slice in the elution pattern was calculated from the refractive index difference and the value of dn/dc, n i Ϫ n 0 ϭ (dn/dc)c i .
The refraction per gram of amino acids has been determined by McMeekin et al. (27). These values, multiplied by the partial specific volumes (28) for each amino acid, should provide an estimate of the value of dn/dc for each amino acid. Values of dn/dc for the polypeptides of L. terrestris Hb, calculated on this basis, together with the weightaverage value for the intact Hb, are given in Table I. We have assumed that the subunits of L. terrestris Hb have the same dn/dc value found for the whole molecule. Although partial specific volumes calculated from amino acid compositions have been widely used in ultracentrifugation, calculation of dn/dc for proteins from their amino acid composition remains to be tested for its possible general applicability.
Interdetector Volume-The volume, V i , between the scattering volume of the photometer and the refractometer cell must be determined so that the light scattering and refractive index data can be properly matched. If the V i value is too large, it will produce a positive slope in a plot of molecular mass versus elution volume for a monodisperse substance, and if too small, the slope will be negative (30). A plot of this slope against various values of V i in one set of experiments with test proteins produced a straight line; the intercept corresponding to a zero slope gave a V i volume of 91 l at a flow rate of 0.3 ml/min. Alternatively, the data from the photometer and refractometer were matched with program "Align" (Wyatt Astra Program Version 2.11). Small changes in flow rate produce changes in the time interval required for a given element of fluid to reach the sample cell of the refractometer from the photometer. This time difference will appear as a change in apparent interdetector volume.
Band Broadening-Some spreading of the chromatographic peak occurs between the light-scattering volume in the flow cell and the refractometer (31). The effect of this dispersion for a sharp peak is to decrease the concentration at the peak maximum below that "seen" in the light-scattering cell so that the calculated molecular weight is FIG. 1. An example of the extrapolation of Kc/R to zero angle. The data shown correspond to a slice at 6.6 ml in the data for L. terrestris HbCO shown in Fig. 3. The extrapolated value of Kc/R corresponds to a molecular mass of 4.06 MDa and the slope of the line shown yields an R G value of 12.3 nm.
slightly too high. Similarly, down the sides of a sharp peak, the concentration increases so that the calculated molecular weight is too low. The overall effect is a small "frown" in the plot of the calculated molecular weight versus elution volume. However, if the weight-average molecular weight is calculated for an entire chromatographic peak for a monodisperse protein, these effects cancel out if the value of the second virial coefficient, B, is zero or if the concentration is sufficiently low so that the Bc term can be neglected. This can be shown as follows. The light scattering Equation 1 gives the relation between the angle-dependent Rayleigh ratio and molecular weight.
The function P() in Equation 1 is omitted here because all data are extrapolated to zero angle. We assume B ϭ 0. The actual concentration determined in the refractometer, c i Ј, differs slightly from c i in the photometer, leading to the calculation of the wrong molecular weight, M i Ј.
Evidently, however, c i ЈM i Ј ϭ c i M i , and the total quantity of protein seen in the photometer and refractometer is the same. Therefore, the weightaverage molecular weight for the entire peak will be This shows that the band-broadening error cancels out if the weightaverage molecular weight is calculated for the entire peak or half-peak for a monodisperse substance at sufficiently low concentrations. However, the light scattering data for most of the test proteins indicate some self association. Use of the trailing half-peak appears to avoid complications from self-association in these proteins (see Table II). A P2 refractometer cell (Wyatt) was used in all of the experiments described in this report. The P10 cell (Wyatt) has been developed to eliminate or greatly reduce secondary band broadening by ensuring complete mixing so that a uniform front enters the cell. Test Proteins-Seven different proteins were examined by light scattering. Five of these were obtained from Sigma and were used without further purification: bovine carbonic anhydrase (catalog no. C7025), yeast alcohol dehydrogenase (catalog no. A8656), sweet potato ␤-amylase (catalog no. A8781), horse spleen apoferritin (catalog no. A3660), and bovine thyroglobulin (catalog no. T9145). Each of these was prepared as ϳ1-2 mg in 1 ml in 50 mM phosphate buffer, pH 6.5-6.6, 1 mM EGTA, and dialyzed overnight at 0°C against 500 -1,000 ml of this buffer. The sample was centrifuged briefly, filtered through a 0.1-m syringe filter (Whatman, no. 6789-0489-0401, nylon), and injected into a 180-l loop of the injection valve of the HPLC system for lightscattering measurements. Filtration caused, however, a variable loss, and therefore centrifugation without filtration was used when the concentration dependence of the light scattering was measured. The dialysis buffer was filtered through a 0.2-m nylon filter (Schleicher & Schuell, no. 00040) and used as a systems buffer for the HPLC column and light scattering. Preparations of carbonic anhydrase and ␤-amylase were also prepared with the same buffer made 100 mM in NaCl. The light scattering results were identical. Ornithine decarboxylase from Lactobacillus sp. strain 30a (32, 33), a gift from Dr. Marvin L. Hackert, was prepared similarly. The concentration of the decarboxylase was determined with an extinction coefficient, E 280 nm 0.1% ϭ 1.34 (32). A ϳ1.6 mg/ml solution in 100 mM phosphate, pH 6.3, 0.5 mM EDTA, was dialyzed overnight at 0°C against the same buffer made 1 mM in freshly prepared DTT just before use. A dn/dc value of 0.190 cm 3 /g was assumed for all of the above proteins. Hemocyanin from Octopus dofleini, a gift from Dr. Karen Miller, was dialyzed at 0°C against 100 mM Tris-HCl, pH 7.6 (at room temperature), 50 mM MgCl 2 prior to use. The dn/dc value was taken to be 0.194 cm 3 /g, a value obtained for several hemocyanins from other species of octopus (34).
Electron Microsopy and Analysis-The protein was diluted to 10 g/ml in 0.25% methylamine tungstate stain and applied by the spray method to the Butvar side of a carbon-coated Butvar film (35). The images were recorded with a JOEL JEM 1200 electron microscope at ϫ 50,000 and 100 kV using conventional irradiation procedures at an under focus of 0.5 m. The micrographs were digitized as 4096 ϫ 4096 ϫ 12 bit arrays using an Eikonix 412 digitizer at a pixel size of 5.7 Å/pixel. The digitally generated power spectrum of the micrographs was utilized to check for drift and astigmatism.
The molecules were in random orientations, and those that were judged to represent 6-and 2-fold projections of the structure were selected for image analysis. Image analyses were performed as de-   (27) combined with the partial specific volumes (28) and the stoichiometric results (16). Both sets of data were determined at 25°C, the same temperature used for L. terrestris Hb. The calculated dn/dc values do not include contributions of the heme and carbohydrate constituents.
scribed previously (36) using Silicon Graphics 4D25 and Indigo work stations and our SUPRIM image-processing software (37).

Molecular Weight Test Proteins
We have evaluated the accuracy of the light scattering methodology with seven test proteins. Fig. 2 shows the 90°lightscattering and refractive index patterns for these proteins. The results of the light-scattering measurements are summarized in Table II.
Carbonic Anhydrase-The molecular weights calculated for the left and right half-peaks ( Fig. 2a) are 33,900 and 32,300 (only the segment to 10.3 ml was used). These values suggest some self-association. Extrapolation of the concentration dependence to c ϭ 0 yielded a molecular weight close to the structural value. The perturbation at the end of the right (trailing) peak suggests a small difference between column buffer (dialysate) and the buffer of the sample. Prolonged (48 h) dialysis against the elution buffer did not eliminate the irregularity in the baseline. However, the perturbation disappeared in a later experiment in which the sample was centrifuged and not filtered.
Yeast Alcohol Dehydrogenase (Fig. 2b)-Substantial nonideality is present in the light-scattering results for alcohol dehyrogenase. A plot of 1/M app versus total quantity in peak was linear and extrapolated to a mass within 1.4% of the structurally determined value. It is curious that the alcohol dehyrogenase migrates in SDS gel electrophoresis much closer to egg albumin (45 kDa) than to glyceraldehyde-3-phosphate dehydrogenase (36 kDa) (data not shown). Nevertheless, the expected molecular mass of the alcohol dehydrogenase in this sample was determined by mass spectrometry (see footnote d of Table II).
Sweet Potato ␤-Amylase-The right-half peak (Fig. 2c) yielded a value of 228 kDa, which is within 1.6% of the structurally derived value.
Horse Spleen Apoferritin-The calculated weight-average molecular weight for the right-half of the major peak ( Fig. 2d), 484,400, is only 1.7% higher than the value expected for a homopolymer of the L-chain of this protein with 24 subunits (44) of molecular weight 19,846.5 (GenBank TM ) (19,846.5 ϫ 24 ϭ 476,316). This result, obtained at a weight-average concentration of only 0.06 mg/ml, indicates that the protein is relatively stable to dissociation, in agreement with prior studies by sedimentation equilibrium (51) and by light scattering (52). The chromatogram (Fig. 2d) shows a minor component of apparent mass 1.15 MDa that is probably a dimer. Horse spleen apoferritin has two types of chain, H (ϳ10%, ϳ21 kDa), and L (ϳ90%, ϳ19 kDa), that form heteropolymers (53). The weight-average molecular mass increases across the major peak (Fig. 2d, right to left) from 484 kDa (right, trailing half) to ϳ540 kDa (left, leading half), as expected from the composition. The amino acid sequence of the H-chain of horse spleen apoferritin has not been determined, but its mass is about 10.5% greater than that of the L-chain. On this basis, the expected molecular mass of the H-chain homopolymer would be 526 kDa (24 ϫ 1.105 ϫ 19847), close to the 540 kDa observed for the left, leading half-peak.
Bovine Thyroglobulin-A shoulder on the leading edge of the peak (Fig. 2e, arrow) suggests some self association). The weight-average molecular weight for the right-half peak, 679,000, is within 1.5% of the 669,000 value obtained by sedimentation (45).
Ornithine Decarboxylase-This dodecameric enzyme has a subunit molecular weight of 82,557 (33) and a total molecular weight of 990,685 (Fig. 2f). Guirard and Snell (32) obtained a molecular mass of 1.04 MDa by sedimentation equilibrium and observed that omission of DTT from the buffer caused variable aggregation to 1.5-3.5 MDa. The inital light-scattering data a All values are based on amino sequences except that for thyroglobulin. b Symbols: R, right (trailing) half of chromatographic peak used; L, left (leading) half of chromatographic peak used; C ϭ 0, data extrapolated to zero concentration. The value of dn/dc was taken to be 0.190 cm 3 /g for all light-scattering measurements of test proteins except octopus hemocyanin for which the measured value of 0.194 cm 3 /g was used (34 gave 1.16 MDa for the right half-peak in the presence of 1 mM DTT (prepared from an old stock solution), suggesting some aggregation. When the experiment was repeated with freshly prepared DTT, the molecular weight for the right-half peak was 978,000, within 1.3% of the structurally determined value.
Octopus Hemocyanin-This O 2 -carrying copper protein is a decamer with a molecular mass of 3.48 MDa measured by sedimentation equilibrium (50) and a mass of 3.44 MDa from amino acid sequence analysis 4 and determination of a carbohydrate content (50). The light-scattering measurements (Fig.  2g, Table II) are in close agreement with these values.
Measurements made with these diverse proteins, summarized in Table II, demonstrate that light-scattering is capable of yielding molecular weights within 2% of those obtained by structural analysis. These results were obtained by assigning the same dn/dc value, 0.190 cm 3 /g, to six of the proteins. An accuracy better than 1% should be achievable with a careful dn/dc determination for each protein combined with measurement of the concentration dependence of the light-scattering and an analysis of the pH-dependent deviations from ideality.

Hemoglobin of L. terrestris
Molecular Weight Fig. 3a shows the 90°light-scattering and refractive index difference patterns for a dilute solution of freshly prepared L. terrestris HbCO. The calculated weight-average molecular mass from these data is shown as a function of elution volume in Fig. 3b. The calculated weight-average molecular mass for the entire peak is 4.10 Ϯ 0.1 MDa on the basis of the measured dn/dc value of 0.185 ml/g. These measurements (Fig. 3) were made at sufficiently low concentrations (Յ84 g/ml) so that correction for the second virial coefficient could be ignored. The value of B (Equation 1) has been estimated to be only 5 ϫ 10 Ϫ9 liter-mol/g 2 (7). This would change our value of the molecular mass by less than 0.3%. Similarly, light-scattering measurements of Hb from the related earthworm (54), O. complanatum 1 , in the range 25-250 g/ml showed no concentration dependence.
The angular dependence of scattering (see Fig. 1) has been used to estimate the radius of gyration, R G , by use of equations (1) and (2) for the same experimental data used for Fig. 3. This yielded a value of 12.7 Ϯ 2.8 nm (n ϭ 22) for R G which is 13% higher than the more precise value of 11.2 Ϯ 0.1 nm determined by low angle x-ray scattering (55). Somewhat higher accuracy may be possible with careful redetermination of the normalization factors for each detector, particularly at low angles (see "Materials and Methods") performed at the same time as experiments such as those shown in Fig. 3. However, the R G value is only 1.8% of the wavelength of the scattering light, and the slope of the line in Fig. 1 is very small so that we are near the experimental limit.
We have compared L. terrestris HbCO samples freshly prepared and frozen with those quickly frozen in liquid N 2 or slowly frozen to Ϫ80°C. The weight-average molecular masses obtained after each of these treatments were indistinguishable:  3. The molecular weight of the CO-hemoglobin of L. terrestris. a, the light-scattering data for 90°and the refractive index values at pH 6.8 for the chromatographic peak; b, the calculated apparent molecular weight across the chromatographic peak. The weightaverage molecular weight for the protein is calculated from the data of the entire peak as described in the text. The protein concentration at the maximum of the peak is ϳ84 g/ml. Buffer: 25 mM bis-tris propane, 100 mM NaCl, 10 mM CaCl 2 . MDa (Ϫ80°C frozen). It may well be that freezing causes damage induced by ice crystals that is not reflected in the total molecular mass. We suspect, however, that previously observed dissociation induced by freezing may be due primarily to oxidation of HbO 2 that readily occurs at high freezing temperatures of Ϫ10°C to Ϫ20°C.
Results very similar to those shown in Fig. 3 for HbCO were obtained for HbO 2 (data not shown). The HbCO was converted to HbO 2 with light and an atmosphere of O 2 at 0°C. metHb was prepared from HbO 2 by oxidation with an excess of K 3 Fe(CN) 6 . Spectral changes that accompany partial oxidation with K 3 Fe(CN) 6 are shown in Fig. 4. Table III summarizes the light-scattering results following storage of the HbO 2 and metHb solutions at 4°C for varying lengths of time. These results show that oxidation causes a decrease of 10 -30% in molecular mass. The lowest mass for metHb, 3.0 MDa, was obtained after prior storage for 9 days at 4°C as HbO 2 . However, no decrease in molecular mass was observed in CN-metHb.
Subjecting Hb to prolonged storage as HbO 2 and then oxidizing it (see Table III) causes a change in the pattern of components seen in reverse phase HPLC: a new peak designated "pseudotrimer" (PT) can be seen that elutes earlier than the abc trimer (Fig. 5). This peak has been identified as trimerderived by reduction with DTT and isolation of chains b, c, and a, together with numerous additional peaks absent from the HPLC profile of the DTT-treated trimer peak (data not shown). The pseudotrimer is markedly heterogeneous and has a variable mass of 40 -45 kDa by SDS-gel electrophoresis and by mass spectrometry (15). No pseudotrimer was detected in any HbCO preparation even after prolonged storage, and none was detected in metHb stored for one day. The lowered molecular mass of the pseudotrimer indicates proteolysis. Proteolytic activity was investigated as follows. HbCO in 25 mM bis-tris propane, 100 mM NaCl, 10 mM CaCl 2 , pH 7.0, was increased to pH 9.5 by 1:3 dilution with the same buffer adjusted to pH 10.5 and containing 50 g/ml leupeptin, 50 g/ml pepstatin A, and 250 M phenylmethylsulfonyl fluoride. Oxidized Hb was made by converting HbCO to HbO 2 with light followed by oxidation with K 3 Fe(CN) 6 at 16°C and then incubated for various times (Table III). No pseudotrimer could be detected with HbCO at either pH 7.0 or 9.5 for at least 19 h with or without the protease inhibitors. However, more pseudotrimer was found with oxidized Hb incubated only 5 min at 16°C at pH 9.5 in the absence of protease inhibitors than in 75 min in the presence of inhibitors (data not shown).
The observations, summarized in Table III, show that the drop in molecular mass from 4.1 to 3.6 MDa upon oxidation is associated with a dissociation of subunits independent of proteolysis. However, dissociation of the Hb exposes new sites for attack by proteases. Molecular masses of 3.7 MDa or less were often associated with proteolysis when the storage of the metHb was prolonged.
The apparent rough correlation between time "as HbO 2 " and quantity of pseudotrimer (Table III) is puzzling. The HbO 2 stored 10 days (Table III) is the same sample from which aliquots were taken for oxidation. Absorption spectra of the HbO 2 stored for various times did not suggest any oxidation. One interpretation is that storage as HbO 2 does result in the cleavage of a few susceptible bonds but that this is insufficient by itself to promote dissociation. However, upon oxidation to metHb, additional proteolysis would be rapid. In any event, fully ligated ferrous Hb is resistant to dissociation.
These observations help to explain the wide variation in molecular mass observed in different laboratories. We have been able to reproduce this variation by oxidizing the Hb. These data support the hypothesis that the variation in molecular mass results from the oxidation-dependent shedding of subunits. Fig. 6a shows the voltage signals from the detectors for 90°light-scattering and the refractometer as a function of elution volume. The calculated weight-average molecu-  lar mass (weight) versus elution volume is given in Fig. 6b. These data show that the ferrous CO-derivative of the abc trimer self-associates extensively at pH 6.8. Although the abc trimer elutes in what appears to be a single peak with a shoulder, the peak is far from monodisperse. The apparent weight-average molecular mass decreases smoothly across the peak from left to right, from ϳ250 kDa to ϳ100 kDa. More than 75% of the abc trimer associates to (abc) 2 (ϳ109 kDa) and larger products. The apparent weight-average molecular mass for the peak between 7 and 9 ml is 206 kDa. A distinct component which elutes at 7.0 -7.5 ml has an apparent mass of 570 Ϯ 22 kDa is within 5% of the mass of (abc) 10 . This appears to be the upper limit to self-association of the trimer. The full-sized Hb eluting at 6 -7 ml can be accounted for by traces of linkers and chain d in the preparation. The data suggest the following self-association equilibrium.

Subunits abc Trimer-
abc º ͑abc͒ 2 º ͑abc͒ 2n º ͑abc͒ 10 abc ASSOCIATION The trailing edge of the abc peak in Fig. 6b suggests that substantial dissociation of (abc) 2 occurs only at concentrations below 0.1 mg/ml. This is confirmed in Fig. 7a which shows the overall weight-average molecular mass of the abc trimer as a function of concentration and pH. The abc trimer associates reversibly to (abc) 2 or larger products at all pH values at concentrations above 0.1-0.2 mg/ml. The trimer associates only to (abc) 2 at pH 9 even at a concentration of 1.8 mg/ml. These results indicate that the abc-abc interface in (abc) 2 is not pH-dependent. The pH dependence of abc trimer association (Fig. 7b) shows an apparent maximal mass of (abc) 10 near ϳpH 6 with half-dissociation at about pH 6.8. However, the data for pH 6.0 (Fig. 7a) extend only to 0.6 mg/ml because higher concentrations saturate the detectors so that the maximal value for abc association cannot be accurately assessed. Nevertheless, the 570 kDa shoulder in Fig. 6b is consistent with a maximum of (abc) 10 in Fig. 7b. Oxidation of the abc trimer (Fig. 8) results in the dissociation of all species larger than (abc) 2 . The only significant association of the met trimer is 2abc [ (abc) 2 . However, ligation with CN Ϫ causes reassociation to larger products. The resulting elution pattern (7.5-8.5 ml in Fig. 8) is similar to that for the oxy abc trimer shown in Fig. 6b. These results indicate that oxidation FIG. 6. The data for CO-derivative of the abc trimer. a, the light-scattering and refractive index data. The voltage scale is for the light scattering; the refractive index data have been scaled to make the comparison clearer. The peaks at 5.0 and 10.7 ml are artifacts in the buffer. b, the distribution of the calculated molecular weight values. Buffer: 25 mM bis-tris propane, 100 mM NaCl, pH 6.8. Protein concentration at the refractive index maximum is 0.12 mg/ml.

FIG. 7.
The concentration dependence of self-association of the CO-derivative of the abc trimer at various pH values. a, the concentration dependence; b, the pH dependence of the weight-average molecular weight at a protein concentration of 0.6 mg/ml. The ordinate is expressed as multiples of the molecular mass of the abc trimer (54.7 kDa). Buffer: 50 mM bis-tris propane, 100 mM NaCl, and either 1 mM EGTA or 25 mM CaCl 2 . These experiments were performed as follows: (a) dialysis against CO-saturated buffer, (b) dilution with dialysis buffer to ϳ2 mg/ml protein, (c) injection of sample at highest concentration, (d) collection of main peak, (e) reconcentration with Centricon 30, and (f) dilution to appropriate concentration. The size-exclusion column was bypassed, and the 1.0 ml injection volume was sufficient to produce a plateau at each concentration. of the heme iron destabilizes the (abc) 2 -(abc) 2 interfaces but has relatively little or no effect on the (abc)-(abc) interface. The association properties of the CN-Met and CO derivatives of the abc trimer are very similar.
Chain d-The CO-derivative of chain d, like the CO-abc trimer, self-associates. The puzzling elution pattern (Fig. 9) indicates that the predominant species must be larger than (d) 2 . The two d components (56), d 1 and d 2 , are treated here together as (d). The same pattern is observed with CN Ϫ met chain d (data not shown). The apparent weight-average molecular mass for the entire double peak between 8.5 and 10.2 ml is ϳ66 kDa, suggesting that (d) 4 is a major species. Our interpretation of the pattern is based on the finding that monomeric CO-chain d occurs only at extreme dilution. Even at pH 9.0, chain d is dimeric (data not shown). This finding suggests that the self-association of chain d involves dimers as the basic unit. The leading (left) half-peak between 8.6 and 8.95 ml gives an apparent weight-average molecular mass of 78 kDa. Although this is close to the mass of (d) 5 , 83 kDa, it could also reflect the process (d) 4 ϩ (d) 2 [ (d) 6 . The smoothly-decreasing molecular weight between 8.95 and 9.4 ml can be interpreted in terms of the dimerization, 2 (d) 2 [ (d) 4 . Unligated ferric chain d does not associate beyond the dimer (d) 2 , but ligation with CN Ϫ results in further self association to (d) 4

(data not shown).
Quantitative Interpretation of Chromatographic Elution Patterns-Much more information is contained in the elution patterns shown in Figs. 6, 8, and 9 than we have attempted to extract. It is important to recognize that these patterns represent transport, not equilibrium, processes. The protein concentrations reflected in the refractive index differences of individual fractions ("slices") do not, in general, correspond to the equilibrium concentrations associated with the weight-average molecular masses determined for each slice. In order to extract dissociation parameters from the patterns, it would be necessary to perform an analysis similar to that we have used for the sedimentation velocity profiles of the associating deoxy Hbs of chicken and frog (57,58). The procedures for the analysis of elution patterns from gel chromatography are examined in detail by Ackers (59) and Cann (60).
abcd Complex-If equimolar amounts of the CO-abc trimer and CO-chain d are mixed at pH 6.8, a well defined complex forms (Fig. 10) with a mass near 280 kDa, corresponding to (abcd) 4 whose molecular mass is 286 kDa. If this mixture is injected onto the size-exclusion column within 1 min. of mixing, the complex elutes in less than 30 min at ϳ8 ml. No further change in pattern occurs after the mixture is incubated overnight at 0°C. The complex also forms at pH 9.0, although the weight-average molecular mass is about 7% lower than that obtained at pH 6.8, indicating some dissociation (data not shown). This result explains why the chromatographic separation of trimer, linkers and chain d at pH 9 requires two passes through a size-exclusion column. A single pass never completely separates the components. The results of experiments with oxidized subunits are given in Table IV. These data show that oxidized abc trimer and chain d do not combine with each other at all: no (abcd) 4 is formed and the oxidized (abc) ϩ and (d) ϩ subunits self-associate only to (abc) 2 ϩ and (d) 2 ϩ . However, when the ferric hemes of an equimolar mixture of (abc) 2 ϩ and (d) 2 ϩ are ligated with CN Ϫ , the abcd complex does form.
A small quantity of linkers is always present in preparations of the abcd complex. It is important to note that HPLC analysis of the fractions depicted in Fig. 10 shows that 100% of these contaminating linkers are sequestered into 4-MDa molecules, and none is left in the fractions of lower molecular mass. This means, as described below, that all linkers are competent to participate in reassembly.
Valence Hybrids-We have also prepared a series of valence hybrids. Addition of ferric (d) ϩ to the CO-(abc) trimer causes the immediate formation of an (abcd) 4 complex. Thus, oxidation of chain d, by itself, does not inhibit the formation of the (abcd) 4 complex. These results are summarized in Table IV. In contrast, no (abcd) 4 forms in the reverse experiment: (abc) ϩ mixed with CO-d. One caveat of these experiments is that the light-scattering measurements must be performed immediately after oxidation and mixing. If the mixed samples are stored 2.5-3.5 days, then a small amount of (abcd) 4 is observed from the mixture of (abc) ϩ and CO-d. We believe that this effect may result from redistribution of oxidation states and CO. Such a redistribution would be aided by the action of CO as a reducing agent (17).
Absorption Spectra of Oxidized Subunits-Oxidation of the abc trimer and chain d with ferricyanide causes the appearance of a characteristic hemichrome spectrum (A in Fig. 11) with maxima at 412 and 532-534 nm and a shoulder at ϳ565 nm. Reduction with dithionite (B in Fig. 11) results in a spectrum with maxima at 428 and 559 nm and a shoulder at 530 nm. Similar results were obtained with chain d (data not shown). The spectra are very similar to those described for hemichrome and hemochrome formation in human Hb (61) and in that of the related annelid, O. complanatum (10). Hemichrome formation in the subunits of L. terrestris Hb appears to be characteristic of the dissociated subunits.
Similar observations have been made in several unrelated invertebrate Hbs in three different phyla: the unligated metHbs of Phoronopsis viridis (62), Scapharca inaequivaluis (63), and Caudina arenicola (64) are monomeric but become dimeric as CN-metHb. This suggests that a common mechanism may exist in L. terrestris Hb and all these Hbs that links dimers with the CN-met form and the monomers in the absence of CN Ϫ . The common mechanism might involve a similar rela-tionship between a critical subunit interface and the heme iron in each of these Hbs.

Recombination of L. terrestris Hb from Trimer, Chain d, and Linkers-
We seek to answer the following questions: (i) How does the yield of recombined Hb vary with the amount of linker chains added? (ii) How does the yield vary with time of incubation? (iii) What is the effect of Ca 2ϩ on the recombination process? Three sets of experiments have been performed to address these questions: Set no. 1: The first experiments were carried out by (a) mixing equimolar quantities of CO-abc trimer and CO-chain d to form (abcd) 4 , (b) adding various amounts of linkers at pH 6.5, then (c) raising the pH to 9.0, followed by (d) dialysis back to pH 7.0 overnight; all steps were performed at 4°C. The CO-saturated buffer at each pH was 50 mM bis tris propane, 100 mM NaCl and 1 mM EGTA. The rationale for the last two steps was that the linkers are known to self-associate, and it was believed that this association might retard formation of the 4-MDa Hb if dissociation of the linkers were a rate-limiting step. Initial dissociation at pH 9 might therefore remove this complication.
The results of adding 50% of the stoichiometric quantity of linkers to the (abcd) complex are shown in Fig. 12a. The measurements were made approximately 18 h after the start of the dialysis. Two features of the weight-average molecular mass distribution are noteworthy: 1) 100% of the linkers added are incorporated into large structures of which more than 90% exceed 3.5 MDa in mass. The material between 5.9 and 8.1 ml was collected and subjected to HPLC analysis. The results were consistent with the values expected on the basis of the stoichiometry already determined (16) (data not shown). This demonstrates that all linkers are competent for reassembly. 2) The weight-average molecular mass decreases in two phases; a gradual decrease from 4 to 2 MDa between 6.0 and 8.2 ml, and a precipitous drop from 2 MDa to ϳ280 kDa between 8.3 and 9.0 ml. 3) A small but definite component comprising about 4% of the total protein, corresponding to half-molecules of ϳ2 MDa occurs at 8.0 -8.2 ml. The concentration of material between 280 kDa and 2 MDa is extremely low, indicating a highly cooperative process of reassembly. We have assumed here that the rate at which equilibrium is reached is slow compared with the time for the chromatography.
Although the addition of 50% of the stoichiometric quantity of linkers to CO-saturated (abcd) 4 yielded approximately 50% of the reassembled molecules, an increase of linkers to 100% increased the yield only to 75-80% of the molecules with mass greater than 3 MDa (see Fig. 12b). The data, however, reveal that all linkers added to (abcd) 4 up to 50% of the stoichiometric FIG. 10. The elution pattern of the (abcd) 4 complex formed by mixing equimolar amounts of CO-abc and CO-d. The peaks at 6.5 and 7.6 ml correspond to 4-MDa and 550-kDa molecules that result, respectively, from the presence of a small quantity of linker chains and a slight excess of the abc trimer. Buffer as in Fig. 6. Protein concentration at maximum refractive index difference (at 8.05 ml), 0.17 mg/ml. a The symbols abc ϩ and d ϩ mean the completely oxidized ferric forms. b "1 min, 25°C" refers to the time after mixing and before injection onto the column. The (abcd) 4 complex elutes in about 27 min. c The variation in the mass of the (abcd) complex is not significant. The mean of the three values obtained after incubation overnight at 0°C is 287 kDa, essentially identical with the mass of (abcd) 4 , 286 kDa. amount are competent to participate in assembly. We conclude that the partial inability to reassemble rests solely with the abc trimer and/or chain d. If the time of exposure of dissociated subunits (trimer, chain d and linkers) to high pH is extended to a week, no reassembly can be detected within 24 h.
Set no. 2: The second set of experiments was similar to the first except that a pH of 6.5 was maintained throughout. Various quantities of linkers were added to the trimer and chain d mixture (all in the CO-form), and the final mixture was incubated at 0°C overnight and up to a month or longer. HPLC analysis was used to determine the subunit composition of each component of the mixture. No calcium was added in these experiments. The results, summarized in Fig. 13, a and b, shows that two distinct kinetic phases are present. When 200% of the stoichiometric quantity of linkers are added, the yield of 4-MDa Hb reaches 50% within 16.5 h, but 71 days are required to reach 78%. These observations reveal that a primary limitation on recombination is kinetic.
Comparison of Figs. 12 and 13 shows that the subunits used in the Fig. 13 experiments reassociated to a much lesser extent than in the experiments shown in Fig. 12. This difference presumably resulted from the use of freshly prepared Hb and subunits in the Fig. 12 experiments and the use of Hb long stored in (NH 4 ) 2 SO 4 in the experiments shown in Fig. 13. Set no. 3: This set of experiments examines the effect of calcium on reassembly. A dramatic increase in the rate of formation of 4-MDa Hb occurs in the presence of added Ca 2ϩ . In one experiment in 25 mM bis tris propane, 100 mM NaCl, pH 6.8, the apparent half-time for the recombination was 30 days. The addition of 10 mM Ca 2ϩ reduced the time to only 5.3 days, corresponding to a 6-fold increase in rate (data not shown). These experiments were done at 0°C to avoid oxidation known to occur at higher temperatures where the rates of recombination would presumably be much faster.
Electron Microscopy-The shape and design of the average images of the 6-and 2-fold projections of the native and reconstituted complex appear identical (Fig. 14). Correspondence analysis followed by hierarchical ascendent classification was applied to the data for each average image. The analysis revealed no significant differences in the cluster averages so that we reasonably conclude that the grand averages are representative of the data sets. These average images are similar to the corresponding average images of the hemoglobin derived from frozen, hydrated material (65). However, it was apparent that the reconstituted preparation consisted of numerous particles that appeared to represent incompletely-formed structures (data not shown). These particles were not chosen for image analysis. The excellent concordance between the two orthogonal projections of the native and reconstituted structures indicates that the subunits have self-assembled to form the native complex. The preparation used is similar to that for the point in Fig. 12b with 100% of the stoichiometric quantity of linkers added and was not fractionated. DISCUSSION Our understanding of the assembly of L. terrestris Hb rests on four sets of observations: 1) the stoichiometry of globin and linker polypeptides: eight globin chains per linker; 2) the determination of the molecular mass of CO-saturated native Hb, close to 4.1 MDa; 3) the finding that the ferrous CO-saturated abc trimer and chain d combine to form (abcd) 4 ; and 4) the stoichiometric titration of CO-saturated (abcd) 4 with linkers to form 4.1-MDa molecules.

Stoichiometric Considerations
Our analysis (16) shows that the weight proportion of linkers is 17% in contrast to the prior reports by Vinogradov and associates (see Ref. 16 and references therein) that the linkers comprise ϳ33% of the total mass. They also propose (66) that the primary structural unit is (abcd) 3 which has a mass of ϳ216 kDa. The consequences of these different stoichiometries are summarized in Table V 4 complex. a, the elution pattern is shown that resulted from the addition of 50% of the stoichiometric quantity of linkers to the CO-saturated (abcd) 4 complex. Approximate protein concentration corresponding to maximum refractive index difference at ϳ6 ml, 0.11 mg/ml. b, the percent recombination as a function of the quantity of linkers added. The point corresponding to the addition of 50% linkers is derived from the data in a. Buffer: 50 mM bis-tris propane, 100 mM NaCl, 1 mM EGTA, pH 6.5.

5-MDa molecules have never been observed and our results
show that a 3-MDa value is only found after oxidation, subunit dissociation and proteolysis. Stoichiometry 1, favored by Vinogradov and colleagues (66), is inconsistent with our determination of the mass of the complex and our measurement of the globin/linker proportions. We conclude that only stoichiometry 4 (17% linkers) and a principal subunit of (abcd) 4 is consistent with the observed molecular mass of 4.1 MDa.

Molecular Mass
The accuracy of the molecular mass determined by light scattering is critically dependent on both the determination of dn/dc and the concentration dependence of the apparent molecular mass. The close correspondence between dry weight and amino acid analysis (16) provides strong support for our value of dn/dc ϭ 0.185 cm 3 /g. Herskovits and Harrington (7) reported a value of 0.193 cm 3 /g but their preparations were not protected from oxidation by saturation with CO. They estimated the protein concentration with a previously determined extinction coefficient (E 1% ϭ 5.95) at 540 nm for HbO 2 (54). The presence of only 8% metHb would be sufficient to reduce the calculated protein content enough to raise the value of dn/dc to 0.193 cm 3 /g because the extinction coefficient of metHb at 540 nm is only half that of HbO 2 (67). Their value of ϳ3.1 MDa for the molecular mass suggests that oxidation has occurred. We can reproduce this value, but only under conditions of oxidation and often proteolysis. The recently reported lower molecular masses of ϳ3.5-3.6 MDa for L. terrestris Hb (5) can be repro-duced by oxidation alone without proteolysis. Although oxidation of the (abcd) 4 complex is accompanied by complete dissociation to (abc) 2 ϩ and (d) 2 ϩ , oxidation of all chains is not required for this dissociation because combination of oxidized chain (d) 2 ϩ with CO-saturated abc trimer still produces (abcd) 4 molecules, whereas the combination of oxidized abc trimer with CO-saturated chain d does not occur (see Table IV).
We suggest that oxidation of only a small fraction of the total The scale bar corresponds to 100 Å, and the gray scale corresponds to high protein density (white) to low protein density (dark). The 6-and 2-fold average images consisted of 368, 186 and 617, 419, respectively, for the native and reconstituted structures, respectively, and their corresponding resolutions were 31, 45, 29, and 42 Å (77). No symmetry has been imposed on the images. The same starting components were used as described in the experiments in Fig. 12. Trimer, chain d, and linkers were combined in 50 mM bis tris propane, 25 mM CaCl 2 , pH 6.5, and then dialyzed overnight against the same buffer with pH increased to 8.8 followed by dialysis for 48 h against the same buffer adjusted to pH 7. All buffers were COsaturated. The native Hb was freshly prepared. The recombined material corresponds to the point at 100% linkers in Fig. 12b. The material used for electron microscopy was not size fractionated. number of hemes is sufficient to cause dissociation. If one chain, for example a or b, forms crucial intersubunit contacts, oxidation of this critical chain would weaken the contacts and cause dissociation. If so, then oxidation of 25% of the total number of hemes might be enough to cause 100% dissociation. Oxidation of 3.8% of the hemes would then suffice to give 15% dissociation and thereby reduce the molecular mass from 4.1 to ϳ3.5 MDa. Furthermore, the effect of oxidation on dissociation of subunits is likely to be cooperative. Just as O 2 binding by the 192 hemes in the Hb is highly cooperative (68), we expect that oxidation has a similar cooperative effect just as it does in human Hb. Preliminary measurements 5 support this suggestion: partial oxidation of only a few percent results in a disproportionately greater dissociation of the Hb. We assume that those interfaces critical for O 2 binding are the same as those involved in oxidation-dependent dissociation. If the cooperative mechanism is also similar, it would mean that the number of oxidized hemes necessary for 15% dissociation may well be much less than 3.8%. Such small levels of oxidation would be hard to detect and could easily be overlooked.
We do not know which chain(s) might be critical for oxidation-induced dissociation. However, isolated chains d and c are relatively resistant to autoxidation whereas chains a and b autoxidize rapidly (16,18). If, as seems likely, similar differences in autoxidation also exist in the intact Hb, chains a and/or b would be prime candidates for the critical chain.
Extent of Recombination-The experiments summarized in Fig. 13, a and b, reveal that a fraction of dissociated subunits reassociates only very slowly. We have shown that the subunits responsible for the slow reassociation are the abc trimer and/or chain d (see "Recombination of L. terrestris Hb from Trimer, Chain d, and Linkers," "Set no. 1"). A possible explanation for the slow rate of reassociation is provided by the concept of conformational drift whereby dissociated subunits of oligomeric proteins, freed of conformational constraints, often appear to change to new conformations and return to the native state very slowly (69 -71). This idea has been fruitful in the analysis of pressure-induced dissociation of subunits of a similar Hb from the worm, Glossocolex paulistus (72).
Recent measurements 6 of the dissociation resulting from oxidation of the Hb and its reassociation upon the addition of CN Ϫ , N 3 Ϫ , or NO 2 Ϫ show that the reassembly of subunits occurs within the time required to add the ligand, mix and perform a light scattering experiment (Ͻ25 min). These experiments indicate that slow reassociation of L. terrestris Hb is characteristic only of subunits that have been dissociated for long periods of time (Ͼ48 h). Our observations are reminiscent of the report by David and Daniel (2) that dissociated subunits of the Hb of a related annelid lost their ability to reassemble after 24 h.
The close correspondence between the images of native and reconstituted hemoglobin (Fig. 14) is consistent with the molecular weight values for reassembled Hb and with the analytical results that indicate that the reconstituted Hb has the same stoichiometry of subunits.

Model of Assembly
A striking feature of the present experiments is that the subunit association processes are strongly dependent on both the ligand and valence state of the iron. The results show that processes at the heme are strongly linked to conformational changes at the interfaces responsible for association of the subunits. This conclusion applies to all globin subunits and their major association products. For all these processes, the presence of ferric subunits is dissociating, and addition of CN Ϫ maintains the associated state. This finding implies, as discussed, that a common mechanism exists that links heme ligand and valence state to the free energy of critical intersubunit contacts which seem likely to be similar in (abc) 2n and (d) 4 . Our understanding of the assembly process rests on the properties of subunits maintained in the ferrous state by saturation with CO with the exclusion of oxygen during storage. We propose the following four steps in the assembly process of CO-saturated ferrous subunits: Step 1-Chain d and abc trimer self-associate to dimers and higher aggregates. Rationale: Both chain d and the abc trimer have a strong tendency to form dimers and higher aggregates when the hemes are ferrous, and CO-saturated (see Figs. 6 -9). Fig. 7a shows that the self-association of abc is reversible because sufficient dilution causes a corresponding dissociation.
Step 2-Chain d and abc trimer combine to form (abcd) 4 .

STEP 2
Rationale: Light-scattering measurements demonstrate that an equimolar mixture of the abc trimer and chain d forms a tight ϳ280 kDa complex that corresponds to the mass of (abcd) 4 (Fig. 10). The accuracy of this determination is strongly supported by the test protein data. The ϳ280-kDa mass of the (abcd) complex lies between those of sweet potato ␤-amylase (224 kDa) and horse spleen apoferritin (484 kDa) for each of which the mass determined by light-scattering is within 1.7% of the structurally derived value (see Table II). The O 2 equilibrium of (abcd) 4 at pH 6.8 is highly cooperative and is identical to that of the intact Hb (18). This identity would presumably occur only if the specific subunit interfaces responsible for cooperativity were also identical to those in the intact Hb. Earlier sedimentation velocity measurements (18) had suggested that the abcd complex might be much smaller than (abcd) 4 , but that was before the drastic effects of oxidation were recognized. The sedimentation coefficient obtained, s 20,w ϭ 5.6 -5.7 S is the value expected for an equimolar mixture of (abc) 2 and (d) 2 after oxidation.
Rationale: Linker chains, like the abc and d subunits, selfassociate. Although a stoichiometric combination of (abcd) 4 and L 2 seems likely, no 340 kDa (abcd) 4 L 2 molecules have been isolated. Any such molecules formed are believed to be removed by a highly cooperative further assembly (see Fig. 12a). 5 H. Zhu and A. F. Riggs, unpublished observations. 6 E. Karpova, C. K. Riggs, and A. F. Riggs, unpublished observations.
Step 4 -The (abcd) 4 L 2 molecules self-associate with high cooperativity to form the complete molecule. 6͑abcd͒ 4 L 2 3 ͑͑abcd͒ 4 L 2 ͒ 6 2͑͑abcd͒ 4 L 2 ) 6 3 ͑͑abcd͒ 4 L 2 ͒ 12 STEP 4 Rationale: The final assembly to 4-MDa molecules from (abcd) 4 and linkers is demonstrated to be highly cooperative by the extremely low concentrations of any intermediates evident in the elution pattern of Fig. 12a The distribution of molecular weights in Fig. 12a suggests that the final self-assembly occurs in two steps. The first entails the highly-cooperative formation of a ring structure corresponding to the half-molecule, ((abcd) 4 L 2 ) 6 , of molecular mass ϳ2 MDa. This half-molecule corresponds to one layer of the hexagonal structure (see Fig.  14). The second step would be a non-cooperative but tight combination of the two half-molecules.
Calcium greatly accelerates the formation of 4-MDa molecules, but this observation does not explain how this effect is mediated. It may be either that Ca 2ϩ accelerates the formation of intermediate species or that the primary effect of Ca 2ϩ is to stabilize the Hb molecule once it is formed. The latter possibility is suggested by the observation that calcium stabilizes the Hb against dissociation at high pH. Thus it remains to be determined whether Ca 2ϩ accelerates any step in assembly or slows the dissociation process. It is possible that Ca 2ϩ affects both processes.
An essential feature of this model is the (abcd) 4 complex. We assume that this structure retains its integrity within the intact Hb. However, it is likely that the conformation of the isolated (abcd) 4 is perturbed to some extent upon binding linkers and incorporation into the larger structure. This may account for the fact that the (abcd) 4 complex has an oxygen equilibrium identical with that of the intact Hb at pH 6.8 but that this match gradually disappears as the pH is raised (18). One reason this may occur is that high pH destabilizes the (abcd) 4 complex more than the intact Hb.
The Dodecamer Model-An alternative model for the Hb of L. terrestris has been advanced by Vinogradov and colleagues (see Refs. 5, 66, and 73, and references therein). They propose that the molecule has 12 (abcd) 3 dodecamer subunits and 36 linkers. This model is based on (a) the lower heme content observed by them, (b) a lower molecular mass of 3.5 Ϯ 0.1 MDa, and (c) the isolation of an apparent dodecamer of 216 kDa from a sizeexclusion column of products dissociated by 4 M urea. We evaluate their measurement of heme content in our companion study (16). Here we make the following observations on their conclusions concerning the lower molecular mass and the dodecamer.
First, comparison of the absorption spectrum 7 of the HbO 2 in Fig. 5A of Martin et al. (5) with our data for HbO 2 freshly prepared from HbCO (Fig. 4, this report) reveals that approximately 30% of their "HbO 2 " is oxidized. This means that their preparation cannot be completely homogeneous. Our finding that oxidation causes a decrease in molecular mass means that the partially oxidized Hb studied by Martin et al. must have a lower weight-average molecular mass than it would have with all its iron atoms ferrous.
Second, Fig. 5B of Martin et al. (5) shows that treatment with ferricyanide causes dissociation. However, the authors attribute the dissociation not to oxidation but to the assumed binding of ferro-or ferricyanide to the Hb. This conclusion was reached because treatment with a 330-fold excess of nitrite oxidizes the Hb without a drop in molecular mass. However, oxidation with nitrite gives aquo-metHb only when Hb is treated with an equimolar amount of nitrite (74). Nitrite is a weak ligand (74,75), and a 330-fold excess should saturate the Hb. We have found that the effects of the ligands, CN Ϫ , N 3 Ϫ , and NO 2 Ϫ , are all very similar. 6 Oxidation with ferricyanide or nitrite will produce low molecular weights, and this can be reversed and prevented with any of these reagents. These experiments show that the binding of ferro-or ferricyanide is not responsible for the observed decrease in molecular weight.
Third, sedimentation equilibrium measurements made on three preparations of HbO 2 by Martin et al. (5) yielded values for the molecular mass of 3.03, 3.25, 8 and 3.95 MDa. Such a variation of 30% indicates an uncontrolled variable which we believe to be oxidation-induced subunit dissociation. These values were obtained with the use of a partial specific volume (V) value of 0.714 cm 3 /g determined with the highly accurate Paar DMA-60 density meter. However, the value of V depends critically on the determination of the concentration of protein.
Oxidation and resultant heme-loss will change any weightbased extinction coefficients at 540 nm for CN-metHb used for these calculations. The net effect will be to increase the apparent concentration so that V is reduced. They also calculated V ϭ 0.734 cm 3 /g from the amino acid compositions of the constituent chains. This value yielded masses of 3.26, 3.50, and 4.25 MDa, the last value being close to that which we have obtained for the mass of completely ferrous HbCO by light scattering.
Fourth, the dodecamer subunit was prepared by "mild" dissociation of HbO 2 with 4 M urea (66,73). However, as discussed above, the starting material is clearly partially oxidized. Our observations show that all subunits (except d) are much more sensitive to oxidation than the original Hb, so the dodecamer is likely to be oxidized to a greater extent. We have found no evidence for a ferrous (abcd) 3 complex of ϳ200 kDa. Our experiments show that oxidation of the abcd complex causes complete separation of the abc trimer from chain d at pH 6.8. Furthermore, the data of Herskovits and Harrington (7) on the effects of 4 M urea on L. terrestris Hb suggest that 5-9% of the Hb may become denatured by this treatment. Because some of the "dodecamer" is oxidized and therefore dissociated, the preparation cannot be completely homogeneous and must contain both free abc trimer and chain d. This conclusion is consistent with the rather broad distribution of molecular masses observed for the dodecamer by scanning transmission electron microscopy (66,76).
Fifth, the composition of the crystals of the putative dodecamer was examined by SDS-polyacrylamide gel electrophoresis and found to contain both abc trimer and chain d (73). However, this preparation must be non-homogeneous because of the dissociation of abc and d as a result of oxidation. The extent to which the crystals were washed with the protein-free crystallizing solution is unclear. The observation would be uninterpretable without meticulous exhaustive washing.
Sixth, Sharma et al. (76) have assumed that high concentrations of guanidinium salts, urea, or heteropolytungstates are required to dissociate the Hb at neutral pH to produce the presumed dodecamer and other products. The dissociation and subsequent reassociation upon removal of the agent were followed by HPLC chromatography with the fitting of the resulting patterns to a series of modified Gaussian distributions. Some of the experiments involved treatment with 8 M urea in which oxidation, denaturation and heme loss occur rapidly. We believe that the heterogeneous kinetics observed for dissociation under these and similar conditions may have resulted from the combined effects of dissociation caused by oxidation and heme loss as well as partial denaturation by treatment with the dissociating agent. Although they concluded that oxidationinduced dissociation is far too slow to be significant, our data clearly show that it can occur rapidly.
We conclude from this analysis that the presence of an (abcd) 3 dodecamer as a principal subunit of L. terrestris Hb is not well supported.