A β_H-spectrin self-interaction domain forms antiparallel dimers that may facilitate non-canonical spectrin membrane skeleton assembly

Abstract

Tetrameric spectrin proteins crosslink filamentous-actin (F-actin) into planar membrane-associated networks known as spectrin-based membrane skeletons (SBMS). The importance of tetramer formation in SBMS formation is amply demonstrated by mutations in human αΙ-spectrin that break the head to head (H2H) association leading to hereditary elliptocytosis (HE). Tetramer formation requires antiparallel αβ-spectrin or αβ_Η-spectrin dimerization via dimer nucleation sites, followed by (αβ)₂ or (αβ_H)₂ tetramer formation via the H2H association of two dimers. The presence of a singular N-terminal F-actin binding site in each β chain makes tetramer formation obligatory for F-actin crosslinking in all structurally defined SBMS. We previously showed that the α-spec^R22S mutation in Drosophila (homologous to the human αI^R28S HE allele), eliminates H2H binding in both conventional (αβ)₂ and heavy (αβ_H)₂ spectrin isoforms in vitro. However, α-spec^R22S mutant flies remain surprisingly unaffected. In (αβ)₂ tetramers, the β chains do not overlap; however, β_H chains are much longer and overlap in an (αβ_H)₂ tetramer, raising the possibility of direct β_H-β_H binding that could ameliorate the α-spec^R22S mutation. Here we use a combination of non-denaturing gel electrophoresis, dissociation isothermal titration calorimetry (ITC), small angle X-ray scattering (SAXS), and structural mass spectroscopy to identify and characterize a β_H self-interaction domain (βHSID) in the overlapping β_H segments 25 to 31 of Drosophila β_H. The βHSID interaction is antiparallel and has a K_d = 187-213 nm at 25 °C, similar to that of the H2H interaction in human (αΙβΙ)₂, making the βHSID physiologically relevant. This interaction suggests that (β_H)₂ dimers may crosslink F-actin to form non-canonical SBMS.

Spectrin-based membrane skeletons (SBMS) are ubiquitous and essential cytoskeletal elements in metazoan cells that regulate key mechanical properties of the plasma membrane (1). These cortical protein networks consist of (αβ)₂-spectrin tetramers that crosslink F-actin to form a network that has been best characterized in erythrocytes (2, 3). In general, cortical SBMS are anchored to the cell membrane by binding directly or indirectly to integral membrane proteins (4, 5). Indirect binding via highly alternatively spliced adapter proteins such as Ankyrin results in interaction with numerous membrane proteins (6). Such interactions play important roles not only in recruiting SBMS to the membrane, but also in stabilizing integral membrane proteins (7, 8, 9). In erythrocytes, the SBMS determines cell shape, while providing membrane resilience and flexibility during circulation (10, 11, 12). In non-erythroid tissues, genetic analysis has revealed a multitude of roles for SBMS in cell structure, neuronal transmission, morphogenesis, protein trafficking, DNA repair, and cancer (13, 14).

In vertebrates, diversity in the SBMS arises from the expression of two α and five β isoforms (15) that are encoded by distinct genes, with further diversity arising through alternative splicing (16). Two distinct classes of β isoforms have been found, conventional β-spectrins and heavy β_H-spectrins (β_H), both of which can form F-actin-crosslinking tetramers with an α chain (Fig. 1; (15)). In Drosophila, there is one gene encoding α-spectrin and one gene encoding each class of β-spectrin, providing a tractable model for dissecting SBMS assembly and function.

Figure 1.

Key interchain interactions permitting F-actin crosslinking by (αβ)₂ and (αβ_H)₂ spectrins and the consequences of the α-spec^R22S mutation.A, β-spectrins (grey filled chain) bind to F-actin through segment β1, a CH1/2 F-actin binding domain (blue box). (αβ)₂ dimers form through antiparallel lateral interactions between four spectrin repeats (segments β2,3 and α20,21; (94)). (αβ)₂ tetramers form between two dimers via the head-to-head (H2H) interaction between segments β18 and α0 (red; (67)) and can crosslink F-actin. The α^R22S mutation disrupts segment α0, preventing the H2H (36), red cell R28S). Since the overlapping segments of α-spectrin do not appear to self-interact (67), the α^R22S mutation prevents F-actin crosslinking. B, β_H-spectrin (grey filled chain) binds to F-actin through segment βH1, also a CH1/2 F-actin binding domain (blue box). The interchain dimer and tetramer interchain binding sites in αβ_H are conserved, indicating dimer formation is via segments βH2,3 and α20,21 and tetramerizing via segments βH32 and α0 (15, 37) to crosslink F-actin. The α-spec^R22S mutation disrupts segment α0, reducing the affinity of the H2H a thousand-fold (36, 37); However, the β_H monomers are longer than α-spectrin monomers and overlap (βΗ24–31) and self-interact in the human βV homolog (49). The Drosophila β_H self-interaction domain (βHSID) is characterized in this paper.

The erythrocyte SBMS consists of an irregular polygonal network of (αβ)₂ tetramers that crosslink short actin filaments at nodes to resemble a fishing net and dynamically rearranges during circulation to accommodate changes in cell shape (17). In contrast, the SBMS in axons, known as the membrane periodic skeleton (MPS), exhibits an orthogonal arrangement of spectrin tetramers stretched between evenly spaced rings of actin filaments (9, 18). The MPS not only confers tensile strength to axons (19, 20) but also maintains tension critical for mechanosensation (21).

Each α- and β-spectrin subunit consists largely of an array of spectrin repeats (Fig. 1). Each repeat (∼106 amino acids) folds into a compact bundle of three helices (NH₂-A,B,C-COOH), with successive repeats linked by a continuous helical C→A linker between segments. Linker flexibility and repeat unfolding confer rope-like flexibility to the spectrins (22, 23, 24, 25, 26). The number of repeats varies from 16 to 30 among chains and isoforms, determining molecular length and extensibility. Non-repetitive domains, some isoform-specific, further diversify spectrin function (15).

All β-spectrin isoforms start with an N-terminal CH1/2 type F-actin binding domain (27) followed by a dimer nucleation site that directs the assembly of antiparallel α/β dimers (28), and a ‘head-to-head’ (H2H) domain near the C-terminus that forms (αβ)₂ and (αβ_H)₂ tetramers (29, 30). In tetramers, one F-actin-binding domain is presented at each end of the complex, allowing for F-actin crosslinking (Fig. 1). Since all currently defined SBMS depend upon crosslinking of F-actin by tetrameric spectrin, tetramerization has long been regarded as essential for SBMS assembly. Indeed, extensive evidence supports the need for canonical network formation in erythrocytes (e.g. (31) and neurons (e.g., (19, 20, 21, 32)), and strong loss-of-function spectrin mutations in Drosophila are generally lethal (33, 34, 35).

A key exception challenges this paradigm. The R28S substitution in human αΙ-spectrin disrupts the α1/β1 H2H interaction and breaks the erythrocyte SBMS to devastating effect (36), yet the homologous Drosophila α-spec^R22S mutation, which eliminates detectable binding to β-spectrin and reduces binding to β_H by 1000-fold in vitro, rescues α-spectrin null flies to viability (Fig. 1; (37)). This finding implies that tetramerization, and thus canonical SBMS network formation, is not essential for the vital functions of spectrin in flies.

Spectrins have a multitude of other documented functions away from the plasma membrane that would not seem a priori to require SBMS formation, including DNA repair (38), dynein anchoring (39, 40), and endosomal trafficking (7, 41). In principle, these could explain the essential nature of spectrin, despite disrupted network assembly. Nevertheless, the mild phenotype of α-spec^R22S mutants raises alternative possibilities: (i) residual tetramers might still form in vivo, or (ii) loss of canonical network formation could be compensated by another interaction or pathway during development (42).

β_H isoforms in Drosophila and chicken can form (αβ_H)₂ tetramers (43, 44, 45), and by implication canonical networks. Consistent with this, ovarian follicle cells in the fly lacking β_H do not retain apical α-spectrin (46). However, several observations suggest that β_H and α-spectrin may have independent roles in the apical domain. For example, apical α-spectrin can accumulate without β_H in nicastrin^agro mutant follicle cells (47), and β_H::YFP does not colocalize with α-spectrin in the Drosophila wing pouch epithelium (48). In addition, the human β_H isoform, βV-spectrin (βV), does not colocalize with α-spectrin in macaque photoreceptor cells (49). It is not known if a network exists in these contexts. It is also possible that indirect crosslinking of F-actin could occur via binding to separate sites on complexes of integral membrane proteins, or via direct binding of repeats to membrane lipids (50, 51), perhaps involving lipid rafts, or via the ubiquitous C-terminal pleckstrin homology (PH) domain (52). Together, these data suggest that β_H might form noncanonical assemblies independent of α-spectrin.

In this article, we identify and characterize a self-interaction domain in Drosophila β_H (βHSID) that mediates antiparallel (β_H)₂ dimer formation and may facilitate α-independent F-actin crosslinking. The βHSID is in a similar region to a dimerization domain that was loosely defined in human βV without further structural characterization (49). Our structural and biophysical analyses show that β_H segments 25 through 31 (βH25-31) form antiparallel dimers with sub-micromolar affinity. This affinity is not as tight as the fly α/β_H H2H interaction (37) but is similar to that of the H2H interaction between αI- and βI-spectrins that is physiologically important for the human erythrocyte SBMS (53). Using AlphaFold2 modeling, zero-length crosslinking with mass spectrometry, small-angle X-ray scattering, and negative-stain EM, we propose a structure for this domain that is one of the longest structural models yet described for a spectrin repeat array. Our results imply that β_H can probably form self-associating dimers, supporting the notion of α-independent β_H dimerization, and suggests a mechanism for non-canonical SBMS formation. Our structural and computational analyses also show that βH25-31 dimers contain a bent and particularly mobile central region, offering insight into the flexible nature of these giant proteins.

Results

A β_H-spectrin fragment containing segments 25 to 31 forms homodimers

A seven-repeat fragment of fly β_H, βH25-31, had good solubility when expressed as a GST fusion protein and released by thrombin cleavage after purification. Purified βH25-31 has a calculated mass of 86 kDa but runs at an apparent mass of ∼200 kDa on non-denaturing polyacrylamide gel consistent with dimer formation (Fig. 2A). In addition, AlphaFold2 predicts that this fragment forms an antiparallel dimer with high confidence (Figs. 2B and S1). The deviation from the predicted dimer mass of ∼170 kDa on a gel is likely due to the way this rod-shaped fragment migrates in comparison with the proteins in the commercial protein ladder, which are mostly globular. An extended rod-shaped dimer should run slower on the gel compared to a compact globular protein of the same molecular weight (see also Fig. S2). The AlphaFold2 model shows that the monomers are offset by one repeat, pairing βH25/30, βH26/29, βH27/28, βH28/27, βH29/26, βH30/25, and leaving segment βH31 unpaired at each end. This register is precisely that expected from the spectrin repeat alignment of a full length (αβ_H)₂-spectrin tetramer set by the dimer nucleation site in segments βH2-3 (Fig. 1; (15)). The repeat-repeat interface involves helices AB-AB throughout and there is a slight coiling of the chains around one another. This seems to be the general rule in published spectrin repeat dimers, and conservation analysis mapped on to the model shows that each repeat is most conserved at this interfacing surface (Fig. S3). We note that the model has complementary electrostatic charges along the length of the dimer interface (Fig. 2C) suggesting that this could be the primary mechanism driving dimer assembly in this register; however, we did not further investigate this hypothesis. Hereinafter we refer to the βH25-31 region as the β_H self-interaction domain (βHSID) in general discussion.

Figure 2.

PurifiedβH25-31fragments form homodimers.A, purified βH25-31 runs as a dimer on a nondenaturing gel. B, the top-ranked predicted AlphaFold2 model for βH25-31 is a dimer with anti-parallel chains (colored from N-terminus in blue to C-terminus in red). C, a surface view of the AlphaFold2 structure colored by electrostatic potential. The two chains have been pulled apart and rotated to show the binding interface to show a pattern of complementary charges.

βH25-31 dimers have a sub-micromolar range dissociation constant

To evaluate the likely physiological significance of the βHSID in vivo, we determined the dissociation constant (K_d) of the βH25-31 dimer protein using dilution ITC experiments in which purified βH25-31 was injected into a cell of matching buffer. With each injection, the heat of dissociation of dimers into monomers will be detected until reaching equilibrium allowing us to derive a K_d. These ITC experiments were done at different βH25-31 syringe concentrations and temperatures. The apparent K_d’s were 187 nM and 214 nM at 25 °C, the optimal temperature used for fly husbandry (Figs. 3 and S4). That this is a physiologically relevant affinity is suggested by comparison to the H2H interaction of erythrocyte spectrin that facilitates (αIβI)₂ tetramer formation. Estimates of that K_d range from 400 nM (36) to 800 nM (30), but its disruption still leads to network breaking and severe red cell defects.

Figure 3.

Dilution isothermal titration calorimetry curve for βH25-31. This experiment was performed at 25 °C in PBS with 1.5 μl injections of 32 μM βH25-31 into a 200 μl cell. The results were fitted to the dimer dissociation model using the TA Instruments NanoAnalyze software to determine that the K_d = 187 nM. Injections at 14 μM (25 °C) and 5.5 μM (4 °C) resulted in similar K_d of 214 nM and 316 nM respectively (Fig. S4).

We conclude that the stronger βHSID interaction is likely to be physiologically important in vivo and capable of supporting F-actin crosslinking by β_H dimers.

Zero-length crosslinking supports antiparallel dimerization of βH25-31

To find structural evidence for the βH25-31 dimer, we used the zero-length crosslinking reagents EDC and sulfo-NHS to covalently join any primary amines (e.g., on Lys) that are near to carboxylic acid groups on side chains of glutamate or aspartate. This chemistry does not add any atoms, and the crosslinked residues must be within 12 Å of each other (54). Running lightly crosslinked βH25-31 on SDS-PAGE gels reveals multiple high molecular weight bands well above the monomer, indicating the presence of multiple crosslinked species (Figs. 4 and S5).

Figure 4.

Zero-length crosslinking ofβH25-31support homodimerization.A, crosslinked βH25-31 run on an SDS-PAGE gel. Bands were excised and analyzed my mass spectrometry to identify crosslinked residues. The asterisked band contains the crosslink pair, K2818 to E3325 (residue numbers refer to full length β_H isoform A (15)). B, residues E3325 (left) and K2818 (right) shown as spheres on the predicted AlphaFold2 model (arrow) are 11.3 Å apart and in appropriate relative positions in the model to form this crosslink, and so this result further supports this arrangement of chains.

As a dimer, any crosslinkable sites in βH25-31 should be present twice in each complex, and at the low levels of crosslinking used we anticipate one or two crosslinks per dimer, each of which can be expected to migrate at a different rate depending on its specific geometry. For example, dimers with a crosslink at each end of the protein would be a compact circle and run faster in a denaturing gel than dimers with a single crosslink across one end that is highly elongated following denaturation. Crosslinks that involve more central residues would generate X-shaped or branched structures that would run differently yet again. Such considerations likely explain the cluster of bands we observe in the dimer region following crosslinking (Fig. 4A). All the bands in this region were separately excised and analyzed by trypsin digestion and tandem mass spectrometry.

Mass spectrometry analysis identified two prominent crosslinked residues that validate the predicted βH25-31 dimer structure. Lysine 2818 (sequence SKIQKHAAF) in βH25 was found to crosslink to glutamate 3325 (sequence QLINEKD) in βH30 (Figs. 4B, S5, S6, and Table S1). This crosslink indicates that the βH25-31 monomers are antiparallel and aligned in the register predicted by the AlphaFold2 model. All other identified crosslinks involve crosslinking of an N-terminal amine with nearby acidic residues in βH26 of the same peptide or with residues in βH30 of the other dimer subunit (Fig. S5B). It is known that the three α-helices within a spectrin repeat (NH-A,B,C-COOH) are connected by flexible turns (55), and these results are consistent with the N-terminal helix A swinging back or unfolding to interact with these residues. These crosslinks must arise as an in vitro artifact of truncating the longer spectrin repeat array and therefore did not factor these crosslinks into our model refinement.

These results further support the notion that βH25-31 forms a dimer in solution. We can also conclude that the AlphaFold2 predicted model for the βH25-31dimer is wholly supported by the positions of two crosslinked residues found by zero-length crosslinking and mass spectrometry.

Structural assessment of βH25-31 dimers using SAXS and negative staining

To further refine the AlphaFold2 model of the βH25-31 dimer structure, we sought to constrain the structure using molecular envelopes derived by small-angle X-ray scattering (SAXS) and negative stain electron microscopy (Fig. 5). SAXS experiments performed at two different concentrations (Table S2) confirm that the dimer structure is extended. In addition, the Kratky plots are not Gaussian as expected for a compact globular protein, but plateau at high q values, indicating that the protein is flexible and curved (Fig. S7). The electron density map for the βH25-31 dimer was generated from the scattering data using the DENSS algorithm (Fig. 5, A and B). This map also indicates that the protein is curved, and the wider central volume reveals that this is a region of flexibility and movement.

Figure 5.

Refined dimer models forβH25-31.A, two orthogonal views of the electron density envelope generated from DENSS using SAXS data from 0.55 mg/ml βH25-31 in PBS overlayed with the refined dimer model. B, a theoretical scattering dataset for the refined dimer model shown in A calculated using CRYSOL and compared to the experimental data has χ² = 1.44. C, representative monomer and dimer negative stain electron microscopy 2D class averages of lightly crosslinked and partially purified βH25-31 dimer. D, two orthogonal views of the negative stain dimer reconstruction with the fitted dimer models. Both analyses point to a region of central flexibility/curvature.

The βH25-31 has a large radius of gyration (Rg) consistent with an elongated dimer structure. The Rg determined by Guinier analysis for each SAXS experiment is 77.73 ± 4.17 Å or 88.07 ± 3.74 Å, whereas the Rg derived from the pair distance distribution P(r) function is 102.90 Å and 105.80 Å. Molecular dynamics simulations using the AlphaFold2 model predict a Rg near 107 Å consistent with the latter results. The maximum diameter (D_max) of the βH25-31 dimer was found to be 358 to 391 Å and is also consistent with the predicted dimer structure.

Negative staining offers an independent method to create a structural envelope for the βH25-31 to check these results. Because the protein prepared for the negative staining grids needs to be very dilute, most of the protein dissociates into monomers. To work around this, we prepared zero-length crosslinked protein that was enriched for the dimer fraction using size exclusion chromatography with multi-angle light scattering (SEC-MALS). The hydrodynamic radius for the dimer fraction determined from the SEC-MALS run (10 nm) was in close agreement with the Rg derived from the SAXS P(r) distance distribution plot, and the presence of crosslinked dimer was verified by SDS-PAGE before being used to prepare the negative staining grids.

After negative staining and imaging (Fig. 5C), βH25-31 dimers are identifiable as their shape is significantly thicker than that of the monomers. A curve in the protein structure is again apparent but is shallower than in the SAXS-derived DENSS envelope (Fig. 5D). The reduced curvature could be caused by the constrained protein interactions with the carbon-coated grids and the dye as opposed to being free to move in solution. The unpaired repeat (βH31) at the C-terminal end of the dimer exhibited very weak staining and did not average well. This may be because it is highly mobile, and so it could not be placed inside the negative stain envelope.

These results provide further support for a rod-shaped βH25-31dimer and indicate that the dimer is prone to bending in the vicinity of the two central segments.

Computational refinement of the βH25-31 dimer model

The unmodified AlphaFold2 model is quite straight and therefore has a relatively poor fit to the SAXS data and negative stain envelopes. Thus, comparing the theoretical scattering based on the AlphaFold2 prediction with the scattering data from the βH25-31 (in PBS at 0.55 mg/ml) using CRYSOL gives a χ² = 4.308. Similarly, comparison of the negative stain envelope with the AlphaFold2 model using Fit in Map in ChimeraX, had an average map value of 1.613. Normal Mode Analysis (NMA) using the iMODS server shows that the first determined mode of motion has the dimer bending at the middle such that the ends of the dimer move towards each other (Fig. 6, S8 and Movie S1). From this analysis, the middle portion of the dimer containing βH27 to 28 of both chains was judged to have continuously higher mobility than the neighboring repeats. βH31 has the highest predicted mobility consistent with our inability to see this region by negative staining, probably because it is not part of the dimer interface. NMA also indicated regions of high deformability at the helix A-B and B-C loops and in the helix A-B linkers throughout the dimer. The unpaired repeat βH31is again predicted to have greater deformability. These analyses indicate that the linker regions are more likely locations for bending to create the slight crescent shape observed.

Figure 6.

Normal mode analysis (NMA) ofβH25-31using the iMODS server.A, a heat map of the βH25-31 dimer model deformability (blue is the lowest, red is the highest) indicates that the loops at helical turns and helix C→A boundaries are likely foci for deformation. B, A heat map of the βH25-31 dimer colored by NMA mobility (blue is lowest, red is highest), indicating that the central βH27 to 28 repeat region is predicted to be more mobile, and that the unpaired βH31 repeat has the highest mobility. C, Arrows overlaid on the βH25-31 mobility model in B indicating the predicted direction of motion. The dimer bends so that the ends move towards each other. Default settings were used for the iMODS server, and a graphical depiction of this data by amino acid location and chain can be found in Fig. S8 and an animation of the predicted mobility in Movie S1.

To refine our dimer model, we therefore allowed for linker bending by specifying regions of disorder in the linker region at the bend between βH26 and βH27 and between βH28 and βH29. Flexibility in this region is also supported by the finding of an intrastrand crosslink in the βH25-31 monomer between repeats βH27 and βH28 (Starred in Fig. S5B, structure iii). This modeling was also constrained by our crosslinking analysis, which puts distance constraints on the K2818-E3325 pair. Following refinement by the ATSAS program, CORAL, using these parameters, the fit was significantly improved, resulting in a χ² = 1.44 (SAXS) with the linker residues modeled ab initio by CORAL. For the negative stain envelope, manually adjusting the model to introduce a similar bend and twist increased the average map value in the negative stain envelope to 1.801. These structural refinements nicely complement the SAXS DENSS envelope and suggest that the βHSID has a region of flexibility and motion at its center.

Discussion

Tetramer formation is central to the formation conventional (αβ)₂-spectrin network, and mutations that disrupt this interaction in αΙ-spectrin (R28S and I24T) or in βΙ-spectrin Providence (S2019P) in humans cause hereditary elliptocytosis (HE; (36)), elliptocytosis/poikilocytosis (56), or fatal hydrops fetalis (57), respectively. We previously showed that the homologous α-spec^R22S mutation in Drosophila was surprisingly benign suggesting that the requirement for a non-erythroid network per se is not as stringent as in the red cell (37). In conventional (αβ)₂-spectrin tetramers, the β-spectrin subunits do not overlap and are dependent on their association with an α-spectrin chain to form tetramers, and crosslink F-actin (see Fig. 1). In contrast, the giant β_Heavy-spectrin (β_H) subunits overlap in (αβ_H)₂ tetramers, offering the possibility of a β_H-β_H dimer interaction that would be capable of crosslinking F-actin. A homotypic interaction in the C-terminal region of the human β_H homolog, βV-spectrin (βV) has been previously identified without structural characterization (49). Here we identify and characterize just such a β_H self-interaction domain (βHSID) in β_H. AlphaFold2 predicts that the βH25-31 overlap region of Drosophila β_H forms an antiparallel dimer. We support this notion through a combination of biochemical analyses (non-denaturing gels, dilution ITC, and crosslinking/MSMS) that show that βH25-31 is capable of a physiologically relevant homodimerization in vitro. Further imaging and biophysical analyses (negative staining and SAXS) generated envelopes that are consistent with the measured and predicted dimensions of this dimer. Computational modeling also suggests that the βHSID may contain a point of increased flexibility. These observations are significant because they suggest that β_H dimerization via the β_H SID may occur in vivo and that (β_H)₂ dimers could crosslink F-actin in the absence of α-spectrin.

The crosslinked residues identified in βH25-31 (K2818↔E3325), agree with the predicted AlphaFold2 model as well as the a priori alignment of the overlapping β_H segments when aligned from the βH2-3 α/β dimer nucleation site in an (αβ_H)₂-spectrin tetramer (Fig. 1). AlphaFold2 analysis of the homologous region in human βV-spectrin indicates that a very similar interaction can be expected in this isoform (Fig. S9, A and B). Moreover, the K2818↔E3325 salt bridge that we were able to crosslink in fly β_H appears to be conserved in βV (K2722↔E3230; Fig. S9C), suggesting that this interaction is conserved and may have many details in common. Close inspection of the previous analysis of the βV homotypic interaction presented by Papal et al. (49), indicates that this is actually a different, possibly species-specific, interaction as their data clearly implicates a primary role for sequences in the non-repetitive C-terminal segment 33 of βV that is not present βH25-31 and has no homology to β_H apart from the pleckstrin homology domain (Fig. S10). It is possible that further SID exist elsewhere in β_Ηeavy isoforms but evidence for these is currently lacking.

That the βHSID exhibits a sub-micromolar dissociation constant that is similar in magnitude to the essential human α1-β1 spectrin H2H (tetramerization) interaction suggests that the βHSID is likely to be physiologically relevant in vivo. The magnitude of the human α1-β1 K_d is thought to facilitate the force-induced rearrangement of F-actin crosslinks during circulation in erythrocytes (17). Since, β_H has been associated with or implicated in several biomechanical/mechanosensory processes such as apical contraction (58, 59), wound healing (60), cell fusion during form muscle syncytia (61), and Hippo[MST]/Warts[LATS] pathway signaling (62, 63), we can speculate that the ability to dimerize and to readily rearrange might also play a significant role in β_H-dimer functionality. The βHSID K_d is significantly weaker than the nanomolar H2H interaction between the fly α and β or β_H chains that we previously determined (K_d = 5 nM for α↔β_H, and K_d = 14 nm for α↔β at 23 °C; (37)), indicating that the H2H is likely to be the dominant interaction in wild-type (αβ_H)₂ tetramer formation. However, we previously showed that an α0-1 protein containing the R22S substitution (homologous to the human R28S HE mutation; (36)) eliminated detectable binding to β17 to 18 and caused a ∼1000-fold reduction in binding to βH31 to 32 (K_d = 7.5 μm; (37)), but that full length α-spec^R22S can rescue α-spec null mutant flies to viability with only minor defects (ibid). Given that we detected low residual binding of α-spec^R22S to β_H, it is possible that the presence of the βHSID may be able to partially compensate for this (and therefore network formation) and contribute to that rescue. However, the resilience of fly development to the loss of all detectable binding of α-spec^R22S to β17 to 18 remains a surprising result since there is no evidence of a strong self-interaction between the overlapping α0-3 region in (αβ)₂ tetramers (Fig. 1; (64)).

Our data also indicate that the βHSID monomers coil around one another and that the SAXS and negative stain envelopes indicate that the βH25-31 dimer is bent and flexible. In trying to refine the fit to these envelopes, our modeling that the bend in βH25-31 probably involves two helix breaks separated by two spectrin repeats, something that could confer a more dramatic local bend in oligomers involving β_H. If this bend is also present in an αβ_H-spectrin dimer, it might also facilitate formation of closed dimers similar to conventional αβ-spectrin isoforms (64).

The bending of the βHSID dimer informs some older results in the spectrin field. Initial imaging of purified spectrin tetramers revealed significant variation between different isoforms in lateral chain interactions, conspicuous flexibility, and the occasional presence of sharp bends (43, 44, 65, 66), leading to the description of spectrins as “rope-like.” Higher order coiling of antiparallel repeat arrays (67) and even ‘supercoiling’ (65) can occur under some conditions. Although existing structures of multi-repeat spectrin fragments tend to have a continuous α-helix in the linker running from helix C into helix A of the next repeat (e.g. PDB files 1U4Q (68) and 3EDV (69)), there is structural evidence for significant bending at these locations (65, 70). Multiple sequence alignments of spectrin repeats indicate that there is conspicuous conservation of the equivalent helix C to A linker between species for the same linker (70). This latter observation, in combination with the deformability of βH25-31, which peaks distinctly in each linker, allows us to speculate that linker flexibility may be one of these conserved functions. The functionality of such large-scale conformational states and differences remains largely unexplored.

There are other examples of lateral association between short spectrin fragments. Repeats 15 to 17 of chicken brain α-spectrin crystallizes as an antiparallel dimer (PDB 1U4Q), but this interaction was not strong enough to be detected by analytical ultracentrifugation (68). There is also weak homodimerization of human red cell spectrin fragments containing the dimer nucleation site (i.e. α-spectrin repeats 20–21 and β-spectrin repeats 1–2) seen in sedimentation equilibrium assays where each fragment was found to be self-binding with K_d of 1.3 and 1.1 mM at 30°C, respectively (71). Doubling the fragment length to include α-spectrin repeats 18 to 21, and β-spectrin repeats 1 to 4 improved the binding affinity to 0.34 mM for each homodimer. Even though this binding is negligible compared to the heterodimer formation between these two fragments, which has low nanomolar affinity (71), it is possible that longer spectrin fragments with larger binding interfaces might have an overall binding affinity similar to that of βH25-31.

Higher order oligomerization mediated by the H2H interaction between 3 or more αβ-spectrin dimers has been observed in vitro by multiple purified conventional spectrins (72, 73) and more recently by cryo-electron tomography of whole red cells (22). Our identification of the βHSID raises the intriguing possibility that this interaction might also facilitate higher order oligomerization of αβ_H tetramers. In fact, rotary shadowed images of (αβ_H)₂ do seem to suggest that pairs of tetramers are able to form crossed or parallel assemblies via this central region (43).

Other possibilities for non-canonical spectrin assemblies remain to be explored. For example, mouse βIV-spectrin dependent clustering of sodium channels may not depend on the MPS per se (8, 74), and the major brain isoforms in both mouse (βIVΣ6) and human (βIVΣ5), is so truncated at the N-terminus that it cannot participate in MPS formation as we currently understand it (74, 75). In fact, the human βIVΣ5 isoform is also truncated at the C-terminus and lacks all the major α-spectrin interaction domains and must therefore be dependent on lateral interactions to associate with other spectrins or do so indirectly via spectrin-interacting proteins (75).

Concluding perspective

The growing list of specific functions and properties ascribed to β_Heavy-spectrins appears to be conserved. Human βV-spectrin, has been shown to have a SID near its C-terminal end and at least part of the cellular βV population does not colocalize with α-spectrin (49). These same authors also identified other binding partners and hypothesized that βV-dimers might have a role in Myosin VII-mediating trafficking of proteins in mammalian photoreceptor cells (49). Here we have demonstrated that Drosophila β_Heavy-spectrin also contains a SID, but in the nearby region where its spectrin repeat array overlaps in (αβ_H)₂ tetramers. We predict that this same SID is also present in βV. Homodimer formation is thus a conserved feature of β_Ηeavy isoforms. Also resembling the human isoform, β_H has recently been localized to the apicolateral region of some fly epithelia in the absence of α-spectrin (48) in what must be a candidate region to find β_H dimers. Finally, we have previously shown that β_H is also involved in protein trafficking through regulation of the endosome pathway in partnership with Annexin B9 (7, 41), and other labs have uncovered further interactions with Sec15, Shot, Patronin and Myosin V (63, 76), proteins associated with exocytic pathways. Together, these observations begin to paint a picture of these β_Heavy isoforms that is quite distinct from their conventional counterparts.

Experimental procedures

DNA cloning, expression and protein purification

Segments βH25-31 and its derivatives were amplified from a preexisting cDNA clone and inserted into the vector pGEX-4-T1. Cloned constructs were sequence verified and expressed in Escherichia coli BL21(DE3)-pLysS, grown to log phase before induction with 1 mM IPTG. Expression was induced for 20 h at 16°C before cells were pelleted and frozen at −80°C. Thawed cells were lysed by sonication on ice in buffer containing 50 mM Tris-Cl, pH 8.0, 50 mM NaCl, 1 mM β-mercaptoethanol, and 1 tablet per 10 ml cOmplete ULTRA Tablets ((64), EASYpack). The lysate was clarified by centrifugation at 12,000g before batch incubation with Glutathione Sepharose 4B beads (Cytiva, Marlborough MA) equilibrated in PBEP (10 mM NaH₂PO₄, pH 7.3, 130 mM NaCl, 1 mM β-mercaptoethanol, 5 mM EDTA, 150 μM PMSF) for 1 h with rotation at 4 °C. The slurry was packed in a column and washed sequentially with ≥10 bed volumes of PBEP and another ≥10 bed volumes of PBE (PBEP without PMSF), then eluted in G-buffer (50 mM Tris, 10 mM reduced glutathione, 1 mM β-mercaptoethanol, 5 mM EDTA, 1 μg/ml pepstatin).

The spectrin fragment was released from the GST using an empirically determined concentration of thrombin (Sigma). Thrombin digestion proceeded for 3 h at 37 °C, after which the digestion was stopped by adding PMSF to 300 μM. Glutathione was removed by dialysis in PBEP, and the protein was again incubated with glutathione beads to remove the GST. The spectrin fragment was collected from the column flow-through and concentrated before being applied to a Sephacryl S300 (Cytiva) column equilibrated in gel filtration buffer (10 mM NaH₂PO₄, pH 7.4, 130 mM NaCl, 1 mM β-mercaptoethanol, 0.5 mM EDTA, 150 μM PMSF) for final purification.

Concentrations of purified proteins were determined by measuring the absorbance at 280 nm and concentrated when necessary, using Vivaspin Turbo 4 and 15 centrifugal filters (Sartorius). Proteins were stored on ice. Purified βH25-31 constructs evaluated by SDS-PAGE as well as non-denaturing PAGE as an initial check for dimerization. Proteins were visualized after PAGE through the inclusion of 2,2,2-trichloroethanol in the acrylamide mixture when casting and the subsequent development of fluorescence with UV light on a transilluminator. Images of these gels were inverted to better see any fainter bands in the images. PAGE gels with samples for mass spectrometry did not use 0.5% 2,2,2-trichoroethanol and were instead stained with NOVEX Colloidal Blue Staining Kit (Thermo Fisher Scientific).

Protein modeling and sequence analysis

The sequence of βH25-31 was used for generating AlphaFold2 models using ColabFold (77) in a Google Colab Notebook with MMseqs2 (78) and AlphaFold-Multimer (79). The input sequence contained residues 2777 to 3517 (Uniprot Q9VZQ3). The Uniclust30/UniRef30 database was used (80). Five top models were generated, and the top structure was relaxed using parmed in AmberTools24 (81). To determine a predicted radius of gyration for the model, the relaxed model from the ParmEd output was used in molecular dynamics simulations in a Google Colab notebook using OpenMM (82). Normal mode analysis was performed on the relaxed structure using the iMODS server (83). Surface electrostatic analysis was done using the APBS Electrostatics plug-in for PyMOL (84). Sequence conservation was examined by uploading the βH25-31 model to the ConSurf server (85).

Zero-length crosslinking and mass spectrometry

Purified βH25-31 protein was dialyzed into PBS (10 mM NaH₂PO₄, 130 mM NaCl, pH 7.3), concentrated, and clarified before being crosslinked with 5 mM 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide (EDC) and 2.5 mM N-hydroxysulfosuccinimide (sulfo-NHS) on ice in the dark for 1 h. Crosslinking reactions were quenched by adding 20 mM dithiothreitol and then kept on ice for 20 additional minutes before adding an equal volume of 2x Laemmli buffer. Prior to separation by SDS-PAGE for mass spectrometry, samples were incubated at 37°C for 5 min. After gels were stained with NOVEX Colloidal Blue, bands were excised, digested with trypsin, and analyzed by mass spectrometry at the Penn State College of Medicine Mass Spectrometry and Proteomics Core. Specifically, peptides were analyzed by using the nanoElute UHPLC and the Bruker timsTOF fleX instrument. Peptides were separated by reversed-phase chromatography using a trapping cartridge (Acclaim PepMap C18 100 Å, 5 mm × 0.3 mm, 160454, Thermo Scientific) and the Bruker15 column (100 Å, 150 mm×0.15 mm). A 30-min gradient elution was performed using 0.1% formic acid throughout with the nanoElute UHPLC at 0.5 μL/min with acetonitrile increasing from 2% to 26% from 0-1500 s, then to 32% at 1800 s, and lastly to 95% at 1 μL/min from 1806 to 2100 s. Crosslinks were identified using pLINK (V2.3.11; cog-genomics.org). Complete mass spectrometry data and protocol is available at the MassIVE repository, where the accession number is MSV000102144.

Isothermal titration calorimetry

Purified βH25-31 that was dialyzed in PBS was concentrated, clarified by centrifugation, and analyzed by isothermal titration calorimetry (ITC) at 4 °C or 25 °C using a MicroCal ITC200 system (Malvern Panalytical Inc) and TA Instruments Affinity ITC Auto instruments (TA Instruments). ββH25-31 at concentrations from 5.5 to 32 μM was injected into a cell of matching dialysis buffer to measure the change in heat from dissociation of the dimer upon dilution. For experiments using MicroCal iTC200, 1.5 μl of protein was injected 25 times with a stir rate in the reaction cell of 750 rpm. With the Affinity ITC Auto instrument, there were 15 injections of 3 μl and a stir rate of 125 rpm. For K_D calculations, the TA Instruments NanoAnalyze (Version 4.0.2.0) dimer dissociation model was used.

Small angle X-ray scattering

Purified βH25-31 protein was dialyzed into the appropriate buffer (PBS, pH 7.40 or PBS, pH 7.40, 2% glycerol, 1 mM β-mercaptoethanol) and clarified by centrifugation after concentration to 0.55 mg/ml or 1.7 mg/ml. Samples were run in triplicate using an autoloader into the capillary which was held at 4 °C into the path of the X-ray beam. Data were acquired using the SAXSlab software on a Rigaku BioSAXS2000nano system, featuring a Rigaku MM007 rotating anode X-ray source and OptiSAXS confocal max-flux optics, coupled with a HyPix-3000 Hybrid Photon Counting detector. The sample-to-detector distance was calibrated to 495.5 mm using silver behenate standard (The Gem Dugout). The momentum transfer range (q) spanned from 0.008 Å⁻¹ to 0.5 Å⁻¹. A Kratky block attenuated the beam by 20%, producing a beam diameter of ∼100 μm. Protein samples were loaded into a quartz capillary flow cell using the Rigaku autosampler, with the sample stage maintained at 4 °C. The beam path was held under vacuum (<1 × 10⁻² torr) to reduce air scatter. Each sample had six scattering images collected for 10 min each that were averaged after confirming no radiation damage. Replicates were further averaged if scattering curves were well-aligned and had averaged replicates of reference buffer scattering subtracted. SAXS data was collected and initially processed using the SAXSLab software to obtain averaged and subtracted scattering curves before being further analyzed in the ATSAS software suite (86). The dimer model from AlphaFold2 was refined using the program CORAL (87), and the model fitted to the SAXS data was checked using CRYSOL (88). During refinement, crosslinked residue pairs identified by mass spectrometry were used as distance constraints. The electron density envelopes were calculated from the scattering data using DENSS (89). The best fitting model was superimposed with the DENSS envelope in PyMOL (https://www.pymol.org/support.html?#citing). Theoretical scattering curves were calculated from the final models using the CRYSOL, yielding low χ² values, indicative of good fits between model and data. SAXS data are available at SASBDB (sasdb.org (90); Accession number SASD2D2 - Beta-Heavy-Spectrin fragment containing segments 25–31).

Negative stain electron microscopy

Protein visualization using negative stain electron microscopy requires substantial dilution that results in extensive dissociation of βH25-31. To preserve the dimer state of βH25-31 for imaging, the protein first was crosslinked with EDC and sulfo-NHS and quenched as described above. Crosslinked dimer was enriched by using size exclusion chromatography coupled to multiple angle light scattering (SEC-MALS) using a Wyatt silica column (Wyatt Technology) equilibrated in PBS, pH 7.3 at 4 °C with 0.5 ml/min flow rate before loading quenched crosslinking reactions. A BSA standard run prior to the crosslinking reactions was used to align the light scattering curves along with the UV and dRI signal. The eluted crosslinked dimer was collected from a peak with larger calculated hydrodynamic radius and mass than the monomer peak and verified by SDS-PAGE. The dimer fraction was then buffer-exchanged to 10 mM HEPES/2% glycerol before preparing the negative stain electron microscopy grids.

The negative stain experiments were conducted using an established protocol (91) using in-house made continuous carbon grids. The grid surfaces were pre-screened for carbon quality using an optical microscope and selected grids were glow-discharged for 30 s at 15 mA using Pelco easiGlow (Ted Pella). 3.5 μl of purified sample was applied to the grid surface and incubated for 30 s. Then, the sample-surface of the grid was stained using 10 drops of 0.75% uranyl formate solution. After the procedure, excess liquid was removed by blotting using Whatman filter paper 1 (Sigma-Aldrich, # WHA1001090) and the grids were stored in a desiccating environment.

Stained grids were imaged at the Penn State Materials Characterization Lab Core Facility. Data was collected using TFS Talos F200C microscope operated at 200 kV with CETA camera at a nominal magnification of 570,00×, corresponding to a physical pixel size of 2.57 Å/pixel.

Both datasets were processed using the same approach: Data were converted to MRC file format using EMAN2 e2proc2d.py (92) and then imported to cryoSPARC and processed using a single-particle approach (93). Particles were picked using Blob Picker, cleaned up using iterative reference-free 2D classification and used to generate 3D classes with ab initio procedure. Selected particles were refined against that 3D template, and the resulting 3D reconstruction was used to generate 2D templates for new particle picking using Template Picker. The newly picked particles were again cleaned up using 2D classification. To avoid observer’s bias, we also analyzed the “rejected” classes. Selected particles were used to generate the final 3D Coulomb potential density map.

The reconstructions were generated based only on their respective data: There was no cross-use of 3D maps from the other dataset.

Data availability

Protein structure models have been deposited in ModelArchive at modelarchive.org/doi/10.5452/ma-ksp5p (βH25-31) modelarchive.org/doi/10.5452/ma-5qpgn (βV25-31). SAXS data are available at SASBDB (sasdb.org; Accession number SASD2D2 - Beta-Heavy-Spectrin fragment containing segments 25–31) Mass spectrometry data is available at the MassIVE repository, where the accession number is MSV000102144. Any additional details about any of our analyses are available upon request from the corresponding author.

Supporting information

This article contains supporting information (15, 49).

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Enrique De La Cruz