Stranger Twins : A Tale of Resemblance and Contrast Between VAP Proteins

Abstract

When considering the vesicle-associated membrane protein-associated protein (VAP) family, major receptors at the surface of the endoplasmic reticulum (ER), it appears that VAP-A and VAP-B paralogs largely overlap in structure and function, and that specific features to distinguish these two proteins hardly exist or are poorly documented. Here, we question the degree of redundancy between VAP-A and VAP-B: is one simply a backup plan, in case of loss of function of one of the two genes, or are there molecular and functional divergences that would explain their maintenance during evolution?

Vesicle-associated membrane protein-associated protein (VAP): VAP-A and VAP-B are ubiquitously expressed type-II endoplasmic reticulum (ER) transmembrane (TM) proteins, with homologs widely distributed from yeast to human, involved in several functions at membrane contact sites (MCS), and implicated in genetic and infectious diseases (Murphy and Levine, 2016; Neefjes and Cabukusta, 2021; Kors et al., 2022). They encompass a N-terminal major sperm protein (MSP) domain, a central coiled-coil region for dimerization and a C-terminal TM domain (Figure 1A). Two intrinsically disordered regions (linkers 1 and 2) connect these different domains together and ensure VAP flexibility (Subra et al., 2023). VAP-A and VAP-B act as receptors for ∼100 proteins containing a short recognition motif called “two phenylalanines in an acidic tract” (FFAT), which consists of the consensus sequence EFFDAxE flanked by acidic residues (Loewen et al., 2003). VAP-A and VAP-B share 63% sequence identity and 83% similarity. Their MSP domain, responsible for FFAT binding, are nearly identical (82% sequence identity) and highly conserved. The FFAT motif binding site within the MSP is positively charged and includes two residues (K94 and M96 in VAP-A and K87 and M89 in VAP-B) that are key for VAP-FFAT interaction (Kaiser et al., 2005). VAP proteins tolerate many variations in the FFAT sequence, and also bind phospho-FFAT (pFFAT) motifs, which must be phosphorylated to interact with the MSP (Di Mattia et al., 2020). For example, STARD3, MIGA2, PTPIP51, and the voltage-gated membrane potassium channels Kv2.1 and Kv2.2 encompass a pFFAT (Alpy et al., 2013; Sun and De Camilli, 2018; Xu et al., 2020; Yeo et al., 2021). Amino acid variations within FFAT motifs may influence affinities between VAP-A or VAP-B and their partners, and thus modulate their respective subcellular localization.

Figure 1.

Similarities and differences between VAP-A and VAP-B. A: VAP-A and VAP-B have similar subcellular distribution and structural organization. B: The graphs show disorder prediction, coiled-coil probability score (grey line), and RMSF values calculated from MD simulations with full-length VAP-A or VAP-B having their TM helices inserted into a lipid bilayer. The two lines represent the RMSF of each monomer of a dimeric protein. Linker regions show larger fluctuation sizes. The conservation score (yellow scale) indicates that sequence divergence between VAP-A and VAP-B occurs primarily in the stem region. The helical structures of the VAP-A and VAP-B stem regions, as predicted by AlphaFold, are shown below. The color code shows the degree of confidence (high, dark blue; medium, light blue; low, yellow; very low, dark orange). C: Quantitative genetic expression of VAP-A and VAP-B among tissues, as determined by RNA sequencing, from the GTEx database (https://www.ebi.ac.uk/gxa/experiments/E-MTAB-5214). Expression levels in TPM (transcripts per million) are indicated according to the scale on the right.

Despite their high sequence similarity, a few genetic mutations are known to affect VAP-B but not VAP-A, including the prevalent VAP-B P56S mutation in its MSP domain, which is linked to a rare dominantly inherited form of amyotrophic lateral sclerosis (ALS-8) (Nishimura et al., 2004). The P56S mutation leads to MSP misfolding, resulting in the formation of aggregates, which are ubiquitinated and eliminated (Borgese et al., 2021). Curiously, proline 56 is conserved in VAP-A, but incorporating the same mutation in VAP-A causes minor effects. It appears that VAP-A is resistant to aggregation because it contains three prolines in a highly conserved sequence of its MSP, called the VAP Consensus Sequence (VCS; aa. 42–63), whereas VAP-B has only two prolines in its VCS. Mutation of a second proline in VAP-A VCS leads to massive aggregation resembling the VAP-B P56S phenotype (Nakamichi et al., 2011). However, this does not appear to be a strict requirement, as the Drosophila protein contains three prolines in the same region, yet the ALS-8-causing mutation can induce aggregate formation (Ratnaparkhi et al., 2008). Of note, MSP domains are also found in ER-resident mobile sperm domain-containing proteins (MOSPD1, 2 and 3), although MOSPD1 and MOSPD3 recognize motifs called “two phenylalanines in a neutral tract” (FFNT) (Cabukusta et al., 2020). MOSPD2 binds to FFAT motifs, but evidence suggests that it has distinct functions, notably via its CRAL-TRIO domain, and a lower abundance compared to VAP-A and VAP-B (Di Mattia et al., 2018; Zouiouich et al., 2022).

The interactomes of VAP-A and VAP-B (also called “VAPome”) have been revealed using affinity chromatography coupled to mass spectrometry (Huttlin et al., 2015), and proximity labeling approaches, either through BioID, which is based on proximity-dependent biotinylation of proteins by a biotin ligase fused to VAP (Cabukusta et al., 2020), or a modified ascorbate peroxidase 2 (APEX2) screen (James et al., 2019). These studies revealed new hits or confirmed known VAP-A and VAP-B interactors, such as OSBP and OSBP-related proteins, PTPIP51, CERT, VPS13A, MIGA2, PITPs and ACBD5, which are involved in lipid exchange at MCS, Rtn4, involved in membrane shaping, and sorting nexins (SNX), involved in membrane remodeling and transport (James and Kehlenbach, 2021). Of note, FFAT-containing proteins represent only ∼50% of the identified putative VAP partners (Murphy and Levine, 2016), suggesting that VAP-mediated interactions are not limited to MSP-FFAT binding. Importantly, the VAP-A and VAP-B interactomes do not fully overlap, as a significant number of hits are listed as specific partners of VAP-A (such as ORP7, ZFPL1, and TMEM38B) or of VAP-B (such as SNX2, ACBD4, and Ric-8) (James and Kehlenbach, 2021). It is unclear, however, what predicts specificity for one or the other protein. Since the MSP domains of VAP-A and VAP-B are very similar, their “stem” regions (comprising linkers, coiled-coil, and TM domain), which show sequence divergence, may promote the formation of specific protein complexes. Such interactions that are independent of the MSP have been reported with different partners (Hamamoto et al., 2005; Saito et al., 2008; Amini-Bavil-Olyaee et al., 2013; Forrest et al., 2013). This is the case for the phosphatase Sac1, which exhibits a better affinity for VAP-A than VAP-B (Wakana et al., 2015). Another example is Protrudin, an ER-resident and FFAT-containing protein that regulates neurite formation and axonal regeneration (Saita et al., 2009; Petrova et al., 2020). Interestingly, Protrudin preferentially associates with VAP-A rather than VAP-B, recognizing not only its MSP domain but also its TM domain (Saita et al., 2009).

The stem region might also confer specific properties to VAP-A that differ from VAP-B, such as oligomerization, flexibility, membrane diffusion, and subcellular localization, which in turn would support association with some partners over others. VAP-A and VAP-B are considered dimeric proteins, mainly due to their coiled-coil region, which consists of six putative heptads that would form a rigid rod of 5.9 nm. The TM α-helices of VAP-A and VAP-B also contain a conserved dimerization GxxxG motif. However, there are conflicting data on the ability of TM domain alone to form dimers (Kim et al., 2010; Cabukusta et al., 2020). Some studies also report that VAP-A and VAP-B are capable of forming heterodimers, but it remains unclear to what extent this occurs (Nishimura et al., 1999; Kim et al., 2010; Cabukusta et al., 2020). What about the role of MSP in dimerization? In the crystal structure, the MSP domain of VAP-A forms a 2:2 complex with FFAT, where MSP domains resemble a flytrap sandwiching two FFAT (Kaiser et al., 2005). By contrast, in NMR studies, VAP-A MSP makes a 1:1 complex with the FFAT motif (Furuita et al., 2010), and no contribution of the MSP to the dimerization of VAP-B was reported (Kim et al., 2010). Nevertheless, future studies will be needed to define the stoichiometry between VAP-A or VAP-B and their partners, as many of them are also organized as dimers, and may occupy the same contact sites. It is also worth noting that chemical crosslinking analyses suggested the existence of other oligomeric forms for VAP-B (monomer, tetramer) (Kim et al., 2010), but their functional impact at the molecular and cellular level remains obscure. This is potentially important when considering the lateral diffusion of VAP along the ER membrane. It is known that the subcellular localization of VAP-A or VAP-B can be modulated upon overexpression of one of its partners. For example, overexpression of OSBP brings VAP-A to ER/Golgi MCS (Mesmin et al., 2013). Recent single particle tracking (sptPALM) analyzes revealed that the trajectory of VAP-B shows localization hubs that correspond to ER/mitochondria MCS (Obara et al., 2022). Indeed, the diffusion of VAP-B along the ER decreases from ∼1.8 to ∼0.25 µm²/s when engaged in a contact site. As expected, overexpression of its partner PTPIP51 concentrates VAP-B at MCS; but without its MSP, VAP-B randomly explores the entire surface of the ER. Interestingly, in the regions that encompasses the TM helix, AlphaFold predicts different α-helix lengths for VAP-A (20 residues) and VAP-B (24 residues) (Jumper et al., 2021; Varadi et al., 2021) (Figure 1B), although other prediction programs (TOPCONS, DeepTMHMM, MEMSAT-SVM, etc.) suggest more similar lengths (∼19 residues on average). Of note, a mismatch between TM helix length and membrane thickness, which is largely determined by its lipid composition, can drive lateral and subcellular protein sorting (Levental and Lyman, 2023). For example, plasma membrane TM helices are on average longer than those of the ER and Golgi (the mean hydrophobic length is 20.3 residues in the ER and 24.4 in the plasma membrane) (Sharpe et al., 2010). Moreover, evidence suggests that proteins and lipids can co-segregate locally in the ER membrane, based on their biophysical properties, which may impact protein enrichment at MCS (Prasad et al., 2020; Zhemkov et al., 2021), and MCS organization (King et al., 2020). Thus, if TM helix lengths are different between VAP-A and VAP-B, they could mediate lateral segregation and/or adapt them to specific MCS.

Another structural feature worth mentioning is VAP linker regions, which show the most significant sequence variation between VAP-A and VAP-B. In a recent study, we demonstrated that these linkers are responsible for VAP-A flexibility, and are necessary for its engagement in the highly dynamic ER/Golgi MCS (Mora et al., 2021; Subra et al., 2023). By contrast, mutants of VAP-A designed to be more rigid only adapt to ER/Mitochondria MCS, which are more stable over time. Intrinsic disorder prediction suggests that VAP-B linkers are unstructured, similar to those of VAP-A. In Figure 1B, we show original molecular dynamics (MD) simulation data of full-length VAP-A versus VAP-B inserted into membranes. The RMSF values clearly indicate a higher mobility of the amino acids present in the linkers, suggesting that like VAP-A, VAP-B has two very flexible regions flanking its coiled-coil domain. It should be noted that there is a less abundant, natural isoform of VAP-A produced by alternative splicing that differs from canonical VAP-A by the insertion of 45 residues in its first linker. Disorder prediction indicates that the inserted sequence is unstructured, thus forming an extra-long flexible linker 1. We have shown that expressing this isoform rescued a range of defects associated with ER/Golgi and ER/mitochondria MCS in VAP-A KO cells (Subra et al., 2023), but the functional implication of this splice variant remains to be determined. As a side note, there is also a splice variant of VAP-B, called VAP-C, whose sequence stops in the middle of the MSP, but its function remains largely unexplored.

A number of studies have demonstrated clear functional differences between VAP-A and VAP-B. For example, it has been shown that VAP-A, but not VAP-B, plays a key role with ORP3 and Rab7 in promoting the entry of lates endosomes in deep nuclear envelope invaginations that penetrate into the nucleoplasm, possibly by facilitating MCS formation between these compartments (Santos et al., 2018). By contrast, a specific function for VAP-B was shown in differentiated motor neurons: in these cells, VAP-B depletion alters phosphoinositide balance and significantly delays neurite extension (Genevini et al., 2019). In rescue experiments, neurogenesis levels recovered to control cells with VAP-B expression, but less so when using the corresponding VAP-A construct. In a recent study, we showed divergent implications of VAP-A and VAP-B on mitochondrial morphology and cardiolipin (CL) levels in hTERT-immortalized retinal pigment epithelial (RPE-1) cells. Our data indicated that knocking out VAP-A, but not VAP-B, results in decreased CL levels associated with defective mitochondrial fusion (Subra et al., 2023). Of note, RPE-1 cells have comparable levels of VAP-A and VAP-B, suggesting that functional differences exist between these proteins. In these cited studies, however, the underlying molecular basis for why one VAP is involved in a given function and not the other remains elusive. Nevertheless, unlike VAP-B, VAP-A KO is lethal at the embryonic stage in mice (Kabashi et al., 2013; McCune et al., 2017), implying that VAP-B is less important and/or less abundant than VAP-A, and further suggesting that VAP-A can replace VAP-B in its functions, but not vice versa. There is indeed a striking difference in the transcript levels of VAP-A and VAP-B genes between tissues and, as a general rule, VAP-B transcript levels are reduced compared to VAP-A (Figure 1C) (Consortium et al., 2015). Of note, they appear to be more balanced in the nervous system than in other tissues. Nevertheless, VAP-A was shown to be at least four times more abundant than VAP-B in mouse motoneuron-like cells (Genevini et al., 2019). Future studies will be necessary to define the molecular basis explaining the preferences of each of the VAPs for different partners, and thus the set of functions that are not shared between VAP-A and VAP-B, as their respective roles may diverge to varying degrees depending on the cell type or tissue considered.

Methods

MD Simulations of VAP-A and VAP-B Dimer Models

MD simulations were performed essentially as described previously (Subra et al., 2023), except that full-length membranous VAP-A and VAP-B proteins were simulated. Simulations of modified VAP-A and VAP-B structures from AlphaFold in a membrane were prepared with CHARMM-GUI using 560 POPC, 560 DOPC, and 56 cholesterol (5% cholesterol), a water thickness of 30 Å at a temperature of 303 K and neutralized with 120 mM NaCl. The TM domains were inserted into the membrane using the PPM 2.0 tool. Energy minimization and simulations were performed as described previously except that two NVT and four NPT equilibrations were performed using the Berendsen thermostat and barostat. Simulations (1 µs) were done in triplicate, using Gromacs 2021.3 (https://www.gromacs.org/) with the CHARMM36 force-field, and the TIP3P water model. MD analyzes of the root mean square fluctuation (RMSF) was performed as described previously (Subra et al., 2023).

Web Servers for Secondary Structure Prediction

For disorder prediction, we used DISOPRED3 (http://bioinf.cs.ucl.ac.uk/disopred). For the coiled-coil probability score, we used COILS (https://npsa-prabi.ibcp.fr/NPSA/npsa_lupas.html). For TM helix prediction, we used the following web servers: TOPCONS (https://topcons.cbr.su.se/), DeepTMHMM (https://dtu.biolib.com/DeepTMHMM/), MEMSAT-SVM (http://bioinf.cs.ucl.ac.uk/psipred/), HMMpTM (http://aias.biol.uoa.gr/HMMpTM/index.php) and CCTOP (https://cctop.ttk.hu/). Predictions from TOPCONS and CCTOP give consensus results from several algorithms.