Curvature sensing amphipathic helix in the C-terminus of RTNLB13 is conserved in all endoplasmic reticulum shaping reticulons in Arabidopsis thaliana

The reticulon family of integral membrane proteins are conserved across all eukaryotes and typically localize to the endoplasmic reticulum (ER), where they are involved in generating highly-curved tubules. We recently demonstrated that Reticulon-like protein B13 (RTNLB13) from Arabidopsis thaliana contains a curvature-responsive amphipathic helix (APH) important for the proteins’ ability to induce curvature in the ER membrane, but incapable of generating curvature by itself. We suggested it acts as a feedback element, only folding/binding once a sufficient degree of curvature has been achieved, and stabilizes curvature without disrupting the bilayer. However, it remains unclear whether this is unique to RTNLB13 or is conserved across all reticulons—to date, experimental evidence has only been reported for two reticulons. Here we used biophysical methods to characterize a minimal library of putative APH peptides from across the 21 A. thaliana isoforms. We found that reticulons with the closest evolutionary relationship to RTNLB13 contain curvature-sensing APHs in the same location with sequence conservation. Our data reveal that a more distantly-related branch of reticulons developed a ~ 20-residue linker between the transmembrane domain and APH. This may facilitate functional flexibility as previous studies have linked these isoforms not only to ER remodeling but other cellular activities.


Results
Phylogenetic, functional, and sequence analysis of RTN isoforms in A. thaliana identifies minimal library of putative RTN APH regions for study. As demonstrated previously 24 , a multiple sequence alignment of all RTN isoforms in A. thaliana revealed that the APH region in RTNLB13 was highly conserved across the family (Fig. 1A). Pairwise sequence alignment of this region in all isoforms against RTNLB13 using EMBOSS Stretcher 28 yielded an average sequence similarity of 51%. Several of the hydrophobic residues that have already been identified as important for the interaction of the APH in RTNLB13 with highly-curved membranes were also the most highly conserved (Fig. 1B) 24 . Here we carried out a phylogenetic analysis of the 21 RTN isoforms in A. thaliana in order to group the isoforms according to their evolutionary relationship. As shown in Fig. 1D, RTNLB1-16 are clearly differentiated from RTNLB17-21. When one interrogates the sequences of these proteins, as shown in Fig. 1C, the most recognizable difference between these two groups is the increased length of the N-terminal domain in RTNLB17-21 (179-390 amino acids) as compared to RTNLB1-16 (23-99 amino acids) 29 .
After careful consideration of the evolutionary relationships revealed via the phylogenetic analysis shown in Fig. 1D, we decided to segregate the 21 isoforms into six clades (denoted here as Clades 1-6) to identify a minimal library of RTNs for further study. RTNLB1-16 were split into four clades: RTNLB1-9 and 14 (Clade 1); RTNLB15 (Clade 2); RTNLB10-12 and 16 (Clade 3); and RTNLB13 (Clade 4). Isoforms within these groups, namely RTNLB1-4 and 13, have been previously reported to share the same topology and general properties 10 , and direct ER shaping and membrane remodelling. In addition to ER shaping, RTNLB1-4 and 8 have been implicated in plant-microbe interactions via interactions with the Agrobacterium tumefaciens VirB2 protein 30,31 , and RTNLB3 and 6 have been localised to the primary plasmodesmata at cytokinesis, indicating that they may be involved in the formation of the desmotubule 32 .
RTNLB17-21 were split into two clades, RTNLB19-20 (Clade 5 in Fig. 1D) and RTNLB17, 18 and 21 (Clade 6), described by Nziengui et al. previously 27 . As well as containing N-terminal domains that are two-tenfold longer than those of RTNLB1-16, and which have been speculated to encode additional protein functionality 29,33 , it has been shown that isoforms in these groups (RTNLB19 and 20) do not constrict ER tubules when overexpressed and are not involved in ER shaping 29 . The same team suggested that the lack of tubule-forming ability may be due to the absence of an APH in the C-terminus of RTNLB20, which has a higher percentage of hydrophobic residues in comparison to RTNLB13. Instead, it was shown that RTNLB19 and 20 are involved in sterol biosynthesis lipid  Fig. 1D suggests that the reticulon "tethers" differentiated from the "structural" reticulons early during evolution 36 . Given that RTNLB19 and 20 have been shown to have no role in constriction of ER tubules and ER shaping 29 and instead have been implicated in other processes, these isoforms were omitted from our minimal library interrogating conservation of a C-terminal APH. Representative RTNs from Clades 1-3 and Clade 6 were selected, and 22-residue peptides corresponding to the putative APH region, as indicated from multiple sequence alignment (Fig. 1A), were synthesized for comparison to previous results for RTNLB13 24 (Clade 4). The identity of each representative RTN and the APH peptide sequences are given in Table 1. All five of these peptides fit the requirements for an APH; when plotted on a helical wheel each peptide contains a polar/charged helical face and a hydrophobic face that yields a high hydrophobic moment (Fig. 2). These peptides, in combination with the RTNLB13 APH peptide reported previously 24 , represent a minimal library of RTN APH constructs that would allow us to understand whether or not the APH module is conserved throughout A. thaliana RTNs involved in ER shaping and, if so, whether the emerging properties that define such curvature sensing regions are retained.  www.nature.com/scientificreports/ RTNLB10 and RTNLB15 contain curvature-sensitive APHs at a location similar to that in RTNLB13, but RTNLB4, RTNLB7 and RTNLB21 do not. The four peptides from Clades 1-3 listed in Table 1 are located at similar positions relative to the RHD in each protein (i.e. immediately following the putative fourth transmembrane domain) and have strong sequence similarities with one another and with the APH in RTNLB13. We have shown that the APH in RTNLB13 folds into an α-helix and binds to membrane bilayers in a curvature-dependent manner 24 , therefore replication of this behavior was investigated here across the library of peptides. Each peptide was exposed to model membranes of increasing curvature, and the resulting impact on secondary structure was measured using circular dichroism (CD) spectroscopy. Figure 2B shows the CD spectra for each peptide from Clades 1-3 in DPC and LMPG detergent micelles, DMPC/DHPC bicelles (q = 0.25), DMPC vesicles, and vesicles of composition 60:1:15 DMPC:DMPG:DHPC prepared with increasing diameter. All hydrodynamic diameters were obtained from dynamic light scattering (DLS) measurements in the absence of peptide and are given in the legend in Fig. 2B. This library of model membranes is a more concise version of that utilized in our recent work 24 designed to include conditions that most clearly demonstrate the response under investigation. CD spectra indicated that all the peptides in Clades 1-3 were soluble and had no defined structure (random coil) in aqueous buffer (Fig. S1). All peptides were also unstructured when the diameter of the model membrane was > 40 nm (Fig. 2B). These results directly reflect the behavior of the APH in RTNLB13 observed in our previous study. However, only the putative APH peptides from RTNLB10 and RTNLB15 demonstrated a transition in fold from random coil to α-helix upon exposure to increasingly curved model membranes (Fig. 2B). The effect is more pronounced in RTNLB15, although both regions demonstrate a clear propensity to fold as the model membranes decrease in size (increase in curvature). Conversely, the peptides derived from RTNLB4 and RTNLB7 remain unstructured under all conditions tested (Fig. 2B). This result was unexpected given that these regions contain many of the features one would expect from a curvature sensor, i.e. a high hydrophobic moment and a shallow hydrophobic helical face, and reinforces the challenge that curvature sensing regions are difficult to identify from sequence alone 24 .
The peptide chosen from Clade 6 (RTNLB21) aligns with the APH in RTNLB13 but has less sequence similarity compared to the other selected peptides (Fig. 2C). One obvious difference is that this peptide contains no acidic residues. The RTNLB21 peptide was also considerably less soluble than the others under investigation, and only a very weak CD signal was observed when this peptide was exposed to the typically most-solubilizing membrane mimetics (DPC, LMPG, DHPC/DMPC bicelles) as shown in Fig. 2C. This lack of solubility hindered www.nature.com/scientificreports/ the acquisition of CD spectra of this peptide with lipid vesicles, and suggests that this region of RTNLB21 has very different chemical properties to those selected from Clades 1-3.
For the two peptides that demonstrated curvature-responsive folding, the same solution-state NMR approach we applied previously 24 was used to map the α-helices within the sequences. 1 H-1 H total correlation spectroscopy (TOCSY) and 1 H-1 H nuclear Overhauser spectroscopy (NOESY) spectra were obtained to sequentially assign both peptides (see Fig. 3A, Tables S1 and S2 for assignments and Figs. S2 and S3 for full spectra) in the presence of deuterated DPC micelles, a condition which resulted in significant helical content for both peptides. 83% of the 1 Hs in the RTNLB10 peptide and 87% of the 1 Hs in the RTNLB15 peptide were assigned. Long-range backbone (i, i + n) H N , Hα and Hβ NOEs indicative of α-helix formation were identified and, together with chemical shift index analysis 37 of the 1 Hα chemical shifts (see Fig. 3B), were used to map the α-helical region in RTNLB10 to a 14 amino acid stretch between Phe164-Leu177, and similarly to a region of the same length in RTNLB15, corresponding to Tyr152-Lys165 (Fig. 3B, C). Both of these regions incorporate the most highly conserved residues in this region of the protein, as shown in the sequence logo in Fig. 1B, including the completely conserved aromatic residues.
The C-terminal domains in RTNLB4 and RTNLB7 contain an APH approximately 20 residues from the putative fourth transmembrane domain. The absence of an α-helix in the RTNLB4, RTNLB7, and RTNLB21-derived peptides did not rule out the possibility that an APH module was present at another location in their C-terminal domains. Given that curvature sensors are difficult to identify from sequence, we decided to investigate the entire C-terminus of RTNLB7 as this was the shortest of the three. A 39-residue peptide corresponding to the entire C-Terminal region of RTNLB7 (Fig. 4A, R7-CTerm, residues 206-244) was synthesized and purified. CD data were collected for this peptide in a range of solution conditions and demonstrated generation of α-helical secondary structure in this larger peptide upon exposure to detergent micelles and DMPC/DHPC bicelles, but not in DMPC or 60:1:15 DMPC:DMPG:DHPC vesicles (Fig. S4). The limited solubility of this peptide precluded high-resolution NMR measurements, as concentrations required for NMR were not attainable, and the location of the helical region could not be defined.
An alternative approach was adopted in which shorter peptides corresponding to the extreme C-termini of RTNLB4 and 7 were synthesized, purified and reconstituted into our set of model membranes. These peptides overlapped with the original sequences (see Fig. 4A), covered the majority of the remaining C-termini, and contained a region of sequence predicted to fold into an amphipathic helix (as evaluated using Heliquest 38 ). Design of such a peptide for RTNLB21, however, was not possible. Our original RTNLB21 peptide was designed to www.nature.com/scientificreports/ follow the putative fourth transmembrane domain of the RHD, however it has been speculated in the literature that members of Clade 6 contain an additional membrane-spanning region within their C-termini 27 . Indeed, topology prediction of RTNLB21 using TOPCONS 39 suggests that this protein contains a fifth transmembrane domain only 13 residues from our peptide between residues ~ 405-426. The C-terminus beyond this putative fifth transmembrane domain is highly polar along its entire length, enriched in acidic residues, and yields no potential APH regions which to target. Therefore, no additional peptide was investigated for RTNLB21. The new RTNLB4 peptide (residues 226-250) and RTNLB7 peptide (residues 223-244) yielded α-helical secondary structure (see CD spectra in Fig. 4B) in the presence of DPC, LMPG and q = 0.25 DMPC/DHPC bicelles while remaining unstructured in DMPC and 60:1:15 DMPC:DMPG:DHPC vesicles of different sizes. Solution-state NMR was used to assign 78% of the 1 Hs in the DPC-solubilized RTNLB4 peptide (Table S3 and 86% of the 1 Hs in the RTNLB7 peptide (Table S4), and chemical shift and NOE data were used to map the location of the helical regions in both peptides (Fig. 4C, D). In the case of RTNLB4, a C-terminal APH is present between residues Glu226-Ser240 and in RTNLB7 a C-terminal APH is present between residues Lys229-Ile244 . These sequences are plotted on a helical wheel diagram in Fig. 5 to demonstrate that both regions fit the description of Survey of NMR-derived sequential backbone NOE connectivities for RTNLB4 and RTNLB7 peptides solubilized in 50 mM DPC-d 38 , classified as strong, weak, or absent by the thickness (or absence) of a bar connecting the residues involved. The non-sequential connectivities listed (i.e. i, i + n) are unique to helices, and were used alongside chemical shift index analyses (CSI) to localize the amphipathic helices to those residues underlined in the sequences. Each displays the characteristic shallow hydrophobic face that restricts how far they can penetrate into a membrane in order to avoid bilayer disruption. Hydrophobic residues are shaded grey, polar residues colored purple, acidic residues colored blue, basic residues colored red-orange and other residues are unshaded. www.nature.com/scientificreports/ an APH, and in both cases the APH is 15-16 residues in length and 20-23 residues from the reticulon homology domain.

Discussion
Recent studies of a yeast and a plant reticulon suggest that an APH following the reticulon homology domain is critical for membrane remodeling and ER shaping by this family of proteins [22][23][24] , however experimental data has only been reported for two proteins, RTNLB13 23,24 and Yop1p 22 . Here we have examined all 21 RTN isoforms from A. thaliana to create a library of peptides in order to establish whether this feature is indeed conserved, or if the APH in RTNLB13 is unique. We have summarized our findings in a structural schematic shown in Fig. 6. Inspection of the phylogenetic tree in Fig. 1A revealed that all A. thaliana RTNLB isoforms share a common ancestor that most likely contained the RHD. RTNLB17-21 branch away from RTNLB1-16 early in evolution, and the RHD in these proteins is thought to simply tether them to the ER for other functions 27,29,35,40 as these proteins are not believed to be involved in shaping the ER. Therefore, isoforms in these two clades are proposed to not contain a curvature-sensitive APH in their C-termini (Fig. 6, bottom panel) and may instead contain other features such as an additional transmembrane domain 27 . RTNLB1-16 all share a common ancestor that RTNLB17-21 do not, therefore a speciation event could have resulted in ER-shaping functionality. RTNLB13 has the closest evolutionary relationship to this common ancestor and, as we have previously shown 24 , contains an essential APH immediately following the RHD that has likely been inherited (along with curvature-generation function) by all the other RTNs in Clades 1-4. RTNLB15 is most closely related to RTNLB13, and our results show that there is a putative APH in this protein that has a similar location, length, sequence, and curvature sensitivity (> 30 nm diameter) to that in RTNLB13. RTNLB10 is the next most-closely related to RTNLB13 according to the phylogenetic tree shown in Fig. 2A, and again contains an www.nature.com/scientificreports/ APH with a similar location, length and sequence, but this isoform appears to have a different curvature sensitivity (only folding in membrane mimetics of diameter < 20 nm). This difference in curvature-sensitivity could be within the error of the measurements, or may be attributed to the fact that ER tubules are highly dynamic, flexible and undergo constant remodeling 41 , and therefore different RTN isoforms may be active at different stages 30 . Additionally, the ER morphology varies between cells so these proteins could have cell-specific roles within A. thaliana. Overall, these results allow us to say with some confidence that the location of the C-terminal APH is likely conserved throughout A. thaliana RTNLB isoforms RTNLB10-13 and RTNLB15-16 (Fig. 6, center panel). RTNLB4 and 7 were selected as representatives of Clade 1 ( Fig. 2A), and both proteins did not contain putative APHs in the same location as those in RTNLB10, 13 and 15; the APHs were instead located ~ 20 residues from the RHD. Proteins with multiple domains often contain short linker regions that are essential in maintaining inter-domain interactions or to facilitate the independent behavior of the two functional domains 42 . While linker regions are challenging to predict from sequence 43 , early studies concluded that linkers lack regular secondary structure, are rich in Ala, Pro and charged residues, and exhibit differing levels of flexibility depending on their biological role [44][45][46] . Both of the linker regions in RTNLB4 and 7 have a higher abundance of negatively charged residues than the equivalent regions of RTNLB10, 13 and 15, especially in regions predicted to lie near lipid head groups (see Fig. 2A), and this may prevent this region from folding into an APH thus shifting it further down the sequence. It is not immediately clear why introducing a longer linker region would be necessary in these isoforms, although members of this group have been shown to localize to other parts of the cell and have functions separate to their roles in ER remodeling, e.g. RTNLB1-4 are thought to also be involved in plant-microbe interactions [30][31][32] . Additionally, a previously mentioned RTN-like protein, Atg40, contains a short helix in its C-term that strengthens binding of this region to a protein-partner 25 . This particular helix is considerably further along the C-term from the transmembrane domains and is rich in acidic residues; these features likely allow this helix to function independently of the RHD. Taken together, these results suggest that the presence of a C-terminal APH is conserved throughout A. thaliana RTNLB1-9 and 14, and that a ~ 20 residue linker between the APH and the RHD was introduced for these isoforms (Fig. 6, top panel).
Consideration of this set of five RTN APH regions (Fig. 5) presents the opportunity to draw comparisons that have not been possible thus far. Our data suggest a range of lipid sensitivities for APH folding across the family of A. thaliana isoforms. The APH regions in RTNLB13 and 15 show no sensitivity to lipid composition and readily fold on highly curved model membranes regardless of charge. This was not the case for the APH regions in RTNLB4, 7 and 10. These APHs did not fold in 60:1:15 DMPC:DMPG:DHPC vesicles which we used to access highly curved bilayer structures 47 . Inspection of the structural models shown in Fig. 5 reveals that these three APHs uniquely contain acidic residues at or near the hydrophobic helical face. Charge-charge repulsion between these acidic residues and negatively-charged PG head groups may be sufficient to destabilize helix formation. Notably, RTNLB13 and 15 contain no acidic residues at this position and experience no charge-dependent destabilization. PG is not a typical constituent of the ER membrane; It is mostly found in the inner mitochondrial membrane where it can also be synthesized 12 and in bacterial membranes 48 . An obvious next step would be measurement of APH folding in the presence of tiny vesicles containing no PG lipids, however as reported by us previously 24 this has not yet been possible due to vesicle instability in the absence of negatively-charged PG 47 . Work is ongoing to better understand the impact of lipid composition on these regions. This work once again highlights how difficult it can be to identify curvature-sensing helical motifs by simply examining amino acid sequences. However, despite the differences in the sequence and location, all five of these APHs share key features that we identified previously as directing curvature-sensitivity and affinity for ER membranes 24 : they are between 14 and 16 residues in length; they contain shallow hydrophobic faces composed of 4-5 residues (Fig. 5); and there is no requirement for the presence of charge for association (in some cases this can prevent interaction). This work provides much needed experimental evidence supporting the presence of a conserved APH module critical for membrane remodeling by RTNs, and more broadly reveals shared features that will facilitate the recognition of other membrane curvature sensors in the future.

Methods
Sequence alignment, phylogenetic analysis and topology prediction. Sequences for all reticulon-like (RTNLB) protein isoforms found in A. thaliana were obtained from UniProt 49 , sequence alignment was carried out using T-Coffee and formatted using Boxshade 28 . A Pairwise Sequence alignment of all the RTNLB isoforms in A. thaliana against RTNLB13 was performed using Emboss Stretcher 28 . Sequence logo indicating the degree of conservation was generated using WebLogo 3 50 . The phylogenetic tree was rendered using the Phylogeny.fr "One Click" mode 51 , with sequences obtained from UniProt 49 . Protein topology was predicted using TOPCONS 39 and helical wheels were obtained from Heliquest 38 .

Synthesis and purification of RTNLB APH peptides.
Five peptides corresponding to putative APH regions of A. thaliana RTNLB4, 7, 10, 15 and 21 were synthesized using F-moc chemistry and purified to 95% purity at Insight Biotechnology Limited (Wembley, UK). The identity, amino acid sequence, and molecular weights for these peptides are shown in Table 1. Additionally, a peptide containing the entire putative C-terminal domain of RTNLB7 (residues 206-244; PMLYEKYEDEIDPIAEKAVIEMK KHYQVFEAKFLSKIPH; M W 4709.50) and the C-terminal 22 residues of this region (residues 223-244; AVIEMKKHYQVFEAKFLSKIPH; M W 2644.17) were prepared. A further peptide derived from the C-term of RTNLB4 (residues 226-250; REIK-KQYAVLDEKVLRKVISKIPRG; M W 2967.59) was also synthesized. The purity of all peptides studied here was confirmed by HPLC and electrospray ionization time-of-flight mass spectroscopy (ESI-TOF-MS microTOF, Bruker). All peptides were stored at − 20 °C as lyophilized powders until use. www.nature.com/scientificreports/ Bicelle and vesicle preparation. All lipids and detergents were obtained from Avanti Polar Lipids (Alabaster, AL) and used without any additional purification. Bicelles with a q value of 0.25 (q = [DMPC]/[DHPC]) and a total lipid concentration of 150 mM were generated using 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) and 1,2-dihexanoyl-sn-glycero-3-phosphocholine (DHPC). An appropriate amount of DMPC was mixed with 25 mM sodium phosphate buffer, pH 6.8, followed by continuously vortexing and centrifuging the mixture until a homogeneous slurry formed, and adding a suitable amount of a 400 mM DHPC solution. This mixture was then subjected to several cycles of centrifugation and vortexing until a clear non-viscous solution was attained. DMPC vesicles were prepared by dissolving an appropriate amount of lipid in 3:1 chloroform:methanol to reach a concentration of 10 mg/mL, and subsequently dried via a rotary evaporator to acquire a thin lipid film. The film was then reconstituted in 25 mM sodium phosphate buffer, pH 6.8, to a final concentration of 3.3 mg/ mL and treated to four freeze-thaw cycles, followed by sonication for 2 min. Prior to use, the DMPC vesicles were extruded through polycarbonate membranes (Avanti Polar Lipids) with pore diameters of 100 nm. Vesicles of diameter < 30 nm containing DMPC, DHPC and 1,2-dimyristoyl-sn-glycero-3-phospho-(1-rac-glycerol) (DMPG) were prepared by adapting a protocol previously described by Yue et al. 47 Briefly, two solutions containing 4:1 DMPC:DHPC and 60:2:15 DMPC:DMPG:DHPC and a total lipid concentration of 50 mg/mL each, were repeatedly vortexed and temperature cycled from 4 to 50 °C in order to dissolve the lipids and then combined to achieve a final composition of 60:1:15 DMPC:DMPG:DHPC. This mixture was then allowed to equilibrate at 4 °C for 24 h before diluting to a final total lipid concentration of 10 mg/mL and subjecting to one freeze-thaw cycle. Vesicles with the same composition but larger sizes were allowed to equilibrate for up to one week instead of 24 h and extruded through either a 50 or a 100 nm polycarbonate membrane before use. All reported diameters were verified via dynamic light scattering.

Circular dichroism. Samples for CD measurements contained peptide at a constant concentration
of ~ 50 mM, as determined by measuring absorbance at 280 nm (A 280 ), in 25 mM sodium phosphate buffer at pH 6.8, and either no membrane mimetic or a membrane mimetic at an appropriate concentration. CD spectra were recorded using a J-1500 spectropolarimeter supplied with a Peltier thermally controlled cuvette holder (Jasco UK, Great Dunmow, UK) and a 1.0 mm pathlength quartz cuvette (Starna, Optiglass Ltd, Hainault, UK). Spectra were measured at 37 °C between 190 and 300 nm, with a bandwidth of 2.0 nm, step resolution of 0.2 nm, scanning speed of 200 nm/min, response time of 1 s and averaged from 16 individual scans after subtraction of the buffer spectrum. To normalize to protein concentration, the machine units of mdeg were converted to mean residue ellipticity. An estimate of reproducibility in our CD measurements is provided in Fig. S5, which contains representative error analysis from three technical repeats for select peptides.
Dynamic light scattering. The hydrodynamic diameter of each membrane mimetic was estimated by dynamic light scattering (DLS) using a Zetasizer Nano-series instrument (Malvern Instruments, UK) at room temperature, with 1 cm path length UV-transparent disposable cuvettes. Mimetic samples were diluted to a concentration of 0.06 mg/ml with 25 mM sodium phosphate buffer at pH 6.8. Each sample was measured six times and each measurement consisted of sixteen accumulations, and were recorded after 300 s equilibration time.
To determine the error in these measurements the standard deviation was calculated for each mimetic; DPC  www.nature.com/scientificreports/