Impact of primary sequence changes on the self‐association properties of mammalian cystathionine beta‐synthase enzymes

Abstract

Cystathionine beta‐synthase (CBS) is an evolutionarily conserved enzyme that plays a key role in mammalian sulfur amino acid biochemistry, mutations in which are the cause of classical homocystinuria (HCU), an inborn error of metabolism. Although there is agreement in the literature that CBS is a homomultimer, its precise structure is a source of confusion. Here, we performed a series of experiments examining the quaternary structure of various wild‐type and mutant CBS enzymes using a combination of native gel electrophoresis, in situ activity assays, analytical ultracentrifugation, and gel filtration. Our data show that recombinantly expressed and purified full‐length wild‐type human CBS enzyme (hCBS) and HCU‐causing variants (p.P422L, p.I435T, and p.R125Q CBS) form high molecular weight assemblies that are consistent with the properties expected of a filament. The filament is enzymatically active, and its size is sensitive to protein concentration. This behavior contrasts sharply with hCBS enzymes containing small deletions within the Bateman domain, which form stable tetramers and octamers regardless of concentration. Examination of liver lysates from humans and mice confirms the existence of enzymatically active high molecular weight aggregates in vivo, but also shows that these aggregates are specific to human CBS and do not occur in mice. Molecular modeling using AlphaFold2 suggests that these experimentally observed differences may be explained by subtle differences in the interaction mediated by the Bateman domains. Our results show that small differences in amino acid sequence can cause large differences in the size and shape of CBS multimers.

1. INTRODUCTION

Cystathionine beta‐synthase (CBS) is a key enzyme in sulfur amino acid metabolism, catalyzing the condensation of serine with homocysteine to form cystathionine. Individuals with mutations in CBS have clinical CBS deficiency (classical homocystinuria, or HCU), an inborn error of metabolism characterized by very high levels of plasma total homocysteine (tHcy) (Mudd et al. 2001). HCU is the most common inborn error of sulfur metabolism, estimated to occur in about 1/100,000 births in the United States, although rates vary significantly in different countries, with the highest rates in the world being found in Qatar (1/1800 births; Zschocke et al. ²⁰⁰⁹). In healthy adults, tHcy concentration in plasma ranges from 5 to 15 μM, but untreated HCU patients often have tHcy in excess of 200 μM. HCU patients suffer from various pathologies including thrombosis, osteoporosis, mental retardation, and dislocated lenses. The major cause of mortality and morbidity in these patients is thrombosis.

The human CBS protein is 551 amino acids in length (63 kDa) and contains three functional domains: a N‐terminal domain that binds heme (a.a. 1–80), a core domain (a.a. 80–386) that contains the catalytic site which binds pyridoxal phosphate (PLP), and a C‐terminal regulatory domain is responsible for positive allosteric regulation by S‐adenosylmethionine (AdoMet) (a.a. 386–551) (Jhee and Kruger 2005) (Figure 1a). This region also contains two repeats of the so‐called CBS domain (Shan et al. 2001), which pair to form a Bateman module, a motif found in a variety of enzymes and membrane proteins involved in protein–protein interactions (Bateman 1997; Ereno‐Orbea et al. ^2013b).

FIGURE 1.

In situ activity and western analysis of hCBS^Wt and hCBS^Δ515–522. (a) Domain structure of human CBS protein and enzymatic reactions. (b) Three pairs of the indicated samples were run in a single native gel, which was divided into three parts. One part was subjected to CBS activity staining in the absence of AdoMet as described in Methods. Arrows indicate two distinct complexes present in the hCBS^Δ515–522 sample. Second panel is identical to first, but the activity staining was performed in the presence of 400 μM AdoMet. Third Panel shows Western blot analysis of same samples using anti‐hCBS polyclonal anti‐serum. Note activity staining lanes contained 800 ng of purified protein, while for Western blots only 8 ng was used.

Although it is widely reported that full‐length mammalian CBS is a homotetramer, careful examination of the literature suggests a more complicated situation. The original data supporting the idea for a tetramer came from a study in which rat liver homogenate was analyzed for CBS activity on a sucrose gradient which estimated the MW ~250,000 kDa (Skovby et al. 1984). However, in the same manuscript, gel‐filtration chromatography estimated the MW at 160,000 kDa. A later study, using recombinant purified human CBS protein, did not confirm that the predominant species was a tetramer, but rather that the protein formed “oligomers and larger aggregates” on native gels (Bukovska et al. 1994). Consistent with this behavior, no group has been able to crystalize full‐length human CBS protein for structural determination by X‐ray diffraction. However, it has been possible to get high‐quality crystals of a truncated CBS enzyme that lacks the entire C‐terminal domain (Δ414–551) (Meier et al. 2001) or one that has a ten‐amino acid deletion in the loop region of the CBS domain (Δ516–525), both in the absence and presence of AdoMet (Ereno‐Orbea et al. ^{2013a, 2014}). Both these engineered enzymes appear to be dimers based on structural and native gel‐electrophoretic behavior. Very recently, it has been shown that full‐length human CBS forms filaments as determined by cryo‐electron microscopy single particle analysis (McCorvie et al. ²⁰²⁴).

In this report, we examined the relationship between CBS quaternary structure and activity using a variety of biochemical, biophysical, and computer modeling assays. Our data show that the full‐length wild‐type human CBS forms enzymatically active elongated large molecular weight assemblies, but that the engineered human CBS constructs having small deletions in the Bateman domain, or the mouse CBS enzyme form smaller, more globular assemblies. Our findings show that small changes in primary amino acid sequence can result in large alterations in homomultimeric assemblies.

2. RESULTS

2.1. Activity versus protein analysis in native gels

We initially used native gel electrophoresis coupled with both an in situ CBS enzyme assay and immunoblot analysis to examine the relationship between enzyme activity and CBS multimer structure. The in situ CBS activity assay was previously developed in our lab and takes advantage of the alternative hydrogen sulfide (H₂S)‐producing activity of CBS which condenses homocysteine with cysteine (Chen et al. 2004). By soaking a native protein gel in a solution containing the substrates and lead acetate, one can visualize the production of H₂S by the formation of insoluble lead sulfide, which shows up as a black spot on the gel where the CBS protein is active.

Comparing the CBS activity stain between full‐length wild‐type human CBS protein (hCBS^WT) and a CBS protein containing an eight amino acid deletion in the second CBS motif of the C‐terminal Bateman domain (hCBS^Δ515–522), we observed a distinct difference in the staining pattern (Figure 1b). The hCBS^WT gave rise to a high molecular weight smear mostly above the 480 kDa marker, while hCBS^Δ515–522 produced a distinct band near 240 kDa, along with a weaker higher molecular weight species near 480 kDa. The intensity of staining in both protein preparations increased when AdoMet was added to the solution, indicating that both enzymes are in basal conformations that can be allosterically activated. Interestingly, immunoblot analysis of the same preparations showed that both had similarly substantial amounts of distinct 240 and 480 kDa species (Figure 1b). Thus, for the hCBS^Δ515–522, protein migration and enzyme activity appeared highly correlated, but for hCBS^WT it was not.

Similar analysis was performed on several other purified CBS proteins (Figures 2 and S1, Supporting Information), including full‐length yeast CBS (yCBS^WT), truncated yCBS lacking the C‐terminal domain (yCBS^Δ347–507), truncated hCBS (hCBS^Δ414–551), hCBS with a 12 amino acid deletion in the second CBS motif of Bateman domain (hCBS^Δ515–526), three different preparations of hCBS^WT, and four different pathogenic missense mutant hCBS proteins found in human HCU patients (p.R125Q, p.G307S, p.P422L, p.I435T). The yCBS^WT, yCBS^Δ347–507, hCBS^Δ414–551, and hCBS^Δ515–526 proteins all behaved similarly to hCBS^Δ515–522, that is, form discrete multimeric species that are enzymatically active and align with the Western blot pattern. Conversely, all three preparations of hCBS^WT, hCBS^R125Q, and hCBS^P422L showed a higher molecular weight smear on the activity gel that did not match up well with the laddering pattern observed in the Western blot. An exception was the hCBS^I435T protein, where the activity and Western patterns are essentially identical. As expected, the stable and inactive hCBS^G307S mutant showed very little or no catalytic activity in either the activity staining or the liquid assay accompanied with a weak Western blot signal possibly indicating a unique conformation that is not well detected by the antibody (Gupta et al. 2018).

FIGURE 2.

Analysis of mutant hCBS proteins. (a) CBS activity staining for the indicated proteins (800 ng/lane). Three different preparations of purified hCBS^WT designated as A, B, and C were assayed. (b) Native gel western analysis for same proteins (8 ng/lane). The panel at the bottom shows denaturing western blot for same preps (32 ng/lane). (c) Solution CBS enzyme activity performed in the presence and absence of 200 μM AdoMet.

We wondered if the discrepancy between native Western blots and activity stains might be related to the difference in protein concentrations between the two different gel systems (8 vs. 800 ng/lane). To explore this, we repeated the experiments using the same concentrations of hCBS^WT and hCBS^Δ515–522 for both the activity staining and protein assay. To prevent oversaturation of signal, we used Coomassie blue staining instead of Western blot to assess protein migration (Figure 3a). We observed that at this higher concentration the hCBS^WT protein pattern closely aligned with the activity stain pattern, that is, both showed a high molecular weight smear. Furthermore, the apparent molecular weight of the activity stain appeared to increase as the amount of protein increased (Figure 3b). This finding suggests that hCBS^WT is forming large molecular weight complexes in a concentration‐dependent manner. However, high concentrations of the hCBS^Δ515–522 protein did not alter its migration pattern, indicating that this version of the protein does not show concentration‐dependent behavior. To confirm the concentration dependence of the high molecular weight aggregates for hCBS^WT, we made triplicate gels in which 25–400 ng of protein were loaded per lane and the gels were examined by activity stain, Western blot, and silver staining (Figure S2). Again, we noted that the size of the complex increased with protein concentration. These results suggest that as the hCBS^WT concentration decreases the high molecular weight complexes fall apart into lower molecular weight species, implying a concentration‐dependent behavior.

FIGURE 3.

Full‐length hCBS^WT, but not hCBS^Δ515–522, shows concentration dependent differences in multimer formation. (a) Native gels stained with Coomassie Blue and CBS activity assay. Amount of protein in each lane is shown. (b) Native activity gel loading indicated amounts of hCBS^WT.

2.2. Migration of hCBS is not affected by addition of substrates and allosteric effector

We examined if the migration and activity pattern on native gels was affected by addition of substrate and the allosteric activator AdoMet (Figure S3). Purified full‐length hCBS^WT was incubated with homocysteine, serine, and AdoMet, or buffer alone, for 1 h at 37°C, and then run on native gel. No significant difference in the behavior was observed in either the Western blot or activity assay. This suggests that formation of high molecular weight aggregate is not affected by either the substrates or AdoMet.

2.3. Analytical ultracentrifugation

To further probe the properties of CBS in solution, we performed sedimentation velocity analytical ultracentrifugation (SV‐AUC) with multiwavelength optics (Edwards et al. 2020). Using first principles, this method provides an experimental measure of size, shape, and density of macromolecules in free solution and is well suited for the characterization of complex mixtures across a broad range of sizes and shapes. In this study, we analyzed our datasets using both the model‐independent van Holde‐Weischet method (Vanholde and Weischet 1978) and a two‐dimensional c(S,f/f _o ) distribution model derived from the Lamm equation, where the Svedberg coefficient (S) is a determined parameter sensitive to both mass (M) and shape (f/f _o ) (Brown and Schuck 2006).

We first examined hCBS^WT via absorbance optics at both 280 nm (total protein) and 430 nm (the Soret peak for the bound heme). Via the van‐Holde Weischet (vHW) analysis, the protein sample displayed disparate sedimentation properties dependent on the absorbance wavelength examined, with a lower S _avg observed for the total protein population versus the heme‐bound species (10.5 ± 1.4 at 280 nm 15.5 ± 4.2 at 430 nm) (Figures 4a and S4). In a two‐dimensional c(S,f/f _o ) analysis of the 280 nm dataset, there were three different populations that were observed with distinctive hydrodynamic properties (Figure 4b). The most abundant population (ii) occurred between 8 and 22 S with very large f/f _o values between ~3 and 6, consistent with a large, highly elongated (filamentous) protein population. A similar distribution of populations was observed at 430 nm, with similar S values but larger f/f _o values between ~5 and 9 (Figure 4c). At both wavelengths, two other less abundant species are observed: a < 8 S population (i) with f/f _o values between ~1 and 2 (consistent with globular macromolecules, e.g., dimers and tetramers of hCBS), and a 22–40 S population (iii) with f/f _o values between ~1 and 2 that we assign as high molecular weight protein (HMW) aggregates. We integrated these signals and observed that at both wavelengths, filamentous species (ii) comprised ~65%–70% of the total signal in the sample (Figure 4d). We surmise that the discrepancies in f/f _o for the filamentous population between both wavelengths is due to significant amounts of CBS protein lacking heme, which is consistent with earlier work showing the heme saturation of recombinant CBS is only around 70% (Kery et al., 1995). Because heme is known to be required for hCBS activity and stability (Majtan et al. 2008), we focused our experiments on the data obtained at 430 nm, representing constituted hCBS.

FIGURE 4.

Sedimentation velocity analytical ultracentrifugation. (a) van Holde‐Weischet (vHW) analysis of sedimentation velocity data, show at two wavelengths: 280 nm (open circles) and 430 nm (closed circles). (b, c) Two‐dimensional c(S,f/f _o ) analysis of sedimentation velocity analytical ultracentrifugation data. Sedimentation data at two different wavelengths (280 nm, Panel B; 430 nm, Panel C) were analyzed using the c(S,f/f _o ) model as implemented in Sedfit. In the uppermost panels one‐dimensional c(S) v S are shown, with S value on the x‐axis and c(S) on the y‐axis. In the lower panels, 2‐D shape and size distribution plots are shown, with sedimentation coefficient plotted along the x‐axis and frictional ratio (f/f _o ) plotted along the y‐axis. Increasing values of f/f _o correspond to highly elongated or non‐globular species. The heat map indicates the species concentration, from lowest population density (blue) to highest (red). (d) Integration of signal from c(S,f/f _o ) analysis. Show as a bar graph is the percent integrated signal corresponding to dimers and tetramers (<8 S, black), filamentous structures (8–22 S, gray), and higher molecular weight (HMW) aggregates (>22 S, cross‐hatch pattern). (e) In the upper panel, a view of the regulatory domain dimer and tetramer interactions as predicted by Alphafold (this study) is shown. Rendered in yellow is the loop that is deleted in hCBS^Δ515–522. Denoted in pink are the positions of P422 and I435. In the lower panel, a view of the hCBS^Δ515–522 crystal structure (PDB 4L3V (cite)) is shown. Shown is a dimer, with the heme domain highlighted in blue and the regulatory domain highlighted in green. In pink, the positions of Arg125 and Gly307 are shown, residing near the heme and PLP binding sites respectively, and distal to the regulatory domain. (h–j) vHW and two‐dimensional c(S,f/f _o ) analysis of hCBS^I435T (panel H), hCBS^R125Q (panel I), and hCBS^G307S (Panel J) are shown. (k) Integration of signal from c(S,f/f _o ) analysis of patient mutation variants. Show as a bar graph is the percent integrated signal corresponding to dimers and tetramers (<8 S, black), filamentous structures (8–22 S, gray), and higher molecular weight (HMW) aggregates (>22 S, cross‐hatch pattern). Values in all data shown were corrected to S_20,w.

Transformation of the two‐dimensional analysis to c(M,f/f _o ) showed that the hCBS^WT filamentous population ranges in mass from between ~2 and 10 megadaltons, and in a S v Stokes Radius (R_s) transformation, radii ranging from ~40–125 nm (Figure S5). The deletion of residues 515–522, which constitute the oligomerization loop responsible for a formation of higher order oligomers and filaments (Ereno‐Orbea et al. ^2013a; McCorvie et al. ²⁰²⁴) (Figure 4e), dramatically decreased the observed S values and heterogeneity (Figure 4f), consistent with the presence of dimers (Figure S5). For reference, the calculated value of S based on the experimentally determined crystal structure of hCBS^Δ516–525 (McCorvie et al. ²⁰¹⁴) is 6.04 with a f/f _o value of 1.27.

We next examined the patient‐derived variants hCBS^P422L and hCBS^I435T (Figure 4g,h) which occupy positions in the regulatory domain at predicted oligomeric interfaces (Figure 4e). Like with the hCBS^∆515–522 variant, substitutions at these positions would be predicted to perturb the interaction between regulatory domains and hence the self‐association properties of the enzyme. While S _avg changed little when compared to wild‐type for hCBS^P422L (Figure 4g), the f/f _o values observed is a bit smaller. The hCBS^I435T shows a modest increase in S _avg and a slight broadening of range of f/f _o values (Figure 4h). Unlike the hCBS^∆515–522 variant, neither single point mutation entirely abrogated the formation of large filament species, but measurably broadened the distribution of species in the mixture.

The patient‐derived variants hCBS^R125Q and hCBS^G307S occur within the catalytic domain of hCBS, near the respective heme and PLP active sites (Figure 4e). Despite not occurring at a predicted oligomeric interface, the hCBS^R125Q mutant displayed a significant broadening in the f/f _o values for the filament population (ii) and a significant increase in the HMW aggregate population (iii), suggesting that this mutant is less stable than the wild‐type enzyme (Figure 4i). The hCBS^G307S displayed a significantly different hydrodynamic distribution than the wild‐type enzyme, showing decreased S _avg values by vHW analysis and a downward shift in f/f _o values below 2 in two‐dimensional analyses (Figure 4j). These observations are recapitulated in integrations of the c(S,f/f _o ) distributions (Figure 4k), and overall the results of the AUC studies are generally consistent with the patterns observed in the native gels.

2.4. Native gel studies using mouse‐liver extracts

To determine if the filament‐like behavior of hCBS occurred in in vivo, we used native gels and activity staining to compare the behavior of liver lysates expressing either hCBS^WT, a nine amino acid N‐terminally hemagglutinin‐tagged hCBS (HA‐hCBS), and endogenous mouse CBS (mCBS). Activity staining indicated that both hCBS^WT and HA‐hCBS gave rise to a high molecular weight smear, but that activity in the mCBS lysates was localized in a single well‐defined complex running just below the 240 kDa marker (Figure 5a). Native western blot analysis showed that the hCBS species form a ladder of species going upward from the 240 kDa marker, but that mCBS only forms a single species at the same location as the activity bands (Figure 5a). For each of the mice, serum tHcy and solution liver CBS activity was also quantified and compared. Interestingly, even though the samples expressing hCBS^WT and HA‐CBS had significantly more CBS activity, tHcy levels were not lower than that found in the mCBS‐expressing mice. These findings suggest that hCBS^WT may be working less efficient than endogenous mCBS in vivo. To explore this idea further, we examined the relationship between liver lysate CBS activity and serum tHcy in a total of 26 mice expressing either mCBS or hCBS (Figure 5b). For both mouse and human CBS, we observed an inverse relationship between liver CBS activity and serum tHcy. However, it is also evident that the regression lines between the mCBS and hCBS samples are not the same, with much lower mCBS activity required for the same levels of serum tHcy as compared to that for HA‐hCBS‐expressing mice.

FIGURE 5.

Comparison of mCBS, hCBS, and HA‐hCBS in mouse liver. (a) Top panel shows native‐gel activity assay of liver lysates containing 60 μg of total protein from three different mice for each construct. Second panel shows native western blot in which 2.5 μg of protein were probed with anti‐CBS anti‐serum. Third panel shows denaturing western probed for CBS, while final panel shows same gel as probed with actin. Note that we suspect that there was a mistake in the dilutions used in lane 2 in the first two panels (both native gels were done at the same time). Serum tHcy concentrations and liver CBS activities of the lysates are shown underneath. (b) Graph of serum tHcy vs. liver lysate CBS activity for mCBS (n = 13) and HA‐hCBS‐expressing mice induced with zinc water (n = 13). Blue and red lines show nonlinear regression (one phase decay) for mCBS and re HA‐hCBS mice, respectively (R ² = 0.16, R ² = 0.64).

We also compared the behavior of endogenous mouse liver and human liver CBS (Figure 6). Whereas mouse liver showed consistency with regards to the locations of CBS activity and protein, hCBS did not, with human liver activity located at a significantly higher molecular weight than what is detected for the protein by western blot. We also confirmed differences in the size of the multimeric structure of mouse vs. human CBS using gel filtration chromatography (Figure S6). Whereas mCBS eluted in a peak centered around 242 kDa, human CBS eluted in the void volume (>670 kDa).

FIGURE 6.

CBS activity staining and native western blot for human and mouse liver extracts. Whole liver lysates from mice or three different human liver preparations were analyzed by native PAGE followed by either western blot or activity stain. Underneath activity stain is activity determined from liquid assay for same samples.

To further confirm these findings, we purchased recombinant 6x‐His‐tagged mCBS produced in S. cerevisiae from a commercial vendor and compared its multimeric behavior on native gels at high concentration (Figure S7a). Similar to what we observed in extracts, the mouse enzyme forms a ladder of distinct oligomeric species as opposed to a high molecular weight smear. We also used this enzyme to confirm the effectiveness of the human CBS anti‐serum we use to detect mCBS with similar sensitivity to hCBS (Figure S7b).

2.5. Structural modeling of C‐terminal domain multimers

To gain potential insight into the observed differences in quaternary structure behavior, we used Alphafold2‐Multimer (AF2‐M) to model the structure of eight copies of the C‐terminal regions of hCBS^WT, hCBS^Δ515–522, and mCBS. Examination of the top AF2‐M models for each of the monomers revealed good structural alignment with the same region for the experimentally determined region in the hCBS^Δ516–525, with the major differences being in the length and size of the beta‐sheet/loop region (Figure 7a,b). For each sequence, we used AF2‐M to generate 100 predicted octamer structures, using 20 random seeds for each of the five different AI algorithms used by AF2‐M (Jumper et al. 2021). For hCBS^WT, we found that 91 of the simulations generated linear multimers consistent with the idea that hCBS^WT forms filaments. These structures had interface predicted template modeling (ipTM) scores ranging from 0.73 to 0.34. All structures with ipTM scores >0.54 (90 of the 91) were almost identical to the cryo‐EM structure of the SAM‐bound filament of CBS domains determined by McCorvie et al. (PDB: 8S5M) (McCorvie et al. ²⁰²⁴) (Figure 7c). In the filament structure the monomers interact via two different surfaces: one surface consists of interactions between two helix–turn–helix regions at the base of the monomers, while a second interaction occurs via a beta‐strand/turn/beta‐strand super‐secondary structural element.

FIGURE 7.

AlphaFold‐Multimer molecular modeling. (a) Amino acid alignment C‐terminal regions of hCBS^WT, mCBS, and hCBS^Δ515–522 containing Bateman domain. Red amino acids show differences between human and mouse sequence. Amino acid number one is relative to position 398 in mCBS, and 403 in hCBS protein. (b) Three‐dimensional alignments of single C‐terminal domain. (c) Comparison of the highest scoring AF2‐M predicted structure of filament of eight C‐terminal domains from hCBS^WT (top) and cyro‐EM derived model. (d) Example of predicted circular structure of eight copies of C‐terminal domain from hCBS^Δ515–522.

In contrast, all 100 simulations using the hCBS^Δ515–522 sequence generated radially symmetric, circular‐like structures (Figure 7d). These structures had relatively low ipTM scores, ranging from 0.33 to 0.29. In these structures only the helix–turn–helix structure is present. The beta‐strand interface is absent because the deletion shortened each beta‐strand from 11 to 7 amino acid residues thus producing a much smaller and unstable interface. While in reality such structures may or may not form, the AF2‐M calculations demonstrate the mutation is likely to significantly alter the interface that creates the filament. The mCBS simulations produced a more varied mixture of the different types of structures, with 72 being linear, 23 that were circular, and 5 that appeared to be chains that folded‐back on themselves (Figure S8). The ipTM scores ranged from 0.62 to 0.28. The mouse sequence has an insertion of 14 amino acids relative to the human sequence in this region, creating a much larger turn and slightly shorter beta sheets (9 vs. 11 amino acids). We speculate that either the shorter beta sheets, or the long loop may interfere with filamentation, especially as loop regions are difficult to model. Overall, these studies suggest that the differences in the multimeric structures detected by experimental methods, may be related to differences in the beta sheet region of the Bateman domain that disrupt back‐to‐back interface, necessary for filament formation.

We also performed AF2 analysis on the C‐terminal regions for two of the human mutant proteins examined earlier. The hCBS^S435T behaved similarly to hCBS^WT, with ipTM ranging from 0.75 and 0.20, with the top 63 models (ipTM >0.62) showing linear chains. The hCBS^P422L had a lower ipTM score range (0.69–0.27) with the first “nonlinear” model occurring at an ipTM score of 0.54 (position 40).

3. DISCUSSION

In the experiments shown above, we provide evidence that hCBS^WT forms elongated filaments in solution, a population not observed for mCBS or hCBS^Δ515–522. The evidence supporting this conclusion arrived from three orthogonal approaches: Native PAGE analysis, multiwavelength SV‐AUC, and gel filtration analysis. The gold‐standard of these techniques, SV‐AUC, shows that the vast majority of recombinant purified hCBS^WT forms large megadalton, elongated polymers. This is also supported by gel filtration analysis, which shows that hCBS^WT elutes in the void fraction (R_s >8.6 nm). Activity‐stained native PAGE gels show that enzyme activity appears to form as a high molecular weight “smear” greater than 420 kDa. The exact centering of the smearing appears to vary depending on the amount of CBS enzyme present, suggesting that complex formation and/or its maintenance are concentration‐dependent. We also suspect that the smear may perhaps represent a mechanical shearing of the CBS filaments as they enter the gel, in much the same way that genomic DNA is sheared when subjected to agarose electrophoresis. Our data is entirely consistent with the recently published cryo‐EM data showing that hCBS^WT forms filaments (McCorvie et al. ^{2014, 2024}).

The behavior of hCBS^WT is in sharp contrast to mCBS, yCBS, or hCBS constructs with small deletions in their C‐terminal region. Our AUC studies of purified hCBS^Δ515–522 show that this enzyme is in a much more globular assembly with a molecular weight consistent with that of being a mixture of mostly dimers. Native activity gel analysis and gel‐filtration also show a smaller complex size. Native PAGE indicates similar behavior for full length yCBS, mCBS, hCBS^Δ414–551, and hCBS^Δ515–526. Interestingly, truncated yCBS^Δ347–507 migrates much faster than the other constructs with an observed molecular weight consistent with that of being a dimer. This result correlates well with the previous report of crystal structures of truncated yCBS, where a similar construct of yCBS lacking the C‐terminal domain formed homodimers (Tu et al. 2018).

The hCBS^WT in filaments appears to be enzymatically active, as judged by native gel activity staining. At first, we thought that these high MW species might be “super” active, as we noticed a majority of the activity staining was occurring in regions of the native gel in which Western blot analysis detected minimal amounts of CBS protein. However, this appears to have been an artifact related to the concentration of protein loaded on the different types of gels. When equivalent amount of protein is loaded, the activity staining and protein staining match quite well. Furthermore, the average size of the enzymatically active complex appears to decrease as the sample becomes more dilute, suggesting concentration‐dependent dynamic balance in the formation of hCBS filaments or higher order oligomers. We did not see any differences between hCBS^WT and hCBS^Δ515–522 with regards to overall specific activity or inducibility by AdoMet, suggesting that elongated fibrils and spherical assemblies are equally active in vitro. This behavior was previously observed for a similar hCBS^Δ516–525 construct, which was activity‐wise indistinguishable from hCBS^WT, but differed in oligomeric state. Dimerization of hCBS^Δ516–525 enabled crystallization and solving of the crystal structures of hCBS in the absence and presence of AdoMet (Ereno‐Orbea et al. ^{2013a, 2014}). Furthermore, the small loop deletion in hCBS^Δ516–525 construct modulated binding of AdoMet and allowed for better understanding of how AdoMet regulates activity and stability of hCBS (Pey et al. 2016). Unexpectedly, mCBS behaved much more similarly to hCBS^Δ515–522 than hCBS^WT with regards to its behavior on native PAGE and gel filtration indicating that it is primarily forming enzymatically active tetramers.

To better understand the molecular basis for these self‐association behaviors, we performed molecular modeling experiments using AF2‐M. AF2‐M is a variant of the original AlphaFold2 program that has been trained to model protein–protein complexes (Evans et al. 2022). Since AF2‐M can reliably predict multimer assemblies, we compared the assemblies of eight copies of the C‐terminal domains from hCBS^WT, mCBS, and hCBS^Δ515–522. We chose to focus on the C‐terminal domain for three reasons: (1) it is known to have a Bateman domain (i.e., a tandem repeat of CBS domains), which is a conserved motif known to regulate functions of a wide variety of proteins (Ereno‐Orbea et al. ^2013b), (2) small deletions within or entire removal of this domain lead to a change in oligomeric properties of hCBS (2001; Shan and Kruger ¹⁹⁹⁸), and (3) using the entire CBS protein would require significantly more computing resources. Because of stochastic processes that occur within the AF2‐M program, we chose to produce 100 assemblies for each of the three different molecules. Interestingly, we saw entirely different results between hCBS^WT and hCBS^Δ515–522. Whereas almost all the models generated by hCBS^WT were linear, all of the assemblies of hCBS^Δ515–522 were circular. This result seemed to agree with the quaternary differences observed experimentally. For mCBS, results were mixed, with 72% being linear and 23% being circular. Thus, it seems like the mCBS has lower propensity to form linear assemblies than hCBS^WT. However, it should be noted that we saw no evidence for mouse linear polymers in our experimental data. It could be interesting to compare experimental and AF2‐M modeling for a series of different CBS proteins from a variety of mammalian species to assess whether AlphaFold2 modeling can be a useful tool in predicting quaternary CBS assemblies.

Another important question is: Do the different quaternary assemblies of hCBS and mCBS have any functional consequences? A hint that they might come from the analysis comparing liver lysate CBS activity versus plasma tHcy in mice expressing either mCBS or HA‐hCBS. In both cases, we observed an inverse relationship between tHcy and CBS activity, but the curves were strikingly different (Figure 5b). For the same amount of mCBS activity, we observed much lower tHcy than we did for HA‐hCBS. This finding suggests that in vivo, hCBS is less “effective” at lowering tHcy. One could imagine that this might be related to the differences in quaternary structure and cellular maintenance of the hCBS filaments versus mCBS tetramers. Furthermore, we showed here that the presence of a pathogenic point mutation in hCBS affects formation of filaments and distribution of populations regardless whether the mutation is located in the oligomerization interface, such as hCBS^I435T (Figure 4h), or catalytic core, such as hCBS^R125Q (Figure 4i). This observation is well in line with previous calorimetric data showing that most of the studied HCU‐causing missense mutations lead to a kinetic destabilization of the C‐terminal regulatory domain regardless of their topology within CBS polypeptide (Pey et al. 2013). Furthermore, a more recent study showed that the cellular turnover of the majority of the studied HCU‐causing CBS variants was significantly faster compared to hCBS^WT (Mijatovic et al. 2024). Together, the data suggest that proper formation of higher order oligomers and filaments is relevant for CBS stability and consequently enzymatic activity. Obviously, future experiments could look at this question as it is potentially of interest because it might suggest novel ways to increase the effectiveness of emerging therapies for CBS deficiency based on gene or enzyme therapies.

4. MATERIALS AND METHODS

4.1. Proteins used for studies

Purified recombinant hCBS^WT, hCBS^Δ515–522, hCBS^Δ515–526, yCBS^WT, yCBS^Δ347–507, hCBS^Δ414–551, hCBS^R125Q, hCBS^I435T, hCBS^P422L, and hCBS^G307S proteins were made as described previously (Ereno‐Orbea et al. ^2013a; Majtan et al. ^{2010, 2014}; Majtan and Kraus ²⁰¹²). A summary of the properties of the different proteins is provided in Table 1. Purified mCBS was obtained from Aviva Systems Biology. It was produced in S. cerevisiae with a 6xHis N‐terminal tag.

TABLE 1.

Analyzed CBS proteins.

	Molecular Wt. (kDa)	Estimated pI	Predicted charge (pH 8.3)
hCBSWT	60.5	6.65	−9.1
hCBS Δ515–522 a	60.8	6.74	−9.1
hCBS Δ515–526 a	60.1	6.74	−10.0
mCBS	61.5	6.48	−9.7
hCBS (1–413)	46.5	6.8	−7.8
yCBS	56.0	6.7	−4.2
yCBS (1–346)	37.8	5.51	−9.0
hCBS I435T	60.5	6.65	−9.1
hCBS R125Q	60.5	6.54	−10.1
hCBS P422L	60.5	6.65	−9.1
hCBS G307S a	61.6	6.77	−10.0

^a Contains 6x‐His tag on C‐terminus.

Mouse livers were obtained from Tg‐hCBS Cbs ^−/− (Wang et al. 2004), Tg‐I278T Cbs ^−/− mice infected with pAAV:hCBS (Lee et al. 2021), or control non‐transgenic control mice. All mice were in C57BL6 background. Anonymous flash frozen human liver samples were obtained from the Fox Chase Cancer Center Biosample Repository. Lysates were made as previously described (Gupta et al. 2009). Mouse serum tHcy was determined as previously described (Gupta et al. 2009).

4.2. Native gel analysis

Samples were diluted with 2x Native Tris‐glycine sample buffer (Life Technologies, LC2673) and loaded on to 4%–12% Tris–Glycine gel (Life Technologies, XP04120BOX) in native running buffer (Life Technologies, LC2672) and run at 100 V for 5 h at 4°C. In all experiments a lane with a molecular weight marker (NativeMark unstained protein, Life Technologies, LC0725) was included.

For Western blot analysis, the protein maker lane was excised for Coomassie staining and remaining gel was transferred to PVDF membrane. After blocking in 5% non‐fat milk in Tris‐buffered saline, the membranes were probed with a 1/5000 dilution of immunopurified rabbit anti‐hCBS serum (Kruger and Cox 1994) followed by a 1/2000 dilution of horseradish peroxidase‐conjugated anti‐rabbit secondary antibody (Cell Signaling Technology, 7074). Signals were detected by chemiluminescence using SuperSignal West Pico Plus chemiluminescent substrate (Thermo Scientific, 34577) and images were captured using the FluorChem SP system (Alpha Innotech).

For the native gel assays, H₂S production by CBS was assayed by reaction with lead acetate (Chen et al. 2004). In brief, after native page the gel was soaked at room temperature overnight in 50 mL of the reaction buffer containing 400 mM sodium bicine (pH 8.6), 50 μM PLP, 0.25 mg/mL BSA, 10 mM L‐cysteine, 20 mM D,L‐homocysteine and 0.4 mM lead acetate in the presence or absence of 0.4 mM S‐adenosylmethionine (AdoMet). Gel images were captured by Alpha Innotech Image system.

Silver staining of native gel was performed using Pierce Silver Stain Kit (Thermo Scientific, 24612) according to the manufacturer's instructions. All steps were performed with constant gentle shaking. Coomassie staining was performed using Simply Blue SafeStain (Life Technologies, LC6060) for 1 h at room temperature with gentle shaking.

For denaturing Western blots, protein samples were reduced by adding 10x NuPAGE Sample Reducing Agent (Life Technologies, NP0004) with 4x NuPAGE LDS Sample Buffer (Life Technologies, NP0007) and denatured by heating for 5 min at 100°C. Proteins were separated by 4%–12% Tris‐Bis gel (Life Technologies, NP0321BOX) in NuPAGE MOPS SDS running buffer (Life Technologies, NP000102) supplemented with NuPAGE Antioxidant (Life Technologies, NP0005). Then, gel was transferred to membrane under reducing conditions by adding antioxidant into transfer buffer (Life Technologies, NP0006).

4.3. Solution CBS activity assay

CBS enzyme activities from purified proteins or mouse liver lysates (30 μg of protein) were carried out in 50 μL reaction volume containing 400 mM sodium bicine (pH 8.6), 100 μM PLP, and substrates (20 mM D,L‐homocysteine and 10 mM L‐serine) in the presence or absence of 0.5 mM AdoMet. Reactions were incubated at 37°C for 1 h and followed by reduction and extraction with 12% dithiothreitol and 10% sulfosalicylic acid, respectively (Wang et al. 2004). The cystathionine was identified and quantitated by an Aracus amino acid analyzer (MembraPure, Germany). One unit of activity is defined as one nanomole of cystathionine formed by milligram of protein per hour.

4.4. Analytical ultracentrifugation and gel filtration

Sedimentation velocity analytical ultracentrifugation experiments were performed in 20 mM Tris pH 8.0, 1% Triton X‐100 at 20°C with an Optima analytical ultracentrifuge (Beckman‐Coulter, Brea, CA) using a An‐50 Ti rotor with two‐channel charcoal‐filled Epon centerpieces and sapphire windows, using both absorbance and interference optics; data collected at 280 nm and 430 nm were used for this study. Complete sedimentation velocity profiles were recorded every 30 s at 20,000 rpm. Data were fit using the c(S) implementations of the Lamm equation as implemented in the program SEDFIT (Dam and Schuck 2004). Resulting S values were corrected for S_20,w. The partial specific volume (ῡ), solvent density (ρ), and viscosity (η) were derived from chemical composition by SEDNTERP (Philo 2023). The data were analyzed using several different models in SEDFIT; the c(s,ff0) model yielded fits with the lowest value for the summed square of the residuals (SSR). The standard c(s) model and c(s,ff0) models in SEDFIT gave final RMSDs of 0.010 or less, and SSR values of 0.5 or less, where a value of zero would indicate a perfect fit with no variance. Fitting of c(s,ff0) models were limited to frictional ratios from 1 to 5. Figures were created using the program GUSSI (Brautigam 2015). Gel filtration was performed using a SUPERDEX 200 column (GE Healthcare) as described previously (Peterson et al. 2006).

4.5. Molecular modeling

For modeling the following sequences were used: hCBS^WT (a.a. 403–551), hCBS^Δ515–522 (a.a. 403–543), and mCBS (a.a. 398–561). Eight copies of each sequence were fed into AlphaFold‐Multimer (Evans et al. 2022) as implemented in a downloaded version of Colabfold (Mirdita et al. 2022) and 100 structures (20 seeds × 5 Alphafold‐Multimer weight sets) for each were generated using the template‐free option. Output structures were visualized using PyMol, and assigned to one of three categories: linear, radially symmetric (circular), or other (branched or folded back).

AUTHOR CONTRIBUTIONS

Hyung‐Ok Lee: Investigation; methodology; data curation; writing – review and editing. Kushol Gupta: Investigation; conceptualization; methodology; formal analysis; resources; writing – review and editing. Liqun Wang: Investigation; methodology. Roland L. Dunbrack: Conceptualization; investigation; resources; formal analysis; writing – review and editing. Tomas Majtan: Resources; writing – review and editing. Warren D. Kruger: Writing – original draft; writing – review and editing; conceptualization; visualization; formal analysis; project administration; supervision; resources; funding acquisition.

Supporting information

Figure S1. Native gel analysis of human and yeast CBS proteins. Purified CBS proteins of indicated type were separated on native PAGE and assayed by activity stain, western blot, and Coomassie stain. Amount of protein used for each experiment is shown. Note that human CBS antibody does not recognize yCBS.

Figure S2. Concentration dependence of hCBS multimer formation. (a) Indicated amount of purified hCBS^WT run on native gels followed by activity staining. (b) Indicated amount of purified hCBS^WT run on native gels followed by Western blotting for CBS. (c) Indicated amount of purified hCBS^WT run on native gels followed by silver staining for total protein.

Figure S3. Effect of pre‐incubation with substrates and AdoMet on multimer formation and activity of hCBS. (a) Samples were diluted to a concentration of 3.7 mg/mL in 200 mM sodium bicine buffer (pH 7.6) and incubated either in the absence or presence of 10 mM D,L‐Hcy, 5 mM L‐serine, 25 μM PLP and 250 mM AdoMet for 5 min at 37°C before performing native PAGE (30 ng/lane) followed by Western blot analysis using anti‐hCBS polyclonal anti‐serum. (b) Samples were diluted to 18.7 μg/mL in 200 mM sodium bicine buffer (pH 7.6) and incubated either in the absence or presence of 10 mM D,L‐Hcy, 5 mM L‐serine, 25 μM PLP and 250 mM AdoMet for 5 min at 37°C before performing native PAGE (465 ng/lane) followed by activity staining.

Figure S4. Raw data collected for each hCBS protein examined in this study by SV‐AUC at 430 nm. For each protein, the uppermost panel shows the fits of the experimental data (circles) to the Lamm equation (lines); the bottom panel shows the residuals from this fitting. Every third boundary and third datapoint are shown for clarity. Measurements were performed at 4.4–11.5 μM loaded monomer concentrations at 20°C.

Figure S5. Transformations of c(S,f/f _o ) distributions calculated by SEDFIT to c(M,f/f _o ) and c(S,R_s) for each of the six hCBS samples examined by SV‐AUC. The heat maps indicates the species concentration, from lowest population density (blue) to highest (red).

Figure S6. Gel filtration of hCBS^WT (purified) and mCBS (mouse lysate). (a) Profile of native gel standards run on Sephadex 200. Molecular weights are indicated above the peaks. Note that the first peak contains all material >600 kDa, which is larger than the size limits the porous spheres (i.e., size exclusion limit). The green and red lines show where the hCBS^WT and mCBS elute. (b) Native Western blot showing fractions where mCBS elutes. (c) Coomassie‐stained denaturing gel showing fractions containing purified hCBS^WT.

Figure S7. Analysis of purified mCBS protein. (a) Native gel stained with commassie blue. Purified hCBS, hCBS^Δ515–522 and three different dilutions of mCBS shown. (b) Western blot comparing sensitivity of anti‐hCBS serum to purified human and mouse recombinant CBS protein.

Figure S8. Examples of three different multimer assemblies predicted by AF2 for mCBS. (a) Linear filament. (b) Folded‐over assembly. (c) Circular octamer.

Lee H‐O, Gupta K, Wang L, Dunbrack RL, Majtan T, Kruger WD. Impact of primary sequence changes on the self‐association properties of mammalian cystathionine beta‐synthase enzymes. Protein Science. 2024;33(12):e5223. 10.1002/pro.5223